ICAF 2009, Bridging the Gap between Theory and Operational Practice: Proceedings of the 25th Symposium of the International Committee on Aeronautical Fatigue, ... Rotterdam, The Netherlands, 27-29 May 2009

ICAF 2009, Bridging the Gap between Theory and Operational Practice M.J. Bos Editor ICAF 2009, Bridging the Gap betw...

Author: M.J. Bos

22 downloads 1082 Views 41MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

ICAF 2009, Bridging the Gap between Theory and Operational Practice

M.J. Bos Editor

ICAF 2009, Bridging the Gap between Theory and Operational Practice Proceedings of the 25th Symposium of the International Committee on Aeronautical Fatigue, Rotterdam, The Netherlands, 27–29 May 2009

Editor M.J. Bos National Aerospace Laboratory NLR Amsterdam The Netherlands

The papers on pp. 15–54, 123–142, 279–299, 643–660 were created within the capacity of an Australian governmental employment and therefore are in the public domain. The papers on pp. 301–327, 811–837 were created within the capacity of a Canadian governmental employment and therefore are in the public domain. The papers on pp. 187–207, 309–353, 407–425, 909–920, 1035–1068, 1215, 1345–1364 were created within the capacity of an US governmental employment and therefore are in the public domain.

ISBN 978-90-481-2745-0 e-ISBN 978-90-481-2746-7 DOI 10.1007/978-90-481-2746-7 Springer Dordrecht Heidelberg London New York Library of Congress Control Number: 2009927714

© 2009 Springer Science+Business Media, B.V. No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work. Printed on acid-free paper 987654321 springer.com

TABLE OF CONTENTS Preface ................................................................................................................. xiii ICAF General Secretary & National Delegates ................................................. xvii FULL-SCALE FATIGUE TESTING OF AIRCRAFT AND AIRCRAFT STRUCTURAL COMPONENTS A test concept for future aircraft fuselage panels R. Best, Th. Fleischer, M. Götze, M. Sachse and M. Semsch ................................. 3 Marker loads for quantitative fractography of fatigue cracks in aerospace alloys S.A. Barter, L. Molent and R.J.H. Wanhill ........................................................... 15 Substantiations of an airborne composite radome mounted on an aircraft dome I. Kressel, U. Ben-Simon, M. Elyahu, D. Peled, A. David, T. Nachshhon, Y. Gary, M. Interator, G. Ghilai and A. Simon .................................................... 55 AIRWORTHINESS CONSIDERATIONS Development of a “low stress criterion” that eliminate a large portion of the aircraft from damage tolerance based maintenance program for structural repairs required by the new part 26 federal regulation Girindra K. Das and Matthew Miller ................................................................... 61 Damage tolerance philosophy for bonded aircraft structures C.D. Rans and R.C. Alderliesten ........................................................................... 73 First diamond: ‘Damage tolerance’ for the structural honeymoon Steve Swift ............................................................................................................. 91 Survey of structural repairs and alterations in transport category airplanes Michel D. Bode, Walter Sippel and John G. Bakuckas ...................................... 109 Enhanced teardown of ex-service F/A-18A/B/C/D centre fuselages L. Molent, B. Dixon, S. Barter, P. White, T. Mills, K. Maxfield, G. Swanton and B. Main .................................................................................... 123 NDI, INSPECTIONS AND MAINTENANCE Glare teardowns from the MegaLiner Barrel (MLB) fatigue test R.J.H. Wanhill, D.J. Platenkamp, T. Hattenberg, A.F. Bosch and P.H. de Haan ...................................................................................................... 143 v

vi

Table of Contents

30 Years of damage tolerance – Have we got it right? Robert G. Eastin and Jon B. Mowery ................................................................. 169 Compilation of damage findings from multiple recent teardown analysis programs Gregory A. Shoales, Scott A. Fawaz and Molly R. Walters ............................... 187 Detection of arrested crack for foam core sandwich structures using optical fiber sensors embedded in a crack arrester Shu Minakuchi, Ippei Yamauchi, Nobuo Takeda and Yasuo Hirose .................. 209 LIFE EXTENSION AND MANAGEMENT OF AGEING FLEETS Concept of the new A320 fatigue test N. Rößler, C. Peters, O. Tusch, G. Hilfer and C. Herrmann ............................. 225 The use of composite material strips to extend the damage-tolerance life of integrally stiffened aluminium panels A. Brot, Y. Peleg-Wolfin, I. Kressel and Z. Yosef ............................................... 237 Structural integrity of a wing upper skin with exfoliation corrosion Andreas Uebersax, Cyril Huber, Guillaume Renaud and Min Liao .................. 245 Helicopter structural integrity program of Polish Mi-24 Hind helicopters Sławomir Klimaszewski, Andrzej Leski, Krzysztof Dragan, Marcin Kurdelski and Mirosław Wrona ......................................................................................... 263 Airframe life extension by optimised shape reworking – Overview of DSTO developments M. Heller, M. Burchill, R. Westcott, W. Waldman, R. Kaye, R. Evans and M. McDonald .............................................................................................. 279 Evaluation, modification and damage tolerance of an in-service aircraft critical area R.S. Rutledge, D.S. Backman and R.J. Hiscocks ................................................ 301 Procedures for aircraft structural teardown analysis: Development of a best practices handbook Gregory A. Shoales ............................................................................................ 329 Fatigue lifetime improvement of arrestment hook shanks by application of laser peening Mike Leap, Jon Rankin, Jim Harrison, Serena Marley, Lloyd Hackel and Joe Nemeth .................................................................................................. 355

Table of Contents

vii

ADVANCED MATERIALS AND INNOVATIVE STRUCTURAL CONCEPTS Design and testing of advanced composite load introduction structure for aircraft high lift devices Tamas Havar and Eckart Stuible ....................................................................... 365 Hybrid structure solution for the A400M wing attachment frames Matthijs Plokker, Derk Daverschot and Thomas Beumler.................................. 375 Control of crack growth rates and crack trajectories for enhanced fail safety and damage tolerance in welded aircraft structures Phil Irving, Yu E Ma, Xiang Zhang, Guido Servetti, Stewart Williams, Gary Moore, Jorge dos Santos and Marco Pacchione ...................................... 387 Durability and damage tolerance of bonded repairs to metallic fuselage structure John G. Bakuckas, Jr., Keith McIver and Ching Hsu ........................................ 407 Investigation on the design of bonded structures for increased damage tolerance Ivan Meneghin, Marco Pacchione and Pascal Vermeer .................................... 427 An experimental investigation on the fatigue performance of riveted lap joints M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel ................................... 449 Improving the fatigue life of aeronautical single lap bolted joints thanks to the hybrid (bolted/bonded) joining technology E. Paroissien, C.T. Hoang Ngoc, H. Bhugaloo and D. Ducher ......................... 475 An experimental approach to investigate detailed failure mechanisms in fibre metal laminates Riccardo Rodi, René Alderliesten and Rinze Benedictus ................................... 493 Experimental determination of energy release rate of CFRP structures by means of transverse crack tension tests Ll. Llopart Prieto, G. Spenninger and H. Wagner .............................................. 513 Central material concept: towards aircraft wing structure insensitive to fatigue Mohamed A.A. Attia ........................................................................................... 529 Practical applications of improvements in FML crack bridging theory Greg Wilson, René Alderliesten, Riccardo Rodi and H.J.K. Lemmen ................ 539

viii

Table of Contents

Simulation of the stress distribution on fuselage structures for the preand post analysis of curved panel tests Matthias Ziegenhorn, Frank Schulze, Holger Sparr, Karsten Wenke and Thomas Fleischer ............................................................................................... 559 Low cycle lifetime prediction of AL2024 alloy A. Vyshnevskyy, S. Khan and J. Mosler .............................................................. 569 Fatigue and damage tolerance aspects of metal laminates J. Sinke and S.A.H. Johansson ........................................................................... 585 Arall and Glare FML’s: Three decades of bridging the gap between theory and operational practice Cees van Hengel and Peter Kortbeek ................................................................ 601 FATIGUE CRACK GROWTH AND LIFE PREDICTION METHODS Detailed strain field analyses of fatigue cracks in friction stir welded joints H.J.K. Lemmen, R.C. Alderliesten and R. Benedictus......................................... 619 A study of interaction and coalescence of micro surface fatigue cracks in aluminium 7050 W. Hu, Q. Liu and S. Barter ............................................................................... 643 Fracture behaviour of skin materials of civil airplane structures Boris Nesterenko, Grirory I. Nesterenko and Valentin N. Basov ...................... 661 Fatigue life prediction of small notched TI-6AL-4V based on the theory of critical distance Yoichi Yamashita, Masaharu Shinozaki, Hiroshi Kuroki and Yusuke Ueda ....................................................................................................... 685 A case study of multiple-site fatigue crack growth in the F-18 Hornet bulkhead B. Andersson, A.F. Blom, U. Falk, G.S. Wang, K. Koski, A. Siljander, J. Linna, A. Miettinen, K. Vaaraniemi and R. Lahtinen ..................................... 707 Experimental and theoretical comparison of some multiaxial fatigue design criteria in the context of life assessment of rotating parts in turboengines V. Bonnand, J.L. Chaboche, H. Cherouali, P. Gomez, P. Kanouté, D. Pacou, P. Paulmier, E. Ostoja-Kuczynski and F. Vogel ................................................ 743

Table of Contents

ix

A contribution of environmental investigations for Glare riveted joint sizing Thomas Beumler, Bob Borgonje and Jos Sinke ................................................. 765 Advances in crack growth modelling of 3D aircraft structures S.C. Mellings, J.M.W. Baynham and R.A. Adey ................................................. 789 Development of advanced risk assessment methodologies for aircraft structures containing MSD/MED Min Liao, Yan Bombardier, Guillaume Renaud, Nick Bellinger and Terence Cheung ................................................................................................. 811 Crack growth in fiber metal laminates under variable amplitude loading S.U. Khan, R.C. Alderliesten and R. Benedictus ................................................ 839 Effect of residual stresses from shot peening on fatigue strength and threshold to crack propagation of AL 7475 alloy components G. Ratti, U. Mariani, M. Giglio and M. Guagliano ........................................... 859 Experimental validation of stress intensity factor solutions for the pin loaded lug D.R. Child, N.J. Moyle and A.F. Grandt ............................................................ 871 Properties of fatigue crack propagation in friction stir welded 2024-T3 aluminum alloy T. Okada, K. Kuwayama, S. Fujita, M. Asakawa, T. Nakamura. and S. Machida ................................................................................................... 899 Fatigue crack growth in thick plate 7050 aluminum Joel Schubbe ...................................................................................................... 909 Fatigue life evaluation on tension type fitting Nicolas Baréa and Marie Massé ........................................................................ 921 Accelerated fatigue testing on hydraulic shaker M. Fressinet, M. Panis and C. Bordes ............................................................... 931 Methods for FEM analysis of riveted joints of thin-walled aircraft structures within the Imperja project Jerzy Kaniowski, Wojciech Wronicz, Jerzy Jachimowicz and E. Szymczyk ........................................................................................................ 939 On the crack growth behaviour under simple compressive loads of 2024-T3 aluminium alloy M. Krkoška, R.C. Alderliesten and R. Benedictus .............................................. 969

x

Table of Contents

An analytical model for load transfer in a mechanically fastened, double-lap joint Ligieja Paletti, Calvin Rans and Rinze Benedictus ............................................ 987 Simulation of fatigue crack growth in the high speed machined panel under the constant amplitude and spectrum loading Petr Augustin ................................................................................................... 1005 Bonded composite patch to repair metallic structures: Disbond propagation testing and modelling P. Madelpech, S. Juaneda and M. Pradels ...................................................... 1019 Three dimensional crack growth prediction Sarah E. Galyon, Saravanan R. Arunachalam, James Greer, Matthew Hammond and Scott A. Fawaz .......................................................... 1035 Hybrid structures for concentrated load transmission in fibre composites: Initial experiments Pascal Vermeer, René Alderliesten and Rinze Benedictus ............................... 1069 STRUCTURAL HEALTH AND STRUCTURAL LOADS MONITORING Pilatus PC-21 – A damage tolerant aircraft Lukas Schmid ................................................................................................... 1085 Introduction to service of an artificial neural network based fatigue monitoring system Steve Reed, Brian McCoubrey and Andy Mountfort ........................................ 1093 Towards automated flight-maneuver-specific fatigue analysis Juha Jylhä, Marja Ruotsalainen, Tuomo Salonen, Harri Janhunen, Tomi Viitanen, Juho Vihonen and Ari Visa ...................................................... 1121 Service history analysis and teardown evidence – Key elements for structural usage monitoring of an ageing fleet K.A. Lucas and M.J. Duffield ........................................................................... 1135 Development of fatigue life monitoring of RMAF fighter airplanes Wahyu Kuntjoro, M. Suhaimi Ashari, M. Yazid Ahmad and Assanah M. Mydin ............................................................................................ 1155

Table of Contents

xi

Research and development of impact damage detection system for airframe structures using optical fiber sensors Hiroaki Tsutsui, Noriyoshi Hirano, Junichi Kimoto, Takahiko Akatsuka, Hirofumi Sashikuma, Nobuo Takeda and Naoyuki Tajima .............................. 1165 Load factors (N/Z) / roll rate: First comparisons between design and inflight recorded data on Eurofighter typhoon Italian fleet Tommaso Giacobbe and Fabio Sardo .............................................................. 1177 Development of a structural health monitoring system to evaluate debonding in composite adhesive structures Hideki Soejima, Noritsugu Nakamura, Toshimichi Ogisu, Hiroshi Wakai, Yoji Okabe, Nobuo Takeda and Yoshihiro Koshioka ......................................... 1187 OLM: A hands-on approach Stephen Willis ................................................................................................... 1199 F-35 Joint strike fighter structural prognosis and health management: An overview Timothy Fallon, Devinder Mahal and Iain Hebden ......................................... 1215 FATIGUE LIFE ENHANCEMENT METHODS AND REPAIR SOLUTIONS Bonded repair for fuselage damage: An overall benefit to commercial aviation D. Furfari, H.J.M. Woerden, R. Benedictus and A. Kwakernaak .................... 1219 Fatigue & damage tolerance of hybrid materials & structures – Some myths, facts & fairytales R.C. Alderliesten .............................................................................................. 1245 High cycle fatigue of laser beam deposited Ti-6Al-4V and inconel 718 E. Amsterdam and G.A. Kool ............................................................................ 1261 The variable expansion process – A new cost efficient method for cold working fastener holes in aluminium aircraft structures Eggert D. Reese, Anthony L. Dowson and Timothy G.B. Jones........................ 1275 Stress measurements with x-ray diffractometry on aluminium alloys. Determination of the most optimized parameters of the measurement Elżbieta Gadalińska, Jerzy Kaniowski and Andrzej Wojtas.............................. 1285 Delamination growth at interfaces in hybrid materials and structures under various opening modes G. Delgrange, R.C. Alderliesten and R. Benedictus.......................................... 1305

xii

Table of Contents

Developments in metal bonding Jarkko J. Aakkula , Kari Lumppio, Olli Saarela and Tapani Haikola ............. 1321 ENVIRONMENTAL EFFECTS On separating the effect of corrosion on inter-lamellar fatigue on thin sheet AA7079-T6 Sandeep R. Shah and S.A. Fawaz...................................................................... 1345 Authors Index.................................................................................................... 1365

PREFACE While writing this preface for the Proceedings of the 25th Symposium – a silver jubilee! – of the International Committee on Aeronautical Fatigue (ICAF) I wish I already knew that it has been a great success and that everything went smoothly, all lecturers had shown up, the number of participants had been record-breakingly high, etc. But alas, owing to our decision to publish the Proceedings beforehand I can only hope that this will have been the case. What I do know, however, is that both the number and quality of the submitted papers is outstanding. It is therefore fair to state that ICAF continues to be regarded as an important platform for the exchange of information concerning aeronautical fatigue and structural integrity, as has been its aim since its foundation. The vitality of ICAF is emphasized by the fact that authors from 21 different countries have responded to our call for papers. This also indicates that ICAF has the potential to expand beyond its present status of 14 member nations. Poland, having become a full member of ICAF this year, has set a much appreciated example in this respect. ICAF was formed in 1951 in response to the growing concerns regarding fatigue problems in metal aircraft structures. The stated aims of ICAF are to exchange information concerning aircraft structural fatigue and to encourage contacts between people active in this field. To this end a Conference and Symposium are organised every two years for attendance by representatives of industry, universities and institutes, regulatory agencies and operators throughout the world. The two-day Conference consists of reviews of aeronautical fatigue activities presented by the National Delegates of ICAF member nations. It is followed by the three-day Symposium that consists of specialist papers presented by authors with backgrounds and expertise in design, manufacturing, airworthiness regulations, operations and research. The ICAF 2009 Conference and Symposium were held in the last week of May in Rotterdam, the Netherlands. ICAF 2009 was hosted by the National Aerospace Laboratory NLR (which celebrated its 90th anniversary in 2009 – another milestone!), under the auspices of the Netherlands Association of Aeronautical Engineers (NVvL), Delft University of Technology and Stork Fokker AESP B.V. The theme of the 25th ICAF Symposium is “Bridging the gap between theory and operational practice”. This theme came to mind during a discussion with F-16 maintenance engineers of the Royal Netherlands Air Force. From their perspective it is not always clear why the fleet should go through a strict and often burdensome inspection programme during which cracks – fortunately – are only rarely found. In this light it is worth mentioning that the NATO Research & Technology Organisation has recently issued a report about future airframe structural lifing methodologies. This report attempts to reconcile the differences between the current damage tolerance requirements as specified by the US Air Force and the xiii

xiv

Preface

safe life philosophy adhered to by the US Navy and helicopter manufacturers. In this report the focus is on risk-based methods that also account for operational damage findings. A good theme should leave room for more topics than one. Judging by the submitted papers this is the case. In particular, a few authors have used this theme as a framework to describe the introduction of new materials and technologies into the latest generation of aircraft. Notable examples are the Airbus A380, Boeing 787, Joint Strike Fighter and NH90 helicopter. The latter three aircraft are especially important in their major use of composite structures, which are increasingly within the scope of ICAF. This is underscored by the 2009 Plantema Memorial Lecture, named in honour of the founder of ICAF, Dr. Frederick Plantema. The 2009 Lecture was given by M. Jean Rouchon, France’s former National Delegate to ICAF, who recently retired as the head of the Technical Centre for Materials and Structures of the Centre d’Essais Aeronautique de Toulouse CEAT, and is now a private consultant. The title of the 2009 Plantema Memorial Lecture is “Fatigue and Damage-Tolerance Evaluation of Aircraft Structures – The Composite Materials Response”. This special lecture was unavailable for publication in the ICAF Symposium Proceedings (this book). It has been decided to publish M. Rouchon’s Plantema Memorial Lecture under the aegis of the organising committee as an NLR Technical Publication. A new feature to ICAF was a session of three parallel workshops on the last morning of the 25th Symposium. These workshops were included to promote dissemination of a wealth of background knowledge and expertise that resides with senior workers. The workshops were led by Prof. Graham Clark (Aircraft Accident Investigation), Prof. Jaap Schijve (Fatigue Problems of Aircraft Structures) and M. Jean Rouchon (Composite Materials in the light of Damage/Fatigue Tolerance). Delft University and the NLR recently established a biennial award for young and talented academics in the field of aeronautical fatigue. The award is named after Prof. Schijve, to celebrate his 80th birthday, and consists of a medal and prize. ICAF agreed to include presentation of the first Jaap Schijve Award on the last day of the 25th Symposium. It was felt that this would amply demonstrate that the subject of aeronautical fatigue is very much alive and well worth a career. The Organizing Committee of ICAF 2009 consisted of Prof. Rinze Benedictus (Delft University of Technology), Marcel Bos (NLR), Tim Janssen (Stork Fokker AESP B.V.) and Henk van Leeuwen (NVvL). The Technical Committee consisted of Marcel Bos, Frank Grooteman and Dr. Russell Wanhill (NLR); Dr. René Alderliesten, Prof. Rinze Benedictus and Prof. Jaap Schijve (Delft University of Technology); Tim Janssen (Stork Fokker AESP B.V.); and Dr. Simon Barter (DSTO, Australia). The Conference Secretariat was provided by Ms. Gemma van der Windt (D&G Partners). Financial support for ICAF 2009 has been received from various organizations. MOOG, FTI and ACRA Control were the main

Preface

xv

sponsors, partnered by Woodhead Publishing, Springer Science & Business Media and Emerald. The contribution by the City of Rotterdam is also gratefully acknowledged. I would like to thank these individuals and organizations, the authors, workshop leaders, National Delegates and all other persons who contributed to the success of ICAF 2009. In conclusion, I would like to direct the attention of the reader to the contents of this book. It includes 77 selected papers on a broad range of structural fatigue topics that include airworthiness, advanced materials and innovative structural concepts, full-scale fatigue testing, life extension and management of ageing fleets, structural health and structural loads monitoring, inspection techniques and maintenance, fatigue crack growth and life prediction methods, fatigue life enhancement methods and repair solutions, environmental effects, probabilistic modelling and risk analysis. I think the papers are an excellent read for anyone with an interest in aeronautical fatigue and aircraft structural integrity, from both a scientific and practical point of view. I hope you enjoy this book as much as I do. Marcel Bos Chairman of the ICAF 2009 Organizing Committee National Aerospace Laboratory NLR

ICAF GENERAL SECRETARY Dr. Anders F. Blom Defence & Security, Systems and Technology Swedish Defence Research Agency (FOI) SE-164 90 Stockholm, Sweden [email protected]

NATIONAL DELEGATES AUSTRALIA

CANADA

Prof. Graham Clark School of Aerospace Mechanical and Manufacturing Engineering RMIT University, PO Box 71 Bundoora VIC 3083, Australia [email protected]

Jerzy P. Komorowski Institute for Aerospace Research National Research Council Canada 1200 Montreal Road, Bldg. M-3 Ottawa, Ontario K1A 0R6, Canada [email protected]

FINLAND

FRANCE

Dr. Aslak Siljander VTT Technical Research Centre of Finland Machines and Vehicles P. O. Box 1000 FI-02044 VTT, Finland [email protected]

Dr. Thierry Ansart Structures Division Toulouse Aeronautical Test Center (CEAT) BP 93123, 47 rue Saint Jean 31131 Balma, France [email protected]

GERMANY

ISRAEL

Dr. Claudio Dalle Donne EADS Innovation Works 81663 München, Germany [email protected]

Abraham Brot Israel Aircraft Industries Dept 4441, Engineering Div. Ben-Gurion Airport 70100, Israel [email protected]

ITALY

JAPAN

Prof. Luigi Lazzeri Universita di Pisa Department of Aerospace Engineering Via G. Caruso, 8 56122 Pisa, Italy [email protected]

Dr. Hiroyuki Terada (retiring) [email protected] Prof. Nobuo Takeda (as of May 2009) Graduate School of Frontier Sciences The University of Tokyo Mail Box 302, 5-1-5 Kashiwanoha, Kashiwa-shi, Chiba 277-8561, Japan [email protected] xvii

xviii

ICAF General Secretary and National Delegates

POLAND

SWEDEN

Dr. Antoni Niepokólczycki NET Institute Institute of Aviation Al. Krakowska 110/114 02-256 Warsaw, Poland [email protected]

Hans Ansell Saab Aerosystems Saab AB SE-581 88 Linköping, Sweden [email protected]

SWITZERLAND

THE NETHERLANDS

Dr. Michel Guillaume RUAG Aerospace P.O. Box 301 CH-6032 Emmen, Switzerland [email protected]

Marcel J. Bos National Aerospace Laboratory (NLR) P.O. Box 153 NL-8300 AD Emmeloord, The Netherlands [email protected]

UNITED KINGDOM

UNITED STATES OF AMERICA

Dr. John E. Moon Airworthiness & Structural Integrity QinetiQ, Building A7, Room 2043 Cody Technology Park, Ively Road Farnborough, Hants, GU14 OLX, UK [email protected]

Dr. Ravinder Chona Structural Sciences Center US Air Force Research Laboratory (AFRL / RBSM) Wright-Patterson Air Force Base Ohio, USA [email protected]

Full-scale fatigue testing of aircraft and aircraft structural components

25th ICAF Symposium – Rotterdam, 27–29 May 2009

A TEST CONCEPT FOR FUTURE AIRCRAFT FUSELAGE PANELS R. Best, Th. Fleischer, M.Götze, M. Sachse, M. Semsch IMA Materialforschung und Anwendungstechnik GmbH Wilhelmine-Reichard-Ring 4, D-01109 Dresden

Abstract: The new developed airplane structures present a collection of several technologies and materials. This tendency is based on the load tailored design of modern airplanes. Especially, the fuselage contains different affected areas. In consequence of the technology substitutions made, the effort of validation and approval activities grows significantly. So the role of curved panel tests has to be redefined. The capability of special task test devices has improved continuously. With the reached higher simulation accuracy the experimental fatigue and damage tolerance analysis has been enabled for the whole panel including every stiffener detail.

INTRODUCTION CFRP is the material chosen for the fuselage of the next generation civil aircrafts. Compared to metal, this material opens a wide range of tailored structures on the one hand. On the other hand, much research work is still to be done in order to obtain design data or to find new solutions feasible for the production process, too. Testing single fuselage panels is one possibility to get results for comparing design variants. Specimens are loaded more realistically than single elements or details only, with medium effort in terms of money and time. This reaches as far as it allows performing a barrel test more effectively.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 3–14. © Springer Science+Business Media B.V. 2009

R. Best et al.

4

AMBITIONS INSIDE THE TEST PYRAMID The development process for high technology products has been accelerated dramatically in the last decade. One aspect is the strong inclusion of composite materials in every kind of mechanical structures. Thereby, the complexity of investigation and decision tasks increases continuously. For the evaluation and certification process many changes came by the use of nonmetallic materials in pressurised fuselages. On one side, fatigue and damage tolerance (F&DT) behaviour of composite materials switched the focus from the crown panel down to the lower as well as the lower side panels. So the load conditions for single panel cyclic tests switched too. The relevant damage initiations like an impact distinguish completely from fatigue problems in metallic panels. For this reason the test capabilities have to meet new conditions in the case of experimental fatigue and damage tolerance simulations at curved fuselage panels. On the other side several combinations are desired by tailored design under aspects of fatigue behaviour, lightning resistance, effort of technology, production tools, requirements of equipment etc. That means the development teams think about different material classes inside one panel. Therefore the investigation of the interaction between the single elements of a fuselage panel got a higher priority in the panel test too. As a result the selection becomes more difficult for the right bundle of technologies.

Figure 1 Redefined purpose for fuselage panel tests

If the decision for a structural concept depends on tests with more complex specimens under conditions close to service conditions, then a fast medium effort

A test concept for future fuselage panels

5

test method is needed. It should be suited for panels around the whole fuselage. With the curved panel test device up to date such an instrument is ready for single curved fuselage panels with or without frames, with monolithic or sandwich skins. That is a possibility to exceed the concept and design data range by means of effort and costs (see Figure 1) up to the substructures, which contain the relevant details. But the components like fuselage sections will be influenced too by the high accuracy of substructure tests. Topics like small cut-outs can be handled competitively with a curved panel test. Therefore, the barrel test may be supported by previously, or even in parallel, running curved panel tests. Summarising, you can say, that the curved panel test offers, on one hand, the possibility with medium effort to obtain test results for details or elements with a more accurate loading. On the other hand, it is accurate enough to make component tests, such as barrel tests, more effective.

FUNDAMENTALS OF PROVEN TEST METHODS The roots of the full-scale panel tests with pressurisation go back to the middle of the last century. These began with two mirrored panels clamped at the longitudinal boundary which were pressurised cyclically. It seems this way was not practicable, because the next step was a pressurisation of the entire fuselage inside the water tank. In parallel several static and fatigue tests had been performed for complementary single loads like tension and shear at plane as well as curved panels. The first step on the development path of technologies for testing single curved fuselage panels had been the test method for superposing hoop stress and longitudinal tension stress. This condition applies for a single crown panel of the forward fuselage of a commercial airplane. The panel has a floating support at the edge of the pressure chamber and several mechanical loads applied to it:

Flong pi Fframe Fskin

longitudinal tension force over-pressurisation frame tension force circumferential skin force

actively controlled actively controlled actively controlled passive reaction

R. Best et al.

6

Figure 2 Test principle for a single crown panel

This test principle has been widely used to assess data for F&DT of metallic panels. The stress distribution within the panel turned out to be more than sufficient for this purpose. Especially, GAG-cycles (ground air ground) inclusive of hoop loads are suitable to investigate longitudinal crack propagation inside a metallic skin under service loads. Propagation of circumferential cracks has been assessed by applying not just GAG-cycles but flight spectra to the panel, to represent the higher influence of longitudinal loading for this damage scenario. Frames keep out of focus for F&DT except that these are cut for the two bay criteria. This test technology was used as the basis for further developments. One major milestone is adding the capability of shear introduction to the test facility described above. The load introduction principles and boundary conditions stay the same. Shear is applied at the test area edges as shown in the figure below.

FS

shear force

actively controlled

A test concept for future fuselage panels

7

Figure 3 Test principle for a single crown panel, including shear This test technology simulates the stress distribution within the skin for damage tolerance purposes quite well. Crack turning due to the influence of shear was shown as expected by numerical simulations. Nevertheless, this family of single curved fuselage panel test technologies, e.g. a floating supported panel and passive constraints at the skin in hoop direction, may reach its limitations. This could be the case, if the level of shear needed to simulate a certain part of the fuselage is too high, or if sandwich panels that are bending sensitive, are going to be tested. That leads to another test principle which includes a closed section, e.g. a single fuselage panel completed by a test device (D-Box [3] and barrel).

CHALLENGE OF A SINGLE PANEL TEST Change of possibilities synchronizes the expectations The general approach for developing a fuselage panel test device is related to the stress distribution within the test structure. Most cases of investigation require a homogenous membrane loading of the shell. In this case, shell means the eccentrically coupling of a 2D-structure (skin) with 1D stiffeners. Local behaviour is resulting out of this even 2D loading in combination with fuselage panel items being the subject of investigation.

R. Best et al.

8

Figure 4 Local effects shown by ARAMIS at a window frame The main challenge is, to load the test panel correctly, as it needs distributed loads along the edges. Furthermore, the specimen does not have any load introduction points, as it is “cut off the fuselage”. The panel, representing a soft, light-weight structure is sensitive to disturbance of load or stiffness. The capability of the load introduction devices has to cover fatigue loading as well as ultimate or residual strength conditions. That means, for instance for internal pressure loading, the burst safety of the whole test facility. The manageable effort for gaining reliable test results for skin related topics when using the curved panel test technology initiated requests in other fields. For instance various joints have been tested in the past. In addition, the need of panel tests for details like windows or small cut-outs has grown. Furthermore, other panel sectors of the fuselage with more complex load conditions should be possible aims for investigations using single curved fuselage panel test technology. Proven test device for actual as well as future complex investigations The test facilities for floating panels meet these requirements. They can be achieved by using relatively rigid structures for load introduction. The pressure box, for instance, may be a solid steel construction, as the panel is “floating” at it. This is not the case for the closed section, where the pressure box is a loaded structure, too. This pressure box has to fulfil the following tasks:

contain the internal over-pressurisation transfer the shear flow from one panel edge to the opposite one transfer hoop loads as smooth as possible into the panel being mechanically soft in longitudinal direction

A test concept for future fuselage panels

9

The solution for IMA GmbH Dresden was a design of a pressure box of GFRP with an optimised geometry.

Figure 5 Soft pressure box

The test principle changes from “floating panel” to “closed section”. Due to the non-circular geometry of a section composed of a curved panel and the pressure box shown above, correction measures have to be taken. These include

the box geometry itself actively controlled correction forces to assure a membrane loading at the panel edges connected to the pressure box passive devices to support the frames

R. Best et al.

10

Figure 6 Principle of a closed section The latest test device for curved fuselage panels of IMA GmbH Dresden is a design according to the above-mentioned items. It provides load support for skin, stringers and frames. The longitudinal force may be applied bidirectionally. The design limit is 1.5% average longitudinal strain. Other parameters are given in the figure below.

Capability Geometriy •

Radius of

1975…2820mm

•

Width of

1550mm

•

Length of

•

Thicknes up to

50mm

•

No. of Frames

0…6

1900…3400mm

Loads •

Longitudinal (average in membran

•

±2800kN ±1600N/mm)

Torque (average in membran

•

(panel)

Over-pressure

±640kNm ±280N/mm)

0…0.15MPa

Figure 7 IMA’s latest test device for cylindrical fuselage panels

A test concept for future fuselage panels

11

Quality demonstrated at a bending sensitive CFRP-sandwich panel Taking into account the general ideas of a curved panel test shown above, the quality of the test technology is related to the force flow within the panel. To ensure a good quality, the distribution within a non-damaged panel without any special items has to be as even as possible. In order to assess the quality of the test device, a bending sensitive sandwich panel has been used.

Figure 8 Test panel for test quality assessment

Loading the panel with the single load components

Longitudinal force Internal over-pressurisation Shear

leads to evenly distributed strains and therefore force flows in the skin area. Skin bending has been negligible. That means that the pressure box layout and active correction forces worked as simulated. The passive frame supports also fulfilled the expectations, as seen in the strain vs. internal pressure loading in the diagrams in Figure 10. The circumferential strains at different positions of the frame section are displayed, as shown in Figure 9.

R. Best et al.

12

Sa Sf Ss Sg

Figure 9 Strain gauges at frames

Sg_1_4

Sa_1-_4_ca

Sf_2-_4_c

Sg_2_4

Ss_2_4_c

Sa_2_4_ca

Sf_2_5_c

Sg_2_5

Ss_2_5_c

Sa_2_5_ca

Figure 10 Comparison of strain behaviours distributed over thickness as well as location (panel loaded by over-pressure)

A test concept for future fuselage panels

13

More intuitive ‚Post Processing‘ for specific experimental simulations For 2D displacement and strain measurements the optical systems, e.g. Aramis, come closer to the comfortable tools of numerical simulations by these principles. But a free view onto the specimen is necessary. Regarding this main restriction it is easy to get an impression of the displacement and strain gradients along gradients of stiffness or load (see Figure 4). A similar presentation of results is unusual for local measurements by strain gauges or optical fibre grids. In Figure 10 a small selection of the strain results is shown in the classic way of presentation. With diagrams arranged according to the positions of the strain gauges you can get an overview about the behaviour of the structure. Furthermore, you can use different colours to show different strain components of one measuring point within a single diagram. A possibility to show more data within a single view is the usage of a 3D environment. The locations of the gauges are represented by the points of origin of a “multi-diagram”. In the case shown here, strain values of a curved panel, the “planes” of the diagrams represent the direction of the strain gauge grid. The load is acquired at a common axis, normal to the surface.

Figure 11 Local strains at multiple points, acquired over load in a 3D environment Values of certain strains are interactively readable. The flexibility of this tool allows to feed measured and FE data to compare both directly in one presentation.

R. Best et al.

14

Being an engineering gadget on one hand, it allows a quick overview to get an idea, where to look closer with more classic tools on the other hand.

OPPORTUNITIES IN NEAR FUTURE The development of curved fuselage panel test technologies will go on, in the future. The latest step has shown, that tests of details like windows within a single panel are possible with the latest technology. One aim will be to increase panel size in order to allow bigger structures to be tested as part of the panel. An example may be an emergency exit cut-out. This direction of development is the consequent continuation of the history of single panel test technology. Until now panel loading is characterised by constant force flows at the panel edges. In order to test wider panels, as described above, correctly, it seems necessary to add further, non-constant load components to the test technology, such as shear caused by vertical forces or linear distributed longitudinal force flow due to fuselage bending.

REFERENCES [1] B. Borgonje, M. Escobedo Medina (2007) in: Lessons learnt from the FullScale Fatigue Test ‘Megaliner Barrel’ – F&DT Analysis of the Glare Structure, Proceedings of the 24th ICAF Symposium, vol. I, p. 400, L. Lazzeri and A. Salvetti (Ed.), Publ. Pacini, Naples [2] L. le Tellier, F. Repiton (2007) in: Full-scale testing and analysis of Falcon 7X curved fuselage panels with Butt-joints, Proceedings of the 24th ICAF Symposium, vol. I, p. 340, L. Lazzeri and A. Salvetti (Ed.), Publ. Pacini, Naples [3] D. R. Ambur, J. A. Cerrof, J. Dickson (1995) in: D-Box Fixture for Testing Stiffened Panels in Compression and Pressure, AIAA Journal of Aircraft, vol. 326, p. 1382 [4] R. W. A. Vercammen, H. H. Ottens (1998) in: Full-scale fuselage panel tests, Proceedings of the 21st ICAS Congress, TP 98148

25th ICAF Symposium – Rotterdam, 27–29 May 2009

MARKER LOADS FOR QUANTITATIVE FRACTOGRAPHY OF FATIGUE CRACKS IN AEROSPACE ALLOYS S.A. Barter1, L. Molent1 and R.J.H. Wanhill2 1

Defence Science and Technology Organisation, DSTO, Melbourne Australia 2 National Aerospace Laboratory, NLR, Amsterdam, The Netherlands

Abstract: The selection of fracture surface marking methods based on exploiting or altering the required fatigue loads is of much interest for many fatigue test programmes. This is particularly true when crack growth measurements during testing are not possible or insufficiently accurate. In such cases, post-test Quantitative Fractography (QF) of the fatigue crack growth may then be needed, and this can be made possible and/or greatly facilitated by fracture surface markers. Here, several examples of fatigue loadings that create fracture surface markings both naturally, as sometimes happens, and intentionally are presented and discussed. While these examples are from fatigue life tests of aircraft alloy specimens and components, particularly high strength aluminium alloys, under normal environmental conditions (air at ambient temperatures), it is probable that some of the fatigue load histories may provide fracture surface markings for other materials and in other environments. The advantages and disadvantages of the various intentional marking methods are detailed with a view to obtaining guidelines and procedures for optimising quantitative fractography of fatigue crack growth. These guidelines are presented in this paper.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 15–54. © Commonwealth of Australia 2009. Published by Springer Science+Business Media B.V. Dordrecht.

Barter, Molent,Wanhill

16

INTRODUCTION The introduction of Damage Tolerance (DT) principals for determining the useful fatigue lives of aircraft structures calls for eliminating the time it takes a fatigue crack to initiate, since it is to be assumed that cracks are already present at the start of the service life. This assumption supposedly simplifies analysis of each critical area, but real structures are usually complex and not easily fully analysed. As a result, in-service fatigue problems that were not predicted during the design DT analyses often occur. This is particularly the case in tactical aircraft, for which unexpected fatigue cracking has resulted in costly structural repairs and unscheduled frequent inspections [1][3] despite extensive analyses and full-scale fatigue tests to validate them. In retrospect these discrepancies should be expected, since in-service load histories can differ markedly from the design assumptions and full-scale test conditions: the service usage may be more severe, the aircraft’s role may change, the overall weight may increase, and the service environment may be different. The situation appears to be less severe for transport aircraft, but similar problems with service load histories, changed mission types and increasing weights do occur. In addition, though this is not confined to transport aircraft, there may be incentives to fly the aircraft well beyond the original design goals. Owing to the many (unexpected) fatigue problems that can occur during the life of an aircraft, there has recently been a shift in the focus of fatigue research from the original DT concept, as set out in MIL-STD-1530A [4], to a more holistic total life approach that includes fatigue crack initiation mechanisms and short crack growth analyses [5], [6]. This approach somewhat belatedly recognises that in many cases most of the fatigue life is spent while fatigue cracks are initiating and small, a fact known for at least 40 years [7]. Fatigue cracking – in both critical and apparently non-critical locations − may start from small discontinuities in the material (constituent particles and voids), discontinuities produced in manufacturing such as scratches, burrs, nicks and excessive pitting from surface treatments, design inadequacies (unexpected stress concentrations), or degradation processes such as corrosion and fretting. Use of an holistic approach makes it necessary to determine (1) the nature of the fatigue crack origins, (2) the times it takes for cracks to initiate and (3) the growth rates both when the cracks are small 1 as well as large enough to be detectable by NonDestructive Inspection (NDI). All three aspects require post-test or post-service study of the fatigue fracture surfaces by Quantitative Fractography (QF). QF 1

Fatigue cracks less than about 0.5 mm in size often grow faster or more erratically than would be predicted from long crack data [8]-[10]. This is the so-called “small crack anomaly”. Be that as it may, in the present context a small fatigue crack is taken to be one that is below the NDI detection limit, which can be well beyond 0.5 mm.

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

17

endeavours to determine the initial crack sizes and shapes and the subsequent small-to-long crack growth rates. QF of short-to-long fatigue crack growth, in normal air at ambient temperatures, is the main subject of this paper. Other environments are not explicitly considered, although at least one example considers elevated temperature fatigue crack growth. The range of crack sizes includes not only cracks detectable by conventional NDI, but also extends down to the microns to millimetres range, where much of the fatigue life of natural cracks may be spent. QF of fatigue cracks requires an ability to match features found on the fracture surfaces with the loading/environmental history of the specimen or component. This is sometimes possible as a natural consequence of test or service load histories that produce well-defined fracture surface markers. However, when natural markers prove insufficient, QF’s main usefulness lies in reconstructing the fatigue initiation and crack growth processes during tests with modified load histories that produce fracture surface markers. Load history modifications may vary from nothing (natural markers) to several different marker load strategies. This paper gives some examples of natural markers before addressing strategies that have been reasonably successful in marking fatigue fracture surfaces. A fuller set of examples may be found in Barter and Wanhill [11]. The marker load strategies fall into five main categories: 1. Reordering the load spectrum 2. Overload additions 3. Underload additions 4. Constant Amplitude (CA) groups or bands of similar load amplitudes 5. Combinations of categories 1 – 4. The advantages and disadvantages of different strategies are discussed with respect to maintaining the representativeness of the load history. Also, a particular strategy may be successful for one type of material, but not another. The discussions are illustrated by images of fracture surfaces for crack sizes in the microns to millimetres range. These images are intended to give the reader an impression of what to expect when looking for fracture surface markers and undertaking QF. We note here that there have been a number of other studies on selecting load history modifications to obtain fracture surface markers, for example those by Schijve [12], Willard [13], Wanhill and Hattenberg [14] and those found in AGARD-CP-376 [15]. Some of the information from these studies is included here.

18

Barter, Molent,Wanhill

FRACTURE SURFACE MARKER REQUIREMENTS For QF the fatigue fracture surface markers should ideally be readily visible, requiring only simple equipment and little technical and interpretative skill. Unfortunately, this is rarely the case. Crack growth measurements usually need high magnifications and experienced personnel, particularly for interpreting the fracture surface images (fractographs). Since optical measurements are generally less costly than SEM examination, it is beneficial − provided the appropriate equipment is available − if fracture surface markers are visible at magnifications ≤ 2,500 over the entire range of the crack sizes of interest. However, there are several caveats: 1. High magnification optical microscopes with long working distances are much less commonly available than SEMs. 2. If crack growth is to be traced back to small sizes, < 0.1 mm, then SEM fractography will most probably be needed anyway. 3. Even for larger cracks the fracture surface markers may be very finely spaced. This is the case for CA loading where measurements of fatigue striation spacings must be made, and also for very low overall crack growth rates, where again, SEM fractography will be needed [14],[15]. In the light of the foregoing remarks, it is no surprise that SEM examination dominates current QF research. Nevertheless, it is sometimes possible to complete crack growth measurements using optical microscopy. Hence examples of fractographs using this technique are also given. Some of these fractographs are compared with SEM fractographs of the same areas. QF of fatigue fracture surfaces is limited to regions where the crack growth rates are high enough for progression markings or striations to be seen. In practice the SEM limit for QF is about 2×10-8 m using a Field Emission Gun SEM (FEG-SEM) and for the optical microscope about 5×10-7 m, so there should be an aim to mark the fracture surface at intervals greater than these and probably at least 5×10-7 m so that the marks may be easily distinguished. It is important to note here that, although progression markings and striations may be found down to very small spacings, they do not necessarily correspond to the macroscopic crack growth rates [16]-[18]. In other words, the use of very finelyspaced progression markings and striations for determining overall fatigue crack growth rates should be carefully considered and evaluated, since the growth rate indicated by the visible striations may differ markedly from the overall growth rate of the crack at that depth. Number of markers Following from the above, a basic consideration is the number of markers to be included in the fatigue load history. Too few markers will result in rather crude

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

19

crack growth curves, while an excessive number will cause measurement difficulties. There is no straightforward answer as to what is too few or excessive. The number of markers to be selected depends on the crack growth regime, the type of test and its purpose, and also the material. Some broad guidelines are given here, but pilot tests may be required. Marking small cracks: Even though small fatigue cracks less than about 0.5 mm in size can grow faster than predicted from long crack growth data, the crack growth rates will generally be low. Also, since most of the fatigue life is spent in initiating and growing small cracks, their study often involves collecting detailed information on their growth. In other words, there should be many markers, but not too closely spaced. If CA loading is being investigated, the markers need to be at intervals producing resolvable spacings as noted above, as will be the case for VA and spectrum loading (although in these cases it will be more difficult to specify the marker intervals a priori). However, there have been VA and spectrum loading studies of small fatigue crack growth that may be helpful [19],[20], in addition to some of the examples in the following sections. Type of test and purpose: Some examples of differing types of test and purposes are: 1. Basic tests with naturally initiating cracks and long periods of crack growth while the crack is small. The chosen marker intervals are likely to be regular, which has obvious analysis advantages. However, resolution of the marker spacings is an issue for small cracks, see the above remarks. This means that the choice of marker interval can depend on the crack growth regime. For small cracks a larger interval between marker applications may be necessary than for cracks in the NDI-inspectable regime. 2. Threshold crack growth tests, which begin from a starter crack and typically run under decreasing ΔK conditions. It may be desirable to progressively increase the number of cycles between markers, such that their spacings are still resolvable close to threshold. Although markers are not specified for threshold tests, they can be useful as a check on the macroscopic measurements of crack growth. They also assist studying the behaviour of crack fronts, their interactions with microstructural features, and environmental effects on crack growth [21]. 3. Full-scale tests produce a few large cracks (those that cause failure or will lead to the first failures) and many smaller cracks [22]-[24]. Some of these smaller cracks, and others that subsequently occur in service, may later become the focus of attention as “new” critical locations needing total life assessment. Between 30 and 50 markers would be ideal for total life assessments of the smaller cracks. Unfortunately, in complex tests it is not possible to estimate the ideal marker intervals beforehand. When the

20

Barter, Molent,Wanhill

cracks are small, some early marker spacings may be unresolvable. On the other hand, if testing is continued after repair of initial and early failures, as is usually the case, the smaller cracks may grow to failure with many more repeats of the markers. For this reason it is sometimes desirable to include several marker load strategies in the one test. A strategy that works well, provided it is acceptable, is to change or reorder the initial load spectrum block. Figure 1 gives an example of the effect of spectrum changes on full-scale fatigue fracture surfaces. The spectra were quite different, and the fracture surface changes produced were found for cracks varying in size from about 1 mm to more than 100 mm. Another possibility is to change the type of marker after an appropriate number of spectrum blocks and at reasonable intervals (10 – 30 blocks) and thereafter. In addition to spectrum alterations, full-scale tests include discrete events, planned or unplanned, that also affect crack growth, such as strain surveys, failures that momentarily increase the loads on other cracks, and subsequent repairs. If needed, particular markers may be applied immediately upon resuming testing, in order to indicate crack sizes when these events occur, although the events themselves may be sufficient.

Figure 1: Optical image of a fatigue fracture surface in AA7050-T7451 from a fullscale test. Three different loading spectra (phases 1 – 3) were applied during the test, resulting in different fracture surface topographies. The arrows indicate visible repeats of the first spectrum. From [25].

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

21

Material: Given that a material will show markers on its fatigue fracture surfaces 2 , then the finer the microstructure, notably the grain size, the harder it may be to distinguish the markers. This is because local fracture surface deflections occur when cracks cross the grain and subgrain boundaries. To compensate for the finer microstructure the marker intervals can be decreased such that several markers should be visible on the relatively flat fractures between boundaries. This aspect of using markers will be very much influenced by the nature of the material and the type of fatigue test. Pilot tests may be required to finalise appropriate marker intervals before extensive coupon and full-scale tests commence. ASSESSING LOAD HISTORIES FOR INTRINSIC QF “READABILITY” So far, we have discussed marker visibility without going into detail about the markers themselves. In this section we discuss the intrinsic QF “readability” of fatigue load histories, i.e. the visibility of natural markers on fatigue fracture surfaces produced by the required loading (and environment) acting on the test material. It is useful to check this readability before deciding on a deliberate marker load strategy, since natural markers allow the most straightforward interpretation of the test results. For both simple and complex load spectra it is often possible (on at least parts of the fatigue fracture surfaces) to detect markers and markings produced by repeated load spectrum blocks, one or several bands or groups of loads, or one or several of the high loads. Certain intrinsic features of loading spectra can provide easily detected repeating patterns or obvious markers. These intrinsic features are usually similar to those deliberately added to produce marker bands, namely overloads, underloads, bands or groups of loads, or various combinations of any of these. For example, aircraft load spectra may have grouped gust or manoeuvre loads, as in the upper two spectra schematically shown in Figure 1. Some further examples of natural markers for simple and complex load spectra are discussed in the following Subsections, followed by a summary at the end of this Section. The examples have been selected (1) to illustrate the types of markers/markings that may be observed when different spectra are applied to different materials, and (2) what the markers/markings may look like at different crack growth rates, and when observed with either the SEM or a high powered light microscope. Two of the examples are from full-scale tests, and some comments about the difficulties of selecting marker strategies for these tests are given in the summary.

2

Markers are usually visible for high strength aluminium alloys in normal air at ambient temperatures. This is not necessarily so for other classes of materials, e.g. titanium alloys and high strength steels, which may show poorly delineated cyclic crack growth on fatigue fracture surfaces.

22

Barter, Molent,Wanhill

Example 1: simple CA + underload spectrum; AA7050-T7451 Figure 2 shows a schematic of a simple fatigue load spectrum with images of aluminium alloy (AA)7050-T7451 fracture surfaces produced by this spectrum. The spectrum consists of increasing numbers of high-R cycles between fully reversed (R = -1) cycles that cause intermittent underloads. This is a very simplified analogy of varying numbers of gust and intermittent Ground-AirGround (GAG) loads for a transport aircraft lower wing. Both SEM and optical images show repeats of the spectrum blocks. These repeats are readily visible at the locations shown, and it can be concluded that intermittent underloads have the potential to produce well-defined fracture surface markers under some circumstances - in this case where the other cyclic loads have a high mean stress. Furthermore, it is clear from this example that underloads could be used to produce markers during CA tests, specifically when regular striations are not resolvable. This would allow QF of very slow CA fatigue crack growth such as in threshold crack growth tests. The reasons why underloads can produce well-defined fracture surface markers are discussed by White et al. [26].

Figure 2: SEM and optical images of an aluminium alloy fatigue fracture surface produced by a simple CA + underload spectrum, shown schematically. The arrows point to repeats of the spectrum blocks. Example 2: complex manoeuvre + buffet spectrum; AA7050-T7451 shows the spectrum block of a manoeuvre + buffet load history (omitting the buffet loads which were applied in bands distributed throughout the spectrum, from the schematic) and images of an aluminium alloy fracture surface produced by the

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

23

complete spectrum. This example is from a full-scale fighter empennage test [22], for which the spectrum was applied without considering QF readability. However, there were considerable periods of buffeting (high frequency near-CA loading) that made recognisable markings and bands on the fatigue fracture surfaces: with buffeting the total number of turning points was more than 106, compared to 1.5×105 for the manoeuvre loads only. The arrows in the SEM and optical images show repeats of the spectrum blocks. The optical image is from a later stage of crack growth, when some buffet load periods were resolvable as bands. In the SEM image these bands appear as single markings with little or no detail between them.

Figure 3: Manoeuvre loads from a manoeuvre load + buffet spectrum, and SEM and optical images of the fatigue fracture surface produced. The arrows point to repeats of the spectrum blocks. Note the different appearances of the repeat markings (see text). Comparison of the SEM and optical images in Figure 3 shows that well-defined markings during early crack growth can become broader and more detailed as the crack grows deeper, and possibly more difficult to recognise. Then the key to determining repeats of the spectrum blocks can change from recognising entire blocks to recognising markings due to repeated sequences or peak loads within the blocks. Since realistic spectra will include several peak or near-peak loads, and may contain large load excursions that mark like peak loads, see Example 3, there is a potential for confusion. In fact, the peak and near-peak spectrum loads in the

24

Barter, Molent,Wanhill

present example did produce similar markings until local tensile tearing by microvoid coalescence made the peak load more obvious. Example 3: simple flight-by-flight spectrum; AA7075-T651 A comprehensive study of simple flight spectra by Abelkis [27] provides the example shown in Figure 4. One of the load histories studied consisted of repeats of the simulated flight shown in the schematic, and one of these flights is indicated on the accompanying transmission electron microscope (TEM) replica image. This example demonstrates the potential for confusion when large load excursions mark like peak loads. The peak load excursion Cmin−Cmax and the large positive load excursion Dmin−Dmax produced similar “giant” striations. The only difference appears to be the strength of the unloading marks (black lines) at Dmin and Emin: since this image is a TEM replica, the black lines probably represent fissures into the original fracture surface.

Figure 4: A single-flight spectrum and TEM replica image of an aluminium alloy fatigue fracture surface produced by this spectrum [27]. If these two giant striations had not been produced by successive large positive load excursions, they could have been confused with each other. This would also have been possible when viewing the fracture surface directly, using either optical or SEM imaging, since the contrast between large and small fissures is usually not strong in such instruments.

Example 4: complex gust spectrum; AA7175-T736 Figure 5 shows the spectrum block of a gust load history applied to the empennage of a transport aircraft during a full-scale test and images of an aluminium alloy fracture surface produced by the spectrum. The spectrum consisted of blocks of 5000 simulated flights separated by GAG cycles. The three severest flight types, A, B and C are also indicated in Figure 5. In the first instance this spectrum was applied without considering QF readability. However, unanticipated cracking of the vertical stabiliser main hinge fitting led to a fractographic investigation and

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

25

recognition of the spectrum blocks and sub-blocks on the fatigue fracture surfaces [28].

Figure 5: SEM and Optica images of the repeat markings on a fracture surface from an AA7175-T736 forging, as the result of applying the spectrum shown above them. The images are from the same area and include the fatigue crack origin (red arrows). The fractographs in Figure 5 are of the same area and include the fatigue crack origin (large arrows). The white arrows in the optical image show spectrum block repeats marked by the A-flight peak loads. There are also sub-block markings due to the B-flight peak loads. These are less clear in the SEM image. Although the spectrum was complex, the high loads were rather evenly spaced. This enabled easy recognition of the spectrum blocks and sub-blocks when the crack was small. At crack depths beyond those in Figure 5 the blocks became more difficult to detect because they spread out over many facets, and steps between facets, angled away from the average crack plane. The facets were due to local differences in crystallographic orientation between grains or subgrains in the material. In addition, as the crack size increased the various high loads started to produce similar-looking striations.

26

Barter, Molent,Wanhill

At larger crack depths, the A-flight peak load once again became obvious, since it started to cause local tensile tearing by microvoid coalescence. This new natural marker occurred very helpfully over much of the later crack growth. A note of caution about examining cracks from full-scale fatigue tests is worth mentioning here. The optical image in Figure 5 shows that the repeating pattern changes after the last of the white arrows. The spectrum was apparently changed by moving one or two of the B-flights from before the next A-flight to after it. Such changes are not uncommon in full-scale fatigue tests, since the task managers may change the order of some of the loads, flights or flight blocks to suit test or operational requirements. Also, other loads such as strain surveys may be brought forward or delayed. Thus when undertaking QF fractographers need accurate and complete descriptions of the loads applied during a full-scale test. Load changes are usually confusing, but sometimes they may be useful in establishing the exact depth of a crack at a specific time during the test. This may be a particular advantage when fracture surfaces have been damaged either at the beginning or end of crack growth, as can occur in a full-scale structure. To make sure that a load change is recognisable, it is of course possible to add a deliberate marker. Summary of intrinsic QF “readability” The above examples of natural markers on fatigue fracture surfaces were presented in order to make some points about the intrinsic QF “readability” of fatigue load histories. In each case, schematics of the load spectra or parts of the load spectra were included to demonstrate which spectrum features produced the fracture surface markings. Naturally, the examples were chosen to clearly show markings that matched the spectrum features. However, in many cases this matching is difficult or may be impossible to achieve. For instance, the simulated flights in the well-known TWIST and mini-TWIST gust spectra [29]-[30] can be very difficult to identify on a fracture surface [32]. Furthermore, even for the examples given, it was usually found that, although markings were identified at some crack growth rates, they could be very difficult to identify at others. The following points can be made from the above examples and their discussion: 1. Many test load histories produce well defined natural markings on fatigue fracture surfaces during some phases of crack growth. However, some do not. 2. The markings may change their appearance as the cracks grow, and this may make spectrum repeats hard to distinguish at some crack growth rates. 3. Positive load excursions of similar magnitude, but with significantly different peak values, may produce similar markings. Notable examples are the upward parts of GAG cycles, which can produce markings very similar to those caused by high positive gust loads. This similarity can

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

27

result in confusion if large positive load excursions are relied upon for crack growth measurements. 4. In some cases individual loads, groups of loads, or simulated flight sequences can be identified, and these may all be important to the full QF analysis. Unfortunately, there is still the risk of confusion, since prominent markings come from several sources: peak loads, large positive load excursions and groups of loads. From these points it is clear that it is always worth considering deliberate marker load strategies. These will be discussed in the following Sections.

REORDERING THE LOAD SPECTRUM If an unaltered spectrum is difficult to ‘read’, i.e. there are no well defined natural markers, then a simple option may be to reorder the spectrum loads. This can improve spectrum readability, but may change the severity of the spectrum with respect to fatigue crack initiation and growth. Unfortunately, the effects of reordering on readability and spectrum severity are often hard to predict. Hence it is usually necessary to carry out a number of trials and have a strategy for qualifying the reordering effects. There are several methods for determining whether changes to a spectrum have had significant effects on its severity. A full discussion of the ways that spectrum changes can be validated is outside the scope of this paper. However, several analytical and experimental methods are available (one such method using only a few coupons is described in Barter et al. [31]), with the severity comparison by coupon testing usually being the most reliable. We note here that spectrum severity changes are not necessarily bad provided they are validated and understood. Three methods of reordering load spectra to improve or enable QF readability have been suggested: (1) single loads can be repositioned, (2) groups of loads can be repositioned, or (3) single loads can be grouped and then re-positioned. These methods are discussed and illustrated in the following Subsections and examples. A summary is given at the end of this Section. Repositioning single loads Repositioning single loads usually means moving the highest (peak) load close to other high loads to help produce an easily observable marker or group of markers. In general, the higher the peak load with respect to the other loads in the spectrum, the more prominent a marker will be. Since the peak load is usually repeated at

28

Barter, Molent,Wanhill

least once in a spectrum block 3 , placing two peak loads close to each other (clustering) may produce a pair of markers that are recognisable as a band at low crack growth rates (about 10-8 m/cycle) or as an easily recognised pair of marks at medium crack growth rates (10-8-10-6 m/cycle). At still higher crack growth rates the peak loads may cause local tensile tearing by microvoid coalescence, as mentioned in Example 4. This tearing mode causes markings that can be useful in tracking crack growth, see Figure 8 in Example 8. This works best when there are only a few high loads in the spectrum block. Many high loads would produce similar markings that make it hard to identify a particular load. The main problem with clustering peak loads is that crack growth retardation caused by the first peak load may alter the effectiveness of the following peak load (or loads). Also, removing a peak load from another part of the spectrum may result in locally increased crack growth. This is likely to be less of a problem to the overall spectrum severity. Since Example 3 showed that prominent markings are also produced by large positive load excursions, even when not reaching a peak load, then moving the lowest turning points to just before the peak turning points to create the largest positive load excursions should result in very prominent markers. Strictly speaking, this is modifying rather than repositioning single loads. Of course, this modification will also affect crack growth, possibly resulting in considerable crack growth retardation. On the whole, however, clustering two or three peak loads or generating large positive load excursions for each peak load are not very effective at producing markers at all crack sizes [32]. In other words, additional modifications to the spectrum may be required. Repositioning groups of loads As illustrated by Example 4, aircraft fatigue load spectra are usually organised into simulated flights. These can be repositioned in the load histories to create well defined marker bands. This procedure is similar to repositioning single loads, except that each flight may consist of several hundred loads. This may mitigate altering the amount of crack growth retardation, but it still can be a problem. The following gives an example of flight-repositioning changes to the standard manoeuvre spectrum FALSTAFF (Fighter Aircraft Loading STAndard For Fatigue) [33].

3

Repeated peak loads are usually due to spectrum “clipping”, whereby the very highest loads in a spectrum are lowered to a specified level. Clipping is done to reduce crack growth retardation after peak loads, thereby ensuring shorter (conservative) crack growth lives.

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

29

Example 5: modification of FALSTAFF; AISI 4340 FALSTAFF consists of a block of 200 simulated flights typical of fighter aircraft usage in the 1970s. There are two peak loads, in flights 31 and 172. In this example FALSTAFF was modified by moving flight 31 to the beginning of the block, which brought flights 172 and 31 (now flight 1 in the block) closer together during repeated applications of the block.

Figure 6: Low magnification optical image of one of the fatigue fracture surfaces from an AISI 4340 steel wing attachment boom tested with the modified FALSTAFF spectrum shown in the schematic. The arrows link the peak loads in each flight block and two of their markers. This figure also shows QF-obtained crack growth curves for some of the other cracks in the boom. Figure 6 shows an optical fractograph for a crack in a high strength steel component subjected to this FALSTAFF modification, together with a schematic of the modified loading block and QF-obtained crack growth curves for other cracks in the same component [34]. The arrows point to two of the fracture surface markers produced by flights 31→1 and 172. The repositioning resulted in sets of two markers, flight 172 followed by flight 31→1, making it easy to undertake QF for larger crack sizes. However, even with this modification it was difficult to find the sets of markers at small crack sizes. This is not unusual for high strength steels, whose fatigue fracture surfaces are harder to mark than those of aluminium alloys.

30

Barter, Molent,Wanhill

When this test and others were completed it was not expected that the markers could be made more obvious for small cracks. However, QF was often successful, with some difficulty, down to crack depths below 0.2 mm, see the crack growth curves in Figure 6. Using an analytical strain-life analysis, with crack growth retardation considerations and the results of supplementary coupon tests of the original and modified FALSTAFF sequences, it was demonstrated that the spectrum severity was unchanged. This useful result was probably due to the overall length of the FALSTAFF flight block, whereby there were still many loads between the peak loads in flights 172 and 31→1. Repositioning single loads Spectra unamenable to simple re-ordering for improved marking usually contain many high loads that are approximately evenly spaced and of similar magnitude. In such cases either low or high loads may be grouped to form CA bands. Low loads are preferred, since grouping high loads may result in significant changes in spectrum severity. Even so, the grouping of low loads can have an effect on severity. The following example of repositioning and grouping similar single loads to obtain CA bands was one of the modifications to the service-generated manoeuvre spectrum BASIC carried out by Van der Linden [35]. Example 6: repositioning similar single loads into bands; AA7075-T651 This type of modification to the BASIC spectrum was based on those made to the Snowbird spectrum [36]. Similar single loads were extracted from BASIC and grouped to form CA bands. These bands were placed just before the peak load in the spectrum. There were three versions, named Snowbird I, II and III after the original Snowbird modifications. These versions had CA bands containing 427, 1177 and 2204 cycles, respectively. The Snowbird I and II versions of BASIC produced very good markers for crack sizes above about 1 mm. However, the Snowbird III version, with a much larger CA band, marked the fracture surfaces well for all crack sizes. Figure 7 shows some fractographs of the Snowbird III markers.

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

31

Figure 7: Optical and SEM images of CA marker bands on the fracture surface of a AA7075-T651 plate specimen [35]. The marker bands were obtained by grouping 2204 cycles extracted from the BASIC spectrum and placing the group just before the peak load in the spectrum. Summary of reordering the load spectrum The preceding Subsections and Examples discussed and presented markers obtained by three ways of reordering the load spectrum. Reordering has the advantage of not increasing the severity of the spectrum by adding extra loads. Nevertheless, spectrum severity changes due to reordering may still occur, and these have to be evaluated. The following points about reordering load spectra can be made: 1. Good markers may be obtained by repositioning the peak loads or the flights containing the peak loads. 2. In general, repositioning one or two single loads will not produce markers that are easy to recognise for all crack sizes: additional modifications to the spectrum will usually be required. 3. An alternative to moving single loads or flights is to generate CA marker bands from the lower loads in the spectrum. These bands can produce well defined markers. 4. Removal of many low loads from a spectrum may reduce the QF readability. For example, it can be more difficult to analyse fracture surfaces produced with mini-FALSTAFF instead of FALSTAFF [23]. (Spectrum severity changes may also be expected from this type of spectrum modification.)

ADDING OVERLOADS OR UNDERLOADS TO THE LOAD SPECTRUM The addition of overloads or underloads is probably the most common strategy for improving QF readability; particularly the addition of overloads. However, it requires considerable care to avoid significant changes in spectrum severity, or at least to minimise these changes. The following Subsections discuss the addition of

32

Barter, Molent,Wanhill

overloads and underloads, respectively. A summary is given at the end of this Section. Overload additions During fatigue crack growth a positive load excursion typically more than 20% higher than the other spectrum loads may produce a good mark but can also result in crack growth retardation, as has been found notably in aluminium and titanium alloys. Similarly, if a peak load is followed by a large number of lower loads before the next peak load, the crack growth rates between the peak loads may have an average growth rate lower than those without the peak loads. Crack growth retardation may be intrinsic to representative load histories, e.g. gust and manoeuvre spectra, and/or it can be enhanced by adding overloads to the spectrum. From a QF readability point of view, the interest in peak loads, overloads and crack growth retardation is the formation of easily recognisable markers and a distinct change in fracture surface topography following them. An overload cycle causes a relatively large crack extension that produces a fracture surface marking sometimes referred to as a “stretch zone” 4 . These stretch zones can be evident at some stages of crack growth. If the overload causes (1) a change in the microscopic or macroscopic fracture mode, or (2) a reduction in any crack closure (probably roughness induced) behind the crack, owing to stretching the crack tip open, or (3) changes the damage state of the material ahead of the crack, then accelerated crack growth may initially occur and persist over many post-overload cycles. Following accelerated crack growth the growth rate may progressively fall and then gradually recover, resulting in a period of retardation. Such events can significantly change the total severity of both CA and VA loading and result in large changes in crack growth lives 5 . Obviously, as mentioned above, these events can be intrinsic to representative load histories, but their importance shows that the addition of overloads to improve QF readability can disturb the intrinsic crack growth acceleration/retardation behaviour and complicate the interpretation of spectrum severity changes. Intrinsic overloads may occur as a natural part of a test spectrum. The transport aircraft gust spectrum TWIST, even when clipped to level III can cause significant intrinsic retardation [10],[37]. Other results, obtained with both TWIST and miniTWIST, showed that the amount of retardation was greatly dependent on the 4

Not to be confused, despite possibly similar appearances, with the stretch zones that occur when fatigue crack growth suddenly changes to final overload failure. 5 The effects on crack initiation lives are uncertain because these have been little investigated, although 15% overloads at the beginning of the testing of low Kt AA7050T7451 specimens made no statistical difference to the total lives or early crack growth rates. However, placing the same overload prior to the start of each of the loading blocks in this test series produced small but significant retardation during early crack growth [39]. (Early crack growth is often considered a part of crack initiation).

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

33

clipping level for the high gust loads [10],[37],[38], and was more pronounced for thinner sheet gauges. Since even intrinsic crack growth retardation depends not only on the load history, but also the material thickness and peak load clipping levels, it is advisable not to alter the spectrum characteristics or add overloads without a thorough understanding of their effects. The following example of an intrinsic overload shows that overloads can in some cases be of considerable use when QF is required. Example 7: overload required in spectrum loading; AA2024-T8 This example concerns the proof tests required for continued service operation of the General Dynamics F-111 aircraft. The in-service load history for the wings of these aircraft was punctuated with limit load proof tests every 2000 service flight hours. The proof tests were for the wing root structure, which was made from an ultra-high strength steel. The proof test load was considerably higher than service peak loads, in part due to flight restrictions on the aircraft. These proof tests resulted in significant crack growth retardation in many locations of the AA2024T8 lower wing structure, i.e. these loads were beneficial to the fatigue life of the wings in service. Figure 8 shows an example fracture surface produced by the service and proof test loads and subsequent full-scale fatigue test and proof test loads. The progression markings due to the proof test loads are very evident because they resulted from local tensile tearing.

Figure 8: Optical image of a fatigue fracture surface from an AA2024-T8 wing plate: the crack grew from a poorly drilled fastener hole. The fracture surface markings due to the proof test loads are arrowed [40] Underload or compression load additions Fatigue cracks are generally thought to advance only during positive load excursions and then only when the crack faces are no longer in contact, i.e. the

34

Barter, Molent,Wanhill

crack is fully open [41]. The latter assumption may not be entirely correct since parts of the crack maybe in contact even when the crack tip is still open, but the general view suggests that compressive stresses arising from crack closure during negative load excursions will not cause significant changes in crack growth rates. However, there is compelling evidence that high compressive loads, for both long and small fatigue cracks, can cause accelerated crack growth, significant markings on the fracture surfaces, and changes in crack growth planes [26],[42]. These markings can be used as fracture surface markers [26],[43] see Example 8 after the following brief review of the literature. Crooker [44] showed that the compression part of a fully reversed, tension/compression cycle could contribute up to 50% to the fatigue crack growth rates in medium-to-high strength steels and titanium alloys. Topper and Yu [45] measured near-threshold crack growth rates in AA2024-T351 specimens using baseline CA cycling interspersed with compressive underloads at varying intervals. The underloads always increased the overall crack growth rates. Others have reported similar results, and an analogy can be made between these simple VA sequences and the complex VA load histories characteristic of transport aircraft lower wings. For example, De Jonge and Nederveen [46] found that when GAG cycles were removed from TWIST the fatigue crack initiation and growth lives of AA2024-T3 specimens increased by factors of 3.4 – 3.5. Crack growth acceleration due to compressive underloads has been explained using crack closure concepts. Topper and Yu [45] suggested that compressive underloads flatten the asperities on fatigue fracture surfaces of aluminium alloys, resulting in less roughness-induced crack closure. This explanation appears reasonable although there are other hypotheses based on the plasticity states ahead of and/or behind the crack tip. Example 8: underloads in simple CA loading; AA7050-T7451 Fatigue fracture surface markers due to underloads as per Example 1 when applied to AA7050-T7451 plate specimens are shown in Figure 9. A schematic of the CA + variable underloads load sequence is shown in Figure 2, which also shows optical and SEM images of a fracture surface region produced by repeats of this sequence when viewed perpendicular to the fracture plane. The underloads were easily identified. If we now look at this fracture at a high angle (about 60o to the average fracture plane) as shown in Figure 9, we can see that the markers consist of both ridges and depressions (which may be associated with fissures). These markers were produced by the underloads and their nature depended on the local angle of fracture facets with respect to the overall crack growth direction. The depressions and fissures have formed on facets orientated such that they are upward-inclined when viewing in the crack growth direction; and ridges have formed on facets downward-inclined when viewing in the crack growth direction.

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

35

Figure 9: SEM image of repeated underloads on the fatigue fracture surface of an AA7050-T7451 specimen subjected to CA loading with intermittent underloads, see the schematic in Figure 2. The underloads produced well defined markers, depressions and fissures or ridges, depending on the local angles of fracture facets with respect to the overall crack growth direction (see text). The ridges and depressions/fissures were previously noted by Abelkis [39] and Beachem [47], respectively. It is not commonly recognised that fatigue cracks can grow in this asymmetric fashion [26], but it may explain the variability of a single marker when it is followed around a crack front. A marker may change from ridges to depressions, and vice versa, depending on local fracture facet orientations, and may even disappear in some areas. This can also be the case for groups of loads, as will be shown in the next example. Since the markings produced by underloads are generally well defined, their addition to CA or spectrum loading with high average mean stresses, or their association with large positive load excursions, would appear to offer promise as a marker strategy. Small groups of underloads should have little effect on spectrum severity provided they are applied before any high loads in the spectrum, rather than directly or shortly afterwards. Example 9: underload groups in a fighter spectrum; AA7050-T7451 Two different underload groups were added to a manoeuvre spectrum, and it was found that five compressive loads associated with five large positive load excursions produced marker bands for most crack sizes when testing AA7050T7451 specimens. Furthermore, there was only a very slight increase in spectrum severity. Figure 10a gives a schematic detail showing the five compressive loads that resulted in marker bands. The marker bands were identifiable as apparently single markers when the cracks were very small, up to about 200 µm in depth. The optical image in Figure 10b illustrates this, where the Kmax values ranged from about 1 – 4 MPa√m. Multiple

36

Barter, Molent,Wanhill

cracking started from etch pits and coalesced rapidly to form a continuous crack front.

a b Figure 10: Schematic of the five compressive loads that resulted in marker bands shown in the included optical image near the crack origins: Kmax ≈ 1 – 4 MPa√m. The bands appear to be single markers. Note that the markers are discontinuous along the crack fronts. The crack started from multiple origins on an etch pitted surface. Figure 11 a & b show progressive changes in the marker band appearances with higher Kmax values. At Kmax ≈ 7 MPa√m the markers became individually recognisable, Figure 11a; and at Kmax values above ≈ 10 MPa√m the markers appeared quite plastic, Figure 11b. At Kmax ≈ 20 MPa√m the markers were often associated with local tensile tearing (not shown).

Figure 11: View a shows an optical image of the underload marker band at a crack depth of about 0.3mm (Kmax ≈ 9 MPa√m). View b shows an optical image of the marker band at a crack depth of 1 mm,(Kmax ≈ 17 MPa√m). This series of images demonstrates that fractographers have to be aware of the changing appearances of marker bands. In the present example the bands first appeared as single markers, then as a group of five markers (relatively large striations). As crack growth continued the individual marker striations became less easy to distinguish from striations created by other loads in the spectrum. Also, as

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

37

can be seen from Figure 11b, the marker striations can have a rippled microtopography, probably resulting from slip deformation. The ripples could easily be confused with striations from smaller loads in the spectrum, though they often interweave, which true striations never do. Finally, the optical image in Figure 10b shows that markers are not always visible along all parts of the crack front. This possibility was mentioned previously and will be shown again in Example 14. Although the underloads in this example were effective for specimen testing, they were well beyond the negative-g design limit load of the fighter aircraft for which the spectrum had been developed. This made it inadvisable to apply these underloads during a full-scale test, since they could have caused buckling and subsequent static failures in those parts of the structure that were not designed to carry high compressive loads. Summary of adding loads to the load spectrum The following points can be made from the discussions and examples given in this Section: Both overloads and underloads can be used to mark fatigue fracture surfaces. 1. Overloads will generally result in crack growth retardation. 2. Retardation may also occur after the application of severe simulated flights. 3. The extent of retardation can depend to some extent on the material, specimen configuration and spectrum clipping level. 4. Progression markings due to overloads are usually evident. They may also be accompanied by local tensile tearing, which can be very obvious. 5. The use of single or multiple underloads to precede an overload can intensify the overload marker. This allows smaller overloads to be used, thereby reducing any retardation. 6. Underloads generally add to the spectrum severity. 7. The absolute influence of underloads on spectrum severity is usually much less than that of overloads. This would appear to favour adding underloads, but.... 8. Underloads will probably be difficult to apply in complex tests, since the underload level needed for producing good markers will usually be well beyond the negative loads in the rest of the spectrum. This means that buckling leading to static failure may occur without appropriate modifications to the test set-up, e.g. the provision of anti-buckling guides.

ADDING CA BANDS TO THE LOAD SPECTRUM Groups or bands of CA loads may be used to obtain markers on fatigue fracture surfaces produced under both spectrum (VA) loading and CA loading. The bands usually contain many CA cycles, whose maximum and minimum loads may differ

38

Barter, Molent,Wanhill

from those of the VA or CA loading to which they are added. Obviously, the choice of CA marker band load levels will depend on the overall load history. The markers produced by CA bands typically depend on: 1. The relative peak load levels of the bands and the overall load history. 2. The stress ratio: R, of the marker bands compared to the mean R of the loads preceding and following the bands. 3. The number of cycles in the bands. 4. The crack sizes and growth rates when the bands are applied. Relative peak load level of a CA band CA bands with peak load levels significantly different to those of the surrounding VA or CA load histories can produce marker bands that are readily visible owing to changes in fatigue fracture topography. This effect occurs on the fracture surfaces of several important alloy groups: high strength aluminium alloys, high strength steels and some titanium alloys [48],[49],[50]. In particular, for aluminium alloys the overall roughness of a fatigue fracture surface typically increases with increasing crack tip stress intensity factors. This means that applying a CA band with reduced Kmax compared to the surrounding load history will generally be expected to produce a marker band that is flatter than the adjacent fracture surfaces, and therefore be observable. This is not always true, since at low growth rates CA marker bands may show a fine river pattern that makes them slightly rougher in comparison to the surrounding VA crack growth. This may be seen in the following example. Example 10: CA bands in a manoeuvre spectrum; AA7050-T7451 Figure 12 shows two SEM images of fracture surface regions produced by repeats of the mini-FALSTAFF spectrum with added CA bands of R = 0 loading where the peak load in the bands was 80% of the peak load in the spectrum. The CA bands consisted of 800 cycles and were added every 15 loading blocks during the Flaw Identification through the Non-representative Application of Loading (FINAL) programme [51]. The number of cycles in the CA bands was chosen to obtain about the same amount of crack growth as a single mini-FALSTAFF loading block.

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

39

a b Figure 12: SEM images of CA bands between flight blocks of mini-FALSTAFF V2 loading on the fatigue fracture surface of an AA7050-T7451 specimen. Some of the flight blocks are arrowed in image a where band is small. Image b shows the band in more detail at a greater crack depth. The CA bands are indicated by brackets. Figure 12a shows repeats of some of the loading blocks (arrowed) followed by a CA band (bracketed). Note the smoothness of the flight blocks fracture surface in contrast to the fine river pattern on the CA band fracture surface. Figure 12b is from a region of high growth rates, where both the flight blocks and a CA band (bracketed) produced fine river patterns on the fracture surface. Even so, the CA band is distinguishable from the flight blocks. In this case, further crack growth allowed striations to become visible in the CA bands. The constant spacing of these striations, unlike the variable striation spacings in the flight blocks, made the CA bands fairly easy to detect at faster crack growth rates. This situation persisted until the spectrum peak loads caused local tensile tearing, which increased the overall fracture surface roughness. In turn, this resulted in local variations in the CA band striation spacings. Stress ratio effects: CA band R compared to load spectrum R Besides fine river patterns, CA bands may produce other fracture surface features that allow their detection against a backdrop of other CA or VA crack growth. For high strength aluminium alloys large changes in the R values of CA loading can cause sudden and quite steep changes in the fracture plane at many places around the crack front. The steepness of this change appears to depend on the R ratio, the crack tip Kmax value, and the crystallographic orientation of the grain that the crack is currently growing in [26]. An example is given in Example 11.

40

Barter, Molent,Wanhill

Another possibility, adding another dimension to marker band detection, is CA band R-induced changes in oxidation colours on the fatigue fracture surfaces of gas turbine alloys tested at elevated temperatures. An example is given in Example 12. For aluminium alloys, changes in the fracture plane can also occur owing to transitions from CA to VA loading and vice versa, notably when the CA band R is 0.5 or higher, and significantly different to the average R of the VA load spectrum. This can be seen in Figure 12a. Example 11: R value changes for CA bands; 7050-T7451 Figure 13 shows a load history schematic consisting of four CA bands with differing R but the same peak load, together with an SEM image of the resulting fatigue fracture topography for an AA7050-T7451 coupon. The changes in the crack growth planes and their appearances are striking. In particular, the changes in crack growth planes often provide excellent markers.

Figure 13: SEM image of R-induced changes in crack growth planes on the fatigue fracture surface of an AA7050-T7451 specimen tested with four CA bands. These bands had R values of 0.7, 0.3, 0.5 and 0 respectively. The large change from R = 0 to R = 0.7, which occurs twice in the image, was particularly effective at producing a marker. Example 12: R value change for CA bands; Inconel 718 Fatigue fracture surface markers due to large changes in R were investigated for the nickel-base superalloy, Inconel 718, tested at 600 ºC in air [49]. The load history consisted of repeats of 1000 cycles at R = 0.1 and 1000 cycles at R = 0.8 with the same peak load, whereby the R = 0.1 load levels represented Low-Cycle Fatigue (LCF). The specimen configuration simulated the blade root notch at the rim of a turbine disc. Figure 14 shows SEM and optical images of one of the fatigue fracture surfaces. The R = 0.8 CA bands resulted in some evident markers owing to changes in the oxidation colours as well as the topography.

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

41

Figure 14: SEM and optical images of CA marker bands on the fatigue fracture surface of an Inconel 718 notched LCF specimen tested at 600 ºC in air. Number of cycles in the CA bands and their periodicity The widths of the CA bands can be important for their recognition. On the one hand, if the bands contain only a few cycles, they may be very difficult to detect near the crack origin and at low growth rates. On the other hand, if the bands contain many cycles they may become a major part of the overall crack growth process and lessen the test validity. This effect can be alleviated by altering the CA band periodicity, i.e. the ratio of CA bands to flight blocks; but it also introduces the complication of variable CA band contributions to the overall crack growth process. Choosing an appropriate number of CA cycles will depend partly on the test requirements. If it is considered necessary to insert or add CA cycles for every flight block, a few (tens to hundreds) of CA cycles can be used to good effect in combination with a few underloads. However, it may not be necessary to add a CA band for every loading block. Adding CA bands after a number of flight blocks will lessen their effect on the overall crack growth, thereby allowing more cycles within the bands and making them easier to detect. Demonstrations of this can be found in Examples 14 and 16. Variable CA band contributions to crack growth The variability of CA band contributions to the overall crack growth process is a complication that may require detailed analysis. The cause of varying CA band contributions to crack growth may result from the change in crack growth plane associated with (1) the introduction of the CA band, (2) changes in crack tip stress intensity factors, and (3) changing crack closure conditions.

42

Barter, Molent,Wanhill

Be that as it may, there is an incentive to choose a CA band width and R value to minimise the variable contributions of the band to crack growth. It would seem best to choose a CA band R value close to the average R of the surrounding loading blocks. On the other hand, from Example 11 we have seen that large differences in R may produce the best markers. One way out of this dilemma is to insert narrow CA bands with differing Rs into broader CA bands whose R is close to the average R of the surrounding VA loading. This should improve the marker band visibility. Crack sizes and growth rates when the CA bands are applied The crack sizes and hence crack growth rates at which CA marker bands are applied can have a strong influence on their relative contributions to the overall crack growth process. This is not the only effect, since the visibility of the CA bands on the fatigue fracture surfaces also varies. For small crack sizes and low growth rates the fracture topography of the markerbands may blend in with the surrounding VA loading, while for long cracks and high growth rates the generally rough topography of the fatigue fracture surface can obscure the more subtle changes due to changing from VA to CA loading and vice versa. For most Damage Tolerance (DT) analyses and assessments the foregoing problems are likely to be minor, since it is the mid-range of crack sizes and growth rates that are of most interest. On the other hand, a total life approach – which is becoming more favoured, as noted in the introduction to this paper − must account for at least the early stages of crack growth. Summary of adding CA bands to the load spectrum As stated at the beginning of this Section, groups or bands of CA loads may be used to produce markers on fatigue fracture surfaces generated under both spectrum (VA) loading and CA loading. From the subsequent discussions and examples the following general points can be made: 1. CA bands with peak load levels significantly different from those of the surrounding VA or CA load histories can produce readily visible markers owing to changes in fatigue fracture topography. 2. CA bands may produce other topographical features allowing them to be easily detected against a backdrop of other CA or VA crack growth. Notable examples are steep changes in the fracture plane. 3. Fracture plane changes may also occur when the CA loading is changed from one R value to another, particularly if the change is large. 4. Large differences between the CA band R values and the average R of the surrounding VA loading are favourable to producing good markers. 5. Conversely, similar R values for the CA bands and VA loading are unfavourable for good markers, although similar values are desirable to minimise varying contributions of the CA bands to crack growth. This dilemma can be alleviated by inserting narrow CA bands with differing R

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

43

within broader CA bands whose R is close to that of the average R of the surrounding VA loading. 6. The crack sizes and growth rates at which CA marker bands are applied can have a strong influence on their relative contributions to the overall crack growth process.

COMBINATIONS OF MARKER STRATEGIES The previous Examples have shown that there are many ways in which reasonably straightforward changes and/or additions to a load history can result in good markers on fatigue fracture surfaces. In this section we shall consider some progressively more complicated combinations of strategies. As before, these strategies pursue the goals of (1) obtaining good markers while minimising their effects on the basic load histories, and (2) being able to account for the effects of the markers on the overall crack growth. Minimising the effects of markers on the basic load histories is particularly important for fighter aircraft components, which are generally highly loaded. The fatigue lives are relatively short, compared to transport aircraft, and small critical crack sizes may be expected. Combinations of CA bands can be used to produce markers that are easier to find than those produced by a single CA band. Some more examples are given in Examples 13, 14 and 15. Example 13: R value changes for CA bands in VA loading; AA7075-T73 This example, Figure 15, concerns the addition of two CA bands with different Rs to a VA load spectrum used in testing an aircraft rudder hinge fitting made from AA7075-T73. There were several constraints on the use of marker loads: 1. The load history had to be all-tensile to be representative. This prevented adding negative underloads. 2. The critical location was shot peened. This prevented adding peak loads (overloads) higher than those in the spectrum, since the overloads might affect the peening residual stresses, making the test results meaningless. 3. The spectrum flight block was relatively short. Adding a marker band every block carried the risk of an excessive number of markers if cracks initially grew very slowly owing to the shot peening. Not only would a large number of closely-spaced markers be difficult to distinguish, but they would also contribute too much to the overall crack growth. 4. Optical QF would be preferable to SEM QF, in order to reduce time and costs. These constraints led to choosing a very specific marker combination:

Barter, Molent,Wanhill

44

• • •

•

The markers were two CA bands with widely different R values. This combination was expected to result in slight changes in fracture plane that would show up well during optical fractography. The CA bands consisted of 500 R = 0.5 cycles followed by 200 R = 0.1 cycles, whereby the maximum load was 0.8 of the peak load in the spectrum flight block. The total number of cycles in the CA bands was chosen to obtain similar amounts of crack growth from the two bands in comparison to one test loading block: the flight block contribution was estimated using an analytical crack growth model. The marker bands were applied every ten flight blocks. This was to ensure that the markers would not be too close together and that errors in the crack growth equivalency of the two CA bands and one flight block would be small in terms of the overall crack growth. In other words, the entire load history was repeats of a “superblock” closely equivalent to 11 test loading blocks. This simplified the QF analysis.

Figure 15 shows a schematic of the “superblock”, with the two CA bands at the end, and an optical image of the repeat marker bands on the fatigue fracture surface of a rudder hinge fitting. Note that the marker bands are evident, as intended, but that the band can disappear at some locations owing to changes in grain orientation with respect to the crack growth direction.

Figure 15: Schematic of the modified spectrum for the rudder hinge fitting, showing the ‘superblock’ consisting of ten flight blocks followed by two CA bands, and an optical image showing the marker bands at the end of two superblocks.

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

45

Combinations of CA bands with counters The idea of including “counters” in CA marker bands has been proposed and used by several investigators [52],[53],[54]. Example 14 began as a development of such a strategy. However, after several pilot tests the variations were abandoned in favour of the combination that gave the best markers at most crack sizes, for the reasons given in this example. Example 14: CA bands with counter cycles; AA7050-T7451 In this example CA bands consisting of 400 R = -1 cycles (CA peak load = 0.5 of the VA peak load) were added to a VA manoeuvre loading spectrum, and counter cycles were added within the CA bands. The counters were varying low numbers of cycles with higher peak loads and lower underloads than those of the CA bands. This strategy worked reasonably well for identifying particular marker bands, albeit over restricted ranges of crack sizes (0.1 – 1 mm) and medium to fast growth rates (2x10-8 – 2x10-6 m/striation for the counter cycles). Figure 16 shows an SEM image of a marker band on the fatigue fracture surface of an AA7050-T7451 specimen tested primarily under VA loading. The marker band topography was generally smoother than that produced by the VA loading, as intended, but there were some interesting effects owing to the inclusion of 7 counter cycles: 1. The 200 CA band cycles after the counters resulted in significantly more crack growth than those before the counters. This suggests that the counters markedly changed one or more parameters (i.e. crack closure level, residual stresses, cyclic plastic zone size, or amount of slip at the crack tip) controlling the local crack growth rates. The consequent increase in the contributions of the CA bands to the overall crack growth is undesirable 2. The generally smoother fracture topography of the CA band was less evident after the counters, making it sometimes less easy to make QF measurements. The first effect means that counters increased the CA band contributions to overall crack growth. This is undesirable, as is the second effect, but both are considered tolerable in view of the following main advantage of counters: • Readable counters allow crack growth increments to be located precisely during the total life of a crack, whereas the positions of straightforward marker bands can be uncertain if only a few are detectable. Owing to this advantage, the CA band + counter strategy should certainly be considered, even if used only two or three times during the growth of a crack. In the light of the comments at the beginning of this Subsection, this possibility would seem particularly relevant to QF for DT investigations of slow growing cracks. We conclude that more work needs to be done on selecting the optimum R values for CA bands and counters.

46

Barter, Molent,Wanhill

Figure 16: SEM image of a CA band containing counter cycles on the fatigue fracture surface of an AA7050-T7451 specimen.

Example 15: spectrum modification + CA bands; AA7050-T7451 In testing of used Centre Barrel (CB) bulkheads from F/A-18 A/B aircraft [51], spectrum modifications were evaluated using low Kt “dogbone” specimens, resulting in selection of a modified mini-FALSTAFF spectrum where the flights with the high loads were moved closer together as was done in Example 5. Although the spectrum modification was considered successful for most crack sizes and growth rates in simple specimens, it was envisaged that the full-scale tests would provide some problems with respect to QF readability: 1. The modified mini-FALSTAFF flight block markings would be very closely spaced on the fracture surfaces near the origins of any slowgrowing cracks. 2. From full-scale test experiences [36],[47],[55],[51] parts of the fatigue fracture surfaces of some, if not many, of the cracks should be anticipated to be in poor condition in comparison to specimen tests. This is because full-scale tests are longer running and much more complicated than specimen tests, with greater opportunities for fracture surface deterioration owing to local corrosion and contamination by e.g. fretting products, sealant, paint, oil and grease.

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

47

3. The tests were to be continued after local failures and subsequent repairs, making it difficult to externally observe any additional cracking and hence essential to be able to do QF measurements. To improve overall QF readability it was decided to add CA marker bands to the spectrum. The CA bands consisted of 800 R = 0 cycles, whereby the maximum load was 0.8 of the peak load in the spectrum loading block. The first CA band was applied after 30 loading blocks and subsequently after every 15 loading blocks. This combined load history was intended to enable QF measurements of slow growing cracks via the CA band positions, and faster growing cracks via the flight blocks. The choice of R = 0 for the CA bands was dictated not only by their ability to provide markers, but also by a more subtle consideration. Many of the cracks occurring in the full-scale tests would likely be generated by secondary bending loads, mainly at integral flanges and stiffeners in the CB bulkheads. To cause cracking, the secondary bending loads in the bulkheads would have to be predominantly tensile, even though the primary loading condition might be dominated by the compressive loads applied to the bulkheads. This reversal of the primary loads would render the CA (R = 0) band cycles ineffective, since they would then be entirely compressive. On the other hand, if the secondary bending loads derived directly and predominantly from the primary tensile loads, then the CA band cycles should cause crack growth. In other words, the presence or absence of CA marker bands on the fatigue fracture surfaces would be diagnostic for the secondary bending loads. Figure 17 gives an example of a fracture surface from the full-scale tests. The first image in this Figure illustrates the usefulness of the CA marker bands, since the block markings were very fine. The second view in Figure 17 shows a fracture surface for a bulkhead crack with an unanticipated marker owing to a transient load caused by failure of an adjacent bulkhead. Such incidental markers can help establish the sizes of any cracks influenced by transient loads at the times of major failures of integral components, or partial failures during tests on multiple load path components. These markers can also aid in determining whether subsequent crack growth was affected by the failures or not. Summary of combinations of marker strategies Combinations of CA bands: 1. A combination of two CA bands with widely different R values can provide very good markers over a wide range of crack sizes, and down to a few tens of microns from the fatigue origin. This combination works very well for high strength aluminium alloys and reasonably well for high strength steels.

48

Barter, Molent,Wanhill

Figure 17: The first optical image shows repeating flight blocks and several CA marker bands (arrowed). In detail the image shows the 2nd, 3rd, 4th and 5th CA marker bands with 15 repeats of the modified mini-FALSTAFF loading block between each. The crack originated from surface pitting under an Ion Vapour Deposition (IVD) coating. The second optical image shows the 1st CA marker band, repeats of loading blocks, some of which are arrowed, and an incidental marker owing to a transient load caused by failure of an adjacent bulkhead.

2. The fracture plane changes between bands result in good optical contrast that makes QF easy using a high powered optical microscope. 3. The contrast between bands is less visible during SEM examination, but the bands are still readily found. 4. The fracture plane changes along any single marker band are intermittent. This indicates that the changes depend on the local crystallographic orientations of the grains through which the crack grows. Combinations of CA bands with counter cycles: 1. The use of counter cycles within CA bands may provide reasonably good markers over restricted ranges of crack sizes (0.1 – 1 mm) and crack growth rates (between about 2x10-5 – 2x10-3 mm/striation) for the counter cycles. These ranges could probably be extended by using counter cycle bands. 2. Counter cycles combined with CA bands tend to contribute too much crack growth unless the combinations are applied infrequently, e.g. after certain numbers of VA flight blocks.

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

49

Spectrum modification + CA bands: 1. CA marker bands added periodically to VA flight blocks can be very useful for QF of fatigue cracks in components undergoing full-scale tests. The CA bands are especially useful for small crack sizes and slow crack growth rates; and also for fracture surfaces partially obscured by local corrosion and contamination (sometimes unavoidable in these long running and complicated tests). 2. Judicious choice of the CA band R value can aid in determining the source of the local secondary bending loads that may be responsible for some of the cracking.

CONCLUSIONS, RECOMMENDATIONS AND SUGGESTIONS The examples and discussions presented have reviewed a number of methods of providing fatigue fracture surface markers to aid QF of fatigue crack growth. These methods are based on load changes, including reordering the basic load histories and/or adding loads to them. The main reason why we have prepared this paper is to provide some guidelines for obtaining recognisable markers for a variety of load histories and crack growth regimes in the locations of interest for full-scale tests. As has been indicated, this may be no easy task. There is much to consider, and not all candidate marker load strategies will result in immediate success. Hence it will often be necessary to carry out pilot tests using simple coupon specimens. The actual approach and choice of marker load strategy that is finally adopted will depend on several factors, including the main purpose of the test, the component or structural configuration, the expected load history, the crack growth regimes of interest, the materials involved, and the test duration. Test plan for providing post-test QF-readable fatigue fracture surfaces The test plan for providing post-test QF-readable fracture surfaces could proceed in the following stages: 1. Examine the spectrum to be applied, noting its form. Spectra with notably different load groups may provide well defined natural markers on fracture surfaces. These spectra may contain large numbers of grouped low loads with periods of higher loading, e.g. the lower wing loads of military transport aircraft in operations where occasional tactical manoeuvring is required, or in many civil transport operations where severe gust loading is included in occasional flights. 2. Consider re-arranging the spectrum to place high loads or high load excursions closer together, or group large numbers of low similar loads. These can be easy modifications if the high loads in the spectrum can be arranged to provide evident markers without changing the validity or severity of the spectrum, or if such changes can be accounted for.

50

Barter, Molent,Wanhill

3. If rearrangement is insufficient or unfeasible, then consider what would be the best type of addition to the spectrum: overloads, underloads, CA loading groups, or combinations of these. All have been successful in producing observable markers on fatigue fracture surfaces. Generally it can be said that: • Overloads may lead to crack growth retardation, which is difficult to account for and may make the test unconservative. • Underloads may cancel crack growth retardation and lead to conservative results, but they are some times impractical to apply. • CA bands, if well chosen, will produce good markers, although to do this they may need to contain many cycles and thus consume a notable percentage of the crack growth life. Additionally they may not have a consistent effect over the full extent of crack growth. A two-phase CA band, when well chosen, will generally give very good marker bands on fracture surfaces from the crack origin to the failure depth: a few tens of microns to tens of millimetres! 4. Determine the number of markers required to achieve the test aims. 5. Determine whether a secondary marking scheme will be required, either to account for long-lived cracks or as an aid to defining the position of the markers that can be observed. 6. When spectrum modifications and the addition of marker loads are contemplated it is always advisable to carry out some simple testing at a representative load level, in order to confirm or determine the effects on the spectrum severity. Here QF is again invaluable. 7. Aim to produce markers that have good optical contrast at relatively low magnifications (compared to the capabilities of an SEM). This will allow relatively easy QF with either a low- or high-powered optical microscope. 8. Remember that even with a very effective marker strategy there will be some areas around any particular marker band where the band is not observable. Marker load strategies Marker loads for CA load histories: The following additions to CA load histories will produce markers on most aircraft alloy fatigue fracture surfaces: 1. Single underloads. These may be considered as single changes in R. Underloads are particularly effective for cracks growing under high-R CA loading. They are not expected to provide good markers for cracks grown at lower R values. 2. Overloads. These will be particularly effective for all CA loadings. Unfortunately they can be expected to change the effectiveness of the CA loading, though there may be situations where this is acceptable. 3. Groups of CA loads with different R. If the R value is significantly different to the basic CA R value, then such a change may be effective. The number of cycles required in a group (CA band) can be estimated by simple fatigue prediction methods, although pilot tests will usually be

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

51

necessary to validate the effectiveness. The effect of the CA band R on the basic crack growth rates may also be questioned. For this reason it may be useful to evaluate the effect of varying the number of basic cycles between the CA bands. Marker loads for VA load histories: The following additions to VA load histories will produce markers on most aircraft alloy fatigue fracture surfaces: 1. Flight repositioning. Placing flights with high loads close together and/or flights with long runs of similar loads together has been found to be effective. Gust spectra usually have many low loads that if grouped will produce good markings. Thus both options are feasible. The extent of the rearrangement will depend on the aims of the testing and the ability to interpret the loading changes. 2. Underloads. Groups of underloads may be useful, while single underloads will probably be unsuccessful. 3. Overloads. These will be particularly effective if above about 120% of the peak load in the spectrum. Unfortunately, they can be expected to change the effectiveness of many subsequent loads, although there may be situations where this is unimportant. 4. Groups of CA loads. If the CA load mean R value is significantly different to that of the spectrum, then the addition of groups of CA loads can mark by causing crack path changes. For CA loads with a similar mean R the crack path change still can be sufficient to provide visible markers. The number of CA cycles required will usually be significant, and this needs to be assessed in view of the aims of the testing – typically, the smaller the group of CA loads, then the harder it is to find the markers. The effect of the CA group on the crack growth rate of the spectrum loading may also be a concern. 5. Combinations of the above. In some cases, combinations of marker strategies may be useful, but these need to be carefully considered in the light of the test objectives.

ACKNOWLEDGMENTS We are much indebted to B. Dixon (DSTO), J. Schijve (TU Delft) and T. Hattenberg (NLR) for assistance in providing some of the figures in this paper.

REFERENCES [1]

[2]

Jones, K., McCowin, R., Harris, B. (2006), In: Proceedings of the USAF Aircraft Structural Integrity Program (ASIP) Conference, San Antonio, Texas, November 28 − 30. Moore, C. D. (2007), In: Proceedings of the USAF Aircraft Structural Integrity Program (ASIP) Conference, Palm Springs, California, December 4 − 6.

52

[3] [4] [5]

[6]

[7]

[8] [9]

[10] [11] [12] [13] [14] [15] [16] [17] [18]

[19]

[20]

Barter, Molent,Wanhill

GlobalSecurity (2007), http://www.globalsecurity.org/military/systems/aircraft/f-16-life.htm MIL-STD-1530A (11) (1975), Airplane Structural Integrity Program, Airplane Requirements, Military Standard, USAF. Glaessgen, E. H., Saether, E., Yamakov, V. (2008), In: Proceedings of the Eleventh Joint FAA/ DoD/ NASA Conference on Aging Aircraft, Phoenix, Arizona, April 21 – 24, 2008 (to be published). Liao, M., Benak, T., Renaud, G., Yanishevsky, M., Bellinger, K., Mills, T., Prost-Domasky, S., Honeycutt, K. (2008), In: Proceedings of the Eleventh Joint FAA/ DoD/ NASA Conference on Aging Aircraft, Phoenix, Arizona, April 21 – 24, 2008. Schijve, J. (1967), Fatigue Crack Propagation, ASTM STP 415, Editors J.C. McMillan and R.M.N. Pelloux, American Society for Testing and Materials, Philadelphia, Pennsylvania, pp. 415-459. Suresh, S., Ritchie, R.O. (1984), Inter. Metallurgical Reviews, Vol. 29, pp. 445-476. Wanhill, R.J.H. (1986), The Behaviour of Short Fatigue Cracks, EGF Pub 1, Editors K.J. Miller and E.R. de los Rios, Mech. Eng. Publications, London, pp. 27-36. Wanhill, R. J. H. (1994a), NLR Technical Publication TP 94177 U, National Aerospace Laboratory NLR, Amsterdam. Barter, S. A., Wanhill, R. J. H. (2008), NLR Client report NLR-TR-2008644. National Aerospace Laboratory, NLR., Amsterdam The Netherlands. Schijve, J. (1973), Eng. Fracture Mechanics, Vol. 5, pp. 269-280. Willard, S. A. (1997), NASA/CR-97-206291, NASA Langley Research Center, Hampton, Virginia. Wanhill, R. J. H., Hattenberg, T. (2006), NLR Technical Publication TP2006-184, National Aerospace Laboratory NLR, Amsterdam. AGARD-CP-376 (1984): Fatigue Crack Topography, Advisory Group for Aerospace Research and Development, Neuilly-sur-Seine. Broek, D., Van der Vet, W.J. (1968), NLR Technical Report TR 68002 U, National Aerospace Laboratory NLR, Amsterdam. Nix, K. J. (1980), PhD. Thesis, Department of Metallurgy and Materials Science, Imperial College London , London. Nedbal, I., Siegl, J., Kunz, J. (1989), Advances in Fracture Research, Proceedings of the 7th Inter. Conference on Fracture (ICF7), Eds. K. Salema, K. Ravi-Chandar, D. M. R. Taplin and P. Rama Rao, Vol. 5, Pergamon Press, Oxford, pp. 3483-3491. AGARD-R-767 (1990): Short-Crack Growth Behaviour in Various Aircraft Materials, Advisory Group for Aerospace Research and Development, Neuilly-sur-Seine. Barter, S. A., Sharp, K. P., Holden, G and Clark, G. (2002), Fatigue and Fracture of Eng. Materials and Structures, Vol 25, pp. 111-125.

Marker loads for quantitative fractography of fatigue cracks in aerospace alloys

53

[21] Wanhill, R.J.H., Barter, S.A., Hattenberg, T. (2009), NLR Technical Publication in preparation, National Aerospace Laboratory NLR, Amsterdam. [22] Simpson, D. L., Landry, N., Roussel, J., Molent, L., Graham, A. D., Schmidt, N. (2002), In: Twenty Third Congress of the International Council of the Aeronautical sciences, Proceedings of the ICAS 2002 Congress, Toronto, September 8 – 13, 2002. [23] Molent, L., Dixon, B., Barter S. (2005), In: Proceedings of the Eighth Joint FAA/ DoD/ NASA Conference on Aging Aircraft, Palm Springs, California, January 31 – February 3. [24] Cutshall, J., Blinn, M., Burnside, H., Whitman, Z. and Howell, R. (2007), In: Proceedings of the USAF Aircraft Structural Integrity Program (ASIP) Conference, Palm Springs, California, December 4 − 6. [25] Barter, S. A., (2005). DSTO report No. DSTO-TR-1745, Department of Defence, Defence Science and Technology Organisation, Air Vehicles Division, Australia. [26] White, P., Barter, S. A., Molent, L. (2008), Inter. Journal of Fatigue, Vol. 30, pp. 1267-1278. [27] Abelkis, P. R. (1974), IRAD Technical Report MDC-J5672, Douglas Aircraft Company, Long beach, California. [28] Hattenberg, T. (1994), NLR Contract Report CR 94123 C, National Aerospace Laboratory NLR, Amsterdam. [29] Jonge, J. B. de, Schütz, D., Lowak, H., Schijve, J. (1973), NLR Technical Report TR 73029 U, National Aerospace Laboratory NLR, Amsterdam. [30] Lowak, H., de Jonge, J. B., Franz, J., Schütz, D. (1979), NLR Miscellaneous Publication MP 79018 U, National Aerospace Laboratory NLR, Amsterdam. [31] Barter, S., Dixon, B. & Molent. L. (2009), Eng. Failure Analysis, 16/3, pp 863-873 doi:10.1016/j.engfailanal.2008.07.014. [32] Siegl, J., Schijve, J. (1990), Report LR-631, Delft University of technology, Delft. [33] Dijk, G. M. van, de Jonge, J. B. (1975, Problems with Fatigue in Aircraft, Proceedings of the Eighth ICAF Symposium, Swiss Federal Aircraft Establishment (F+W), Emmen, pp. 3.61/1 − 3.61/39. [34] Barter, S. A., Clark, G., Goldsmith, N. T. (1993), Durability and Structural Integrity of Airframes, Proceedings of the 17th ICAF Symposium, Stockholm, Editor A.F. Blom, Engineering Materials Advisory Services, Warley, West Midlands, pp. 281-304. [35] Linden, H. H. van der (1984), Fatigue Crack Topography, AGARD Conference Proceedings No.376, Advisory Group for Aerospace Research and Development, Neuilly-sur-Seine, pp. 11-1 − 11-17. [36] Dainty, R. V. (1984), Fractography of Ceramics and Metal Failures, ASTM STP 827, Editors J.J. Mecholsky and S.R. Powell, Jr., American Society for Testing and Materials, Philadelphia, Pennsylvania, pp. 285-308. [37] Wanhill, R. J. H. (1994), Inter. Journal of Fatigue, Vol. 16, pp. 99-110.

54

Barter, Molent,Wanhill

[38] Schijve, J., Vlutters A.M., Ichsan, S.P., ProvoKluit, J. C. (1985), Inter. Journal of Fatigue, Vol. 7, pp. 127-136. [39] Pell, R. A., Molent, L., Green, A. (2004), DSTO Report DSTO-TR-1547, Defence Science and Technology Organisation, Melbourne. [40] Molent, L., Boykett, R., Walker, K. (2008), Research and Technology Organisation Meeting Proceedings RTO-MP-AVT-157, Montreal, Canada 13th – 16th October. [41] Elber, W. (1971), Damage Tolerance in Aircraft Structures, ASTM STP 486, American Society for Testing and Materials, Philadelphia, Pennsylvania, pp. 230-242. [42] Dawicke, D. S. (1997), NASA Contract Report NASA/CR 201668, NASA Langley Research Center, Hampton, Virginia. [43] Barter, S. A., Price, J. (2000), In: Proceedings of Symposium 2000 on Structural Integrity and Fracture, Edited by G. Heness., University of Technology, Sydney, Published by the Institute of Metals and Materials Australasia, Melbourne, June 29 30, pp. 140-153. [44] Crooker, T. W. (1971), Journal of Eng. for Industry, Vol. 93, pp. 893-896. [45] Topper, T. H., Yu, M. T. (1985), Inter. Journal of Fatigue, Vol. 7, pp. 159164. [46] Jonge, J. B. de, Nederveen, A. (1980), Effect of Load Spectrum Variables on fatigue crack Initiation and Propagation, ASTM STP 714, American Society for Testing and Materials, Philadelphia, Pennsylvania, pp. 170-184. [47] Beachem, C. D. (1967), ASM Transactions Quarterly, Vol. 60, pp. 630-643. [48] Rusk, D, T., Arocho, A., Pierce, J., Abfalter, G. (2008), Naval Air Systems Command, Structures Division, Patuxent River, Maryland. [49] Wanhill, R. J. H., Hattenberg, T., Ten Hoeve, H.J. (1996), NLR Contract Report CR 96451 C, National Aerospace Laboratory NLR, Amsterdam. [50] Huynh, J., Molent, L., Barter, S. A. (2003), DSTO report DSTO-TR-1484, Defence Science and Technology Organisation, Melbourne. [51] Molent, L., Barter, S. A., White P., Dixon B. (2005), In: Proceedings of the Eleventh Australian International Aero Space Congress, Melbourne, Victoria March 13 – 17. [52] Conley, F. M. and Sayer, R. B. (1977), In: Proceedings of the AAIA/ASME 18th Structures, Structural Dynamics & Material Conference, Volume A, San Diego, California, March 21 − 23, pp. 119-130. [53] Sunder, R. (1983), Proceedings of the International Conference on Digital Techniques in Fatigue (SEECO ’83), Editor B.J. Dabell, Society of Environmental Engineers, London, pp. 197-218. [54] Prakash, R. V., Sunder, R. (2007), Proceedings of the Inst. of Mech. Eng., Part L: Journal of Materials: Design and Applications, Vol. 221, pp. 53-62. [55] Barter, S. A., Molent, L., Landry, N., Klose, P., White, P. (2003), In: Proceedings of the Tenth Australian International Aero Space Congress, Brisbane. Queensland July 29– August 1.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

SUBSTANTIATIONS OF AN AIRBORNE COMPOSITE RADOME MOUNTED ON AN AIRCRAFT DOME I. Kressel, U. Ben-Simon, M. Elyahu, D. Peled, A. David, T. Nachshhon, Y. Gary, M. Interator, G. Ghilai, A. Simon IAI, Ben Gurion International Airport 70100 Israel

Abstract: A dome mounted composite structure Radome was design and substantiated for an early warning surveillance system. This Radome is the biggest composite part that has been manufactured at IAI. Appropriate facilities were built including a new clean room, autoclave, assembly hangar and a dedicated NDT system. The substantiation process includes development of material design allowables, coupons and elements testing and a full scale static test to the maximum ultimate design loads.

DESIGN AND MANUFACTURING The Composite Radome is a Honeycomb structure shell with fiberglass skins. The thickness of the Honeycomb and the skins vary according to electrical requirements. The general dimensions of the Radome can be seen in Figures 1. A steel lay-up tool with uniform wall thickness was designed and manufactured to achieve specific requirements – thermal expansion coefficient as close as possible to that of the composite, minimized weight, optimal thermal transfer and uniform temperature distribution. Advanced techniques were utilized for the production of this very big Radome: automated plies cutting, lay-up of each ply with laser projectors and assembly using laser tracker location. A special process was developed for production of the honeycomb core (with very complicated shapes and variable thicknesses) by forming and NC machining of several segments with perfect fit between them.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 55–58. © Springer Science+Business Media B.V. 2009

I. Kressel et al.

56 9952

3259

2024

Figure 1: Overall Radome Dimensions (mm) Special T.T.U C-Scan system was developed for automatic scanning of the Radome. Due to its large size and complex curvature, the Radome is inspected using six scanning segments. The analysis of the C-Scan data is performed using NDTEX, advanced software developed in IAI.

CERTIFICATION IAI’s approach to certification of the composite Radome structure is presented. The certification is analytical, supported by extensive test program in compliance with guidelines issued by the FAA and industrial common practice. The Radome structural test program was developed through a "building-block" methodology. The "building-block" approach includes tests at the coupon, element and component levels, combined with full-scale limit and ultimate load tests as the final proof of structure. The tests at coupon, element and component levels were used for obtaining material characterization, design allowables, design concept verification, substantiation of analysis methods and manufacturing processes. The environmental effects on the composite structure were characterized at the element and component levels and were accounted for in the structural analysis. Experience with similar structure was important in developing of the composite Radome certification program. The resin constituent is a well-known material system, which has been in wide use in the world for civil Aircraft applications. Extensive service experience has been gained with these parts, which shows that this material system is suitable for such applications. A test program at a coupon level provided laminate-level allowable design strain values covering each failure mode and environmental condition. Design allowable values are based on saturated laminates and sandwiches. Effect of defects and cyclic loading were also included. Corrections for material variability followed approved procedures in MIL-Handbook 17.

Substantiations of an airborne composite Radome

57

At the component level, Leading edge and Dome-Radome attachment structures were tested. These specimens were exposed to 85% RH & 60°C, accounting for the effects of environment. The leading edge component, after subjecting to fatigue test of 20 life times, withstood 225% design limit load without failure or damage growth. This test component included barely visible damages and was also exposed to environmental aging during test. Similar test was performed on the Radome-Dome interface component. This specimen, with embedded defects, also withstood 225% design limit load without failure or damage growth. The residual strength testing of the leading edge and root elements verified that the structure was capable of carrying the required loads with existing damages, taking into account severe environmental aging. This strength testing substantiated the analytical predictions and empirical results based on coupon tests characterization. The test set-up of the leading edge is seen in Figure 2.

Figure 2: Component test. (a) Dome-Radome attachment; (b) Leading edge The two primary damage tolerance requirements were addressed: damage growth characterization and residual strength-capability. Considering the applied strains, IAI selected a “no-growth” approach for the composite Radome. This philosophy states that any damage that is visually undetectable should not be critical. Structure with this type of damage must be capable of carrying ultimate load for the operational life of the airplane. This approach was validated through a series of tests at coupon and component levels. To do this, impact damages were inflicted on a leading edge element specimens at the barely visible level, combined with environmental aging followed by fatigue testing under load cycles representative of twenty (20) design service lifetimes. Ultrasonic inspection revealed the absence

I. Kressel et al.

58

of damage growth. IAI fatigue evaluation of the Radome-Dome metal splice structure was based on established methods. Since the composite Radome is a shell body of revolution the Root specimen and the leading edge are the only important design details. The two test elements, representing the root and leading edge, are large enough to assure proper similarity in failure modes to the full-scale Radome. The Extensive test program described above demonstrated the radome structure robust design with adequate margin of safety. Due to the high margin of safety demonstrated in the component testing it is concluded that repetition of the environmental effects on a full scale Radome is not required. One of the benefits of a building block approach is that the extent of full scale testing can be reduced based on the test results from lower levels of testing and validation of analytical methods by comparison to those results. Based on this methodology, a limited number full scale test load cases were tested, under ambient temperature/moisture only. The other temperature/moisture conditions can be cleared by analysis or by direct comparisons of strain data to element test results. Similarly, other load cases can be cleared by analysis. Full scale Limit load strain surveys and ultimate load testing demonstrated the predictive capability of the FEA model. A loads enhancement factor was added in order to compensate for temperature effect. This factor was established in a series of element tests, representing the honeycomb to solid glass interface at the radome root. This load increase is the maximum possible for the metal parts. Any load increase beyond this value may introduce plasticity in the aluminum parts and as a result the internal loads distribution may change to an unrealistic one. The Radome for Full scale test was identical to a serial production Radome including the DomeRadome attachment. Since such radome is manufactured from at least two batches of materials the material properties variation is inherent.

SUMMARY The full-scale test, along with a comprehensive test program at coupon, element and component levels, met all of the design requirements. It provided full-scale validation of the design and analysis methodology, fabrication processes, and damage tolerance capability of the composite Radome.

Airworthiness considerations

25th ICAF Symposium – Rotterdam, 27–29 May 2009

DEVELOPMENT OF A “LOW STRESS CRITERION” THAT ELIMINATE A LARGE PORTION OF THE AIRCRAFT FROM DAMAGE TOLERANCE BASED MAINTENANCE PROGRAM FOR STRUCTURAL REPAIRS REQUIRED BY THE NEW PART 26 FEDERAL REGULATION Girindra K. Das and Matthew Miller Structural Damage Technology, Boeing Commercial Airplanes Seattle, Washington, USA

Abstract: The main purpose of the paper is to outline a major criterion developed for the new Part 26 Code of Federal Regulation (CFR) that require aircraft manufacturers to support operator compliance to damage tolerance based maintenance program for aircraft structural repairs and alterations. The paper focuses on this newly developed “low Stress Criterion” that eliminates large areas of the aircraft from damage tolerance evaluation and the resulting damage tolerance inspections (by the operators) if stresses are below a certain level. In order to comply with this Aging Airplane Safety Rule (AASR) Aircraft manufacturers are required to provide damage tolerance data for repairs and alterations to the operators so that they can develop inspection program to maintain airworthiness of their aircrafts. As the focal of this activity, Structural Damage Technology (SDT) group of Boeing is actively engaged in developing methods and criteria to support the Part 26 rule. The “Low Stress Criterion” is one of the major criteria developed recently to support the Part 26 rule. This paper describes in detail this criterion that eliminates large areas of the aircraft from damage tolerance evaluation by Boeing and resulting inspections by the operators if stresses in the repaired structure are below a certain level. Low Stress Criterion is based on both fatigue crack initiation and propagation considerations. The criterion provides a stress level M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 61–71. © Springer Science+Business Media B.V. 2009

62

Girindra Das and Matthew Miller

below which the fatigue life of the part will be many times the design service objective of the airplane. At the same time this stress level will also provide a large crack growth life and the required damage tolerance based inspection interval will be much longer that the routine inspection already being carried out on the airplane. The paper describes in a comprehensive manner how the criteria has been developed which in the long run will save the world airlines significant maintenance costs by eliminating inspections on repairs in less critical low stress areas, thereby allowing them to focus on more critical structural repairs

INTRODUCTION For past 43 years airworthiness standards for transport category airplanes has been governed by Part 25 of Title 14 of the Code of Federal Regulation (CFR). Specifically, section § 25.571 covered Damage Tolerance and Fatigue Evaluation of Structure which requires “an evaluation of the strength, detail design, and fabrication to show that catastrophic failure due to fatigue, corrosion, manufacturing defects, or accidental damage, will be avoided throughout the operational life of the airplane”. Figure1 shows how § 25.571 is related to the Title 14 of the CFR. This regulation also states that “This evaluation must be conducted --- for each part of the structure that could contribute to a catastrophic failure (such as wing, empennage, control surfaces and their systems, the fuselage, engine mounting, landing gear, and their related primary attachments). Based on the evaluations, inspections or other procedures must be established to prevent catastrophic failure”.

Figure 1: Section § 25.571 of Federal Regulations

Low stress criterion for part 26 regulations

63

A new Part 26 rule and an Advisory Circular (AC) was published in the Federal Register on December 12, 2007 and becomes effective January 11, 2008 that directs Type Certificate Holders (TCH) like Boeing to support operator compliance to the operational rule 14 CFR §121.1109 and §129.109 that directs operators to have a damage tolerance based maintenance program for all structural repairs and alterations. This rule affects all past, present and future commercial transports and includes in-certification, in-production and out-of-production airplanes. Figure 2 shows how Part 25 and Part 26 are related to the Title 14 of the CFR.

Figure 2: Part 25 and Part 26 of the Title 14 of the CFR

DETAILS OF PART 26 REGULATION Part 26 regulation directs Type Certificate Holders (TCH) like Boeing to support operator compliance to the operational rule 14 CFR §121.1109 and §129.109 by

64

Girindra Das and Matthew Miller

providing them the necessary data to develop a damage tolerance based maintenance program for all structural repairs and alterations. Operational Rule 14 CFR §121.1109 ~ Supplemental inspections With certain exceptions, this section applies to transport category, turbine powered airplanes with a maximum passenger seating capacity of 30 or more; or a maximum payload capacity of 7,500 pounds or more. After December 20, 2010, the rule requires that a certificate holder may not operate an airplane under this part unless the following requirements have been met: (1) Baseline Structure. The certificate holder's maintenance program for the airplane includes FAA-approved damage-tolerance-based inspections and procedures for airplane structure susceptible to fatigue cracking that could contribute to a catastrophic failure. (2) Adverse effects of repairs, alterations, and modifications. The maintenance program for the airplane includes a means for addressing the adverse effects repairs, alterations, and modifications may have on fatigue critical structure and on the inspections required. The means for addressing these adverse effects must be approved by the FAA Oversight Office. (3) Changes to maintenance program. The required changes made to the maintenance program and any later revisions to these changes, must be submitted to the Principal Maintenance Inspector (PMI) for review and approval. Operational Rule 14 CFR §129.109 ~ Supplemental inspections for U.S.-registered aircraft Part 129 deals with the operations of foreign air carriers and foreign operators of U.S. registered aircraft engaged in common carriage. Subpart b of Part 129 deals with continued airworthiness and safety improvements. Specifically, section § 129.109 provides for supplemental inspections for U.S.-registered aircraft and the requirements are identical to those for14 CFR §121.1109, stated above in paragraph 2.1.

CRITERIA DEVELOPMENT FOR THE PART 26 REGULATION Part 26 regulation requires aircraft manufacturers to support operator compliance to the operational rules §121.1109 and §129.109 by providing them the necessary data to develop a damage tolerance based maintenance program for all structural repairs and alterations.

Low stress criterion for part 26 regulations

65

Type Certificate Holders (Boeing, Airbus), Supplemental Type Certificate Holders (Repair Stations, Mod Centers other than Original Equipment Manufacturers), Operators, and Lessees will be significantly impacted by new requirements for Damage Tolerance based maintenance program for Repairs, Alterations and Modifications. Criteria and methods to conduct damage tolerance analysis of the basic airplane structure has been in existence for a long time. Many of the methods and criteria can be directly applied to the repairs to the basic structure as well. However, there have been situations where existing criteria and methods could not be applied directly to the repairs. New criteria and methods for the DT Analysis of repairs had to be developed in addition to the existing ones used for the basic airplane. Structural Damage Technology (SDT) group at Boeing has been heavily engaged in developing these criteria, with a focus on reducing the scope of the program, without compromising the structural safety of the aircraft. Some Examples of Methods and Criteria being developed are: z z z z z z

Develop new Stress Intensity Factors (SIFs) for unique repair configurations for which existing SIFs are not applicable. Develop Fatigue Critical Baseline Structure (FCBS) for all commercial models. Only repairs on the FCBS need Damage Tolerance (DT) evaluation and inspection. Criteria and method for Widespread Fatigue Damage of repairs. Define Industry Standard repairs. Method to determine Flight Length Sensitive (FLS) items and hour cutoff in addition to the flight cycle limit for the threshold of DT inspections for repairs on these items. Low stress criterion which reduces the scope of the program significantly and described in detail in this paper.

LOW STRESS CRITERION FOR STRUCTURAL REPAIRS The Low Stress Criterion defines the stress levels below which no damage tolerance evaluations and damage tolerance inspections for structural repairs to FCBS need to be carried out to satisfy the requirements of the Part 26 regulation. Boeing over the years has assured its customers that there will be minimal structural fatigue cracking in the aircraft within its Design Service Objective (DSO). This is achieved by using appropriate fatigue design methods validated by large number of tests including full-scale fatigue tests. Fatigue analysis based crack initiation data for the repairs are also required to establish the threshold for the damage tolerance inspections. Often, in certain fatigue prone areas, stresses are so low that the fatigue lives could be several multiple of the DSO. From the crack

66

Girindra Das and Matthew Miller

initiation perspective, this means that the fatigue crack initiation could be predicted to be long after an airplane has been retired. Using this concept a low stress level is determined by using standard Boeing analysis methods. Large number of analysis is conducted on many typical repairs covering most of the major component of the aircraft and covering several major models. The details of the analysis and how the final levels of Low Stress are arrived at based on crack initiation are given in the following sections. Low Stress Criterion is also looked from the point of view of the repeat inspection requirements to maintain the structural safety of the aircraft. Repeat inspections are based on fracture mechanics based damage tolerance analysis. The DTR (Damage Tolerance Rating) method developed by Boeing evaluates the goodness of the repeat inspections by obtaining the DTR for the structure and comparing them to the required DTR approved by the FAA for the same. The DTR system, which has been successfully used by the airlines for the basic airplane for past 25 years, is also applied for developing the Low Stress Criterion for the repairs. Damage tolerance analysis is conducted on the same repairs that underwent fatigue analysis and DTR plots are developed. This is done for each repair for various stress levels gradually reducing the stress. The stress level at which a routine inspection provides many times the required DTR was established as the Low Stress value. The details are described in the following sections. Finally, lower of the fatigue based low stress and damage tolerance based low stress is used to develop the Low Stress Criterion. Low Stress criterion based on fatigue (crack initiation) To develop this criterion a large number of fatigue analyses were conducted on Boeing recommended typical repair configurations detailed in the Service Repair Manuals (SRMs) for various models. The SRMs are provided to the airlines so that they can develop their own repairs if they are similar to ones given in the SRMs. Besides maximum stress (Fmax) other parameters considered that provide fatigue life are the Stress Ratio, the DFR (Quality parameter related to the local increase in stress, manufacturing quality etc) and the Ground Air Ground (GAG) Damage Ratio that accounts for spectrum loading. Using these parameters standard analysis methods were used to calculate N 95/95 Fatigue Life (95% reliability and 95% confidence) for typical repairs from the SRMs using actual operating loads. Keeping all other parameters the same Fmax is then gradually reduced and N 95/95 Fatigue Life is calculated for these reduces stress levels. N 95/95 Fatigue Lives are then normalized in terms of Design Service Objectives (DSO). DSO is the number of flights all fatigue critical structures are designed to, during which practically no fatigue crack initiation is anticipated. Finally, the stress levels are plotted in terms of N 95/95 Fatigue Life normalized as a multiple of DSO and shown in Figure 3.

Low stress criterion for part 26 regulations

67

Analysis of Results: The figure 3 shows that the fatigue life of the repairs on the 7000 series Aluminum is considerably lower than those for the 2000 series Aluminum. Based on the results it is apparent that a lower level of “Low Stress” needs to be chosen for the 7000 series Aluminum. Based on this, for the 7000 alloys (example: wing upper skin repairs shown in Figure 3) a stress level that gives fatigue life more than 3 to 4 times the DSOs was chosen as the “Low Stress Level”. For the rest of the repairs which were on the 2000 series Aluminum a stress level that gave fatigue life of more than 5 DSOs was chosen as the “Low Stress Level”. Interestingly, repairs on the “Pressure Critical Structure” like the fuselage longitudinal lap splices did not show any specific reduction in fatigue life compared to the other items. Therefore, from the point of view of crack initiation, two levels are chosen for the “Low Stress” threshold, one for the 2000 Aluminum and the other for the 7000 Aluminum, each providing fatigue life many times the DSO of the airplane. Thus, the criterion provides that the fatigue cracks will not initiate in areas with the “Low Stress Levels” until a time way beyond the operational life the airplane.

Figure 3. Plot of Normalized Maximum Stress against Fatigue life as a multiple of DSO

68

Girindra Das and Matthew Miller

Low Stress criterion based on Damage Tolerance (crack propagation) In the damage tolerance philosophy airworthiness of the fleet is maintained by continued directed inspections based on damage tolerance analysis which involves residual strength analysis and crack growth analysis. In addition, Boeing has the unique method of developing inspection program called DTR system (Ref 1) which combines probability of detection and the damage tolerance analysis. Inspection method and the interval chosen provide a DTR which must exceed the required DTR set by the regulatory requirements. Parameters influencing crack growth usually are maximum operating stress, material crack growth rate parameter and configuration dependent geometric modification factor that determines the Stress Intensity Factors. In the event there is spectrum loading Effective Stress Rating “S” rather than the maximum operating stress is used in the analysis. For residual strength analysis parameters required are the maximum limit stress and the configuration dependent geometric modification factors. Using the above parameters complete DT analyses was conducted on typical repairs (from SRM) using actual loads. The repairs analyzed were the same that were used in the fatigue analysis. Detailed DT analysis involved (a) determining critical crack length using Residual Strength analysis; (b) calculating detectable to critical crack growth using crack growth parameters specific to repair analyzed and (c) Developing the DTR forms for each repair. DTR forms provide DTR for various inspection intervals for a specific type of inspection. Inspection interval must provide DTR that exceeds required DTR set by the regulatory requirements. Using the DTR forms, baseline DTR is obtained using routine inspections normally conducted by world airlines like “D” or “4C” Surveillance. Then, keeping all the other parameters same, effective Stress Ratings “S” is gradually reduced and DTR forms are developed for each of these reduced “S”. As in the baseline analysis, using routine inspections like “D” or “4C” Surveillance, the DTR values obtained for each level of “S” for each repair. The DTR value for each level of “S” is then normalized in terms of the required DTR for each of the repairs.

Low stress criterion for part 26 regulations

69

Figure 4: Normalized Stress Rating Vs Multiple of required DTR for D/4C Surveillance Inspection for Pressure Critical Structure

Finally, the Stress Rating levels are plotted against DTR for routine inspections like “4C” or “D” Check Surveillance normalized as a fraction of Required DTR. These are shown in Figures 4 and Figure 5. Low Stress criterion is based on the Stress levels that provide many times the required DTR by routine inspections like “D” or “4C” Checks Surveillance that is regularly conducted by the airlines.

Analysis of Results: The figures 4 and 5 show that the pressure critical structure is more critical than the rest of the structure for repeat inspections. One of the reasons is that the required DTR is much higher for pressure critical structure than that for the non pressure critical ones. This result is different than the results of the fatigue analysis where no difference was observed. On the other hand, from the DT perspective 7000 alloys did not show any significant reduction in the DTR. Based on the data for pressure critical structure, a “Low Stress Level” was chosen that gave at least about 2 times the required DTR for the routine inspections. For rest of the structure the “Low Stress Level” chosen gave more than 2 times the required DTR for routine inspections.

Girindra Das and Matthew Miller

70

Figure 5 : Normalized Stress Rating Vs Multiple of required DTR for D/4C Surveillance Inspection for Non Pressure Critical Structure

Discussions: Most analyses were conducted on wide-body typical repairs obtained from the SRMs. A few narrow-body repairs were also analyzed and similar trends were observed. Since the data is based on standard SRM repair analysis this criterion applies to Industry Standard Repairs.

SUMMARY A “Low Stress Criterion” has been developed that provides a stress level below which no analysis or inspection is required for a structural repair. The criterion is based on both fatigue (initiation) and crack growth (propagation) considerations and lower of the two “Low Stress” values is used as the final “Low Stress” level and will apply to all Industry Standard Repairs.

Low stress criterion for part 26 regulations

71

BENEFIT TO THE INDUSTRY The Criterion will save the industry significant resources without compromising safety. Large area of the airplane will be exempt from (a) the required analyses by the Original Equipment Manufacturer and (b) the required inspections by the operators. Airlines can then focus their attention and resources on the more critical structures which are above the stress levels defined by this criterion.

REFERENCES [1] Goranson, U.G. titled “Damage Tolerance - Facts and Fiction”, In: the 17th ICAF Symposium, Stockholm, Sweden, June9, 1993.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

DAMAGE TOLERANCE PHILOSOPHY FOR BONDED AIRCRAFT STRUCTURES C.D. Rans and R.C. Alderliesten Delft University of Technology, Delft, 2629 HS, The Netherlands

Abstract: This paper presents a damage tolerant philosophy for bonded aircraft structures and repairs. The approach presented utilizes a strain energy release rate approach for predicting bond line delamination growth and a stress intensity factor approach for predicting adherent cracking. The novel addition to the method lies within the coupled analysis of bond line and adherent failure using a displacement compatibility approach. An overview of the methodology and the necessary experimentation to implement it is presented. Additionally, two case studies are presented to illustrate the power and performance of the proposed methodology.

NOMENCLATURE Symbol δ displacement [mm] γ shear strain τ shear stress [MPa] a adherent half-crack length [mm] b bond line delamination length [mm] C Paris relation coefficient E Young’s modulus [MPa] GT total strain energy release rate [MPa mm] GI mode I strain energy release rate [MPa mm] M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 73–90. © Springer Science+Business Media B.V. 2009

C.D. Rans and R.C. Alderliesten

74

GII GIII K n N R S t v x

mode II strain energy release rate [MPa mm] mode III strain energy release rate [MPa mm] stress intensity factor [√(MPa mm)] Paris relation exponent number of cycles [cycles] stress ratio [dimensionless] stress [MPa] thickness [mm] crack opening displacement [mm] distance along adherent crack length [mm]

Subscripts 1 adherent 1 2 adherent 2 I mode I II mode II III mode III ∞ far field ad adhesive br bridging cg crack growth d delamination eff effective st stringer

INTRODUCTION As the demand for lower aircraft operating costs increases, the improvements in structural weight and operational maintenance offered by adhesive bonding technologies over mechanical fastening technologies becomes an important factor in aircraft design. The increase in service life offered by a bonded structure over that of a riveted structure makes adhesive bonding technology desirable for aircraft structures. The assessment of the structural integrity of an aircraft, however, is not simply defined by service life. To ensure the structural integrity of an aircraft structure, its response to both incidental (fatigue, limit/ultimate loads) and accidental (overloads, impact damages, manufacturing flaws, etc.) need to be assessed.

Damage tolerance philosophy for bonded aircraft structures

75

Currently, the widespread use of mechanically fastened technologies continues due to the relative ease of predicting the types of damages and assessing the response in the presence of damages within such structures. The so called damage tolerance analysis of mechanically fastened structures is well established. In order to further advance the use of adhesive bonding technologies in the aircraft industry, an equally simple and effective means of assessing the damage tolerance of bonded structures is needed.

CURRENT STRUCTURAL ASSESSMENT METHODOLOGY Structural analysis of bonded structures in general and bonded lap joints in particular has been a prevalent research topic in open literature [1, 2]. The developed analysis methods can be categorized into two main classes: •

•

Stress/strain based methods o Peak peel and shear stresses for bond line failure o Peak tensile/compressive stresses over normal stresses for adherent failure Fracture mechanics based methods o Energy balance methods for bond line damage growth o Stress Intensity Factor for adherent damage growth

It could be argued that the analysis of bond line failure and adherent failure should be addressed separately in the design. However, the authors believe that several designs, especially bonded patch repairs, influence both aspects and therefore warrant the simultaneous assessment of these aspects in a similar approach. The reason that most approaches presented in the literature tend to examine either bond line failure [1, 2] or adherent failure [3-5] can be attributed to the complexity of a generic structural analysis method for design. Nevertheless, the current methods consider static failure criteria only.

PROPOSED DAMGE TOLERANCE METHODOLOGY The proposed methodology for the design of bonded structures is to adopt a damage tolerance approach whereby the safety of a structure is ensured through regular inspections (determine on damage growth behaviour) for damage and an assessment of the criticality of detected damages. In the case of bonded structures, these damages include delamination along the bond line and adherent fracture. To achieve this, tools which can predict the growth of damages within a structure and final fracture of a damage structure are required.

C.D. Rans and R.C. Alderliesten

76

As mentioned earlier, most of the analysis methods currently used for bonded structures are based on the calculation of the local shear and peel stresses in the expected delamination region [1]. These stresses are used to determine minimum overlap lengths and are related to joint strength. More sophisticated methods include elastic and plastic adhesive behaviour for further detailed joint optimization [6]. These methods, however, are meant to calculate the static strength of the structure. Damage accumulation is often only considered by applying sufficient overlap lengths to avoid risking bonding integrity. Based on experience gained through the development of damage tolerance assessment tools for crack growth in Fibre Metal Laminates (FMLs), the application of a fracture mechanics based approach to delamination growth has proven quite successful [7-9]. FMLs consist of alternating metallic and fibrereinforced plastic layers. Fatigue cracks initiate and grow within the metallic layers; however the growth of these cracks is influenced by the bridging effect of the adjacent intact fibre layers. In turn, the bridging effect of the fibre layer is related to the size of the delamination between the fibre and adjacent cracked metallic layer. Thus, an accurate prediction of crack growth in FMLs requires an accurate prediction of delamination growth along the interfaces of the fibre and metal layers. Such predictions of delamination growth in fibre metal laminates have been achieved by correlating delamination growth rate to the strain energy release rate using a Paris type relation. It is believed that the success in coupling the damage tolerant analysis of bond line delamination growth and adherent crack growth in FMLs can be successfully applied to a wider range of bonded aircraft structures. The proposed damage tolerance approach requires a simultaneous analysis of adherent failure, bond line failure, and interaction between the bond line and adherent failure. Each of these elements is briefly described below. Bond Line Delamination Growth Bond line delamination growth behaviour is characterized and predicted using a Paris type relation based on Mode II strain energy release rate. Only the Mode II strain energy release rate is used due to the fact that most bonded interfaces are designed to transfer load primarily through shear. This assumption will be further discussed later in the paper. Thus, delamination growth rates can be determined by the following relation:

db = Cd dN

(

GII max − GIImin

)

nd

(1)

where b is the delamination length, N is the number of cycles, G is the mode II strain energy release rate and Cd and nd are the Paris relation fit parameters.

Damage tolerance philosophy for bonded aircraft structures

77

The formulation of the strain energy release rate range in eqn. (1) requires some explanation. Typically, delamination growth is often characterized in the literature using the maximum strain energy release rate, Gmax [10-13], or the arithmetic difference between the maximum and minimum strain energy release rates, ΔG = Gmax - Gmin [14]. The use of Gmax stems from the importance of this parameter for checking for the onset of static delamination failure; however, it fails to include information about the minimum load and thus fails to describe the entire load cycle applied. The use of ΔG attempts to overcome this shortcoming, but fails to adhere to the rules of superposition of G which state that:

GI = ⎡⎣ GI (1) + GI (2) + GI (3) + "⎤⎦

2

GII = ⎡⎣ GII (1) + GII (2) + GII (3) + "⎦⎤

2

GIII = ⎡⎣ GIII (1) + GIII (2) + GIII (3) + "⎤⎦ GT = GI + GII + GIII

(2) 2

The reason for the square roots in eqn. (2) arises due to the fact that G is proportional to the square of the applied stress. Thus, the proper formulation for the applied Mode II strain energy release rate range reduces to:

ΔGII =

(

GII max − GII min

)

2

(3)

This is the definition of strain energy release rate range adopted in eqn. (1), with the squared term in eqn. (3) simply being combined into the Paris exponent, n. Adherent Crack Growth Crack growth within the adherents is analyzed using a standard linear-elastic fracture mechanics approach. Similar to delamination growth prediction, crack growth behaviour is predicted using a Paris relation for crack growth such that:

da = Ccg ( ΔK eff dN

)

ncg

(4)

where a is the crack length, N is the number of cycles, ΔKeff is the effective mode I stress intensity factor and Cd and nd are the Paris relation fit parameters. ΔKeff is calculated using the Schijve correction [15]:

ΔK eff = ( 0.55 + 0.33R + 0.12 R 2 ) (1 − R ) K max

(5)

C.D. Rans and R.C. Alderliesten

78

Damage Interaction The novel addition to the proposed methodology is in the treatment of the interaction between bond line delamination damage and crack growth in the adherent material. This interaction is critical in many bonded aerospace structures such as bonded patch repairs and stiffened panels made with bonded doublers, stringers, and frames. In such structures, crack growth within an adherent (such as the fuselage skin) is influenced by the presence of an alternative load path provided by the bonded adherent (stiffener or patch) which bridges the crack. The effectiveness of this bridging mechanism is in turn influenced by the integrity of the bond between the two adherents. By defining the load carried by the alternate load path as a bridging load (Pbr), the problem can be broken up into two parts using the principle of superposition, as illustrated in Figure 1, where P1 and P2 are the components of the applied load carried by adherents 1 and 2 in an undamaged state. In this way, crack growth in adherent 1 can be calculated by a superposition of the effects of the far field load, P1, and the bridging load, Pbr. Similarly, bond line delamination growth can be calculated including the effects of Pbr on the strain energy release rate. Thus, given Pbr, the crack and delamination growth rates can be determined by means of the methods outlined in the previous two sections.

P1

P2

P1

A

Pbr

P2

2a b x

= 2v

b

+

P2 + Pbr

b

P2 A Section A-A

Figure 1: Division of bonded structure analysis into adherent crack growth and bond line delamination growth problems. Determination of Pbr is achieved through displacement compatibility between the cracked and uncracked adherent. At any point along the length of the crack, the amount of crack opening due to the far field and bridging load in the cracked

Damage tolerance philosophy for bonded aircraft structures

79

adherent can be equated to the elongation of the uncracked adherent over the delamination length, b, plus the shear deformation of the adhesive layer:

v∞ − vbr = δ 2 + δ ad

(6)

Each of these deformation and crack opening components can be defined at any point, x, along the crack length. To aid in this exercise, the bridging load (Pbr) and loads carried by adherent 1 and 2 (P1 and P2) will be defined as stresses Sbr, S1, and S2. The crack opening displacements can be calculated using linear elastic fracture mechanics solutions [16]. The opening component due to the far field stress can be obtained from the solution for an infinite monolithic sheet subjected to a far field stress:

v∞ ( x ) = 2

S1 a2 − x2 E1

(7)

To determine the component due to the bridging, the bridging stress is assumed to be represented by a distributed number of infinitesimal point loads located at xP along the length of the crack. The opening due to bridging can then be determined by integrating the contributions due to the infinitesimal point loads.

vbr ( x ) = ∫ v ( x, xP ) dxP a

(8)

0

Solutions for the opening resulting from a point load, v(x, xP), are available in the literature [7, 16]. Deformation of the uncracked adherent along the delamination length is written as:

δ2 ( x) =

S2 + Sbr ( x ) ⋅b ( x) E2

(9)

where b(x) is the delamination length at location x. The displacement resulting from adhesive shear deformation can be obtained by the product of the adhesive thickness (tad) and shear deformation angle (γ), assuming the shear deformation angle is small:

δ ad ( x ) = γ ( x ) ⋅ tad =

τ ad ( x ) Gad

⋅ tad

(10)

C.D. Rans and R.C. Alderliesten

80

Using eqns. 6 through 10, the bridging stress at any point x along the crack length can be determined. Once the bridging stress distribution is known, a crack tip stress intensity reduction factor (Kbr) can be calculated using the same distributed point load assumption employed for the crack opening calculation. Similarly, the strain energy release rate can be calculated including the bridging stress component. Thus, the inputs to eqns. 3 and 5 become:

K = K ∞ − Kbr

GII = f ( S1 , S2 , Sbr )

(11)

allowing simultaneous calculation of crack and delamination growth.

IMPLEMENTING THE DAMAGE TOLERANCE METHODOLGY The previous section presented an overview of the proposed methodology for predicting the damage tolerant behaviour of bonded aircraft structures. This section reviews some of the necessary experimentation and simplifications required to implement the methodology. Characterizing Delamination Resistance Characterizing the delamination resistance of an interface is not a trivial matter. The delamination resistance can be affected by a number of processing variables and variations in the adhesive bonding process. Rather than trying to characterize the delamination resistance of an adhesive system independently from these processing variables, it is suggested that an interface delamination resistance be defined and characterized which is dependant on the process and adhesive system used. For aerospace products where strict control of adhesive processing is enforced, such a simplification is advantageous. For a given adhesive system and bonding process, the delamination growth behaviour of the adhesive interface can be determined using simple test specimens. Numerous test specimen configurations exist for studying Mode II delmaination growth behaviour; the more commonly used being the End Notched Flexure (ENF) and End Loaded Split (ELS) specimens. The lack of consensus on a standardised test specimen is a result of the requirements for studying static Mode II delamination growth. Of primary interest are stable delamination propagation under quasi-static loading, purity of the shear loading, and the ability to apply a compliance calibration analysis for the purposes of determining delamination tip location. None of the specimen configurations meet all the requirements, with

Damage tolerance philosophy for bonded aircraft structures

81

some introducing significant levels of friction along the fracture surface during loading, and others having complicated fixtures which highly influence the compliance calibration analysis. These requirements, however, are related to quasistatic delamination testing. For the purposes of determining fatigue growth behaviour, the requirements for a test specimen are different. Since growth behaviour in relation to strain energy release rate is desired, an ideal test configuration would allow delamination growth measurements to be made at a constant strain energy release rate (such a configuration would be result in unstable delamination propagation under quasistatic loading). In this way, several growth measurements could be made to obtain the growth behaviour for a single strain energy release rate. Additionally, eliminating the dependency of strain energy release rate reduces the importance of determining a precise delamination length, as long as a change in length can be measured. A tension loaded specimen configuration, illustrated in Figure 2, has been extensively used to characterize the delamination growth behaviour of FMLs [7, 8, 17-20]. The specimen, containing a fully cracked adherent (metal layer in the case of FMLs), results in the formation of delaminations between the cracked and intact adherent layers. In this specimen configuration, the strain energy release rate is independent of delamination length and can be controlled by the applied load as shown in eqn. (12), where subscripts al and f refer to the aluminum and fibre layers [7]. Due to the presence of four delaminations in the specimen, it must be remembered that this strain energy release rate is divided amongst all the delaminations (j = 4 in eqn. (12)). 2 σ app Ef 2 ⎡ 2 ⎤ 2 GII = (γ − λ 2 ) n f t f ⎥ ⎢γ ( nal − 1) tal − λ nal tal +

γ=

2 jEal ⎣ ttotal

( nal − 1) tal +

Eal

Ef Eal

; λ= nf t f

ttotal E nal tal + f n f t f Eal

⎦ (12)

Furthermore, the fact that the specimen is tension loaded reduces the likelihood of significant frictional effects along the fracture surface and greatly simplifies the relation between specimen compliance and delamination length. This test specimen configuration can easily be modified by replacing the metal and prepreg layers with the desired adherent materials of interest.

C.D. Rans and R.C. Alderliesten

82

30

1.4 Delamination

275

550

Metal

Prepreg Dimensions in mm

Figure 2: Illustration of delamination growth specimen used for FML delamination characterization. 10-2

db/dN [mm/cycle]

10-3

-40 oC o 20 C o 70 C

10-4

10-5

10-6

10-7

10-8

0.2

0.4

0.6

√G max - √G min [MPa mm]

Figure 3: Variation in delamination growth behaviour with temperature for Aluminum-glass fibre FML variant [21].

Damage tolerance philosophy for bonded aircraft structures

83

Other delamination behaviour dependencies can also be characterized with such a simple test. Of particular interest for aerospace structures are environmental effects such as temperature and humidity. Figure 3 shows a typical family of curves showing the effects of temperature on delamination growth behaviour for the aluminum-glass fibre epoxy prepreg interface in the variant of FMLs known as Glare. Thus, with a relatively small series of tests, the delamination growth behaviour for a specific interface can be characterized for a range of service loads and environments. The Mode II Loading Assumption In a previous section, an approach for predicting delamination growth based on a Mode II strain energy release rate was proposed. The assumption of Mode II loading is based on a combination of observed phenomenon and design practice. First, adhesive joints are designed to transfer load primarily in shear. This is a result of the higher efficiency of an adhesive joint in carrying shear over tensile (or peel) loads. Thus adhesive interfaces are designed primarily for shear. Secondly, even in the presence of limited mixed mode loading, experimental evidence has shown that delamination growth under fatigue loading is dictated primarily by the Mode II loading component [9].

10-2

db/dN [mm/cycle]

10-3 10-4 10-5 10-6 10-7 10-8 10-9

0.2

0.4

0.6

0.8

1

√G max - √G min [MPa mm]

Figure 4: Delamination growth curves for two Glare specimen configurations: Mode II centre crack in 5/4 lay-up (right) and Mixed-mode surface cracks in a 2/1 lay-up (left) [8].

C.D. Rans and R.C. Alderliesten

84

Figure 4 shows fatigue test results for a 5/4-lay-up (pure Mode II, Ψ≈90° [22]) of Glare containing a cracked central layer and a 2/1-lay-up of Glare containing cracked outer layers (mixed mode, Ψ≈53° [22]). Results are plotted using the calculated GII. No clear contribution of the calculated mode I can be observed in the delamination growth tests. This conclusion is also supported by the work of Marissen [9]. Although fatigue delamination growth has been observed to be independent of the Mode I loading contribution, this does not preclude the influence of Mode I loading on delamination initiation. The presence of peel stresses in an adhesive joint is known to promote the initiation of delamination damages and may have some initial influence on early delamination growth. For the purposes of a damage tolerant analysis where an initial delamination is assumed, this influence can be neglected. Further study, however, is required into the extent which Mode I loading contributions can be neglected for delamination growth under fatigue loading. Strain Energy Release Rate Determination Several methods exist for the determination of strain energy release rates, ranging from analytical formulations for specific geometries to finite element methods such as the virtual crack closure technique for more complex scenarios. In applications of the described analysis approach to FMLs, success has been met with describing delamination growth with a series of 1-dimensional delamination growth zones of width, w, as illustrated in Figure 5. Determination of the strain energy release rate for these 1-dimensional zones can be determined with a simple analytical formulation. For a 2-adherent system, this relation takes the form [7]:

GII =

⎞ t2 ⎛ t1 E1 2 ⎜ ⎟ ( S 2 + Sbr ) 2 E2 ⎝ t1 E1 + t2 E2 ⎠

(13)

where subscript 1 denotes the cracked adherent and subscript 2 denotes the intact adherent. This form of descretizing the problem only permits prediction of delamination growth perpendicular to the crack growth direction. Delamination growth in the direction of crack growth is achieved by assuming that the delamination tip is fixed to the crack tip; thus, a new 1-dimensional delamination zone is added for every w increment of the crack growth. Use of the proposed methodology is not limited by the use of the 1-dimensional discretization of delamination growth. The method can be used for a more generalized 2-dimensional planar delamination growth prediction. The 1dimensional discretization merely simplifies the problem making it possible to calculate the strain energy release rate for complex delamination shapes using simplified analytical models. This simplification could easily be removed and

Damage tolerance philosophy for bonded aircraft structures

85

finite element techniques could be used to determine the strain energy release rate distribution along the delamination front.

w 1 2

w b

2b

2a Figure 5: Discretization of a planar delamination into fixed-width delamination zones with 1-dimensional delamination growth.

CASE STUDIES To illustrate the potential of the proposed damage tolerance approach, several case studies in which the methodology has been used are presented. Delamination Growth in a Bonded Patch Repair Beumler [23] reported delamination growth test results for Glare panels containing bonded Glare patches. Panels were tested at applied stress levels of 106 and 120 MPa with R = 0.1. Delamination growth measurements were made for a fatigue initiated delamination at the edge of the bonded patch as it grew towards the centre of the patch. A simplified prediction of this process was made by predicting delamination growth along the centre line of the bonded patch using the 1dimensional delamination growth zone simplification discussed in the previous section. Using this simplification, the strain energy release rate along the centre of the patch is a constant (S2 + Sbr = Sapp) resulting in a constant delamination growth rate predicted using eqn. (1). Figure 6 shows a comparison of the experimental and predicted delamination growth. Firstly, use of a mixed-mode strain energy release rate for predictions of delamination growth is shown to result in gross over-predictions of the results. Use of the Mode II strain energy release rate only resulted in more accurate results.

C.D. Rans and R.C. Alderliesten

86

S max = 106 MPa S max = 120 MPa mixed-mode prediction

103

b [mm]

b

1-dimensional approximation

experimental measurement

102

101

100

mode II prediction

0

20

40

60

80

100

120

140

N [kcycles]

Figure 6: Delamination growth along bonded patch centre line obtained from experiment and from simplified 1-dimension delamination growth model. Comparing the Mode II predictions to the experimental results, it is observed that the predictions were conservative for the higher applied stress and nonconservative for the lower applied stress. This result can be accounted for by differences in the initiation behaviour of the two specimens. In the predictions, it is assumed that a delamination is initiated in the first cycle and continues to grow at a constant rate. In reality, initiation occurs over a period of time and the initial growth rate of a delamination is higher until a steady state is reached [7, 8]. Combined with scatter in initiation life, variation in the prediction and experimental results is expected; however, by considering an initial delamination size, this scatter could be minimized. An alternative scenario for delamination growth in a bonded patch repair is the formation of a delamination in the centre of the patch where the underlying adherent is damaged. In such a case, the analysis would include the growth of and load bridging around the cracked adherent. This would be similar to crack propagation and delamination growth in an FML panel [7, 8]. Crack Growth with an Intact Bonded Stringer Rodi [24] employed the proposed methodology to predict the bridging effect of an intact stringer on the crack growth of a centre crack tension (CCT) panel with a central stringer. The panel comprised of a Glare sheet with a titanium stiffener bonded to the centre of the panel. The proposed methodology was used to predict both delamination growth and crack propagation in the Glare sheet as well as

Damage tolerance philosophy for bonded aircraft structures

87

delamination growth between the stiffener and underlying sheet. Thus, the compatibility equation given in (6) was modified to:

v∞ − vbr − vst = δ 2 + δ ad

(14)

where vst is the crack opening due to the load bridging effect of the intact central stiffener. Using this modified compatibility equation, the bridging effect of the intact stringer could be predicted. Figure 7 and Figure 8 show a comparison of the predicted crack growth and predicted delamination growth to experiment. Excellent agreement in both delamination and crack growth was observed. The proposed methodology was also extended by Rodi [24] for prediction of the effects of broken stiffeners and double lateral stiffeners on crack growth in a CCT type stiffened panel. Similar success in applying this methodology was also achieved for these cases.

0.0002

Glare3 – 5/4 – 0.4 f = 10Hz 2a0 = 25 mm W = 200 mm Wst = 25 mm

0.00018 0.00016

Slam=120 MPa R=0.05

Da/DN [m m /c y c le]

0.00014 0.00012 0.0001 0.00008

Slam=100 MPa R=0.05

0.00006

0.00004

Measurements Prediction

0.00002 0 12.5

17.5

22.5

27.5

32.5

37.5

42.5

47.5

52.5

57.5

Crack length [mm]

Figure 7: Comparison between the predicted and measured crack growth rate of Glare3-5/4-0.4 stiffened by intact central titanium straps, W=200 mm, Wst=25mm, a0=12.5 mm [24].

C.D. Rans and R.C. Alderliesten

88 12

Glare3 – 5/4 – 0.4 2as = 25 mm W = 200 mm Wst = 25 mm Slam =120 MPa Sst =232 MPa

10

b [mm]

8

Measurement

6

Prediction

4

2

0 12.5

front left front right Rear Left Rear Right Prediction

22.5

32.5 42.5 Location along the crack length [mm]

52.5

62.5

Figure 8: Comparison between the predicted and measured delamination shape of Glare3-5/4-0.4 stiffened by intact central titanium straps, W=200 mm, Wst=25mm, a0=12.5 mm, Slam=120 MPa and R=0.05 [24].

CONCLUSIONS A generic damage tolerance approach to bonded structures which permits prediction of adherent crack growth, adhesive delamination, and the interaction between adherent crack and interface delamination has been presented. This methodology was developed for predicting the coupled delamination and crack growth phenomenon in fibre metal laminates, but has shown promise for being applied to a wider range of bonded aircraft structures. The power of such an approach, particularly for cases where adherent cracking and adhesive delamination are expected simultaneously such as in bonded patch repairs, is evident. The methodology builds upon established methods for predicting crack growth and delamination growth under fatigue loading conditions and provides a sound description of their interaction. Using such an approach, numerous damage scenarios can be analyzed permitting a thorough damage tolerance analysis of bonded aircraft structures. The validity of this approach has only been demonstrated for a narrow range of applications. Although it is anticipated that the approach has a wide reaching generic applicability, the validity of the approach still needs to be verified for a wider range of materials and interfaces, including pure composite materials and cocured vs. secondary bonded interfaces.

Damage tolerance philosophy for bonded aircraft structures

89

REFERENCES 1. 2. 3.

4.

5. 6. 7.

8.

9.

10. 11.

12.

13.

14.

15. 16.

L.J. Hart-Smith, Adhesive-bonded single lap joints, Langley Research Center, NASA, 1973. J.W. van Ingen, Stress analysis of adhesively bonded single lap joints, Delft University of Technology, 1993. R.S. Fredell, Damage Tolerant Repair Techniques for Pressurized Aircraft Fuselages. Ph.D. dissertation, 1994, Delft University of Technology: Delft. L.R.F. Rose, Theoretical analysis of crack patching, in Bonded Repair of Aircraft Structures, A.A. Baker and R. Jones, Editors. 1980, Martinus Nijhof Publishers. L.R.F. Rose, An application of the inclusion analogy for bonded reinforcements. Int. J. Solids Struct., 13 (1981) 827-838. D.M. Gleich, Stress Analysis of Structural Bonded Joints. Ph.D. dissertation, 2002, Delft University of Technology: Delft. R.C. Alderliesten, Fatigue Crack Propagation and Delamination Growth in GLARE. Ph.D. dissertation, 2005, Delft University of Technology: Delft, the Netherlands. R.C. Alderliesten, J. Schijve, and S. van der Zwaag, Application of the SERR approach for delamination growth in Glare. Eng. Fract. Mech., 73 (6) (2006) 697-709. R. Marissen, Fatigue Crack Growth in ARALL, A Hybrid AluminiumAramid Composite Material. Ph.D. dissertation, 1988, Delft University of Technology: Delft. A. Hosoi, et al., High-cycle fatigue characteristics of quasi-isotropic CFRP laminates. Adv. Compos. Mater, 16 (2) (2007) 151-166. J.J. Muñoz, U. Galvanetto, and P. Robinson, On the numerical simulation of fatigue driven delamination with interface elements. Int. J. Fatigue, 28 (2006) 1136-1146. A. Sjögren and L.E. Asp, Effects of temperature on delamination growth in carbon/epoxy composite under fatigue loading. Int. J. Fatigue, 24 (2002) 179-184. D. Tumino and F. Cappello, Simulation of fatigue delamination growth in composites with different mode mixtures. J. Compos. Mater., 41 (20) (2007) 2415-2441. M. Beghini, L. Bertini, and P. Forte, Experimental investigation on the influence of crack front to fibre orientation on fatigue delamination growth rate under mode II. Compos. Sci. Technol., 66 (2006) 240-247. J. Schijve, Fatigue of Structures and Materials Kluwer Academic Publishers, Dordrecht. H. Tada, P.C. Paris, and G.R. Irwin, The Stress Analysis of Cracks Handbook. 3rd ed The American Society of Mechanical Engineers, New York, 2000.

90

17.

18.

19. 20.

21. 22. 23.

24.

C.D. Rans and R.C. Alderliesten

R.C. Alderliesten, Analytical prediction model for fatigue crack propagation and delamination growth in Glare. Int. J. Fatigue, 29 (2007) 628-646. C.T. Lin and P.W. Kao, Delamination growth and its effects on crack propagation in carbon fiber reinforced aluminum laminates under fatigue loading. Acta Mater., 44 (3) (1996) 1181-1188. C.T. Lin and P.W. Kao, Fatigue delamination growth in carbon fibrereinforced aluminium laminates. Composites Part A, 27 (1996) 9-15. R. Marissen, Fatigue crack growth in ARALL: A hybrid aluminiumaramid composite material. Ph.D. dissertation, 1988, Delft University of Technology: Delft, the Netherlands. J.E. Schut, Glare: Delamination growth at low temperatures. Masters thesis, 2006, Technical University of Delft: Delft. A.S.J. Suiker and N.A. Fleck, Crack tunnelling and plane-strain delamination in layered solids. Int. J. Fract., 125 (2004) 1-32. T. Beumler, Flying Glare, A contribution to aircraft certification issues on strengths properties in non-damaged and fatigue damaged Glare structures. Ph.D. 2004, Delft University of Technology: Delft. R. Rodi, The effect of external stiffening elements on the fatigue crack growth in Fibre Metal Laminates. Masters thesis, 2007, Technical University of Delft: Delft.

25th ICAF Symposium – Rotterdam, 27-29 May 2009

FIRST DIAMOND ‘Damage tolerance’ for the structural honeymoon Steve Swift Civil Aviation Safety Authority (CASA), Australia 1

Abstract: This paper offers a fresh and simple approach to setting the threshold for structural inspections. It shows how damage tolerance and the diamond work the same as for any other interval, and how they would work even better with risk management, human factors and less prescriptive regulation.

INTRODUCTION I sorrow that all fair things must decay

⎯ American poet, Fitz-Greene Halleck Aircraft structure decays. So, it needs inspection. A big issue is the threshold. When does the ‘honeymoon’ end so the structure needs its first inspection? This paper offers a fresh and simple approach. The threshold is a big issue because it affects: •

safety for air travellers

•

sales for aircraft manufacturers

•

costs for airlines

1

The views in this paper are personal and not necessarily CASA policy

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 91–107. © Springer Science+Business Media B.V. 2009

Steve Swift

92

Getting it right is hard: •

tension between economy and safety

•

uncertainty and variability

•

sensitivity to small changes

•

prescriptive regulation

Disputes are common. Most are behind closed regulatory doors. But, some have emerged at ICAF and other fora. What is the best and fairest way to set thresholds? For the answer, this paper looks: •

anew at damage tolerance

•

afield at risk management and human factors

It continues the ‘diamond’, which, in 2007, won the Whittle Safety Award from the International Federation of Airworthiness.

Figure 1 The paper’s pragmatism comes from 25 years of putting threshold theory into practice for Australia’s 13,000 civil aircraft, acknowledged as one of the hardestworking fleets in the world. Its scope is metallic structure in civil aircraft, but some issues are universal.

First Diamond

93

DAMAGE TOLERANCE AND THE DIAMOND First, what is, ‘damage tolerance’? As Rough Diamond [1] explained at ICAF 2005, for civil rules it is not fracture mechanics or a design property, but aviation’s application of a universal method for maintaining anything prone to damage or decay. 2 In aviation, we use it to set the interval between structural inspections. Reliabilitycentred maintenance uses it to set the P-F interval for on-condition tasks for machines. [2] Health professionals use it to set the interval for check-ups, for our teeth to our bowels. It can even tell you how often to weed your lawn.

Figure 2: Damage tolerance is not just for aviation Rough Diamond proposed a simple teaching tool and memory aid for the method, called the diamond:

where? size?

site detectable

scenario dangerous

duration

what? size? detectable to dangerous

Figure 3 2

In the jargon (and in this paper), ‘damage’ includes the natural and gradual (like fatigue and corrosion), which we might also call ‘decay’, as well as the unnatural and sudden (like a scratch or a ding), its more normal sense

Steve Swift

94

Then, at ICAF 2007, a USAF paper [3] mentioned the importance of ‘human factors’ for inspection, including the detectable size. Some are: •

skill variation – ask about the worst inspector, not the best

•

environment – ask about the worst conditions, not the best

•

optimism – ask about the worst missable, not the best detectable

•

pride – ask about the ability of others, not their own

So, human factors would prefer the diamond’s ‘detectable’ facet be ‘missable’. At the same ICAF, Rusty Diamond [4], a paper that applied damage tolerance and the diamond to corrosion, reminded us the issue is not always safety. Often, it is repairability. So, flexibility would prefer the ‘dangerous’ facet be ‘permissible’, as Broek recommends in [5]. Third, memorability would prefer the ‘duration’ facet be ‘interval’ (to rhyme more with ‘missable’ and ‘permissible’). So, the diamond evolves. For this paper, it is:

where? size?

site missable

scenario

what?

permissible

size?

interval

missable to permissible

Figure 4 If the diamond works for the interval, including for corrosion, will it also work for the threshold?

First Diamond

95

DOES THE DIAMOND WORK FOR THE THRESHOLD? For each site and scenario, here is how the diamond works for the interval:

damage

permissible

missable time

interval inspection

inspection

Figure 5 The interval must assure that damage missable by one inspection stays permissible until the next.

damage

Isn’t the threshold the same, except the first inspection is in manufacture, not maintenance?

permissible

missable threshold

inspection (manufacture)

time

inspection (maintenance)

Figure 6 So, the diamond works for the threshold too. But, will it always work in the form we call the initial flaw method? We will soon see.

Steve Swift

96

USING THE DIAMOND FOR THE THRESHOLD Site We should think more broadly than we often do. Sites are: •

not just stress peaks on a finite or boundary-element mesh, but also…

wet areas (corrosion)

•

exposed areas (accidental damage)

not just holes, but also…

near holes

•

assembled areas (fretting and pre-loading)

free edges [6]

internal flaws

not just their corners, but also…

down their bores, like in this wing attach fitting (left) from a Strikemaster (right)

For an old aircraft type, compare service bulletins with predictions. It is sobering how often we get the site wrong. [7] has several examples.

First Diamond

97

Scenario We should think past fatigue to the other damage FAR 25.571 [8] lists: •

corrosion, such as…

exfoliation

•

manufacturing defects, such as…

nicks (like in this helicopter TR blade)

•

stress corrosion

faulty material (like in this landing gear)

accidental damage, such as…

scribe lines (from removing sealant with the wrong type of

changing a bush (the damaged hole fatigued and broke the wing on this Viscount)

Of course, some will interact with or transition to fatigue, as happened with some of the above. And, like the site, we often get the scenario wrong.

Steve Swift

98

Missable This is where we question the range of applicability of the initial flaw method. For each scenario at each site, missable helps us size the damage for the start. In FAR 25.571, it is the ‘maximum probable size that could exist as a result of manufacturing’. 3 So, isn’t it also the maximum size probably missable by manufacturing inspections? 4 Either way, for the initial flaw method, it is the size that fracture mechanics analytically ‘grows’ to the threshold. The other method in FAR 25.571, the fatigue method, does not need to know the missable size, but does need a random sample that has damage typical of what is missable for manufacture. It starts by testing it, and then statistics account for uncertainty and variability, by a probability density function, to get the threshold. 5

probability density

Figure 7 shows the two methods graphically:

statistics Initial Flaw Method: Fatigue Method:

variability uncertainty

damage size

mean threshold

permissible

Finish fracture mechanics

test threshold test

Start = maximum missable Start = random sample

threshold

time

Figure 7: Comparing the initial flaw and fatigue methods 3

‘Probable’ is similar to ‘likely’ in MSG-3 [9]. Civil rules do not expect us to anticipate the ‘improbable’, ‘unlikely’ or ‘rogue’. 4 We refer to POD (probability of detection) more than its complement, POM (probability of missing). But, missable damage, which could keep growing, is more important than the detectable. 5 Common probability density functions for fatigue are log-normal and Weibull

First Diamond

99

Even for crack-like initial flaws, neither method is intrinsically better. They should both calculate the same threshold. But, it is often shorter for a so-called ‘rogue’ flaw, which is improbably big. We could shorten a fatigue threshold too, if we wanted, with an improbably big scatter factor. Both methods influence design: they keep stresses low, and the initial flaw method favours tough materials. The initial flaw method is less likely to influence manufacturing quality, because the initial flaw is often prescribed. Both methods have uncertainty. For the initial flaw method, it is in the fracture mechanics and the data it uses, especially for the short cracks common for thresholds. For the fatigue method, it is in the probability density function, if the structure tested is similar but not the same. The FAA agree with Emmerson [10] when they accept both methods for: •

new small aircraft (FAA Advisory Circular 23-13A) 6 [11]

•

all old aircraft (FAA Advisory Circular 91-82) [12]

•

some structure in new large aircraft

For the third, for new large aircraft, why do the FAA only accept the initial flaw method for: Single load path structure and multiple load path ‘fail-safe’ structure and crack arrest ‘fail safe’ structure, where it cannot be demonstrated that load path failure, partial failure, or crack arrest will be detected and repaired during normal maintenance, inspection or operation of an airplane prior to failure of the remaining structure. 7

If such partiality is not necessary, is it even desirable? As Airbus reminded us at ICAF 2005 [13], a problem with exclusively requiring the initial flaw method is the tendency to neglect non-crack-like manufacturing damage, which is just as dangerous and likely, just because fracture mechanics cannot analyse it. Others have aired similar concerns in the past. Fortunately, for large airliners, Maintenance Review Boards (MRB), working to MSG-3, look more broadly than does Certification, working to FAR 25.571.

6 AC 23-13A has fatigue curves from World War 2 aircraft, which should be conservative for modern civil manufacturing quality 7 FAR 25.571 (a) (3) (i) and (ii). It is often problematic sorting this structure from the rest. Except for fracture mechanicists, who prefer the initial flaw method for all structure.

Steve Swift

100

Over-inspecting holes cannot compensate for under-inspecting elsewhere

Fortunately, for large airliners, Maintenance Review Boards look more broadly than does Certification

Figure 8 So, when using the diamond, remember that missable is not just for cracks and the initial flaw method. The concept holds for all damage scenarios and applications of damage tolerance and the diamond, whether by analysis or test. Permissible Residual strength should end the threshold, just like any other interval, because strength is what FAR 25.571 says our inspections must assure. But, for example, [2] and [14] suggest inspectability. It would have two problems. First, inspectability is no measure of strength (sometimes damage is dangerous before detectable). Second, it would discourage improving inspectability (because improving it would shorten the threshold!). Interval This is the time from missable to permissible. The threshold is just a special interval – the first. That is exactly how MSG-3 defines it. 8 So, like any other interval, we should factor it for uncertainty and variability from all sources, not just initial damage size and inspectability. The simple factors of 2 or 3 by which we often divide the interval seem more arbitrary than the scatter factors we use for traditional life limits. See [11] for example. There is still uncertainty and variability. Is there still statistical rigour?

8

From MSG-3’s Glossary: “Threshold: See ‘Interval – Initial’”

First Diamond

101

RISK MANAGEMENT Risk management is part of any Safety Management System. From ICAO’s Safety Management Manual [15], the steps are simple and common sense:

Figure 9: Risk management process Although for an organisation, could it not also be for structure, with the hazards being the scenarios at the sites? Could it not broaden the range of sites and scenarios we assess, for the ‘holism’ Hoeppner [16] and others advocate, by broadening the input, as we do for MRB [17]? With MRB, experienced operators team their wisdom with the regulator and the manufacturer. It helps when the manufacturer’s representatives are from the materials, production and product support departments, as well as from design. Then, safety management systems require the process to continue. In this context, it means what we often call ‘continuing airworthiness’.

Steve Swift

102

HUMAN FACTORS We have already considered human factors for the diamond. There are also human factors for thresholds. For example, conservatism is good. But, not too much. Too short a threshold can cause: •

disbelief

•

disengagement

•

delusion

Disbelief The Fairchild Metro is a 19-passenger turboprop airliner. First, Fairchild did the thresholds by the fatigue method. Years later, the FAA paid Fairchild to redo them by the initial flaw method. [18] The latter are shorter – so short no one believes them. Not even the FAA. They do not think exceeding them is ‘unsafe’ (FAR 39) enough for an Airworthiness Directive. So, in the USA at least, they languish. The story is similar for most of the initial flaw based inspection thresholds for Cessna’s propeller-driven twins.

Figure 10: Fairchild Metro (left) and Cessna 402C (right) Disengagement Too short a threshold means too many inspections finding too little. It fosters complacency and disengagement for future inspections, reducing their effectiveness. Delusion Too many inspections finding too little means too much chance of the delusion to extend the interval instead of the threshold. You will even see it in international maintenance publications, like [19]. See A Collective Approach To Aircraft Structural Maintenance Programs [20]. It includes an example of a turboprop airliner that threw a blade soon after take off.

First Diamond

103

SPECIAL ISSUES Realism The aim for a threshold is often the Design Service Goal (DSG), often 20 years. 9 But, if the optimism survives certification, how long will it survive service?

Figure 11: A380 flap track fairings Sometimes, not long. Flap track fairings on Airbus’ A380 cracked unexpectedly, so early, the threshold is now only 100 flights! [22] This is not to criticise Airbus. It is just to remind us again about the difficulty of prediction. Extension Service experience can extend a threshold. But, statisticians warn that it takes hundreds of nil findings, not the handfuls many think suffice. And, it takes confidence that findings, if any, have been reported. Often, by the time extension is justifiable, it may be too late if too few aircraft are still young enough to benefit. Efficiency Is a threshold always efficient? Sometimes it costs more to calculate and certify than the inspections it saves. It could if the fleet is small or the inspection is easy. But, beware too much conservatism. Is a threshold always safe? MSG-3 warns it is not if accidental damage is likely. Cold expansion 10 Can a threshold take credit for cold expansion? Has its quality risen faster than its quantity per aircraft, so that error is now no more ‘probable’ than for other manufacturing processes we trust? Sometimes, regulators allow cold expansion for fatigue tests, but not for analyses. The inconsistency comes from their wavering as they struggle to balance encouragement with caution. 9

For a definition of DSG, see FAA Advisory Circular 25.571-1C, page 2 [21] Cold expansion is a fatigue-improvement process for holes and other structural features. A modern airliner has thousands of cold-expanded holes.

10

Steve Swift

104

Safe life Some civil rules, like FAR 23.573 [23], allow ‘safe life’ or ‘damage tolerance’. Imagine certifying a structure with a safe life. Then, imagine recertifying it later for damage tolerance. This is common for civil aircraft, for life extensions. Why should the old safe life not become the new inspection threshold? Intuitively, why should it change the inspection-free period (the ‘structural honeymoon’) if all we change is the maintenance after it ends? Also, why do regulators require ultimate strength at the safe life, but only limit strength at the threshold? Grace What do we do with a new threshold for an old fleet? Safety says inspect immediately; economy says inspect conveniently. How should we balance the two? A present problem – and a challenge for regulators – is that there are no rules.

Figure 12: Ansett Boeing 767 New thresholds, with no grace, grounded seven of Ansett’s Boeing 767s (some of the oldest in the world) until they could inspect them. [24] Structural Health Monitoring (SHM) SHM inspections are often free and frequent, if not continuous. So, they should not need a threshold. But, can we trust SHM alone? The SHM method would have to prove it is adaptive and extensive, as well as reliable. So far, most methods are so fine-tuned to one scenario at one site they would be unforgiving if we got either of them wrong, as we often do.

First Diamond

105

CONCLUSIONS Yes, damage tolerance works for the ‘structural honeymoon’, so the diamond can help us set the threshold, the time when inspections must start. 11 This first diamond is the same as the diamond for later intervals, except sizing the ‘missable’ damage depends on the inspections in manufacture, not maintenance. But, a particular form of damage tolerance, the initial flaw method, does not always work, because not all damage is crack-like and analysable by fracture mechanics. And, sometimes it is too conservative to be credible and enforceable. Its prescription, by policy if not by rule, hinders safety and economy.

RECOMMENDATIONS We should: •

unify and simplify our regulation and management of thresholds and intervals

•

specify principles and outcomes and not prescribe methods

•

learn from the fields of risk management and human factors

•

anticipate surprises in service by designing structure that does not need sensitive local NDI 12 , and by having a robust continuing airworthiness system

•

encourage Structural Health Monitoring that is adaptive and extensive, so a threshold is no longer necessary

ACKNOWLEDGEMENTS I thank the Australian CASA for supporting my preparation and presentation of this paper. I also thank Prof Graham Clark, Australia's ICAF delegate, and Bob Eastin, the FAA’s Chief Scientific and Technical Adviser for Fracture Mechanics. They inspire and encourage me every time we speak. Finally, I thank Kathy for still loving me while I again thought more about another 'diamond' than hers.

11

Purists will note that this paper uses ‘threshold’ as both a period and a point in time. Strictly, it is only the second. MSG-3 has the same ambiguity. Practically, since the reference is zero, it makes little difference. At times, the first sense is more helpful. 12 The structure needs ‘fail-safety’ or ‘large damage capability’

Steve Swift

106

REFERENCES [1]

Eastin, R. and Swift, S. (2005), Rough Diamond, Proceedings of the 23rd ICAF Symposium, Hamburg, Germany

[2]

Moubray, J. (1997), RCM II, Reliability-centred maintenance, Second edition, pages 145 and 162, Butterworth-Heinemann, UK

[3]

Gallagher, J. (2007), A Review of Philosophies, Processes, Methods and Approaches that Protect In-Service Aircraft from the Scourge of Fatigue Failures, Proceedings of the 24th ICAF Symposium, Naples, Italy

[4]

Swift, S. (2007), Rusty Diamond, Ibid.

[5]

Broek, D. (1988), The Practical Use of Fracture Mechanics, page 7, Kluwer Academic Publishers, Dordrecht, Boston and London

[6]

Swift, S. (1995), The Aero Commander Chronicle, Proceedings of the 18th ICAF Symposium, Melbourne, Australia, EMAS, UK

[7]

Swift, S. (1999), Gnats and Camels, Proceedings of the 20th ICAF Symposium, Bellevue, Washington, USA

[8]

FAA (1998), Damage-tolerance and fatigue evaluation of structure, FAR 25.571, Amendment 25-96, USA

[9]

Air Transport Association (2007), ATA MSG-3, Operator/Manufacturer Scheduled Maintenance Development, Revision 2007.1, Washington DC, USA

[10]

Emmerson, A. (1992), Airworthiness Control in the Face of Structural Fatigue, Proceedings of the Ageing Commuter Aircraft Conference, Canberra, Civil Aviation Safety Authority, Australia

[11]

FAA (2005), Fatigue, Fail-Safe, And Damage Tolerance Evaluation Of Metallic Structure For Normal, Utility, Acrobatic, And Commuter Category Airplanes, AC 23-13A, USA

[12]

FAA (2008), Fatigue Management Programs for Airplanes with Demonstrated Risk of Catastrophic Failure Due to Fatigue, AC 91-82, page 17, USA

[13]

Medina, M, Kimmins, S, Rodrigo, P. (2005), Using the Initial Flaw Concept in Current and Future Aircraft Certifications, Proceedings of the 23rd ICAF Symposium, Hamburg, Germany

[14]

Broek, D. (1995), Manual for the Damage Tolerance Analysis of Repairs and Modifications of Aircraft Structures, FracturResearch TN 9501, page 79, Galena, Ohio, USA

First Diamond

107

[15]

ICAO (2006), Safety Management Manual (SMM), Doc 9859, AN/460, First Edition, page 6-2, Montreal, Canada

[16]

Hoeppner, D. (2002), From Safe Life to Holistic Structural Integrity Design, Presentation for NRC IAR Structures Workshop, Ottawa, Ontario, Canada

[17]

FAA (1997), Maintenance Review Board Procedures, AC 121-22A, USA

[18]

FAA (2000), Development of Supplemental Inspection Report for the Fairchild Metro SA226 and SA227 Airplane, DOT/FAA/AR-00/18, Office of Aviation Research, Washington DC, USA

[19]

International MRB Policy Board (2008), Evolution/Optimization Guidelines

[20]

Swift, S. (2008), A Collective Approach to Aircraft Structural Maintenance Programs, Proceedings of the International Air Safety Seminar, Honolulu, Flight Safety Foundation, USA

[21]

FAA (1998), Damage Tolerance and Fatigue Evaluation of Structure, AC 25.571-1C, USA

[22]

EASA (2008), Wings – Movable Flap Track Inspection/Replacement, AD 2008-0216, Cologne, Germany

[23]

FAA (1996), Damage tolerance and fatigue evaluation of structure, FAR 23.573, Amendment 23-48, USA.

[24]

Australian Transport Safety Bureau (2002), Investigation into Ansett Australia maintenance safety deficiencies and the control of continuing airworthiness of Class A aircraft, Report BS/20010005, Canberra, Australia

[25]

Clark, G. (2005), A Review of Australian and New Zealand Investigations on Aeronautical Fatigue During the Period April 2003 to March 2005, DST-TN-0624, DSTO Platforms Sciences Laboratory, Australia

Fairing

–

25th ICAF Symposium – Rotterdam, 27–29 May 2009

SURVEY OF STRUCTURAL REPAIRS AND ALTERATIONS IN TRANSPORT CATEGORY AIRPLANES Michel D. Bode1, Walter Sippel2, and John G. Bakuckas3 1

Sandia National Laboratories Airworthiness Assurance NDI Validation Center (AANC) 2 Transport Standards Staff Airframe and Cabin Safety Branch, ANM-115 Federal Aviation Administration 3 William J. Hughes Technical Center Airworthiness Assurance Branch, AJP-6360 Federal Aviation Administration

Abstract: The work presented here details a survey of structural repairs, alterations, and modifications (RAM) on transport airplanes that was conducted over the past year and a half as part of a two-year project. The goal of this effort is to provide data to better understand the risks that RAMs may pose for developing widespread fatigue damage (WFD). Surveys were conducted on retired airplanes at aircraft salvage locations and on in-service airplanes at the operator’s heavy maintenance locations and will be compared to a similar survey conducted by the Airworthiness Assurance Working Group in the 1990s. Additionally, specimens from retired airplanes were acquired and in-depth teardown inspections were performed to look for the presence of damage indicative of WFD. A database was developed to analyze the data for WFD risk assessments. Once completely populated, recorded data will be examined for trends to quantify the risks that RAMs may pose for developing WFD.

INTRODUCTION Structural fatigue has long been recognized as a significant threat to the continued airworthiness of airplanes. This is because even small fatigue cracks can M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 109–121. © Springer Science+Business Media B.V. 2009

110

M.D. Bode et al.

significantly reduce the strength of airplane structure. A phenomenon referred to as widespread fatigue damage (WFD) is identified as a severe degraded condition that threatens the continued airworthiness of airplanes, is theoretically inevitable, and will be reached at some point in the life of a structure. The Federal Aviation Administration (FAA) has defined WFD as the simultaneous presence of cracks at multiple structural locations that are of sufficient size and density such that the structure can no longer maintain its residual strength. A major concern of WFD is that fatigue cracks are initially so small that they cannot be reliably detected with existing inspection methods. The small undetectable cracks then “link-up” and grow together. This growth may be very rapid and may result in catastrophic structural failure of an airplane. To address this safety concern, the FAA issued a Notice of Proposed Rulemaking (NPRM) on the subject of WFD in April 2006. The WFD NPRM contained extensive requirements pertaining to certification and operation of certain transport category airplanes to preclude WFD in those airplanes. Among other things, the proposal included a requirement for design approval holders to evaluate baseline airplane structure and certain repairs, alterations, and modifications (RAMs) for WFD. Commenters to the NPRM suggested changes to the requirements for evaluating RAMs for WFD. Although there is a technical possibility of a WFD-related accident involving RAMs, commenters stated that there are no recorded accidents attributed to WFD occurring in properly installed RAMs. All adverse service experience to date has been limited to baseline airplane structure. Additionally, RAM structure that may be susceptible to WFD may be limited to only one or a few airplanes, whereas baseline structure typically exists on each airplane. Under the Aging Airplane Safety Rule [1] and the Damage Tolerance Data for Repairs and Alterations Rule [2], surveys and damage tolerance evaluations will be performed on RAMs. Those surveys and evaluations should provide some degree of mitigation of risk. Based on this, there was general agreement among industry stakeholders that there is a low risk of WFD occurring in RAMs. For these reasons, the FAA determined that WFD assessments should be focused on baseline structure for only establishing a limit of validity of the engineering data that supports the structural maintenance program for airplane models. This is a change that the FAA made to the original proposal as a result of the comments it received. This and other changes are described in a technical document that was published in the Federal Register on November 7, 2008 [3]. The public was allowed the opportunity to comment on the technical document. The comment period closed on December 22, 2008. The FAA is further assessing whether additional regulatory actions are needed to address RAMs that may pose a risk of developing WFD. To support this assessment, the FAA is funding additional research at the William J. Hughes Technical Center.

Survey of structural repairs and alterations in transport category airplanes

111

This paper provides details on that research project, which is in progress. The project was initiated in December 2007 and is scheduled to be completed in December 2009. There are two goals of the research project. The first goal is to determine the structural integrity of RAMs by sampling high-time airplanes with RAMs. The second goal is to determine if there are trends indicating increased risk of WFD occurrence and subsequent risk of catastrophic structural failures in high-time RAMs. It is anticipated that the FAA will use the results to conduct additional analysis and assess the risk of RAMs developing WFD in aging transport category airplanes.

PREVIOUS INDUSTRY SURVEYS In the early 1990s, the Airworthiness Assurance Working Group (AAWG) was tasked with assessing the general condition of repairs on transport airplanes and developing recommendations concerning whether new or revised requirements and compliance methods for structural repair assessments of existing repairs should be instituted and made mandatory for several specific transport category airplanes. In 1996, a final report was released detailing the data they collected during two surveys of repairs done in 1992 and 1994 [4]: •

•

1992 Survey: A survey of lower fuselage repairs on stored airplanes was conducted at aircraft storage and salvage locations in Mojave, CA and Amarillo, TX. The work was conducted by teams of engineers from FAA Aircraft Certification Offices, FAA Flight Standards Offices, operators, and original equipment manufacturers (OEMs). The teams surveyed 30 airplanes and assessed a total of 356 repairs. 1994 Survey: A survey of repairs on in-service airplanes was conducted at various operators’ maintenance bases by only engineers from OEMs, which included Airbus, Boeing, Douglas, and Lockheed. The teams surveyed 35 airplanes and assessed a total of 695 repairs.

The scope of the studies was limited to an external visual observation of external lower fuselage plating repairs. The objectives were to gain first-hand observations of typical repairs, and to sample the numbers, types, proximities, and conditions of repairs. The quality of each repair was assessed based on AAWG repair criteria and with OEM size and proximity limits. Each repair was examined and classified into three categories [4]: •

Category A: A permanent repair for which the baseline zonal inspections are adequate to ensure continued airworthiness

M.D. Bode et al.

112

• •

(inspectability) equivalent to unrepaired surrounding original structure. Category B: A permanent repair that requires supplemental inspections to ensure continued airworthiness. Category C: A temporary repair that will need to be reworked or replaced prior to an established time limit. Supplemental inspections may be necessary to ensure continued airworthiness prior to this limit.

Table I shows a breakdown of the models and numbers surveyed in 1992 and 1994. In all, a total of 65 airplanes and 1051 repairs were assessed. The study found that about 40% of repairs met the criteria for a Category A repair. The 60% that fell into Category B or C were primarily due to size and/or proximity criteria.

3

Table I. AAWG fuselage survey statistics [4].

Survey of structural repairs and alterations in transport category airplanes

113

The overall goal was to develop a qualitative opinion of repairs as a safety concern. In general, the AAWG concluded there was a need for repair assessment evaluations; although no immediate safety concern was observed. They cited a need for the following, (1) updates to structural repair manuals and new guidance materials for repairs, (2) OEM developed repair assessment procedures for existing and new repairs, and (3) more training for persons doing repair assessments. In addition, the AAWG found the majority of repairs were on the fuselage pressure shell and older airplanes generally have more repairs. As a result of the work, the AAWG recommended certain regulatory changes that have led to implementation of the Repair Assessment Program [5 and 6]. In comparison, the objective of the current project is to determine the risk of RAMs developing WFD. While the objective of the current project is more specific, the net effect is similar in that the results will be used to support decisions on rulemaking activities. Because the AAWG work provided a benchmark for earlier rulemaking activities, the initial goal of the project was to survey at least as many RAMs identified as being surveyed in the 1996 report [4]. It was anticipated that the bulk of the RAMs would be repairs, and most of those would be found on fuselage skins, as it has been the structure of most concern in terms of WFD occurrence. While the fuselage pressure boundary was the primary concern, other structures including wing and empennage were considered. The primary output for this project is a Field Survey and Teardown Database detailing all the data collected for each RAM on each airplane examined. Details of the database are described later in this paper. The database includes information to support the following determinations: • • • • • • •

Average number or distributions of repairs per airplane for each model and type of repair. Size distributions of repairs. Age distributions of repairs (number of cycles or number of hours relative to Design Service Goal). Locations of repairs relative to fatigue sensitive structure. Repair source data and ranking of repair workmanship. Reason for repairs. Detailing damage subsequent to the repairs, especially cracking.

FAA FIELD SURVEYS OF RETIRED AIRPLANES One objective of this project is to conduct a field survey of RAMs that exist in the U.S. registered fleet of transport category airplanes that have been certified under part 25 of the Code of Federal Regulations or 4b of Civil Air Regulations. The field survey applies to older U.S. registered airplanes that have usually

M.D. Bode et al.

114

accumulated numerous flight cycles or flight hours or both and are still being operated or have been recently retired. This section will discuss field surveys on retired airplanes, which were performed early in the project. The survey examines the condition of RAMs and surrounding structure for evidence of WFD. RAMs and surrounding structure being surveyed include the fuselage pressure boundary, wing skins and spars, alterations due to gross weight or pressure increases, alterations by addition or removal of antennas, replacements of entire lap joints, and passenger-to-cargo conversions. The FAA compiled a list of airworthiness directives and service bulletins for airplanes of interest to use as a guide in selecting retired airplanes for field surveys. A broad cross-section of airplane models were selected and included: Boeing 727, 737, 747 (specifically flat areas), 757, 767, DC-9, MD-80, MD-11 models, and Airbus A-300, A-310, and A-320 series models. However, selection was influenced by availability at various salvage locations. About a dozen different salvage companies were queried as to airplane availability. The test bed airplanes available at the FAA’s Airworthiness Assurance NDI Validation Center (AANC) were also included in the survey. Data was collected at various field locations, and the listing of aircraft actually surveyed is shown in Table II. During one field survey trip, a subset of specimens from a retired B737-200 was returned to the AANC inspection laboratory for disassembly and inspection. Teardown inspections are discussed in a later section.

Airplane Model B727 B737 B747 B767 DC-9 Total

Number of Retired Airplanes 8 4 2 1 1 16

Number of RAMs Surveyed 48 136 88 5 44 321

Table II. Summary of retired airplane field surveys.

The survey process involved field inspections using general visual inspection, measurement of the RAM structural dimensions, and photographic documentation (including both sides of a RAM whenever possible). An “Aircraft Information & Documentation Form” was used to ensure all necessary data was collected for each RAM as discussed above. When available, associated maintenance records were acquired and married to the field survey documentation to put an age and engineering design basis on each RAM. In general, all areas of the airframe were considered in the field surveys, but in some cases, only portions of airplanes were

Survey of structural repairs and alterations in transport category airplanes

115

surveyed due to logistical reasons. For example, surveys were not performed on certain areas when upper areas were not accessible due to lack of high lift equipment or there were data taken from teardown specimens previously acquired. As of the publication date of this paper, the majority of repairs surveyed have been found to be of sufficient quality. Of particular interest are large RAMs. Figure 1 shows two large repairs from a retired airplane. Note, the two large repairs are outlined in black and are just below a lap joint replacement modification.

Figure 1. Example of adjacent large RAMs from a retired airplane. An “Aircraft Information & Documentation Form” was used to ensure all necessary data was collected for each RAM, as previously discussed. Data was manually entered into the Field Survey and Teardown Database after the field survey was complete. As a minimum, the following information was gathered for each RAMs installation: • • • • • • •

Proximity to other RAMs, WFD susceptible structure and/or other fatigue critical structure. The size, shape, fastener spacing, edge distances, and doubler thickness(s). The general condition of the RAMs. The materials used including fastener and sealant types. Photos and/or simple engineering drawings. The location of the RAMs on the airplane. If maintenance records were available, determined the engineering basis for the RAMs.

M.D. Bode et al.

116

FAA FIELD SURVEYS OF IN-SERVICE AIRPLANES While there was no shortage of relevant retired airplanes to survey, the data concerning age of the RAMs and the engineering basis under which they were designed were usually not available. In general, the required information was more readily available for in-service airplanes than for retired airplanes. Thus the project also focused on surveying in-service airplanes. A listing of the aircraft surveyed as of this writing is shown in Table III. Airplane Model B727-Cargo B737 B757 MD-88 MD-10-Cargo Total

Number of In-Service Airplanes 4 8 4 1 2 19

Number of RAMs Surveyed 75 238 42 7 58 420

Table III. Summary of in-service airplane field surveys. The same survey process that was used for retired airplanes was also followed for in-service airplanes, except that the field surveys were performed while the airplanes were undergoing routine heavy maintenance. Thus, the survey teams typically had access to both sides of most RAMs and a greater likelihood of obtaining maintenance records to derive the age and engineering basis for the RAMs. This was especially true for airplanes that were being surveyed under the Repair Assessment Program during the heavy maintenance visit. Since both sides of most RAMs were accessible, more data were collected including the repair Category (A/B/C) as detailed in FAA Repair Assessment Guidelines [5 and 6] and a workmanship rating score. Examples of in-service repairs are shown in Figure 2. a.

b.

Figure 2. Examples of repairs on an in-service airplane: a. exterior surface and b. interior surface.

Survey of structural repairs and alterations in transport category airplanes

117

Again, an “Aircraft Information & Documentation Form” was used to ensure all necessary data was collected for each RAM. However, for in-service airplanes, an electronic form was used in conjunction with a tablet-type lap-top computer. This method ensured accurate, real-time collection of the physical features and digital photos for each RAM surveyed. The electronic form contained lists and drop down boxes to speed data input and reduce errors. The form interfaced with the documentation database so that records were automatically entered. Maintenance records were then matched to specific repairs so that data on age and engineering basis could be acquired. Although the initial sample plan was based on the 1996 AAWG survey [4], it was subsequently decided that a more representative survey of the U.S. registered fleet was needed. An airplane sample plan was developed to target the number and models of airplanes that would represent the in-service U.S. registered transport category fleet. A total of 148 airplanes will be surveyed to provide a representative sampling of the population of approximately 4600 airplanes in the U.S. fleet. The project to date has surveyed 35 airplanes. There remain approximately 113 airplanes to be surveyed before the end of the calendar year 2009. The numbers of each airplane type to be surveyed are shown in Table IV. Airplane Model

Number Completed

Number to be Surveyed

Total Airplanes

Passenger Airplanes B727 8 0 B737-NG 0 26 B737-Classic 12 9 B747 2 0 B757 4 11 B767 1 8 A300 0 3 A318/321 0 18 B717/MD-80/DC-9 2 26 Cargo and Supplemental Type Certificate Airplanes B727 4 0 B747 0 2 B757 0 3 A300 0 3 MD11/DC10 2 4 Total 35 113 Table IV. Sample plan for remaining in-service aircraft in RAMs field survey of operators.

8 26 21 2 15 9 3 18 26 4 2 3 2 4 148

M.D. Bode et al.

118

TEARDOWN INSPECTIONS OF RAMS Another objective of this project is to collect a variety of RAMs from the retired U.S. domestic fleet to perform teardown inspections and fully characterize their condition. Of particular interest are specimens with large repairs covering large areas, and specimens with multiple smaller repairs accumulated relatively closely over a large area. Specimens of interest were identified for collection and returned to the AANC for teardown inspection during all field survey trips. These specimens typically include surrounding structure. A protocol for disassembly and inspection was developed by DiMambro [7] and is being used for this study with the exception of performing certain emerging nondestructive inspections. Generally, the disassembly process starts with a cleaning of the specimen and a determination of the most appropriate method for disassembly. Detailed visual inspection or other nondestructive inspection(s) are performed before disassembly by removal of fasteners and doubler layers. Nondestructive inspection (NDI) methods can include, but are not limited to, liquid penetrant, eddy current, ultrasonic, and thermographic methods. The specific methods used are determined by the situation at the time the part is dissembled. If defects are identified, they undergo destructive fractographic analysis to determine details that might explain how the defect was formed and how long it had been propagating in the structure. Figure 3 provides a summary of the teardown inspection process.

Figure 3. Typical teardown and inspection process for RAMs: a. clean and catalog, b. NDI for damage and cracks, c. disassembly and cleaning, and d. measure crack size, shape and distribution.

Survey of structural repairs and alterations in transport category airplanes

119

Data collected from teardown specimens are added to the Field Survey and Teardown Database. These data should provide an in depth review of the occurrence of smaller fatigue cracking in RAMs. The number of specimens collected is driven by availability of candidate specimens observed during field survey trips. The goal is to obtain a representative sampling of large area RAMs or clusters of RAMs that have been removed during maintenance such that they can be inspected with appropriate NDI. Additionally, there is interest in obtaining one or more retired cargo door modifications, particularly from an airplane that underwent a passenger to cargo modification. Specimens are still being collected at this time for later assessment, but the main focus of the project is on the field survey of in-service airplanes. A set of specimens from a retired B737 has been acquired that included large lap joint modifications as well as several large repair patches in critical fuselage areas as shown in Figure 1. Disassembly and inspection of the first large window belt lap joint modification is nearing completion. Additional specimens have been identified during field survey work and will be collected in the near future. Of particular interest are certain cargo door modifications. Results of teardown inspections will yield a smaller quantity of data, but due to the more intensive inspection they receive, there will be a higher likelihood of detecting smaller fatigue cracks should they be present.

DATABASE OF FIELD SURVEY AND TEARDOWN RESULTS The objective of this final task is to document all procedures, approaches, and results from the work outlined above, and compile it in an electronic database for further analysis by the FAA. The idea was to develop a database that allows rapid retrieval and detailed searching for the purpose of determining if an appreciable risk of catastrophic failure exists in repaired, altered, or modified airframe structures. Microsoft® Excel® provides a common operating environment that is used in this project to store, sort, query, and retrieve all accumulated information. While most users think of MS Excel as an array of numerical data in rows and columns, relevant data that can be entered into the database through the use of electronic links includes scanned documents, digital photographs, and other forms of information converted to digital format. Links from specific RAM citations take the user to pictorial documentation of each repair, with annotations to highlight important information. Thus, the RAMs Field Survey Database consists of multiple data sets generated from aircraft part information, inspection results, aircraft history, and significant findings. Individual records are associated through primary keys (such as airplane model) and relational links (such as repair category) of all parameters allowing records with common data to be associated by any combination of search criteria. Queries by FAA end-users will be inclusive of all

M.D. Bode et al.

120

related data by the use of specified fields and will allow users to search individual records or to simultaneously access multiple records with similar characteristics across any of several data fields. Data input began at the initiation of the project, and will continue throughout the life of the work. The final database will be delivered to the FAA along with the Final Report in December 2009. The primary electronic database deliverable will provide 72 categories of information on RAMs that can be sorted based on any number or combinations of categories. Categories of interest might be airplane model, certain aircraft structural zones (e.g., zone 42 belly skins), operational type (e.g., passenger versus freight), or other factors. Measurement data is a vital part of the data that will be used to assess the risk of WFD occurring associated with RAMs. Continuous input by the potential end users of the data has allowed the final database product to include a wealth of factors related to airframe structures. While the overall basic information to be collected has been determined at the outset of this project, there is capacity to add more fields for types of information as the project progresses. In some cases, it may even be possible to retroactively include data from surveys performed months prior, depending on the type of information.

PROJECT SUMMARY This project is well on its way to providing a good source of data for understanding the risks that RAMs may pose for developing WFD. Field survey data include not only numerical measurements, but internal visual inspections, digital photos, and more extensive records reviews than those performed in the prior field survey. The field survey progress to date, in terms of the number of RAMs and airplanes, is detailed in Table V. Additional data regarding RAM quality and damage characterization is being generated through teardown inspections. Teardown specimens are still being collected at this time for later assessment. The field survey and teardown inspection database is being populated in real-time as data is collected. Additional and significant work remains to acquire maintenance records for specific RAMs. Once completely populated, the database will be examined for trends to quantify the risks that RAMS that may pose for developing WFD.

Survey of structural repairs and alterations in transport category airplanes

Airplane Model B727 B737 B747 B757 B767 DC-9 MD-88 MD-10 Total

Number of Airplanes 11 10 2 4 1 1 1 2 35

Number Retired 7 2 2 0 1 1 0 0 16

Number InService 4 8 0 4 0 0 1 2 19

Number of RAMs

Retired RAMS

120 293 88 42 5 44 7 58 741

45 55 88 0 5 44 0 0 321

121

InService RAMs 75 238 0 42 0 0 7 58 420

Table V. Numbers of airplanes and RAMs collected during field surveys as of March 2009.

ACKNOWLEDGEMENTS The authors wish to acknowledge Colin McConnell, Ciji Nelson, Justin Newcomer, Bryce Smith, and Todd Sokol of Sandia Nationals Laboratories for their diligent work performing field surveys, teardown inspections, and database construction during this project. In addition, thanks are due to William Scott, Robert Eastin, and Doug Ostgaard from the Federal Aviation Administration for their substantive technical guidance.

REFERENCES [1] 70 Federal Register (FR) 5518, February 2, 2005. [2] 72 FR 70486, December 12, 2007. [3] 73 FR 66205, November 7, 2008. [4] Airworthiness Assurance Working Group Final Report dated 12/12/96 Continued Airworthiness of Structural Repairs. [5] Title 14, Code of Federal Regulations (CFR) 121.1107, Repair Assessment for Pressurized Fuselages [6] Federal Aviation Administration (FAA) Advisory Circular (AC) 120-73, Damage Tolerance Assessment if Repairs to Pressurized Fuselages. [7] DiMambro, J. and Nelson, C. (2007), In: Survey of Retired Boeing 727 Metallic Structural Repairs, Proceedings of the 50th Annual Air Transport Association (ATA) NDT Forum, Orlando, Florida.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

ENHANCED TEARDOWN OF EX-SERVICE F/A18A/B/C/D CENTRE FUSELAGES ∗ L. Molent1, B. Dixon1, S. Barter1, P. White1, T. Mills1, K. Maxfield1, G. Swanton1 and SQNLDR B. Main2 1

2

Defence Science and Technology Organisation, Melbourne Australia Directorate General Technical Airworthiness, RAAF, Melbourne, Australia

Abstract: The Flaw IdeNtification through the Application of Loads (FINAL) is a fatigue testing and destructive inspection program for ex-service F/A-18A-D wing attachment centre fuselage structure, known as the centre barrel (CB). It is a joint program between the Defence Materiel Organisation (DMO), Directorate General Technical Airworthiness (DGTA) and the Defence Science and Technology Organisation (DSTO). High life ex-service F/A-18 A-D CBs are cycled under a wing root bending moment spectrum until cracking from pre-existing defects can be readily found. The FINAL CB test program was originally conceived to satisfy engineering caveats associated with development of the Hornet Structural Refurbishment Program 1 (SRP1) package. SRP1 extends the fatigue life of the CB via a number of inspections and modifications to discrete CB locations. At the new CB fatigue life limit, the CB was to be replaced or the aircraft retired. At the time of SRP1 development, concerns were held that any potential combination of in-service mechanical damage, environmental degradation and widespread fatigue damage may jeopardise the useful life of SRP1 CBs. FINAL was successful in mitigating against these threats. Most recently the results from FINAL, in part, have extended the induction times for CB replacement, leading to increased aircraft availability and cost savings to the RAAF. ∗

This paper has been published previously under the title of “The F/A-18 FINAL Fatigue Testing Program” in the proceedings of the Australian International Aerospace Congress, Melbourne 2009.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 123–142. © Commonwealth of Australia 2009. Published by Springer Science+Business Media B.V. Dordrecht.

L. Molent et al.

124

This paper describes the FINAL program and its outcomes as well as test result interpretation.

INTRODUCTION The Boeing F/A-18 Hornet is employed by the Royal Australian Air Force (RAAF) as a land-based front-line fighter aircraft. It was originally designed, built and tested for the requirements of the United States Navy (USN) who operate it mainly as a carrier based fighter/attack aircraft. Because of the operational differences between RAAF and USN, the RAAF in collaboration with the Canadian Forces (CF), who had similar needs, developed a full-scale fatigue test program to determine the safe-life of the F/A-18 in their service. This program, known as the F/A-18 International Follow On Structural Test Project (IFOSTP) [1], included a fatigue test of the centre fuselage called FT55. This test showed that the safe–life of the centre barrel (CB) is insufficient to meet the RAAF planned withdrawal date for some aircraft in the fleet. The F/A-18 CB, shown in Figure 1, carries wing loads into the fuselage through its three main structural elements, the Y453, Y470.5 and Y488 bulkheads. These bulkheads are fracture critical and loss of structural integrity in any of them is considered a safety-of-flight issue. For this reason, they are the main focus of the Flaw IdeNtification through the Application of Loads (FINAL) program described herein. The three bulkheads are made of 7050-T7451 aluminium alloy, which has been coated for corrosion protection with a thin layer of almost pure aluminium by the Ion Vapour Deposition (IVD) process. As a precursor to this coating, the items to be coated are acid etched to improve coating adhesion. This etching leaves numerous tiny pits in the surface of the parts over which the coating is applied. The preferred logistics solution to address deficiencies highlighted by FT55 is a centre barrel replacement (CBR) program. This strategy has already been commenced by the USN and CF. In 2002, two main problems with the CBR program for the RAAF were highlighted; the program may be difficult to run incountry because of logistical concerns, and the availability of aircraft during the program may be insufficient to meet the operational needs of the RAAF. For these reasons, combined with the expected expense of such a program, the RAAF considered alternative strategies to delay the induction of and minimise the number of aircraft to undergo the CBR program. This alternative solution, referred to as Structural Refurbishment Program - Plus (SRP1+) [2] (part of Hornet Up Grade (HUG) phase 3), allows the CBR

Enhanced teardown of ex-service F/A-18A/B/C/D centre fuselage

125

incorporation time to be delayed from a fatigue life expended index (FLEI) 1 of 0.72 to 0.78. The strategy comprises a series of modifications made at critical locations mid-way through the life of the F/A-18, and will allow some aircraft in the fleet to reach planned withdrawal date without CBR. The actions that will be taken in SRP1+ are based mainly on service experience from the early life of the fleet and the results of the F/A-18 fatigue tests including IFOSTP. Without data from high life fleet aircraft, a number of potential risks were envisaged in implementing a SRP1+ program, including: 1. 2. 3. 4.

Potential for cracking from in-service defects including mechanical damage and corrosion. Influence of widespread fatigue damage (and thus potential new fatigue critical locations not previously seen in F/A-18 fatigue tests). Ineffective repairs. Inferior quality material used in early build CBs.

Figure 1 - An ex-CF CB used in the FINAL program. The teardown and inspection of ex-service CBs was agreed to be a method of reducing the risks involved in the SRP1+ program. A number of ex-service CBs from the USN and CF CBR programs where made available for teardown. Since the sizes of the average largest cracks present on ex-service CBs were expected to 1

Wing root strain peak and valleys are monitored on each RAAF F/A-18 and the individual FLEI is the calculated damage in comparison to a baseline value.

L. Molent et al.

126

be less than 1 mm ([4],[5],[6]), which is below the threshold of current practical Non-Destructive Inspection (NDI) methods, it would be difficult to gain service data (size and distribution of cracking) from ex-service CBs in their as-received condition since detection of cracking by NDI alone would be difficult. To overcome this obstacle, the FINAL program was implemented. This program involves the application of representative Wing Root Bending Moment (WRBM) fatigue cyclic loads to ex-service CBs in a test rig. This process grows existing cracks or cracks originating from other existing damage/discontinuities (e.g. corrosion, mechanical damage and inherent discontinuities in the material etc) to a size where they may be readily detected under laboratory conditions. After fatigue cycling of each CB has been completed, a teardown including thorough inspection and Quantitative Fractography (QF) is performed. The data from this was used to address the following aims of FINAL: 1.

2. 3.

4.

To determine if in-service aircraft contain CB damage not accounted for in the fatigue test program, including mechanical damage and corrosion that are the result of the service environment. To gain a more complete picture of the types of discontinuities, defects or degradation that lead to cracking in the fleet [3]. To ensure that future decisions on the CBR program are based on as much relevant information about the structural integrity of the in-service CBs as possible; and To provide data that will enhance current risk and reliability method deliberations with regard to the F/A-18 aircraft (see White et al. [4],[5] for example).

More recently, the FINAL results have been used in part to further delay the induction time for CBR by re-assessing the necessity for some refurbishment modifications and inspections. The results of this effort have reduced the risk of a capability shortfall during a CBR program and resulted in considerable cost savings. This paper details the cycling and teardown process of the FINAL program as well as some of the results 2 .

RAAF F/A-18 LIFING METHODOLOGY The F/A-18 A/B entered into RAAF service as a Safe Life managed aircraft having been certified by the USN to the crack initiation requirements of MIL-8866 [8]. It soon became evident however on the basis of Canadian and RAAF usage, that the Hornet was being flown differently to the USN and outside the scope of the assumptions to which it was originally certified. The RAAF and Canada therefore 2

Adapted from [7].

Enhanced teardown of ex-service F/A-18A/B/C/D centre fuselage

127

undertook the International Follow On Structural Test Project (IFOSTP) to recertify the Hornet and in doing so adopted the fatigue management requirements of DEFSTAN 00-970 [9], see [10]. The core fatigue requirements of DEFSTAN 00-970 are: • •

•

•

Full-scale fatigue testing of a representative structure under representative flight-by-flight sequence of service loading; Ideally fatigue testing should be conducted to the equivalent of 5 service lifetimes (3.3 to account for potential material fatigue and build quality variability or scatter 3 and 1.5 for those components whose loads are not monitored in service [9]). A Residual Strength (RS) criterion where failure (total life) is defined as the test time at which the structure is no longer capable of withstanding 1.2 times Design Limit Load (DLL) for all major loading actions and all primary structure comprising those load paths. In many instances the structure does not fail catastrophically, and artificial damage may be progressively introduced in order to evaluate the fail-safety of the structure (see for example [11]). A continuous program of structural condition and usage assessments to ensure the validity of the basis of certification.

Specifically for the FINAL program, as the CBs are not complete built up structures, much care was taken to ensure that strain distributions adjacent to critical locations during RS tests matched those of the IFOSTP CB certification test, FT55. Crack locations and morphology were also matched in detail to the performance of the full scale test and, as an additional conservatism; only critical crack sizes under RS loads were used from the FINAL program in the safe-life extension exercise. All crack growth rates utilised in fatigue certification were those observed on the full scale test article. These measures necessitated a wide range of additional verification techniques including detailed strain surveys, thermographic determination of stress distributions, finite element analysis and individual component tests of cracked structure excised from wing carry though bulkheads. Throughout, a heavy emphasis was placed on DSTO’s state of the art QF capabilities.

3

3.33 for 2 items tested, 2 items installed. 4.0 for 1 item tested, two items installed. A scatter factor of 3.33 is insufficient to demonstrate the safe life of high variability features i.e. high frequency low stress spectrum, in which case the Safe SN process [9] is to be used and appropriate allowances made during substantiation.

L. Molent et al.

128

Test Interpretation In essence, the RAAF methodology requires the establishment of the test time to the 1.2DLL failure criterion. Whether the structure fails below 1.2DLL or survives 4 , a method is required to determine the full fatigue life as defined by a location’s ability to survive 1.2DLL. In effect, the test time (equivalent RS life) to the critical crack length/depth (aRST) at the 1.2DLL point is required. To determine this DSTO, has developed a suite of crack growth tools [12]. These are based on the following observations and assumptions: 1.

2.

3. 4.

Critical cracks commence growing shortly after the aircraft (particularly for highly structurally optimised aircraft like the F/A-18) is introduced into service [6],[14]. Any period of crack nucleation can be ignored; Critical surface initiating cracks in an ambient environment grow approximately in a linear log crack depth versus time fashion (see for example, [3],[5],[6],[14]) in the absence of residual stresses and/or load shedding. The small fraction of life involved in fast fracture or tearing can be ignored; Typical initial flaws for fighter aircraft material are approximately equivalent to a 0.01 mm deep crack [3],[14]; and The critical crack depth is small for the highly loaded structure of the F/A-18, being typically about 10 mm [3],[14]. (The actual critical crack size for each location examined is determined on a case-by-case basis).

This behaviour is summarised graphically at Figure 2. As can been seen it is a relatively simple exercise to extrapolated or interpolate as necessary to determine the equivalent time to the estimated, calculated or test determined acritical. The DSTO methods are heavily dependant on QF. By using QF, crack growth data can be obtained for the period close to crack nucleation through to failure. For highly optimised structure the period up to detectable using conventional NDI can represent up to two-thirds of the total life [3],[14].

FATIGUE CYCLING OF CENTRE BARRELS The ex-service CBs are cycled in the test rigs shown in Figure 3 prior to teardown. The rigs were designed to simulate wing loads at the wing attachment lugs. Pairs of actuators apply equal and opposite loads to the ends of beams that are attached to the sides of each bulkhead. The WRBM produced by the actuators is transferred as a couple at the wing attachment lugs. The CBs are rotated by 90 degrees to allow them to sit on one set of beams. The rigs are self-reacting so that the top and

4

This point could be conservatively used without further analyses.

Enhanced teardown of ex-service F/A-18A/B/C/D centre fuselage

129

bottom beams apply equal and opposite bending moments to opposite sides of each CB. Each actuator is controlled separately, making it possible to proportion the bending moment applied to each bulkhead. This allows the loading of the CB to be tailored to match in-flight load distributions. It is also possible to continue cycling after bulkheads have failed by switching off the actuators attached to the failed bulkheads. As a result, each bulkhead is cycled until failure, maximising the amount of growth to existing flaws and thus increasing the chances of finding the maximum number of locations prone to cracking.

Figure 2: Schematic of the growth of a typical crack (see [6], [14]). Example shows the depth versus time history for blocks of loading as determined through QF and then extrapolated to the estimated critical crack depth for the residual strength load. Possible continued crack growth under quasi-static failure modes near the end of life are conservatively ignored.

A modified version of the mini-FALSTAFF (Fighter Aircraft Loading STAndard For Fatigue evaluation) sequence is generally applied to the test articles. The miniFALSTAFF sequence is a truncated version of FALSTAFF, which was developed to represent the standard load history at the wing root of a fighter aircraft [15]. The sequence is equivalent to 200 flights of a standard fighter of the time of its development. A normalised mini-FALSTAFF sequence was generated using the NLR developed software Genesis 4 Fatigue [16]. The sequence was normalised by dividing each load by the maximum load in the sequence. The FINAL fatigue sequence also had some of the mini-FALSTAFF flights rearranged to make it easier to interpret during QF without significantly affecting its severity. This was confirmed by a small coupon test program.

130

L. Molent et al.

The WRBM sequence applied to the whole CB was derived by multiplying the normalised sequence by the peak IFOSTP centre-fuselage (FT55) WRBM of 6462 in-kip. The WRBM applied to each bulkhead was then determined from the horizontal lug loads at the design load cases [17]. The proportion of the total WRBM applied to each bulkhead, as inferred by the horizontal lug loads, was used to scale the loads applied to each bulkhead. The relatively small size of the mini-FALSTAFF blocks allowed sufficient points (repeats) for fast growing cracks to make the QF useful and to enhance crack surface “readability”. Additionally, the repeating nature of the surface markings produced by this loading aided the detection of the demarcation between the service loading and the fatigue enhancement on those crack surfaces that were found to have service cracking. This allowed the size of the cracks present at the end of service life to be determined. Instrumentation in the form of strain gauges is applied to each test article to assess the distribution of load between the three bulkheads and to compare the applied loads to previous F/A-18 certification fatigue tests.

Figure 3 – Two F/A-18 CBs mounted in the FINAL rigs. Inspections Two inspections are performed on each of the teardown articles. An inspection is performed before fatigue cycling to allow prompt reporting of information regarding significant existing flaws. Another full inspection is performed when the CB has been dismantled at the completion of fatigue cycling. Parts exhibiting cracking may be subjected to QF where they are broken into fragments and the crack surfaces examined and measured under a microscope. The QF examination gives information relating to the type of defect causing cracking and the rate of crack growth from that defect. The procedures for the pre- and post-test inspections are basically the same. The first stage of inspection is a close visual inspection of the entire structure, followed

Enhanced teardown of ex-service F/A-18A/B/C/D centre fuselage

131

by eddy current inspection of potentially cracked locations in the main bulkheads, including those areas where cracks were detected in previous fatigue tests. The post test inspection is usually more comprehensive, since eddy current inspection can include all bulkhead holes and fillet radii that were not exposed in the built-up structure. This level of inspection is required to find cracks in areas where no cracking has been found in previous fatigue tests. After the final eddy current inspection, the three main bulkheads may be inspected using fluorescent dye penetrant as a final check for new cracking sites. Some other areas that are known to have cracked during previous fatigue tests, but where no cracking has been found during NDI of FINAL, are broken open to assess whether there is cracking below the threshold of NDI. Careful destructive inspections of this type can expose fatigue cracks that are only tens of microns deep.

PRELIMINARY RESULTS Thus far fatigue cycling and teardown of 10 CBs has been completed (see Table i) and the teardown down inspection results for four of the CB’s are available [18], [19], [20] & [21]. Many dents and scratches were found during the initial inspection of all aircraft CBs (with the exception of CB4, which was never in service). These were mostly consistent with CB removal, although some damage particularly around the wing attach holes appeared to be service related. Failure locations were found to be quite consistent between the CBs tested, albeit at different failure times due to variations in service-times, initiating discontinuity size and spectra used. A summary of common locations is given in Table ii and depicted in Figure 4. Cracking found on FINAL has also been found on previous full-airframe fatigue test articles of the F/A-18 at these locations. These results provided sufficient confidence to commence the SRP1+. Whilst a detailed description of each failure is beyond the scope of this paper, the following examples from the first two CBs tested illustrate some pertinent features. Each CB generally failed in up to five locations during fatigue cycling, rather than the three that may have been expected – one for each bulkhead. More than one failure was induced in some of the Y488 bulkheads because they were repaired to allow further cycling. This was carried out to reveal more flaws since this bulkhead usually failed first, early in the FINAL loading life. Generally, notwithstanding this, cycling of a bulkhead ceased after it failed. An additional failure location occurred occasionally on the Y453 bulkhead, where it failed at two locations simultaneously. The failure locations from CB to CB have also been quite consistent. The Y488 bulkhead’s lower flange below the air inlet ducts has failed three times, twice on one CB and once on another. Several CBs have also failed in

L. Molent et al.

132

the vicinity of the Y488 bulkhead upper duct flange and Y453 bulkhead upper duct flange. Furthermore, the completed teardown inspection results for the first few CB’s showed that crack indications were found at the locations mirroring the failure locations (i.e. other side of the bulkheads) in most cases. Table i: Summary of CBs tested to-date Origin Service Hours Spectrum CF 3536 Mini-FALSTAFF USN 5198 Mini-FALSTAFF USN 6169 Mini-FALSTAFF Ex-FT46 0 Mini-FALSTAFF (IFOSTP) USN 2799 Mini-FALSTAFF USN 4704 Mini-FALSTAFF USN 4724 Static (no teardown completed) USN 3677 Mini-FALSTAFF USN 6124 Mini-FALSTAFF RAAF 3620 FT55 RAAF 3297 FT55

Test Article CB1 CB2 CB3 CB4 CB5 CBH1 CBH2 CB6 CB7 CB8 CB9

Table ii: Common failure locations on each bulkhead. Bulkhead

Descriptor (See Figure 4)

Y453

UDF SLA

Y453 SDLA Y470.5

Failure location

Upper duct flange. Outboard flange at drag longeron attachment location. Outboard flange at side longeron attachment location.

Y470.5

X19

Lower flange at X19.

Y470.5

FC

Upper beam above fuel cavity

Y488

ILF

Inboard lower flange and kick point

Y488

CB012

Web above upper duct flange.

Y488

CB007

Main landing gear uplock pocket

Y488

CB041

Upper mould line flange.

QF has been completed on the fracture surfaces of each of the failure sites. The first failure to occur on the CF CB1 was the result of fatigue cracking in the lower

Enhanced teardown of ex-service F/A-18A/B/C/D centre fuselage

133

surface of Y488 bulkhead’s lower flange. Figure 4 indicates the location of this fracture (ILF) and Figure 5 shows the fracture. This fracture (Figure 6) was examined after it was excised.

Figure 4 - Some cracking locations on the bulkheads (acronyms defined in Table ii and Table iii)

An examination of the cracks that are shown in Figure 6, including the measurement of the crack growth by QF show (Figure 7) that the cracking was well represented by a log-linear model. Therefore, these types of cracks in this type of structure under these loading conditions (in the absence of significantly fluctuating residual stresses and/or load shedding) appear to be well modelled by an equation of the following forms [14]:

Ln(a ) = ϕN + Ln(a 0 ) or a = a 0 e ϕN

(1)

where N is the “fatigue life”, ϕ is a parameter that is geometry, material and load dependent, a is the crack depth at time N and a0 is the initial crack-like flaw size (depth of the crack at the start of loading).

134

L. Molent et al.

Figure 5 - Location of the first failure of the CF CB1 at the Y488 bulkhead’s lower flange.

Figure 6 - Fracture surface at the failure of the CB1’s Y488 bulkhead lower bulkhead flange. The four cracks upon which QF was carried out (see Figure 7) are labelled. All of the cracks thus far examined (where significant residual stress fluctuations, load shedding or load variation with crack depth were not an issue) have exhibited crack growth that approximately follows the relationship given in Equation 1. It has been possible in most cases to identify the start of FINAL block loading on the fracture surface and thus estimate the extent of service crack growth. It was found that the crack depths at the end of service were generally less than 0.1 mm. There were, however, some notable exceptions to this. The crack that caused

Enhanced teardown of ex-service F/A-18A/B/C/D centre fuselage

135

failure of the Y470.5 bulkhead of CB2, shown in Figure 8, was almost 1 mm deep at the end of service. Three of the sub-cracks that joined to form this crack were examined with QF. Figure 9 shows a closer view of one of these sub-cracks. The region of growth caused by service loading is darker in colour and clearly visible. The crack growth curves measured for the three sub-cracks, shown in Figure 10, indicate that the sizes of the cracks at the start of block loading were between 0.4 mm and 0.9 mm. This cracking occurred in a region where prescribed in-service shot peening had not been performed on this CB. This resulted in significant service crack growth and consequently a failure early in FINAL fatigue cycling of this bulkhead (after only twenty blocks of the mini-FALSTAFF spectrum, as shown in Figure 10). This cracking was not representative of the expected condition of high time teardown aircraft where this area has been shot peened, since the peening, when performed correctly and in a timely manner has been shown to greatly retard early crack growth [22]. Large service cracks were also found in the main landing gear uplock holes in the Y488 bulkhead. These were generally at least 0.5mm deep. This area is amongst the most highly stressed in the F/A-18 CB. The consistent discovery of relatively large service cracks is evidence that locations that have had their fatigue resistance improved by cold working rely heavily on the effectiveness of these production/repair processes to retard crack growth. In the case of the uplock holes it was found that the retardation begins at deeper crack depths than with shot peening. Cracks can reach depths greater than 1 mm before crack growth retardation becomes significant, so in these cases cracking below this level is not an indicator that safety has been compromised [23]. Notwithstanding the large crack sizes found in the two locations discussed above, the QF results confirm that it would be extremely difficult to find cracks at the end of service without a comprehensive and very difficult destructive examination. The fatigue cracks shown in Figure 6 on the fracture surface show the extent of multiple cracking typically found in failures of the fatigue tested CB bulkheads. The largest crack, on the right hand side in the Figure, caused the final fracture to occur in this case. This multiple crack failure characteristic of the bulkhead material is a product of the failures usually occurring at areas of low stress gradient and as a result of the numerous small surface pits that exist on the bulkheads from the pre-IVD etching. Over four hundred cracks from high-risk locations in the CB have been examined with QF and it has been found that the majority of them have initiated from etch pits formed during the IVD coating process. Often multiple microscopic cracks had initiated from different etch pits, and these cracks have later joined to form a single crack. Other causes of crack initiation that have been found include corrosion pits, surface breaking porosity, manufacturing defects in holes such as scratches and laps, and peening induced defects. The largest of these discontinuities were generally found to be much less effective at growing a fatigue crack in comparison to cracks that grew from the smaller etch pits. This is possibly

L. Molent et al.

136

related to flaw aspect ratio issues or their far less frequent occurrence. It is notable that cracking from corrosion was only found in the absence of the protective IVD coating, indicating the importance of the coating in maintaining the corrosion protection of critical areas. 10

1

0.1

CB1 Y488 CB1 Y488 CB1 Y488 CB1 Y488

0.01

C1 C2 C3 C4

0.001 0

10

20

30

40

50

60

70

No. of Blocks

Figure 7 - Crack growth curves for the four largest fatigue cracks at the first Y488 bulkhead failure of CB1. The curves were derived from QF measurement of the crack depth after the application of each block of loading and include an estimate of the physical initial discontinuity size. The curves are labelled according to the names given to the cracks in Figure 6.

Figure 8 - The crack that caused failure of the USN CB2's Y470.5 bulkhead. The origins of the sub-cracks that were investigated with QF have been labelled. The arrows above these origins indicate the approximate directions in which crack growth measurements were made.

Enhanced teardown of ex-service F/A-18A/B/C/D centre fuselage

137

Figure 9 - A view of the origin of C1 (see Figure 8) from the crack that failed the USN CB2’s Y470.5 bulkhead. The brown discoloured region was thought to be service related crack growth. The end of service and the start of FINAL block loading has been highlighted. Some progression lines, representitive of FINAL block loading have been highlighted with arrows. The teardown inspection of each CB generally reveals approximately eighty sites where fatigue cracking is detected. While most of these were at sites similar to the cracking found during the IFOSTP fatigue test of FT55, there were a number of new locations where no cracking had been found on FT55. Assessment of these locations has indicated that they should not pose a threat through to the planned withdrawal date of the aircraft in Australian service. However, as mentioned above, the most significant cracks that resulted in failures have all been identified on previous fatigue tests. It may be hypothesised that since the etch pits cover all of the structure and cracks initiate very early in the structure’s service life, then these pits will result in a population of growing cracks at all areas of high stress. These cracks will normally dominate test article and specimen failures in a test environment. Assuming that the distribution of etch pit sizes is equal in all areas, it follows that the most highly stressed areas are likely to grow cracks from these etch pits most quickly, which implies they should be detected or cause failure prior to any service induced damage since they had a head start. The exception may occur when service induced damage either occurs very early in the service life or is very large – neither of which have been shown to have occurred in the ten CB’s tested. This is supported by the observation that most of the significant cracking has occurred at identical locations on both sides of each of the bulkheads. It also supports the consistency in cracking sites between fatigue test articles. It appears that while there is widespread cracking in the CBs, the locations where significant cracking has led to failure have been in discrete locations of high stress.

L. Molent et al.

138

So far no cracking from mechanical damage, except for the manufacturing defects in fastener holes cited above has been detected. The mechanical damage on most CBs appeared severe, but as yet no evidence has been found to support any connection between significant cracking and this mechanical damage. The results from the ten CB teardowns have allowed the identification of some rare and potentially very damaging initial defects such as corrosion, although this corrosion had most likely not occurred during service, but between the time the CB’s were removed and its arrival at DSTO . Information pertaining to the causes and likely locations of these defects has and will be considered as part of the SRP.

Figure 10 - Crack growth curves derived with QF for the sub-cracks C1-C3 at the crack that failed the USN CB2’s Y470.5 bulkhead.

LIFE EXTENSION ACTIVITIES As previously stated, the results of the IFOSTP FT55 full-scale fatigue test was used to life the RAAF CBs, resulting in the SRP and CBR requirements. The FT55 test was managed conservatively. When there was an indication of damage, or when very small damage was detected, remedial action would be undertaken with the aim of maximising the over-all life of the test article and the resulting test clearance. To expedite testing, the criticality of the damage was not assessed prior to modification. Thus, there were potentially a large number of FT55 cracking locations that could have been cycled further without the need for modification. Furthermore, in many cases suspected damage was blended out, which precluded the possible use of QF measurements in life extension activities.

Enhanced teardown of ex-service F/A-18A/B/C/D centre fuselage

139

In fact, there were many cases where the only details of damage available were the time it was found, its location and the NDI estimated surface length of cracking. Furthermore, in many instances the NDI threshold value (approximately 1mm) was all that was provided. Table iii: Locations where Residual Strength was Investigated Damage Item Location (DIL) see Figure 4

CB003 CB028 CB072 CB007 CB008 CB010 CB011 CB012 CB014 CB041 CB102

Description

Data Source for Life Extension

Y453 Crease Longeron RST CB6 Y453 Holes in Upper Outb’d Longeron RST CB6 Y470 Crease Longeron RST CB6 Y488 Main Landing Gear Uplock Holes CB3 & CB8 Y488 Main Landing Gear Uplock Pocket CB8 Y488 Control Cable Holes RST CB6 Y488 Crease Longeron RST CB7 Y488 Upper Duct CB1 Y488 Lower Hydraulic Hole FEM (thermography CB8) Stand alone test (& crack Y488 Drag Longeron data from several CBs) Y488 Lower Kick Point CB2, CB7 & CB9 (strain Flange survey)

FINAL allowed the possibility to demonstrate the largest crack size that could have safely sustained the RS test (RST) requirement, (aRST). With this demonstrated RS crack size, the FT55 crack size at the time of modification could then be grown forward using the log-linear crack growth model to give an equivalent FT55 life had the modification during FT55 not been carried out 5 . This model was only used where log-linear growth was shown in FINAL or FT55 to be valid. Fatigue cracks were allowed to grow naturally or were grown from introduced defects at critical locations under FINAL loading and then the FT55 RS loading was applied to the FINAL rig to demonstrate fracture resistance with the larger crack size. This new crack size effective demonstrated that a new test life could be used for fleet lifing purposes.

5

As the crack size at the time of modification was extrapolated to a test determined aRST no additional factor on the standard scatter factor was applied to determine the safe-life, as would have been required had the process relied solely on analytical modelling, see [9].

L. Molent et al.

140

To make this process successful, it was required to be demonstrated that local FT55 RST conditions could be replicated in the FINAL rig. Identical strain gauge locations in the vicinity of critical locations were used as a cross-reference between FT55 and FINAL loading conditions. In some cases where the FINAL rig was incapable of reproducing FT55 conditions, stand alone component tests (excised from the FINAL CBs) were fabricated for isolated areas containing high-risk locations. Special jigs were manufactured and strain gauges were used to confirm that FT55 conditions were reproduced in the region local to the critical cracking. Cracks were then grown in these component tests under RST loads until failure occurred thus determining the aRST. Finally, in a few cases where no test based clearance was possible, detailed finite element models (FEM) of the critical areas were developed and the maximum allowable crack depth under FT55 RST conditions was found by analysis. These FEM were validated by comparison to strain gauge responses under FINAL and FT55 conditions, which were reproduced in the FEM. In some cases, the FEM was also validated though comparison with stresses obtained through thermography (see Figure 11). For the locations (DILs) investigated (see Table iii) it was found that FT55 could have cycled significantly longer without modification. This information was used to effectively increase the fatigue life of the CB by 10%.

Figure 11: Stress Distribution from Thermography (left) compared to Finite Element Analyses for the Y488 lower hydraulic hole.

CONCLUSIONS DSTO is currently undertaking teardown inspections of ex-service F/A-18 Hornet centre barrels (CB) to support RAAF deliberations concerning a number of

Enhanced teardown of ex-service F/A-18A/B/C/D centre fuselage

141

planned modifications to critical structure, including the complete replacement of the fracture critical CB. These CBs are fatigue cycled prior to teardown to increase the amount of service induced damage to the point that it can be readily found. The pre-teardown fatigue cycling activities have also provided valuable opportunities for life extension activities. Simple rigs have been built to apply wing root bending moment loads to the CBs, so that they may be fatigue cycled generally with a modified mini-FALSTAFF spectrum. Thus far ten CBs have been fatigue cycled and torn down. The locations of failures and other significant cracking appear to be consistent across the CBs and with the previous F/A-18 IFOSTP FT55 fatigue test. The wide sample provided by the teardown and inspection of ten CBs has provided valuable information about rare causes of cracking in the structure such as corrosion and surface breaking porosity. FINAL was initial instigated to ensure that no previously unidentified critical locations would exist after the implementing a series of discrete modifications, designed to delay the replacement of the CB. FINAL has also been used to revaluate the need for the implementations of modifications to the certification test article at specific test lives. In doing so it was determined that these modifications could have been delayed. The determination of a effectively new certification test end points, has lead to a significant life increase of numerous locations on the CB thus delaying the need for CB replacement. This has resulted in significant cost savings and increased aircraft availability. Thus the revised implementation time of the discrete modification package will allow the RAAF to operate the F/A-18 Hornet to its planned withdrawal date, with the minimum number of CB replacements. FINAL testing is expected to continue to support fatigue studies, verify and validate RAAF usage monitoring algorithms, determine defects types leading to fatigue in the material of interest, and to allow the further study of other specific locations where concerns may exist.

ACKNOWLEDGEMENTS The authors gratefully acknowledge the RAAF, USN and Canadian Forces for supplying the centre barrels being tested in this program, and also to the RAAF Tactical Fighter System Program Office for financial support and encouragement.

REFERENCES [1]

Simpson, D.L., Landry, N., Roussel, J., Molent, L., Graham, A.D. and Schmidt, N. (2002). In: Proceedings of ICAS 2002 Congress, Grant, I (Editor), Toronto, Canada.

142

L. Molent et al.

[2]

RAAF (2002). SRP1+ Project – Interim Technical Scope of SRP1+, ASI/A2102020 Pt 1 (21), Melbourne. Molent, L., Sun Q. and Green A. (2006). Fat. Fract. Eng. Mat. Struct., Vol 29; p 916-937. White, P, Barter, S and Molent, L. (2002). In: Proceedings of 6th Joint FAA/DoD/NASA Aging Aircraft Conference – San Diego, Sep. 16-19. White, P., Molent, L. and Barter, S. (2005), Fatigue, Vol 27 n 7, p 752-767. Molent L. and Barter S.A. (2007). Fatigue; Vol 29, p 1090-1099. Molent L., Dixon B., Barter S., White P., Mills T., Maxfield K., Swanton G. and SQNLDR B. Main. In: Proceedings AIAC13, Melbourne, 10-11 Mar 2009. Anon. (1975) Detail Specifications for Model F/A-18 Aircraft Weapon System, SD-565-1-5, Naval Air Systems Command, USA. Anon. (1999). Defence Standard 00-970, Design and Airworthiness Requirements for Service Aircraft, Issue 2, UK. Anon. (2007) F/A-18 Hornet Aircraft Structural Integrity Management Plan Vol 1; RAAF, Directorate General Technical Airworthiness, Australia. Barter, S., Molent, L., Landry, N., Klose, P. and White, P. (2003). In: Proceedings of Australian International Aerospace Congress, Brisbane Aust. Molent, L., McDonald, M., Barter, S. and Jones, R. (2007). Fatigue, Vol 30. n 1, p 119-137. Anon. (1998). JSSG-2006, Joint Service Specification Guide, Aircraft Structures, Department of Defense, USA. Barter, S., Molent, L., Goldsmith, N. and Jones, R. (2005). Eng. Fail. Anal., Vol 12, n 1, p 99-128. Van Dijk, G.M. and De Jonge, J.B. (1975). In: proceedings 8th ICAF Symposium, Lausanne, J. Branger and F. Berger (Ed). Anon. Genesis 4 Fatigue. (2001). CDROM, National Aerospace Laboratory NLR, Netherlands. Knight, J.W. (1986). F-18 Fuselage Stress Analysis – Volume 2 (Centre Fuselage), MDC A5163 Vol 2, Revision A, McDonnell Aircraft Company, USA. Dixon, B., Molent, L., Barter, S.A. and Mau V. (2004). DSTO-TR-1660, Melb., Aust. Dixon, B., Molent, L., Barter, S.A. and Mau, V. (2005). DSTO-TR-1740, Melb., Aust. Dixon, B., Molent, L., Barter, S.A. and Mau, V. (2007). DSTO-TR-1980, Melb., Aust. Dixon, B. and Molent, L. (2008). DSTO-TR-2199, Melb., Aust. Barter S.A. (2003). DSTO-TR-1477, Melb., Aust. Barter, S. and Dixon B. (2009). Eng. Fail. Anal., Vol 16, n 3, p 833-848.

[3] [4] [5] [6] [7]

[8] [9] [10] [11] [12] [13] [14] [15] [16] [17]

[18] [19] [20] [21] [22] [23]

NDI, inspections and maintenance

25th ICAF Symposium – Rotterdam, 27–29 May 2009

GLARE TEARDOWNS FROM THE MEGALINER BARREL (MLB) FATIGUE TEST R.J.H. Wanhill, D.J. Platenkamp, T. Hattenberg, A.F. Bosch and P.H. de Haan National Aerospace Laboratory NLR, Amsterdam, the Netherlands

Abstract: The MegaLiner Barrel (MLB) pressure cabin fatigue test was part of the Airbus A380 development programme. The NLR carried out teardowns of GLARE (GLAss REinforced aluminium laminate) structures from three key locations of the MLB: a window area, a beam above the passenger door, and some stringer couplings. The teardowns began with Non-Destructive Inspection (NDI), and were followed by fractographic investigation of the longest NDI-indicated cracks in the window and door beam locations. The main objectives were to verify the NDI techniques and capabilities, determine the fatigue initiation and crack growth behaviour, and provide data to check fatigue crack growth models for GLARE. The overall results demonstrated very good NDI teardown capabilities and high fatigue damage tolerance by the GLARE structures. The window area cracks grew under variable amplitude loading, while the door beam cracks grew under almost constant amplitude loading. The window area cracks were too small to check model predictions, but a significantly longer door beam crack had a constant growth rate, which agrees with model predictions.

INTRODUCTION The MegaLiner Barrel (MLB) full-scale pressure cabin test was begun in the mid1990s to study the fatigue behaviour of a double-deck transport aircraft M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 145–167. © Springer Science+Business Media B.V. 2009

Wanhill et al.

146

configuration. As part of the Airbus A380 development programme, the MLB was used to investigate several design solutions, structural materials and joining methods, including the use of GLARE (GLAss REinforced aluminium laminates). The applied fatigue loads were set high enough to obtain fatigue damage. The test was done by Airbus Deutschland in Hamburg, Germany, and was discontinued after 45,402 simulated flights. Stork/Fokker Aerospace in Papendrecht, the Netherlands, then specified a teardown programme for GLARE panels and components from several key areas of the MLB. The NLR carried out this programme under contract to the Netherlands Agency for Aerospace Programmes (NIVR). The present paper surveys the teardowns of a window area, a beam above the passenger door, and some stringer couplings. The teardowns began with NonDestructive Inspection (NDI), and were followed by fractographic investigation of NDI-indicated cracks in the window and door beam locations. The general objectives were: (1) Verification of teardown NDI techniques and capabilities. (2) Establish the patterns of cracking in the GLARE aluminium layers. (3) Determine the fatigue initiation locations and likely causes. (4) Estimation of fatigue “initiation” lives and crack growth behaviour in the GLARE aluminium layers. (5) Provision of data to check fatigue crack growth models for GLARE.

THE MLB Figure 1 shows the MLB in the test hall in Hamburg, and the general loading conditions applied in the test rig. These were pressurization cycles combined with longitudinal and transverse bending and ground loads. Figure 2 shows a schematic of the MLB with key-codes of its construction and the GLARE locations investigated by the NLR. F4 is the window area, F6 is the stringer coupling area, and F7 is the passenger door beam area. The colour-shading codes refer to the GLARE and aluminium alloy skin materials. The GLARE, 2024 and 2524 panels were assembled with mechanical fasteners. The 6013 and 6056 panels were welded.

MLB FATIGUE LOAD SPECTRUM/HISTORY The MLB fatigue load spectrum was defined by Wagner [1]. The spectrum was based on a 6.25 hour mission and included basic ground and flight loads, with incremental loads for taxiing, rotation, landing, vertical and lateral gusts, and coordinated turns. There were eight basic flight types, ranging from severe turbulence (A) to calm air (H). Flight type A was combined with one ground load condition, type B with two ground load conditions, and types C – H with three ground load conditions, resulting in a total of 21 flight types. These occurred with

MegaLiner Barrel GLARE teardowns

147

differing frequencies in a block of 2150 flights, which was repeated until the end of testing. Table I gives the positions of the severest flight types A – C in each flight block. Table I

Severest flights in the MLB fatigue load history

A

B

C

1379

58, 127, 196, 1094

139, 148, 366, 494, 671, 956, 1026, 1392, 1549, 1785, 1854, 1928

THE F4 WINDOW AREA Construction Figure 3 shows the F4 GLARE window area just after its removal from the MLB. There is a circumferential butt joint (C64, see figure 2) just to the right of the removed area. The basic structure of the window area was a GLARE 3-7/6-0.3/0.4 countersunk skin fastened to die forged 7175-T73 aluminium window frames by 4.76 mm press fit Hi-Loks. The GLARE code means seven 2024-T3 aluminium layers 0.3 mm or 0.4 mm thick and interleaved with six glass fibre layers 0.25 mm thick; the outer two aluminium layers were 0.4 mm thick. U

Teardown procedure Full details of the teardown and NDI are given in Refs. [2–4]. The teardown was done in several stages: (1) NDI Removal of fasteners around windows and eddy current rotor inspection of fastener hole bores. Removal of window frames and eddy current pencil probe inspection of fastener holes in the GLARE skin and aluminium window frames on the faying surface sides. Disassembly of window frames and eddy current pencil probe inspection of aluminium rebates. (2) Optical fractography NDI-indicated cracked fastener holes in the GLARE skin and window frames forcibly opened. Low-magnification fractography to verify and map fatigue cracks in the fastener holes. (3) Field Emission Gun Scanning Electron Microscope (FEG-SEM) detailed fractography Fractography of the largest fatigue cracks in either window frame and the GLARE skin.

148

Wanhill et al.

NDI results 216 fastener holes were inspected. 45 crack indications were obtained for the GLARE fastener hole bores in the partially disassembled condition, and 41 when completely disassembled. 8 crack indications were obtained for the aluminium window frames. The crack indications were almost equally divided between the windows. Figure 4 shows 21 crack indications for the GLARE skin and 3 crack indications for the aluminium window frame of the C65-C66 window. Both of the windows had most crack indications in the B and D quadrants. These were the quadrants largely subjected to tensile shear loads during the MLB test. NDI verification Opening up the GLARE fastener holes with crack indications revealed 4 false calls, which for 216 inspected holes is less than 2 %. Optical fractography measurements of the crack sizes were used in a Probability Of Detection (POD) analysis [5]. This showed that for the 90 % probability + 50 % confidence level the detectable crack length was only 0.25 mm [3–5], which is an excellent teardown NDI capability. Optical fractography Mapping the GLARE skin fatigue cracks showed that most were in the fastener hole bores, with only a few in the countersinks [3]. Figure 5 gives examples of the cracking patterns in the aluminium layers. The shapes of smaller cracks indicated that cracking generally began at aluminium layer corners, most probably because of the stress concentrations provided by the corners. However, this does not explain why most cracks were in the bores rather than the countersinks, where there were relatively severe “knife edge” stress concentrations. The explanation, confirmed by fractography of the window frame cracks, is that local secondary bending favoured cracking in the fastener hole bores. FEG-SEM detailed fractography This detailed fractography had several objectives: (1) Check the “readability” of the MLB fatigue load history. (2 Estimate and compare the fatigue “initiation” lives and crack growth behaviour of the largest “readable” fatigue cracks in either window frame and the GLARE skin. (3) Provide data to check fatigue crack growth models for GLARE. Load history “readability”: Figure 6 gives an overview and a low-magnification detail of the largest crack in the window frames. Its position is arrowed in figure 4b. Higher magnifications showed crack front markings due to severe simulated flights, and the “readability” was generally very good. Figure 7 is an example of identifying the severest flight types A, B and C. Fractographic analysis of the largest window frame crack: Figures 8a and 8b show the crack lengths, a and a*, perpendicular to the fastener hole bore, plotted against the number of simulated flights (N) and the crack growth rates, da/dN, where a* is

MegaLiner Barrel GLARE teardowns

149

the mean crack length for each growth interval used to calculate da/dN. Both figures show the effects of severe simulated flights, pointed out explicitly in figure 8b. Some effects appeared to be transient, but a significant period of crack growth retardation began at a = 0.6 mm. This could be due to termination of a “short crack effect”. Such effects are generally attributed to a lack of fatigue crack closure in cracks smaller than about 0.5 mm [6]. Thus it is likely that once the window frame crack grew beyond about 0.5 mm the peak loads in severe simulated flights caused closure-induced retardation. Back-extrapolation of the data in figure 8a suggests a fatigue “initiation” life of zero and an initial crack size of about 0.06 mm. In fact, the fatigue crack began at a fretting scar caused by fastener movement against the bore of the hole, see figure 6b. Since fretting rapidly induces fatigue-initiating damage [7] and promotes early crack growth [8], it explains the (effectively) zero “initiation” life and also the suggestion of an initial crack size. Fractographic analysis of the largest “readable” GLARE skin crack: The position of this crack is arrowed in figure 4a. The crack was 0.91 mm long. Preliminary examination showed that the initial 0.2 mm was obscured by debris and sealant. Figures 9a and 9b show the crack lengths, a and a*, perpendicular to the fastener hole bore, plotted against N and da/dN. The data are limited but sufficient to show a maximum crack growth rate about 50 % of that in the aluminium window frame at similar crack lengths. Back-extrapolation of the data in figure 9a is unfeasible, owing to the limited data. Hence an estimate of the fatigue “initiation” life was not possible. Also, this crack (and, of course, all the other GLARE skin cracks) was too small to provide a check on fatigue crack growth models for GLARE: see, for example, the models proposed by De Koning [9], Alderliesten and Woerden [10], Beumler [11], Alderliesten and Homan [12] and Alderliesten [13]. However, on the positive side this teardown result demonstrates the high fatigue damage tolerance capability of the GLARE skin.

THE F7 DOOR BEAM AREA Construction Figure 10 shows the F7 GLARE door beam area before removal of a rectangular sample to be pulled to failure. The basic structure of the door beam area was a GLARE 3-9/8-0.4 countersunk skin reinforced with seven GLARE doublers to make a total of 34 aluminium layers. The GLARE skin code means nine 2024-T3 aluminium layers 0.4 mm thick and interleaved with eight glass fibre layers 0.25 mm thick. The doublers also consisted of 2024-T3 aluminium layers 0.4 mm thick and interleaved with glass fibre layers 0.25 mm thick.

150

Wanhill et al.

Teardown procedure The teardown was done in several stages: (1) NDI: Removal of fasteners and eddy current rotor inspection of fastener hole bores by Airbus Deutschland. (2) Sample testing: Removal of the rectangular sample and pulling it to failure. (3) Optical examination Fractography for fastener hole 33. Fastener holes 31 – 35. (4) FEG-SEM examination Fractography of the largest fatigue crack in fastener hole 33. Fastener holes 31 − 35. NDI results The Airbus Deutschland results are in Ref. [14]. The largest crack indication was 7 mm in fastener hole 33. Rectangular sample testing The sample indicated in figure 10 was pulled to failure in a 900 kN machine. The arrows in figure 10 point to the failure position, and figure 11 shows this position after failure. The sample failed across fastener holes 12 and 33, with some fibre pullout and delamination. Note the cracks at the nearby fastener holes 11, 32 and 34. These cracks were made visible (opened up) by the testing. The sample breaking load exceeded the Limit Load (LL) requirement despite the prior presence of fastener hole cracks. Optical examination Fractography for fastener hole 33: Figure 12 shows the best visible fracture surfaces. The aluminium layer fatigue cracks generally appeared silvery-white, although there were blackish areas near the fastener hole bore and on much of the fatigue fracture surfaces of the bottom aluminium layer. The overall contours of the fatigue cracks indicate a strong influence of local bending. This bending was due to pressurization-induced bulging of the door cutout area during the MLB fatigue test [14]. The largest fatigue crack was in aluminium layer 31 (the first layer of the last doubler), see the right-hand image in figure 12. The crack length was 6.54 mm, which is close to the Airbus Deutschland NDI indication of 7 mm. Fastener holes 31 − 35: The left-hand image in figure 12 shows considerable black debris and/or damage in the bore of fastener hole 33. Consequently some nearby fastener holes were sectioned perpendicular to the visible or NDI-indicated cracks. All the holes had circumferential scoring (grooves) suggesting poor initial hole quality [15].

MegaLiner Barrel GLARE teardowns

151

FEG-SEM examination Fractographic analysis of the largest crack in fastener hole 33: The fatigue fracture surface was first checked for “readability”. This was generally excellent: most of the fracture surface, including the very beginning, see figure 13, was covered with uniformly spaced fatigue striations. There were occasional larger striations, but the load history was essentially constant amplitude cycling. This must have been due to the predominance of pressurization loads at the F7 location. As a first approximation the uniform striation spacings were taken to represent the crack growth in each simulated flight. This meant obtaining an a versus da/dN plot first, and then deriving the a versus N plot, i.e. the reverse procedure to that for the F4 window area cracks. Figure 14 shows the derived a versus N plot based on a fatigue “initiation” life of zero. The trend line for the data points shows that the estimated total life is too long by about 18 %. This is most probably because severe simulated flights (occasional larger striations, ignored in the first approximation) accelerated the overall crack growth. Consequently, figure 14 also includes a corrected trend line. Figure 15 is a compilation of the a versus N results for the F7 door beam and F4 window area cracks. There are two main points to make: The data for the door beam crack shows the original a versus da/dN plot and a trend line derived from the corrected plot in figure 14. There is an evident trend of nearly constant crack growth rates. Bearing in mind that the load history was almost constant amplitude cycling, this result is an encouraging validation of the more sophisticated crack growth models for GLARE [9, 10, 13]. There are significant differences in crack growth rates for the F7 and F4 locations and also differences in the trends. These differences are attributable to differing structural geometries, local load levels and load histories, and also different materials (GLARE compared to a monolithic aluminium alloy). Fastener holes 31 − 35: Detailed viewing of the cracked fastener hole bores was enabled by the FEG-SEM, owing to its capability of operating at low kV to minimise charging-up the non-conducting glass fibre layers. Figure 16 gives an example of the observed damage in the bores. The following characteristics of cracking were ascertained: Fatigue cracking began in the aluminium layers, generally at or near corners. However, the higher magnification micrograph in figure 16a shows a kinked rack that initiated both at a corner and heavy scoring. This proves that (a) the scoring was present during the MLB fatigue test and was not a result of fastener removal before the NDI by Airbus Deutschland, and (b) the initial hole quality was indeed poor.

Wanhill et al.

152

As sequential aluminium layers became through-cracked the intermediate glass fibre layers began to protrude into the hole bores. More severe cracking of the aluminium layers led to cracking of the interleaved glass fibre layers as well as further protrusion. Black debris in the hole bores, see the left-hand image in figure 12, was most probably due to cyclic displacements of the protruding glass fibre layers during the MLB fatigue test. In other words, the debris was a kind of fretting product. These results suggest it is difficult to drill good quality fastener holes in a thick GLARE laminate. For the F7 door beam area this meant that fatigue crack growth in some fastener holes began as soon as the MLB test commenced. Even so, the longest fatigue crack was less than 7 mm at the end of the test. Since the applied fatigue load history was set at a conservatively high level, this teardown result once again demonstrates the high fatigue damage tolerance capability of GLARE.

THE F6 STRINGER COUPLING AREA Construction Figure 17 is a schematic of part of the F6 stringer coupling assembly. The materials used in this assembly were: (1) 2524-T351 aluminium skin, 5.25 mm thick, panel B1. (2) GLARE 4A-5/4-0.4 skin, 4.15 mm thick, panel D1. The GLARE code means five 2024-T3 aluminium layers 0.4 mm thick and interleaved with four glass fibre layers 0.25 mm thick. (3) GLARE 2B-10/9-0.4 butt strap, 6.25 mm thick. The Glare code means ten 2024-T3 aluminium layers 0.4 mm thick and interleaved with nine glass fibre layers 0.25 mm thick. (4) 7349-T7651 aluminium, 5.25 mm thick, for the forward stringers. (5) 4 × GLARE 2A-2/1-0.3 bonded by 3 × 0.15 mm adhesive layers, total thickness 3.85 mm, for the aft stringers. The GLARE code means two 2024-T3 aluminium layers 0.3 mm thick and interleaved with one glass fibre layer 0.25 mm thick. (6) 3 thicknesses, 5.85 mm, 4.85 mm, 2.85 mm, of GLARE 2A-2/1-0.3, decreasing in two steps outwards from the coupling mid-points, bonded by 0.15 mm adhesive layers. As before, the GLARE code means two 2024-T3 aluminium layers 0.3 mm thick and interleaved with one glass fibre layer 0.25 mm thick. Teardown procedure The teardown was done for NDI only: Removal of fasteners to enable disassembly of the GLARE stringers and butt strap from the skins and the stringer couplings from the stringers,

MegaLiner Barrel GLARE teardowns

153

butt strap and skin. The aluminium stringers and skin could not be separated because they had been adhesively bonded as well as mechanically fastened. Eddy current rotor inspection of fastener hole bores for the skins, butt strap, stringers and stringer couplings. Eddy current pencil probe inspection of faying surfaces (not possible for the aluminium skin/stringer faying surfaces) to verify the eddy current rotor inspection and estimate crack lengths. Full details of the teardown and NDI are given in Ref. [16]. A summary of the NDI results is given next. NDI results Figure 18 classifies the NDI-indicated crack lengths for the components of the F6 stringer coupling assembly, with the exception of the 2524-T351 aluminium skin, for which there were 4 crack indications. There were many crack indications for the GLARE components, but they were all less than 4.5 mm. However, two aluminium stringers contained cracks with indicated lengths of 6.3 mm and 7 mm [16]. These differences are consistent with GLARE’s susceptibility to fatigue crack “initiation” in the aluminium layers of the laminates, but its increasing resistance to crack growth, owing to fibre bridging [12]. Support for the above interpretation comes from comparing the NDI-indicated crack lengths for GLARE and aluminium stringer couplings. The maximum indicated crack length for the GLARE stringer couplings was 3.4 mm after 45,402 simulated flights [16]; but aluminium stringer couplings gave fifteen indications with crack lengths 6 − 25 mm after about 20,000 simulated flights, and another twenty-four indications with crack lengths 6 – 16 mm after less than 40,000 simulated flights [14].

CONCLUDING REMARKS Teardowns of GLARE structures from three key locations of the MLB have demonstrated very good NDI capabilities and high fatigue damage tolerance by the GLARE structural features. The NDI capabilities were verified for the F4 window area by opening up crack-indicated fastener holes in the GLARE skin, measuring the crack sizes, and carrying out a POD analysis. The high fatigue damage tolerance of the GLARE structural features is inferred from (a) the NDI-indicated and measured crack lengths in GLARE fastener holes from the F4, F6 and F7 locations, (b) their comparison with NDI-indicated crack lengths in aluminium alloy components, and (c) the applied fatigue load history being set at a conservatively high level. The GLARE crack lengths were less than 7 mm after 45,402 simulated flights, when the MLB fatigue test was discontinued.

Wanhill et al.

154

However, some of the aluminium alloy components, notably stringer couplings, had NDI-indicated crack lengths up to 25 mm after “only” about 20,000 simulated flights. The high fatigue damage tolerance of GLARE is due to fibre bridging, which causes increasing resistance to crack growth in the aluminium layers of GLARE. This was indirectly shown by the largest crack in the F7 door beam sample. This crack experienced almost constant amplitude loading and had a nearly constant growth rate, which agrees with the more sophisticated model predictions that account for fibre bridging and its interaction with delaminations at the interfaces between aluminium and glass fibre layers.

REFERENCES [1] [2]

[3]

[4]

[5] [6]

[7] [8]

[9]

Wagner, M. (2001): Fatigue loads program for MegaLiner Barrel, Airbus Deutschland GmbH, Hamburg. Platenkamp, D.J., Bosch, A.F., Wanhill, R.J.H., Hattenberg, T. (2005): MegaLiner Barrel teardown programme, Part I: NDI test results for the F4 window panel, NLR Contract Report NLR-CR-2005-665, National Aerospace Laboratory NLR, Amsterdam. Wanhill, R.J.H., Hattenberg, T., Platenkamp, D.J., Bosch, A.F. (2006): MegaLiner Barrel teardown programme, Part II: fractographic examination and analysis of fatigue cracks for the F4 window panel, NLR Contract Report NLR-CR-2005-666, National Aerospace Laboratory NLR, Amsterdam. Wanhill, R.J.H., Platenkamp, D.J., Hattenberg, T., Bosch, A.F. (2006): Teardown of a GLARE window area from the Airbus MegaLiner Barrel fatigue test article, NLR Technical Publication NLR-TP-2006-183, National Aerospace Laboratory NLR, Amsterdam. Grooteman, F.P. (2005): Personal Communication, National Aerospace Laboratory NLR, Amsterdam. Suresh, S., Ritchie, R.O. (1984): The propagation of short fatigue cracks, International Metallurgical Reviews, Vol. 29, pp. 445-476. Waterhouse, R.B. (1981): Theories of fretting processes. In: Fretting Fatigue, pp. 203-219, Waterhouse, R.B. (Ed.), Applied Science Publishers Ltd, London. Edwards, P.R., Ryman, R.J., Cook, R. (1977): Fracture mechanics prediction of fretting fatigue under constant amplitude loading, RAE Technical Report 77056, Royal Aircraft Establishment, Farnborough. Koning, A.U. de (2001): Analysis of the fatigue crack growth behaviour of “through the thickness” cracks in Fibre Metal Laminates (FML’s), NLR Contract Report NLR-CR-2000-575, National Aerospace Laboratory NLR, Amsterdam.

MegaLiner Barrel GLARE teardowns

[10]

[11]

[12]

[13]

[14] [15]

[16]

155

Alderliesten, R.C., Woerden, H.J.M. (2004): Load history effects during fatigue crack propagation in Glare. In: ICAF 2003, Fatigue of Aircraft Structures as an Engineering Challenge, Vol. I, pp. 509-530, Guillaume, M. (Ed.), EMAS Publishing, Sheffield. Beumler, T. (2004): Flying GLARE. A contribution to aircraft certification issues on strength properties in non-damaged and fatigue damaged GLARE structures, Doctoral Thesis, Delft University Press, Delft. Alderliesten, R.C., Homan, J.J. (2006): Fatigue and damage tolerance issues of Glare in aircraft structures, International Journal of Fatigue, Vol. 28, pp. 1116-1123. Alderliesten, R.C. (2007): Analytical prediction model for fatigue crack propagation and delamination growth in Glare, International Journal of Fatigue, Vol. 29, pp. 628-646. Borgonje, B., Meier, Th. (2006): MegaLiner Barrel – teardown of the Glare structure, Test Analysis Report, Airbus Deutschland GmbH, Hamburg. Wanhill, R.J.H., Hattenberg, T. (2008): MegaLiner Barrel GLARE location F7: fractographic analysis of fastener holes just above the passenger door in panel B3, NLR Technical Report NLR-TR-2006-433, National Aerospace Laboratory NLR, Amsterdam. Platenkamp, D.J., Bosch, A.F., Wanhill, R.J.H. (2007): MegaLiner Barrel teardown programme: NDI test results for the F6 Glare stringer coupling programme, NLR Contract Report NLR-CR-2007-629, National Aerospace Laboratory NLR, Amsterdam.

156

Wanhill et al.

Figure 1 The MLB configuration and general loading conditions

MegaLiner Barrel GLARE teardowns

Figure 2 “Opened out” view of the MLB

157

158

Wanhill et al.

Figure 3 The F4 GLARE window area just after its removal from the MLB

Figure 4 NDI crack indications for the GLARE skin and aluminium window frame fastener hole bores of window C65-C66: DOF = Direction Of (simulated) Flight

MegaLiner Barrel GLARE teardowns

159

Figure 5 Examples of fatigue cracking patterns in the GLARE aluminium layers

[a]

[b]

Figure 6 Largest window frame fatigue crack: (a) overview with fatigue origin arrowed; (b) detail with fretting scar in fastener hole bore arrowed

Figure 7

Example of identifying severe simulated flights from crack front markings due to peak loads

Figure 8

Crack growth curves for the largest window frame crack. The material was die forged aluminium alloy 7175-T73

160 Wanhill et al.

MegaLiner Barrel GLARE teardowns

161

Figure 9 Crack growth curves for the largest “readable” GLARE skin crack

Figure 10 The F7 GLARE door beam area before removal of the rectangular sample indicated by the white border. The arrows point to the subsequent failure position

162

Wanhill et al.

Figure 11 Failure position of the rectangular door beam sample. Arrows in the lower photograph point to cracks at the nearby fastener holes 11, 32 and 34

MegaLiner Barrel GLARE teardowns

163

Figure 12 Macrofractographs for fastener hole 33. The arrow points to the largest fatigue crack

Figure 13 Fatigue crack growth (striations) commencing directly from the bore of fastener hole 33. The arrow points to a fatigue striation “plateau”

164

Wanhill et al.

Figure 14 Crack growth plot, a* versus N, derived from fatigue striation spacings for the longest crack in the door beam GLARE sample

Figure 15 Comparison of crack growth rates for the longest cracks in the F7 and F4 locations

[b]

Figure 16 Example of fatigue cracking and damage in the bore of fastener hole 34

[a]

MegaLiner Barrel GLARE teardowns 165

Figure 17 Schematic of partly disassembled F6 stringer coupling assembly: DOF = Direction Of (simulated) Flight

166 Wanhill et al.

MegaLiner Barrel GLARE teardowns

167

Figure 18 Classification of NDI-indicated cracks for the F6 stringer coupling assembly

25th ICAF Symposium – Rotterdam, 27–29 May 2009

30 YEARS OF DAMAGE TOLERANCE – HAVE WE GOT IT RIGHT? Robert G. Eastin and Jon B. Mowery Federal Aviation Administration

Abstract: This paper examines how the damage tolerance requirements for large transport airplanes have changed over the last thirty years with respect to (1) reliance on inspection, (2) operational life, and (3) inspection threshold. In considering each subject area, the relevant damage tolerance requirements as originally adopted with Amendment 25-45 [1] in 1978 are reviewed. Following this, the changes that were made with Amendment 25-96 [2] in 1998 are identified and their significance discussed. A qualitative assessment of the current requirements at Amendment 25-96 is made followed by suggestions for future revisions. It is concluded that relative to “reliance on inspection” only relatively minor revisions are needed. Relative to “operational life” it is suggested that a change should be made to make what is arguably an implicit requirement to establish an operational life limit an explicit one. Lastly, relative to “inspection threshold” it is concluded that the current requirements are problematic and a major change is needed.

INTRODUCTION Damage tolerance requirements were first introduced into Title 14 of the Code of Federal Regulations (14 CFR) just over 30 years ago with Amendment 25-45 in 1978. This was done in response to concerns with total reliance on a structure’s “fail-safe” capability as shown and found in accordance with the fail-safe requirements of CAR 4b [3] and 14 CFR part 25 prior to Amendment 25-45. The original damage tolerance requirements were put into § 25.571 of 14 CFR and remained virtually unchanged for 20 years. The current damage tolerance requirements were put into place in 1998 with Amendment 25-96. The changes made with Amendment 25-96 were very significant from both a philosophical and M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 169–186. © Springer Science+Business Media B.V. 2009

Eastin and Mowery

170

practical perspective. In the discussion that follows, the original and current requirements are examined relative to the threat of fatigue, and their impact, or lack thereof, on “reliance on inspection”, “operational life” and “inspection threshold”. Observations are made on the effectiveness of the current requirements relative to meeting the stated safety objective of avoiding catastrophic failures due to fatigue during the operational life of an airplane. In closing, suggestions are made for revisions to the current requirements. The views and opinions expressed in this paper are those of the authors and should not be interpreted as official FAA position, policy or rules interpretation.

RELIANCE ON INSPECTION The degree to which inspections are relied upon to prevent catastrophic failures due to fatigue has changed significantly since the adoption of the damage tolerance approach in 1978. In the beginning, there was an expectation, consistent with the rule, that with a few minor exceptions, airframe structure could be practically and reliably kept safe indefinitely with inspections alone. Today the rule actually precludes the exclusive use of inspection for certain structural details that are common to, and represent a significant amount of structure in, typical metallic airplane type designs. Amendment 25-45 to the Aloha Incident As discussed by Eastin and Mowery [4], Amendment 25-45 removed the fail-safe option for dealing with the threat of fatigue and specified damage tolerance as the approach that had to be used unless shown to be impractical. In the context of the rule, using the damage tolerance approach meant that there would be total reliance on inspection to prevent catastrophic failures due to fatigue. Although an allowance was made in the rule for proactive replacement or modification before fatigue cracking became likely to occur (i.e. use of “safe-life” or the safety-byretirement approach), there was a clear expectation that this would rarely be done. This expectation was reinforced by preamble information contained in the Notice of Proposed Rulemaking (NPRM) [5] and the Final Rule [1]. On the other hand, the corresponding advisory circular [6] contained guidance that indicated that there was some airframe structure where inspections might not be practical. For example, the following was included in section 3 of [6]: “(4) Assessing the fatigue characteristics of certain structural elements, such as major fittings, joints, typical skin units, and splices, to ensure that the anticipated service life can reasonably be attained, is needed for structure to be evaluated under § 25.571(c).”

30 Years of damage control – Have we got it right?

171

However, this sub-paragraph was removed in 1986 with publication of AC 25.5711A [7] and was overshadowed with other guidance contained in section 3 of [6] that read as follows: “(2) Damage-tolerance design is required, unless it entails such complications that an effective damage-tolerant structure cannot be achieved within the limitations of geometry, inspectability or good design practice. Under these circumstances, a design that complies with the fatigue evaluation (safe-life) requirements is used. Typical examples of structure that might not be conducive to damage-tolerant design are landing gear, engine mounts, and their attachments.” It should be noted that in 1986, when sub-paragraph (4) was removed, the preceding text was revised to remove the reference to “engine mounts”, thus further reinforcing the expectation that safety-by-inspection would be the norm for virtually the entire airframe. Consistent with expectations, and the fairly explicit guidance provided, compliance with § 25.571 was shown and found almost exclusively based on safety-byinspection (i.e. § 25.571(b)). This includes structure that in hindsight and by today’s standards would be considered not amenable to total reliance on inspection. This same philosophy was adopted for the Supplemental Inspection Documents (SIDs) that were voluntarily developed by the type certificate holders of certain aging transport category airplanes that were certified to the fail-safe requirements of [3]. This was in spite of guidance contained in AC 91-56 [8] that indicated that replacement or modification might also be a likely outcome of the evaluation performed. For example, section 5 of Appendix 1 of [8] contained the following guidance: “5.1 The Supplemental Inspection Document should contain the recommendations for the inspection procedures and replacement or modification of parts or components necessary for continued safe operation of the airplane.” Nevertheless, the SIDs that were created relied almost exclusively on inspections. Additionally, this same philosophy was extended to in-service fatigue cracking problems wherein it was common practice to implement and allow continued inspection of areas that had been found to be cracked in other airplanes. Repair or modification was on condition, with terminating actions (e.g. replacement, improved material, etc.) left as optional.

172

Eastin and Mowery

Aloha Accident to Amendment 25-96 The much discussed and well-known Aloha accident that occurred in April of 1988 resulted in an industry-wide reexamination of past practices, especially with respect to total reliance on inspection for continued airworthiness. This resulted in some significant changes in FAA policy. One of the best examples of this is the policy change implemented for airworthiness directive (AD) actions addressing unsafe conditions. This is documented in FAA Memoranda from Keith [9, 10]. In [9] it states that: “As a result of extensive industry and FAA efforts following the Aloha Airlines 737 accident in Hawaii in April 1988, it has become apparent that repetitive long term inspections or special procedures may not be providing the degree of safety assurance necessary for the aging transport airplane fleet. This, coupled with a better understanding of the human factors associated with numerous repetitive inspections or procedures, has led us to consider placing less emphasis on repetitive inspections and special procedures, and more emphasis on design improvements. Thus, we find it necessary to shift our direction in AD actions addressing unsafe conditions. In lieu of our previous position, we now acknowledge that long term continued operational safety will be better assured by design changes to remove the source of the problem, rather than by repetitive inspections or special operating procedures. It is essential that, beginning immediately, we initiate this policy in our AD rulemaking process.” In [10] several criteria are presented to be used to determine when continued reliance on inspection is unacceptable. One criterion focuses squarely on inspection practicality and reliability and recognizes that there are many areas where inspection should only be relied on as an interim action until terminating actions can be accomplished. Consistent with this, all proposed ADs that did not include a mandated terminating modification required supporting rationale and special approval. It follows that total reliance on supplemental inspections (e.g. as included in a SID or Airworthiness Limitations Section (ALS) of the Instructions for Continued Airworthiness (ICA)) of these kinds of areas would have been similarly questionable and that proactive replacement or modification would be the proper action. The Aloha accident also led to formation of the Airworthiness Assurance Task Force that was the predecessor of the Airworthiness Assurance Working Group (AAWG) of the Aviation Rulemaking Advisory Committee (ARAC). This was an industry group chartered by the FAA to consider the general issue of aging aircraft structures’ continued airworthiness and make recommendations to the FAA. This group met repeatedly throughout the 1990’s and into the early 2000’s. One of the

30 Years of damage control – Have we got it right?

173

main focuses of this group was concurrent fatigue cracking at multiple sites in the same structural elements (MSD) and in multiple adjacent elements (MED). Over time, a consensus was reached that total reliance on inspection for this kind of cracking was very questionable and additional actions were needed for structures prone to crack in this fashion. Although the AAWG didn’t recommend that there should be rulemaking until 1999 in [11] and didn’t finalize their recommendations on the subject until 2003 in [12], the concern with MSD and MED caused the FAA to move forward unilaterally with Amendment 25-96 in 1998. Amendment 25-96 Amendment 25-96 revised § 25.571(b) to include the following: “Special consideration for widespread fatigue damage must be included where the design is such that this type of damage could occur. It must be demonstrated with full scale fatigue test evidence that widespread fatigue damage will not occur within the design service goal of the airplane.” A definition of widespread fatigue damage is included in the corresponding guidance material [13]. Also included are definitions of MSD and MED, which are defined as sources of widespread fatigue damage. The supporting rationale for the revision is discussed in the preamble information of the NPRM [14] and Final Rule [2]. It can be summarized as follows: (1) Inspections alone cannot be relied on for structural areas that are susceptible to widespread fatigue damage. (2) Because of (1) it must be demonstrated that widespread fatigue damage will not occur while the airplane is in service. (3) (2) can only be achieved with full-scale fatigue test evidence that shows that widespread fatigue damage will not occur within the design service goal of the airplane. The preamble information in [2] and guidance material in [13] advises that at least two design lifetimes of full-scale fatigue testing are needed, along with extensive teardown inspection of the test article to satisfy the requirement. Have we got it right? Not quite, but we have come a long way. The rule finally recognizes that there may be a significant number of areas in an airplane’s type design where safety-byinspection is impractical. However, there are several problems with the existing rule. One is that it completely rules out inspection for certain types of areas. Another is that those areas are identified using two very narrow and problematic definitions of hypothetical cracking scenarios.

Eastin and Mowery

174

It is inappropriate to completely rule out inspection in the rule. Needed cautions and/or qualifiers against reliance on inspection can be included in guidance material. However, the results of the required damage tolerance evaluation should be used to make the final determination as discussed by Eastin and Swift [15], and those results depend on many factors and should not be presupposed. There are many examples where it has been demonstrated by analysis, test and/or service experience that inspections are not sufficiently practical or reliable to be used to insure continued airworthiness. Many of these examples do not fit the definitions of MSD and MED provided in [13]. Use of these definitions to make the determination of where inspections cannot be relied on only serves to artificially and incorrectly limit the areas where reliance on inspection should not be allowed. A way forward. The rule should be revised to be more pragmatic about determining how potential fatigue should be managed during the operational life of an airplane. It should require the applicant to perform damage tolerance evaluations but not presuppose what the outcome will be based on problematic definitions of two different fatigue cracking scenarios. As currently written, the rule does not allow inspection if fatigue cracking fits the definition of MSD or MED, and requires it (if not demonstrated to be impractical) if it doesn’t. This is too prescriptive and not realistic or correct. Additionally, the implied, if not explicit, message that special directed inspection is preferred should be eliminated. This is incongruous with the dictate that inspection cannot be used to manage MSD or MED. The rule requires damage tolerance evaluations be performed and based on the results, inspections or other procedures must be established to prevent catastrophic failures due to fatigue. This requirement is objective and appropriate and should be retained. Also, the requirement to perform full-scale fatigue testing should be retained, but the purpose should be changed from just addressing MSD/MED to addressing all fatigue scenarios that could result in a catastrophic failure that cannot be managed with inspections. Sufficient full-scale test evidence should be required to demonstrate that any fatigue that cannot be managed with inspections will be unlikely to occur for a defined (by the applicant) period of operation. The period demonstrated should then be established as a qualified operational limit. This last point is discussed in some detail in the following section.

OPERATIONAL LIFE The damage tolerance requirements adopted in 1978 did not include any explicit requirement to establish a finite operational life for the type design. There is still no such requirement in the current rules although it can be argued that one is implied and that it is equal to the design service goal.

30 Years of damage control – Have we got it right?

175

Amendment 25-45 and Past Industry Practice Due to several accidents, most notably the Lusaka accident that has been discussed by Bristow and Eastin [16], Amendment 25-45 introduced the following requirement into § 25.571(a): “An evaluation of the strength, detail design and fabrication must show that catastrophic failure due to fatigue, corrosion, or accidental damage, will be avoided throughout the operational life of the airplane.” This same section under subparagraph (a)(3) stated that: “Based on the evaluations required by this section, inspections or other procedures must be established as necessary to prevent catastrophic failure….” But there was no requirement to establish a limit on the operational life of the airplane. “Inspections or other procedures” were to be put in place that, as long as they were accomplished, would insure the safety for as long as the airplane was operational. In order to comply with the letter of the law there appear to be two viable approaches, described as follows: (1) Leave the operational life indefinite and establish all the “inspections or other procedures” needed to insure safety for an indefinite period of time. (2) Establish a finite operational life and establish the “inspections or other procedures” needed to insure safety to that point but not necessarily beyond. The first approach is unrealistic. The problem of defining all maintenance actions that will ever be required to operate indefinitely is unbounded and impossible given a finite amount of time, money and engineering data. The second approach has typically been avoided for many reasons that are beyond the scope of this discussion. In fact, for many type designs neither approach was used. A commonly accepted practice was to establish all the maintenance actions deemed to be necessary for the period of time that the airplane was expected to remain in service. A service life was established pragmatically by the manufacturer for marketing purposes and became a design life objective that was used for any fatigue analyses and any testing performed. It also became a de facto life limit for determining what was needed to prevent catastrophic failures due to fatigue. However, this number was not required to be submitted as part of the type design. The lack of a requirement to formally declare a service life limit is consistent with the notion of total and indefinite reliance on inspection that was previously discussed. This can be seen in the preamble to Amendment 25-45 where public comments to the NPRM are dispositioned. One commenter recommended that the

Eastin and Mowery

176

applicant be required to determine the “time to first failure of critical structure.” The FAA replied in part that: “The FAA recognizes that this procedure may be necessary for some critical safe-life components but not for those evaluated by means of the damage tolerance approach covered in this amendment.” Amendment 25-96 In Amendment 25-96, the FAA addressed the issue of how long an inspection program will be effective but didn’t follow through with a requirement for an operational life. As previously noted, one of the changes made with this Amendment is the inclusion of a requirement to demonstrate with full-scale fatigue testing that widespread fatigue damage will not occur within the design service goal of the airplane. Also significant with this Amendment is the addition of the requirement for inspection thresholds that will be addressed later in this paper. In the preamble information of [2] one of the questions that was raised in explanation for the need for this change was: “When in an airplane’s life can safety no longer be effectively maintained by the damage tolerance inspection program prescribed at the time of certification of the airplane type (the onset of widespread cracking)?” Also in the preamble is the stated objective of the rulemaking: “…the objective of this rulemaking is to ensure that transport category airplanes will remain free of widespread fatigue damage within their design service goal.” From both of these statements it can be inferred that the idea of an unbounded life, with safety being maintained by damage tolerance based inspections alone is no longer acceptable. The changes in the rule require that full-scale test evidence be used to demonstrate that widespread fatigue damage not occur within the design service goal. The rule goes on to require that if the full scale test is not complete that no airplane can be operated to more than ½ of the number of cycles accumulated on the test article until the test is complete. However, there is no requirement to identify what the final design service goal is. In fact, even though airplane operation is limited by regulation to ½ the number of test cycles until the test is finished, there is no regulation limiting the operation of the airplane after the test is finished! The guidance material of [13] says in part:

30 Years of damage control – Have we got it right?

177

“In general, sufficient full-scale test evidence consists of fatigue testing to two or more times the design service goal, followed by specific inspections and analysis …” Again there is a reference to a “design service goal” but no regulatory requirement to define it or stop airplanes from over-flying it. Only a few type designs have been certified to Amendment 25-96. A notable example is the Airbus A380. Although not explicitly required by the rule, Airbus published a Limit of Validity (LOV) in the ALS of the ICA for the A380. This is a limit on how long an airplane can remain in service with the fatigue management program that was established by the manufacturer as part of type design certification. Have we got it right? Almost. With the inclusion of fail-safety in the 1950’s as an option to the safe-life approach, airplane manufacturers could achieve type design certification without defining an operational limit. Amendment 25-45 did not add any such requirement even though it is implied in the rule that there is an “operation life.” This appears to be based on the expectation that airplanes would be retired due to economic or advancing technological factors prior to reaching the point where inspections could no longer achieve the level of safety required. Amendment 25-96 changed the rule significantly to address the fact that in some cases fatigue can and does reduce the strength of the structure to an unacceptable lever prior to becoming detectable. Amendment 25-96 requires the use of full-scale test evidence to demonstrate that such structure will not degrade to an unsafe state within the design service goal. Although Amendment 25-96 could be interpreted to mean that airplanes should not operate beyond their design service goals unless additional testing and analyses (over and above what was required for type design certification) are performed, there is no regulatory requirement in place to stop operation of the airplane at any defined point in its life. A Way Forward A defined operational life is needed to maintain the enviable level of safety currently enjoyed by the aviation system. This has not been much of an issue in the past as airplanes were being retired for economic reasons or due to advances in technology. The airplanes simply didn’t remain in service long enough for fatigue wear-out to be a general problem. The notion of a finite operational life has always been with us. Whenever a manufacturer defined their fatigue test cycles, planned their inspections, or designed their safe life parts they were working with an idea of how long they expected their airplane to be in service. Indeed, with Airbus defining an LOV for the A380 identifying how long the type design can be safely operated with the

Eastin and Mowery

178

fatigue management programs that have been established, the industry is leading the way in defining bounds that should not be over-flown. It must be reiterated exactly what this limit is and is not. It is based on considering the inherent fatigue and damage tolerance characteristics of the type design, establishing an appropriate fatigue management program and determining how long the type design can be safely operated with confidence. The regulations should require this number be defined, supported by test evidence and made generally available. As service information becomes available, more testing and analysis is accomplished and additional maintenance actions are added to the ALS of the ICA (as needed to manage fatigue), the limit can be adjusted. This is effectively what happens when the Type Certificate is issued prior to the completion of the fatigue test.

INSPECTION THRESHOLD When must damage tolerance based inspections start? We’ve gone from no requirement or guidance at Amendment 25-45 to a very prescriptive but incomplete requirement that is problematic and guidance that is paradoxical. Amendment 25-45 and Past Industry Practice There was no specific mention of inspection threshold in the final rule [1] or guidance material [6] when the damage tolerance approach was adopted in 1978. Applicants were allowed to adopt their own philosophy and methods to establish thresholds. Among major airframe manufacturers the approach used was generally the same. It is commonly referred to as the fatigue life approach. Using fatigue life to set inspection thresholds appears reasonable if the primary objective of the inspection is to detect the first signs of symptomatic fatigue in a fleet of airplanes. In applying this approach, experience-based variations in manufacturing quality were typically considered and inspections started when there was a certain probability of cracks of a detectable size being present. Goranson reported in [17] that detectable cracking probabilities between 1 in 50 and 1 and 250 were used when applying this approach. It was also reported in [17] that this approach had been used successfully for many years and suggestions that it needed to be abandoned for another approach were challenged. Concepcion Escobedo Medina et al [18] reported more recently that the fatigue life approach has also been used satisfactorily by Airbus over the years to set inspection thresholds. Successful application of the fatigue life approach requires reliable prediction of the fatigue life of the structure being addressed, which is very dependent on having sufficient component and full-scale fatigue test data and service experience. Due

30 Years of damage control – Have we got it right?

179

to lack of this information and other reasons another approach was sometimes adopted. This approach has often been referred to as the “rogue” flaw approach. The rogue flaw approach used for establishing thresholds can be traced back to the United States Air Force (USAF) damage tolerance philosophy that was adopted and applied starting in the 1969/1970 time period. The application of this philosophy in the assessment of existing USAF airplanes was discussed by Tiffany [19] at the 9th ICAF Symposium. The figure below depicts the safety criteria used in the assessment process. An initial crack is assumed to be present when the structure first enters service and the time it takes to grow to a point that the strength falls below an acceptable level is determined. This time is established as the “safety limit” for the structure being assessed. The initial crack or rogue flaw size was specified along with the crack growth scenario to be assumed and the residual strength that had to be maintained. The safety criteria used requires the structure to be inspected initially at 1/2 the safety limit.

FAIL

SAFETY LIMIT FLAW SIZE INSPECT

ai FLIGHT HOURS .03 TO .05 INCH FLAW SIZES AND SHAPES:

USAF Safety Criteria (adopted from [19]) The safety criteria described above is fundamental to the USAF damage tolerance design requirements contained in MIL-A-83444 [20]. These requirements were discussed in some detail by Wood [21]. In accordance with [20] the requirements that a new USAF design must meet if it is qualified as slow crack growth inservice non-inspectable category structure follow directly from the safety criteria. For this category the structure must be designed so that the specified initial cracks

180

Eastin and Mowery

will not grow to the point that residual strength is reduced below the required level in less than two design service lifetimes. For structure designed accordingly it follows that no special inspections for fatigue are required prior to one design service life (i.e. the inspection threshold is greater than the service life). The rogue flaw approach became an acceptable and popular default method for determining inspection thresholds. There were several reasons for this. First it came with built-in credibility because it was being used by the USAF on existing aircraft and for new USAF aircraft designs. Secondly its application required relatively few resources and little imagination or further clarification. The applicant did not need access to fatigue performance data. The initial crack sizes and growth scenario to be used were defined in detail in [20] and software applications needed to perform the calculations were readily available. Lastly and perhaps more importantly, the use of the approach satisfied the strongly held views of some that inspection thresholds for civil aircraft should account for the presence of initial rogue flaws. This view was promoted and discussed by Tom Swift who was the FAA’s Chief Scientific/Technical Advisor for Fracture Mechanics/Metallurgy up until mid-1997. In [22] he commented that when the fatigue life approach is used “the true damage tolerance philosophy is diluted with the safe life approach to some extent.” He further states that using the rogue flaw approach discussed previously “is still a damage tolerance approach undiluted by a safe life philosophy.” Amendment 25-96 Amendment 25-96 introduced inspection threshold requirements into § 25.571 for the first time. The NPRM [14] that preceded Amendment 25-96 noted that the existing regulations did not prescribe criteria for establishing thresholds for the detailed structural inspection program. In discussing the rationale behind the proposed new requirement, the need to “account for premature failures due to undetected manufacturing defects (rogue flaws)” was sited. This implied a relatively high degree of pessimism about the state of the finished product given that the aforementioned USAF rogue flaws were considered to be relatively rare as discussed in [19] and [21]. Reference [14] goes on to say that “Initial inspection thresholds should be established based on cracks growing from likely defects developed during manufacture such as machining marks, improper installation of fasteners, etc.” It is noted that this implies a somewhat different level of pessimism. The requirement as proposed in [14] read: “These procedures must include threshold for inspections that are based on analyses and test considering the damage tolerance design concept, manufacturing quality, and susceptibility to inservice damage.” However, this relatively objective requirement did not make it into the final rule. Based on comments received, the FAA replaced it with the following:

30 Years of damage control – Have we got it right?

181

“Inspection thresholds for the following types of structure must be established based on crack growth analyses and/or tests, assuming the structure contains an initial flaw of the maximum probable size that could exist as a result of manufacturing or service induced damage: (i) Single load path structure, and (ii) Multiple load path “fail-safe” structure and crack arrest “fail-safe” structure, where it cannot be demonstrated that load path failure, partial failure, or crack arrest will be detected and repaired during normal maintenance, inspection, or operation of an airplane prior to failure of the remaining structure.” Have we got it right? The current threshold requirement is problematic at best. The major reasons are discussed below. Requiring that a specific analysis methodology be used and thereby ruling out others is counter to FAA rulemaking policy that promotes objective requirements. This precludes allowing the applicant to propose a means to comply with a stated objective. Further, requiring that a crack growth approach be used infers that this is the only way that structural discrepancies can be properly addressed when it comes to determining when to start an inspection. What is the basis for this? Even though the USAF chose crack growth, the fatigue life approach shouldn’t be ruled out. For example, if the fatigue life approach was based on fatigue life data from structure with the discrepancies of concern present (e.g. joints lacking proper clamp-up, machine marks, and improper surface finish) then those “flaws” would be inherently accounted for. In fact, from an inspection threshold perspective it can be argued that the fatigue approach is more appropriate since based on service experience most of the discrepancies of concern are not initial cracks (e.g. examples sited in [19] and [21]). Their primary impact is on fatigue crack initiation life and in most cases they have little or no impact on subsequent crack growth life. So while a crack growth approach may by defined that will bound certain discrepancies it doesn’t automatically make it a superior approach. However, if a crack growth approach must be used there has to be more information and direction on the details. Just specifying that a certain generic approach be used almost insures a non-uniform approach to compliance by applicants. For example, there is the problem of establishing the initial crack size(s) to be used to start the analysis and the cracking scenario to be followed. This is in itself is a non-trivial problem. The USAF effort to establish a crack size to be used to bound manufacturing defects in fastener holes in aluminum structure has been discussed in [19] and by Rudd [23]. This was a very time consuming

182

Eastin and Mowery

effort involving extensive teardown inspection of retired aircraft and analysis of crack findings. In the end, the results were inconclusive and the final rogue flaw size was based largely on engineering judgment. Beyond size(s) and scenario, what should be the end point of the analysis and what factor if any should be applied to the predicted life? None of these details are provided in either the rule or the advisory material. While the rule may not provide the details, the advisory material should at least give some guidance. As it stands now, making a compliance finding to the rule is left up to a negotiated settlement between the applicant and the agency, which does nothing to assist in standardizing the approach to regulatory compliance. The USAF rogue flaw approach discussed previously has been used and accepted as a default standard, but there is no evidence to suggest that using this approach will meet the safety objective for all materials, structural details, and “maximum probable….damage” that might be encountered. Another problem is requiring the applicant to assume “the structure contains an initial flaw of the maximum probable size that could exist as a result of manufacturing or service induced damage”. How does one quantify this? Based on many years of experience, there has been very large damage inflicted during manufacturing and even larger in service that has gone undetected. Does this make it probable? Are there any limits on what has to be accounted for? This element of the requirement by itself could easily be a show stopper if interpreted literally. Unfortunately the guidance provided in [13] does little to shed light on the issues discussed above. The guidance material simply parrots the rule and adds what is considered to be a paradoxical statement that reads as follows: “Where it can be shown by observation, analysis, and/or test that a load path failure in multiple load path “fail-safe” structure, or partial failure in crack arrest “fail-safe” structure, will be detected and repaired during normal maintenance, inspection, or operation on an airplane prior to failure of the remaining structure, the thresholds can be established using either: (i) Fatigue analysis and tests with appropriate scatter factor; or (ii) Slow crack growth analyses and tests, based on appropriate initial manufacturing damage.” What does this mean? It appears to allow the use of the fatigue life approach if it can be shown that fatigue cracking will be found and corrected in the course of normal maintenance, inspection and operation before it creates a safety issue. But, if this is the case the rule would not require establishment of any special directed inspections in the first place so the method used to establish a threshold would be a non-issue!

30 Years of damage control – Have we got it right?

183

Lastly, it is inappropriate to include “service-induced damage” as something that must be addressed when establishing inspection thresholds. For civil airplane certification, inspections for service-induced damage, which is more commonly referred to as accidental damage (AD), are established using the MSG-3 [24] process and documented in the Maintenance Review Board Report. For these inspections, there is no allowance for thresholds in the same sense as there is for inspections for fatigue damage (FD). This is because AD can occur at any time starting with the first day of operation. It follows that AD inspection requirements, unlike those for FD, must be available to operators of, and are applicable to all airplanes when they first enter service. A Way Forward. The rule should include objective requirements for setting inspection thresholds and guidance material should be provided to address acceptable means of compliance. It is proposed that inspection thresholds for all structure that requires special directed inspections to insure continued airworthiness account for variations in manufacturing quality only. Also, it is not practical or appropriate to account for the effects of “maximum probable….service-induced damage.” Service-induced damage is, in general, too varied and unpredictable to describe and bound with any degree of confidence and allowing for a threshold is unacceptable. Inspections for AD, developed in accordance with [24], are the practical and appropriate way to deal with it. The variations in manufacturing quality that should be accounted for are those that could be considered likely to occur on all airplanes. This does not encompass the USAF rogue flaw as characterized by in [19, 21 and 23]. The USAF rogue is a “one off” which is unlikely and not part of the normal manufacturing defect population. Accounting for such a rogue in the establishment of inspection thresholds is unnecessary. This is because special directed inspections target only the very highest stressed, most fatigue sensitive areas that are only a small percentage of the overall structure. By definition, rogue defects can occur anywhere independent of stress level. The probability of such a rogue defect occurring and also occurring in an area that will be subject to a special directed inspection for fatigue cracking is considered sufficiently remote as to not merit further attention from an in-service fleet-wide inspection program perspective. Past civil airplane service experience validates this position. So, based on the preceding discussion, what should the inspection threshold requirements be? ARAC recommendations for revisions to § 25.571 contained in [25] include proposed new inspection threshold requirements. It was proposed that the current requirement be replaced with the following:

Eastin and Mowery

184

“When special inspections are required to prevent catastrophic fatigue failure, inspection thresholds must be established to ensure that cracking in a PSE will be detected before it results in a catastrophic failure. The inspection thresholds must account for the variations of manufacturing quality.” This proposed requirement is considered acceptable and satisfies the goal of having objective rules. The ARAC recommendations referenced above also include proposed advisory material. The guidance proposed allows the applicant to select a methodology. Both the crack growth and fatigue life approaches are recognized as potentially acceptable for all types of structure. However, it emphasizes that there should be validation regardless of the methodology selected. It also notes that a conservative estimate of a lower bound threshold for fastener hole inspection might be obtained for aluminum structure if a single initial corner crack of radius .05” is assumed and the life to critical size is divided by 2. In the end, the burden is on the applicant to satisfy the objective of the rule, which is to start the inspection early enough to detect cracking before it becomes unsafe in any airplane in the fleet given likely variations in manufacturing quality.

CONCLUSIONS This paper considered 14 CFR 25.571 changes that have occurred over the last 30 years relating to reliance on inspection, operational life and inspection thresholds. In general it is concluded that the changes that have occurred have been for the better. We haven’t got it right yet, but for the most part we are moving in the right direction. Specific conclusions about the three areas considered are summarized below. Reliance on Inspection We have moved away from an almost total reliance on inspection. For certain types of structure there is now a requirement to achieve the level of fatigue performance required to insure that fatigue cracking will be unlikely prior to the design service goal. This is because it was recognized that there are circumstances when safety-by-inspection is neither practical nor sufficiently reliable. This is a positive change and results in a more realistic and balanced approach to fatigue management. However, there are some revisions that are needed. Categorically ruling out inspection as an acceptable fatigue management strategy for structural areas determined to be susceptible to MSD and MED, as defined in the advisory material, is questionable. Conversely, implying that safety-byinspection will be appropriate for all other areas is equally questionable. Selection of appropriate maintenance actions should be based on the results of damage tolerance evaluations and these results should not be presupposed. Consistent with this, the need for providing sufficient full-scale fatigue test evidence should be

30 Years of damage control – Have we got it right?

185

broadened to address all structural areas where it has been determined that total reliance on inspection is questionable. Operational Life Although an operational limit is implied by the current regulations, there is still no requirement to establish a bound. A rule revision is needed to require that the period of time for which it has been demonstrated that fatigue cracking is unlikely to occur in structural areas, where safety-by-inspection has been determined to be impractical, be identified as the end of permissible operation. This period of time would be subject to revision based on additional data generated from analyses, testing and service experience. Inspection Threshold We’ve gone from no requirement or guidance to a very prescriptive but incomplete requirement that is problematic and guidance that is paradoxical. A significant revision is needed. An acceptable proposal was submitted to the FAA in [25]. The proposal includes an objective requirement and corresponding advisory material.

REFERENCES [1] FAR Final Rule, Federal Register: October 5, 1978 (Volume 43, Number 194), 14 CFR Part 25 (Docket No. 16280; Amendment No. 25-45). [2] FAR Final Rule, Federal Register: March 31, 1998, (Volume 63, Number 61), 14 CFR Part 25 (Docket No. 27358; Amendment No. 25-96). [3] Civil Air Regulations, Amendment 4b-3, Airplane Airworthiness Transport Categories, March 13, 1956. [4] Eastin, R. and Mowery, J. (2007). In: Durability and Damage Tolerance of Aircraft Structures: Metals vs. Composites, Proceedings of the 24th ICAF Symposium, vol I, pp. 55-72, Lazzeri, L. and Salvetti, A. (Eds.), Pacini Editore, Pisa. [5] FAR Notice of Proposed Rulemaking, Federal Register: August 15, 1977 (Volume 42, Number 157), 14 CFR Part 25 (Docket No. 16280; Notice No. 7715). [6] FAA Advisory Circular AC 25.571-1, Damage Tolerance and Fatigue Evaluation of Structure, 9/28/78. [7] FAA Advisory Circular AC 25.571-1A, Damage Tolerance and Fatigue Evaluation of Structure, March 5, 1986. [8] FAA Advisory Circular No. 91-56, Supplemental Structural Inspection Program for Large Transport Category Airplanes, May 6, 1981. [9] Keith, L., Guidance Concerning Continued Operational Safety and Airworthiness Directives, Internal FAA Memorandum, 8 May 1989, Seattle, USA.

186

Eastin and Mowery

[10] Keith, L., Additional Guidance Concerning Continued Operational Safety and Airworthiness Directives, Internal FAA Memorandum, 16 July 1990, Seattle, USA. [11] N.N., Recommendations for Regulatory Action to Prevent Widespread Damage in the Commercial Airplane Fleet, A Report of the Airworthiness Assurance Working Group, final report, revision A, June 1999 [12] N.N., Widespread Fatigue Damage Bridging Task, A Report of the Airworthiness Assurance Working Group, July 23, 2003. [13] FAA Advisory Circular AC 25.571-1C, Damage Tolerance and Fatigue Evaluation of Structure, 4/29/98. [14] FAR Notice of Proposed Rulemaking, Federal Register: July 19, 1993 (Volume 58, Number 136), 14 CFR Part 25 (Docket No. 27358; Notice No. 93-9). [15] Eastin, R. and Swift, S. (2005). In: Structural Integrity of Advanced Aircraft and Life Extension for Current Fleets – Lessons Learned in 50 Years after the Comet Accidents, Proceedings of the 23rd ICAF Symposium, vol. I, p.43-54, Dalle Donne, C. (Ed.), DGLR-Bericht. [16] Eastin, R., Bristow, J. (2003), Looking at Lusaka’s Lessons, Proceedings of the 2003 USAF Aircraft Structural Integrity Program Conference. [17] Goranson, U. (2007). Damage Tolerance Facts and Fiction, International Conference on Damage Tolerance of Aircraft Structures, Delft, The Netherlands. (http://dtas2007.fyper.com) [18] Concepcion Escobedo Medina, M. and Kimmins, S. and Rodrigo, P. (2005). In: Structural Integrity of Advanced Aircraft and Life Extension for Current Fleets – Lessons Learned in 50 Years after the Comet Accidents, Proceedings of the 23rd ICAF Symposium, vol. I, pp. 189–195, Dalle Donne, C. (Ed.), DGLR-Bericht. [19] Tiffany, C. (1977). In: Fatigue Life of Structures under Operational Loads, Proceedings of the 9th ICAF Symposium, pp. 4.4/1-4.4/31, Buxbaum, O. and Schutz, D. (Eds.), Fraunhofer-Gesellschaft. [20] Mil-A-83444 (USAF), Airplane Damage Tolerance Requirements, July 1974. [21] Wood, H., (1975), Eng. Fract. Mech., Vol. 7, pp. 557-564, Pergamon Press. [22] Swift, T. (1994). In: Fatigue, vol. 16 n.1, pp. 75–94, ButterworthHeinemann Ltd., Oxford. [23] Rudd, J. and Gray, T. (1978) Journal of Aircraft, vol. 15, n. 3, pp. 143-147. [24] MSG-3, Operator/Manufacturer Scheduled Maintenance Development, Revision 2007.1, Air Transport Association of America, Inc. [25] Attachment ‘A’ to letter from A. Kasowski (General Structures HWG Chairperson) to C. Bolt (Assistant Chair, TAEIG), General Structures Harmonization Working Group Report – Damage Tolerance and Fatigue Evaluation of Structures FAR/JAR §25.571, July 2, 2003.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

COMPILATION OF DAMAGE FINDINGS FROM MULTIPLE RECENT TEARDOWN ANALYSIS PROGRAMS Dr. Gregory A. Shoales, P.E., Dr. Scott A. Fawaz, P.E. and Molly R. Walters1 Center for Aircraft Structural Life Extension HQ USAFA/DFEM 2354 Fairchild Dr., Ste. 2J2A USAF Academy, CO 80840-6240 USA

Abstract: Fleet managers depend upon various tools to assure the safe operation of their aircraft. Performing a structural teardown analysis program of one or more aircraft with known service history provides precise damage data resulting from a given usage. These data can then be used to validate NDI methods, update damage models and reduce uncertainty in the damage condition assessment of the remaining fleet. In the past six years the USAF Academy Center for Aircraft Structural Life Extension (CAStLE) has been involved, at various levels, in multiple teardown analysis programs. CAStLE’s level of involvement has ranged from consultant through failure analysis support to planning and executing all elements of an entire program. In each program large regions of selected aircraft were disassembled by precision means, cleaned of all coatings and inspected by a variety of NDI techniques. These NDI indications were then evaluated by failure analysis methods to determine the root cause of failure—the type of damage which resulted in the NDI indication. Failure analysis findings which were a result of aircraft operations were further analyzed to obtain detailed damage characteristics. This work presents a summary of findings from teardown programs conducted on eight aircraft ranging from small military M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 187–207. © US Government 2009. Created within the capacity of an US governmental employment and therefore public domain. Published by Springer Science+Business Media B.V. Dordrecht.

188

Gregory A. Shoales, Scott A. Fawaz and Molly R. Walters

trainers to large transports. Comparisons are presented for damage types and categories of aircraft. Damage type and scale is also presented categorized by the type of damaged component (skin, stiffener, fitting, etc.). Data are also provided with regard to measurable features within regions of damage initiation.

AIRCRAFT STRUCTURAL TEARDOWN ANALYSIS PROGRAMS To better understand the teardown program findings presented in this work it is important to understand the motivation and character of the programs which produced these data. Destructive analysis of aircraft structure by what is commonly referred to as a “teardown” has long been considered a requirement in satisfying many of the sustainment issues that arise in aging aircraft [1]. As both military and civilian fleets continue to age, the number of teardown programs continues to increase. In the United States Air Force (USAF) alone there have been more than a dozen such programs in the past 8 years. In fact, the unique methodologies employed in teardown programs prompted the USAF Aging Aircraft Program Office in 2007 to task The USAF Academy Center for Aircraft Structural Life Extension (CAStLE) to prepare a best practices guide for conducting such programs. This handbook was published in 2008 [2] and its development is the subject of an additional publication in this [3] and other conferences [4]. As detailed in the best practices handbook, all teardown programs share specialized tasks which are tailored to the unique program requirements. It is therefore useful here to present the most common teardown program goals/requirements and a summary of teardown program tasks used to satisfy those requirements. Common Teardown Analysis Program Goals USAF Military Standard [1] requires execution of a teardown program to assess the damage state of an aircraft after a known period of usage. The same standard also suggests using such a program to assess and potentially revise an aircraft’s damage prediction models. The United States Federal Aviation Administration (FAA) and Delta Airlines conducted a teardown of a retired Boeing B727 in order to assess the damage state of high time aircraft fuselage structure [5]. The stated goal of a USAF program conducted 2004 and 2005 on a retired C-5A was to determine damage state and use this data to validate and adjust, if required, the aircraft’s structural models [6]. In such programs the damage mechanisms and damage morphology is compared with applicable predictive models. In most cases predictive models only exist for fatigue crack growth and therefore model validation is limited to findings of this type. In any event it is important to characterize the damage mechanism, all relevant dimensions and the location of each damage finding.

Compilation of damage findings from multiple recent teardown analysis programs

189

A further and somewhat related teardown program goal is to assist in the validation of the non-destructive inspection (NDI) methods employed on operational fleets. NDI techniques are used the world over to ascertain indications of possible damage in operational aircraft while minimizing the impact to readiness of those aircraft. The desire is that NDI indications can be correlated to a damage type and scale in order to support fleet management decisions. Unfortunately, the only possible means to obtain an actual damage finding to perform such correlation is by the destructive means ordinarily referred to as failure analysis. Therefore, conducting typical operational NDI prior to a teardown analysis affords a unique opportunity to correlate indications of damage with the corresponding failure analysis results. As shall be shown in the next section, NDI techniques are further used to support the teardown tasks themselves. Teardown Analysis Program Tasks As outlined in the aforementioned teardown program best practices handbook [2], all teardown programs share a common set of tasks. While many seem outwardly similar to standard aircraft maintenance operations, the unique requirements of the teardown program drive similarly unique requirements to each task. These unique requirements are descried in great detail in the teardown handbook and are briefly summarized here. Define Program Goals and Requirements. The first and foremost task in any teardown program is to define the goal and therefore the requirements of that program. A properly defined teardown program goal subsequently defines the requirements for aircraft type, number of aircraft subjects, subject structure within each aircraft, desired damage types and the characterization data required of each damage type from failure analysis. These requirements serve as the guide for all tasks which follow. Prepare the Teardown Program Subjects for NDI. Given a defined goal which in some way requires the determination of damage findings in the teardown subject structure, fiscal and schedule constraints dictate tasks which facilitate the efficient focusing of program resources. Since it is not practical to apply destructive failure analysis methods to every element of structure in a given aircraft or aircraft component, the next tasks are directed at focusing failure analysis to the most likely damage sites. There can be an increasing degree of fidelity uncertainty when applying NDI methods to increasingly complex assembled structures. However, when inspections are performed on properly prepared disassembled parts, indications of damage are far more confidently obtained. The following teardown program tasks effectively “stack the deck” in favor of the employed NDI technique giving teardown program analysts the most reliable data. After the focuses of the teardown program or its “subjects” are identified, the first task to prepare those subjects for NDI is to extract a region of assembled structure which encompasses each subject. By separating these assemblies from the parent

190

Gregory A. Shoales, Scott A. Fawaz and Molly R. Walters

structure, the probability of inducing additional damage to the subjects is minimized. The assembled structure is then disassembled by precisions means which further prevents inducing damage, particularly to the fastener holes, as a result of the disassembly process. Disassembly in support of teardown differs somewhat from that of normal maintenance operations. While it might not be desirable to damage a fastener hole during maintenance operations it is also common practice to oversize and refinish the hole during reassembly. Therefore minor damage induced during maintenance disassembly might easily, and rightly so, be considered inconsequential. On the other hand, any damage induced during teardown disassembly is wholly unacceptable as such damage may at worst destroy desired damage data and at the least result in needless analysis of the NDI indication which might result from the induced damage. After the assembled structure is fully separated into component parts, all coatings must be removed. Cleaning the metallic surfaces of all coatings not only enhances probability of damage indications detection for most NDI techniques, it is required for techniques such as florescent penetrant inspection (FPI). Here too, coating removal in support of teardown differs from that done in support of maintenance. Aircraft structure is commonly stripped of coatings to aid depot maintenance and inspections. At the completion of the depot maintenance program the structure is repainted. A minor etching of a metal part surface prior to painting is not necessarily detrimental and indeed is frequently highly desirable to enhance the quality of the repaint operation. In a teardown, however, any removal of the part surface risks the loss of critical program failure analysis data. A surface etch which removes only a few microns can destroy damage nucleating features, fatigue crack striations and further limit the complete characterization of damage. Like all teardown program tasks the level of sensitivity to surface damage depends upon the damage characterization goals of the teardown program. Nondestructively Inspect Subject Parts. Depending upon the program requirements each teardown subject may be inspected by a variety of NDI techniques. Ordinarily the first techniques applied are visual in nature. The visual techniques range from a macroscopic un-aided eye inspection to various techniques designed to enhance visual detection such as magnified visual, FPI and magnetic penetrant inspection (MPI). Requirements to find small scale damage usually drives program NDI requirements to employ eddy current and even various ultrasonic techniques. NDI data are derived from either the exposed surface features or electronic signals which may indicate damage or other anomalies in the inspected regions. Failure Analysis. All NDI indications are potential sites of damage which may be of interest to the aircraft maintainers and fleet managers and more importantly satisfy the program requirements. Accordingly, each site is considered a possible candidate for failure analysis evaluation. The first objective of failure analysis is to determine the root cause of the NDI indication. If the cause is from damage

Compilation of damage findings from multiple recent teardown analysis programs

191

morphology which is of interest to the program goal then the remaining failure analysis objective is to fully characterize the damage. It follows that such characterization includes all parameters required by the program goal.

OVERVIEW OF SUBJECT AIRCRAFT CATAGORIES In ICAF 2007 a summary of results from the C-130 teardown program was presented by this author. The present work addresses the culmination of all failure analysis findings from eight subject aircraft. In all, the teardown programs conducted on these eight aircraft resulted in 711 failure analysis investigations. These aircraft may be described by three categories and as such shall be referenced to these categories as their data is presented. After the description of each category the goal or focus of each teardown program within that category is also presented. This program focus is presented to help better understand the data presented herein. Aircraft A: Small trainer/attack class aircraft. Teardown programs conducted on this category of aircraft evaluated multiple wing sets in complete detail. The subjects of these programs included all structural elements of the wing and wing to fuselage attach structure. Additional programs were conducted which focused on all structural elements which, by fleet management practices, were considered fatigue critical. All aircraft evaluated had reached the end of their service life. CAStLE performed all failure analysis for these programs. Aircraft B: Medium scale transport aircraft. Teardown activities conducted on this category of aircraft focused on the center wing structure. This category included a single aircraft type. The center wing structure is considered the most critical structural element of this aircraft type and is the primary element used to determine aircraft life. While not established at program inception, according to operational limits (in equivalent flight hours) set by fleet management during the teardown program’s execution, this center wing had exceeded its allowable life. CAStLE had oversight on all failure analysis for this program and performed 75% of all analyses. Aircraft C: Large scale transport aircraft. This category also included a single aircraft type. The primary focus of this teardown program was fatigue critical structure throughout the aircraft. While some corrosion damage was characterized, the damage morphology of greatest interest to the program was fatigue cracking. This aircraft was one of the first production aircraft of its type and had been retired by the using command. CAStLE performed 55 failure analyses for this program.

192

Gregory A. Shoales, Scott A. Fawaz and Molly R. Walters

FAILURE ANALYSIS FINDINGS These data shall be presented as a grouped summary of the failure analysis from all subject programs. Where appropriate, each data set shall also be presented by aircraft, damage, or other relevant category. Damage Finding Type and Correlation to NDI Indications As previously stated, the first goal of failure analysis is to determine the cause of a given NDI indication. Whereas failure analysis is a time consuming and relatively costly portion of any teardown program, assignment of an NDI indication to failure analysis is usually preceded by an analysis of the NDI indication data. Accordingly, the NDI indications which are referred to in depth analysis by failure analysis methods are generally those which have the highest confidence of coinciding with actual damage. Depending on the schedule and fiscal resources available, each program discussed in this work employed some system to evaluate and prioritize NDI indications for failure analysis. For example, NDI indications of surface anomalies should be further examined under high magnification before forwarding them to failure analysis. This step helps to differentiate between defects such as surface-breaking cracks which are of keen interest to aircraft managers and the more superficial defects such as light surface scratches which would be of less interest to the same group. Similarly, NDI indication from eddy current techniques are most useful if the strength of the signal is captured with the indication data. A common metric here is, after the eddy current device has been calibrated to a relevant standard, to record the strength of a given indication relative to this standard. This relative measurement is reported as a percentage of full screen height or %FSH. Prioritization systems may then give first priority to indication with higher %FSH. Conversely, eddy current indications less than 20-30% FSH are usually not evaluated further. It follows from the preceding discussion that a first priority examination of failure analysis data might be to evaluate the findings relative to the NDI indications that led to the failure analysis requirement being levied. Performing such an analysis allows the investigator to assess the criteria used to evaluate and prioritize NDI indications. All failure analysis findings presented here were the result of a NDI indication prioritization scheme whose goal was the evaluation of the most likely sites of relevant damage. The first comparison between findings and NDI indications is qualitative and shows the overall damage finding types resulting from all 711 failure analysis evaluations. Figure 1 shows the number of findings in each type while Figure 2 shows the same information but broken out by teardown program aircraft category. As is evident in both figures, despite careful consideration of NDI indication data during the prioritization phase, no damage was found at just over 100 indication sites. In all cases of a “no damage” finding, the CAStLE failure analysis practices [7] ensured that there was no damage larger than 0.5 mm at the indication site.

Compilation of damage findings from multiple recent teardown analysis programs

193

These practices have been further documented in a failure analysis protocol for the C/KC-135 teardown program [8]. The development and validation of all C/KC-135 teardown program protocols have been previously presented elsewhere [9]. Referencing the damage types listed in these figures it is useful to distinguish between findings from operational usage damage and findings which are not related to operational usage. In the later case, strong NDI indications often result from mechanical damage such as deep gouges in the bore of a fastener hole. Even if a mechanical gouge is confirmed through the initial visual examination of failure analysis the investigation must be completed in depth. The reasoning here is that this sort of damage can serve form a highly localized stress concentration and in fact is often analyzed as part of the initial discontinuity state or IDS [10]. A further example of a non-operational usage related finding is a material defect. The four material defects findings shown in Figure 1 are from either material processing pores or interstitial particles which were aligned in the radial hole bore direction— thus mimicking a pattern followed by a crack. The last non-operational damage type applies to a damage finding which, after analysis, was determined to have been caused by loads not related to normal aircraft usage. Most of this damage type was from the two category A aircraft which had sustained damage while in storage and prior to being selected for the teardown analysis program. While a detailed analysis of each finding is presented in the applicable teardown program report, a brief explanation of those damage types classified as resulting from usage is warranted. The “bore corrosion” type refers to corrosion pitting which was found on the bore of a fastener hole. This type of damage was of keen interest to several programs and as such has been separated from other surface corrosion. The “IGC” type is intergranular corrosion. This is distinguished from the “exfoliation” finding category where the IGC progressed in a laminar fashion such that layers of material exfoliated. In-plane cracks again represented a unique type found throughout the category A aircraft programs and were observed in the parts originating from the bore walls and propagating parallel to the part surface. However, due to their extremely small size, opening was not practical. Therefore, specific damage mechanisms could be positively identified for these cracks. To be conservative, findings which could not be positively associated with any damage mechanism and therefore listed as “unknown” were considered to be from operational usage. Corrosion-fatigue, stress corrosion cracking (SCC), fatigue and overstress follow accepted definitions of those damage types [11].

Gregory A. Shoales, Scott A. Fawaz and Molly R. Walters

1000

100

10

Damage findings resulting from operational usage

Non-operational Damage

No Damage

Mechanical Damage

Material Defect

Unknown

Overstress

Fatigue

SCC

In-plane Cracks

Corrosion-Fatigue

IGC

Exfoiliation

Corrosion

1 Bore Corrosion

Number of Findings _

194

Finding Type

Figure 1: Damage finding types for all teardown analysis programs.

For clarity of presentation the finding types shown in Figure 1 have been group according to their source for Figure 2. All corrosion mechanisms are therefore grouped under the “environmental” damage source. As given in Figure 1, these include bore corrosion, corrosion, exfoliation and IGC. Damage which stems from environmental factors but also has a stress component includes in-plane cracks and SCC. Damage types from stress alone include fatigue and overstress. The nonoperational sources include the previously described non-operational damage along with mechanical damage. Failing any definitive source, the unknown damage type is retained separately in Figure 2.

195

Unknown

No Damage

Nonoperational

Stress

Category A Category B Category C

Environment & Stress

45% 40% 35% 30% 25% 20% 15% 10% 5% 0%

Environmental

Percentage for Each Category __

Compilation of damage findings from multiple recent teardown analysis programs

Primary Damage Source

Figure 2: Damage finding types for all teardown programs broken out by aircraft category.

The next comparison between findings and NDI indications is quantitative. In this comparison the data set is limited to fatigue cracks which were found at the site of eddy current indications. Figure 3 shows maximum damage dimension compared to eddy current %FSH. For this comparison the maximum dimension of the confirmed fatigue crack, either in the thickness direction or the radial direction, is taken as the maximum damage dimension. As is evident from this figure, the strongest indication (highest %FSH) does not necessarily directly correlate to largest damage dimension. As may be expected however, the largest number of confirmed cracks occur with stronger (higher %FSH) eddy current signal. Although it is a somewhat weak correlation in the mid %FSH range, this graph shows a trend of confirmed crack size decreasing with decreasing %FSH. The category B aircraft teardown programs used 20% FSH as the prioritization cut-off. Even so, only one crack of less than 0.4 mm was confirmed below 30% FSH.

Gregory A. Shoales, Scott A. Fawaz and Molly R. Walters

Maximum Damage Dimension (mm) _

196

100

10

1

0.1

0.01 0

20 40 60 80 Relative BHEC Indication Strength (%FSH)

100

Figure 3: Maximum fatigue crack damage dimensions as a function of eddy current indication in %FSH.

Damage Scale In the previous comparison the maximum fatigue crack dimension found during failure analyses evaluation was compared with the strength of the eddy current indication which led to that evaluation. It follows that this comparison was only made for those findings which could be correlated to eddy current NDI indication locations for which corresponding %FSH data was available. The data presented in this section is also the maximum damage scale but includes all findings of operational usage crack damage regardless of the available NDI indication data. As in the previous comparisons the maximum dimension determined during the damage evaluation is used in the presentations which follow. Figure 4 presents a histogram of the maximum damage dimension for all damage which was judged to be the result of operational usage, either environmental or stress, from all teardown programs. Figure 5 presents the same data broken out by the same damage source categories used in Figure 2. Lastly, Figure 6 presents these data separated out by aircraft category (A, B or C).

Number of Findings _

Compilation of damage findings from multiple recent teardown analysis programs

197

80 60 40 20 0 < 0.6

< 1.27 to 0.6 < 3 to 1.27 Maximum Damage Dimension (mm)

>3

Figure 4: Histogram of the maximum damage dimension for all operational usage crack damage characterized in the subject teardown programs.

Number of Findings_

35 30 25

Environment Env& Stress Stress Unknown

20 15 10 5 0 < 0.6

< 1.27 to 0.6 < 3 to 1.27 Maximum Damage Dimension (mm)

>3

Figure 5: Histogram of maximum damage dimension for all operational usage damage cracks characterized in the subject teardown programs broken out by primary damage source.

Gregory A. Shoales, Scott A. Fawaz and Molly R. Walters

Number of Findings _

198

40 35 30 25 20 15 10 5 0

Category A Category B Category C

< 0.6

< 1.27 to 0.6 < 3 to 1.27 Maximum Damage Dimension (mm)

>3

Figure 6: Histogram of maximum damage dimension for all operational usage damage cracks characterized in the subject teardown programs broken out by aircraft category.

The preceding 3 figures highlights how little damage was found in any of eight retired aircraft which exceeds the 1.27 mm flaw size used in most damage tolerance analysis. Of 711 findings, 119 (less than 17%) are operational damage with a dimension of 1.27 mm or greater. Referencing Figure 5, damage due to stress dominates the findings of all aircraft while damage which combines the affect of stress and environment were found at the higher end of the dimension range. A further presentation of findings focuses on regions of surface corrosion damage. The characteristic from such damage which are normally of greatest interest to fleet managers is the amount of material lost. The surface corrosion damage data from all teardowns are summarized in the scatter diagram of Figure 7. Each data point plotted in this diagram represents a single evaluated site of surface corrosion. The maximum thickness lost expressed in percentage of thickness of the affected part is plotted against the total surface area of the site evaluated.

Surface Area (sq mm)

Compilation of damage findings from multiple recent teardown analysis programs

199

10000 1000 100 10 1 0.1 0

10

20

30 40 % Thickness Loss

50

60

Figure 7: Percentage of maximum part thickness lost due to surface corrosion damage plotted against the total surface area affected at the evaluated site.

Damage Initiation Features Also of keen interest to the aircraft sustainment engineers and researchers is the feature or features which initiate the damage mechanism. While fractographic analysis of laboratory specimens to identify such characteristics may be somewhat routine, obtaining such data from operationally damaged surfaces is frequently problematic. For example, fatigue cracks which occur in service may be subject to compressive stresses which can smear and otherwise damage the fracture surface. Sufficient damage to the fracture surface can cause a loss of damage morphology data such as the identification and measurement of the initiation feature. The fracture surface of an open crack may also be damaged by corrosion—resulting in deposition of corrosion product and loss of the substrate material. While proven methods exist to remove corrosion product from the facture surface without damaging that surface [8], substrate material lost due to in-service corrosion causes a corresponding loss of all data contained on that portion of the fracture surface. Even so, every effort was made to measure the region of damage initiation and identify the feature(s) within that zone. The damage initiation zone is defined as the region between the smallest resolvable fatigue striation and the specimen edges. Observed corrosion pits or other features within this zone were recorded as such. The authors recognize that within this damage initiation zone, as defined, there could be very slow fatigue crack growth but no visible fatigue striations. Table I presents a summary of all identified damage initiation features from all three teardown aircraft categories. Measured dimension distribution parameters for each damage initiation zone and the percentage of initiation zones located on part joining or faying surface are also given in this table.

Gregory A. Shoales, Scott A. Fawaz and Molly R. Walters

200

Table I: Summary of all identifiable damage initiation features, initiation zone dimension distribution parameters and faying surface location percentage of these zones.

Initiation Feature Corrosion Pit Mechanical Damage

% 80% 20%

Dimensions (mm) Minimum Maximum Average 0.022 0.624 0.135 0.040 0.326 0.156

Percentage of Initiation Sites on Faying Surface

31%

Number of Occurances

The study of damage initiating feature is also relevant to structural teardowns in that it in part drives the requirement to evaluate sites where nothing can be identified but mechanical damage. As discussed previously, mechanical damage such as drill marks, gouges and deep scratches potentially represent stress concentration features. A long held view is that such stress risers will nucleate fatigue cracking and other continuing damage mechanisms. Mechanical damage indications are therefore frequently prioritized for failure analysis to ascertain whether further damage exists. While Table 1 indicates that 20% (a total of 15) of all identifiable initiations sites were from mechanical damage it is worth recollecting that 216 of findings of 711 total (or 30%) evaluations were mechanical damage which did not nucleate any continuing damage. Figure 8 illustrates the distribution of initiation feature size along with the number of occurrences of each. As shown here, mechanical damage features were found only within the lower size range of the corrosion pits feature.

Corrosion Pits Mechanical Damage

0

0.1

0.2

0.3 0.4 0.5 Maximum Dimnesion (mm)

0.6

0.7

0.8

Figure 8: Distributions of measured initiation features by feature type. Damage Location and Proximity to Other Damage The last presentation of teardown findings to be made concerns the location of damage. The first area of interest is the component type where relevant damage

Compilation of damage findings from multiple recent teardown analysis programs

201

was found. The next area of concern is the accumulation of damage in close proximity to other damage. Damaged component. For the purposes of this work the structural components are divided into five categories. Those are skin, skin stiffeners, rib cap, spar cap and fitting. Skins webs are simply the planar parts which generally cover the outer surfaces of an aircraft but also included webs. Stiffeners are linear elements which are mechanically fastened to skins and serve to increase their stiffness. Caps and fittings on the other hand are typically components of three dimensions which serve the function of transferring load, often times between multiple elements. A summary of damage types broken out by component category is given in Table II. Fatigue cracking seems to be slightly more prevalent in skin and spar caps but is otherwise evenly distributed. There is a predominance of corrosion findings in spar caps. This result was predominately from extensive spar corrosion found in the category A aircraft. The in-plane cracks were exclusive to the category A aircraft and were thought to be in part due to a somewhat unique material specification in that structure.

Component Type

Corrosion

CorrosionFatigue

SCC

In-Plane Cracks

Fatigue

Overstress

Unknown

Table II: Number of finding of a particular damage type broken out by structural component category.

Skin & Webs

3

1

1

0

1

0

1

Skin Stiffenner

0

1

0

0

1

0

3

Rib Cap Spar Cap Fitting

2 2 1

0 0 1

0 1 1

0 1 1

1 1 1

1 2 1

0 1 1

Proximity between damage findings. When assessing structural risk, the proximity of damage to other damage, regardless of the size, frequently has equal emphasis as the damage morphology of individual findings. This emphasis was a result of concerns raised by fleet managers about widespread fatigue damage (WFD). WFD is characterized by the simultaneous presence of cracks at multiple structural details that are of sufficient size and density whereby the structure will no longer meet its damage tolerance requirements; for example, not maintaining required residual strength after partial structural failure [12]. Indications which, after evaluation reveal multiple fatigue cracks along a common wing station in a single wing panel are an example of multi-site damage (MSD). Similarly, multiple fatigue crack findings in common holes of joined structural elements are an example of multi-element damage (MED). WFD, MSD and MED are currently prime structural concerns in many aircraft fleets.

Gregory A. Shoales, Scott A. Fawaz and Molly R. Walters

202

aft

Fuselage Station _

It is for this reason that post teardown analysis must consider the proximity of damage finding locations. Given the wide variation of aircraft category scale and structural configurations this comparison is only useful between teardown findings from the same aircraft type. The lower center wing of teardown aircraft category B at the usage level of the subject aircraft was thought to be prone to widespread fatigue damage. Figure 9 illustrates the location of all damage findings after a complete inspection of that structure. Left and right side damage are distinguished by symbols. While the numeric scale has been omitted for release purposes, the vertical and horizontal scales are shown at the same relative size to one another. The damage grouping at the aft inboard location corresponds to the wing life tracking point used by most operators of this aircraft. The grouping along the entire outboard wing station represents the outboard attachment which is currently undergoing a redesign due to damage propensity in this location.

Left Side Right Side

outboard Wing Station

Figure 9: Wing station (WS) and fuselage station (FS) location of all operational usage damage findings in the center wing of the category B aircraft teardown program. The data in Figure 9 represents the findings from a single teardown aircraft. As with any attempt to represent a population (such as an aircraft fleet) with a sample (such as teardown aircraft) it is far more meaningful to combine the findings from multiple aircraft subjects. The wing damage findings from the six category A teardown aircraft are combined in Figure 10. Like Figure 9 this figure shows the locations of all operational usage damage findings but, due to the larger sample size, with arguably far more statistical significance. Here again, left and right side damage is distinguished by symbols and the vertical and horizontal scales are shown at the same relative size to one another. Clearly evident in this figure is the predominance of damage on both spars and the aft spar in particular. A concentration of damage at an approximately mid-span rib is also evident.

aft

Fuselage Station

Compilation of damage findings from multiple recent teardown analysis programs

203

Left Side Right Side

outboard

Wing Station

Figure 10: Wing station (WS) and fuselage station (FS) location of all operational usage damage findings in the wing of all category A aircraft teardown programs. Both of the preceding examples demonstrate how the morphology of individual damage findings must be considered in concert with the proximity to other damage.

DISCUSSION AND CONCLUSIONS The findings presented here represent the wide variety of damage morphologies that may be expected when conducting a teardown analysis program of aged aircraft structure. Given the specialized inspection of structure required by the goals of most teardown programs, the scale of damage findings should also be expected to be far smaller than can be detected by in service inspections. Even so, without careful application of NDI procedures teardown program resources may be unnecessarily taxed. Recalling Figure 2, the number of NDI indication failure analysis evaluations which resulted a “no damage” finding varied greatly by teardown aircraft category. The most obvious difference is the comparatively small percentage of these findings for the category A aircraft programs. This result is not entirely by coincidence. In fact, the large number of these findings in the earlier occurring category B and C aircraft teardown programs served as a lesson learned for future programs. NDI process improvements were published in 2008 by Air Force Research Laboratory [13]. A draft version of these new best practices were applied to the category A teardown programs. The benefit of reducing the number of needless evaluations and thus controlling program cost and schedule is made clear by this comparison. A further NDI observation comes from Figure 3. The maximum damage dimensions at each value of %FSH are directly proportional to

204

Gregory A. Shoales, Scott A. Fawaz and Molly R. Walters

%FSH. This result supports the practice defining some lower bound to these indications when prioritizing fro failure analysis. Caution is advised however, to carefully consider the program’s damage fidelity requirements before defining an overly aggressive cut-off value just to save cost—lest valuable data may never be evaluated. As already stated, the comparisons of damage scale presented in Figures 4-6 show how few findings from any of the eight programs even exceed the assumed damage tolerant flaw size (1.27 mm quarter circular corner crack at a hole). Also observable here is that stress is the primary source of damage. It may be thought that this conclusion is skewed by the fatigue damage emphasis on the category B and C aircraft programs. Failure analysis of the six category A aircraft however, accounted for 56% of all evaluations performed. If the aforementioned “no damage” findings are removed this number increases to 66%. Furthermore, despite the fatigue emphasis on the category C aircraft teardown, corrosion related damage findings accounted for 2/3 of all CAStLE evaluations. The corrosion damage depiction of Figure 7 shows what appears to be a predominantly mild attack to the evaluated structure. This is based upon the fact that most repair manuals permit surface corrosion grind-out in excess of the depth shown. The permissible loss of material is indicative of design margins in the part sizing. If these sites were left unchecked the pitting corrosion could possibly be followed by more aggressive corrosion mechanisms such as exfoliation. Exfoliation could easily result in unacceptable levels of material loss. Such likelihood may not seem relevant in aircraft that had reached their service limit. The danger to safety comes from uninspectable corrosion sites in the presence of a more aggressive environment or an earlier onset to corrosion damage. All corrosion sites, regardless of size must be fully addressed by the fleet’s corrosion prevention and control program. The most critical sites should be considered for additional inspections in the remaining fleet. The next characteristics presented in this work were the sites of damage initiation and the identifiable features within them. The majority of initiation features observed as the source of damage are corrosion pits. Brooks Peeler, Honeycutt, and Prost-Domasky made this observation when forming their holistic life prediction models [14]. In their analysis they state that damage begins with corrosive pitting attacks at material discontinuities. This was the experience of some of the analysis performed in the subject teardowns. Many more of the pits formed when protective coatings broke down and permitted the contact of dissimilar metals. In one example the analysis revealed that a steel fastener had lost a portion of its cadmium coating, either in service or during an aggressive installation process, allowing contact with the aluminum part. In some installations these fasteners are further protected by installing them with sealant, commonly referred to as a “wet” installation. Analysis of these installations where pitting corrosion formed anyway revealed that the sealant had become brittle with age and subsequent cracking

Compilation of damage findings from multiple recent teardown analysis programs

205

Probability of Occurrence

permitted the pitting corrosion to take place. A further example is where the failure analysis showed evidence that a part edge was machined, most likely to accommodate installation, thus eliminating the protective aluminum clad layer (Alclad). Whatever the source, a discontinuity system was established and corrosion pits formed on the part surface. After pits form the next phase in their proposed holistic life modeling is that cyclic loading at the corrosion pits results in fatigue crack nucleation and growth. This observation is also consistent with the findings presented in this work. Of all corrosion pits identified as the source of damage, 78% led to continuing damage by fatigue mechanisms. Furthermore, of the mechanical damage features identified in initiation sites, 93% led to fatigue damage mechanisms. In the previously cited work of Brooks and company [10], the relative scale of various initial discontinuity states are compared to one another. Figure 11 is reprinted from the presentation of that work at the 1999 ASIP Conference. Manufacturing defects in this figure equate to what is called the mechanical damage in Figure 8. The distributions of mechanical and corrosion IDS compare favorably between both figures.

As-Built

Bulk Material Quality

As-Is

To-Be

Too Late!

Predicted Corrosion range of flaws at and Corrosion future Fatigue and/or Fatigue depot intervals

Manufacturing Defects

Flaw Size Figure 11: Comparison of various IDS dimensions from C. Brooks 1999 ASIP Conference presentation titled “Correlation of Life Prediction Methods with Corrosion-Related Tests” The last set of findings presented were that of damage finding location. These data are most useful to fleet managers when considered in whole. For example, a majority of corrosion findings in one component type versus another may illuminate a problem in the protective strategies. Any grouping of findings would of course be compared with predictive models to ensure damage is being modeled at those locations. Recalling the previous discussion of WFD, MSD and MED, a complete analysis of teardown program findings requires equal consideration of damage type, damage size, affected component(s), aircraft coordinates and of course the damage density. Significant findings in unexpected locations require that predictive models be enhanced to include these new areas.

206

Gregory A. Shoales, Scott A. Fawaz and Molly R. Walters

One goal suggested in the introduction of this work is the assessment of fielded inspection programs. To answer this requirement the precise tracking of each finding’s location back to its original location is essential. It is only by doing so that indications from the fielded inspection may be analyzed with respect to the damage findings. The data and comparisons presented here, rather than addressing the goal of any specific program, were aimed at presenting a wide variety of possible analysis that may be performed on teardown findings. It is offered to show how findings from diverse aircraft types with differing program goals can both be different and similar. For any one teardown analysis program it is of course essential that the analysis performed on the findings satisfy the stated program goals. It therefore follows that clear program goal(s) identification must always be the first step to program planning. This step is then to be followed by the definition of comprehensive data requirements to answer that goal. Having done so, the best way to ensure program validity is to use the defined program data requirements to guide each and every teardown program task.

REFERENCES [1] MIL-STD-1530C, DoD Standard Practice, Aircraft Structural Integrity Program (ASIP) (2005), Aeronautical Systems Center, Wright-Patterson AFB, Ohio. [2] Shoales, G. (2008), Procedures for Aircraft Structural Teardown Analysis, USAFA TR-2008-02, USAF Academy, Colorado. [3] Shoales, G. (2009), Procedures for Aircraft Structural Teardown Analysis: Development of a Best Practices Handbook, Proceedings of the 25th International Conference on Aeronautical Fatigue Symposium, Rotterdam, Netherlands. [4] Shoales, G. (2008), Procedures for Aircraft Structural Teardown: Best Practices Handbook Development, 2008 Aircraft Structural Integrity Program Conference, San Antonio, Texas. [5] Steadman, D. and Bakuckas, Jr., J., (2007), DOT/FAA/AR-07/22, Destructive Evaluation and Extended Fatigue Testing of Retired Transport Aircraft, U.S. Department of Transportation, Federal Aviation Administration, Washington, D.C. [6] Final Report, C-5A Structural Risk and Model Revalidation Program, (2007), 730th Aircraft Sustainment Group, Robins AFB Georgia.

Compilation of damage findings from multiple recent teardown analysis programs

207

[7] Shoales, G., Shah, S., Rausch, J., Walters, M., Arunachalam S., and Hammond, M. (2006), C-130 Center Wing Box Structural Teardown Analysis Final Report, USAFA TR-2006-11, USAF Academy, Colorado. [8] Arunachalam S. and Shoales, G. (2009), C/KC-135 Teardown Analysis Program Protocols, Protocol 8: Failure Analysis, USAFA TR-2009-06, USAF Academy, Colorado. [9] Shoales, G. (2009), C/KC-135 Teardown Analysis Program Protocol Development and Validation,” Aging Aircraft 2009 Conference, Kansas City, Kansas. [10] Brooks, C. Honeycutt, K. and Prost-Domasky S. (2000), Case Studies for Life Assessments With Age Degradation, Aging Aircraft 2000 Conference, St. Louis, Missouri. [11] Powell, G., (Ed.) (1986), ASM Handbook, Volume 11: Failure Analysis and Prevention, ASM International, Materials Park, Ohio. [12] Swift, T. (1993), Widespread Fatigue Damage Monitoring –Issues and Concerns, Proceedings of the 5th International Conference on Aging Aircraft, Hamburg, Germany. [13] T.O. 33B-1-2, Nondestructive Inspection General Procedures and Process Controls (2008), Tinker AFB, Oklahoma. [14] Brooks, C., Peeler, D., Honeycutt, T. and Prost-Domasky, S. (1999), Predictive Modeling for Corrosion Management: Modeling Fundamentals, Aging Aircraft 1999 Conference, Albuquerque, New Mexico.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

DETECTION OF ARRESTED CRACK FOR FOAM CORE SANDWICH STRUCTURES USING OPTICAL FIBER SENSORS EMBEDDED IN A CRACK ARRESTER Shu Minakuchi1, Ippei Yamauchi2 Nobuo Takeda1, and Yasuo Hirose3 1

Department of Advanced Energy, Graduate School of Frontier Sciences, The University of Tokyo, 5-1-5 Kashiwanoha, Kashiwa-shi, Chiba, 277-8561, Japan 2 Department of Aeronautics and Astronautics, School of Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8656, Japan 3 Commercial Aircraft Project Engineering Division, Aerospace Company, Kawasaki Heavy Industries, Ltd., 1 Kawasaki-cho, Kakamigahara-shi, Gifu, 504-8710, Japan

Abstract: A crack arrester has been recently developed to suppress crack propagation along the interface between a facesheet and a core in a foam core sandwich structure. The crack arrester has semi-cylindrical shape and is inserted into the interface. The crack arrester decreases an energy release rate at the crack tip by suppressing local deformation around the crack. If the arrested crack can be instantaneously detected, the reliabilities of the structures are dramatically improved. This study establishes an innovative crack detection technique using two fiber Bragg grating (FBG) sensors embedded at both edges of the arrester. The change of the strain distribution in the crack arrester induced by arresting the crack is evaluated using reflection spectra from the FBG sensors. The proposed technique enables an effective application of the crack arrester and significantly improves the reliability of the foam core sandwich structures.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 209–221. © Springer Science+Business Media B.V. 2009

Shu Minakuchi et al.

210

INTRODUCTION Carbon fiber reinforced plastic (CFRP) has been used for almost all modern commercial aircrafts as a primary structural material. However, the potential capability of CFRP cannot be maximized under the conventional structural design concept, consisting of skins, stringers and frames. One of the innovative structural concepts is a foam core sandwich panel structure [1-3]. The integral construction consists of two thin facesheets and a lightweight foam core, and can considerably reduce the weight and the number of parts compared to conventional structures. However, it has been pointed out that the crack propagation along the interface between the facesheet and the core is the critical issue. The interface crack originates from the impact damage or the fatigue shear cracks in the foam core, and seriously degrades the structural integrity. In this context, Hirose et al. developed a crack arrester (Fig. 1) [4]. The crack arrester has semi-cylindrical shape and is inserted into the interface. When the crack approaches the arrester, it decreases the energy release rate at the crack tip by suppressing local deformation around the crack. The suppression of the crack propagation has been evaluated under various loading conditions, confirming that the crack arrester can dramatically improve damage tolerance of the foam core sandwich structures. In view of practical use, the arrested crack must be instantaneously detected and appropriate measures need to be taken against the damaged area in order to maintain the structural reliability. However, the crack below the facesheet is difficult to detect using conventional non-destructive inspection techniques. This study establishes an innovative technique using FBG sensors embedded in the cocured semi-cylindrical crack arrester. We begin by proposing crack detection technique and a numerical analysis is then conducted to predict the change of reflection spectra from the FBG sensors due to the crack propagation. Finally, the technique is verified by experiments.

Fig. 1 Crack arrester.

Detection of arrested crack for foam core sandwich structures

211

CRACK DETECTION TECHNIQUE Figure 2 depicts a schematic of the crack detection technique. Hirose et al. [4] demonstrated that stress was induced at the crack side edge of the arrester when the crack approached the arrester. As a result, stress concentration at the tip of the crack was reduced, preventing the crack propagation. In this study, two FBG sensors are embedded at both side edges of the crack arrester where the strain increases as the crack approaches the arrester. The change of the strain distribution induced by suppressing the crack propagation is evaluated using reflection spectra from the FBG sensors. When the stress is distributed at the edge of the arrester, non-axisymmetric strain arises at core of the FBG sensor and the reflection spectrum from the sensor splits into two peaks due to a birefringence effect [5, 6]. The difference between the central wavelengths of the two peaks, λp and λq, is calculated using the following equation: n2λ (1) Δλ = λ p − λ q = 0 0 ( p12 − p11 ) ε1 − ε 2 2 where λ0 is the center wavelength of the initial reflection spectrum, n0 is the initial refractive index of the optical fiber core, p11 and p12 are the photoelastic constants, and ε1 and ε2 are the maximum and minimum principal strains in the crosssectional direction of the FBG sensor. This equation shows the difference between the center wavelengths of the two peaks is in proportion to the value of the difference between the maximum and minimum principal strains. Since the change in the magnitude of the stress and thus the principal strain is greater at the crack side edge than at the opposite edge, the crack propagation can be detected comparing reflection spectra from the two FBG sensors.

Fig. 2 Schematic of crack detection technique.

Shu Minakuchi et al.

212

In this study, Mode I and Mode II type cracks were evaluated. First, finite element analysis (FEA) was conducted on double cantilever beam (DCB) and end notch flexure (ENF) sandwich specimens to calculate the change in the difference between the maximum and minimum principal strains. Based on the calculated strain values, the reflection spectrum is simulated.

FINITE ELEMENT ANALYSIS Finite element model Figure 3 presents 2-D finite element models with cracks modeled in ABAQUS 6.6. Plane strain was assumed and thermal deformation was considered. Due to restriction of the experimental facilities, the DCB and ENF specimens had different material configurations. The DCB specimen consisted of CFRP facesheets (UT500/#135, Toho Tenax Co.,Ltd., curing temperature: 130 oC, [(+45, -45)/(0, 90)/(0, 90)/(+45, -45)], thickness: 1.68 mm), foam core (PMI Rohacell WF-110, Evouik Rohm GmbH, thickness: 35 mm) and semi-cylindrical crack arrester (UT500/#135, Toho Tenax Co.,Ltd., unidirectional prepreg, radius: 10 mm). The fiber direction of the arrester was perpendicular to the direction of the crack propagation. The distance between the crack tip and the crack arrester, L, was set as 10, 5, and 0 mm to investigate the influence of the crack propagation on the change in the principal strains. On the other hand, the ENF specimen consisted of CFRP facesheets (T700S/2500, Toray Industries, Inc., curing temperature: 130 o C, [0/90]5S, thickness: 2.5 mm), foam core (PMI Rohacell WF-110, Evouik Rohm GmbH, thickness:35 mm) and semi-cylindrical crack arrester (HC 9872 SynCore, Hysol Aerospace Products., radius: 10 mm). The value of L was set as 20, 10, and 0 mm. The load applied to each specimen was determined based on fracture load measured in the preliminary tests and 2.0N/mm and 60N/mm were applied to the DCB and ENF specimens, respectively. Two FBG sensors were embedded at both side edges of the arrester along the specimen width direction. “Sensor A” is the one embedded at the crack side and “Sensor B” is the one at the opposite side. Results The changes in the principal strains in the FBG sensors are presented in Fig. 4. In the DCB specimen (Fig. 4 (a)), as the crack approached the arrester, the difference between the maximum and minimum principal strains increased in the sensor at the crack side. Figure 5 illustrates a schematic of the sensor deformation obtained from the FEA. The facesheet tensile deformation due to the local bending induced the increase of the maximum principal strain (Fig. 5 (a)). In the sensor at the opposite side of the crack, however, both the maximum and minimum principal strains were almost constant. It is important to note that the strain induced in the sensor at the opposite side of the crack was not the mechanical strain, but the thermal residual strain. These results indicate that only the arrester edge near the crack contribute to the suppression of the crack propagation. In the ENF specimen, on the other hand,

Detection of arrested crack for foam core sandwich structures

(a) DCB specimen.

(b) ENF specimen. Fig. 3 Finite element models.

213

Shu Minakuchi et al.

214 1500

1500

Sensor A

1000 Principal Strain [με]

Sensor B

Arrester Principal Strain [με]

L 500

0

Maximum Principal Strain Minimum Principal Strain

-500

Arrester

L

1000

500

0

Maximum Principal Strain Minimum Principal Strain

-500

-1000

-1000

10

5

0 L [mm]

-5

-10

10

5

0 L [mm]

-5

-10

(a) DCB specimen. 750

Arrester

Sensor B

L

500 Principal Strain [με]

500 Principal Strain [με]

750

Arrester

Sensor A

L

250

Maximum Principal Strain Minimum Principal Strain

0

250

Maximum Principal Strain Minimum Principal Strain

0

-250

-250

-500

-500 20

15

10

5 L [mm]

0

-5

-10

20

15

10

5 L [mm]

0

-5

-10

(b) ENF specimen. Fig. 4 Principal strains in core depending on L. the crack arrester was compressed by the core (Fig. 5 (b)). As a result, the difference between the maximum and minimum principal strains again increased in the sensor at the crack side. The reflection spectra I(λ) were then calculated using Eq. (1) and the following simple equation: 1 1 (2) I (λ ) = I 0 (λ ) + I 0 (λ − Δλ ) , 2 2 where I0(λ) is the initial reflection spectrum and λ is the wavelength. Figure 6 gives the simulated spectra. The initial spectrum of the FBG sensors used in the verification tests (next section) were defined as I0(λ). The intensity of each spectrum was normalized by the intensity of the highest component in the initial

Detection of arrested crack for foam core sandwich structures

215

(a) DCB specimen.

(b) ENF specimen. Fig. 5 Schematics of sensor deformation demonstrated in FEA.

1

Before embedding

1.2

Sensor A Normalized Intensity

Normalized Intensity

1.2

0.8

10 mm

0.6

5 mm 0 mm

0.4 0.2 0 -0.4

-0.2

0

0.2

0.4

0.6

0.8

Sensor B

1 0.8 0.6

5 mm 0 mm

10 mm

0.4 0.2 0 -0.4

Wavelength Detuning [nm]

-0.2

0

0.2

0.4

0.6

Wavelength Detuning [nm]

(a) DCB specimen. Fig. 6 Simulated reflection spectra.

0.8

Shu Minakuchi et al.

216 1.2

1.2

Normalized Intensity

Normalized Intensity

Before embedding 1 0.8

20 mm 10 mm

0.6

0 mm

0.4 0.2 0 -0.1

0

0.1

0.2

0.3

1 0.8

0 mm 10 mm 20 mm

0.6 0.4 0.2 0 -0.1

Wavelength Detuning [nm]

0

0.1

0.2

0.3

Wavelength Detuning [nm]

(b) ENF specimen. Fig. 6 Continued. spectrum and the wavelength is expressed by the detuning from the initial center wavelength. In both DCB and ENF specimens, the reflection spectrum from the FBG sensor at the crack side clearly split into two peaks as the crack approached the arrester and thus the difference between the principal strains increased. However, the reflection spectra from the FBG sensor opposite to the crack hardly changed. It was expected that the crack propagation can be detected comparing the spectra obtained from the FBG sensors embedded at both edges of the arrester. In the next session, verification tests using the DCB and ENF specimens are conducted.

VERIFICATION TEST Materials and Methods The experimental setups is shown in Fig. 7. The configurations of the specimens were the same as the ones used in the FEA. The widths of the DCB and ENF specimens were 10cm and 5cm, respectively. The facesheets and the arrester were co-cured with the foam core in an autoclave. Two FBG sensors (grating length: 15 mm) were embedded in the arrester in contact with the facesheet. Initial crack was introduced by inserting a piece of 0.01 mm thick polyimide film. The length of the initial crack was 80 mm and the distance between the crack tip and the crack arrester, L, was 20 mm. The specimens were loaded at a constant crosshead speed of 2.0 mm/min using a universal testing machine (AG-50kNI, Shimazu Co.). Once the crack propagated, the test machine was stopped and the reflection spectra from the two FBG sensors were recorded. Then, the test was resumed and this procedure was repeated until the crack reached the arrester edge.

Detection of arrested crack for foam core sandwich structures

(a) DCB test.

(b) ENF test. Fig. 7 Experimental setups.

217

218

Shu Minakuchi et al.

Results Figure 8 presents reflection spectra obtained in the DCB test. The intensity of the spectra before embedding is normalized by the intensity of the highest component. Also the intensity of the spectrum at each L is normalized by the maximum intensity of the spectrum just after embedding (L = 20 mm). For a clear comparison, the wavelength is expressed by the detuning from the peak wavelength of each spectrum. Before the test, the spectrum from the FBG sensor at the crack side had two peaks due to thermal residual strain. When the crack approached the arrester, the spectrum clearly split into two peaks (L = 6.5 mm). On the other hand, the spectrum from the FBG sensor opposite to the crack hardly changed. These changes of the two spectra are consistent with the result of the FEA, confirming the validity of the proposed crack detection technique in DCB specimens. However, when the crack reached the arrester edge (L = 0 mm), the spectrum from the crack side sensor abruptly recovered its original shape; the two peaks united into one. FEA could not estimate this spectral response and thus it was expected that internal damage, which the simulation did not consider, was induced in the sensor embedded area. ENF test results are shown in Fig. 9 Only the spectra obtained from the sensor at the crack side edge are shown, since the opposite side sensor showed almost no changes in the spectrum shape as in the DCB test. Intensity of the spectra is normalized by the maximum intensity of the spectrum before embedding. As the crack approached the arrester, the difference between the central wavelengths of the two peaks gradually enlarged. The spectral response well agreed with the ENF simulation results, confirming the validity of the proposed technique in Mode II type crack detection. Again, however, the spectrum recovered its shape when the crack reached the arrester edge. After the test, cross-section of the embedded FBG sensor was observed by an optical microscope. Figure 10 shows the observed crack between the facesheet and the arrester. The interfacial crack passed by the FBG sensor, indicating that the crack released the strains in the FBG sensor. Finally we conducted FEA considering the interfacial crack. The spectrum simulated by using the obtained strain is presented in Fig. 11 The thermal residual strain and the mechanical strain was almost completely released and, consequently, the spectrum recovered its original shape. This result indicates that a crack penetrating the arrester can be detected from the recovering of the spectrum shape. The penetrating crack critically degrades the structural integrity and leads to catastrophic damages. Thus the spectral shape recovery can be a danger signal. To conclude the verification tests, we can detect the crack approaching the arrester from the splitting of the spectrum and the crack penetrating the arrester from the recovering of the spectrum shape, respectively. The proposed technique enables an effective application of the crack arrester and significantly improves the reliability of the foam core sandwich structures.

Detection of arrested crack for foam core sandwich structures

(a) L=20 mm.

(b) L=6.5 mm.

(c) L=0 mm. Fig. 8 Reflection spectra measured in DCB test.

219

Shu Minakuchi et al.

220

(a) L=20 mm.

(b) L=10 mm.

(c) L=0 mm. Fig. 9 Reflection spectra measured from Sensor A in ENF test.

Fig. 10 Cross-sectional micrograph of interfacial crack.

Detection of arrested crack for foam core sandwich structures

221

1.2

Normalized Intensity

Before embedding 1 0.8 0.6 0.4 0.2 0 -0.2

-0.1

0

0.1

0.2

Wavelength Detuning [nm]

Fig. 11 Simulated reflection spectrum of Sensor A after considering interfacial crack in FEA. It well agreed with the spectrum in Fig. 9 (c).

CONCLUSIONS The technique for detecting arrested cracks in the foam core sandwich structure was developed. First, FEA was conducted to predict the change of the reflection spectra from the FBG sensors embedded at both edges of the arrester. The verification tests clearly demonstrated that the crack approaching the arrester and the crack penetrating the arrester can be separately detected using the spectral responses. The proposed technique enables an effective application of the crack arrester and significantly improves the reliability of the foam core sandwich structures.

REFERENCES [1] Zenkert, D., editor (1997), The Handbook of Sandwich Construction, EMAS Publishing, Warrington, UK. [2] Herrmann, A.S., Zahlen, P.C., and Zuardy, I. (2005), In: Proceedings of the 7th International Conference on Sandwich Structures (ICSS-7), p 13-26. [3] Zahlen, P.C., Rinker, M., and Heim, C. (2008) In: Proceedings of the 8th International Conference on Sandwich Structures (ICSS8), p 606-623. [4] Hirose, Y., Hojo, M., Fujiyoshi, A., and Matsubara, G. (2007), Advanced Composite Materials, vol. 16, n. 1, p. 11-30. [5] Gafsi, R. and El-Sherif, M.A. (2000), Optical Fiber Technology, vol. 6, n. 3, p. 299-323. [6] Zhang, A.P., Guan, B.O., Tao, X.M., Tam, H.Y. (2002), Optics Communications, vo. 206, n. 1-3, p. 81-87.

Life extension and management of ageing fleets

25th ICAF Symposium – Rotterdam, 27–29 May 2009

CONCEPT OF THE NEW A320 FATIGUE TEST N. Rößler1 , C. Peters1, O. Tusch1, G. Hilfer1, C. Herrmann2 1

2

IABG mbh, Ottobrunn, Germany Airbus Deutschland GmbH, Hamburg, Germany

Abstract: Within the scope of the Extended Service Goal (ESG) for the A320 family fleet fatigue tests (NEF) have been decided by Airbus in order to support analysis/justification for an extension of the limit of validity (LoV) of the maintenance program to ensure continuation of safe aircraft operation. Up to 180,000 SF have to be performed to simulate the fatigue and damage tolerance behaviour of the airframe. The concepts of the new fatigue tests have been developed based on the experience and cognitions of all former aircraft tests for Airbus and in particular the A320 EF tests about 20 years ago, when EF2 was accomplished by IABG. Since the new tests have been taken into Airbus` program at very short notice, the realisation of the test set-up required unique and very time efficient solutions in all needs. The paper presents the test concept for the A320 NEF2 & NEF3 and highlights some improvements. The paper thus gives an overview of efficient and state-of-the-art full-scale fatigue testing currently being practised by IABG.

HISTORY OF AIRBUS FULL-SCALE FATIGUE TESTS BY IABG Aircraft structural integrity must be ensured from the first to the last flight. The comprehensive services provided by IABG’s aeronautic experts contribute substantially to the safety of aircraft structures. Structural tests are indispensable for the development and certification of aircraft, to guarantee efficient and safe operation in service. IABG’s fatigue and strength testing activities are based on the experience which IABG has gained during more than 40 years of aircraft structures M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 225–236. © Springer Science+Business Media B.V. 2009

226

N. Rößler, C Peters, O. Tusch, G. Hilfer and C. Herrmann

testing. Especially the full scale fatigue tests of all main AIRBUS types have been carried out successfully by IABG meeting all requirements of AIRBUS as contractor. The first full-scale fatigue test started 1973 with the A300. All following AIRBUS aircraft A310, A320, A330/ 340, A340-600 [2], A380 [3] have been tested by IABG (see fig. 1). Currently, A320 is undergoing testing again in the frame of the extend service goal (ESG) program. Thus, the A320 centre fuselage and wings will have been tested twice (as EF2 and NEF2) in the course of 20 years. For the first time ever a test of the A320 rear fuselage (NEF3) is currently being performed by IABG simultaneously with the NEF2 test. While these two tests are performed in Ottobrunn near Munich, the test of the A380 is running at the Dresden site of IABG. Moreover, the test for the A400M is under preparation in Dresden.

BACKGROUND OF A320 ESG PROGRAM On the strength of 20 years’ experience in monitoring more than 3,600 single-aisle aircraft now in service, Airbus is looking to take the A320 Family further. Work towards an extended service goal (ESG) for the family is now well under way with a development test programme, including major full-scale fatigue tests [1]. The overall aim of the A320 Family ESG is to enhance its maintenance programme from the current specification of 48,000 flight cycles (FC) and 60,000 flight hours (FH), up to 90,000FC and 180,000FH respectively. This extension will progress in two steps referred to as ‘ESG1’ and ‘ESG2’. ESG1 is an initial step planned to become effective in 2010 that will target a service goal extension of 60,000FC and 120,000FH. This programme will achieve a balanced development of structural potential and optimised maintenance activities for the aircraft family through its analyses. The development test phase for ESG1 is being conducted until July 2009. It will involve 120,000 simulated flights, each characterised by operational data monitored and collected to date. Subsequent testing through to 2011 will aim to validate ESG2’s ultimate goal of 90,000FC and 180,000FH, and will involve up to 180,000 simulated flights and a total of 360,000 simulated flight hours [1]. To achieve approval for the Extended Service Goal package, Airbus is carrying out full-scale fatigue tests on new production standard aircraft sections, enhanced with special structural features. The original A320 configuration will be taken into consideration, as well as the specificities of other family types. All of the results will be compiled to show the fatigue behaviour of the complete aircraft family. These tests will take into account 20 years of experience in A320 family operations.

Concept of the new A320 fatigue test

227

A300: 1973 – 1980

A310: 1981 – 1985

A320: 1986 – 1988

A330/ A340: 1990 – 1996

A340-600: 1998 – 2002

A380: 2003 – 2010

A320 NEF2: 2007 – 2011

A320 NEF3: 2007 – 2011

Figure 1: Airbus Fatigue Tests at IABG

228

N. Rößler, C Peters, O. Tusch, G. Hilfer and C. Herrmann

TIMELINE OF THE A320 NEF2/ NEF3 The timeline requirements in the A320 ESG program of the preparation phase for the test set-up and fatigue testing phase posed a challenge on IABG by the extremely short lead time. After the project start in March 2007 with the engineering, design and manufacturing phase the first major project milestone was reached with the delivery of both test specimens beginning of January 2008 by Beluga transportation to Munich Airport. After the follow-on works for the final assembly of the test specimens the two test set-ups were completed and the commissioning phase was started as scheduled beginning of July 2008 for NEF3 and September 2008 for NEF2. The NEF3 fatigue test was started in August 2008 in order to reach the ESG1 milestone of 120,000 simulated flights in July 2009. The NEF2 fatigue test was started in November 2008 in order to reach the ESG1 milestone of 120,000 simulated flights in July 2009 as well. The ESG2 milestone of 180,000 simulated flights shall be completed prior to July 2011 for both NEF2 and NEF3. Both NEF2 and NEF3 are progressing well according to plan.

THE CONCEPT OF THE A320 NEF2 & NEF3 TEST SET-UP Test Set Up The test articles are installed in a test set up consisting of restraint systems, several loading rigs, load distribution and introduction system (loading trees), inspection rigs, the hydraulic and pneumatic loading systems as well as the control and data acquisition systems. Due to the fact that both tests are using the same data acquisition system, hydraulic and pneumatic power, the tests could be placed tightly together in one test laboratory hall.

The NEF2 test article is supported statically determined by 6 hinged rods of the restraint system blocking all 6 degrees of freedom. The rods are connected at the forward bulkhead constrained in X-, Y- & Z-direction, at the rear bulkhead constrained in Y-direction and at each main landing gear dummy constrained in Zdirection. The NEF3 test article is rigidly fixed at the bulkhead. This was achieved by welding the bulkhead to a steel ring, which is bolted to the fuselage end. It was necessary to design the restraint rig considering stiffness requirements to avoid large deflection of the 19m freely suspended tail.

Concept of the new A320 fatigue test

229

A modern test and load introduction concept was developed, which fulfils all given requirements. In total 61 (NEF2) and 33 (NEF3) dual acting servo-hydraulic jacks are installed to the test set up to simulate aerodynamic and inertia loads of the aircraft structure.

Figure 2 – Isometric view of the A320 NEF2 and NEF3 test set up The loads are introduced via loading trees to seat and cargo rails, fittings riveted to the structure, pads bonded onto the wings or via dummy structures. Scaffolds are used as inspection rigs. They provide access to all parts of the test article and test systems during inspection, maintenance and repair works. The advantages of using scaffolds instead of a designed inspection rig are: • minimal effort and costs during the planning (only the inspection levels need to be defined) • highly flexible rig adaption to the current requirements during the installation phase • fast dismantling and reinstallation to get access for repair works at the test article • reuse for future projects A Comparison between A320 EF2 and NEF2 Test As mentioned above, IABG performed the former A320 EF2 test during the late eighties. The dimensions of the former EF2 test article are equal to the NEF2 test

230

N. Rößler, C Peters, O. Tusch, G. Hilfer and C. Herrmann

article. The past experiences of IABG were used for the dummy design and as concept guideline for the NEF2 test Therefore the design of the dummies for main landing gear, engine-pylon, slattrack, flap-track, aileron and the bulkheads are based on former EF2 design, taking into account the actual load requirements. The manufacturing costs of fuselage loading trees could be drastically reduced by using welded design instead of former casting design. There is a significant change in the support of the test article between the former EF2 and today’s NEF2 test. The EF2 test article was supported statically determined in the test rig at both bulkheads and after test shut down additionally supported in Z-direction at the main landing gear by a separate loading system thus avoiding undesired specimen stresses. At the NEF2 the Z-support rods were moved from the rear bulkhead to the main landing gear dummies. In this way, no further support system was required. The wing box loading was adjusted from five jacks at EF2 test to seven jacks at NEF2 test per wing. Two additional jacks are located at the inner wing area to achieve a better torsion moment simulation. The outer wing loading has been shifted from the top wing skin at EF2 test to the lower wing skin at NEF2 test. This concept reduces the moving mass to a minimum at outer wing, which is necessary to achieve a high test speed. At NEF2 test the volume reduction of the fuselage by styrofoam blocks was not realised. This gives now the flexibility to perform quick inspection in the fuselage without necessity to remove the blocks out of the fuselage for getting access. Load Pad Arrangement The arrangement of load pads has been customized in terms of structural demands and efficiency. The pad groups had been defined such that they follow spars and ribs. Therefore rectangular pad groups were not acceptable. Nevertheless, the pad group still follow a certain standard in order to save time and maintain a high quality. The figure 3 below shows a pad group with an optimized load distribution onto rib and spar.

Figure 3: Customized pad layout at the A320NEF2

Concept of the new A320 fatigue test

231

The tough time schedule of the overall project ended up in a narrow time gap for pad bonding. Therefore the time consuming turning of the wing and/or exclusive provision of the wing at Airbus for pad bonding was not possible. The bonding had to take place after final assembly of the wing to the fuselage and while the assembly of the test rig continued in parallel. In order to fulfil the task the pad bonding had to be done from below with an adapted bonding device to host the individual pad layouts with a high reliability and quality.

Figure 4: Pad Bonding at the A320 NEF test specimen

CONTROL & MONITORING SYSTEM, DATA ACQUISITION SYSTEM For actuator control of the tests two different control & monitoring systems (CMS) are used. Based on the complexity the NEF2 is operated by an IABG developed system (70 channels) based on Logidyn components and NEF3 uses a SmarTEST Elite system (40 channels). These systems also control the pressure inside the fuselage. Each control channel is equipped with various monitoring features to guarantee safe test operation. For additional safety, the NEF2 CMS monitors the reactive loads measured by 6 struts and NEF3 CMS the displacements in x- and ydirection of the tail cone. Compared to the first A320 test, the loading program was much more extended for NEF2 and NEF3: • NEF2: 165 different flight types with about 7,800 different load cases • NEF3: 270 different flight types with about 10,000 different load cases.

232

N. Rößler, C Peters, O. Tusch, G. Hilfer and C. Herrmann

An FE-model for each of the test structures was used to generate appropriate and optimized control parameters for the load cases within the flights. Due to the introduction of the innovative IABG-CMS on A340-600 and A380 structural tests, which allows increased test speed, the A320 NEF2 is performed as well with superior speed and excellent load reproduction accuracy. For both test facilities extremely short commissioning times could be achieved, particularly for the optimisation procedure to reach the final test speed. The central data acquisition system (HBM MGCplus) serves both of the NEF2 and NEF3 tests independently. For cost-efficient use of the DAS-Hardware, up to 3072 strain gauges can be connected via a 3-way connector board to the DAS. The DAS itself is equipped with 1024 (expandable to 1344) strain gauge inputs and 70 inputs for deflection transducers. Changing between 3 configurations – each of it can contain up to 1024 strain gauges– is done by patch cables with multi pin connectors. Data Acquisition is managed by the state of the art software system catman Enterprise, designed especially for multi channel purposes. The modern nature of fatigue testing demands comprehensive measurements results tailored on the requests of analysis. Therefore different kind of measurement campaigns can be performed. Depending on customer requests all patched sensors can be measured simultaneously and automatically following a list of load cases and load steps or following a flight cycle. Here, both types of measurement yield a data file containing the measured peak values of each load case. For particular requirements, a continuous measurement of all patched sensors can be performed as well. Then, the data file contains not only the peak values of the load cases but includes additionally the entire information during ramp-up and ramp-down of the load.

Pneumatics, Hydraulic IABG is using its in-house hydraulic power supply with totally 4,000 l/min capacity. A consumption of 1.100 l/min will be needed when both tests are running. The dimensioning of the hydraulic pipes was performed according to the experience of former Airbus tests (A300 – A380 EF), due to the fact that the loading program was not available during the conception phase of this challenging time schedule. A complete new pneumatic power system for both tests was built up within three months. In this time period the conception, the facility purchase (compressor, pipes, etc.), the build of a new hall, the installation and the function tests were performed. Now the pneumatic system fills each of the NEF2 and NEF3 fuselage up to 564 mbar differential pressure in less than 10 sec.

Concept of the new A320 fatigue test

233

TEST PERFORMANCE Test Speed The loading program consists of 106 loading points for an average NEF2 flight and 83 loading points for an average NEF3 flights. After optimisation of the complete test set up an average test speed of 700 flights/day at NEF2 and 640 flights/day at NEF3 could be reached. The NEF2 test is a factor 2.1 faster compared to the former EF2 test. Inspection Procedure Figure 5 below shows schematically the basic procedure of the inspection process. The inspection process is managed by the inspection administration which is responsible for the whole inspection process and the preparation of the inspection job cards.

Figure 5: Schematic view of the Inspection Process Non Destructive Testing The aircraft structure of the A320 NEF2 and A320 NEF3 is intensively inspected by IABG (up to level 3/ DIN EN 4179) in regular intervals with latest inspection equipment. The task is to find damages in an early state and to monitor damage propagation closely.

234

N. Rößler, C Peters, O. Tusch, G. Hilfer and C. Herrmann

The inspections of the A320 NEF2 and A320 NEF3 are organized in the same manner and are divided in four intervals with different inspection focus: - A inspection / daily visual inspection combined with monitoring of damages or special areas by visual inspection or NDT techniques - B inspection / in scheduled intervals with a mix of visual inspection and NDT inspection - C inspection / alternating with the B-Inspection with an increased NDT inspection and visual inspection effort - D inspection / intensive visual and NDT inspection at test end

Figure 6: Eddy current inspection on the A320 NEF2 wing The used NDT methods have been selected based on the inspection task and the structure requirements. The following methods are being practised: - General and Detailed Visual Inspection (GVI & DVI) - Ultra Sonic Testing (UT) including Phased Array Ultrasonic Testing (PAUT) - High and Low Frequency Eddy current Testing (HFET & LFET) including usage of four frequency eddy current testing for lap joints (LFET) and bore holes with rotating probes - X-Ray Testing (RT)

Concept of the new A320 fatigue test

235

Figure 7: Ultra sonic inspection (PAUT)

Figure 8: Lap joint inspection with Eddymax 4U (LFET) Damage Detection and Documentation The inspection results are documented using the IABG software DamDocV3. The SQL server based development is tailored for fatigue testing and seamlessly connects the inspection database with the inspection results in order to provide a complete damage documentation solution.

236

N. Rößler, C Peters, O. Tusch, G. Hilfer and C. Herrmann

The database is multi user capable and generates fast and economic documentations and overview of inspection results in order to give Airbus an early view for in time evaluations and conclusions.

Figure 9: Damage Documentation Database DamDocV3

CONCLUSION For the Extended Service Goal (ESG) program of the A320 Family, two major airframe tests for the center fuselage with wings (NEF2) and the rear fuselage (NEF3) have been erected at IABG Ottobrunn, Germany. Within 9 months, the entire engineering, design and a considerable part of mechanical test set-up build was completed in order to integrate the two test structures in time. Further integration, instrumentation and commissioning could be completed in a second compact time frame. Since July 2008 (NEF3) and September 2008 (NEF2), respectively, both tests are progressing at a fast rate. Test completion is planned for 2011. The extraordinarily successful achievement of densely packed milestones up to the current date was enabled by an efficient mix of proven technologies and procedures and newly established advancements.

REFERENCES [1] Airbus Letter November 2008 [2] Tusch, O. and Woithe, K., A340-600 Full Scale Fatigue Test: A Further step forward into a efficient structure qualification, In: Proceedings of the 27th ICAF symposium, Volume II, pp. 27-29, Toulouse, France, 2001 [3] Schwarberg, F. and Eichelbaum, F., An efficient load introduction concept for the A380 full scale fatigue test, In: Proceedings of the 29th ICAF symposium, Volume I, pp. 6-10, Hamburg, Germany, 2005

25th ICAF Symposium – Rotterdam, 27–29 May 2009

THE USE OF COMPOSITE MATERIAL STRIPS TO EXTEND THE DAMAGE-TOLERANCE LIFE OF INTEGRALLY STIFFENED ALUMINUM PANELS A. Brot, Y. Peleg-Wolfin, I. Kressel and Z. Yosef Engineering Division Israel Aerospace Industries Ben-Gurion Airport, Israel e-mail: [email protected]

Abstract: This paper describes testing and analysis performed on integrally stiffened aluminum panels reinforced by carbon-epoxy or boron-epoxy bonded strips. Testing was performed at roomtemperature and at -50°C. The test results identify a large potential for increasing significantly the damage-tolerance life of these panels. The analytical results, based on finite-element models, correlate very well with the test results.

INTRODUCTION Israel Aerospace Industries (IAI) has studied the damage-tolerance behavior of integrally stiffened metallic structures, reinforced by composite strips, as part of an international project called DaToN (Innovative Fatigue and Damage Tolerance Methods for the Application of New Structural Concepts), which was partially funded by the European Commission (EC). IAI has performed both numerical and experimental studies of integrally stiffened metallic structures, in the framework of this EC project. This paper describes testing performed under the DaToN framework, where composite material strips were used to enhance the crack growth resistance of the panels. The paper also describes the analytical calculations supporting the experimental results.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 237–244. © Springer Science+Business Media B.V. 2009

238

A. Brot, Y. Peleg-Wolfin, I. Kressel and Z. Yosef

TESTING OF INTEGRALLY STIFFENED PANELS A total of six integrally stiffened panels were crack-growth tested under constant amplitude loading. The panels were machined from 2024-T351 aluminum alloy. The overall dimensions of the panels were a 450 mm width and a 1000 mm length. Each panel was manufactured with two integral stringers. The first three panels were crack-growth tested without any reinforcing strips, at several stress levels and stress-ratios. Figure 1 shows a two-stringer panel in the test rig, before and after failure. An artificial crack of ±15mm length was inflicted at the panel centerline. The panels had crack propagation gages bonded back-to-back to the panels, along the expected crack path, in order to monitor the crack growth. Figure 2 shows the measured results of an unreinforced panel at a maximum stress level of 80 MPa, with R = 0.1. The results shown in Figure 2 represent the mean value of the growth of the front and back, right and left crack tips. It is very clear from Figure 2 that the stringers offered almost no resistance to the advance of the fatigue crack. As such, their value as a damage-tolerance enhancer was found to be minimal. The results shown in Figure 2 were used as a baseline in order to evaluate the effect of the panel reinforcement using composite materials.

Figure 1: A Two-Stringer Panel in the Test Rig and a Panel after Failure In recent years, there has been much discussion of the advantages of a "hybrid" stiffened panel which has composite materials bonded to the aluminum [1] - [5]. The composite material reinforces the aluminum panel and serves to bridge any cracks that may develop in the aluminum panel. This bridging effect was proven during the last 30 years in many composite bounded repairs of aging aircraft [4]. In order to improve the performance of the two-stringer integral panel, two 35mm wide strips, made from Hexcel Vicotex 913 unidirectional carbon-epoxy material

The use of composite material strips to extend the damage tolerance life

239

were co-bonded to the panels at 120°C, using 3M AF 163-2 adhesive, as is shown in Figure 3. Each strip consisted of three plies of carbon-epoxy material. The purpose of the strips was to reduce the stress-intensity of a crack that grows under it, thereby increasing the crack growth life of the panel. On another identical panel, two 35 mm wide strips of Textron 5521 F/4 boron-epoxy were bonded. Each strip consisted of two plies of boron-epoxy material. For both reinforcement schemes, the composite material strips were bonded only on the stringer side of the panels. 140

120

U n r e in f o r c e d S p e c im e n C a r b o n - E p o x y R T Sp e c im e n B o r o n - E p o x y R T Sp e c im e n

half crack length a (mm)

100

80

S tiff e n e r C e n te r L in e 60

40

B e g in n in g o f R e in f o rc m e n t

20

0 0

20000

40000

60000

8 0 0 (Cycles) 00 100000 Life

L if e ( c y c le s )

120000

140000

160000

Figure 2: Measured Crack Growth Curve for the Two-Stringer Panel Both hybrid panels were tested at room temperature, under a 7% higher loading than what was used for the unreinforced panel (80 MPa at R = 0.1). The purpose of the 7% increase was to compensate for the additional EA cross-section contribution of the reinforcing strip.

\

Figure 3: Panel with Carbon-Epoxy Reinforcing Strips, and after Failure

A. Brot, Y. Peleg-Wolfin, I. Kressel and Z. Yosef

240

Figure 4 shows the crack growth test results of both hybrid panels at roomtemperature. The results clearly show that both hybrid panels had a significantly slower crack growth rate than the unreinforced panel. Figure 4 also shows that the crack growth life of the three-layer carbon-epoxy strips gave somewhat better results than the two-layer boron-epoxy strips.

half crack length a (mm)

It should be noted that no debonds between the composite strips and the metal substrate, or delaminations between the layers, were observed up to failure for all the tested panels.

140

Unreinforced Specimen

120

Carbon-Epoxy RT Specimen

100

Boron-Epoxy RT Specimen

80 Stiffener Center Line

60 40

Beginning of Reinforcment

20 0 0

20000

40000

60000

80000

100000

120000

140000

160000

Life (cycles)

Figure 4: Comparison of the Crack Growth of Both Reinforced Panels to the Unreinforced Two-Stringer Panel. (The loading on the reinforced panels were increased by 7% relative to the unreinforced panel.) The crack propagation rate of all the reinforced panels seems to be constant, almost up to failure. This phenomenon is in good agreement with the Rose Model [4] that predicts a constant stress-intensity factor under a bonded composite patch. Detrimental residual thermal stresses exist in the aluminum panels reinforced by composite material patches, induced by the thermal expansion coefficient mismatch between the carbon-epoxy or boron-epoxy materials and the aluminum substrate. These residual stresses may be significant because of the difference between the curing temperature 120°C and the operating temperature. When tested at room-temperature (approximately 25°C), finite-element studies show that the residual stress in the aluminum panel will reach approximately 8 MPa for both the three-layer carbon-epoxy strips and the two-layer boron-epoxy strips, a relatively insignificant value. It should be noted that the compressive residual stress in the composite reinforcement strips was significantly higher than that of the aluminum substrate.

The use of composite material strips to extend the damage tolerance life

241

Figure 5: Test set-up for the Carbon-Epoxy Reinforced Panel Tested at -50°C This residual stress phenomenon was shown to be more pronounced at the reduced temperatures that occur at higher altitudes. Finite-element studies showed that the residual stress in the aluminum panel will reach approximately 14 MPa for the carbon-epoxy strips at -50°C. On the other hand, the inherent crack growth rate in the 2024-T351 aluminum panel is much slower at -50°C than at room temperature. Therefore, an additional test was performed with a carbon-epoxy reinforced panel at an ambient temperature of -50°C. Figure 5 shows the test setup and the refrigeration unit that was used to cool the test chamber to -50°C. Also for this test, 7% higher loading was used, compared to what was used for the unreinforced panel (80 MPa at R = 0.1)

half crack length a (mm)

140

Unreinforced Specimen Carbon-Epoxy RT Specimen Carbon-Epoxy -50ºC Specimen

120 100

Stiffener Center Line

80 60 40

Beginning of Reinforcment

20 0 0

100000

200000

300000

400000

500000

Life (cycles)

Figure 6: Comparison of Carbon-Epoxy Reinforced Panels at Room-Temperature and at -50°C (results are compared to an unreinforced panel)

A. Brot, Y. Peleg-Wolfin, I. Kressel and Z. Yosef

242

Figure 6 shows that the crack grew significantly slower at -50°C than at room temperature, showing that the reduced crack growth rate of aluminum at -50°C was more decisive than the presence of tensile residual stresses. It should be noted that, as in the previous tests performed at room temperature, no debonds between the composite strips and the metal substrate, or delaminations between the layers, were observed up to failure, for all the panels tested.

CRACK GROWTH ANALYSIS OF THE PANELS

K [MPa*SQRT(mm)]

A NASTRAN finite-element model (FEM) was built to study the effect of the composite material reinforcement strips on the stress-intensity factor. The model was composed of CQUAD4 shell elements representing the skin and the reinforcement strips. 3D HEXA elements were used for the adhesive. Due to symmetry, only a quarter-model was analyzed. A nonlinear analysis was performed for several crack configurations in order to examine the contribution of the composite strips on the stress-intensity values. The first step was to build the finite-element model for the unreinforced panel. The next step was to add the composite material strips and adhesive to the model. The final step was to calculate the stress-intensity of the cracked panel, for a range of crack lengths from 15mm to 100mm using the displacement-extrapolation method. The results of this stressintensity analysis are shown in Figure 7 for the carbon-epoxy and boron-epoxy reinforced panels. It should be noted that the stress-intensity of the cracked aluminum panel is much lower at the bonding surface interface than at the free surface of the aluminum panel, as is shown in Figure 7. This means that the effect of the reinforcements is to introduce both tensile and bending effects on the aluminum panel. Figure 7 also showed a convergence of the mean stress-intensity factors to a nearly constant value beneath the strip, verifying the good agreement with the Rose Model [4] that predicts a constant stress-intensity factor under a bonded composite patch. 1800 1600 1400 1200 1000 800 600 400 200 0

K inner (Carbon Strip) K outer (Carbon Strip) K inner (Boron Strip) K outer (Boron Strip)

0

10

20

30 40 50 60 70 half crack length a (mm)

80

90

100

Figure 7: Stress-Intensity Results for Cracked Panels ("inner" refers to the bonding surface while "outer" refers to the free surface)

The use of composite material strips to extend the damage tolerance life

243

The reduction at the stress-intensity due to the reinforcement was taken into account by the NASTRAN analysis. The stress-intensity factors were extracted from the FEM, as a function of crack length, for the unreinforced panel, and for the panel with the reinforcing strips (at the interface between the panel and the reinforcement, and at the free surface of the panel). The stress-intensity results, as obtained from the FEM, were input into NASGRO ver. 5 (crack growth software) as a data table, in order to compute the predicted crack growth characteristics. The effects of the stress-intensity variation (between the free edge and at the interface) were accounted for by this analysis. The results are shown in Figure 8 for the carbon-epoxy reinforcing strips and in Figure 9 for the boron-epoxy reinforcing strips. The results shown in Figure 8 and Figure 9 indicate a very good agreement between the test and analytical results.

Figure 8: Crack Growth of the Carbon-Epoxy Reinforced Panel

Figure 9: Crack Growth of the Boron-Epoxy Reinforced Panel

A. Brot, Y. Peleg-Wolfin, I. Kressel and Z. Yosef

244

SUMMARY AND CONCLUSIONS 1.

This experimental and analytical study demonstrated the large potential that exists by use of carbon-epoxy or boron-epoxy patches to increase the damage-tolerance life of integrally stiffened aluminum panels.

2.

Further testing and analysis is needed to quantitatively confirm these results.

3.

The analytical results, derived from finite-element models, correlate very well with the test results.

4.

The effect of tensile residual stresses in the aluminum panels at low temperatures, introduced by the coefficient of thermal expansion mismatch between the aluminum and composite materials, is not detrimental to the crack growth rate since the reduced crack growth rate of aluminum at low temperatures more than offsets the effect of the tensile residual stresses.

5.

No debonds between the composite strips and the metal substrate, or delaminations between the layers, were observed up to failure for all the panels tested.

REFERENCES [1] Zhang, X. et al, "Improving Fail-Safety of Aircraft Integral Structures through the Use of Bonded Crack Retarders", Proceedings of the 24th ICAF Symposium, Naples, May 2007. [2] Heinimann, M., "Validation of Advanced Metallic Hybrid Concept with Improved Damage Tolerance Capabilities for Next Generation Lower Wing and Fuselage Applications", Proceedings of the 24th ICAF Symposium, Naples, May 2007. [3] Baker, A. "Repair of Cracked or Defective Metallic Aircraft Components with Advanced Fibre Composites – an Overview of Australian Work", J. of Composite Structures, vol. 2, 1984. [4] Baker, A., Rose, F. and Jones, R. "Advances in the Bonded Composite Repair of Metallic Aircraft Structure", Elsevier, 2002. [5] Baker, A. "Bonded Composite Repair of Fatigue-Cracked Primary Aircraft Structure", Composite Structures, 47:431-443, 1999.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

STRUCTURAL INTEGRITY OF A WING UPPER SKIN WITH EXFOLIATION CORROSION Andreas Uebersax1, Cyril Huber1, Guillaume Renaud2 and Min Liao2 1

2

RUAG Aerospace, Switzerland National Research Council Canada

Abstract: This paper presents the modelling and testing of the effects of exfoliation corrosion on the structural integrity of an aluminium 7075-T651 wing upper skin. Finite element and fatigue strength analyses were carried out to quantify the impact of exfoliation corrosion and grind-out repairs at fastener holes on a wing upper skin under a fighter aircraft spectrum loading. The analytical models were compared and validated by test results. The tests were performed on 27 fatigue coupons in a pristine, naturally and artificially corroded and grinded condition. The models provide an engineering tool for quick assessment of the effects of exfoliation corrosion damage on wing upper skins as well as guidance on possible grind-out repair actions.

INTRODUCTION Exfoliation Corrosion is a severe form of intergranular corrosion. It is commonly found in aging aircraft wing skins made of aluminium alloys, especially in precipitation hardened aluminium alloys such as 7075-T6. Aircraft operators often face uncertainties about the effects that exfoliation corrosion and subsequent repairs have on the structural integrity and subsequent residual life of corroded parts. The current ‘find-it-fix-it’ approach to corrosion maintenance requires that even the smallest corrosion damage be removed by grind-out. In addition to being costly, grinding away exfoliation degrades the static strength of the structure M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 245–261. © Springer Science+Business Media B.V. 2009

246

Andreas Uebersax, Cyril Huber, Guillaume Renaud, Min Liao

unnecessarily. A new corrosion management philosophy has been proposed with the intent of anticipating, planning and managing corrosion. Although some conservative assumptions due to a lack of comprehensive data may be required, structural integrity can often be maintained for a certain service life without repair. This information may be invaluable for decision-making in fleet management or aircraft maintenance. A collaborative project between RUAG Aerospace and the National Research Council of Canada (NRC) was undertaken to study the effects of exfoliation corrosion at fastener holes on the structural integrity of a wing upper skin. The project was based on the example of an F-5E fighter aircraft wing. The analytical approach used to model exfoliation corrosion by the soft inclusion method used in this study was presented in detail in a paper by G. Renaud et al. [1]. This paper compares the updated analytical and experimental results of the local damage model and determines the structural strength criteria for the wing upper skin.

BACKGROUND In general, wing upper skin loading is not Mode I fatigue critical. However, it may become critical for certain wing skin details (e.g. CF116, which is the Canadian version of the F-5A [2] and the F-15 [3]) and for certain loading conditions. To assess the structural integrity of a wing upper skin, other failure modes must be considered in addition to fatigue and corrosion fatigue failure, such as: − Bearing failure at a fastener hole − Local buckling (e.g. inter-rivet buckling) − Global buckling (e.g. panel buckling) − Fastener failure In the work of T.B. Mills et al. [4] the depth of the exfoliation is considered to be the primary critical parameter. The extent of the exfoliation damage in the L and T material directions has little effect once exfoliation has degraded a fastener hole. The analytical techniques used to assess the structural integrity of exfoliated aircraft structures have focused mainly on the thickness loss due to the grind-out of the exfoliation. As some prior work of the NRC has shown [5], grinding away exfoliation further degrades the ‘as-is’ condition of the structure. In a previous study performed at RUAG Aerospace, the most critical section for panel buckling of the F-5E wing upper skin was identified. The work within this study refers to this specific wing section (see Fig. 1).

Structural integrity of a wing upper skin with exfoliation corrosion

247

Figure 1. Analyzed wing section To be conservative, the most severe flight spectrum was chosen for the investigation: the spectrum of the solo display aircrafts of the Patrouille Suisse (Swiss Air Force Display Team). The Nz spectrum shows a considerable amount of negative Nz events that implicate tension stresses in the wing upper skin.

Figure 2. Nz usage spectrum of the solo display aircrafts of the Patrouille Suisse compared to the FALSTAFF spectrum

248

Andreas Uebersax, Cyril Huber, Guillaume Renaud, Min Liao

EXPERIMENTAL PROCEDURE A retired Swiss Air Force F-5E wing with corrosion on the upper skin was used for the experimental investigation. Twenty-seven dog-bone coupons were cut from the wing upper skin for fatigue testing. These tests were performed on specimens that were pristine, naturally corroded, artificially corroded and repaired by grind-out. Additional coupons were cut from the wing skin for material characterization and environmental corrosion testing. Environmental corrosion tests To gather data on the corrosion growth rates for a 7075-T651 aluminium alloy in Switzerland’s meteorological conditions, as well as on surface protection degradation, specimens in multiple configurations were placed on a test rig for outdoor exposure. The test included specimens with natural corrosion from the disposed wing skin and pristine specimens with different surface treatments. Coupon preparation For fatigue testing, coupons with differing amounts of corrosion were prepared. The naturally corroded coupons had small amounts of corrosion and were comparable to pristine specimens. Artificially corroded coupons were created using a modified ANCIT corrosion protocol to generate higher corrosion levels (see Fig. 3, 4). The corrosion morphology showed better results with the modified ANCIT corrosion protocol than with EXCO. Although EXCO produced more severe pitting, it did not produce an intergranular attack as was seen in naturally corroded specimens.

Figure 3. Coupon No. 17 after artificial corrosion

Figure 4. Coupon No. 26 after artificial corrosion

Structural integrity of a wing upper skin with exfoliation corrosion

249

For the grind-out repairs, it was assumed that the original HL13-6 fastener would be removed for grinding and would be replaced by a PLT-154-6 type (see Fig. 5). Hi-Lok fasteners can not be used for repairs on a real wing skin where there is no access from the backside. For analysis and testing, three different grind-out depths were chosen that were deeper than that of the natural corrosion (see Fig. 5 and Table I). For accuracy and reproducibility reasons, the grind-out shapes were circular and centred on the fastener hole.

Figure 5. Coupons with three different grind-out depths: 2.0 mm, 1.3 mm, 0.5 mm (from left to right) Specimen Number 03,16, 18, 24, 25, 27 01, 04, 05, 10, 19, 23 06 08 11 02 12 13 07 09 14 15 17 20 21 22 26

Type Pristine Natural corrosion 0.5mm grind-out 0.5mm grind-out 0.5mm grind-out 1.3mm grind-out 1.3mm grind-out 1.3mm grind-out 2.0mm grind-out 2.0mm grind-out 2.0mm grind-out Artificial corrosion Artificial corrosion Artificial corrosion Artificial corrosion Artificial corrosion Artificial corrosion Table I. Test specimen list

Maximum Thickness Loss 0% <4% (not modelled) 8% 10% 11% 18% 39% 21% 26% 24% 29% 7% 12% 26% 5% 34% 26%

250

Andreas Uebersax, Cyril Huber, Guillaume Renaud, Min Liao

Because the coupons were cut from a wing skin, they were linearly tapered in the longitudinal and transverse direction. Therefore bi-linear shims were required to force parallel surface contact in the grip area. For the pristine and naturally corroded coupons, the original fastener at the test hole was left in place, in order not to disturb the local tribological contact condition. The stiffener (spar cap) was removed from the back-face of the coupon, leaving only a small portion of the flange around the fastener. The fastener holes adjacent to the test hole were cylindrically reamed, cold-worked and filled with interference fit steel plugs to prevent premature failure. It can be seen from Table I that the maximum thickness loss percentage, based on the average thickness of the tapered specimen, ranged from 0% (pristine) to 39% (grind-out). Moreover the shape and size of the damage were different for each specimen. Static testing Static tensile and compression tests were carried out with the fatigue test coupons in order to verify the lateral deflection, i.e. the secondary out-of-plane bending due to the material loss (grind-out repairs and exfoliation corrosion) of the coupons. Those results were used to verify the finite element models of the coupons. Fatigue testing Load levels were determined for each specimen to ensure the same nominal stress in the coupons. The spectrum valleys and peaks ranged between -293 MPa (-42.5 ksi) and +142 MPa (+20.6 ksi) in the nominal gross section (see Fig. 6).

Figure 6. Nominal gross section of coupon

Structural integrity of a wing upper skin with exfoliation corrosion

251

The coupons were tested in fatigue under the compression-dominant wing upper skin loading spectrum of the solo display aircrafts of the Patrouille Suisse (see Fig. 2). This loading spectrum showed higher and more frequent tension loads than the average fleet spectrum. The tests were carried out in laboratory air at a frequency of 12 to 15 Hz. The coupons were tested to failure with crack growth data being recorded. The initial requirement was to test the coupons for up to 4 service lives, i.e. 16,000 simulated flight hours, if no premature failure occurred. Subsequently, it was decided to continue testing beyond four service lives until failure. The life and fatigue damage induced prior to the test start was not considered for the present analysis. The wing had 1,800 flight hours when it was withdrawn from service, which is a minor portion of the subsequently tested lives. Additionally, the in-service loading spectrum was not as damaging as the one used for the analysis and the testing.

MODELLING The local damage model developed to analyse exfoliation corrosion in this research project was presented in detail in a previous paper [1]. This paper presents improvements to the model based on a correlation with test results. An example of how the model can then be used to assess the necessity of a repair is shown for an F-5 wing skin. Local damage model A parameterized geometric and Finite Element (FE) model of medium complexity was developed to represent the local corrosion damage in MSC/MARC as seen in Fig. 7. In this model, damage is represented as an off-centre, rotated, truncated ellipsoid enclosing a fastener hole. The corrosion damage is modelled as a ‘soft inclusion’ of adjustable stiffness that maintains some load bearing capability. If the inclusion is removed, the model is representative of a grind-out repair. In the event that the corrosion is not uniform around the hole, the 8 ‘quadrants’ of the corrosion ellipsoid can be individually selected as corroded or pristine, as shown in Fig. 8.

252

Andreas Uebersax, Cyril Huber, Guillaume Renaud, Min Liao

Figure 7. Soft inclusion model of a dogbone coupon

Ultrasound image

Coupon No. 22

Visual image

Coupon No. 21

Figure 8. Corrosion damage and corresponding soft inclusion models

Test strain measurement results were used to calibrate the soft inclusion parameters. In work performed previously by the NRC, a reduction of 70% in corroded material E modulus showed a good correlation with test results [5]. Wing section finite element model The most critical section of the wing upper skin with respect to buckling is known from previous work performed at RUAG Aerospace. PATRAN/NASTRAN FE shell models were used to analyse the wing section with a wing upper skin of nominal thickness and with grind-out repairs (see Fig. 9). The analysis provided buckling factors (see Fig. 10) and local stresses for inter-rivet buckling and bearing strength calculation for three different damage configurations. The three different simulated grind-out depths corresponded to the depths used for the coupon testing. In the model, the area around each fastener which represented corrosion damage was removed. This idealizes the model geometry as a skin with strips of reduced thicknesses over the spars and ribs (see Fig. 9). For modelling convenience, the width of the strips was kept constant at 30mm for the three different grind-out depths. Only the more conservative cases with the grind-outs were simulated.

Structural integrity of a wing upper skin with exfoliation corrosion

253

upper skin

Figure 9. Thickness distribution of wing upper skin with grind-out strips around the fasteners

Figure 10. Result of linear buckling analysis for 2.0 mm deep grind-out repairs

Fatigue analysis model All but the naturally corroded specimens were modelled and analyzed under static tension, static compression and fatigue loading. This resulted in 20 distinct FE models. The principal stress curves in the lateral direction of all the models were compared with that of an AFGROW base configuration in order to calculate the normalized stress curve to be used in AFGROW. A summary of the stress concentration factors, defined as the maximum and minimum principal stress values divided by the nominal gross section applied stress, is presented in Fig 11. Results are presented for an assumed 70% reduction in the E modulus of the corroded material, as well as for the more conservative case of a 100% reduction in E.

Stress concentration factor

4 3.5 tension

3 2.5

compression

2 1.5

Grind-out Corrosion 70% E reduction Corrosion 100% E reduction

1 0.5 0 0%

10%

20%

30%

40%

Maximum thickness loss

Figure 11. Stress concentration factor vs. maximum thickness loss, [1]

254

Andreas Uebersax, Cyril Huber, Guillaume Renaud, Min Liao

Fig. 11 shows that the stress concentration factors correlate linearly with the grindout thickness loss. Note that for an equivalent loss in thickness, the stress concentration factors of the corroded specimens are lower than those of the grindout specimens. This is mainly due to the different shapes of the grind-out and corroded areas, as well as the assumed residual stiffness in the exfoliation area for the corroded specimens. Fatigue analysis model calibration The crack growth model was calibrated to the test results by adjusting the equivalent initial flaw size (EIFS) and the bending stress fraction. These parameters were adjusted until the total lives and crack growth rates of the coupons that failed or cracked during the test matched the predicted lives. AFGROW allows superposition of tension and bending loads. Therefore the ratio of the tension and bending stress to the reference stress must be input for each load case to be modelled (i.e. tension and bending stress fraction), [6]. It was observed that in the analysed specimens where cracks nucleated in the centre hole, the cracks grew from both sides of the hole. Therefore, the fracture mechanics model selected in AFGROW was a quarter-circular double corner crack at a hole. This crack model is presented in Fig. 12 for a grind-out geometry. The thickness assumed in AFGROW was the total thickness of the coupon.

Figure 12. AFGROW model for a grind-out geometry A methodology to determine the bending stress fraction used in AFGROW for all coupons was established. It was found that the FE models underestimated the lateral deflection of the coupons due to the secondary bending by a factor close to 2. By multiplying the value of the FE generated bending fractions close to the countersink by a factor of 2, a good match with the crack growth test data was obtained.

Structural integrity of a wing upper skin with exfoliation corrosion

255

RESULTS Environmental Corrosion Tests The environmental corrosion tests were started in 2006. To date, only minor corrosion propagation on specimens with pre-existing corrosion was observed. Therefore an insufficient quantity of data is currently available to determine the corrosion growth rates. The tests are still on-going. Static analysis of the wing section Within the considered wing section, panel buckling of the upper skin occurs for all configurations (pristine and grind-out) at the same location, i.e. approximately at wing station 70 between the 27% and 33% spar, see Fig. 10. The margin of safety for panel buckling decreases from 1.49 for the pristine condition to 0.49 for the condition with 2.0mm deep grind-out repairs. Fatigue The available crack growth and total life data of the coupons that cracked from the centre hole were used for comparison with analysis results. The grind-out coupons No. 07, 09 and 11 provided usable crack growth information until failure. Grindout coupons No. 12, 13 and 14 and artificially corroded coupon No. 17 provided one crack growth data point each. Coupon No. 12 failed unexpectedly with a fatigue life much shorter than that of the other coupons, as seen in Fig. 13.

Figure 13. Total fatigue test lives of tested coupons (coupons are in the same order as in Table I)

Andreas Uebersax, Cyril Huber, Guillaume Renaud, Min Liao

256 20

Coupon 07 - Test (left) 18

Coupon 07 - AFGROW Model

16

Crack length c (mm)

26% max. thickness loss

Coupon 07 - Test (right) Coupon 09 - Test (left)

24% max. thickness loss

Coupon 09 - Test (right)

14

Coupon 09 - AFGROW Model

12 10 8 6 4 2 0 0

0.5

1

1.5

2

2.5

3

Cycles (Millions)

Figure 14. Experimental and analytical crack growth for two grind-out coupons It can be seen that when the total test lives are matched, the overall analytical crack growth behaviour is, in most cases, similar to what was observed in the test. It should be noted that the crack lengths shown in Fig. 14 were adjusted with respect to the crack origin found by fractography.

4.5

20

4.0

18

3.5

16 Test life (Mcycles)

Analytical life (Mcycles)

Using the appropriate bending stress fractions, the equivalent initial flaw sizes averaged around 0.457 mm (0.018 inch) for coupons with grind-out repairs and 0.279 mm (0.011 inch) for coupons with artificial exfoliation corrosion. These resulting EIFS are not directly related to the actual physical damage dimensions. They are calibration values that can include effects not included in the model such as the countersink geometry, the actual crack origin, residual stresses, etc.

3.0 2.5 2.0 1.5 1.0

Grind-out

0.5

Corr. 70% E reduction

Grind-out Corrosion

14 12 10 8

`

6 4

0.0

2

0%

10%

20%

30%

40%

Maximum thickness loss

Figure 15. Analytical life vs. maximum thickness loss

Coupon No12

0 0

2

4 6 8 10 12 14 16 18 20 Analytical life (Mcycles)

Figure 16. Analytical and test life comparison

Structural integrity of a wing upper skin with exfoliation corrosion

257

A summary of the analytical lives of all the modelled coupons, plotted against maximum thickness loss, is presented in Fig. 15. Only one pristine coupon (0% thickness loss) was included in Fig. 15 to avoid an undue influence of this type of coupon on the linear regression curve. Testing of half of the coupons was stopped after 6 service lives, because fatigue cracks nucleated in the clamping area of the coupons. Therefore limited data is available to evaluate the quality of the calibrated model. However, it can be seen that the model estimated analytical total lives within 30% of the test lives (dashed lines in Fig. 16), except for the outstanding short life of coupon 12. Fractography was performed on cracked and failed specimens to determine the locations of crack nucleation. It was found that for grind-out coupons, most of the cracks nucleated in the countersink. Corroded coupons showed multiple crack nucleation sites in the cylindrical part of the hole, in the countersink and on the top surface as seen on Coupon No. 26 (see Fig. 17). Detailed fractography of the specimens showed that crack nucleation features were essentially pits or pit-like features.

Coupon surface Fastener hole Figure 17. Scanning electron microscope (SEM) picture of coupon No. 26 shows multiple crack initiation

DISCUSSION Curves like those presented in Fig. 11 and Fig. 15 could be used in an engineering context for a quick assessment of the effects of exfoliation corrosion and guidance on possible grind-out actions. Using such curves, calibrated for various initial crack sizes, an engineer could carry out a quick damage tolerance analysis. For example, the reduction in crack growth lives, with respect to the pristine state, based on a 1.27mm (0.05 inch) standard initial crack size assumption and a thickness loss of 10% are listed in Table II.

Andreas Uebersax, Cyril Huber, Guillaume Renaud, Min Liao

258

Scenario Corrosion (70% reduced E mod.) Grind-out

Stress concentration factor increase 2.5%

Life reduction

7.6%

13.5%

3.1%

Table II: Effects of a 10% thickness loss (1.27 mm initial crack)

The results in Table II were based on the assumption that the grind-out geometry, that of coupon No. 08 in this example, is identical to the corrosion geometry. This is a conservative assumption since the grinding out of corrosion damage always removes more material. Another application of the local damage model for defining repairs is shown in Fig. 18. This curve summarizes the structural strength criteria for the analyzed wing upper skin section with grind-out repairs. The fatigue strength criterion assumes that crack growth life must fulfil twice the projected service life of 4,000 flight hours, see Eqn. 1. M .S . fatigue (thickness) =

2 x 4'000 flighthours −1 life CG (thickness)

(1)

As it can be seen in Fig. 18, the structural strength/integrity envelope is defined by three static strength criteria: fastener strength, inter-rivet buckling and panel buckling. A similar graph can be worked out for skin thickness reduction due to exfoliation corrosion.

Structural integrity of a wing upper skin with exfoliation corrosion

259

Figure 18. Strength criteria as a function of local skin thickness reduction due to grind-out repair in the considered section of the wing upper skin. Fractography of the exfoliated coupons confirmed that exfoliation corrosion does not only consist of grains lifting off. The damage also appeared to form pit-like discontinuities. In the studies by Sharp, Mills and Clark, [7], the pitting associated with EXCO exfoliation in 7075-T651 aluminium that caused fatigue failure was 60-100 m in depth regardless of the continued depth of exfoliation. In the present study the average EIFS for exfoliation corrosion damage was found to be 0.279 mm. As already noted, these EIFS are not directly related to actual physical damage dimensions. They are calibration values that can include effects not included in the model. In order to determine an initial crack size which is based on actual physical damage sizes, adequate crack growth material data (da/dN), including data for the short crack growth regime, would be required as well as more accurate modelling (e.g. countersink geometry, etc.). The main step towards developing a holistic model would be to integrate the timedependent aspect of the corrosion pitting and corrosion exfoliation rates in the model. The structure must be analyzed for adequate static margins of safety in its ‘as-is’ condition, as well as for a ‘time-to-critical’ state using exfoliation and pitting rates. Even if conservatively high corrosion rates need to be used, a structure can sometimes show an adequate margin of safety for a certain service life without repair.

260

Andreas Uebersax, Cyril Huber, Guillaume Renaud, Min Liao

For the present case, one of the key issues in allowing exfoliation to remain in-situ is to determine its extent and to monitor its growth. Still, the non-destructive inspection (NDI) technique to measure the maximum depth of exfoliation corrosion without access from the backside is a challenge and is not yet available to be used in field applications.

SUMMARY AND CONCLUSION A combination of static and crack growth analysis models were developed to evaluate the impact of exfoliation corrosion and grind-out repairs on remaining fatigue life and residual strength of a wing upper skin. For this purpose a total of 20 dog-bone coupons were modelled, analyzed and validated by test results. A local damage model, based on a soft inclusion concept, was used to model the corroded and grinded coupons. Results showed that the models provided a good estimation of the performance reduction due to the exfoliation or grind-out. For the considered centred circular grind-out repairs, the stress concentration and analytical crack growth life could be correlated well with the maximum thickness loss. Such a relationship was not observed as clearly with the corroded coupons. For future work a combined severity parameter, which includes the effects of corrosion pits and intergranular corrosion, maximum thickness loss, total corrosion volume and damage location, is needed to give a better estimation of the performance of the corroded coupons. The results from the local damage models were applied to a finite element model of an F-5E wing upper skin section. It was shown that for the analyzed wing section, the model could be used to define the most important structural strength criteria, which were the fastener strength, the inter-rivet and the panel buckling behaviour. This information is important when defining repair actions for corrosion damaged wing skins. It was also shown that leaving the corrosion in place gives an advantage in static residual strength, [5] as well as in fatigue strength over grindout repairs. This is true as long as corrosion growth is not a determining factor. These results and conclusions are specific for the considered wing upper skin section under the compression-dominated loading spectrum. A similar approach may be envisaged for the analysis of a tension-dominated lower wing skin where fatigue is of greater concern. The study shows that the models can provide an engineering tool for quick assessment of the effects of exfoliation corrosion damage on wing skins and guidance on possible grind-out repair actions. Further work to develop more holistic models will focus on including short crack growth behaviour and the time-dependent aspect of corrosion pitting and corrosion exfoliation rates.

Structural integrity of a wing upper skin with exfoliation corrosion

261

ACKNOWLEDGEMENT This project is a RUAG-NRC collaborative project. The authors would like to thank the following people at RUAG Aerospace: M. Wyss for fatigue testing, J. Lussi and S. Zehnder for NDI support, B. Bucher for FE modelling and I. Kongshavn for reviewing the paper. Thanks to Ron Gould, Tom Benak, Richard Desnoyers, Mike Brothers, David Backman, and Weijie Chen of NRC for their involvement in corrosion testing, fatigue testing, fractography, NDI, and various data analyses and test measurements.

REFERENCES [1] Renaud, G., Liao, M., Uebersax, A. and Huber, C. (2008), Static and fatigue analysis of an upper wing skin with exfoliation corrosion, 26th International Congress of the Aeronautical Science. [2] Raizenne, M.D., et al. (1993), CF1165 Upper wing skin compression induced fatigue cracking: a case study, Proceedings of the 17st ICAF Symposium, vol. I, pp. 367-388, Blom, A.F. (Ed.), EMAS, Warrington. [3] Meyer, E.S., et al. (2005), Fatigue cracking in structure dominated by compression loading, Proceedings of the 23rd ICAF Symposium, vol. II, pp. 473479, Dalle Donne, C. (Ed.), DGLR-Bericht 2005-03. [4] Mills, T.B., et al. (2002), Managing Damage in the wing: Modeling the interaction of exfoliation with static and fatigue loads, Proceedings of the 6th joint NASA/FAA/DoD Conference on Aging Aircraft. [5] Liao, M., Renaud, G., Bellinger, N.C., Backman D., Forsyth D.S. (2005), Effects of exfoliation corrosion on static and fatigue behaviour of aircraft materials and structures – testing and modelling studies, Proceedings of the 8th joint NASA/FAA/DoD Conference on Aging Aircraft. [6] Harter, J.A. (2008), AFGROW Users Guide and Technical Manual, http://www.afgrow.net/. [7] Sharp, P.K., Mills, T.B., Clark, G. (2001), Modeling of fatigue crack growth from pitting and exfoliation corrosion, Proceedings of the 21st ICAF Symposium, vol. I, pp. 485-498, Rouchon, J. (Ed.), Cépaduès Editions, Toulouse.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

HELICOPTER STRUCTURAL INTEGRITY PROGRAM OF POLISH Mi-24 HIND HELICOPTERS Sławomir Klimaszewski, Andrzej Leski, Krzysztof Dragan, Marcin Kurdelski, Mirosław Wrona Air Force Institute of Technology, Poland

Abstract: In this paper selected problems connected with developing Helicopter Structural Integrity Program (HSIP) of Polish Army Mi-24 Hind Helicopters have been presented. In the paper abilities of IT system, the SAMANTA system used in Polish Armed Forces for collecting and analysing operational and maintenance related information have been discussed. Advantages from application of relational data base technology implemented in the SAMANTA system and based on the Oracle platform have been shown. In order to develop FEM model of Mi-24 structure reverse engineering (RE) methods have been used. The RE methods made possible helicopter structure to be digitally reconstructed and transferred to FEM software. Two of RE methods have been used: - a digital photogrammetry method by means of TRITOP system, - 3D scanning by means of ATOS III system. Service loads spectrum is obtained from operational loads monitoring program. The stresses and loads at selected points where measured during several test flights. Moreover the procedures of helicopter structure fatigue life estimation based on safe S-N curve concept and crack growth concept has been included in the paper. A modified NDE/NDI program is developed in order to support HSIP. Special attention was paid on detection of subsurface defects such as hidden corrosion and cracks by means of DSight Aircraft Inspection System (DAIS II), Magneto-Optic Eddy Current Imager (MOI-308) and MAUS V C-scan system. M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 263–277. © Springer Science+Business Media B.V. 2009

264

S. Klimaszewski, A. Leski, K. Dragan, M. Kurdelski, M. Wrona

Above mentioned HSIP was developed by Polish Air Force Institute of Technology under research project founded by Polish Ministry of Science and Higher Education and Polish Ministry of National Defence.

INTRODUCTION Conditions of Polish HSIP The majority of military aircrafts used in Poland come from the former Soviet Union. The term ASIP, in reference to aircrafts used in Poland, formally did not exist until the purchase of F-16 aircrafts. The existing principles do not describe in a precise and detailed way tasks and responsibilities connected with structures integrity of aircrafts as it is described in such normative documents as [1], [2]. However, undertaken tasks in order to assure safe maintenance of aircrafts have many common features to those used by other NATO countries. Thus, the terms ASIP (HSIP) concerning polish aircrafts should not be understood as implementation of particular documents analogical to [1], [2]. The terms have been used analogically particularly to activities described in Task V of ASIP [1]. The aims of HSIP Mi-24 program in Poland Helicopters Mi-24 are used in Polish Air Force since 1978 (D version) and since 1986 (W version). Basically, helicopters are maintained according to the principle safe life. The basic service time (calendar service life) given by the manufacturer is 26 years. Permissible flight hours in this time is 4000 hours of flying. Serviceoverhaul system expects conducting overhaul of a helicopter every 7 years or 1000 flight hours. Creating HSIP program for Mi-24 in Poland aims at increase of exploitation safety of these helicopters in the context of their aging. Experiences gathered so far are a good base for extending the previously planned service life of these helicopters in the Polish Air Force. The funds for works on HSIP Mi-24 program come from the Ministry of Defense and Ministry of Science and Higher Education (research grant).

THE SYSTEM OF MAINTENANCE AND OVERHAUL INFORMATION COLLECTION The process of gathering, processing and analyzing the information about Mi-24 damages as well as their maintenance data, analogically to other military aircrafts, is conducted with the use of IT systems such as SAMANTA and NEFRYT. SAMANTA system is intended, among other tasks, to gather, process and analyze the data about damages of all types of aircrafts exploited in Polish Air Force. Thanks to the process of verification and unification of the information about damages which come from different sources a uniform data bank was created which identifies:

Helicopter structural integrity program of Polish Mi-24 Hind helicopters

1. • • 2. •

265

the object (aircraft): the type and the version of the aircraft, the user of the aircraft (number of the Military Unit), the summary of work to damage: since the beginning of maintenance: - the number of hours, - the number of landings, • after overhaul, - the number of hours, - the number of landings, 3. the circumstances of damage appearance • the date of damage, • the circumstances of damage symptoms, • the conditions of the flight, • who found the damage, 4. the effects for flight safety, 5. the effects for tasks conducting by the crew, 6. the way of damage fixing, the time of fixing, how work consuming and the time of not using the aircraft because of this damage, 7. the element of the aircraft which was damaged, 8. the symptoms of the damage: • the code and description of the damage, • the code and description of the symptoms of the damage of the device or unit, • the code and description of the element damage, • the causes of the damage: • the code and description of the damage causes, • damaged unit or device: • the serial number, • the year of production, • the number of overhauls, • the work since the beginning of maintenance, • the work after the last overhaul, • the code of work hours, 9. maintenance data of the damaged engine: • the code of the version of the engine of the particular type, • the serial number of the engine, • the time of work since the beginning of maintenance, • the number of the last overhaul, • the time of work after the last overhaul. The gathered data about Mi-24 helicopters damages are from 1979. The gathering of the data at the beginning meant writing the information down on a paper in a form of a description. Later on, the information coding systems were introduced.

266

S. Klimaszewski, A. Leski, K. Dragan, M. Kurdelski, M. Wrona

SAMANTA system (based on Oracle) for gathering and processing the information has been used since 1996. Operation data collection system Polish Air Force, Army and Navy, uses the unified system SAN to collect, process, and analyze the operational and maintenance related information regarding military aircraft [2]. SAN system consists of: • main system modules with subsystems for recording and processing the aircraft operation data, • cumulative operation data bank with processing subsystems, • central operation and reliability data bank with processing subsystems, • module for analysis of military aircraft operation and reliability rate, The system allows : • to collect, process and transmit data regarding the process of military aircraft operation, according to regulations on reporting procedures of the Polish Armed Forces, regulations of aviation-engineering service and military aircraft onboard documentation, • to inspect the military aircraft operation processes including current and periodical tracking of their technical condition, also investigate the readiness and reliability. As a result of verification and unification of information regarding the damages of Mi-14 helicopters which were stored in older operation data collection systems, the coherent data bank was achieved. Following items were identified: 1. Object (aircraft): • aircraft type and version, • aircraft user (Military unit number), 2. Total amount of work performed before failure: • since commissioning: - how many hours, - number of landings, • since maintenance, - how many hours, - number of landings, 3. Circumstances causing the failure • date of failure, • circumstances of failure symptoms noticed, • flight conditions, • who found the failure, 4. Flight safety consequences, 5. Consequences for task performance by the crew, 6. Way the failure was repaired, repair time, labor demand and down-time, 7. Faulty aircraft part, 8. Failure symptoms:

Helicopter structural integrity program of Polish Mi-24 Hind helicopters

267

• code and description of aircraft failure symptoms, • code and description of failure symptoms regarding the device or unit, • code and description of part failure, 9. Cause of failure: • code and description of failure cause, 10. Faulty device or unit: • serial number, • year of production, • number of repairs, • work units since commissioning, • work units since last repair, • code of work unit, 11. Operation data of a faulty engine: • version code of the particular engine type, • engine serial number, • work time since commissioning, • last repair code number, • work time since last repair. Recorded data are used to prepare analyses, forecasts and reports. Predictions of reliability and safety factors is an addition to their statistical analyses. Typical forecast time is a two year period. Predictions are carried out using the following techniques: statistical forecasting, analysis of time sequences and artificial neural networks. The statistical analysis of damages of Mi-24 airframe was elaborated on the basis of the information about damages which were recorded in the SAMANTA system and its predecessors. In the time period analyzed, from 1979 to 2006, 531 damages of the Mi-24 helicopters airframe were noted down, which is 23% of all recorded damages.

Fig. 1. Percentage setting-up of the number of damages of the particular elements of Mi-24 helicopter

268

S. Klimaszewski, A. Leski, K. Dragan, M. Kurdelski, M. Wrona

The Figure 1 presents the percentage share of helicopter elements in the global number of helicopters. Figure 2 shows the bar chart of average flight hours to damage Tu of the helicopter structure

Fig. 2. The average flight hour to damage of Mi-24 helicopter structure

MONITORING OF OPERATIONAL LOADS General notes One of the tasks of HSIP program is to monitor operational loads. Monitoring of operational loads for helicopters not equipped in digital flight data recorder is a difficult task to perform. In practice it often means examining the profile of helicopter maintenance understood as percentage setting-up of shares of particular maneuvers per 100 flight hours. [The rotary wing aircraft usage spectrum consists of maneuver durations and occurrences per 100 flight hours]. Indicating individual operational profiles for helicopters not equipped in digital flight data recorder is a very work consuming and expensive task. Indicating an average operational profile is a more economical approach. The methodic of indicating usage spectrum of Mi-24 maintenance Usage spectrum indication of Mi-24D and Mi-24W is based on/means: • Indicating a group of tasks / exercises, on the basis of an aerospace training program, which characterize in similar service loads values; • Choosing representative flights of Mi-24D and Mi-24W helicopters for each group of tasks / exercises; • Decoding flight parameters recorded by flight recorder SARPP-12G from representative Mi-24D and Mi-24W helicopter flights. (SARPP-12G recorder is an old generation recorder – information is written on a photo film) • Decoding was conducted on the basis of : - photo film of SARPP-12G recorder,

Helicopter structural integrity program of Polish Mi-24 Hind helicopters

269

- scaling graph of information transmitters. • Building data files of representative flights for each exercise group with one second resolution on the basis of decoded flight parameters values recorded by the flight recorder SARPP-12G; • Dividing flight maneuvers representative for each exercise group into clasess of helicopter airframe structure, where the criteria of division are conditioned by the values of recorded parameters; • Classifying all flights to exercise groups; • Standardizing the time of exercise conducting to the time of flight for each exercise group; • Indicating for each helicopter number percentage share of particular flight maneuvers normalized to 100 flight hours; • Indicating for the whole helicopters population percentage share of flight maneuvers normalized to 100 flight hours – this share is the average profile of Mi-24D and Mi-24W maintenance. Forces and deformations measurements in flight All the discussion about the durability of helicopter or usage spectrum assessment should be preceded by the stress in structure analyses. In order to indicate loads of main structural components of Mi-24 helicopters during different flight phases, flight tests of internal loads and strain in chosen structure places are currently conducted (while this article is being written). Mi-24 Flight test instrumentation. Measurement devices building was made in Military Workshops WZL-1 during a helicopter overhaul. This solution lowered the costs of research a lot and shortened the time of helicopter operational unavailability. The concept of measurement includes strain gauges measurements in approximately 70 points of the structure. The measurement device consist of: • KAM-500 (ACRA Control) recorder, • 72 strain gauges, • Sensors to measure shock absorber line feed, • Nz sensors. Some of strain gauges sensors were built in such a way as to allow the measurement of internal loads in helicopter structure. The remaining sensors were used to local measurement of strain. The general strain gauges sensors distribution is presented on the figure 3.

270

S. Klimaszewski, A. Leski, K. Dragan, M. Kurdelski, M. Wrona

Fig. 3. Measurement points spread [strain gauges sensors]

The measurement of internal loads and strain The measurement of internal loads and strain was conducted during research flights. The program of research flights was especially prepared in such a way as to include all helicopter flight states during maintenance (defined during helicopter operational profile). Before the research flights, physical scaling of chosen measurements channels were made. Elements possible to disassemble assigned to measurement were scaled with the usage of fatigue test in ITWL laboratory.

Fig. 4. Scaling of angle strut of main gearbox support

Helicopter structural integrity program of Polish Mi-24 Hind helicopters

271

Helicopter structure was also scaled. Loads were put to tail bone, landing gear, stabilizer and wings of helicopter in a controlled way. The scaling results make it possible to interpret the results gained during the research.

DEVELOPING OF THE FE MODEL OF THE HELICOPTER STRUCTURE Initial notes Computer model made in FE technique has primary meaning for different types of calculations and assessments. Thanks to FE calculations the distribution of stress field and displacement of all helicopter elements under the work of various loads is possible. In the global model, which embraces the whole helicopter body, the introduction of a number of simplifications is necessary. In order to gain proper solutions for smaller structure elements, the local models are used. In such a case the global model is a source of boundary conditions for local models and in many cases the global model is necessary to conduct local models analyses correctly. Currently, the global model of aircrafts is created during the process of construction. In the case of structures designed in the last decades, building computer model is the effect of reverse engineering. Having technical documentations of an aircraft it is possible to built a computer model. In case of a lack of such a documentation, the building of a model requires a number of measurements of a real object. These measurements include the external shape as well as arrangement of elements inside the structure of an aircraft. It is also necessary to weight different types of elements, which weight cannot be indicated on the bases of the shape (for example units). The process of creating computer model without good technical documentation is difficult and not effective. Building the model of Mi-24 was conducted on the basis of a measurement of a real object. The following phases of work may be distinguished: • the measurement of helicopter geometry, • the built of shell model, • the built of FE model. Photogrammetry measurements Digital photogrammetry technique and optical 3D scan were chosen to measure the shape of a helicopter. Integrated measurement systems were used and made by the following manufactures: GOM: TRITOP – photogrammetry measurement and ATOS III – optical 3D scan. The basis of photogrammetry technology is substituting the direct measurement of an object by the measurement on photos, which in reality are the reflections of an object’s middle projection on the plane. The advantages of this method are: • Transferring some of the work from making a series of photos taken in a proper way to work conducted outside the place of an object of measurement; • Keeping all the photographical documentation of an object has archival meaning, as well as enables an extra measurement when needed,

272

S. Klimaszewski, A. Leski, K. Dragan, M. Kurdelski, M. Wrona

• Possibility to determine the shape and the features of objects inaccessible to direct measurement. As a result of fuselage photogrammetrical measurement, it was possible to acquire the space arrangement of 3920 points. 2,5 m long patterns were used during the measurement. Mid error of patterns length indication was 0,0336 mm, and mid error of measurement in pixels was 0,0993 pixel. Points easy to unique identification as well as crucial from structural point of view were chosen for the measurement. In particular, these points were registered which position reflects run of bulkheads, frames and stringers.

Fig. 5. Elaboration of measurement results – system TRITOP

The structure of fe model of helicopter The proper modeling, in the sense of describing the fuselage’ skin shape with the planes spread on measured points, was conducted with the usage of the CAD program. Figure 6 shows the middle phase of work on computer model of middle part of Mi-24 helicopter fuselage.

Helicopter structural integrity program of Polish Mi-24 Hind helicopters

273

Fig. 6. Creation of the computer model of Mi-24 helicopter – middle phase

On the basis of geometrical data, global helicopter model, its shell and beam FE model was delivered. During creation of FE model some of the solutions offered in other helicopters, such as AH-64A (McDonnell Douglas Helicopter Co.) [3] i UH60A (Sikorsky Aircraft) [4], were used. The tool used for building this model was MSC.Patran program. Figure 7 presents MES model of Mi-24 tail boom.

Fig. 7. MES model of helicopter’s tail boom

Numerical model of the back part of a helicopter was created by the connection of tail beam and tail rotor pylon. The base for the models were photogrammetrical measurements. Similar to real structure the model consists of longitudinal and

274

S. Klimaszewski, A. Leski, K. Dragan, M. Kurdelski, M. Wrona

vertical frames and stringers as well as skin. The model was divided into groups reflecting technological division of the structure and into groups which come from differentiation of material properties. Due to the size and complicated structure, global model was made with in a moderately detailed way. Riveted and bonded skin links as well as smaller holes were not taken into consideration. Sub-elements which do not take place in load transition, but important from the inertial point of view was modeled in approximate way. Material density declared in properties of these sub-elements was taken in such way to be in accordance with real summary weight of that elements. In the model there is majority of four-node shell elements QUAD4 type. Construction reinforcements and such as bulkheads and transmission box was modeled with the use of brick element HEX8 type. Beam elements BAR2 was designed during reconstruction of longerons and horizontal stabilizer. In places where it was possible regular mesh was applied with the use of IsoMesh. In the places of complicated geometry finite elements was made manually.

MODERN NON DESTRUCTIVE EVALUATION TECHNIQUES IMPLEMENTATION Corrosion Assessment of Mi-24 Helicopters On the base of data analysis the following areas which could be intensively endangered by the corrosion were determined: • substructure skin, fuselage skin in the hot temperature affecting area in proximity of engines; • skin of the front, middle and rear section of the fuselage; • refueling points; • battery container; • attachment points of main landing gear; • attachment points of front landing gear; • cantilevers for attaching of cables anti-icing installation for main rotor blades; • cantilevers for attaching of cables anti-icing installation for tail rotor blades; • blades of rotor fan; • leading edge of horizontal stabilizer; • cantilevers of mounting of horizontal stabilizer; • attachments point of tail skid angle struts; • attachments point of shock absorber skid angle struts • attachments point of engines; • undercarriage strut lever; • skin of landing gear housing; • pull roads of steering landing gear mechanism; • landing gear bay; • attaching point of undercarriage strut lever;

Helicopter structural integrity program of Polish Mi-24 Hind helicopters

275

• wing skin; • attaching points of wing to fuselage. For the corrosion assessment of fasteners (pillowing phenomena) the system DAIS is used [5]. After the assessment of number helicopters population the following conclusions can be made: • in the selected areas corrosion required repair work was not detected; • the most corrosion vulnerable area of the helicopter was the tail boom. Non Destructive Testing of airframe elements The use of nondestructive testing (NDT) for the aerospace application goes from the necessity of assuring of flying safety, reliability of work and high technical readiness of the aircrafts. Moreover economical reasons influence applying NDT during aircraft maintenance. These techniques enabling technical assessment of multiple elements with no or limited access to visual inspection. NDT are able to detect and determine diversity of failure modes which occur during aircraft maintenance. From that reasons use of NDT is very important during aircraft maintenance in accordance with condition based maintenance. There is no general technique. Existence of few techniques is the best example of that assumption. The most often used techniques are: visual, penetrant, magnetic particle, ultrasonic and eddy current. Nondestructive testing are performed according to the following scheme: 1. Preliminary testing. Preliminary tests of inspected element and area around the element. 2. Tests with the use of basic testing technique. Tests with the use of basic method and data collection and evaluation. 3. Tests with the use of additional technique. Tests to confirm indication received during the inspection with the use of basic technique. At present more often automated systems are used for data collection which enabling much faster inspection. These systems use digital data collection which enabling data comparison of received data. One of the example of such system is the DAIS™ system (Fig. 8). It is used for corrosion assessment based on pillowing phenomena. One of other examples is the MAUS® system used for automated eddy current and ultrasonic inspection (Fig. 9).

276

S. Klimaszewski, A. Leski, K. Dragan, M. Kurdelski, M. Wrona

Fig. 8. Inspection with the use of DAIS system

Fig. 9. Inspection with the use of MAUS system

Use of such systems considerably decrease time of inspection with the keeping of accuracy. However for the helicopters there is many of raised head fasteners which makes automated data collection difficult for some places. Solution could be manual hand scanning but with the use of data collection. Most of the failure modes detected in the Mi-24 helicopter structure was cracks, corrosion and deformation. These damages were detected with the use of several techniques. Moreover so called ‘hot-spots’ were determined affecting structural integrity. For the purposes of monitoring of the technical condition NDT service bulletins were elaborated in the AFIT.

SUMMARY AND CONCLUSIONS Mi-24 helicopters used In the Polish ARMY are maintained on the basis of national overhaul and R&D background. Unfortunately common modernization plans of the Mi-24 helicopters by the former Warsaw Pact (at present NATO members and such as Poland, Czech Republic and Hungary) has failed. Created rules for helicopters maintenance as well as associated with this research programs are only Polish activity. The main purpose of that activity is to keep maintenance of that helicopters for the longer time. In the article the main activities of the Air Force Institute of Technology has been presented to supply structural integrity of Mi-24 helicopters. Many of activities has got similar applications in other programs introduced in Armed Forces of other nations with relation to aircraft used in that countries. Authors are not familiar if there are any other similar activities related to Mi-24 helicopters in other countries. Program realization brought the following conclusions so far: • Mi-24 helicopters are durable structure;

Helicopter structural integrity program of Polish Mi-24 Hind helicopters

277

• aging of helicopters may cause corrosion problems however at that stage of helicopters service life that problem is not crucial; • there are exist potential reserves of fatigue durability for the Mi-24 helicopters used in Poland (small flying time, medium intensive use).

REFERENCES 1. 2. 3.

4.

5.

6.

7.

Defense Standard 00-970, Design and Airworthiness Requirements for Service Aircraft, Ministry of Defence, Issue 2, Dec 1999. Air Force MIL-STD-1530C Aircraft Structural Integrity Program (ASIP) Christ R., Ferg D., Kilroy K., Toossi M., Weisenburger R., Plan, Formulate, and Discuss a Nastran Finite Element Model of the AH-64A Helicopter Airframe, McDonnell Douglas Helicopter Company, Mesa, Arizona, October 1990. Dinyovszky P., Twomey W. J., Plan, Formulate, and Discuss a Nastran Finite Element Model of the UH-60A Helicopter Airframe. United Technologies Corporation, Sikorsky Aircraft Division, Stratford, CT 06601, February 1990. Klimaszewski S., Leski A., Zurek J., The Role of AFIT in the Polish Aging Military Aircraft Programs, Proceedings of 7th Joint FAA/DoD/NASA Conference on Aging Aircraft, New Orleans, 9-11 September 2003. Klimaszewski S., Leski A., Zurek J. Helicopter Structural Integrity Program of Polish Navy Mi-14 Haze Helicopters. Proceedings of ASIP 2004 Conference. Memphis 29.11-03.12.2004. Dragan K. NDE activities connected with Service Life Extension of Main Rotor Blades of Helicopters used in Polish Armed Forces. 7th Australian Pacific Vertiflite Conference on Helicopter Technology, 9 - 12 March, 2009, Melbourne.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

AIRFRAME LIFE EXTENSION BY OPTIMISED SHAPE REWORKING – OVERVIEW OF DSTO DEVELOPMENTS M. Heller1, M. Burchill1, R. Wescott2, W. Waldman1, R. Kaye1, R. Evans1, M. McDonald1 1

Defence Science and Technology Organisation, Australia 2 QinetiQ AeroStructures, Australia

Abstract: The Defence Science and Technology Organisation Australia (DSTO) has developed a unique life extension approach, where optimised rework shapes are designed and implemented at critical airframe features. These shapes are used to remove local cracking, and also to minimise the peak stresses. The free-form optimal shapes are obtained using an iterative gradientless finite-element method, which is based on an analogy with biological growth. This paper covers the robust design approach, lessons learned from practical applications, fatigue trends, and improved and efficient in situ manufacturing methods. An important aspect of the this work has been the transition from theory and generic developments to practical applications. It is shown that large increases in fatigue lives and inspection intervals can be achieved with appropriate design.

INTRODUCTION Typically during the working life of an airframe, a few key stress-concentrating locations can become fatigue critical, and an effective repair option is needed. An appropriate repair can provide significant economic benefits, by avoiding the need for component replacement, and/or typically increasing the interval between costly periodic in-service inspections. M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 279–299. © Commonwealth of Australia 2009. Published by Springer Science+Business Media B.V. Dordrecht.

280

M. Heller et al.

The Defence Science and Technology Organisation Australia (DSTO) has developed a unique life extension approach, where optimised rework shapes are designed and used. Two distinct scenarios for optimal shape reworking are: (i) a repair, where damaged material is removed and (ii) as a pre-emptive repair measure to avoid cracking. In both cases, apart from removing material, the optimised shapes provide significantly reduced stress peaks, thus typically leading to increased fatigue life, inspection intervals and airframe availability. For example Figure 1 shows the concept for optimised shape reworking, for the crack repair scenario. It can be seen that in this case the shape of the traditional rework is significantly different to the optimised shape. The optimisation approach is based on an analogy with biological growth, and is implemented as an iterative gradientless finite element procedure, [1-3]. The optimal shapes are unique and depend on local loading and practical geometric constraints, [4-11]. An important aspect of the DSTO work has been the transition from theory and generic developments to practical applications. Here, lessons learnt from the practical experience have also been used to enhance the theoretical methods, leading to an improved practical capability. It is important to note that apart from minimising peak stress, the design requirements for optimal rework shapes can be significantly different to the case where a new component is being designed (which is more often addressed in the literature). For example, in the rework context: (i) the optimal hole must be larger than the existing hole; (ii) different manufacturing constraints operate, because the rework must typically be done in situ; (iii) the shapes need to be robust to some load direction uncertainty and/or manufacturing misalignment, and/or geometric variability in the aircraft fleet; (iv) inspection issues may need to be taken into account; and (v) weight saving is usually irrelevant. Hence, in the present paper the following work is presented. First in Section 2 the numerical method for shape optimisation is given, which accounts for multiple stress peaks around the stress concentrators, and consider stress robustness and geometric constraints. In Section 3 a key practical application is presented involving reworking of critical features of the wing pivot fitting (WPF) in the General Dynamics F-111. Section 4 then considers generic fatigue life trends for optimised reworks, relevant to safe-life and safety-by-inspection philosophies. Methods for post-processing FEA co-ordinates to achieve improved manufacture are discussed in Section 5, while Section 6 focuses on new developments in compact in-situ machining. Then Section 7 summarises a range of other generic and practical investigations, while issues related to transitioning the technology are given in Section 8.

Airframe life extension by shape optimisation – Overview of DSTO developments

281

NUMERICAL METHOD FOR SHAPE OPTIMISATION The stress optimisation approach used here is an iterative gradientless approach, yielding free-form shapes (fully automated). It is used since it is simple to implement, readily amenable to extension, and has been shown to achieve very good solutions for stress minimisation for realistic geometrically constrained problems, which is often not a focus of other investigators. The approach is based on the aim of achieving a minimum peak stress by directly attempting to make the stresses uniform along the longest segment of a stress-concentrating boundary. This idea is well known, and has been successfully used in various analytical [12,13], numerical [2,7,14] and experimental [15] studies. The different numerical algorithms have various advantages and disadvantages in their implementation. The present approach is to add material to local regions of high stress and/or remove material from local regions of low stress on the stress concentrator boundary. Early work typically focused on minimizing a single dominant stress peak (tensile or compressive) on boundary segments of stress concentrators [1,4,5]. More recently the approach has been extended to multiple peaks [3,16] since the most fatigue critical location on a hole boundary may not be the one with the greatest tensile peak. In DSTO work both generic and unique optimal shapes have been determined, for example a typical unique shape is shown in Figure1b, while two generic shapes are given in Figure 2, with different aspect ratios h/w. Multi-peak method Consider a hole in a remotely loaded large 2D plate, as shown in Figure 3a. There are k nodes distributed around the hole boundary, and the local tangential stress at the i-th node is σi. The optimisation is subject to a minimum radius of curvature constraint ρ ≥ ρmin, and an optional geometric bounding constraint consisting of a polygon with m sides. The former is activated to ensure that sharp corners, which are commonly found in ideal optimal shapes [3,11] do not develop. The peak stress distribution around the hole is shown in Figure 3b. There are four distinct stress peaks (i.e. σp1, σp2, σp3 and σp4), which in general are not equal to one another, together with four zero crossings. The zero crossings are used to identify the set of n subregions of positive and negative stress, each subregion lying between a consecutive pair of zero crossings. Consistent with the work of Vigdergauz and Cherkayev [13], as well as prior semi-automated multi-peak shape optimisation [7], it is taken that in general a constrained optimal shape will have n zones of peak uniform stress. During each iteration, the stresses are computed at all the nodes around the hole boundary. Then, the i-th boundary node, which is located in the qth stress subregion, is moved in the direction of the local outward normal to the boundary by a distance d iq , given by: ⎛ σq − σq th ⎜ i d iq = ⎜ q ⎜ σ th ⎝

⎞ ⎟ ⎟s ⎟ ⎠

q where σ th = max σ iq , q = 1,2,..., n

(1)

M. Heller et al.

282

where σ iq is the tangential stress at node i in the q-th stress subregion, σ thq is the stress threshold corresponding to the peak positive or negative stress occurring in the q-th stress subregion (updated each iteration), s is a constant step-size scaling factor. Hence, at each iteration only material at low stress regions is removed. This process is repeated until the optimal is reached. This occurs when the stresses in each subregion converge to a reduced constant value, within some prescribed tolerance, over the maximum possible length of arc as: n

μT = ∑ μ q q =1

( μ T ≤ 1)

(2)

Here μT represents the normalised total combined length of the constant stress zones around the hole boundary. Special algorithms are also used to avoid mesh distortion. All analyses are linear elastic. Extension to in-built robustness The aim here is to determine a 'robust' notch shape that renders the peak stress constant and minimal for a range of load orientations, along a boundary segment. It is useful to initially explain this for only one subregion for the 2D case. Figure 4a shows an arbitrary hole in a plate, subjected to a remote uniaxial load that can be applied over a range of angular orientations. The angular orientation, α, of that load is discretised into an appropriate number of individual load cases l = 1...m, where l is the load case identifier and m is the total number of load cases. For an arbitrary initial hole shape, the stress distribution σl,i around the hole, as well as the magnitude and position of the peak stress, will usually be different for each load angle, as illustrated in Figure 4b. The locus of the maximum peak stress distribution σ imax, q = max σ lq=1…m,i for all load cases is also shown in Figure 4b. To render the peaks minimal and constant, we can calculate the nodal movements in a similar fashion to the basic method in Eqn. (1), except here the stress term σi is replaced with the locus of the maximum stress term σ imax, q . Hence, for the q-th subregion, at each iteration the boundary node movements for the robust stress method are given by: q ⎞ ⎛ σ max, q − σ th ⎟ s, where σ q = max(σ imax,q ), q = 1,2,…, n diq = ⎜ i th q ⎜ ⎟ σ th ⎝ ⎠

(3)

Figure 5 shows the typical results from optimisation where the range of the load angle variation is changed. It can be seen that there are differences in shape in Figure 5a. The key result shown in Figure 5b is that within the given load orientation design range, the peak stress does not degrade for the robust hole, when the load orientation, α, is perturbed. Conversely when the applied load orientation is perturbed for the standard optimal (which is designed for α = 0) , there is a significant stress increase.

Airframe life extension by shape optimisation – Overview of DSTO developments

283

From a manufacturing perspective, it is desirable to have the same 2D shape throughout the plate thickness. However in the optimisation analysis, the 3D variation of stress through the plate thickness can be taken into account, see Figure 6. Hence at any angular position i around the hole boundary, the maximum nodal stress σ imax,q, j = max σ lq=1…m, j =1…v ,i through the thickness, j = 1…v, is used as the basis for the optimisation calculation, [3,16]. The nodal movements at a given angular position i, are then given by: q, j ⎛ σ max, q, j − σ th d iq, j = ⎜ i q, j ⎜ σ th ⎝

⎞ ⎟ s where σ q, j = max(σ imax,q, j ), q = 1,2,…, n and j = 1,2,…, v th ⎟ ⎠

(4)

EXAMPLE - OPTIMISATION OF F-111 WPF FEATURES Background The critical stress concentrating features in the WPF are managed via safety-byinspection due to the possibility of manufacture-induced flaws in the high strength, but flaw intolerant, steel which is used. This has resulted in a significant ongoing maintenance burden due to the regular inspections that are required. The key to easing the maintenance costs is through reducing the frequency of the inspections, i.e. by increasing the inspection interval. Shape optimisation via material removal only was selected as the repair approach based on; ease of implementation, lower long-term maintenance costs, lowest risk, and the ability to increase flight envelopes [7]. A local region of the finite element model used to undertake the final development of the optimal shapes (semi-automated) is shown in Figure 7a. For each wing, four optimal stiffener runouts (SROs) were determined, along with four optimal fuel flow vent holes (FFVHs). Typical optimised FFVH shapes are shown in Figure 6b. Also as an example, initial and machined optimised FFVHs are shown in Figure 8a and 8b respectively. Here the optimal provides a peak elastic stress reduction at the critical location off 53%, (it offers a 38% reduction as compared with a nominal rework). A typical optimised SRO is shown in Figure 8c. Lessons learned Large stress reductions were predicted by using the optimised shapes, where only material removal was allowed. The stress reductions have been demonstrated by full-scale static strain surveys, and fatigue tests have shown that the optimal shapes are highly durable [17] and can tolerate significant corrosion and mechanical damage [18]. Overall, the scope for significantly increased inspection interval, and/or reduction of continued crack growth has been predicted [19] and demonstrated [17,18]. A complex but robust manufacturing process was devised which involved an electrical discharge machining process to achieve: (i) material removal with minimal heat affected zone; (ii) in situ measurement of the WPF geometry to account for individual dimensional variations between WPFs; and (iii) a custom-built support arm to position the probes and machining instruments deep

M. Heller et al.

284

inside the structure via restrictive access holes. To date a total of 96 stress concentrating features have been reworked with only a few requiring additional engineering assessment. Some positional tolerances could not be met due to larger-than-expected variations in the geometry between various WPFs. A more comprehensive survey of the variations in WPF geometry in the fleet may have lead to a more robust and flexible manufacturing process, however this would have been difficult to achieve without affecting the availability of aircraft. Even though the optimal shapes were designed to be robust to deviations, and that a relatively flexible manufacturing process was devised, (e.g. in situ measurements), it is evident that some additional effort to further increase these aspects may have mitigated some of the additional engineering assessments that were required. While the optimal shapes were developed to minimise the amount of material removed, a number of size-related issues remained, including: (i) the electrical discharge process was time consuming, with a single rework taking around 8 hours to machine; and (ii) due to the magnetisation process, it was difficult to use the preferred non-destructive inspection (NDI) method (magnetic rubber inspection) if the optimal shape was large. Hence these lessons have focussed research efforts, as discussed in the subsequent sections

FATIGUE LIFE TRENDS Key aspects as they relate to the estimation of fatigue lives in the context of optimised reworks are discussed below. 4.1 Safe-life approach The result of calculating the total life under the safe-life approach is the stress-life curve, which is shown schematically in Figure 9a. One approach to derive this curve is by coupon testing. Hence a simple empirical expression for rewriting the stress-life relationship under a given flight spectrum is: ln(σK t ) = Α ln( t t ) + Β

(5)

Here A and B are derived constants, σ is a reference remote stress level for the variable-amplitude spectrum of interest, tt is the total life. Hence shape optimisation is used to increase the total life by minimizing the geometric concentration factor Kt. This is shown schematically in Figure 9a, where the decrease in Kt from Kt1 to Kt2 results in an increase in total fatigue life from tt1 to tt2. The quantitative relative safe-life benefits can be estimated by re-arranging Eqn. (5) and evaluating for Kt1 and Kt2. This leads to: 1/ A

tt 2 ⎛ K t 2 ⎞ ⎟ =⎜ tt1 ⎜⎝ K t1 ⎟⎠

(6)

Airframe life extension by shape optimisation – Overview of DSTO developments

285

In Figure 9b both generic and specific [16,20] values of A are used to show the generic trends of the relative life increase. The benefits of reducing Kt are clear. Safety by inspection The basis for calculating the residual fatigue life under a safety-by-inspection, approach is the crack growth curve as shown schematically in Figure 10a. The residual life, tr, is taken as the time in flight hours to grow the fatigue crack from the NDI maximum undetectable size andi to a critical size ac. Shape optimisation typically increases the residual life by reducing the stress intensity factor for cracks up to certain size. This effect is highlighted in Figure 10a. There is a relatively higher crack growth rate for smaller crack lengths for the non-optimal case compared to the optimal case. However as the crack length increases, a stage may be reached where the growth rate for an optimal shape is greater than the corresponding nominal hole. A convenient approach to investigate the generic trends for crack growth lives, obtained by minimizing Kt, is the effective block approach [21]. The essence is to derive crack growth data from coupon tests under a representative variableamplitude flight spectrum for a particular location. An equation of the following form is fitted: da / dN = C ( K ref )m ,

where

K ref = σ ref β test π a

(7)

where: (i) C and m are experimentally determined constants; and (ii) Kref is the reference stress intensity factor, σref is the reference stress, and β test depends on the crack geometry and size. Taking crack growth data for aluminium 7050 coupons, under an F/A-18 FT55 spectrum [21], we have C = 1.5726E-9, m = 2.14, and σref = 100 MPa. Consider an idealised case of a circular or optimised hole in a large uniaxially loaded plate with a through-crack. Taking the stress decay from the hole edge along the prospective crack path, σy for the corresponding uncracked body, the stress intensity factor for the cracked case can be estimated as: K = 1.12(σ y / σ remote )σ ref π a

(8)

Hence from Eqns. (7) and (8), we can determine the life for circular and optimal holes having different Kt values. This is done by varying the length of the optimal hole in a large plate, while the minor axis of all holes is kept the same, i.e. w = 20 mm. The crack growth life trends are shown in Figure 10b. As expected the greatest benefits are achieved for smaller crack lengths. It is clear that depending on the Kt reduction there are large benefits for asset preservation using shape optimisation, and in some cases the inspection intervals can be doubled. Similar trends occur when a more realistic initial corner-crack is assumed instead of a through-crack, [16]. Very large reductions in crack growth rates have also been

M. Heller et al.

286

predicted by other means for F-111 fleet aircraft [19], and have also been demonstrated during full scale fatigue tests, [18]. Optimal hole stress decay analogy with elliptical holes For the fracture analysis of plates containing optimised holes the stress decay of elliptical holes is very useful. In recent unpublished work by the authors at DSTO, it has been shown, for a wide range of cases, that the stress decay at optimal holes and elliptical holes in large plates is the same if the peak stresses are equivalent. (and they are under the same plate finite width conditions). Of course their aspect ratio’s and local radius of curvature will be different. It is noted that if instead the local radius of curvature is matched, then the stress decay is poorly matched. The geometry and notation for the elliptical hole case is shown in Figure 11a. The stress decay, σy, for an elliptical hole in a large plate loaded by a uniaxial remote stress, σ r , is given by: σ y = γK t σ r

where Kt is the peak stress concentration factor at the hole, and

(8)

γ is the normalised

stress decay at a distance x′ from the edge of the hole, which is given, [22], by :

γ =

1 + ( w / h) + ( w / h) tanh ξ (( w / h) 2 − ( w / h) − 3 + (tanh ξ ) 2 ) (1 + 2( w / h))(1 + ( w / h))(1 − ( w / h)) 2

(9)

in which: ξ = asinh(( x′ + w) / c) and c = h 2 − w 2 Figure 2 shows two different optimal holes having approximately the same Kt (i.e. Kt = 2.15 for the 2.1:1 optimal and Kt = 2.17 for the 1:1 optimal). Figure 11b shows the stress decay of these holes compared to an ellipse with a Kt of 2.17. The agreement is very good. This demonstrates that the stress decay (or stress intensity factor) for an optimised hole can be well estimated from published solutions of elliptical holes. This feature does not appear to have been identified in other published work.

PROCESSING OF FEA CO-ORDINATES FOR REWORK MANUFACTURE. Improving the manufacturing specification of optimal rework shapes is important for the wider application of the shape optimisation technology to RAAF airframes. In most of the previous DSTO investigations, cubic splines have been fitted to the FEA shape co-ordinates. While this worked, this approach was not automated and can be improved upon. For example monotonic variation in the radius of curvature was not enforced.

Airframe life extension by shape optimisation – Overview of DSTO developments

287

Recent developments for co-ordinate smoothing For the effective manufacture of optimal shapes (or optimal templates) algorithms have been developed to smooth the variation in radius of curvature of optimal shapes that are represented by points output by finite element based optimisation codes. The equations for smoothing are based on iteratively adjusting the radius of curvature of multiple circular arcs joining the points, [23]. For example Figure 12a shows the undesirable variation in radius of curvature around typical robust 2.1:1 optimal shape, as determined from co-ordinates output by the FEA optimisation code. The variation in radius of curvature after smoothing algorithm is shown in Figure 12b. The shape of the smoothed optimal hole is virtually indistinguishable from the original (not shown), and there is negligible change in shape optimality, see [23]. For NC machining the smoothed shapes are defined in IGES file format. Recent developments for morphed shapes Equations have also been developed to efficiently create intermediate optimal shapes by morphing, [23]. Here the morphed intermediate optimal shapes are generated by interpolating between two bounding optimal shapes. It has been shown that the morphed shapes have negligible difference in peak stress as compared to a shape generated by FE optimisation. The morphing approach enables a family of different sized reworks to be efficiently created, without the need for separate optimisation analyses. Generally it is desirable to use the smallest suitable optimal for a given crack size to be removed.

MACHINING METHODS AND ISSUES Previous approaches As discussed in Section 3.1 a relatively complex EDM method was used for the manufacture of reworks in the F-111 WPF [7,17,18]. Figure 13a shows the electrode used for EDM. Figure 13b shows NC machining of an optimised upper wing skin cutout for the F-111 F-WELD test article, [17], to repair cracking of the initial circular shape. New compact in-situ methods For transitioning of this technology, simpler methods than those used previously are highly desirable, particularly for locations which have difficult access. This has been a focus of recent work. Appropriate methods have now been achieved, all based on the use of precise templates, in conjunction with air-powered tooling. Figure 14 shows the arrangement for the manufacture of a typical FFVH in a steel coupon. The new optimal shape of length 48.65 mm, a width of 25.9 mm and ρmin = 7.1 mm. A coarse sanding drum of 14.5 mm diameter and then a fine abrasive drum of 14 mm diameter was used. The effectiveness of this arrangement has also been demonstrated on actual F-111 WPF’s. A final step was sanding with 400 grit ‘wet or dry’ paper. Excellent shape conformity was gained with the maximum deviation between the specified and manufactured shapes being approximately

M. Heller et al.

288

0.07 mm. This robust shape was designed using the procedures given in Section 2 and the co-ordinates were processed as outlined in Section 5. The geometry of the original FFVH is shown in Figure 15a along with the optimised shape. The stress decay along the prospective crack paths are shown in Figure 15b for the blueprint and optimised shapes respectively. Good stress reductions are achieved. A similar arrangement has been developed for the reworking of circular fuel flow holes in closely spaced aluminium stiffeners. These represent the geometry of the lower wing skin of the P-3C aircraft, which contains some fatigue critical holes. A reworked panel section containing the optimal hole is shown in Figure 16a The optimal hole is quarter symmetric, with an aspect ratio of 2.44:1. It has a length of 36.21 mm, width of 14.85 mm and ρmin = 4.41 mm. In this case a cutting tool of 3 mm radius is used as shown in Figure 16b, while for final polishing a diamond coated abrasive tool of radius 4 mm is used as shown in Figure 16c. Excellent shape conformity was gained with the maximum deviation between the specified and manufactured shapes being approximately 0.06 mm. It is noted that similar tooling has been successfully demonstrated for reworking stiffener runouts, and for conveniently over sizing circular holes, in closely spaced stiffeners. Stress effect due to machining accuracy Even though the reworks are very accurate, there will always be some discrepancy between the specified and actual local radius of curvature at locations around the hole boundary. Thus it is useful to quantify how this effects Kt. We note that the critical regions in an optimised shape have a close to uniform local stress. Hence a conservative estimate for the stress increase is represented by the geometry and loading shown in Figure 17a, representing a shallow circular notch in a uniaxially loaded plate. The Kt for this case is given as, [24]: Kt = 1 + t / ρ

(10)

where t is the depth of the notch and ρ is the radius of the notch. The corresponding stress increases are shown graphically in Figure 17c, and compared to an earlier solution, [25]. For a less conservative estimate, we consider the case of repeating shallow circular arc notches shown in Figure 17b, and compare the Kt change between the specified and manufactured ρ cases. A parametric FEA study was undertaken and the results are also shown in Figure 17c. A convenient equation representing the behaviour is given by K t = 1 + 1.815 ( t / ρ ) 0.505

(11)

For the FFVH case considered in Section 6.2, the worst-case values were from t/ρ = 0.00008 to t/ρ = 00022, for the optimal to manufactured cases respectively. This represents an increase in Kt of only 1.0%. For the symmetric robust hole considered in Section 6.2 , the worst case values were from t/ρ = 00012, to t/ρ =

Airframe life extension by shape optimisation – Overview of DSTO developments

289

0.00022, for the optimal to manufactured cases respectively. This represents an increase in Kt of only 0.6%.

OTHER DESIGN STUDIES, APPLICATIONS AND DEMONSTRATORS In broad terms two types of optimal shape have been addressed in DSTO work, (i) generic transferable shapes (either ½ or ¼ symmetric) and non-symmetric unique optimals. In most cases the shapes are provided as a set of non-dimensional coordinates for ease of further application. Apart from the WPF features discussed in Section 3, some other key cases are summarised as follows. Generic Cases The following cases have been considered; (i) Holes; a wide range of standard and robust shapes have been determined, covering single and interacting holes. Here uniaxial and biaxial loading conditions have been considered [2,3,5,10,11]. For example Figure 18 shows optimised interacting holes,. [11]. It can be seen in Figure 18c that irrespective of the interaction distance that Kt is reduced to Kt = 2 (when based on bulk stress), (ii) Stiffener runouts and fillets: shapes have been determined as a function of relative length and section thickness change, for tensile and bending cases, [4]. In unpublished work minimum length transition lengths have also been determined which give Kt = 1, (iii) Edge notches and shapes for surface damage removal: Shapes have been determined which depend on the relative length and depth of the rework for plates loaded in tension, [6,9]. These shapes are also useful for the design of low Kt specimens. Other design studies The following cases have been considered:(i) F/A-18 vertical tail attachment stub: A rework option was designed based on the use of generic fillet shapes, (ii) F/A-18 aileron hinge: Optimised blends were designed and experimentally validated to remove prospective cracking and minimise stresses, (iii) P-3C fuel flow hole in lower wing skin stiffener: Both symmetric and asymmetric rework shapes have been designed, [8,16], and (iv) F-35 bulkhead access/lightening hole: A generic robust optimal was successfully investigated for use in a shear loaded bulkhead panel, [10]. Other Practical Application There have been successful applications of optimised reworks on the F-111 FWELD test article to repair cracking and extend life, so that test down time could be avoided, as follows: (i) fatigue critical WPF bushes were manufactured with an optimised undercut fillet shape, and (ii) upper wing skin holes were optimised. An example is shown in Figure 13b.

M. Heller et al.

290

TECHNOLOGY TRANSITIONING ISSUES To date the DSTO shape optimisation technology has been developed and applied on a case by case basis to ADF aircraft. The current aim is to efficiently transition this technology more widely to other aircraft in service with the ADF. The transition management for this technology is still an evolving process; however one issue identified was the need for a codification and specification document (SD). This has now been achieved through reference, [26,27], where the approach for implementation can be very broadly summarised as follows. Once the potential application of shape optimisation is identified, it is followed by a feasibility analysis. The feasibility analysis may include cost-benefit and risk analysis consistent with the Aircraft Structural Integrity Program (ASIP), for that specific airframe. Once the feasibility analysis indicates a likely positive outcome, the shape optimisation repair development can be initiated, either on a single airframe and/or on a fleet of aircraft. Here a one-off repair may be designed using the method outlined in Sections 2, or existing generic optimal shapes can be used, see Section 7. In the repair development process key aspects to consider are stress reduction, fatigue life, non-destructive inspection methods, geometric constraints, and manufacturing methods. Then a package of engineering documents can be reviewed by the AEO for acceptance. For certain rework applications it is important to confirm that there is still adequate buckling strength. For example this was done during the design of the optimised F111 SRO's. In the unlikely event that cracking recurs at a shape optimised feature, there are two recommended repair approaches. The preferred option is to machine an oversized “morphed” optimal with good stress reduction properties. Alternatively a local shallow blend can be done using hand tools, where the blend radius is chosen to be similar to the radius at the crack location. Here the stress effects are quantified as shallow blends on circular arcs in [28], while Eqn. (11) gives the effect for very shallow blends.

CONCLUSION In the context of airframe life extension, some key points regarding the practical implementation of rework shape optimisation arising from the present paper, are as follows. Robust 2D and 3D optimal shapes can be designed to provide significant stress reductions, for multiple peaks around the stress concentrator boundary. This applies even when the shapes are subject to significant geometric constraints, and where only material removal is possible. The application of these shapes generally leads to increases in safe-life and inspection intervals. For fatigue life estimation the effects of the key design parameters and practical constraints has been quantified. In the context of crack growth analyses, a convenient analogy based on

Airframe life extension by shape optimisation – Overview of DSTO developments

291

the behaviour of elliptical holes is demonstrated, This provides for accurate prediction of stress distributions and stress intensity factors based on the peak stress at the hole edge. More efficient methods for accurate in-situ manufacture of the optimal shapes have been developed and validated. This includes post processing of numerically determined shapes that are smoothed and can be morphed. The morphing approach enables a family of different sized reworks to be efficiently created, without the need for separate optimisation analyses. In broad terms two types of optimal shape have been addressed, generic transferable shapes, and non-symmetric unique optimals. The generic shapes have been defined as a set of non-dimensional coordinates for ease of further application. Finally recommendations for transitioning the technology for wider application.

REFERENCES [1] Heller, M., Kaye, R. and Rose, L. R. F. (1999). A gradientless procedure for shape optimisation. J. Strain Anal. Eng. Des., v. 34, n. 5, pp. 323–336. [2] McDonald, M. and Heller, M. (2004). Robust shape optimization of notches for fatigue-life extension. Struct. Multidisc. Optim., v. 28, n. 1, pp. 55–68. [3] Waldman, W. and Heller M. (2006). Shape optimisation of holes for multipeak stress minimisation. Australian J. Mech. Eng., v. 3. n. 1, pp. 61–71. [4] Waldman, W., Heller, M. and Chen, G. (2001). Optimal free-form fillet shapes for tension and bending. Int. J. Fatigue, v. 23, n. 6, pp. 509–523. [5] Burchill, M. and Heller, M. (2004). Optimal free-form shapes for holes in flat plates under uniaxial and biaxial loading. J. Strain Anal. Eng. Des., v. 39, n. 6, pp. 595–614. [6] Burchill, M. and Heller, M. (2004). Optimal notch shapes for loaded plates. J. Strain Anal. Eng. Des., v. 39, n. 1, pp. 99–116. [7] Heller, M., McDonald, M., Burchill, M. and Watters, K. C. (2002). F-111 airframe life extension through rework shape optimisation of critical features in the wing pivot fitting. In: Proceedings of the 6th Joint FAA/DoD/NASA Conference on Aging Aircraft, 16–19 September, San Francisco, CA, USA. [8] Evans, R., Braemar, R. and Heller, M. (2005). Aircraft life extension via shape optimization of holes in lower wing skin stiffeners. In: Proceedings of the 4th Australasian Congress on Applied Mechanics, pp. 801–806, Xie, M. et al. (Eds.), 16–18 February, Melbourne, Australia. [9] Wescott, R., Semple, B. and Heller, M. (2005). Stress analysis of near optimal surface notches in 3D plates. J. Mech. Des., v. 127, pp. 173–1182. [10] Kaye, R. and Heller, M. (2006). JSTAB – Shape optimization for structural detail design: Activity 2 report: Optimal shape design for JSF demonstrator geometry, DSTO-TR1881, DSTO, Melbourne, Australia. [11] Waldman, W. and Heller, M. (2003). Shape optimisation of two closely-spaced holes for fatigue life extension. DSTO-RR-0253, DSTO, Melbourne, Australia. [12] Neuber, H. (1972). Zur optimirung der spannungskonzentration. Continuum Mechanics and Related Problems of Analysis, Nauka Publishing House, Moscow, pp. 375– 380.

292

M. Heller et al.

[13] Vigdergauz, S. B. and Cherkayev, A. V. (1986). A hole in plate, optimal for its biaxial extension-compression. Journal of Applied Mathematics and Mechanics, v. 50, n. 3, pp. 401–404. [14] Mattheck, C. and Burkhardt, S. (1990). A new method of structural shape optimisation based on biological growth. Int. J. Fatigue, v. 12, n. 3, pp. 185–190. [15] Heywood, R. B. (1969). Photoelasticity for Designers (Chapter 11: Improvement of Designs), Pergamon Press. [16] Heller, M., Braemar, R., Waldman, W., Evans, R., McDonald, M. and Kaye. (2005). Local Shape Optimization for Airframe life extension. In Proceedings of the 6th World Congress of Structural and Multidisciplinary Optimization, 30 May-03 June, Rio de Janeiro. [17] Diab, H. and Tatangelo, A. (2009). F-111F wing economic life determination test report. DSTO-TR-2274, DSTO, Melbourne, Australia. [18] Swanton, G. (2009). 40 years of F-111 aircraft structural integrity support by the Defence Science and Technology Organisation. In: Proceedings of the 13th Australian International Aerospace Congress, 9–12 March, Melbourne, Australia. [19] Walker, K., Weller, S. and Walker, J. (2005). F-111 wing pivot fitting upper plate critical features fatigue assessment – post wing optimisation modification. DSTO-TR-1776, DSTO, Melbourne, Australia. [20] Molent, L., Ogden, R. and Pell, R. (2000). F/A-18 FS488 bulkhead fatigue coupon test program. DSTO-TR-0941, DSTO, Melbourne, Australia. [21] Huynh, J., Molent, L. and Barter, S. (2007). Fatigue crack growth predictions for 7050 aluminium alloy with different stress concentration factors. DSTO-RR-0330, DSTO, Melbourne, Australia. [22] Schijve, J. (1980). Stress gradients around notches. Report LR-297, Department of Aerospace Engineering, Delft University of Technology, Netherlands. [23] Wescott, R. and Heller, M. (2009). Algorithms for improved transformation of optimised blend shapes for effective numerically controlled manufacture. DSTO-RR-0340, DSTO, Melbourne, Australia. [24] Noda, N.A. and Takase, Y. (2002). Stress concentration factor formulas useful for all notch shapes in a flat test specimen under tension and bending. J. Test. Eval., v. 30, n. 5, pp. 369–381. [25] Neuber, H. (1946). Theory of notch stresses – Principles for exact stress calculation. JW Edwards, Ann Arbor, Michigan. [26] Ang, D. and Clark, B. (2007). Shape optimisation specification document. ESSTRUCT-51-ASM071, Aerostructures Technologies, Melbourne, Australia. [27] Janardhana, M., Heller, M. and Davis, C. (2007). Transitioning rework shape optimisation technology for ADF airframe applications. In: Proceedings of the 5th Australasian Congress on Applied Mechanics, pp. 723–729, Veidt, M. et al. (Eds.), 10–12 December, Brisbane, Australia. [28] Burchill, M. and Heller, M. (2007). Stress and fracture analysis of circular arc blends for repair of cracked metallic components. In: Proceedings of the 5th Australasian Congress on Applied Mechanics, pp. 74–81, Veidt, M. et al. (Eds.), 10–12 December, Brisbane, Australia.

Airframe life extension by shape optimisation – Overview of DSTO developments

293

Crack

Blueprint geometry Traditional rework Optimised rework

(a)

(b)

Figure 1. Concept of optimised shape reworking life extension: (a) typical stress concentrators and (b) crack repair with shape optimised for minimum Kt.

h

h

w w

(a)

(b)

Figure 2. Two extreme cases of optimised holes: (a) 2.1:1 robust optimal hole, (ρ/w = 2.91), and (b) 1:1 standard optimal (ρ/w = 3.36).

M. Heller et al.

294

Figure 3. Multi-peak method: (a) notation for constrained hole optimisation, and (b) idealised multi-peak stress distribution around initial and optimised hole.

Figure 4. Robust optimisation for one region: (a) plate with load of varying orientation α, and (b) stress distributions for different load condition cases. 3.0

0.25

α = 0º, ρ = 0.15 -2º ≤ α ≤ +2º, ρ = 0.15 -5º ≤ α ≤ +5º, ρ = 0.15 -10º ≤ α ≤ +10º, ρ = 0.15

2.6

0.15 α = 0º, ρ = 0.15

0.10 0.05

-2º ≤ α ≤ +2º, ρ = 0.15 θ

-5º ≤ α ≤ +5º, ρ = 0.15 -10º ≤ α ≤ +10º, ρ = 0.15

0.00 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 Horizontal position, x/L

(a)

Peak Kt

Vertical position, y/L

2.8 0.20

2.4 2.2 2.0 1.8 1.6 -10

-8

-6

-4

-2

0

2

4

6

8

10

Load angle, α (degrees)

(b)

Figure 5. Stress minimisation results for typical 2D hole showing effect of design robustness range, α: (a) optimal hole shapes, and (b) peak stress response.

Airframe life extension by shape optimisation – Overview of DSTO developments

295

j=v

:

j=2 j=1

i-1

i i+1

(a)

(b)

Figure 6. Extension to 3D applications: (a) notation for definition of location around bounday, and (b) typical optimal hole shape in a wing skin stiffener.

(a)

(b)

Figure 7. FEA for optimisation of features in F-111 WPF: (a) large scale model, and (b) range of optimised FFVH shapes.

(a)

(b)

(c)

Figure 8. Typical critical features in F-111 wing pivot fitting: (a) nominal FFVH, (b) optimised FFVH, and (c) optimised stiffener runout.

M. Heller et al.

296

14 12

tt2 tt2 / tt1

σKt

σKt1

A= -0.2 A= -0.227, F18 APOL A= -0.314, AP-3C A= -0.4

10

tt1 Non-optimal

6 4

Optimal

σKt2

8

2 0

Flight time, t (hours)

0.6

0.7

0.8

0.9

1.0

Kt2 / Kt1

(a)

(b)

Figure 9. Life extension benefits of optimised reworks for ‘safe-life’ approach based on stress-life curve: (a) life trend, and (b) typical relative lives. 15000 Kt 3.0

ac

t r1 Non-optimal Optimal

Residual life (hrs)

Crack size, a

t r2

2.5

10000

2.0 1.5

5000

andi Flight time, t (hours)

0 0.01

0.1

1

10

100

ai (mm)

(a)

(b)

Figure 10: Life extension benefits of optimised reworks for ‘safety-by-inspection’ approach: (a) generic trends, and (b) trends for residual crack growth lives.

Airframe life extension by shape optimisation – Overview of DSTO developments

297

3.0

σr y

σy/σr

2.5

h

σy w

x/

1.00:1 standard optimal 2.10:1 robust optimal 1.71:1 ellipse circle

2.0

x

1.5

1.0 0.00

σr

0.25

0.50

0.75

1.00

1.25

1.50

1.75

2.00

x' / w

(a)

(b)

2.0

2.0

1.8

1.8

Normalised radius of curvature

Normalised radius of curvature

Figure 11. Sress decay for two optimal holes and elliptical hole (ρ/w = 2.93) when Kt is identical: (a) notation for elliptical holes, and (b) stress decay comparison.

1.6 1.4 1.2 1.0 0.8 0.6 0.4

1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2

0.2

0.0

0.0 0

15 30 45 60 75 Angle about hole, θ (degrees)

(a)

90

0

15 30 45 60 75 Angle about hole, θ (degrees)

90

(b)

Figure12. Variation of ρ/h for typical optimal shape FEA coordinates, case is h:w = 2.1:1 for ±20° load perturbation: (a) no smoothing, and (b) smoothed.

M. Heller et al.

298

Electrode Plate

Finishing Electrode

Locating Probe

Roughing Electrode

(a)

(b)

Figure 13. Machining examples for optimised shapes: (a) electrode for EDM of WPF features, and (b) NC machining of wing skin cutout for F-WELD test.

(a)

(b)

Figure 14. Demonstration of in-situ machining jig for new F-111 FFVH robust shape: (a) reworked holes with assembled jig, and (b) key components. 3.5 3.0 Optimal FFVH 2w Blueprint FFVH

Blueprint FFVH

σn/σr

2.5

Optimal FFVH

2.0

Positions of potential cracks

x' 1.5

x' σn

(a)

1.0 0.0

0.5

1.0 x' / w

1.5

2.0

(b)

Figure 15. FEA for new F-111 FFVH robust shape used in in-situ machining jig: (a) geometry, and (b) stress decay comparison.

Airframe life extension by shape optimisation – Overview of DSTO developments

(a)

(b)

299

(c)

Figure 16. Demonstration of in-situ machining of symmetric robust shape for P-3C stiffener holes: (a) holes in stiffeners, (b) cutting tool, and (c) finishing tool. σ

1.07

σ

1.06

ρ

1.05

ρ

1.04 Kt

t

1.03

t σ

Multiple Notches FEA Multiple Notches Fit to FEA Single Notch Neuber Single Notch Noda & Takase Single Notch FEA

1.02

σ

1.01

1.00 0.0000 0.0002 0.0004 0.0006 0.0008 0.0010 t /ρ

(a)

(b)

(c)

Figure 17. Analysis for stress increase due to shallow edge notches in a large plate: (a) single notch case, (b) repeating notch case, and (c) comparison of stress results. 4.0

2.0 1.5

σ2

1.0

y

3.0

x σ1

Kt

σ1

y/h

0.5

e

Circular holes Optimal holes

3.5

0.0

2.5

-0.5

2h

2w

σ2

2.0

-1.0 -1.5 -2.0 -1.5

1.5 -1

-0.5

0

x/h

(a)

(b)

0.5

1

1.5

0.0

0.5

1.0

1.5

2.0

e /2h

(c)

Figure 18. Kt variation for optimal interacting holes as a function of e/h for a tensile biaxial stress field where σ1 = σ2, (a) geometry, (b) shapes, and (c) Kt results.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

EVALUATION, MODIFICATION AND DAMAGE TOLERANCE OF AN INSERVICE AIRCRAFT CRITICAL AREA R. S. Rutledge, D. S. Backman and R. J. Hiscocks National Research Council Canada Institute for Aerospace Research.

Abstract: In-service component damage on the wing-fold sheartie of a Canadian Forces (CF) fighter aircraft, led to a full scale localized fatigue test of a critical area using loads derived from inservice usage. The aim was to generate a crack in the test article, at the correct location and in a similar manner to that found on inservice aircraft inner wings. Then the structure was modified to the repair configuration and fatigue testing of the modified component was carried out to define the life of the modification. The project scope required: generation of nucleated cracking damage to a size typically found in the field with substantiation via non-destructive inspection; excision of the in-service premodification crack; introduction of the planned in-service hand blended repair to the test article; measurement of both surface strain and the altered geometry in the blend region; and use of quantitative fractography to develop the test crack growth curves for pre-modification and post-modification. Testing applied the equivalent of seven life times of simulated usage in the two phases of the fatigue testing. This included the required factor of five lifetimes, post-modification, for an unmonitored dynamically loaded component. The applied dynamic spectrum alternated between blocks of loads for a tip mounted store with fins-off and fins-on. By the end of the test a significant size crack had developed. To complete the test, a residual strength test was carried out to failure. Post-test quantitative fractography was used to obtain the crack growth rate information. This data and the residual strength test failure load were then used to generate a residual strength curve. Based on the loading spectrum, a Gumble extrapolation was carried out to establish the once in a lifetime M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 301–327. © National Research Council Canada 2009. Published by Springer Science+Business Media B.V. Dordrecht.

302

R.S. Rutledge, D.S. Backman and J. Hiscocks

design limit load. This was then scaled to obtain the residual strength, end of life, 120% load and an end of life critical crack size was estimated. The client and the repair and overhaul contractor utilized this test and the derived data to establish fleet repairs, scheduling and inspection requirements for the modified structure.

INTRODUCTION The full-scale wing fatigue test, FT245, was the primary means of wing structural certification for a Canadian Forces (CF) fighter aircraft. This test applied three repeated load lives of simulated flight hours to prove the wing could safely fly for the design life flying objective using CF missions. Individual aircraft tracking revealed usage of the inner and outer wing components was more severe than the original certification testing. The test spectrum used for the wing test was for an average aircraft in a severe squadron based on maintenance signal data recording set (MSDRS) data statistics developed for the year 1989-1990. This spectrum was designated IARPO3a and included dynamic loads caused by trailing edge and wing tip buffet, particularly at high angles of attack and in the wing tip store with fins-on configuration. In the wing test, the dynamic loads at the control surface attachments on the trailing edges had to be severely truncated in order to complete the fatigue testing within the client desired time frame. Consequently, some areas of the wing susceptible to dynamic loads may have been under tested. In particular, cracks were found in the wing-fold aft-spar shear-ties of in-service aircraft. Since these were not found in the FT245 fatigue test it was assumed that the spectrum was not severe enough or the surface preparation associated with the installation of strain gauges on the shear-tie may have removed the acid etched surface damage that is typically present at the surface of all aluminium on this aircraft. This damage was a by product of the cleaning process that was carried out prior to the application of the corrosion prevention ion vapour deposition process used during the manufacture of these aluminium parts. To repair the cracks in the fleet wing shear-ties, various blends were proposed to remove damaged material. Due to the severity of the problem within the fleet, the National Research Council Canada was tasked to perform a local fatigue test on the aft-spar shear-tie of the inner wing (S/N A12-0312) to simulate the initial crack nucleation; work with the repair and overhaul contractor to blend the damaged area; and then test the modified structural element under fatigue loading to determine if the structural modification was adequate to meet the estimated life objective for this aircraft in accordance with the lifing policy for this aircraft [1].

TEST OBJECTIVES The test objective was to locally fatigue test the wing-fold aft-spar shear-tie. Initial fatigue cycling was performed to generate cracking similar to that found in-service. This was followed by an in-service repair typical of a structural modification

Evaluation, modification and damage tolerance of an in-service aircraft critical area

303

proposed for this area. After the structural modification, fatigue testing was continued to determine the post-modification life for an average aircraft using the representative spectrum [2] and to generate test crack growth data for management and provision of fleet modifications. The plan was to apply the equivalent of five repeated load lifetimes (RLLs) of measured aircraft IARPO3a [3] usage.

METHODOLOGY Test Specimen The test specimen for FT312 was retained from the original FT245 wing test. This wing component, serial A12-0312, was retired, having accumulated approximately 60% of one RLLs of usage. Note that the wing fold shear-tie area of this wing had not been loaded during FT245 testing. Test Rig Configuration The test article was installed in its original position on the left side of the fuselage. A reaction system consisted of a contour board with solid struts at BL88 and the FT245 reactions as illustrated in Figure 1a. A single actuator, mounted on the existing FT245 grillage, was used to apply loads directly to the shear-tie. The actuator to test article attachment lug was designed to apply the load offset from the centre of the shear-tie (Figure 1c). Three different offsets were considered in the spectrum development and a pre-test strain survey was used to obtain strain distributions. Based on this data [4], the configuration with an offset of 7.62 mm (0.3 inches) aft was selected and used in the fatigue test. The measured strain distribution is provided in Figure 2. Note that the tensile first principal strains were measured for up and down loads at 47% of the design limit load (DLL) and the principle directions are illustrated. FS525.275 BL162.758 WL104.323

FS525.675 BL158.306 WL19.369

Figure 1 FT312 Rig Configuration a) Wireframe on Grillage, b) Actuator Load Cell Installation and c) Loading Lug

R.S. Rutledge, D.S. Backman and J. Hiscocks

304

Figure 2 First Principal Strains 47% DLL Down (Left) and -47% DLL Load (Right) Control System and Actuators The actuator and load cell used for this test was previously used on FT245. The actuator system consisted of a double-ended actuator, servo valve, load cell, mechanical load limiter and load abort module [5]. The location of the actuator, along with the coordinates of both end connectors, is illustrated in Figure 1a. The geometry in Figure 1a represents the position from which the test loads were applied. The hydraulic system was powered by the facility’s MTS pump. This system supplies hydraulic pressure through a controllable service manifold. The test was run using an MTS Aero-ST multi-channel controller with an MTS AeroPro software interface. The actuator/load cell/servo valve combination was configured as a control channel and tuned for closed-loop control. The load cell was a dual bridge with A/B feedback comparison on every end level. The control system was configured with digital input/outputs from an ultrasonic (UT) crack detection system to the Aero-Pro controller. A high signal indicated the UT equipment was on and the loss of this signal would trigger the controller to hold the test. This ultrasonic testing (UT) technique was programmed to stop the load application when the ultrasonic signal exceeded the level established on a calibrated test specimen. The calibrated specimen had been extracted from a separate wing aft spar and an electrical discharge machining notch was introduced in the critical area to create the inspection test piece. A crack with a depth of 0.381 mm (0.015 inches) to 0.508 mm (0.020 inches) was the intended target size, which was typical of damage found in the CF fleet aircraft. Load Conditions and Spectra The spectrum contained two blocks, each representing a year’s flying, one block (even numbers) assumed a wing tip missile had fins fitted (fins-on), and the other (odd numbers) assumed the fins had been removed (fins-off). These blocks were applied alternately throughout the fatigue testing because the fins-on and fins-off

Evaluation, modification and damage tolerance of an in-service aircraft critical area

305

condition could not be determine from the MSDRS monitoring system. The two blocks combined contained 14,828 end-levels. Four additional zero-g lines were added, one at the start, and one at the end of each block. The pre-modification spectra used a truncation level of 15% maximum manoeuvre range (MMR) to remove the smallest dynamic loads. A plot of the two block spectrum is illustrated in Figure 3. In order to obtain the correct damage configuration, this fatigue spectrum [6] was applied until a 0.508 mm (0.020 inch) deep crack was generated. This occurred after approximately 72% of one RLL after test start. The time to generate this size crack was found using a combination of optical strain measurements (digital image correlation), strain gauges, and an automated ultrasonic inspection of the test area. After these techniques indicated that an appropriate size crack was potentially present in the critical area, fatigue testing was suspended and both eddy current and liquid penetrant inspections of the area were conducted. Inspection confirmed that an appropriate size flaw, typical of in-service damages, was present. The crack was excised for post-test quantitative fractographic examination. The proposed fleet structural modification was then introduced. Then a short time after test restart, high first principal strains were found to occur at peak loads on the modified part. Testing was suspended and the altered geometry was measured using digital image correlation. This measured geometry was provided to both the client and repair and overhaul contractor for geometry comparison to the planned modification and finite element modelling (FEM) stress analysis. Pre-Modification Spectrum FT312 Wing Fold Shear Tie Two Block Spectrum Peak Load / DLL Critical Strain Conditions Pre-Alex 66.2

1

0.8 0 to 7421 Lines No Fins Block

7422 to 14832 Lines Fins Block

Load / Design Limit Load

0.6

0.4

0.2

0

-0.2

-0.4

-0.6

-0.8 0

2000

4000

6000

8000

10000

12000

Sequence Number

Figure 3 Pre-Modification Spectrum Sequence Plot

14000

R.S. Rutledge, D.S. Backman and J. Hiscocks

306

Based on the FEM analysis of the measured blend using the peak loads resulted in stress values that were higher than the intended blend predicted. To adjust the spectrum to the stress level intended the pre-modification load spectrum was shortened and peak loads were reduced [7]. The peak down load was lowered by approximately 8% in order to apply the intended peak strain in the stress concentration. This spectrum was derived in a similar fashion to the premodification spectra, but used a pegged stress level in the upper flange hot-spot area of the blend. The altered spectrum had 14,656 end levels for the fin-on and fins-off sequence, as illustrated in Figure 4. The peak positive load was reduced to 68% DLL and the peak negative load was reduced to 63% DLL.

Post-Modification Spectrum FT312 Wing Fold Shear Tie Two Block Spectrum Peak Load / Design Limit Load Critical Strain Conditions Post-Alex 66.2 0.8

0 to 7356 Lines No Fins Block

0.6

7357 to 14660 Lines Fins Block

Load / Design Limit Load

0.4

0.2

0

-0.2

-0.4

-0.6

-0.8 0

2000

4000

6000

8000

10000

12000

14000

Spectrum Line Number

Figure 4 Post-Modification Spectrum Sequence Plot Transition Times To estimate the test completion time, a base transition time for each load line was estimated. The total estimated time for the double block was 13.8 hours; the actual achieved time was about 8 hours for the same double block. Note that a minimum of 0.2 seconds was applied to avoid incurring dynamic strain errors during loading. Instrumentation A total of eight strain gauges were installed on the shear-tie to record strain data. These strain gauges replicated those installed on the FT245 right hand wing sheartie. Illustrations of each gauge have been included in Figure 5.

Evaluation, modification and damage tolerance of an in-service aircraft critical area

307

Figure 5 Strain Gauges 952S14L1-10 Upper Cap (left), 952S14L1-11 Near Lower Cap and 952S14L1-4-A,B,C (middle), and 952S14L1-5A,B,C (right) An ultrasonic sensor was located on the shear-tie to monitor flaw size. The sensor used was a Panometrics 5.0 X 0.25 A5435, on a Panometrics 60° Aluminum Shoe ABWM-4088T. The signal from this sensor was connected to a Reflectoscope Type S80 Style SOD900 ultrasonic tester. Both are illustrated in Figure 6. The strain gauges and dual bridge load cell data were recorded on a Hottinger Baldwin Messtechnik (HBM) data acquisition unit using an MTS Aero-Pro software interface. The MTS Aero-Pro control software allows digital inputs and outputs. This test setup was configured for load control on bridge A with verification on bridge B and programmed to stop the testing each time the Reflectoscope Ultrasonic gate was triggered.

Figure 6 Panometrics Ultrasonic Sensor Installed Opposite Critical Radius (left), and Reflectoscope S80 Ultrasonic (right) Digital Image Correlation Data Acquisition System A three dimensional digital image correlation system [8] was installed, to measure surface strains during peak loading conditions as shown in Figure 7. The digital image correlation system consisted of two AVT Marlin cameras (AVT Inc, Newburyport, MA) controlled using VicSNAP and VIC-3D software (Correlated Solutions Inc, West Columbia, SC) with illumination to the wing-fold shear-tie region provided by two high intensity LED grid arrays (Siemens NerLite, Weare, NH). The cameras were mounted on a custom fixture designed and manufactured by NRC. The fixture was used to ensure camera stability during the fatigue loading phase of the test.

308

R.S. Rutledge, D.S. Backman and J. Hiscocks

Figure 7 Digital Image Correlation Camera Set-Up The theoretical basis behind image correlation revolves around the fact that reference images taken of a high contrast surface can be matched to similar images of the object after loading. Using a cross correlation function, the displacement of small subsets on the surface of the object can be determined, and the derivatives of these displacements can be converted to strain. By adding a second camera to the image correlation system, it is possible to calibrate the cameras and use photogrammetric techniques to measure surface displacement as well as surface strain [9-12].

TEST OPERATION During regular operation, there was daily monitoring of the test, including: rig inspections; periodic examination of digital image correlation data at the critical test section on the test article; trend monitoring; and automatic error monitoring. Error monitoring detected shut downs and residual strength test failure, while the trend monitoring detected less rapid changes in the structure based on the strain gauge data. The test was planned to operate for eight hours per day, five days per week with an elapsed test time of sixteen weeks to complete 94 blocks and apply five times the planned life sequence of loads. The fatigue test actually ran from April 27th 2006 to September 27th 2006 (25 weeks) and applied 133 blocks, applying over seven times the planned life sequence of loads. The longer duration could not be anticipated since the post-modification testing was intended to be a crack growth test to failure. The crack growth was found to be slow and stable. When the fatigue testing reached the seventh life time, fatigue testing was stopped and a residual strength test to failure was carried out to complete the testing.

Evaluation, modification and damage tolerance of an in-service aircraft critical area

309

Data Acquisition System Results Full sequences of load and strain data were recorded during blocks 1, 2, 3, 4, and after the spectrum was changed starting in blocks 15 and 16. A sample strain response sequence for a gauge has been included in Figure 8.

Figure 8 Sample Two Block Strain Gauge Response During all other blocks, a limited number of end levels were recorded to enable trend monitoring of the structure. In this way, changes in strain level were detected and inspections initiated if deemed appropriate. The trend lines recorded all loads at 73%, 47%, -47% and -69% of design limit load (DLL) during the pre-modification spectrum and at 68%, 47%, -47% and -63% DLL during the post-modification blocks. MTS Aero Trend Monitor v 4.65 was used, with additional off-line analysis as required. The trends monitored during the duration of the test for one gauge have been included in Figure 9. Note the change in tensile strains during the premodification crack growth, after the modification blend and the three other structural changes due to loading system component failures. Beyond the fourth RLL, the strain changes were found to be more severe, and this can be seen in the trend data which corresponded well with measured crack growth.

310

R.S. Rutledge, D.S. Backman and J. Hiscocks

Figure 9 Sample of Trend Strain Monitoring

During testing under the peak down load causes the highest strains in the component. At these occurrences in the spectrum the digital image correlation system was triggered to record the strain distribution [8]. This measured distribution varied as the crack started to grow. A typical strain distribution for this condition soon after the start of testing at 0.6 RLL is illustrated in Figure 10 and the measured strain within the highlighted circular zone, was approaching 7200 microstrain. At 1.0 RLL in the same load condition, the strain distribution was very similar. These strains are illustrated in Figure 11 and the strain within the circular zone was slightly lower at 7010 microstrain. This indicated that there was likely little crack growth during this period of fatigue testing.

Evaluation, modification and damage tolerance of an in-service aircraft critical area

311

Figure 10 Measured Strain Distribution at 0.6 RLL

Figure 11 Measured Strain Distribution at 1 RLL

By 1.26 RLLs, the strain within the circular zone was approaching 7650 microstrain and a zone of higher strains could be clearly seen in the distribution. Figure 12

312

R.S. Rutledge, D.S. Backman and J. Hiscocks

illustrates the strain distribution captured by digital correlation prior to the modification.

Figure 12 Measured Strain Distribution Prior to Modification at 1.26 RLLs Condition 280 During post-modification, the critical area moved due to the aircraft life extension modification (ALEX) hand blend. After sixty five simulated flight hours after the modification, at 1.33 RLLs, the test was stopped due to high strains measured at condition 280 and due to ultrasonic triggering of the non-destructive test instrumentation. The strains measured in the critical area were of similar magnitude to the strains measured just prior to the introduction of the ALEX 66.2, as illustrated in Figure 13. Note that the elevated strain zone is much larger after modification than in the original configuration. A strain survey was carried out to document the strain distribution change due to the modification [13]. After the evaluation of this data and subsequent finite element modelling by the repair and overhaul contractor [14], the spectrum was changed. The strain distribution under the peak load condition of the post-modification spectrum condition 288 at 2 RLLs is shown in Figure 14.

Evaluation, modification and damage tolerance of an in-service aircraft critical area

313

Figure 13 Post-Modification Condition 280 First Principal Strains at 1.33 RLLs

Figure 14 Post-ALEX 66.2 Modification Strain Distribution Critical Area Trend Monitoring Condition 288 at 2 RLLs

R.S. Rutledge, D.S. Backman and J. Hiscocks

314

Specimen Inspection and Damages A pre-test inspection of the test article was carried out at which time some corrosion damage was identified. There were no planned inspections during the test. However, inspections were carried out based on: the automated ultrasonic inspections; information obtained from triggered strain measurements using digital image correlation [8]; and trend monitoring of strain gauges. The inspections using the ultrasonic test set were configured to detect a 0.13 to 0.25 mm (0.005 to .010 inch) deep crack, at which point, the test was stopped. The first triggering of the ultrasonic sensor occurred at 1.0 RLL. After several confirmed triggers, the ultrasonic gate was changed to detect an increased crack depth. This ultrasonic equipment triggered 49 test stops between 1.0 RLL and 1.3 RLLs. At 1.32 RLLs the digital image correlation paint was removed from the test specimen and a liquid penetrant inspection was performed, which confirmed the presence of a crack, as shown in Figure 15. Eddy current inspection estimated the crack to be of similar depth as found in-service within the CF fleet.

Figure 15 Liquid Penetrant Inspection of Shear-Tie Prior to Modification The crack was then excised for examination. The overall view of the two adjacent and partly overlapping cracks is provided in Figure 16 [15]. The arrows point to the primary origins of the two cracks. Closer examination revealed repeat patterns that corresponded to the loading block sequence applied. An overall view of the lower crack is provided in Figure 17. The arrow points to the primary origin and secondary origins while the inclined white arrows point to some of the repeat patterns. Repeat patterns can be associated with crack growth during a loading block. At the beginning of the fatigue test, after 0.6 RLL, the test article had a semicircular lower crack that was about 0.050 mm (0.002 inches) long and slightly less than 0.025 mm

Evaluation, modification and damage tolerance of an in-service aircraft critical area

315

(0.001 inches) deep while the upper crack had a semi-elliptical shape and was approximately 0.125 mm (0.005 inches) long and slightly less than 0.025 mm (0.001 inches) deep. Using the repeat patterns the crack growth rates of both cracks could be measured on the excised piece, and are included as Figure 18.

Figure 16 Excised Shear-Tie Piece

Figure 17 Lower Crack Primary and Secondary Origins and Repeat Patterns

R.S. Rutledge, D.S. Backman and J. Hiscocks

316

After the cracks were excised at 1.32 RLLs the ALEX modification was carried out. This modification was a hand blend and polish of the cracked area as shown in Figure 19. After the eddy current and liquid penetrant inspection, this area was repainted and the digital image correlation system was used to extract the blended geometry for finite element evaluation, which resulted in a spectrum modification as described previously. Thereafter, the ultrasonic inspection and digital image correlation systems were used to monitor the test specimen. Excised Crack at 1.32 lifetimes Crack Depth (inches) 0.016

Crack Depth (mm)

0.4

0.35

0.014

0.3

0.012

0.25

0.010

0.2

0.008

0.15

0.006

0.1

0.004

0.05

0.002

0.000

0 0

0.2

0.4

0.6

0.8

1

1.2

1.4

Limetimes

Figure 18 Crack Growth Curves for Lower and Upper Cracks in the 68% Spar Shear-Tie Since the remainder of the test was intended to obtain crack growth information, the automated ultrasonic inspection system was only used to monitor the test specimen. At 1.5 RLLs the ultrasonic inspection system was disabled from automatically stopping the test due to the large number of triggers that were occurring. The digital image correlation system and trending of the strain gauges were used from this point on to monitor the structure.

Figure 19 Blended Area of the 68% Spar Shear-Tie

Evaluation, modification and damage tolerance of an in-service aircraft critical area

317

Fatigue testing continued post-modification with a slightly modified fins-off finson spectrum and both trend strain data and digital image correlation data were recorded periodically throughout the remainder of the fatigue testing. A sequential series of extracted images and processed strain distributions at each RLL were recorded. Figure 20 shows the progression of the strain field post-modification.

Figure 20 Post-Modification Peak Strains Measured at 1.33 RLLs and at 2 RLLs Visible cracking could be seen through the digital image correlation paint at approximately 2.3 RLLs under the peak applied spectrum tensile load, as shown Figure 21. From this point forward in the testing, the surface strains were processed excluding the crack.

Figure 21 Visible Crack in the 68% Spar Shear-Tie at 2.3 RLL

318

R.S. Rutledge, D.S. Backman and J. Hiscocks

Figure 22 Post-Modification Images at 3, 4, 5, 6, 7 and 7.75 RLLs The fatigue testing was carried out to the target 5 RLLs post-modification. Figure 22 shows the visible cracking at 3, 4, 5, 6, 7 and 7.75 RLLs under peak spectrum loading. At 6.33 RLLs and with significant visible damage present, the client requested an additional lifetime to generate crack growth and potentially fail the specimen. Testing was extended to 7.75 RLLs when it was stopped and a residual strength test (RST) was carried out.

Evaluation, modification and damage tolerance of an in-service aircraft critical area

319

The crack in the spar prior to the residual strength testing is shown in Figure 23 under no load. A view of the inboard post-RST fracture surface is provided in Figure 24. The fracture surface was then excised and post-test quantitative fractography was carried out on the failed component.

Figure 23 Visible Crack in the 68% Spar Shear-Tie Pre-RST at No Load

Figure 24 Inboard Fracture Surface in the 68% Spar ShearTie Post-RST The failure surface used for the quantitative fractography [16] is shown in Figure 25. In the middle of this image is a ledge (extending from right to left in the image), which separates the two nucleation points. The crack surface in the upper half of the image was several millimeters lower than the crack surface in the lower part of the image. Hence, it is referred to as the lower terrace and the crack in the lower half of Figure 25 is referred to as the upper terrace (Note: the specimen is shown inverted from its normal orientation on the wing).

320

R.S. Rutledge, D.S. Backman and J. Hiscocks

Figure 25 Quantitative Fractography with Surface Origins and Crack Paths To obtain the crack growth rate information on the modified part the fracture surface was examined using a scanning electron microscope (SEM). Two origin points and crack paths were identified and are shown in Figure 25. The close up of the origin point on the lower terrace had a particle at the edge of the sample as shown in Figure 26. This constituative particle appears to have served as the nucleation source for the lower crack while the origin point on the upper terrace did not show any particle and it was not clear what served as the nucleation source for the crack.

Figure 26 Quantitative Fractography Post-Modification Surface Origins Quantitative fractography was carried out using multiple images of the same magnification from the SEM. These images were stitched together and features on the surface were used to align the images. Repeat pattern distances were extracted along the crack path for both the upper and lower terrace cracks and a crack growth curve was generated. The upper terrace linear and semi logarithmic crack growth curves are provided in Figure 27. Using this data, one can conclude that the blend of the radii from the modification delayed the cracking of this fitting by 43% of one RLL or one third the original time to create a 0.35 mm (0.14 inches) deep crack.

Evaluation, modification and damage tolerance of an in-service aircraft critical area

321

Therefore, the improvement in time to the redevelopment of the crack was not significant. However, it was very encouraging to find that the crack growth rate was slow and stable.

FT312 Sequential Linear Crack Growth Data for the Upper Terrace 16

Crack Length (mm)

14 12 10 8 6 4 2 0 0

1

2

3

4

5

6

7

Lifetimes

FT312 Sequential Semi-Logarithmic Crack Growth Data for the Upper Terrace Inches

Crack Length (mm)

100

3.937

10

0.3937

1

0.0394

0.00394

0.1 0.01

0.0003947

0.001

0.0000394 7 0.0000039 47

0.0001 0

1

2

3

4

5

6

7

Lifetimes

Figure 27 Quantitative Fractography Upper Terrace Crack Growth Plots

RESIDUAL STRENGTH TEST A pre-residual strength test inspection found no additional damage, to those observed previously. The residual strength test (RST) was carried out in the fatigue test configuration. Load was applied in 10% increments until the peak spectrum load in the test sequence was reached. Additional load was then applied in 2% increments until failure. The final failure load during the RST was 87.5% of the calculated DLL

R.S. Rutledge, D.S. Backman and J. Hiscocks

322

and 128% of the peak spectrum load found in the post-modification spectrum. The strains measured during this residual strength static test did not exceed 3000 microstrain and non-linear behaviour was observed. Post-test examination of the crack found the final semi-circular fatigue crack size to be 14.65 mm (0.577 inches) deep at the shear-tie critical location [16]. This crack measured 16.55 mm (0.652 inches) across the upper cap and 13.2 mm (0.516 inches) down. Using the RST failure load, the fracture residual strength failure was estimated. Based upon the failure stress a residual strength curve was generated for the shear-tie and this is included as Figure 28. The critical crack size was estimated to be 13.2 mm (0.516 inches). This residual strength curve was estimated based on post-test quantitative fractography measurements of crack depth and length of the through thickness arc to obtain an estimate of crack area. The load for the residual strength curve was estimated based on intact area remaining to carry the load. For the first 2.54 mm (0.1 inches) of crack depth the area was estimated assuming a semi-circular shape. Thereafter, the crack area was determined based upon the incremental growth and estimate for the arc length. Figure 29 shows the crack front progression on the fracture surface.

Residual Strength Based on Test End Failure Load 3

2.5

DLL

2

1.5

Residual Strength Load 120% DLL

Design Limit Load 1 lifetime

1

RST Failure at 0.875 DLL

0.5

0 0

2

4

6

8

10

12

Crack Depth (mm)

Figure 28 Residual Strength Curve

13.2 mm 14

16

Evaluation, modification and damage tolerance of an in-service aircraft critical area

323

Figure 29 Crack Front Progressions on Wing-Fold Shear-Tie

Note that the peak load pre-modification was 75% DLL and occurred 4 times in the 652 simulated flight hours SFH double block pre-modification (326 SFH fins-on and 326 SFH fins-off). In order to define the design limit load, a Gumble extrapolation of the applied test spectra was carried out using Berens Methodology [17]. This was used to estimate the once in a lifetime design limit load, as shown in Figure 30. Note that the number of points used for the Gumble extrapolation yields different once in a life time design limit loads so one must be careful to select a conservative DLL. The variation between 150 points, 100 points and 50 points was only 2% of the DLL so the most conservative 100 point DLL load was selected. To obtain a residual strength target load, the DLL was increased by 20% to 120% DLL and the residual strength curve, Figure 28, was used to determine the critical crack depth for the shear-tie. At 120% DLL, the critical size was estimated to be 13.2 mm (0.516 inches). If we assume that the crack growth portion occurs between a 0.51 mm (0.020 inches) size and the 13.2 mm (0.516 inches) critical crack depth from the residual strength curve, a threshold inspection interval can be calculated using the crack growth curve. If the detectable size is assumed to be 1.27 mm (0.050 inches) both an initial and repeat inspection sizes can be calculated. An example calculation is included below

324

R.S. Rutledge, D.S. Backman and J. Hiscocks

Figure 30 DLL Gumble - Berens Extrapolation of the Applied Test Spectra

Evaluation, modification and damage tolerance of an in-service aircraft critical area

325

with a scatter factor 2 included to account for unknowns, like the inability to monitor this location individually on every wing.

The crack growth curve can then be used to identify the time at which these inspections should be carried out with the proviso that the prescribed threshold and repeat inspections never exceed half the aircraft life regardless of the results obtained from this simplified analysis. Since the threshold inspection would result in an estimated 1.26 RLLs with the repeat inspection at 1.21 RLLs, both threshold and repeat inspections were recommended to be set to half the intended life of the aircraft if a damage tolerance approach is to be used to guarantee the structural integrity of this area.

CONCLUSION NRC tested the wing-fold shear-tie in the pre-modification configuration to 1.32 RLLs at which point a 0.51 mm (.020 inches) deep crack developed. If this defines the durability life and is factored by three for a component that is tracked and individually monitored, an average aircraft shear-tie life to obtain a fatigue crack would only be 0.44 RLL. In the post-modification configuration tested, a crack similar to the pre-modification 0.51 mm (0.20 inches) deep crack was generated within 0.46 to 0.57 RLLs after the modification was introduced. If this defines the durability life of the modification and is factored by three, then an average modified component life would only be 0.16 to 0.19 RLLs. This indicates that this

326

R.S. Rutledge, D.S. Backman and J. Hiscocks

modification from a durability perspective was not very successful. The original decision to carry out the fatigue test to 5 RLLs and the decision to extend this to 6 RLLs post–modification followed by a static residual strength test provided an opportunity to collect crack growth data which can be exploited if one chooses to use a damage tolerance philosophy for managing this critical area of the aft spar. After fatigue cycling for the equivalent of 7.15 RLLs and cracking visible to the naked eye, a residual strength static test was carried out. This RST failure occurred at 87.5% DLL. As per the lifing policy, it was required that the component sustain 120% DLL at test end. Since the component failed prior to achieving this load, a residual strength curve was developed using quantitative fractography data to determine the critical crack size at the required load. This critical crack size was found to be 13.21 mm (0.516 inches). If a safety by inspection damage tolerance methodology is applied instead of the durability safe life approach, the component inspection was recommended at half a lifetime post-modification, with any subsequent repeat inspections at half a lifetime. These conservative inspections intervals were recommended even though the experimental evidence indicates that each of the inspection intervals could be over 1.2 RLLs. NRC does not recommend defining an inspection interval greater than half the design life of the wing for a modified part even if conservative assumptions, such as the use of accelerated crack growth curves including the effect of aggressive salt water environment, humidity and other unknowns, were used.

SUPPORT AND PARTICIPATION NRC-IAR recognizes the support and participation of L3 Communications (MAS) Canada and DND Directorate of Technical Airworthiness and Engineering Support (DTAES) in this test programme.

REFERENCES [1] DND, (2001), ‘Canadian Forces Lifing Policy, CF-18 Structural Integrity Standard for Continued Airworthiness, Appendix 9 to Annex A, CF-18 CMATC’. [2] Hiscocks, R., (2006), ‘Test Plan for FT312 CF-18 Wing-Fold Aft-Spar Shear-Tie’, LTR-SMPL-2006-0051, National Research Council Canada. [3] Zavitz, D. et al., (1993), ‘CF/RAAF IFOSTP Post-LEX Period Mission Block (IARPO3) Definition and Analysis’, LTR-ST-1928, National Research Council Canada. [4] Rutledge, R., Hiscocks, R., and Backman, D., (2006) ‘Strain Survey for FT312 CF-18 Wing-Fold Aft Spar Shear-Tie’, LTR-SMPL-2006-0072, National Research Council Canada. [5] Rutledge, R., Hiscocks, R., and Backman, D., (2007), ‘FT312 CF-188 Wing-Fold Aft-Spar Shear-Tie Fatigue Test Final Report’, LTR-SMPL-2007-0103, National Research Council Canada. [6] Macleod, G., (2006), ‘Aft Spar Shear-Tie Component Test – Spectrum Delivery’, L3 Communications (MAS) Canada, GDM/M06/05.

Evaluation, modification and damage tolerance of an in-service aircraft critical area

327

[7] MacLeod, G., (2006), ‘Aft Spar Shear-Tie Component Test – Post-Mod Requirements’, SES13980.CP2.WNG.ALEX66.CT.4 ALEX 66 Component Test, L3 Communications (MAS) Canada, GDM/M02/06. [8] Backman, D., Vesely, P., Laliberte, J., (2004), ‘Validation of Three Dimensional Digital Image Correlation for Aerospace Applications’, LTR-SMPL2004-0031, National Research Council Canada. [9] Chu, A., Ranson, A., Sutton, M., and Peters, A., (1985), "Applications of digital image correlation techniques to experimental mechanics," Experimental Mechanics, 25 (3), 232-244. [10] Sutton, M., McNeill, S., Helm, J., and Chao, Y., (2000), "Advances in twodimensional and three-dimensional computer vision," Topics in Applied Physics, 77, 323-372. [11] Sutton, M., Wolters, W., Peters, W., Ranson, W., and Mcneill, S., (1983), "Determination of displacements using an improved digital correlation method," Image and Vision Computing, 1 133-139. [12] Helm, J., McNeill, S., and Sutton, M., (1996), "Improved three dimensional image correlation for surface displacement measurement," Optical Engineering, 35 Rutledge, R., Hiscocks, R., Backman, D., Chisholm K. (7), 1911-1920. [13] and Petitjean, J., (2006), ‘3434 Simulated Flight Hour Strain Survey for FT312 CF-18 Wing-Fold Aft Spar Shear-Tie’, LTR-SMPL-2006-0118, National Research Council Canada. [14] MacLeod, G., (2006) ‘Aft Spar Shear-Tie Component Test – Post-Mod Requirements’, GDM/M02/06, SES13980.CP2.WNG.ALEX66.CT.4 ALEX 66 Component Test, L3 Communications (MAS) Canada. [15] Roth, M. and Sander, K., (2006), ‘F/A-18 INTERNATIONAL FOLLOWON STRUCTURAL TEST PROJECT FRACTOGRAPHIC ANALYSIS REPORT NSD31201-1-1 FT312 68% Spar Shear-Tie Component Fatigue Test’, QETE A006004-1 FT312-001, Quality Engineering Test Establishment Canada. [16] Underhill, R., (2006), ‘F/A-18 INTERNATIONAL FOLLOW-ON STRUCTURAL TEST PROJECT FRACTOGRAPHIC ANALYSIS REPORT NSD31201-2-1 FT312 68% Spar Shear-Tie Component Fatigue Test’, RMC-2006001, Royal Military College Canada. [17] Berens, A., Hovey, P., and Skinn, D., (1984), ‘Risk Analysis for Aging Aircraft Fleets, vol. I, Analysis’, Technical Report WL-TR-91-3066 OH 454336553, Flight Dynamic Directorate, Wright Laboratory, Air Force Systems Command, Wright-Patterson AFB.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

PROCEDURES FOR AIRCRAFT STRUCTURAL TEARDOWN ANALYSIS: DEVELOPMENT OF A BEST PRACTICES HANDBOOK Dr. Gregory A. Shoales, P.E. Center for Aircraft Structural Life Extension HQ USAFA/DFEM, 2354 Fairchild Dr., Ste. 2J2A, USAF Academy, Colorado 80840-6240 USA

Abstract: Safely sustaining the ever-increasing numbers of aging aircraft in the United States Air Force (USAF) has brought a correspondingly increased demand to determine the true condition of service-aged aircraft structural components. The only means available to precisely determine the damage state of a given structure is by what is commonly referred to as a “teardown inspection” of that structure. A very large number of these teardown programs have been executed in recent years. These include: A-10, AV-8B, B-2, B-52, B-727, C-5, C-17, C-130, C141, EC-135, F-15, F-16, F-22, F-111, KC-130, P-3, T-37, T-38, VC-10 and numerous civil and general aviation aircraft. Unfortunately, there is little to no cross talk between organizations performing teardown programs. The lack of communication limits the sharing of lessons learned and best practices developed during the course of each program. The USAF Academy’s Center for Aircraft Structural Life Extension (CAStLE) was tasked by the 77th Aeronautical System Wing, Wright-Patterson AFB, Ohio to prepare a handbook which documents task-by-task procedures for planning and executing an aircraft structural teardown program. CAStLE was further tasked to, wherever possible; include lessons learned from past programs in this handbook. The resulting handbook was published as M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 329–353. © US Government 2009. Created within the capacity of an US governmental employment and therefore public domain. Published by Springer Science+Business Media B.V. Dordrecht.

330

Gregory A.Shoales

“USAFA-TR-2008, Procedures for Aircraft Structural Teardown Analysis” in early 2008. The handbook chapters address all teardown tasks from setting program goals and requirements through the analysis of the resulting teardown program findings. This work presents a brief summary of the best practices handbook chapters and highlights some of the more critical aspects of each.

BACKGROUND AND INTRODUCTION Safely sustaining the ever-increasing numbers of aging aircraft in the United States Air Force (USAF) has brought an ever-increasing requirement to determine the true condition of service-aged aircraft structural components. The only means available to precisely determine the damage state of a given structure is by what is commonly referred to as a “teardown inspection” of that structure. Many such programs have been conducted by operators of both military and civil aircraft fleets. Programs have ranged from complete teardown of a given structural component to teardown of all major structural elements in a single aircraft or even the same elements in several aircraft. Teardown analysis program (TAP) reports yield valuable data for aircraft fleet managers. However, cross-platform sharing of successful teardown procedures and other lessons learned is often lacking. Recently, The Technical Cooperation Program (TTCP) between the Governments of Australia, New Zealand, Canada, United Kingdom and United States published a handbook to document best practices for aircraft teardown programs [1]. The TTCP handbook focused on policy decisions and other programmatic issues as they applied to the international military and civil community. However, the procedures and requirements to be considered during each task of a teardown program were not addressed in great detail in the TTCP handbook. The USAF Academy’s Center for Aircraft Structural Life Extension (CAStLE) has had various levels of involvement in six different teardown programs. CAStLE’s participation has ranged from membership in technical review panels to planning, directing and executing entire teardown programs. Given their experience across a variety of recent teardown programs, CAStLE was tasked by the 77th Aeronautical System Wing, Wright-Patterson AFB, Ohio to document task-by-task procedures for planning and executing a teardown program and wherever possible capture lessons learned from past programs. In order to answer the task of capturing lessons learned, CAStLE requested input from both military and U.S. Air Force (USAF) fleet operators. Teardown reports were received and analyzed from Ogden and Tinker USAF Air Logistics Centers (ALC) as well as the U.S. Navy, U.S. Coast Guard, Delta Airlines and Federal Aviation Administration (FAA). In all, the reports of more than two dozen teardown programs were analyzed for this effort. The resulting teardown best practices handbook was published in early 2008 as “USAFA TR-2008-02, Procedures for

Procedures for aircraft structural teardown analysis development

331

Aircraft Structural Teardown Analysis” [2] but has become informally known by its acronym PASTA. The chapters of PASTA are organized in order of task accomplishment during the execution of an aircraft structural TAP. The solid line in Figure 1 illustrates the flow of these tasks in a typical program. The dotted line from the “Identify Purpose and Requirements” block to each of the other task blocks emphasizes the inter-dependence between this first critical task and all tasks which follow. The remaining sections in this work describe the chapters of and highlight the important points of each.

Figure 1: Typical teardown program task flow shown by the solid line. The dashed line indicates the inter-dependence of the first task on all tasks which follow.

IDENTIFY PURPOSE AND REQUIREMENTS This chapter of PASTA presents a variety of issues that must be addressed before embarking on any TAP. These issues set the tone for the entire program and will therefore govern the level of detail for each program task. Identify the Purpose There are a wide variety of reasons why a teardown program may be required or desired. The TTCP Handbook [1] addresses a number of economic and life cycle issues that could drive a program manager to consider some level of a teardown program. PASTA assumes that the decision to pursue a teardown has already been

332

Gregory A.Shoales

made and focuses on the primary technical justifications for making such a decision. The first two technical reasons for conducting a are mandated by the USAF Aircraft Structural Integrity Program (ASIP) Standard [3]. The Standard requires that an assessment of damage state be conducted at the conclusion of full scale durability and damage tolerance testing. The teardown data obtained after these tests are part of the process of identifying critical structural details, establishing inspection intervals and developing damage prediction models. Simulated usage tests such as these are generally based on assumed load spectra. As a result, a teardown program conducted on an aircraft removed from service after a period of known operational usage still provides the most reliable data to validate these models. The Standard also requires a condition assessment after known usage if the aircraft is expected to operate beyond its service life or when it is suspected that service damage may jeopardize the aircraft’s structural integrity. A further reason to conduct a teardown is to validate non-destructive inspection (NDI) procedures. While this purpose may not be the sole reason to justify a teardown, validation of inspection procedures is frequently a “target of opportunity” during TAPs. Incorporating such inspections into any TAP impacts both the program schedule and cost. However, adding such a task provides a unique opportunity to evaluate the effectiveness of the established procedures on actual aircraft structure. Since a part level inspection followed by failure analysis is the only way to verify the true damage state at the site of any NDI indication, most program managers elect to incorporate TO inspections into their teardown programs. Required Fidelity of Findings Easily, the most important decisions to be made in any teardown program are what type of findings will the program be focusing upon and what degree of fidelity will be required in these findings. It is vital that these decisions be made at the very onset of the TAP as it will drive all program tasks and impact the budget and the schedule. It is important to ensure that this decision is based upon what is “required” rather than what is “desired.” Generally speaking, it is human nature to want to know everything and to obtain as much data as possible. Therefore, this decision of finding fidelity is arguably the hardest one to make in any teardown program. An easy, and ill advised, way out is to decide that the required fidelity is “whatever can be found”. From the technical point of view of the NDI vendor or the failure analysis lab, this is the correct choice since it involves the most work in their respective fields. Laboratory NDI techniques can be applied to look for defects that could not be found by any field deployed method. Failure analysis labs can identify fatigue cracks, complete with striation spacing at multiple locations, which are less than 100 nm in length. While any good lab might enjoy being tasked to this level of effort, it may not best serve the program’s available budget or timetable for achieving the results.

Procedures for aircraft structural teardown analysis development

333

The required fidelity must be tied back to the purpose of the teardown program. If the goal is to validate damage prediction models, there should be consideration of what minimum size findings would be useful for such validation. Since damage tolerance analysis begins from the rouge flaw size, 1.27 mm for most structure (0.127 mm for cold-worked holes and 0.5 mm for durability analysis), there may be no utility in data for cracks which are dramatically smaller. Other Considerations There are further considerations that are best addressed at the very beginning of any teardown program. The program must establish what will be needed for analysis and for long term archival. Most current programs opt to electronically retain all data generated given the availability of database systems. The data not only includes the products of NDI and failure analysis but also any program management documents which could help explain why program decisions were made. Databases may be used for program management such as tracking teardown subjects or task progress as well as providing easy access to analysis results. Consequently, the database must be designed to permit easy access to any stored data required to satisfy program requirements. A further consideration of any teardown program management team is to not operate in a vacuum. Given the number of programs, any organization seeking to undertake a new teardown should have access to other organizations with teardown experience. Teardown program managers are always advised to seek expert assistance. Teardown subject size is ultimately a decision based on both cost and schedule considerations. Even so, it is important to bear in mind the impact of sample size decisions upon the results. As with any experimental process, it is desired to represent a population with a sample. In other words, the ASIP manager would like the data from the TAP to help make fleet decisions. A logical question would always be, “Do the teardown data represent the fleet or are they unique results?” Therefore, it is always worthwhile to consider the sample size at the beginning of the program and ensure the chosen subjects satisfy the program requirements.

SELECT VENDORS The selection of an appropriate vendor to conduct the various teardown tasks is critical to the success of the entire teardown program. Careful and prudent selection at the beginning of the program will pay huge dividends in terms of the quality and fidelity of the results in addition to meeting cost and schedule requirements. Each vendor must fully demonstrate their capacity and competence to complete the tasks they intend to support, whether that task is disassembly, part preparation, NDI, failure analysis, or data assembly and analysis. The vendor

334

Gregory A.Shoales

selection strategy depends upon the scope and timeline of the program. There are a number of factors which should be considered whether individual vendors are chosen for each task or whether vendors are chosen which can execute all tasks. Past Experience More so than with many tasks typically assigned to vendors, teardown offers many opportunities to destroy critical and unique information. Improper disassembly, cleaning, coating removal, polishing, and crack opening are just a few examples of tasks which, when performed improperly, can result in the loss of one of a kind data with no way to recover. For this reason, potential vendors considered for teardown program tasks must demonstrate previous success in conducting work of a similar nature. Facilities and Equipment Potential vendors must have the appropriate facilities, tools, instrumentation and analytical equipment, as described in the applicable chapters of PASTA, to perform the specific tasks assigned. Qualifications and Certifications Various recognized certifications are applicable to demonstrate the qualifications of vendors seeking to execute teardown program tasks. For example, potential vendors seeking to perform NDI tasks must have a written practice as well as a training and certification program meeting the requirements of NAS 410 or SNT-TC-1A. Since recent experience and skill are critical to a robust inspection process, all individuals conducting NDI processes must be certified to Level 2 or Level 3 in the method to be performed and must have actively performed the specific methods and techniques within the past six months. Proficiency and Constancy Given the criticality of each teardown task, vendors should be expected to demonstrate their proficiency. If qualification to perform any task is dependent upon the individuals present at the time of the qualification, it is worth exploring the vendor’s retention rate. Furthermore, all program managers should include periodic performance audits of all their vendor program participants. Cost Cost will always be considered in the final vendor selection. However, it must not become the primary consideration. The quality of the resulting analysis is paramount and must not be compromised merely by selection of the lowest bidder. Source selections typically result in choosing the fully qualified vendor having the lowest cost. Therefore, it is imperative that selection criteria are written such that unqualified or even marginally qualified vendors are identified and eliminated during the selection process.

Procedures for aircraft structural teardown analysis development

335

IDENTIFY TEARDOWN SUBJECTS AND EXTRACT The selection of the teardown subject is nearly as important as setting the required fidelity of the findings decision. Accurately setting the fidelity of findings to address the program requirements will ensure the data provided meets the teardown purpose. Proper selection of the teardown subject will address the utility of that data. Teardown subjects are both the components removed from the aircraft for teardown analysis as well as the aircraft from which they are extracted. The following sub-sections discuss the most important aspects of subject selection from considerations of service history to the separation of those subjects from nonprogram aircraft structure. Known Service History Regardless of the teardown program purpose, the findings obtained will need to relate back to the applicable service history. Depending upon the damage mechanism, the relevant service history of a selected teardown subject may be in terms of operational hours, ground air ground (GAG) cycles, number of certain mission types flown, calendar time or a combination of factors. Therefore , like every other task, the program requirements are essential in identifying an appropriate teardown subject or at least ensuring the subject chosen is accompanied by a complete set of relevant service history. Operational hours, Cycles and Missions. If the primary interest of the program is fatigue cracking then the relevant service history is stress spectra which would result in damage. Such service history is not to be confused with simple total flight hours. Fatigue relevant history requires more detail depending upon the structure being evaluated. Fatigue cracking in wing structure would require stress spectra at the point of evaluation. Fatigue cracking in fuselage structure is more likely dependent upon fuselage pressure cycles. A complete load history must not only include normal operational missions, but also any special use periods of time which may have included periods of non-standard missions. The key to service history is completeness. Any uncertainty in mission history will be carried forward to program analysis. Calendar Time and Environment. In contrast to fatigue damage which depends on cyclic application of load, most corrosion damage mechanisms only require the presence of a corrosive environment over a period of time. The period of time is measured from the original manufacture of the teardown subject parts to the time of the teardown analysis. The severity of the corrosive environment depends mainly upon the aircraft’s geographic basing locations. Additionally, special missions which result in unique exposure to chemicals or agents, such as fire fighting aircraft or sprayers, can have a significant impact as well. In the simplest terms, a corrosive environment refers to the presence of an electrolyte enhanced by higher activation energy. Therefore, hot, moist climates are more corrosive than cold, dry climates. Sources of moisture can be from pooled water or simply from

336

Gregory A.Shoales

the humidity in the air. Furthermore, the electrolytic nature of seawater promotes corrosion faster than pure water. Therefore, some of the most corrosive environments include air bases located directly on ocean coasts. Repair History. A thorough review of repair history is required to fully understand any potential teardown subjects. Repair history represents an essential element of service history which must not be overlooked. If the program objective is to evaluate high time components, then it follows that the service history of the individual components should meet that requirement. As an example, the teardown analysis of the C-130 center wing conducted by CAStLE [4] required evaluation of all lower surface structure, as well as the center to outer wing joint regions from a high time aircraft. A review of the maintenance records, however, revealed that one of the four wing joints had been replaced in the very recent past. This joint region had relatively little usage and did not meet the program requirement. Based on the assessment of maintenance data, the low time wing joint was removed from the list of teardown subjects. Any uncertainty in mission history will naturally be carried forward to program analysis. Failure to have complete and reliable service history will make analysis of teardown program findings difficult at best and useless at worst. Extract Teardown Subjects After the subject aircraft has been selected for a teardown program the next task is to remove the components which constitute the focus of the program. A key consideration in the extraction of teardown program subjects is to avoid inducing incidental damage by the extraction process itself. Such incidental damage would cloud the inspection results and waste program resources as it would ultimately need to be distinguished from damage incurred during service. For smaller aircraft, sub-assemblies may be extracted directly from the airframe with minimal special provisions. Larger airframes, such as transports and bombers, require additional considerations. In such structure, the weight involved with the individual extracted components, as well as the dependency of one section of structure to support the weight of another, requires significant preparation prior to any sub-sectioning and component extraction. In these cases, the aircraft original equipment manufacturer (OEM) is often a useful resource to provide component weight data, support requirements and plans to stabilize the remaining structure as component assemblies are removed. Support and stabilization requirements often dictate the order in which component assemblies are removed from a given teardown aircraft. For example, if the stabilization plan makes use of the main landing gear structure then it follows that structure near the landing gear will be the last region extracted from the subject aircraft. Similarly, if certain components, such as empennage, are readily removed using established maintenance procedures then, regardless of program analysis priorities, these assemblies are likely to be the first to be removed.

Procedures for aircraft structural teardown analysis development

337

Figure 2 shows an example of extracting major assemblies from a KC-135R beginning with the empennage and working progressively closer to the main landing gear structure.

a

b

Figure 2: Extraction of major assemblies from a KC-135R beginning with a) the removal of the vertical tail and b) progressing towards the main landing gear (Tinker AFB image). Prior to making any of the section cuts, a detailed cut plan should be developed and approved by the program manager. This plan not only includes the aforementioned stabilization requirements but ensures the inclusion of all teardown subjects. Considerations when defining section cuts include: Provisions for handling of removed sections in terms of size and weight Cuts made sufficiently far away from teardown subject locations so that the cutting process prevents incidental damage to the subjects themselves Minimizing cutting required through complex, thick or otherwise difficult structure Combining a number of teardown program subjects included into one assembly to maximize removal efficiency Preserving remaining structure for potential follow-on programs

DISASSEMBLY After component assembly (with their teardown program subjects) extraction, the process begins to execute the program evaluation requirements. As discussed in previous sections, all programs have specific structure designated for damage evaluation. Additionally, the types of findings and desired fidelity have been defined to meet the program goals. Rather than attempting to perform metallographic failure analysis investigations of all designated structure, teardown programs make use of established NDI techniques to focus these investigations. Proven NDI methods are used to obtain indications of potential damage to ensure failure analysis resources are concentrated on the most appropriate portions of the

338

Gregory A.Shoales

subject structure. Since it is unacceptable to miss damage indications during these NDI inspections, measures are taken to ensure the highest practical quality in the NDI results. One of these measures includes removing all mechanical fasteners and performing the teardown program NDI only on single layer parts, rather than attempting to inspect multi-layer part assemblies. Disassembly Requirements Component disassembly for teardown programs differs from that of aircraft maintenance in that joined parts must be separated from one another without inducing damage to the joined parts themselves. When fasteners are removed during structural maintenance operations, it is customary to oversize or in some way refinish the bore of the fastener hole to prepare for fastener replacement. In contrast, for structural teardown analysis, the hole bores are often primary targets of the investigation; preserving the condition of the hole bore while removing the fastener is of paramount importance. Any damage induced during the disassembly process will create the extra task of distinguishing that damage from the operational damage data sought by the program. Since disassembly results in the creation of separate parts and the exposure of many faying surfaces, part identification, tracking and documentation take on additional importance during this task. Disassembly Tasks The sub-sections of this chapter present the details of the primary teardown program disassembly tasks. This chapter of PASTA gives detailed procedures and examples of recommend tools wherever applicable. Component Photo Documentation and Teardown Subject Part Identification. Before beginning the disassembly of an extracted component, the teardown subject parts must be identified. Since the precision fastener removal procedures described in the disassembly chapter of PASTA are designed to minimize damage to joined parts, it is important to identify the parts which require the most attention. If a program subject part is attached to a non program subject part, then this information will guide the disassembly technician’s selection of tools and their application. For example, the component extracted from a T-37B teardown aircraft shown in Figure 3a contains a single teardown subject part. Figure 3b shows the wing carry-through fitting after disassembly. By knowing this information beforehand, the disassembly technicians could focus their efforts on efficiently preventing incidental disassembly damage to the critical component while not wasting effort on non critical scrap parts.

Procedures for aircraft structural teardown analysis development

a

339

b

Figure 3: Images from T-37B teardown program showing a) component assembly as removed from the aircraft and b) the teardown subject part after disassembly from surrounding structure (CAStLE image). Prior to disassembly, the inspection requirements for each teardown subject part should have been identified based upon the program requirements. This step ensures all parts are properly tracked throughout the subsequent program tasks. Accordingly, upon disassembly each program part should be tracked with a parts log of appropriate identifying data. Some examples of information which should be recorded in the part tracking log are as follows: Assembly identification Part identification Part nomenclature Inspection requirements Assembly photograph identification information Part photograph identification information Visual inspection observation and disassembly notes To aid accountability of disassembled parts all program subject parts should be identified via part engraving or affixed tag. It is essential that the identification system be designed to survive all subsequent program tasks. Coating removal, described in the next section, puts the highest demands on identification system survivability. Past programs have had success with aluminum or stainless steel tags affixed by an aluminum or stainless steel wire. Fastener Removal. Several aircraft specific and general USAF maintenance documents provide guidance for performing fastener removal in aircraft structure [5]. The development of this chapter of the teardown best practice began with this guidance but emphasizes practices which are essential for teardown programs. The resulting PASTA disassembly chapter is too detailed for inclusion in this work. However, the following points summarize the philosophies used in the disassembly

340

Gregory A.Shoales

chapter to ensure the highest precision during fastener removal and to minimize incidental damage. Ensure the work surface provides adequate stability while working on the part Ensure adequate lighting exists to identify the important feature of each fastener Teardown subject parts should be secured while removing fasteners such that the: o Part will not move o Clamping system will not induce damage to critical areas of the part o Regions of the part around the fastener hole are backed up with thicker structure to prevent deformation. (This is particularly important for thin structure.) Removal techniques must emphasize preservation of the surface of the fastener hole bore: o Removal tools must not contact the hole bore or countersink surface o Fasteners should not spin during the removal process Incidental Damage Reporting. It is imperative that incidental damage be reported and tracked in any teardown program. These damage reports should be reviewed by program leadership in order to judge the impact on remaining program tasks. A history of incidental damage severity and frequency can also serve as a valuable tool to monitor vendor performance. It is also important that reports of damage reach the program technical management as soon as possible after the incident occurs. Overly frequent or severe instances may require retraining of that vendor’s disassembly technicians or in extreme cases reassignment of the remaining work. Reports of incidental damage should include, at minimum, the location, type, extent and cause of the damage as well as a detailed photographic record. The location should be given in terms of part number and aircraft coordinates to include any program specific location categories, such as a hole number scheme. If the location is near a fastener hole, it is important to note whether the damage is on the surface or in the fastener hole bore. The extent should be noted in terms of size and orientation if applicable. The photographic record must be sufficient to fully define the damage type, extent and location on the part or assembly.

CLEAN PARTS AND REMOVE COATINGS Upon disassembly of all extracted assemblies which contain teardown subject parts into those component parts, the parts must be prepared for NDI. The level of cleaning required, as well as which features must be cleaned is entirely dependent upon the TAP’s NDI requirements.

Procedures for aircraft structural teardown analysis development

341

Initial Cleaning and Sealant Removal To properly prepare a subject part for initial visual inspections and damage photographic documentation, the surfaces should be cleaned to remove all loose grease, sealant, dirt and particles. Specific guidance for cleaning can be found in Technical Order (TO) 1-1-691, Aircraft Weapons Systems Cleaning and Corrosion Control, Appendix E-2 and E-3 [6]. Remove sealants with sharp plastic scrapers or other removal tools that will not scratch or otherwise damage the substrate surfaces. Metallic tools should not be used. Chemicals that can assist in this task suggested by TO 1-1-691 are listed in this chapter along with their application. Coating Removal To complete surface preparation for NDI using methods such as close-visual, fluorescent penetrant and eddy-current inspections, it is important that all coatings be removed from parts after the initial visual inspections. To this end, it is critical to select and implement coating removal processes that: Thoroughly remove all surface coatings Result in a chemically clean substrate surfaces without residues, providing optimum surface conditions for subsequent inspections Do not attack the substrate materials and therefore preserve fractographic evidence There are two general categories of coating removal processes: chemical removal and media removal. The selection and mix of removal process will depend on the criticality or fidelity of the required analysis, the NDI methods that will be subsequently applied and local or regional environment regulations. Chemical Removal. Chemical stripping is the method of choice for removal of sealant, topcoat and primer, as these processes result in chemically clean surfaces which are optimum for accepting penetrant inspection. Properly applied chemical stripping will not modify the component substrate surfaces as will often occur with many mechanical stripping methods. Many of the conventional organic topcoats and primers can be readily removed using processes defined in TO 1-1-8, Application and Removal of Organic Coatings, Aerospace and Non-Aerospace Equipment [7], Chapter 2. Nylon bristle brushes and nonmetallic scrapers can be used to facilitate coating removal. However, some primers, particularly those used within fuel cells, will be resistant to chemical removal. When these primers are encountered, elevated temperature stripping processes may be required. All stripping processes should first be demonstrated for the coating types and substrate materials to be encountered in any given program to ensure effective coating removal without substrate attack. PASTA gives several examples of coating removal process successes as well as failures and strategies to ensure success. Because of the environmental concerns related to hazardous materials handling and waste disposal, particularly when removing chromated coatings, the use of chemical stripping processes may be restricted. A thorough review of local environmental regulations must be conducted prior to application of these

342

Gregory A.Shoales

processes. While some chemical choices are less hazardous than others, the majority of USAF coatings will contain elements which require special handling and disposal. Therefore, regardless of the hazardous material handling requirements of the chemical stripping agent, the user must consider the combined end product of the coating removal process. Media Blast. Media blast coating removal methods, (such as plastic media, grit blast and wheat starch) while attractive for their environmental reasons, are not recommended for use in detailed teardown programs. Media-blast methods are mechanical stripping processes that can modify (peen or smear) substrate surfaces and therefore negatively impact the performance of subsequent NDI processes. Prior to performing subsequent visual and/or fluorescent penetrant inspections of surfaces modified by media blasting, a chemical etch process is required to remove smeared surface material and reopen surface breaking discontinuities. Unfortunately, chemical etching can result in the destruction of key surface features, such as corrosion pitting or even the loss of fracture surfaces features such as fatigue striations. Media blasting of parts is only recommended for use on non-critical parts where inspections methods will be applied that are not affected by surface modification (i.e. eddy current inspections). Fastener Hole Preparation Fastener holes are generally given greater scrutiny, as they are often locations of crack initiation and propagation due to load transfer and stress concentration. The most common and effective method for inspecting fastener holes is rotary bolt-hole eddy current (BHEC). The fidelity of BHEC inspections is greatly dependent on the quality and cleanliness of the hole. Fastener removal, even when conducted by skilled technicians, can cause hole damage (scratches, gauges, nick and burrs) that, if not addressed, will prevent an effective eddy current inspection. Furthermore, non-relevant scratches, nicks and burrs can result in false positive indications (an indication when no detrimental defect is present) and result in additional, unnecessary and expensive fractographic analysis. Two primary methods are recommended for bolt-hole preparation: emery cloth and flex-hones. Both methods are described in greater detail in PASTA and are intended for use by a qualified inspector during the NDI task. The NDI task is presented in detail in the next section.

NONDESTRUCTIVE INSPECTION (NDI) NDI, also sometimes referred to as nondestructive evaluation (NDE) or nondestructive testing (NDT), is a process employed to interrogate structures and materials for discontinuities, defects and damage without causing damage to the component being examined or resulting in loss of vital metallographic evidence. The selection of the appropriate NDI method to employ depends on many factors including:

Procedures for aircraft structural teardown analysis development

343

Component geometry, size and material type Surface condition including coatings, plating, and finishes Expected damage, location, orientation and type of interest Part criticality Detection fidelity (capability) Cost

Establishing Teardown Program NDI Requirements Inspection processes should be implemented in an efficient fashion to maximize inspection fidelity for critical structures and minimize the cost for inspection of non-critical or secondary/tertiary structures, while ensuring the resulting teardown analysis is sufficiently robust to support reliable system management decisions. The scope and fidelity of the teardown effort will largely be driven by the program budget and schedule. Therefore, to optimize resources and manage cost and schedule risk, a tiered approach should be implemented. Such an approach focuses program resources in proportion to the structural criticality of the part or component being evaluated. The following sub-sections provide TAP NDI recommendations based on part category and/or criticality. Fracture Critical, Durability Critical, Mission Critical, Safety-of-Flight Components o Close visual inspection (CVI) o Bolt-hole eddy current (BHEC)–fastener holes and bore holes o Eddy current surface scan (ECSS)–edges, radii, critical details o Fluorescent penetrant inspection (FPI)–full field with attention to critical details o Enhance visual inspection (EVI)–only critical fastener holes and details as identified by the teardown program manager o Magnetic particle inspection (MPI)–critical ferromagnetic components such as fasteners, pins, rods and brackets Non-Critical Load Bearing Components o CVI o FPI o Limited BHEC or ECSS o MPI Secondary, Non-Load Bearing Components o CVI Establishing a Teardown Program NDI Level III Due to the complexity of the inspection technique selection process, a qualified NDI professional must be consulted when selecting the array of inspections methods to be applied on individual components. The NDI professional must have experience in applying an array of NDI methodologies, (i.e. an NDI Level III), Because of the criticality of this selection and the importance of ensuring robust NDI processes are established, it is recommended that the Teardown Program identify a responsible Teardown Program NDI Level III. This individual must be

Gregory A.Shoales

344

qualified, and possess demonstrated experience in implementing the various inspection methodologies to be utilized in the teardown program. The Teardown Program NDI Level III is responsible for: Reviewing and approving inspection call-outs for individual parts Developing and/or approving all NDI procedures and processes Review and qualification of all inspectors and inspection vendors Conducting periodic audits of inspection documentation and vendor inspection processes Providing guidance for resolution of audit findings and implementation of corrective actions Providing expertise for implementation of additional inspection processes. Typical Aircraft Teardown Analysis Program NDI Techniques The typical NDI methods that have been historically employed on aircraft TAPs are summarized in this chapter and listed previously in this section the present work. The data given in the PASTA’s NDI chapter includes the applicability and the current 90/95 probability of detection (POD) of each technique. A 90/95 POD defines the size flaw that will be detected with 90% probability within the 95% confidence bound. The PASTA NDI chapter includes sub-sections for each technique, each of which contains detailed descriptions, required equipment, training and procedures. The NDI chapter also provides advantages and limitations of each method from a number of resources including: TO 33B-1-2, Nondestructive Inspection Methods, Basic Theory [8] and the ASM Handbook, Vol. 17, Nondestructive Evaluation and Quality Control [9].

PRIORITIZE NDI INDICATIONS By the very nature of destructive TAPs, the number of subjects for each of the tasks just described tends to continuously increase. To satisfy the program analysis requirements a single teardown aircraft may generate many extracted assemblies. Each of these extracted assemblies will result in one or more teardown subject parts. Each part will need to be prepared for NDI by cleaning and coating removal. NDI of each part may result in multiple NDI indications. Each NDI indication must be considered for further failure analysis investigation. This growth in the number of subjects for each task as the teardown program progresses is illustrated simplistically in Figure 4. With the ever increasing number of subjects, prioritization is a powerful tool in any program to ensure finite program resources are best focused on achieving desired results.

Procedures for aircraft structural teardown analysis development

345

Figure 4: Simplistic representation of the growth in number of subjects for each task which occurs during any TAP (CAStLE image) Program requirements often serve to prioritize the extraction order of assemblies from a teardown airplane. Such priority would naturally carry through the subsequent tasks of disassembly, coating removal and NDI. However, especially with large aircraft, this input is often subordinate to jacking and shoring requirements as discussed in the previous subject extraction section. Recall that jacking and shoring requirements not only satisfy safety concerns but also are designed to minimize incidental damage to teardown subjects during the extraction process. Therefore, prioritizing the order of assembly extraction based on jacking and shoring also satisfies program requirements. As the number of subjects for each subsequent teardown program task increases the need to bring program data requirements into the prioritization scheme also increases. This fact becomes clear when extending Figure 4 to a realistic program scope. Each aircraft can produce hundreds of extracted assemblies. Depending on their size, each assembly likely results in dozens of disassembled parts. Each part

346

Gregory A.Shoales

could result in dozens to hundreds of indications. Investigating all indications by failure analysis techniques may be an unacceptable burden on both schedule and fiscal program constraints. Therefore, it is after the NDI task where prioritization based on analysis requirements of the teardown program becomes most important. Program managers are advised to have the prioritization system established before NDI indications present themselves. Even so, the prioritization scheme should also be a living system that can evolve throughout the program as data becomes available. Each TAP seeks to obtain a body of data for a given set of structures and frequently for a specific type of damage. As a result, NDI indication prioritization, like all tasks, is governed by the program requirements. This chapter is devoted to considerations and methodology for prioritizing NDI indications for further evaluation by failure analysis. The prioritization methods presented are tailored to the different types of program requirements.

PERFORM FAILURE ANALYSIS OF INDICATIONS Metallographic failure analysis is a destructive failure analysis technique, wherein the specimen is sectioned and broken in parts to identify and visualize the indications identified by the various NDI techniques discussed in the previous section. Since this is a destructive analysis technique, appropriate care and documentation is warranted at every step of analysis. The aim is to identify the source of the indication documented by the NDI inspector. If the indication is verified as a defect, then it is necessary to analyze the root cause and failure mode of the defect in accordance with the program requirements. This PASTA failure analysis chapter begins with a typical failure analysis task flow path. The flow chart depicts two major finding categories typical to most aircraft structure failure analyses; namely cracks and corrosion. Both of these defects are the result of aircraft usage, lead to reduced load bearing capacity of the structural member and can ultimately result in catastrophic failure if allowed to propagate unchecked. With current USAF emphasis on anticipation and management of damage rather than finding and fixing damage, complete and detailed characterization of these findings is a vital result for any TAP. Such detailed characterization includes the type, origin, propagation mechanism, propagation path, propagation rate, final size and final shape of the damage. Such data is necessary to establish the database required to support predictive model validation, develop and/or modify preventive maintenance schedules and evaluate the remaining life of the aircraft fleet. The remainder of this chapter is dedicated to a detailed description of how a program would obtain the required data. Personnel Qualifications and Equipment Requirements This chapter of PASTA begins by describing what should be considered minimum requirements for any program participants performing failure analysis evaluations. Metallographic failure analysis is both objective and subjective. Tell-tale signs of fracture features help to make the objective evaluations required during the failure

Procedures for aircraft structural teardown analysis development

347

analysis process. However, the presence of microscopic features often requires subjective interpretation. It is always helpful to have two pairs of eyes look at these subjective features to arrive at a consistent and reasonable conclusion. Successful interpretation of fracture surfaces and metallurgical observations also requires familiarity with the features under investigation. Therefore, there is no substitute for experience in correctly interpreting the fractograph or micrograph. Failure analysis must always be performed by a trained failure analysis engineer/technician who is fully conversant with failure mechanisms and features. Before a report can be finalized based on fractographic evaluation, it is necessary for a second failure analysis engineer to review and evaluate the data independently as a verification of the findings. Specifically the supervising failure analysis engineer should possess the following credentials: A technical degree in metallurgy, materials science, materials engineering, mechanical engineering or related field, with some coursework on physical and mechanical metallurgy Have completed a recognized failure analysis course offered by an accredited university, ASM or a similar institution Have at least 2 years experience working on the failure analysis of aircraft structures during the previous 5 years Be familiar with all relevant processes associated with the failure analysis tasks described in this chapter and theories behind them A working knowledge of the equipment used to execute the failure analysis tasks described in this chapter and be able to point out artifacts resulting from defective equipment or improper use of that equipment Equipment Requirements Similar to personnel requirement recommendation, the failure analysis chapter further identifies the minimum equipment requirements of any teardown program participants performing failure analysis evaluations as follows: Minimum equipment o Digital camera with 8 mega pixel (MP) resolution o Measuring equipments such as calipers, micrometer, mass balance, thermometers, thermocouples and thermistors o Precision saw o Specimen mounting o Specimen grinding and polishing equipment to ¼ micron finish o Fume hood for chemical etching of specimens o Chemical handling ability for acids and alcohols o Ultrasonic bath o Autoclave or similar oven capable of heating up to 400°F o Stereo microscope with at least 40X magnification o Optical microscope with at least 500X magnification o Scanning electron microscope (SEM) with energy dispersive Xray spectroscopy (EDXS) capability for elemental analysis

Gregory A.Shoales

348

All the equipment and measurement tools must be maintained in accordance with manufacturer’s guidelines and USAF TO-33K-100-1 and TO-33K-100-2. All the measuring equipment must be calibrated with standards traceable to NIST at least once a year or as per manufacturer’s guidelines. During periodic usage the equipment must be tested to make sure that calibration is still valid. Proper documentation of calibration/service/repair should be maintained at a central location. If a piece of equipment has a limited calibration certificate, that range should be clearly marked on the equipment. Personnel must be properly trained to use the equipment. Reference material such as ASM Handbooks Vol. 11, Failure Analysis and Prevention [10] and Vol. 12, Fractography [11] should be available for comparing failure surface micrographs to interpret the failure mechanism.

Failure Analysis Tasks The next section of the failure analysis chapter describes in great detail the specific tasks associated with conducting a proper failure analysis investigation on a given NDI indication. The tasks and sub-tasks included in this chapter are as follows: Visual examination Enhanced (high magnification) examination of the NDI indication site o Stereoscopic Analysis o Optical microscopy Specimen preparation o Sectioning o Mounting o Polishing Crack opening o Three-point bend method o V-notch method Scanning electron microscopy o Damage mechanism characteristics o Identification techniques Corrosion product evaluation Fracture surface corrosion product removal The sequence and selection of each of these tasks is dependent upon the result of the preceding failure analysis task which is illustrated by various flow charts and other guidance. PASTA’s failure analysis chapter cautions that assignment of these individual failure analysis tasks should not be a program management decision but rather left to the qualified failure analyst.

Procedures for aircraft structural teardown analysis development

349

Documentation of Findings After the analysis of an NDI indication has been completed, the details of the finding must be fully documented. This task also includes entry of pertinent data into the program database. Since failure analysis findings are the data which answers the program requirements, proper documentation of the findings is an essential outcome of any teardown program. While this documentation can take on many forms, the required content will remain the same. The report content must include a complete description of the indication leading to the failure analysis evaluation requirement, a description of analysis tools, and a finding summary to include the analysis methods, macroscopic description of the finding, and supporting evidence of the finding. This section of PASTA’s [2] failure analysis chapter summarizes the minimum content required in a report to document a finding and ends with examples of a possible summary report format. Indication Description. This first section of the report of finding fully describes the NDI indication which resulted in the requirement for a failure analysis investigation. It includes precise structural location and pertinent NDI details. This section is essential in order to correlate the finding back to the subject structure for analysis purposes. Failure Analysis Investigation and Finding Summary. This section summarizes the investigation by detailing the methodology and finding result. Macroscopic Indications and Finding Images. The images in this section help to identify the indication and support the finding. Particularly useful in this portion of the report are schematics which help the program management to visualize the damage. These schematics can also help to describe crack opening methods and other analysis details. Figure 5 shows examples of such schematics. Supporting Finding Evidence. This last section contains data which supports the finding. Figure 6 shows and example of such data used to support a fatigue crack finding. Programmatic Details. For programmatic purposes additional information may also be desired in the report of findings.

350

Gregory A.Shoales

Figure 5: Examples of schematic diagram used to help describe a failure analysis finding and the analysis methodology (CAStLE image).

Figure 6: Example of supporting data given in a report of a fatigue crack finding (CAStLE image). One last point emphasized in PASTA’s [2] failure analysis chapter is that a finding is the result of the failure analysis investigation of an NDI indication. A finding of damage which was not caused by operational usage such as flight loads or a corrosive environment is still a finding which must be documented. Findings which document the cause of an NDI indication also include material defects and

Procedures for aircraft structural teardown analysis development

351

mechanical damage. Recalling the program purpose of validating NDI techniques, findings of no detectable damage at an indication site are just as valuable to support the program requirements as would be a fatigue crack or any other type of damage.

ANALYZE AND REPORT FINDINGS In this chapter, the possible uses of teardown and failure analysis data are discussed. In concert with the typical teardown program goals discussed previously these include: The correlation of non-destructive inspection (NDI) indications and findings is investigated Inspecting assets other than the teardown article to understand the nature of damage found The fidelity of the operational fleet NDI procedures Fatigue life prediction model validation Reparability of the findings Corrosion prevention and control programs The results of a teardown program have many uses in fleet management activities. For damage found during the teardown program, the fleet manager must understand the nature of the finding in order to properly use the information to manage the fleet. For example, if a crack or corrosion (for brevity, the term “damage” is used to indicate either cracking or corrosion) is found, questions to be answered are: Is the cracking only found in the teardown article? o If so, what events, basing, load history, mission mix, maintenance history; or characteristics, manufacturing quality, structural modifications resulted in the teardown article being unique as compared to the fleet? Was the damage found in the full-scale or component fatigue tests during engineering and manufacturing development (EMD)? o If so, are the fatigue tests loads similar to service loading? Are there any other differences between the fatigue test and teardown article that would indicate they should not be part of the same population? Was damage found in fatigue tests conducted post-development? o Same questions as the comparison above to fatigue tests conducted during development. Was damage found in the fleet resulting in an entry in the Force Structural Maintenance Plan (FSMP)? o If so, was the damage found in just one aircraft or multiple? Was the usage of the teardown article similar to that of the aircraft for which the FSMP entries were generated? Are there any other

Gregory A.Shoales

352

differences between the FSMP entry aircraft and teardown article that would indicate that they should not be part of the same population? All of the questions above are guiding the fleet manager to decide if the damage is systemic and should be expected in most of the fleet or anomalous and only occurring in the teardown article. An example of systemic damage could be a poor detailed design of a particular part which resulted in high stresses from normal usage. Thus, if a crack was found on one aircraft in this detail, it is reasonable to assume other aircraft will develop the same cracking problem since the root cause is the high stresses due to the poor detailed design. Systemic damage requires a fleet-wide solution; whereas, anomalous damage does not have fleet-wide implications. The remainder of PASTA’s analysis and report findings chapter describes in great detail the methodologies employed to answer all of the questions offered and otherwise use the TAP findings to satisfy the program goals and requirements.

CONCLUSIONS AND FUTURE WORK This worked has described in some detail the chapters which comprise a USAF best practices handbook for planning and executing an aircraft structural TAP. PASTA is currently on contract as guidance for two current USAF teardown programs. It has also served as guidance for the C/KC-135 TAP’s development of teardown task protocols [12-18]. These protocols address the step-by-step requirements of each task include program specific requirements and provide a direct vehicle for capturing lessons learned from that program. Lessons learned from the C/KC-135 teardown and others along other with evolutions in best practices shall be incorporated into a future revision of Procedures for Aircraft Structural Teardown Analysis.

REFERENCES [1] Handbook for Best Practice in Teardown of Aircraft Structure (2005), 2002 Technical Co-Operation Program, Technical Panel AER-TP-4. [2] Shoales, G. (2008), USAFA TR-2008-02, Procedures for Aircraft Structural Teardown Analysis, USAF Academy, Colorado. [3] MIL-STD-1530C, DoD Standard Practice, Aircraft Structural Integrity Program (ASIP) (2005), Aeronautical Systems Center, Wright Patterson AFB, Ohio. [4] Shoales, G., Shah, S., Rausch, J., Walters, M., Arunachalam S. and Hammond, M. (2006), USAFA TR-2006-11, C-130 Center Wing Box Structural Teardown Analysis Final Report, USAF Academy, Colorado. [5] TO 1-1A-8, Engineering Manual Series, Aircraft and Missile Repair, Structural Hardware (2003), OC-ALC/LIRF, Tinker AFB, Oklahoma.

Procedures for aircraft structural teardown analysis development

353

[6] TO 1-1-691, Cleaning and Corrosion Prevention and Control, Aerospace and Non-Aerospace Equipment (2003), WR-ALC/LEET, Robins AFB, Georgia. [7] TO 1-1-8, Application and Removal of Organic Coatings, Aerospace and NonAerospace Equipment, Change 2 (2007), AFRL/MLS-OR, Robins AFB, Georgia. [8] T.O. 33B-1-2, Nondestructive Inspection General Procedures and Process Controls (2008), Tinker AFB, Oklahoma. [9] Davis, J., (Ed.) (1989), ASM Handbook, Volume 17, Nondestructive Evaluation and Quality Control, ASM International, Materials Park, Ohio. [10] Powell, G., (Ed.) (1986), ASM Handbook, Volume 11, Failure Analysis and Prevention, ASM International, Materials Park, Ohio. [11] Mills, K., (Ed.) (1987), ASM Handbook, Volume 12: Fractography, ASM International, Materials Park, Ohio. [12] Shoales, G. (2009), C/KC-135 Teardown Analysis Program Protocol Development and Validation,” Aging Aircraft 2009 Conference, Kansas City, Kansas. [13] Shoales, G. and Klahn, C. (2009), USAFA TR-2009-01, C/KC-135 Teardown Analysis Program Protocols, Protocol 3: Teardown Section Extraction, USAF Academy, Colorado. [14] Shoales, G. (2009), USAFA TR-2009-02, C/KC-135 Teardown Analysis Program Protocols, Protocol 4: Teardown Section Disassembly, USAF Academy, Colorado. [15] Galyon S. and Shoales, G. (2009), USAFA TR-2009-03, C/KC-135 Teardown Analysis Program Protocols, Protocol 5: Part Coating Removal, USAF Academy, Colorado. [16] Shah, S. and Shoales, G. (2009), USAFA TR-2009-04, C/KC-135 Teardown Analysis Program Protocols, Protocol 6: Lap Joint and Corrosion Preventive Compound (CPC) Evaluation, USAF Academy, Colorado. [17] Shoales, G. (2009), USAFA TR-2009-05, C/KC-135 Teardown Analysis Program Protocols, Protocol 7: Nondestructive Inspection (NDI), USAF Academy, Colorado. [18] Arunachalam S. and Shoales, G. (2009), USAFA TR-2009-06, C/KC-135 Teardown Analysis Program Protocols, Protocol 8: Failure Analysis, USAF Academy, Colorado.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

FATIGUE LIFETIME IMPROVEMENT OF ARRESTMENT HOOK SHANKS BY APPLICATION OF LASER PEENING* Mike Leap1, Jon Rankin2, Jim Harrison2, Serena Marley2, Lloyd Hackel2 and Joe Nemeth3 1

Naval Air Warfare Center, Patuxent River MD, Metal Improvement Company, Livermore CA, 3 Concurrent Technologies Corporation, Johnstown PA 2

Abstract: We have demonstrated in large scale geometrically modeled coupons an ability to significantly improve by a factor of more than 2.5x over conventional shot peening the fatigue lifetime of Navy aircraft arrestment hook shanks by applying laser peening to the key highly stressed area where the arresting shoe attaches to the shank. The deep level of residual stress (2.5 mm depth) provided by the laser peening is key to providing the fatigue improvement to this component loaded beyond the plastic limit.

LASER PEENING OF T-45 ARRESTMENT HOOK SHANKS During an aircraft arrest on a carrier, the heavy load transferred through the cable to the aircraft induces an equivalent axial and bending tensile stress load in the inner notched area of the hook shank end. In this application, we employ laser peening to induce compressive residual stress to depths of 0.100 inches (2.5 mm) to the fatigue prone notched end of the shank. This intense and deep level of compressive stress superimposes with the heavy load stress to significantly reduce the highest surface and near surface tensile loading and significantly improves the fatigue lifetime of laser

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 355–361. © Springer Science+Business Media B.V. 2009

Mike Leap et al.

356

Figure 1. T-45 aircraft used for Naval aviation training

peened components. Figure 1 shows the T-45 aircraft and Figure 2 shows the arrestment hook shank as mounted. The high stress area is at the hook end where the arrestment shoe fits on to the hook shank. Figure 3 shows more detail of the hook end and the high stress area of the shank.

T-45 arrestment hook shank

Figure 2. Arrestment hook shank as deployed on T-45 aircraft

Fatigue lifetime improvement of arrestment hook

357

Shank

High stress region

Figure 3. Details of arrestment hook shank showing high stress region

We focused our initial work on the T-45 aircraft hook shank and received crosssectional stress load data for its shank end from the OEM. We designed a large scale fatigue coupon using finite element stress analysis and iterated the design details to closely match those generated in actual T-45 shanks. We then built approximately 60 samples for surface treatment and fatigue and corrosion testing testing. Figure 4 shows the sample design. The offset geometry enabled an axial pull test rig to generate the representative axial and bending stress associated with hook shank loading. The raised boss in the center of the gage section is representative of the actual hook design. Fatigue cracks in actual hook shanks and those observed in testing our specimens, occur in the region between the boss and the outer radius of the shank giving us confidence that our design was representing the hook shank situation. In the case of our fatigue specimen we have a dual, left and right side region that replicates the shank geometry thus giving two points for failure in each test.

358

Mike Leap et al.

Figure 4: Specimen design to replicate arrestment hook shank

RESIDUAL STRESS AND FATIGUE LIFE IMPROVEMENT For the T-45 work samples were made of HyTuf steel, a high strength steel with yield strength of 185 ksi (1276 MPa). Samples were test laser peened and residual stress measured to assess the intensity and depth of the induced residual stress. Figure 5 shows the deep depth of residual stress induced by the laser process with compression extending to depths of 0.100 inches (2.5 mm) approximately 10 times deeper than achieved by conventional shot peening of the hook shanks. The surface compressive residual stress of approximately 200 ksi (1400 MPa) is instrumental in limiting the plastic deformation at the hook shank notch surface during peak arrestment loading. The notation indicates that 10 GW/cm2 irradiance was used with 18 ns pulse duration. The green curve shows results for 3 layers of laser peening whereas the red curve shows results for 4 layers of layer peening.

Fatigue lifetime improvement of arrestment hook

359

Figure 5: Residual stress induced by laser peening of Hy-Tuf steel

Samples were divided into three classes, unpeened, conventionally shot peened, and laser peened followed by conventional shot peening. The 10 GW/cm2 treatment using 3 layers of peening (300% coverage) was chosen for the process. Samples to be shot peened were processed with PS 14023 6 to 10A 230H shot 100% coverage. For the laser peening process, it was decided to apply the laser peening prior to an overall shot peening treatment. This was chosen so that the laser peening process could be inserted into the more standard method of hook shank fabrication and to eliminate any need to mask off the laser peened area during shot peening processing.

Mike Leap et al.

360 RELATIVE CRACK INITIATION LIFE 0

1

2

3

4

0.90

UNPEENED SPECIMENS (BASELINE)

CONVENTIONALLY SHOT PEENED SPECIMENS

CUMULATIVE PROBABILITY

0.50

LASER SHOCK PEENED SPECIMENS

0.10

0.05 0

1000

2000

3000

4000

CRACK INITIATION LIFE (SIMULATED ARRESTMENTS)

Figure 6: Laser peened samples provide a 2.5x lifetime increase over shot peened After surface treatment preparation, fatigue samples were checked for residual stress and then fatigue tested on a 100 kip test frame at NAVAIR in Patuxent River MD. A test load spectrum replicating that experienced during actual arrests was used. The testing procedure determined failure to be the initiation of an observable crack so testing was routinely stopped to examine for initiation and restarted with this procedure repeated until and initiation was observed. Figure 6 shows the test results clearly indicating that shot peening improves lifetime over unpeened specimens but that laser peening exhibits a lifetime 2.5 times greater than the shot peening. We are completing a final series of tests that are demonstrating fatigue life improvement of laser peened samples subjected to corrosion and the fatigue and

Fatigue lifetime improvement of arrestment hook

361

also to further lifetime improvement enabled by re-peening after 70% of fatigue life is expended. This work is being completed prior to qualification work. In production laser peening cost will be a few percent of the overall cost of replacement while enabling a 250% increase lifetime and also extending reliability and system availability.

Advanced materials and innovative structural concepts

25th ICAF Symposium – Rotterdam, 27–29 May 2009

DESIGN AND TESTING OF ADVANCED COMPOSITE LOAD INTRODUCTION STRUCTURE FOR AIRCRAFT HIGH LIFT DEVICES Tamas Havar1 and Eckart Stuible2 1

EADS Innovation Works, Department: TCC3, 81663 Munich – Germany Phone: 0049-89-607-27126, Fax: 0049-89-607-27796 Email: [email protected] 2 EADS Military Air Systems, Department MAE111 86179 Augsburg– Germany, Phone: 0049- 821 801-63470 Email: [email protected]

Abstract: This paper focuses on innovative composite load introduction structures for high lift devices and their design philosophy. In a joint project between different EADS Business Units, a new composite flap support system is developed for which design principles are investigated. The load introduction of these composite high lift elements is of most interest. The design of single pin load introduction can be divided into composite lugs and load introduction loops. Both designs are analyzed in detail with finite elements using non-linear boundary conditions (e.g. gap elements, deformable body contact definitions) and afterwards are verified by static and dynamic tests. It is shown that force introduction loops have advantages at in-plane tensile loads, but lead to poor performance at out of plane loads in comparison to composite lugs. Therefore load introduction loops can be used for limited design space and in-plane tensile loads. For multiple loadings in various directions composite lugs should be used instead. The investigations further show that friction within the bearing leads to a major load increase. Advanced composite linkage structures of the flap support system are designed and verified by full scale fatigue tests.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 365–374. © Springer Science+Business Media B.V. 2009

Tamas Havar, Eckart Stuible

366

INTRODUCTION Innovative composite structures are increasingly being used in the aircraft industry. A critical point of these composite parts is their attachment to the surrounding aircraft structure. In cooperation between different EADS Business Units, new advanced composite load introduction structures are developed to minimize weight and manufacturing costs (Fig. 1). The kinematics system consists of several advanced composite link structures with single pin load introductions, which are investigated in detail. Some studies are carried out for composite lugs with tension loads, which are defined as a composite lay-up with a circular hole and a distinct edge distance (Fig. 2, left) [1], [2]. For composite load introduction loop structures mad of fibres which revolve around the pin (Fig. 2, right), only little is known. Most investigation [2], [3], [4], [5] deal with pure loop structure without any middle laminate, which is usually placed between the loop fibres. Previous studies on composite load introduction loops with a middle laminate [6], [7] have shown high strength values for pure tension loads, but a poor performance under bending loads due to high interlaminar stresses. These relate only to the specific designs of the respective studies. Therefore, detailed numerical investigations are necessary to compare the different characteristics of both load introduction designs.

Figure 1: Advanced composite flap support system

Design and testing of advanced composite load introduction structures

Laminate with lug

367

Load introduction loop

Figure 2: Composite load introduction designs

FAILURE BEHAVIOUR OF CFRP LOAD INTRODUCTIONS The failure behaviour of the two different composite load introduction concepts is very different. The composite lug design fails at the flanges due to a fracture within the plies, which leads to an immediate total failure of the lug (Fig. 3, left). First ply failure can be concluded from a starting non linearity of the test curve, representing first matrix failure in single plies at the flange. The force introductions loops show a first failure by debonding (inter fibre failure IFF) of the loop plies from the middle laminate (Fig. 3, right). After the loop plies are debonded, the force introduction loop is able to carry further tension as well as compression loads, since both parts remain individually intact. A total failure occurs at much higher load levels due to a failure of the loop plies at the flanges similar to the lug design.

Figure 3: Failure of composite load introduction loops

Tamas Havar, Eckart Stuible

368

CFRP LOAD INTRODUCTIONS Two composite load introduction designs are analyzed with the finite element method using non-linear boundary conditions (e.g. gap elements, deformable body contact definitions) for different edge distances and three specific loadings (Fig. 4). Afterwards they are verified by static and dynamic tests.

Tension 0°

Combined loading 45°

Side load 90°

Figure 4: Loading of CFRP lugs and loops The different load introductions considered in this paper are modelled with 3D solid composite elements. One major advantage compared to common 2D investigations is that the resulting stresses are available in all three directions enabling a 3D failure analysis. Two basic lug designs are investigated. The first one is a lug design with quasi isotropic properties and the second one is a force introduction loop with Unidirectional (UD) plies in the loop. The designs are distinguished by their fibre orientation in the model allowing for equal material properties at each element for both designs. Therefore a direct comparison of the two designs is possible. Each element has transversal isotropic material properties: one layer of elements represents one CFRP layer. Laminate failure is analyzed individually for the matrix (first ply failure) and the fibre (residual) strength. Fibre fracture is calculated with the maximum stress criterion. The maximum bearable load for the matrix is investigated with a 3D failure analysis called simple parabolic criterion (SPC) [8], [9]. The investigations demonstrate that for pure tension loading, the load introduction loops show the best performance (Fig. 5), since the fibres are aligned parallel to the loading direction. For bending loadings that are perpendicular to the tension loads, the performance of the load introduction loop drastically decreases, because the load is carried internally primarily by the matrix and not the fibres. On the contrary, the lug shows a better performance for the bending load case (Fig. 5), since part of the fibres are aligned with the loading direction. For a mixed mode loading both designs show similar performance (Fig. 5). According to the design rules, composite load introduction loops should therefore only be used for structures, which are primarily loaded in tension. For structures with various loadings, lug designs are a better compromise.

Design and testing of advanced composite load introduction structures

369

Failure of CFRP Single Pin Load Introductions at various anglea Lug Design vs. Loop Design 700,0 600,0

Loop Failure Load

500,0 400,0 300,0

Lug

200,0 100,0 0,0

0

15

30

45

60

75

90

Load Angle

Figure 5: First ply failure of loop and lug design at different loadings

BEARING FRICTION ANALYSIS In addition bearing friction is considered in the analysis of the load introduction. Bearing friction occurs due to small movements during loading between two connected parts such as the flap and the wing, which are connected through a rear link. Former analysis methods consider equal friction in all directions based on the applied load, inner ball radius and the friction coefficient within the bearing (Eqn. 1). M = M1 = M 2 = M 3 = F ∗ r ∗ μ (1) 2

Fs rot 2

r

F1

Ro ta tio n

Ax is

M

Friction moment

2 3

rot 3

1

FEA model of bearing

3

rot 1

1

Rotation axes

Figure 6: Friction moment and rotation axes Further investigations show that friction occurs through shear forces at the liner within the bearing in rotation direction (Fig. 6). The rotation of the bearings can be

Tamas Havar, Eckart Stuible

370

divided in three directions, which are named after the effect on the link, which they are mounted in.

Figure 7: Friction moments of bearings Each moment is a product of the corresponding shear forces and lever arm. For the bending moment (M3, Fig. 7, left) small shear forces have small corresponding lever arms and large forces have large lever arms (Fig. 8, left) leading to a moderate friction moment. The largest friction moment is the rotation moment (M2, Fig. 7, middle), since small and large moments are multiplied by constantly large lever arms (Fig. 8, middle). The torsion moment (M1, Fig. 7, right) is the smallest, because large shear forces are multiplied with small lever arms and small shear forces are multiplied with large lever arms (Fig. 8, right).

Bending moment (M3)

Rotation moment (M2)

Torsion moment (M1)

Figure 8: Shear forces and lever arms leading to friction moments within bearings Based on non linear calculations, approaches for bearing friction analyses are developed. The analyses show that the friction moments vary for different rotation axes up to 42% (Table I). The highest moment occurs around the rotation axis of the bearing, which the bearing is normally tested by the supplier. For bending (swivelling of the bearing), which is the most critical loading for linkages, the friction moment is 15% lower than for rotation (Table I). For torsion (tilting of the bearing), the friction moment is smallest with 58% of the rotation moment (Table I).

Design and testing of advanced composite load introduction structures

371

Table I Load Case Rotation Factor ϕ i

Torsion

Rotation

Bending

0,575

1,000

0,843

These rotation factors can be applied to the bearing friction analysis, allowing for a more precise calculation depending on the rotation axis (Eqn. 2). M i = F ∗ r ∗ μ ∗ ϕi (2) These investigations show that the former approaches, where the rotation friction moment was applied for all three directions, are far too conservative.

FLAP SUPPORT SYSTEM A high lift support system for future aircrafts is developed based on the former investigations. It consists of a hybrid flap track functioning as load carrying fairing (composite skin) with an aluminum frame and an innovative composite kinematics system. The kinematics system has several composite linkages and fittings, which connect the flap to the flap track. The linkages with two attachment points are primarily exposed to tension and compression loads. Therefore, these linkages are designed as load introduction loops. Linkages with three attachments points and fittings typically carry multiple loadings and are consequently designed as composite lugs. Each linkage element is analyzed using non linear finite element methods showing sufficient strength for the occurring loads (Fig. 9).

Drive Strut

Front Link

Swing Link

Figure 9: Design of composite linkages Innovative manufacturing concepts for the composite linkages and the flap track are developed to minimize manufacturing costs. The complete composite flap support system is manufactured and verified on a prototype level (Fig. 10).

Tamas Havar, Eckart Stuible

372

Swing Link Prefrom

Drive Strut preform

Front Link UD braiding

Figure 10: Manufacturing of composite linkages

FATIGUE TESTING OF FLAP SUPPORT SYSTEM The flap support system is validated with a full scale fatigue test in accordance with Airbus specifications. Prior to the test, barely visible impact damages (BVID) are applied to the structure at critical areas. The occurring damages such as delaminations are verified and monitored with non destructive testing methods (NDT). At first, a static test is performed, simulating the maximum loads that can occur during flight. Afterwards, a full fatigue spectrum is applied, which demonstrates a complete aircraft life cycle (Fig. 11). Since the fatigue loads are much lower than the static loads, no fatigue damages occur during the fatigue test. The fatigue cycle is finished with a static test with the maximum occurring loads for the selected aircraft. In addition, safety factors are applied as load enhancements at the static tests. Prior to the second phase (damage tolerance phase) visible impact damages (VID) are applied to the critical areas. Afterwards, a limited fatigue spectrum is applied to demonstrate the robustness of the design. At the end, a final static test with the maximum occurring loads is applied. Safety factors and environmental knock down factors are applied as load enhancements at the static tests.

Design and testing of advanced composite load introduction structures

373

Figure 11: Full scale fatigue test The fatigue test proved that the composite flap support system is able to withstand all fatigue as well as static loads even with predefined defects.

REFERENCES [1]

Kassapoglou, Townsend, (2003), Failure Prediciton of Composite Lugs under Axial Loads, AIAA Journal Vol 41, No. 11.

[2]

Schwarmann, (2000), Handbuch Luftfahrttechnisches Handbuch.

[3]

Schürmann, (2000), Konstruieren mit Faserverbundwerkstoffen, SpringerVerlag.

[4]

Conen, (1966), Deformation und Versagen von GFK-Strangschlaufen, Journal Kunststoffe.

[5]

Bansemir (1981), Einleitung konzentrierter Lasten in Faserverbundstrukturen, Haus der Technik, Tagung: Krafteinleitungsprobleme bei Leichtbaustrukturen.

[6]

Siemetzki, Havar, Breuer, (2003), Structural Development of Mini-TrailingEdge-Devices in CFRP, SAMPE Europe Conference, Paris.

[7]

Havar (2007), Beitrag zur Gestaltung und Auslegung von 3D verstärkten Faserverbundschlaufen, Dissertation, Universität Stuttgart.

für

Strukturberechnung,

374

Tamas Havar, Eckart Stuible

[8]

Juhasz, (2003), Ein neues physikalisch basiertes Versagenskriterium für schwach 3D verstärkte Faserverbundlaminate, Dissertation, Technische Univeristät Carolo-Wilhelmia zu Braunschweig.

[9]

Cuntze, Deska, Szelinski, Jeltsch-Fricker, Meckbach, Huybrechts, Kopp, Kroll, Gollwitzer, Rackwitz, (1997), Neue Bruchkriterien und Festigkeitsnachweise für unidirektionale Faserverbundwerkstoffe unter mehrachsiger Beanspruchung – Modellbildung und Experimente, VDI 5 (596), Fortschrittberichte, VDI Verlag.

[10]

Brandt, Gessler, Filsinger, (2002), New Approaches in Textile and Impregnation for the Cost Effective Manufacturing of CFRP Aerospace Components, International Council of the Aeronautical Sciences Congress, Toronto.

[11]

Havar, Schouten, Roth, (2006), Composite Linkage System, Materialica – Composite Congress, Munich.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

HYBRID STRUCTURE SOLUTION FOR THE A400M WING ATTACHMENT FRAMES Matthijs Plokker, Derk Daverschot, and Thomas Beumler Airbus Deutschland GmbH

Abstract: The mainframes in the centre fuselage of the A400M have the vital function of introducing the wing loads into the fuselage. The lift distribution among the wing causing a torque and the frame curvature causing a bending moment, result in high running stress and limit loads in the frames below the rear wing attachment. These frames are made of 7000 series aluminium alloy and are primarily dimensioned by fatigue reasons, especially crack growth. Consequently, three different options for the A400M mainframes have been investigated. The first option is to reinforce the frames integrally. The second and third options are a titanium or GLARE® strap. In this paper, a trade off between the different options is made with respect to fatigue, maintenance, and weight targets.

INTRODUCTION The A400M is a military transport aircraft launched by Airbus in 2003. The aircraft is designed for replacement of the C-160 Transall and the C130 Hercules fleet. The A400M has a payload of 37 tons and a range of 1780 nautical miles with maximum payload. It can operate as “in theatre” tanker, tactical aircraft including areal delivery and as military transporter of paratroopers, tanks, helicopters and medical evacuation. The aircraft has a length of 45 metres and a wing span of 42 meter. The fuselage is divided in three sections: nose/forward fuselage, centre fuselage and rear fuselage. The centre fuselage is characterized by a centre cut out, where the high-wing is attached to the fuselage shown in Figure 1.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 375–385. © Springer Science+Business Media B.V. 2009

Plokker, Daverschot, and Beumler

376

A400M Center Fuselage

Aluminium

FML bonded strap

Figure 1: The A400M fuselage with the centre section and the division of the centre section Both the wing and the main landing gear are attached to the fuselage centre section. The loads coming from the wing are introduced by discrete attachments into the fuselage structure. Where the frames at the rear attachment are subjected to the loads originating from wing bending and wing torque (see Figure 2). These loads are for the main part of the fatigue spectrum tensional. Due to this load introduction and the curvature of the frame, the inner flange has to carry high tensile loads. The combination of the significance of the frames, the required inspection interval and the high running stresses makes these subjects of an intensive concept study.

Direction of Flight

Rear wing attachment frames

Figure 2: The side shell of the centre fuselage of the A400M with the rear wing attachment frames

Hybrid structure solution for the A400M wing attachment frames

377

CONCEPT STUDY The concept study was focused on the reinforcement of the inner flanges of the frames (Figure 3). The inner flange is the highest loaded and critical part of these frames. Three reinforcement options were investigated: 1. Thickening of the aluminium inner flange to reach an acceptable stress level. 2. Application of a riveted Titanium strap attached to the inner flange 3. Application of a bonded GLARE® strap with anti-peeling rivets In the options 2 and 3, the strap acts as a damage containment feature that in case of a damaged frame, redistributes the load retarding the damage growth in the frame.

Loft Outer flange

web Inner flange

Figure 3: cross section of frame with loft line The first part of the investigation was a comparison of the three design options on fatigue behaviour, damage tolerance and weight. Finally the costs of the favourised solution were determined. The integral reinforcement of the aluminium frame (thickening) is the heaviest solution. A thickness of around 22 mm would be required to obtain sufficient crack growth life to fulfil the inspection interval. Additionally, the crack growth rate of this single load path solution increases exponentially, which leads to critical inspection intervals means that it is harder to detect a crack before becoming critical. For the titanium and GLARE® strap, the product of the area and stiffness was kept the same compared to the initial design, to prevent global load redistribution or load path change.

Plokker, Daverschot, and Beumler

378

The titanium strap is riveted on the aluminium inner flange and sized to have an even load distribution between the constituents. The crack growth in the aluminium inner flange is retarded by the second load path of the strap. This retardation function can be taken from Rooke and Cartwright [1]. The GLARE® strap is adhesively bonded to the inner flange. The thickness ratio between the aluminium inner flange and the GLARE® strap is 1 to 2, to realize an optimum of damage tolerant behaviour of the reinforced frame inner flange: low crack growth rate and a long critical crack length. To determine the correction function for a bonded GLARE® strap a coupon test programme was conducted at the company GTM-Advanced Structure [2]. Test coupons from Al-7075-T6 and Al 7475-T761 bonded with GLARE®2A ranging from 11/10 to 16/15 aluminium/fibre layers with 0.4 mm aluminium were tested on crack growth. The test results showed that GLARE® reinforced aluminium has a low and constant crack growth rate for long ranges of crack lengths compared to normal aluminium or a titanium strap under same loading (see Figure 4). The improved crack propagation performance of the frame can be explained due to the “crack bridging” capability of the bonded strap: when the crack in the frame flange reaches a certain initial crack length, the crack opening is constrained by the strap. Crack growth life in Aluminium inner frame flange from 4mm flaw with different reinforcements 50 45

Crack length [mm]

40 35 30 25 20 15

Aluminium Integral

10

Titanium strap reinforcement Glare strap reinforcement

5 0 0

5000

10000

15000

20000

25000

Life [cycles]

Figure 4: Crack growth in the aluminium inner flange for the three different reinforcement options

Hybrid structure solution for the A400M wing attachment frames

379

The total weight of the titanium and the GLARE® strap including the aluminium inner flange was approximately the same. However, due to the riveting the titanium strap solution is heavier. The main advantage of the GLARE® strap compared to the titanium strap is the constant and lower crack growth rate and longer critical crack length. Additionally, the GLARE® strap is bonded which avoid fatigue sensitive locations. In Table 1, a qualitative comparison of the weight opportunity and the inspection effort to detect cracks can be seen for the different reinforcement options.

Option

Table 1: Crack growth in the aluminium inner flange for the three different reinforcement options Weight opportunity Effort to detect a crack (before becoming critical)

Integral Aluminium Titanium strap GLARE® strap

+ ++

-++

The A400M inspection requirement could easily be fulfilled with the GLARE® strap reinforcement. The GLARE® strap “bridges” the inner flange crack and offers a second load path (see Figure 5). Additionally, the strap offers a significant weight opportunity compared to the integral aluminium frame.

Figure 5: Aluminium frame with FML reinforced inner flange Final part of the trade study was the calculation of the recurring- and non-recurring cost for the option with the GLARE® strap. The non-recurring costs exist of extra design cost, extra cost for justification and cost for production equipment. The non-recurring costs were divided of the number of aircraft fitted out with the GLARE® strap (today 188). The recurring costs are cost for materials and assembly of the strap on the frame and transportation costs. The cost per kilogram

Plokker, Daverschot, and Beumler

380

weight save per aircraft were about 20% of the A400M programme specified limits.

MANUFACTURING The manufacturing concept is based on the existing GLARE® manufacturing techniques. The straps are milled out from a GLARE® sheet. The frames are Chromic Acid Anodised (CAA) and coated with a primer. The GLARE® straps are bonded to the inner flange in a second bonding cycle. The flanges of the strap are 1 mm smaller than the inner flange to get adhesive squeeze out on both sides. The process is schematically shown in figure 6.

Figure 6: Aluminium frame with GLARE® reinforced inner flange Due to the different coefficients of thermal expansion between the GLARE® strap and the aluminium frame a small contraction takes place after curing. If needed, this correction can be corrected by grid blasting. Completing the manufacturing is an anti-peeling fastener primarily installed in the run-out area.

QUALITY ASSURANCE AND INSPECTION The production quality assurance is again based on standard GLARE® qualification techniques. After lamination and curing, the GLARE® sheet is

Hybrid structure solution for the A400M wing attachment frames

381

inspected with ultrasonic through transmission with C-scan visualization. To inspect the frame-strap bond line, manual ultrasonic through transmission is used The maximum acceptable defect levels were defined in a document. The maximum allowable defect levels are tested in structural qualification test as well as in the full-scale fatigue test of the A400M.

FATIGUE AND DAMAGE TOLERANCE VALIDATION In this section the fatigue and damage tolerance validation is explained. First the fatigue justification is discussed. In the second part of the section the crack growth and residual strength analysis are dealt with. All calculations concentrate on the aluminium inner flange, since the GLARE® strap has much better damage tolerance properties than the aluminium. Fatigue calculation The first part of the fatigue justification is done with an allowable stress for the aluminium inner flange. The equivalent stress from the stress spectrum should not succeed the allowable stress for the aluminium frame. The fatigue behaviour of the assembly of inner flange with adhesively bonded GLARE® strap and with at some location fasteners is tested in coupon tests. The second part of the fatigue justification is the calculation of crack initiation from a rivet hole in the inner flange. Damage Tolerance calculation The damage tolerance calculation is based on an initial flaw at a rivet hole and from the edge. The first scenario is crack growth from an edge crack with a length of 1.27 mm (1/2 inch) to critical crack length. The second scenario is crack growth from a detectable edge crack to critical crack length. The third scenario is the crack growth from an initial flaw of 1.27 mm at a bore hole (used for anti-peeling fastener or bracket attachment) to critical crack length. The fourth and final scenario is the crack growth from a detectable crack at a bore hole to critical crack length. In all scenarios it is assumed that the 1.27 mm rogue flaw is a through crack. The stress spectrum is multiplied with the stress multiplication factor for the aluminium inner flange as an overall factor. The stress intensity at the crack tip is a function of the common parameters, like the stress range, the crack length and a geometry factor β, but also a “strap bridging” factor α (see Equation 2).

ΔK = α ⋅ β ⋅ ΔS ⋅ πa

(1)

Plokker, Daverschot, and Beumler

382

The “strap bridging” factor α is determined from the coupon test series and reduces the stress intensity at the crack tip. The factor is a function of crack length and has an exponential decay, which can be seen in Figure 7. Strap bridging versus crack length Strap bridging function

Expon. (Strap bridg ing function)

0.7 0.6

Alph a

0.5 0.4 0.3 0.2 0.1 0 0

10

20

30

40

50

60

70

80

90

100

crack leng th a ( m m)

Figure 7: The correction function alpha as function of crack length The crack growth analysis should show that the crack growth life from a detectable crack to critical crack length is longer than the inspection interval multiplied with a scatter factor. With high frequency eddy current in a special detailed inspection, a crack length of 4 mm can be detected. Residual Strength Analysis The residual strength analysis consists of the determination of the critical crack length and the maximum delamination size. The critical crack length in the aluminium inner flange is the longest crack length that can fulfils limit load conditions. The limit load is again the product of the limit load of the Finite Element (FE) and the specific stress multiplication factor. The second part of the analysis is to determine the maximum delamination size between the strap and frame that can sustain limit load. This was done with a fine mesh FE model by the company Advanced Lightweight Engineering (ALE) [3]. It has been proved by FE-analysis that with a delamination over the full width of the flange, with a length of 105 mm under limit load (tension or compression) condition, will not lead to delamination growth. By linearly scaling a maximum theoretical delamination of 1100 mm would lead to failure under limit load conditions. Nevertheless, an anti-peeling fastener will be installed with pitches >105mm, completing the design concept.

Hybrid structure solution for the A400M wing attachment frames

383

STRUCTURAL JUSTIFICATION The local metal bonding process on a large part with complex contour is considered as challenging. Therefore, the structural justification of the GLARE® reinforced inner flange was done with an own test pyramid consisting of coupon, component, and full-scale fatigue tests. The coupon test where crack growth was tested were performed to determine the “strap bridging” effect and to investigate the effects of the interface layers. Additional coupon test will be performed to investigate crack initiation and repair solutions. For the standard justification of the A400M mainframes, three static frame bending tests with standard production defects are performed (see Figure 8). These tests consist of a positive and negative bending moment on the run-out of the GLARE® strap and a negative bending moment on a continuous GLARE® strap.

Figure 8: Test set up of a frame bending test Additionally, four static frame bending tests will be performed that validate the maximum delamination size, determined in the FE-analysis. These test consist of the specimens with a continuous GLARE® strap and run-out of the GLARE® strap, loaded with a negative and positive bending moment. All the static test specimens are built up of a skin, stringers and three frames (see Figure 9).

384

Plokker, Daverschot, and Beumler

Figure 9: Test specimen for static frame bending test To validate the fatigue and damage tolerance behaviour of the reinforced frames, two frame bending test are performed. The first consists of a fatigue test of the runout including artificial delaminations. The second test investigates crack growth from initial flaws, at the edge of the inner flange and at the borehole, with a continuous strap. The specimens are loaded with in-service spectrum loading and after three design service goals with limit load. The fatigue frame bending tests specimens are performed on just a frame, representing the A400M frame design (see Figure 10).

Figure 10: Configurations of fatigue frame bending test specimens Finally, at the top of the A400M test pyramid is the full-scale fatigue test that has bonded GLARE® straps on six frames.

Hybrid structure solution for the A400M wing attachment frames

385

CONCLUSIONS The application of GLARE® bonded on the inner flange offers a good opportunity to improve the damage tolerance behaviour with a reduction of structural weight. The cost of production and justification of this application were well within the A400M programme boundaries. After the development of the production process and an inspection technique, a test and analysis programme is performed for structural justification. The first flying frames with bonded strap are built in the A400M.

REFERENCES [1]

Rooke D.P. and Cartwright, D.J., Compendium of Stress Intensity Factors, H.M. Stationary Office, London, Great Britain, 1975.

[2]

Hooijmeijer P.A., Roebroeks, G. and Gunnink, J.W., A400M Frame Reinforcement, Confidential report: GTM 2007-009, Den Haag, The Netherlands, 2007.

[3]

Witasse, R., Boer H. de, Koppert, J.J.M., Simulation of Bonding Strength for A400M Reinforced Mainframes, 07204/RWI, Delft, The Netherlands, 2007.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

CONTROL OF CRACK GROWTH RATES AND CRACK TRAJECTORIES FOR ENHANCED FAIL SAFETY AND DAMAGE TOLERANCE IN WELDED AIRCRAFT STRUCTURES Phil Irving1, Yu E Ma1, Xiang Zhang1, Guido Servetti1, Stewart Williams1, Gary Moore2, Jorge dos Santos3, Marco Pacchione4 1

Cranfield University, Cranfield, Bedford, MK43 0AL, U.K. 2 Bombardier Aerospace, Belfast, U.K. 3 Institute of Materials Research, GKSS, Germany, 4 Airbus, Hamburg, Germany

Abstract: The paper describes work that has been undertaken in controlling residual stresses in welded structures and in predicting crack growth rates and crack trajectories under the influence of friction stir weld residual stress fields. Experiments to explore the ability of stressing during welding to modify post-weld residual stress fields associated with the weld are described. They demonstrate that residual stresses can be successfully modified by this treatment. C(T) and ESE(T) samples in three different sizes, together with SEN samples to test crack deviation behaviour were machined from friction stir welded 2198 and 2195 material. Crack growth rates in a range of locations and orientations with respect to the weld were determined. To predict crack growth rates in the welded samples, stress intensity factors associated with the residual stress field were calculated and used to modify the values of the stress intensity factors due to the applied loads. Fatigue crack growth rates were calculated from parent plate data using the Harter T-Method. Crack trajectories were predicted using the maximum tangential stress criterion. Good agreement of model predictions with experiments for all crack paths and residual stresses was obtained. The implications for aircraft fail safety and damage tolerance of a capability to control and predict fatigue crack development in welded aircraft structures are discussed. M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 387–405. © Springer Science+Business Media B.V. 2009

388

Irving, Ma, Zhang, Servetti, Williams, Moore, dos Santos, Pacchione

INTRODUCTION Manufacture of metallic sections in future hybrid aircraft structures could involve use of adhesive bonding and welding to replace riveting for the joining of parts and sections. Weld processes selected for aircraft construction are high integrity ones such as friction stir welding and laser welding. In common with other weld processes, use of welds creates significant local inhomogeneities in material microstructure, in strength, toughness, and residual stress fields – all located in and around the line of the weld. Initiation and growth of fatigue cracks in the vicinity of the weld line is markedly affected by these inhomogeneities. Welding in addition creates large integral sections which require design and analysis techniques in order to improve and enhance their fail safety and damage tolerance. Weld residual stresses are tensile along the weld line and are associated with microstructure changes which act together to accelerate fatigue crack growth rates and are therefore detrimental to fatigue life and damage tolerance capability [1, 2]. However, residual stress fields always contain balancing compressive stresses. The stress field itself can be modified via post weld machining operations and /or local cold work. These operations result in reduced tension or even compression stress fields surrounding potential crack initiation sites. These could reduce crack growth rates and enhance fatigue and damage tolerance behaviour. It is also observed that fatigue cracks propagating in and around welds frequently deviate from the simple planar propagation perpendicular to the applied global alternating stress field [3]. The reasons for this are believed to be partly local microstructural inhomogeneities in the weld and partly interaction between weld residual stresses and global alternating ones. Thus the possibility exists of optimising damage tolerance and fail safety in welded structures via control of crack growth rates and crack trajectories. This attractive prospect is contingent on two key capabilities. • The ability to control development of microstructure and residual stress during and after welding so that desirable distributions with reduced crack growth rates are achieved. • The ability to predict crack fatigue growth rates and crack trajectories in response to material property distributions and to internal as well as externally imposed stresses. Residual stress modification and prediction of crack trajectories and growth rates Recent research has made significant progress in both control of weld residual stress and prediction of fatigue crack growth rates in the presence of local residual stress fields. The capability to modify in weld tensile residual stresses and produce even compressive stress fields via the application of cold work during and after the weld process has been demonstrated for a number of years [4, 5, 6], most recently in [7].

Control of crack growth rates & crack trajectories in welded aircraft structures

389

Previous research on prediction of fatigue crack growth rates in the presence of weld residual stress fields has employed two approaches. The first is the linear elastic fracture mechanics superposition method, which was originally proposed by Glinka [8] and Parker [9]. Under cyclic loads, the total SIF range (ΔKtot) and effective SIF ratio (Reff) are calculated as:

ΔKtot = ( Kapp,max + Kres ) − ( Kapp,min + Kres ) = ΔKapp

(1)

Reff = ( K app ,min + K res ) /( K app ,max + K res )

(2)

Thus, the Reff ratio changes due to the presence of residual stresses. Note that Reff is not the same as the nominal applied stress ratio (R). Kres the stress intensity factor at the crack tip due to the residual stress is the integrated contribution from the entire residual stress field in the sample, and can be either positive or negative increasing or decreasing Reff accordingly. As the residual stress field will locally relax as the crack grows through it, Kres will change with crack growth, in turn modifying Reff. In the simplest approach to life calculation valid for positive Reff, crack growth rates are those measured for parent material determined at R values equal to the current local value of Reff. Fatigue crack growth rates can then be calculated by integrating the Paris law in the usual way. The success of this approach has been demonstrated [8, 9]. For Kres values resulting in negative Reff the simplest approach is to consider that only the tensile part of the load cycle contributes to crack growth, following the crack closure approach of Elber [10]. The second method is based on Elber’s crack closure concept [10] that crack closure occurs even for fully tensile cyclic stresses. Newman’s crack closure model [11] was employed to calculate the crack opening stress So: S0 (3) = A0 + A1R + A2 R 2 + A3 R 3 (R ≥ 0) S max

S0 (-1 ≤ R < 0) (4) = A0 + A1R S max The expressions for the coefficients A0, A1, A2, and A3 can be found in [11]. Then the effective stress intensity factor range ΔKeff is given by: S (5) ΔK eff = ΔK (1 − 0 ) /(1 − R ) S max In the present work, Reff calculated via Eqn (2), was used in Eqn (3-5) instead of the nominal stress ratio R to account for the effect of residual stresses. After ΔKeff is found, fatigue crack growth rate can be predicted. It is now widely accepted that different da/dN vs. ΔKapp curves for respective R ratios can be correlated to a single “master” curve by using the effective stress intensity factor range ΔKeff. In the presence of welding residual stresses, a further complication arises, using either method, due to the fact that the weld microstructure is greatly changed form

390

Irving, Ma, Zhang, Servetti, Williams, Moore, dos Santos, Pacchione

that of the parent metal, and even in the absence of residual stresses, there may be differences in growth rate arising from this cause e.g. [6]. A number of studies have found that crack paths respond to mixed mode stress fields such as are found in welded structures via changes in their trajectory. Rubinstein [12] studied the effect of stress fields surrounding holes on crack paths and the effect of hole position. Fersini [13] used FRANC2D to calculate the mixed load crack growth of FSW lap joints. Tilbrook and Hoffman approximated curved crack paths under mixed mode loading [14]. To predict crack trajectories, two criteria have been suggested. These are the maximum energy release rate criterion [14] and the maximum tangential stress criterion [15]. In this study the maximum tangential stress criterion was used to predict crack trajectories. 1 3 1 1 (6) σ θθ = cos θ ( K I cos 2 θ − K II sin θ ) 2 2 2 2π r

1 (7) cos θ [K I sin θ + K II (3 cosθ − 1)] 2 2πr For maximum tangential stress criterion, that is ∂σ θθ = 0 or τ rθ = 0 , then the ∂θ angle of the crack plane θ can be calculated as: ⎛ 3K 2 + K 4 + 8 K 2 K 2 ⎞ I I II ⎟ (8) θ = cos −1 ⎜ II 2 2 ⎜ ⎟ K I + 9 K II ⎝ ⎠

τ rθ =

1

After calculating the crack increment and angle θ, crack trajectories can be predicted. This paper describes recent work that has been undertaken in controlling residual stress in welded structures, together with prediction of fatigue crack growth rates and crack trajectories under the influence of weld residual stress fields.

EXPERIMENTAL Aluminium alloys, sample dimensions and friction stir welding procedures The investigations used friction stir welds in a range of second and third generation aluminium-lithium alloys to investigate both residual stress field modification and prediction of fatigue crack growth rates and crack trajectories in welded samples. For residual stress modification experiments using the Global Mechanical Tensioning (GMT) technique [16, 17] the alloy 2199 was selected. The composition is given in Table I and the mechanical properties in Table II [18, 19]. Measurements of residual stress fields and fatigue crack growth rates to validate models for prediction of fatigue crack growth rates and crack trajectories were made on samples of friction stir welded 2195 Al-Li alloy. Crack path samples 3.2

Control of crack growth rates & crack trajectories in welded aircraft structures

391

mm thick were made of 2198-T8 alloy. Composition and mechanical properties of these three alloys are shown in Tables I and II.

wt. %

Si (max)

2195

0.12

Table I. Chemical compositions [18, 19] Fe Mn Cu Mg Cr Zn Zr (max) (max) 0.15

3.74.3

0.75

2198

0.08

0.10

3.5

0.5

0.8

2199

0.05

0.07

2.5

0.3

0.4

2195 [18] 2198 [19] 2199 [20]

0.35

0.08 0.16 0.18

0.6

0.12

0.36 0.05

Table II. Mechanical properties [18 -20] TYS (MPa) UTS (MPa) E (GPa) 580 615 79 436 491 77 390 428 78.5

Li

Ag

0.8-1.2

0.25 -0.6

1.10

0.50

1.6

Elongation (%) 9 14 11

Friction stir welding was performed using parameters specially developed for these alloys. Originally the friction stir welded plates were 1000 mm x 250 mm x 12.7 mm thick. Machining samples from the welded originals will reduce the stresses via removal of the constraint. A series of fatigue test sample geometries were machined from welded plates, as shown in Figure 1; the sample sizes are listed in Table III; in the process the sample thickness was reduced from 12.7 to 8 mm. Additional fatigue tests were performed to validate the capability of the tangential stress criterion to predict fatigue crack deviation in mixed mode stress states. The sample geometry used is shown in Fig 1(b). It consists of a simple plate 500 mm X 160 mm. The thickness was 3.2 mm in the region around the weld; elsewhere 1.6 mm. An edge crack was introduced at the mid point, and a hole of diameter 60 mm was located as shown in the figure. The effect of hole position (hence stress state) near the crack plane and residual stress field on crack trajectory was investigated in samples manufactured with and without a friction stir weld. The weld was located transversely at the sample mid point with the crack growing parallel to the weld line in the weld nugget. In order to investigate the influence of the sample dimensions on maximum residual stresses and fatigue properties, residual stress measurements were performed on ESE(T) and C(T) samples using neutron diffraction at Geesthacht Neutron Facility (GeNF) at GKSS. Neutrons of 0.164 nm wavelength were chosen with a silicon (311) monochromator. The scans were performed parallel to the notch. The scan line was placed at the height of the notch tip, and the first point was at a distance of 3 mm from the tip. The distance between the measuring points was also 3 mm. Each specimen was scanned three times, to measure the three

Irving, Ma, Zhang, Servetti, Williams, Moore, dos Santos, Pacchione

392

independent strains in longitudinal (LD), normal (ND), and transverse (TD) directions. Residual stress profiles were calculated from the strains.

160mm C(T)

ESE(T)

y

a0

x

h

500m

Weld

(a) Schematic of C(T) and ESE(T) (b) The offset hole SE(T) specimen samples & the location of the weld. Fig. 1 Fatigue test samples used in this work. Table III. Dimensions of C(T) and ESE(T) samples. Relationship with weld

Type

Sample size (mm): length x width

Crack parallel to weld Crack perpendicular to weld

C(T) ESE(T)

84 x 87.5; 120 x 125; 240 x 250 148 x 40; 185 x 50; 370 x 100

Control of residual stresses The Global Mechanical Tensioning (GMT) technique was selected as a means of modifying friction stir weld residual stresses. In this technique load in the weld direction is applied during welding. The effect of the applied longitudinal load is to reduce the amount of comprtessive plastic flow during the heating cycle and to induce tensile plastic flow during cooling [21]. The result of this is a linear dependence of the resultant longitudinal stress on the load applied [16, 17]. Samples of 2199 material 1 m x 0.459 m X 5 mm thick were friction stir welded parallel to the long axis. Samples were loaded to 17.5%, 35% and 50% of the material nominal room temperature yield strength during welding. To measure residual stress neutron diffraction measurements were performed on the SALSA beam at Institut Laue-Langevin (ILL , France. A description of the experimental details is given in [17].

Control of crack growth rates & crack trajectories in welded aircraft structures

393

Fatigue tests – experimental details Fatigue crack growth tests were performed on all samples to procedures in accordance with ASTM E647, in laboratory air at R = 0.1, with a fatigue load frequency of 10Hz. Additional fatigue crack growth tests were conducted at R = 0.6. Stress intensity factors for all specimens were calculated using the expressions recommended in ASTM E647. The electric potential method was used to monitor crack growth for C(T) and ESE(T) samples. The accuracy of crack length measurements was ±0.1 mm. To test the fatigue crack deviation samples, the samples were clamped at the ends rather than being pin loaded. Fatigue loading at an R ratio of 0.1 and a frequency of 10 Hz was applied. Crack lengths and crack trajectories were monitored using an automated optical video system which recorded the position of the crack tip in two dimensions. Residual stresses – ESET samples Fig 2 a & b shows measured residual stress profiles in the three ESE(T)sample sizes. There are the double peak tensile residual stresses along the weld, with maximum tensile stresses over 100 MPa in the largest sample, reducing to 30-40 MPa maximum for the smallest samples. Away from the weld the samples are subject to compression, approaching over 130 MPa compression around the notch tip in the largest sample (30 MPa in compression in the smallest sample). Residual stresses perpendicular to the weld are much smaller than parallel to it; as Fig 2b shows even this smaller stress component is negative around the notch tip. In the CT samples stresses perpendicular to the weld on the weld line were also compressive around the notch, becoming tensile at 10-20 mm from the notch root achieving a maximum of 40-50 MPa before becoming compressive again near to the far sample surface. On the weld line itself stresses parallel to the weld were roughly the same as the ones perpendicular, leading to an almost equi-biaxial field. Away from the weld line stresses parallel to it were tensile and much larger. Residual stress modification – GMT residual stress measurements Fig 3 shows the changes in residual stress which it is possible to produce by applying GMT during welding. The largest stresses in the original welded sample are in the region of 200 MPa, and are of the typical double peak form. Application of a stress 17% of the nominal material yield strength reduces this to 100 MPa, and application of 35% of the yield stress reduced the weld stress to zero. Application of 52% of yield takes weld residual stresses into compressionhowever this is at the expense of tensile residual stress development in the regions outside the weld which will have an adverse effect on fatigue crack growth rates in those regions.

Irving, Ma, Zhang, Servetti, Williams, Moore, dos Santos, Pacchione

394 150

Retreating side

Residual stress (MPa)

100

148x40-x 185x50-x

50

370x100-x

0 -50 -100 -150 -50

-30

-10

10

30

50

Distance from weld centre (mm)

(a) Effect of ESE(T) sample size on residual stresses parallel to weld; in each curve the notch tip is coincident with the first point on the left hand side; crack perpendicular to weld; 2195 alloy, 8 mm thick. 150

Residual stress (MPa)

Retreating side

370x100-x

100

370x100-y

50 0

Notch

-50 -100 -150 -50

-30

-10

10

30

50

Distance from weld centre (mm)

(b) Residual stresses parallel and perpendicular to weld line of 370x100 mm ESE(T) sample; crack perpendicular to weld; 2195 alloy 8 mm thick.

Residual stress perpendicular to weld MPa

Control of crack growth rates & crack trajectories in welded aircraft structures

395

80 60 40 20 0 -20 -40

w=84 mm w=120 mm w=240 mm

-60 -80 0

20

40

60

80

100

Distance from notch root mm

(c) Residual stresses perpendicular to weld line of 3 sizes of CT samples; crack parallel to weld; 2195 alloy, 8 mm thick. Fig. 2 Effects of sample size on residual stress profiles in ESE(T) & CT samples.

250 0% 200

17% 34%

Stress [MPa]

150

52% 100 50 0 -50 -100 -50

-40

-30

-20 -10 0 10 20 Lateral distance from weld line [mm]

30

40

50

Fig 3 Changes in residual stress produced by global mechanical tensioning during friction stir welding [21] 2199 alloy

396

Irving, Ma, Zhang, Servetti, Williams, Moore, dos Santos, Pacchione

EXPERIMENTAL RESULTS – FATIGUE CRACK GROWTH Effect of residual stresses on crack growth rates Tests under fatigue loads of R = 0.1 on the two largest ESE(T) samples could not initiate fatigue cracks from the notch and demonstrate the power of local compressive residual stresses in inhibiting fatigue failure. As shown in Fig 2 local fields around the notch root in all the ESE(T) samples are compressive. The smallest sample had the smallest compressive residual stresses and slow fatigue crack growth rates were produced at R=0.1 in this sample. Fatigue crack growth could be produced in all samples by testing at an R ratio of 0.6. The crack growth rates obtained are shown in Fig 4, plotted against ΔK and are compared with growth rates obtained in parent 2195 material at the same R value. The results show that data for the smallest sample and smallest residual stresses are almost coincident with parent material growth rates. Growth rates for both the larger samples are slower with the largest sample producing slowest growth rates. Both samples exhibit discontinuities in the curve at ΔK values of 10-15 MPa √m. Three sizes of C(T) samples were tested at R = 0.1, all with the crack growing in the weld centre line – the nugget region of the weld. Crack growth rates versus ΔK for all samples sizes together with parent material rates are shown in Fig 5. Fatigue crack growth rates for all welded samples were slower than in the parent material. Growth rates in the largest sample are slowest and growth rates in the smallest sample are most similar to parent plate growth rates because the effect of residual stress. However, the difference in growth rates decreases at larger ΔK values corresponding to longer crack lengths. Effect of residual stress and stress concentrations on crack trajectory Initial tests were performed in which the distance h between sample crack line and hole centre was 63 mm. Both with and without a weld on the crack line, there was no crack deviation. With the hole centre distance from the crack reduced to h = 51 mm, deviation of the crack into the hole was observed. Fig 6 shows the crack trajectories observed for the samples with h = 51 mm, with and without the weld. It can be seen that in the case of the samples with the weld, crack deviation occurred at a smaller crack length than in the samples without the weld.

PREDICTIONS Predictive models for influence of residual stress on fatigue crack growth rates Both the superposition and crack closure approaches have been used for predicting fatigue crack growth rates. The first method employs Eqn (1-2). Welding residual stress results in changes in the effective crack tip stress ratio Reff.

Control of crack growth rates & crack trajectories in welded aircraft structures

397

1.00E-04 370x100-R06 (Large) 185x50-R06(Middle)

1.00E-05

da/dN(m/cycle)

148x40-R06(Small) 2195-T8 Parent material-R06 1.00E-06

1.00E-07

1.00E-08

1.00E-09 1

10

100

deltK (MPam^0.5)

Fig. 4 da/dN vs. ΔK for three different sizes of welded ESE(T) samples at R = 0.6 compared with parent material.

1.00E-03

250x240

da/dN (m /cycle)

1.00E-04

125x120

1.00E-05

87.5x84

1.00E-06

2195-T8 Parent material

1.00E-07 1.00E-08 1.00E-09 1

10 ∆K (MPa√m)

Fig. 5 Crack growth rate in three sizes of C(T) samples; R=0.1

100

Irving, Ma, Zhang, Servetti, Williams, Moore, dos Santos, Pacchione

398

50 40

2.8-40-weld 2.8-40-no weld

30

Distance 20 from 10 g 8 The effect of weld on crack path with the hole Distance across sample mm weld line 0 mm 20 40 60 80 100 120 140 -10 0

160

-20 -30 -40 -50

Fig 6 The influence of weld residual stress field on crack trajectory on samples containing hole with the centre 51 mm from crack line; 2198 alloy, 3.2 mm thick.

The key task is to determine the Kres. The finite element method (FEM) is used in this work. It is more flexible and robust when weight function solutions cannot be found due to the complexity in either the geometry or loading conditions. It is also more convenient to model the residual stress redistributions due to crack extension. FEM in conjunction with the virtual crack closure technique (VCCT) is used to determine the Kres values by inputting measured residual stress fields into the FE model. Details can be found in [22]. Once the variation of Kres with crack length is known, a local effective stress ratio Reff can be found by Eqn (2). Crack growth rates in the weld residual stress field may then be derived using empirical representations of parent plate crack growth rates such as the Harter T-Method [23] or the Walker equation [24]. Both equations need at least two sets of crack growth rate data for different R ratios. In this work the material da/dN data at different R ratios were obtained by testing the parent material under three R ratios of 0.1, 0.35 and 0.6. The second method is based on the crack closure concept originally proposed by Elber [10]. Welding residual stresses will affect the crack closure behaviour and hence change the crack opening stresses. In the work presented in this paper, ΔKeff is determined by the analytical method proposed by Newman [11]. In Eqn (3-5), the nominal stress ratio R is replaced by the effective stress ratio Reff determined by Eqn (2). Crack growth rates are calculated by using the ΔKeff and the “master” curve of the material crack growth law, i.e. da/dN vs. ΔKeff. Prediction results – small ESE(T) sample at R = 0.1 and 0.6 Predicted crack growth rates using the superposition and crack closure methods are shown in Fig 7(a). For R = 0.6 the superposition approach predicts growth rates slightly faster than experiments, while the closure based approach predicts da/dN values slightly slower than experimental; the accuracy of both methods is better

Control of crack growth rates & crack trajectories in welded aircraft structures

399

than a factor of 2. At R = 0.1, superposition has an error of a factor of 3-4, while the closure approach, which uses a modified ΔKeff instead of the nominal ΔK used by the superposition model, is much better -within a factor of 2. Because of the compressive residual stresses at the notch root, there will be significant crack closure in the early crack growth for R = 0.1. Crack closure behaviour is currently being further investigated by finite element modelling using a very fine element size to model crack closure behaviour in the compressive residual stress field and coupling with the plasticity in the crack wake. The prediction for the R = 0.6 case gives reasonably good results because the higher applied mean stress can overcome the compressive residual stresses around the initial notch root so that the crack is fully open throughout the entire load cycle in spite of the compressive residual stresses. The two larger ESE(T) samples had much larger residual stresses, and their reduced growth rates shown in Fig. 4 is evidence of a reduced ΔKeff in spite of the large tensile mean at R = 0.6. This case has demonstrated that the superposition method is accurate if Reff >0.4 and there is no significant crack closure. Fig. 7(b) shows the calculated Kres and Reff for R = 0.1 and R = 0.6 loading on the ESE(T) small sample test. Compressive residual stress fields near the notch root have resulted in negative Kres values over the entire crack length span. For nominal R = 0.1 case, Reff values are also negative for crack length less than 17 mm. Consequently, crack growth rates under compressive residual stress fields are greatly reduced and also are less well predicted by simple superposition models. At nominal R values of 0.6, Reff values increase to between 0.2 and 0.6, there is little or no crack closure and simple superposition models work well as Fig 7(a) shows. Cracks parallel with the weld – CT samples The crack growth rate data shown in Fig. 5 show some differences between the three sample sizes, which once again reflect the samples sizes, with the largest sample having the slowest growth rates, and parent plate the fastest. The differences are at their greatest in the region around the notch root in the early stages of growth and least at long crack lengths when the relaxation of residual stress will be almost complete. This suggests that even the small compressive stresses of 30-50 MPa found perpendicular to the weld line and crack are sufficient to slow the crack significantly. Even in the absence of residual stresses at long crack lengths, there are still differences in growth rate between parent and weld, implying that the very fine grained and ductile nugget region has greater intrinsic resistance to fatigue crack growth than the parent plate. Further work is progress to quantify these hypotheses. Prediction results – crack deviation To predict crack deviation and subsequent crack trajectories, a 2D Finite element model of the experimentally tested samples was built then meshed with 4-node quadrilateral elements in ABAQUS. The maximum tangential stress criterion was used to predict the crack path. The SIGINI code was used to build a subroutine to input example residual stress profiles into the models. The effect of hole position

400

Irving, Ma, Zhang, Servetti, Williams, Moore, dos Santos, Pacchione

and residual stress on crack paths was investigated. As shown in Fig. 8, for hole distance h = 63 mm, a small crack deviation is predicted, but deviation into the hole is avoided. For h = 51 mm, the crack is predicted to deviate into the hole. This shows reasonable prediction accuracy in that experimentally it was observed that cracks did not deviate at all at 63 mm, whereas for 51 mm deviation was observed and predicted. The influence of the weld residual stress field in promoting earlier crack deviation than for unwelded samples is also predicted by the model.

(a) Prediction by superposition and by crack closure; R=0.1 and R = 0.6

(b) Calculated Kres and Reff; comparison with nominal R for tests at R=0.1, R= 0.6. Fig. 7 Predicted fatigue crack growth rates and R ratios for 148 x 40 ESE(T) sample compared with experimental results.

Control of crack growth rates & crack trajectories in welded aircraft structures

401

50 H51R30-with residual stress

40

H51R30-without residual stress

30 H63R30-with residual stress

Distance 20 from the 10 weld line 0 mm mm

H63R30-without residual stress

0

10

20

30

40

50

60

70

80

90

100 110 120 130 140 150 160

-10 -20

h=63 mm

h = 51 mm

-30 -40 -50 mm

Fig. 8 Predicted crack paths using FE analysis and maximum tangential stress criterion; h=51 and h=63 mm with and without residual stress. The numerical calculation results so far have shown that the distance between hole and weld line influences crack path significantly, and determines whether a crack deviates. Weld residual stresses further modify crack trajectory, by reducing the crack length at initial deviation.

DISCUSSION Control of residual stresses The technique of global mechanical tensioning described in the present paper has been applied previously to reduce residual stress fields in various guises for a number of years [e.g. 6]. It clearly is very effective at modifying residual stress fields around welds, changing a peak tensile stress field of 200 MPa to a compressive one of 70 MPa, albeit with an associated tensile field remote from the weld. GMT is only one of a number of techniques for residual stress modification which are contemplated by weld process engineers. These include local mechanical tensioning, thermal mechanical tensioning and thermal tensioning using cooling. Some of these have been proven to reduce weld line stresses- the capability of others has yet to be investigated. Data in the present paper on the influence of residual stress fields arising from post weld machining operations on fatigue crack

402

Irving, Ma, Zhang, Servetti, Williams, Moore, dos Santos, Pacchione

initiation and propagation demonstrate that the generation of compressive fields at potential crack initiation sites is a powerful means of avoiding crack development; hence maximum benefit of post weld stress field modification will be gained via processes which can place compressive residual stresses at the best position. Prediction of crack growth rates through residual stress fields Work reported in this paper and others previously [25] shows that the approach of calculation of Kres followed by Reff and then da/dN from the relevant Walker or Harter T-Method for the influence of R on crack growth rates works well for tensile residual stress fields. Residual stress fields which are in part significantly compressive are still problematic. The simplest approach to this situation is to consider only parts of the load range, which are in tension, rejecting the compression parts. Although this approach goes some way to predicting the observed fatigue crack growth rates agreement is not ideal. The relative lack of success of the tension only approach may be to do with the difference between crack opening & closing behaviour in an inhomogeneous residual stress field reflecting the entire field in the component; and crack opening in a homogeneous field, equivalent to an applied mean stress. Better models of opening and closing behaviour in inhomogeneous compression going fields may improve predictions. A further complicating factor influencing prediction accuracy is the role of weld microstructure separately from residual stress influences. While cracks are propagating perpendicular to the weld line, the possible effect on crack growth rates will be transitory as shown in Fig 4. However, the crack growth tests on C(T) samples, where the crack grows parallel to the weld line in small tensile residual stress fields, show that significant crack growth rate changes occur, and use of parent plate data in the Walker equation to calculate lives will be inappropriate. Calculation of crack trajectories The results of the limited tests and finite element predictions using the maximum tangential stress criterion show that both the influence of mode II stress intensity component and the influence of residual stress fields may be satisfactorily predicted for the geometries and stress fields considered here. Much further work is necessary to explore a wider range of geometries and models before the approach can be considered sufficiently mature to be used with confidence. However the success of the present limited approach is very satisfactory. Implications for design of damage tolerant and fail safe aircraft structures Welding fabrication of the metallic parts of aircraft is considered a benefit to economic manufacturing and a deficit in terms of prediction of damage tolerant lives coupled with increased challenges for the fail safety of the welded integral structure. The developments described in this paper in control of residual stresses, prediction of fatigue crack growth rates, and prediction of crack trajectories offer a

Control of crack growth rates & crack trajectories in welded aircraft structures

403

different perspective. With suitable stress control treatment the possibility exists of designing structures in which crack initiation and propagation performance is optimised and is predictable, this including specification of locations for crack paths to deviate and promote crack arrest. Such control over the initiation and growth behaviour of cracks would permit a step change in damage tolerance and fail safety of metallic aircraft structures. Techniques to control weld residual stress fields are key, and while the results shown in this paper are promising, there is clearly much to be done to further develop approaches for residual stress control. The approaches for crack growth rate predictions look now to be well established but significant research is still required on the practically important case of cracks approaching local compressive stress regions. The success of the simple tangential stress approach for predicting crack trajectories is gratifying, but considerable further research is required.

CONCLUSIONS (1) Techniques such as Global Mechanical Tensioning can modify weld residual stress fields, converting tensile fields to compressive ones and raising tensile balancing stresses outside the weld. Machining operations can modify weld stresses and introduce intense compressive residual stress fields sufficient to completely inhibit fatigue crack growth at low applied mean stresses. (2) Fatigue crack growth rates through friction stir weld tensile residual stress fields may be accurately predicted by using the Kres approach coupled with empirical models for representing the R ratio effects. Crack growth rates under residual stress fields causing negative Kres and Reff values are less well predicted by simple models. (3) Crack trajectories under mixed mode stresses and in weld residual stress fields are accurately predicted by a numerical model using the maximum tangential stress criterion. (4) The techniques describes under 1-3 above together offer a new opportunity to tailor damage tolerance and fail safety capability via modification of weld residual stresses using post weld processing.

404

Irving, Ma, Zhang, Servetti, Williams, Moore, dos Santos, Pacchione

ACKNOWLEDGEMENTS Grateful thanks are given to R Maziarz & M Poad of Airbus for much technical advice and discussions & for manufacture of 2195 and 2199 alloy weld samples used in this work. Part of the work reported in this paper was funded by the EU under contract No AST5-CT-030825 COINS “Cost effective integral metallic structures”. The Global Mechanical Tensioning work was funded by UK DTI under the “SEALS” project.

REFERENCES [1] Maddox, SJ. (2003) International Journal of Fatigue 25, 1359-1378. [2] Edwards, L, Fitzpatrick ME, Irving PE, Sinclair I, Zhang, X, Yapp, D. (2006) Journal of ASTM International (JAI), Vol. 3. available online at www.astm.org. [3] Pouget, G, Reynolds, AP (2008), International Journal of Fatigue 30, 463–472. [4] Gray, TGF, Spence J, North TH (1975). In: Rational Welding Design. Newnes – Butterworths, London. [5] Cheng X, Fisher JW, Prask HJ, Gna¨upel-Herold T, Yen BT, Roy S. (2003). Int J Fatigue, 25: 1259-1269. [6] Bussu, G, Irving, PE. (2003), Inl J of Fat 25, 77-88. [7] Altenkirch, J, Steuwer, A, Peel, MJ, Richards, DG, Withers, PJ, (2008) Mat Sci & Eng A 488 16-24. [8] Glinka, G. (1979). In: Fracture Mechanics, ASTM STP 677, 198-214. [9] Parker, AP (1982). In: Residual Stress Effects in Fatigue, ASTM STP 776, 1331. [10] Elber, W. (1971). In: Damage Tolerance in Aircraft Structures, ASTM STP 486, 230-242. [11] Newman, JC Jr. (1984). Int J of Fract. 24, R131-R135. [12] Rubinstein, AA. (1991) Int J of Fract 47, 191-305. [13] Fersini, A. Pirondi. (2007) Engng Fracture Mechanics 74, 468-480. [14] Tilbrook, M, Hoffman, M. (2007) Engng Fracture Mechanics 74, 1026-1040. [15] Qian J, Fatemi A (1996). Engng Fracture Mechanics, 55, 969-990. [16] Altenkirch J, Steuwer, MJ, Peel, MJ, Withers, PJ, Williams, SW, (2008) Metall. Trans A 39 3246-3259. [17] Moore, G, Altenkirch, J. (2008) DTI SEALS Project deliverable D2, Bombardier Aerospace report RAM 1007. [18] www.alcanaerospace.com accessed 020409 [19] Cavaliere P, De Santis A, Panella F, Squillace A, (2008), Eng Fail Anal; in press. [20] Hirsch J, Gottstein G, Skrotzki B, “Aluminium alloys their physical and mechanical properties” pub Technology & Engineering (2008) [21] Price, DA, Williams, SW, Westcott, A, Harrison, CJC, Rezai, A, Steuwer, A, Peel, M, Staron, M, Kocak, M, (2007) Sci & Tech of Welding and Joining, 12 620-633.

Control of crack growth rates & crack trajectories in welded aircraft structures

405

[22] Servetti, G, Zhang, X. (2009), Eng Fract Mech. In press. [23] Walker, K. (1970) In: Effects of environment and complex load history on fatigue life, ASTM STP 462, vol. 462: 1-14. [24] Harter, JA. (2006) In: AFGROW users guide and technical manual, AFRLVA-WP-TR-2006-XXXX, Version 4.0011.14. Section 3.2.2.2, pp. 34-36. http://www.siresearch.info/projects/afgrow/downloads/afgrow/ddownload.php. [25] Lilljedahl CDM, Brouard J, Zanellato O, Lin J, Tan ML, Ganguly S, Irving PE, Fitzpatrick ME, Zhang X, Edwards L. (2009). Int J of Fatigue, 31, 1081–1088.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

DURABILITY AND DAMAGE TOLERANCE OF BONDED REPAIRS TO METALLIC FUSELAGE STRUCTURE John G. Bakuckas, Jr.,1 Keith McIver2 , and Ching Hsu2 1

FAA William J. Hughes Technical Center, Atlantic City International Airport, NJ 08405, USA 2 Boeing Research & Technology, 5301 Bolsa Ave., Huntington Beach, CA 92647, USA

Abstract: The durability and damage tolerance of boron/epoxy and aluminum bonded repairs to a metallic fuselage structure was characterized under simulated flight load conditions up to one typical design service goal of 60,000 cycles. Bonded repairs to two damage scenarios were considered, namely, a mid-bay through-the-thickness crack and a lap joint scribe. Several methods were used to monitor and record damage formation and growth of cracks and disbonds, including visual inspections, eddy current, and flash thermography. There were no indications of damage development in the form of crack growth or disbonding for the lap joint scribe line repair patches. For the mid-bay repair patches, no disbonding occurred; however, slow crack growth was measured. Test and analysis results revealed that the installation of the mid-bay repairs resulted in eccentric loading and inward deformation of the mid-bay region, resulting in higher tensile strains on the inner skin surface and compressive strain on the outer patch surface. All repairs were effective in containing damage during the post-fatigue residual strength test to the damage tolerance requirements in Title 14 Code of Federal Regulations 25.571.

INTRODUCTION Adhesive bonding technology, using composite and metallic patches, offers an efficient and cost-effective approach to airplane structural repairs. Compared to M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 407–425. © US Government 2009. Created within the capacity of an US governmental employment and therefore public domain. Published by Springer Science+Business Media B.V. Dordrecht.

408

John Bakuckas, Keith McIver and Ching Hsu

conventional, mechanically fastened metallic repairs, bonded repairs have no stress concentrations due to holes, are less damaging to the parent material since no drilling or machining are required, and are more aerodynamically and structurally efficient. The application of bonded repairs to metallic aircraft structure has been extensively studied where durability and damage tolerance aspects have been demonstrated [1-4]. However, several technical challenges need to be addressed before bonded repair technology will be generally accepted and implemented in both military and commercial primary structural applications. Currently, credit is typically not provided in certification programs of bonded repairs for slowing crack growth or restoring residual strength. Of primary concern is the ability to predict the fatigue behavior and ensuring the durability of the repair bond. In an effort to gain a better understanding of the durability and damage tolerance aspects of bonded repair technology, the Federal Aviation Administration (FAA) and The Boeing Company have partnered through a cooperative research and development agreement. A phased approach is being undertaken with the initial study focusing on the test and analysis of bonded repairs on a B727 metallic fuselage structure using the FAA Full-Scale Aircraft Structural Test Evaluation and Research (FASTER) facility. The objectives are to characterize the long-term durability of bonded repairs under simulated flight load conditions up to one typical design service goal (60,000 cycles) and then determine if the repair patches meet damage tolerance requirements in a residual strength test. Bonded repairs to two damage scenarios were considered, namely, a mid-bay through-the-thickness crack (fatigue presharpened) and a lap joint scribe. Both boron/epoxy and aluminum patches were used to repair this damage. A photogrammetry method was used to obtain full-field displacement and strain measurements at the patch regions. The acoustic emission (AE) method was used to monitor for damage growth in real time and serve as an early warning for imminent failure. Several nondestructive inspection (NDI) methods were used, including flash thermography and computer-aided tap techniques to scan for patch disbonds and eddy current to monitor crack growth. An overview of the results obtained from this study is provided in terms of data to validate analytical predictions of bonded repair patch durability and residual strength, and an assessment of several NDI techniques in detecting disbonds and fatigue cracking.

EXPERIMENTAL PROCEDURE Testing was conducted using the FASTER facility as summarized in this section. Test Panel and Repair Configurations The fuselage panel was removed from a retired passenger service B727-232 aircraft, serial number 20751 and line number 1000. The panel dimensions were 3175 mm by 1854 mm with radius of 1880 mm, Figure 1. The substructure

Durability and damage tolerance of bonded repairs to metallic fuselage structure

409

included six stringers in the longitudinal direction with 241-mm spacing and six frames in the hoop direction with 508-mm spacing. A longitudinal lap joint was located along stringer S-4L. The panel thickness varied from 1.02 mm between frame station (FS) 720C and 720D to 2.04 mm between FS 720F and 740. The panel was reinforced with doublers on the four edges. Holes of 12.7-mm diameter were drilled through the doublers to attach the panel to the load actuator arms of the loading fixture. Strain gages were installed on the skin, stringers, and frames. Bonded repairs to two damage scenarios were considered, namely, a mid-bay through-the-thickness crack (fatigue presharpened) and a lap joint scribe, as shown in Figure 1. Both boron/epoxy and aluminum patches were used to repair this damage. For the composite repairs, the patch material used was boron/epoxy 5521 prepreg tape. The precured boron/epoxy patches had a taper ratio of 20:1 for all eight sides. For the aluminum patches and fillers, 2024-T3 bare aluminum sheet was used. All patches had a 45° corner truncation angle. BMS5-101 Type II Grade 10 film adhesive was used as the bonding adhesive.

Figure 1. Test panel dimensions, initial damage scenarios, repair designations, and location of rosette strain gages. The aluminum skin panel surface preparation included paint removal, solvent cleaning, leak testing, sol-gel application, and primer application. The patches were pre-cured using the calibration fuselage section as the tool surface in order to match the curvature of the test part. The patches were then bonded to the test panel by portable composite repair equipment, vacuum bag, and heat blankets at a cure temperture of 120ºC for 60 minutes, per the manufacturer’s recommendation.

John Bakuckas, Keith McIver and Ching Hsu

410

Details of the repair patches are shown in Figures 2 and 3 for the mid-bay cracks and the lap joint scribes, respectively. For the mid-bay patch, C1BE, the fibers were oriented in the hoop direction to provide a stiff repair. For repair C2A, aluminum was used to match the properties to the skin material. The effect of stiffness was studied using the lap joint scribe repairs. As described in Figure 3, fibers were oriented in the axial direction for repair S1BE to yield a low stiffness, compliant patch in the hoop direction. For repair S2BE, the fibers were directed in the hoop direction to provide a high stiffness patch. For repair S3A, aluminum was used to match the properties of the skin material.

a. Mid-Bay Crack 1 Boron/Epoxy Patch (C1BE)

45°

Ply Stack Sequence (W x H, θ) Hoop

0° -20° 20°

H

76.2 mm θ

Axial

P1: P2: P3: P4: P5: P6:

177.8 x 76.2 mm, +20° 174.2 x 72.6 mm, -20° 170.7 x 68.6 mm, 0° 167.1 x 65.5 mm, 0° 163.6 x 61.9 mm, -20° 160.0 x 58.4 mm, 20°

W

45°

b. Mid-Bay Crack 2 Aluminum Patch (C2A) Sheet Sequence (W x H x T) S1: 152.4 x 76.2 x 0.5 mm S2: 137.2 x 60.9 x 0.5 mm S3: 121.9 x 45.7 x 0.5 mm

H 76.2 mm

Aluminum 2024-T3

W

Figure 2. Mid-bay crack repair patch configurations: a. crack 1 boron/epoxy patch (C1BE) and b. crack 2 aluminum repair patch (C2A).

Durability and damage tolerance of bonded repairs to metallic fuselage structure

411

a. Lap-Joint Scribe Patches b. Aluminum Fillers Hoop H Axial W

152.4 mm

Filler Sheet Sequence (W x H x T) F1: 304.8 x 86.4 x 0.8 mm F2: 292.1 x 73.7 x 0.8 mm Aluminum 2024-T3

45°

c. Scribe 1 Boron/Epoxy (S1BE)

-20°

20° -20°

θ

Ply Stack Sequence (W x H, θ) P1: P2: P3: P4:

d. Scribe 2 Boron/Epoxy (S2BE)

279.4 x 254.0 mm, +20° 275.6 x 250.2 mm, -20° 271.8 x 246.4 mm, -20° 267.9 x 242.6 mm, +20°

e. Scribe 3 Aluminum (S3A)

20° θ

Ply Stack Sequence (W x H, θ) P1: 279.4 x 254.0 mm, +20° P2: 275.6 x 250.2 mm, -20° P3: 271.8 x 246.4 mm, -20° P4: 267.9 x 242.6 mm, +20°

Sheet Sequence (W x H x T) S1: 279.4 x 254.0 x 0.5 mm S2: 275.6 x 250.2 x 0.5 mm Aluminum 2024-T3

Figure 3. Lap-Joint Scribe Repairs: a. configuration, b. aluminum fillers, c. boron/epoxy patch with low stiffness (S1BE), d. boron/epoxy patch with high stiffness (S2BE), and e. aluminum patch (S3A). Applied Loads The panel was subjected to the applied loadings listed in Table I for the baseline strain survey, fatigue precracking, fatigue, and residual strength tests. The loads used for the fatigue test simulated the in-service loading conditions, including cabin pressurization and fuselage vertical bending, and was represented by an equivalent constant-amplitude spectrum. For the residual strength test, the loads corresponded to the damage tolerance requirements of the Title 14 Code of Federal Regulations (CFR) 25.571: Maximum value of the normal operating pressure including the expected external aerodynamic pressure during 1 g level flight multiplied by 1.15. The magnitude of the loads used in the strain survey and precracking was 75% of the loads used for the fatigue test. For the baseline strain survey tests, quasi-static loadings were applied in ten increments up to the maximum loads listed in Table I. For the precracking and fatigue test, constantamplitude loading was applied at a frequency of 0.33 Hz with an R ratio (minimum to maximum load) of 0.1.

John Bakuckas, Keith McIver and Ching Hsu

412

Test Phase

Load Type

Strain Survey Fatigue PreCrack Fatigue Residual Strength

Pressure (KPa)

Maximum Load Hoop Axial (KN/m) (KN/m)

Frame (KN)

QuasiStatic

46.03

75.92

83.49

5.04

Cyclic (R=0.1)

46.03

75.92

83.49

5.04

61.37

101.22

111.32

6.72

65.89

108.68

111.32

7.22

Cyclic (R=0.1) QuasiStatic

Table I. Maximum Applied Loads Inspection and Measurement Procedures Several NDI methods were used to monitor and record damage formation and growth of cracks and disbonds as indicated in Table II. Crack growth was monitored continuously using high-magnification visual inspections, particularly for mid-bay patches C1BE and C2A. Detailed inspections were made every 5000 cycles: eddy current was used to inspect for crack growth and flash thermography and computer-aided tap tester were used to inspect for disbonds. In addition, the AE method was used to continuously monitor activity at selected intervals. The full-field strain and displacements of the patches and the area surrounding structure were recorded using, non-contact, 3-D deformation measuring system, ARAMISTM. The system uses two 4-megapixel cameras, capable of measuring full-field strains within 50με. Baseline measurements were made at the beginning of the test and then at 5000 cycle intervals. Method

Inspection

Interval

High-magnification visual from interior surface Eddy current from internal surface Eddy current from external surface Flash thermography from exterior surface Tap tester from exterior surface AE from external surface

Crack growth under mid-bay patches C1BE and C2A Crack growth under mid-bay patches C1BE and C2A Crack growth under mid-bay patch C1BE Disbonding of all patches

Continuous

5000 cycles

Disbonding of all patches

5000 cycles

Monitor damage from AE activity in the vicinity of patch C1BE

Continuous

Table II. Inspection Method

5000 cycles 5000 cycles

Durability and damage tolerance of bonded repairs to metallic fuselage structure

413

ANALYTICAL PROCEDURE Geometric nonlinear finite element analyses were conducted to predict the strain distributions and to compute the stress-intensity factor (SIF) solutions using the commercial finite element code ABAQUS. The panel was modeled using shell elements with each node having 6 degrees of freedom. Figure 4 shows the global view of a typical finite element model of the panel. Four-noded shell elements were used throughout to model the skin, frames, stringers, and clips. In the immediate vicinity of the crack tips, a highly refined mesh was used where the element dimension was approximately 1 mm. Beam elements were used to model the rivets. The boron/epoxy and aluminum patches were modeled using fournoded shell elements (not shown). Three-dimensional solid elements were used to connect the shell elements using the average thickness of the bonded parts. The model contained the major geometric details and dimensions of the panels, including the cross-section properties of the substructure (frames, stringers, and clips) and the load attachment doublers. Typically, the panel model had 250,000 degrees of freedom. The load conditions specified in Table I was simulated in the analysis. For the hoop, frame, and longitudinal loads, nodal point forces were applied at the load application points, as shown by the arrows in Figure 4. Internal pressure was applied to the inner surface of the skin.

Full Assembly

Skin

Axial Load

Crack Area

Hoop Load

Substructure

Frame Load

Figure 4. Finite element model showing full assembly, substructure, and boundary conditions.

414

John Bakuckas, Keith McIver and Ching Hsu

The Modified Crack Closure Integral (MCCI) approach [5] was used to obtain the four components of SIF. The four components of SIF are the Mode I SIF caused by tensile load, K1, the Mode II SIF cause by in-plane shear load, K2, the SIF due to symmetric bending loads, k1, and the SIF due to out-of-plane shear and twist loads, k2, as shown in Figure 5.

6

u~ top = Symmetric Membrane Loading, KI

∑u

top i

i =1

Anti-Symmetric Membrane Loading, KII

Δa

Δa

a

a 6

~ bot

u

=

∑

uibot

~ F close =

6

∑F

i

close

i =1

i =1

2 Symmetric Bend Loading, k1

5

Anti-Symmetric Bending and Shear Loading, k2

3

4 6

1 D.O.F

Figure 5. Definition of four stress-intensity factors for thin cracked shells and crack tip element and nodes for computing the rate of work done to close a crack using the MCCI method.

RESULTS AND DISCUSSION Tests and analyses were performed to study the durability and damage tolerance of bonded boron/epoxy and aluminium patches to metallic fuselage structure. Representative results are outlined in the subsequent sections. Initial Damage Scenarios and Fatigue Precracking Defects were first introduced in the panel, which consisted of mid-bay through notches at two locations and lap joint scribes at three locations, Figure 1. The midbay through notch lengths were approximately 76.2 mm, designated Notch 1 and 2. The lap joint scribes were approximately 152.4 mm long and 0.3 mm deep, designated Scribe 1-3. The mid-bay notches were fatigue presharpened using cyclic loads that were 75% of applied fatigue loads, as shown in Figure 6 for Notch 1. Note the symmetric and co-linear crack extension from the notch tips. Fatigue precracking was applied until a natural crack extended from the notches to a length of 1 to 4 mm.

Durability and damage tolerance of bonded repairs to metallic fuselage structure

Pre-Cracking Loads: Pressure = 46.03 KPa Hoop = 75.92 KN/m Axial = 83.49 KM/m Frame = 5.05 KN

415

Notch 1 76.2 mm Notch 1

Notch 2

Hoop Axial

AFT

FWD

1.04 mm 0.26 mm 0.93 mm

2.08 mm

1500 cycles

2500 cycles

2.09 mm

2.91 mm 2.82 mm

3250 cycles

Figure 6. Photographs of crack growth from Notch 1 during fatigue precracking. The measured and predicted crack extension for both mid-bay notches is shown in Figure 7. The mode I SIF was used in the analysis since it was the dominate SIF. After fatigue precracking the mid-bay notches, boron/epoxy patch, C1EP, was installed over Notch 1 and aluminium patch, C2A, was installed over Notch 2. In addition, boron/epoxy patches, S1BE and S2BE, were installed over Scribe 1 and 2, respectively, and aluminium patch, S3A, over Scribe 3. Details of the patch configurations are provided in Figures 1-3. Strain Survey Strain surveys were performed and several NDI methods used to baseline the patches. The strain distribution was measured and predicted under quasi-static load conditions. Three tests were conducted where the load was applied in ten increments up to the maximum values listed in Table I. In general, the strain distributions measured from all three tests were nearly identical. Figure 8 shows representative results for three quasi-static tests of the strain in the hoop direction, which was measured using a rosette strain gage (S2-H) located in a skin mid-bay between FS 720C and FS 720D. As shown in figure, the strains were highly repeatable for all three tests. In addition, good agreement was obtained between the test results and predictions using finite element analysis (FEA), which is indicated by the solid line.

John Bakuckas, Keith McIver and Ching Hsu

416

Notch 1

Crack Extension, Δa (mm)

4

3

Aft Fwd Analysis

Notch 2

Δa

Aft Fwd Analysis

Pre-Cracking Loads: Pressure = 46.03 KPa Hoop = 75.92 KN/m Axial = 83.49 KM/m Frame = 5.05 KN

2

1

0 1000

1500

2000

2500

3000

3500

Cycles Figure 7. Crack growth from Notch 1 and 2 during fatigue precracking.

Figure 8. Test and analysis results from strain survey at a skin mid-bay gage. The distribution of hoop strains in the mid-bay regions of the panel is shown in Figure 9. Note that the strains are not constant because the skin thickness varied

Durability and damage tolerance of bonded repairs to metallic fuselage structure

417

from 1 to 2 mm. Good agreement was obtained between the test results and predictions using FEA.

Figure 9. Measured and predicted strain distribution. The full-field surfaces hoop strains in the vicinity of all five patches, which was measured using the ARAMIS system, is shown in Figure 10. In the figure, the patch boundary and initial defect are indicated. The hashed regions are areas where data could not be processed because of interference from strain gage wires. In general, the strain increased in the panel skin along the edges of the repair patches (indicted by the fringe patterns having lighter shades of gray) and decreased in the repair patch region. Comparing the lap joint scribe repairs, the low stiffness repair, S1BE, carried less strain and attracted less load in the skin region along the repair boundary in the hoop direction. For the high stiffness repair, S2BE, the strains were lowest in the patch and highest in the skin along the boundary. The aluminum patch, S3A, had stiffness comparable to the skin.

For the mid-bay patches, the higher stiffness patch, C1BE attracted more load to the skin (indicated by the white fringe patterns) along the patch boundary compared to C2A, Figure 10. Note that both mid-bay patches were both under compressive loading in the center along the notch region, as indicated by the darkgray fringe patterns. FEA revealed that the mid-bay patch installations resulted in eccentric loading and inward deformation of the mid-bay region, Figure 11a. As a

418

John Bakuckas, Keith McIver and Ching Hsu

result, there are higher tensile stresses on the inner patch surface and compressive stresses on the outer patch surface, Figure 11b.

Figure 10. Full-field hoop strains in the vicinity of repair patches.

Figure 11. FEA results showing effect of eccentric loading resulting from patch installations at mid-bay regions: a. Deformed mesh of skin displayed without the patch and b. strains in the patch. The measured and predicted hoop strains from several strain gages used to monitor patch C1BE is shown in Figure 12. As shown, gage S9 is located in the center of

Durability and damage tolerance of bonded repairs to metallic fuselage structure

419

the patch on the outer surface and is under compression. Gages S8 and S10 are located on the outer surface of the skin 19 mm from the patch boundary. Gages IS26 and IS27 were at the same location on the inner skin surface. Comparing the back-to-back gages, there is a large amount of bending along the patch boundary with the outer surface in tension and inner surface in compression.

Figure 12. Strain measurements showing bending effect at mid-bay patch C1BE.

Fatigue Tests After the baseline strain survey test, the panel was subjected to fatigue loads that simulated in-service loading conditions, including cabin pressurization and fuselage vertical bending at the levels shown in Table I. Strains and damage growth was monitored and recorded during the fatigue test which lasted 60,000 cycles. Representative results are outlined in the subsequent paragraphs. In general, the global strains remained relatively constant throughout the fatigue test. Results are summarized in Figures 13 and 14 for skin mid-bay regions and repair patch C1BE.

420

John Bakuckas, Keith McIver and Ching Hsu

Figure 13. Strains at skin mid-bay locations during fatigue test.

Figure 14. Strains at repair patch C1BE during fatigue test.

Durability and damage tolerance of bonded repairs to metallic fuselage structure

421

There were no indications of damage development in the form of crack growth or disbonding for the lap joint scribe line repair patches. Strain gages mounted on the inner surface of the skin to monitor crack growth from the scribe lines remained relatively constant throughout the fatigue test. For the mid-bay patches, there was no gross or measurable disbonding; however, a slow stable crack growth was observed, particularly the crack under patch C1BE, as shown in Figure 15. The images in this figure were taken using a flash thermograghy system.

C1BE Flashed Thermography Images: • •

Temp. vs. Time: 2nd Derivative Time Slice = 3 Seconds

10,000 Cycles

Fatigue Loads: • • • •

Axial Hoop

35,000 Cycles

35 K

20,000 Cycles

Pressure = 61.37 KPa Hoop = 101.22 KN/m Axial = 111.32 KM/m Frame = 6.72 KN

a

Mid-Bay Notch

Figure 15. Flash thermography inspections of patch C1BE during fatigue test. The observed crack growth was measured during the fatigue test using visual and internal eddy-current inspections for patches C1BE and C2A, as shown in Figures 16 and 17, respectively. For the composite mid-bay patch, C1BE, the crack growth was symmetric and co-linear, Figure 16. The crack growth rate was nearly constant. The crack lengths measured using internal eddy current were slightly longer than those measured visually, indicating that crack tunneling occurred. The visual measurements were made during the fatigue test with a high-magnification camera. The crack tip could be located only when the crack opening was maximized at the upper load levels of the fatigue cycle. At the lower levels of the fatigue cycle, it was difficult to distinguish the crack tip. The internal eddy-current inspections were made under no load conditions. For the crack under the aluminum patch C2A, the extension was much less than under patch C1BE, Figure 17. Visual measurements could not be reliably made

John Bakuckas, Keith McIver and Ching Hsu

422

since the growth was so small. Crack length measurements were conducted solely with internal eddy-current inspections.

Half Crack Length,a (mm)

65

Aft (Visual) Fwd (Visual) Aft (Internal Eddy Current) Fwd (Internal Eddy Current)

60

55

C1BE

50

Fatigue Loads: Pressure = 61.37 KPa Hoop = 101.22 KN/m Axial = 111.32 KM/m Frame = 6.72 KN

45

40 0

10000

20000

30000

40000

50000

60000

Cycles Figure 16. Crack length measurements from visual and internal eddy-current inspections of patch C1BE.

Crack Extension, Δa (mm)

6 Aft (Internal Eddy Current) Fwd (Internal Eddy Current) 5

4

3

C2A Fatigue Loads: Pressure = 61.37 KPa Hoop = 101.22 KN/m Axial = 111.32 KM/m Frame = 6.72 KN

2

Δa 1

0 0

10000

20000

30000

40000

50000

60000

Cycles Figure 17. Crack length measurements using eddy-current inspections of patch C2A.

Durability and damage tolerance of bonded repairs to metallic fuselage structure

423

Figure 18 shows a comparison of three methods of measuring crack growth under composite patch C1BE: visual, internal eddy current, and external eddy current. As shown, the measurements made by the three methods were in good agreement.

Half Crack Length, aAVG (mm)

65 Visual Internal Eddy Current External Eddy Current 60

55

C1BE

50

Fatigue Loads: Pressure = 61.37 KPa Hoop = 101.22 KN/m Axial = 111.32 KM/m Frame = 6.72 KN

45

40 0

10000

20000

30000

40000

50000

60000

Cycles Figure 18. Crack length measurements from visual, internal eddy current, and external eddy-current inspections of patch C1BE.

Residual Strength Test After the fatigue test, the panel was then subjected to quasi-static loading to a level simulating the damage tolerance requirements defined in 14 CFR 25.571. The load levels are listed in Table I for the residual strength test. All five repair patches were effective in preventing failure of the damaged panel. No disbond or crack growth was observed during inspections made after the residual strength test. In addition, there was no evidence of load redistribution: strain survey results after residual strength test were similar to the baseline strain survey results.

SUMMARY Tests and analyses were performed to study the durability and damage tolerance of bonded boron/epoxy and aluminium patches to metallic fuselage structure. Defects were first introduced in the panel, which consisted of mid-bay through notches and lap joint scribes. These defects were fatigue presharpened using cyclic loads that were 75% of applied fatigue loads. Boron/epoxy and aluminium patches having varying stiffness were then installed. Afterwards, strain surveys were performed and several nondestructive inspection methods used to baseline the repair patches.

424

John Bakuckas, Keith McIver and Ching Hsu

The panel was then subjected to fatigue loads that simulated in-service loading conditions, including cabin pressurization and fuselage vertical bending. Damage formation and growth was monitored and recorded during the fatigue test, which lasted 60,000 cycles. Finally, a residual strength test was performed to the damage tolerance requirements in 14 CFR 25.571. Results from the baseline strain survey indicted that the stiffer boron/epoxy patches attracted more load in the skin along the repair boundary. A detailed finite element analysis revealed that the mid-bay patch installations resulted in eccentric loading and inward deformation of the mid-bay region. There was a large amount of bending in the skin along the patch boundary, with the outer skin surface in tension and the inner skin in compression. In the middle of the patch along the notch centerline, the strains were highly compressive on the patch outer surface and in tension on the inner skin surface. During fatigue loading to 60,000 cycles, the global strains remained relatively constant. From inspections made using flash thermography and tap tester, there were no indications of disbond occurring in any patch. For the lap joint patches, strains remained constant during fatigue from inner skin gages used to monitor the crack growth from the scribe lines. For the mid-bay repairs, slow and stable growth was measured. There was good agreement in the crack length measurements made from the visual and eddy-current inspections. Results from the residual strength test demonstrated that the repair patches were effective in containing damage. There was no observed disbond or crack growth. In addition, there was no evidence of load redistribution: results from a strain survey conducted after the residual strength test were similar to the results of the baseline strain survey.

AKNOWLEDGEMENTS The authors wish to acknowledge Reewanshu Chadha, Yongzhe Tian, and Jeff Panco for their diligent efforts running the tests using the FASTER facility at the FAA William J. Hughes Technical Center and Bud Westerman, Cong Duong, and Carly Schlottman of The Boeing Company for their support and contributions.

REFERENCES [1] Advances in the Bonded Composite Repair of Metallic Aircraft Structure, Volume 1 and 2. Baker, A. A., Rose, L. R. F., and Jones, R., (ed.) Elseviar 2002.

Durability and damage tolerance of bonded repairs to metallic fuselage structure

425

[2] Roach, D. and Rackow, K., Development and Validation of Bonded Composite Double Repairs for Commercial Aircraft, Sandia Report SAND20074088, July 2007. [3] Kressel, I., Botsev, Y., Green, A. K., Ben-Simon, U., Ghiai, G., Tur, M., Banks-Sills, L., and Gali, S., Damage Tolerance of Bonded Composite Repairs: Analysis and Testing, Proceedings of the 23rd ICAF Symposium, vol. II, pp. 585588, Dalle Donne. C. (ed.), DGLR-Bericht 2005-03, ISBN 3-932182-42-1, Hamburg, 2005. [4] Baker, A., Issues in the Certification of Bonded Composite Patch Repairs for Crackeed Mettalic Aircraft Structures, Proceedings of the 20th ICAF Symposium, vol. I, pp. 299-320, Rudd, J. L., and Bader, R. M. (eds.), Electric Print Imaging Corporation, Bellevue, 1999. [5] Viz, M. J., Potyondy, D. O., and Zehnder, A. T., Computation of Membrane and Bending Stress-Intensity factors for Thin, Cracked Plates,” International Journal of Fracture, Vol. 72, pp. 21-38, 1995.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

INVESTIGATION ON THE DESIGN OF BONDED STRUCTURES FOR INCREASED DAMAGE TOLERANCE Ivan Meneghin1, Marco Pacchione2, Pascal Vermeer3,4 1 2

University of Bologna, Aerospace Engineering, Forlì, Italy Generic Design and Research, Airbus, Hamburg, Germany 3 Faculty of Aerospace Engineering, Delft University of Technology, Delft, Netherlands 4 Metallic Technologies & Surface Engineering, EADS Innovation Works, Munich, Germany

Abstract: A test campaign has been performed to investigate the design of fuselage stiffened panels with different metallic skin concepts (monolithic vs. Metal Laminates), stringer materials and local reinforcements from aluminum, titanium and glass-fiber. Observations were made on the crack growth retardation mechanisms. Bonded doublers and selective reinforcements confirmed to be outstanding tools to improve the damage tolerance properties of structural elements with a minor weight increase.

INTRODUCTION By means of adhesive bonded doublers it is possible to add reinforcements to a fuselage skin panel in order to guarantee slow crack propagation and increased residual strength. When damage occurs in the panel due to fatigue, corrosion or accidental causes, the internal bonded doubler provides an alternate load path. Key is then the load redistribution that takes place and causes the damage growth to be retarded thus allowing the damage to be discovered and repaired before its dimensions gets critical. Adhesive bonding is applied in Airbus metallic primary structures since the 70’s with the beginning of A300 manufacture and its use is continued with A310 / A330 M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 427–447. © Springer Science+Business Media B.V. 2009

428

I. Meneghin, M. Pacchione, P. Vermeer

/ A340 and the latest A380 program. Metal bonding had in the past a bad reputation regarding corrosion and durability: Airbus experience has shown no major in-service problems after early modifications to prevent bond-line corrosion. The robustness of the current process is confirmed by an outstanding in service records. In various Airbus fuselages aluminum doublers are bonded to the skin in longitudinal and circumferential joints as well as in the window belt area. A particularly significant application is the direct bonding of stringers on fuselage skin. Main drivers for the manufacture of adhesive bonded stiffened panels are the improved damage tolerance characteristics as well as the higher compression stability loads [1]. The experience gained with use of bonded aluminum panels has shown that one limit of the aluminum straps is the partial crack arresting capability due to the premature failure of the reinforcement caused by early nucleation of fatigue damages. To further improve the damage tolerance performances the reinforcing doubler or strap should preferably be from a fatigue insensitive material having also high stiffness and static strength. Such kind of reinforcement like titanium or glass fibers has been defined as “selective” reinforcement [2] to distinguish it from standard aluminum reinforcements. Bonded stiffened panels that combine thin metallic skin and selective reinforcements (i.e. fatigue insensitive) could guarantee slow crack propagation, crack arrest and large-damage capability combined with low structural weight. The first example of selective reinforcements used on aircraft fuselages has been the use of titanium bonded doublers. Adhesive bonding of titanium reinforcements on aluminum door structures was applied on the Douglas DC-9 in the 60s’, nevertheless, long term stability issue driven by moisture absorption lead Douglas to discontinue the use of adhesive bonding of Titanium. Few years later Lockheed introduced titanium circumferential straps in the fuselage of the L-1011. The adhesive bonded titanium crack stoppers suffered similar long term stability issues. After this learning experience adhesive bonding of titanium has not been used in fuselage primary structures. In the 80s, investigations were started to evaluate composite patches adhesive bonded on metallic fuselage skin as an easy to apply repair [4]. Few years later, in the 90s, Schijve [5] investigated the use of ARALL (Fiber metal laminate made of aluminum foils and aramid-fiber/epoxy plies) as crack stopper reinforcement and, in the late 90s, GLARE® (Fiber metal laminate made of aluminum foils and glass fibers) was developed and proposed for local reinforcements [7]. GLARE® is today used as skin material in the Airbus A380 program and as reinforcing doublers for some heavy loaded frames of the A400M. Recent research from various authors [2, 3] has explored in details the benefits to wing and fuselage damage tolerance of crack retardation techniques such as bonded straps or selective reinforcement of GLARE® or other materials bonded to the aluminum substrate. Considerable work has been performed on developing robust modeling strategies [6, 8]. The effects of bonded doublers and selective reinforcements is very difficult to be predicted numerically or analytically due to

Investigation on the design of bonded structures for increased damage tolerance

429

the complexity of the underlying mechanisms and failures modes acting: debonding at the interface between the doubler and the skin around the crack tip, load redistribution between damaged part and intact reinforcements, fatigue damage of the metallic reinforcement eventually causing its premature failure, crack bridging of long crack in case of fatigue insensitive doublers. In addition, secondary effects like tensile residual stresses from the bonding process and secondary bending due to the eccentric doublers increase the complexity of the phenomena. Reliable predictions of crack growth and residual strength on bonded structures can only be based on sound empirical and phenomenological considerations strictly related to the specific structural concept. In this paper the experimental investigation made in Airbus in the period ranging from 2002 to 2007 on methods to reinforce bonded stiffened fuselage panels is reported. Large flat bonded stiffened panels have been tested to quantitatively investigated the role of different metallic skin concepts (monolithic vs. metal laminates), of aluminum, titanium and glass-fiber reinforcements, of stringers material and of the geometry and location of doublers / selective reinforcements.

EXPERIMENTS A total of 35 stiffened panels, representative of a typical fuselage skin of a longrange family aircraft, were manufactured and tested in the laboratories of EADS-IW Ottobrunn. The fatigue crack propagation (FCP) was investigated for 24 of them. The residual strength (RS) was measured for the remaining 11 panels. Section A-A (a)

(b)

Figure 1. Test coupon: “seven stringers” panel with (a) doublers bonded between and below the stringers, with (b) additional glass fiber reinforcement The coupons, shown in Figure 1, consisted of a flat skin (1200mm wide and 1455mm long) with seven equally spaced bonded stringers. In addition to the

I. Meneghin, M. Pacchione, P. Vermeer

430

stringers, bonded doublers were placed below and/or between the stringers to act as additional reinforcing elements. All doublers were oriented in the stringer direction and orthogonal to the propagating crack. The combinations of skins, stringers and doublers tested are given in Table 1. The skin thickness was in the range 1.2mm to 1.6mm. Both monolithic and metal laminate (produced by adhesive bonding of thin sheets to obtain the required thickness) skins were manufactured. The metal laminate skins, see Figure 2, consisted of two layers of 0.8mm and 0.6mm thick sheets of similar and dissimilar materials. 2024-T3, 2524-T3 and 7475-T7 aluminum alloys were employed to manufacture the skins as described in the Table 1.

Figure 2. Cross section of a specimen with metal laminate skin The stringers were “J”-shape extruded profiles made of 7075-T73511 and high strength 7349-T76511 aluminum alloys. Different thickness, widths and materials (2024-T3, 2524-T3 and the titanium alloy Ti-6Al-4V) were chosen for the bonded doublers. In three panels aluminum doublers were reinforced by means of two or three layers of unidirectional S2 glass fiber prepreg lying between the strap and the skin as sketched in Figure 1(b). For two panels glass prepreg was used also to bond the doublers under the stringers. Type Monolithic [M]

Skin

Metal Laminates [L]

Monolithic [M]

Double r

Stringe r

Monolithic with reinforcement

Type “J“

Materials 2024-T3 2524-T3 2024-T3/2024-T3 2524-T3/2024-T3 2524-T3/7475-T7 2024-T3 2524-T3 Ti-6Al-4V Doubler Reinforcement 2024- Unidirectional Glass-Fiber T3 2524- Prepreg (S-2 glass-fibers) T3 7075-T73511 7349-T76511

Dimensions Thickness (mm) 1.2; 1.4; 1.6

Adhesives -

0.8 + 0.6

FM73M0. 3

0.7; 0.8; 1.2

FM73M0. 3

Doubler

Reinforcement

0.6

2 layers x 0.125 3 layers x 0.125

Cross-sectional area (mm2) 124, 126

FM94/S-2

FM73M0. 6

Table 1. Overview of the tested materials The aluminum pre-treatment before bonding consisted of Chromic Acid Anodization (CAA) and was followed by application of Primer BR 127. The

Investigation on the design of bonded structures for increased damage tolerance

431

specimens were bonded in autoclave using film adhesive FM 73 or S2 glass prepreg (adhesive FM94), both systems have a curing temperature of 120°C. The tests were performed by means of a servo-hydraulic INSTRON 8805 machine with a 1MN load cell. The clamping was specifically designed for the 7-stringers panels (see Figure 3/a). An anti-bending device was installed to prevent the out of plane deflection of the panel during the test (see Figure 3/b). (a)

(b)

Figure 3. (a) clamping system and (b) anti-bending device Fatigue crack propagation test The panels were provided with a through the thickness machined notch 50mm long (2a0). The notch was centered at the middle stringer; both the stringer and the relative doubler (when present) were cut. The loading parameters were the same for all the tested configurations, constant amplitude (CA), R=0.1. The tests ended when the crack was “four bay” long or in case of panel failure. The FCP over broken stringer has been analyzed for all the variants within the first two bays of the panel, Figure 4. The crack growth period that the crack spends below the bonded stringer is significant and has been measured during the tests; anyhow it will not be discussed in the following comparison of different reinforcing strategies with the assumption that an inspection interval for such a structure will be based on the crack growth period necessary to reach the next stringer foot.

Figure 4. Fatigue crack propagation over broken stringer within the 2 bays The observed crack length within the two bays were practically symmetrical: this has permitted to draw the crack propagation curves [a = f(N)] considering the

I. Meneghin, M. Pacchione, P. Vermeer

432

average crack length between the left and right crack tip displacement. The crack growth rate curves [da/dN = f(a)] were calculated with the equation: da a −a = i +1 i dN N i +1 − N i

(1)

where ai is the ith crack length measurement performed. Given that all the specimens were tested with the same load, dissimilar stress levels were induced for the different reinforcement configurations; however all results presented are based on comparisons among specimens that were characterized by equivalent stress conditions. Residual Strength test The residual strength (RS) test specimens were provided with a central saw-cut that, before the test, was fatigue propagated up to a two-bay crack length. Both the central stringer and the doublers bonded within the two-bays adjacent to the central stringer were cut. The tests were performed by continuously increasing the loading up to the occurrence of unstable crack growth and subsequent final failure.

RESULTS - Fatigue Crack Propagation tests Effect of the skin material In Figure 5 are reported tests performed on coupons with identical design (aluminum doubler between the stringers), geometry and loading conditions, differences are only in the material constituents. The tests show that the skin material itself has a significant effect on the FCP performances; in particular the beneficial effect of the 2524-T3 aluminum alloy is evident in comparison to the 2024-T3. The number of cycles cumulated for the crack propagation within the two bays for panels with the 2524-T3/2024-T3 and 2524-T3/7475-T7 laminate skins was respectively 60% and 100% higher than for the panel with a 2024-T3 laminate skin. On the other hand all the specimens made with 2024-T3 skin, either monolithic or laminate, achieved similar crack propagation performances. This shows that the metal laminates skin does not bring significant benefits on the constant amplitude crack propagation behavior compared to the monolithic skin. Effect of the bonded doublers Doublers bonded on the cracked substrate can retard the crack propagation by means of two mechanisms: i.e. Local Stiffening and Crack Bridging (see [6]). The bonded doubler is a stiffening element that enhances the load carrying capability of the cracked substrate. The load transfer between the skin and the doubler relieves the stress at the crack tip reducing the crack growth rate. This effect, known as local stiffening effect, starts working when the crack is close to the doubler and as long as it does not pass beyond the stiffener (Figure 6/a).

Investigation on the design of bonded structures for increased damage tolerance

433

Figure 5. Effect of skin material on FCP

(b)

(a)

Local Stiffening

Crack Bridging

Figure 6. Crack retardation mechanisms the bonded strap work with An intact bonded doubler that bridges the crack exerts a traction force on the crack lips and reduces the crack opening displacement. As a consequence, the crack driving force is decreased as well as the crack propagation rate. This mechanism,

434

I. Meneghin, M. Pacchione, P. Vermeer

known as crack bridging, begins exerting its action when the crack tip enters in the doubler covered area and go on up to the doubler failure (Figure 6/b). The effectiveness of the crack retardation mechanisms described above depends mainly on the doubler stiffness. The stiffness mismatch between the doubler and the substrate and the stress concentration at the doubler free-edges are the causes of a local progressive debonding at the interface around the advancing crack tip; this may reduce both the local stiffening effect and the crack bridging traction. These mechanisms and phenomena play a role whenever a doubler of whatsoever material is used to reinforce a substrate, nevertheless the amount of local stiffening and crack bridging contribution depends on the particular design solution adopted for the bonded doubler. The effect of an aluminum doubler in the middle of the bay can be appreciated observing the crack propagation curves in Figure 7 showing two similar panels with and without bonded doublers. The presence of the doubler (0.7mm thick and 35mm wide out of 2524-T3) increased by 80% the crack propagation period within the two bays; the weight increase of the panel was about 4%. Figure 7 highlights that the crack retardation effect induced by the aluminum bonded doubler is mainly concentrated in the first part of the doubler covered area, where the difference between the crack growth rates of the two panels is more pronounced. The crack retardation effect is quite limited when the crack is not yet at the edge of the bonded reinforcement and ends when the crack is still under the doubler itself. When the lead crack approaches the bonded doubler, the crack tip stress field is relieved by means of the local stiffening effect induced by the doubler, and the propagation rate decreases. When the crack starts growing under the bonded doubler the constraint on the crack propagation becomes more significant: the bonded strap bridges the crack tip and thus the crack opening becomes increasingly restrained. The combination of the local stiffening and a sort of a local crack bridging drives the crack growth rate to its minimum value (see point [▲] in Figure 7). From this point the crack growth rate starts increasing again even if the crack has propagated through a small fraction of the doubler covered area. This happens because, at the point indicated by the red triangle [▲] in Figure 7, the bonded strap nucleates a crack that drives the doubler itself to its premature failure. The new crack nucleates at the doubler free-edge notch as a consequence of the high magnitude cyclic strains induced locally (Figure 8).

Investigation on the design of bonded structures for increased damage tolerance

435

Figure 7. Effect of aluminum bonded doubler in the middle of the stringer bay

Figure 8. Crack nucleation in the free-edge of the bonded aluminum doubler induced by the underneath lead crack in the skin After the crack nucleation in the doubler the crack in the skin is less constrained and can accelerate (as seen in Figure 7) while the crack in the doubler lags behind it. As a consequence the complete aluminum doubler failure appears when the crack in the substrate has just gone beyond the doubler edge (see [♦] in Figure 7). In accordance with this observation an adhesively bonded aluminum doubler

436

I. Meneghin, M. Pacchione, P. Vermeer

works as an effective crack retarder only as long as it is perfectly intact and no crack is nucleated at its free-edge. Since the fatigue resistance of the doubler is a fundamental characteristic for an effective crack stopper its detail design plays an important role. A thick aluminum doubler is less prone to crack nucleation than a thin one when the lead crack is growing in the skin underneath it, as shown in Figure 8. As a result, in the first case the lead crack can propagate longer under the intact doubler than in the second case while its growth rate continuously decreases.

Figure 9. Effect of the thickness of the aluminum doubler bonded in the middle of the bay

Investigation on the design of bonded structures for increased damage tolerance

437

In Figure 9 the FCP behavior of two panels with 2024-T3 bonded doublers between the stringers with similar cross-sectional area but different thicknesses and widths have been compared. The other features of the two panels were identical. The crack propagation rates, reported in the Figure 9, show that the minimum of the crack growth rate under the thick doubler (1.2mm thick, 25mm wide) is lower and occurred at a longer crack length than in the case of thin doubler (0.8mm thick, 35mm wide), as it is represented by the shifting of the point indicated by▲. The crack propagation rate is given in the magnification as a function of a normalized space variable (named “anormalized”) so that, independently from the doubler’s effective width, the inner doubler edge corresponds to anormalized = 0 and the outer doubler edge to anormalized = 1. The test results in Figure 9 show that the crack propagation periods within the two bays increased by 47% substituting the 0.8mm thick aluminum doubler bonded between the stringers with 1.2mm thick aluminum doublers. This confirms that, for the same stiffness, thick and narrow aluminum doublers are more effective crack retarders than thin and wide ones. This behavior of the aluminum reinforcements has been confirmed testing a design with two adjacent doublers, 25mm wide, 0.8mm thick bonded between the stringers and 35mm spaced each other, as sketched in Figure 10 panel number 3.

Figure 10. Effect of the thickness of the aluminum doubler bonded in the middle of the bay The total cross-section of the two doublers between the stringers was 40mm2, that is a third more that the cross-section of the single doublers. Despite of the crosssection increase the crack propagation within the two bays was 14% shorter than the case of the thicker doubler (panel number 2 in Figure 10). A benefit was obtained only in comparison to the doubler having same thickness but smaller cross-section (panel number 1 in Figure 10). In conclusion it does appear that the effectiveness of an aluminum strap as crack stopper is driven by its thickness more that its width.

438

I. Meneghin, M. Pacchione, P. Vermeer

Outstanding crack propagation performances have been achieved when the aluminum doublers were reinforced by means of unidirectional (UD) glass-fibers placed at the interface between the doublers and the skin, as sketched in Figure 1. The results plotted in Figure 11 show that the crack propagation period within the two bays in the case of panels with aluminum doublers reinforced and bonded by means of three prepreg layers of UD glass-fibers was about two times longer than the case of similar specimens with aluminum doublers only.

Figure 11. Effect of glass-fiber reinforcement under the aluminum doublers This result has been achieved without weight increase, compensating the fiber weight with a lower thickness of the aluminum doubler (0.6mm thick in comparison to the 0.8mm thick adhesively bonded doublers). At the base of the good performance there is a different and more effective way of working of the glass-fibers reinforcement. The crack propagation rates in Figure 11 show that the retardation effect exerted by the glass-fibers reinforced doubler is not concentrated

Investigation on the design of bonded structures for increased damage tolerance

439

only under the bonded doubler (as it has been observed in the case of aluminum doublers bonded directly on the substrate) but its action is extended all along the two bays. This means that the glass-fibers prevent the fatigue nucleation of the new crack in the aluminum doubler and therefore avoid the subsequent premature doubler failure when the lead crack emerges underneath it. As a consequence the glass-fibers reinforced doubler is still intact when the growing crack tip is leaving the doubler itself toward the stringer, keeping on exerting its crack retardation effect by means of an effective crack bridging contribution (see Figure 11). This retardation mechanism is completely absent if the aluminum doubler is directly bonded on the skin surface, since the crack leaves a severed aluminum doubler behind it. The status of a 7-stringer panel, manufactured with glass-fibers reinforced aluminum doublers, at the end of the FCP test is shown in Figure 12. At the crack tip, as shown in the detail, the doubler appears still intact but totally debonded. Glass-fibers reinforced straps develop large delaminations being the prepreg peel strength lower than that of pure film adhesive.

Figure 12. Specimen with glass-fiber reinforced aluminum doubler bonded between the stringers at the end of the FCP test (final failure) and sketch of local debonding developed around the crack tip at the interface between the doubler and the skin during the FCP During the crack propagation under the doubler a local debonding is developed around the advancing crack tip as sketched in Figure 12. The local debonding does not affect the crack retardation effect of the bonded doubler and prevents the aluminum doubler from nucleating fatigue crack. The local debonding under the glass-fibers reinforced aluminum doubler spreads the doubler strain over the delaminated area, reducing the relative cyclic peak value; the fatigue crack initiation in the aluminum glass-fibers reinforced strap and its premature failure is thus avoided. From this point of view the prepreg layers create a barrier that preserves the aluminum layer from the crack nucleation. In conclusion, if a thick and narrow aluminum strap was the best solution for the

440

I. Meneghin, M. Pacchione, P. Vermeer

unreinforced doublers, a thin and wide aluminum strap is the best choice for a glass-fibers reinforced doubler. In view of the fact that the most significant crack retardation effect is obtained when the lead crack is still under the intact doubler it is a significant advantage to have a wider doubler without increasing the weight. Additional benefits of a thin and wide doubler are the reduced secondary bending and the lower stress concentration at the run-out. The specimen with titanium doublers bonded in the middle of the bays achieved the best crack propagation performance among all the panels of the experimental campaign. Figure 13 shows that the number of cycles recorded for the crack propagation within the two bays was 160% longer in comparison to the panel with aluminum doublers. Anyhow, it was the heaviest reinforcing solution being the weight of the titanium doubler respectively 13% and 22% heavier than the aluminum doublers and the glass-fibers reinforced doublers. The titanium doubler induces a crack retardation effect which is not only concentrated under the doubler itself but its effect is evident up to the stringer with the same behavior already observed for the glass-fibers reinforced aluminum doubler. The titanium doubler do not nucleate a fatigue crack and as a consequence it keeps on exerting the retardation effect after the lead crack passage in the skin underneath it. The ARAMIS strain map given in Figure 14 shows the crack tip beyond the first stringer and the intact titanium doubler still carrying load. It is evident in Figure 14 that the first stringer is broken as well: there was not doubler below the stringer. The specimen with the titanium doublers between the stringers with a four bays long crack was still having all the doublers intact and bonded on the skin (see Figure 15); this means that the doublers could still carry the fatigue load with the skin completely severed. The crack retardation mechanisms induced by the titanium doublers are qualitatively the same already discussed for the glass-fibers reinforced aluminum straps (i.e. local stiffening followed by crack bridging) but more effective due to the high titanium stiffness. According to the titanium high Young modulus a lower crack growth rate was expected in the early propagation stage, i.e. before the crack tip reaching the first doubler, in comparison to the panel with aluminum strap. As shown in Figure 13 this was not the case; a possible reason is the high tensile residual stresses that the bonding process set-up in the skin as a result of the different thermal expansion coefficients of titanium and aluminum. In fact, the epoxy adhesive used to bond the doublers requires a curing temperature of 120 °C resulting, after cooling, in tensile residual stresses in the skin. The tensile residual stress field can affect significantly the effectiveness of the bonded straps reducing the local stiffening contribution and, as a consequence, the number of cycles cumulated for the crack propagation up to the bonded strap. The residual stresses have a negligible effect on the crack bridging contribution.

Investigation on the design of bonded structures for increased damage tolerance

Figure 13. Effect of the titanium doublers bonded in the middle of the bay

Figure 14. Strain on a panel with titanium doubler in the middle of the bay

441

442

I. Meneghin, M. Pacchione, P. Vermeer

Figure 15. Specimen with intact titanium doublers at the end of the FCP test The possible presence of rivet holes in the titanium bonded doubler has not been investigated but it is anticipated that it may significantly affect the results. A fastener in proximity of the lead crack would prevent the debonding of the doubler, would reduce the cross section and introduce a stress concentration: all the three factors would contribute to an earlier fatigue crack nucleation. Bonded stringers The tests have shown that the fatigue initiation mechanism described for the aluminum doublers is the same that occurs for the bonded stringers when the fatigue crack propagates in the skin underneath them. The main difference between the bonded straps and bonded “J” stringers is that the latter have an unsymmetrical geometry: the foot side of the stringer is much less stiff than its head side (Figure 16). As a consequence the number of cycles cumulated for the crack initiation in the stringer, since the lead crack starts propagating under it, it is different according to which side the lead crack approaches the stringer itself. A stringer approached by the lead crack from the foot side (Figure 16/b) takes less cycles to nucleate a new crack than the stringer approached from the head side (Figure 16/c); as end result the first one fails before the second. The crack propagation on the stiffened panel beyond the first two bays could be highly unsymmetrical. In nine panels out of eleven the crack propagation after the first stringer was leaning toward always the same direction, i.e. the foot side of the stringers (see Figure 17).

Investigation on the design of bonded structures for increased damage tolerance

443

(a) Aluminum Bonded Doubler

(b) Bonded Stringer approached by the Foot Side

(c) Bonded Stringer approached by the Head Side

Figure 16. Fatigue crack initiation in the bonded doublers and stringers

Figure 17. Highly unsymmetrical crack propagation at the end of the FCP test

RESULTS - Residual Strength tests The residual strength test results are compared in terms of failure stress. The calculated values have been reported in Figure 18 and Figure 19 in normalized terms as a function of the failure stress of a tested panel taken as reference. The reference panel had a 2024-T3, 1.4mm thick, monolithic skin, 7075-T73511 bonded stringers and 2024-T3 doublers.

444

I. Meneghin, M. Pacchione, P. Vermeer

Effect of the stringer and skin material The panel design investigated, due to the high stiffening ratio (defined as (Astiffener+ Adoubler) / (Astiffener + Adoubler + Askin)), leads to a residual strength behavior driven by the stringer failure. As shown in Figure 18 the stringer material has a remarkable effect on the RS performances: panels (number 3, 4 and 5) with 7349-T76511 stringers failed respectively, at 4%, 5% and 8% higher stresses than the reference panel with 7075-T73511 stringers. The three specimens, assembled with the 7349-T76511 stringers, had different skin materials, which despite the high stiffening ratio seem to play a role on the RS performances; noticeable is the effect of the 2524-T3/7475-T7 laminate skin.

Figure 18. RS tests: effect of the stringer material The comparison of panels 3 and 4 with an identical stiffening configuration and respectively, a 2024-T3 monolithic skin and a 2024-T3/2524-T3 laminate skin shows that metal laminate skin does not bring any substantial advantage in comparison to the monolithic skins. Panel 2 had a lower failure stress than panel 1, despite of the higher toughness of the 2524-T3 monolithic skin; it confirm that the

Investigation on the design of bonded structures for increased damage tolerance

445

lower stiffening ratio that characterizes panel 2 with a 1.6mm thick skin has a larger effect than the skin material. Effect of the doubler under the stringer In the configuration tested the doublers placed under the stringers enhance significantly the RS of the panels, as shown in Figure 19. The failure stress for the panel 4 with 0.7mm thick aluminum doublers below the stringers was 6% higher than the panel 6 without the additional reinforcement; this beneficial contribution increased to 8% in case of 1.2mm thick aluminum doublers. The highest RS has been reached with 0.8mm thick, titanium doublers: 18% higher than for the panel without doublers below the stringers. The stiffness characteristics of the doublers are reported in Figure 19 in terms of the product between the Young modulus (E) and cross-sectional area (A). The comparison of panels 7 (aluminum doublers) and 8 (titanium doublers) having similar EA shows that the increase of residual strength is not only driven by the stiffness characteristics of the doubler.

Figure 19. RS tests: effect of doubler’s material under the stringer

446

I. Meneghin, M. Pacchione, P. Vermeer

DISCUSSION In the range of thickness investigated (monolithic skin thinner than 1.6mm) metal laminate skins (foils with thickness lower than 1 mm) have shown no benefits on the constant amplitude crack propagation and residual strength performances in comparison to the respective monolithic skins from the same material and thickness. The combination of different alloys 2524-T3/7475-T7 laminated to create a ML skin seems to influence positively both the fatigue crack growth rate and the residual strength. However due to the limited test experience with this material combination it is not possible to draw any definitive conclusion. Bonded straps confirmed to be an efficient solution to increase significantly the damage tolerance properties with reduced weight penalties. The effectiveness of bonded doublers to increase the crack propagation and the residual strength performances of a stiffened panel depends greatly on the material and geometrical features of the bonded doublers; a key factor is the relative fatigue resistance. Aluminum doublers bonded directly on the skin are prone to early nucleation of fatigue cracks when the lead crack is propagating in the skin under the doubler itself; this drives the aluminum doubler to a premature failure before the lead crack emerges beyond it. The retardation effect is thus limited and the aluminum doubler does not deliver any crack bridging contribution. A thick and narrow aluminum doubler has a longer crack nucleation period than a thin and wide doubler resulting, for the same weight, in a more effective crack retarder. A fundamental improvement is observed when the aluminum doublers are reinforced by means of unidirectional glass-fibers placed at the interface between the doubler and the skin. The glass-fibers prevent the fatigue nucleation in the doubler and thus avoid the potential aluminum doubler premature failure. The reinforced doubler continues exerting its retardation effect even after the passage of the lead crack beyond it by means of an effective crack bridging contribution. It can be concluded that the glass-fiber reinforced aluminum doubler act as a selective reinforcements. The stiffened panels with bonded Ti-6Al-4V doublers have achieved the best performances among all tested panels. Titanium doublers are resistant to fatigue crack nucleation and with their high modulus exert a strong crack bridging contribution. Thanks to the high static strength titanium doublers provide the highest residual strength performances and bring to the conclusion that titanium doublers are very effective selective reinforcements.

CONCLUSIONS A careful choice of the reinforcement materials and their design is fundamental to achieve an effective design solution for fuselage stiffened panels.

Investigation on the design of bonded structures for increased damage tolerance

447

Bonded doublers and selective reinforcements confirmed to be outstanding design tools to improve the damage tolerance properties of structural elements with a minor weight increase. The crack retardation mechanisms induced by selective reinforcements (titanium doublers or glass-fibers reinforced aluminum straps) is similar and consists of local stiffening followed by crack bridging. Tests results on stiffened panels show that fatigue insensitive selective reinforcements offer, in respect to standard aluminum doublers, significant improvement of the damage tolerance characteristic: longer propagation lives, or correspondingly higher allowable stress levels and enhanced residual strength performances.

ACKNOWLEDGEMENTS The metal laminate investigation has been performed in frame of the DIALFAST project (European Specific Targeted Research Project – contract No. AST3-CT2003-502846). The financial support of the EU 6th Framework is acknowledged. Grateful thanks are given to E. Hombergsmeier and V. Holzinger of EADS-IW Ottobrunn for technical advice, manufacture of the specimens and performing the tests.

REFERENCES [1] Pacchione, M., Hombergsmeier, E. (2008), Hybrid metal laminates for low weight fuselage structures, Proceedings of the 1st International Conference of Engineering Against Fracture, University of Patras, Patras, Greece [2] Heinimann, M.B., Bucci, R.J., Kulak, M., Garratt, M. (2005), Improving damage tolerance of aircraft structures through the use of selective reinforcement, Proceedings of the 23rd ICAF Symposium, Ed. Dalle Donne C, pp. 197-208 [3] Heinimann, M., Kulak, M., Bucci, R., James, M., Wilson, G., Brockenbrough, J., Zonker, H., Sklyut, H. (2007), Validation of advanced Metallic hybrid concept with improved damage tolerance, Proceedings of ICAF 24th, Ed Lazzeri L, Naples, pp 206-223. [4] Baker, A., Rose, F., Jones, R. (2002) Advances in the Bonded Composite Repair of Metallic Aircraft Structure, Vol. 1 - 2, Elsevier Science Ltd, Oxford, UK [5] Schijve, J. (1989), Crack stoppers and ARALL laminates, Tech. Report LR-589, Delft University of Technology, Delft, Netherlands. [6] Zhang, X., Boscolo, M., Figueroa-Gordon, D., Allegri, G., Irving P.E. (2009), Fail-safe design of integral metallic aircraft structures reinforced by bonded crack retarders, Engineering Fracture Mechanics, Vol. 76, pp. 114-133 [7] Lynch, W.J., Vlot, A. (1995), GLARE crack stoppers, Memorandum M718, Delft University of Technology, Delft, Netherlands [8] Rodi, R., Alderliesten, R., Benedictus, R., The effect of external stiffeners on the fatigue crack growth in fibre metal laminates, Proceedings of ICAF 24th, Ed Lazzeri L, Naples, pp. 858-875.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

AN EXPERIMENTAL INVESTIGATION ON THE FATIGUE PERFORMANCE OF RIVETED LAP JOINTS M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel Faculty of Mechanical Engineering & Robotics AGH University of Science & Technology, Al. Mickiewicza 30, 30-059 Krakow, Poland

Abstract: Results of an experimental research on the influence of several factors on the fatigue behaviour of simple riveted lap joint specimens under constant amplitude loading conditions are presented. The variables considered are the rivet type and material, sheet material and the squeeze force. Also, the effect of sheet thickness staggering in the overlap region on the joint fatigue life is studied in the context of secondary bending. The measurements of the driven head dimensions for a range of squeeze force levels and fatigue test results for rivets installed with various squeeze forces indicate that the squeeze stress rather than the rivet driven head dimensions is a safe standard for the quality of the rivet installation. The superior fatigue performance of a rivet with the compensator compared to the round head and universal rivet is noted. The underlying reason is the better hole filling achieved due to the presence of the compensator. Fatigue lives observed for the staggered thickness specimens are consistently longer than for the standard specimens. The analysis of these preliminary results suggests that a primary reason for the improved fatigue performance is the reduction of secondary bending.

INTRODUCTION Riveted joints are recognized as a fatigue critical element in metallic airframe construction. Among the factors contributing to the relatively low fatigue M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 449–473. © Springer Science+Business Media B.V. 2009

450

M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel

properties of riveted joints with built-in eccentricities, like for example the lap joints, is the occurrence of secondary bending. Fatigue properties of a riveted joint are affected by a number of aspects related to the rivet type and material, sheet material, production process and the joint geometry. In view of these complex dependencies fatigue life predictions can only be made employing a semiempirical approach, e.g. [1, 2]. Obviously, the reliability of the prediction heavily depends on the quality of the experimental data bank behind this type concepts. Though the stress-strain conditions in the real structure deviate from those in simple uniaxially loaded specimens employed in typical laboratory tests, the latter can be extremely useful for a detailed investigation of the effect of specific variables, for comparative studies or for the verification of numerical solution. In the present paper results of an experimental research on the influence of several factors on the fatigue behaviour of simple riveted lap joint specimens under constant amplitude loading conditions are presented. The variables considered are the rivet type and material, sheet material and the squeeze force. Also, the effect of sheet thickness staggering in the overlap region on the joint fatigue life is studied in the context of secondary bending.

SPECIMEN CONSIDERATIONS Specimen configuration Two types of riveted specimens were applied in the experiments. The dependence between the rivet squeeze force and the driven rivet head dimensions was obtained using a specimen shown in Fig. 1, as first proposed by Müller [3] and subsequently applied by others, e.g. [4]. Two sheet strips are riveted together with 10 pairs of rivets, one rivet type being used in each specimen. The fatigue tests were carried out on riveted lap joint specimens of two different configurations presented in Fig. 2. Besides the “standard” specimen geometry shown in Fig. 2a, a staggered thickness configuration proposed by Schijve [5] and shown in Fig. 2b was considered. Thickness reduction of the sheets from t=1.9 to 0.95 mm was made by face milling. In order to avoid the effect of the type of the specimen clamping on the stress conditions in the joint the distance between the outer rivet row and the neighbouring clamping edge always exceeded 50t. Sheet materials The sheet materials used are two aircraft aluminium alloys, namely D16CzATWH and 2024-T3. The Russian material D16CzATWH with a thickness of 1.9 mm was delivered in the Alclad condition by a Polish aircraft industry PZL Mielec. It is a high purity alloy the chemical composition of which is the same as of the 2124 alloy. As detailed elsewhere, the mechanical properties and the fatigue crack growth behaviour of D16 are similar to those reported in the literature for 2024-T3 Alclad [6]. The 2 mm thick 2024-T3 sheets were delivered by a Czech aircraft industry Evektor in both the Alclad and Alclad & anodized (sulfuric acid)

An experimental investigation on the fatigue performance of riveted lap joints

451

condition. The mechanical properties of both materials determined perpendicular to the rolling direction, are compared in Table I. The result for each batch of the material represents the mean value from the tests on six specimens. It is seen in Table I that the properties for either sheet material are above the required minimum values. Note that the required levels apply to the properties determined in the rolling direction because those in the perpendicular direction, which are known to be lower, are not available in the standards. The data in Table I indicate that compared to 2024-T3 the D16 material is slightly stronger and, at the same time, somewhat more ductile.

t t

Fig. 1. Rivet squeeze specimen.

Table I. Mechanical properties of the sheet materials (perpendicular to the rolling direction). Yield stress Ultimate strength Elongation Material Batch Su, MPa e, % S0.2, MPa A 324 468 22 B 318 462 20 D16CzATWH C 317 452 22 mean 320 461 21 276 420 13 required* D 291 433 18 2024-T3 276 420 15 required** * **

GOST 4784-97 MIL-HDBK-5H, 1 October 2001

M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel

452

a)

p

16

p p/2

16

t O8

s s=5D

12.5

s/2 p

16 t

O8

p

p p/2

16

t/2

25

s s=5D

0

6s

OD0

12.5

s/2

Rolling direction

b)

25

0

6s

Rolling direction

OD0

Fig. 2. Specimens for the fatigue tests: (a) standard and (b) staggered thickness configuration. Rivets The nomenclature used here for the characteristic rivet dimensions and the geometry of three protruding head rivet types considered in the present investigation is shown in Fig. 3. Two series of the rivets with the compensator (Fig. 3b) were used, either one of a different Polish aluminium alloy, namely PA24

An experimental investigation on the fatigue performance of riveted lap joints

453

and PA25. As seen in the latter figure, the compensator is a small protrusion on the mushroom rivet head. The PA24 material is supposed to be an equivalent of the 2117-T4 aluminium alloy (designation AD). The latter is the material of the universal rivets MS 20470-AD6-7 (Fig. 3d). The round head rivets (Fig. 3c) were only made of the PA24 alloy. Note in Fig. 3 b-d that the protruding rivet length Ho for the rivets of the PA alloys is 6.2 mm whilst for the MS rivet it is 7.1 mm. This difference follows from different rules for choosing the rivet length according to the PZL Mielec and Evector specifications.

c)

O10 O5

d)

O8

O9.5

7.13

D0

O5

O5

O4.77

D

O7.5

O7.5

O7.13

7.1

2.72

6.2

6.2

H0

H

2.62

t2

t1

2.3

2.7

2.9

b)

Dh Hh

a)

Fig. 3. Nomenclature for the rivet dimensions (a) and rivet types used in the experiments: (b) rivet with the compensator (OST1 34040-79 Russian standard); (c) round head rivet (BN-70/1120-03 Polish standard); (d) universal rivet MS 20470-AD6-7 (U.S. military specification). The mechanical properties were measured in tensile tests on PA24 and PA25 rivet material wire specimens. Average results presented in Table II indicate that PA25 is a harder material than the PA24 material. Engineering stress vs. strain plots were recorded during the compression tests on 10 mm long cylindrical samples machined from the PA24 and PA25 wire and from all rivets used in the present investigation. The diameter of the specimens from the MS 20470-AD6-7 rivet was 4.8 mm whilst all the other specimens had the diameter of 5 mm. Each of the curves shown in Fig. 4 represents the average data from several tests. It is seen that the plots for the PA24 wire, the PA24 round head rivets (notation PA24r) and the universal rivets (notation MS) almost coincide. Similarly, a good agreement is shown between the results for the PA25 wire and the rivet (with the compensator) of that material (notation PA25c). However, though the material of the other series of the rivets with the compensator is designated by the producer as PA24, the results for the corresponding samples (notation PA24c) indicate a more ductile material. Actually, the data in Fig. 4 imply that the difference between the PA24c rivet material and the PA24r rivet material is of the same order as between the PA24r and the PA25c rivet material.

454

M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel

Table II. Mechanical properties of the PA24 and PA25 rivet material wire. Mechanical properties Yield stress S0.2, MPa Ultimate strength Su, MPa Elongation e, %

Rivet material PA24 PA25 202 266 361 435 30 28

Fig. 4. Results of the compression tests on samples machined from the rivets and the rivet material wire. Specimen production The specimens were cut out of the sheets in such a way that the direction of loading in the fatigue tests be perpendicular to the rolling direction. The rivet holes in the specimens from Figs 1 and 2 were drilled according to the process specifications of PZL Mielec for the D16 alloy and Evektor for 2024-T3. The hole diameters (d) and tolerances are given in Table III where the survey of the specimens used in the fatigue tests is presented. Each of the four squeeze force specimens (Fig. 1) was filled with one specific type of the rivet. Thus, the D16 sheets with the thickness t of 1.9 mm were joined with the PA24r, PA25c and PA24c rivets whilst the MS rivets were used for joining the t=2.0 mm thick 2024-T3 sheets.

An experimental investigation on the fatigue performance of riveted lap joints

455

Table III. Riveted specimens for the fatigue tests. Specimen type Sheet material

Standard D16CzATWH

Staggered 2024-T3

Surface Alclad & Alclad Alclad condition anodized Sheet 1.9 2 thickness t, mm* Rivet PA24r PA24c PA25c MS 20470-AD6-7 type Rivet 5D0 5D0 row pitch 5D0 7.5D0 p, mm Rivet 5 4.8 diameter D0, mm Rivet 4.9H11 5.05H12 hole (5.05+0.12) (4.9+0.075) d, mm *

D16CzATWH Alclad 1.9 PA24r 5D0

5 5.05H12 (5.05+0.12)

see Fig. 2

A force controlled riveting was applied using a squeezer shown in Fig. 5 mounted in the grips of a MTS 810 fatigue machine. The spring ensured the preliminary clamping of the sheets and a good contact between the manufactured rivet head and the sheet. The device effectively prevents overlap curvature and rivet tilting. For each squeeze force specimen the squeeze force level was stepwise increased for the following rivets from a relatively low to a relatively high value. This enabled to obtain the driven rivet head dimensions over a range of squeeze loads for each combination of the sheet material and rivet type. When riveting the specimens for the fatigue tests, measures were taken to avoid the so called edge effect. Within the overlap the lateral contraction is smaller because the average stress is lower than outside the overlap. This causes shear loads on the rivets in the direction transverse to loading which are highest in the edge rivets of the critical (outer) rows. Because these edge rivets carry a load larger by some 10% than the inner rivets [3] in the row, premature crack initiation occurs at their location. These cracks grow to the edges of the specimen and cause a general collapse before significant crack nuclei at the inner rivets could develop. The

456

M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel

above phenomenon referred to as the edge effect was successfully prevented in the present fatigue tests by installing all edge rivets with a squeeze force increased by 10% compared to the other rivets.

Fig. 5. Set used for riveting.

EFFECT OF THE SQUEEZE FORCE ON THE DRIVEN RIVET HEAD DIMENSIONS The diameter D and height of the driven rivet head H (see Fig. 3a) were measured with an accuracy of 0.01mm using a micrometer gauge. Rather than perfectly cylindrical the driven rivet head is of a barrel shape. The D-value was obtained as the average of the maximum barrel diameter measured in the longitudinal and transverse direction of the squeeze force specimen. All H-values measured at the rivet head centre were found to be very close to those computed from the equation

D 2 H = D02 H 0

(1)

which confirms the validity of the assumption that the volume of the rivet head remains constant during the rivet squeezing, consistent with the results of de Rijck et al [4]. Fig. 6 shows the dependence of the normalized driven head diameter D/Do on the squeeze force level Fsq. As could be anticipated, at a given Fsq-value the smallest driven head diameter is always obtained for the hardest material PA25 and the

An experimental investigation on the fatigue performance of riveted lap joints

457

largest diameter occurs for the most ductile material of the PA24c rivet. The coincidence of the results for the PA24c and MS rivets is purely accidental in that it stems from the addressed earlier differences in the length of either rivet type. Consequently, the Ho/Do ratio equals 1.24 for the PA24c rivet and 1.49 for the MS rivet. Schijve [7] has shown that the shorter the protruding length of a rivet the smaller the driven head diameter at a given squeeze force level. This implies that increasing the Ho/Do ratio for the PA24c and PA24r rivets would cause a shift upwards of the corresponding D/Do vs. Fsq curves. Consequently, the PA24r rivet data would shift closer to and the PA24c rivet data would shift more away from the MS rivet results.

Fig. 6. Normalized driven rivet head diameter as function of the squeeze force. In Fig. 7 the normalized rivet head dimensions are presented against the squeezing stress defined as

σ sq = Fsq /(πD 2 / 4)

(2)

Also plotted in Fig. 7 is the scatter band of data for the 2117-T4 aluminium alloy rivets reported by de Rijck et al [4]. It is seen that the results for the MS and PA24r rivets made of the same material fall within the de Rijck’s data scatter band. For the PA24c rivet, the D/Do and H/Ho data falling above and below respectively the scatter band of de Rijck’s results confirm the higher ductility of this type of rivet

458

M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel

material. The opposite trend is shown in Fig. 7 by the results for the PA25c rivet made of the hardest material.

Fig. 7. Normalized driven rivet head dimensions as function of the squeeze stress together with the scatter band of de Rijck et al [4] for 2117-T4 rivets.

RIVET HOLE EXPANSION MEASUREMENTS The favourable influence of a high squeeze force on the fatigue behaviour of riveted joints is well documented in the literature, e.g. [3, 8-10 ]. Among beneficial aspects of increasing Fsq is the improved hole filling associated with the interference fit and the hole expansion. This prevents rivet tilting, causes a pretension stress distribution around the rivet hole which is favourable for the load transmission, and diminishes the sensitivity for the surface finish of the hole. Results of the hole expansion measurements for the PA24c and PA24r rivets installed in the squeeze force specimen (see Fig. 1) of the D16 sheets are presented in Fig. 8. The hole expansion (he) is defined as

he =

de − d ⋅ 100% d

(3)

An experimental investigation on the fatigue performance of riveted lap joints

459

where d and de is the rivet hole diameter prior to riveting and the expanded rivet diameter respectively.

Fig. 8. Comparison between hole expansion for PA24r and PA24c rivets in D16 1.9 mm thick sheets: (a) D/Do=1.3; (b) D/Do=1.5.

460

M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel

Both d and de were measured in several directions with an accuracy of 0.01mm using an optical microscope. The results shown in Fig. 8 are based on the average values. The de-diameter measurements were taken in the rivet cross sections at two locations for each sheet, namely 0.2 mm and 1 mm below each rivet head. To this end both rivet heads were first cut away. Then for either sheet two layers of the material were consecutively removed by milling in order to obtain the desired locations. The machined area was subsequently polished to facilitate the observations of the rivet/hole boundary. As expected, the results in Fig. 8 demonstrate that for either rivet type he is larger at the higher squeeze force, i.e. for D/Do=1.5, than at the lower Fsq-level (D/Do=1.3). The measurements indicate considerable differences between the hole expansion behaviour of both rivet types. For the rivet with the compensator he below the manufactured head is much larger than below the driven head. The reverse is true for the round head rivet but the differences between he in either location are much less significant. Generally, due to the presence of the compensator he for the PA24c rivet is larger than for the PA24r rivet, the advantage of the rivet with the compensator being particularly pronounced within the sheet adjacent to the manufactured head for D/Do=1.5. For both D/Do-values the average results for the round head rivet fall between shown in Fig. 8 measurement data reported by Müller [3] for countersunk DD rivets with Do=4.8 mm and 2024-T3 sheet thicknesses of 1.6 and 2.2 mm. A reasonable agreement is also revealed between the present measurements for the round head rivet and the FE results of Rans [11] derived for the universal Do=3.2 mm AD rivet and 2024-T3 sheet thickness of 1 mm. In Figs 9 and 10 measurement results on he for two sheet and PA24r rivet configurations occurring in the staggered thickness specimen are compared with the data for the standard specimen considered in Fig. 8. When joining the 0.95 mm thick and the 1.9 mm thick sheet the same driven rivet head dimensions were obtained as in the case of joining two 1.9 mm thick sheets at the same squeeze force level when the protruding rivet length (Ho=6.2 mm) was the same for all configurations. The de measurements in the 0.95 mm thick sheet were taken for z=0.1 and 0.6 mm. In the 1.9 mm thick sheet de was measured at the same locations as previously in Fig. 8. Again, the hole expansion for the lower D/Do ratio of 1.3, Fig. 9, is smaller than for the larger driven head diameter (D/Do=1.5), Fig. 10. Other trends revealed in Figs 9 and 10 are common for both D/Do ratios. For the sheet adjacent to the driven rivet head he in the thinner sheet is always larger than in the thicker sheet. With the sheet adjacent to the manufactured head, hole expansion remains not affected by the sheet thickness configuration. The scatter of the results revealed in Figs 8-10 is due to the scatter in the d-values rather than in the de-values. It was observed that at a given squeeze force level and the rivet type the expanded rivet diameters showed only minute differences. At the same time, due to the relatively large tolerance of the rivet hole diameter allowed

An experimental investigation on the fatigue performance of riveted lap joints

461

for the PA24 and PA25 rivets (see Table III) the d-diameters for nominally identical holes could differ by even 0.1 mm.

Fig. 9. Comparison between hole expansion for PA24r rivet and three D16 sheet thickness configurations. Sheet adjacent to: (a) rivet driven head; (b) rivet manufactured head. D/Do=1.3.

462

M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel

Fig. 10. Comparison between hole expansion for PA24r rivet and three D16 sheet configurations. Sheet adjacent to: (a) rivet driven head; (b) rivet manufactured head. D/Do=1.5.

An experimental investigation on the fatigue performance of riveted lap joints

463

FATIGUE TEST RESULTS The fatigue tests were carried out on a MTS 810 fatigue machine. A pin-hole connection between the steel plates clamping the specimen ends and the attachment to the hydraulic grips of the fatigue machine effectively prevented the in-plane bending of the specimens. All specimens were tested under constant amplitude loading with a stress ratio (R) of 0.1 and a load frequency of about 20 Hz. Effect of the production variables on the fatigue life In Fig. 11a the fatigue lives (Nf) observed in the tests with the maximum applied cyclic stress level Smax=120 MPa are presented in terms of the driven head diameter to rivet shank diameter ratio. Except for the PA25c rivet for all combinations of the rivet type and sheet material Nf increases with increasing D/Do. This trend is expected because for a given specimen type D/Do increases with increasing the squeeze force level (see Fig. 6) which, as said earlier, improves the fatigue performance of the riveted joint. The form of presentation as in Fig. 11a implies up to twice longer lives for the PA24r rivet specimens than for the MS rivet and 2024-T3 Alclad sheet specimens. The only difference between the PA24r and MS rivet is the protruding length Ho because the material (see Fig. 4) as well as the manufactured head shape and dimensions (see Figs 3c and d) are nearly identical. Due to the different Ho-lengths, however, the same D/Do value is obtained for the PA24r rivet at a higher squeeze stress level than for the MS rivet, see Fig. 7. If Nf is presented against σsq, as in Fig. 11b, the data for both specimen types addressed above almost merge. Altogether, a closer insight into the results for the PA24r rivet and the MS rivet in 2024-T3 sheet presented in Figs 7 and 11 leads to some important observations and conclusions. Obtaining the same driven head diameter D at different protruding lengths Ho requires different squeeze stress levels. Similarly, the same height H of the driven head is obtained at different σsq-values if different Ho-lengths are used. Fatigue lives for rivets with different Ho-values but of the same driven head diameter or the same driven head height can be quite different at different σsqvalues. However, for the same σsq magnitude the fatigue lives observed are very similar independent of Ho. It can be concluded based on the present results and in accordance with Müller’s investigation [3] that σsq rather than the rivet driven head dimensions like D/Do or H is a safe standard for the quality of the rivet installation. A conspicuous result in Fig. 11 is the superior fatigue performance of the PA24c rivet. Evidently, the longest fatigue lives observed for this type rivet are due to the beneficial influence of the compensator which enables a hole expansion much higher than in the case of the PA24r rivet, as shown previously in Fig. 8. As revealed in Fig. 11 by the result denoted as PA24c-rem, removing the

464

M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel

compensator by machining causes the fatigue life to drop to the level observed for the PA24r rivet and the MS rivet in 2024-T3 Alclad sheet.

Fig. 11. Fatigue lives of riveted specimens as function of: (a) the rivet driven head diameter; (b) the squeeze stress.

An experimental investigation on the fatigue performance of riveted lap joints

465

Slightly lower fatigue lives observed in Fig. 11 for the Alclad and anodized 2024T3 sheet than for the Alclad material are consistent with the literature evidence, e.g. [12]. The present results further suggest that the small differences between the monotonic mechanical properties of the D16 and 2024-T3 material (see Table I) do not yield noticeable differences in the fatigue responses of both materials when in the Alclad condition. As seen in Fig. 11, the results for the PA25c rivet show an extremely large scatter. When installing this type rivets in the squeeze force specimen the rivet tilting and a cracking of the driven head occurred. This tendency was already observed for the D/Do ratio slightly below 1.4 and became more pronounced when the higher squeeze force was applied. Obviously, such a quality of the riveting is unacceptable. Altogether, the present results suggest that the PA25 alloy, though designated in the OST1 34040-79 standard for the rivet with compensator, is not an appropriate rivet material. Effect of staggering of the sheet thickness For joints with eccentricities, like lap joints or single strap butt joints, bending of the sheets, referred to as secondary bending, occurs under nominally tensile loading. The bending stresses Sb induced at a given location under the remotely applied tensile stress S can be estimated using a simple model proposed by Schijve [13]. For a lap joint with three rivet rows, the location of the maximum bending moment is always at the load path eccentricities, namely at the outer rows I and III. The largest positive bending stresses are produced along the faying surface at location A of sheet 1 and location B of sheet 2, Fig. 12. At the same time, outside the end rivet rows either sheet carries the full load P. Consequently, A and B are also the locations of the maximum combined tensile stresses (S+Sb) and, therefore, the most commonly observed initiation sites of the fatigue cracks.

Fig. 12. Locations of the maximum tensile stresses in a lap joint with three rivet rows. Two somewhat antithetic design concepts to provide for the high stresses at the outer rivet rows of the lap joint have been proposed. With an approach referred to as a “padded” lap splice design the sheets are locally thicker in the joint region [1]. Padding can be achieved either by bonding additional skins in the overlap area or milling the sheets down to their final thickness away from the joint, thus leaving

466

M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel

thicker pads in the joint region. The larger sheet thickness in the overlap area leads to a reduction of the nominal tensile stresses on the joint but, at the same time, yields elevated bending stresses due to the increased eccentricity at the outer rivet rows. Therefore, some of the beneficial effect of the pads may be negated by the bending moment they introduce. A totally different option of improving the fatigue quality of the lap joint proposed by Schijve [5] is to locally decrease the sheet thickness around the outer rivet rows, as shown in Fig. 2b. On one hand, the concept is conceived to reduce the secondary bending by decreasing the eccentricity of the crack path at the end rivet rows. Halving the original thickness, as shown in Fig. 2b, yields the eccentricity twice lower compared to the standard geometry from Fig. 2a. At the same time, it is presumed that a locally lower sheet stiffness around the outer rivet rows can release the load transmission by these rows, in a similar way as it occurs when the end of the sheet is of a finger tip form [14]. So far, the utility of staggering the thickness has not been verified experimentally. The staggered thickness specimens with two different round head rivet row spacings (p), namely of 5Do (25 mm) and of 7.5Do (37.5 mm), have been considered in the present fatigue tests, see Fig. 2b and Table III. Because staggering of the thickness implies a small weight reduction, the staggered thickness specimen is lighter compared to the reference standard specimen at the same p=25 mm. If p for the staggered thickness configuration is increased by a factor of 1.5 to p=37.5 mm, the weight of the staggered lap joint will be the same as of the reference standard joint. The contribution of secondary bending at a given applied stress amplitude Sa can be quantified by the bending factor defined as

K b(a) =

( S + S b )a − Sa Sa

(4)

where (S+Sb)a is the combined tensile stress amplitude at the critical location (A or B, Fig. 12). Application of Schijve’s model [13] referred to above yields for the considered specimen configurations the following Kb(a)-factor values and this factor reduction compared to the reference standard geometry: Reference standard, p=25 mm: Kb(a)=1.06 to 1.18 Standard, p=37.5 mm: Kb(a)=0.89 to 0.97, Kb(a)-reduction=17 % Staggered thickness, p=25 mm: Kb(a)=0.99 to 1.11, Kb(a)-reduction=6.2 % Staggered thickness, p=37.5 mm: Kb(a)=0.81 to 0.89, Kb(a)-reduction=24.3 %.

An experimental investigation on the fatigue performance of riveted lap joints

467

For each geometry the higher and lower Kb(a)-value corresponds to the lowest and the highest applied stress amplitude respectively considered in the fatigue tests because, according to Schijve’s model, Sb is a nonlinear function of S. To reduce locally the D16 sheet thickness from 1.9 to 0.95 mm the milling operation was applied. The same protruding rivet length Ho=6.2 mm was maintained independent of the combined sheet thickness. Two sets of specimens were produced, namely with the D/Do ratio of 1.3 and 1.5. For each set the rivets were installed with the same squeeze force for the standard and staggered specimens selected according to the data in Fig. 6. No difference between the rivet driven head dimensions for the thicker and thinner overlap was observed at the same Fsq level. The fatigue test results in the form of Sa vs. Nf data and the corresponding trend lines (Basquin equation) are presented in Figs 13a and b for D/Do of 1.3 and 1.5 respectively. It is evident that the staggered thickness configuration with p=37.5 mm yields consistently longer fatigue lives than the reference standard configuration (p=25 mm). The data obtained so far suggest that at D/Do=1.3 more benefits are seen at the higher applied stresses whilst the reverse is true for D/Do=1.5. The staggered thickness specimens with p=25 mm show longer fatigue lives compared to the reference standard specimens only for D/Do=1.3. Note in Fig. 13b that two available results for the standard geometry with p=37.5 mm fall between the data for the reference standard specimens and for the staggered thickness specimens with p=37.5 mm. This result is fully consistent with the Kb(a)values produced above. When the fatigue lives are presented in terms of the combined tensile stress amplitude, as shown in Fig. 14, a considerable consolidation of the data is obtained compared to Fig. 13. This is quantified by the high correlation coefficient values (R2) which amount 0.94 and 0.96 for D/Do=1.3, Fig. 14a and for D/Do=1.5, Fig. 14b respectively. It can be concluded that the reduction of secondary bending is the important reason underlying the beneficial influence of thickness staggering on the lap joint fatigue performance. Fractographic observations Observations on the crack initiation sites and the visible crack growth are in general consistent with the literature evidence, e.g. [3, 15]. The fatigue cracks always started at the mating surface at one or sometimes at both of the outer rivet rows. At the lowest squeeze force levels corresponding to D/Do of 1.3 the cracks nucleated at the rivet holes which gave rise to a quarter-elliptical crack, Fig. 15a. With increasing Fsq-magnitude the crack initiation location tended to move away from the hole edge and a semi-elliptical crack resulted, as shown in Fig. 15b. In all specimens the cracks propagated through the rivet holes in a plane slightly shifted away from the minimum (net) section, as illustrated in the top of Fig. 16. The amount of shift was not observed to be related to the squeeze force level. Typically, the MSD occurred. Exemplary results on visible crack growth for one of the specimens are shown in Fig. 16. In none of the tests the part of the fatigue life

468

M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel

from when the cracks were visible until failure exceeded 15%, again in agreement with the literature results, e.g. [3, 16].

Fig. 13. Fatigue lives for standard and staggered thickness riveted specimens as function of the applied stress amplitude: (a) D/Do=1.3; (b) D/Do=1.5.

An experimental investigation on the fatigue performance of riveted lap joints

469

Fig. 14. Fatigue lives for standard and staggered thickness riveted specimens as function of the combined tensile stress amplitude: (a) D/Do=1.3; (b) D/Do=1.5.

470

M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel

All specimens with the PA24c rivets fractured in the sheet adjacent to the driven head. The obvious reason was the much smaller hole expansion under that head compared to the manufactured head side, see Fig. 8. Such a preferred location of cracking was not observed for the specimens with the round head or universal rivets for which hole expansion of a similar magnitude occurs within either sheet, as indicated in Figs 8 to 10.

Fig. 15. Effect of squeeze force on the crack initiation site: (a) corner crack at rivet hole edge; (b) semi-elliptical crack slightly away from hole

Fig. 16. Fatigue crack path and visible crack growth record for a riveted specimen. D16 sheet, PA24r rivet, D/Do=1.5, Sa=36 MPa.

An experimental investigation on the fatigue performance of riveted lap joints

471

CONCLUSIONS The experiments presented in this paper lead to the following conclusions: 1. The results on the driven head dimensions for the rivets of the PA24 aluminium alloy presented in terms of the squeeze stress confirm that this material is equivalent to the AD rivet material. 2. The present results suggest that the small differences between the monotonic mechanical properties of the D16 and 2024-T3 sheet material do not yield noticeable differences in the fatigue responses of both materials when in the Alclad condition. 3. Hole expansion for the rivet with the compensator is larger than for the round head rivet. This effect becomes more pronounced below the rivet manufactured head and at a larger squeeze force. The hole expansion behaviour is different for either rivet type. With the rivet with the compensator, expansion is always higher in the sheet adjacent to the manufactured head whilst for the round head rivet a tendency for slightly higher expansion under the driven head is observed. 4. When two sheets with a different thickness are joined with the round head rivet, hole expansion in the thinner sheet is larger than in the thicker sheet if the thinner sheet is adjacent to the driven rivet head. With the thinner sheet adjacent to the manufactured head, hole expansion remains not affected by the staggered sheet thickness configuration. 5. For the round head and universal rivets (same material and manufactured head diameter, slightly different rivet shank diameter, different protruding length) the same fatigue lives are observed at the same squeeze stress level. When the fatigue lives are correlated using the rivet driven head-to-rivet shank diameter ratio different results for either rivet type are obtained. The above shows that the squeeze stress rather than the rivet driven head dimensions is a safe standard for the quality of the rivet installation. 6. Within the range of squeeze force values considered the fatigue performance of the rivet with the compensator of the PA24 alloy is far superior to the performance of the round head rivet and universal rivet made of the same alloy. Machining away the compensator results in the drop in the fatigue life to the level observed for the round head and universal head rivets. The underlying reason behind the improved fatigue properties of the rivet with the compensator compared to other two rivet types is the better hole filling achieved due to the presence of the compensator. 7. The present results suggest that the hard PA25 alloy, though designated in the Russian standard for the rivet with the compensator, is not an appropriate rivet

472

M. Skorupa, A. Skorupa, T. Machniewicz and A. Korbel

material. Fatigue lives for rivets with the compensator of this alloy show extremely large scatter and are much lower than for the same type rivet of the PA24 material. 8. Preliminary fatigue tests on the staggered thickness specimens suggest that fatigue lives of these specimens are consistently longer than observed for the standard specimens. It is noteworthy that the improved fatigue performance is achieved without a weight penalty. A good consolidation of the fatigue test results obtained when the fatigue lives observed for several configurations of the staggered thickness and standard specimens are presented in terms of the combined tensile and bending stress amplitude implies that the benefits from thickness staggering come mainly from the reduction of secondary bending. A further verification of this concept is recommended.

REFERENCES [1] Das, G.K., Miller, M. and Sovar T. (2001). In: Design for durability in the digital age, Proceedings of the 21st ICAF Symposium, vol. I, pp. 124–138, Rouchon, J. (Ed.), Cépaduès Editions, Toulouse. [2] Homan, J.J. and Jongebreur, A.A. (1993). In: Durability and structural integrity of airframes, Proceedings of the 17th ICAF Symposium, vol. I, pp. 173190, Blom, A.F. (Ed.), EMAS, West Midlands, UK. [3] Müller, R.P.G. (1995). An experimental and analytical investigation on the fatigue behaviour of fuselage riveted lap joints. The significance of the rivet squeeze force, and a comparison of 2024-T3 and Glare 3. PhD. Thesis, Faculty of Aerospace Engineering, Delft University of Technology, The Netherlands. [4] Rijck de, J.M.M., Homan, J.J., Schijve, J. and Benedictus, R. (2007), Int. J. Fatigue, vol. 29, pp. 2208-2218. [5] Schijve, J. (2006), Riveted lap joints with a staggered thickness in the overlap of the joint. Calculations of secondary bending. Doc. B2-06-02, Faculty of Aerospace Engineering, Delft University of Technology, The Netherlands. [6] Schijve, J., Skorupa, M., Skorupa, A., Machniewicz, T. and Gruszczyński, P. (2004), Int. J. Fatigue, vol. 26, pp. 1-15. [7] Schijve, J. (1998), Some considerations on the correlation between the rivet squeezing force and the dimensions of the driven rivet head. Memorandum M-847, Faculty of Aerospace Engineering, Delft University of Technology, The Netherlands. [8] Schijve, J. (1992). In: Proceedings of International Workshop on Structural Integrity of Ageing Airplanes, pp. 2-27, Atlanta Technical Publications. [9] Schra, L., Ottens, H.H. and Vlieger, H. (1995), Fatigue crack growth in simulated Fokker 100 lap joints under MSD and SSD conditions. NLR CR 95729 C, National Aerospace Laboratory NLR, The Netherlands. [10] Harish, G., Farris, T.N., Wang, H.L. and Grandt, A.F. (1999). In: Proceedings of the USAF Aircraft Structural Integrity Program Conference, San Antonio, Texas, pp. 1-14.

An experimental investigation on the fatigue performance of riveted lap joints

473

[11] Rans, C.D. (2007), The role of rivet installation on the fatigue performance of riveted lap joints. PhD Dissertation, Carleton University, Ottawa, Canada. [12] Wanhill, R.J.H. (1996), Some practical consideration for fatigue and corrosion damage assessment of ageing aircraft. NLR TP 96253 L, National Aerospace Laboratory NLR, The Netherlands. [13] Schijve, J., Campoli, G. and Monaco, A. (2009), Int. J. Fatigue, in press. [14] Swift, T. (1990), Repairs to damage tolerant aircraft. Technical Report FFAAIR-90-01, Federal Aviation Administration, U.S.A. [15] Schijve, J. (2009), Fatigue of structures and materials, 2nd Edition, Springer. [16] Ottens, H.H. (1995), Multiple crack initiation and crack growth in riveted lap joint specimens. NLR TP 95049 L. National Aerospace Laboratory NLR, The Netherlands.

ACKNOWLEDGEMENTS Thanks are due to Prof. Jaap Schijve, TU Delft, for his valuable advices in the course of this research and comments on the manuscript. Useful information from Ir. Johannes Homan, Fatec Engineering, is gratefully appreciated. The financial support from the governmental research funds within the years 20062009 via the Eureka project No. E!3496 is acknowledged

25th ICAF Symposium – Rotterdam, 27–29 May 2009

IMPROVING THE FATIGUE LIFE OF AERONAUTICAL SINGLE-LAP BOLTED JOINTS THANKS TO THE HYBRID (BOLTED/BONDED) JOINING TECHNOLOGY E. Paroissien, C.T. Hoang Ngoc, H. Bhugaloo, D. Ducher SOGETI HIGH TECH, PE3, Department of Fatigue and Damage Tolerance, Blagnac, France

Abstract: It has been experimentally shown [1-5] the possibility to obtain with hybrid (bolted/bonded) joining technology higher static failure load and a longer fatigue life than the corresponding bolted or bonded joints by using a suitable adhesive. This paper aims at comprehensively showing, by both a simplified analytical approach and an accurate three-dimensional finite element analysis that the application of hybrid (bolted/bonded) joining technology instead of the classical bolted technology allows for a possible improvement of fatigue life. A simplified theoretical analysis is presented to understand the mechanical behaviour of such joints and to provide possible elastic mechanical properties of a suitable adhesive. Then, an accurate three-dimensional Finite Element model is developed to demonstrate the possible benefit on fatigue life.

INTRODUCTION In the frame of aircraft performance efficiency enhancement both by reducing manufacturing costs and increasing weight saving, the design of longitudinal metallic joints of civil aircraft is under consideration. These joints are mainly single-lap bolted joints, the criterion of design of which is the fatigue strength. Moreover, between jointed sheets, a layer of sealant is applied to ensure the sealing of the pressurized cabin and the protection against galvanic corrosion.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 475–492. © Springer Science+Business Media B.V. 2009

476

Paroissien, Hoang Ngoc, Bhugaloo and Ducher

Hybrid (bolted/bonded) – quoted HBB – joining technology allows for associating a discrete transfer mode with a continuous transfer mode, each one having its own stiffness. The bolted system (discrete transfer mode) generates a high stress concentration around the holes, which is penalising to the fatigue strength. The bonded part allows a better distribution of the load transfer between the adherends. In regard to aircraft assembly, HBB joining could be interesting because it could reduce the load transferred by the fasteners to improve the fatigue life, while ensuring the static strength under extreme loads. HBB joining technology was presented as a relevant concept of fail-safe structures by Hart-Smith [6] in 1982. According to this study, HBB joints with aerospace configurations and material systems do not offer any significant increase in strength compared to bonded joints, which could be explained by the low fraction of load transferred by the fasteners. In 1995, Imanaka [1] showed that the fatigue strength of bonded joints can be improved through the combination with a rivet. Since 2000, along with the development of adhesive materials as well as the increasing use of composite structures in industrial applications, some studies have been pursued to analyse the mechanical performance of HBB joints. Fu and Mallick [2] experimentally demonstrated that single-lap HBB joints with structural injection moulded (SRIM) composite as the adherends and epoxy material as the adhesive possess a higher static strength and longer fatigue life as the corresponding bonded joints. Kelly [3,4] or Paroissien [5] experimentally showed that as the load can be shared between the adhesive and the bolt by using low modulus adhesive, HBB joints can have greater static strength and fatigue life than the corresponding bonded or bolted joints respectively. Paroissien et al. recently developed analytical one-dimensional (1D) [7] and two-dimensional (2D) [8] models which allow for the investigation of balanced single-lap HBB joints with linear elastic material systems. It was shown that a low modulus adhesive should be used, in order to suitably distribute the load between the adhesive layer and the fasteners. In order to demonstrate the possible benefits of HBB joining technology in fatigue life comprehensively, a reference configuration of a two fastener lines single-lap HBB joint is chosen, as illustrated in Figure 1. The first part of this paper presents the approach employed. The second part deals with the formulation of a 1D analytical approach, which, in particularly, allows the computation of the bolt load transfer rates. The third part presents the accurate three-dimensional (3D) finite element (FE) model developed to understand the mechanical behaviour of HBB joints and to show the possible benefits expected in fatigue life.

Improving the fatigue life of aeronautical single-lap bolted joints

477

Figure 1: Reference configuration and labelling

b

d

e

e1

24

12

0.1

e2

1.6

hh

hn

L

7.2

10.08

48

l1

l2

178

p

φ

φh

φn

24

4.8

1.155

6.6

Table 1: Geometrical parameters of the reference configuration in mm

Ef 110000

E1

E2

73100

G

Gf

variable

41353

G1

G2

27481

K 5000

Table 2: Mechanical parameters of the reference configuration in MPa

478

Paroissien, Hoang Ngoc, Bhugaloo and Ducher

APPROACH In the frame of the analysis of bolted joints, the fatigue life is mainly dependent on the stress concentration factor (SCF). The lesser the SCF, the more the fatigue life is. Experimental curves in [9] show with higher stress concentration generated at the fastener holes, the fatigue life becomes lesser. This SCF corresponds to the sum of the bypass load and the transferred load contributions. As a result, in order to increase the fatigue life of a bolted joint, it is sufficient to decrease the bolt load transfer rate, such as performed by increasing the number of the fastener lines. Indeed, the maximum bolt load transfer rate is located at the extreme fastener lines. These extreme fastener lines represent the critical failure sites. When the number of fastener lines increase, the maximum bolt load rate transfer tends to decrease, with the consequence of decrease in critical SCF on the extreme fastener lines and thus fatigue life is increased. The idea to use the HBB joining technology is to decrease the bolt load transfer and the bypass load in order to decrease the SCF, without increasing the number of fastener lines. In order to simplify the analysis, two hypotheses H1 and H2 are considered. H1 consists in assuming that no preload is applied to the bolts, so that no additional stress concentration is set. H2 supposes that the adhesive layer does not fail under fatigue load, whichever the adhesive stiffness under consideration. The validity of H2 is referred to the adhesive strength and stiffness. In this study, the bore holes are found to be the critical sites. The fatigue strength of the adhesive layer is not addressed in this paper. The fatigue performance analysis is then restricted to HBB joints, for which the critical zones are located at bore holes rather than at the overlap ends. The adhesive is considered to be flexible, meaning that it is able both to sustain large deformation and to have low stiffness, so that HBB joints do not fail at the overlap end regions in the adhesive as well as in the adherends.

ONE-DIMENTIONNAL ANALYTICAL ANALYSIS FOR THE BOLT LOAD TRANSFER RATE COMPUTATION Description of the analysis The computation of bolt load transfers is based on an improved approach of the general method developed in details in [7] applied to the reference configuration. The improvement consists in taking into account to the adherend shear stress varying linearly with the adherend thickness.

Improving the fatigue life of aeronautical single-lap bolted joints

479

In this study, existing one-dimensional methods for bonding is combined with bolting analysis, so that the overlap is meshed in three bays linked by both fasteners (Figure 2). Each of the three bonded bay is analysed with the help of the improved Volkersen’s bonding approach [10], while the link between two consecutive bays is performed using the electrical analogy approach employed in the bolted joints analysis [11, 12].

Figure 2: Meshing of the reference configuration

Hypotheses The model is based on the following hypotheses:

the component (adhesive, adherends, fasteners) materials are linear elastic and isotropic;

the adhesive layer thickness is constant all along the overlap;

the adhesive shear stress is constant through the adhesive layer thickness;

the adherend shear stress varies linearly with the adherend thickness.

The 1D beam theory is employed, so that both adherends are simulated by unrestricted bars. Each fastener is classically simulated by a shear spring, the stiffness of which is quoted Cf. Finally, the adhesive layer is simulated by an infinite number of shear springs linking both adherends. In this study, it is

Paroissien, Hoang Ngoc, Bhugaloo and Ducher

480

underlined that the possible adherends bending and the adhesive peel stress are not considered in the presented model. Computation of bolt load transfer rates The index i represents the number of a bay included between 1 and 3. Equilibrium. Considering the global equilibrium of the structure allows to write in the bay i the relationship between the adherend normal forces and the applied load: N 1,i + N 2 ,i = f

(1)

The local equilibrium of both adherends in the bay i provides both following equations (cf. Figure 3): dN 1,i dx dN 2 ,i dx

= −bTi

(2)

= bTi

(3)

Adhesive shear stress definition. A linear shear stress (thus strain) distribution through the thickness adherend is considered in the same way as described in [8]. Since zero shear stress state exists at the top surface of the upper adherend and at the bottom surface of the lower adherend, both adherend shear stresses are expressed as: y1 Ti e1

(4)

⎛ y ⎞ T2 ,i = ⎜⎜ 1 − 2 ⎟⎟Ti e2 ⎠ ⎝

(5)

T1,i =

where y1 et y2 are local coordinates as defined in Figure 3. The longitudinal displacements are then computed from the adherend shear stress by:

u1,i (x , y1 ) = u1,i (x ,0 ) +

e1

∫ 0

2

y1 Ti 1 Ti ⎛ y1 ⎞ ⎜ ⎟ e1 dy1 = u1,i (x ,0 ) + e1 G1 2 G1 ⎜⎝ e1 ⎟⎠

(6)

Improving the fatigue life of aeronautical single-lap bolted joints

481

e2

⎛ y ⎞T u 2 ,i (x , y 2 ) = u 2 ,i (x ,0 ) + ⎜⎜ 1 − 2 ⎟⎟ i dy 2 e2 ⎠ G 2 0⎝

∫

T = u 2 ,i (x ,0 ) + i G2

⎛⎛ y ⎜⎜ 2 ⎜⎜ e ⎝⎝ 2

⎞ 1 ⎛ y2 ⎟⎟ − ⎜⎜ ⎠ 2 ⎝ e2

⎞ ⎟⎟ ⎠

2

(7)

⎞ ⎟e ⎟ 2 ⎠

In order to compute the adhesive shear stress defined by:

Ti =

G (u2 ,i (x ,0 ) − u1,i (x ,e1 )) e

(8)

it is useful to express the longitudinal displacement of the upper adherend as: 2

1 Ti 1 Ti ⎛ y1 ⎞ ⎜ ⎟ e1 u1,i (x , y1 ) = u1,i (x ,e1 ) − e1 + 2 G1 2 G1 ⎜⎝ e1 ⎟⎠

(9)

since (y1=e1 in Eqn. 6): u1,i (x ,e1 ) = u1,i (x ,0 ) +

1 Ti e1 2 G1

(10)

Moreover, the average values of both adherend longitudinal displacements are given by:

u1,i (x ) =

e1

1 1 Ti u1,i (x , y1 )dy1 = u1,i (x ,e1 ) − e1 e1 3 G1

∫

(11)

0

u 2 ,i (x ) =

e2

1 1 Ti u 2 ,i (x , y 2 )dy 2 = u 2 ,i (x ,0 ) + e2 e2 3 G2

∫

(12)

0

Thus the adhesive shear stress can be expressed as: Ti =

1 G (u 2 ,i (x ) − u1,i (x )) 1+ β e

(13)

1 G ⎛ e1 e2 ⎞ ⎜ ⎟ + 3 e ⎜⎝ G1 G2 ⎟⎠

(14)

where:

β=

Paroissien, Hoang Ngoc, Bhugaloo and Ducher

482

y1

1 N1,i+dN1,i

-N1,i

Tibdx

y2 -N2,i

N2,i+dN2,i 2

dx Figure 3: Local equilibrium of both adherends in the bay i

Differential equation. The normal forces in the adherends are written as: e1

N 1,i (x ) = b E1

∫

0 e2

N 2 ,i (x ) = b E 2

∫ 0

∂u1,i ∂x ∂u 2 ,i ∂x

⎛ ∂u1,i ⎞ (x ,e1 ) − 1 e1 dTi ⎟⎟ 3 G1 dx ⎠ ⎝ ∂x

(x , y1 )dy1 = be1 E1 ⎜⎜ (x , y 2 )dy 2

⎛ ∂u 2 ,i e dT ⎞ (x ,0 ) + 1 2 i ⎟⎟ = be 2 E 2 ⎜⎜ 3 G 2 dx ⎠ ⎝ ∂x

(15)

(16)

By differentiation of Eqn. 8, it comes with Eqn. 15 and Eqn. 16: ∂u dTi G ⎛ ∂u 2 ,i = ⎜⎜ (x ,0 ) − 1,i (x ,e1 )⎞⎟⎟ dx e ⎝ ∂x ∂x ⎠ N e G⎛ N 1⎛ e = ⎜ 2 ,i − 1,i − ⎜⎜ 1 + 2 ⎜ e ⎝ be2 E 2 be1 E1 3 ⎝ G1 G2

⎞ dTi ⎟⎟ ⎠ dx

⎞ ⎟ ⎟ ⎠

(17)

or: N ⎞ dTi 1 G ⎛ N 2 ,i ⎜⎜ = − 1,i ⎟⎟ dx 1 + β e ⎝ be2 E 2 be1 E1 ⎠

(18)

Improving the fatigue life of aeronautical single-lap bolted joints

483

The introduction of Eqn. 1 and Eqn. 3 in Eqn. 18 provides the following second order linear differential equation: d 2 N i ,2 dx

2

− η 2 N i ,2 = −

χ

G 1 f 1 + β e e2 E 2

(19)

where:

η2 =

1 G⎛ 1 1 ⎜⎜ + 1 + β e ⎝ e1 E1 e2 E 2 e E χ= 2 2 e1 E1

⎞ 1+ χ G 1 ⎟⎟ = ⎠ 1 + β e e2 E 2

(20) (21)

Differential equation solution and boundary conditions. The solution of Eqn. 19 is under the shape: N i ,2 = Ai e −ηx + Bi eηx − λf

(22)

where Ai and Bi are integration constants and:

λ=−

χ

(23)

1+ χ

Thus, using Eqn. 3, the adhesive shear stress is given by:

Ti =

η b

(− A e η

− x

i

+ Bi eηx

)

(24)

For each of the three bays, two integration constants have to be found. As a result six boundary conditions (A1, B1, A2, B2, A3, B3) are required. The first boundary condition corresponds to a zero load at x=0, which leads to: N 2 ,1 (0 ) = 0 ⇔ A1 + B1 = λf

(25)

The second boundary condition corresponds to complete load transfer at x=L, which leads to: N 2 ,3 (L ) = f ⇔ e −ηL A3 + eηL B3 = (1 + λ ) f

(26)

Paroissien, Hoang Ngoc, Bhugaloo and Ducher

484

Two additional boundary conditions are given by considering the continuity of the shear stress at each of both fasteners:

T1 (d ) = T2 (d ) ⇔ e −ηd A1 − eηd B1 − e −ηd A2 + eηd B2 = 0

T2 (L − d ) = T3 (L − d ) ⇔ e

−η ( L −d )

η (L−d )

A2 − e

B2 − e

(27)

−η ( L −d )

η ( L −d )

A3 + e

B3 = 0 (28)

Both last boundary conditions are obtained by writing the load transfer at each of both fasteners: N 2 ,2 (d ) = N 2 ,1 (d ) + τ 1 f

(29)

N 2 ,3 (L − d ) = N 2 ,2 (L − d ) + τ 2 f

(30)

But, the bolt load transfers are given by:

τ 1 f = C f (u 2 ,1 (d ) − u1,1 (d ))

(31)

τ 2 f = C f (u 2 ,2 (L − d ) − u1,2 (L − d ))

(32)

By using the definition of the adhesive shear stress in Eqn. 13, the bolt load transfers are expressed under the shape:

(

τ 1 f = C f (1 + β ) T1 (d ) = κ − A1e −ηd + B1eηd

)

e G e τ 2 f = C f (1 + β ) T2 (L − d ) = κ − A2 e −η (L−d ) + B2 eη (L−d ) G

(

(33)

)

(34)

where:

κ = (1 + β )

eη Cf Gb

(35)

Both last boundary conditions are explicitly given using Eqn. 33 (Eqn. 34) in Eqn. 29 (Eqn. 30):

qe −ηd A1 + reηd B1 − e −ηd A2 − eηd B2 = 0 qe

−η ( L −d )

η ( L −d )

A2 + re

B2 − e

−η ( L −d )

(36) η ( L −d )

A3 − e

B3 = 0

(37)

where:

q = 1−κ r = 1+κ

(38) (39)

Improving the fatigue life of aeronautical single-lap bolted joints

485

Finally, the integration constants are found by solving the following linear system: ⎛ 1 ⎜ −ηd ⎜e ⎜ qe −ηd ⎜ ⎜ 0 ⎜ ⎜ 0 ⎜ 0 ⎝

1

0

ηd

−e

−e − e −ηd

ηd

re 0 0 0

0

−ηd

e −η (L −d )

qe −η (L −d ) 0

0

ηd

e − eηd − eη (L −d )

0 0

− e −η (L −d ) − e −η (L −d )

reη (L −d ) 0

⎞⎛ A1 ⎞ ⎛ λf ⎞ ⎟⎜ ⎟ ⎜ ⎟ 0 ⎟⎜ B1 ⎟ ⎜ 0 ⎟ ⎟⎜ A ⎟ ⎜ 0 ⎟ 0 ⎟⎜ 2 ⎟ = ⎜ ⎟ η (L−d ) ⎟⎜ B2 ⎟ ⎜ 0 ⎟ e ⎟⎜ ⎟ ⎜ ⎟ − eη (L −d ) ⎟⎜ A3 ⎟ ⎜ 0 ⎟ eηL ⎟⎠⎜⎝ B3 ⎟⎠ ⎜⎝ (1 + λ ) f ⎟⎠ (40) 0

e −ηL

Resolution of the linear system for a balanced reference configuration. The system is solved in the case of a balanced reference configuration, leading to χ=1 and thus λ=-0.5. The idealization then performed on the balanced reference configuration (cf. Figure 2) implies that both bolt load transfer rates are equal. This analytical model presented does not take into account possible imperfect geometrical and mechanical symmetry conditions. This additional condition allows for simplifying the resolution of the linear system, since only A1 and B1 have to be found, and provides the following equation:

− e −ηd A1 + eηd B1 = −e −η (L−d ) A2 + eη (L−d ) B2

(41)

The linear system is then simplified in the following one: ⎛ 1 ⎜ −ηd ⎜e ⎜ qe −ηd ⎜ ⎜ e −ηd ⎝

1 − eηd reηd − eηd

0 − e −ηd − e −ηd − e −η (L −d )

⎞⎛ A1 ⎞ ⎛ − 0.5 f ⎟⎜ ⎟ ⎜ ⎟⎜ B1 ⎟ ⎜ 0 ⎟⎜ A ⎟ = ⎜ 0 ⎟⎜ 2 ⎟ ⎜ eη (L −d ) ⎟⎠⎜⎝ B2 ⎟⎠ ⎜⎝ 0 0 e eηd ηd

⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠

(42)

Then, by linear combinations, it comes: 1 ⎛ ⎜ ( )e −ηd q 1 + ⎜ ⎜ (q − 1)e −ηd ⎜ ⎜ 0 ⎝

1

(r − 1)eηd (r + 1)eηd 0

0 − 2e −ηd 0

e −ηd − e −η (L −d )

⎞⎛ A1 ⎞ ⎛ − 0.5 f ⎟⎜ ⎟ ⎜ ⎟⎜ B1 ⎟ ⎜ 0 ⎟⎜ A ⎟ = ⎜ 0 ⎟⎜ 2 ⎟ ⎜ + eη (L −d ) ⎟⎠⎜⎝ B2 ⎟⎠ ⎜⎝ 0

0 0 − 2eηd − eηd

⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠

(43)

Finally, the six by six linear systems is reduced to the two by two linear system:

Paroissien, Hoang Ngoc, Bhugaloo and Ducher

486

⎛ 1 ⎜⎜ ~ −ηd ⎝ qe

1 ⎞⎛ A1 ⎞ ⎛ − 0.5 f ⎟⎜ ⎟ = ⎜ ~ r eηd ⎟⎠⎜⎝ B1 ⎟⎠ ⎜⎝ 0

⎞ ⎟⎟ ⎠

(44)

where: q~ = 1 − cosh(η (L − 2 d )) + q sinh(η (L − 2 d )) ~ r = −1 + cosh(η (L − 2d )) − r sinh(η (L − 2 d ))

(45) (46)

Finally, A1 and B1 are given by: ~ r eηd 1 f 2 q~e −ηd − ~ r eηd q~e −ηd 1 B1 = ~ ηd ~ −ηd f 2 r e − qe A1 =

(47) (48)

Thus, by introducing Eqn. 47 and Eqn. 48 in Eqn. 33, the bolt load transfer is provided:

⎛

~ ~

⎞

1 r +q τ 2 = τ 1 = κ ⎜⎜ ~ ηd ~ −ηd ⎟⎟ 2 ⎝ r e − qe ⎠

(49)

Application on the reference configuration The above mentioned formula (Eqn. 49) of the bolt load transfer rate is applied to the reference configuration for various values of the adhesive Coulomb’s modulus. The result is plotted in Figure 4 with a fastener stiffness equal to 50000 N.mm-1. Results show that the bolt load transfer rate tends to zero when the adhesive stiffness increases, so that at high adhesive modulus the fasteners do not participate in the load transfer. The bonded joint and the bolted joint could be seen as both limit cases of the HBB joint.

Improving the fatigue life of aeronautical single-lap bolted joints

487

50 analytical model

45 bolt load transfer rate in %

40 35 30 25 20 15 10 5 0 0.01

0.1

1

10

100

1000

adhesive Coulomb's modulus in MPa

Figure 4: Bolt load transfer rate as a function of the adhesive Coulomb’s modulus for the reference configuration

THREE-DIMENTIONNAL FINITE ELEMENT ANALYSIS Description of the analysis A 3D FE analysis is performed on the reference configuration in order to interpret SCF as a function of the bolt load transfer rate; This analysis aims at checking the possibility of fatigue life improvement of bolted joints thanks to the HBB joining technology. Presentation of the model The developed model is fully detailed in [13] and is based on the model as presented in [8]. A brief description is given hereafter. The FE model is developed using three-dimensional brick elements by using the SAMCEF FE code [14]. More precisely, the adherends are modelled with eight-node element (24 DoF), while the adhesive layer and the fasteners are modelled with twenty-node elements (60 DoF). The mesh around the holes and at the overlap ends is refined, in order to ensure the convergence of computations (see Figure 5). An isotropic linear elastic behaviour of adherends and fasteners is considered, so that the applied stress is equal to 80 MPa. The adhesive layer is considered to be

488

Paroissien, Hoang Ngoc, Bhugaloo and Ducher

non linear elastic and quasi-incompressible; it is simulated by a two-coefficient (C10 and C01) Mooney-Rivlin hyperelastic potential with a bulk modulus of 5000 MPa. The selective integration is employed to overcome the incompressibility problem. Only a half of the joint is modelled using the symmetrical boundary condition. One end is clamped whereas the opposite end is free to move in the longitudinal direction only. No preload is applied to the fastener, and no clearance between the fasteners and the adherends is considered. The contact between the adherends and the fasteners is considered without friction. The geometrical non linearity is considered in the analyses. The mechanical and geometrical parameters are given in Table 1 and Table 2.

Figure 5: View of the mesh of the single-lap HBB joint

Measurement of the SCF as a function of the bolt load transfer rate An equivalent adhesive Coulomb’s modulus is defined as a function of both hyperelastic material parameters, C10 and C01, as:

G eq = 2(C10 + C 01 )

(50)

In order to obtain different values of the bolt load transfer rate at the fixed geometry, the adhesive equivalent Coulomb’s modulus is varied between 0.1 and 120.6 MPa with respect to C01=1.6C10. The bolt load transfer is numerically

Improving the fatigue life of aeronautical single-lap bolted joints

489

measured according to a special method developed in [4] and validated in [5], which consists in summing the nodal forces at the bolt mid-plane. The stress concentration is computed as the ratio between the first principal stress, which is considered to be responsible for the initial crack, at the critical locations with respective applied stress. The curves of the SCF as a function of the bolt load transfer rate measured for each of both bolts are given in Figure 6. It appears that the stress concentration factor varies almost linearly and increasingly with an increasing bolt load transfer rate, in a wide range of bolt load transfer rate included between around 5% and 45%. By considering that the HBB joints are bolted joints for which all the applied load is transferred by the bolts, and with application of hypothesis H2, results show that the fatigue life performance of HBB joints are higher than bolted joints.

3.6 Bolt 1

stress concentration factorrrrr

3.2

Bolt 2

2.8

2.4

2

1.6

1.2 0

5

10

15

20

25

30

35

40

45

50

bolt load transfer rate in %

Figure 6: Stress concentration factor as a function of the bolt load transfer rate

CONCLUSION In this study, with a three-dimensional FE analysis, a linear relation between bolt load transfer and SCF is demonstrated. Results also show improvement of fatigue life using HBB technology without increasing fastener lines. Here an onedimensional analytical approach is presented and applied for a particular case, in

Paroissien, Hoang Ngoc, Bhugaloo and Ducher

490

total load transferred by both fasteners in %azer

order to compute the bolt load transfer rate of HBB joints. The fastener stiffness could be considered as a parameter useful to calibrate the 1D model [5]. This calibration could be performed by finding the value of the fastener stiffness which minimizes the sum of the quadratic difference between the total load transferred by both fasteners computed by the analytical and the numerical approaches. On the reference configuration, the fastener stiffness is found equal to 29302 N.mm-1 (see Figure 7). This conclusion is sustainable under the hypothesis of critical sites located at the fastener lines and without considering any preload of fasteners.

100 1D analytical model 3D FE model

90 80 70 60 50 40 30 20 10 0 0

25

50

75

100

125

equivalent adhesive Coulomb's modulus in MPa

Figure 7: Comparison of the total load transferred by the fasteners computed by the 1D analytical model and the 3D FE model with Cf = 29302 N.mm-1

NOMENCLATURE Aj b Bj Cf C01 C10 DoF e e1

integration constant number j=1..3 overlap width or transversal pitch integration constant number j=1..3 fastener stiffness adhesive material parameter adhesive material parameter degree of freedom adhesive thickness upper adherend thickness

[N] [mm] [N] [N.mm-1] [MPa] [MPa] [mm] [mm]

Improving the fatigue life of aeronautical single-lap bolted joints

491

e2 E1 E2 f FE G G1 G2 HBB hh hn l1 l2 K Kt L N1,i N2,i p q q~ r ~ r SCF Ti T1,i T2,i u1,i u2,i u 1,i

lower adherend thickness upper adherend Young’s modulus lower adherend Young’s modulus in-plane applied force finite element adhesive Coulomb’s modulus upper adherend Coulomb’s modulus lower adherend Coulomb’s modulus hybrid (bolted/bonded) height of fastener head height of fastener nut upper outside the overlap length lower outside the overlap length bulk modulus of the adhesive stress concentration factor overlap length upper adherend normal force in the bay i=1..3 lower adherend normal force in the bay i=1..3 longitudinal pitch characteristic parameter characteristic parameter characteristic parameter characteristic parameter stress concentration factor adhesive shear stress in the bay i=1..3 upper adherend shear stress in the bay i=1..3 lower adherend shear stress in the bay i=1..3 upper adherend longitudinal displacement in the bay i=1..3 lower adherend longitudinal displacement in the bay i=1..3 average value of u1,i through e1 in the bay i=1..3

[mm] [MPa] [MPa] [N]

u 2,i

average value of u2,i through e2 in the bay i=1..3

[N]

x y y1 y2 z β χ η κ λ τ τi

structural rectangular Cartesian x-coordinate structural rectangular Cartesian y-coordinate local rectangular Cartesian y-coordinate local rectangular Cartesian y-coordinate structural rectangular Cartesian z-coordinate characteristic parameter characteristic parameter characteristic parameter characteristic parameter characteristic parameter bolt load transfer rate bolt load transfer rate of the fastener number i=1,2

[mm] [mm] [mm] [mm] [mm] [-] [-] [mm-1] [-] [-] [-] [-]

[MPa] [MPa] [MPa] [mm] [mm] [mm] [mm] [MPa] [-] [mm] [N] [N] [mm] [-] [-] [-] [-] [MPa] [MPa] [MPa] [N] [N] [N]

492

Paroissien, Hoang Ngoc, Bhugaloo and Ducher

ACKNOWLEDGMENTS The authors gratefully acknowledge the engineers of the Methods and Research team of SOGETI High Tech Fatigue & Damage Tolerance Department in Toulouse for their support and advice, in the frame of the development of JoSAT (Joint Stress Analysis Tool) internal research program.

REFERENCES [1] Imanaka, M., Haraga, K., Nishikawa, T. (1995), Journal of Adhesion, vol. 49, pp. 197-209 [2] Fu, M., Mallick, P.K. (2001), International Journal of Adhesion and Adhesives, vol. 21, pp. 145-159 [3] Kelly, G. (2006), Composites Structures, vol. 73, pp. 119-129 [4] Kelly, G. (2004), PhD Dissertation, KTH Aeronautical and Vehicle Engineering, Stockholm, SW [5] Paroissien, E. (2006), PhD Dissertation (in French), 2006TOU30133, Institut de Génie Mécanique de Toulouse, FR [6] Hart-Smith, L.J. (1982), Technical Report, AFWAL-TR-81-3154 vol. 1, Douglas Aircraft Company, Long-Beach, CA [7] Paroissien, E., Sartor, M., Huet, J. (2007), Trends and Recent Advances in Integrated Design and Manufacturing in Mechanical Engineering II, Springer Eds., pp. 95-110 [8] Paroissien, E., Sartor, M., Huet, J., Lachaud, F (2007), AIAA Journal of Aircraft, vol. 44, n. 2, pp. 573-582 [9] Niu, M.C.-Y. (1999), Airframe Structural Design – Stress Analysis and Sizing, 2nd Eds., Conmilit Press Ltd., Hong Kong [10] Tsaï, M.Y., Oplinger, D.W., Morton, J. (1998), International Journal of Solids Structures, vol. 35, n. 12, pp. 1163-1185 [11] Tate, M.B., Rosenfeld N.J. (1946), Technical Note, n. 1051, National Advisory Committee for Aeronautics Washington, DC [12] Ross, R.D. (1947), Technical Note, n. 1281, National Advisory Committee for Aeronautics Washington, DC [13] Hoang Ngoc, C.T. (2008), Technical Note (in French), PE3/1190NT08/EP, Sogeti High Tech, Blagnac, FR [14] SAMCEF, Ver. 11.1-04, 31st August 2006, Samtech Group, Liège, BE

25th ICAF Symposium – Rotterdam, 27–29 May 2009

AN EXPERIMENTAL APPROACH TO INVESTIGATE DETAILED FAILURE MECHANISMS IN FIBRE METAL LAMINATES Riccardo Rodi1, René Alderliesten2 and Rinze Benedictus2 1

Material Innovation Institute -M2i-, Mekelweg 2, 2628CD, Delft, Netherlands 2 Delft University of Technology, Faculty of Aerospace Materials, Kluyverweg 1 2629 HS, Delft, Netherlands

Abstract: This paper propose an experimental approach, based on digital image correlation, which enables detailed quantitative description of the most important failure mechanisms occurring in Fibre Metal Laminates during fatigue and static load. Digital image correlation provided measurement of the crack tip plasticity, fatigue delamination shape and fibre bridging in different FMLs. Digital image correlation showed to be the most versatile and suitable non-destructive and in-situ technique for detailed full-field strain measurements in FMLs.

INTRODUCTION Fibre Metal Laminates (FMLs) have been developed in the past to increase the fatigue resistance of laminated aluminium structures by adding fibres in the bond line. The fibres are insensitive to the occurring fatigue stresses in FMLs and bridge the fatigue cracks in the metal layers by restraining the crack opening. This results in a complex mechanism of crack growth in the metal layers and delamination growth at the interface between metal and fibre layers that, if in optimal balance, results in the excellent fatigue characteristics for which FMLs are known [1].

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 493–512. © Springer Science+Business Media B.V. 2009

494

R. Rodi, R.C. Alderliesten, R. Benedictus

Figure 1 Typical lay-up for a cross-ply Fibre Metal Laminate [8]. Quite an amount of research [3-10] has been carried out in the past to investigate both the residual strength and fatigue behaviour of Fibre Metal Laminates. It has been pointed out that the composite nature of FMLs (metal/fibre combination) increases the complexity of the problem because more constituents work together to redistribute the applied load, especially when large scale yielding develops in the metal layers. Understanding of the behaviour of structural materials under both static and fatigue load is important when assessing the damage tolerance of the structural design of an aircraft. Understanding the failure mechanisms is thus of primary importance. In the last two decades several measurement approaches were adopted in order to investigate the complex fracture mechanisms involved in the failure sequence of FMLs [3-10]. The significant amount of knowledge about the behaviour of FMLs is the result of intensive experimental investigations and measurements carried out using different measurement techniques [9], such as C-scan, chemical etching, shearography, strain gauges, etc. All these techniques provided information regarding fatigue delamination shape and growth, crack propagation, plastic zone extension, etc. This paper propose an experimental approach based on digital image correlation (DIC) which enables detailed quantitative description of the most important failure mechanisms occurring in different FMLs structures. Using DIC it is possible to measure the metal crack tip plastic zone, fatigue delamination shape and fibre layers deformation in-situ and in a non-destructive manner.

FAILURE MECHANISMS IN FMLs The failure process in FMLs under static load involves complex phenomena which increase the difficulty of analysis. Vermeeren [3] and De Vries [4] provided a detailed description of the quasi/static failure sequence in Glare. They subdivided the failure process in four main mechanisms:

An experimental approach to investigate detailed failure mechanisms in FMLs

495

Crack tip blunting and initiation in the metal layer Because of the presence of a through the thickness crack, or saw-cut, the stress increases at the crack tip. An increment of the applied load results in the formation of a local plastic zone in front of the crack tip in the metal layer, see Figure 2-b, with subsequently crack tip blunting [4]. Crack initiation occurs when locally the strain at failure of the metal is reached, see Figure 2-a. Static delamination Both metal and fibre layers are equally strained in the far field, this means that the occurrence of plasticity induces a complex shear stress state at the metal/fibre interfaces in front of the crack tip. Delamination can occur when the shear stress at these interfaces reaches a critical value. Vermeeren [3] describes the shear stress system as the superposition of the shear stress due to the blunting of the crack tip and the shear stresses due to the plasticity ahead of the crack tip. Stable crack extension in the metal layer When the crack extends some millimetres, the fracture mechanisms change. De Vries [4] relates this change to the Crack Tip Opening Angle (CTOA). Within his study, it has been demonstrated that the critical CTOA, which represents the amount of crack tip blunting, reduces for large crack extension. Therefore, the shear stresses resulting from blunting reduce. Under these conditions static delamination can hardly occur. Local fibre failure and dynamic delamination Fibre failure can occur when locally the ultimate strain of the fibres is reached. This generally happens in those areas where high stress concentrations are present, e.g. the delamination boundary or in front of the crack tip. Vermeeren [3] explained that fibre failure can occur either after static delamination or without static delamination. In both cases, the local fibre failure induces stress redistribution towards the still intact part of the specimen. Once fibre failure occurs, the large amount of elastic energy stored in the fibres is released into the matrix causing delamination along the rich-resin layer at the metal/fibre interface. This behaviour is defined as dynamic delamination, an example is provided in Figure 2-d. The fatigue behaviour of FMLs has been extensively investigated by several researchers [5-8]. For fatigue under constant amplitude loading, three main mechanisms can be pointed out: Metal crack growth and fibre bridging As result of the cycling load, fatigue crack propagation occurs in the metal layers. The intact fibres along the wake of the crack act as second load path carrying part of the load, reducing the stress intensity at the metal crack tip

R. Rodi, R.C. Alderliesten, R. Benedictus

496

Delamination at the metal/fibre interface The cyclic shear stresses at the metal/fibre interface induce delamination growth. The amount of delamination is related to the strength of the adhesion between metal and fibre layer and to the applied load. An example is showed in Figure 2-c. Although the metal cracking and delamination mechanisms under quasi-static loading and in fatigue loading are different from each other, the experimental approach proposed in this paper applies to both damage scenarios.

Fatigue crack

Crack tip blunted

Static crack extension a) Crack tip blunting and propagation

b) Crack tip plastic zone

Dynamic delamination

Fibres failure

c) Fatigue delamination

d) Dynamic delamination and fibre failure

Figure 2 Illustration of the most important failure mechanisms within FMLs

AVAILABLE MEASUREMENT TECHINIQUES FOR FMLs Because of the hybrid nature of FMLs, several empirical approaches are used to characterize the main failure mechanisms. In this section a brief survey and comparison of the most common measurement techniques is provided. Fatigue delamination shape investigation The investigation of delamination shape can be performed in several ways. Two main techniques are ultrasonic C-scanning and chemical etching of the metal layers.

An experimental approach to investigate detailed failure mechanisms in FMLs

497

Ultrasonic C-scanning is a non-destructive inspection technique in which a short pulse of ultrasonic energy is transmitted through a sample. Measurement of the received pulse indicates the attenuation of the transmitted pulse caused by defects. For fatigue delamination investigation purpose, the specimen can be taken from the test frame and C-scanned. After, it can be inserted in the testing frame again to continue the crack propagation [8]. Chemical etching is a semi-destructive chemical process where the outer metal layer is removed at a controlled speed. Once the fatigue crack propagation tests is finished, the aluminium layers can be etched away and the delamination shapes can be evaluated [4,6,8]. An example of result obtained with C-scan and chemical etching is illustrated in Figure 3

Figure 3 Example of fatigue delamination shape using C-scan (left) and using chemical etching (right) Crack tip strain field and plasticity The crack tip strain field can be evaluated in several ways. Possible techniques are shearography, strain gauges measurement or digital image correlation. Shearography is a method based on laser interferometry that measures the displacement derivatives of a material surface, which can be transformed into strain measurements. With this technique, the object is illuminated with a coherent monochromatic light (laser) to obtain interference. If the object surface has an average roughness larger than the wavelength of the illuminating laser, the reflected image will have a typical granular appearance that is called speckle. By comparing the deformed and undeformed speckles, it is possible to obtain an image containing fringes that represent equal deformation [4], see Figure 4-a. Shearography instrumentation includes automated processing of fringes to generate real time strain maps [11] and to automatically detect defects [12]. Strain gauges are commonly used for local strain measurement. The most common type of strain gauge consists of an insulating flexible backing which supports a metallic foil pattern. The gauge is attached to the object by a suitable adhesive. As the object is deformed, the foil is deformed, causing its electrical resistance to

498

R. Rodi, R.C. Alderliesten, R. Benedictus

change. This resistance change is related to the strain by the quantity known as the “gauge factor”. Digital image correlation is a technique which provides displacements and displacement gradients (strain) on an object surface [14-15]. The modern digital image correlation uses digital grid points which mark specific positions (pixels) on the specimen surface when no deformations are present (source image). Those positions are traced back in each image (target image) captured during the test execution. The post-processing of the obtained displacement fields provides the relative strain field, see an example in Figure 4-b.

(a) (b) Figure 4 Example of crack tip strain field measured using shearography [4] (a) and DIC [19] (b) Fibre layer deformation Fibre Bragg Grating (FBG) is an optical strain sensor that can be formed within an optical fibre by modulation of the refractive index of the fibre core. An FBG acts as a band-stop wavelength filter in transmission and a band-pass wavelength filter in reflection [16]. The principle of a strain sensor based on a FBG is that the wavelength of the reflected light shifts when the object containing the sensor is deformed. This wavelength shift is measured and the strain can be calculated from the sensor calibration data. The wavelength shift is proportional to the axial strain applied to the FBG sensor. Embedded FBG sensors are usually introduced during manufacturing. An example of application of this promising measurement is in composite materials [17,18], where optical fibres containing FBGs are introduced during the manufacturing of the composite. They then become a permanent sensor within the material, which can be used to study and monitor locally the strain. Table 1 provides a comparison of the mentioned measurement techniques. The comparison is based on few simple characteristics: the non-destructive characteristic, the possibility of an in-situ application and full-field application. In addition, the measurement process and equipment complexity is also compared.

An experimental approach to investigate detailed failure mechanisms in FMLs

499

Table 1 Comparison of measurement techniques Type of measurement

Complexity of the process

Complexity of the equipment

Delamination / disbond

low

high

9

Delamination/ fibre failure

low

low

9

9

Strain

high

high

9

9

8

Strain

low

low

Fibre Bragggrating

-

9

8

Strain

medium

medium

DIC

9

9

9

Strain

medium

medium

NDI

In-situ application

Full-field application

C-scan

9

8

9

Chemical etching

8

8

Shearography

9

Strain gauges

Table 1 shows that the DIC appears to be the most versatile and suitable nondestructive and in-situ technique for detailed full-field strain measurements in FMLs. Moreover, as will be explained in the next section, DIC can also provide information on the fatigue delamination shape and the magnitude of the bridging stress.

EXPERIMENTAL APPROACH Macro-scale failure in structural materials generates redistribution of stress and strain. DIC has been employed to investigate different failure mechanisms in FMLs, such as crack tip plasticity, fatigue delamination and bridging stress, by measuring the strain redistribution due to local failures. Several tests have been performed on CCT specimen made of different types of FMLs to explore the applicability and limitations of DIC. The experiments consisted of three types of test: Test type 1: Investigating crack tip plasticity Two crack configurations were considered: saw-cut crack, representing an accidental damage where no bridging fibres are present, and fatigue crack where the bridging fibres in the wake of the crack act as second load path. The measurement of the strain field was performed under static load using DIC to study the effect of the bridging fibres on the metal crack tip plasticity. Test type 2: Investigating the fatigue delamination shape With DIC it is possible to visualize the fatigue delamination shape by measuring the difference in strain between the delaminated and the non-delaminated area while the specimen is under static tension. Several FMLs were tested to explore the applicability of such technique on different materials.

500

R. Rodi, R.C. Alderliesten, R. Benedictus

Test type 3: Investigating the deformation of fibre layers To measure the deformation of the fibres layer, inverted lay-ups were manufactured containing prepreg layers on the outside. The strain measured with DIC provided information on the fibre bridging stress. To be able to relate this inverted lay-up to the standard lay-up (FML 2-2/1-0.4) the MVF was kept the same. Figure 5 illustrates both the standard and inverted lay-up.

Figure 5 Illustration of the ‘inverted lay-up’ and the ‘standard’ lay-up

Test execution A digital camera system was used to perform the digital image acquisition of the specimen surface of interest. Figure 6 illustrates schematically the load sequences used to perform the three test types described above. Here the triangular symbols denote the image acquisition. The images were captured and stored while the specimen was under static load. For test type 1, two load sequences were used: quasi-static load increments for the saw-cut configuration, and fatigue loading plus quasi-static load increments for the fatigue crack configuration. An image of the test set-up and specimen geometry is provided in Figure 7

An experimental approach to investigate detailed failure mechanisms in FMLs

Force

501

Force control ramp-up

Force control ramp-up

Fatigue crack growth

TEST TYPE 1

Delamination growth by fatigue cycles

TEST TYPE 2

Force control ramp-up

Fatigue crack growth and delamination growth

TEST TYPE 3 Image acquisition

Time

Figure 6 Load sequences used during the tests 1

Ø2

Digital cameras 20 290

SAW-CUT Ø2

140

5 20 FATIGUE CRACK

Resolution=1600 by 1200 px Scanned area = 20 by 15 mm ÷ 42 by 31.5 mm, depending on the test type Ratio mm/px = 0.0125 ÷ 0.0262, depending on the test type

Figure 7 Test set-up (left) and specimen geometry (right). All dimensions in mm

R. Rodi, R.C. Alderliesten, R. Benedictus

502

In order to reach fixed load/stress levels, the quasi-static load increments were performed under force control, while the images where captured under displacement control. This procedure enabled control the crack tip deformation avoiding crack propagation, especially when high load were applied. Digital image correlation method Figure 8 schematically illustrates the correlation procedure performed using the DIC tool [14]. Post-processing of the obtained displacement fields provides the relative strain field. In order to obtain a good correlation, a speckle pattern had to be applied. As described in [14, 19] the importance of such speckle pattern is twofold: first, it ensures a high correlation, increasing thus the accuracy, and second it prevents changes in the surface light reflection when high deformation occurs.

Strain measurement

y Detail of the “source” image

x

“Target” image with strain εyy visualization

Figure 8 Schematic overview of the correlation process for one pixel and strain measurement [14,19]

An experimental approach to investigate detailed failure mechanisms in FMLs

503

EXPERIMENTAL RESULTS AND DISCUSSION The strain fields data obtained in the experiments contain information on plasticity in the metal layer, the shape of fatigue delamination and the stress distribution in the fibre layers. The experimental results are presented for each of these categories individually. Test type1: Investigating crack tip plasticity Two specimens made by Glare 3-3/2-0.4, one containing a saw-cut and the other a fatigue crack, were tested. The fatigue crack was obtained by applying fatigue load with σmax=120 MPa and R=0.05. In both cases the length of the crack was 20mm, as illustrated in Figure 7. Both configurations were quasi-statically loaded and the images were captured at several stress levels, following the approach illustrated in Figure 6. An example of strain measurement is showed in Figure 9, where the strain fields of the saw-cut (a) and fatigue crack (b) configurations are compared to each other while a static load of 180 MPa was applied. The bridging fibres in the wake of the fatigue crack act as a second-path load attracting load and reducing the stress intensity at the crack tip of the metal layer.

(a) (b) Figure 9 Comparison of the strain field ahead of a 10 mm saw-cut (a) and a 10 mm fatigue crack (b) in Glare 3-3/2-0.4 under an applied stress of 180 MPa (Scale in %) The plastic zone, defined as the part of the strain field with strain values equal to or beyond the yield strain (ε0,2), has been evaluated form the measurement data. An example is provided in Figure 10, where the plastic zone of both saw-cut and fatigue crack configurations are compared under the same applied load. The size of the plastic zone ahead of the fatigue crack is about 80% smaller in x-direction and 65% in y-direction compared to the saw-cut configuration.

R. Rodi, R.C. Alderliesten, R. Benedictus

504 10 8

Fatigue crack configuration Saw-cut configuration

6 4 Y [mm]

2 0 -2 -4 -6 -8 -10 5

7

9

11

13

15

17 19 X [mm]

21

23

25

27

29

Figure 10 Comparison of the plastic zone ahead of a 10 mm saw-cut and a 10 mm fatigue crack in Glare 3-3/2-0.4 under an applied stress of 180 MPa From the images represented in Figure 9 it is possible to obtain the strain and stress gradients in front of the crack tip, see Figure 11 and Figure 12. Due to the bridging fibres, the strain at the tip of the fatigue crack configuration is 80% less than in the saw-cut configuration. The strain in front of the crack can exceed the strain at failure of the fibre prepreg when a high load is applied. In the example shown in Figure 11, the strain measured on the metal surface exceeded the failure strain of the S2-glass fibres in the area in front of the saw-cut. This is a possible indication either of static delamination or local fibre failure. 0.1 Fatigue crack configuration

Saw-cut configuration

Strain y-direction [-]

0.08

S2-glass fibre failure at 4.5%

Possible fibres failure in front of the crack tip

0.06

0.04

0.02

0 0

5

10

15

20

25

Position in front of the crack tip [mm]

Figure 11 Comparison of the strain gradients ahead of a 10 mm saw-cut and a 10 mm fatigue crack in Glare 3-3/2-0.4 under an applied stress of 180 MPa

An experimental approach to investigate detailed failure mechanisms in FMLs

505

500

Fatigue crack configuration 450

Saw-cut configuration 2024-T3 yield stress

Stress y-direction [MPa]

400 350 300 250 200 150 100 50 0 0

5

10

15 20 25 Position in front of the crack tip [mm]

30

35

40

Figure 12 Comparison of the stress gradients ahead of a 10 mm saw-cut and a 10 mm fatigue crack in Glare 3-3/2-0.4 under an applied stress of 180 MPa Figure 12 shows the stress gradients ahead of the crack tip. These gradients are obtained by relating the measured strain to the stress with the stress-strain curve of the metal, including the residual stress. The size of the plastic zone due to the applied stress is defined as the distance from the crack tip where the stress is larger or equal to the yield stress of the metal. Figure 13 illustrates the comparison between the Irwin’s and Dugdale’s plastic zone corrections, equations 1 and 2, and the experimental measurements under different applied loads.

rIrwin rDugdale

a ⎛σ = ⎜ metal 2 ⎜⎝ σ ys

⎞ ⎟⎟ ⎠

a ⎛ πσ = ⎜ metal 2 ⎜⎝ 2σ ys

2

(1)

⎞ ⎟⎟ ⎠

2

(2)

Both predictions and measurements account for the residual stress in the metal layer. The predictions show good agreement with the measurements only for the saw-cut configuration. Both Irwin’s and Dugdale’s corrections enable good results as long the plastic zone size is no larger than the crack size (rp ≤ a). On the other hand, for the case of fatigue crack, both analytical tools fail in predicting the correct size of plastic zone. A reformulation of both equations based on the actual stress intensity factor, accounting for the fibre bridging, could provide better results.

R. Rodi, R.C. Alderliesten, R. Benedictus

506 40 Irwin's correction

35

Dugdale correction

30

DIC fatigue crack

Plastic zone size rp [mm]

DIC Saw-cut

25 20 15 10 5 0 50

100

150 200 250 Applied stress in the laminate [MPa]

300

Figure 13 Comparison of the plastic zone extensions ahead of a 10 mm saw-cut and a 10 mm fatigue crack in Glare 3-3/2-0.4 between measurements and analytical predictions Recently, the described approach has been used in variable amplitude fatigue tests in Glare [20], investigating the effect of over and under loads on the plastic zone size and delamination shape.

Test type 2: Investigating the fatigue delamination shape Fatigue tests were performed on several FMLs to explore the potentiality of DIC in measuring the fatigue delamination shape. Several images were captured at different crack lengths by applying statically the maximum fatigue load.

delaminated area

delaminated area 23.5 mm

Figure 14 Fatigue delamination shape obtained with DIC in a FML 3-3/2-0.4 2024T3/Zylon, σmax_fatigue =120 MPa, R=0.05, a=23.5mm (scale in %)

An experimental approach to investigate detailed failure mechanisms in FMLs

507

Figure 14 shows an example of fatigue delamination in a 2024-T3/Zylon laminate obtained by measuring the strain on the metal surface. The applied load induces a full opening of the crack and a complete elongation of the intact fibres in the wake of the crack, which attract load from the cracked metal layers. Thus, the disbonded part of the metal layer strains less than the still bonded part Since the load is transferred to the bridging fibres gradually, in the vicinity of the actual delamination front a transition area is present; see Figure 14 and Figure 16. The extension of the transition area is related with the applied load and with the relative stiffness of the constituents. The actual delamination front lies within this area and in Figure 14, denoted by white dotted lines, is visible and distinguishable from the rest of the specimen. Monitoring the fatigue delamination growth during a fatigue test on a FML is one of the main issues to assess the fatigue behaviour of such hybrid materials. Fatigue cycles with a maximum stress of 150 MPa and R=0.05 were applied to a 2024T3/M30-carbon laminate. Images were captured at different crack lengths while the maximum fatigue load was applied statically. Figure 15 shows the evolution of the fatigue delamination together with a comparison with chemical etching, which was performed only at the end of the test 16 DIC measurements

position along y-direction [mm]

14

Chemical etching

σ =150 Mpa R=0.05

12 10 8 6 4 2 0 5

7

9

11 13 15 17 19 Position along the fatigue crack [mm]

21

23

25

Figure 15 Fatigue delamination growth measured using DIC and comparison with chemical etching. FML 2-2/1-0.4 2024-T3/M30-carbon, σmax_fatigue =150 MPa, R=0.05 The application of DIC to measure the fatigue delamination becomes very suitable when it is difficult to perform chemical etching, like in case of FMLs made with titanium alloys or stainless steel. In Figure 16 the comparison between DIC and Cscan is illustrated for a FML 3-3/2-0.4 made by Ti-6-al-4V/M30-carbon.

R. Rodi, R.C. Alderliesten, R. Benedictus

508

Respect to the C-scan, DIC is not affected by the ‘kissing’ bond, which is a phenomenon occurring when the disbonded surfaces remain in contact to each other thus reducing the attenuation of the signal.

Transition area

Position along y-direction [mm]

9

4

-1

-6

C-SCAN DIC -11 0

5

10

15

20

25

30

35

Position along x-direction [mm]

Figure 16 Comparison of the fatigue delamination shape between DIC and C-scan for a FML 3-3/2-0.4 Ti-6Al-4V/M30-carbon. σmax_fatigue =170 MPa, R=0.05, a=29.5mm. The power of this approach lies in the fact that with one single test is possible to investigate the fatigue delamination growth in-situ and in a non-destructive manner. Investigating the fibres layer deformation Fatigue tests were performed on laminates with prepreg on the outside to measure the deformation of the fibre/prepreg layer in presence of a fatigue crack. Figure 17a shows the strain field of the prepreg layer of a 2024-T3/ S2-glass laminate with a fatigue crack in the metal layers.

Figure 17 Illustration of the strain field of the fibres layer measured with DIC in a 2024-T3/S2-glass laminate with prepreg outside with a fatigue crack of 17.5 mm (left). Qualitative illustration of the stress distribution in the fibres layer (right)

An experimental approach to investigate detailed failure mechanisms in FMLs

509

The delamination has a triangular shape and the strain values in the disbonded area present an average constant value, except in the vicinity of the crack tip, where a peak is present. The calculation of the stress gradient is possible if the measured strain is related to the stress with the stress-strain curve of the prepreg. Figure 17-b qualitatively illustrates the stress distribution in the fibre layer, both ahead and behind the crack tip. Several images were captured at different crack lengths by applying statically the maximum fatigue load. Figure 18 illustrates the evolution of the total stress in the fibre layer due to the fatigue load. The shape of the measured strain, recalculated in stress, along the fatigue crack accords to the shape of the total stress analytically calculated as reported in [8-10]. The quasi-constant value of the stress in the wake of the crack is the result of a balancing between crack opening and delamination extension. In the vicinity of the crack tip, due to the small amount of delamination, a stress peak is present. In addition, the strain compatibility ahead of the crack tip must be respected, if static delamination does not occur, thus the stress peak represents the link between the stress behind and ahead of the crack tip. 500

a=17 mm a=10 mm

450

a=20 mm

Total stress [MPa]

400 350 300 250 200

σmax =150 MPa R=0.05

150 100

Analytical calculation

50

DIC measurements

0 4

9

14 19 Position along the crak length [mm]

24

Figure 18 Fibre stress measured with DIC on the external prepreg along three fatigue cracks in a 2024-T3 /S2-glass inverted laminate, and comparison with analytical calculations. When the fatigue crack reached a length of 20 mm, the load was increased quasistatically. Figure 19 illustrates the evolution of the stress in the fibres layer when high static loads were applied. Despite the absence of crack propagation in the load range between 110 MPa and 230 MPa, the stress peak shifts slightly to the right. This is explained by the high deformation developed in front of the crack tip, which increases the strain of the fibres layer ahead of the tip.

R. Rodi, R.C. Alderliesten, R. Benedictus

510

Crack propagation in the metal layers was observed when the load reached 230 MPa. This induces the stress peak to shift further to the right, following the crack extension.

1400

Total stress [MPa]

1200

280 MPa

1000

255 MPa

800 230 MPa

600 210 MPa

400

185 MPa 160 MPa

200

140 MPa 110 MPa

0 4

6

8

10

12 14 16 18 20 Position along the crackline [mm]

22

24

26

Figure 19 Fibre stress measured with DIC on the external prepreg along a fatigue crack in a 2024-T3 /S2-glass inverted laminate under several static loads. A Modification of the available analytical prediction model [8-10], which could account for the quasi-static load sequence, would provide calculation of the bridging stress also in case of high static loads.

CONCLUSIONS The hybrid nature of FMLs increases the complexity of the analysis of the failure mechanisms. So far different measurement techniques have been employed to assess the failure mechanisms within FMLs, such as fatigue delamination, plastic zone extension, fibres layer behaviour. An experimental approach able to provide detailed measurements of the most important failure mechanisms within FMLs has been proposed. It has been described how the use of digital image correlation enables to perform nondestructive measurements of fatigue delamination, plastic zone shape and fibres layer deformation directly during the test.

An experimental approach to investigate detailed failure mechanisms in FMLs

511

Using DIC the plastic zone shapes of both saw-cut and fatigue crack configurations have been compared. The plastic zone size has been analytically calculated with Irwin’s and Dugdal’s approximations, pointing out the limitations of such analytical tools. Following the same approach, the delamination shape has been measured describing the potentiality of DIC during fatigue testing of FMLs. FMLs panel manufactured with prepreg on the outside provided an insight on the fibres layer deformation. DIC provided detailed measurement of the strain/stress in the fibres layer, including the bridging stress.

REFERENCES [1]

Vlot, A., Gunnink, J.W. (2001), Fibre Metal Laminates an Introduction, Kluwer Academic Publishers. [2] Anonymous (1997), Damage tolerance and fatigue evaluation of structure, Federal Aviation Regulations Part 25 – Airworthiness Standards: Transport category airplanes, Section 25.571, Federal Aviation Administration, Department of Transportation, Washington D.C. [3] Vermeeren, C.A.J.R. (1995), The Residual Strength of Fibre Metal Laminates, PhD thesis, Faculty of Aerospace Engineering, Delft University of Technology, The Netherlands. [4] Vries, T.J. de (2001), Blunt and sharp notch behaviour of Glare laminates, PhD Thesis, Delft University of Technology. [5] Marissen, R. (1988), Fatigue Crack Growth in ARALL, A hybrid AluminiumAramid Composite Material, crack growth mechanisms and quantitative predictions of the crack growth rate, PhD Thesis, Delft University of Technology. [6] Roebroeks, G.H.J.J. (1991), Towards GLARE - The Development of a fatigue Insensitive and Damage Tolerant Aircraft Material, PhD thesis, Delft University of Technology. [7] Beumler, Th. (2004), Flying Glare®, - A contribution to aircraft certification issues on strengths properties in non-damaged and fatigue damaged GLARE® structures, PhD Thesis, Delft University of Technology. [8] Alderliesten, R.C. (2005), Fatigue Crack Propagation and Delamination Growth in Glare, PhD Thesis, Delft University of Technology. [9] A. Fahr, C. E. C., D.S. Forsyth, C. Poon, J.F. Laliberte' (2000). Nondestructive evaluation methods for damage assessment in Fibre Metal Laminates. Polymer composites 21(4): 9. [10] Alderliesten, R.C. (2007), Analytical prediction model for fatigue crack propagation and delamination growth in Glare, International Journal of Fatigue 29 (4) (2007) 628-646.

512

R. Rodi, R.C. Alderliesten, R. Benedictus

[11] Groves RM, Osten W, Doulgeridis M, Kouloumpi E, Green T, Hackney S, Tornari V, Shearography as part of a multi-functional sensor for the detection of signature features in movable cultural heritage, Proc. SPIE 6618-10, 2007. [12] Moser E, Detection Capabilities of State-of-the-Art Shearography Systems. 17th World Conference on Non-destructive Testing, 25-28 Oct 2008, Shanghai, China [13] R.Rodi, (2007), The effect of external stiffeners on the fatigue crack growth in fibre metal laminates, 24th ICAF Symposium,, Italy [14] Lemmen, H.J.K., Alderliesten, R.C., Benedictus, R., Hofstede, J.C.J., Rodi, R., The power of Digital Image Correlation for detailed elastic-plastic strain measurements. EMESEG ’08 conference, Engineering mechanics, structures, engineering geology, Creta, 2008. [15] Po-Chin Hung and A.S. Voloshin, In-plane strain measurement by Digital Image Correlation, J. of the Braz. Soc. of Mech. Sci. & Eng., Vol 25, no.3, 2003. [16] Groves RM, Chehura E, Li W, Staines SE, James SW and Tatam RP, Surface Strain Measurement: A comparison of speckle shearing interferometry and optical fibre Bragg gratings with resistance foil strain gauge”, Meas. Sci. Technol., 18, pp. 1175-1184, 2007. [17] Xuefeng Zhao, Jihua Gou, Gangbing Song, Jinping Ou, Strain monitoring in glass fiber reinforced composites embedded with carbon nanopaper sheet using Fiber Bragg Grating (FBG) sensors, Composites Part B: Engineering, Volume 40, Issue 2, March 2009, Pages 134-140. [18] Austin, T. S. P., M. M. Singh, et al. (2008). Characterization of fatigue crack growth and related damage mechanisms in FRP-metal hybrid laminates, Composites Science and Technology 68(6): 1399-1412. [19] R.Rodi, G.Campoli, R.C. Alderliesten, R.Benedictus (2009), Characterization of the crack tip behaviour in fibre metal laminates by means of digital image correlation, 50th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, AIAA-20092586 Palm Springs, USA . [20] S.U. Khan, R.C. Alderliesten, R.Benedictus (2009), Fatigue crack growth in Fibre Metal Laminates under variable amplitude loading, 25th ICAF Symposium, Rotterdam, The Nederlands.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

EXPERIMENTAL DETERMINATION OF ENERGY RELEASE RATE OF CFRP STRUCTURES BY MEANS OF TRANSVERSE CRACK TENSION TESTS Ll. Llopart Prieto 1, G. Spenninger 1 and H. Wagner 2 1

EADS Premium AEROTEC GmbH, Manching, Germany 2 Institut für Flugzeugbau (IFB), University of Stuttgart, Germany

Abstract: The use of cost effective infused materials like Vacuum Assisted Process (VAP) needs of splicing technique when manufacturing large fuselage components and structures. Only resin fills the area between these butt joined semi-finished products, where cracks generate under high loading. Afterwards these cracks develop to delamination. Delamination and their susceptibility to growth are normally characterized using strain energy release rate parameter (G). Compression After Impact (CAI) test has become a key experiment to gather damage tolerance performance data during the design or certification phase of a new structure or material, involving composites. However, this procedure cannot specify information about damages in varying depths and the kind of damage. An alternative procedure is the use of the Transverse Crack Tension (TCT) specimens. This work presents the results obtained for G under mode II loading (GIIc) by means of TCT-tests for different infusion and Prepreg materials used at the present time in aeronautical industry. It is shown that results using standard specimens as Double Cantilever Beam (DCB) and End-Notch Flexural (ENF) specimen and non-standardised specimens, TCT, agree. Furthermore, the use of TCT method let reconstruct the initiation and development of damages by means of delamination under static loading, which will help on opening the floodgates to the next generation of sizing criteria and numerical tools. M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 513–528. © Springer Science+Business Media B.V. 2009

514

Ll. Llopart Prieto, G. Spenninger, H. Wagner

Extensive static test sets were accomplished, which made a fundamental understanding available of initiation and damage process.

INTRODUCTION Resin transfer moulding technology has been used from the last decades as a process to manufacture composites. However, this technique is limited to relatively small parts due to exploding tooling costs. Since then, the technology have been developed to replace the expensive tooling with an autoclave or simply using vacuum pressure to infuse dry fabrics. One of the latest developments - the Vacuum Assisted Process (VAP) by EADS has found its way into workshop to manufacture large aerospace components. In this process, dry semi-finished layers in form of Non Crimp Fabrics (NCF) or woven fabrics are used [1]. When manufacturing large components and structures, splice configurations result due to the limited widths of the semi-finished products. An application of the splice building method is the flap track concept represented in figure 1. The U-formed shell has a too large extension in order to be covered by only one semi-finished layer. The black spots depicted in the figure 1, schema on the bottom left, represent the place were the splices are butt jointed and here resin nests are generated.

Figure 1: Example of a structural part made by VAP-process with splicing characteristics

Experimental determination of energy release rate of CFRP structures

515

During excessive load cracks initiate and propagate in these resin reach zones leading to delamination. Current design philosophy - based on the damage tolerance principle - excludes the occurrence of delamination, because the temporal and local moment of this kind of failure is not reliably predictable yet. This gap is fulfilled by tests allowing to macroscopic factor to be applied in the stress analysis. Compression after impact test has become a key experiment to gather damage tolerance performance data during the design or certification phase of a new composites structure or material. On the other hand, splice details can be dimensioned and secured via analytical and numerical procedures, only if delamination is predictable. Delamination and their susceptibility to growth are normally characterized using fracture mechanics principles and the strain energy release rate parameter. The evaluation and results presented in this work are the first step to extend sizing procedure for splice joints to mixed mode (peeling and shear) conditions. In first line and with the aim to class VAP-materials regarding its resistance to delamination and to compare conventional Prepreg materials, different standard specimens and tests were used as the Double Cantilever Beam (DCB), the EndNotch Flexural (ENF) and the CAI. However, new specimen shape for the evaluation of delamination is necessary, since standard samples, specially DCB and ENF, exhibit uncertainties regarding the interpretation of results and are difficult to be manufactured. The presented work have the objective to prove the validity of TCT- specimens for the determination of GIIc in spliced structures, to complete the material database for basic fracture mechanical parameters (GIIc) based on the TCT- and CAIspecimens, and to compare the evaluation of fracture toughness by means of TCT and CAI specimens.

STATE OF THE ART Different standard specimen shapes exists to determine the critical strain energy release rate under different loading modes. These are, as already introduced, the DCB for Mode I evaluations and ENF for Mode II, see figure 2.

Ll. Llopart Prieto, G. Spenninger, H. Wagner

516

DCB-test for Mode I after Airbus standard

ENF-test for Mode II after Airbus standard.

Figure 2: Standard specimens for the determination of the strain energy release rate under Mode I and II. As described in [2], the TCT supplies, by means of a simple tensile test, similar data as complex ENF- and DCB-test. Damage tolerance, regarding delamination, can be determined by targeted positioning of an imperfection or a pre-damage. Between the interrupted fibres or layers a resin gap is generated. This gap is characterised with pure matrix characteristics and thus only limited load can be carried. From the notch effect of this gap results, in case of gaps inside the laminate, a nearly pure Mode II loading. If these gaps are placed symmetrically at the upper and lower layer the TCT external gap specimen is configured. This specimen experiments a loading mixture of Mode I and II, whereby Mode II is larger than Mode I.

σII h, t

Tensile stress Geometrical properties

Figure 3: Characteristic of TCT internal gap specimen A fracture mechanic approach can be found in [2] for the calculation of the stress delamination, σII,del, of TCT specimens made of unidirectional lay-up as characterised in figure 3. For multi-directional lay-up an approach was derived in [3]. This relation is given in Eqn. 1.

Experimental determination of energy release rate of CFRP structures n ⎛ n ⎞ ⎜ ⎟ ⋅ + ⋅ E t E t ⋅ E t ∑ ∑ dg dg dt dt ∑ dg dg ⎜ ⎟ 1 1 = = dg dt ⎝ ⎠ = 2 ⋅ x ⋅ Gc ⋅ dgn=1 h2 ∑ Edt ⋅ tdt

517

n

σ II , del

(1)

dt =1

where E is the Young’s Modulus, x the number of crack surfaces, Gc the critical energy release rate, dg index of non interrupted layers and dt index of interrupted ones. Experimental determination of strain energy release rate under Mode II and I by means of TCT-specimens is not widespread contrary to the conventional standards. Only data on TCT testing could be found in [2] by Eurocopter and the German Aerospace Centre (DLR). One of the main factors against composite materials use, is their poor damage tolerance, particularly their lack of resistance to impact damage, and the reduced compression strength in the presence of the impact damage. Therefore, the determination of compression testing in low velocity impacted specimens, the compression after impact, is a key experiment to measure damage tolerance performance during the design or certification phase of a new structure. As shown in [4], the differences in impact and CAI response of the laminates are largely a consequence of the impact damage created at the Damage Threshold Load (DTL), rather than of the differences in delamination growth. This threshold force is related to initial damage in the form of matrix cracking, fibre breakage and mainly delamination and it is identifiable with the first load drop on the load history of the impact event. Within these evaluations a strong correlation between GIIC values measured by delamination tests and those calculated from measurements of DTL were found. Investigations prove that impact load-time histories can be used to characterise impact events, see figure 4. The graphical representation of load versus time shows a sudden drop at specific contact force. This point corresponds to the DTL and indicates the initiation of detectable damage in the laminate. Before the maximum, further oscillations and load drops happen. These are generated by the development of additional damages [5].

Ll. Llopart Prieto, G. Spenninger, H. Wagner

518

Figure 4: Force history of a CAI specimen of HTA/922 impacted at 20J with detectable damage after impact [4] Critical strain energy release rate GIIc can be determined using the DTL as described in Eqn. 2.

GIIc

9 ⋅ DTL2 ⋅ (1 −ν 2 ) = 8π 2 Eh3

(2)

where ν is the Poisson's ratio and h the laminate thickness.

EXPERIMENTAL TESTING Fifteen different sets of TCT specimens were tested, see table I. They were manufactured with both biaxial NCF of the type (0/90) or Unidirectional fabrics (UD). Due to the thickness of the semi-finished layers larger thickness are generated than TCT-specimens made of Prepreg material. Further sets were conceived to be able to compare the results with Prepreg specimen from [2] and to characterise UD material G1157 from Hexcel.

Experimental determination of energy release rate of CFRP structures

519

Table I: Definition of TCT-specimens tested Set

1

2

3

4

5

6

7

8

9

Configuration

Lay-up Material Notch location

(0)10 UD HTS/977-2 TCT- internal gap (0)10 UD IM7/8552 TCT- internal gap [(0/90/90/0/90/0/90/0/0/9 0)]s HTS/977-2 TCT- internal gap [(0/90/90/0/90/0/90/0/0/9 0)]s IM7/8552 TCT- internal gap (0/90)5s HTS/977-2 TCT- internal gap

(0/90)5s IM7/8552 TCT- internal gap (0)5 UD HTA/RTM6 Bin. 12g/m² TCT- internal gap (0)12 UD HTA/RTM6 Bin. 6g/m² TCT- internal gap (90,0)5s IMS/RTM6 TCT - internal gap

Total thickness [mm]

Interrupted thickness [mm]

1.25

0.250

0º – 0º

1.25

0.250

0º – 0º

2.50

0.250

0º – 0º

2.50

0.250

0º – 0º

2.50

0.250

0º – 90º

2.50

0.250

0º – 90º

1.25

0.250

0º – 0º

3.15

0.525

0º – 0º

5.00

1.000

0º – 90º

Crack configuration

Ll. Llopart Prieto, G. Spenninger, H. Wagner

520

Table I (continuation) Set

Configuration

Lay-up Material Notch location

Total thicknes s [mm]

Interrupted thickness [mm]

Crack configuratio n

10

(90,0)5s IMS/RTM6 Bin. 12g/m² TCT - internal gap

5.00

1.000

0º – 90º

11

[(90/0)(90/0)(90/0)(90/0) (0/90)]s IMA/970-20 TCT- internal gap

5.00

1.000

0º – 90º

12

(0/90)5s HTA/RTM6 Bin. 12g/m² TCT - internal gap

5.00

1.000

0º – 90º

13

[(0/90)/{(90/0)}3/(0/90)] s HTA/RTM6 Bin. 12g/m² TCT-internal gap

5.00

1.000

0º – 0º

14

(0)10 UD HTA/RTM6 Bin. 6g/m² TCT- internal gap

3.20

0.640

0º – 0º

15

(0/90)5s HTA/RTM6 Bin. 12g/m² TCT - external gap

5.00

1.000

0º – 90º

The instrumentation of the TCT-specimens took place via one knife extensometer monitoring a length of 50 mm over the interrupted zone, see figure 5. Within, the exact strain by the initiation of the delamination can be evaluated at the perturbed area and on the basis of characteristic discontinuities in the measures. Furthermore, the tests were carried out with displacement control and the perturbed area was monitored with a microscope camera. This last allowed to compile the failure mechanism accurately and to assign the individual failure features to the elongation values.

Experimental determination of energy release rate of CFRP structures

521

Figure 5: Instrumentation of the TCT-test Regarding the CAI-test, the impacts were performed with a standard falling weight impact machine with an hemispherical tip of 16 mm diameter. Impact velocities were kept between 1 and 3 m/s. The force applied to the specimen is recorded together with the initial velocity at the time of impact. Energy, velocity and deflection are numerically deduced from the force/time data. CAI specimens have dimensions of 150 x 100 mm and a thickness of about 4 mm. Originally four different sets of CAI-specimens were aimed to be tested and compared with the TCT - results. However, only the set described in table II could be closed and evaluated. Table II: Specimen set definition for CAI-test Layup Fibre areal NCF weight [g/m²] {(0/90)(0/90)}2s 556 16 HTA RTM6 12 g/m² Biaxial (0/90)

Set Fibre Resin Binder

Thickness [mm] 4.14

Impact Energy [J] 9, 16, 25, 30, 40

The fibre volume content of all specimens presented in this work were set to be 60 ± 5 %. The environmental condition during testing was Room Temperature (RT) and the specimens were stored As Received (AR).

RESULTS TCT-specimens The TCT-test series presented in this paper served to the determination of material properties as GIIc and GMIXc and made important findings on the failure mechanism of this kind of specimens. Table III represents a typical failure process of a spliced laminate. As expected cracks initiate on the resin nest between interrupted fabrics and leads to delamination and fracture. Transverse tension cracks produce a

522

Ll. Llopart Prieto, G. Spenninger, H. Wagner

noticeable degradation of the mechanical properties and lead, in further failure, to large peeling of the fibre layers from one to another (delamination). On gaps larger than 1 mm fibre waviness were generated. This has strong influence on the generation of matrix damages as resin cracks and transverse tension cracks, which induced an early failure of the specimen. Small waviness had no influence on the resistance to delamination. Table III: Failure mechanism of spliced laminates

1. Resin cracking

- Cracking inside the resin nest - No influence on the total strength - The generation depend on the boundary conditions (e.g. Waviness)

2. Transverse tension cracks

- Cracks develop perpendicular to tensile load - Influence on the total strength, the reason is a classic developed σ22 – stress

3. Delamination

- Large peeling of the fibre layers from one to another - Clear degradation of mechanical performances - Delamination depends on the geometrical characteristic of the notch and also the lay-up

4. Fibre fracture

- Fracture of the bearing fibres - Complete destruction of the specimen - Depending on the boundary conditions, fibre failure took place before or after delamination.

Based on test results, it can be stated that Gc is a material constant which depends on fibre orientation at the cracked region, see table IV. If not varying material indices wants to be used during the generation of technical data for splices in a

Experimental determination of energy release rate of CFRP structures

523

multi-directional laminate, the conservative case involve to generate splices interrupted between 0° and 90° fabrics. Table IV: Critical strain energy release rate and delamination stress for the sets 7 to 8 and 12 to 15 presented in table I. Set 7 8 12 13 14 15

Delamination stress [MPa] 1261.8 N. a. 433.4 536.2 N. a. 511.9

Gc [N/mm] 1.19 1.90 0.99 1.50 1.85 1.37

Contrarily as expected, the critical strain energy release rate for external gaps is higher than for internal TCT specimens. A possible cause is the lower waviness produced on the TCT with external gap.

CAI-specimen The load history of set 16 for an impact energy of 30J is depicted in figure 6.

Figure 6: Load history of specimens set 16 impacted with 30 J From the curve in figure 6 the DTL corresponds to 4666 N. The statistical Emodulus and the Possion's ratio of the tested lay-up corresponds to ca. 64000 MPa

524

Ll. Llopart Prieto, G. Spenninger, H. Wagner

and 0.45 respectively. The thickness of the specimens were 4.14 mm. According to [4,6-9] and Eqn. 2 the critical strain energy release rate corresponds to 0.44 N/mm. The correspondent C-scan inspection together with 16J impact pictures are illustrated in figure 7.

Figure 7: C-scan of specimens set 16 with 16 and 30 J impact

Direct comparison of different Prepreg- and VAP-systems regarding the strain energy release rate under Mode II loading is possible and represented in figure 8, with the test results of all sets dealing with TCT internal gap, and available GIIcvalues in the literature [4,6-8]. For the classification of the represented values, the materials type, conventional Prepreg or resin infiltration materials (LRI), if the semi-finished layer is composed of unidirectional layers or multiaxial and the orientation of the layers on the crack interface are indicated. As can be observed, HTS/RTM6 behave a very good interlaminate fracture toughness which is noticeably superior to older Prepreg systems. Only the Prepreg system IM7/977-2 nears the specific values obtained with VAP-systems. Also the influence of binder fleece becomes noticeable, since using 6 g/m² an increase of GIIc is observed.

Experimental determination of energy release rate of CFRP structures

525

Figure 8: GIIc comparison of different material systems

COMPARISON OF STANDARD AND TCT - TESTS For the determination of strain energy release rate for pure Mode I and/or II the presented standardised test specimens have been manufactured and tested in an earlier technology program launched at EADS [10]. Table V shows, as example, a listing of the actual strain energy release rates for VAP-systems for comparison determined by means of these standard specimens and the TCT. When comparing tests with same boundary conditions, TCT-internal and ENF, it can be noticed that ENF results in a substantially more progressive value. This lays on the fact that fibre fracture occurred before delamination in the TCT-specimen. TCT generates less standard deviation than ENF or DCB. Table VI, summarise the critical strain energy release rates evaluated by means of both the TCT-specimens and CAI-specimen for sets 12, 13 and 16, which were made of identical lay-up and materials. As can be seen in table VI, the value of GIIc calculated by means of the load histrory correspond to more than the half of the value obtained by means of the TCT-specimens. Compared with the ENF results, the results calculated with CAI are not plausible.

Ll. Llopart Prieto, G. Spenninger, H. Wagner

526

Table V: Overview of interlaminar fracture toughness obtained for the material system HTA, HTS/ RTM6 with different specimen types Property Test method

ENF

Crack interface

0º - 0º

Environment al conditions Binder 12g/m2 Standard deviation [%] Value [N/mm]

GIc

GIIc TCT-in 0º - 0 º -55º C AR

RT AR

DCB

0º - 90 º RT AR yes

0º - 0º

Mixed Mode TCT-ex t 0º - 90º

RT AR

-55º C AR

RT AR

yes

yes

yes

no

yes

yes

yes

16.2

13.5

14.6

10.3

5.9

11.9

10.5

6.9

1.19 2

2.23 8

1.40 3

1.500

0.989

0.82 2

0.92 8

1.37

Table VI: GIIc for sets 12, 13 and 16 Set 12 13 16

Test TCT TCT CAI

GIIc [N/mm] 0.989 1.500 0.440

Eqn. 2 were derived in [4,6-9] with a simplified model, whereby all the delaminations were collapsed onto a singular circular patch and the quasi-isotropic laminate was treated as isotropic. As can be seen in figure 7 damage do not occurred in a circular patch but as elliptical one. Furthermore, no isotropic behaviour can be assumed for the tested lay-up. This last result shows that more investigations and development on the calculated GIIc by means of the load history of infused materials is required.

DISCUSSION AND OUTLOOK The main target of the project which this paper belongs, is to find out whether present used strain limits can be increased when implementing damage tolerant methods. Based on the results obtained here, it can be noticed, that delaminations occur far from present strain limits. Particularly for (0/90) lay-ups, the strain delaminations by tensile loading are very close to the fibre strength, which would authorise related to delamination failure - an increase on the strain limit.

Experimental determination of energy release rate of CFRP structures

527

Furthermore following statements can be concluded: - TCT specimens are proved to be convenient for the determination of critical strain energy release rate under Mode I and Mixed Mode I+II. - TCT specimens allow to be tailored in order to reproduce real splice configurations on large structures. - Multiple fibre layers with lower fibre areal weight NCF would allow to reach higher resistance to delamination. - GIIc is affected by crack interface configurations only by high fibre areal weight. Ultrasonic C-Scan projects all inside damage in a laminate onto a single plane. Hence, no information is given about damages in varying depths and on the kind of damage, i.e. delamination, fibre failures, or matrix cracking. By means of TCT specimens the damage can be monitored. - Eqn. 2 is not applicable for not quasi-isotropic layers and materials made of dry fabrics with high areal weigh. Further works shall concentrate: - on the development of Eqn. 2 taking into account different forms of damaged areas on CAI-testing and introduce non isotropic behaviour. - on analysing TCT-results with external gap configurations and compared with mixed mode bending results.

ACKNOWLEDGEMENTS Mr. H. Bansemir is gratefully acknowledged to insist by means of his publications, discussions and presentations about the advantages of the TCT-specimens. Part of this work were carried out when part of the authors worked for EADS Military Air Systems.

REFERENCES [1] Berchtold, G. (2006), SAMPE California. In: “The potentials of vacuum assisted CFC-Manufacturing process and its impact on large aircraft components”. [2] Burghagen, S. (2005) Master Thesis at Delaminationsverhalten von Unidirektionalverbunden“.

Eurocopter.

In:

“Das

[3] Pahl, A. (2005) Master thesis at EADS-Military Air Systems. In “Untersuchung des Delaminationsverhaltens von Spleißkonfigurationen in Multiaxialgelegen auf Basis der Bruchmechanik.“ [4] Cartié D. D. R. and Irving P. E. (2002), Comp. Part A, vol. 33, p. 483-493. In: “Effect of resin and fibre properties on impact and compression after impact performance of CFRP”.

528

Ll. Llopart Prieto, G. Spenninger, H. Wagner

[5] Wagner, H., Bansemir, H., Drechsler, K. and Weimer, C. (2007), SAMPE EUROPE Technical Conference. Madrid. In: “Impact Behaviour and Residual Strength of Carbon Fiber Textile based Materials” [6] Bansemir, H. (1973), Archive Appl. Mech. vol. 42, n. 2, p. 127-140. In "Krafteinleitung in versteifte orthotrope Scheiben" [7] Bansemir, H. and Emmerling S. (2000), ECD-0096-99-PUB. In: "Fatigue Substantiation and Damage Tolerance Evaluation of Fiber Composite Helicopter Components" [8] Spenninger, G., Bansemir, H. and Schulz M. (2007), Seminar Hochleistungsstrukturen im Leichtbau; In: "Dimensionierung von Flugzeug- und Hubschrauberstrukturen in VAP- und Prepreg-Bauweise mit Hilfe bruchmechanischer Daten" [9] Davies, G. A .O. Zhang, H., Zhou, G. and Watson, S. (1994), Composites, vol. 25, nr. 5, p. 342-350. In: "Numerical modelling of impact damage". [10] Neumaier, R. (2006), Technology test program. Internal report (EADS Military Air Systems).

25th ICAF Symposium – Rotterdam, 27–29 May 2009

CENTRAL MATERIAL CONCEPT: TOWARDS AIRCRAFT WING STRUCTURE INSENSITIVE TO FATIGUE Mohamed A.A.Attia University of Salford The Crescent M5 4WT Abstract: Over the last two decades, some groundbreaking research concerning materials has been published. Major improvement of material is still, nevertheless, robust manifested in the recent development termed CentrAl. The work reported in this research serves two purposes. First, in response to concerns regarding fatigue problems in aircraft structures, it is shown how to realize the benefits of this concept. The second part of the work aims to contribute to the existing dearth body of knowledge on CentrAl concept to bridge the gap between theory and operational practice. Grounded in state-of-the-art literature, current insight for the deficient research in CentrAl are identified and presented n the form of required experimental process. The ultimate research goal aims to analyze and summaries the present state and future tendency in the different approaches of metal and fibre composite layer combination. The achieved results with the support of CentrAl are discussed in detail and compared whenever possible with previously published results on Glare to lay out the applicability of CentrAl concept to aircraft wing structures insensitive to fatigue.

INTRODUCTION Aluminium alloys have been used for over fifty years in aircraft structures. Driven by weight saving of the total wing structure and the required resistance to crack growth for this part of the structure, the development of composite materials initiated. Composites offer many advantages when compared to metal alloys, M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 529–537. © Springer Science+Business Media B.V. 2009

530

Mohamed A.A. Attia

especially where excellent fatigue properties are concerned. Aiming to reduce weight and tolerate damage, a new lightweight fibre metal laminates (FML) family has been developed. In principle, three parties are involved in structural design: the aircraft industry, the aircraft operator and the airworthiness authorities. The demand and capabilities are typically evaluated against each other and through a design process comprising methods and analysis, validation tests, and regulations. Put simply, designing of structures is a guided interactive process aimed at achieving a practical balance between the state of the art and the intended usage requirements. To this end, fatigue cracking is considered to be major threat to structural integrity. The understanding of fatigue damage problems increased significantly in the last five decades. But in spite of so much progress, it must be admitted that we are still facing problems. Major improvement of materials is still, nevertheless, robust manifested in the recent material termed CentrAl. Recently, the U.S. aluminum giant Alcoa, materials-technology company GTM Advanced structures and scientists at Delft University of Technology in the Netherlands have patented a FML called CentrAl reinforced aluminum. The new, CentrAl concept comprises a central layer of FML, sandwiched between one or more thick layers of high-quality aluminium. It is interesting to note that the development of new technologies and materials take many years before they become mature enough to apply them in civil aircraft due to stringent qualification requirements by way of extensive tests required. The development of new materials is painstakingly slow due to their need to perform better than the currently being used counterparts. It ensues naturally that the need for probation of fatigue and damage tolerance of the new CentrAl material is an essential area, not merely from research and academic purposes, but for practical applications as well offering a challenging field for innovative ideas to achieve new solutions and concepts for old problems. Having said that the stringent qualification requirements make the development of new material daunting by way of extensive tests required, and given that they can not be satisfied without relevant tests, the work in this paper aims to serve this purpose. Thus, the present research aims to explore and look more closely at the CentrAl for application in aircraft wing structure. How well it works and what the typical Fatigue and Damage Tolerance issues related with CentrAl is discussed in this paper with an eye on the experimental process opening this new arena of research to reach the technology readiness of CentrAl for application in aircraft wing structure. To narrow the defined gap between the development of new technology and the CentrAl technology readiness, this poster may be considered as the footsteps preparing for the CentrAl probation. In order for to achieve the full potential of CentrAl and include it in the next generation category of fibre material, this is a call for other universities to start research activities on the important area of fatigue resistance that lacks knowledge for this material.

Central material concept: Towards aircraft wing structure intensive to fatigue

531

THE PRESENT SITUATION FMLs consist of alternating layers of uni-directional impregnated fibre lamina and thin metallic sheets adhesively bonded together. This technique of coupling the metal with fibre shows improvements over the properties of both aluminium alloys and composite materials individually, let alone the fact that alternating metallic and fibre layers of FML provide the potential to tailor it to its application. In fact, the combination of metal and fibres composite laminates can create a synergistic effect on many properties. The impact of the presence or the lack thereof of the amount of layers and their orientations (the optimal metal fraction relative to the fibre content) appeared to be crucial for a satisfactory fatigue performance. Difference between material properties is what gives rise to the gap between the Glare and CentrAl. Before reviewing the testing methods, it would be best to see what factors affect this concept and sometimes even restrict it reaching the technology readiness. Design: Stress concentrations caused by sharp corners, sudden changes of section, or undercuts are all classified as ‘incipient cracks’ from which a fatigue crack may spread. Surface Finish: Scratches in highly finished surfaces or tooling marks are left when machining also represent stress concentrations which can lead to fatigue failure in highly stressed components. Surface discontinuities left by heat-treatment processes, hot working and cold working can also cause fatigue failure. Temperature: Changes of temperature at which the test is carried out, and at which the material is sunsequently used in service, can have a significant effect upon the aftigue resistance of that material. Residual Stresses: Such stresses left by processing can also substantially affect the fatigue resistance of the work piece. Processes which leave compressive stresses in the surface of the material improve its fatigue resistance, whilst processes which leave tensile stresses in the surface of the material reduce its fatigue resistance. Corrosion: This may be atmospheric corrosion, oxidation during heat treatment or saline attack due to marine environments. Surface treatment by galvanizing and painting prior to saline exposure would prevent corrosion and would result in normal fatigue characteristics for the material. Damge tolereance: The ability of structure to sustain anticipated loads in the presence of fatigue , corrosion or accidental damage until such damage is detected through inspections or malfunction and repaired. Durability: Ability of the structure to sustain degradation from such sources as fatigue, accidental damage and environmental deterioration to the extent that they

Mohamed A.A. Attia

532

can be controlled by economically acceptable maintenance and inspection programs. Crack Growth (Damage): The interval of damage progression from lengths below which there is negligible probability of detection to an allowable size determined by residual strength requirements. Damage Detection: A sequence of inspection in a fleet of airplanes with methods and intervals selected to achieve timely damage detection.structural inspection programs are typically developed by use of rating systems for each of the three major form of damage.

FINDINGS OF THE LITERATURE

METHODOLOGY FOR MECHANICAL CHARACTERISATION In view of the foregoing it can be discerned that in order to be capable of reaching the technology readiness of CentrAl for application in wing structure, testing is the fundamental feature that distinguish the material and helps to establish it in solid grounds. It can be easily deduced as such keeping in view that in aerospace applications the requirement is to reach full maturity and be at a stage where there is zero risk associated with using the material, namely; to reach initial operating

Central material concept: Towards aircraft wing structure intensive to fatigue

533

capability. For the purposes of the research, consultation with experts in Delft University of Technology (Assistant Professor in Fatigue and damage tolerance of metallic and hybrid structures) provided essential knowledge about the subject matter to carry out an authentic research. The present work discusses the process for the CentrAl experimental to open this new arena of research in the industry to reach the technology readiness of CentrAl for application in aircraft wing structure. Tensile testing Tensile testing has made it possible to determine ultimate tensile strength and elongation, yield point, yield strength, modulus of elasticity, resilience, and toughness. The mechanical properties from the tensile test are illustrated by plotting a stress-strain curve. From this plot, the deformation behaviour of the material can be seen. Thus the material is characterized by its yield strength, ultimate tensile strength, stiffness, ductility, resilience and toughness. The data are used to compare CentrAl with Glare and, in conjunction with the other factors derived for the context of this research, it assist in the replacement or improvement of Glare. Compression Testing Compressive test give information about CentrAl under a load. From this test the yield point, yield strength, modulus of elasticity and ultimate compression strength are obtained. Compression testing is the same in many ways as tension testing. The specimen shortens, however, instead of lengthening. Up to, and through the yield point, the compression stress-strain curve is similar to that for tension; beyond this point, however, the compression curve becomes steeper because of the increased cross-sectional area in compression. Torsion Testing The torsion test is useful for the purpose of this research to determine shear modulus, yield shear strength, ultimate shear strength, modulus of rupture in shear, and ductility. The shear stress versus shear strain curve is determined from simultaneous measurements of the torque and angle of twist of the test specimen over a predetermined gage length. Bending Testing When CentrAl, initially straight, is stressed in bending, the relationships between the applied bending moment and the deformation, on the one hand, and between the deformation and the stress at any point in the material on the other hand, are given by the ordinary elastic formulae of strength of materials. The concern is the beams of simple cross-section, and generally with beams of uniform cross-section. Fatigue Tests The types of stress system employed in the fatigue testing are namely; Direct stresses, Plane bending, Rotating bending, Torsion, Combined stresses. For the purpose of this research, these stresses are applied in three different fluctuating stress-time modes. Thus, fatigue ultimately occurs regardless of the magnitude of the stress. For CentrAl, the fatigue response is specified as fatigue strength, which is defined as the stress level at which failure occur for some specified number of cycles. It is the number of cycles to cause failure at a specified stress level, as

Mohamed A.A. Attia

534

taken from the S-N plot. Unfortunately, there always exists considerable scatter in fatigue data. This may lead to significant design uncertainties. The scatter in results is a consequence of the fatigue sensitivity to a number of test and material parameters that are impossible to control precisely. Impact Testing Impact tests are performed, because it is recognized that the resistance of CentrAl to shock is dependent upon factors other than those which control its resistance to a steady or slowly applied load. Resistance to a slowly applied load is measured in terms of stress, but resistance to impact involves in addition to the capacity for developing stress, the capacity of CentrAl for being deformed without damage.

ANALYSIS The development of CentrAl has been based on minimizing the fatigue crack growth rates in centre cracked specimens loaded with a mini-TWIST fatigue spectrum. The final result of this work is discussed in detail and compared whenever possible with previously published results on Glare to lay out the applicability of CentrAl concept to aircraft wing structures insensitive to fatigue [1]. The final result is a reduction of the fatigue crack growth rates with up to two orders of magnitude for the larger fatigue crack lengths. The wide range of fatigue crack growth curves that can be obtained for CentrAl is displayed in figure 1. After the specimens were pulled to failure, the fatigue damage was determined in the net cross section for all layers of the material. Reference fatigue tests were performed on 4mm thick aluminium 2024-T3.

Figure 1. reduced crack growth in CentrAl with symmetric bondpregTM

Central material concept: Towards aircraft wing structure intensive to fatigue

535

For all fatigue results in this investigation the maximum fatigue crack length occurred in one of the two outside layers of the material. So, the residual strength of the specimen is determined from the fatigue crack length observed from the specimen outside, as found during visual inspection of the material. The number of layers in CentrAl is significantly reduced relative to Glare. The thick aluminium sheets do not only build up panel thickness fast, they are also much easier to handle during pre-treatment and lay-up eliminating the chance of denting the metal, compared to the thin aluminium layers used in Glare. The thick metal sheets applied in CentrAl can be shaped to double curvature using autoclave forming techniques. So, the reduction of the width of the thick aluminium sheets of the material is not needed, unlike the use of thin aluminium layers used in double curved fuselage Glare panels. The Glare reinforcing layer in CentrAl is produced as flat sheets. Shear tests have shown that the butt splice in the Glare layer of CentrAl does not influence the shear stiffness nor the shier yield strength. Tests on static and fatigue performance derived excellent results. Towards the wing tip, there will be no need for the additional central Glare reinforcement. This can be accomplished by applying the Glare reinforcement over only approximately 50 to 70% of the wing span; an option that can not be realised for a Glare lower wing panel. The CentrAl material concept can benefit from the latest alloy developments such as Aluminium-Lithium. These alloys do not need to be rolled to the relatively thin sheet thickness needed for Glare like materials. Generation of allowables for the CentrAl concept can be based on the allowables made available for the alloy in a thickness range between 1 and 4 mm and the available allowables for Glare 2; today’s choice for the reinforcing layer, using current calculation tools; including the MVF approach. In view of the foregoing, it can be discerned that in order to further reduce the fatigue crack growth rates in Glare the issue has consumed a great deal of time and energy. It might be useful here to point out that the dearth of academics who wrote about CentrAl agreed upon three additional steps in order to further reduce the fatigue crack growth rates in Glare: The first additional step to further improve the fatigue properties is to reduce the thickness of the outer aluminum sheet. This minimum thickness is the one that can be readily obtained for the rolling process of the aluminum alloys that are believed to be most suitable for CentrAl. This is approximately 1 mm. By using several of those aluminum sheets on each side of the central Glare reinforcement, all bonded together with bondpreg layers, the total required material thickness is obtained. There is no doubt that the second material variable influencing the fatigue crack growth rates in CentrAl is the choice of the Glare type. Glare 1 has better fatigue properties in comparison with Glare 27. The influence of this difference is provided in figure 2.

Mohamed A.A. Attia

536

Figure 10 the influence of the Glare type used in CentrAl on fatigue crack growth under mini-TWIST fatigue loading All results presented above are based on the bondpregTM composition. The adhesive film in this product is applied on the bondpregTM interface towards the outside of the laminate only. The other interface of the bondpregTM; towards the centre of the laminate is comparable to the prepreg / metal interface of standard Glare and has more or less the same delamination resistance as found for the prepreg in Glare. This non-symmetrical composition gives unequal delamination sizes on the two interfaces of the bondpregTM in CentrAl. One could argue that such claims are not flawless. Clearly, the possibilities for layup are numerous, intelligently choosing the order, the number and the thickness of metal layers, bondpregTM and Glare, depending on local stress conditions including bending and considering wing panel tapering from wing root to wing tip. Further testing is recommended on the thick outer aluminum sheets preferably (2024-T3), and thus, improved fatigue behaviour and static strength will be obtained for CentrAl versions based on modern aluminum alloys.

CONCLUSIONS FML have been selected for aircraft structures because of their fatigue and damage tolerance after extensive test evidence. However, in spite of this progress, it must be admitted that fatigue and damage tolerance of aircraft structures are still problematic. Major improvement of materials is still, nevertheless, robust manifested in the very recent concept termed CentrAl. The work involving CentrAl was initiated in the U.S. aluminium giant Alcoa and materials-technology company GTM Advanced structures. Later in 2008, scientists at Delft University of

Central material concept: Towards aircraft wing structure intensive to fatigue

537

Technology in the Netherlands began to work with CentrAl material where they realised that fatigue and damage tolerance of CentrAl are still problematic. Emphasis on the work since then is on extensive test evidence. The poster shows how it is possible to predict fatigue and damage tolerance for the CentrAl concept and has put forward an experimental process to open this new arena of research. This need is of extreme importance at this point in time to narrow the gap between the development of new technology and the CentrAl technology readiness. In order for CentrAl to be included in the next generation category of fibre material, it is very much necessary that other universities start research activities on the important area of fatigue resistance that lacks knowledge.

REFERENCES [1] Roebroeks, G. H.J.J, Hooijmeijer, P.A., Kroon E.J. & Heinimann, M.B. (2007), Proceeding of First International Conference on Damage Tolerance of Aircraft Structures, TU Delft.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

PRACTICAL APPLICATIONS OF IMPROVEMENTS IN FML CRACK BRIDGING THEORY Greg Wilson,1,2 René Alderliesten,2 Riccardo Rodi,2 and H.J.K. Lemmen2 1

Alcoa Technical Center, Alcoa Center, PA, USA. Corresponding author: [email protected] 2 Delft University of Technology, Faculty of Aerospace Engineering, Chair of Aerospace Materials, Delft, The Netherlands

Abstract: Fiber metal laminates (FMLs) have excellent crack growth performance compared to monolithic metals thanks to crack bridging by intact fibers in the wake of a fatigue crack. Calculating the distribution of bridging loads in the fibers is key to analyzing and predicting the crack growth of FMLs. Most analytical approaches to modelling this phenomenon do so by imposing compatibility between the deformation of cracked metal layers and the elongation of the bridging fibers. In doing so, they assume that the crack opening displacement is equal to the displacement of the metal sheets at the boundary of the delamination between the cracked and the bridging layers. This paper derives a solution to the crack bridging problem that accounts for the deformation of the metal between the crack flanks and the delamination boundary. The results of doing so show that neglecting that deformation is acceptable for FML crack growth prediction, but the solution incorporating the exact displacement of the metal layers enables the application of the crack bridging method to a variety of additional situations as well as the extension of its applicability to more complex FMLs. This paper surveys a number of such applications, including examples of how to apply crack bridging theory to these problems.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 539–558. © Springer Science+Business Media B.V. 2009

540

G. Wilson, R. Alderliesten, R. Rodi, and H. J. K. Lemmen

NOMENCLATURE a as b E K Ktip K∞ Kbr R v∞ vbr δf δpp ν σ∞,al σmax

Crack length Saw-cut or notch length Delamination height Young’s modulus Stress intensity factor K of the cracked metal layers in an FML Portion of Ktip due to farfield metal layer stress Contribution of crack bridging loads to the reduction of Ktip Stress ratio of fatigue loading Vertical displacement due to farfield stress Vertical displacement due to crack bridging loads Elongation of reinforcing fibers Shear deformation of adhesive at delamination boundary Poisson’s ratio Farfield stress in metal layers Maximum applied cyclic stress in fatigue loading

INTRODUCTION Fiber metal laminates (FMLs) such as Glare [1], ARALL [2], and CentrAl [3], consist of alternating layers of metals and fiber-reinforced polymer composites. FMLs are known for exceptionally slow fatigue crack growth compared to monolithic metallic materials, while still exhibiting many desired properties of metals, such as formability, impact resistance, machinability, and reparability. The slow crack growth of FMLs is a result of crack bridging by the fibers. While the metal layers may develop fatigue cracks over the life of a structure, the fibers in the composite layers remain intact. Load that would otherwise contribute to the stress intensity of the cracks in the metal is borne by the fibers, reducing the stress intensity and therefore the crack growth rate. Another way of thinking about crack bridging is to consider the fibers to be resisting the opening of the cracks. Delamination between the fiber and metal layers develops around cracks due to the high degree of load transfer over this interface. As cracks grow across the metal layers, perpendicular to the main loading direction, the delamination grows in the loading direction, away from the cracks. This delamination behaviour is critical in understanding crack bridging, since large delamination size and shape reduces the effectiveness of fiber bridging. Small delaminations bridge better than large ones. Several models [4, 5, 6] have been developed to characterize the bridging effect of the fibers in FMLs. These models all employ a compatibility constraint in their

Practical applications of improvements in FML crack bridging theory

541

bridging solutions, noting that since the cracked metal layers and the intact fiber layers are attached at the delamination boundary, the deformation of the metal layers at the delamination boundaries must equal the elongation of the fibers in the wake of the crack. It is possible to calculate what the bridging loads must be in order to satisfy this condition. Once the bridging loads are known, the influence of the bridging loads on the crack tip stress intensity factor of the metal layers and on the energy release rate at the delamination boundary can be determined. These values then give a prediction for the crack and delamination growth behavior of the laminates. One feature common to these models is that they avoid an exact calculation of the displacement of the cracked metal layers at the delamination boundary. The models of Alderliesten [4] and Wu and Guo [5] use the crack opening in the compatibility solution, assuming that the additional deformation of the metal between the crack flanks and the delamination boundary is negligible. Marissen [6] solves compatibility with two bounding scenarios, one in which the crack opening is used, with the stiffness of the fibers contributing to this displacement, even in the delaminated region, and one in which the displacement at the edge of a delamination-shaped hole in the laminate is calculated. It is indeed possible to calculate the exact displacement of the metal layers along the delamination boundary. The first section of this paper will briefly explain how, then show that the decision to approximate this displacement by using the crack flank opening displacement had little effect on the outcome of the calculation. While it is therefore not necessary to go to the extra effort to include the exact calculation in these models, the exact displacement calculation adds additional versatility to these tools. The rest of this paper focuses on several examples of new practical uses for a crack bridging model using the exact displacement solution.

EXACT DISPLACEMENT SOLUTION Brief review of the Alderliesten method In order to appropriately apply fracture mechanics-based crack growth models to FMLs, the stress intensity factor of the crack tip in the metal layers, Ktip, must be determined. By superposition, the stress in the cracked metal layers can be separated into components due to remote loading and internal residual stress in the laminate, and the stress reduction caused by the bridging action of the fibers. The former is treated entirely as a remote load, and the latter is analyzed as a line load acting along the boundary of the delamination between the metal and fiber layers. Similarly, the stress intensity factor is split by superposition, giving

K tip = K ∞ − K br ,

(1)

542

G. Wilson, R. Alderliesten, R. Rodi, and H. J. K. Lemmen

where K∞ represents the stress intensity factor due to the farfield stress applied as a remote load, and Kbr is the stress intensity factor due to the bridging load. In order to determine Kbr, the distribution of bridging load along the delamination boundary must be known. Compatibility is employed as a means of calculating the bridging load. Since the metal and fiber layers are connected at the delamination boundary, the displacement there must be the same in each layer. The left side of Eqn. 2 gives the crack opening of the cracked metal layer, which Alderliesten assumes is nearly equivalent to the displacement at the delamination boundary. Here, superposition is employed to separate that due to the farfield stress, v∞, and that due to bridging, vbr. The right side of Eqn. 2 gives the displacement of the reinforcing fiber layers at the delamination boundary. The term, δf, represents that displacement due to the elongation of the fibers, and δpp accounts for shear deformation in the prepreg layer at the delamination boundary. All of the terms in Eqn. 2 are functions of x since the bridging stress, crack opening, and fiber elongation may vary along the boundary, and compatibility must hold at all x.

v ∞ ( x) − vbr ( x ) = δ f ( x) +δ pp ( x )

(2)

The vbr(x) and δf(x) terms are both functions of the bridging load, and Equation (2) can therefore be solved for the bridging load distribution. In turn, Ktip can be determined from these results, and the growth of a crack in an FML can be predicted. For the crack opening due to the farfield stress in the metal layers Alderliesten uses the following equation, from [7], in which σ∞,al represents the farfield stress in the metal layers and Eal is the metal stiffness:

v ∞ ( x) = 2

σ ∞ ,al Eal

a2 − x2

(3)

The crack opening due to the bridging load is expressed as an integral of the crack opening due to point loads along the delamination boundary: a

vbr ( x) = ∫ v( x, xP ) dxP

(4)

s

In Eqn. 4, s is the width of the saw cut, where no fiber bridging force is present, and v(x,xP) is the crack opening displacement at horizontal location x due to a point load applied at location xp. Alderliesten approximates v(x,xP) as

Practical applications of improvements in FML crack bridging theory

543

4 P ( xP ) ⎛ ⎜ tanh −1 π E ⎜⎝

(1 + ν ) b a2 − x2 + 22 2 2 2 a − x +b xP − x 2 + b 2

⎞ ⎟ a 2 − x 2 + b 2 ⎟⎠

(5)

4 P ( xP ) ⎛ ⎜ tanh −1 π E ⎜⎝

(1 +ν ) b a2 − x2 + 22 2 2 2 a − xP + b x − xP2 + b 2

a2 − x2 a − xP2 + b 2

⎞ ⎟ ⎟ ⎠

(6)

v( x, xP ) =

1

2

a 2 − xP2

for x<xP, and v( x, xP ) =

1

2

2

for x>xP. In Eqns. 5 and 6, P represents the point bridging load, normalized by the thickness of the metal sheet. This approximation combines the solutions for crack opening displacement due to point loads above and below the crack plane on the centerline and point loads on the crack flanks to the left and right of the centerline from [7] in such a way that Eqns. 5 and 6 correspond exactly to those handbook solutions in the limits of x→0 and b→0, respectively. The stress intensity factor due to the bridging loads is similarly approximated by combining the solutions for the stress intensity factor due to point loads above the crack flanks along the centerline and due to point loads along the flanks, symmetric about the centerline.

K I ( xP ) =

2P

πa

⎛ 1 ⎞ b2 1 + (1 +ν ) 2 ⎜ 2 2 ⎟ 2 2 2 a − xP + b ⎠ a − xP + b ⎝ 2 a

(7)

Because the integral in Eqn. 4 includes P, an unknown function of xP, Eqn. 2 cannot be solved explicitly for the bridging stress distribution. Instead, it is solved numerically by treating the reinforcing material as a series of “bar elements” with width, wi, height, bi, and location at the center, xi, as shown in Figure 1.

G. Wilson, R. Alderliesten, R. Rodi, and H. J. K. Lemmen

544

Figure 1 – Bar element scheme for solving the compatibility equation for the bridging stress Under this scheme, the crack opening due to bridging at a given bar element, i, is given by:

vbr ( x) = ∑ v( x, xi )wi

(8)

i

Eqn. 2 can now be written as a series of linear equations – one equation for each bar element, in which its crack opening is a linear function of the bridging load at every bar element, as well as the farfield stresses – and solved with linear algebra. Derivation of the exact displacement method It is possible to derive exact solutions for the displacement along the delamination boundary due to point loads along the boundary and for the stress intensity factor due to the same point loads, using the Westergaard stress functions given on page 5.7 of Refs. [7] and [8] 1 . These functions relate to the loading conditions depicted in Figure 2.

1

It should be noted that Ref. [8] includes extra terms in the Westergaard functions that affect the displacement, but not the stress intensity factor. Referring only to the Westergaard functions given in [7] will give erroneous results.

Practical applications of improvements in FML crack bridging theory

545

Figure 2 – Symmetric point load scenario from [8] The Westergaard stress functions describing this scenario are

ZI =

⎡ ⎤ 2 2 a 2 − z02 ⎞ ∂ ⎤ ⎢⎛⎜ a − z0 P⎡ 1 ⎟ z z z − + − 1 , , α y q ( 0 0 )⎥⎥ ⎢ ⎥ 0 ∂y0 ⎦ ⎢⎜ ( z 2 − z02 ) ( z 2 − z02 ) ⎟ 1 − a 2 π ⎣ (z) ⎠ ⎢⎣⎝ ⎥⎦

(9)

and ZI =

∂ ⎤ ⎡ −1 P⎡ ⎢1 − α y0 ⎥ ⎢ tan ∂y0 ⎦ ⎢ π ⎣ ⎣

z2 − a2 a 2 − z02

+ tan −1

⎛ z ⎞⎤ z − ⎜ tan −1 i − tan −1 i ⎟ ⎥ z0 ⎠ ⎥ z0 a 2 − z02 ⎝ ⎦ (10) z 2 − a2

with z = x + iy , z0 = x0 + iy0 , and z0 = x0 − iy0 , where (x,y), and (x0,y0) are the coordinates of the stress function and of the point of load application, respectively; with

⎧⎪ 1 (1 + ν ) plane stress ; 1 ⎪⎩ 2 ( 1−ν ) plane strain

α = ⎨ 21

(11)

G. Wilson, R. Alderliesten, R. Rodi, and H. J. K. Lemmen

546

and with

⎞ z y0 y0 1⎛ z q ( z , z0 , z0 ) = ⎜ 2 0 2 − 2 0 2 ⎟ = + . 2 2 i ⎝ z − z0 z − z0 ⎠ ( z − x0 ) + y0 ( z + x0 )2 + y02

(12)

The Westergaard functions of Eqns. 9 and 10 can be used to solve for the vertical displacement of the metal sheets using Eqn. 13 for plain stress.

vbr =

1 − 1ν+ν y Im Z I − Re Z I 2G G

(13)

Similarly, Westergaard functions can be used to find the exact displacement at the delamination boundary due to the farfield stress in the metal layers, using those for a remote biaxial stress:

ZI =

σ 1 − ( az )

2

ZI = σ z2 − a2

(14)

(15)

The displacement due to the remote load can be calculated using a formula similar to Eqn. 13, but corrected for the fact that the loading of interest is uniaxial, rather than biaxial [9]:

v∞ =

κ − 3 σ∞ ⎞ 1 ⎛ κ +1 Im Z − y Re Z − y⎟ ⎜ 2G ⎝ 2 2 2 ⎠

(16)

where κ = ( 3 −ν ) / (1 +ν ) for plane stress. Substituting the exact expressions for displacement at the delamination boundary, from Eqns. 13 and 16, into the compatibility equation, Eqn. 2, in place of Eqns. 8 and 3, respectively, allows the bridging stress to be solved with the exact displacement method. Comparison of results Figures 3, 4 and 5 show the results of an example calculation using this method. The results of a calculation using the original crack flank opening method of Alderliesten are compared to those of a calculation in which the exact solutions for vbr, Kbr, and v∞ at the delamination boundary are used.

Practical applications of improvements in FML crack bridging theory

547

The example calculation is for a laminate of Glare 3-4/3-0.4, meaning the reinforcing fiber layers are composed of one 0˚ and one 90˚ S2-glass prepreg ply each, there are four metal layers and three fiber layers, and each metal layer is 0.4 mm thick. The initial notch size, as, is 1.5 mm and the starting crack size for the calculation is 2 mm, with an initially elliptical delamination. The prediction was run with an applied loading of σmax = 80 MPa and R = 0.05. The bridging stress distributions of the initial scenario at the maximum applied load calculated with each method are shown in Figure 3. Here, the bridging stress is defined as the total bridging load carried in a given bar element divided by the total thickness of fiber layers. The bridging stress calculated with the exact displacement method is significantly higher than that calculated with the original crack flank opening method of Alderliesten along the entire delamination boundary, with the discrepancy highest at the crack tip.

Figure 3 – Fiber bridging stress calculated for initial parabolic delamination shape in crack growth prediction of Glare 3-4/3-0.4 at an applied stress of 80 MPa The prediction was carried out, allowing the crack to grow from 2 mm to 14 mm, using both the original crack flank opening method and the new exact displacement

548

G. Wilson, R. Alderliesten, R. Rodi, and H. J. K. Lemmen

method proposed here. Figure 4 shows the bridging stresses calculated with each method for the maximum applied load and the final delamination shape from the respective predictions. Again, the bridging stress calculated by the exact displacement method is consistently higher, but the differences between the two solutions are smaller. The final predicted delamination shapes are also compared in Figure 4. Since the exact displacement method calculates a higher bridging stress along the delamination boundary throughout the prediction, the delamination grows to a larger size.

Figure 4 – Fiber bridging stress calculated for the respective final predicted delamination shape of each method, after crack growth prediction of Glare 3-4/30.4 at an applied maximum stress of 80 MPa and a stress ratio of R = 0.05

Practical applications of improvements in FML crack bridging theory

549

Figure 5 – Predicted crack growth of Glare 3-4/3-0.4, σmax = 100 MPa, R = 0.05 The crack growth predictions obtained using the two methods are shown in Figure 5, compared with test data of Glare in the same configuration and loading. The exact displacement prediction is slightly more conservative than the prediction using the original method. The latter part of the test data has been transposed to the right, aligning it with the two predicted curves, to illustrate that at both the beginning and second half of the test both methods predict the behaviour of the crack accurately. The only discrepancy comes in the middle of the testing. Figure 6 compares the predicted crack growth rates using both method with test data for the above scenario, Glare 3-4/3-0.4 with a maximum applied stress of 80 MPa and R = 0.05, as well as for Glare 3-6/5-0.4 with a maximum applied stress of 100 MPa and R = 0.05. Both methods result in predictions that match the experimental crack growth rates once the cracks have grown to around 10 mm. Because of the small difference between the predictions in these regions compared to the scatter in the data, it is not possible to conclude that either method is significantly better than the other from these comparisons.

550

G. Wilson, R. Alderliesten, R. Rodi, and H. J. K. Lemmen

Figure 6 – Comparison of crack growth predictions against experimental results

In the end, the crack growth prediction obtained using the exact displacement formulas is similar to that using the original assumptions of Alderliesten, and slightly closer to the experimental results in the two scenarios examined here. The added complexity of including the exact displacement solution may not be worth enduring for the purposes of FML crack growth prediction alone. As the following sections demonstrate, however, there is a great deal more that can be done by employing the exact displacement solutions in ways beyond the crack growth of Glare.

ARBITRARY LAMINATE A fiber metal laminate is made up of a number of layers, and describing it mathematically by assuming all the metal layers have the same displacement, and all the fiber layers have the same elongation in the wake of the crack, limits the applicability of the model to situations in where this is the case. Of course, there are many situations in which it may be beneficial to consider each lamina separately. Bending of the laminate, a corner flaw in one sheet at a hole, a laminate with layers of different thickness or composition, and many more scenarios are not adequately described using Eqn 2. A more complex approach is required. Cracked metal layer between two bridging layers Consider the scenario depicted in Figure 7, in which a single cracked metal sheet is reinforced with one fiber layer on each side, but the delaminations at each interface are not identical. In this case, the assumption that the crack opening displacement can be used in place of the actual displacement at the delamination boundaries yields a poor solution.

Practical applications of improvements in FML crack bridging theory

551

Figure 7

In this scenario there are two unknown functions, the distribution of bridging load along delamination 1 that the first reinforcing layer carries and the distribution of bridging load along delamination 2, carried by the other reinforcing fiber layer. Compatibility can be imposed along each delamination boundary, since the metal sheet and the respective fiber layers must be in contact there. Therefore, it is reasonable to write Eqn. 2 twice, once for each delamination boundary, giving two equations which can be solved for the two unknowns. However, applying Eqn. 2 using only the crack opening for v∞ and vbr for compatibility means that the left side of both compatibility equations must be equal. This gives the result that the displacement of the bridging layers at both delamination boundaries, plus their respective adhesive shear deformations, must be equal. Two, more independent, equations can be written and solved if the compatibility equation (Eqn. 2) is written once for the displacement of the metal layer and the first bridging layer along the boundary of the first delamination and once for the displacement of the metal layer and the second bridging layer along its delamination. Eqn. 17 demonstrates the form this solution takes. The top equation enforces compatibility along delamination 1, and the bottom equation does so along delamination 2. Notably, since the bridging stresses along both delamination boundaries affect the displacement of the single metal sheet, the displacement due to the bridging stress transferred at delamination 2 must be included in the calculation of the deformation of the metal sheet along delamination 1, and vice versa.

G. Wilson, R. Alderliesten, R. Rodi, and H. J. K. Lemmen

552

v ∞ ( x) b ( x ) − vbr ( x, σ br ,1 ( x) )

b1 ( x )

v ∞ ( x) b ( x ) − vbr ( x, σ br ,1 ( x) )

b2 ( x )

1

2

− vbr ( x, σ br ,2 ( x) ) − vbr ( x, σ br ,2 ( x) )

= δ f 1 ( x, σ br ,1 ( x) )

b1 ( x )

b1 ( x )

= δ f 2 ( x, σ br ,2 ( x) )

b2 ( x )

+δ pp ( x)

b2 ( x )

b1 ( x )

+δ pp ( x)

b2 ( x )

(17) By discretizing the distribution of bridging stress with the same scheme shown in Figure 1, Eqn. 17 can be rearranged to solve for the bridging stress: ⎛ ⎡ v*br ( x1 , x1 , b1,1 ) b1 ( x ) ⎜⎢ ⎜⎢ # ⎜⎢ ⎜ ⎢ v*br ( xn , x1 , b1,1 ) b1 ( x ) ⎜⎢ ⎜ ⎢ v* x , x , b ⎜ ⎢ br ( 1 1 1,1 ) b2 ( x ) ⎜⎢ # ⎜⎢ ⎜ ⎢ v* ( x , x , b ) ⎜ br n 1 1,1 b2 ( x ) ⎝⎣

" v*br ( x1 , xn , b1,n ) %

#

v*br ( x1 , x1 , b2,1 )

b1 ( x )

#

" v*br ( xn , xn , b1,n ) " v*br ( x1 , xn , b1,n ) %

b1 ( x )

b1 ( x )

b2 ( x )

#

v*br ( xn , x1 , b2,1 ) v*br ( x1 , x1 , b2,1 )

% b1 ( x )

b2 ( x )

#

" v*br ( xn , xn , b1,n )

b2 ( x )

v*br ( xn , x1 , b2,1 )

" v*br ( x1 , xn , b2,n ) " " %

b2 ( x )

"

⎤ ⎥ ⎥ # ⎥ ⎥ v*br ( xn , xn , b2,n ) b1 ( x ) ⎥ ⎥ v*br ( x1 , xn , b2,n ) b2 ( x ) ⎥ ⎥ # ⎥ ⎥ v*br ( xn , xn , b2,n ) b2 ( x ) ⎦ b1 ( x )

⎧ δ * f 1 ( x1 ) ⎫ ⎞ ⎧ v ∞ ( x1 ) ⎫ − δ f 1,∞ ( x1 ) b1 ( x ) b1 ( x ) b1 ( x ) ⎪ ⎪ ⎟ ⎧ P1 ( x1 ) ⎫ ⎪ ⎪ ⎪ ⎪ ⎟⎪ ⎪ ⎪ # # ⎪ ⎪ ⎪ ⎟⎪ # ⎪ ⎪ ⎪ ⎪⎪ δ * f 1 ( xn ) b ( x ) ⎪⎪ ⎟ ⎪⎪ P ( x ) ⎪⎪ ⎪⎪ v ∞ ( xn ) b1 ( x ) − δ f 1,∞ ( xn ) b ( x ) ⎪⎪ 1 1 1 n +⎨ ⎬I⎟ ⎨ ⎬=⎨ ⎬ ⎪δ * f 2 ( x1 ) b2 ( x ) ⎪ ⎟ ⎪ P2 ( x1 ) ⎪ ⎪ v ∞ ( x1 ) b2 ( x ) − δ f 2,∞ ( x1 ) b2 ( x ) ⎪ ⎟ ⎪ ⎪ ⎪ # ⎪ ⎪ ⎪ # # ⎪ ⎪ ⎟⎪ ⎪ ⎪ ⎪ ⎟ P ( x ) ⎪ ⎪ ⎪ ⎪ ⎩ 2 n ⎭ ⎪ ⎪ δ x v ( x ) x * δ ⎟ − ( ) ( ) f 2 n n f 2, n ∞ ∞ ⎪ ⎠ ⎪ b2 ( x ) b2 ( x ) ⎭ b2 ( x ) ⎭ ⎩⎪ ⎩⎪

(18) In Eqn. 18 above, the large matrix can be divided into four quadrants. The upperleft quadrant represents the component of the vertical displacement of the metal sheet along delamination 1 due to the bridging load of delamination 1. The upperright quadrant represents the component of the vertical displacement of the metal sheet, measured along the boundary of delamination 1, due to the bridging stress of delamination 2. Likewise, the lower-left and -right quadrants give the displacement of the sheet along the boundary of delamination 2 due to the bridging loads transferred across delamination 1 and delamination 2, respectively. These components are defined as

v*br ( x n , x m , bk ,m ) b ( x ) = l

wk v br ( x n , x m , bk ,m ) b ( x ) l

t metal Pk ( x m )

(19)

where vbr is that from Eqn. 13, xn and bl(x) refer to the horizontal and vertical coordinates where the displacement is measured (in bar element n and at the height

Practical applications of improvements in FML crack bridging theory

553

of delamination l), xm and bk,m are the coordinates where load Pk is applied (at bar element m and at the height of delamination k). The wk term is the width of bar element k, and tmetal is the thickness of the central metal sheet. Also in Eqn. 18, the δ * fk ( xn ) b ( x ) terms added to the diagonal of the matrix (by multiplication with the k

identity matrix) represent the component of the elongation of fiber layer k at bar element n due to the bridging force, normalized by the bridging force. On the right of Eqn. 18 are the metal layer deformation due to the farfield stress, from Eqn. 16, and the component of fiber elongation due to the farfield stress in the fibers layers. Eqn. 18 can be solved for the unknown Pk(xn)’s, which are the bridging forces at delamination k and bar element n. Example crack bridging results for such a scenario are depicted in Figure 8. There is a clear difference between those using the adapted Alderliesten method, with crack flank opening, and using the exact displacement at the boundary layer. The calculated Kbr from each method are 570 Mpa-mm1/2 and 659 Mpa-mm1/2, respectively.

Figure 8 – Fiber bridging stress with elliptical delaminations of unequal size

554

G. Wilson, R. Alderliesten, R. Rodi, and H. J. K. Lemmen

Extending this solution to an arbitrary laminate and other practical applications Complex FMLs, with metal and fiber layers of differing thickness and composition, have been observed to have cracks of different lengths in every layer and different delamination sizes and shapes throughout [10]. Accurate modelling of such situations depends on obtaining the best estimates of bridging stresses possible. Using the exact displacement method, it is possible to apply compliance between the metal and reinforcing composite layers at every delaminated interface. In conjunction with a means of describing the extension of the bridging material in an arbitrary laminate, which can include the metal as well as fiber layers, the exact displacement solution provides a route to a complete description of the bridging problem in any fiber metal laminate.

SUPPLEMENTING DIGITAL IMAGE CORRELATION Digital image correlation (DIC) is an experimental technique which measures displacements and strain on an object’s surface [10-11]. In DIC, grid points are selected in an initial image of an unloaded test specimen. In subsequent images, when the specimen is subject to load, the positions of those grid points are tracked. By comparing the relative positions of each grid point throughout the test, the displacement at each point can be found, and the strain can be calculated. DIC allows full field surface strain measurements to be made simply. DIC has been used recently to study cracked FMLs [12], with the demonstrated capability to measure the crack tip strain fields, the delamination shape, and even the fiber bridging stress of ‘inverted’ FMLs with the fibers on the outside. Since DIC provides data on the displacement fields of FMLs, it is natural to compare DIC results to predictions of displacement made with the methods of this paper. Several comparisons were made against DIC data from cracked FMLs. First, the delamination shapes were estimated using the method described by Rodi et al. [12], in which the difference of the strain between the delaminated and undelaminated areas of the outer metal sheets makes the shape of the delamination recognizable. The measured delamination shape was used in a bridging compatibility calculation for each FML, the result of which was the predicted bridging load along the delamination boundary. With Eqns. 13 and 16, the displacements of the metal layers due to bridging and the farfield stresses, calculated using classical laminate theory for the applied loads and thermal residual stresses, were calculated and summed to get the displacement over a grid of points. In order to compare the results to the DIC measurements, in which the photos taken of the specimens under load are compared with those taken with no applied load, a second bridging and displacement calculation was made with no applied load, but with residual stresses,

Practical applications of improvements in FML crack bridging theory

555

and the displacement results were subtracted from those calculated with an applied load. Figures 9, 10, and 11 show the results of three comparisons between the predicted displacement and DIC, made by comparing displacement along horizontal lines at given distances, y, above the crack plane. In Figure 9, a comparison is made with Glare 3-2/1-0.3. In Figure 10, a comparison is made with an FML composed of three 0.4 mm thick aluminum sheets and two layers of carbon fiber composite, each made up of one prepreg with fibers in the loading direction and one with fibers in the transverse direction. Figure 11 compares the predicted and measured displacements in Glare 3-5/4-0.4. In all three cases, the predicted trends along each line agree with the DIC-measured displacement, however the exact values are not quite the same. In general, predictions along lines closer to the crack plane are in better agreement with the experimental results. Notably, the prediction fails to capture the observed behavior that, in the delaminated region, points closer to the crack plane have greater vertical displacements than those above them, meaning there is negative strain in the loading direction in this region. This discrepancy is likely a result of the way residual stress is treated in the model. Since residual stress due to curing is treated as part of the remotely applied stress of the metal layers, the relaxation of the residual stress in the delaminated region is seen in the model as an increase in the bridging stress along the delamination boundary, whereas in reality, the residual stress is introduced to the metal sheets via shear stress from the fiber layers, making relaxation of the residual stress observable as shortening of the metal, negative strain, in the delaminated region. Further work may allow the residual stress to be more realistically incorporated into the model.

556

G. Wilson, R. Alderliesten, R. Rodi, and H. J. K. Lemmen

Figure 9 – Glare 3-2/1-0.3, a = 11.7 mm, σ = 100 MPa, comparison of experimental and predicted displacements

Figure 10 – Aluminum/Carbon 3-3/2-0.4, a = 10 mm, σ = 120 MPa, comparison of experimental and predicted displacements

Practical applications of improvements in FML crack bridging theory

557

Figure 11 – Glare 3-5/4-0.4, a = 14 mm, σ = 110 MPa, comparison of experimental and predicted displacements

CONCLUSION The exact displacement solution presented in this work adds accuracy to calculations of the bridging stress in cracked fiber metal laminates; however, the similarity between the crack growth solutions with and without the exact displacement shows that the assumption of previous models, that the deformation of the metal layers between the crack tip and delamination boundary is negligible, is acceptable. The true value in the exact displacement solution comes not from improving old bridging models, but from adding flexibility to those models, allowing crack bridging solutions to be applied in new ways to a variety of situations. This work provides an overview of several such situations, with examples demonstrating how crack bridging analysis, enhanced with the exact displacement solution, can be applied to advance the study of FMLs. Each case can and should be explored in much greater depth, both to provide the confidence that the approach is appropriate and to gain further insight into ways the exact displacement method can improve the analysis of cracked structures and materials.

558

G. Wilson, R. Alderliesten, R. Rodi, and H. J. K. Lemmen

REFERENCES [1] Vermeeren, C. (2002), Around Glare: a new aircraft material in context, Kluwer Academic Publishers. [2] Wu, H. F., Bucci, R. J., Wygonik, R. H., and Rice, R. C. (1993). AIAA J. of Aircraft, Vol. 30, no. 2, pp. 275–282. [3] Roebroeks, G. H. J. J., Hooijmeijer, P. A., Kroon, E. J., and Heinimann, M. B., (2007). In: First International Conference on Damage Tolerance of Aircraft Structures, Benedictus, R., Schijve, J., Alderliesten, R. C., and Homan, J. J. (Eds.), Delft, The Netherlands. [4] Alderliesten, R. C. (2007), Intl. J. of Fatigue, Vol. 29, pp. 628-646. [5] Wu, X. R. and Guo, Y. J. (2002), Fatigue Fract. Eng. Mat. Struct., Vol. 25, pp. 417-432. [6] Marissen, R., Fatigue crack growth in ARALL. A hybrid aluminium-aramid composite material: Crack growth mechanisms and quantitative predictions of the crack growth rates, Ph.D. thesis, Delft University of Technology, Delft, The Netherlands, 1988. (available at http://repository.tudelft.nl/file/755663/375263) [7] Tada, H., Paris, P.C., Irwin, G.R. (2000), The Stress Analysis of Crack Handbook, 3rd ed., The American Society of Mechanical Engineers, New York. [8] Tada, H., Paris, P.C., Irwin, G.R. (1973), The Stress Analysis of Crack Handbook, 1st ed., Del Research Corporation. [9] Sun, C. T., Farris, T. N. (1989), Intl. J. Fract., vol. 40, pp. 73-77. [10] Lemmen, H.J.K., Alderliesten, R.C., Benedictus, R., Hofstede, J.C.J., and Rodi, R. (2008), The power of Digital Image Correlation for detailed elasticplastic strain measurements. EMESEG ’08 conference, Engineering mechanics, structures, engineering geology, Crete. [11] Po-Chin Hung and A.S. Voloshin, In-plane strain measurement by Digital Image Correlation, J. of the Braz. Soc. of Mech. Sci. & Eng., Vol 25, no.3, 2003. [12] Rodi, R., Alderliesten, R. C., and Benedictus, R. (2009), In: Design for durability in the digital age, Proceedings of the 21st ICAF Symposium.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

SIMULATION OF THE STRESS DISTRIBUTION ON FUSELAGE STRUCTURES FOR THE PRE- AND POST ANALYSES OF CURVED PANEL TESTS Matthias Ziegenhorn1, Frank Schulze1, Holger Sparr1, Karsten Wenke1 and Thomas Fleischer2 1 2

Hochschule Lausitz IMA GmbH Dresden

Abstract: The testing equipment for fuselage panels of aircrafts at the IMA GmbH Dresden is designed for fatigue tests as a preliminary stage to the full aircraft fatigue test. The natural disturbance at the boundaries of these panels is going to be minimized due to a well-adapted test rig design. The main advantage of this testing scenario lies in the lower testing effort and is therefore used to assess new design or material concepts. A numerical simulation of this process by means of a finite element (FE) analysis is the most important tool to model the actual setup and to adjust the loads in terms of a varying panel configuration. The objective in this project is to extend the current testing strategy to panel configurations with an open set of geometrical parameters. The consideration of cut-outs or new materials as well as a new variety of fuselage diameters and non-circular crosssections gets into focus. The results coming from these analyses can be used to draw conclusions for the loading situation of future detailed structure tests having refinement and sub-modelling strategies of an FE model in mind.

INTRODUCTION The current state of the testing setup was designed with respect to cylindrical symmetry of the fuselage for panels which can be subjected to internal pressure, M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 559–567. © Springer Science+Business Media B.V. 2009

560

Matthias Ziegenhorn et al.

torsion and longitudinal tension/pressure. The stress state in a fuselage panel produced by pure internal pressure is then dominated by the membrane stresses in the skin. The symmetry assumption simplifies the requirements to complete setup in terms of loading and geometrical boundary conditions. Nonetheless the noncylindrical boundary condition effects in the setup must be reduced by inserting sophisticated additional elements to increase the effective measurement area on the panel.

Figure 1: Principle sketch out of an FE mesh The applied elements are characterized by a coordinated interaction and can be identified in the FE-plot of Figure 1. The used pressure box has a small bending stiffness about the aircraft y -axis and a small stiffness in the aircraft x direction. To achieve a membrane stress dominated state horizontal correction forces are applied at the pressure box close to the panel interface in a self-equilibrated manner. The resulting bending effect within the panel gets adjusted by passive elements at the frame positions. The pressure box is closed at both ends with basically rigid bulkheads. Other concepts of testing fuselage panels can be found in [1] and [2]. The described principle provides a stress field in a relatively large area of the panel which is close to the expected one of the top fuselage section in a complete undisturbed loading situation. The geometrical parameters of the mentioned elements were identified by analytical calculation and FE simulations. Also the measurements of panel strains and displacements were verified by FE simulations. To assess the limits of the applied principles and assumptions one can subject panels with an asymmetric design like window panels to the current set-up (see Figure 2).

Simulation of the stress distribution on fuselage structures

561

Figure 2: Testing rig with window panel The obtained results in FE simulations suggest that the available parameter set is sufficient to control the stress state in the panel since the resulting asymmetry is still moderate.

FULL MODEL ANALYSIS The asymmetry increases dramatically when you take curvature changes or larger cut-outs into account. While the cut-out affects the stress state in its vicinity, the curvature change is responsible for a completly different deformation state which can be seen in Figure 3 for a barrel simulation with a semi-realistic door cut-out and floor structure loaded with internal pressure. The barrel is characterized by two different radii in the top and bottom section and a transition region below the floor.

Figure 3: Displacement plots for a FE simulation of a barrel with cutout and curvature change

Matthias Ziegenhorn et al.

562

Going straight forward in developing a testing strategy for panels we have to consider only the part of the barrel of interest – the section around the door. By employing the cut boundary interpolation feature in ANSYS we used a simple and effective tool to analyze the reaction forces and moments at the new boundary which we refer to as the submodel. When we concentrate resulting reactions at the frames and neglect the reactions (forces in x-direction and moments about y resp. z-axis) which would cause a displacement out of the frame plane (s. Figure 4), we achieve a simplified submodel. s=0mm Frame 1

s=4760mm

Frame 4 Simplified loading

Figure 4: Submodel and simplified loading condition In Figure 5 and Figure 6 the circumferential skin stress in the top layer can be seen at two different frame positions. The closer the frame is positioned to the door the more the general stress state gets influenced by the localized effects with respect to the loading situation (door joints), wall thickness profile in the skin and adapted stiffeners. The drop of the circumferential stress in Figure 5 is dominated by the overall bending state. In Figure 6 the circumferential stress in the top layer lies within a relatively narrow range. Both submodels show a reasonable agreement with the barrel simulation results which holds true for the other components of the stress tensor.

Simulation of the stress distribution on fuselage structures

563

140 120

Submodel simplified Submodel

100

Barrel

[N/mm²]

60

σ

80

20

40

0 -20 -40 -60 0

500

1000

1500

2000

2500

3000

3500

4000

4500

s [mm]

Figure 5: Circumferential skin stress along the frame 1 position 200

[N/mm²]

100

σ

150

50 Submodel simplified Submodel 0

Barrel

-50 0

500

1000

1500

2000

2500

3000

3500

4000

4500

s [mm]

Figure 6: Circumferential skin stress along the frame 4 position

SIMPLIFICATION TO A TWO-DIMENSIONAL OPTMIZATION APPROACH Especially the stress state in frame 4 position justifies a further simplification with respect to an optimization approach. As a first step we consider a two-dimensional problem. Therefore, it is assumed that a stress state which does not depend on the

Matthias Ziegenhorn et al.

564

aircraft longitudinal axis and can be modelled with beam elements. The loading consists of pure internal pressure. The analysis of the simplified submodel with a loading based on resulting reactions in the frames suggests the introduction of simplified compensation elements. These compensation elements consist of a pressure box (also as a beam representation) and various loading parameters acting on selected beam positions. The exclusion of the floor preserves our optimization approach at a first step to deal with singularities in the stress resultant function along the circumference. The input for the objective function in the optimization procedure is retrieved by two-dimensional barrel solutions with different properties (see Figure 7).

non-homogeneous cross-section and material properties

homogeneous cross-section and material properties

Figure 7: Two different bending moments along the circumference due to different stiffness properties The development of a bending state is caused by the already mentioned curvature change. To show the optimization capability we selected a region with a bending moment sign change (depicted in Figure 7). For the actual calculation we used the FE package ANSYS again. As the optimization technique the build-in first order gradient method produced the best results and showed the most robust behaviour with respect to the starting parameter set. Four major configurations with a different set of design parameters are used to show some details of the results obtained.

RESULTS In Table 1 the bending state in the two-dimensional models for a homogeneous property distribution is summarized. The green part on the left stands for the panel properties and the blue section is connected to the pressure box properties. With variation 1 we reflect the current state of the testing rig with three design parameters. An additional moment can be introduced with a third horizontal force.

Simulation of the stress distribution on fuselage structures

565

1

2

3

4

Table 1: Variation type and optimized bending state Allowing different moments and forces at the panel boundaries 7 design parameters can be adjusted. Leaving the horizontal force balance untouched we can generate a moment by two separated lines of actions, which leads likewise to a parameter set of 7 variables in variation 3. The last variation operates with two additional force components having a design space of 8 variables.

566

Matthias Ziegenhorn et al.

1

2

3

4

Table 2 Variation type and optimized bending state for a nonhomogeneous panel Table 2 contains the optimization results for a modified panel setup where the thinner section symbolizes a stiffness reduction. In both cases the first two variations do not meet the bending state calculated by the barrel solution and show

Simulation of the stress distribution on fuselage structures

567

high bending moments in the pressure box. In variation 3 and 4 the stress state of the barrel solution is almost identically fulfilled with the illustrated pressure box model. The lowest stress resultants are produced with variation 4 in both cases.

CONCLUSION AND PROSPECT This work shows that a straight forward simplification from a barrel solution to an easily optimizable two-dimensional beam representation is employable. Two variants can map the objective bending state in the panel almost exactly, while the other two set-ups show larger deficiency. The successful mapping of the bending state has two complete different effects of the stress situation in the pressure box. The mechanical properties of the pressure box in terms of a functional stiffness distribution and geometry design comes into focus. Subsequently the applied moments and forces need to be reflected into a design for the testing rig. We like to mention two different aspects of future work. Firstly, we want to examine the limits of this approach having qualitatively different bending states in mind. And secondly, the extension to a three-dimensional optimization procedure is of major interest where the presented results serve as basic starting parameters for individually optimized frame sections.

ACKNOWLEDGEMENT This work is AiF funded by the BMBF of Germany.

REFERENCES: [1] L. le Tellier, F. Repiton (2007) In: Full-scale testing and analysis of Falcon 7X curved fuselage panels with Butt-joints, Proceedings of the 24th ICAF Symposium, vol. I, p. 340, L. Lazzeri and A. Salvetti (Ed.), Publ. Pacini, Naples [2] D. R. Ambur, J. A. Cerrof, J. Dickson (1995) In: D-Box Fixture for Testing Stiffened Panels in Compression and Pressure, AIAA Journal of Aircraft, vol. 326, p. 1382

25th ICAF Symposium – Rotterdam, 27–29 May 2009

LOW CYCLE LIFETIME PREDICTION OF AL2024 ALLOY A. Vyshnevskyy1, S. Khan2, and J. Mosler3 1,2,3

GKSS Research Centre Geesthacht GmbH, Max-Planck-Strasse 1, 21502, Geesthacht, Germany

Abstract: The 2024-T351 aluminum alloy is extensively used for fabricating aircraft parts. This alloy attains relatively low ductility at room temperature and is generally heat treated in various conditions to suit particular applications. The present study experimentally and numerically analyzes the damage mechanism of an Al2024-T351 plate (short transverse direction) subjected to multi-axial stress states. The purpose of this work is to predict the cyclic lifetime of the considered alloy, based on the local approach of damage evolution using continuum damage modeling (CDM). The experimental program involves different kinds of specimens and loading conditions. Monotonic and cyclic tests have been conducted in order to measure the mechanical response and also to perform micromechanical characterization of damage and fracture processes. The cyclic plasticity behavior has been characterized by means of smooth cylindrical specimens. For analyzing the evolution of plastic deformation and damage under multi-axial stress conditions, cyclic loading tests in the low cycle regime have been conducted on different round notched bars. The predictions of the CDM were compared to the experimentally observed mechanical response and to the micromechanical characterization of damage. Emphasis was laid on the prediction of the number of cycles to failure.

INTRODUCTION Low cycle fatigue (LCF), in contrast to high cycle fatigue (HCF), is generally characterized by failure in less than 104 cycles showing a pronounced plastic mechanical response. These loading conditions can occur in airplane structures M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 569–583. © Springer Science+Business Media B.V. 2009

570

A. Vyshnevskyy, S. Khan, and J. Mosler

during or after unpredictable mechanical impacts e.g. hard landing, bad weather conditions, operational errors, failure of structural integrity etc. Today’s airplane design is based on numerous tests on specimens of various sizes for the estimation of the residual strength of certain components such as the fuselage. These tests programs are time consuming and expensive. The development of predictive models for ductile crack initiation is thus essential for improving structural design and maintenance, and it accelerates the introduction of new materials or assembling methods. Crack initiation as the result of LCF is a complex process that is influenced by many factors: stress (strain) history, rate of loading, environmental influence, temperature, hold times etc. Apart from the influence of these extrinsic factors, numerous features of the micro-structure of the material undergoing cycling loading influence crack initiation. Fatigue damage in ductile metals and alloys usually is associated with plastic deformation. Therefore, it is reasonable to examine stress singularities as possible locations showing material yielding which leads finally to crack initiation. Such locations can be metallurgical accidents like casting pores, impurity inclusions of even, irreversible microstructure changes due to manufacturing process. The CDM approach used within this paper was proposed by Lemaitre (1985). It considers the evolution of the thermodynamic state variables associated with plastic and damage mechanisms. Conceptually, damage affects the mechanical response through the degradation of the stiffness. Clearly, for complex engineering components and structures subjected to cyclic loading this is a highly challenging task. In this case the design and validation of damage evolution laws are required. Particularly, the influence of multi-axial loading on the fatigue life of materials and structures under operational conditions is of utmost importance. This paper is devoted to the extension and validation of the CDM approach by Lemaitre and Desmorat [1] to aluminium alloys. A very common aluminium alloy used for years in the aerospace industry is studied here as a reference (Al2024). This alloy is preferred because of its good damage tolerance. The work includes both experimental and simulation parts. The experimental program involves different kinds of specimens and testing conditions: ¾ Monotonic tensile tests to investigate the ductile damage and fracture behavior. The respective results serve as a reference for the cyclic mechanical response and the respective micromechanical characterization. ¾ Cyclic plasticity tests on smooth round bars (Damage Low Cycle specimens (DLC)) and round notched bars (RNB) with different notch geometries to study plasticity and damage evolution under simple and multi-axial stress conditions. The contribution concerns the comparison of the results predicted by the aforementioned CDM to experiments. Special emphasis is laid on the prediction of the number of cycles to final failure including the locations of crack initiation.

Low cycle lifetime prediction of AL2024

571

MODELLING Axi-symmetric FE models of the specimen geometries were developed using 4noded standard elements (standard integration). Symmetry conditions are imposed in order to increase computational efficiency. The mesh size in the region of expected failure is of the same order as a representative volume element in metals [1]. The simulations are run using ABAQUS v6.7 software with Z-Mat as material library. The constitutive behavior of the material was modeled by means of the CDM of Desmorat-Lemaitre [2]. This CDM is based on rate-independent elastoplasticity (isotropic and kinematic hardening) combined with anisotropic damage evolution (with micro-defects closure effect). Further details can be found in [2]. For the sake of comprehensibility, an overview of the governing equations is presented in the next sections. Part I: Elasto-plasticity The model is based on the assumption of additive decomposition of the strains into elastic and plastic parts. More precisely,

(

)

~ = C : ε e = C : ε − ε p , σ

(1)

~ - is the effective Cauchy stress tensor (rate) and C - is the 4-th order where σ stiffness tensor.

A classical von Mises flow rule is adopted. The evolution of isotropic hardening and that of the back stress are governed by an exponential law and an ArmstrongFrederick relation respectively [1, 2]. Part II: Damage evolution For the description of anisotropic damage evolution, a 2-nd order damage tensor D is introduced. Without going too much into details, the effective stresses are computed from

(

~ = H pσ D H p σ +

) + (H σ D

n

H ijP = (1 − D)ij

D −

Hn

−1 2

)

D

⎡ σH − σH ⎤ +⎢ − ⎥I , ⎣⎢1 − ηDH 1 − ηha DH ⎦⎥

and H nij = (1 − ha D)ij−1 2

(2) (3)

where σ D+ , σ −D - are the positive and the negative part of the deviatoric stresses (based on eigenvalues); σ H - is the hydrostatic stress; • - are McCauly brackets;

A. Vyshnevskyy, S. Khan, and J. Mosler

572

η and ha - are the closure effect and hydrostatic damage sensitivity parameters respectively. The damage evolution law is a function of the plastic strain rate ε p and the effective elastic energy density Y : s

⎛Y ⎞ D ij = ⎜⎜ ⎟⎟ ε p , ij ⎝S⎠

(3)

Clearly, Y depends further on the stress state (tri-axiality).

An important consideration within the aforementioned damage framework is a threshold value of damage initiation. It is assumed, that damage starts only after the plastic stored energy [3] t

ws = ∫ R∞ (1 − exp(−br ) ) 0

A (1− m ) m 3 2 r rdt + X m 4C

(4)

reaches the value wD and continues until one of the eigenvalues of the damage tensor D reaches the critical damage Dc . In Eqn. 4, r is the accumulated plastic strain, X is the back stress tensor, A, m are parameters defining the plastic stored energy function, and R∞ , b, C are additional parameters describing the cyclic plasticity model. The mechanical response during the strain controlled symmetric cyclic loading can be divided into two stages: ¾ Initially, the gradual increasing of the stress peaks is observed (see Fig. 6b). It continues till the maximum value. In line with other metallic materials, stabilization of the cyclic mechanical response is observed. At the end of this stage ws = wD is fulfilled. The respective number of cycles is denoted as N D (Fig. 6b). ¾ During the second stage, the stress peaks start to decrease and mechanical degradation is observed (Fig. 6b). This stage is associated with the initiation and progress of damage ( 0 ≤ D ≤ Dc ). The number of cycles during this stage is denoted as N E . Beyond, these stages the (meso-)macro-crack in the critical region will be initiated. The modeling of crack extension was not considered in the present framework. Summarizing the aforementioned stages, rupture occurs at

Low cycle lifetime prediction of AL2024

573

NR = ND + NE

(5)

number of cycles (Fig. 6b). Reference modeling with the Coffin-Manson relation A standard representation of LCF life time for uni-axial cyclic loading is the relation between plastic strain amplitude Δε p and number of cycles to rupture N R (in logarithmic scale).This results in the so-called Coffin-Manson relation: ⎛C N R = ⎜ MC ⎜ Δε p ⎝

⎞ ⎟ ⎟ ⎠

γ MC

(6)

where CMC and γ MC are material constants [2, 4].

EXPERIMENTS The material under consideration is Al2024-T351. Specimens have been taken from the thickness direction, also referred to as the S-direction. The material was supplied in form of a rolled sheet in the T351 temper (solution heat treated, airquenched, stress-relieved by cold stretching). The strong anisotropy of the microstructure, caused by the manufacturing process, almost does not affect the mechanical properties but impacts the fracture resistance significantly. The main alloying element in the 2XXX series is copper, while magnesium and manganese are introduced in order to improve the quenching properties. The material composition is given in Table I below: Element Weight percentage

Cu

Mg

Mn

Si

Fe

Cr

Al

4.11%

1.12%

0.46%

0.048%

0.05%

0.003%

Rest

Table I: Chemical composition of Al2024 alloy Mechanical testing Three types of specimens were tested (see Fig. 1). Initially, a so-called DLCspecimen (Damage Low Cycle), was designed for tension-compression tests showing a stress state close to the uni-axial one (see Fig. 1a). Subsequently, a circumferential round notch was designed to achieve a variation of the stress state in the middle cross section of the specimen (see Fig. 1b). All specimens were cut from the same plate of Al2024-T351 in S direction. The entire specimens were

A. Vyshnevskyy, S. Khan, and J. Mosler

574

grinded and mechanically polished in the direction of loading. The final polishing was performed with a diamond paste showing 1 μm grains. b)

32.5

100 35

Ø 6.5

M 20

Ø 18

M 20

R

10

26

26

100 27.5

R

a)

R

Figure 1: Specimens geometries: (a) DLC-specimen; (b) specimens with circumferential notches R = 10 mm and R = 2 mm The entire experiments were conducted on a 60 kN servo-hydraulic testing machine. Specially designed gripping of the testing setup and shoulders of the specimens allow for tension-compression reverse loading. The loading in all cases was performed displacement-controlled. The displacement was measured using extensometer, attached to the working distances of specimens. A triangular wave shape symmetric cycle ( R = −1 ) was imposed with various amplitudes and identical time period T = 100 s . The amplitudes were varied to achieve different life times (number or cycles to failure). All tests were conducted until complete rupture. Time-force-displacement responses during the tests were recorded with a digital acquisition device. Micro-structure investigations In order to successfully predict the life time of a material, it is important to understand the damage evolution. Damage in Al-alloys usually nucleates at large and brittle intermetallic particles (cluster of particles) or coarse precipitates, and sometimes also at smaller particles such as dispersoids or precipitates [5-7]. In the specific case of fracture in Al-alloys, damage is essentially induced by the breaking or de-bonding of coarsened precipitates located inside precipitate free bands (PFB) [8]. Optical microscopy, scanning electron microscopy and X-ray microtomography have been used to obtain quantitative information [9] about microstructural characteristics. The particles are aligned in a network-like structure which separates the matrix into domains. The latter have a disc-like shape, with the S-direction as the shortest axis and almost identical dimensions in L and Tdirections. When loaded in S-direction the damage rate is higher in the PFB. A macroscopic crack can extend throughout the material perpendicular to the loading direction. To systematically characterize fatigue fracture, some monotonically fractured surfaces were compared to cyclic ones. 2D and 3D snaps of the microstructure have been extracted. Monotonous fracture reveals a dense structure, which is characteristic of a ductile failure. Some multiple cracks could also be seen on the surface of the notch (Fig. 2a). Slight necking in the notch root is also observed. By way of contrast, in

Low cycle lifetime prediction of AL2024

575

cyclic loading large dimples appear near the edge of the specimen, referring to ductile failure originating from the surface. Fatigue lines are highly visible. a)

b)

Figure 2: SEM of fracture surface close to the notch root of (a) monotonically loaded specimen, and (b) cyclically loaded specimen (703 cycles) In the monotonic case, the fracture surface goes through the particles of level I, just under the fractured surface in the longitudinal polished specimen (Fig. 3a,b). The density of the broken particles is high in the middle of the specimen (Fig. 3a). a)

b)

Figure 3: SEM of longitudinal micro-sections close to the fracture surface of the monotonically loaded RNB specimen (a) middle of the specimen, (b) region close to the notch root Imaging software combined with SEM was used to study the 3D shape of the fractured surfaces. Compared to monotonic experiments, the crack initiation behavior changes in the LCF-regime in a qualitative way. In Fig. 4a, it can be seen that the fracture initiates at various sites, creating an uneven structure. This kind of ductile damage occurs at various stages within the material’s layers. Finally, the small scattered micro-cracks coalesce and form a macro-crack. The bright planes (shear lips) in Fig. 4a represent the sudden shearing failure. In the LCF-regime (Fig. 4b), the mechanisms change to mixed-mode failure. Consequently, uneven ductile failure areas, fatigue lines, striations and bright planes are visible. Additionally, some micro-cracks can be seen near to the notch root. Failure starts within clusters of particles. Subsequently, with increasing accumulation of plastic strain, dimple structures appear and finally, all small cracks coalesce.

A. Vyshnevskyy, S. Khan, and J. Mosler

576

In this alloy, the heterogeneous particle distribution and large grain size cause long free paths for dislocations forming persistent slip bands. The deformation is localized in these bands, enabling irreversible deformation, and causing crack initiation mostly at the slip bands and at the interfaces between inclusions and the matrix after a high cumulative plastic strain [10-12]. b)

a)

Figure 4: 3D re-constructed fractured surface (a) monotonous and (b) cyclic Evidently, a complete understanding of the aforementioned mechanisms requires particularly a careful analysis of damage evolution at the micro-scale. Consequently, non-destructive micro-structural techniques such as synchrotron Xray micro-tomography (XTM) have also to be applied. During the last 10 years, significant progress has been made in terms of resolution. More precisely, the spatial resolution is now close to that of an optical microscope ( 2 − 3 μm ). In the field of damage/fracture characterization, micro-tomography has been used for visualizing cracks in metals also for Al-alloys [13-16]. a)

b)

c)

Figure 5: Section of reconstructed XTM 3D-image: (a) as a full solid, (b) as a semi visible volume with the cracks, and (c) projection of the crack front on the plane perpendicular to the axis of specimen In Fig. 5, first results obtained from XTM are shown (geometry: inner diameter 4 mm , notch radius 4 mm ). With the help of SEM and XTM micrographs, the results of the CDM reported in the next sections can be proven (particularly, the evolution of damage), cf. [17]. So far, the XTM measurements have been performed after loading the specimen. In the future, true in-situ investigations will be conducted. By doing so, the evolution of damage mechanisms can be investigated at a fine temporal and spatial resolution.

Low cycle lifetime prediction of AL2024

577

SIMULATION RESULTS Material parameters for the plasticity model For the simulation of the cyclic mechanical response, a combined isotropickinematic hardening plasticity model is used. Since no damage occurs within the first N Dexp cycles (see Fig. 6a), the material response is purely elasto-plastic (without damage). The material parameters were adjusted using experimental cyclic stress-strain curves of DLC-specimens (round bar specimen) with the experimental response for Δε = 0.02 . The resulting material parameters are summarized in Table II. Elasticity Plasticity γ E,[MPa] σ y ,[ MPa ] R∞ ,[ MPa ] ν b 5 7·10 0.3 284 150 4 80 Table II: Elasto-plastic parameters model

Δεp -0.01

0

-600

0.01 Strain, [-] Δε

b)

1.7·104

600 Stress peaks, [MPa]

600 Stress, [MPa]

a)

C,[MPa]

400

N exp D N exp R

N exp E

200 tension (experiment) compression (experiment) CDM (simulation)

0

0

50 100 150 Number of cycles, [-] Figure 6: (a) Mechanical hysteresis and (b) Experimental vs. simulated absolute maximum value of tension and compression stresses in a particular strain loading amplitude, DLC-specimen ( Δε = 0.02 ; Δε p = 0.0063 ; N Rexp = 138 ) Since a von Mises flow rule is adopted, the tension-compression asymmetry is not captured. However, this can easily be improved by using a more realistic yield function. Material parameters for the damage model The material parameters are calibrated starting with those for the evolution and the threshold value of the stored plastic energy function Eqn. 4. Frequently, an analytic method is applied, cf. [18]. This method is based on the explicit calculation of

A. Vyshnevskyy, S. Khan, and J. Mosler

578

function values for different number of cycles. It is assumed, that the isotropic hardening function reaches the saturation prior to damage initiation. This is a common observation in ductile steels. Al2024-T351 does not exhibit the stabilization and thus other parameterization methods have to be used. In the present work the following objective function [19]

{A, m, wD }* =

(

)

⎛M ⎞ arg min ⎜⎜ ∑ log N Dexp i − log N Dsim i ( A, m, wD ) ⎟⎟ {A, m , w }∈R ⎝ i =1 ⎠ 3 +

D

(2)

is minimized, where N Dexp i and N Dsim i are the experimentally observed cycles and the numerically computed counterparts; M is the number of experiments (with different loading amplitudes). The problem is solved by implementing an algorithm in MATLAB. The final parameters are summarized in Table III: Stored plastic energy 3

3

Damage η s

ha Dc wD ,[ MJ / m ] S ,[ MJ / m ] A 3.9 0.0113 0.897 1.3 1.0 2.8 0 0.1 Table III: Parameters of the model governing the evolution and the threshold values of the stored plastic energy and damage evolution m

The choice of the damage parameters is motivated by the macroscopic material behavior. In several experiments with DLC-specimens no degradation of the mechanical properties was observed. The decreasing of stresses showed an abrupt or “quasi-brittle” character. A fast evolution of damage and a small value of critical damage Dc suit such a material behavior well. The parameters for abrupt fracture and gradual degradation of the material behavior (e.g. Fig. 6b) are given in Table III. The damage parameter was calibrated without any optimization technique. Simulation of fatigue life time of DLC-specimens Fig. 7 shows the results of FEM simulations performed on one single 3D element with the set of CDM-parameters taken from Tables II, III. For the sake of comparison, experimental results, together with the Coffin-Manson model, are shown as well. The parameters of the Coffin-Manson model are CMC = 0.196 and γ MC = 1.4 .

Low cycle lifetime prediction of AL2024

579

Δεp, [-]

0,1

0,01 experimental (DLC) Desmorat-Lemaitre Coffin-Manson

1E-3

1

10

100

1000

NR, [-]

Figure 7: Experimental vs. simulated fatigue response of DLC-specimens According to Fig. 7, the CDM fits almost perfectly the experimental data. For all loading amplitudes the agreement between the experimental observations and the numerically predicted response is remarkable. Lifetime prediction An example of a typical force-displacement and force-time response of the notched specimen is presented in Fig. 8.

Figure 8: (a) Stress-strain and (b) Stress-time responses of the notch specimen ( 2 mm notch, Δl = 0.02 mm ). Point 1 indicates the formation of a crack on the surface of the specimen The curves are similar to those of the DLC-specimen. The two stages of material hardening and degradation as mentioned formerly are evident. In contrast to the DLC-specimens, the stage II is significantly longer. This requires further discussion.

A. Vyshnevskyy, S. Khan, and J. Mosler

580

The results of FE simulations (not included in the present article) reveal that with decrease of loading amplitude and increase of notch curvature the plastic deformation will be localized in the notch root. At the same time the remaining volume of the specimen deforms purely elastically. If loading is further increased, this leads to crack propagation in an almost fully elastic specimen. Thus, from a structural point of view, this problem is mostly related to fracture mechanics. Evidently, the aforementioned CDM is not well suited for the aforementioned cyclic crack propagation. Therefore, crack initiation (point 1 in Fig. 8b) is chosen as the state defining N R , see Eqn. 5. The results of the FEM simulation are compared to those of the experiments in Fig. 9. b)

Force peaks, [kN]

Force peaks, [kN]

a)

40

20 10 mm notch Δl = 0.375 mm

0

0

10 Number of cycles, [-]

10 mm notch Δl = 0.15 mm

0

1000 2000 Number of cycles, [-]

d)

Force peaks, [kN]

Force peaks, [kN]

20

0

20

c)

40

40

20 2 mm notch Δl = 0.2 mm

0

40

20 2 mm notch Δl = 0.16 mm

0 0

10 20 Number of cycles, [-]

30

0

50 100 150 Number of cycles, [-]

Figure 9: Experimental (□ - tension, ○ - compression) vs. simulated (  CDM) force peaks. The simulation presented in (b) was stopped due to saturation of stored plastic energy function (2000 cycles) The results, presented in Fig. 9, show the mechanical response for extreme combinations of notch geometries and loading amplitudes. The best prediction was achieved for the 10 mm notch and the highest loading amplitude (Fig. 9a). Although the numerical results are very promising, some differences compared to the experimental data can be seen. Two possible reasons for this are the calibration

Low cycle lifetime prediction of AL2024

581

method, i.e., the material parameters have been adjusted by using DLC-specimen and the problem associated with the modeling of crack propagation for cyclic loading (see point 1, Fig. 8b). Additionally, further investigations concerning the limits of the model (small strain amplitude, high stress tri-axiality) are required. An interesting loading case is observed in Fig. 9a. There the amplitude of the imposed displacement is so small, that the accumulated plastic strain concentrated at the notch root has saturated after some cycles. This is related to the redistribution of stresses in the vicinity of the notch root. The saturation of accumulated plastic strain causes the saturation of stored plastic energy and as the result the threshold value cannot be achieved in any number of cycles ( ws < wD ). The comparison of the predicted fatigue life time is summarized in Table IV: Notch radius, [mm]

Δl ,[mm]

N Dexp

N Dsim

N Rexp

N Rsim

10 0.375 14 15 15 16 10 0.200 204 378 431 383 10 0.150 2010 -* 2345 -* 2 0.20 19 22 26 24 2 0.18 36 43 55 45 2 0.16 63 79 122 84 Table IV: Comparison of the simulation results to the experimental data for the notched specimens. * - stopped after 2000 cycles Two damage distributions in loading direction at the rupture time point N = N R are shown in Fig. 10: a)

b)

Figure 10: Distribution of damage in loading direction at the rupture time point ( N = N R ). (a) Notch radius 10 mm , N R = 431 ; (b) Notch radius 2 mm , N R = 55 The region of crack initiation is predicted very accurately. It is in excellent agreement with similar investigations for ductile steels [20]. The major difference as compared to the distribution of damage in steel is the pronounced localization of damage in the notch root.

582

A. Vyshnevskyy, S. Khan, and J. Mosler

CONCLUSIONS Low cyclic behavior of Al2024-T351 alloy was investigated using mechanical testing, micromechanical characterization of fracture surfaces, and X-ray micro tomography for various specimen geometries and loading regimes. The predictive capability of the Desmorat-Lemaitre CDM model [1] was analyzed for high load reversals under various stress states and life times. The results of the simulations with CDM have shown significantly better descriptive capability as compared to the classical approaches, especially for low number of cycles. Even for complex geometries, the lifetime prediction is in good agreement with experiments. It was shown, that with elevated tri-axiality (curvature of notch) and number of cycles to rupture the initiation of damage will be over estimated and the number of cycles to rupture will be underestimated. Further investigations in this direction are necessary for lifetime predictions of complex structural components.

ACKNOWLEDGMENTS The authors gratefully acknowledge D. Steglich, J. Herrens for useful advises, J. Knaack for mechanical testing, P. Fischer for metallography and V. Ventzke for SEM-analysis.

REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9]

Lemaitre, J. and Chaboche, J.-L. (1987), Mechanics of solid materials, Oxford University Press. Lemaitre, J. and Desmorat, D. (2005), Engineering Damage Mechanics, Springer, Heidelberg. Chrysochoos. A. (1987), Dissipation et blocage d’energie lors d’un ecrouissage en traction simple, Thèse d’état de l’université, Paris. Coffin, L.F. (1986), Jour. Vibration, Acoustics, Stress and Reliability in Design, Trans. ASME, vol. 108, pp. 241-248. Pardoen, T. and Pineau, A. (2007). In: Comprehensive structural integrity encyclopedia, Chapter 6, vol. 2., Elsevier, Amsterdam. Lassance, D., Fabrègue, D., Delannay and F., Pardoen, T. (2007), Prog. Mater. Sci., vol. 52, pp. 62-129. Garrett, G.G. and Knott, J.F. (1978), Metall. Trans. A, 9A, p. 1187. de Hass, M. and De Hosson, J.Th.M. (2001), Scripta Mater., vol. 44, pp. 281-286. Quan, G., Heerens, J. and Brocks, W. (2004), Prakt. Metallogr., vol. 41., pp. 304-313.

Low cycle lifetime prediction of AL2024

[10]

[11]

[12] [13] [14] [15] [16] [17] [18]

[19]

[20]

583

Meier, B. and Gerold, V. (1987). In: Fatigue 87, Proceedings of the 3rd International Fatigue Conference, vol. 1, Ritchie, R.O. and Starke, E.A. (Eds.), Charlottesville Virginia, USA. Bomas, H. and Mayr, P. (1987), In: Collected abstracts of the 4th International Conference on Age-Hardenable Aluminium Alloys, p. 42., Balatonfured, Hungry. Hunsche, A. and Neuman P. (1986), Acta metal., vol. 34, pp. 207-217. Guvenilir, A., Breunig, T., Kinney, J.H. and Stock, S.R. (1997), Acta Mater., vol. 45(5), pp. 1977-1987. Guvenilir, A., Breunig, T., Kinney, J.H. and Stock, S.R. (1999), Philos. Trans. R. Soc. Lond. A, vol. 357, pp. 2755-2775. Khor, K., Buffière, J.Y., Ludwig, W. and Sinclair, I. (2006), Scripta Mater., vol. 55(1), pp. 47-50. Toda, H., Sinclair, I., Buffière, J.Y., Marie, E., Connolley, T., Joyce, M., Khor, K. and Gregson, O. (2003), Philos. Mag., vol. 83(21), pp. 2429-2448. Braun, R., Steglich, D. and Beckmann, F. (2006), DESY Annual Report, pp. 497-498, Hamburg. Otin, S. (2007), .Lois d’endommagement incrementales isotrope/ anisotropies pour applications thermomecaniques complexes, Thèse d’état de l’ecole normale superieure de Cachan, Cachan. Vyshnevskyy, A., Khan, S. and Mosler, J. (2009). In: 8th International Conference on Fracture and Damage Mechanics, 8-10 September 2009, Malta (In press). Pirondi, A., Bonora, N., Steglich, D., Brocks, W., Hellmann D. (2006), Int. J. Plast., vol. 22, Issue 11, pp. 2146-2170.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

FATIGUE AND DAMAGE TOLERANCE ASPECTS OF METAL LAMINATES J. Sinke1 and S.A.H. Johansson2 1 2

TU Delft, Faculty of Aerospace Engineering Linkoping University, Institute of Technology

Abstract: In the DAILFAST program (EU KP6 program), a number of new metal and hybrid technologies have been investigated. Main objectives were to improve the properties of structural materials and concepts and to reduce the manufacturing costs. Part of this research was aimed at the improvement of the fatigue and damage tolerance properties of selected Metal Laminates and structures made of these laminates. The constituents of these Metal Laminates were state-of–art aluminum alloys, having several thicknesses, and adhesives. In the program a number of static properties have been tested on coupon level. The results of most tensile dominated tests showed that, as expected, the static properties of the Metal Laminates are equal to the properties of monolithic materials. For some other properties that are related to shear and/or compression loading some improvement is achieved. Other properties related to damage tolerance showed the potential for improvement, one is the impact resistance. Coupons tested in a drop tower test showed that most laminates perform significantly better than monolithic materials. Also the fatigue resistance of the metal laminates is higher when compared to monolithic alloys: in general, the increase of fatigue life is in the order of 10-20%. These results are based on tests with specimen with through the thickness holes. Further improvement of the fatigue and damage tolerance behavior of Metal Laminates structures is achieved by proper M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 585–599. © Springer Science+Business Media B.V. 2009

J. Sinke and S.A.H. Johansson

586

design and manufacture of those structures. Metal Laminates can be tailored for its purpose by selecting the right combinations of constituents and the right local thickness. This tailoring is achieved by the natural application of doublers, local reinforcements and other local features bonded onto the. To demonstrate the effects of tailoring, a number of small and medium size panels have been designed, manufactured and tested. At coupon level a number of joint types and joint configurations were tested statically and in fatigue. These tests showed that the best joining method is adhesive bonding. The panel tests showed that a weight reduction of 10 to 13% is possible when compared to riveted structures. The manufacture of these panels however, is more expensive.

INTRODUCTION Within the aerospace community there is a constant awareness and drive for the development of new materials and structural concepts in order to reduce the costs and weight of aerospace vehicles. The current paper, which presents some results of the DIALFAST 1 program, focuses on fatigue and impact of Metal Laminates. These laminates, made of metal sheets and having a thickness below 1 mm, are bonded together. The adhesive has no reinforcing elements like fibers, as is the case for Fiber Metal Laminates. The main objectives of the research program were to decrease the manufacturing costs, and in particular the weight of the structures by 20% [1]. To achieve that, the properties of the metal laminates should be improved with respect to monolithic sheets by applying the metal layers and reinforcing elements in an efficient way. In this paper some results of the DIALFAST program on Metal Laminates are presented. The paper hat its focus on fatigue aspects of metal laminates and on another damage tolerance property: impact. Since the entire research program comprised of a large number of tests, only a summary of test results will be given. First, in the next section, a few comments are made about the materials that have been used in the test program. After preliminary research some metal alloys and adhesives were selected, which have been combined in a number of Metal Laminates. Those laminates are briefly described in the next section. The main part of the paper is dedicated to the description of a number of tests. The following tests are discussed: the tensile test is briefly mentioned, just to give some insight in the static properties of the selected Metal Laminates; fatigue tests and impact tests. The fatigue tests are divided in fatigue life tests performed on coupon 1

DIALFAST stand for: Development of Innovative and Advanced Laminates for Future Aircraft Structures. This project was a EU KP6 project, project no. 502846.

Fatigue and damage tolerance aspects of metal laminates

587

specimens; fatigue crack growth tests, fatigue tests performed on panels with reinforcing doublers, and fatigue tests on joints in Metal Laminates. For each test a brief description is given of the test and how the tests were performed. In a subsequent section the tests results are presented. After the section on the discussion of the test results, in which the results are briefly discussed, the last section summarizes the presented information in a number of conclusions. During the sections about the tests, the sequence of the type of tests is the same. Selected materials Metal Laminates have two constituents: sheets made of metal alloys and adhesives to bond the sheets together. For the test program the well-known 2024-T3 material (t = 1.6 mm) is used as reference material. As candidates to be used in the Metal Laminates a number of alternatives can be found amongst Aluminum alloys of the 2xxx-, 5xxx-, 6xxx- and 7xxx-series (e.g. Al-2524, -6013, -6156, -7085, and -7475). Other metal alloys that could be used are Aluminum-Lithium alloys (like Al-1424, 1441, 2098), aluminum based Metal Matrix Composites (MMC), and Titanium alloys. The requirements used for the selection were: availability in thin gauges, and no experimental alloys that needed many years for certification, the alloy should be available within a couple of years. Many candidate materials failed on one of these two criteria. Other, less stringent criteria that have been used, are related to the necessary surface treatment and subsequent bonding process. The metal alloys that remained were: 2024-T3 (t =1.6 mm) as reference material; 2024-T3 (t = 0.8 mm), 2524-T3 (t = 0.8 or 0.7 mm (bare)), and 7475-T7 (t = 0.4 and 0.8 mm) used in the laminates (all Clad-materials, unless stated otherwise) Ti-6Al-4V (t = 0.8 mm), used for doublers only. For the selection of the adhesives, the Cytec FM-73 (noted as adhesive A) adhesive is used as the reference. The selected adhesives have at least one specific property which is different from this reference, which should contribute to the objective of low cost manufacturing. The selected adhesives are: the Henkel EA-9696 (B) adhesive, which is a non-autoclave curing adhesive, which should avoid expensive autoclave costs; the Henkel EA-9380 (C) adhesive, a paste, which is an adhesive cured at a low temperature (700 C); this will reduce the required curing energy; and the Henkel EA 9686 (D) adhesive, which has a high service temperature (1480 C).

TESTS PERFORMED During the DIALFAST program a large number of different tests has been performed, like tensile tests, blunt notch tests, bearing tests, impact tests, tests for corrosion, fatigue tests to determine the fatigue life and crack propagation for laminates, panels with crack stoppers, and joints in Metal Laminates [2, 3]. Only those tests, related to fatigue and impact, are reported in this paper. In the end, also

588

J. Sinke and S.A.H. Johansson

some general remarks are made about the influence of architecture and manufacturing on the damage tolerance of Metal Laminate structures. Tensile tests Tensile tests have been performed according to EN 10.002 using a Zwick 1466 testing machine. The test section of the specimens was 12.5 mm wide and 75 mm long. The tests were performed with a test speed of 3mm/min and an extensometer of 25 mm was used. Fatigue tests Significant attention was given to the fatigue properties of Metal Laminates. If these laminates could offer an increase of the fatigue resistance [4], this could be translated into a weight benefit. Several fatigue tests have been performed: Fatigue tests on open hole coupon specimen. The fatigue tests on notched coupons have been performed on specimens 330 mm long, 40 mm wide and with an 8 mm hole in the center. For this test a large number of different laminates were tested, both double (2-layer) laminates (20/20, 25/25, 74/74 and mixed 25/74) 2 with different adhesives, as well as triple (3-layer) laminates (74/25/74 and 25/74/25 combinations). All laminates but one (the 25/74/25 combination) had a total metal thickness of 1.6 mm. The specimens were tested by constant amplitude fatigue which was performed on a 40 kN Instron servo hydraulic testing machine applying 10 Hz cyclic loading at R = 0.1. The first test of each material was performed at the largest applied fatigue stress, based on the metal sheets only, which was chosen as 60% of the (maximum) static strength measured in blunt notch tests. Other, lower stress levels were chosen in order to obtain a S-N-curve for the different laminates. The tests were performed an Linkoping University, Sweden, where also a few additional laminates were tested, which had not been selected for the DIALFAST program. Tests to evaluate the fatigue crack propagation of laminates. The fatigue crack propagation tests were carried out (at EADS, Germany) according to ASTM E-647. The used specimen were 400 mm long, 160 mm wide and had a small center notch (hole plus saw cut). The applied stress level was 100 MPa and R = 0.1, at a frequency of 2 or 4 Hz. As before, the stress was calculated on the metal constituents only. The crack measurements were carried out by a potential drop method and a few optical measurements for calibration. 2

The coding of the laminates is related to the combination of used constituents. “20” refers to 2024-T3 alloy, “25” to 2524-T3 alloy, “74” to 7475-T7 alloy. Total thickness of the laminates is 1.7–2.0 mm, keeping in mind that bond line has a thickness of 0.1 mm. The sheets for all 2-layer laminates are 0.8 mm thick; for the 3-layer laminates the 74-layers are 0.4 mm thick and the 25-layers are 0.7 mm thick (outside) or 0.8 mm thick (inside). The adhesives are coded by A (FM73), B (EA9696), C (EA9380), and D (EA9686).

Fatigue and damage tolerance aspects of metal laminates

589

Tests to evaluate the fatigue crack propagation of doubler reinforced laminates. The third type of fatigue tests that was performed are skin panel tests with bonded doublers. The 875 x 400 mm symmetrical panels were made with two crack stoppers bonded on the panels. The crack stoppers had a width of 35 mm and were made of different materials: aluminum 2524- and 7475-alloys and the Titanium alloy Ti6Al4V. The fatigue crack propagation tests were carried out similar to ASTM E-647 (at EADS, Germany). The specimen were tested with a maximum stress of 100 MPa and R = 0.1. The test frequency was 2 Hz. The stresses were calculated using the nominal skin thickness without the thickness of the doublers. During testing a suitable anti-buckling device was used. The crack measurements were performed as for the crack propagation tests: potential drop with some calibrating optical measurements. Fatigue tests on joints For a limited number of Metal Laminates several joints have been fabricated to be tested under fatigue loading until failure (determination of fatigue life). Two groups were tested: the first group consisted of common lap joints having two fasteners (solid rivet or titanium rivet/bolt) or an adhesive bonded lap joint with similar overlap dimensions. The specimens were 38.4 mm wide (related to a 4.8 mm fastener) and had an overlap length of 48 mm. Both double and triple laminates were tested. These joints were tested at different stress levels, 60, 80 and 90 MPa (R = 0.1) for the fastener type joints, and 90 and 120 MPa for the adhesive joints. The second group consisted in a number of alternative bonded joints, which were derived from the splice configuration as used for Fiber Metal Laminates (see figure 1). These joints were also tested in fatigue at stress levels of 100, 150 and 200 MPa, These tests were primarily explorative and the specimens were fatigued until failure. All tests were performed at the TU Delft, the Netherlands. A

B

C

D

E

Figure 1. Splice like adhesive bonded joints used for Metal Laminates.

J. Sinke and S.A.H. Johansson

590

Impact tests In addition to the fatigue resistance also the impact properties of the laminates were investigated. For these tests a number of materials were selected: Two monolithic aluminum alloys (2024-T3 and 7075-T6); two double laminates with different metal layers and adhesives (A and B); and two triple laminates with the high strength 7475-alloy in the center. The specimens, each 125 x 125 mm, were tested in the impact drop tower of the TU Delft. For each laminate about 10-12 specimen were available. Testing at different energy levels gave the failure limits of the different materials. The monolithic materials had one failure limit: cracking, starting at the convex surface. The Metal Laminates could have more than one failure mode: first a crack in one layer, and subsequent cracking at the second layer or full failure. During testing the damage (cracks) and the dent depth are recorded and based on the boundary values of the failure modes, the maximum impact energy is determined as illustrated in figure 2.

Figure 2. Method to determine the maximum impact energies.

TEST RESULTS Tensile test The test results for the tensile test are presented in figure 3. Three parameters have been recorded: the yield strength (Rp 0.2), the failure strength (Rm), and the failure strain (A).

Fatigue and damage tolerance aspects of metal laminates

591

As can be seen in this figure, the static strength of the laminates is not as large as for a monolithic material when the constituent metal alloy is the same. Only in those cases where the laminate is made of the stronger 7475-alloy, some improvement in static properties can be recorded.

Figure 3. Tensile test results for a large number of Metal Laminates.

Fatigue tests S-N-curves The test series performed on the coupon specimens, were fatigue tests to determine a S-N-curve for a number of different Metal Laminates. Although a large number of S-N-curves have resulted from the tests only a few curves are presented in this paper (see figure 4). Since the effect of the type of adhesive is negligible, the plot shows four curves fitted through the test data of specimen with the same metal constituents but different adhesives. the four curves in figure 4 are: the reference curve based on Airbus data; a curve for triple laminates which is closest to the reference curve, and two curves for 2024- and 2524-laminates which had the best properties. The mixed double laminates, which are not represented, have results which are similar to the triple laminate. When in the triple laminate the alloys are switched (2524 at the outside and 7475 at the inside) the fatigue resistance is improved and like the 2024- and 2524- double laminates.

592

J. Sinke and S.A.H. Johansson

Figure 4. S-N-curves for different laminates tested with open hole coupon specimens

Crack propagation Also a large number of laminates have been tested in crack propagation tests. However, as for the coupon tests, the results showed little variation between the different laminates. The effect of the type of adhesive was negligible and the effects of the type of alloy were slightly visible although not significant: the 2524based laminates had a (slightly) slower crack growth rate than the 7475-based laminates; and the triple laminates were slightly better than the double laminates. Therefore, only a representative plot is presented (see figure 5). The plot shows the best combinations: the triple laminates and the 2524 double laminates. All other laminates have slightly higher crack growth rates and their curves are above the curves in figure 5.

Fatigue and damage tolerance aspects of metal laminates

593

Figure 5. Representative plot of the crack growth curves of Metal Laminates.

Crack stopper panels The test results for the fatigue tests on the crack stopper panels are summarized using figure 6. The figure represents the crack length versus the number of fatigue cycles for the tested Metal Laminate/doubler combinations.

594

J. Sinke and S.A.H. Johansson

All curves show a (small) shift in the fatigue crack curve to the right, which is the effect of the doubler. The crack growth reduction in the vicinity of the doubler is followed by delamination of the doubler and the initiation of a fatigue crack in the doubler itself. Once the doubler has failed, the fatigue crack growth is accelerated again until failure of the panel. From the figure is becomes clear that the different aluminum doublers had a rather small effect on the fatigue curve (three curves on the left), but for the Titanium doublers the crack retardation effect is significant (two curves on the right).

Figure 6. Fatigue curves for Metal Laminate panels with crack stoppers. {The crack stopper is located between a = 150 -200 mm)

Fatigue of joints The results of the fatigue tests of the conventional joints are presented in the figures 7 and 8: figure 7 showing representative results for a double laminate, and figure 8 showing similar results for triple laminates. Although both figures show the results for laminates based on the B-adhesive (EA9696), it should be mentioned that the type of adhesive had little effect on the test results.

Fatigue and damage tolerance aspects of metal laminates

595

Fatigue lives ML 25/74 (B)

Max. Stress [MPa]

150

100

solid solid-EA9686 (D)

50

Ti-bolt Adhesive

0 10

100 fatigue life [kCycles]

1000

Figure 7. Fatigue tests results for riveted and bonded joints in double Metal Laminates.

Fatigue lives ML 25/74/25 (B)

max. stress [MPa]

150

100

solid

50

solid- EA9686 (D) Ti-bolt adhesive

0 10

100

1000

10000

fatigue life [kcylces]

Figure 8. Fatigue tests results for riveted and bonded joints in triple Metal Laminates.

J. Sinke and S.A.H. Johansson

596

The alternative joints were tested at different stress levels. Not for all specimen a fatigue life had been recorded; some of the specimen were very fatigue resistant.

For those specimen that failed, the failure location was used to rank the different configurations. The failure locations of the alternative bonded joints are given in figure 9: two failures occur in the joint area, three in the parent material, one of which is a reference concept (E).

A t1

t2

t3

B t1

t2

C t2

t1

D t1

t2

E t1

Figure 9. Fatigue failure locations in different joint concept for Metal Laminates. Impact tests The test results of the impact tests are presented in figure 10. In this figure the relative impact energies of the materials are plotted as function of the parameter “dent depth x thickness”. The latter parameter is a measure for the plastic deformation that remains after impact. The relative impact energy is the impact energy divided by the density and thickness of the material. The plot shows that the relationship between the relative impact energy and this parameter is almost linear. For each material there are two data points: the lower impact energy represents the maximum recorded impact energy without a crack in the material; the highest value represents the minimum recorded impact energy with full failure.

Fatigue and damage tolerance aspects of metal laminates

597

Maximum relative impact values

Relative Impact Energy

250

7075 2024-REF

200

25/74A 25/74B 74/25B

150

25/74/25A 25/74/25B

100 50 0 0

5

10

15

20

dent depth x thickness

Figure 10. Relative Impact energies for some monolithic alloys and Metal Laminates

DISCUSSION OF RESULTS Tensile test From the tensile tests it is clear that the static strength is mainly dependent on the metal constituents of the laminates. The laminates based on the 2xxx-alloys have strength properties slightly lower than the reference material: the adhesive does not contribute to the strength of the laminate, although it adds to the thickness of the laminate. For the laminates containing 7475-alloy the strength properties are sometimes slightly higher. Main reason for this are the higher strength values of the 7475-alloy over the 2024-T3 reference alloy. Fatigue tests The improvements in fatigue life that can be achieved by Metal Laminates are limited; the fatigue tests in the coupon specimens showed, in general, improvements of the fatigue life of about 10-20% with respect to the reference material. Some laminates had a longer fatigue life, up to 50-100% w.r.t. the reference material. The difference in fatigue properties is more pronounced at lower life. At those fatigue lives the effect of scatter of the crack initiation has a larger impact on the fatigue life than at larger fatigue lives, where the crack growth has more influence. However, the main reason for the limited improvement is the fact that the coupons had through the thickness holes. During the fatigue, cracks

598

J. Sinke and S.A.H. Johansson

are initiated at both (or all three) layers, shortly one after the other. Once the cracks have been initiated, crack growth will take place and ultimately the specimen will fail. During these tests the retardation by the adhesive layer does not take place. The adhesive layers could act as an additional barrier in case of a surface crack. The crack propagation tests showed very little variation. Some laminates had a slightly better behavior, but these improvements are within the range of scatter of the results. Since the crack propagation data for the 0.8 mm sheets seems to be equal to the propagation behavior of the 1.6 mm sheets, no improvement was found for the crack propagation in Metal Laminates with respect to the monolithic materials. This might be different for Metal Laminates with partially through cracks: in those cases the crack growth may slow down for a while, until the crack fronts run parallel again. However, these partly through cracks have not been tested. In general each doubler has a positive effect on the crack growth propagation behavior due to the local increase of the stiffness which resulted in a reduced stress intensity at the crack tip when this crack enters the area of the doubler. In this respect the stiffness (material and geometry) of the doubler is decisive. For the aluminum doublers the fatigue life could be increased by about 10%, for the titanium doubler the life could be doubled. Main reason for this difference is the crack initiation in the doubler: the aluminum doublers showed rather fast crack initiation, but titanium doublers offered more resistance to crack initiation. For the fatigue life of the joints the riveted and bolted joints showed no improvement with respect to the tested reference specimen. The S-N values were at the same level. The bonded joints however, for the Metal Laminates as for the monolithic (reference) material, showed much higher S-N values. These joints have a better load distribution, no in-plane stress concentrations, and therefore a better fatigue resistance. Since bonding is part of the concept of Metal Laminates the bonded joints is a natural choice for such structures. The alternative joints, related to FML-splices, showed a mixture of results. The best results were achieved for the concepts A and D, which failed in the parent material and not in the joint area. They also had the highest fatigue lives, although not all specimens had been tested until failure (number of cycles became too high – tests too expensive). Impact test The linear relationship in figure 10 is a logic one since the materials have a comparable thickness (ranging from 1.6 – 2.0 mm). The most important result however is that the laminates offer far better impact resistance than the monolithic materials. During impact the adhesively bonded interfaces are loaded in shear and contribute to stress redistribution and energy absorption. This is also underlined by the fact that the 3-layer laminates performed better than the 2-layer laminates. The concept of Metal Laminates Finally some general remarks are made about Metal Laminates as a innovative structural concept that could improve the Damage Tolerance behavior (incl. fatigue

Fatigue and damage tolerance aspects of metal laminates

599

and impact) of aircraft structures. The concept is based on bonded structures which can be very durable as has been demonstrated in the past (e.g. Fokker aircraft). Since the development of Fiber Metal Laminates (FML), several manufacturing and design concepts can be applied to Metal Laminates as well. The alternative joints that are described in this paper, can be regarded as derivatives of the splicing concept as used for FML. These joints should enable the manufacture of large (long and wide) skin panels, including reinforcing elements like doublers and crack stoppers. Such large panels could reduce the number of (riveted) joints in the structure and thereby reducing the weight.

CONCLUSIONS The test results and their discussions as presented in this paper are summarized in the following conclusions: • The static strength of Metal Laminates is not as high as the strength of their monolithic counterparts. Only the use of higher strength alloys could improve the strength values over the selected reference material (Al-2024T3). • The fatigue crack initiation and the fatigue crack growth of Metal Laminates give only minor improvements. • The fatigue lives can be increased when doublers or crack stoppers are added onto the laminates; however, this is more a design improvement than a strict material improvement. Titanium crack stopper were by far the best. • Metal Laminates may have high fatigue lives and joint strengths in case of bonded joints. Bonded joints are a natural way of joining for Metal Laminates. • To optimize the application of Metal Laminates one should focus on both the design and selection of constituents • The impact properties of Metal Laminates can be significantly better than for monolithic materials.

REFERENCES [1] [2] [3] [4]

--, (2003) DIALFAST, Part B, No. 502846, 6th European Framework Program Hombergsmeier, E., Johansson, S.A.H., Sinke, J., Vichniakov, A., Pacchione, M., (2006), DIALFAST deliverable D2-1 Sinke, J., Vermeer, P., (2008), DIALFAST deliverable D2-6. Schijve, J. (2009), Fatigue of structures and materials, Springer, ISBN: 978-1-4020-6807-2.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

ARALL AND GLARE FML'S: THREE DECADES OF BRIDGING THE GAP BETWEEN THEORY AND OPERATIONAL PRACTICE Cees van Hengel and Peter Kortbeek

FMLC, Delft, The Netherlands Abstract Since their conception in the late 70's of the last century, Fiber Metal Laminates (FML's) have bridged the gap between theory and practice several times, beginning from ARALL on the C17 aft cargo door, and more recently as (standard) Glare and as HSS Glare on the A380 fuselage. This paper summarizes these evolutionary cycles, and outlines some of the activities currently in progress in preparation for the next cycle.

ARALL Although it is not always remembered today, the first generation of FML’s to successfully bridge the gap was ARALL. ARALL, for Aramide Reinforced Aluminium Laminate, was conceived of in the late 1970's and early 1980's. Fatigue of aluminum was known to be an important material weakness to consider in the design of aircraft structures, and so a "fatigue insensitive aluminium" was a highly desirable material. The theory to achieve this was to use the fatigue resistance of the newly developed high strength Aramid fibres, and combine it in an optimal way with the time proven production technique of adhesive metal bonding. Major development efforts for several applications followed, such as for lower wing skins of the Fokker 27 and later Fokker 50 commuter airplane [1]. Ultimately "the gap" was bridged when ARALL was qualified for Mil Handbook 5 and was applied as the skin material for the aft cargo door on the McDonnell Douglas C17 military air lifter [1], see figure 1. So, FML's had completed their first evolutionary loop: starting from theory, ARALL had gone through material specification, material qualification and allowables development, design and manufacturing, and finally ending at operational usage. M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 601–615. © Springer Science+Business Media B.V. 2009

Cees van Hengel and Peter Kortbeek

602

However, as soon turned out, "survival of the fittest" for aerospace material demanded more than "just" good performance on properties: good performance on cost was also imperative. ARALL was manufactured as flat sheet laminates which had to be prestretched to counter unfavorable curing stresses. The double-curved skin of the C17 aft cargo door part was manufactured with metal-like processes: the 27 separate sheets were individually formed by rolling or stretching, and then cut and assembled with additional titanium straps with over 10.000 fasteners [2]. In the long run, the weight saving achieved did not justify these costs, and so, after a limited number of ship sets, ARALL was again replaced by conventional aluminium.

GLARE Some 10 years later, the second generation of FML's managed to bridge the gap again, this time in the form of Glare on the A380. Glare stands for Glass Fiber Reinforced Aluminium. The original concept behind this FML generation had "just" been to make an FML with biaxial reinforcement, which required the use of glass fibres to avoid the need for prestretching. The new material did indeed show the good fatigue life expected from theory, but also unexpectedly turned out to have favorable properties for resistance to impact, lightning strike and fire (burn through). Nonetheless, as with ARALL before, Glare's establishment as a viable aircraft material was seriously threatened by cost constraints. Thanks to the new fibre ingredient however, this time a new manufacturing concept was possible: different from ARALL, Glare laminates could be produced directly into the shape and thickness of the final part, even in double curved form [2], as shown in figure 2. This "composites-like" manufacturing approach enabled large cost reductions compared to the previous "metal-like" approach of ARALL. Thus Glare parts also became cost competitive, especially for double curved parts. Additionally, as shown in figure 3, the new approach to laminate manufacture allowed integrating local reinforcing elements at very low cost and weight, further improving part performance [3], and so Glare FML technology managed to earn its way onto the A380. Soon after, it was realized that for certain parts of the A380 fuselage, the 2024based "standard Glare" did not provide the optimal mix of static, fatigue and residual strength properties. Thus a third evolution took place, resulting in the 7475-based "High Static Strength" (HSS) Glare. The excellent performance of the Glare parts in the A380 full scale static and fatigue tests was recently released to the public domain [4], and so it appears that these second- and third generation FML's have indeed successfully bridged the gap from theory to operational performance.

Three decades of bridging the gap between theory and operational practice

603

NEW DEVELOPMENTS However, the aerospace industry is not a static environment and standing still soon amounts to being overtaken by advancing competitors such as friction stir welded (FSW) or laser beam welded aluminium structures on one side, and CFRP structures produced in highly automated processes on the other. FML's proponents are acutely aware that FML's still are only a niche technology and that it is therefore vulnerable. Accordingly, continuous development projects have been underway under various industrial and/or academic, national and European frameworks. In the Netherlands, some of the most involved parties are the NLR, TU Delft (faculty of Aerospace Engineering) and Stork Fokker AESP. These organizations coordinate their FML development through the FMLC. The government-related agency NIVR is a frequent sponsor, and Airbus usually the intended key benefitor. An overview of the main projects is shown in figure 4. Example programs that were participated in are the European DIALFAST program (6th Framework Program) and the Dutch national Strategic Research Program (SRP). Some results of these programs are highlighted hereafter. The DIALFAST project, which ran from 2004 to 2008, targeted the development of the development of a new generation of fibre metal laminates (FML) and metal laminates (ML), with the aim of weight and cost reductions. The project included a weight study [5] on the fuselage of a generic wide-body airplane for which load data were available (the aft fuselage was chosen since this was the most highly loaded). New FML-types were defined incorporating advanced, high strength metals and new fibre types, such as Zylon and their static and fatigue allowable stresses were estimated from initial test data or using calculation methods (figure 5). Using sizing methods typical for preliminary design, the necessary skin thicknesses were then determined for each panel (area between two adjacent stringers and frames) based on the applied loads, failure criteria and allowable stresses. Figure 6 shows an example of the distribution over the surface of the critical (i.e. sizing) failure modes. The predominant failure modes in this study were residual strength (axial and hoop), and stability related failure modes ("buckling"). Important also was the minimum thickness requirement. Fatigue crack growth was very rarely critical. The resulting weight saving potential of the various advanced FML's is shown in figure 7. They are in the order of 10%, when conventional design rules are applied. However, if more advanced designs are allowed (i.e. deviating from the standard Glare types such as Glare 3, Glare 4 etc.), and lower density alloys, then weight savings up to 20% are anticipated. Concerning cost, the study concluded that expensive ingredients such as new fibres are not necessarily cost effective. An insight gained from this DIALFAST study, is that property balancing is a potentially powerful technique for optimizing the weight of FML structures. Property balancing is defined as the concept of designing an FML in such a way

604

Cees van Hengel and Peter Kortbeek

that the mechanical properties of the laminate (yield strength, ultimate strength, F&DT etc) are matched as closely as possible to loads that act on the structure under the various design load conditions (such as static Ultimate Loads -UL-, static Limit Loads -LL-, fatigue cases and damage tolerance cases, etc.). Property balancing becomes practical because in many structures, such as fuselages, there are fairly constant ratios between the loads acting under various load conditions. In those cases, property balancing comes down to selecting FML ingredients and configurations in such a way that the ratios of the properties of the laminate match as closely as possible the ratios of the applied loads in the various load conditions.

The SRP program started in 2005 and is a multi-year Dutch national research program aimed at developing a broad range of technologies for aerospace applications. Many of the specific topics are chosen in consultation with Airbus. The research by NLR, TUDelft and Stork Fokker on FML's is typically coordinated through the FMLC. The research covers a diverse range of topics and TRL's (Technology Readiness Levels), as sketched below. The three strategic aims for FML work in SRP are: -

improved capabilities for existing applications, for example by testing of bonded window frames (fig, 8) [6]

-

supporting the competitiveness of FML's on future narrow body airplanes, for example by design, manufacture and testing of a full scale fuselage panel and associate stringer couplings made of Glare (fig. 9, 10)

-

cost reduction of FML development and manufacturing, for example by dedicated automation developments (fig. 11)

COST REDUCTION Cost reduction is pursued for both non-recurring as well as recurring activities. Examples for the non-recurring part are increased use of simulation for determination of static material allowables [7], and knowledge-based design methods to -substantially- automate detail design of FML fuselage skins [8]. For recurring cost reduction, topics under consideration include continuous (coil-tocoil) metal pretreatment [9] and development of robot-based tape-laying techniques. [fig. 11]. Concerning automation, it may be recalled that the original industrialization concept of FML production was chosen around the year 2000. At that time, the decision was made to rely on manual work for critical processes such as lay-up in view of the high cost of automated tape-laying equipment at the time, their uncertain performance and their inflexibility in case of changes in production rates. The wisdom of this decision is born out by the experience of recent years.

Three decades of bridging the gap between theory and operational practice

605

It is understood that even taking into account only a limited number of the above improvements, FML fuselage panels were and still are very cost competitive compared to the "should" price of CFRP structures of similar performance [10]. Nonetheless, considering the strong developments in automated composite manufacturing, a new look is justified and is currently under way. In this respect it is interesting to note that FML's are a hybrid technology, and thus can take advantage of many of the improvement of the competing technologies of composites and metals. For example, the obvious need for automation of the production of composites structures has reduced the price of tape-laying and fibre placement equipment. Therefore today, quite differently from several years ago, investing in these machines for production of FML's may now start making economic sense. Similarly, the emergence of new alumium alloys, be-it high strength or low density, may always offer opportunities for cost effective property improvements.

CONCLUSION It was shown that FML technology has a long track record of successful evolution to meet the exacting performance and cost needs of aerospace applications. As shown by the results of the more recent developments, FML's may be expected to remain very competitive materials for the realization of high performance, low risk and low cost structures.

REFERENCES [1]

[Vlot 2001] “Glare; history of the development of a new aircraft material” Ad Vlot, Delft University of Technology, Faculty of Aerospace engineering, Delft, the Netherlands, ISBN 1-4020-0124-X, 2001

[2]

[Vlot 2001] “Glare; history of the development of a new aircraft material” Ad Vlot, Delft University of Technology, Faculty of Aerospace engineering, Delft, the Netherlands, ISBN 1-4020-0124-X, 2001

[3]

[Pleitner 2006] "Airbus customers benefit from Fiber Metal Laminates; Application of smart structures in the aircraft industry" Jürgen Pleitner, Airbus Deutschland, presented at ILA 2006, May 2006. Av. online at: http://www.ltt.de/uploads/pdfs/Events/ILA2006/ILA06_Praesentat_AIRB US_Juergen_Pleitner.pdf

606

Cees van Hengel and Peter Kortbeek

[4]

[Pleitner 2006] "Airbus customers benefit from Fiber Metal Laminates; Application of smart structures in the aircraft industry" Jürgen Pleitner, Airbus Deutschland, presented at ILA 2006, May 2006. Av. online at: http://www.ltt.de/uploads/pdfs/Events/ILA2006/ILA06_Praesentat_AIRB US_Juergen_Pleitner.pdf

[5]

[van Hengel 2007] “Cost & Weight study” Cees van Hengel, Giel van der Kevie, DIALFAST Deliverable report 1.8, Stork Fokker AESP, December 2007

[6]

[Ubels 2006] “Shear test of a bonded window frame on a Glare skin for application in the Airbus A380-800”, L.C. Ubels, NLR report NLR-CR2006-698, December 2006 (restricted)

[7]

[van Hengel 2008] “On FML property prediction for cost reduction of allowables determination”, Cees van Hengel, Fokker report F-DRSP-CHR08-01, April 2008 (restricted)

[8]

[Vermeulen 2007] “Knowledge based method for solving complexity in design problems”, Brent Vermeulen, Dissertation TU Delft Aerospace faculty, june 2007

[9]

[Kwakernaak 2007] “Overview of candidate anodising processes for continuous surface treatment for structural bonding”, A. Kwakernaak, TU Delft Adhesion Institute Report: HI 2362, July 2007

[10]

Personal communication, mr. H. Rooimans, Stork Fokker AESP Marketing and Sales dept, March 2009

Three decades of bridging the gap between theory and operational practice

Figure 1: ARALL section in MMPDS and C17 rear cargo door made of ARALL; the skin is manufactured from 27 ARALL stretch formed sheets assembled with 10,500 rivets [1,2]

607

Cees van Hengel and Peter Kortbeek

608

Manufacturing of two double curved Glare panels Double curved lay-up tool

Panel after lay-up

Bonding of curved stringers

Lay-up

Finished double curved sheet

Manufacturing of large curved Glare panels is not significantly more difficult than for small flat sheet, while providing value increase! However further product / process optimization is possible

Figure 2: Early demonstration of manufacturing of double curved Glare panels [2]

Three decades of bridging the gap between theory and operational practice

Figure 3: Options for laminate optimisation in FML's [3]

609

Cees van Hengel and Peter Kortbeek

610

Technology phases and maturity Research & Technology Programmes R&T spend profile

Number of Validate Number of subjects CPP

Adapt

Understand Discover NG FML ~ 10% R&T Portfolio

DLS DP2L SPF INDT Deploy BGR ISGL TBOX CRG DFST ANSA2ANSA1 GSC GWF ~ 50% R&T Portfolio ~ 40% R&T Portfolio

Long term (after EIS 2017): Medium term: • radical concepts • Emerging technologies

M5

M6

•optimise and develop new FML with improved properties

Short term: •Cost reduction of Glare

- ?? years

- 10 years

- 5 years

- 2 years

Application

(EIS =+3 years)

Project abbreviati on DP2L CPP INDT GWF CRG KBE GSC SHPR DFST NFML DLS BGR TBOX ISGL ANSA

Project name

Machining of thin sheets Continuous pretreatment Improved NDT (esp. C-scan) Bonding of window frames on Glare Cost reduction of Glare Application of KBE Glare stringer coupling Pre-shaping of sheets Dialfast New Generation FML Improved adhesive properties Bonded Glare repairs Toolbox for prediction of Glare properties Impact simulation of Glare LE Glare/FML technology for ANSA

Figure 4: Overview of some recent Glare development projects

Three decades of bridging the gap between theory and operational practice

Material comparison Glare 3 type, 3/2-0.3 lay-up 600

500

Sy (=S0.2) Sbn_gross* S_bn_nett

Stress (MPa)

400

300

200

100

0 Standard Glare (2024 + S2)

HSS Glare (7475 + S2)

FML Standard Glare (2024 + S2) HSS Glare (7475 + S2) 2024 + Zylon HTLS + S2 7475 + Zylon

2024 + Zylon

HTLS + S2

Sy (=S 0.2 )

Sbn_gross*

200 261 275 398 335

308 367 399 451 437

* factor for cross section reduction (assuming P/d= 5):

7475 + Zylon

S_bn_nett 385 459 499 564 546 0.8

Figure 5: Static properties of material used for Dialfast weight study [5]

611

612

Cees van Hengel and Peter Kortbeek

Figure 6: Example output showing map of failure modes, left hand fuselage half, (Standard Glare upper fuselage, Alumium lower fuselage) [5]

Three decades of bridging the gap between theory and operational practice

613

practical weight, total (kg)

weight comparison

pract weight, alu bottom panels (skin + stringers; kg)

927

948 var 6

459

957

var 5 b

459

981

980 459

-20%

943

958

996

944

200

FML concept overall potential

400

459

600

459

800

aluminium baseline (1995 technol.)

weight (kg)

1000

989

1200

0 var 1

var 1b

var 2

var 2b

Std Std HSS Gl. HSS Glare Glare top, Glare top, (top & AlMgSc (top & AlMgSc bottom) bottom bottom) bottom

var 4

var 4b

Zylon/ Zylon/ 2024 2024 top, (top & AlMgSc bottom) bottom

var 5

var 6b

HTLS/ HTLS/ Zylon/ Zylon/ S2 top, S2 (top 7475 7475 AlMgSc & top, (top & bottom bottom) AlMgSc bottom) bottom

Figure 7: Weight savings from basic properties. Centre bar is overall weight saving potential taking into account ingredient properties, improved design options, and ingredient density effects) [5]

614

Cees van Hengel and Peter Kortbeek

Figure 8: Shear test of a bonded window frame on a Glare skin [6]. Note the warping of the window frame due to the skin buckling.

Figure 9: Test panel with stringer couplings made of Glare

Three decades of bridging the gap between theory and operational practice

Figure 10: Glare stiffened panel ready for mounting into the test bench

Figure 11:Pictures of early trials of automatic tape laying for Glare

615

Fatigue crack growth and life prediction methods

25th ICAF Symposium – Rotterdam, 27–29 May 2009

DETAILED STRAIN FIELD ANALYSES OF FATIGUE CRACKS IN FRICTION STIR WELDED JOINTS H.J.K. Lemmen, R.C. Alderliesten and R Benedictus Delft University of Technology Faculty of Aerospace Engineering

Abstract: This paper presents the results from Fatigue Crack Growth (FCG) tests on Friction Stir (FS) welded aluminium alloys. To obtain more understanding how the reduced yield strength and the residual stresses in the FS weld affect the FCG behaviour, welds were rotated under an angle of 0º, 45º and 90º with respect to the applied load. The goal of this study was to observe the change in FCG direction due to the presence of the FS weld. A new measurement technology, Digital Image Correlation (DIC), was used to measure the strain field in the vicinity of the crack tip. The tests were highly successful resulting in changes in FCG direction and rate, all recorded by DIC. One specimen showed a fatigue crack which turned 90º upward, parallel to the applied load and continued to grow in this direction. Moreover the effect of yield strength and residual stress was measured successfully. The main conclusion is that the relation between the yield strength and residual stress exist, but that the microstructure in the FS weld has also an influence.

INTRODUCTION In a previous study on Fatigue Initiation (FI) in Friction Stir (FS) welded AA2024-T3, remarkable Fatigue Crack Growth (FCG) behaviour was observed. In a test configuration with the FS weld under an angle of 45º, the direction of FCG changed dramatically in the FS weld [1,2]. However, the FI specimens used for that research had only a width of 50 mm and thus the FCG phase was too small for proper FCG measurements. Therefore, dedicated FCG centre cracked specimens M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 619–642. © Springer Science+Business Media B.V. 2009

620

H.J.K. Lemmen, R.C. Alderliesten and R. Benedictus

with FS welds in different orientations, were tested with the purpose to repeat and to find an explanation for the unusual FCG behaviour. The tests were successful and remarkable fatigue behaviour like crack turning was observed in several specimens. This paper presents the results from these FCG tests and possible explanations for the observations are discussed. FS welding introduces residual stresses and reduces the yield strength in the FS weld. The authors believe that the change of yield strength and the residual stresses in the FS weld play an important role in the change of FCG rate and direction. Therefore, the purpose of this study is to reveal the relation between the FCG behaviour and the macro material properties in the FS weld. The residual stresses in the FS weld change the magnitude of the principle stress at the crack tip, which leads to a change in FCG rate. Besides, if the fatigue crack enters the FS weld with an angle different from 0º and 90º, the orientation of the principle stress at the crack tip will be changed, and thus the direction of FCG. The variation in yield strength in the FS weld is different for each alloy, depending on the strengthening mechanism and the temper of the alloy prior to FS welding. The yield strength of the FS weld has an influence on the fatigue crack growth rate, because it affects the size of the plastic zone around the crack tip, and thus the amount of crack closure in the wake of the crack. Moreover, the yield strength affects indirect the effectiveness of the residual stress on the principle stress at the crack tip. A large plastic zone leads to a distribution of the residual stresses over a larger region; this will reduce the magnitude of the principle stress at the crack tip. Besides, the different regions with low or high yield strength in the FS weld have an influence on the geometry of the plastic zone. The geometry of the plastic zone influences the distribution of the residual and applied stresses and thus the magnitude and orientation of the principle stress. The yield strength profiles and residual stress profiles from the FS welds used for this research, were obtained in a previous study [3]. This information was used to choose the FCG test configurations, i.e. FS welds under an angel of 0º, 45º and 90º with respect to the applied load, such that the effect of the yield strength and residual stress is controlled. To visualize the strain field and thus the geometry of the plastic zone in the vicinity of the crack during a fatigue test, a new measurement technology was developed and applied; Digital Image Correlation (DIC) [4]. DIC is a measurement technology which uses images of a test specimen to analyse the deformations during a test. DIC made it possible to measure and visualize the strain distribution around a crack tip in an FS weld. The behaviour of the strain field together with the FCG rate and direction, contain information on whether the effects of yield strength and residual stress as mentioned above are indeed reality.

Strain field analyses of fatigue cracks in FS welded joints

621

The tests described in this paper cover fatigue crack growth through three FS welded alloys; AA2024-T3, AA7075-T6 and AA6013-T4, which are tested in three test configurations, i.e. FS welds under an angel of 0º, 45º and 90º with respect to the applied load. The initial centre crack was placed such that optimal effect of the FS weld on the FCG behaviour was expected for the different configurations. At certain fatigue intervals the test was interrupted to measure the crack length and to take images from the surface of the specimen. The results presented in this paper give more understanding of fatigue crack propagation in FS welded joints. Especially the influence of the yield strength on the magnitude and geometry of the plastic zone in the vicinity of the crack tip is quantified using the results obtained by DIC.

EXPERIMENTS Friction stir welding FS welding is seen as a possible light weight solution for joints in aircraft fuselage structures [5]. Moreover, FS welding is a robust process without emission of dangerous gasses or radiations for which protection is required and it can reduce lead times and manufacturing costs. Since FS welding is a solid state process, it is possible to join high strength aluminium alloys which are considered to be nonweldable. FS welding is performed by pressing a non-consumable, rotating tool at the work piece (Figure 1a) [6]. The FS weld tool consists of a non-consumable rotating cylinder with a ‘pin’ at the centre of the lower circular surface (the ‘shoulder’). The frictional heat generated between the shoulder and the work piece enables it for the pin to stir two parts together.

Downward welding force

a α

Welding direction

b front side

Shoulder

z

Pin

RS

AS

Tool rotation

x L y

HAZ

LT weld Advancing Side (AS)

TMAZ

Weld Nugget

HAZ

back side

Retreating Side (RS)

Figure 1, a: FS welding process, b: cross-section of FS weld Three aluminium alloys (AA2024 T3, AA7075 T6 and AA6013 T4) were used for this research, in both welded and un-welded (base material) condition. Welding

H.J.K. Lemmen, R.C. Alderliesten and R. Benedictus

622

was performed at EADS in Munich with an ESAB FSW machine using for AA2024-T3, AA7075-T6 and AA6013-T4 respectively, a welding speed of 350, 300 and 1000 rpm, a rotational speed of 550, 280 and 1500 rpm and a force of 19, 18 and 14 kN. A tool with a shoulder and pin diameter of respectively 13 and 5 mm was used for all three alloys with a tool angle of α = 2º. Due to the rotation of the FSW tool, the process is asymmetric with at one side a higher speed difference between the tool and the work piece than the other side, respectively the Advancing Side (AS) and the Retreating Side (RS) (Figure 1a). According to the orientation of the axes in Figure 1a, ‘y’ indicates a location in the FS weld with y = 0 mm as the centre of the FS weld and the advancing side being the positive side of the y-axis. As a result of welding, the microstructure of the base material is changed resulting in three zones with a different thermodynamic and/or mechanical history (Figure 1b) [7-10]. FS welding affects the yield strength, but this effect is highly dependent on the strengthening mechanism of the base material and the thermodynamic and mechanical processes during FS welding. The local yield strength profiles as were measured in a previous study are shown in Figure 2 (upper three curves) [3]. Important to note is that the yield strength of the artificially aged alloy AA7075-T6 is more affected by FS welding than naturally aged AA2024-T3, while AA6013-T4 shows no significant reduction of the yield strength.

shoulder of the FSW Tool

500 Base AA7075-T6

RS

Pin of the FSW tool

AS

stress [MPa]

400

300

200

Base AA2024-T3 AA7075-T6 yield stress AA2024-T3 yield stress AA6013-T4 yield stress AA7075-T6 residual stress AA2024-T3 residual stress AA6013-T4 residual stress

Base AA2024-T3

100

0

-100 -20

-15

-10

-5

0 5 distance from weld centre [mm]

10

15

20

Figure 2, yield strength profiles (upper 3 graphs) and longitudinal residual stress profiles (lower 3 graphs) of three FS welded materials. Beside the yield strength, also the residual stress profiles are shown in Figure 2 (lower three curves). The residual stresses were measured using X-ray diffraction [3] and represent therefore the residual stress profiles at the surface of the welded sheet. The residual stresses have an anisotropic character, which means that the residual stresses are high parallel to the weld, but low perpendicular to the weld. Figure 2 shows only the residual stresses parallel to the FS weld.

Strain field analyses of fatigue cracks in FS welded joints

623

Digital Image correlation To obtain the strain field around the crack tip, Digital Image Correlation (DIC) was used. The software tool to process the images was developed by the author [4]. The images of 1600 by 1200 pixels were captured, using a CCD camera which was positioned in front of the specimen and aligned perpendicular to the specimen’s surface (Figure 3). To obtain a high accuracy (large amount of pixels/mm2) but also a large measurement area, a grid of images were taken with a small overlap using an x-y-z-controlled rack. To prepare the surface of the specimen for image correlation, an irregular speckle pattern was paint brushed on the measurement area of the specimen. The speckle pattern in combination with the DIC tool resulted in an accuracy in strain of at least εerror = 0.167 % (for a distance between the initial grid points of L0 = 60 pixels) [4]. However, the strain is calculated for at least 780 locations in each image, therefore the trend of the strain field can be distinguished from the scatter with a much larger accuracy.

Figure 3, Schematic fatigue test setup for taking images for DIC Before the test was started (fatigue life N = 0), three images were taken, the first at 1 kN (~3.0-4.0 MPa), the second at minimum load (σmin = 13.3 MPa) and the last at maximum load (σmin = 133.0 MPa). The image at 1 kN was used as a reference to calculate the strains in the subsequent images. It is better to take the reference image at a load of 0 kN, but due to welding the sheets were slightly curved and to avoid the influence of the curvature on the measurement, a small load of 1 kN was applied. Afterwards the strain is corrected for the initial strain by a calibration. The average strain in the image at maximum load was used to calibrate the measurement with theoretical data. During the test, the load sequence was

H.J.K. Lemmen, R.C. Alderliesten and R. Benedictus

624

interrupted at certain fatigue life intervals to capture images at maximum load until the specimen failed. After calculation of the x- and y-strain in all the images, the strain data from different locations in the image grid is combined resulting in the total strain field around the fatigue crack in the measurement area. In this paper mainly the strain in y-direction will be discussed because this is also the orientation of the applied load. Fatigue crack growth specimens For this research Centre Cracked Tension (CCT) fatigue specimens were used with a hole (diameter 3 mm) from which a saw cut was made manually. Three different test configurations were used with a different orientation of the FS weld with respect to the applied load, 0º, 45º and 90º (see also Figure 4). To measure a maximum effect of the FS weld on the FCG behaviour in the 0º and 45º test configuration, the saw cut was only made at the side of the hole where also the FS weld is situated. The presence of a saw cut at one side of the hole results in asymmetric FCG, but theory exist to compare the asymmetric crack configuration with a symmetric crack configuration. Because some configurations are not interesting to be tested for all alloys, and because the FS welded material was limited, not all test configurations were applied for the three FS welded alloys (Table 1). To evaluate the results, equivalent tests were performed on base material specimens having the same geometry as the welded FCG specimens. The material for those reference tests was obtained from the same sheets from which the welded material was taken.

σ∞

σ∞

LT

LT

σ∞

L

AS

RS

AS

L

a = 2.5mm

2a = 5mm y

AS

RS x

x

y

H >> 2w

0º

45º

90º

RS

d = 3mm L

y

w LT

σ∞

σ∞

x

σ∞

Figure 4, Configurations of FS welded CCT specimens The surfaces of the FS welds were machined and ground up to P1000, to remove the typical residual FS weld surface. This process reduced the thickness of the specimens 0.1 to 0.2 mm.

Strain field analyses of fatigue cracks in FS welded joints

625

Table 1, configurations of FS welded CCT specimens Material: AA2024-T3 AA7075-T6 AA6013-T4

Configuration: 45º FS weld 90º FS weld 0º FS weld 45º FS weld 90º FS weld 0º FS weld

Distance hole to weld centre [mm]: 8.3 (for all) 0.0; 3.0; 6.5; 6.5 21.6; 16.6 8.3 (for all) 7.0; 10.0 19.0; 10.9

Saw cut: 1-side 2-side 1-side 1-side 2-side 1-side

All tests were performed at the 6 kN hydraulic fatigue test machine of the TU Delft Aerospace Structures & Materials Laboratory. The tests were conducted at 10 Hz with an applied stress amplitude of 60 MPa and a stress ratio R = 0.1. During the fatigue test at certain fatigue life intervals, the crack length was measured at one side of the specimen. At the other side the images for the DIC were taken in which afterwards the crack length was measured. From the crack length data the FCG rate (da/dN) was determined using the ASTM 7-point incremental polynomial method (ASTM E647). The choice for the different configurations of FS welds and distances between the initial hole with saw-cut and the FS weld centre, is based upon the yield strength and residual stress profiles (Figure 2) and previous research [1,2]. The different test configurations made it possible to include or exclude the effect of the residual stress or the effect of yield strength in the FS weld on the FCG properties. For the 0º test configurations the residual stresses are acting in the same direction as the applied load and thus have an effect on the FCG rate. The distance between the hole and the weld centre for these specimens were determined by the location of the FS welding in the limited available material. Therefore the location of the hole is different for all the specimens, resulting in different crack lengths at which the effect of the FS weld on the FCG properties is manifested. Interesting for this configuration is the comparison between the FCG rates in AA7075-T6 and AA6013-T4, because the latter material has no change in yield strength in the FS weld, but only a residual stress field. For AA7075-T6, in contrary, the yield strength is reduced more than 30 % in the FS weld. Furthermore, no significant change of crack path is expected because the residual stresses are perpendicular to the crack growth direction. In the 45º test configuration an effect of both residual stress and yield strength is expected. Moreover, it is expected to see some change of FCG direction because the principle stress at the crack tip is rotated due to residual stresses. Besides, a rotation of the strain field is expected because the regions with low or high yield strength have an orientation of 45º with respect to the crack path.

626

H.J.K. Lemmen, R.C. Alderliesten and R. Benedictus

In the 90º test configuration the FCG properties are only affected by the yield strength because the residual stresses are oriented parallel to the crack and thus perpendicular to the applied load. The effect of the residual stresses on the lateral contraction is not relevant because the stresses are in the elastic region, and thus the lateral strain is the same as without residual stresses. Off course the real lateral stresses in the specimen are reduced, but that does not affect the plastic zone. Because the yield strength is the important parameter for this test, the locations of the hole and the saw-cut are for both AA2024-T3 and AA7075-T6 chosen at typical points of the yield strength profiles (Figure 2). It is expected that regions with high yield strength lead to small plastic zones and thus less crack closure resulting in higher FCG rates. Visa versa, low yield strength should lead to lower FCG rates. For AA2024-T3 the initial crack was positioned at y = 0 mm, at y = 3 mm and at y = 6.5 mm in the FS weld, for which the latter is tested twice to enable DIC measurements from both sides of the FS weld. Because the trapezium-shaped FS weld through the thickness of the alloy (see Figure 1b), also the yield strength profiles are different through the thickness of the FS weld [3], resulting in different strain fields. For AA7075-T6 the centre crack was placed at the location with minimum yield strength; y = 7 mm, while in the other test the centre crack was positioned at the location with the largest gradient of yield strength y = 10 mm.

RESULTS In this section the experimental results are presented using two types of figures. In the first type the FCG rate (da/dN) versus the crack length (a) are plotted for a certain specimen configuration and alloy. The other type presents the crack tip strain field measured by DIC, using contour plots. The first type does not need an explanation, but for the latter the procedure to compose these figures will be explained before the actual results are presented. The main goal of the contour plots is to compare qualitatively the plastic zone geometry from a fatigue crack in an FS weld with the plastic zone geometry from a fatigue crack in the base material. Therefore, contours from different tests are plotted which are obtained from the strain field measured by DIC. Each contour represents a strain level in the strain field around a fatigue crack with a certain crack length. To enable the comparison, the contours belonging to the same crack lengths are aligned. Along with the contours from the tests on FS welded material, always one contour from the representative base material test is plotted. Together with the contours, circles are plotted at each crack length, to indicate the theoretical size of the plastic zone according to the theory from Irwin and Dugdale.

Strain field analyses of fatigue cracks in FS welded joints

627

In most figures the strain level of the contour represents the engineering yield strain of the base material. However, sometimes another strain level is presented because the contours from the engineering yield strain is either too large to be visible in the image, or too small with respect to the resolution of the DIC measurement. In general, it does not matter which strain level is plotted as long as all the contours in one figure represent the same strain level, because they are only presented for comparison purposes. In all the contour plots a schematic notch with saw cut is plotted together with a crack path. This crack path belongs to the test which is first mentioned in the legend. For clarity, not all the tests presented in the figure followed the same crack path, but the contours are plotted on top of each other to make the comparison easier. Furthermore, although the contours do not always represent the yield strain, they are an equivalent for the plastic zone, and thus in the following text is spoken of ‘plastic zone geometry’ or ‘plastic zone size’. In the figures where this is appropriate the centre of the FS weld is shown together with the location of the edge of the weld tool. The latter is not the edge of the FS weld, because the HAZ extends beyond this line, but this is the edge of the nugget at the front side of the FS weld. The specimen numbers indicated in the contour plots correspond to the specimen numbers in the FCG rate figures with the same test configuration and alloy. Results 90º specimens Figure 5 shows the FCG rates measured in FS welded AA2024-T3 in 90º test configuration with the hole and saw cut at different locations in the FS weld. The FCG rates measured at y = 0 mm and 3 mm from the centre of the FS weld are initially equal to the FCG rates measured in the base material. The fatigue crack at y = 3 mm shows FCG retardation at a crack length of a = 8.5 mm after which the FCG rates increase rapidly at a crack length of a = 10.5 mm. The FCG rates at the location y = 6.5 mm from the weld centre are greater than the base material FCG rates for all crack lengths and the same for both tests up to a = 6.5 mm after which the curves deviate from each other. The large difference in FCG rate is remarkable because both test specimens have the same test configuration and geometry. Figure 6 displays the contour plots from the 90º configuration tests at locations y = 0 mm and y = 3 mm together with the base material. The crack lengths for which the contours are plotted correspond to the typical locations in Figure 5, i.e. the undisturbed FCG rate at a = 5 mm, the point of FCG retardation at a = 8 mm and the point where the FCG rate of specimen 2 increases at a = 10.5 mm.

H.J.K. Lemmen, R.C. Alderliesten and R. Benedictus

628

5.E-03 base AA2024-T3 FS weld, 90º config, spec 1, y = 0.0mm FS weld, 90º config, spec 2, y = 3.0mm FS weld, 90º config, spec 3, y = 6.5mm FS weld, 90º config, spec 4, y = 6.5mm

σ∞ m

3.E-03

6. 5m

da/dN [mm/cycle]

4.E-03

2.E-03

mm 0.0 e bas

m 6.5m

90º conf. σa=60MPa R=0.1

1.E-03

3.0mm

σ∞

0.E+00 0

5

10

15

20

a [mm]

Figure 5, da/dN versus a for FS welded and base AA2024-T3; 90º configuration

a=5.0mm Crack path a=8.0mm Welding direction

a=10.5mm

Figure 6, Contour plots FS welded AA2024-T3; 90º configuration; location y = 0.0 mm and y = 3.0 mm; front side FS weld At a = 5 mm the contours are hardly larger than the resolution of the DIC measurement, but the small butterfly wings are visible in the data. At a = 8 mm the contour for specimen 2 is larger than the contour for the base material, which could correspond to the subsequent FCG retardation. A larger plastic zone will result with a small delay in FCG retardation, because the fatigue crack has to grow into the plastic zone before crack closure can occur. The contour of specimen 1 shows a more circular shape instead of the butterfly shape visible for the other two

Strain field analyses of fatigue cracks in FS welded joints

629

contours. However, the height is the same as for the base material, which could explain why specimen 1 has the same FCG rate as the base material. At a = 10.5 mm the contours of all three specimen have the same height, although the shape of specimen 1 is still not a butterfly shape. The contours from specimen 3 and 4 with the fatigue crack at location y = 6.5 mm are shown in Figure 7 for three crack lengths. There are no significant differences observed at the three crack lengths, except that the size of the contours is increasing with the crack length. The geometry of both specimen 3 and 4 are asymmetric, indicating a gradient in the yield strength. For both specimens the lower butterfly wing is larger than the upper. The sizes of both contours in the FS welded specimens are larger than for the base material test, even though the FCG rates are higher. This seems to contradict with the statement that a large plastic zone leads to a lower FCG rate.

Crack path Welding direction

a=6.5mm a=9mm

a=11mm

Figure 7, Contour plots FS welded AA2024-T3; 90º configuration; location y = 6.5 mm; front & back side FS weld Figure 8 shows the FCG rates for FS welded AA7075-T6 in 90º test configuration. Specimen 2 (y = 10 mm) has equal FCG rates as the base material up to a = 9 mm where the FCG rate increases. Specimen 1 shows a lower FCG rate than the base material for almost all crack lengths. However, at a = 13 mm the FCG rate becomes higher than the base material FCG rates.

H.J.K. Lemmen, R.C. Alderliesten and R. Benedictus

630

5.E-03 base AA7075-T6 FS weld, 90º config, spec 1, y = 7.0mm

da/dN [mm/cycle]

4.E-03

FS weld, 90º config, spec 2, y = 10.0mm

σ∞

3.E-03

m m .0 10

se ba

2.E-03 90º conf. σa=60MPa R=0.1

mm 7 .0 1.E-03

σ∞ 0.E+00 0

5

10

15

20

a [mm]

Figure 8, da/dN versus a for FS welded and base AA7075-T6; 90º configuration

a=5mm Crack path Welding direction

a=9.0mm

a=12.5mm a=20mm

Figure 9, Contour plots FS welded AA7075-T6; 90º configuration; location y = 7.0 mm and y = 10.0 mm; front side FS weld Figure 9 presents the contour plots for FS welded AA7075-T6 in 90º test configuration. The geometry of the plastic zone for the base material is not visible within the resolution of the DIC, but the circles give a good representation of the plastic zone size in the base material. Specimen 2 in the figure does not have a significant larger plastic zone than the base material except for the crack length a = 20 mm. Specimen 2 in contrary, shows a significant larger plastic zone, which was expected because the crack grows through a region (y = 7 mm) with a 30 % lower yield strength compared to the base material. The geometry of this contour, which can only be distinguished for the two largest crack lengths in the figure, has a butterfly geometry at a = 12.5 mm, but a more asymmetric geometry at a = 20 mm. The asymmetric geometry is also found for the specimen 1 at a = 20 mm.

Strain field analyses of fatigue cracks in FS welded joints

631

Results 0º specimens Figure 10 and Figure 11 display the FCG rates versus the crack length a, for FS welded AA7075-T6 and AA6013-T4 respectively. For both alloys a change in FCG rates is observed when the fatigue crack grows towards the FS weld. Even though the effect is significant larger for FS welded AA7075-T6, FS welded AA6013-T4 shows the same trend. For both alloys, the FCG rate increases at 10 to 10.5 mm from the FS weld centre. After a few millimetres (for AA7075-T6 at 6.5 and 7 mm from the weld centre; and for AA6013-T4 at 5.0 and 5.5 mm from the weld centre) the FCG rates decrease below the FCG rate of the base material. After this initial decrease, the FCG rates in FS welded AA7075-T6 increase rapidly before the FS weld centre is reached. In AA6013-T4 the FCG rates do not suddenly increase, but follow a trend comparable to the base material.

da/dN [mm/cycle]

Tool edge, spec 2, right

3.E-03

Weld centre, spec 1

FS welded AA7075-T6; 0º conf; spec 2

σ∞

Weld centre, spec 2

FS welded AA7075-T6; 0º conf; spec 1 4.E-03

Tool edge, spec 1, left

base AA7075-T6-3, 1 saw cut

Tool edge, spec 2, left

5.E-03

0º conf. σa=60MPa R=0.1

2.E-03

σ∞ 1.E-03

0.E+00 0

5

10

15

20

25

a [mm]

Figure 10, da/dN versus a for FS welded and base AA7075-T6; 0º configuration For both alloys, the location of the FS weld has a large impact on the fatigue life of the specimen, because a FCG rate increase at small crack lengths, decreases the fatigue life much more than the same FCG rate increase at large crack lengths. The AA6013-T4 specimen with the FS weld at y = 11 mm from the notch, had a 20 % shorter fatigue life than the specimen with the FS weld at y = 19 mm from the notch. Figure 12 shows the contour plots for specimen 2 in FS welded AA7075-T6 at two strain levels. For both strain levels the contours, and thus the plastic zone, in the FS welded material are larger than the contours from the base material. The plastic zone at a = 7.5 mm, which is outside the FS weld, is already larger. Once the fatigue crack reaches the other side of the FS weld at a = 22 mm, the shape of the plastic zone is changed from a butterfly into an oval geometry. It must be noted that the size of the plastic zone for FS welded AA7075-T6 is much larger in the 0º test configuration than in the 90º test configuration shown in Figure 9.

H.J.K. Lemmen, R.C. Alderliesten and R. Benedictus

632 2.E-03

da/dN [mm/cycle]

Weld centre, spec 1

Tool edge, spec 2, left

0º conf. σa=60MPa R=0.1

1.E-03

Tool edge, spec 1, right

FS welded AA6013-T4; 0º conf; spec 1

σ∞

Tool edge, spec 1, left

Weld centre, spec 2

base AA6013-T4; 1 saw cut FS welded AA6013-T4; 0º conf; spec 2

σ∞

0.E+00 0

5

10

15

20

a [mm]

Figure 11, da/dN versus a for FS welded and base AA6013-T4; 0º configuration

Weld centre

Tool edge

Welding direction

Tool edge

a=5mm

a=7.5mm

Crack path

a=10mm

a=22mm a=15mm

Figure 12, Contour plots FS welded AA7075-T6; 0º configuration; back side FS weld Figure 13 shows the contour plots for specimen 2 in FS welded AA6013-T4 in 0º test configuration. Just as for FS welded AA7075-T6 the plastic zone for FS welded AA6013-T4 is larger than the base material plastic zone, but the differences are smaller. Because the yield strength of AA6013-T4 is low, the size of the plastic zone is large, almost equal to the width of the FS weld. Such a large plastic zone results in a different, more equally, redistribution of residual stresses than for an alloy with high yield strength and thus a small plastic zone like AA7075-T6.

Strain field analyses of fatigue cracks in FS welded joints

633

Tool edge

Welding direction

a=5.0mm

a=9.0mm Tool edge

a=16.0mm

Weld centre

Crack path

Figure 13, Contour plots FS welded AA6013-T6; 0º configuration; front side FS weld Results 45º specimens Figure 14 present the contours measured in the 45º test configuration for FS welded AA2024-T3. The crack path shown in this figure was not observed in all three specimens to the same extent. However, the geometries of the plastic zones show a high degree of similarity. In all three specimens the plastic zone size in the FS weld was larger than the plastic zone of the base material; moreover, they all changed from a butterfly shape into a single winged butterfly in the centre of the FS weld. Towards the edge of the FS weld, the geometry changes back to a butterfly shape with a strong orientation parallel to the FS weld. The FCG direction rotates perpendicular to the FS weld in the nugget region. Once it reaches the other side of the nugget region, the fatigue crack plane rotates back to the original direction, perpendicular to the applied load.

H.J.K. Lemmen, R.C. Alderliesten and R. Benedictus

W

el di ng

di re ct

io n

634

a=5mm

a=7.5mm

a=12mm

W

To ol

el d

ed ge

ce nt re

Crack path a=20mm

Material flows in Nugget

W

el di ng

di re ct io n

Figure 14, Contour plots FS welded 2024-T3; 45º configuration; front and back side FS weld

To ol

ed ge

a=7.5mm

a=20mm a=5mm

W

el d

ce n

tre

a=10mm a=13mm

a=15mm Crack path

Figure 15, Contour plots FS welded AA7075-T6; 45º configuration; front side FS weld Figure 15 and Figure 16 present the contour plots measured in FS welded AA7075-T6 in 45º test configuration. Initially, the same behaviour of crack path

Strain field analyses of fatigue cracks in FS welded joints

635

and plastic zone geometry is observed in this test configuration as for FS welded AA024-T3. The plastic zone is larger than the plastic zone of the base material and changes through the FS weld from a butterfly shape into a single wing. Also both specimens showed a change of FCG direction close to the centre of the FS weld, into a direction perpendicular to the FS weld. However, a large deviation was observed in one specimen when the fatigue crack grew out of the nugget region (Figure 16). In that case the fatigue crack did not turn back perpendicular to the applied load, but continued to grow upwards (more than 30 mm) parallel to the applied load. The specimen finally failed statically because the crack at the other side of the hole reached the edge of the specimen. The main difference between the specimens in Figure 15 and Figure 16, is the direction of welding which is opposite for both specimens. Therefore, the orientation of the onion rings in the nugget is different where the fatigue crack exits the nugget region (see circular pattern in the figures), which could mean that the onion rings determine in which direction the fatigue crack leaves the nugget.

A

B W

W

el di ng

el d

ce n

Speckle pattern for DIC

Material flows in Nugget a=20mm

di re ct io n

tre

Crack path

Weld centre

Crack path

a=15mm

a=5mm

To ol

ed

ge

a=7.5mm a=10mm

a=13mm

Notch and saw cut

Figure 16, A: Contour plots FS welded AA7075-T6; 45º configuration; back side FS weld; B: image of specimen with the fatigue crack

DISCUSSION The goal of the study presented in this paper was to repeat and investigate the remarkable FCG behaviour through an FS weld. Especially the change in FCG direction was one of the phenomena which draw attention in previous studies. As stated in the introduction, the authors believe that the change of yield strength in

636

H.J.K. Lemmen, R.C. Alderliesten and R. Benedictus

the FS weld and the presence of residual stresses play an important role in the choice from the fatigue crack to grow in a certain direction. Besides, the yield strength affects the FCG rate directly through the size of the plastic zone and thus the amount of crack closure in the wake of the crack tip. This section discusses first the results from the three test configuration individually. Initially the relation between the yield strength, residual stress and FCG behaviour is discussed. However, in each test configuration some behaviour was observed which does not comply with what was expected, like the influence of the microstructure in the nugget on the FCG rate. These observations will be discussed in the last paragraph. Discussion 90º specimens The results from the specimen with the FS weld in 90º test configuration should confirm the effect of yield strength on the FCG rates because the residual stresses are negligible perpendicular to this direction. As can be seen in Figure 5 and Figure 8, the FCG rates correspond initially with the expected effect of yield strength on the FCG rate, i.e. the fatigue cracks situated at a peak of the yield strength profile (Figure 2) grow faster than the fatigue cracks in regions with lower yield strength. In general this behaviour is supported by the size of the plastic zones in the contour plots (Figure 6, Figure 7 and Figure 9), i.e. low yield strength results in large plastic zones and thus more crack closure resulting in lower FCG rates. However, in several specimens the FCG rates change suddenly in a way that cannot be explained by the yield strength or the residual stress profile, for instance specimen 2 in Figure 5 at a = 10.5 mm. The next paragraphs will discuss the individual tests and describe the expected and unexpected behaviour. The behaviour of specimen 2 in Figure 5 from FS welded AA2024-T3 (y = 3 mm) can only partly be explained with the contour plots in Figure 6. The FCG retardation at a = 8.5 mm is preceded by an increase of the plastic zone at a = 8 mm. The FCG retardation is explained by this increase of plastic zone, but the behaviour of the plastic zone is not. One possibility could be the crack path which turns towards a region with a lower yield strength, however, this is not the case for this specimen. In this specimen, the crack path turned towards the centre of the FS weld where higher yield strength is present. After a few millimetres of FCG retardation the, the FCG rates increase suddenly at a = 10.5 mm, without any warning in the plastic zone geometry. The fracture surface shows at a = 10.5 mm a sudden change in appearance; the crack plane orientation is bended in different directions and the roughness of the surface has been changed. Moreover, black residues are present indicating fretting, which can be an indication of FCG in shear. In fact all specimens with the fatigue crack near or in the nugget show black residues at the fracture surface and sudden changes of crack plane orientation.

Strain field analyses of fatigue cracks in FS welded joints

637

Both specimen 3 and 4 in Figure 5 at y = 6.5 mm in FS welded AA2024-T3 have a much higher FCG rate than the base material, even though the plastic zones in Figure 7 are larger. However, the geometries of the plastic zones are highly asymmetric, especially for specimen 4 which developed only one butterfly wing. Apparently an asymmetric plastic zone does not lead to effective crack closure. Most likely is the compressive region during unloading for an asymmetric plastic zone not at the crack tip, but above or below the crack plane. This also explains the difference between the FCG rates of the two specimens beyond a crack length of a = 6.5 mm. The crack path of specimen 3 (shown in Figure 7) grows through the largest wing of the butterfly shape, whereas specimen 4 grows above the largest butterfly wing. This asymmetric geometry of the plastic zone is a result of the yield strength profile because the location y = 6.5 mm is at a gradient in the profile (Figure 2) with at one side a peak and at the other a valley. Clearly, the butterfly wing of the plastic zone at the valley side is larger than at the peak side. The lower FCG rates measured for specimen 1 in Figure 8 (y = 7 mm) in FS welded AA7075-T6, is indeed a result of the lower yield strength at that location. The plastic zone size in the FS weld is at least twice as large as the theoretical plastic zone of the base material (Figure 9). Unfortunately the plastic zones are small with respect to the resolution of the DIC, but the theoretical plastic zone size is representative for the plastic zone in the base material. The plastic zone of specimen 1 is in reality even larger than the contour which is plotted because the plotted engineering yield strain of the base material is larger than the reduced yield strain in the FS weld. The sudden increase of FCG rates for specimen 2 at a = 9 mm (Figure 8) coincide again with changes in the appearance of the fracture surface. The changes in the fracture surface include change of crack plane and roughness of the surface. Besides, the fatigue crack starts to grow towards the centre of the FS weld. The plastic zone for this specimen shows a highly asymmetric shape pointed towards the centre of the FS weld. The question now rises whether the fatigue crack grows towards the FS weld centre due to the steep gradient in yield strength or another reason. In principle, the fatigue crack tip does not “feel” a gradient in yield strength. The yield strength only affects the strain field and thus the stress field. It is not unlikely that an asymmetric plastic zone, with high deformations at one side and almost no deformation at the other side, results in high shear stress and thus a rotation of the principle stress. However, the results from the DIC measurement showed only strain in one direction and it is not possible to obtain the principle stress with this measurement technique. Some more insight can be obtained by calculating the shear angle and principle strain orientation from the DIC data. However, the only way to show how the principle stress behaves in this situation is by Finite Element Analyses (FEA). At the end of this section it is further explained how FEA can be useful.

638

H.J.K. Lemmen, R.C. Alderliesten and R. Benedictus

In general, it can be concluded for the 90º test configuration, that the yield strength has an influence on the FCG rates, especially for small crack lengths and for alloys with low ductility like AA7075-T6. However, beside the size also the geometry of the plastic zone plays an important role. An asymmetric plastic zone due to a gradient in the yield strength profile, can results in an effect on the FCG rate which is the opposite of what was expected from the size of the plastic zone. Finally, an effect of the microstructure on the FCG behaviour is observed, which is not taken into account yet. In the last paragraphs of this section this phenomena is discussed further. Discussion 0º specimens The 0º test configuration is an interesting case for the effect of residual stresses on the FCG behaviour. Especially because two alloys were tested, AA7075-T6 and AA6013-T4, which have both significant residual stresses, but for the latter the yield strength is unchanged. Both alloys show FCG rate acceleration at 10 to 10.5 mm from the weld centre (Figure 10 and Figure 11). After this acceleration the FCG rates decline for both alloys. For FS welded AA7075-T6 the FCG rate decreases below the base material FCG rate, while for AA6013-T4 the FCG rates return to the base material FCG rates. The initial acceleration observed for both alloys, can only be attributed to the residual stress, because that location coincides with the edge of the residual stress field (Figure 2). Moreover, the yield strength is not yet reduced for AA7075-T6 at that location and for AA6013-T4 it does not change at all. The magnitude of the change in FCG rate is much smaller for AA6013-T4 than for AA7075-T6 (Figure 10 and Figure 11). The differences are related to the ductility of the two alloys. A ductile material as AA6013-T4 is less affected by residual stresses, because the residual stresses are redistribution over a larger region resulting in a lower stress level at the crack tip than for an alloy which is less ductile, like AA7075-T6. The reduction of yield strength in the FS welded AA7075-T6 implies more ductility and thus a positive effect on the FCG rates in the FS weld. Probably this is one of the reasons why the FCG rates decrease once the fatigue crack enters the FS weld. The increase of crack closure due to the reduced yield strength in FS welded AA7075-T6, is the other reason why FCG retardation occurs. As was seen in the 90º test configuration at y = 7 mm, a small increase of plastic zone has a large effect on the FCG rate in AA7075-T6. The increase of plastic zone observed in Figure 12 is much larger than in Figure 9, which could imply that the plastic zone formation is for the largest part a result of the residual stress. The sudden increase of FCG rates at 5 mm from the FS weld centre in AA7075-T6 cannot be explained by the presence of the residual stresses or the yield strength. Especially not, because the combination of residual stresses and reduced yield strength initially led to FCG retardation. Taking a closer look at the fracture surfaces of the specimens reveals that, at the point of acceleration, the fracture

Strain field analyses of fatigue cracks in FS welded joints

639

surface is changed. In one specimen the fracture plane follows a path which has a similar shape as the circular onion rings in the nugget, which suggests that this microstructure has low FCG resistance. In general, the residual stresses indeed influence the FCG rate, but may be compensated by plasticity as a result of low ductility. However, once the fatigue crack enters the nugget, other phenomena affect the FCG rate and path. This behaviour is further discussed in the last paragraph. Discussion 45º specimens For the 45º test configuration several interesting observations have been made. First of all, many similarities have been observed in the FCG behaviour between FS welded AA2024-T3 and AA7075-T6. In both alloys the geometry of the plastic zone develops in a similar way. The crack paths shown in Figure 14 and Figure 15 have a lot in common. Furthermore, the onion ring structure in the nugget seems to have a large influence on the crack path. The change of FCG direction can be a result of two phenomena; one is the orientation of the principle stress at the crack tip, which is rotated in the direction of the FS weld due to the residual stresses. The other explanation is that the rotation of the crack path is a result of the onion ring microstructure in the nugget. Especially the geometry of the crack path in Figure 16 together with the geometry of the onion rings in the nugget, confirm a relation between the two. However, the current test results do not explain why this FCG behaviour is observed. The main conclusion drawn from the 45º test configuration is that it is possible to change the FCG direction. Whether this is a result of the microstructure in the nugget or the yield strength and the residual stress field in the FS weld cannot be answered based upon the current results only. Despite the lack of an explanation, the results imply that the geometry of an FS welded joint has a large effect on the FCG behaviour. Influence of microstructure on FCG behaviour That fatigue cracks follow the circular microstructure has been observed before in a previous study on fracture surfaces in FI specimens [2]. Figure 17 show a fatigue crack growing through the microstructure of the nugget in FS welded AA2024_T3. The onion rings are clearly visible in the microstructure of the nugget. Moreover, it is clearly visible that the fatigue crack path follows the onion rings in the nugget. Apparently a path exists along the onion rings with a low fatigue crack growth resistance. Two possibilities for the low FCG resistance can be identified; one is the existence of brittle grain boundaries between the different material flows. This implies that the fatigue crack growth is inter-granular. The other possibility is related to the low angle grain orientation in the material flow. The colours in Figure 17 indicate the orientation of the grains, which means that grains with the same colour have the same orientation, and thus the slip planes in the crystals are

H.J.K. Lemmen, R.C. Alderliesten and R. Benedictus

640

equally oriented. This is the case for the material flows in the nugget and thus the slip planes in the grain structure are aligned which implies a low FCG resistance. This can explain the observed crack paths in several specimens in combination with the increased FCG rate. To investigate whether one of those presented solutions is indeed the reason for the FCG behaviour through the nugget, a detailed study of the fracture surfaces is required. It is possible to obtain the orientation of the grain structure at the fracture surfaces using Electron Back Scattered Diffraction (EBSD). Optical and Scanning Electron Microscopic (SEM) can answer which FCG mode is present, or whether there is trans-granular of inter-granular FCG.

Path of fatigue crack through nugget

Hole

Crack path

Nugget with Circular micro structure (“onion rings”) due to material flows around the pin

Direction of crack growth

Figure 17, Fatigue crack through micro structure of FS welded AA2024-T3, 0º test configuration, source: [2] The effect of an FS weld on the FCG direction In this study DIC was used to obtain the behaviour of the material in front of the crack tip. However, those results did not always explain the FCG behaviour. First of all, because no distinction can be made between an increase of plastic zone due to residual stress, or due to lower yield strength. Both have the opposite effect on the FCG rate. Another, more fundamental question is how the plastic zone affects the FCG direction. In general the orientation of the principle stress at the crack tip determines the direction of FCG. In homogeneous materials the plastic zone is a result of the stress field and thus the applied load. The applied load determines the orientation of the principle stress at the crack tip. However, in these tests the plastic zone geometry is not only a function of the stress field but also the yield strength profile in the FS weld. Therefore the yield strength affects the geometry plastic zone and with that the stress field, including the orientation of the principle stress. This means that the relation between the applied load and the orientation of the principle stress at the crack tip is changed. Especially at high yield strength gradients, the plastic zone is highly asymmetric and thus the orientation of the principle stresses is affected most.

Strain field analyses of fatigue cracks in FS welded joints

641

To provide an answer to the last question, FEA can be a useful tool. to understand how the yield strength and residual stresses in the FS weld affect the stress and strain field around the crack tip. An FS weld can be modelled with a yield strength profile and a residual stress profile enabling to investigate the effect of each individually. If the yield strength and the residual stress are indeed the only responsible parameters for the FCG rate, than the same geometries of the plastic zone should be obtained as for the tests. Moreover, FEA gives additional information about the orientation of the principle stress at the crack tip, which could also explain the FCG direction in the 45º test configuration. For clarity, it is not proposed to perform an FCG analyses using FEA.

CONCLUSIONS 1- The yield strength in an FS weld affects the FCG rate. However, the magnitude of the effect is highly dependent on the orientation of the FS weld with respect to the FCG direction. 2- The microstructure in the FS weld, indicated by the onion rings, creates a path with low FCG resistance. This leads to higher FCG rates and is probably also responsible for the change of FCG direction in some test configurations. 3- DIC is a useful technology to observe the strain field in the vicinity of a fatigue crack which provides understanding of the FCG behaviour. 4- FEA can be used to understand the effect of the residual stress field and the yield strength profile on the stress field in the vicinity of a crack tip in an FS weld.

RERENCES [1]

[2]

[3]

H. J. K. Lemmen, R. C. Alderliesten, J. J. Homan, R. Benedictus, (2007), Fatigue crack initiation behaviour of friction stir welded joints in aluminium alloy, in: International Committee on Aeronautical Fatigue, Napels. H. J. K. Lemmen, R. C. Alderliesten, J. J. Homan, R. Benedictus, The influence of residual stresses in friction stir welded joints on the fatigue crack growth properties, in: D. G. Pavlou (Ed.) Computational & Experimental Analysis of Damage Materials, Transworld Research Network 2007, pp. 127-144. H. J. K. Lemmen, R. C. Alderliesten, R. R. G. M. Pieters, R. Benedictus, (2008), Influence of Local Yield Strength and Residual Stress on Fatigue in Friction Stir Welding, in: 49th AIAA/ASME/ASCE/AHS/ASC Structures, Structures dynamics, and Materials Conference, AIAA, Schaumburg IL, p. 22.

642

H.J.K. Lemmen, R.C. Alderliesten and R. Benedictus

[4]

H. J. K. Lemmen, R. C. Alderliesten, R. Benedictus, J. C. J. Hofstede, R. Rodi, (2008), The power of Digital Image correlation for detailed elasticplastic strain measurements, in: M.-K. Nikolinakou, G. Tsekouras, V. Gekas, D. Pavlou (Eds.), Engineering Mechanics, Structures, Engineering Geology (EMESEG '08), WSEAS Press, Heraklion, pp. 73-89. M. Pacchione, S. Werner, N. Ohrloff, (2007), Design principles for damage tolerant butt welded joints for application in the pressurized fuselage, in: ICAF, Napels. R. S. Mishra, Z. Y. Ma, (2005), Materials Science and Engineering, vol. 50, p. 1-78. M. J. Jones, P. Heurtier, C. Desrayaud, F. Montheillet, D. Allehaux, J. H. Driver, (2004), Scripta Materialia, vol. 52, p. 693-697. K. V. Jata, (2000), Materials Science Forum vol. 331-337, p. 1701-1712. A. F. Norman, I. Brough, P. B. Prangnell, (2000), Materials Science Forum vol. 331-337, p. 1713-1718. M. Guerra, C. Schmidt, J. C. McClure, L. E. Murr, A. C. Nunes, (2003), Materials Characterization, vol. 49, p. 95-101.

[5]

[6] [7] [8] [9] [10]

25th ICAF Symposium – Rotterdam, 27–29 May 2009

A STUDY OF INTERACTION AND COALESCENCE OF MICRO SURFACE FATIGUE CRACKS IN ALUMINIUM 7050 W Hu, Q Liu and S Barter Defence Science and Technology Organisation, 506 Lorimer Street, Fishermans Bend, Australia 3207

Abstract: Multi-site cracks may occur in structural components in shallow notches, in corroded areas, or as a result of multiple notches such as in fuselage lap joints, and they are inherently a stochastic phenomenon. The interaction and coalescence of multiple cracks pose a challenge to the accurate prediction of fatigue crack growth rates and the residual strength of these components. It is, therefore, important to quantify the interaction and coalescence of multiple surface cracks. In this paper we study the interaction and coalescence of multiple surface cracks in 7050-T7451 aluminium plates. Experiments have been carried out under constant amplitude and spectrum loading, using coupons containing artificial micro-cracks with a surface length of ~50 µm. Within the framework of linear elastic fracture mechanics, three-dimensional finite element analyses were conducted to determine the interaction between cracks. The interaction factors were subsequently used for crack growth analysis of coplanar cracks. It was found that the numerical results correlated well with the experimental data for both constant amplitude and spectrum loading. It is envisaged that the outcome of this study will form a part of a stochastic model, leading to improved fatigue life assessment in the short crack regime for the material considered.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 643–659. © Commonwealth of Australia 2009. Published by Springer Science+Business Media B.V. Dordrecht.

W. Hu, Q Liu and S. Barter

644

INTRODUCTION Multi-site cracks may occur in structural components in shallow notches, in corroded areas, or as a result of multiple notches such as in fuselage lap joints, and they are inherently a stochastic phenomenon. The interaction and coalescence of multiple cracks pose a challenge to the accurate prediction of fatigue crack growth rates and the residual strengths of components, as the interaction of cracks may accelerate or retard their growth, and the resulting combined crack may lead to a lower residual strength more quickly than cracks that grow individually. It is, therefore, important to quantify the interaction and coalescence of multiple small surface cracks. Fatigue life of a structure depends on, among other factors, (1) the size and distribution of fatigue crack initiating discontinuities present in the structure and the how cracks initiate at these discontinuities; (2) the growth and coalescence of these multiple cracks to become a dominant single crack and (3) the propagation of the single crack to cause final failure. Research shows that a relatively large part of the fatigue life is spent in the initial formation and propagation of micro and small cracks, regardless of the level of applied stress and the type of loading history [1]. The work of Newman et al [2] also indicated that crack propagation from a microstructural discontinuity to small cracks represented 50 ~ 90% of the total life of specimens. DSTO research on 7000 series aluminium alloys also resulted in a similar conclusion [3, 4]. Thus, the initiation, growth, interaction and coalescence of multiple small surface cracks will play a crucial role in the fatigue performance of aluminium alloys in airframe components. A number of researchers have investigated the growth, interaction and coalescence of multiple cracks and their implication on fatigue life prediction. Kamaya [5, 6] studied the growth of interacting multiple cracks in the context of stress corrosion cracking for boiler vessels in the long crack regime. The stress intensity factors were calculated using the body force method [7]. He noted that the interaction between cracks is influenced not only by the relative position but also by the relative lengths of cracks. A model was developed in [8] to account for crack coalescence in low cycle fatigue an approximate solution of J . Based on the postulation that the crack growth acceleration just before coalescence was cancelled out by the deceleration immediately after coalescence, it was suggested that, for the cases they considered, the long crack growth law using a throughthickness crack may be used for the analysis of multiple crack link-up, as long as the cracks were re-defined after coalescence. In this paper we present the results of a study on the growth, interaction and coalescence of multiple micro surface cracks in plates of 7050-T7451 aluminium alloys. A more comprehensive coverage of the experimental details may be found in reference [9]. Experiments have been carried out under constant amplitude (CA) and spectrum loading, using coupons containing crack-like slots to mimic micro-

Interaction and coalescence of micro surface fatigue crack in aluminium 7050

645

cracks with surface lengths of ~50 µm and depths of ~15 µm. The initial shallow surface “cracks” were arranged as coplanar with their planes perpendicular to the loading direction, with a short distance separating their near ends, or with their planes vertically offset while being perpendicular to the loading direction, with again a horizontal separation between their near tips (see the Section “Microcrack layouts”). Within the framework of linear elastic fracture mechanics, threedimensional finite element analyses were performed to determine the extent of interaction between cracks. This was investigated by comparing stress intensity factors at the adjacent crack tips to those at the outer tips for different combinations of crack lengths, horizontal separations and vertical offsets. The interaction factors were subsequently used for crack growth analysis for coplanar cracks. Although the initiation, propagation and coalescence of multiple surface cracks are inherently a stochastic process with regard to metallurgical variables, in this report we consider it as a deterministic process. It is envisaged that the outcome of this study will later be used to develop a stochastic model.

MATERIALS AND EXPERIMENTS Materials and Specimens The material used in this study was aluminium alloy 7050-T7451, a material widely used in commercial and military aircraft. The nominal mechanical properties of this material are: yield stress 470 MPa, ultimate strength 525 MPa, modulus of elasticity 72.0 GPa, plane strain fracture toughness 35 MPa m and Poisson ratio 0.33. The geometry of the plate specimens used in this study is shown in Figure 1. This specimen has a test section width of 25 mm and a thickness of 6.35 mm. In order to clearly observe crack initiation and coalescence, the areas of interest on the surface of the specimen were highly polished. Micro-Crack Layouts Natural multiple surface cracks initiate stochastically in location and size, but their orientations are more or less perpendicular to the applied tensile load, as demonstrated in Figure 2 [9]. The circle in the figure indicates where two micro cracks were in the process of linking up when this test was stopped. These natural micro cracks that were prepared to help understand the nature of linking micro cracks were generated in a coupon of the type noted above. The coupon was etched all over to produce small etch pits that would act as fatigue crack initiators. It was then peened all over except for a small “window” on the face of the specimen. This was done to confine the cracks of interest to the window making the finding of the cracks manageable, although this was still difficult. The specimen was then fatigue loaded to produce the cracks shown.

W. Hu, Q Liu and S. Barter

646

x 100

6.35

26 40

25 126 Unit: mm

Figure 1 The geometry of the specimens and the dimensions.

Figure 2 Natural surface cracks in aluminium 7050-T7451. The focus shows where two cracks link up. The loading direction is vertical. To study the behaviours of small cracks in the vicinity of other small cracks in a more systematic manner, crack-like slots were produced were produced in the coupon type noted, using the focus-ion beam technique [10]. A finely focused beam of gallium ions of about 20 nm was rastered over a sample surface, resulting in controlled sputter-removal of material. These slots are very crack-like and can be produced precisely with a controlled position and depth.

Interaction and coalescence of micro surface fatigue crack in aluminium 7050

647

Figure 3 (a) and (b) show the layouts of the surface “cracks” prior to cycling. These “cracks” are shown in two configurations referred to as “coplanar cracks” and “offset cracks”, respectively. The crack lengths are denoted a1, a2, and the horizontal and the vertical separations of the cracks are denoted s and h, respectively. Reference [9] contains more crack layouts including three, four and five co-planar cracks and non-coplanar cracks. Only the two configurations presented in Figure 3 are discussed here.

a1

a2 s

a2 a1

h

s

two co-planar cracks

two offset cracks

Figure 3 Configuration of artificial surface cracks. Loading Sequences The fatigue testing was carried out under load control on a digitally-controlled MTS 100 kN test machine. Two types of loading were applied, a fully-reversed CA loading with a peak stress of Speak=390 MPa and a spectrum load sequence. The spectrum loads were derived from the wing root bending moment sequence of a military fighter aircraft. The loading spectrum consists of 13,475 turning points, with a peak stress of Speak=390 MPa. In order to provide easily identifiable microfeatures on the fracture surfaces, the spectrum was modified slightly by adding five large compressive load cycles immediately before the peak stress. Studies have shown that these additional compressive load cycles have only a negligible effect on the total fatigue lives [11] of these coupons, with this spectrum and at this stress level. The loading cycles were applied at a rate of 2 Hz. Crack Measurement The crack length was measured using an optical microscope. Images of the surface of the specimens about the crack site were recorded at a predetermined number of cycles for CA loading and after the application of a number of blocks for the spectrum loading. It should be noted that only measurements of the surface length of the cracks are considered in this study, and these measurements were projected

648

W. Hu, Q Liu and S. Barter

back to the initial crack plane if the crack path deviates from this plane. Further analyses of the crack surface will be carried out reported separately.

EXPERIMENTAL RESULTS Coplanar Cracks Figure 4 shows the layout of two coplanar surface cracks and their link up after 1000 cycles of CA loading. It can be seen that the inner tips of the cracks do not approach each other in the anticipated crack plane; rather, they deviate from the crack plane and coalescence takes place outside the crack plane. Other images from [9] show that this is the typical fashion in which coalescence takes place. Melin in [12] suggested this should be the case. He thought that it was energetically unfavourable for coplanar fatigue cracks to remain coplanar at the commencement of coalescence [13].

Figure 4 Crack propagation and coalescence of two coplanar surface cracks under CA loading after 1000 cycles. Speak=390 MPa and R=-1. a1=a2=50 µm, s=50 m. Offset Cracks Figure 5 shows the link up of two offset cracks under CA loading after 3800 cycles. The initial crack lengths were about 50.6 m, and the horizontal and the vertical separation were 15.2 m and 26.8 m, respectively. After 600 cycles, a small amount of crack extension was noted at the adjacent tips. The lengths of

Interaction and coalescence of micro surface fatigue crack in aluminium 7050

649

these crack extensions were 7.2 and 5.6 m, respectively. It was also observed that after commencement of the crack extension, the cracks immediately deviated from their original direction. After 2800 cycles, both crack tips continued to grow at about 45 degrees to the loading direction. At 3800 cycles, the upper crack had passed over the lower crack’s extension and intersected the lower crack’s initial slot. It was noted that the lower crack had stopped propagating after 2800 cycles, probably due to the load shielding effect of the upper crack.

Figure 5 Crack propagation behaviour of two non-coplanar surface cracks after 3800 cycles under CA loading. Speak=390 MPa and R=-1. a1=a2=50 µm, s=15.2 µm and h=26.8 µm. Figure 6 shows a schematic of two coplanar initial cracks (slots) and the coalesced crack, where a represents the crack extension attained at the outer crack tips when the two inner crack tips link up. It should be noted that the crack propagation was measured by projecting the actual crack extension onto the anticipated crack plane. Figure 7 plots the experimental crack growth curves for the coplanar and the offset cracks subjected to CA loading. The crack growth data for the individual cracks before coalescence are represented by the hollow and filled symbols: the hollow and filled circles representing the two coplanar cracks, and the hollow and filled triangles representing the two offset cracks. The corresponding merged cracks are shown by the half-filled circles and triangles, respectively. For comparison, growth data for a single crack [9] are also plotted in the figure by stars.

W. Hu, Q Liu and S. Barter

650

Coalesced crack slot 2

slot 1

a1

a

s

a2

a

Figure 6 Schematic of two coplanar initial cracks and the coalesced crack. The crack propagation was measured by projecting the actual crack extension onto the anticipated crack plane.

Crack Length, microns

Single crack Coplanar crack 1 Coplanar crack 2 Offset crack 1 Offset crack 2 Coplanar coalesced Offset coalesced

Material: 7050-T7451 Load: CA, S=390MPa, R=-1

400

300

200

100

0 0

500

1000

1500

2000

2500

3000

3500

4000

Number of cycles

Figure 7 Experimental crack growth curves for coalesced and individual cracks subjected to CA loading. a1=a2= 50 m, s=50 m. For the coplanar cracks, coalescence occurred when the total crack length reached about 180 µm. Let Nc denote the number of cycles when coalescence occurs, and 2∆a the combined crack extension at the two outer crack tips at Nc, then when coalescence occurs we have

a1 + a 2 + s + 2∆a = 180 ,

(1)

Interaction and coalescence of micro surface fatigue crack in aluminium 7050

651

which gives ∆a = 15 µm. On the other hand, the average crack extension at each of the inner tips at that time would be s/2=25 µm. That is, there was an accelerated growth at the inner crack tips before coalescence, which is caused by the interaction between the cracks. This feature has been confirmed by experimental data for other cases reported in [9]. After coalescence, the merged crack appeared to behave the same as a single crack of the same length, as indicated by the half-filled circles and the stars. The effect of interaction on crack growth for the offset cracks is much more complicated. No attempt was made to analyse this further other than noting that the faster growth at one inner tip may retard the growth at the other inner tip, or even shield it from further growth. The combination of crack lengths, the separation and the offset may affect the path of crack growth, as hinted in Figure 5. Figure 8 shows the growth and coalescence of two coplanar cracks under spectrum loading. The results are very similar to those observed for CA loading. 1400 1200

Crack Length, microns

Failed at 6.26mm

Spectrum: IARPO3a, Speak=390 MPa Single crack Coplanar crack 1 Coplanar crack 2 Coplanar coalesced Offset crack 1 Offset crack 2 Offset coalesced

1000 800 600 400 200 0

0

5

10

15

20

25

30

35

40

Number of Blocks

Figure 8 Crack growth behaviour of two coplanar cracks under spectrum loading. Speak=390 MPa. a1=a2= 50 m, s=50 m.

W. Hu, Q Liu and S. Barter

652

NUMERICAL ANALYSES In this section, the interaction between two coplanar cracks and that between two offset cracks are quantified, based on the LEFM parameter KI. Using the numerically determined interaction factor and the material crack growth rate, the crack growth behaviour of the coplanar cracks are simulated and compared to experimental results. Stress Intensity Factors The stress intensity factors for the multiple micro cracks are calculated using a finite element method package, StressCheck [14]. The parametric modelling capability of StressCheck is particularly suited for the type of analysis considered here, as all the variables, including the crack lengths and crack tip separation, may be parameterised to reduce the work involved in model creation. StressCheck has an advanced method, the contour integral method [15, p. 227], implemented for the computation of mode I and mode II stress intensity factors in linear elastic fracture mechanics. This method is super-convergent, allowing the use a of relatively coarse mesh and low degree of polynomials for the accurate computation of stress intensity factors [16]. For three-dimensional analyses, the stress intensity factors for mode I and II are computed using the contour integral method over a circular path around the crack front at the point of interest. The procedure determines a cutting plane normal to the tangent to the crack edge at the point of interest, as shown in Figure 9, and extracts the global components of the stresses and displacements along a circular path contained in the cutting plane. These stresses and displacements are projected onto the cutting plane and integrated with the extraction function to compute the contour integral. The details of the method can be found in [16].

.

Point of interest Integration path and radius

Cutting plane normal to the crack front at the point of interest.

Crack front

Figure 9 Schematic of integration path for a given point on the crack front, for the calculation of stress intensity factors using the contour integral method.

Interaction and coalescence of micro surface fatigue crack in aluminium 7050

653

The overall geometry of the finite element model was a rectangular plate, as shown in Figure 10, with dimension corresponding to the specimen tested. The surface cracks were modelled as semi-circular cracks. For coplanar surface cracks, a quarter of the coupon was modelled, as shown in Figure 10, but for cracks with offsets the full coupon was modelled. The model is discretized using tetrahedral solid high order p-elements [15], with a boundary layer mesh attached to the crack front. Within the boundary layer, the mesh is refined in a geometric progression with predefined ratios to facilitate the accurate determination of stresses and strains in the vicinity of the crack front.

A-A

(a) Overall model.

(b) Finite element mesh, A-A view.

Figure 10 The finite element model for two coplanar surface cracks. Due to symmetry in geometry and loading, only a quarter of specimen section was modelled. Crack Interaction Finite element analyses were performed for a number of crack configurations, with or without vertical offsets, to investigate the interaction of two cracks within the framework of linear elastic fracture mechanics. Here, the interaction factor is defined as the ratio of stress intensity factors with and without the presence of another crack, i.e.,

f =

K I ,m K I ,s

,(2)

where the superscripts “m” (for multi crack) and “s” (for single crack) indicate whether K was calculated with or without the presence of another crack, respectively, and f denotes the interaction factor.

W. Hu, Q Liu and S. Barter

654

The interaction factors calculated for the two coplanar cracks and two offset cracks, are plotted in Figure 11 and Figure 12, respectively. The half crack length for the coplanar cracks was 0.05 mm. For the offset cracks, the half crack length was 0.5 mm and the offset was 0.2 mm.

Interaction factor f

1.07

Half crack length: 0.05 mm Inner Outer

1.06 1.05 1.04 1.03 1.02 1.01 1.00

0.00

0.05

0.10 0.15 Half separation, mm

0.20

Figure 11 Factor of interaction for two coplanar surface cracks. a1/2=a2/2=0.05 mm, s=0.05 mm.

Interaction factor

1.15 Offset: 0.2 mm Half crack length: 0.5 mm Inner Outer

1.10

1.05

1.00 0.00

0.05

0.10 0.15 0.20 Half separation, mm

0.25

0.30

Figure 12 Factor of interaction for two surface cracks with an offset. a1/2=a2/2=0.5 mm, h=0.2 mm.

Interaction and coalescence of micro surface fatigue crack in aluminium 7050

655

Crack Growth Analysis The crack growth analyses were carried out using CGAP [17], an in-house code based on the FASTRAN crack closure model [18]. The surface cracks were modelled as semi-elliptical cracks, with the major axis lying along the surface direction. The aspect ratio, which is the ratio of the crack length in the depth direction to that in the width direction, was initially taken as ~0.22, to simulate the shallow slots. The crack was then allowed to growth in both depth and width direction. As a first approximation, a three-step analysis was conducted using the long crack growth data reported in [19], supplemented by the short crack data developed in the current study. In the first step, one of the two surface cracks was analysed, taking into consideration the interaction factors developed above. When the cracks coalesce, satisfying Eqn (1), the life is recorded as Nc. In the second step, the same crack was analysed without considering the interaction factor. The crack length attained at Nc is recorded as ∆a , which represents the crack increment at the outer tips when coalescence took place as shown in Figure 6. In the third step, the crack is redefined by setting the initial crack size to a1+a2+s+2 ∆a , and resetting the aspect ratio to simulate a new shallow surface crack. Figure 13 shows a comparison of the numerical and experimental crack growth data for two coplanar cracks, before and after coalescence, subjected to CA loading. Figure 14 shows the same comparison for the cracks subjected to spectrum loading. A reasonable correlation was achieved for both the individual cracks and the coalesced cracks, although it must be noted that this is for crack length not depth. 400 Load: constant amplitude

Material: 7050-T7451

Crack Length, microns

σ = 390MPa, R= -1 300

200

Test: crack 1 Test: crack 2 Test: coalesced Analysis: individual Analysis: coalesced

100

0 0

500

1000

1500

2000

2500

Number of cycles

Figure 13 Comparison of numerical and experimental crack growth data for two coplanar cracks subjected to CA loading.

W. Hu, Q Liu and S. Barter

656

Crack Length, microns

1200

Spectrum: IARPO3a, S peak=390 MPa Single crack Coplanar crack 1 Coplanar crack 2 Coplanar coalesced Analysis: invidual Analysis: coalesced

1000 800 600 400 200 0 0

5

10

15

20

25

30

35

Number of Blocks

Figure 14 Comparison of numerical and experimental crack growth data for two coplanar cracks subjected to spectrum loading.

DISCUSSION The numerical results in Figure 11 show that if the distance between the inner tips of the two coplanar cracks is greater than 1 mm, the interaction factor becomes 1, indicating that the two cracks will grow independently of each other. A comparison of the results shown in Figure 11 and those from [9] that are not included in this paper further show that for larger cracks, the interaction starts at a large separation. When the ratio s / a ≤ 1 the interaction will be greater than 1.02, indicating a 2% boost in the crack driving force, and the interaction becomes stronger as s / a approaches zero. This result is consistent with the “non-interaction criteria” adopted in [20] for coplanar cracks. In contrast to coplanar cracks, the variation of the interaction factor for the offset cracks is more complicated, and further analyses are needed. From the viewpoint of crack growth prediction, coalescence may be considered to have taken place when the plastic zones at the adjacent crack tips meet [21]. The numerical results confirm the experimental observation that cracks grow slower in the surface direction for shallow cracks (when the aspect ratio is low), as the major portion of the externally supplied energy was spent on growing the crack in the depth direction. This is consistent with the experimental data. This happened at the

Interaction and coalescence of micro surface fatigue crack in aluminium 7050

657

initial stage of crack growth and after crack link up, as evidenced by the relatively flat parts of the crack growth curves plotted in Figure 13. Quantitatively, the numerical result deviates from the experimental ones, with a higher growth rate, probably due to the error introduced by the LEFM assumptions when the crack size is small and stress levels are high. Furthermore, the assumed crack aspect ratio at the early stage of crack growth is higher than that of the slots, and a higher aspect ratio would lead to faster crack growth in the surface direction. Another source of error came from the way in which the experimental data were acquired: the crack lengths were measured by projecting the actual crack length onto the anticipated crack plane, rather than being measured along the actual crack path. Hence, the actual crack growth rate would be faster than the experimental data plotted in Figure 13 if the lengths were measured along the actual crack path, although it must be noted that this would also require a correction for the crack growth direction relative to the loading in the FE analysis. Reasonable numerical results were obtained for two coplanar cracks subjected to CA loading based on the numerically determined interaction factors. The interaction between two offset cracks remains to be characterized. It is hoped that further analyses of the experimental data contained in [9] will help to reveal the relationship between the offset, separation, crack lengths and the path of crack growth. In addition, since the crack path severely deviates from the anticipated crack plane, mixed mode growth may need to be considered. Furthermore, for cracks of the dimension of grain sizes, it is always a challenge for the continuum mechanics approach. Micro mechanics approaches such as that advocated in [22] may need to be considered to obtain a fuller description of the complex effect of micro features on crack growth at this scale.

CONCLUSION The growth, interaction and coalescence of micro surface cracks in aluminium 7050-T7451 have been studied experimentally and numerically. Based on the experimental data and the three-dimensional finite element analysis, it has been found that the interaction accelerates the crack growth at the inner tips for two coplanar cracks, while its effect on two offset cracks is more complex. Based on the numerically-determined interaction factors, the growth of two coplanar cracks, before and after coalescence, was calculated using a crack closure model. The numerical results correlated well with the experimental data for both the constant amplitude loading and the spectrum loading. Further analyses need to be carried out to characterize the interaction between parallel surface cracks with vertical offsets.

W. Hu, Q Liu and S. Barter

658

ACKNOWLEDGEMENT The authors wish to acknowledge Ron Wescot of QinetiQ Australia for conducting some of the finite element analyses.

REFERENCES 1. Suresh, S. and R.O. Ritchie, Propagation of short fatigue cracks. International Metals Reviews, 1984. 29(6): p. 445-476. 2. Newman, J.C., Jr., Phillips, E.P., Swain, M.H. and Everett, R.A., Jr., Fatigue mechanics, an assessment of a approach to life prediction. Advances in Fatigue Lifetime Predictive Techniques, ASTM STP 1122, Eds., M.R. Mitchell and R.W. Landgraf, 1992: p. 5-27. 3. Barter, S.A., B. Boishop, and G. Clark, Defect assessment on F/A-18 488 bulkhead tested at ARL. 1991, Defence Science and Technology Organisation. 4. Barter, S.A., and Price, J., Effect of surface preparation treatments on fatigue life of 7050-aluminium alloy. Proceedings of the Structural Integrity and Fracture 2000 Symposium, 2000: p. 140-153. 5. Kamaya, M., A crack growth evaluation method for interacting multiple cracks. JSME International Journal, 2003. Series A, 46(1): p. 15-23. 6. Kamaya, M., Growth evaluation of multiple interacting surface cracks. Part I: Experiments and simulation of coalesced crack. Engineering Fracture Mechanics. In Press, Corrected Proof. 7. Nishitani, H. and Y. Murakami, Stress intensity factors of an elliptical crack or a semi-elliptical crack subject to tension. International Journal of Fracture, 1974. 10: p. 353-368. 8. Hoshide, T., M. Miyahara, and T. Inoue, Life prediction based on analysis of crack coalescence in low cycle fatigue. Engineering Fracture Mechanics, 1987. 27(1): p. 91-101. 9. Liu, Q., W. Hu, and S.A. Barter, Interaction and coalescence of multiple surface fatigue cracks aluminium 7050 and 2024. 2009, Defence Science and Technology Organisation. 10. Bunis, C.B., and Novaris, K.L., The utilization of focused ion beam (FIB) and field emission electron microscopy (FE-SEM) in failure analysis and process characterization. Corporate Reliability, Technical Brief, 1999. 11. Molent, L., S. Barter, and A. Green, Comparison of two F/A-18 aluminium alloy 7050-T7451 bulkhead coupon fatigue tests. 2004, Defence Science and Technology Organisation. 12. Melin, S., Why do cracks avoid each other? International Journal of Fracture, 1983. 23: p. 37-45. 13. Melin, S., Why do cracks avoid each other? Inter.J Fract., 1983. 23: p. 37-45. 14. ESRD, StressCheck. 2008, Engineering Software Research and Development Inc.

Interaction and coalescence of micro surface fatigue crack in aluminium 7050

659

15. Szabó, B.A. and I. Babuska, Finite Element Analysis. 1991: John Wiley and Sons, Inc. 16. ESRD, StressCheck Master Guide. Vol. 4. 2008: Engineering Software Research and Development Inc. 17. Hu, W. and K.F. Walker. An analysis of fatigue crack growth of a notched aircraft component under compression-dominated spectrum loading. in SIF2004 Structural Integrity and Fracture. 2004. Brisbane, Australia: Australian Fracture Group Inc. 18. Newman, J.C., Jr., FASTRAN II - A fatigue crack growth structural analysis program. 1992, NASA. 19. Sharp, P.K., R. Byrnes, and G. Clark, Examination of 7050 Fatigue Crack Growth Data and its Effect on Life Prediction. 1998, Aeronautical and Maritime Research Laboratory DSTO. p. 48. 20. Wanhill, R.J.H. and L. Schra, Short and long fatigue crack in 2024-T3 under Fokker 100 spectrum loading, in AGARD-R767. 1998. 21. Kuang, J.H. and C.K. Chen, Alternating iteration method for interacting multiple crack problems. Fat. Eng.Mat. and Struct., 1999. 22: p. 743-752. 22. Navarro, A. and E.R. de los Rios, Short and long fatigue crack growth: a unified model. Philosophical Magazine A, 1988. 57(1): p. 15-36.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

FRACTURE BEHAVIOUR OF SKIN MATERIALS OF CIVIL AIRPLANE STRUCTURES Boris Nesterenko1, Grirory I.Nesterenko2, Valentin N.Basov2 1

Institute of Mechanical Engineering (IMASH RAN) 2 Central Aerohydrodynamic Institute (TsAGI)

Abstract: This paper presents the results of analytical and test study on fracture behaviour of aluminium alloys applied in lower and upper wing surfaces and in fuselage skin. Characteristics of static strength, fatigue, residual strength and crack growth had been studied. Fatigue and crack growth had been analysed at cyclic loads and at standardized quasi-random load spectra. Fatigue life had been analyzed basing on linear hypothesis of damage accumulation. Crack growth was calculated using linear as well as retardation\acceleration models. Comparison of fracture behaviour of different aluminium alloys had been carried out.

INTRODUCTION Obviously, one of the ways to increase the operation life of airplane structure and to improve its damage tolerance characteristics is the application of advanced materials. For the purpose of profound material selection during the airplane design and the development of effective technical maintenance plans static and fatigue strength characteristics are being studied for structural materials Significant results on fatigue and crack resistance principles and regularities in aluminium alloys are stated in [1], [2]. The basis for these behaviours is the experimental data obtained on sheet specimen tests of 2024-T3 and и 7075-T6 types. Specifications on semi products of different aluminium alloys are given in [3] - [7].

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 661–683. © Springer Science+Business Media B.V. 2009

662

B. Nesterenko, G.Nesterenko, V. Basov

This paper presents the results of analytical and test study of fatigue and crack resistance characteristics of aluminium alloys applied in wing and fuselage skins of various airplane types. Tested in TsAGI were the following specimens made of sheets, plates and extrusions of the following Al-alloys: 7075 Т6, 7178 Т6, 7055 Т7751, 2024 Т3, 2024 Т351, 2324 Т39, 2524 Т3, V95ATV1, V95osT2, V96z 3pchT12, D16ATV, 1163ATV, 1163RDTV, 1161T, 1163T, 1163T7, 1441PT1. All the tests had been carried out according to the USSR\Russia aviation industry standards and ASTM standard. Fatigue life and crack growth duration under load spectra with variable amplitude had been investigated at four load programmes: block (typical flight) loading, TsAGI spectrum, Truncated TWIST [9] and Boeing load spectrum [10]. Crack growth analysis at random load had been carried out using Willenborg retardation model [11]. Basing on test results typical fracture behaviour of aluminium alloys applied in the structures of airplanes manufactured by different companies. More details on test and analysis methods used are given in [12], [13], [14]

TEST PROCEDURE Static and fatigue strength properties and crack resistance characteristics of Al -alloys had been determined in flat specimens’ tests. Geometry of the specimens is shown in Fig. 1. The specimens were cut out of full-scale airplane wing and fuselage skins, as well as from semi products from the metallurgical plants. Spectral analysis of the alloy composition was conducted in Russian Institute of Aircraft Materials (VIAM) and in Russian Institute of Light Alloys (VILS). Initial notches of the central cracks in crack growth test specimens have been made by electrical erosion. The characteristics of strength, fatigue strength, crack growth rate and residual strength of the alloys listed above have been determined experimentally in TsAGI during the specimen testing in electrohydraulic test machines having loading range from 10 to 1000 tons. Crack growth was observed visually using optical microscope and was also analyzed by means of quantitative fractography using electronic microscope. Residual strength Kapp has been determined on sheet specimens without elimination of buckling near the cracks. In additional TsAGI experiments on specimens with and without anti-buckling devices correction factors had been obtained for recalculation of Kapp values. These factors are equal approximately to 1.3 for 1.5-1.8 mm thick sheets and to 1.15 for the 4 mm thick sheets. For the sheets of thickness more than 8 mm this factor is about 1.0 . This paper contains Kapp values for the sheets without anti-buckling device. The values of Kapp for the alloys 2524-Т3, 7475-Т651 and 1441RТ1 were taken from [7], [6] and [5] correspondingly. The parameters of typical loading programs of specimens’ tests on fatigue and crack growth are shown in Fig.2 and Table IV.

Fracture behaviour of skin materials in civil airplane structures

663

Generalized results of conducted test-analytical study are presented in Figs.3-24 and Tables I – V.

MECHANICAL PROPERTIES The following static strength characteristics had been determined: ultimate strength σB, yield strength σ02, relative elongation δ5 after breakage. Test results are shown in Tables I - III. These tables also give percentage of lithium (Li), zirconium (Zr), ferrum agents (Fe) and silicium (Si) additives. Relative elongations δ5 in Al-alloys applied for wing lower surface skin and for fuselage significantly exceed relative elongations of the alloys applied for upper wing skin. The values of yield strength σ02 of the alloy 2324-Т39 are much higher than those of the rest alloys applied for the lower wing surface. Mechanical properties of Al - Li-alloy 144RT1 are at the same level as those of Al-alloys applied for fuselage skin. Ultimate and yield strengths of the alloys 7055-T7751 and V96ts-3pchT12 containing additives of zirconium are much higher than the corresponding properties of Al-alloys widely applied for upper wing skin of contemporary airplanes. It should be noted that structural materials are developed differentially for upper and lower wing surfaces and for fuselage. The level of design stresses in the upper wing skin is assumed basing on static strength of the wing upper surface under compressive loads. Stress levels in the wing lower surface and fuselage are defined mainly by requirements on fatigue strength and damage tolerance of the skin.

FATIGUE CHARACTERISTICS Fatigue curves for the alloys applied in lower wing skin, upper wing skin and fuselage are shown in Figs. 3 - 5. Tables I -III give the values of power m in the fatigue curve equation (σ0)m•N=Const These tables also give the values of fatigue live N133 showing the mean life values for the given alloy under maximum gross stresses σmax=133 MPa and cycle ratio factor R=0. The following could be concluded from the given test data: •

The life of modern Al-alloys is several times as much as that of the sheets made of outdated Al alloys.

B. Nesterenko, G.Nesterenko, V. Basov

664

•

The value of parameter m in fatigue equation for advanced alloys is bigger than one for outdated alloys. When plotted in the logarithmic coordinates the “σ - N” relation for the sheet specimens (wing and fuselage skins and high-strength alloys) is well correlated by one straight line. Relations “σ - N” for the plates made of advanced alloys (1163Т, 1163Т7) have some salient point at life value about 2•105 cycles and described by two lines.

Figs. 6 – 8 show the effect of loading cycle aspect ratio and stress values on specimen lives for different Al-alloys. These factors have significant effect on fatigue lives of different alloys. The heaviest fatigue damage of the material during one flight is generated while test by Truncated TWIST program. Boeing program results in the least severe fatigue damage, and thus the longest fatigue life could be obtained. TsAGI program gives the intermediate values of fatigue life. Test and analytical results have shown significant dependence of accumulated fatigue damages Σ(ni/Ni) on the alloy type, loading program and mean stresses σm in a cruise flight (Figs. 9 – 10, Table V). Fatigue life calculation in case of quasirandom loading programs had been carried out basing on Palmgren-Miner linear hypothesis of fatigue damage accumulation. Presented data show that Boeing program with cruise flight mean stress σm=85 MPa gives accumulated damage sum Σ(ni/Ni) > 1. These values are < 1 for Truncated TWIST and TsAGI programs. Experimental values of life standard deviation SlgN describing life scatter are shown in Fig. 11. For clad D16AT sheets this parameter does not exceed 0.1. Scatter is much higher for 1163Т7 and 2324-Т39 alloys.

RESIDUAL STRENGTH Critical values of stress intensity factors Kapp estimated during specimen tests without buckling elimination are presented in Tables I -III. The residual strength of advanced alloys sheets had been increased on 25-30% comparing to the sheets of 2024-Т3 (equivalent of D16АТV) alloys while the yield strength remained almost the same. The residual strength of Russian advanced Al - alloys, applied for lower wing skin had been increased on 45-75% comparing to D16ATV sheets, while the yield strength has changed slightly. Significant increase in the yield strength was reached in 2324-T39 plates as compared to 2024 - T351 plates with residual strength remained on the same level.

Fracture behaviour of skin materials in civil airplane structures

665

Simultaneous increase in residual strength (more than 40%), yield strength (20%), fatigue life (several times) has been reached in advanced high-strength Al-alloys 7055-T7751 and V96ts 3pchT12 (containing zirconium additives) as compared to the alloys like 7075-T6 (V95AT1).

CRACK GROWTH UNDER REGULAR LOADS Kinetic fracture diagrams representing the crack growth rate da/dN vs. stress intensity factor ΔK have been obtained for all the alloys mentioned. Tables 1-3 give the mean values of crack growth rates V31= da/dN at ΔK =31 MPa √m and R=0. Here a is half-length of central crack. The following difference in crack growth rates listed in Tables I-III can be observed when different alloys are compared: • The difference is about 1.5 times for the fuselage skin materials. Such a difference is verified by the crack growth data shown in Fig. 12. • The difference in growth rates is about 2 for the materials applied in lower wing skin. • Sufficient difference in crack growth rates is observed in the materials applied for upper wing surface (up to 5 times). Low crack growth rates are found in the materials with zirconium additives (Fig. 13). Crack growth in different alloys depends on the cycle ratio factor R (Figs. 14-15). Compressive loads have different effects on crack growth in different alloys.

CRACK GROWTH UNDER QUASI-RANDOM LOADING PROGRAMS The study of crack growth in lower and upper wing surfaces loaded by quasirandom spectra has been performed. Parameters of loading programs are shown in Fig.2 and Table 4. To compare specimen load levels the values of maximum stresses σequiv of equivalent puls cycle have been determined at mean cruise flight stress σm=85 MPa. The damage of the specimen (structure) at one equivalent cycle (R=0) is equal to damage accumulated during one mean statistical (in damage rate) flight. Damage rate of mean statistical flight was defined by dividing damage rate of the complete flight block by the number of flights in this block. To determine the value of σequiv, the aggregate full cycles obtained from flight block stress cyclogram were replaced by the aggregate equivalent at damage rate pulse cycles with maximum stresses being calculated as following:

B. Nesterenko, G.Nesterenko, V. Basov

666

σ 0i = 2σ aiσ max i

(1) th

where: σa i - stress amplitude; σmax i – maximum stresses of the i cycle in the flight block. Maximum stress in equivalent pulse cycle is:

σ экв = m

1 K

∑ (σ )

m

0i

i (2) where: K – number of flights in a flight block; m – exponent in fatigue equation , (it is assumed here m=4).

Summation is performed in all particular equivalent cycles of the flight block. Equivalent stresses σequiv for four loading programs at mean stresses σ m = 85 MPa are equal to: block program – 170MPa; TsAGI program – 170MPa; Truncated TWIST – 220MPa; Boeing program – 151MPa. Tests of the advanced alloys specimens under Truncated TWIST program has shown that crack growth duration in 1161T extruded panels significantly exceeds crack growth duration in 1163T7 and 2324-T39 plates (Fig. 16). The value of crack growth in the specimens from 1973T2 extruded panels tested under Truncated TWIST loading is much smaller than that in the specimens of the same alloy loaded by TsAGI and Boeing programs (Fig. 17). Crack growth duration in 1163T specimens under TsAGI loading program is much higher than that in the specimens of the same alloy under block loading program (Fig. 18). Crack growth durations under block loading program and under cyclic loading (σ max =175 MPa, R=0.02) are close to each other (Fig. 18). The accuracy of crack growth calculation in case of “random load spectrum” programs has been studied for 2324-T39 and 7055-T7751 alloys, which are applied in lower and upper wing skins, correspondingly. Similar Russian alloys are 1163T7 and V96ts-3pchT12 Crack growth rates in the alloys mentioned for different cycle ratios R are presented in Figs. 19-20. To describe the crack growth rate under regular loads Walker equation was taken in the form of : p da / dN = C ⋅ ( Z ⋅ K max ) , (3) where a is a crack half-length, N – number of cycles, K max – maximum stress intensity factor, and factor Z =(1-R)α for 0
Fracture behaviour of skin materials in civil airplane structures

667

Based on approximation of test data constants C, p, α of Walker equation have been found. Crack growth was studied under typical (standartized) quasi-random loading programs of Boeing company [10]. Fragments of load sequence are presented in Figs. 21-22. Comparison of analytical and experimental values of crack growth duration is shown in Figs. 23-24. Analysis of the results shows that crack growth duration calculated by the linear model would be conservative in case of wing lower skin. Calculations of crack growth in the wing upper skin by linear model are very unconservative. Widely used Willenborg model [11] gives adequate results of crack growth duration when stress spectrum mainly consists of tensile stresses and compressive stresses are not significant in magnitude (lower wing skin). For the loading that includes large number of high compressive stresses (upper wing surface) a modified Willenborg model is valid with strict limitations.

CONCLUSION Experimental study of strength, fatigue and crack resistance has been performed for Al-alloys applied in the wing and fuselage structures of the airplanes of different aircraft companies. Significant improvement of modern alloys is shown as compared to outdated alloys. This improvement is the result of decreasing percent content of silicium and iron; as well as additives of zirconium, lithium; alongside with improved alloy manufacturing technology. The effects of cycle ratio and stress magnitude of regular loading and type of standardized quasi-random loading program on fatigue life had been correlated. Numerical values of relative life sums calculated by Palmgren-Miner hypothesis had been obtained for different standardized quasi-random loading programs, mean stress in cruise flight for different Al-alloy semiproducts. Scatter factor of fatigue life n different alloys for different standardized quasi-random loading programs had been defined. Critical values of stress intensity factors are outlined for different alloys. Fatigue crack growth rates and durations under regular loads are defined for advanced alloys applied in the wing and fuselage skins. Crack growth durations in different allows are compared when loaded by different standardized quasi-random loading programs.

668

B. Nesterenko, G.Nesterenko, V. Basov

Kinetic diagrams of crack growth rates were determined for various cycle ratios in high-strength and high-durable plates applied in the skin of upper and lower wing surfaces, correspondingly. Experimental and analytical values of crack growth had been compared for the different standardized quasi-random loading programs of the upper and lower wing surfaces.

REFERENCE [1] Jaap Schijve. Fatigue of Structures and Materials. Second Edition with CD-ROM. Springer, 2008, 621 p. [2] A.Z. Vorob’ev, B.I. Olkin, V.N. Stebenev, T.S. Rodchenko. Fatigue Strength of Structural Elements. Moscow, Mashinostroyeniye, 1990, 199p. (in Russian). [3] S.I. Kishkina. Fracture Resistance of Al-alloys. Moscow, Metallurgiya, 1981, 279 p. (in Russian). [4] I.N. Fridlyander.Al-alloys in Flying Vehicles in the Periods of 1970-2000 and 2001-2015. Journal: Tehnologiya Legkih Splavov, N 4, 2002, VILS, p.12-17 (in Russian). [5] I.N. Fridlyander , V.V. Sadkov, V.S. Sandler, T.P. Fedorenko. Semiproduct Properties of High-Tech Al-Li-Alloy 1441. Journal: Tehnologiya Legkih Splavov, N 4, 2002, VILS, p.24-26 (in Russian). [6] United States Patent, 4,294,625. Oct. 13, 1981. [7] Airliner, April-June 1996, Boeing. Boeing Structural Design and Technology Improvements. [8] V.N. Basov, G.I. Nesterenko, V.Ye. Strizhius. Typified Loading Program for Heavy Transport. Troudy TsAGI, issue 2642, Moscow, 2001, pp.26-34 (in Russian). [9] J.B. de Jonge, D. Schütz, D. Lowak and J. Schijve. A Standardized Load Sequence for Flight Simulation Test on Trasport Aircraft Wing Structures. LBF Bericht FB-106, NLR 73029U, 1973. [10] R.F. Fowler and R.T. Watanabe. Development of Jet Transport Airframe Test Spectra. Boeing Commertial Airplanes, Seattle, Wahington, USA, May 1989, 16 p. [11] J.P. Gallagher, P.C. Miedlar, C.W. Cross, M.L. Papp. Cracks 93 System User’s Manual. University of Dayton Research Institute, UDR-TR-93-107. [12] V.N. Basov, G.I. Nesterenko. Fatigue Strength of Al-Alloys under Irregular Loading. Troudy TsAGI, issue 2669, Moscow, 2005, pp.119-125 (in Russian). [13] V.N. Basov, B.G. Nesterenko, G.I. Nesterenko. Fracture of High-Strength Al-Alloys. All-Russian Scientifica Journal: Polyet, 2008, “MashinostroyeniePolyet”, pp. 87-92. (in Russian). [14] V.N. Basov, B.G. Nesterenko, G.I. Nesterenko, V.G. Petrusenko. Experimental Study of Airplane Long Operation Effect on Degradation of Mechanic properties, fatigue strength and crack resistance of applied Al-alloys. Troudy TsAGI, issue 2669, Moscow, 2005, pp.126-142 (in Russian).

Fracture behaviour of skin materials in civil airplane structures

Semi product D16АТV sh. 4-6

σb σ02 δ5 Fe Si m MPa MPa % % % 457 343 19 0.38 0.20 3.9

669 Kapp N133 V31 cycle mm/kcycle MPa√m 95 000 4.0 100

2024-Т3 sh. 3-6 D16chТ extruded panel 1163Т p. 20

474

372 20 0.25 0.10 4.3 125 000

2.0

110

500

400 15 0.16 0.13 4.6 195 000

2.5

140

455

340 20 0.07 0.04 4.8 205 000

2.5

175

1163Т7 p. 30 1161Т extruded panel*

502 474

3.0 1.4

163 155

2024-Т351 p. 20

500

2.0

136

2024-Т351 p. 20

490

392 14 0.12 0.06 5.0 200 000 324 16 0.10 0.03 5.3 220 000 650 000 380 15 0.08 0.05 6.8 125 000 390 15 0.23 0.14 5.1 115 000

2.0

135

2324-Т39 p. 20 * Zr=0.11%;

494

458 12 0.08 0.04 5.9 275 000

2.5

148

Table I

Mechanical, fatigue strength and crack resistance parameters of Alalloys applied for transport lower wing skin

Semiproduct V95АТ1 sh. 3.5 7178-Т6 sh. 3.5 7075-Т6 sh. 1.7-3.5 V95pchТ2 extruded panel V95оchТ2 p. 30 7475-Т651 p. 25 В96ts-3pchТ12 p. 25* 7055-Т7751 p. 25* * Zr=0.11 - 0.12%

Table II

σb MPa 540 600 570

σ02 δ5 MPa % 504 10.5 569 9 520 8

Fe Si N133 Kapp V31 m % % cycle mm/kcycle MPa√m 0.5 0.5 3.2 47 000 15 52 0.20 0.06 3.8 40 000 15 55 0.22 0.13 4.3 80 000 7,5 77

540 480

12

0.15 0.06 4.0 125 000

3,0

135

514 555 635 618

9.6 10 10 6.7

0.12 0.12 0.12 0.13

3,25 2,75 5,0 3,5

175 135 70 90

455 505 595 594

0.07 0.10 0.03 0.12

4.7 7.4 5.9

170 000 105 000 320 000 300 000

Mechanical, fatigue strength and crack resistance parameters of highstrength Al-alloys applied for transport upper wing skin

σb MPa D16АТВ sh. 1.2-2 448 2024-Т3 sh. 2 470 1163АТВ sh. 1.5-2 431 1163РDТВ sh. 1.5-2 460 2524-Т3 sh. 1.6 429 2524-Т3 sh. 3-6* 448 1441РТ1 sh. 2.0** 426 * milled sheet 3.8 mm thick; Semiproduct

σ02 δ5 Fe MPa % % 321 20 0.38 345 21 0.19 315 24 0.12 340 23 0.14 326 20 0.09 346 19 0.09 337 13,4 0.05 ** Li – 1,9%;

Si % 0.20 0.17 0.05 0.03 0.05 0.04 0.03

m 3.5 3.5 3.45 4.0 3.35 4.1 4.05

N133 Kapp V31 cycle mm/kcycle MPa√m 80 000 3.0 100 85 000 3.0 110 100 000 2.0 120 115 000 2.0 129 100 000 2.0 132 168 000 1.7 85 000 2.7 100

Table III Mechanical, fatigue strength and crack resistance parameters of Alalloys applied for transport fuselage skin

B. Nesterenko, G.Nesterenko, V. Basov

670

Program

σmax/σm

σmin/σm

σpark/σm

nair cycle

Truncated TWIST

2.150

-0.500

-0.500

98

Flight number in flight block 4000

TsAGI

1.910

-0.875

-0.250

70

4000

Block

1.268

-0.571

-0.250

31

1

Boeing

2.100

-0.402

-0.300

24

5000

σmax and σmin – maximum and minimum stress in flight block; σm – mean stress of cruise flight; σpark – parking stress; Nair – number of cycles in the program simulating airplane flight in air.

Table IV Principal parameters of loading program

N, flight / Σ(ni / Ni) Material

Program title*

D16АТВ sh. 6.0 1163Т plate 1163Т7 plate D16chТ extruded panel (t=4 mm) 1161Т extruded panel 2024-Т3 sheet** 2024-Т351 plate 2024-Т351 plate 2324-Т39 plate * **

Trunc. TWIST 9 240 0.87 32 280 0.76 18 650 0.41

TsAGI 21 460 0.72 59 190 0.52 113 000 0.92

Block 18 220 0.63 71 880 0.70 105 150 0.95

Boeing 49 620 1.04 375 100 2.07 272 000 1.4

18 790 0.69

39 740 0.56

29 030 0.45

132 690 1.16

26 290 0.50 22 320 0.80 33 430 0.88 17 960 0.68 25 330 0.31

71 800 0.53

-

498 900 2.26

-

-

-

53 840 0.54

70 990 0.79

348 800 2.18

-

-

-

104 150 0.53

78 090 0.42

885 900 2.63

- mean stress of cruise flight σm was 85 MPa; - milled sheet about 4 mm thick.

Table V

Fatigue life and relative life sums during specimen testing by the programs typical for transport wing lower surface

Fracture behaviour of skin materials in civil airplane structures

a) Static strength

b) Fatigue

с) Crack resistance Fig. 1. Flat specimens for strength, fatigue and crack resistance tests

671

B. Nesterenko, G.Nesterenko, V. Basov

672

150

σ, MPa

120

σmax =133 MPa

σ, MPa

90 60

100

R =0; f =3 Hz 50

30 0 -30

0

150

-60

а)Pulse loading σ, MPa

b)Block

125 100 75 50 25 0 -25 -50

c)Truncated TWIST 160

σ, MPa

120 80 40 0 -40 -80

d)TsAGI 140 120 100 80 60 40 20 0 -20 -40

σ, MPa

е) Boeing (Boeing 767) Fig. 2. Typified loading programs for specimen tests

Fracture behaviour of skin materials in civil airplane structures

673

σmax gross, MPa -

220 200 180 160 140

2024-T351 plate; 2324-Т39 plate; 1163T7 plate; 1163Т plate; 1161Т extruded panel; D16chT extruded panel; D16chT extruded panel corrosion; 2024-Т3 sheet t=4; 2024-T351 plate; D16AT sheet t=4

120

R=0; f=2-3 Hz 100

80

2

4

6

8

10

5

2

4

6

8 10

6

2

4

N, cycle

Fig. 3. Fatigue curves for Al-alloy specimens applied for transport lower wing skin σmax gro ss, MPa 220

- V95AT1 sheet 3.5; - 7178-Т6 sheet 3.5; - V95ochТ2 plate; - 7075-Т6 sheet 3.5; - 7055-Т7751 plate; - V96ts-3pchТ12 plate

200 180 160 140 120 100

R=0; f=2-3 Hz 80

60 10 4 2

2

4

6

8 10 5

2

4

6

8 10

6

2

N, cycles

Fig. 4. Fatigue curves for Al-alloy specimens applied for transport upper wing skin

B. Nesterenko, G.Nesterenko, V. Basov

674

σmax g ross, MPа 200 180 160

R=0; f=2-3 Hz

140 120 100 - 2524-T3 sh. 4.0; - 1163RDTV sh. 1.5-2.0; - 1441RТ1 sh. 2.0; - 1163АТV sh.1.5-2.0; - 2524-T3 sh. 1.6; - 2024-T3 sh. 2.0; - D16АТV sh. 2.0

80

60 4

6

8

10

5

2

4

6

8

N, cycles

Fig. 5. Fatigue curves for Al-alloy specimens applied for transport fuselage skin

Fig. 6. Loading cycle asymmetry effect on different material lives

Fracture behaviour of skin materials in civil airplane structures

675

N , flights

867000

1 000 000

- D16AT sh. 6.0; - 1163Т7 p. 30;

σm= 85 MPa

- 2324-Т39 p. 20

272000 93300 104200

100 000

45900 25330 18650 10 000

21500

9240

1 000

Truncated TWIST

TsAGI

Boeing (B-767)

Fig. 7. Loading program effect on different material lives

Fig. 8. Stress effect on life relation in different materials

B. Nesterenko, G.Nesterenko, V. Basov

676 Σ(n ι/Ν ι) 3

- Tr. TWIST - D16AT; - Tr. TWIST - 1163Т7; - Tr. TWIST - 2324-Т39; - TsAGI - D16AT; - TsAGI - 1163T7; - TsAGI - 2324-T39; - Boeing (B-767) - D16AT; - Boeing (B-767) - 1163T7; - Boeing (B-767) - 2324-T39

2

1

0 60

70

80

90

100

σm , MPa

110

120

Fig. 9. Effect of mean stress in cruise flight (ny=1) on relative life sum for different materials

Σ(n i /N i) 3.0

- Truncated TWIST; - TsAGI; - Boeing (В-767)

2.5

2.0

2.57

σm= 85 MPа

1.52

1.5 1.05 1.0

0.87

0.76

0.72

0.53 0.40

0.5

0.31

0.0

D16AT

1163Т7

2324-Т39

Fig. 10. Effect of loading program on relative life sum in different materials

Fracture behaviour of skin materials in civil airplane structures

677

Fig. 11. Loading program effect on life logarithm scatter in different materials

2а , mm 100 - 1163AT sh. 1.5; - 1163АТ sh. 2.0; - 1163RDTV sh. 1.5; - 1163RDTV sh. 2.0; - 2524-Т3 sh. 1.6 (T-L); - 2524-Т3 sh. 4.0; - 1441RТ1 sh. 2.0

80

60

σmax = 133 MPa; R = 0.02

40

20

0 0

5000

10000

15000

20000

25000

30000

35000

40000

45000

N , cycle

Fig. 12. Comparison of fatigue crack growth duration in the sheets of different Alalloys applied in fuselage skin

B. Nesterenko, G.Nesterenko, V. Basov

678 da / dN , mm/kcycl e 40.0

R =0, f =1 Hz 20.0

V96ts-3pchT12 10.0

4.0

7055-Т7751 2.0 1.0

0.4

V95ochТ2 0.2 0.1 8

10

20

40

60

Fig. 13. Comparison of fatigue crack growth rates in the specimens of upper wing skin made of improved high-strength Al-alloys 2а , mm 100

σm ax = 66 MPa; R = -0,6 ; f = 5 Hz

80

- 7055-T7751 plate;

60

- 2324-T39 plat e 40

20

0 0

5000 10000 15000 20000 25000 30000 35000 40000 45000 50000 55000 60000

N , f li ghts

Fig. 14. Comparison of fatigue crack growth duration in 7055-T7751 and 2324T39 alloys under regular loading with negative value of cycle asymmetry factor

Fracture behaviour of skin materials in civil airplane structures

679

2а , мм 80

σmax = 119 MPa; R = 0.1; f = 5 Hz

60

- 7055-T7751 plate; - 2324-T39 plate

40

20

0 0

5000

10000

15000

20000

25000

30000

35000

40000

N , flight

Fig. 15. Comparison of fatigue crack growth duration in 7055-T7751 and 2324T39 alloys under regular loading with positive value of cycle asymmetry factor

120

2а , mm - 2324-T39 плита; - 1163Т7 плита; - 1161T прес. панель

100 80 60 40 20 0 0

2000

4000

6000

8000

10000

12000

N, flight

Fig. 16. Comparison of fatigue crack growth durations in different Al-alloys under loading by Truncated TWIST

B. Nesterenko, G.Nesterenko, V. Basov

680

2а , mm 100

σm = 85 MPa 80

- Truncated TWIST; - TsAGI; - Boeing

60

40

20

0 0

2000

4000

6000

8000

10000

12000

14000

N , flight

Fig. 17. Loading program effect on fatigue crack growth duration in 1973T2 extrude panel (zirconium added)

2а , mm 100

Block σm=85 MPa σeqw=170 MPa

80

R =0.02 σmax=175 MPa

TsAGI σm=85 MPa σeqw=170 МПа

60

40

20

0 0

2000

4000

6000

8000

10000

12000

14000

16000

18000

N , flight

Fig. 18. Loading program effects on fatigue crack growth duration in 1163T plate

Fracture behaviour of skin materials in civil airplane structures

10

-1

10

-2

681

da /dN , mm/cycle R = -0.2 σmax = 90 MPa R = 0.1 σmax = 120 MPa

-3

10

10

-4

10

-5

10

-6

R = -0.6 σmax = 65 MPa

10

20

30

40

50

ΔK , MPа·m1/2

60

70

80 90 100

Fig. 19. Fatigue crack growth rate in the specimens of lower wing surface made of high-endurance 2324-Т39 alloy

101-1

da /dN , mm/cycle

R = -0.2 σmax = 90 MPа 10-2 0.1 R = 0.1 σmax = 120 MPа 10-3 0.01

0.001 10-4

R = -0.6 σmax = 65 MPа

0.0001 10-5 10

20

30

40

50

60

70

80

90 100

ΔK , MPа·m1/2

Fig. 20. Fatigue crack growth rate in the specimens of upper wing surface made of high-strength 7055-Т7751 alloy

B. Nesterenko, G.Nesterenko, V. Basov

682 σ , MPa 120

Type "E" flight

Type "D" flight

Type "Е" flight

90 60 30 0 -30

Fig. 21. Sequence in typical flights (E and D) of Boeing program (B-767) as applied to lower wing elements

40

σ , MPa Type "E" flight

Type "D" flight

Type "Е" flight

0

-40

-80

-120

-160

-200

Fig. 22. Sequence in typical flights (E and D) of Boeing program (B-767) as applied to upper wing elements

Fracture behaviour of skin materials in civil airplane structures

80

683

2а , mm

60

Analysis by Willenborg mode using refined parameters of Walker equation l

50

Analysis by Willenborg model

70

40

Analysis by linear model

30 20

- experiment

10 0 0

5000

10000

15000

20000

25000

30000

35000

40000

N, flights

Fig. 23. Comparison of analytical and experimental values of fatigue crack growth duration in the specimens of lower wing skin made of high-endurance 2324-Т39 alloy under loading by Boeing program

90

2а , mm

80

- experiment

70 60

Analysis by linear model

50 40

Analysis by Willenborg model

30 20 10 0 0

100000

200000

300000

400000

500000

600000

700000

N , полет

Fig. 24. Comparison of analytical and experimental values of fatigue crack growth duration in the specimens of upper wing skin made of high-strength 7055-Т7751 alloy under loading by Boeing program

25th ICAF Symposium – Rotterdam, 27–29 May 2009

FATIGUE LIFE PREDICTION OF SMALL NOTCHED TI-6AL-4V BASED ON THE THEORY OF CRITICAL DISTANCE Yoichi Yamashita, Masaharu Shinozaki, Hiroshi Kuroki and Yusuke Ueda IHI Corporation

Abstract: This study investigated the estimating method for small notched fatigue strength of Ti-6Al-4V alloy using the theory of critical distance that employ the stress distribution in the vicinity of the notch. Circumferential notched round bar axial-loading fatigue tests were conducted to investigate the effect of notch radius and notch depth on fatigue strength. The fatigue tests show that the larger notch radius gives the increase of fatigue strength and the larger notch depth gives the decrease of fatigue strength. The theory of critical distance assumes that fatigue damage can be correctly estimated only if the entire stress field damaging the fatigue fracture process zone is taken into account. Critical distance stress is defined as the average stress within the fatigue fracture process zone where satisfies the area in the critical distance, a0, from the notch root. It has been found that there exists the good correlation between critical distance stress and crack initiation life of small notched specimen if the critical distance, a0, is determined with 107 cycle plain fatigue limit stress range, ∆σ 0 , and threshold stress intensity factor range, ∆K th using the stress intensity factor equation of semi-circle surface crack.

INTRODUCTION The high cycle fatigue (HCF) of titanium alloy turbine engine components remains a principal cause of failures in aircraft engines. HCF can result from vibration, M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 685–706. © Springer Science+Business Media B.V. 2009

686

Y. Yamashita, M. Shinozaki, H. Kuroki, Y. Ueda

forced response, unsteady aerodynamic loads, or other fluctuating loads. Gas turbine engine rotor blades and stator vanes are subject to all types of loads and are particularly vulnerable to HCF failures. Rotating blades in gas turbine engine are often subjected to impact from debris ingested into the engine during takeoff and landing which is called foreign object damage (FOD). Such debris may result in small notches at the leading or trailing edges of airfoils which in turn, reduces the fatigue strength of the material. Therefore, FOD-induced HCF is one of the significant themes in fatigue problems of aero-engine component [1-3]. Figure 1 shows the photographs leading edge notches by FOD and it is much concerned whether these defects may cause HCF or not. In these view points, notched fatigue problem is important to investigate the effects of FOD on fatigue strength of Ti6Al-4V alloy.

(a) fan blade (b) small FOD damage Fig. 1 Schematics of FOD damage at leading edge of airfoil. This study investigated the estimating method for notched fatigue strength of aeroengine fan blade and compressor blade of forged Ti-6Al-4V alloy. Fatigue predictions for notched specimens based on the maximum peak stress at the notch root are often inaccurate. It has been recognized that the simple use of the maximum peak stress at the notch root is often inadequate for characterizing the fatigue life of notched components. The stress state and stress distribution in the vicinity of the notch root are contributing factors to the fatigue process. The stress gradient at the surface of the notch root has been used in several early investigations, resulting in the fatigue predictions of Neuber [4], Peterson[5] that are utilized in current engineering design. Several more recent investigations have produced models that predict fatigue failure by employing the stress distributions in the region of the notch root [6-13]. The physical reasoning behind these investigations is that a critical volume of material must be subjected to a critical stress level for fatigue failure to occur at a given failure life. If one considers this volume to remain constant for notches of any size or shape within the same material, the observation that sharp notches have higher fatigue strengths than mild

Fatigue life prediction of small notched Ti-6Al-4V

687

notches for the same magnitude stress concentration factor is partial evidence for such a theory [4,5]. In this study, the method for fatigue life prediction of small notched Ti-6Al-4V specimens using the theory of the critical distance (TCD) [6,7] has been investigated. The TCD applied to fatigue problems assumes that fatigue damage depends on the stress field distribution in the vicinity of the stress concentrator. In other words, the TCD assumes that fatigue damage can correctly be estimated only if the entire stress field damaging the fatigue process zone is taken into account. Conventionally, critical distance stress is defined as the average stress in the material characteristic length as the distance from the notch-tip. And material characteristic length, a0, may be calculated with the threshold value of the stress intensity factor range, ∆K th , and the smooth specimen’s fatigue limit, ∆σ 0 . Taylor [6] and Susmel [7] were reviewed the most interesting findings in the use of the theory of critical distances to predict fatigue strength of notched mechanical components. They investigated the physical meaning of the theory and tried to verify its accuracy in predicting notch fatigue strength under different loading conditions and concluded that the theory is a powerful engineering tool suitable for assessing real mechanical components in situations of practical interests. And they reviewed that the theory of critical distance formalizations based on the use of linear-elastic stresses. They suggest that such theory can successfully be used to post-process simple linear-elastic finite element (FE) models reducing time and costs of the design process. Wang [8] used a critical distance calculated from the fatigue limit stress and crack growth threshold, ∆K th . The maximum principal stress at this distance from the notch root of a stress concentration was then compared to another notch geometry and loading condition and used to predict fatigue failure. Lanning [9] studied the critical distance at which the notched specimen stresses correspond to the stresses in a smooth bar that produces the same fatigue limit stress have resulted in modest success using Ti-6Al-4V alloy. The results of their investigation seem to indicate that a single parameter involving a critical distance may be a viable approach for predicting fatigue limit stresses in notched Ti-6Al-4V components. But they concluded that a size effect may exist with very small notches, in which case a single parameter may not be applicable. A lot of studies in the theory of critical distances have been conducted as mentioned above. But the unified recognition in determining the critical distance is not constructed so far in the practical engineering use especially for very small notches. In this study, the practical and appropriate approach in determining the critical distance and the critical distance stress has been investigated using simple

688

Y. Yamashita, M. Shinozaki, H. Kuroki, Y. Ueda

linear-elastic finite element (FE) results. The critical distance stress is defined as the average stress in the critical distance region. In the experimental study, small notched round bar axial-loading fatigue tests have been conducted with forged Ti-6Al-4V alloy to investigate the effect of notch radius and notch depth on fatigue strength of the material. Notch radius is varied from 0.05 to 0.2mm and notch depth is varied as 0.1, 0.3 and 0.5mm. And in the analytical study, the investigation into the appropriate determination method of the critical distance and the critical distance stress to predict crack initiation life of notched round bar specimen has been conducted using forged Ti6Al-4V alloy. This study shows that the appropriate method to evaluate fatigue life of Ti-6Al-4V material with various notch radius and notch depth can be constructed from the relationships between the critical distance stress and fatigue crack initiation life.

EXPERIMENT Materials and specimen Axial-loading fatigue tests were conducted to investigate the effects of small notches on the fatigue strength of Ti-6Al-4V alloy. Table I and Table II show the chemical composition and mechanical properties of Ti-6Al-4V alloy used, respectively. Figure 2 shows the bimodal microstructure. The material was a forged round bar. The material was subjected to the following heat-treatment, 960℃ for 1 hour, water-cooling, 705℃ for 2 hour and air-cooling. The result was an alpha-beta titanium alloy microstructure with acicular Widmanstätten structures. The plane area are equiaxial microstructures of α phase and the lamellar parts are β phase including acicular structures of α phase. Figure 3 shows the specimen size and geometry for axial loading fatigue tests. Before introducing notches, all cylindrical fatigue specimens were machined using a low-stress grinding technique from Ti-6Al-4V forged bar. The Ti-6Al-4V bar was forged from a parent bar produced in accordance with AMS 4928. Subsequently, hydrofluoric acid, nitric acid and hydrogen peroxide were used for etching. After the etching, small notches are introduced by machining. Specimens were 110mm in length with a grip section diameter of 15mm and a V-notch angle of 45°. A diagram of notch is shown in Fig. 3 (b), (c) and corresponding dimensions for the various notched specimens are provided in Table III. Three types of notch depth d=0.1mm, 0.3mm and 0.5mm with notch root radius, ρ = 0.05mm ,were designed to have different dimensions of elastic stress concentration factors, Kt = 3.92, 5.71, and 6.61 which were determined by ANSYS FE-analyses of elastic Ti-

Fatigue life prediction of small notched Ti-6Al-4V

689

6Al-4V material. And also, notched specimen with d=0.3mm and ρ = 0.2mm of Kt = 3.16 was used to investigate the effects of notch root radius on notched fatigue strength. It was verified that these stress concentration factors of FEA solutions have good agreement with the literature solutions [14,15]. Smooth un-notched specimens were machined with a gauge section diameter of 7.5mm and all other nominal dimensions same as those of circumferential notched specimens. Fatigue test method All the tests were carried out under constant stress amplitude using a MTS servohydraulic test system at a nominal frequency of 20Hz. Specimens were cycled under 0-tension fatigue loading of stress ratio, R = σ min / σ max = 0 where σ min is minimum stress and σ max is maximum stress. Generally, in an axial loading fatigue test, bending misalignment of the specimen can easily lead to underestimation of the fatigue limit. To avoid this difficulty, four strain gauges were applied to the smooth section near the gripping fixture and alignment was carefully adjusted before the fatigue testing.

Table I Chemical composition of Ti-6Al-4V material used (wt %). C 0.002

V 4.16

Al T i N 6.3 bal. 0.004

Fe 0.16

O 0.2

H 0.0053

Table II Mechanical properties of Ti-6Al-4V material used. σY0 σuts εf (MPa) (MPa) (%) 935 1006 18.4

ψ E (%) (GPa) 44.5 110

Notes; σ Y 0 : Yield stress, σ uts : Tensile strength, ε f : Elongation (Gauge length = 50mm), ψ : Reduction of area, E : Young’s modulus Table III Fatigue test conditions. Notch depth, d (mm) 0 0.1 0.3 0.5 0.3

Notch root radius, ρ (mm) ∞ 0.05 0.05 0.05 0.2

Note Smooth specimen Notch specimen Notch specimen Notch specimen Notch specimen

Y. Yamashita, M. Shinozaki, H. Kuroki, Y. Ueda

690

Φ 15

R60

Φ 7.5

Fig. 2 Microstructure of Ti-6Al-4V alloy used.

Small notch

10 110

(a) Specimen size and geometry

45°

ρ

d

(b) Schematics of circumferential notch (c) Definition of notch size and geometry Fig. 3 Shape and dimensions of the small notched round bar fatigue specimen. Results Figure 4 shows S-N data. Solid marks, ■ , denotes the stress range, ∆σ 0 , of smooth specimens. Open marks denotes the stress range, ∆σw , of the small notched specimens. It has been found that the fatigue tests shows that the larger notch radius gives the increase of fatigue strength and the larger notch depth gives the decrease of fatigue strength.

1200 1000

Stress range, Δ σ 0 or ⊿ σ

w

(MPa)

Fatigue life prediction of small notched Ti-6Al-4V

Smooth specimen

R =0

d=0.1mm, ρ =0.05mm

800

d=0.5mm, ρ =0.2mm

600

d=0.3mm, ρ =0.05mm

400

d=0.5mm, ρ =0.05mm

200

691

Runout

Ti-6Al-4V, R.T. Round bar tensile fatigue specimen

0 1.E+01

1.E+03

1.E+05

1.E+07

Number of cycles to failure, Nf

Fig. 4 Results of fatigue tests using round bar specimen.

THORY OF CRITICAL DISTANCE Theory The present study has been investigated the method to evaluate the efficiency of the theory of critical distance to determine fatigue strength for use with small notched Ti-6Al-4V components that is applicable to a broad range of geometries and loading conditions. The theory of critical distance applied to notched fatigue problems assumes that fatigue damage depends on the stress field distribution in the vicinity of the stress concentrator. In other words, it assumes that fatigue damage can correctly be estimated only if the entire stress field damaging the fatigue process zone is correctly taken into account [4-13]. The simplest approach conventionally determines the distance inward from the notch root to the location at which the stress range required for a fatigue failure life considered for the notched specimen is equal to the stress range required for the fatigue failure life for smooth specimens [4,5]. This simple approach is called the “point method (PM)”. In this case, the critical distance may be given by a0=LPM, where ∆σ yy = ∆σ 0

(1)

where a0 is the critical distance to be evaluated, LPM is the distance inward from the notch root in the radial direction normal to the loading where ∆σ yy = ∆σ 0 . ∆σ yy is the axial stress range at a0=LPM of the notched specimen as shown in Fig. 5(a). The reference value of ∆σ 0 was obtained from the smooth specimen test data. In some works [4,5], according to the point method, a notched component is in its fatigue limit condition when the range of the axial stress at a distance from the notch tip of LPM = L 2 equals the plain fatigue limit where L is defined by Eq.

692

Y. Yamashita, M. Shinozaki, H. Kuroki, Y. Ueda

(2). 107 cycles plain fatigue limit stress, ∆σ 0 , is 538MPa determined by smooth specimen fatigue test results and threshold value of stress intensity factor range is adopted as ∆K th = 3.86MPa m which is referenced for NASGRO material database [16]. Therefore, L equals 0.16mm for Ti-6Al-4V alloy used. L=

1 ∆K th 2 ( ) π ∆σ 0

(2)

The following approach is called the “line method (LM)” to determine the critical distance of the material. The line method approach conventionally calculates the average values of the quantities in question instead of point values. The condition to determine a0 over which the average stress range in the notched specimen is equal to the stress range in the smooth specimen may be given by a0=LLM where

1

LLM

∫ ∆σ yy ( x)dx = ∆σ 0 LLM 0

(3)

where ∆σ yy ( x) is the axial stress range distribution along x direction. In some works [4,5] the way the LM works, as expressed by Eq. (3), is summarized in Fig. 5(b), where LLM = 2L. Figure 6 provides the critical distances computed for the simple approach with Eq. (1) and Eq. (3) as well as the number of tests at each test condition. Figure 6 provides all critical distances computed for the approaches of the point method and line method using the stress distribution in the vicinity of the notch tip obtained by FE-analyses. Crack initiation life was determined by the round bar axial loading fatigue fracture life and calculation life of fatigue crack growth as mentioned below of Eq. (10). It has been found that the critical distances by the point method for small notched specimens with ρ = 0.05mm have the similar values of 0.02 to 0.035mm. The average critical distances of the line method are 0.08mm for small notched specimens with ρ = 0.2mm . The critical distances of the small notched specimens with ρ = 0.2mm are much larger than those with ρ = 0.05mm by both the point method and the line method. From this point of view, the approach in determination of the critical distances as the distance inward from the notch root to the location at which the stress range required for a fatigue failure life considered for the notched specimen is equal to the stress range required for the fatigue failure life for smooth specimens may not give an appropriate unified value, if the notch tip radius of the notched specimens is different from each other. And also, the critical distances of the point method are much larger than the conventional determined method using LPM = L 2 = 0.008mm . The critical distances of the line method are also much larger than the conventional determined method using LLM = 2 L = 0.032mm . From these results, LPM = L / 2 of the

Fatigue life prediction of small notched Ti-6Al-4V

693

point method and LLM = 2 L of the line method in the conventional approach [4, 5] of determination of the critical distances are much smaller than the critical distances as the distance inward from the notch root to the location at which the stress range required for a fatigue failure life is equal to the stress range required for the same fatigue failure life for smooth specimens. ⊿ σ

yy

⊿ σ

⊿ σ

Linear-Elastic Stress Distribution

⊿ σ

0

Notch

1

LLM

LLM

0

∫ ∆σ yy ( x )dx = ∆σ 0

0

Notch

x

LPM

Linear-Elastic Stress Distribution

yy

x

LLM

(a) Point method

(b) Line method

0.12 0.1

Ti- 6Al- 4V, R.T. Small notched specimen

0.08 0.06

d=0.1, ρ =0.05 d=0.3, ρ =0.05 d=0.5, ρ =0.05 d=0.3, ρ =0.2 L/ 2=0.008mm

0.04 0.02 0 1.E+04

1.E+05 Crack initiation life, Ni (cycles)

1.E+06

Critical Distance, L LM (mm)

Critical Distance, L PM (mm)

Fig. 5 Concept of the conventional theory of critical distance [4, 5]. 0.4 0.3

Ti- 6Al- 4V, R.T. Small notched specimen

0.2

d=0.1, ρ =0.05 d=0.3, ρ =0.05 d=0.5, ρ =0.05 d=0.3, ρ =0.2 2L=0.032mm

0.1 0 1.E+04

(a) Point method

1.E+05 Crack initiation life, Ni (cycles)

1.E+06

(b) Line method

Fig. 6 Critical distances based on the conventional point method and line method based on Eq. (1) and Eq. (3), respectively.

Investigation into definition of appropriate critical distance In this study, the critical distance is calculated from the plain fatigue limit stress under R=0, ∆σ 0 , and threshold value of stress intensity factor range, ∆K th . In this view point, critical distances, a0, are studied by two equations to investigate the appropriate calculating method for the critical distance in Ti-6Al-4V alloy used. One is the type based on the stress intensity factor range equation for through crack in infinite plate as Eq. (4).

∆K = ∆σ πa

(4)

Y. Yamashita, M. Shinozaki, H. Kuroki, Y. Ueda

694

And another equation is the type based on the stress intensity factor range equation for semi-circle surface crack in semi-infinite plate as Eq. (5)

∆K = 1.1215 ⋅ (2 / π ) ⋅ ∆σ πa

(5)

Therefore, the critical distances, a0 , are defined by Eq. (6) and Eq. (7) based on Eq. (4) and Eq. (5), respectively using 107 cycles fatigue limit stress, ∆σ 0 . a0 = (1 / π ) ⋅ (∆Kth / ∆σ 0 ) 2

(6)

a0 = (1 / π ) ⋅ (π ⋅ ∆Kth /(1.1215 × 2 × ∆σ 0 )) 2

(7)

It has been found that the critical distance of the line method determined based on Eq. (3) by adjusting to the stress range in the smooth specimen at the same level of fatigue failure life is larger than that determined by Eq. (7) of a0=0.032mm as shown in Fig. 6 (b). But the critical distance of the line method by a0=LLM=2L is almost the same as that determined by Eq. (7). The critical distance stress, σ CD , is defined as Eq. (8) which takes into account of the volume effects on fatigue strength due to circumferential notch of the round bar tensile specimen. In Eq. (8), critical distance stress is given by the integrated value, a0 ∫0 σ yy ( x ) ⋅ 2πx ⋅ dx , of highly critical axial stress in fatigue fracture process zone

of circumferential notch tip region divided by critical distance area, ∫0a 0 2πx ⋅ dx , of notch tip region. Sheppard [6] and Taylor [7] have observed that the range of the effective stress could also be calculated by averaging the range of the maximum principal stress over a semi circular area, formalising the so-called area method (AM). For the notched round-bar tensile specimen adopted in this study, the highly stress region in the vicinity of the notch tip has the volume effect due to having the circumferential notch as shown in Fig. 7. The critical distance stress, σ CD , can be calculated using the axial stress distribution at maximum fatigue loading in Eq. (8). a0

a0

0

0

σ CD = ∫ σ yy ( x) ⋅ 2π ⋅ ( R0 − x) ⋅ dx

∫ 2π ⋅ ( R0 − x) ⋅ dx

(8)

Fatigue life prediction of small notched Ti-6Al-4V

y

σ peak,max

d

695

y σ

σ yy ρ

o

yy

R0 x

x

a0

a0

Fig. 7 Schematics of critical distance concepts taking into account of the volume effects due to the circumferential notch.

FATIGUE LIFE PREDICTION Determination of fatigue crack initiation life The critical distance stress may be good correlation with fatigue initiation life. In the experiment, the fatigue failure life was measured. Generally, it is considered that fatigue failure life, N f , is the sum of fatigue initiation life, N i , and fatigue crack growth life, N P , as Eq. (9). The fatigue crack initiation life of the notched specimen can be determined by the round bar axial loading fatigue test data and calculation life of fatigue crack growth. Fatigue crack growth life is calculated based on the stress intensity factor range and Paris’ type crack growth property of Ti-6Al-4V material. N f = Ni + Np

(9)

Fatigue initiation life is determined by Eq. (10). N i = Nf − Np

(10)

Fatigue crack growth life is calculated based on Paris’ law of Eq. (11) and initial crack size of a0 as Eq. (7). Paris’ law constants were determined by making reference to NASGRO material database [16]. In the calculation of fatigue crack growth life, the stress concentrations of the specimens in the vicinity of the notch root were taking into account in determination of stress intensity factor range, ∆K . da dN = C (∆K ) m , C = 2.82 × 10 −11 , m = 3.2 unit in MPa, m

(11)

Y. Yamashita, M. Shinozaki, H. Kuroki, Y. Ueda

696

Maximum stress approach Fatigue predictions for notched specimens based on the maximum stress at the root of the notch, frequently termed maximum peak stress, are often inaccurate. Figure 8 shows the relationship between maximum peak stress, σ peak, max , at the notch tip and fatigue initiation life. The results of the notch root radius of ρ =0.05mm and notch depth of d=0.1, 0.3 and 0.5mm have a coincidence each other but the results of the notch root of the notch depth of 0.3mm with notch radius of ρ =0.2mm is different from those of the radius of 0.05mm. It can be seen that fatigue life prediction based on maximum peak stress approach gives inaccurate results for the notch fatigue life of various notch root radiuses.

Maximum peak stress at notch tip, σ peak,max (MPa)

Figure 9 (a) and (b) show the axial stress distribution in the vicinity of the notch tip region at fatigue failure lives of Nf=3.0×104 cycles and Nf=1.0×105, respectively. These stress distributions are calculated by FE-analyses using axisymmetric solid elements in ANSYS FE-code. There is good coincidence for the axial stress distributions of the notched specimens of d=0.1, 0.3 and 0.5mm with ρ =0.05mm. But, the axial stress distribution of the specimen of d=0.3mm with ρ =0.2mm is evidently different from that of ρ =0.05mm. From these results, it has been found that the stress distribution in the vicinity of the notch root are contributing factors to the fatigue process. The stress gradient in the vicinity of the notch root has to be taken into account in fatigue life prediction for notched specimens. Therefore, by employing the stress distributions in the region of the notch root, critical distance stress, σ CD based on Eq. (8), is adopted as the efficient parameter for predicting notched fatigue initiation life.

3000 2500

Ti-6Al-4V, R.T.

2000 1500 1000 500 0 1.E+03

d=0.1mm, ρ=0.05mm d=0.3mm, ρ=0.05mm d=0.5mm, ρ=0.05mm d=0.3mm, ρ=0.2mm 1.E+04 1.E+05 1.E+06 Crack initiation life, Ni (cycles)

1.E+07

Fig. 8 S-N curve (Vertical axis : maximum peak stress at notch root).

Fatigue life prediction of small notched Ti-6Al-4V d=0.1mm, ρ =0.05mm d=0.3mm, ρ =0.05mm d=0.5mm, ρ =0.05mm

3000

Axial stress, σ yy(x) (MPa)

697

2500 2000

d=0.3mm, ρ =0.2mm

1500 1000

Stress distribution at maximum loading 4 at Nf=3× 10 cycles Ti- 6Al- 4V, R.T.

500 0

0

0.02 0.04 0.06 0.08 Distance from notch root, x (mm)

0.1

(a) Axial stress distribution at fatigue failure of Nf=3.0×104 cycles d=0.1mm, ρ =0.05mm d=0.3mm, ρ =0.05mm d=0.5mm, ρ =0.05mm

Axial stress, σ yy(x) (MPa)

2500 2000

d=0.3mm, ρ =0.2mm

1500 1000 500 0

Stress distribution at maximum loading 5 of Nf=10 cycles Ti- 6Al- 4V, R.T. 0

0.02 0.04 0.06 0.08 Distance from notch root, x (mm)

0.1

(b) Axial stress distribution at fatigue failure of Nf=1.0×105 cycles Fig. 9 Axial stress distributions in the vicinity of notch root at fatigue failure of Nf=3×104 cycles and Nf=1×105 cycles. Sensitivity of critical distance Figure 10(a)-(c) show the relationships between critical distance stress of Eq. (8) and crack initiation life using the critical distances of a0 = 0.016mm, a0 = 0.032mm and a0 = 0.064mm, respectively. The crack initiation life prediction of a0 = 0.032mm gives the best correlation for all fatigue test data. In this view point, a0 = 0.032mm calculated based on Eq. (7) is adopted for calculating the critical distance stress of Eq. (8). This critical distance is almost the same as that determined by a0=LLM=2L based on Eq. (3). And results of Fig. 10 can be understood by the axial stress distribution of the notch tip region for each fatigue test specimens as shown in Fig. 9. Critical distance stress is defined as the average stress of Eq. (8) in the critical distance area

Y. Yamashita, M. Shinozaki, H. Kuroki, Y. Ueda

Critical distance stress, σ (MPa)

CD

698

d=0.1mm, ρ=0.05mm d=0.3mm, ρ=0.05mm d=0.5mm, ρ=0.05mm d=0.3mm, ρ=0.2mm

3000 2500 2000 1500 1000 500

Ti-6Al-4V, R.T. Round bar tensile fatigue specimen

0 1.E+03

1.E+04

1.E+05

1.E+06

1.E+07

Crack initiation life, Ni (cycles)

Critical distance stress, σ (MPa)

CD

(a) Critical distance a0= 0.016 mm ( a0 = (1 / π )(∆Kth / ∆σ W0 ) 2 ) d=0.1mm, ρ=0.05mm d=0.3mm, ρ=0.05mm d=0.5mm, ρ=0.05mm d=0.3mm, ρ=0.2mm

3000 2500 2000 1500 1000 500

Ti-6Al-4V, R.T. Round bar tensile fatigue specimen

0 1.E+03

1.E+04

1.E+05

1.E+06

1.E+07

Crack initiation life, Ni (cycles)

Critical distance stress, σ (MPa)

CD

(b) Critical distance a0= 0.032 mm ( a0 = (1 / π )((π∆Kth) /(2 × 1.1215 × ∆σ W0 )) 2 ) d=0.1mm, ρ=0.05mm d=0.3mm, ρ=0.05mm d=0.5mm, ρ=0.05mm d=0.3mm, ρ=0.2mm

3000 2500 2000 1500 1000 500

Ti-6Al-4V, R.T. Round bar tensile fatigue specimen

0 1.E+03

1.E+04

1.E+05

1.E+06

1.E+07

Crack initiation life, Ni (cycles)

(c) Critical distance a0= 0.064mm = 0.032×2 mm Fig. 10 Relationships between critical distance stress and crack initiation life for various critical distances.

Fatigue life prediction of small notched Ti-6Al-4V

699

Critical distance stress, σ (MPa)

CD

600

Ti-6Al-4V, R.T.

d=0.1mm d=0.3mm d=0.5mm

500 400 300 200 100

Nominal stress in notch section, σ n = 100 MPa

0 0.001

0.01

0.1

1

Notch root radius, ρ (mm)

Fig. 11 Notch root radius dependencies on critical distance stresses. 10 fatigue strength, ⊿ σ w (MPa)

800

ρ =0.05mm

700 600 500

d=0.1mm

400

d=0.3mm

300

7

200

Ti- 6Al- 4V, R.T. Round bar tensile specimen

100 0

0

d=0.5mm

10 20 30 40 Stress concentration factor, Kt

50

Fig. 12 Effects of stress concentration factors on fatigue limit stress at 107 cycles fatigue life. as the fatigue fracture process zone. It can be concluded that there exists the good correlation between critical distance stress of Eq. (8) and crack initiation life of small notched specimen if the critical distance, a0, is determined with 107 cycles plain fatigue limit stress range, ∆σ 0 , and threshold stress intensity factor range, ∆K th , using the stress intensity factor equation of semi-circle surface crack. Effects of notch root radius Figure 11 shows the relationships between critical distance stresses, σ CD , and notch root radius, ρ , of each notch depth, such that d=0.1, 0.3 and 0.5mm, respectively, under the nominal stress in notch section of σ n =100 MPa. In Fig. 11, critical distance stress is calculated by the axial stress distribution obtained by FEA based on Eq. (8). It has been found that the very small notch root radius in the range of ρ = 0.001～0.01mm gives the constant critical distance stress value. From these results, it can be seen that the minimum notch fatigue strength exists for each notch depth of d=0.1, 0.3 and 0.5mm. Figure 12 shows the relationships between the 107 notched fatigue strength and stress concentration factor for each notch depth of d = 0.1, 0.3 and 0.5mm,

Y. Yamashita, M. Shinozaki, H. Kuroki, Y. Ueda

700

respectively. 107 cycles notched fatigue strength was determined from the relationship between the critical distance stress and crack initiation life as shown in Fig. 10(b). And stress concentration factor was determined by FEA. In Fig. 12, it can be also seen that the minimum notch fatigue strength exists for each notch depth. It has been found that the minimum notch fatigue strength is determined for each notch depth of d=0.1, 0.3 and 0.5mm from the critical distance stress approach. Fatigue strength reduction curve From the investigation into the critical distance stress and crack initiation life, it has been found that reduction level of notched fatigue strength against smooth specimen can be determined for various notch depth under a fatigue failure life. The mark of ◆ in Fig. 13 denotes the non-dimensional fatigue strength with ρ =0.05mm for various notch depths of d=0.1, 0.3, and 0.5mm at 107 cycles fatigue failure life. Non-dimensional fatigue strength is determined by ∆σ w ∆σ 0 , where ∆σ w is the notch fatigue strength at 107 cycles fatigue failure life and ∆σ 0 is the plain fatigue strength at the same fatigue life.

7

Non- dimensional 10 fatigue strength, ⊿ σ w/ ⊿ σ 0

Due to the results in Fig. 11 and Fig. 12 that there exists minimum fatigue strength in a notch depth under a fatigue failure life, the minimum fatigue reduction curve of non-dimensional fatigue strength can be determined for various notch depths with ρ = 0.001 mm as mark ■ shown in Fig. 13. 1.2

ρ =0.05mm

1 0.8 0.6 0.4

Minimum notch fatigue strength (ρ =0.001mm)

0.2

Ti- 6Al- 4V, R.T.

0

0

0.1

0.2

0.3

0.4

Notch depth, d (mm)

0.5

0.6

Fig. 13 Non-dimensional fatigue strength for various notch depths with ρ =0.05mm and minimum non-dimensional fatigue strength for various notch depths with ρ =0.001mm.

DISCUSSION SEM observation Figure 14 (a)-(c) show the macro photographs of fracture surface of round-bar circumferential notched specimen and the schematics of fatigue crack initiation origins and fatigue crack growth paths. Figure 14(a) is corresponding to 104 cycles

Fatigue life prediction of small notched Ti-6Al-4V

701

low cycle fatigue (LCF) range as shown with group mark ○ in Fig. 15. Figure 14(b) is corresponding to medium LCF range as shown with group mark △ in Fig. 15. Figure 14(c) is corresponding to HCF range as shown with group mark □ in Fig. 15 . In 104 cycles LCF range, multi fatigue crack initiation origins can be seen at circumferential notch root region. From these sites, multi site fatigue crack growth and coalescence may be occurred. In medium LCF range, multi fatigue crack initiation origins can also be seen at circumferential notch root region and multi site fatigue crack growth and coalescence may be occurred. But the number of origins of fatigue crack initiation was reduced in comparison with the case of 104 cycles LCF range. In HCF range, only one origin of fatigue crack initiation can be seen and from this origin, fatigue crack growth was occurred. From these behaviors, the volume effects due to circumferential length of notch should be taken into account in calculating critical distance stress. Therefore, the critical distance stress equation based on Eq. (8) may be appropriate because of takeing into account of the volume effects due to the circumferential length of notch and stress gradients in radial direction of the specimens. Figure 16 (a) and (b) show the SEM observation of smooth specimens at fatigue failure of Nf= 5.67×104 cycles and 3.86×104 cycles, respectively. At the origin of fatigue crack initiation site, crystallographic facets can be seen. There is not any inclusion observed by SEM in the vicinity of fatigue crack initiation site. The size of the crystallographic facet is 0.03mm for smooth specimen of Nf=5.67×104. The size of the crystallographic facet is 0.035mm for smooth specimen of Nf=3.86×104. The shape of the crystallographic facets may be approximately semi-circle. From these results of the detailed SEM observation, the determination procedure of the critical distance (Eq. (7)), calculated from the 107 cycle fatigue limit stress, ∆σ 0 , and the equation (Eq. (5)) of stress intensity factor range, ∆K , of semi-circle crack, may be appropriate for the forged Ti-6Al-4V alloy. Some literatures [17-19] concluded that these observed crystallographic facets were corresponding to the α phase grain size. And the size of facet are almost the same as the average size of one α phase grain size which is recognized by the microstructure as shown in Fig. 2. From SEM observation of fatigue fracture surface, critical distance stress calculation of Eq. (8) may be appropriate in terms of taking into account of the volume effects and critical distance determined by Eq. (7) which is corresponding to the facet size of the crack initiation site.

Y. Yamashita, M. Shinozaki, H. Kuroki, Y. Ueda

702

(a) 104 LCF range (b) medium LCF range (c) HCF range Fig. 14 Macro photographs of fatigue fracture surfaces. d=0.1mm, ρ=0.05mm

Stress range, Δ σ (MPa)

1200

Group : very low cycle fatigue Group : medium low cycle fatigue ＊＊＊＊＊

1000 800

d=0.3mm, ρ=0.05mm d=0.5mm, ρ=0.05mm Smooth specimen

600

Runout

400 200

＊

0 1.E+03

Group : high cycle fatigue Group : multi sub-crack of smooth specimen

1.E+04 1.E+05 1.E+06 1.E+07 Number of cycles to failure, Nf

1.E+08

Fig. 15 Group divisions of fracture surface aspects on S-N curves. Figure 17 shows the macro photograph of the side surface of a fatigue fractured smooth specimen. The multi-sub-cracks can be observed. These multi-sub-cracks were observed in the fatigue fracture specimens which are shown in Fig. 15 with the mark *. This may be the evidence that in the smooth specimen which has the gauge length of 10mm, the volume effects of fatigue fracture process zone extend throughout the gauge length. On the other hand, the fatigue fracture process zone of the circumferential notch fatigue specimen is only limited in the vicinity of the notch tip region. This can be the difference of the volume effects between the round-bar smooth specimen and the circumferential notched round-bar specimen. From this experimental fact, it can be recognized that the approach in determination of the critical distances as the distance inward from the notch root to the location at which the stress range required for a fatigue failure life considered for the notched specimen is equal to the stress range required for the fatigue failure

Fatigue life prediction of small notched Ti-6Al-4V

703

life for smooth specimens may not give a appropriate unified value as shown in Fig. 6. The smooth specimen of round bar specimen can have the much larger volume effects than the circumferential notched specimen.

0.035mm

(a) Smooth specimen of Nf=5.67×104 0.03mm

(b) Smooth specimen of Nf=3.86×104 Fig. 16 SEM photographs of smooth specimen’s fatigue fracture surface.

Fig. 17 Multi sub-crack in side surface of smooth specimen. Estimation of residual stress effects A further study may be needed to predict FOD-induced HCF based on critical distance approach. Residual stress introduced by FOD can have significant effects on FOD-induced HCF life prediction. Fatigue strength design curve for FOD-

Y. Yamashita, M. Shinozaki, H. Kuroki, Y. Ueda

704

induced HCF life may be proposed by taking into account of the residual stress effects on minimum notch fatigue strength determined as Fig. 13. The schematics of the reduction curve concept of FOD induced notch fatigue strength is shown in Fig. 18 with broken line. Stress ratio ( = σmin / σmax ) effect on notch fatigue strength and residual stress level introduced by FOD will be needed to taking into account of residual stress effects on FOD-induced HCF.

7

Non- dimensional 10 fatigue strength, σ w/ σ 0

ρ =0.05mm minimum notch fatigue strength (ρ =0.001mm) 1.2

Effects of residual stress introduced by FOD ?

1 0.8 0.6 0.4

?

0.2 0

0

0.1

0.2 0.3 0.4 Notch depth, d (mm)

0.5

0.6

Fig. 18 Schematics of the notch fatigue strength reduction concept by residual stress effects which are introduced by FOD.

CONCLUSIONS This study investigated the estimating method for small notched fatigue strength of Ti-6Al-4V alloy using the theory of critical distance that employ the stress distribution in the vicinity of the notch. The following conclusions can be made based on this study. (1) The approach in determination of the critical distance as the distance inward from the notch root to the location at which the stress range required for a fatigue failure life is equal to the stress range required for the same fatigue failure life for smooth specimens may not give an appropriate unified value if the notch tip radiuses of the notched specimens are different from each other. This is due to that the smooth specimen of round bar specimen can have the much larger volume effects than the circumferential notched specimen. (2) Critical distance stress is defined as the average stress of Eq. (8) in the critical distance area of high stresses region in the vicinity of the circumferential notch tip as the fatigue fracture process zone. It has been found that there exists the good correlation between critical distance stress of Eq. (8) and crack initiation life of small notched specimen if the critical distance, a0, is determined with 107 cycles plain fatigue limit stress range, ∆σ 0 , and threshold stress intensity

Fatigue life prediction of small notched Ti-6Al-4V

705

factor range, ∆K th , using the stress intensity factor equation of semi-circle surface crack. (3) From SEM observation of fatigue fracture surface, critical distance stress calculation of Eq. (8) may be appropriate in terms of taking into account of the volume effects and critical distance determined by Eq. (7) which is corresponding to the facet size of the crack initiation site.

ACKNOWLEDGEMENTS This study is conducted under the contract with New Energy and Industrial Technology Development Organization (NEDO) as a part of the civil aircraft basic technology program of Ministry of Economy, Trade and Industry (METI).

REFERENCES [1] [2]

S. R. Thompson, J. J. Ruschau, T. Nicholas, Int. J. of Fatigue 23 (2001), S405-S412. J. Ruschau, S. R. Thompson, T. Nicholas, Int. J. of Fatigue 25, (2003), 955962. D. Nowell, P. Duó, I.F. Stewart, Int. J. of Fatigue 25, (2003), 963-969. H. Neuber, Theory of notch stresses 2nd ed. Berlin: Springer Verlag; 1958. R.E. Peterson, Notch sensitivity. Metal fatigue. McGraw Hill; 1959. 293-

[3] [4] [5] 306. [6] L. Susmel, Engng Fract. Mech. 75 (2008), 1706-1724. [7] D. Taylor, Engng Fract. Mech. 75 (2008), 1696-1705. [8] D. Taylor, G. Wang, In : J.C. Newman, R.S. Piascik, editors. Fatigue crack thresholds, endurance limits, and design. ASTM STP 1372, 1999. [9] S.D. Sheppard, Trans ASME J. Mech. Design 1991; 113:188-194. [10] D. Taylor, Int. J of fatigue 1999;21:413-420. [11] D. B. Lanning, T. Nicholas, G. K. Haritos, Int. J. of Fatigue 27 (2005) 4557. [12] D. B. Lanning, T. Nicholas, A. Palazotto, Int. J. of Fatigue 25 (2003), 835841. [13] K. Tanaka, Int. J. of Fracture 1983;22:R39-45. [14] H. Nishitani, N. Noda, Engng Fract. Mech. 1984, 20(5/6), 743-766. [15] N-A Noda, M. Sera, Y. Takase, Int. J. of Fatigue 1995, 17(3), 163-178. [16] NASA, FATIGUE CRACK GROWTH COMPUTER PROGRAM “NASGRO” VERSION 3.0 Reference Manual, National Aeronautics and Space Administration, Lyndon B. Johnson Space Center, 2000. [17] S. Nozue, T. Okubo, J. Japan Inst. Metals, Vol. 52 No. 11, 1988, 1057-1062. (in Japanese)

706

[18] [19]

Y. Yamashita, M. Shinozaki, H. Kuroki, Y. Ueda

H. Oguma, T. Nakamura, S. Yokoyama, T. Noguchi, J. Soc. Mat. Sci., Japan, Vol. 52 No. 11, 2003, 1298-1304. (in Japanese) O. Jin, R.W. Hamm, W.S. Johnson, Fatigue Fract. Engng Mater. Struct. 25, 2002, 563-574.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

A CASE STUDY OF MULTIPLE-SITE FATIGUE CRACK GROWTH IN THE F-18 HORNET BULKHEAD B. Andersson1, A. F. Blom1, U. Falk1, and G. S. Wang1 K. Koski2, A. Siljander2, J. Linna3, A. Miettinen3 , K. Vaaraniemi4, R. Lahtinen4 1

FOI (Swedish Defence Research Institute), Sweden VTT (Technical Research Centre of Finland), Finland 3 Patria Aviation Oy, Finland 4 FINAF (Finnish Air Force), Finland

2

Abstract: The research goal of the present study was to develop and validate a tool for reliable analysis of fatigue crack growth following initiation from several small surface flaws that grow and coalesce when subjected to service loading. Both analytical and experimental methods were applied in a case study of an isolated detail of an F-18 Hornet bulkhead structure selected for this purpose. The experimental part consisted of integral I-type beams, designed to closely simulate a critical part of an F-18 Hornet fuselage bulkhead, which were subjected to actual inflight bending and tension loads in the critical section. The variable amplitude spectrum used in the tests was created based on the operational load measurement results of the F-18 Hornet of the Finnish Air Force.

INTRODUCTION The usage, i.e. the structural life consumption, of the F-18 Hornet fleet of the Finnish Air Force (FINAF) is more severe than the original design target. Hence, there is a need to further develop and apply a systematic methodology to cope with structural degradation effects in the fleet. An overview of the on-going structural integrity related research efforts is highlighted in the Finnish ICAF National Review [1]. One aspect of the structural integrity efforts is related to early crack M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 707–741. © Springer Science+Business Media B.V. 2009

B. Andersson et al

708

initiation due to the micro-notches (etch pits) on the surface of metallic aircraft parts. The corrosion protection of the aluminium structures of the F-18 Hornets has been improved by the OEM during manufacturing of the aircraft by introducing an Ion Vapour Deposition (IVD) coating. This process involves surface cleaning by etching (pre-IVD etch) prior to the IVD coating process. The etching, which is required for surface cleaning prior to the IVD, has the drawback of inducing severe micro-notches (pits) at the grain boundaries. These surface flaws (pits) adversely affect the fatigue strength of the structure [2]. To quantify the physical phenomena of multiple crack initiation, coalescence and growth characteristics of fatigue cracks emanating from the etch pits, FOI (the Swedish Defence Research Agency), Patria Aviation Oy and VTT teamed up to develop and validate a tool for reliable fatigue analysis. Both analytical and experimental methods were applied in the present study of a representative FINAF F-18 Hornet structural detail subjected to in-flight service loading.

EXPERIMENTAL PROGRAMME I-beams simulating the F-18 Hornet bulkhead For the present study, both fatigue testing and numerical modelling are performed on integral I-type beams of a geometry simulating a critical part of an F-18 Hornet fuselage bulkhead subjected to actual in-flight bending and tension loads in the critical section, see Figure 1. The fatigue tests were part of a test series in which a strengthening method (proactive repair patching) of a critical part of the bulkhead was studied [3]. The I-beams were designed and machined from 7050-T7451 aluminium plate by Patria Aviation. The mechanical properties of the 7050-T7451 aluminium are provided in Table I. Table I. Physical and mechanical properties of aluminium 7050-T7451. α

=

2,3×10-5

1/ ˚C

Ec Et Emean ν Sy Su

= = = = = =

73084 71016 72050 0,33 455 531

MPa MPa MPa

MPa MPa

Thermal expansion coefficient (20 ˚C → 100 ˚C) Young’s modulus (compression) Young’s modulus (tension) Young’s modulus (average) Poisson’s ratio Yield strength (L-T direction) Ultimate strength (L-T direction)

A case study of multiple-site FCG in the F-18 Hornet bulkhead

709

Six of the I-beams were chemically etched prior to the fatigue tests. The etching was conducted in a manner similar to the pre-IVD etching process of the original bulkheads by the OEM. In addition to the six etched specimens, there were two unetched specimens for other purposes, e.g. calibration and set-up of the test rig and associated data acquisition systems as well as for the residual strength evaluation.

Figure 1. – Geometry of studied test object. Picture courtesy of Patria Aviation (b, c).

a) Schematic geometry of the bulkhead Y488 of the F/A-18 Hornet [4], b) An isometric view of the I-beam specimen simulating a detail of the bulkhead, c) An example of one I-beam specimen equipped with strain gauges.

B. Andersson et al

710

Of the total of eight specimens which were instrumented with strain gauges, three fatigue specimens without any strengthening served as reference specimens and they were used for the analytical model’s experimental validation purposes, Figure 2. A summary of the I-beams of the study is provided in Table II. Table II. Description of the I-beam specimens and the experiments, [3]. Specimens 3, 4 and 5 were used for the validation purposes. Machined Proactive IMachined and un-etched and etched repair beam surface patching no. surface 1

2

X

X

X

3

X

4

X

5

X

6

X

X

7

X

X

8

X

X

Experiments performed

Test rig

a. static calibration b. spectrum fatigue test c. static calibration d. UTS test a. other tests b. static calibration c. UTS test a. static calibration b. spectrum fatigue test a. static calibration b. spectrum fatigue test a. static calibration b. spectrum fatigue test a. static calibration b. spectrum fatigue test a. static calibration b. spectrum fatigue test a. static calibration b. spectrum fatigue test

MTS – 250 kN MTS – 250 kN Instron – 1000 kN Instron – 1000 kN Instron – 1000 kN Instron – 1000 kN MTS – 250 kN MTS – 250 kN MTS – 250 kN MTS – 250 kN MTS – 250 kN MTS – 250 kN MTS – 250 kN MTS – 250 kN MTS – 250 kN MTS – 250 kN MTS – 250 kN MTS – 250 kN

Fatigue load spectrum The load spectrum used for both numerical modelling and spectrum fatigue testing was created by Patria Aviation. Using the FE model and flight data from the two FINAF F-18C Hornets equipped with the onboard HOLM instrumentation suite, the stress/strain spectrum was modified to the external load spectrum representing the average usage of the FINAF at structural locations of interest, Ref. [5]. Marker loads were added to the spectrum, in order to aid fractographic examinations but without markedly affecting the actual crack growth behaviour. The creation of the marker loads was emulated from Ref. [2]. Consecutive turning points (instantaneous load values) were combined with a half sine wave (representing an average of 0.5 Hz test frequency run at strain rate control) to define the test spectrum. There were 3619 turning points in the spectrum, see Figure 3. To speed up the spectrum fatigue tests, slightly higher load values were used by multiplying all turning point values with a scalar value of 1.3.

A case study of multiple-site FCG in the F-18 Hornet bulkhead

711

Figure 2. Instrumented I-beam subjected to spectrum loading. Non-destructive inspections In addition to the in-situ strain gauge monitoring, see Figure 2, various NDI methods were employed prior to and during the spectrum fatigue cycling to detect fatigue cracking characteristics: Eddy current (ET), ultrasonic (UT), penetrant (PT) and visual (VT). ET (calibrated on chosen test samples containing known artificial flaws) was the primary method to quantify the size of the cracks from the I-beams. Table III summarises the NDI methods used for each I-beam specimen.

B. Andersson et al

712

250 200

Voima, F[kN] [kN] Force

150 100 50 0 -50 -100 -150 0

500

1000

1500

2000

2500

3000

3500

4000

Käännepisteiden lkm, N [kpl]

Number of turning points 250 200

Voima, F [kN] [kN] Force

150 100 50 0 -50

"Akuperäinen signaali" Original signal Markerit Markers

-100 -150 1230

1235

1240

1245

1250

1255

1260

1265

1270

Käännepisteiden lkm., N Number of turning points

Figure 3. The turning point load spectrum used in the spectrum fatigue tests. Experimental results The three specimens aimed for validation of the numerical methodology, i.e. Ibeams numbers 3, 4, and 5, broke after 108, 130, and 106 blocks, respectively. Failure was defined as a predetermined stroke (displacement) exceedance. The fracture surface of specimen no. 3 is shown in Figure 4 and the upper right hand

A case study of multiple-site FCG in the F-18 Hornet bulkhead

713

side nucleation point is shown separately in Figure 5. It becomes apparent from Figure 5 that crack initiation occurs at several points. Multiple-site nucleation also takes place in the other two test specimens, see Figures 6-7. As has already been mentioned crack initiation is facilitated due to micro pits induced by the etching applied prior to the IVD-coating process. However, it should also be mentioned already here that the stress distribution is locally very homogenous which increases the number of likely stress initiation points. The actual stress distributions will be shown below in the section on numerical modelling. The local crack growth behaviour has not yet been able to back-track to very small crack sizes on the basis of fractography, but the NDI techniques used has been able to detect surface crack lengths from roughly one or two mm depending on test specimen. Such data is shown together with numerical predictions in a following section. According to the preliminary fractographic examinations it seems that in the future, different marker techniques should be studied in order to be able to derive, in more detail, the early crack growth behaviour in order to validate the prediction methods even further. Table III. Summary of the NDI methods used. I-beam no. 1 2 3 4 5 6 7 8

Prior to fatigue tests ET, PT ET, UT ET ET ET, UT ET, UT ET, UT

During fatigue tests

ET, VT ET, VT ET, VT ET, UT, VT ET, UT, VT ET, UT, VT

NUMERICAL MODELLING Stress Analysis of Undamaged I-beam The hp-version of the finite element method, [6], is used in all stress analyses. So called blended function mapping is used to model all fillets with an exact radius. Figure 8a shows the finite element mesh used for linear and nonlinear stress analysis. The mesh has about 23000 HEX elements with nodes per element varying between 8 (p=1) and 105 (p=6) during convergence studies. For the statistical analysis a locally very fine mesh is used in the direction where MSD cracks might line up. Figure 8b shows a detail of the local mesh design. Note

714

B. Andersson et al

that mesh refinement is only needed in one direction in order to be able to separate cracks in quite arbitrary MSD-distributions. Figure 8c shows the calculated stresses σ y obtained from a geometrically nonlinear analysis with high-accuracy finite elements (105-noded HEX elements). The solution shown is converged, i.e. the errors in point wise stresses are less than 1 MPa, the maximum stress being just over 400 MPa for the maximum load 204.1 kN. In Figure 8e experimentally determined strains at two gauge locations, see Figure 8d, are compared with virtually exact calculated strains (averaged over the gauge areas). The measured strains show a significant hysteresis effect. At the load 100 kN the measured strain at gauge OL1 is 2447 ± 123 μstrain while the corresponding computed strain is 2572 μstrain, a difference of +5% compared to the average experimental strain (at 100 kN load). For the maxim load, i.e. 203 kN, the (nonlinear) analysis overestimates the strain with 7%, see Figure 8e. The reasons for the difference between experimental and numerical strains are not known. If the reason is related to mechanical effects, such as friction etc, calculated stress intensity functions are overestimated by about 5% too. The relative error in calculated stress intensity factors is much smaller than 1%.

Figure 4. Fracture surface of test specimen no. 3 showing at least three nucleation sites. The area inside the rectangle is shown in Figure 5.

A case study of multiple-site FCG in the F-18 Hornet bulkhead

715

Figure 5. Crack nucleation site at the upper right hand location of Figure 4.

Figure 6. Fracture surface of specimen no. 4 showing multiple-site crack initiation.

716

B. Andersson et al

Figure 7. Fracture surface of specimen no. 5 showing multiple crack initiation locations.

Figure 8a. Finite element mesh of the test beam for linear and non-linear analyses.

A case study of multiple-site FCG in the F-18 Hornet bulkhead

Figure 8b. Mesh design for statistical fatigue analysis

Figure 8c. Calculated stress distribution σ y [MPa] for maximum load used during testing. Y = longitudinal axis of the I-beam.

717

B. Andersson et al

718

Figure 8d. Strain gauge locations.

Figure 8e. Measured and calculated strains at two gauge locations (OU1 and OL1) in test beam.

A case study of multiple-site FCG in the F-18 Hornet bulkhead

719

Fast and reliable fracture mechanics analysis in case of MSD The cracks are considered to be placed arbitrarily in several critical section of the critical detail. At each location, the stress intensity factor is affected by all the crack sizes involved in the model. (1) K = f (a , c , x , a , c , x , ) 1

1

1

2

2

2

The change for each individual dimension of any crack will affect the value for all the other cracks. For the fatigue crack growth calculation, the cycle-by-cycle computation method is used so that the method can be applied for the multi-channel arbitrary spectrum loading condition. For an arbitrary crack increment along any dimension of the crack front, the stress intensity factor is determined with:

K = K 0 + ΔK

(2)

The increment in the stress intensity factor can be written as N ⎛ ∂K ⎞ ∂K ∂K ΔK (a1 , c1 , x1 , a2 , c2 , x2 , ) = ∑ ⎜⎜ Δai + Δci + Δxi ⎟⎟ + 0( ∂ci ∂xi i =1 ⎝ ∂ai ⎠

)

(3)

By omitting the high order terms in the above relation, the first order approximation the stress intensity factor may be calculated according to the crack increment using

⎛ ∂K

N

ΔK (a1 , c1 , x1 , a 2 , c 2 , x 2 , ) =

∑ ⎜⎜⎝ ∂a i =1

i

Δa i +

⎞ ∂K ∂K Δc i + Δx i ⎟⎟ ∂c i ∂x i ⎠

(4)

The partial derivatives can for example be estimated from difference quotes. However, we will elaborate on a more general and accurate scheme below. A multiple-site crack situation of the type shown in Figure 9 is considered. Stress intensity functions for interacting cracks of arbitrary size and location in the frame is needed for the statistical fatigue crack growth analysis presented below. In the present paper we describe a fast and very accurate method for calculation of a

~

highly accurate approximation space Kα (ϕ , a , c , xij , y ), α = I, II or III for the

(0 ≤ ϕ ≤ π / 2) or semi-elliptical (0 ≤ ϕ ≤ π ) shape of size (ai , ci ) , i = 1,..., n . Cracks are located in y -sections

case

of

totally n

cracks

of

quarter-

B. Andersson et al

720

y j and at x -coordinates xij where ( ai , ci , y j ) are stored in the vectors a, c and y , respectively and xij is stored in the matrix xij . For cracks defined by

(ak , ck , xkl , yk ) with i ≠ k and/or ij ≠ kl interpolation will be used with methods described below. The number of points in the space

~ Kα easily becomes of order 10 20 to 1030 even

for moderate n, say of order 20. In one statistical fatigue analysis, of the order

~ 10 6 - 10 7 3D multiple-site stress intensity factor solutions will be used in Kα ,

hence a very effective computational scheme has to be developed.

Figure 9. Quarter- and semi-elliptical cracks at inner and outer flange at different sections y. The method used is the so-called splitting method described in Ref. [7]. The splitting scheme is using the finite element method where of the order of thousands of local meshes (in our case) and one global mesh are used in a kind of nested analysis. The method is exact in the sense that if the finite element method would provide exact solutions, exact stress intensity factor data would be obtained also for arbitrary crack configurations. In the present case we adopt an hp-version of the finite element method which provides practically exact solutions. Hence the

~

calculated K-data are also exact for any crack configuration in the space Kα . The

A case study of multiple-site FCG in the F-18 Hornet bulkhead

721

scheme for stress intensity factor calculation is reviewed below together with the splitting method for fast fracture mechanics analysis.

Calculation of stress intensity factors Fast, accurate and reliable calculation of stress intensity factors K is critical for the fast solution scheme to be used in the probabilistic analysis. We therefore briefly review the method for reliable and accurate extraction of K’s for general smooth crack fronts in 3D domains. Consider a smooth crack front. Denote by r, the (short) distance from the edge to a point, and θ (-π ≤ θ ≤ π ) the polar angle, see Fig. 10. The displacements u at a point x3 on the crack front can be written in the form,

u( r ,θ , x3 ) =

∑ Kα ( x3 )r 1 / 2Ψα ( θ ) + smoother terms

(5)

α = I ,II ,III

Where KI (x3), KII (x3), and KIII (x3), are the mode I, mode II, and mode III stress intensity functions, respectively.

Figure 10. Domain Ωe used for calculation of stress intensity functions K

For smooth edges, the edge intensity functions Kα(x3) are analytic on open intervals sk ≤ x3 ≤ sk+1. Hence, we approximate the edge intensity functions Kα with the polynomials:

B. Andersson et al

722 p

K α ( x3 ) = ∑ kαn Pn ( s), n =0

s=

2( x3 − xk ) −1 xk +1 − xk

(6)

where kαn are unknown coefficients, p is the polynomial order, and Pn the Legendre polynomials. Figure 10 illustrates a domain Ωe used for calculation of the coefficients kαn . The local domain used has three cylindrical surfaces with circular cross sections (ρ2 > ρ1 > ε) perpendicular to the edge considered. The two outer cylindrical surfaces are exactly modeled in the finite element analysis. By applying the Maxwell-Betti reciprocity theorem the accuracy of the calculated K’s thus will depend only on a weighted average of the finite element solution inside the extraction domain. This gives exponentially fast convergence, with increasing p to the exact K-values. The splitting method A mathematical splitting method will be used to efficiently and reliably calculate the millions of K- solutions that will be needed in a statistical analysis. For clearness of notation, we review the solution of the crack in a 2D context. For a more complete discussion of the mathematical theory, see Refs. [7-9]. In the splitting method, the 3D fracture mechanics problem is split into three subproblems. The simple multi-site cracking scenario shown in Figure 11 is used to exemplify the decomposition of the problem into one global un-cracked problem and local problems with a single crack.

A case study of multiple-site FCG in the F-18 Hornet bulkhead

723

Figure 11. Schematic description (in a 2D setting) of the splitting method The three major solution steps in the splitting scheme are summarized below. I. The Global Crack Free Problem: The solution of the global crack free ( 0)

problem is U G and the only result required from the finite element analysis is the stress distributions on the surfaces where the cracks are to be located. Thus, this sub-problem is independent of the number and size of cracks under consideration. This gives computational efficiency. III A Set of M Local Problems: A local mesh is developed for each crack size parameter. The applied load consists of L different normalized crack surface tractions with the solutions denoted as

{U

( m ,l ) L

}

m = 1, 2… M , l = 1, 2,… L . The local models contain a single

crack. Close to the crack front, see Figure 10, another 4-6 cylindrical layers of elements are introduced. So-called blended mapping is used in the analysis to describe the exact shape of the crack and the radices at fillets etc. The results required from the local analyses are the tractions and displacements on an arbitrarily selected surface, Γ (dashed line in Figure 11) which encloses the entire crack in addition to the stress intensity functions KI(s), KII(s), and KIII(s) for all the crack face loadings.

B. Andersson et al

724

V A Set of Global Crack Free Problems: The global mesh is re-used in and analyzed with prescribed jumps in the tractions and displacements at the surfaces Γ used in the local problems. Thus, the output from the local problems is input to the load calculations in the global model on level V.

{

( m ,l )

The solutions are denoted U G

}

m = 1, 2 … M , l = 1, 2,… L .

The approximate displacement solution, U to the exact 3D solution, U is:

M

L

M

L

U = U G( 0 ) + ∑∑ α ( m ,l )U G( m ,l ) − ∑∑α ( m ,l )U L( m ,l ) m =1 l =1

(7)

m =1 l =1

where α(m,l) are scaling factors determined by solving a small set of linear (0)

( m ,l )

equations. Thus, with the known U G , U G

( m ,l )

, and U L

, the solution U can be

calculated with virtually no computational cost per MSD-configuration. The computational efficiency of the strategy devised makes it feasible to perform Monte Carlo type studies of 3D multiple-site fatigue crack growth. Figure 12 exemplifies types of local meshes used at level III. Note that the FEmeshes used for local (III) and global analyses (I and V) can be created completely independently of each other. This simplifies mesh design. Mesh generators have been created to generate the thousands of local meshes used on level III.

Figure 12. Two of the four types of local finite element meshes used at level III.

A case study of multiple-site FCG in the F-18 Hornet bulkhead

725

The global mesh needed for statistical analysis must be designed after certain principles. The mesh density in the x-direction must be 2-3 times denser than the minimum spacing between crack initiation sites of interest. Figure 13 exemplifies the meshing used in the present study. The element spacing used is 0.5 mm in the x-direction in the region with MSD-cracks. The fine part of the mesh is tied to the coarser part with multi-point constraints. Highly accurate stress distributions are derived by employing the hp-version of FEM in the stress analysis. If a high resolution database

~ Kα is used, typically the following crack sizes (in

mm) are analyzed for each of the four types of meshes.

ai = 0.1,0.2 … 1.2,1.3,1.4 ... 1.6,1.8, … 2.8, 3.0, 3.5,4.0, … 7.5, 8.0

(8)

ci = 0.1,0.2 … 1.2,1.3,1.4 ... 1.6,1.8, … 2.8, 3.0, 3.5,4.0, … 7.5, 8.0

(9)

(flange thickness is 8.2 mm) The xij -locations, for given j, of the cracks are selected with a spacing of typically 1.0 mm, hence of the order 40 cracks are considered on the inner flange and 57 on the outer flange. If three y-sections are analyzed, a space

~ Kα with 10 250 points

corresponding to unique multiple-site crack scenario with non-overlapping cracks is obtained. For smaller cracks than 0.1 mm, very accurate stress intensity data can be obtained after proper scaling (square-root dependence) for actual ai / ci ratio.

B. Andersson et al

726

Figure 13. Proper mesh design for probabilistic multiple-site fatigue analysis.

Computational efficiency in K-calculations The steps I-V in the splitting scheme requires large computational efforts, typically of the order of one thousand CPU-hours for the complex crack patterns of interest here. As described above, the stress intensity functions for any crack pattern can then (i.e. after step V) be obtained by solving a set of linear equations. The system of equations has typically in the order of 40 ⋅ n unknowns (for high-accuracy Ksolutions), where n is the number of cracks in the MSD scenario. The time t needed to derive stress intensity functions on a modern personal computer hence is only about ,

t = 20 ⋅ (n / 100)3

(10)

seconds. Derivation of virtually exact stress intensity functions for one 10-crack scenario hence takes about 0.02 seconds and 1 million solutions takes less than 6 hours on a PC. Large scale statistical analysis in this project is performed on a 32 CPU HP-computer with a software design that exhibits perfect scalability, i.e. one million exact K-solutions can be calculated in about 10 minutes.

A case study of multiple-site FCG in the F-18 Hornet bulkhead

727

Figure 14 exemplifies calculated stress intensity functions for c = 2.0 mm and 0.2 ≤ a ≤ 8.0 mm for a corner crack in the inner flange of the beam. The points plotted in the figure have been calculated from the polynomials approximations with 11 points per finite element edge (for element edges at the crack front). The figure shows that the point sets changes smoothly with ϕ and with no discontinuities at finite element interfaces (calculated relative discontinuity from two neighbouring polynomials is typically 0.0001). Close to the two ends of the crack fronts the solutions do capture the vertex behaviour well too. Note that Kfunctions of the type shown are used to calculate the weighted values defined by Equations 13 to 15, see below. The Simpson’s rule is used. Since the number of points used is 140-280 for the crack types considered, the accuracy of the numerical integration is very high (4-5 digits accuracy).

K-calculation in case of fatigue analysis For the cycle-by-cycle fatigue crack growth analysis performed in the present paper, stress intensity function approximations have to be known for all cracks in analytical form. Equation 4 shows the simplest approximation. The solution

~

set Kα is used to create a more accurate analytical approximation. This approach

~

overestimates the stress intensity functions except at points in Kα . A much less conservative algorithm is the following, for simplicity firstly described in the simplest possible 2D context. Consider a single crack of initial size (a0 , c0 ) located at a coordinate x. An analytical approximation for K is sought for using a

~

~

space K 2 D . An ( a, c ) sub-plane of the space K 2 D is shown in Figure 15. Approximations to K will be derived in a sequence of rectangular domains dependent on the crack size ( a ( N ), c ( N )), N being the number of load cycles.

~

The fatigue life analysis starts by identifying the rectangle in K 2 D such that

ai ≤ a0 ≤ ai + Δai , ci ≤ c0 ≤ ci + Δci , and calculation of stress intensity functions for crack sizes corresponding to the four corner nodes in the rectangle. Stress intensity functions for points inside the rectangle are obtained using interpolation. If an approach inspired by finite element technology is used, an analytical expression for K in the rectangle considered becomes, L

K (a, c) = ∑ kl ⋅H l (ξ ,η ) l =1

Where:

(11)

B. Andersson et al

728

L = 4, H1 = (1 − ξ ) ⋅ (1 − η ) / 4, H 2 = (1 + ξ ) ⋅ (1 − η ) / 4, H 3 = (1 + ξ ) ⋅ (1 + η ) / 4, H 4 = (1 − ξ ) ⋅ (1 + η ) / 4 , ξ = 2 ⋅ (a − ai ) / Δai − 1, η = 2 ⋅ (c − ci ) / Δci − 1 and kl are the values of the stress intensity functions (for example the complete function, or weighted averages) at the four corners in the rectangle ai ≤ a0 ≤ ai + Δai , ci ≤ c0 ≤ ci + Δci considered. In the fatigue analysis the analytical expression Eqn. 11 is used until a becomes larger then ai + Δai or/and c becomes larger than ci + Δci . When this happens a new approximation is calculated for the neighbouring rectangle (c.f. C1, C2 and C3 in Figure 15). The process is repeated until cracks coalesce or grow through the thickness or similar.

At corners in the rectangle, the approximation is virtually exact since accurately calculated K-solutions are used. For the case with n cracks defined by two n

parameters (ai , ci ) the rectangle in Figure 15 would correspond to a 2 2n

dimensional polyhedron with 2 corners (nodes). Stress intensity functions for cracks located at quite arbitrary x -coordinates are needed in the present work. For this case when each crack is defined by three parameters ( ai , ci , xi ) a polyhedron n

3n

in a 3 -dimensional space with 2 corners is used to create an analytical expression for K inside the polyhedron. With increasing n , the computational work

t K (in seconds on a n modern personal computer) for calculation of an approximation to K in a 3 increases exponentially. An estimate for the computational cost dimensional polyhedron, is obtained from Eqn. 10 as:

t K = 20 ⋅ 23n ⋅ (n / 100)3

(12)

For a 6-crack scenario, the computational cost is about 1000 seconds for the more than ¼ of a million crack patterns that need be analyzed to form the K-

~

approximation in a polyhedron in the 829-dimensional space Kα used. This is prohibitively much unless the calculations are performed in parallel on a computer cluster. Note also that corner data need to be computed only once, the derivatives with respect to x need not always be computed which leads to saving of many orders of magnitude.

A case study of multiple-site FCG in the F-18 Hornet bulkhead

729

Figure 14. Calculated stress intensity functions for a corner crack at the 3/ 2 inner flange (in N / mm )

~

Figure 15. Part of space K 2 D showing the two crack sizes.

B. Andersson et al

730

Averaged stress intensity functions for fatigue analysis The value of stress intensity factor changes dramatically along the crack front near the areas (see Fig 14) when the crack front intercepts the free surface. It has been considered that the local value near vertices of the crack front of the stress intensity may not provide a reliable reference to determine the growth of a crack.

Figure 16. Geometrical quantities and schematic of the virtual crack growths.

Three conditions are considered for a partial elliptical crack as shown in the above figure. The virtual crack growth increment, respectively along either side of the crack long the surface (δc1, δc2), or in depth (δa), see the schematic in the above figure, are considered. The equivalent stress intensity factors at c1, c2 and a are determined by

Ka =

π

K (ϕ )sin π∫ 2

2

2

ϕdϕ

(13)

0

and

K c1 = and

4

π

π /2

∫ K (ϕ )cos 2

0

2

ϕdϕ

(14)

A case study of multiple-site FCG in the F-18 Hornet bulkhead

K c2 =

π

K (ϕ )cos π π∫ 4

2

2

ϕdϕ

731

(15)

/2

The Link-up of Fatigue Cracks Inevitably, the link-up of cracks should be considered for the multiple crack problems. Ideally, the new crack after the link-up of adjacent cracks should be determined with an equivalent crack that has the same stress intensity factors at both ends on the surface and the deepest point in depth. The algorithm for this solution is however subjected to convergence and efficiency problem in the data base since the iterations must be used. Two alternatives for the linkup of cracks are used instead for the computation efficiency and algorithm stability. The first alternative is to assume the new crack has the same crack size on the surface as the schematic in the following figure shows.

Figure 17. The first linkup criterion.

For this solution, the new crack area is assumed to be the sum of the old crack areas so that the loss of the load carrying capability of the material is not changed. This assumption gives

Area new = Area1 + Area 2

(16)

To assume that the crack size on the surface is the same after the link-up such as:

c new = c1 + c 2

(17)

B. Andersson et al

732

The crack size in the depth becomes

a new = (a1 × c1 + a 2 × c 2 ) /(c1 + c 2 )

(18)

The centre of the new crack is the computed as:

x new = [(x1 − c1 ) + (x 2 +c 2 )] / 2

(19)

This assumption is suitable when the stress on the surface is significant for the places where the stress concentration gradient is high, or the bending is significant. For the other cases, the second solution is used to assume that the depth of the crack after the linkup is kept the same as the largest crack, and the dimension of crack on the surface is determined based on the same principle that the crack area is kept the same. The schematic of this solution is shown in the following figure.

Figure 18. The second linkup criterion For this assumption, the size of the new crack in depth direction is determined with:

a new = max(a1 , a 2 )

(20)

The size of crack on the surface becomes:

c new = (a1 × c1 + a 2 × c 2 ) / a new

(21)

When Crack 2 is the larger one, the centre of the new crack is determined by:

x new = x 2 + c 2 − c new

(22)

A case study of multiple-site FCG in the F-18 Hornet bulkhead

733

When Crack 1 is the larger one, the centre is determined by:

x new = x1 − c1 + c new

(23)

VALIDATION The computational scheme described above has to be validated. Specimens 3 and 5 were singled out for this purpose. The first specimen is number 3. Figure 4 shows the crack pattern observed at final failure. Two elliptical cracks (semi-quarter) nucleated on the inner flange and a corner crack on the outer flange. The corner crack did grow in another y-section prior to final fracture. Experimental results for cracks at locations A and B, based on NDI observations on the free surface, are shown in Figure 19 as circles. For predictions, material properties and the basic relation between stress intensity range and fatigue crack growth rates from the fatigue crack growth computer code NASGRO, Ref. [10], were used. The actual predictions were made by the in-house code CLOTEST, Ref. [11]. However, all calculations are based on linear elastic theory, and no crack growth retardation was assumed. Calculated crack sizes shown in Figure 19 are based on the assumption that initially ai = ci = 0.1 mm. With these assumptions the experimental crack growth is somewhat slower than the predicted ones. However, as discussed in the section on stress analysis of the un-cracked beam the calculated stresses were roughly 5% larger than the measured ones. Redoing the same calculations with 5% reduced stresses, and also computing the crack growth behaviour for the corner crack, results in the curves shown in Figure 20. These results agree well with the few experimental results available.

734

B. Andersson et al

Figure 19. Calculated and observed crack sizes during fatigue loading of test specimen no. 3.

Figure 20. Calculated and observed crack sizes during fatigue loading of specimen 3. Stress levels decreased by 5% relative predictions in Fig. 19. Corner crack predictions have also been added.

A case study of multiple-site FCG in the F-18 Hornet bulkhead

735

Performing the same type of calculations on specimen 5 gives the results shown in Figure 21. The initial corner crack size is again assumed to be 0.1 mm. The two predicted curves are for the same crack, but with the original stress levels, from FE analysis, and with a 5% reduction due to the effect discussed above. Again, the predictions with reduced stress level compare very well with the experimental observations. The initial crack size assumption of 0.1 mm for the calculations was based on typical sizes of the initial etching pits. As is clear from all figures of predicted crack growth behaviour, the crack speed increases rapidly after the crack sizes grow to the order of one mm. Hence, elimination of the etching process can be expected to lead to a rather large increase in fatigue life, as essentially the early crack growth observed here, would be taking place after a traditional crack initiation phase.

Figure 21. Calculated and measured data (locations ABD) for corner cracks in test specimen no. 5.

736

B. Andersson et al

As mentioned above, the present numerical methodology makes it possible to study both crack interaction and statistical modelling of crack growth in detail, as the large number of K-computations needed for such purposes can be obtained virtually exactly in very short time. For test specimen no. 3 an additional analysis has been performed that takes into account the interaction of all three cracks previously studied individually. As before, the stresses were lowered by 5% in these calculations. Figure 22 shows the size of the three different cracks that were previously studied individually, after 94 blocks when crack interaction is considered. In Figure 23 the predicted crack growth behaviour, when the crack are interacting with each other, is shown. Here, the initial crack sizes were chosen to 0.1 mm for the semi-elliptical cracks and 0.127 mm for the quarter-elliptical crack. Reducing the latter to 0.1 mm also leads to another 4 blocks of total fatigue life. Crack growth is very rapid after the cracks have grown to the sizes shown in Figure 22. Failure occurs within the next block of fatigue loading. The predicted sizes at failure, of the three interacting cracks, correspond rather well with the observed experimental results, see Figure 22.

Figure 22. The three individual cracks that are interacting with each other are shown with solid lines, after 94 blocks of fatigue loading.

A case study of multiple-site FCG in the F-18 Hornet bulkhead

737

Figure 20. Predicted crack growth behaviour for specimen no. 3, accounting for interaction of the three individual cracks.

STATISTICAL ANALYSIS The efficiency of the computational scheme devised in the present paper admits a very advanced statistical analysis of complex crack geometries in complex structural parts. Only quarter- and half-elliptic crack-shapes have been discussed above but the theory is general and hence admits for general cracks shapes, broken beam flanges etc. We will here consider the statistical fatigue life of test specimen no. 3. Because the cracking has been found to be fairly rapid we use the statistical analyses to assess the potential benefits of reducing the initial flaw sizes. This could be achieved in practice by removing the etching process, which on the other hand might influence corrosion properties. Clearly, it is hard to guarantee that even

B. Andersson et al

738

with different surface conditions, no flaws larger than a certain size would be prevalent. Hence, a probabilistic model provides a much better feeling for the general effects of small crack growth behavior than a discrete analysis starting with one single flaw size does. In our example we will assume that three cracks of random size and with 0.0127 mm < a0 = c0 < 0.127 mm initially exist at the three initiation locations observed in experiments ( x =0.0, 10.0, 18.0 mm, respectively). Two initial flaw distributions are considered. The first is the normal probability function adjusted at the two tails. The probability density for the stochastic function a0 is: 2 f e −(( a0 −μ ) / σ ) / 2 for − 2 ⋅ σ < a0 − μ < 2 ⋅ σ 2 ⋅π | a0 − μ |≥ 2 ⋅ σ 0 for

Where, c0

(24)

= a0 , μ =0.06985 mm is the mean value and σ = 0.028575 mm the

standard deviation (in agreement with initial flaw sizes being randomly distributed between 0.0127 and 0.127 mm, respectively) and f a constant accounting for the tail areas removed. The second initial flaw distribution considered is uniform over the interval 0.0127 to 0.127 mm. The fatigue life up to final failure was calculated explicitly for over four thousand beams, using Monte Carlo simulation and the distribution given by Eqn.24. By fatigue life is here meant the number of load blocks needed to first reach the fracture toughness K Ic at one of the three crack fronts. Figure 24 shows the fatigue life distribution functions for the probability of failure after N load blocks for the two initial flaw size distributions considered. The time until failure observed in experiments, i.e. 108 load blocks for specimen 3 is also indicated in the figure. This figure shows that the calculated probability that failure (for the Normal initial distribution) shall have taken place after 108 blocks is 47%. This number is of course dependent on the assumptions, Eqn.24. For the uniform initial distribution, the corresponding value is 63%.

A case study of multiple-site FCG in the F-18 Hornet bulkhead

739

Hence, even with three initial flaw sizes that are random from one tenth of the crack length used in the previous discrete predictions, up to that very length, i.e. 0.127 mm, failure will occur after roughly the same overall number of loading blocks as for the discrete analyses. The variation in life from a 10% failure probability up to a 90% failure probability is around a factor of two. Once again, the effect of the initial flaw size becomes very prevalent. From a damage tolerance point of view the time difference between the time when a crack can be observed and the time of final structural failure is very important. If this time slot is too short, costs for regular inspection will be prohibitive. The statistical analysis above shows that this time difference varies between 12 and 19 load blocks.

Figure 24. Calculated distribution functions showing the number of load blocks until failure for two initial crack size distributions at the three crack locations in specimen 3.

CONCLUSIONS A novel methodology for fast and accurate calculations of stress intensity factors for multiple-site crack scenarios has been derived. This methodology makes it possible to study multiple-site cracking and also statistical modelling of crack growth in detail, as millions of stress intensity factors can be computed virtually exactly, in short time on parallel computers.

B. Andersson et al

740

The fracture mechanics modelling has been used to predict cracking in an I-beam type of specimens representing the FINAF F18 Hornet bulkhead. For this purpose, fatigue crack predictions have so far only been done without considering any crack retardation phenomena due to load interaction effects under spectrum loading. Such features may be included in the future. When accounting for the 5% difference in local strain levels that were observed when comparing accurate stress analysis of the un-cracked structure to actual strain measurements, predicted crack growth behaviour agreed well with experimental observations. Further validation of the derived methodology will be possible after new experiments with marker loads facilitate back tracking of the crack morphology down to small crack sizes. Modelling including the interaction of multi-site cracks showed that the interaction effects only take place rather late in the overall life. This effect, however, depends on geometry, load level, and crack patterns, and may be quite different in other situations. The etching pits, in the order of 0.1 mm, were found to shorten the fatigue life substantially, by removing the crack initiation phase. A statistical analysis that considers three initial flaws located at the observed experimental initiation points showed how such analyses can be used to improve the understanding of early crack growth behaviour. By assuming that the three initial flaw sizes are randomly distributed from one tenth of the crack length used in the previous discrete predictions, up to that very length, i.e. 0.127 mm, failure will occur after roughly the same overall number of loading blocks as for the discrete analyses. The variation in life from a 10% failure probability up to a 90% failure probability is around a factor of two. The performed analyses constitute the first application of the new methodology. In the future such modeling may certainly be taken further on the present bulkhead structure, in conjunction with more detailed fractographic work on the behaviour of early crack growth. Another obvious application of the methodology derived here is the aging aircraft problem. For all aging aircraft fleets, the present procedure may assist in taking well informed decisions on life extension programs, inspection intervals, or removing the aircraft from service.

REFERENCES [1]

Siljander, S. (Ed.). A review of aeronautical fatigue investigations in Finland May 2007 – April 2009. ICAF Doc.No. 2418. In: Bridging the gap between theory and operational practice, Proceedings of the 31 ICAF Conference, 2009.

[2]

Barter, S.A.. Fatigue crack growth in 7050-T7451 aluminium alloy thick section plate with a glass bead peened surface simulating some regions of

A case study of multiple-site FCG in the F-18 Hornet bulkhead

741

the f/a-18 structure, 2003, Australian Government, Department of Defence, Defence Science and Technology Organisation, DSTO-TR-1477. [3]

Koski, K., Juntunen, J., Sarkimo, M., Merinen, S., Teittinen, T., Pitkänen, I. The fatigue tests of the I-beams. Report № VTT-R-01979-09. Espoo: VTT Technical Research Centre of Finland, 2009 (in Finnish, classified).

[4]

White, P., Molent, L., Barter, S., Interpreting fatigue test results using a probabilistic fracture approach. International Journal of Fatigue 27(2005)7, pp. 752–767.

[5]

Kettunen, J.and Linna J. Design of the I-beam (Y488). Report no. HN-L0142. Halli: Patria Aviation 2008 (in Finnish, classified).

[6]

Andersson B, Falk U, Babuska I and Petersdorff T , ''Reliable stress and fracture mechanics analysis of complex components using a hp-version of FEM'', Int. J. Num. Meth. Eng., Vol 38, 1995, pp 2135-2163

[7]

Babuska I, Andersson B, “The Splitting Method as a Tool for Multiple Damage Analysis”, SIAM Journal Scientific Computing, Vol 26, No 4, 2005, pp 1114-1145

[8]

Andersson B, Babuska I and Falk U, ''Accurate and reliable determination of edge and vertex stress intensity factors in three-dimensional elastomechanices'', Proceedings of 17th Congress of the International Council of the Aeronautical Sciences, Paper No: ICAS904.9.2, 1990, pp. 1730-1746.

[9]

Nilsson, Karl-Fredrik and Börje Andersson. “Analysis Methodology for Fatigue Crack Propagation and Residual Strength of Joints with Widespread Fatigue Damage.” Proceedings of the 1999 Aircraft Structural Integrity Program Conference, San Antonio, TX 30 November – 2 December 1999.

[10]

Forman, R.G., Shivakumar, V., Mettu, S.R., Newman, J.C., Fatigue crack growth computer program; “NASGRO” version 3.0. Reference Manual, JSC-22267B. Engineering directorate, August 2000, National Aeronautics and Space Administration, Lyndon B. Johnson Space Center.

[11]

Wang, G. S. and Blom, A. F., “A strip model for fatigue crack growth predictions under general load conditions,” Engng. Fracture Mech. Vol.40, No.3, 1991, pp.507-533.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

EXPERIMENTAL AND THEORETICAL COMPARISON OF SOME MULTIAXIAL FATIGUE DESIGN CRITERIA IN THE CONTEXT OF LIFE ASSESSMENT OF ROTATING PARTS IN TURBOENGINES V. Bonnand(1), J.L. Chaboche(1), H. Cherouali(2), P. Gomez(3), P. Kanouté(1), D. Pacou(1), P. Paulmier(1), E. Ostoja-Kuczynski(4), F. Vogel(4) 1

ONERA, Châtillon, France Snecma, Villaroche, France 3 CEAT, Toulouse, France 4 Turbomeca, Bordes, France 2

Abstract: An experimental study is presented for two alloys commonly used in compressor and turbine discs, that includes classical uniaxial tests together with multiaxial ones, both in tension-compression torsion, in tension-internal pressure and in biaxial loads applied on a specially designed cruciform specimen. The results are analysed by comparison with many different multiaxial fatigue criteria, including those based on octahedral shear stress amplitude combined with hydrostatic pressure, and more recent models based on critical plane approaches. The study shows that none of them is able to correctly correlate all the considered uniaxial and multiaxial conditions. A new combined criterion is then proposed that offers a possible compromise and significantly improves the predictions.

INTRODUCTION Fatigue is recognized to be the principal cause of many mechanical failures. Complex geometrical structures are often involved and complex multixial loadings are occurring. Rotating parts in turboengines, like turbine or compressor discs, that are subjected to severe and complex cyclic loads, are typical examples. They M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 743–764. © Springer Science+Business Media B.V. 2009

V.Bonnand et al.

744

involve the interaction of Low Cycle Fatigue (ground-to-ground cycles) and High Cycle Fatigue (vibratory) in regions with high biaxialities and high mean-stresses. Improved understanding of the damage mechanisms associated to fatigue has lead to the development of numerous fatigue models. Historically, the research work was based on experiments involving uniaxial stress states. Nowadays, many multiaxial fatigue models and data are available in the literature. In spite of the number of proposed criteria, no universally accepted approach yet exists. Several good reviews of multiaxial fatigue criteria are available in the literature (Garud [1], Brown and Miller [2], You and Lee [3], Wang and Yao [4], Papadopoulos [5] or Macha and Sosino [6]). . These approaches can be divided into three categories. One popular approach has traditionally been to extend the static yield criteria to fatigue by combining octahedral shear stress amplitude with hydrostatic pressure. By similarity, several strain based criteria have been proposed. Another class of approach uses energy as a correlating parameter for the multiaxial data. Energy criteria have been proposed and tested by several researchers (Garud [7], Radakrishan [8] and Ellyin [9]) but present some difficulties for quasi elastic cycles. A third method for multiaxial fatigue life evaluation has been to use critical plane approaches. In all case, as it will be underlined in the present paper, the largest difficulty with any such fatigue criterion is to take simultaneously into account both the mean-stress effects (in uniaxial conditions) and the multiaxiality effects. With the objective to assess current design methodologies for turboengine discs, an important program that involved four main partners : Snecma, Turboméca, Onera and CEAT, has been conducted on two classes of classical disc materials, a titanium alloy and a nickel-based alloy. The experimental program has been established with the objective to obtain various multiaxial states of stress. In addition to standard LCF strain controlled fatigue tests (with various strain ratio) and HCF tests (under stress control), a significant number of complex multiaxial tests has been performed:

tension / torsion in-phase and out-of-phase conditions on tubular specimens, tension / internal pressure in-phase and out-of-phase conditions on tubular specimens, biaxial fatigue loadings on a specially designed compact cruciform specimen (in-phase and out-of-phase conditions).

Several existing multiaxial fatigue have been evaluated to determine their suitability at estimating the fatigue damage of the multiaxial fatigue data generated in the program. A simple Basquin power law has been considered in this study to correlate the lifetime to an effective stress. Several formulations of this effective stress have been analysed regarding multiaxiality effects, namely Sines [10], Crossland [11], the critical plane models of Brown-Miller [12], Fatemi-Socie [13], and Smith Watson Topper [14]. The maximum principal stress criterion, as used by

Experimental and theoretical comparison of some multiaxial fatigue design criteria

745

Gonçalvès et al. [15] is also evaluated. A combination of this last criterion with the Sines one is also proposed with the objective to estimate correctly torsion and fatigue conditions together with a correct description of mean stress effects.

EXPERIMENTAL PROGRAM Classical uniaxial fatigue tests have been performed in order to complement existing data bases on the two materials considered in the program: the Titanium alloy Ti-6Al-4V, at room temperature and at 200°C, and the Nickel base alloy INCO718-DA at 550°C. These tests involve stress controlled and strain controlled conditions under various stress or strain ratio. Tests have also been made at high frequencies and large stress ratio. They are not presented here. Only some of the results are used below, when determining the mean stress effects and the uniaxial life relationships. The multiaxial test conditions are detailed a little more as follows. The axial/torsional tests on tubular specimens allow combining tension-compression and shear stresses. In order to cover a large domain of stresses multiaxiality, and more precisely the equibiaxial configuration, tension / internal pressure tests have been developed. A drawback of these is that the biaxial loading can be only repeated, strongly positive. The last testing configuration is this one requiring cruciform specimens. It allows also having different ratio between the two loading axes, but it presents some difficulties for fatigue test. The two significant ones are, firstly to avoid initiation in the attachment part, failure must occur in the central gage section, and secondly, the geometry requires Finite Element calculations to estimate the state of stresses, as will be shown below. The above mentioned multiaxial fatigue tests were performed in two partner laboratories, ONERA and CEAT. The investigation of multiaxiality effect was mainly conducted at 200°C for the titanium alloy and 550°C for the nickel-based alloy. All specimens were machined from blanks taken from turbine discs. Specimen geometries are different according to the material and the testing system. Table I describes the multiaxial testing devices required for this investigation and their general capabilities. Figures 1a and 1b present respectively the biaxial fatigue test device from ONERA and the Tension-Internal pressure fatigue device from CEAT, these two kinds of tests will be described more in details below. Figure 2 illustrates the various multiaxial techniques and the main loadings considered in the program.

V.Bonnand et al.

746 Multiaxial testing devices Kinf of Partner loading Biaxial tension TensionONER Torsion A TensionInternal pressure

CEAT

Mechanical Parameters Maxima of loading

Thermal Parameters Freq. max.

Kind of heating system Induction with susceptor

Temp. max

F1= ±200 kN

F2= ±200 kN

5 Hz

F= ±63 kN

T=±500 N.m

2 Hz

Induction

1200°C

F= ±63 kN

P=1500 bars

0,2 Hz

Induction

950°C

TensionTorsion

F= ±250 kN

T=±2000 N.m

0,5 Hz

TensionInternal pressure

F= ±250 kN

P=1200 bars

2 Hz

Induction and Joule effect (or furnace) Induction and Joule effect (or furnace)

600°C

1200°C

400°C

Table I: Multiaxial fatigue testing devices.

a. Biaxial tension testing device for cross-like specimen (ONERA)

b. Tension – Internal Pressure for tubular specimen (CEAT)

Figure 1: Biaxial Fatigue testing devices.

Experimental and theoretical comparison of some multiaxial fatigue design criteria

747

Figure 2 : Possible stresses domain and corresponding experiments. (Presentation of some applied loading paths) The biaxial fatigue testing device for cruciform specimen was designed and manufactured in a horizontal configuration at ONERA in 1997 [16]. The generation of two perpendicular forces which would require four jacks to be free of geometry and position modifications of the loaded specimen is obtained here more simply by the use of two servo-controlled hydraulic jacks. Mechanical guides, flexible along one direction only, ensure an effective uncoupling of the two perpendicular forces and the overall displacement, reduced by the composition of two mechanical translations, and do not introduce any parasitic moment within the specimen. The associated heating is carried out by induction with an additional susceptor. Due to the confinement of the heating system in the gage section, a temperature calibration was carried out to introduce a temperature map in the Finite Element calculation, in order to obtain the thermal compressive stresses induced by the thermal gradient and the geometry. One asset of the local thermal loading is to impose compressive thermal stresses even if the device is able to function with a positive force, as evoked by figures 3a and 3b. Another important point consists in evaluating the « ring effect » interpreted as the structural version of Poisson’s effect. That is why in a first time an elastic F.E. simulation was performed to

V.Bonnand et al.

748

choose correctly the loading in order to obtain locally the wanted state of stresses. More precisely, this evaluation allows us to impose actual uniaxial or biaxial stresses in the middle of the specimen by compensating the stress induced by Poisson’s effect.

a. Temperature (°C)

b. Compressive stresses thermally induced (MPa)

Figure 3 : Influence of the thermal gradient on cruciform specimen. The two fatigue devices from ONERA and CEAT combining axial loading and internal pressure loading have some similarities. The most important point is to take into account the “cap end effect”. That means it is necessary to apply an axial force in order to compensate the force resulting of the action of the internal pressure on the internal section of the specimen. For the two devices, this compensation is included in the servo-loop to aim to apply a pure circumferential tension. One can notice that the internal pressure device from ONERA requires a minimum value of 90 bars, and the CEAT device loading requires a loading ratio superior or equal to 0.3. Different methods are employed for crack detection to obtain as far as possible the number of cycles given for a crack length or crack depth close to 300 microns. The most common ones are the potential drop technique and the optical monitoring with a long distance microscope. Furthermore all the tested specimens are subjected to fractographic investigation by light microscope and SEM. Therefore the observed initiation zones have been compared with the prediction issued of Finite Element Calculations. Figures 4a and 4b show some examples of obtained cracks, depending on the loading configuration.

Experimental and theoretical comparison of some multiaxial fatigue design criteria

749

Equibiaxial test

F1=F2=9500 daN

Compensated unixial test

F1=9240daN / F2=3168daN

a. Biaxial tension tests on cross-like specimen (ONERA)

b. Tension – Internal Pressure test on tubular specimen (CEAT)

Figure 4 : Examples of observed cracks on Titanium alloy.

EXPERIMENTAL RESULTS AND FINITE ELEMENT ANALYSIS Table II shows an example of multiaxial test program for the titanium alloy at 200°C. A similar test program exists for the nickel-based alloy at 550°C. As shown on Figure 3, most of the loading conditions have been chosen to be proportional. Some additional tests, indicated in Table II, were performed under out-of-phase conditions. All these multiaxial test conditions have been calculated by various methods: - strength of materials rules - (thermo-)elastic Finite Element analysis - full cyclic inelastic Finite Element analysis The constitutive equations used involve a multi-kinematic hardening rule including thresholds [17] in order to correctly reproduce the cyclic curve, the stabilized stress-strain loop and the corresponding mean-stresses. Moreover the monotonic curve and the cyclic softening of the two materials are modeled by means of one or two isotropic hardening variables.

V.Bonnand et al.

750 Experimental Program Specimen

Partner

Biaxial Tension

Cruciform

ONERA

Tension / Torsion

Tubular

CEAT

Tension / Internal Pressure

Tubular

CEAT

Kind of loading and associated values

TYPE / Biaxiality Ratio (F1max/F2max)

Normalized Life Duration

F1max/ F1min (daN) F2max/ F2min (daN) 9500 / 475 10500 / 525 11480 / 574 12510 8090 / 388 9098 / 426 10097 / 487 9660 / 484 11400 / 570 8400 /420 9240 / 462

9500 / 475 10500 / 525 11480 / 574 12510 8090 / 388 9098 / 426 10097 / 487 4830 / 242 5700 / 285 2880 / 144 3168 / 158

10080 / 504

3456 / 173

7800 / 390

0/0 Tmax/ Tmin (N.m) 65 / -65 72 / -72 80 / -80 120 / 10 53,2 / -53,2 59,3 / -59,3 60 / -60 63 / -63 67,1 / -67,1 70 / -70

Fmax/ Fmin (daN) 2020 / -2020 2310 / -2310 1542 / -1542 1711,7 / -1711,7 1736 / -1736 1817 / -1817 1937,5 / -1937,5 2018,5 / -2018,5 F (daN)

IP (bars)

3550 / 1065 3800 / 1064 3820 / 1146 4000 / 1000

725 / 217 740 / 222 760 / 228 800 / 240 691 / 207 740 / 208 745 / 223 780 / 195

Equibiaxial (in-phase) Rb=1 Equibiaxial (out-of-phase) Rb=1 Biaxial (Rb=0,5) Compensated tension

91,58 39,44 22,13 13,40 53,36 35,22 10,23 20,33 12,34 180,12 26,49

(Rb=0,29) Uniaxial tension

19,06 52,71

Alternated tension

5,12 1,00 164,21 9,94 4,95 1,36 1,04 2,13 5,18 4,50 3,07 1,42

Alternated torsion Repeated torsion Tension/torsion (in-phase) Tension / torsion (out of phase)

Internal Pressure

Tension / Internal pressure

3,19 34,30 19,83 6,74 47,39 5,11 11,89 7,05

Table II : Multiaxial fatigue tests (Titanium alloy Ti-6Al-4V- T=200°C).

Experimental and theoretical comparison of some multiaxial fatigue design criteria

751

As expected, for tubular specimen, the simplified elastic rules deliver already correct stress and strain results, not significantly modified by plasticity. Nevertheless, all the tests are calculated to describe precisely the stabilized cycle. Figure 5 shows an example of deformed geometry resulting of an internal pressure loading after 500 cycles.

Figure 5 : Von Mises equivalent stress for an internal pressure loading with a high magnification on displacement, the maximal value is located at the internal radius (Nickel-based alloy).

stress

For the compact cruciform specimens, the full cyclic inelastic analysis was absolutely necessary to obtain the local multiaxial cyclic stress-strain responses and their stabilized mean-stress level in critical regions. F.E. simulations have shown that the minimum number of cycles to calculate is 100. One can remark the local loading ratio observed, from -0.25 to -0.1.

strain

a. Von Mises equivalent stress (equibiaxial loading)

b. σ11-ε11 loops in the critical region (100 cycles)

Figure 6 : F.E. calculation of Equibiaxial loading on cruciform specimen (Titanium alloy)

V.Bonnand et al.

752

MULTIAXIAL FATIGUE MODELS The models evaluated in the present work can be classified in two categories: equivalent effective stress models and critical plane models. Models based on an effective stress amplitude Effective stress models are essentially extensions of static yield criteria, such as the Von Mises criterion. The multiaxial stress amplitude is then reduced to an effective uniaxial cyclic value. Mean stress effects are traditionally described in uniaxial conditions by Goodman, Gerber or Soderberg relations. For multiaxial loadings, the hydrostatic stress is often introduced to take into account these mean stress effects. This is the case for Sines and Crossland criteria. The three effective stress models, Sines [10], Crossland [11] and Goncalves [15] considered in this study are respectively recalled in the equations below. Sines criterion:

σ aeff = σ aeq + bσ D 0 (Trσ )mean (1)

σ aeq + α (Trσ )mean = β

where σ aeq is the octahedral shear stress amplitude, (Trσ )moy the mean value of the first stress invariant during the cycle. In this expression. α and β are material parameters that will be identified from uniaxial tests with mean stress effects. The two given expression are equivalent. On the right, the effective stress amplitude involves parameters b and σ D 0 , where σ D 0 represents the fatigue limit for reversed cycle at 107 cycles. This effective stress amplitude is independently related to the fatigue life (under periodic conditions). Crossland criterion:

σ a eq + α (Trσ )max = β

(

σ a eff = σ a eq + bσ D 0 (Trσ )max − σ a eq

Gonçalvès criterion :

σ a eq + α σ p max = β

(

σ aeff = σ aeq + bσ D 0 σ p max − σ aeq

)

)

(2)

(3)

It is easy to check that these three criteria can be written in the same format:

(

σ aeff = σ aeq 1 + b σ D0 t eq

)

(4)

where the multiaxiality factor t eq takes one of the three following forms associated respectively with Sines, Crossland and Gonçalvès criteria :

Experimental and theoretical comparison of some multiaxial fatigue design criteria

t eq = t F =

(Trσ )mean σ aeq

t eq = s F − 1 =

σp

t eq =

,

max

(Trσ )max − σ aeq σ aeq

− σ aeq

753

,

(5)

σ aeq

where R is the uniaxial stress ratio. t eq in Eqn. 4 may be replaced by the t eq − 1 . multiaxial stress ratio defined by Rmult = t eq + 1 Moreover, a specific combined model is proposed in the present work that allows, as it will be shown later, to obtain better predictions both for pure shear and equibiaxial loading conditions. The proposed criterion is written as defined in Eqn. 4 but with the following expression for the multiaxiality factor t eq : s t t eq = ξ F F + ( 1 − ξ ) (s F − 1) 1 + tF

(6)

where ξ is a material parameter. This new model gives results similar to the Sines one when t F tends to zero and similar to a model with t eq = s F when t F takes high values. For uniaxial loadings, we have for this four criteria :

t eq =

1+ R

(7)

1− R

These criteria are then identical. In the present work they have been correlated to the lifetime by using a simple Basquin type function involving the fatigue limit as follows: −β

σ aeff = BN f

+ σ D0

(8)

where B and β are material parameters. The identification of these criteria on uniaxial fatigue tests, more specifically the mean stress effects, is explained later on.

Critical Plane models Critical plane models were mainly developed on the basis of phenomenological observations of fatigue crack development. It is generally recognised that fatigue

V.Bonnand et al.

754

cracks often nucleate and propagate on critical planes. The models assume then that crack nucleation on a particular plane is a function of the normal and /or shear stresses and strains on that plane. The various proposed formulae are different, but the process to follow is merely the same. One must firstly, found the critical plane and secondly check if the criterion is satisfied on this plane. Critical plane approaches have received a great attention these last few years. Reasonable correlations of multiaxial fatigue data have been obtained using critical plane approaches (Findley [18], Mcdiarmid [19], Kandil [20]). One of the first such approach is the Dang-Van criterion, much popular in France [21]. Three other popular critical plane models are evaluated in this paper, the critical plane approach of Brown and Miller [12], Fatemi-Socie [13 ] and Smith-Watson-Topper [14]. Dang Van G Proposed in 1972, it combines the current shear stress amplitude on any plane n with the current value of the hydrostatic pressure : G G Max nG Maxt τ ( t ) − τ 0 + α Trσ ( t ) = β ( N f ) (9)

[

]

α and β are adjustable material parameters, where last one is depending on the considered life. Because of his general definition, and the consistency between the criterion and the critical plane selection, it can be used under complex loadings conditions encountered in fatigue life analysis of components. Brown and Miller The concept of critical planes was first proposed by Brown and Miller. The critical G plane is the plane, oriented as n * on which the shear strain amplitude γ a reaches its maximum value. The criterion is expressed as follows:

γ a + α ε nmax = β ( N f )

(10)

where γ a is the shear strain amplitude on the critical plane, ε amax is the maximum value of the normal strain acting on the critical plane. Smith-Watson-Topper The Smith-Watson-Topper model is a quite popular uniaxial one, involving the product of the strain amplitude by the maximum stress in the cycle. A similar critical plane interpretation is considered in the present work., where the critical plane is defined as the plane experiencing the maximum value of the overall G G G G criterion, i.e. the maximum value of W ( n ) = γ a ( n ) σ nmax ( n ) when varying n . Then, the criterion on the critical plane is defined by:

γ a σ nmax = β ( N f )

(11)

Experimental and theoretical comparison of some multiaxial fatigue design criteria

755

σ nmax is the maximum value of the normal stress on the critical plane. In this criterion there is no parameter for the adjustment to mean-stress effects. Fatemi and Socie To better predict multiaxial fatigue life under both in phase and out-phase loading conditions, Fatemi and Socie proposed a modification to Brown and Miller’s critical plane approach. The critical plane is still the one where the shear strain amplitude reaches its maximum, but the model uses the shear strain amplitude and the maximum normal stress on this critical plane as follows: ⎛

γ amax ⎜⎜ 1 + kσ n

max

⎞

( )

⎟ σ y ⎟⎠ = f N f

(12) ⎝ where k is a constant that can be identified by fitting the uniaxial data against the pure torsion data. The maximum normal stress during the loading cycle is normalised by the yield strength σ y to maintain the dimensionless feature of strain.

LIFE PREDICTIONS The presented multiaxial models were evaluated based on their ability to correlate both uniaxial data and multiaxial data. All the models have been identified on the same uniaxial data base, namely strain and stress fatigue test at several loading ratio. Identification of mean stress effects on unixial tests As mentioned previously, the experimental program includes also LCF (strain or stress control) and HCF tests (stress control). A particular attention has been addressed on the identification of the mean stress effects. Figure 7 represents Haigh’s diagram obtained on the titanium based alloy (TA6V) at two different temperatures. Room temperature results well correlate literature data, showing for this material an extremely high dependency on the mean stress, but its saturation for large mean stresses. The material parameter b that appears in Eqn.4 corresponds then to the slope of the straight line that fits to the best the experimental data in the region −1 < R < 0.45 . However, in order to reproduce correctly the experimental tendency above, a bilinear fitting is introduced. For high stress ratio or for high multiaxility factor

t eq , that means t eq > t* , the expression given in Eqn.4 is replaced by

(

σ aeff = λ* σ aeq 1 + b* σ D0 t eq

)

(13)

V.Bonnand et al.

756

where λ* = ( 1 + b σ D0 t * ) /( 1 + b* σ D0 t * ) . The limit value t * is defined as t * = ( 1 + R* ) /( 1 − R* ) where R* is the limit loading ratio, in uniaxial, for which a change of the slope of the straight line occurs in Haigh’s diagram. In what follow, at 200°C, the red bilinear line in figure 7 has been used. These mean stress effects have also been validated for LCF stress fatigue tests.

normalized stress amplitude

1,5

MILLN-60Hz MILLN-70Hz MLS 350 Hz Hamrick 350Hz Mill - 400 Hz TA6Vpq 10^7 axial TA6Vpq 10^7 radial TA6Vpq 10^7 200°C TA6Vpq 10^7 20°C

1,0

0,5

0,0 -0,5

0,0

0,5

1,0

1,5

2,0

2,5

3,0

normalized mean stress

Figure 7 : uniaxial Haigh’s diagram for TA6V at room temperature and 200°C (Rectangular red boxes indicate the overall uncertainty of results at 200°C) However, when considering strain control tests, the cyclic mean stress relaxation is an important phenomenon that has to be taken into account. Figure 8 shows the typical results observed on materials for strain controlled fatigue test at different strain ratio. During the cycles, there is a continuous relaxation of the mean stress until a non zero stabilised value which depends on the applied strain amplitude. In this paper, for simplicity reasons, the stabilised mean stress versus the applied strain amplitude is described by bi-linear relations as presented in Military Handbook [22]. Once these evolutions are described, we can derive the fatigue strain life relation as illustrated in the figure 9. Here, as mentioned earlier, it has been chosen to correlate the effective stress to the fatigue life as presented in Eqn. 8. We need then to derive the stress amplitude from the applied strain amplitude. It

Experimental and theoretical comparison of some multiaxial fatigue design criteria

757

is performed by using the cyclic hardening behaviour of the material, namely described by a power law type function Δσ

n ⎛ p ⎞ ⎟ . = K ⎜⎜ Δε 2 2 ⎟ ⎠ ⎝

In the figure 9, there are two regimes. We can observe the great influence of the non completely relaxed mean-stress in the HCF regime, quite well predicted by this simple model. Above a certain amplitude, all the lives tend to coincide independently of the strain ratio, due to the complete relaxation of the mean stress. Clearly, it means that the mean strain effect is in fact a mean stress effect. Moreover, in the range of 1000 -10000 cycles, the fatigue model shows an inflexion point which corresponds to the transition in the figure 8 for which we have a complete relaxation of the mean stress to zero. This can be perfectly explained by the fact that just before this transition, we have a mean stress that will reduce the lifetime and just after there is mean stress and consequently the lifetime can be higher.

Figure 8 : stabilized mean stress as a function of applied strain amplitude, for various strain ratio Rε on TA6V at 200°C, and fitting by a bilinear expression

758

V.Bonnand et al.

Figure 9 : strain range as a function of the observed lives for TA6Vpq at 200°C for various Rε and prediction by the simple model used in the present paper Using uniaxial strain and stress data, the mean stress effects are then identified and consequently the multiaxial mean stress effects considered in most criteria are also completely characterised. Multiaxial Evaluation The comparisons will be made on biaxial stress diagrams ( σ 3 = 0 ), for proportional loading conditions, very useful to compare various criteria on the same basis, all of them giving identical results for uniaxial conditions. In Figure 10 we have two examples, for reversed conditions ( R = −1 ) on the left and for an elevated stress ratio R = 0.3 . Here R denotes the applied stress ratio, not to be confused with the multiaxial one Rmult defined above. Let us note that in the last case, when σ 1 < 0 or σ 2 < 0 or both, the shown envelop corresponds to minimum values of stress components instead of maximum ones. The envelops are

Experimental and theoretical comparison of some multiaxial fatigue design criteria

759

plotted for equals fatigue lives of 10 5 cycles, consistently with experimental results 2

2

1

1

0 -3

-2

-1

0

1

2

0 -2

-1

0

1

2 -1

-1 combi linear Sines Crossland Goncalvès

combi linear

Sines

-2

Crossland

Goncalvès

-2

-3

Figure 10: multiaxial fatigue envelops for TA6Vpq at 200°C, for 3 classical criteria and the new proposed one; left : R = −1 ; right: R = 0.3

Provided the mean stress influence is extremely high, the above figure shows very large differences between the 4 first criteria, both in the pure shear regime (second bissectrice) and in the equibiaxial one for R = 0.3 . In this second case, we observe surprising shape changes for Sines and Crossland criteria in the biaxial regime. This is because, for high loading ratio and high biaxiality, the equivalent multiaxial stress ratio overpasses the value R* = 0.45 that corresponds to the slope change in Haigh’s diagram. From these comparisons, Crossland criterion must be rejected, as confirmed when plotting experimental results as in Figure 11. Dang-Van criterion must also be rejected, provided it is identical to Crossland ones in equibiaxial, and at the middle distance between Sines and Crossland in the pure shear regime. This is a clear an surprising result of the present study. Though less pronounced mean stress effects are observed for INCO 718, the same kind of results is qualitatively obtained. Normalized experimental results in Figure 11 are obtained by roughly interpolating the true experimental data (measured life as a function of loading levels) for a fatigue life of N *f = 10 5 (at crack initiation). As they are plotted in a normalized way, existing data at 250°C from Gomez [23] were also added. All results very clearly demonstrate that only Sines and the modified combination proposed in this work are able to correlate results in the shear regime, though only Gonçalvès and the proposed combination may be acceptable for repeated equibiaxial conditions.

V.Bonnand et al.

760 2

0,4

TA6Vpq 200°C 4.10^4 INCO 718 550°C 10^5 0,2

INCO 718 550°C 10^4

0 0

0,2

0,4

0,6

0,8

1

1,2

1

1,4

-0,2

-0,4

-0,6

-0,8

-1

-1

Sines Crossland Gonçalvès combi linear

0 T-IP Gomez 250°C 0 T-IP PRC CEAT 200°C Sines Crossland Gonçalvès combi linear csi=0.7 -1

1

2

Figure 11: multiaxial fatigue envelops for 3 classical criteria and the new proposed one; with interpolated results on INCO 718 and TA6Vpq. Left : R = −1 ; right: R = 0 .3 Other criteria based on critical plane approaches are presented on Figure. 12, both for the reversed conditions (on the left) and for R = 0.3 (on the right). These models do not show true improvements in what concerns the simultaneous description of the shear regime and the repeated equibiaxial regime. Fatemi-Socie is acceptable for the equibiaxial conditions (slightly non conservative) but totally inefficient for shear. Brown & Miller is acceptable for reversed shear but very non conservative for the repeated equibiaxial conditions. SWT model is conservative for reversed shear, but, as there is no material parameter for the mean stress effect, both uniaxial and equibiaxial conditions are non conservatively predicted. Even if readjusted on the uniaxial conditions, this criterion will greatly suffer of a too large mean shear stress effect in the shear regime (for the R = 0.3 case).

Figure 12: predicted multiaxial fatigue envelops for TA6Vpq at 200°C, with 3 critical plane approaches. Left : R = −1 ; right: R = 0.3

Experimental and theoretical comparison of some multiaxial fatigue design criteria

761

Figure 13 confirms previous results for the various configurations tested in the present work for TA6V alloys, including cruciform specimens. The corresponding criteria are adjusted for the average value of stress ratio R = −0.15 . On the left figure we observe the great conservatism of Sines criterion for equibiaxial conditions, and the slight non conservatism of Gonçalvès one, both for shear conditions and for équibiaxial ones on cruciform specimens. On the right, we observe the quite good compromise with the combined criterion proposed in the present paper, both for shear conditions and equibiaxial ones (where it is only with a limited conservatism). 2

2

1

1 T-IP Gomez 250°C R = 0.3 T-IP CEAT 200°C R=0.3 Cruciform 200°C R=-0.1 Sines R = -0.1 Gonçalvès R=-0.1 Sines R = 0.3 Gonçalvès 0R=0.3 -1

T.T. CEAT 200°C R=-1 0 Sines R = -1 Gonçalvès R = -1

-1

T-IP Gomez 250°C R = 0.3

T-IP CEAT 200°C R=0.3 Cruciform 200°C R=-0.1

combi linear R=-0.1 combi linear R=0.3

1

2 -1

T.T. CEAT 200°C R=-1 0 combi linear R 0 = -1

1

2

-1

Figure 13: predicted multiaxial fatigue envelops for TA6Vpq at 200°C, and comparisons with 3 kinds of loading conditions; left : Sines and Gonçalvès; right: new proposed criterion

DISCUSSION AND CONCLUDING REMARKS The experimental program performed on two very different alloys used in rotating parts of turboengines, especially in compressor and turbine discs, have shown a series of important results for the fatigue lifetime analysis of those critical components. They include both uniaxial stress controlled and strain controlled fatigue tests, that give precise information on the mean stress effects, together with multiaxial conditions (in-phase and out-of-phase) both for cyclic shear (reversed) and for equibiaxial repeated loadings. The consideration in this work of many of the available lifetime prediction models and multiaxial fatigue criteria allows the following main observations: - among all criteria evaluated, no one is able to correctly predict the observed multiaxiality effects for both pure shear and equibiaxial conditions. This is true for both the stress invariant based models and for the critical plane approaches.

V.Bonnand et al.

762

- this is mainly due to the imposed choice to simultaneously reproduce , with the same criteria, the mean stress effects as observed under uniaxial conditions; - the fact that some of the classical multiaxial criteria used in the literature are often considered as good ones is very surprising. Most probably it is because the corresponding works do not consider simultaneously multiaxiality effects and mean stress effects; - a new special combined form of a multiaxial criterion has been proposed in the present work, that allows to simultaneously respect the pure shear conditions (both reversed and repeated ones), the equibiaxial regimes and the mean stress effects for uniaxial and multiaxial conditions. - the experimental conditions of equibiaxial loadings have been obtained with two different procedures, tension-compression with internal pressure on tubular specimens and biaxial loads on cruciform specimens. The observations made against classical criteria are consistent with the two configurations, for two different loading ratio. Classical strain controlled tests have confirmed the known result of a persistent, non completely relaxed mean stress in the intermediate range of amplitude for repeated strain controls. The clear correlation with the effect on the corresponding lives underlines that this is a mean stress effect, instead of a mean strain effect. In the same experimental programs, out-of-phase loading conditions were performed experimentally. Not considered in the present paper, specific procedures were developed in order to account for the detrimental out-of-phase effects and for defining effective stress amplitude that take into account temperature evolution during the cyclic loading, and therefore to be able to predict lifetime of components under complex non isothermal loading conditions.

Acknowledgements: The financial support of DGAC (Délégation Générale à l’Aviation Civile) is gratefully acknowledged. The fruitful advice of Bernard Lalanne and the help of Serge Kruch have also been greatly appreciated.

REFERENCES [1] Garud YS. (1981), Multiaxial fatigue: a survey of the state of the art. J Test Evaluat; 9(3):165-78. [2] Brown MW, Miller KJ. (1982), Two decades of progress in the assessment of multiaxial low cycle fatigue life. In: Amzallag C, Leis B, Rappe P, editors. Low-cycle fatigue and life prediction, ASTM STP 770. Philadelphia: ASTM; p 482-99.

Experimental and theoretical comparison of some multiaxial fatigue design criteria

763

[3] You BR, Lee SB. (1996), A critical review on multiaxial fatigue assessments of metals. Int J Fatigue; 18(4):235-44 [4] Wang YY, Yao WX. (2004), Evaluation and comparison of several multiaxial fatigue criteria. Int J Fatigue; 26(1) :17-25 [5] Papadopoulos IV, Davoli P, Gorla C, Fillippini M, Bernasconi A.(1997), A comparative study of multiaxial high-cycle fatigue criteria for metals. Int J Fatigue; 19(3):219-35. [6] Macha E, Sonsino CM. (1999), Energy criteria of multiaxial fatigue failure. Fatigue Fract Eng Mater Struct; 22, 1053-70. [7] Y.S. Garud. (1981), A new approach to the evaluation of fatigue under multiaxial loadings. Trans.ASME JEMT, 103, 118-125. [8] Ellyin, F.(1974), A criterion for fatigue under multiaxial states of stress, in Mechanics Research Communications, vol 1, No. 4, 219-224. [9] Radakrishnan, V.M. (1980), An analysis of low cycle fatigue based on hysteresis energy. Fatigue Fract. Engng Mater. Struct. 3, 75-84. [10] Sines G. (1959), Behavior of Metals under Complex Static and Alternating Stresses, in Metal Fatigue, Sines G. and Waisman J.L. Editors, McGrawHill. [11] Crossland B. (1956), Effect of Large Hydrostatic Pressures on the Torsional Fatigue Strength of an Alloy Steel, Proceedings of the International conference on Fatigue of Metals, Institution of Mechanical Engineers, London. [12] Brown M.W. and Miller K.J. (1973), A theory for fatigue failure under multiaxial stress-strain conditions. Proc Inst Mech Engrs, 187, 745-55. [13] Fatemi A. and Socie D.F. (1988), A critical plane to multiaxial fatigue damage including out-of-phase loading, Fatigue Fract Eng Mater Struct, 11(3), 149-165. [14] Smith K.N., Watson P. and Topper T.H. (1970), A stress-strain function for the fatigue of metals, Journal of Materials Science, 1970, 5(4), 767-776. [15] C. A. Gonçalvès C.A., Araujo J.A., Mamiya E. N., Multiaxial fatigue : a stress based criterion for hard metals, Int. J. of Fatigue, Vol. 27, pp. 177-187, 2005. [16] Chaudonneret M., Gilles P., Labourdette R.,Policella H. (1997) Machine d'essais de traction biaxiale pour essais statiques et dynamiques, La Recherche Aérospatiale., No 1977-5, pp. 299-305. [17] Chaboche, J.L., Jung, O. (1998), Application of a kinematic hardening viscoplasticity model with thresholds to the residual stress relaxation, Int. J. Plasticity, Vol.13, no 10, pp.785-807.

764

V.Bonnand et al.

[18] Findley W.N. (1957) Fatigue of metals under combinations of Stresses, Transactions of the American Society of Mechanical Engineers, vol 79. [19] McDiarmid D.L.(1991), A general criterion for high cycle multiaxial fatigue failure, Fatigue Fract Engng Mater Struct, 14, 429-53. [20] Kandil F.A., Brown M.W., Miller K.J. (1982), Biaxial low cycle fatigue fracture of 316 stainless steel at elevated temperatures. The metal Society, London, 280, 203-210. [21] Dan Van K. (1973), Sur la résistance à la fatigue des métaux, Thèse de Doctorat ès Sciences , Sci.Techniq.l’Armement, 47, 647. [22] Military Handbook, (1998), MIL-HDBK-5H, December 1998. [23] Gomez V. (2001), Etude en fatigue biaxiale à haute température d’alliages métalliques pour disques de turbomachines aéronautiques, Thèse de Doctorat, Univ. Paul Sabatier.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

A CONTRIBUTION OF ENVIRONMENTAL INVESTIGATIONS FOR GLARE RIVETED JOINT SIZING Thomas Beumler1,2, Bob Borgonje1,2, Jos Sinke2 1

2

Airbus Deutschland GmbH Delft University of Technology

Abstract: An extensive experimental research program has been initiated by Delft University in the year 2000 for potential influences of environmental conditions on Standard-GLARE. The investigation covers the whole range of the classical test pyramid from elementary-, via material-, coupon- and component level up to full scale size. The main focus of the exercise is the investigation of potential influences of both, variable temperature and/or environmental exposure on the structural strength justification for static and fatigue properties. This outdoor exposure program is performed in cooperation with Airbus, who contributes to several investigations, e.g the variable temperature tests and the full scale tests are performed at Airbus. Two structural details which have been part of Airbus full scale tests have been selected for the environmental exposure, i.e. a circumferential joint with a GLARE2 butt strap which has been justified in the A340-600 center section test article and a riveted repair on the full scale research test article “Megaliner Barrel”, made of GLARE4A (see figure 1). Delft University shipped about 200 GLARE specimens of different sizes and shapes to Innisfails in Queensland, Australia, for environmental exposure. Specimens have been returned after 1, 2 and 6 years, respectively, in order to perform mechanical tests and to compare them with both the as-received specimen and artificially aged specimen. A particular question which is dealt with is, whether the selected artificial ageing process (3000h exposure in 85%/70°C humid air) is representative for the M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 765–788. © Springer Science+Business Media B.V. 2009

766

Thomas Beumler, Bob Borgonje, Jos Sinke

simulation of natural ageing – and if so, for which kind of specimens or structures this might be the case. The specimens which have been exposed for 6 years in Australia returned in the year 2008. They have been inspected and mechanically tested, i.e. fatigue crack initiation curves for riveted joints in the as-received condition and after 6 years outdoor exposure are now available. However, the inspection procedure, the test procedure and the evaluation of results is often time consuming, especially if micro-fractografic analysis are involved. Therefore not all mechanical test results belonging to this project have been evaluated yet, but some are available for this paper. Selected items will be presented in this paper, which support the strength analysis of the above mentioned GLARE4A repair with special regard to residual strength.

Figure 1, riveted repair on full scale test “Megaliner Barrel” upper fuselage panel, GLARE4A on GLARE4A

FML JOINT SIZING PRINCIPLES AND INVESTIGATED STRUCTURE Riveted joints in Fiber Metal Laminates (FML) are usually designed for bearing (static strength) if the Design Service Goal (DSG) of the structure is low, or for fatigue if the DSG is high. To be considered for fatigue is the second load path character of the material, i.e. the design principle that the fibers are not allowed to fail under fatigue conditions, even if fatigue cracks are present in particular aluminium layers of the laminate. The riveted joint strength justification has been extensively discussed in ref. [1] and also presented in ref. [2,3]. For fatigue related

A contribution of environmental investigations for Glare riveted joint sizing

767

strength, potential influences on crack initiation, fatigue crack propagation and residual strength have to be investigated. Different than for monolithic aluminium, environmental influences might not just cause strength degradation due to corrosion effects, but potentially also due to moisture absorption in the epoxy of the fiber prepreg. Fortunately the outer aluminium layers of FML’s prevent any diffusion in the prepreg in the undisturbed structure, but the locations where holes are drilled for fastener installation as well as high loaded sheet edges, e.g. at aircraft fuselage door corner cutouts, might be exposed if the surface treatment allows access of moisture to diffuse. Any ageing experiment is a compromise compared with true operational exposure conditions of an aircraft component. Because the glass fiber prepregs do not contribute much to carry loads below yield strength, e.g. on fatigue load levels, it is unlikely to expect other than corrosion related influences due to environmental exposure (note that the transition between crack initiation and crack propagation is defined at 1mm crack length for FML’s). However, once fatigue cracks are propagating, they are accompanied by the very valuable fatigue delamination zone between aluminium and prepreg. It might be considered that, if cracks initiate for example after 0.5 DSG, the prepreg around the hole is already aged due to moisture diffusion. It is then the question which is faster, the moisture diffusion rate or the fatigue crack, and both may influence each other. Especially when the crack opens, much more prepreg might be directly exposed to the environment. This problem can not be solved by any accelerated test and even not by outdoor exposure tests, because specimens would require continuous mechanical loading during the outdoor exposure. The compromises for the here discussed outdoor exposure program are, that some specimens have not been fatigued before exposure and they are available for static, crack initiation-, crack propagation- and residual strength tests after returning from the outdoor exposure site. For these specimens, diffusion in the prepregs is limited through the non-fatigue holes. Other specimens have been fatigued before exposure. After returning from the test site they have been available for both crack propagation and residual strength tests. Moisture and even rain water could attack the material also through the fatigue cracks. Another compromise has been accepted for the outdoor exposure specimens. Because diffusion rates increase with temperature, it was necessary to paint the outdoor exposure specimens with dark paint as a conservative approach. Selected was the dark blue paint used by U.S. Airways at the time. However, with perfectly sealed riveted structures it might have been the case that a lot of investigations would be performed without any valuable information. In order to cover the worst case, it was therefore decided to paint the specimens before fastener installation and to install the fastener without anti fretting compound. It was considered that moisture and rain would get access to the prepregs through the filled holes under these conditions.

Thomas Beumler, Bob Borgonje, Jos Sinke

768

Figure 1 shows the Standard-GLARE4A-5/4-.4ssc repair patch, riveted with 5.6mm ASNA 2026 fasteners and ASNA 2528 collars to the fuselage skin made of exactly the same material. The main loading direction is the fuselage length direction, 4 rivet rows are available to carry the load. Fasteners have been installed with clearance fit in all specimens. Details of this joint are discussed in the following.

WEIGHT GAIN RESULTS Specific specimens have been developed for weight gain measurements in order to gain information about the amount of moisture which diffuses through the drilled holes into the prepreg. The so-called Moisture Reference Specimens (MRS) contained 20 prepreg layers sandwiched between 2 corrosion resistant aluminium layers. Specimens have been manufactured without holes (figure 2), with open holes (figure 3) and with aluminium Lok-Bolts installed in the holes. Most of the specimens received the dark blue paint, similar as all mechanical specimens. All MRS have been shielded by an epoxy glass cover in order to avoid their direct exposure to rain. An exposure of these specimens to rain would have beared the risk to transport dirt into the holes, which would influence the weight gain measurements. Notice that all specimens due for mechanical testing have been exposed in the open field, i.e. they got wet when it rained. The first observation was a high sensitivity of the paint system to the ambient conditions. The lower two curves in figure 4 show the weight gain behaviour of the two specimens without hole, i.e. the weight gain of the epoxy based paint system. Details of this discovery are not further discussed here but need to be taken into account for the calculation of the weight gain through the holes in the fiber prepreg.

Figure 2, MRS without hole

Figure 3, MRS with open holes

A contribution of environmental investigations for Glare riveted joint sizing

769

The upper curves in figure 4 show the weight gain in [g] of two painted open hole specimens. In order to calculate the weight gain of the fiber prepreg obviously the paint effect on weight has to be subtracted from the total effect. The result of this calculation is shown in figure 5, together with 2 additional open hole specimens which have been exposed with a bond primer surface, only (lower 2 curves).

0,12 0,10 weight gain [g]

0,08 open hole

0,06 0,04 0,02 0,00 -0,02 -0,04 -0,06

no hole

-0,08 0

12

24 36 48 exposure time [month]

60

72

Figure 4, MRS weight gain in [g] For comparison the weight gain curve obtained from a specimen which was exposed in a climate chamber is included in figure 5. Note that figure 5 contains the weight changes in the unit [% prepreg]. The often applied accelerated ageing time of 3000 hours in 85%RH/70°C humid air is obviously quite conservative compared with the weight gain obtained from outdoor exposure. The moisture distribution in the prepregs, however, is expected to be quite different because of the different diffusion rates (depending on temperature) and moisture contents of the environment. Estimations about moisture concentrations around the hole can be found in reference [1]. Interesting enough, the specimen weights did not increase continuously during the outdoor exposure, instead it followed the wet and the dry seasons in Australia. 6 years exposure has been defined in order to extrapolate expected weight gain curves

Thomas Beumler, Bob Borgonje, Jos Sinke

770

weight gain [% prepreg]

0,70 0,60

85%RH, 70°C, accelerated 0,50 0,40

3000 h

0,30

outdoor, open hole, corrected

0,20 0,10 0,00 -0,10 0

12

24 36 48 exposure time [month]

60

72

Figure 5, comparison outdoor / accelerated to the maximum DSG of an aircraft in [years]. In fact this is not necessary and the maximum moisture content measured during the 6 years period might be used as reference for future investigations. Other experiments performed within this TU Delft / Airbus project have shown that the maximum moisture content can be most realistically represented by 9500 hours ageing at 30°C in 70% RH humid air.

τ-γ RESULTS Some elementary tests have been conducted in order to investigate any potential influence of humidity on the shear strength between the aluminium and the glass prepreg. For example the relevance of this investigation is of importance for the fatigue crack propagation case, when high shear forces are acting locally between the cracked aluminium foils and the fiber prepregs, which causes a controlled fatigue delamination growth. A GLARE2-2/1 lap shear specimen was developed with an overlap size equal to one rivet pitch (of the above mentioned circumferential joint) in both directions. Each 2024 sheet had a thickness of 8mm. A hole was drilled in the center in order to allow diffusion into the prepreg, but filled with a fastener during exposure.

A contribution of environmental investigations for Glare riveted joint sizing

771

Figure 6, sketch of tau-gamma specimen from the side

FEM calculations have shown 2 locations with high stress concentrations, the edge of the hole but also the edges of the particular sheets. Both locations are on the surface where moisture can easily diffuse. However, the edges have been painted, which should prevent the diffusion into the prepreg at the edge location for some time.

Figure 7, photo tau-gamma specimen Fasteners have been removed after ageing and the specimens have been pulled to failure. All loads had to be transferred by shear through the prepreg in the center of the specimen between the saw cuts. The average failure load of the artificially aged specimens is 7% below the average value of the dry specimens (as-received), see figure 8. The average failure load after 6 years out-door exposure is 1% above the dry value, however, the scatter after 6 years exposure is higher. It might be concluded that 6 years outdoor exposure has no significant detrimental influence on those failure modes, which are shear dependent between the aluminium and the fiber prepreg. 3000 hours artificial ageing leads to a higher strength reduction than outdoor exposure. It might be usefull if minimum shear values shall be derived.

772

Thomas Beumler, Bob Borgonje, Jos Sinke

Figure 8, shear test results tau-gamma

RIVET STRENGTH Rivet strength tests according to NASM 1312-4 are conducted in order to determine the allowable tensile yield strength and the allowable tensile failure strength of a fastener, in combination with the material in which the fastener is installed. Details of the test procedure are available in reference [4]. For the support of the repair joint analysis, the rivet strength is evaluated for the combination GLARE4A-5/4-.4ssc with 5.6mm ASNA 2026 countersunk fastener. The final failure is in any case rivet pull through of the countersunk head, as indicated in figure 9.

Figure 9, rivet strength specimen made of GLARE4A

A contribution of environmental investigations for Glare riveted joint sizing

773

Five specimens for each exposure condition have been available. All specimens have been tested without fatigue damages, because rivet strength is a pure static sizing property. Some tensile tests with exposed specimens have been performed at 70°C. The results are contained in figure 10 although not relevant for sizing, since the position (and hence the load and the design) of the to be justified riveted repair is the upper shell of the rear fuselage, where sizing loads do not occur at high temperatures. It is common rule to size for static tension either against the failure load or against 1.5 x yield load, whatever is lower. For that reason figure 10 contains these two values in the unit [load per fastener]. 16000

load per fastener [kN]

14000 12000 10000 8000 6000 4000 1,5x

2000 0

fail yield

1,5x

fail yield

fail

1,5x yield

as received 3000h acc. 2 years RT RT RT

fail

1,5x yield

6 years RT

fail 1,5x

yield

fail 1,5x

2 years 70°C

6 years 70°C

yield

Figure 10, rivet strength, ASNA 2026 fastener in GLARE4A

Obviously the yield load is sizing the structure in any case and there is always no scatter on the room temperature mean values. This result can be expected knowing that the aluminium foils provide the yield strength and that the local shear properties due to local bending are obviously not effected by exposure (see section tau-gamma tests). One specimen belonging to the 6 years exposure / room temperature (RT) yield group provided the lowest value, 9178 N, which is 14% below the average. The application of rivet strength mean values is common sense, but no particular value should drop more than 10% below the mean value. Obviously this is the case for one particular specimen. Its failure load (precisely: 1.5 x yield load) of 9178 N is therefore used for the further strength justification.

774

Thomas Beumler, Bob Borgonje, Jos Sinke

BLUNT NOTCH STRENGTH Specimens with a row of holes are frequently used for multiple FML investigations. Without fatigue they provide information about the failure strength of open or filled holes (blunt notch, BN), for fatigue they can be used for the generation of SN-data, da/dN-data and residual strength. The three latter properties include some scatter related questions which can be answered by the row-of-hole specimens as well. Consider that a FML specimen with 5 holes in a row nicely fits in any common test machine and offers 20 independent crack initiation locations. The outdoor exposure project encountered several times problems to load the GLARE2 specimens, which belong to the project but which are not discussed in this article, to failure. Due to the high failure strength of the GLARE2 material speci-mens often fail in the suspension/clamping area, as they did in frame of the outdoor exposure project. BN specimens have been fatigued with the original width and 5 bore holes followed by ageing treatments for some of them. After exposure and prior to test-to-failure, all specimens, GLARE2 and GLARE4, have been quickly milled to a dogbone shape. A specimen is shown in figure 11 after failure.

Figure 11, dogbone specimen for blunt notch and residual strength testing

Specimen 3-A-22 belongs to the series of specimens which was located 6 years in Australia. It should be noticed that an inspection of 6 year exposure specimens shows corrosion on the aluminium collar and also in some of the GLARE holes. The non-fatigued specimens have been pulled to failure and provide simple information. All specimens obtain almost an identical failure load and all values are above the material B-value (526 MPa), see figure 12. Note that three as-

A contribution of environmental investigations for Glare riveted joint sizing

775

received specimens have been available but just one specimen for each of the exposure types. Consider also that BN-stresses are always net section stresses.

Figure 12, filled hole blunt notch results for GLARE4A Residual strength is obtained from the specimens which have been fatigued to different fatigue damage rates RD prior to exposure.

RD =

Aalu min iumwithfatiguedamage Aalu min iumwithoutfatiguedamage

The fatigue crack lengths, which are required for the calculation of the fatigue damage rates, have been measured on the surface aluminium layers but calculated for the center layers. After crack length measurements and prior to exposure, bolts have been installed in the holes, with clearance fit and without anti fretting compound. Figure 13 contains the residual strength results after the various ageing procedures. Results up to 50% damage rate are located in a very narrow band. However, if linear regressions are performed through all data which belong to the same ageing treatment, different slopes are found. The lowest reduction of strength with increasing fatigue damages is found for the dry specimen, followed by 2 year outdoor exposure, then 3000h artificial ageing and finally the 6 year outdoor exposure.

Thomas Beumler, Bob Borgonje, Jos Sinke

776

Residual strength [MPa]

600 500 400 300 as received

200

3000h 70dC / 85% RH 2 years outdoor

100

6 years outdoor

0 0

20

40

60

80

100

RD [% ]

Figure 13, residual strength of filled hole specimens, net stresses It can be concluded that 6 years of exposure have a more damaging environmental effect for high fatigue damage rates than the 3000 hour artificial ageing. The physical reason for this behaviour is not yet identified. A speculation is the presence of water due to rain in the outdoor exposure specimens, which gets a better access to the prepreg / metal interface with increasing crack length. Previous researchers discovered the quite linear relationship between fatigue damage rates and residual strength decrease. The investigation was pioneered by Richard Müller, who provided a formula for the residual strength prediction with fatigue damages which is linked to the Metal Volume Fraction (MVF) approach [5]:

σ Re sidual , BN = σ BN ,Glare,min − σ BN , Al × MVF × RD with:

σ BN ,Glare ,min = 526 MPa

MVF = 0.571

The Müller equation was developed for riveted joints and supposed to be applicable for a row of holes as well. Morris assessed the Müller equation for a row of holes, based on additional test results. His equation accounts for the more than proportional drop of strength for short crack length [6]:

A contribution of environmental investigations for Glare riveted joint sizing

777

σ Re sidual , BN = (σ BN ,Glare − σ BN , Al × 0.5MVF × R D )× 0.77 with: σ BN ,Glare = 579 MPa MVF = 0.571 Both design curves are included in figure 14. The Müller curve matches neatly with most of the test results and follows the slope of the exposed specimens. The Morris curve matches with the slope of the as-received specimens and is more conservative compared with all results for damage rates up to 60%. The relevance for riveted joint sizing is discussed in the next paragraph.

Residual strength [MPa]

600 500 400 300

as received 3000h 70dC / 85% RH 2 years outdoor 6 years outdoor MVF min (Müller) MVF min (Morris)

200 100 0 0

20

40

RD [% ]

60

80

100

Figure 14, residual strength of filled hole specimens, net stresses

RIVETED JOINT STRENGTH Obviously the above discussed rivet strength specimens are mainly loaded by bearing, while the failure of the closed hole (BN) specimens is mainly dependent on the by-pass stress. A step upwards on the test pyramid, riveted joint coupons and rivet patch panels, which both copy the repair design of the Megaliner Barrel, have been manufactured and tested under various conditions. It must be considered that, different than for the blunt notch specimens, the riveted joints are subjected to bending, which leads to a stress distribution through the thickness and causes the mating aluminium layer of the laminate to fatigue first. Various specimens have

Thomas Beumler, Bob Borgonje, Jos Sinke

778

been investigated microfractografic in order to record the fatigue crack length of the various aluminium layers in the joint after fatigue. An example is shown in figure 16. It provides one fatigue crack pattern through the 5/4 lay-up at one fatigue sensitive location (each arrow indicates the transition from fatigue crack to static crack). Note that the mating aluminium layer is cracked completely (link-up).

Figure 15, riveted joint coupon specimen after failure

mating surface between sheets

countersunk

Figure 16, fatigue cracks through the thickness of GLARE4A-5/4-.4ssc top sheet of one fatigue sensitive location at riveted joint coupon specimen

If a number of specimens are inspected that detailed, the relation of the fatigue crack length in sub-surface aluminium layers depending on the fatigue crack length of the surface layers can be determined. This function allows to assess the fatigue damage rate for the entire cross section by calculation, if just all surface layer crack

A contribution of environmental investigations for Glare riveted joint sizing

779

lengths are measured. Now, also for the riveted joint a relation between failure strength and fatigue damage rate becomes available, see figure 17. 600

net stress [MPa]

500 400 300

coupons not exposed, RT coupons exposed 1 year tropic, RT coupons exposed 2 years tropic, RT coupons exposed 6 years tropic, RT RSmin (Müller) RSmin (Morris) detectable crack length

200 100 0 0

5

10

15

20 25 RD [% ]

30

35

40

45

50

Figure 17, residual strength of riveted joint coupons

Note that the stresses in figure 17 are related to the net section and that the X-axis extends to 50% fatigue damage rate. Proposed design curves obtained from both, Müller [5] and Morris [8], are included in figure 17 as well as the detectable crack length (8mm for the surface layer). Interesting enough, the slopes of the as-received specimens and of the 2 years outdoor exposed specimens are similar, while the slope of the residual strength curve obtained from specimens after 1 year outdoor exposure follows a different, slightly steeper slope. One specimen with 47% fatigue damage rate drops significantly below the design curves. It is noteworthy that all fatigue cracks in the highest loaded aluminium layer already linked up in this specimen, see also figure 16. Detailed investigations of the fatigue delamination zone and the fiber condition after ageing and residual strength are discussed in reference [1]. The next specimen on the test pyramide is the unstiffened panel according to figure 18. Comparing both, crack initiation and crack propagation of riveted joints in coupon-, panel- and full scale size requires consideration of two contradicting aspects:

Thomas Beumler, Bob Borgonje, Jos Sinke

780

• •

secondary bending decreases with specimen size → improving fatigue life test frequency decreases with specimen size → decreasing fatigue life

Strain gauges have been bonded at the edge of the single shear overlap at one of each involved specimens. The comparison of stresses on the outside and the inside of the skin just close to the repair patch is shown in figure 19. The secondary bending of the skin of the Megaliner Barrel, measured at load case 1g+Δp+10ft/s vertical gust is almost negligible. For the un-stiffened panel the secondary bending is in the order of 50% of the normal stress and for the coupon it is even higher. This information is of importance if stress analysis in FML are considered, because the crack length of the various aluminium layers through the thickness of the laminate differ considerably

Figure 18, unstiffened riveted repair panel

A contribution of environmental investigations for Glare riveted joint sizing

781

Figure 19, stresses E*ε in aluminium layer of repaired skin Figure 20 contains the relevant design parameter for tensile stress. Consider that the basic skin is designed with a cut-off level of 230 MPa limit load for Large Damage Capability and repair reasons. The ultimate stress level is obviously 345 MPa.

gross stress [MPa]

400

300

200 limit stress ultimate stress (large damge limit x 1.5) rivet strength min, RT RSmin (Morris, filled hole & joint coupons) panel 12-RR-01, not exposed, 70°C panels -02 & -03, 2 years exposed tropic, RT detectable crack length

100

0 0

5

10

15

20 25 RD [% ]

30

35

40

45

50

Figure 20, tensile design parameter obtained from tests, gross stress

782

Thomas Beumler, Bob Borgonje, Jos Sinke

Calculating a stress allowable based on the rivet strength result, i.e. with 9178 N per fastener, would lead for four rivet rows to a gross stress of 375 MPa which is obvi-ously non-conservative. The residual strength curve proposed by Morris, which covers both the flat filled hole specimen results (blunt notch) and the repair coupons, is crossing the Y-axis at 341, i.e. almost the ultimate stress level (calculated static reserve for the repair: 0.99). Under these conditions, the ultimate load capability would be left immediately after fatigue crack initiation. Three data points from panels tested under various conditions are included in figure 20 as well. As expected, the throught-the-thickness crack distribution is different in the panels compared with the coupons, which required fatigue crack measurements in the particular aluminium layers of the panel laminate as well. However, all strength data obtained from the panels are above the Morris-curve. Two data points are located exactly on each other, for panel 12-RR-01 which was fatigued, not exposed and pulled to failure at 70°C, and for panel 12-RR-02, which was fatigued, exposed for 2 years and finally subjected to residual strength (6 year results not yet evaluated). A significant number of different fatigue tests have been performed by Delft University and Airbus, which all contributed to the evaluation of a computer software for fatigue strength evaluation, the FML F&DT Toolbox. Riveted joints in GLARE are always calculated in a Multiple Site Damage (MSD) scenario because of the early fatigue crack initiation. For the here discussed riveted joint, a Toolbox prediction is performed with 150 MSD scenarios for the 12 fastener which are oriented in circumferential direction on the Megaliner Barrel. The calculation results are factorised and full scale test validated. Figure 21 contains the fatigue crack initi- ation distribution, which is linked to the lead crack of each of the 150 calculated scenarios, and the distribution of the detectibility, which is linked to the crack propagation of the various lead cracks.

A contribution of environmental investigations for Glare riveted joint sizing

783

Figure 21, crack initiation- and detectable crack length distribution

Crack initiation is expected with 50% probability after 16000 flights. Because this value is already factored and because, as explained above, the residual strength would drop below ultimate load for short crack length, the inspection threshold for this repair would equal 16000 flights. Figure 22 shows one of the 150 calculated MSD scenarios. Included is the crack propagation curve for the lead crack and the average crack propagation curve obtained from all cracks. The latter is directly linked to the residual strength curves. The minimum residual strength curve is recommended for strength justification in the range of ultimate load capability, when no inspections are necessary. For cases like discussed in this paper, i.e. a mandatory inspection task, the mean residual strength curve shall be applied.

Thomas Beumler, Bob Borgonje, Jos Sinke

784

mean. R.S. min. R.S. detectable crack length

lead crack

average crack

Figure 22, fatigue strength plot, scenario 1 from 150 According to the information provided by figure 21, the inspectable crack length is reached after 46000 flights (factored). The critical crack length, however, will be reached after more than 100000 flights. The calculated inspection interval would then be larger than (100000-46000) = 54000 flights. First inspection: value) Second inspection: value)

16000 flights after embodiment of repair (factored 70000 flights after embodiment of repair (factored

Check: After 70000 flights link-up of cracks in the highest loaded aluminium layer of the repaired skin is expected. For the reasons discussed in section “Riveted Joint Strength”, the second inspection should not be performed later than after 70000 flights. The Megaliner Barrel was representing a center section of a fuselage with a double deck cross section, not very different than later designed into the A380. Since the

A contribution of environmental investigations for Glare riveted joint sizing

785

present verion of the A380 is designed with GLARE panels just in the front and the rear section, this investigation is a theoretical one. However, the general principle of relatively early, controlled fatigue crack initiation and relatively slow fatigue crack propagation is shown in the analysis above. Taking the A380-DSG of 19500 flights into account, either no or just one inspection of the discussed repair would be required.

COMPRESSION FILLED HOLE RESULTS The presence of small fatigue cracks during a part of the aircraft life requires also local investigations under compression. A known failure mode under compression is the local buckling of the external aluminium layer of GLARE outboards, e.g. at the riveted attachment of skin and stringer. This failure mode happens at very high loading and is a secondary failure mode. The outer aluminium layer escapes by superseeding the local peel strength between 1st prepreg layer and external aluminium layer. This kind of failure mode can be found in compression panel tests and is not easy to generate with elementary specimen size. A similar, but not equal failure mode is the kinking of the entire laminate at a single hole specimen, see figure 23, right.

Figure 23, compression filled hole specimen Details of the test set-up and about specimen sizes can be obtained from reference [1]. Similar as for the blunt notch / residual strength investigation, specimens have been fatigued, exposed and loaded to failure after exposure. The results are summarised in figure 24. It turned out that also the compression filled hole residual strength can be calculated with the MVF approach.

Thomas Beumler, Bob Borgonje, Jos Sinke

786

CFH failure stress [MPa]

600 500 400 300 GLARE4A, as received, RT GLARE4A, 1 year outdoor, RT GLARE4A, 1 year + 550h/70°C/85%RH, RT GLARE4A, 2 yers outdoor, RT GLARE4A, 6 years outdoor, RT

200 100 0 0

20

40 60 RD [%]

80

100

Figure 24, compression filled hole residual strength, gross stresses The high tension loads which are considered to size the repair, are typical for the upper fuselage of a center/rear fuselage. Compression loads at this location usually do not exceed 50% of the tension load level. It is therefore obvious that compression filled hole is not the sizing criterion for the repair.

CONCLUDING REMARKS A fatigue strength analysis for a riveted repair on a GLARE4A fuselage skin with a GLARE4A repair patch has been performed under consideration of various outdoor exposure test results, which support the residual strength part. Not considered in these analysis are potential effects of corrosion and variable temperature on crack initiation and crack propagation. The presented results represent a fraction of the TU Delft / Airbus environmental research program. It must be considered that for economic reasons mechanical testing can be never superposed with real time exposing. Instead, as a compromise, a sequence of mechanical testing and specimen exposure must be performed. Weight gain measurements obtained from open hole specimens, which have been exposed under a plexy glass cover in the tropic part of Australia, did not show a continuous moisture absorption. It was expected that a continuous weight gain due to diffusion through the holes into the prepreg would allow to extrapolate a maximum moisture content at the end of the aircraft life (e.g. after 30 years). Instead the specimen weight followed the seasons of Australia, i.e. moisture

A contribution of environmental investigations for Glare riveted joint sizing

787

diffused in and out of the epoxy prepreg. The maximum weight gain is below 0.1% prepreg weight. This value can be obtained by 9500 hours artificial ageing in 70% relative humid air at 30°C. The 3000h accelerated ageing treatment in 85% RH / 70°C humid air, known from the composite world, leads to 0.32% prepreg weight increase and is very conservative if compared with the tropic climate weight gain. Consequently, concerning the moisture content in the prepreg, all mechanical tests can be considered as worst case, if the tropic climate is representative for the structure which is analysed. However, the mechanical specimens have been further exposed due to the rain water, which could access the holes. Elementary shear tests with the FM94/S-2 glass prepreg have been performed, using modified tau-gamma specimens. In the average, all outdoor exposed strength data exceed the dry test results, but the shear strength after artificial ageing drops by 7% percent. However, the average value obtained from the artificially aged specimens almost equals the minimum value of the outdoor exposed population. The test results do not indicate any noteworthy strength reduction due to ageing of the prepreg under in-plane shear condition. Static rivet strength tests according to NASM 1312 determine both the ultimate fastener strength and the yield strength of the investigated sheet/fastener combination. Because GLARE4 has a low Metal Volume Fraction, the ultimate failure mode is rivet-pull-through and the yield load is relatively low. Hence the sizing criterion is 1.5*yield and not ultimate strength. With the exception of one specimen all results at room temperature for the dry condition, after 3000h artificial ageing and after outdoor exposure are similar. It does not appear to be necessary to perform artificial ageing before conducting NASM 1312 tests with Fiber Metal Laminates in the future. Flat filled hole specimens are used for both, static tensile tests without fatigue damage (BN) and residual strength tests with fatigue damages. The blunt notch strength is similar for all ageing conditions and all values are above the B-value. Artificial ageing does not seem to be required for this test. The residual strength tests indicate a tendency to deeper slopes for higher fatigue damage rates after ageing, compared with the non-aged specimens. The artificial ageing (3000h) fits between 2 years outdoor exposure and 6 years outdoor exposure strength results. All test results fit very close to design curves proposed by Müller and Morris, respectively. Riveted joints, in coupon size and in the shape of un-stiffened panels, are subjected to secondary bending, which changes the fatigue damage distribution through the laminate thickness completely, compared with the filled hole specimens. By-pass loads and bearing loads are superposed. Except for one specimen, the residual strength test results after fatigue and ageing are close to the prediction curves of Müller and Morris, respectively. The outlaw specimen was fatigued beyond the status of a complete link-up of the fatigue cracks in the highest loaded aluminium

788

Thomas Beumler, Bob Borgonje, Jos Sinke

layer and after the rupture test it is impossible to determine, whether even fiber damage was present due to fatigue and before residual strength test. The detailed failure mode of this specimen is still under investigation, it is believed that the high additional bending of an un-stiffened coupon with one aluminium layer of the laminate completely fatigued plays an important role. Compression filled hole strength is not effected by any ageing treatment. On elementary test scale this property follows the Metal Volume Fraction approach, if tested with fatigue cracks. A complete tensile F&DT analysis is performed for the present repair joint. The results are not influenced by the test results provided by the outdoor exposure residual strength analysis and they fit to the results obtained from an Airbus full scale test article

REFERENCES [1]

[2] [3] [4] [5]

[6] [7] [8]

Beumler, Th., Flying GLARE, a contribution to aircraft certification issues on strength properties in non-damaged and fatigue damaged GLARE structures, Delft University of Technology, PhD Thesis, 2004 Beumler, Th., Controlling the Damage with FML Structures, DTAS Conference, Delft 2007 Beumler et. al., Means of Compliance for A380 Hybrid Structures, ASIP Conference, San Antonio 2008 Fastener test methods, Method 4, Lap joint shear, MIL-STD-1312-4, Department of Defense, Washington D.C., 1984. Müller, G., An Experimental and Analytical Investigation on the Fatigue Behaviour of Fuselage Riveted Lap Joints, Delft University of Technology, PhD Thesis, 1995 Morris, G. and Rodrigo, Ph., GLARE Open/Filled Hole Residual Strength, Airbus France Report RP0506955, 2005 Schott, J., Werkstoffkundlicher Untersuchungsbericht Nr. 9697, Airbus Deutschland, 2003 Morris, G. and Rodrigo, Ph., GLARE Riveted Joint Residual Strength, Airbus France Report RP0504106, 2005

25th ICAF Symposium – Rotterdam, 27–29 May 2009

ADVANCES IN CRACK GROWTH MODELLING OF 3D AIRCRAFT STRUCTURES S.C.Mellings, J.M.W.Baynham, R.A.Adey BEASY, Southampton, UK

Abstract: This work is focussed on the 3D simulation of crack growth in metal structures which are exposed to complex multiaxial loadings, as regularly occur in aircraft. Simulation of these cracks under the applied loading can be used by the design engineers to investigate changes in the design, loading or materials. Given an initial crack, the aims of the simulation are to identify the stress intensity factors occurring at any stage of the loading cycle, and to predict the time taken for the crack to grow through the structure. This means that vulnerability to fast crack growth is assessed, and that fatigue growth calculations of growth direction and distance are performed and accumulated. The result is that crack life predictions are made of the crack size variation with number of cycles. The use of multi-scale techniques allows sub-models to be created from any part of the model. This enables large-scale FE models to be used for a detailed BE analysis. The paper describes the crack growth in one such sub-model using a variety of initiated crack shapes and positions, and for a range of different loading regimes.

INTRODUCTION The analysis of cracks in airframe structures is a critical part of the design process. In this paper various crack growth analyses are presented in which a crack is grown within a single model, a machined stiffened panel. In the analyses various crack growth scenarios are presented.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 789–809. © Springer Science+Business Media B.V. 2009

790

S.C.Mellings, J.M.W.Baynham and R.A. Adey

Initially in this paper, details of the analysis methodology are presented, followed by descriptions of the model creation, from the initial FE model, and the automatic crack growth process. The paper then goes on to apply the described method to a number of crack growth scenarios for the given model. These examples present cracks initiated at different locations and under a variety of loading regimes.

BOUNDARY ELEMENT ANALYSIS The numerical analysis presented in this paper is based on the Boundary Element Method (BEM) rather than the ubiquitous Finite Element Method (FEM). The BEM has significant advantages for mathematical modelling of fracture specifically that only the surfaces of the part need to be meshed. The crack, as a surface of the model, is meshed as well, using a layer of elements on both the top and the bottom surfaces of the crack. Figure 7 shows a view from the inside of a structure, showing the elements on the crack surfaces and those on external surfaces. The quarter-penny-shaped crack has a roughly quarter-circular crack front. The simplicity of the mesh required to represent behaviour of cracks is a major advantage when using BEM for fracture analysis. The method allows very refined meshing near the crack front without any difficulty at all. By contrast FE meshes often suffer from the unwanted side-effect that the refinement tends to propagate through the volume of the structure. The simplicity provided by the BEM can especially be appreciated during the growth of a crack, since the only parts of the model that are affected are the surface mesh on the crack and the mesh on the immediately adjacent surfaces. In particular with BEM, it is not necessary to recreate the mesh of the entire structure. Additional advantages of the BEM are first that it is a mixed method which calculates both displacement components and stress components directly, and secondly that these values are calculated at positions on the surface of the structure. Some other methods derive stresses by differentiation of displacement, and often calculate results at “gauss points” inside the volume, thereafter extrapolating to the surface. It is well known that differentiation tends to dilute accuracy, and in pressure-vessels, the highest stresses are generally at the surfaces, with sometimes very rapid variation near the surface, which means that extrapolation to the surface can be very misleading. The end result is that BEM provides a methodology which fits very conveniently with fracture and crack growth simulation requirements, and makes it feasible to perform automatic fatigue crack growth modelling.

Advances in crack growth modelling of 3D aircraft structures

791

In this paper a specific form of the BEM is applied, involving use of dual elements on the crack. These allow the greatest simplicity of modelling, since the same element geometry can be used to represent the behaviour on both sides of the crack, hence the name “dual”. From a mathematical viewpoint, a node (at which problem variables are defined) on one side of the crack is at the same position as the node on the corresponding dual element on the other side of the crack. This situation would ordinarily prevent a solution from being obtained, but the dual BEM avoids this difficulty by using an influence equation (for the dual node) which is derived in a different way [1-6]. The examples shown in this paper were modelled using the BEASY simulation package, which additionally allows selection of an initial crack from a library, and automatic insertion of the crack into a BEM model of an un-cracked structure. The benefits of mathematical modelling are obtained when the method is applied to situations which are not covered either by standards, previous experience, reference solutions, or experiment.

CREATING BEM MODELS FROM FE GEOMETRY AND RESULTS The methodology presented above describes how the boundary element analysis is performed. However in most cases, the parts to be analysed have already been modelled with finite elements, rather than boundary elements. The multi-scale approach used involves extracting a sub-model from part of the larger scale FE model. In the case presented here, a finite element model of the part is available in ABAQUS, as shown in Figure 1. The part is a section of a curved panel that is stiffened by two ribs.

Figure 1 ABAQUS model of a stiffened panel

792

S.C.Mellings, J.M.W.Baynham and R.A. Adey

This part is modelled in ABAQUS using tetrahedral elements and 3 load cases are modelled. One load case generates a uniform tension load in the panel, aligned with the stiffener direction. The other two load cases generate non-uniform tension loading along the panel edge, again with the loading aligned with the stiffener axis. For the purposes of the crack growth analysis it is not necessary to analyse the complete stiffener model, therefore the BEM model that will be generated will be a sub-model of this part. The sub-model section used is shown in Figure 1.

Figure 2 Sub-model selection in ABAQUS This group of finite elements is used to generate the BEM model using a “skinning” process. In this process the selected elements are used to define the extent of the BEM sub-model. The external surfaces of this region are then used to define the BEM elements which are then loaded from the FE results. This process is controlled using a JAVA based tool, the BEASY model generation wizard, which extracts the required information from the ABAQUS result file and then creates and loads the required boundary element model. The resultant BEM mesh is shown in Figure 3.

Figure 3 Sub-model selection in ABAQUS

Advances in crack growth modelling of 3D aircraft structures

793

This tool allows for BEM sub-models to be created from finite element meshes, generated using a variety of FE tools and the loading on the sub-model can be generated from the realistic stress results computed in the FE analysis. This allows for complex stress fields to be represented without the requirement to manually interpolate the boundary conditions. In some cases, however, the FE mesh may not be suitably refined in the area to be investigated. For example, if the FE model is represented with linear elements, then facetted rather than true curved surfaces will be generated. An example of this is shown in Figure 4. On the left is a bolt that has been represented using linear elements in FE, this will give a very different local stress field to the quadratic representation shown on the right, although the applied loading to the model is the same. In addition, features may not be included in a more global model that will be required in a more detailed crack growth, such as fillets or inclusions in the model, all of which will affect the crack growth rate and direction.

Figure 4 Linear and quadratic representation of a bolt

If an FE model is not directly suitable for a detailed crack growth study then the model can be recreated directly as a BEASY model. However the loading from the global FE model can still be used and provides valuable data on the stress fields that are applied to the part. Additionally, residual stress fields may be applied to the part and these can be obtained by reading the FE stress results in the vicinity of the crack. These stresses are then automatically applied to the crack faces to provide loading either opening or closing the crack.

794

S.C.Mellings, J.M.W.Baynham and R.A. Adey

THE AUTOMATIC CRACK GROWTH PROCESS In this paper the automatic crack growth process uses a toolkit for fracture analysis in which a crack can be initiated into an existing, un-cracked, BEM mesh. In the examples presented here the mesh used was the BEM sub-model created from the larger scale FE model of the stiffened panel. An example of the initial mesh is shown in Figure 3 where the mesh of boundary elements has been created, but the crack has not yet been initiated in the part. The simulation process continues with selection of the required initial crack shape from a library of shapes. This is achieved within a GUI which leads the user through the various necessary tasks, such as choosing the size of the initial crack and its position and orientation within the component. Figure 5 shows an example of the appearance of the GUI during selection of a crack shape. Other necessary tasks include selection of a “growth distance” used during simulation of fatigue growth, and selection of appropriate fatigue growth data and a crack growth rate equation. Optionally, further settings can be adjusted to control mesh density, number of J-integral calculations to be performed along each crack front and so on. Having defined all the necessary data, the simulation process is started, the GUI can be closed, and the steps required for fatigue crack growth then proceed automatically. The very first step involves placement of the required initial crack into the mesh of the un-cracked component, and modification of the mesh on the crack and nearby surfaces of the component so that it is suitable for the fracture calculation. The mesh after completion of this step is as shown in Figure 6, with detail around the crack shown in Figure 7.

Advances in crack growth modelling of 3D aircraft structures

Figure 5 Appearance of the GUI during selection of initial crack shape from the crack library

Figure 6 Mesh of corner crack modelled into stiffener

795

796

S.C.Mellings, J.M.W.Baynham and R.A. Adey

Figure 7 Wireframe close-up view of the model with the initial crack In the next step of the process, the new mesh is first used to compute stresses and displacements, and then to determine the stress intensity factors, typically using a J-integral approach. Next, the fatigue behaviour is simulated, to “grow” the crack. This involves use of the stress intensity factors and load variation details to select appropriate direction and distance of growth. This calculation is made at multiple positions along the crack front, and a new position of the crack front, for the grown crack, is thus predicted. This crack growth allows for variable crack growth rates and directions along the crack front, due to the predicted stress intensity factor values computed in the analysis. In the next step, the predicted new crack front position is used together with the old crack, to define the surface of the new grown crack. To complete the cycle, the final step involves placement of the new grown crack into the mesh of the un-cracked component, and as before, modification of the mesh on the crack and nearby surfaces of the component so that it is suitable for the fracture calculation.

Advances in crack growth modelling of 3D aircraft structures

797

The process, shown diagrammatically in Figure 8, is continued until one of several stopping criteria is met. These criteria include for example occurrence of critical stress intensity factor, or the attainment of the required total growth distance.

Figure 8: Flowchart outlining the process of fatigue crack growth simulation

Using this process it is simple to change the crack that is initiated into the model. Figure 9 show the same model that can be re-used with different crack shapes, orientation, sizes or initiation positions. In this paper three different initial cracks, in the same base model, will be examined.

S.C.Mellings, J.M.W.Baynham and R.A. Adey

798

Example 2 Example 1

Example 3

Figure 9 Crack growth studies 1-3 using the BEM sub-model.

EXAMPLE 1: THROUGH CRACK IN A STIFFENER For the initial study, a through crack will initiated into the stiffener itself. The crack is grown using an initial cycling of the uniform tension load in the panel. Figure 10 shows a cross section of the stiffener model with the through crack added into the part.

Figure 10 Initial crack modelled into the stiffener

Advances in crack growth modelling of 3D aircraft structures

799

The crack is now grown under the cyclic loading. Figure 11 and Figure 12 show the crack after 8 and 10 crack growth increments respectively. As can be seen the crack shape given is not pre-defined, rather it is computed directly from the stresses in the model.

Figure 11 Crack grown though the stiffener

Figure 12 Crack growing into the base panel As the crack continues to grow, under the loading used in this model, the crack grows faster along the breakout edge and slower though the depth of the panel. However it eventually breaks though the base panel and a transition can be made to a crack that can grow in two directions along the base panel.

800

S.C.Mellings, J.M.W.Baynham and R.A. Adey

Figure 13 Crack just before transition to a through crack

Figure 14 Crack transitioned to a through crack

EXAMPLE 2: CORNER CRACK IN A STIFFENER The analysis presented above assumes that the crack starts as a through crack that is modelled at the top of the stiffener. However, since the initial geometry does not contain the initial crack and this is added in from a library of pre-defined cracks, it is easy to start the analysis either with a different crack or with the crack in an alternative location or orientation. In this second analysis a crack is grown from a corner crack in the stiffener to identify the grown crack shape.

Advances in crack growth modelling of 3D aircraft structures

801

Figure 15 Initial corner crack added to the model Initially the crack growth proceeds as a through crack, however when it reaches the corner of the stiffener it transitions to become a “through” crack as seen in Figure 16. This initially has the same circular crack front, as shown in Figure 15, however as the crack grows through the stiffener this quickly transitions to become a through crack which is very similar to that grown from the initial through crack.

Figure 16 Corner crack transition to a through crack and then growing into base panel

802

S.C.Mellings, J.M.W.Baynham and R.A. Adey

EXAMPLE 3: CRACK IN THE PANEL The cracks grown above show a crack growing from the stiffener into the panel. Alternatively the crack may initiate in the panel itself and grow though into the stiffener. In this analysis a crack is initiated as a though crack starting at the edge of the sub-model. This analysis assumes that there is a similar crack growing into the remainder of the part that has not been modelled. Note that in this analysis the model used in the previous studies has been re-used and only the selected crack and initiation points needed to be changed.

Figure 17 Initial crack in panel

Figure 18 Crack grown into fillet

Figure 19 Crack grown into thickened section

Advances in crack growth modelling of 3D aircraft structures

803

EXAMPLE 4: FATIGUE ANALYSIS WITH MULTI-AXIAL LOADING The example presented above show various examples of uni-axial fatigue crack growth in the part. Each stage of the analysis yields a set of stress intensity factors at each mesh point along the crack front. This is used to grow the crack to the next increment, as shown in the previous examples, but will also allow a fatigue life to be computed. During the model build phase, it was noted that three load cases were included in this model; a uniform tension and two non-uniform load cases. In the analysis above, the fatigue load applied was a cyclic load of the uniform tensile load case from full to zero loading. Using the crack shapes from example 1, investigations into effects of changes of loading can be performed. Three fatigue cases will be studied looking at different combinations of the non-uniform loading using three different load spectrum files. In the first case the loads will act 180 degrees out of phase, in the second they will act in phase and in the last case they will be 90 degrees out of phase. 20.0000

Crack Size

15.0000

10.0000

5.0000 0

10000

20000

30000

40000

50000

60000

70000

80000

90000

100000

Cycles 180 deg out of phase

In Phase

90 deg out of phase

Figure 20 Comparison of crack growth rates As can be seen, the crack growth rate when the two loads are applied 180 degrees out of phase gives a longer crack life than when they either apply in phase or 90 degrees out of phase. These fatigue life results were evaluated without re-growing the crack. This situation is typical of forensic studies where crack shapes may be known from photographic data. The way in which the multi-axial loads are combined and

804

S.C.Mellings, J.M.W.Baynham and R.A. Adey

cycled may however affect the crack shape, and so now we will re-grow the crack to investigate these effects. The crack has therefore been re-grown in each case for a short distance to compare the crack growth results. In each case the resultant grown crack for the 90 degree and 180 degree out of phase load cases are show in the pictures below.

Figure 21 Crack growth with loading 90 degrees out of phase

Figure 22 Crack growth with loading 180 degrees out of phase

Advances in crack growth modelling of 3D aircraft structures

805

The analysis shows very similar grown crack shapes in the planar directions but, as can be seen, the case with the loads applied further out of phase generates a higher “twist” in the resultant crack. Using this data to re-compute the fatigue life we see that the fatigue life, shown in Figure 23, gives a similar behaviour to that estimated from the planar analysis, shown in Figure 20, but the fatigue calculation is now higher and generates a faster crack growth for the larger crack sizes. This is because the twist in the crack causes it to be aligned normal to the maximum principal stress direction. This results in higher computed SIF values and therefore a reduced life computation. This removes some of the uncertainty in the results resulting in a more realistic (and more conservative) life. 20.0000

Crack Size

15.0000

10.0000

5.0000 0

10000

20000

30000

40000

50000

60000

70000

80000

90000

100000

Cycles In phase

90 deg out of phase

180 deg out of phase

Figure 23 Comparison of crack growth rates with full analysis

EXAMPLE 5: CONTACT BETWEEN THE CRACK FACES In the models, examined tensile loading was applied and the crack growth was dominated by mode 1 behaviour. However in many realistic cases, the cyclic loading may, at one extreme, introduce compression across the crack face. If the resulting contact is not taken into account, the calculated stress intensity factors will not be realistic. For this example an alternative load case has been applied to this model, in which the loading creates a compressive load across the crack faces. In a standard BEM analysis, the crack faces are assumed to be load free and therefore in a compressive field the crack faces are allowed to interfere. This causes a negative stress intensity factor in the mode 1 direction.

S.C.Mellings, J.M.W.Baynham and R.A. Adey

806

3.0000E+02 2.0000E+02

SIF

1.0000E+02 0.0000E+00 0.0000 -1.0000E+02

K1 0.2000

0.4000

0.6000

0.8000

1.0000

K2 K3

-2.0000E+02 -3.0000E+02 -4.0000E+02 Local position along crack front

Figure 24 SIF values computed from a compressive loading In reality however, the crack faces would not interfere but would transfer loading. This is modelled with contact applied to the crack faces. This allows for the mode 2 and 3 results to be accurately computed. 3.0000E+02 2.0000E+02

SIF

1.0000E+02 0.0000E+00 0.0000 -1.0000E+02

K1 0.2000

0.4000

0.6000

0.8000

1.0000

K2 K3

-2.0000E+02 -3.0000E+02 -4.0000E+02 Local position along crack front

Figure 25 SIF values computed with contact on crack faces The application of contact on crack faces can best be seen with a simple example. In this model an angled though edge crack has been modelled into a plate, which is subjected to a compressive loading.

Advances in crack growth modelling of 3D aircraft structures

807

Figure 26 Initial crack in plate (front & back surfaces removed) At the first increment the contact load generates a sliding along the crack faces which generates a shear loading in the part. This causes the crack to turn as shown in Figure 27.

Figure 27 Crack after first growth step As can be seen this now generates a tensile (opening) load on the crack face which leads to SIF values that are dominated by mode 1. The crack continues to grow in this direction until a “pit” is generated.

808

S.C.Mellings, J.M.W.Baynham and R.A. Adey

Figure 28 Pit generated after 5 crack growth steps

CONCLUSION A methodology has been presented in this paper which can be applied to crack growth in real components. This process takes into account the real stress fields which occur in practise, including effects of section thickness change, use of the real component and crack geometry, and multi-axial loading. The method has been applied in a multi-scale fashion, in which sub-models have been extracted from larger FE models. It has been shown that the methodology predicts how a crack will grow without being constrained by standard reference solutions. By extension, it can be concluded that this technique allows investigations to be performed into the grown crack shapes and enables effects of design changes to be studied in a realistic manner. Although not discussed, the technique of applying loading to the crack faces can be extended to allow effects of residual stresses to be studied. This is necessary for the study of effects on crack growth of different surface treatments or different methods of welding. Additionally, multi-cracks can be analysed together in the same part and the interference of the cracks will affect the crack growth paths.

Advances in crack growth modelling of 3D aircraft structures

809

REFERENCES [1]

BEASY User Guide, Computational Mechanics BEASY Ltd, Ashurst, Southampton, UK, 2008.

[2]

Portela, A.; Aliabadi M.H; Rooke, D.P., “The Dual Boundary Element Method: Efficient Implementation for Cracked Problems”, International Journal for Numerical Methods in Engineering, Vol 32, pp 1269-1287, (1992).

[3]

Mi Y; Aliabadi M.H., “Three-dimensional crack growth simulation using BEM”, Computers & Structures, Vol. 52, No. 5, pp 871-878, (1994).

[4]

Prasad, N.N.V.; Aliabadi M.H; Rooke, D.P., “The Dual Boundary Element Method for Thermoelastic Crack Problems”, International Journal of Fracture, Vol 66, pp 255-272, (1994).

[5]

Dell’Erba D.N.; Aliabadi M.H; Rooke, D.P., “BEM analysis of fracture problems in three-dimensional thermoelasticity using J-Integral”, International Journal of Solids and Structures, Vol. 38, pp. pp.4609-4630, (2001).

[6]

Neves A; Niku S.M., Baynham J.M.W., Adey, R.A., “Automatic 3D crack growth using BEASY”, Proceedings of 19th Boundary Element Method Conference, Computational Mechanics Publications, Southampton, pp 819-827, 1997.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

DEVELOPMENT OF ADVANCED RISK ASSESSMENT METHODOLOGIES FOR AIRCRAFT STRUCTURES CONTAINING MSD/MED Min Liao1, Yan Bombardier1, Guillaume Renaud1, Nick Bellinger1, Terence Cheung2 1

Institute for Aerospace Research (IAR), National Research Council Canada (NRC) 2 Directorate Technical Airworthiness and Engineering Support (DTAES), Department of National Defence of Canada

Abstract: This paper presents the results of IAR/NRC research on developing risk assessment methodologies and tools for aircraft structures containing multiple site fatigue damage (MSD) / multiple element damage (MED) to support the structural life cycle management for Canadian Forces air fleet. The lower surface panel of the CC-130 centre wing structure was used as a case study to demonstrate the methodologies and tools developed by NRC. At first, a crack growth analysis program was developed to simultaneously grow the MSD cracks, considering the effects of crack interaction and adjacent structure failure (MED). A beta correction library for MSD/MED cracks was developed and verified with p-version finite element (FE) analysis (StressCheck) and other closed-form solutions. The residual strength of the MSD/MED structures was analyzed using a global-local FE modelling technique, which was developed to quantify the load redistribution among the cracked panel and adjacent components. For the quantitative risk analysis, an efficient Monte Carlo Simulation technique was developed and used in NRC’s crack growth program. The Monte Carlo simulated crack size distribution was then used in NRC’s in-house risk analysis code, ProDTA, to calculate the probability of failure (PoF) for the MSD/MED structures. The results of the case study showed that the presence of MSD/MED significantly increases M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 811–837. © National Research Council Canada 2009. Published by Springer Science+Business Media B.V. Dordrecht.

812

M. Liao, Y. Bombardier, G. Renaud, N. Bellinger and T. Cheung

the PoF. Therefore, the maintenance actions should be adjusted according to the MSD/MED risk analysis, crack growth, and residual strength analysis results.

INTRODUCTION A risk based management approach/tool has been adopted by many military air fleets. In the past few years, the Canadian Forces (CF) have been introducing and revising a Record of Airworthiness Risk Management (RARM) process to manage technical and operational airworthiness for all CF aircraft [1,2]. Today, the RARM has become the single most critical decision making tool in the CF air fleets [3]. In RARM, both qualitative (defining hazard probability as ‘frequent’, ‘remote’, ‘extreme improbable’, etc.) and quantitative (defining hazard probability as ‘10-3’, ‘10-5’, ‘10-8’, etc., per flight hour) risks are defined for all CF aircraft platforms, including unmanned air vehicles (UAVs) and helicopters. When there are sufficient data available, a quantitative risk assessment (RA) can be performed to substantiate the assignment of a risk index number. When a qualitative RA indicates a high/medium risk, a detailed quantitative RA is often demanded to support better decision-making. However, some challenges remain for a quantitative RA, such as how to establish methodologies to get sufficient/meaningful input data, particularly for a complicated damage scenario such as multiple site fatigue damage (MSD)/multiple element fatigue damage (MED)/widespread fatigue damage (WFD).

Upon request by the Department of National Defence (DND) of Canada, NRC was tasked to develop advanced methods for carrying out risk assessment of build-up structures. The lower surface panel of the CC-130 centre wing, shown in Figure 1, was selected as a case study for this work. The material of the panel is 7075T7351. It is known that the structural failure of the centre wing caused a catastrophic accident on a C-130A during a firefighting mission in Walker, California, in 1997 [4,5]. As investigated by the National Transport Safety Board (NTSB), inspection of the failed area revealed MSD [4] and MED [5] in one of the lower surface panels.

Development of advanced risk assessment methodologies for aircraft structures

813

Figure 1: CC-130 centre wing box, and the lower surface panel (circled, panel No. 3) studied in this paper Since the Aloha accident in 1988, extensive research has been carried out on MSD/WFD. A brief survey of available MSD/WFD analysis guidelines and methods was performed by focusing on the major authorities and aircraft industries. In general, some useful technical guidelines were found in working documents from AAWG (Airworthiness Assurance Working Group) [6] and FAA (Federal Aviation Administration) [7] for MSD/WFD analysis/evaluation. In 1994, Airbus developed a Monte Carlo method for the WFD evaluation of the A300 fuselage, where random initial crack scenarios were determined by a lognormal distribution developed from simple coupon fatigue lives, and subsequent crack growth and failure was analyzed using fracture mechanics models [8]. In 2006, Lockheed Martin (LM) Aerospace developed an equivalent initial flaw size (EIFS) based Monte Carlo method to perform risk analyses of C-130 build-up structures [9]. On the subject of risk assessment as it might apply to managing MSD/MED situations, it was acknowledged that considerable advances had occurred in this area recently. However, additional studies would need to be conducted to establish risk approach capability in this area [6]. This paper presents the NRC’s effort on the development of advanced methodologies for MSD/MED crack growth analysis, including MSD/MED crack beta solutions, MSD/MED residual strength analysis, and Monte Carlo simulation, for quantitative RA of build-up structures.

814

M. Liao, Y. Bombardier, G. Renaud, N. Bellinger and T. Cheung

RISK ASSESSMENT METHODOLOGY AND TOOL According to [1,2], risk analysis is an evaluation of a potential hazard severity and probability of occurrence. For aircraft structural applications, the potential hazards include structural failures that can cause injury or death to personnel, damage to or loss of the aircraft, or reduction of mission readiness/availability. The most important task in risk assessment is to calculate the probability of failure (PoF) of aircraft structures. Similar to US DoD (JSSG2006, MIL-STD-1530C), the CF use single flight hour PoF (hazard probability per flight hour) to measure the risk level at critical locations [2]. In collaboration with the Defence Research and Development Canada (DRDC), NRC has been carrying out research on structural risk analysis since 1999. NRC initially evaluated the USAF (US Air Forces) tool PROF (Probability of Fracture) for PoF study. Because of the limitations of PROF for some applications, especially related to corrosion risk assessment, NRC developed an in-house tool, ProDTA (Probabilistic Damage Tolerance Analysis), for structural risk analysis by taking into account both conventional fatigue damage and age related environmental damage (i.e. corrosion) [11,12]. Figure 1 presents the major inputs of ProDTA, which are grouped into fatigue and corrosion related inputs.

Figure 2: Main inputs for NRC in-house tool ProDTA

Development of advanced risk assessment methodologies for aircraft structures

815

Two methods were developed in ProDTA to calculate the PoF [12]. The first method, i.e. the Master Curve approach, grows an initial crack size distribution (ICSD) based on a single, master crack growth curve, and then calculates the PoF using numerical integration. This method is similar to that of PROF [11], but ProDTA employs different numerical techniques/algorithms to calculate the PoF. In addition, ProDTA has the flexibility to use various statistical models for different inputs, depending on the available fleet data. The second method, i.e. the Monte Carlo approach, grows the ICSD using a Monte Carlo crack growth program, including retardation and MSD/MED crack interaction. The Monte Carlo approach is suitable for both single and MSD/MED crack growth problems. More importantly, this method also allows ProDTA to use more random variables including age degradation parameters such as corrosion growth rate, thickness loss, pit depth, and corrosion protection breakdown time [12]. In ProDTA, the following three types of PoF can be calculated separately or jointly, depending on the application: In ProDTA, the following three types of PoF can be calculated separately or jointly, depending on the application: ∞ ∞ K C criterion : PoF (t ) = ∫ 0 f ( a ) ∫0 f K ( K C )(1 − Hσ [σ C ( a, K C )]) dK C da C ∞ Residual strength criterion : PoF (t ) = ∫ 0 f ( a )(1 − Hσ [σ RS ( a )])da a Critial crack size criterion : PoF (t ) = f ( a ) /[1 − F ( a )], F ( a ) = 1 − ∫ 0 c f ( a )da

(1) (2) (3)

where PoF(t): f(a): fKc(KC): H (σ): σC (a, KC): σRS (a): F(a):

Single flight hour PoF at time t Probabilistic density function (pdf) of crack size a, at time t Probabilistic density function (pdf) of fracture toughness KC Distribution of the maximum stress per flight hour Critical stress at a given crack size a, stress intensity factor related beta factor β(a), and KC σC (a, KC) =KC/[β(a)√πa]: Residual strength as a function of crack size a Cumulative distribution function (CDF) of crack size at time t, i.e., the probability that the crack length is smaller than a, at time t

The PoF(t) of Eqn. 1 accounts for the failure probability that the crack length is smaller than a certain size but the stress level encountered during the single flight hour is greater than σC (a, KC), dominated by KC and a. This probability is the same as a hazard rate for a single flight hour.

816

M. Liao, Y. Bombardier, G. Renaud, N. Bellinger and T. Cheung

The PoF(t) of Eqn. 2 accounts for the failure probability that the crack length is smaller than a certain size, but the stress level encountered during the single flight hour is greater than σRS (a) dominated by crack size a. This probability is the same as a hazard rate for a single flight hour. The PoF(t) of Eqn. 3 accounts for the failure probability that the crack length is greater than a critical crack length aC, on condition that there is no failure before time t. It should be noted that, for extreme remote probability (F (a)<10-7), the PoF(t) (hazard rate) is almost the same as f(a), since the non-failure condition term is approximately 1.0 ([1 - F(a)] = 1.0 - 10-7 ≈ 1.0). From Figure 1, it is shown that, without the corrosion effects, the major inputs for a risk analysis are the ICSD, the crack growth curve including geometry factor (beta), the maximum stress distribution, the probability of detection (POD), and the residual strength or Kc distribution. The generic methods for preparing all these inputs can be found in previous papers from the authors [13,14]. This paper mainly describes the development of the crack growth curves, the residual strength curves, and the ICSD for MSD/MED structures.

MSD CRACK GROWTH ANALYSIS MSD crack growth beta library Due to computational time constraints, numerical approaches involving direct computation using boundary and/or finite element methods were rejected in favour of closed-form and tabular solutions. The developed methodology aimed to address MSD problems in panels with a large number of holes and cracks emanating from the holes, while providing fast and accurate β-solution results for every crack tip. This was achieved by using a combination of closed-form and table look-up βfactors, including the finite element (FE) modelled β-factors. A β-factor library was developed to calculate the β-solutions using the principle of superposition. A minimum set of seven β-factors, illustrated in Figure 3, was found to be sufficient to cover all possible MSD cases in the CC-130 wing panel case study. Some β-factors illustrated in Figure 3, such as the surface crack [15], the collinear cracks [16], and the crack approaching a hole [17], were directly obtained from the literature. Other β-factors were obtained by the principle of superposition of various published β-factors such as the edge crack, the unsymmetrical crack, and the radial or diametrical cracks emanating from a hole with and without pin load [18-24]. In addition, NRC developed a more accurate correction factor for cracks emanating from an offset loaded fastener hole (Figure 3b and 3c when B≠W/2).

Development of advanced risk assessment methodologies for aircraft structures

817

The table look-up method uses pre-computed β-factors stored in a table to estimate the β-factor by spline interpolation. This method was required to ensure fast execution of the Monte Carlo simulation. In addition to the pre-defined β-library, user-defined β-factors can be used to model additional effects, such as the effect of intact or failed adjacent components and the load redistribution to other structures caused by a long crack. Crack link-up events are determined using the “intuitive link-up criterion” proposed by Swift [25]. This link-up criterion stipulates that ligament failure would occur when the plastic zone of the lead crack touches the plastic zone of adjacent cracks. The β-factor compounding method requires proper identification of the crack information. An algorithm was developed to identify and update the crack information, which includes extracting the location and size of a crack, determining if the crack is an edge or a centre crack, determining if the crack is a part-through or through crack, determining the type of interaction with adjacent structural elements (edge, hole, or other cracks), verifying if crack link-up occurs, and merging the cracks as required. The β-factors compounding algorithm is provided in Appendix A. To make the analysis easy and user-friendly, the β-factor compounding algorithm was implemented in the crack growth program, called CGCC130MSD, which was specifically developed at NRC to analyze the MSD problem in the CC-130 wing panel case study. Before the CC-130 case study, a small scale MSD panel was used to verify the proposed MSD β-factor compounding methodology. The baseline results, i.e., the stress intensity factors (SIF) at every crack tip, were obtained using the p-version FE code, StressCheck from ESRD, based on the contour integral method. To efficiently extract the data, a Visual Basic for Applications (VBA) interface was developed to output the SIF at every crack tip as a function of the element p-level. The SIF error estimated by StressCheck was less than 0.5% for the models used to generate the data presented in this paper. The small MSD panel case study is presented in Figure 4. Diametrical cracks with initial flaw size of 0.254 mm (0.010 inch) were assumed for the five holes, except for the lead crack (a11) that was assumed to be 1.27 mm (0.050 inch). All cracks were assumed to be through-the-thickness.

818

M. Liao, Y. Bombardier, G. Renaud, N. Bellinger and T. Cheung

Figure 3: Schematic representation of the β-factors used to solve MSD problem by the principle of superposition

Development of advanced risk assessment methodologies for aircraft structures

819

Figure 4: Parametric MSD panel case study

Bypass stress was applied on the upper edge of the panel as well as bearing loads at every hole. The amount of bearing load was determined by a constant bearing (σpin) / bypass (σby) stress ratio (BBR=σpin/σby; σpin=Pi/(diT)). For the studied case, a BBR of 1.587 was assumed for hole No. 1 and 0.65 for holes No. 2 to 5. The bearing load was reduced to zero once a link-up occurred with an edge or with another crack. The β-solutions for all crack tips were calculated from the SIF provided by StressCheck using the following equation: β=

K σ

by

πa

(4) 0

where K is the SIF, σby is the bypass gross stress, and a0 is the half crack length for centre cracks and the total length for an edge crack. The β-solutions computed using NRC’s CGCC130MSD program and StressCheck are compared in Figures 5 and 6 for the lead crack defined by a11 and a12 (a11/a12). Overall, good agreement was obtained between CGCC130MSD and StressCheck’s β-solutions for the lead crack (a11/a12).

820

M. Liao, Y. Bombardier, G. Renaud, N. Bellinger and T. Cheung

Figure 5: Comparison between the lead crack (a11 / a12) β-solutions obtained using NRC’s CGCC130MSD program and StressCheck (a0 ≤ 50 mm)

Figure 6: Comparison between the lead crack (a11 / a12) β-solutions obtained using NRC’s CGCC130MSD program and StressCheck (50 mm ≤ a0 ≤ 300 mm)

Development of advanced risk assessment methodologies for aircraft structures

821

Effects of adjacent structures and MED Load redistribution β-factors, developed using finite element models, were used to quantify the effects of adjacent structures on cracked components. For the CC130 wing panel (Figure 1) case study, this correction factor was determined considering two levels of interactions: the first interaction is the local load redistribution between panel and stringers/caps, as a function of crack length. The load taken by the stringers/caps directly decreases the SIF of the panel crack. The second interaction is the progressive load transfer from the panel-stringers/caps assembly to the wing box. This reduction of load also reduces the SIF of the panel crack. The first load redistribution β-factor was derived using a simplified StressCheck finite element model of the local panel-stringers/caps assembly. It consisted of two-dimensional (2D) models of the various components, attached with fastener elements that considered nonlinear contact in the vicinity of the holes. For each crack length considered, two models were built, one with the stiffening elements, one without. The first correction factor, βas1, was calculated by taking the ratio of the SIF obtained from the two models: β β πa β πa K β as1 = Tu as1 = Ts = s βTu πa βTu πa K u

(5)

where Ks is the crack tip SIF of the stiffened panel, Ku is the crack tip SIF of the unstiffened panel, βTs and βTu are the total correction factor of the stiffened and unstiffened panel, respectively, and a is the crack length. The second correction β-factor, corresponding to the load redistribution between the panel-stringers/caps assembly and the wing box, was obtained using a coarse global NASTRAN finite element model. A wing bending load case was used to calculate the total load reduction measured in the panel, spar cap, and stringers, by disconnecting nodes to simulate the presence of a lead crack. This load reduction, normalized to 1.0 when a=0, was defined as the correction factor βas2. The total effects of the adjacent structures on the lead crack tip SIF, βas, was defined as the product of βas1 and βas2. In addition, MED was also considered by assuming the possibility of stringer failures for a long lead crack. To simulate stringer failure, the βas factors were modified by either deleting or breaking the stringers in the local and global FE models. The adjacent structure correction factors for the lead crack of the CC130 wing panel case study are shown in Figure 7. As expected, breaking stringers increased the βas factors, which increased the crack tips SIF. It should be noted that crack growth analysis for the stringers were not performed in this study.

822

M. Liao, Y. Bombardier, G. Renaud, N. Bellinger and T. Cheung

Figure 7: Adjacent structure β-factors (lead crack) MSD/MED crack growth modelling To analyze the MSD problems, a crack growth algorithm, CGCC130MSD, was developed by NRC to propagate multiple cracks simultaneously. The CGCC130MSD program currently uses Forman equation and Hsu retardation model [28]. A flowchart of the analysis procedure used in CGCC130MSD is provided in Appendix B. Using the CGCC130MSD program, the CC130 wing panel (Figure 1) with 17 holes along the crack plane was modelled. Two scenarios were simulated: the standard crack scenario (Figure 8a) and the MSD scenario (Figure 8b). The standard crack (SC) scenario includes multiple cracks propagating simultaneously in the panel, but does not include crack interactions (i.e. no cracks grow toward each other). This scenario is commonly used for durability and damage tolerance analysis (DaDTA) which normally simplifies the analyses by assuming that every crack growth is independent. The MSD scenario, however, includes the βcorrection factor to model the crack interaction. A crack tip aij is identified by the hole number i, starting at 1 from the left side of the plate, and by the side j of the crack tip (j=1 for the left side, j=2 the right side). The lead crack was assumed to be a11/a12 (illustrated in Figure 8) for both scenarios with an initial crack size of 1.27 mm (0.050 inch) for both a11 and a12. All other cracks were assumed to have an initial crack size of 0.127 mm (0.005 inch), in the deterministic MSD/MED analysis. Using a loading spectrum developed by L3-Spar, a series of crack growth analyses were carried out for the CC-130 wing panel, with both standard crack and MSD scenarios. As discussed in the previous section, all these analyses included the

Development of advanced risk assessment methodologies for aircraft structures

823

effects of adjacent structural elements such as the wing cap and stringers and the effects of load redistribution within the wing box structures.

Figure 8: Standard crack scenario versus MSD scenario The β-solutions for the standard crack and MSD scenarios are presented in Figure 9 for the lead crack a11/a12. It is seen that the crack interaction only affects the βsolutions when crack tips are sufficiently close. The crack interaction effect increases the SIF at the crack tips, and then increases the crack growth rate (i.e. faster link-up). In addition, the presence of the interacting cracks (e.g. a21, a31, …, ai1) reduces the net section area of the ligaments between the holes and therefore reduces the time required to break the ligaments, even if the crack interaction βfactor is not included in the model. The crack growth curves for the standard crack and MSD scenarios are compared in Figure 10, which presents the lead crack (a11/a12) length versus number of cycles. For both cases, the lead crack became unstable around 400 mm (15.75 inch). Due to this double effect (β amplification and ligament net section loss), the MSD scenario provided about 25% shorter life than the standard scenario. The crack growth curves (crack tip position along the panel versus cycles) for all crack tips are provided in Figure 11 for the standard scenario and in Figure 12 for the MSD scenario. A comparison between the crack growth curves of the secondary flaws (e.g. a41 and a42) reveals an additional factor that speeds up the MSD crack growth, i.e. diametrical cracks grow faster than single radial crack emanating from a hole.

824

M. Liao, Y. Bombardier, G. Renaud, N. Bellinger and T. Cheung

Figure 9: β-solutions of the lead crack (a11/a12) for the standard crack and MSD scenarios

Figure 10: Crack growth curves for the standard crack and MSD scenarios (lead crack only)

Development of advanced risk assessment methodologies for aircraft structures

Figure 11: Crack growth curves for the standard crack scenario (all cracks)

Figure 12: Crack growth curves for the MSD scenario (all cracks)

825

826

M. Liao, Y. Bombardier, G. Renaud, N. Bellinger and T. Cheung

MSD/MED RESIDUAL STRENGTH ANALYSIS Several failure criteria can be used to calculate the residual strength of a structure in the presence of a growing crack [18,19]. To date, the methodologies/tools developed at NRC consider the yield (or ultimate) strength criterion and the fracture toughness (KC) criterion for abrupt fracture. In this paper, the residual strength of the stringers and/or fasteners were not analyzed. The time of the adjacent structure failure (MED) was assumed based on in-service data. In this case study, the first stringer was assumed severed when the lead crack reached 304.8 mm (12.0 inch), and the second stringer severed when the lead crack reached 431.8 mm (17.0 inch). The residual strength curve as a function of crack length σRS(a) is defined using the following equation by combining the the yield strength and the fracture toughness criteria: σ

⎛

RS

(a) = min⎜⎜ σ ys ,

⎝

⎞ ⎟⎟ β(a) πa ⎠ KC

(6)

where σys is the yield strength of the material, β(a) is the total β-solution for the SIF calculation, and a is the crack length. If present, MSD and MED effects are included in the compounded β-solution, which can include FE-based β-factors for the effects of adjacent structures. Considering continuing damage growth, the β(a) values corresponding to part-through cracks are discarded from the residual strength calculation. The part-through crack β-factor would result in a localized peak residual strength close to the holes, but it would quickly decrease once the crack becomes a through crack. A set of guidelines was established to develop the monotonically decreasing residual strength curves for the standard crack and MSD/MED scenarios. For instance, if the residual strength of the panel increases when the crack approaches a stiffening element, it is assumed that the strength at this point is the same as its maximum value under the stringer. Passed the stringer, the strength continues to decrease. The residual strength curves are illustrated in Figure 13 for the CC130 wing panel case study. In this case, a design limit stress criterion was applied and the residual strength was expressed as the ratio to this design limit stress.

Development of advanced risk assessment methodologies for aircraft structures

827

Figure 13: Residual strength curves: standard crack scenario and MSD/MED scenario

RISK ASSESSMENT FOR STRUCTURE WITH MSD/MED Initial crack size distribution (ICSD/EIFSD) In this paper, a direct method (see [14] for details) was applied to regress (back calculate) the in-service crack data to obtain the EIFS values and then a best-fit statistical distribution was determined for the EIFS values In this regression, two approaches were used to calculate the EIFS. The first approach, makes use of a master DaDTA crack growth curve. As a DaDTA curve often starts from an initial crack size of 0.127 mm (0.005 inch) or smaller, an exponential function can be used to extrapolate the curve to smaller EIFS values. The second approach uses the NRC crack growth program. The program was used to find an EIFS through an iteration process. Although some very small EIFS values were calculated, the material model, i.e., da/dN-ΔK curve, used only long crack growth data, no short/small crack growth data, Therefore, it should be noted that all these EIFS values do not represent the material intrinsic discontinuity state. Figure 14 presents the EIFS results as a function of the probability of exceedance, using the two approaches and 16 in-service crack findings, for both standard crack and MSD scenarios. It is shown that both approaches provided similar EIFS values. Recall that the NRC’s crack growth program takes into account crack growth retardation effect, but the master curve approach does not. This comparison indicates that the crack retardation effect is not significant under the spectrum loading applied.

828

M. Liao, Y. Bombardier, G. Renaud, N. Bellinger and T. Cheung

A goodness-of-fit test was carried out and a two-parameter (2P) lognormal distribution was found to fit the EIFS values well, as shown in Figure 14.

Figure 14: EIFS values obtained by the regression of in-service finding crack data using two approaches: NRC crack growth program and a master curve

Monte Carlo MSD/MED crack growth simulation To carry out MSD/MED risk analysis, the CGCC130MSD program was enhanced with the capability to generate random initial crack sizes using the Monte Carlo algorithm. For the CC130 case study, the same random initial crack sizes were generated for both sides of the 17 holes based on the lognormal EIFS distribution (EIFSD) (Figure 14). In the Monte Carlo simulation, each MSD/MED crack growth analysis was performed with MSD/MED crack interaction, in a similar way as shown in Figure 12, until the lead crack reached 508 mm (20 inch). To verify the Monte Carlo program and the EIFSD, the crack size distribution F(a) at a certain time t*, obtained from the Monte Carlo simulation, was compared to the inservice findings. The comparison, presented in Figure 15, indicates that the crack

Development of advanced risk assessment methodologies for aircraft structures

829

size distribution obtained from the Monte Carlo simulation matched the in-service findings well. It should be noted that because the in-service data were obtained at different inspection times, they were regressed to the same time t* using the master DaDTA crack growth curve. Since the studied wing panel has 17 holes and 34 crack tips, each crack growth analysis (starting from an EIFS of 0.127 mm (0.005 inch)) takes approximately 3.5 minutes on an AMD processor Opteron 250 running on Linux, which would require 6-7 years to perform a Monte Carlo simulation with 106 trials. To reduce the computation time, a tail sampling technique was used for the Monte Carlo simulation. Figure 15 shows that the crack size distribution (F(a)) obtained using the Monte Carlo simulation with a 10% tail sampling, is very similar to the one obtained with full distributional sampling.

Figure 15: Monte Carlo simulated crack size distribution (F(a)) matched with the in-service findings

Calculate PoF using ProDTA After the MSD/MED Monte Carlo simulation is completed, the lead crack size distributions, at different times, were processed in the NRC risk analysis code ProDTA. Figure 16 presents a part of the crack size distributions generated by the Monte Carlo simulation.

830

M. Liao, Y. Bombardier, G. Renaud, N. Bellinger and T. Cheung

To calculate the single flight hour PoF, a maximum stress distribution was determined from the stress exceedance data in a single flight hour, which is shown in Figure 17. Depending on the available stress data, ProDTA can take either Gumbel distribution or a table look-up format for the maximum stress distribution. Figure 17 shows that a table look-up format fitted the stress data better than a Gumbel distribution, so the table look-up format was used for the CC-130 wing panel case study.

Figure 16: Monte Carlo simulated crack size distribution (part) for the MSD scenario

Development of advanced risk assessment methodologies for aircraft structures

831

Figure 17: Maximum stress distribution for single flight hour

Based on the Eqn. 2, ProDTA calculated the single flight hour PoF using the lead crack size distribution (Figure 16), maximum stress distribution (Figure 17), and residual strength curve (Figure 13). The single flight hour PoF results are presented in Figure 18, for both the standard crack and MSD scenarios, using the Monte Carlo approach. The comparison indicates that the PoF for the MSD scenario is significantly higher than the PoF for the standard crack scenario, especially in the end of life stage. Therefore the maintenance schedule should be revised for MSD scenario, based on the MSD/MED risk analysis, the MSD/MED DaDTA and the residual strength analysis.

832

M. Liao, Y. Bombardier, G. Renaud, N. Bellinger and T. Cheung

Figure 18: Single flight hour PoF results for the standard crack and the MSD scenarios

CONCLUDING REMARKS In this paper, advanced methodologies were presented for carrying out durability and damage tolerance analysis for built-up structures with MSD/MED. The methodologies included 1) a beta library for MSD/MED cracks in which most beta solutions were verified with StressCheck FE and closed form solutions; 2) crack growth analysis with MSD/MED interactions; 3) MSD/MED residual strength analysis; and 4) MSD/MED crack growth Monte Carlo simulation. All these advanced methods were implemented in a crack growth software, CGCC130MSD, developed by NRC. To carry out the MSD/MED risk analysis, an EIFS distribution was developed based on in-service inspection data. Finally, an NRC risk analysis code, ProDTA, was revised to process the crack size distribution from the MSD crack growth program in order to calculate the PoF for build-up structures. In this wing panel case study, the PoF for the MSD scenario is significantly higher than the PoF for the standard crack scenario, especially when the structure is near the end of its service life. Future work includes the use of large-scale structure test results to validate the NRC MSD/MED analysis tools, in collaboration with other organizations.

Development of advanced risk assessment methodologies for aircraft structures

833

ACKNOWLEDGEMENTS This work was performed with financial support from DRDC (Defense Research and Development Canada) and NRC (National Research Council Canada) through project 13ph11: Quantitative Risk Analysis of Aircraft Structures, under 13ph: Economic Life Assessment for CF Air Fleets, and other DTAES contracts. Thanks to Major Y. Caron and J. Gaerke of DTAES/DND, Capt. B. Tang, Mr. K. McRae of DRDC/DND, Dr. Gang Li and Mr. M. Yanishevsky of NRC for providing some data and technical discussions.

REFERENCES [1]

Department of National Defense of Canada (August 2007) Review of Open Records of Airworthiness Risk Management, Department of National Defense of Canada, Airworthiness Directive (AD) DND-AD-2007-01, 5 p.

[2]

Department of National Defense of Canada (July 2007) Technical Airworthiness Manual (TAM), Department of National Defense of Canada, Document no. C-05-005-001/AG-001, 530 p.

[3]

Komorowski, J.P., Bellinger, N.C., Liao, M. and Fillion, A. (2007) Application of the Holistic Structural Integrity Process to Canadian Forces Challenges, Proceedings of the 2007 USAF ASIP conference.

[4]

National Transportation Safety Board LAX02GA201, NTSB Identification: LAX02GA201, http://www.ntsb.gov/ntsb/brief2.asp?ev_id=20020621X00954&ntsbno=LA X02GA201&akey=1

[5]

National Transportation Safety Board (September 2002) Update On Investigations Of Firefighting Airplane Crashes In Walker, California And Estes Park, Colorado, NTSB press release, http://www.ntsb.gov/pressrel/2002/020924.htm

[6]

Federal Aviation Administration – Regulations and Policies, Task 6 – Widespread Fatigue Damage, Aviation Rulemaking Advisory Committee, Transport Airplane and Engine Issue Area, http://www.faa.gov/regulations_policies/rulemaking/committees/arac/media /tae/TAE_AA_T6.pdf

[7]

Advisory Circular (DRAFT), Widespread Fatigue Damage, AC 120-YY, Initiated by: ANM-100 and AFS–300, Federal Aviation Administration, Flight Standards Service, Washington, D.C.

[8]

Schmidt, H. J., Schmidt-Braddecker, B., Trey, H. (1999) Airbus A300 Fuselage Program For Life Extension and WFD Evaluation, Proceedings of

834

M. Liao, Y. Bombardier, G. Renaud, N. Bellinger and T. Cheung

the 1999 Aging Aircraft Conference. [9]

P. Christiansen, J. Bateman, and A. Navarrete (2006) C-130 Center Wing MSD/MED Risk Analysis, Proceedings of the 2006 USAF ASIP conference, San Antonio, USA.

[10]

Liao, M., and Xiong, Y. (December 2000) Risk assessment of aging aircraft structures using PRISM and PROF, Canadian Aeronautics and Space Institute Journal, vol. 46, no. 4.

[11]

Liao, M., Forsyth, D.S., and Bellinger, N.C. (December 2003) Risk Assessment of Corrosion Maintenance Actions in Aircraft Structures, Proceedings of the 2003 USAF ASIP conference, Savannah, GA, USA.

[12]

Liao, M., Bellinger, N.C, Forsyth, D.S., and Komorowski, J.P., (June 2005) A new probabilistic damage tolerance analysis tool and its application for corrosion risk assessment, Proceedings of the 23rd ICAF Symposium, Hamburg, Berlin, EMAS Publishing, pp 241-252.

[13]

Liao, M., Renaud, G., Bellinger, N.C., Bombardier, Y., Cheung, T. (April 2008) Quantitative Risk Assessment for Aircraft Structures, Proceedings of the 11th Joint NASA/FAA/DOD Conference on Aging Aircraft, Phoenix, USA.

[14]

Liao, M., Renaud, G., Bombardier, Y., Bellinger, N.C. (October 2008) Development of Initial Crack Size Distribution for Risk Assessment of Aircraft Structures, Proceedings of the RTO-MP-AVT-157 Military Platform Ensured Availability Symposium, Montreal, Canada.

[15]

Newman, J. C., Raju, I. S., Jr. (August 1981) Stress-intensity factor equations for cracks in three-dimensional finite bodies, NASA Langley Research Center, NASA TM 83200, 54 p.

[16]

Rooke, D. P., Cartwright, D. J. (1976) Compendium of Stress Intensity Factors, Her Majesty’s Stationery Office, 330 p.

[17]

Isida, M. (1970) On the Determination of Stress Intensity Factors for some Common Structural Problems, Eng Fract Mech, vol. 2, pp. 61-79.

[18]

Gallagher, J. P., Giessler, F. J., Berens, A. P., Engle, R. M., Jr. (May 1984) USAF Damage tolerant design handbook: guidelines for the analysis and design of damage tolerant aircraft structures, Wright-Patterson AFB, Ohio, Air Force Wright Aeronautical Labs, report AFWAL-TR-82-3073, 835 p.

[19]

Miedlar, P.C., Berens, A. P., Gunderson, A., Gallagher, J. P. (November 2002) USAF Damage tolerant design handbook: guidelines for the analysis and design of damage tolerant aircraft structures, Wright-Patterson AFB, Ohio, Air Force Wright Aeronautical Labs, report AFRL-VA-WP-TR-20033002, 716 p.

[20]

Tada, H., Paris, P. C., Irwin, G.R. (1985) The stress analysis of cracks

Development of advanced risk assessment methodologies for aircraft structures

835

handbook, Paris Productions Inc. and Del Research Corp., St-Louis, Missouri. [21]

Murakami Y. (Ed.) (1992) Stress intensity factors handbook, The Society of Materials Science, Japan, Pergamon Press, New York.

[22]

Harter, J. A. (December 1998) An alternate closed-form stress intensity solution for single part-through and through-the-thickness cracks at offset holes, AFRL-VA-WP-TR-1999-3001, 29 p.

[23]

Brussat, T. R. (1981) Technical Note: Mode I Stress Intensity for a Radial Crack at a Hole with Arbitrary Pressure Distribution, Eng Fract Mech, vol. 14, issue 1, pp. 233-235.

[24]

Cartwright, D. J., Parker, A. P. (January 1982) Opening Mode Stress Intensity Factors for Cracks in Pin-Loads Joints, Int J Fract, vol. 18, no. 1, pp. 65-78.

[25]

Swift, T. (January 1994) Damage tolerance capability, Int J Fatig, vol. 16, issue 1, pp. 75-94.

[26]

Newman, J. C., Dawicke, Jr, D. S., Sutton, M. A., Bigelow, C. A. (June 1993) A fracture criterion for widespread cracking in thin-sheet aluminum alloys, Proceedings of the 17th ICAF Symposium, pp. 443–468.

[27]

L. Wang, L., Chow, W. T., Kawai, H., Atluri, S. N. (1996) Prediction of widespread fatigue damage threshold, Proceedings of the 1996 FAA/NASA Symposium on Continued Airworthiness of Aircraft Structures, Atlanta, GA.

[28]

Harter, J. A. (June 2006) AFGROW User Guide and Technical Manual (AFGROW for Windows 2K/XP, version 4.0011.14), AFRL-VA-WP-TR2006-XXXX, 270 p.

836

M. Liao, Y. Bombardier, G. Renaud, N. Bellinger and T. Cheung

APPENDIX A: GENERIC β-SOLUTION COMPOUNDING ALGORITHM FOR MSD PROBLEMS

Development of advanced risk assessment methodologies for aircraft structures

APPENDIX B: MSD CRACK GROWTH ANALYSIS ALGORITHM

837

25th ICAF Symposium - Rotterdam, 27-29 May 2009

CRACK GROWTH IN FIBRE METAL LAMINATES UNDER VARIABLE AMPLITUDE LOADING S.U. Khan, R.C. Alderliesten, R. Benedictus Aerospace Materials and Structures, Faculty of Aerospace Engineering, Delft University of Technology, P.O. Box 5058, 2600 GB Delft, The Netherlands.

Abstract: Fatigue crack growth tests have been performed on three types of material (i.e., monolithic aluminium 2024 − T 3, laminated aluminium sheets and fibre metal laminate (F MLs)). Quantitative analysis of plastic zone sizes under variable amplitude loading in these materials is performed. Plastic zone sizes have been calculated using Irwin’s relation and compared with the measurements done using digital image correlation (DIC) technique. For monolithic metal and laminated sheets large difference has been observed between the calculation and measurement of plastic zone, while in case of F MLs, smaller difference is observed. In addition, delamination shapes in F MLs, under variable amplitude load sequences, are investigated. A change in delamination shape has been observed due to overloads. However, no significant effect of this change in delamination shape has been identified on fatigue crack growth.

INTRODUCTION The fatigue crack growth behaviour of Fibre Metal Laminates (F MLs) has been extensively investigated in the past decades. The majority of papers being published on this topic report fatigue and fracture experiments on various coupon specimens, analysis of observations and development of understanding by means of empirical or theoretical models [1–14]. The main advantage of F MLs being identified is the very slow and approximately constant crack growth under Constant Amplitude (CA) M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 839–858. © Springer Science+Business Media B.V. 2009

840

S. U. Khan et al.

loading as result of the bridging by the intact fibers. The crack growth is accompanied by delamination growth in the wake of the crack, which is for these materials to some extent considered favourable, opposite to for instance fibre reinforced polymer composites. Although various crack growth experiments on F MLs have been reported where selective load sequences, generalized or actual flight spectra have been applied [8–11], the detailed understanding of the interaction effects in F MLs is not yet fully understood. At least not to the same level of understanding as has been developed for crack growth under CA loading. In the research presented here, the F ML behaviour under V A loading has been investigated to enable further extension of the prediction capabilities of the CA model towards V A and flight spectrum loading. To make the step towards a V A loading prediction model from the CA prediction model, certain aspects have been identified for further study. These aspects are the quantitative information on plastic zone formation and its size and delamination shapes. The size of plastic zone is vital information required to investigate the effect of overloads. Smaller or constant plastic zone sizes will reveal the presence of limited interaction effects. Delamination growth rate under V A has already been investigated and no interaction effects were observed under V A loading. However, changes in delamination shape may affect the bridging stress intensity factor which directly influence the crack tip stress intensity factor. To study the effect of overloads on bridging stresses, delamination shapes under V A loading had to be investigated. To quantify the above mentioned parameters, three types of materials have been considered for the test program, namely monolithic aluminium, Metal Laminates (MLs) and F MLs. The similarities between behaviour of metal layers in F MLs and monolithic metal under V A loading has been studied. Plastic zone sizes have been calculated using Irwin’s relation and compared with the measured plastic zone using Digital Image Correlation (DIC). The effect of VA loading on delamination growth rate in F MLs has been investigated already, but the effect of V A loading on delamination shape is presented here. Finally, the aspects are addressed necessary to modify the current CA model to predict for V A cases.

FIBRE METAL LAMINATES After ARALL [15], GLARE (GLAss-REinforced) is the second member of the FML concept. Unlike ARALL, GLARE has good fatigue properties in combination with compressive loading [16]. Beside the excellent fatigue characteristics, GLARE also has good impact and damage tolerance characteristics [17]. GLARE is quite durable in case of corrosion, impact and high temperature. The fibre/epoxy layers act as barriers against corrosion of the inner metallic sheets, whereas the metal layers protect

Crack growth in FMLs under VA Loading

841

the fibre/epoxy layers from picking up moisture. The laminate has an inherent high burn-through resistance as well as good thermal insulation properties. FMLs are built up of alternating metal and fibre layers. For standard GLARE, aluminium 2024 − T 3 sheets and S 2 − glass fibres are bonded together with F M94 epoxy adhesive to form a laminate. This stack is cured in an autoclave at 120 ◦C and 6 bar for 1 21 hour. The fibre orientation is defined with respect to the rolling direction of the aluminium layers and each orientation represents a prepreg layer of 0.133 mm nominal thickness. Detailed description of the ARALL and GLARE grades are given in [16]. Since 1980, the FML concept has been investigated, developed and tested especially at Delft University of Technology. The development was speeded up in 1996 after the decision of Airbus to apply GLARE on the Airbus A380 [18]. In April 2005 this newly-build aircraft with GLARE on major structural fuselage parts, performed its first maiden flight. Part of the load is “BRIDGED” over the crack crack

METAL

Crack

D

FIBRE METAL

D

D

D=Delamination Boundary

Figure 1: Crack bridging of the fibres and delamination of the layers [19]

Fatigue Mechanisms FMLs being materials made of metal and composite exhibit the properties of both metals and composites. The metallic layers show fatigue crack growth similar to monolithic metals and the composite lay-up shows delamination at the metal-composite interfaces. The fibres in FMLs transfer load over the fatigue crack in the metal layers

842

S. U. Khan et al.

and restrain the crack opening. This phenomenon is called fibre bridging (figure 1). Delamination at the metal-fibre interface in the wake of the crack is also shown in figure 1. The cyclic shear stresses at the interface as result of the load transfer from the metal to the fibre layers induce delamination growth. Both the fatigue crack growth in the metal layers and the delamination growth at the interfaces form a balanced and so-called coupled process. For the development of structural FMLs, the main keys are the crack growth behaviour of the metal layers and the delamination resistance of the fibre/adhesive layers. The choice of the aluminium alloy in GLARE determines to a great extent the fatigue behaviour. If the metal type is determined by other (static) requirements, the fatigue behaviour can only be influenced by controlling the delamination resistance of the fibre/adhesive layers [20]. Increase in delamination resistance results in better fibre bridging and thus slower crack growth in the metal layers. However, too high delamination resistance will induce too high stresses in the fibre layers causing fibre failure. This means that control of the delamination resistance requires knowledge of the fibres and the adhesive.

(a) Single OL

(b) Block of OL

(c) Multiple OL

(d) Period Blocks of OL

(e) Two Level Block Loading – Const. Load amplitude

(f) Two Level Block Loading – HI-LO

(g)Two Level Block Loading – LO-HI

Figure 2: Examples of selective variable amplitude load spectra [21]

SELECTIVE VARIABLE AMPLITUDE LOADING A selective variable amplitude load spectrum (figure 2) is defined as a constant amplitude spectrum with few load variations like single overload/underload, multiple overload/underload and their combinations in different sequences. These selective load variations are the building blocks of the (aircraft) service spectrum [22]. In the research presented here, the selective variable amplitude spectrum has been utilized

Crack growth in FMLs under VA Loading

843

to develop the understanding of crack growth and delamination behaviour of FMLs under aircraft spectra. These loading spectra can be categorized for understanding in the following groups:

Crack length, a

aD

ΔaOL

aOL NCA ND OL-Cycle No. of cycles, N

Crack growth rate da/dN

ND

aOL

ΔaOL

Crack length, a

Figure 3: Schematic of delayed retardation of crack growth following a single OL in a K-controlled test on metals [21]

Single and Multiple overloads Application of a single overload in the constant amplitude (CA) baseline cycles causes a load interaction effect detailed schematically in figure 3. This entire mechanism is known as crack growth retardation. The magnitude and extent of the retardation is measured by the parameters defined in figure 3, i.e. the number of delay cycles ND , the delay distance aD and the OL-effected crack growth increment ∆aOL . Similar results and phenomena are present in case of multiple overloads. In case of single/multiple overloads the main parameter influencing crack growth retardation is the overload ratio (ROL = S OL /S max ). Thus, increase in ROL will result in an increase in ND , aD and ∆aOL , and a reduction in the da/dN level.

844

S. U. Khan et al.

Blocks of overloads A block of overloads give more severe crack growth retardation during following CA baseline cycles than a single overload. The retardation also occurs more rapidly than for a single overload. Similar to single overloads, the retardation effect is amplified by increasing the ROL . On the other hand, an increase in the number of overload cycles NOL enhances the retardation by extending the ND period, while the ∆aOL distance remains similar to a single overload [23, 24]. Two-level block sequences A high-low load sequence produces similar results as those obtained under a block of overload cycles, except that retardation in fatigue crack growth is always immediate and not preceded by the acceleration phase while in case of low-high sequence, the start of the high amplitude block is typically accompanied by crack growth acceleration followed by a gradual reduction until the new steady state is attained [25, 26]. In some fatigue crack growth tests (figure 2-e), only the mean load level is changed step-wise, whereas the load amplitude remains the same in each block. In such a case, a reduction of the mean load causes an immediate drop in crack growth rate below the level related to CA loading with the lower mean and crack growth retardation or even a temporary crack arrest [27].

DIGITAL IMAGE CORRELATION Digital image correlation (DIC) is a image evaluation technique used to track the surface displacements of deforming materials. DIC is a fully non-contact and nondestructive measurement technique. DIC is based on the comparison of two images acquired using a CCD camera at different defined states, one before the deformation (known as reference image) and the other one after the deformation (known as deformed image) [28, 29]. DIC computes the displacement of grid points on the specimen surface using these two images. DIC has been widely used to examine the deformations of engineering materials including monolithic metals, F MLs and adhesive. For current research, a DIC tool developed at Delft University of Technology is used. This tool is programmed in Matlab and is a multi-purpose tool. It is independent of type of test, the amount of images per test or the size and shape of the grid as long as only in-plane deformations. This DIC tool is described in detail in [30, 31].

EXPERIMENTAL PROGRAM Fatigue crack growth experiments on monolithic aluminium 2024 − T 3, laminated aluminium 2024−T 3 sheets and GLARE 3−5/4−0.4 with cross-ply fibre orientation

Crack growth in FMLs under VA Loading

845

Figure 4: Cross section of materials used

have been performed. The thickness of specimen in these materials was selected to be equal (i.e., 2 mm). The thickness of monolithic metal specimens was 2 mm while for metal laminate and GLARE, metal layer thickness was 5 × 4 = 2mm, excluding adhesive and fibres. The cross section of these three materials is given in figure 4. Center-crack tension (CCT) specimens, illustrated in figure 5, are used for the three above mentioned materials. The starter notches are made by drilling a hole of 3 mm diameter with two saw cuts oriented perpendicular to the loading direction. The total length of the starter notch (2a0 ) is approximately 5 mm.

Test Matrix Load variations are applied on a CA baseline spectrum with a maximum stress S max = 100 MPa and a stress ratio R = 0.1. The single overload spectrum has an overload SOL = 145 MPa at aOL =11-15 mm. The multiple overload spectrum has three overloads i.e. SOL1 = 145 MPa; SOL2 = 130 MPa and SOL3 = 115 MPa at 9, 12 and 19 mm respectively. Block loading spectra are constituted of two stress levels S max1 = 100 MPa and S max2 = 140 MPa and vice versa with stress ratio R = 0.1. Detail of these tests are given in table 1.

Test Equipment & Procedure The tests were conducted in lab air at room temperature on a closed loop mechanical and computer controlled servo-hydraulic testing system with a load capacity of 25 metric tons. The test frequency was 10 Hz. The testing setup is shown in figure 6.

846

S. U. Khan et al.

Ø10 3

240

5

120

All dimensions are in mm 25

70

70

Figure 5: Center-crack tension specimen geometry

RESULTS & DISCUSSION Delamination Study Delamination shapes of different overload cases were measured using the DIC technique. Figure 7-(a) shows the comparison of delamination between CA and single overload case. For the CA case, an elliptical shape was observed during and at the end of the test, while for the single overload case, a certain change in the shape was observed after the application of overload. Both shapes (i.e., CA and single OL) matched at the end near the crack but the the case of single overload, the shape showed a sharpness. Similar behaviour was observed in case of multiple overloads, shown in figure 7-(b). This change in delamination can be due to the S OL which is illustrated in figure 8. After the application of OL, crack growth stopped for a certain number of cycles ND while the specimen was under fatigue loading. The delamination already present around the crack kept growing perpendicular to the crack due to the cyclic fatigue loading. After ND cycles, crack started to grow and new delamination was formed around this crack length increment∆a. The shape of new delamination was different than the delamination shape before OL. The differences between both the shapes mainly depends on ROL and ND . Similarly, the delamination shape of the block load case (HI − LO) (figure 7-(c)) shows the sharp edge after changing the load from HI − LO. When the load is changed from HI − LO, crack growth retardation occurs and the affect the delamination shape as illustrated in the figure 8. On the other hand, the block load case with LO − HI sequence acts as two CA load cycles. The delamination shape in this case remained elliptical till the end of the test. One more observation is that the delam-

Crack growth in FMLs under VA Loading

A1 A2 A3 A4 A5

B1 B2 B3 B4 B5

C1 C2 C3 C4 C5

Monolithic Aluminium CA cycles Load variation Maximum stress Stress ratio [MPa] Constant amplitude 100 0.1 Single overload 100 0.1 Multiple overload 100 0.1 Block loading-LO-HI 100 0.1 Block loading-HI-LO 145 0.1 Laminated Aluminium sheets CA cycles Load variation Maximum stress Stress ratio [MPa] Constant amplitude 100 0.1 Single overload 100 0.1 Multiple overload 100 0.1 Block loading-LO-HI 100 0.1 Block loading-HI-LO 145 0.1 GLARE3 5/4-0.4 CA cycles Load variation Maximum stress Stress ratio [MPa] Constant amplitude 100 0.1 Single overload 100 0.1 Multiple overload 100 0.1 Block loading-LO-HI 100 0.1 Block loading-HI-LO 145 0.1

847

Load variation [MPa] 145 145,130,115 145 100 Load variation [MPa] 145 145,130,115 145 100 Load variation [MPa] 145 145,130,115 145 100

Table 1: Fatigue crack growth test matrix

ination shape for the HI sequence is larger than the LO sequence. This can be the result of the higher S max . Alderliesten et. al [14] has reported a change in delamination shape after application of an OL. They have done few tests using single and multiple over load cases. During their tests, sharpness in the delamination shape was quite clear due to the higher overload ratio ROL = 1.7 compared to present case (i.e., ROL = 1.45). They explained this change in delamination shape due to stress reversal. Plokker [12] has also observed changes in delamination shape after the application of the OL. Delamination shape of CA test (figure 9-(a)) is given in order to highlight the effect of V A load sequences. Figure 9-(b) shows his etched specimen with delamination shape, after the application of single OL. The sharpness and change due to the load application is visible in figure 9-(b). In case of block load sequences no change was observed for LO − HI sequence (Figure 9-(c)). For HI − LO and block of overloads cases a change in shape has been observed after the load transition (Figures 9-(d) and 9-(e)). Now the question arises whether this change in delamination shape and area has any effect on the crack growth and whether it is a significant phenomenon to be considered during fatigue crack growth prediction. Effect of change in delamination

848

S. U. Khan et al.

Figure 6: Testing setup shape is illustrated using figure 10. The delamination shape after the overload is smaller than the the one observed in case of CA. However, in case of crack growth prediction for OL case (discussed later), even with the elliptical delamination shape (as observed in CA case), a match is observed between the test data and prediction results. In order to fully understand the effect of delamination shape change, further investigation is required. Plastic Zone Study Plastic zones created by the overload cycles were calculated for monolithic metal, metal laminate and F MLs. The plastic zone size is calculated using the plastic zone relation (Equation1) given in [32]. Subsequently, the DIC technique is used to measure the plastic zone sizes for all of these materials. The output of the DIC software is shown in figure 11 for all three materials. Figure 12-(a) shows the method used to measure the plastic zone size from the data acquired with the DIC. Figure 12-(b) shows the comparison of plastic zone contours acquired using DIC technique for monolithic 2024 − T 3, laminate aluminium sheets and F MLs for CA and OL cases. The detailed plastic zone size comparison of calculation and measurement is given in table 2. !2 1 Ktip ry = · (1) π σy In case of monolithic metal and metal laminates the calculated plastic zone is smaller than the measured one using DIC. This means that predictions made using only the

Delamination length - b [mm]

Crack growth in FMLs under VA Loading

849 CA CA (Final) S OL (Before S OL(At OL) S OL (Final)

4

2

0

10

5

20

15

Crack length - a [mm]

(a) Delamination length -b[mm]

6 5

M OL (Before OL) M OL (After OL1) M OL (After OL2) M OL (Final)

4 3 2 1 0

15

10

5

Crack length -a[mm]

Delamination length -b [mm]

(b) 10

HI-LO (HI Sequence) HI-LO (LO Sequence) LO-HI (LO Sequence) LO-HI (HI Sequence)

8

6

4

2 5

10

15

Crack length -a [mm]

20

25

(c)

Figure 7: Delamination growth tests results: a. Constant amplitude and single overload, b. Multiple block loading, c. Two block loading

DELAMINATION SHAPE

DELAMINATION EXTENSION AFTER OVERLOAD

DELAMINATION EXTENSION

SAW-CUT

CRACK

PLASTIC ZONE DUE TO OVERLOAD

CRACK AFTER OL DELAMINATION SHAPE AFTER OVERLOAD

Figure 8: Illustration of delamination extension after application of OL

850

S. U. Khan et al.

DELAMINATION SHAPE

FATIGUE CRACK

(a) DELAMINATION SHAPE (before OL)

OVERLOAD APPLIED DELAMINATION SHAPE (After OL)

(b)

DELAMINATION SHAPE LO-HI TRANSITION

FATIGUE CRACK

(c) DELAMINATION SHAPE (HI Cycles)

HI-LO TRANSITION DELAMINATION SHAPE (LO Cycles)

FATIGUE CRACK

(d)

DELAMINATION SHAPE (before OLs)

OVERLOADS APPLIED DELAMINATION SHAPE (After OLs)

FATIGUE CRACK

(e)

Figure 9: Delamination shape: a Constant amplitude, b Single overload, c Block load (LO-HI), d Block load (HI-LO), e Block of loads

Crack growth in FMLs under VA Loading

851

CA DELAMINATION SHAPE APPLICATION OF OVERLOAD DELAMINATION EXTENSION

DELAMINATION SHAPE AFTER OVERLOAD

SAW-CUT

FATIGUE CRACK

Figure 10: Effect of VA load sequence on delamination shape

plastic zone calculation combined with Paris’s crack growth relation (Wheeler) will be erroneous. This mismatch between the calculation and measurement also show the existence of other overload related phenomena. On the other hand, for F MLs the difference between the calculated and measured plastic zone size is quite small. This may allow to use a plastic zone relation in combination with a crack growth law. Other observation from the DIC measurement is the almost constant plastic zone size of F MLs, while in the case of monolithic and metal laminate plastic zone kept increasing with the increase in crack length. From equation 1, it is known that plastic zone is directly related to crack-tip stress intensity factor (Ktip ). In monolithic metal and metal laminate, Ktip depends on crack length and increases with the corresponding increase in crack length. But in case of F MLs the plastic zone size remained constant. Increase in the crack length, for the same stress, increases the Ktip . However, increase in crack length results in the increase of number of fibres bridging the crack and fibre bridging stress intensity factor Kbr . According to equation 2, Ktip will remain almost constant, resulting in constant plastic zone size. Ktip = K f f − Kbr

(2)

Fatigue Crack Growth Predictions The comparison between single OL prediction and the tests result is shown in figure 13. The LDA, being a non-interaction model, is unable to predict the retardation effect whereas the yield zone model is able to predict crack growth retardation in agreement with the experimental data. An interesting observation from figure 13 is that there is hardly any difference between the yield zone model’s prediction and experimental data after the application of overload. In other words, the number of delay cycles are almost equal to the plastic zone size, which seems to indicate that it is only the plastic zone that cause the retardation. In case of pure monolithic metals,

852

S. U. Khan et al.

Figure 11: DIC Outputs: (a) Monolithic Metal - CA, (b) Monolithic Metal - OL, (c) Laminated Aluminium sheets - CA, (d) Laminated Aluminium sheets - OL, (e) FML - CA, (f) FML - OL

Crack growth in FMLs under VA Loading

853

8

Distance from crack - [mm]

6

4

2

Plastic zone size measured

0

2

4

Crack location 14

12

10

8

Distance from Center of Specimen - [mm]

(a) 8.

Constant Amplitude Overload

Metal Laminate

Distance from crack - [mm]

6

Monolithic 2024-T3 4

2

FML 0.

2

Crack location

4 8

9

10

11

12

13

14

15

Distance from Center of Specimen - [mm]

(b) Figure 12: Plastic zone size : (a). Measurement technique, (b). Comparison of Plastic zone measurement for CA and OL (Monolithic, Aluminium laminate and FML)

854

S. U. Khan et al.

A1 A1 A2 A2 A3 A3 A3 A3

Monolithic Aluminium 2024-T3 Crack length Plastic zone size Calculated Measured (DIC) [mm] [mm] [mm] 11.25 0.95 1.25 13.5 1.134 2.2 11.25 0.95 1.5 11.25 (at OL) 1.987 3.0 6 0.504 1.25 6 (at OL1 ) 1.06 3.0 9 0.76 2.0 9 (at OL2 ) 1.28 3.0

B1 B1 B2 B2 B3 B3 B3 B3 B5 B5

Laminated Aluminium Sheets 13 1.092 16 1.344 15 1.26 15 (at OL) 2.65 9 0.84 9 (at OL1 ) 1.77 11 1.04 11 (at OL2 ) 1.76 12.6 2.226 12.6 (HI-¿LO) 1.0586

1.6 2 1.8 4.0 1.2 2.2 1.5 3 4.6 2.2

GLARE3 5/4-0.4 C1 C1 C2 C2 C3 C3 C3 C3 C5 C5

12.75 19.25 11.1 11.1 (at OL) 9.25 9.25 (at OL1 ) 12.3 12.3 (at OL2 ) 13.2 13.2 (HI-¿LO)

0.2 0.2 0.2 0.41 0.2 0.37 0.2 0.25 0.35 0.2

0.25 0.25 0.25 0.5 0.25 0.4 0.25 0.35 0.4 0.25

Table 2: Plastic zone measurement

predictions with a yield zone model are quite inaccurate [33]. However, for FMLs, the error is limited because of the intact fibres which restrain the crack opening leading to the small plastic zone formation which is comparable to the one calculated by yield zone model. In single overload case (figure 13), the crack growth rate gets back to its original level depending on the magnitude of S OL and ROL . It is known from the literature that in metals the retardation region is highly influenced by the magnitude of S OL , and similar behaviour is seen in case of FMLs. For the predictions, semi-elliptical delamination shape (observed in CA case) is assumed around the crack. The match between the test results and prediction show that there is hardly any effect of change in delamination shape due to load variation.

Crack growth in FMLs under VA Loading 0

Crack growth rate [µm/cycle]

10

855

Experimental data

LDA’s predictions

10-1

Yield zone model predictions 10

-2

0

6

12

18

Crack length [mm]

Figure 13: Fatigue crack growth tests prediction - Single overload

CONCLUSION DIC, a relatively new technique is used for measuring plastic zone sizes and delamination shapes in F MLs. The load variations slightly change the delamination shape. This change in delamination shape is very insignificant and while fatigue crack growth prediction, even assuming a semi-elliptical delamination shape, a good correlation is obtained between the test results and predictions. Plastic zones created by overloads are predicted using a simple theory and measured using DIC. In case of monolithic metal and metal laminates, a big difference is observed between the predicted and measured sizes. On the other hand, a small difference is present between the predicted and measured plastic zone size. The plastic zone size calculated as well as measured for F MLs is almost constant at any crack length which is corresponding to the constant Ktip observed for F MLs. Finally, a simple interaction model having plastic zone calculation embedded in crack growth law has been used to predict the fatigue crack growth for F MLs. The correlation between test data and prediction shows that change in delamination shape has only minor or insignificant effect on the crack growth. Linear damage accumulation predictions are also compared with the yield zone model predictions just to highlight that by adding the plastic zone calculation, the model is able to predict well for the selective load variations.

856

S. U. Khan et al.

References [1] Marissen, R. (1988) Fatigue crack growth in arall - a hybrid aluminiumaramid composite material. Technical report. Delft University of Technology, Delft,LR-574. [2] Takamatsu, T., Matsumura, T., Ogura, N., Shimokawa, T., and Kakuta, Y. (1999) Fatigue crack growth properties of a glare3-5/4 fiber/metal laminate. Engineering Fracture Mechanics, 63, 253–272. [3] Takamatsu, T., Matsumura, T., Ogura, N., Shimokawa, T., and Kakuta, Y. (2003) Evaluation of fatigue crack growth behaviour of glare3 fiber/metal laminates using compliance method. Engineering Fracture Mechanics, 70, 2603– 2616. [4] Guo, Y. J. and Wu, X. R. (1998) A theoretical model for predicting fatigue crack growth rates in fibre-reinforced metal laminates. Fatigue & Fracture of Engineering Materials and Structures, 21, 1133–1145. [5] Guo, Y. J. and Wu, X. R. (1999) Bridging stress distribution in center-cracked fiber reinforced metal laminates: modelling and experiment. Engineering Fracture Mechanics, 63, 147–163. [6] Alderliesten, R. C. (2007) Analytical prediction model for fatigue crack propagation and delamination growth in glare. International Journal of Fatigue, 29, 628–646. [7] Guo, Y. J. and Wu, X. R. (1999) A phenomenological model for predicting crack growth in fiber-reinforced metal laminates under constant-amplitude loading. Composite Science and Technology, 59, 1825–1831. [8] Schijve, J., Wiltink, F. J., and Van Bodegom, V. J. W. (1994) Flight-simulation fatigue tests on notched specimens of fiber-metal laminates. Technical Report Report No. LRV-10. Delft University of Technology, The Netherlands. [9] Guo, Y. J. and Wu, X. R. (2002) Fatigue behaviour and life prediction of frml under ca and va loading. Fatigue & Fracture of Engineering Materials and Structures, 25, 417–432. [10] Marissen, R. (1984) Flight simulation behavior of arall. Engineering Fracture Mechanics, 19, 261–277. [11] Kawai, M. and Hachinohe, A. (2002) Two stress level fatigue of unidirectional fml hybrid composites: Glare2,. International Journal of Fatigue, 24, 567–580. [12] Plokker, H. M. (2005) Crack closure in glare. Master’s thesis. Delft University of Technology, The Netherlands.

Crack growth in FMLs under VA Loading

857

[13] Woerden, H. J. M. (1998) Variable amplitude fatigue of glare 3. Preliminary master thesis. Delft University of Technology, The Netherlands. [14] Alderliesten, R. C. and Woerden, H. J. M. (2003) Load history effects during fatigue crack propagation in glare. In Guillaume, M. (ed.), Fatigue of Aeronautical structures as an Engineering Challenge, pp. 509–530. [15] Marissen, R. (1988) Fatigue crack growth in arall - a hybrid aluminiumaramid composite material. Technical report. Delft University of Technology, Delft,LR-574. [16] Roebroeks, G. H. J. J. (1991) Towards GLARE - The Development of a fatigue insensitive and damage tolerant aircraft material. PhD thesis Delft University of Technology, Delft. [17] Vlot, A. and Gunnink, J. (2001) Fibre Metal Laminates-An introduction. Kluwer Academic Publishers, Dordrecht, The Netherlands. [18] Vlot, A., Vogelesang, L. B., and Vries, T. F. (1999) Toward application of fibre metal laminates in large aircraft. Aircraft Engineering & Aerospace Technology, 71, 558–570. [19] Khan, S. U., Alderliesten, R. C., and Benedictus, R. (2008) Post-stretching induced stress redistribution in fibre metal laminates for increased fatigue crack growth resistance. Composite Science and Technology, 69, 396–405. In Press. [20] Alderliesten, R. C. (2005) Fatigue crack propagation and delamination growth in Glare. PhD thesis Delft University of Technology, Delft. [21] Skorupa, M. (1998) Load interaction effects during fatigue crack growth under variable amplitude loading-a literature review. part i: Empirical trends. Fatigue & Fracture of Engineering Materials and Structures,, 21, 987–1006. [22] Schijve, J. (1981) Prediction of fatigue crack growth in 2024-t3 alclad sheet specimen under flight-simulation loading. Technical report. Delft University of Technology, Delft,LR-574. [23] Mills, W. J. and Hertzberg, R. W. (1976) Load interaction effects on fatigue crack propagation in 2024-t3 aluminum alloy. Engineering Fracture Mechanics, 8, 657–667. [24] Chen, G. L. and Roberts, R. (1985) Delay effects in aisi 1035 steel. Engineering Fracture Mechanics, 22, 201–212. [25] Sehitoglu, H. and McDiarmid, D. L. (1980) Effect of load stepdown on fatigue crack arrest and retardation. International Journal of Fatigue, 2, 55–60.

858

S. U. Khan et al.

[26] Ward-Close, C. M., Blom, A. F., and Ritchie, R. O. (1989) Mechanisms associated with transient fatigue growth under variable-amplitude loading : an experimental and numerical study. Engineering Fracture Mechanics, 32, 613– 638. [27] Glinka, G. and Molski, K. (1980) Fatigue crack growth retardation under constant amplitude and variable mean stress. International Journal of Fatigue, 2, 105–111. [28] Sutton, M. A., R., M. S., Helm, J. D., and Chao, Y. D. (2000) Advances in twodimensional and three-dimensional computer vision. Photomechanics, Topics Applied Physics, 77, 323–372. [29] Corr, D., Accardi, M., Graham-Brady, L., and Shah, S. (2007) Digital image correlation analysis of interfacial debonding properties and fracture behavior in concrete. Engineering Fracture Mechanics, 74, 109–121. [30] Lemmen, H. J. K., Alderliesten, R. C., Benedictus, R., Hofstede, J. C. J., and Rodi, R. (2008) The power of digital image correlation for detailed elasticplastic strain measurements. In Nikolinakou, M. K., Tsekouras, G., Gekas, V., and Pavlou, D. G. (eds.), New aspects of engineering mechanics, strucutres and enginereing geology, pp. 73–89. WSEAS WSEAS Press, Athens. [31] Rodi, R., Campoli, G., Alderliesten, R. C., and Benedictus, R. (2009) Characterization of the crack tip behavior in fibre metal laminates by means of digital image correlation. 50th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, Palm Springs, CA., pp. 1–18. AIAA. [32] Schijve, J. (2009) Fatigue of Structures and Materials, second edition edition. Springer Science + Business Media, B.V. Amsterdam. [33] Khan, S. U., Alderliesten, R. C., Schijve, J., and Benedictus, R. (2007) On the fatigue crack growth prediction under variable amplitude loading. In Pavlou, D. G. (ed.), Computational and experimental analysis of damaged materials, pp. 77–105. Research Signpost, Transworld Research Network, Kerala.

25th ICAF Symposium – Rotterdam, 27–29 May 2009 Leave Four Blank Lines Before the title

EFFECT OF RESIDUAL STRESSES FROM SHOT PEENING ON FATIGUE STRENGTH AND THRESHOLD TO CRACK PROPAGATION OF AL 7475 ALLOY COMPONENTS Leave two Blank lines G. Ratti1, U. Mariani1 , M. Giglio2, M. Guagliano2 1 blank between names and affiliations 1 Fatigue Department, AgustaWestland 2 Politecnico di Milano, Department of Mechanical Engineering Leave three Blank lines Before the abstract Abstract: Shot peening is a mechanical process which cold works the surface of a structural part by means of a propelled stream of spherical shots. It is used to improve the fatigue properties of the part by introducing on the surface and in a small layer underneath beneficial compressive stresses which retards or sometimes prevents fatigue cracking. In the presence of shot peening a surface flaw is partially or totally embedded in the compressive stress field. When considering the traditional LEFM methodology, this compressive stress field is the main source of the closure of the crack, which therefore grows at a highly reduced rate. The same closure effect is also present when considering the threshold stresses for non propagation of cracks emanating from the original flaw, so that the detrimental effect due to the presence of the flaw is reduced. The present paper shows the effect of shot peening for a typical aluminum alloy (Al7475-T7351) on the fatigue limit and the threshold to propagation of a semicircular surface flaw 0.36 mm deep, for two different peening intensities (A6 and A12) and for three different stress ratios (-1, 0.1, 0.5) of the external load. The results have been obtained experimentally and are part of a wider program whose aim is the optimization of the shot peening parameters versus fatigue strength and threshold to crack propagation of small defects. According to the current activity the beneficial effect of the shot peening reduces with increasing stress M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 859–870. © Springer Science+Business Media B.V. 2009

G. Ratti, U. Mariani, M. Giglio, M. Guagliano

860

ratio. Furthermore, while the fatigue limit of pristine material is not affected by the Almen intensity, the threshold stress to propagation of surface cracks increases with increasing peening intensity. Leave two blank Lines before first section

INTRODUCTION One blank line Shot peening is a mechanical treatment that consists of impacting the metal surface of a part with a propelled stream of thousands of high speed small spheres [1, 2]. This cold working treatment allows to increase the fatigue limit of the material, thanks to the residual stress field and work-hardening induced on the surface and in a small layer underneath. The impact of the spheres generates stresses that, within this small layer, exceed the yield stress of the material and, as a consequence, cause its permanent plastic deformation. The other layers are stressed below the yield value and, at the end of the treatment, try to return in their initial position, but this process is inhibited by the plastic deformation of the plastic-deformed layer. Because of this, a compressive residual stress field is generated on the surface and in a small layer underneath, while tensile residual stresses are present in the other layers of material. Globally, the residual stress field is self-equilibrated. The most important problem related to the shot peening treatment consists of the difficulty relating the main parameters of the process with the residual stresses induced and to the beneficial effects induced in the peened materials. Currently the only parameters used in industry to calibrate the process are: •

•

the peening intensity, which is expressed as the arc height of a shot peened test strip at saturation and is called Almen intensity. The intensity is a function of the mass of the shot, its hardness, its velocity and its angle of impact on the peened surface of the strip to the shot stream. The arc height is measured on standard test strips made of spring steel. the coverage, which is the amount of the surface that has been impacted by the peening media. The measure of the coverage is given in terms of percentage of surface impacted. Complete coverage or saturation means that 100% of the surface has overlapping peening dimples. Coverage above 100% is a multiple of the exposure time necessary to achieve saturation. A treatment duration of two times the time needed to have 100% coverage means that the coverage is 200%. In this investigation this latter value was chosen and fixed.

Almen intensity and coverage are not directly related to the residual stress field induced by shot peening and two elements can be peened with the same Almen intensity (obtained by using different shots and impact parameters) but having different induced residual stresses [3]; that is to say that the Almen intensity cannot give useful information about the residual stress profile in the material and, consequently, gives only an approximate indication about the fatigue strength

Effect of shot peening on fatigue and threshold to crack propagation of Al 7475 alloy

861

improvement [4-6]. Besides, it is well known that, to date, a quantitative and general criterion able to take into account the residual stress induced by shot peening on fatigue strength has not been developed. This is also true when specimens containing surface defects acting as pre-crack are considered. A recent paper about nitrided and shot peened steels containing surface defects shows that the improvement of the fatigue behaviour induced by shot peening can be attributed not only to residual stresses but also to surface work-hardening [7] and there is not much data that refers to shot peened parts with surface defects acting as pre-cracks. In this paper an Al alloy (Al7475-T7351), typically used for helicopter parts manufacturing, is considered and the effect of shot peening on the fatigue strength and short crack propagation threshold is investigated. Small micro-holes obtained by means of electro-erosion acting as pre-existing defects were introduced in the specimens and fatigue tests considering different combinations of shot peening parameters - flaw sizes were carried out, the aim being to assess the influence on the fatigue behaviour of the in-depth residual stress trend with respect to the depth of the defect. In particular, a test campaign was carried out to show, for two different peening intensities (A6 and A12) and for three different stress ratios (-1, 0.1, 0.5) of the external load: • the effect of shot peening on the fatigue strength of the pristine material and on the threshold to propagation of a semicircular surface flaw 0.36 mm deep; • the measurement of the shape and the magnitude of compressive stress field on the surface and in the small layer underneath of cycled and non cycled coupons in order to evaluate the extent of relaxation of the value of the residual stresses induced from the treatment after different load cycles. This test campaign is part of a wider program whose aim is the optimization of the shot peening parameters versus fatigue strength and threshold to crack propagation of small defects.

TEST DETAILS One blank line Measurement of the compressive residual stress field The measurement of the residual stresses was carried out by using an X-ray diffractometer AST X3000. The stress measurements were carried out with a 2 mm collimator, using n° 5 positive angles ψ and n° 5 negative angles ψ, plus ψ = 0°. The cathode used for the measures is a Chromium one. The radiation is, in both cases, the Kα. The current intensity is equal to 6 mA. The measurements were conducted in geometry Ψ (Psi): the angles θ and Ψ are, respectively, set up and measured in the same plane; the diffraction peaks were found with 2 sensors PSD symmetrically disposed regarding the X-ray source: for every measure it was

862

G. Ratti, U. Mariani, M. Giglio, M. Guagliano

measured the peak of diffraction in 11 angle positions, for sin2 Ψ = 0°, ±16.7°, ±24.0°, ±29.9°, ±35.1°, ±40.0°. The determination of the diffraction peaks was executed with the "Cross Correlation" method. The time of exposure to the X-ray is equal to 20s for each angle of exposure. Determination of the fatigue strength and the threshold to propagation of defects Fatigue tests were carried out on smooth specimens using an axial resonant testing machine with 100 kN maximum load (see Figure 1). Tests were carried out at constant amplitude loads (CAL) according to ASTM E466 and the test frame was verified to guarantee the alignment according to ASTM E1012. The fatigue specimens were hourglass shaped with circular ends. Figure 2 shows the fatigue specimen drawings and a picture of the same specimen.

Figure 1

Figure 2

Effect of shot peening on fatigue and threshold to crack propagation of Al 7475 alloy

863

RESULTS OF THE TESTS Compressive residual stress field Table I summarises the value of the measured principal stresses and the angle Φ of the principal directions with respect to the directions of measure for non cycled specimens. Shot peening intensities are 6 and 12 Almen respectively.

Table I As shown, Almen intensities higher than 6A do not imply larger residual stresses, both on the surface and in the depth. The parameter that is mainly influenced by Almen intensity is the depth of the compressive stress field induced by the treatment and the residual stresses distribution. This effect confirms for this alloy what is widely reported in the literature for steel and titanium alloys [4-6, 8, 9]; as shown in these papers, the stress distribution presents, for higher Almen intensity, a maximum compressive residual stress more close to the external surface and with higher intensity, not directly reported in the measured values in Table I. Tables II and III summarise the values of the measured compressive principal stresses for specimens submitted to 30000 loading cycles for stress ratios of R=-1 and R=0.1 and shot peening intensities of 6 and 12 Almen respectively.

Table II

Table III

G. Ratti, U. Mariani, M. Giglio, M. Guagliano

864

Also shown in this case, higher Almen intensities do not imply necessarily larger residual stresses on the surface (Table III) , but in Table II the different stress distribution at different Almen intensity is clear, in particular in depth measures. Furthermore, by comparing the results obtained with and without cycling the parts (see Figure 3), it is possible to see that the relaxation due to cycling has a negligible effect for this alloy within the range of stress ratios.

Figure 3 Fatigue strength and threshold to propagation of defects Fatigue limit tests were carried out using 'stair-case' sequences. Specimens were considered ‘run-out’ if they survived 12·106 fatigue cycles. Tests were carried out at different stress ratios, namely R=-1, R=0.1 and, for the flawed specimens, also R=0.5, for two shot peening intensities (6 and 12 Almen). Results of stair-case sequences were analysed with the method of Hodge and Rosenblatt. According to this method, the mean fatigue limit is determined as the average value of the stress amplitude levels during the ‘up and down’ sequence. Tests of “not peened” specimens (pristine and flawed) were also carried out in order to define a baseline for the results of the activity. Figure 4 shows, for pristine material, the trend of the fatigue endurance limit with varying the stress ratio R and the Almen intensity. The following results can be highlighted for this alloy: 1.

the beneficial effect of the peening reduces with increasing the value of the stress ratio, it is negligible at positive R ratios, in agreement with literature results;

2.

the improvement of the fatigue strength is no longer present with increasing the Almen intensity beyond the Almen intensity of 6A.

Effect of shot peening on fatigue and threshold to crack propagation of Al 7475 alloy

865

Figure 4 The result highlighted in the first point can be explained on the basis of the redistribution of the stress induced by shot peening during the fatigue load application, more evident for high applied stress [10, 11]. The result highlighted in the second point can be explained considering that, as shown in Table I, Almen intensities higher than 6A do not imply larger residual stresses, both on the surface and in the depth. Since fatigue strength is dominated by the status of the surface, this being similar in the two cases, no difference in the endurance limit can be expected. Figure 5 shows, for a material with a semicircular surface flaw 0.36mm deep, the trend of the fatigue endurance limit while varying the stress ratio R and the Almen intensity. The following results can be summarized for this alloy: 1.

the beneficial effect of the peening is present up to very high values of the stress ratio, even though it tends to reduce with increasing the value of R;

2.

the fatigue strength increases with increasing the Almen intensity.

Figure 5

G. Ratti, U. Mariani, M. Giglio, M. Guagliano

866

The result highlighted in the first point can be explained considering that the compressive stress field contribute to sustain the plasticity-induced fatigue crack closure at the crack tip. A simple approach present in literature [12] suggests to consider this contribution as a modification of the stress ratio applied, which becomes:

R new =

σ min − σ res σ max − σ res

(1)

where σres is the value of the compressive residual stress field. This approach considers a linear superposition of the residual stress field on the actual stress cycle, obtaining an effective stress cycle with a new R ratio (eq. 1); this simple theory does not consider the stress redistribution and the crack tip plasticity, which modifies the stress field. In particular, as the value of the stress ratio increases, the plasticity-induced fatigue crack closure at the crack tip is reduced (no closure for R→1) and the contribution of σres becomes less and less. The result highlighted in the second point can be explained considering that the threshold to propagation of the crack emanating from the original flaw is dominated by the value of the ΔKTH along the crack tip. It is embedded in the compressive stress field, whose magnitude is different in the depth and therefore its contribution to sustain the plasticity-induced fatigue crack closure is different along the crack tip. The higher Almen intensity being characterized by a thicker compressive layer allows to contribute better than the lower Almen intensity to the closure of cracks whose size is such that the bottom part of the crack tip is embedded in the part of the compressive layer where the higher Almen intensity is still effective. This assumption is in good agreement with the hypothesis of different residual stress distribution at high Almen intensity. This is shown in Figure 6, where the residual stress trends due to 6A and 12A Almen intensity are shown. In this case: -

crack A is not influenced by the different Almen intensity since it is embedded into a compressive stress layer which is not influenced by its depth;

-

crack B is deeply influenced by the peening intensity since the bottom part if its tip is embedded into a effective compressive stress layer for the higher intensity and beyond the compressive stress layer for the lower intensity;

Effect of shot peening on fatigue and threshold to crack propagation of Al 7475 alloy

6A

867

12A

Figure 6

-

crack C is little influenced by the Almen intensity since even though part of its tip is embedded into compressive stress layers which are different for the various peening intensities, its bottom part is in both cases beyond the compressive stress layer.

For this alloy and Almen intensities investigated, the crack size object of the test, being characterized by a depth of 0.36 mm, is a B type, for which the difference in intensity has the highest effect on its threshold to propagation. Test results confirm this fact. As shown in Figure 7 in fact, the compressive stress layer finishes in correspondence of the bottom of the tip of the initial crack and the growth of the crack initiates from that point.

G. Ratti, U. Mariani, M. Giglio, M. Guagliano

868

Initial defect size

Crack growth nucleation

Compressive stress layer

Figure 7

CONCLUSIONS The present paper shows the effect of shot peening on the fatigue limit and on the threshold to propagation of a semicircular surface flaw 0.36 mm deep for a typical aluminum alloy (Al7475-T7351), for two different peening intensities (A6 and A12) and for three different stress ratios (-1, 0.1, 0.5) of the external load. The results have been obtained experimentally and are part of a wider program whose aim is the optimization of the shot peening parameters versus fatigue strength and threshold to crack propagation of small defects. The most important problem related to the shot peening treatment consists in fact of the difficulty relating the main parameters of the process with the beneficial effects induced in the peened materials. Currently the only parameters used in industry to estimate the effect of the treatment are the Almen intensity and the coverage. A coverage of 200%, which corresponds to twice the exposure time to have a complete surface coverage, is a typical value for critical applications. This parameter is therefore fixed to 200% and does not become an object of the investigation.

Effect of shot peening on fatigue and threshold to crack propagation of Al 7475 alloy

869

The first step of the investigation activity was the measurement of the shape and the magnitude of compressive stress field on the surface and in the small layer underneath of cycled and non cycled coupons with varying Almen intensity in order to evaluate the extent of relaxation of the value of the residual stresses induced from the treatment after different load cycles. The activity showed that the shape and the magnitude of the residual stress field is not dependant on the Almen intensity between 6A and 12A, and that the relaxation of these stresses due to cycling is negligible. The only parameter influenced by the different peening intensity is the thickness of the residual stress field and this has been demonstrated by the results of the second step of the investigation. The second step of the investigation activity, the evaluation of the effect of peening intensity on the fatigue properties of the Al7475-T7351 alloy, showed that: • •

for the pristine material the beneficial effect of the shot peening reduces with increasing stress ratio, it is negligible at positive R ratios and is independent of the peening intensity in the range 6A – 12A. for the flawed material the beneficial effect of the peening is present up to very high values of the stress ratio, even though it tends to reduce with increasing the value of R and increases with increasing Almen intensity.

As a conclusion of this activity it is possible to affirm that using peening intensities higher than necessary does not imply a higher beneficial effect on the fatigue properties of typical Al alloys and that this effect reduces with increasing the mean stress, becoming negligible for positive stress ratios. A different behaviour can be seen in the case when a flaw is present in the material. In this case the threshold to propagation of cracks emanating from the flaw is deeply influenced by the relationship between the thickness of the compressive stress field and the flaw depth. Within this context the peening intensity being the main parameter to define the thickness of the compressive stress layer, a relationship between this intensity and the threshold to propagation of defects can be found. For the material and the peening intensities investigated it can be seen that the peening intensity has an influence on the threshold to propagation of surface defects whose depth is in the range 0.2 < a < 0.7mm, which is the typical range of small defects of helicopter dynamic components.

ACKNOWLEDGMENTS The authors would like to thank the specialists of AgustaWestland and Politecnico di Milano Test Laboratories and in particular Massimo Ciprandi for his high contribution in carrying out the experimental work.

870

G. Ratti, U. Mariani, M. Giglio, M. Guagliano

REFERENCES 1.

K.J. Marsh: Shot Peening: Techniques and Applications. (EMAS, London 1993). 2 Almen, J. O.; Black, P.H., Residual stresses and fatigue in metals (McGraw-Hill Publ. Company) (1963). 3. R. Herzog, W. Zinn, B. Scholtes, H. Wohlfahrt, The significance of Almen intensity for the generation of shot peening residual stresses, Proceedings of the Sixth International Conference on Shot Peening, ICSP6, San Francisco, CA, 1996, 270-281. 4. M. Guagliano: Relating the Almen intensity to residual stresses induced by shot peening: a numerical approach, J Mater Proc Tech, 110, 277-286, 2001. 5. W. Cao, R. Fathallah, L. Castex: Correlation of Almen arc height with residual stresses in shot peening process, Mater Sci Tech, 1995, 11, (9) 967-3. 6. M. Guagliano, L. Vergani, M. Bandini, F. Gili “An approach to relate the shot peening parameters to the induced residual stresses”, Proc. ICSP7 (7th International Conference on Shot Peening) (Ed. A. Nakonieczcy), IMP – Warsaw, 1999, pp. 274.282 7. I. Fernández Pariente, M. Guagliano “About the role of residuals tresses and surface work hardening on fatigue ΔKth of a nitrided and shot peened low-alloy steel, Surface and Coatings Technology, 202, 2008, pp. 30723080. 8. B. Scholtes, O. Vöhringer “Ursachen Ermittlung und Bewertung von Randschichtveränderungen durch Kugelstrahlen, Mat.- wiss u. Werchstofftech 24, 421, 1993. 9. B. Scholtes “Eigenspannungen in mechanisch Randschichtverformten Werkstoffzuständen”, DGM Informationgesellschaft, Verlag, Obersursel, 1991. 10. M.L. Aggarwal, V.P. Agrawal, R.A. Khan.”, A stress approach model for predictions of fatigue life by shot peening of EN45A spring steel” International Journal of Fatigue, 28, 2006, pp. 1845-1853. 11. M.A.S. Torres, H.J.C. Voorwald “An evaluation of shot peening, residual stress and stress relaxation on the fatigue life of AISI 4340 steel”, International Journal of Fatigue, 24, 2002, pp. 877-886. 12. Y. Murakami, Metal Fatigue: Effects of Small Defects and Nonmetallic Inclusions (Elsevier) (2002) 35-74.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

EXPERIMENTAL VALIDATION OF STRESS INTENSITY FACTOR SOLUTIONS FOR THE PIN LOADED LUG D.R. Child1, N. J. Moyle2 and A.F. Grandt3 1

Royal Australian Air Force, Directorate General Technical Airworthiness 2 Royal Australian Air Force, Tactical Fighter Systems Program Office 3 School of Aeronautics & Astronautics, Purdue University, USA

Abstract: Predicting crack growth behaviour is an important element of managing the structural integrity of aircraft fleets. It is common for fleet managers to employ software packages, such as AFGROW, to automate the prediction of crack growth behaviour. Stress intensity factor solutions employed in these software packages must be appropriately verified in order for results to be utilised in structural integrity management. Two independent research programs were conducted at Purdue University to review stress intensity factor solutions for cracked, pin-loaded lugs. The purpose of the research was to experimentally validate stress intensity factor solutions derived from representative StressCheck® Finite Element Models (FEMs). Component fatigue tests were conducted on corner and through cracked, pin loaded aluminium lugs and non-dimensional stress intensity factor (geometry factor) solutions calculated utilising a backtracking method. For the through crack configuration, experimental and StressCheck® geometry factors closely correlated and were an improvement to the solutions provided in AFGROW at the time. Based on the results of this research, StressCheck® pin-loaded lug geometry factors have since been incorporated into the AFGROW software. In the case of the corner crack configuration, the correlation between experimental and StressCheck® results was dependant on the choice of pinM. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 871–898. © Springer Science+Business Media B.V. 2009

872

D. R. Child, N.J. Moyle and A.F. Grandt

loading boundary conditions in the FEM. The results of this research have been utilised to define the default AFGROW software pin-loaded lug boundary condition assumptions.

INTRODUCTION Pin-loaded lug joints are widely used in aircraft structure, often at critically loaded components and where ease of structural disassembly is required. Examples within the aircraft industry include fuselage to wing connecting joints, engine mounting connections and control rod assemblies. In this scenario the likelihood of crack initiation and crack growth due to manufacturing defects, maintenance damage, and in-service stresses is high, thus the determination of fatigue lives for these components is an important consideration to adequately manage the structural integrity of the aircraft. The originating requirement to conduct this research was driven by the US Air Force Research Laboratory (AFRL) who, at the time of the research, was responsible for the ongoing development of ‘AFGROW’, a fracture mechanics program used for crack growth analysis [1]. Discrepancies with pinloaded lug geometry factors were known to exist, in particular for geometries approaching that of a pin loaded plate such that the solution was only considered accurate for a small range of W/D ratios (approximately 2–4) [2]. Here, as indicated in Figure 1, W is the plate width and D is the pin-hole diameter. In an effort to rectify these shortcomings, stress intensity factor solutions were developed for the through and corner cracked pin-loaded lugs using the finite element program StressCheck®. This program had been proven successful when used for previous AFRL research relating to components with multi-site damage (MSD). The objectives of the research reported in this paper were to validate, via fatigue test, the geometry factor solution derived from StressCheck® for through and corner cracked pin-loaded lug geometries, and to prepare the geometry factor solutions for inclusion into the AFGROW program. Fatigue testing also considered the oblique crack case although FEM comparison of the oblique crack case was not conducted. This work was undertaken as two independent Masters of Science research programs at Purdue University, USA during the period 2003-2006. Full details of the research work is provided in [3] and [4]. Research Approach for Through-Cracked Lugs Stress intensity factors for through cracked pin-loaded lugs were obtained experimentally by cyclically loading cracked specimens of four different W/D ratios. Geometry factors were derived using a backtracking method based on specimen material properties and observed crack propagation rates. This process allowed for the comparison of experimental geometry factors to those derived using StressCheck®. Equations were fitted to the StressCheck® data to give through-crack geometry factors as a function of the non-dimensional crack length

Validation of stress intensity factors for the pin loaded lug

873

and W/D ratio. These equations have since been included in the AFGROW code so that crack growth can be predicted for a single-cracked pin-loaded lug of any W/D ratio with any specified initial crack size. Research Approach for Corner-Cracked Lugs As for the through crack case, stress intensity factors for corner cracked pin-loaded lugs were obtained experimentally by cyclically loading pre-cracked specimens with two different W/D ratios. However, unlike the through crack case, corner crack propagation could not be observed visually during the test as the crack path was obstructed by the pin. As such, a marker banding technique was utilised to measure crack propagation rates. Geometry factors were derived using a backtracking method based on specimen material properties derived from Compact Tension (CT) specimen tests and crack propagation rates calculated from measurement of the marker bands from the fracture surface. Corner crack front geometry factors were compared to StressCheck® for two different pin to lug loading and boundary condition assumptions. As a result of the research, default pin to lug boundary conditions and boundary condition controls, including user guidance have been applied to the AFGROW code. StressCheck® Finite Element Software StressCheck® by Engineering Software Research & Development Inc. is a finite element (FE) program having the functionality to model cracks and estimate fatigue parameters for a specified geometry [5]. StressCheck® utilises p-version finite elements and the contour integral method to calculate Mode I and Mode II stress intensity factors. The research reported in this paper supported verification of StressCheck® models that were created by AFRL and Analytical Processes Engineered Solutions (APES) for the through-cracked and corner-cracked lug.

NUMERICAL MODEL FOR THROUGH-CRACKED LUGS StressCheck® Model StressCheck® FE models were created by AFRL with the geometry shown in Figure 1 and key characteristics detailed in Table I. The loading pin was modelled using a pre-defined ‘Spring’ loading condition which can be used for the analysis of fastener load transfer. In this case, the loading pin is modelled as a fixed rigid core surrounded by a neat fitting spring with stiffness derived from the elastic solution of a disc under radial compression [6]. A unit load is applied to the upper edge of the lug as shown in Figure 1. The StressCheck® solution accounts for local yielding of the lug and models the interaction between the pin and lug as nonlinear, thus an iterative process is required to obtain a converged solution. A circular contour path was chosen for all FE models which surrounded ‘sacrificial’ triangular elements located at the crack tip and passed through an adjacent layer of quadrilateral elements.

D. R. Child, N.J. Moyle and A.F. Grandt

874

σo

• • •

Loading-pin diameter 12.7mm (0.50in) Width varied to give W/D ratio Unit stress on upper edge Pin-load using ‘Spring’ boundary conditions (cosine squared)

Boundary Conditions

• •

‘Neat’ fitting pin Core of loading-pin fixed

Analysis

• •

Plane stress condition No pin bending or friction effects

•

Steel Pin. E = 206.8GPa (30Msi), ν = 0.3 Aluminum Lug. E = 68.9GPa (10Msi), ν = 0.33 P-type element analysis Triangular elements at crack tip surrounded by quadrilateral elements Contour path through quadrilateral elements

Loads

Material Properties

• • •

Elements •

Table I – Properties of the StressCheck® model for the through-cracked lug

8W

• Geometry

W Rigid Core Fixed BC

c max

c θ

D

Figure 1 – Geometry of the StressCheck® model for the through-cracked lug

StressCheck® Solution Mode I stress intensity factors were calculated by AFRL using the StressCheck® model for W/D ratios of: 1.5, 1.75, 2, 2.5, 3, 4, 5, 6, 7, 9.5, 12, 16, 20, 40 and for non-dimensional crack lengths ranging from 0.025cmax to 0.98cmax. Geometry factors were calculated using Eqn. 1 where KI is the Mode I stress intensity factor, σb is the bearing stress as defined in Eqn. 1, c is the crack length and β is the geometry factor. Geometry factors were defined as a function of the bearing stress as this was the standard format being used in AFGROW. Geometry factors for a sample of W/D ratios plotted against non-dimensional crack length are shown in Figure 2. K I = βσ b πc , σ b = σ o W

D

(1)

Validation of stress intensity factors for the pin loaded lug

875

12 W/D = 1.5

10

Beta

W/D = 2.0 8

W/D = 3.0

6

W/D = 4.0

4 2 0 0.0

0.2

0.4

c/cmax

0.6

0.8

1.0

Figure 2 – Geometry factors derived from the StressCheck® model for the through-cracked lug. AFGROW Implementation Estimates of geometry factors were required at the lower and upper limits of the crack growth path (crack length approaching zero and crack approaching the free edge) to ensure AFGROW could be appropriately coded and that crack growth could be predicted from any initial crack length to component failure. Using the same approach as in [7], [8] and [9] the crack geometry was considered equivalent to an edge-crack in a semi-infinite plate as the crack length approached zero. As a function of the stress at the crack initiation site, this configuration yields a geometry factor of 1.1215. This was then converted to a function of bearing stress by multiplying by a stress concentration factor (maximum stress to bearing stress). As the crack approaches the free edge of the lug the crack growth rate is expected to increase asymptotically. This has been predicted by the finite element method in [8] and superposition of fundamental solutions in [10].Guidance from AFRL was to set a geometry factor intercept at the free edge to a very large value; a procedure used for other AFGROW geometry factor solutions. This guidance was taken into account when curve-fitting the StressCheck® solution as detailed below. Range of W/D ratios considered. For large W/D ratios the lug geometry approaches that of a pin-loaded plate of finite width. A solution developed by Ball [10] was used to estimate the lug geometry at which the W/D ratio has negligible influence on the Mode I geometry factors. This was found to occur at a W/D ratio of approximately 100. For the purposes of implementing geometry factor solutions into AFGROW a lower limit of 1.5 and upper limit of 40 was set as this was expected to be the maximum range of any W/D ratio used for aircraft components. Curve fitting the StressCheck® solution. The preferred method for AFGROW implementation was to derive a set of two-parameter equations to give the geometry factor as a function of W/D and non-dimensional crack length. To reduce

D. R. Child, N.J. Moyle and A.F. Grandt

876

residual error the StressCheck® data set was split into seven regions and an equation fitted to data in each region. Equation constants were automatically determined by the curve fitting software so as to minimize residual error. For the majority of data points this was less than 1% and reached a maximum of approximately 3.5%. Equations were checked to ensure a smooth transition as the non-dimensional crack length approached the free edge and that there was a definite intercept at the free edge. Results of the curve fitting process including equations, constants, residual error and a surface plot for each region are detailed in [3].

EXPERIMENTAL PROCEDURE FOR THROUGH-CRACKED LUGS Specimen Material Lug specimens were manufactured from Al7075–T651 and Al2024–T3 aluminium alloy with material properties shown in Table II. Crack growth properties for Al7075–T651 specimens were taken from previous research in [11]. Crack growth properties for Al2024–T3 specimens were based on a one segment Paris equation fit to Handbook data in [12], supplemented by material property tables used in AFGROW for the HarterT crack growth method.

Material Ultimate Strength Yield Strength Young’s Modulus Stock Thickness Loading Direction

7075–T651 Bare 475.7MPa (69ksi) 7.95mm (0.313in) L–T Walker

da = C[ K max (1 − R ) m ] n dN Crack Growth Properties [coefficients a function of in/cycle]

2024–T3 Bare 441.3MPa (64ksi) 330.9MPa (48ksi) 67.6GPa (9.8Msi) 6.35mm (0.250in) L–T Paris

da = C (ΔK ) n dN

2.07e-8 < da/dN < 4.70e-6 in/cycle C = 3.99e-13, n = 8.60412, m = 0.6

Fitted to Fig 7.5.3.1.11 of [12] (R=0.05)

4.6185e-6 < da/dN < 9.5e-5 in/cycle C = 6.46e-08, n = 2.26572, m = 0.6

1.0e-6 < da/dN < 1.5e-4 in/cycle C = 2e-9, n = 3.3469

9.5e-5 < da/dN < 1.436e-3 in/cycle C = 5.41e-11, n = 4.46639, m = 0.6

HarterT, R = 0.05 Pre-defined crack growth rate table within the AFGROW program.

Table II – Material properties for through-crack lug specimens. Specimen Geometry A total of twelve specimens with four different geometries (designated ‘Lug 1’ to ‘Lug 4’ as shown in Table III) were tested to failure. Specimens were

Validation of stress intensity factors for the pin loaded lug

877

manufactured of sufficient height to ensure a uniform stress field remote from the lug ligament and were all pre-cracked with an initial single through-crack approximately 1.27mm (0.050in) in length at the location shown in Figure 1. All specimens were polished for improved resolution of crack tip measurements via optical microscope. Specimens were loaded using precision ground dowel pins with a diameter tolerance of less than +0.025mm (+0.001in). Pin clearances for Lug 1, Lug 3 and Lug 4 specimens was 0.051mm (0.002in) allowing the loading pin to be easily pushed into the lug-hole without any movement in the transverse direction. Pin clearance for Lug 2 specimens was 0.127mm (0.005in) allowing a small amount of transverse movement once the pin was located in the lug-hole. A small amount of Dow-Corning Molykote™ paste was applied to the lower surface of the lug-hole prior to insertion of the loading-pin to delay the onset of fretting cracks. Despite lubrication of the loading-pin, fretting cracks initiated during fatigue testing for Lug 3 and Lug 4 specimens. This is discussed in further detail below.

Lug Specimen W/D Ratio Width Lug-hole Diameter [mm] Lug-hole Diameter [in] Pin-diameter Thickness Material Quantity Tested

Lug 1

Lug 2

Lug 3

Lug 4

1.765

3.0

8.0

12.0

44.83mm (1.765in) 25.400 (-0.000, +0.051) 1.000 (-0.000, +0.002) 25.400mm (1.000in) 6.223mm (0.245in)

57.15mm (2.25in) 19.050 (-0.000, +0.127) 0.750 (-0.000, +0.005) 19.050mm (0.750in) 6.223mm (0.245in)

50.80mm (2.00in) 6.350 (-0.000, +0.051) 0.250 (-0.000, +0.002) 6.350mm (0.250in) 6.223mm (0.245in)

76.20mm (3.00in) 6.350 (-0.000, +0.051) 0.250 (-0.000, +0.002) 6.350mm (0.250in) 5.715mm (0.225in)

7075–T651

7075–T651

7075–T651

2024–T3

3

3

3

3

Table III – Geometry details for through-crack lug specimens Loading Frame A specialized loading frame was used to apply cyclic loads to the loading-pin whilst allowing for the optical measurement of fatigue cracks. The viewing window had a maximum width of 127mm and sufficient height to allow for the observation of the entire lower portion of the lug specimen. Loading frames were placed on either side of the specimen, which were connected at the base via a single fixture. This was in turn connected to a double pivoted ‘alignment fixture’ to minimize any out of plane loading. Spacer blocks were placed between the loading frames to prevent contact with the specimen, and shaft collars were attached to

D. R. Child, N.J. Moyle and A.F. Grandt

878

either end of the loading pin to prevent its movement during fatigue testing. A schematic of the loading frame is shown in Figure 3. Fatigue testing Pin-loading was calculated so that a full range of crack growth rates would be observed during testing. In conjunction with this requirement, loads were reduced as much as possible to minimize the likelihood of fretting crack initiation whilst achieving specimen fracture within a reasonable number of cycles. Fatigue testing was conducted in accordance with ASTM Standard E647 [13] under constant amplitude loading with an R ratio of 0.05. Crack lengths were measured on both the front and back face of the specimens using 40X traversing Bausch and Lomb optical microscopes. Crack length measurements were made at crack growth intervals of approximately 0.254mm (0.01in) to an accuracy of 0.0254mm (0.001in). Measurements were taken with the specimen under 15% maximum cyclic load to assist with crack tip resolution. Three specimens of each lug geometry were fatigue tested to failure, which was considered to occur when the crack reached the free edge.

Upper Hydraulic Grip Lug Specimen Pin-support Blocks (Front & Rear) Loading Pin Loading Frames (Front & Rear)

Alignment Fixture Hydraulic Grip Fixture Lower Hydraulic Grip

Figure 3 – Loading fixture for fatigue testing of through-cracked lug specimens.

Figure 4 – Typical primary crack (right side) and fretting crack (left side) geometries for through-cracked lug specimens.

Validation of stress intensity factors for the pin loaded lug

879

EXPERIMENTAL RESULTS FOR THROUGH-CRACKED LUGS Fatigue cracks generally propagated in a negative θ direction (as defined in Figure 1) with an average slope of –10o. Since this follows a direction perpendicular to principal stresses, it can be concluded that fatigue cracks propagated under Mode I loading. Crack growth measurements were taken as the horizontal component, which induced an error in crack length measurements of less than 3%. For Lug 3 and Lug 4 specimens, secondary (fretting induced) cracks initiated at the lug-hole opposite the primary crack prior to specimen fracture. Fretting cracks typically initiated at negative θ angles on the edge of the lug hole (as seen on the left hand side of Figure 4). Despite the initiation of fretting cracks for Lug 3 and Lug 4 specimens, crack growth data was successfully obtained for all lug geometries. Determination of Experimental Geometry Factors Crack growth rates were calculated for each specimen using the five point incremental polynomial method detailed in [13]. Material crack growth rate properties specified in Table 2 were used to determine ΔK values for each crack growth rate data point. A back-tracking method proposed by James & Anderson [14] was used to determine geometry factors as a function of non-dimensional crack length (c/cmax) for each lug specimen. For Al2024–T3 specimens, ΔK values were calculated using the Paris equation constants shown in Table 2. Experimental geometry factors were then calculated using Eqn. 2, where β is the experimental geometry factor, ΔK is stress intensity factor range, Δσbr is the applied bearing stress range, and c is the crack length corresponding to the experimental data point.

β=

ΔK Δσ br πc

(2)

For each lug geometry a plot of the geometry factor (β) as a function of nondimensional crack length is shown in Figure 5 to Figure 8. In each of these plots, experimental geometry factors are compared to the StressCheck® solution, and to the geometry factors that were used in AFGROW at the time of the research (hereafter referred to as the ‘Original AFGROW Solution’). The point at which fretting cracks initiated in Lug 3 and Lug 4 specimens is shown in the plots where relevant.

D. R. Child, N.J. Moyle and A.F. Grandt

880

4.0 3.5 3.0

Beta

2.5 2.0 1.5

W/D 1.765

Specimen L1-001 Specimen L1-002

1.0

Specimen L1-004 StressCheck

0.5

Original AFGROW

0.0 0.0

0.2

0.4

0.6

0.8

1.0

c/c max

Figure 5 – Comparison of Experimental geometry factors to Original AFGROW and StressCheck® solutions for Lug 1 specimens.

1.6 1.4 1.2 1.0 Beta 0.8 0.6 Specimen L2-001

0.4

Specimen L2-002

0.2

StressCheck

W/D 3.0

Specimen L2-004

Original AFGROW

0.0 0.0

0.2

0.4 c/c max 0.6

0.8

1.0

Figure 6 – Comparison of Experimental geometry factors to Original AFGROW and StressCheck® solutions for Lug 2 specimens.

Validation of stress intensity factors for the pin loaded lug

881

0.7 Specimen L3-001

W/D 8.0

Specimen L3-003

0.6

Specimen L3-002 StressCheck

0.5

Original AFGROW

0.4 Beta

Fretting Crack Fretting Crack

0.3 0.2 0.1 0.0 0.0

0.2

0.4

c/c max

0.6

0.8

1.0

Figure 7 – Comparison of Experimental geometry factors to Original AFGROW and StressCheck® solutions for Lug 3 specimens. 0.40 Specimen L4-002 Specimen L4-003

0.35

W/D 12.0

Specimen L4-004 StressCheck

0.30

Original AFGROW

0.25

Fretting Crack

Beta

Fretting Crack

0.20

Fretting Crack

0.15 0.10 0.05 0.00 0.0

0.2

0.4

c/c max

0.6

0.8

1.0

Figure 8 – Comparison of Experimental geometry factors to Original AFGROW and StressCheck® solutions for Lug 4 specimens.

D. R. Child, N.J. Moyle and A.F. Grandt

882

Accounting for Fretting Cracks The presence of fretting cracks during fatigue testing effected the crack growth of the primary crack and thus had an impact on the experimental geometry factors calculated for each specimen. A comparison of experimental and predicted geometry factors can be conducted to the point of fretting crack initiation, although past this point one can not confirm whether the Original AFGROW solution or StressCheck® values are a more accurate prediction. For this reason, experimental geometry factors were modified to account for the presence of fretting cracks. Experimental geometry factors were modified using a correction factor based on a solution proposed by Ball [10]. The Ball solution was derived using the superposition method to determine the geometry factor for a point-loaded lug with either a single or double asymmetric through-crack. From this solution, the ratio between the single and double asymmetric geometry factors was applied as a correction factor to the experimental geometry factor, giving a modified experimental factor which could be compared to the StressCheck® solution. Modified experimental geometry factors to account for fretting cracks in Lug 3 and Lug 4 specimens are shown in Figure 9 and Figure 10. 0.7 Specimen L3-003 Modified

0.6

Specimen L3-002 Modified

W/D 8.0

StressCheck

0.5 Original AFGROW

0.4 Beta 0.3 0.2 0.1 0.0 0.0

0.2

0.4

0.6

0.8

1.0

c/c max

Figure 9 – Comparison of Modified experimental geometry factors (to account for fretting cracks) to Original AFGROW and StressCheck® solutions for Lug 3 specimens.

Validation of stress intensity factors for the pin loaded lug

0.40

Specimen L4-002 Modified

0.35

Specimen L4-003 Modified

883

W/D 12.0

Specimen L4-004 Modified

0.30

Beta

StressCheck Original AFGROW

0.25 0.20 0.15 0.10 0.05 0.00 0.0

0.2

0.4

0.6

0.8

1.0

c/cmax

Figure 10 – Comparison of Modified experimental geometry factors (to account for fretting cracks) to Original AFGROW and StressCheck® solutions for Lug 3 specimens. Crack Growth Predictions Crack growth predictions were made for each lug geometry using the StressCheck® and Original AFGROW solution to compare experimental to predicted fatigue lives (Figure 11 to Figure 14). StressCheck® fatigue life predictions used the two-parameter equations that were fitted to StressCheck® geometry factors. The presence of fretting cracks for larger W/D ratios were also accounted for using the geometry factor modification process as described above, to give a ‘double-asymmetric’ crack fatigue life prediction. The initiation of fretting cracks was assumed to occur at a number of cycles equivalent to the experimental average.

D. R. Child, N.J. Moyle and A.F. Grandt

884

1.0

W/D 1.765

0.9 0.8 0.7

c/c

0.6 max

0.5 0.4 0.3

Specimen L1-001 Specimen L1-002

0.2

Specimen L1-004 StressCheck

0.1

Original AFGROW 0.0 0

10000

20000

30000

40000

50000

60000

Cycles 1.0

W/D 3.0

0.9 0.8 0.7 0.6

c/c

0.5

max

0.4 0.3

Specimen L2-001 Specimen L2-002

0.2

Specimen L2-004 StressCheck

0.1

Original AFGROW 0.0 0

20000

40000

60000

80000

Cycles

Figure 11 and Figure 12 – Comparison of observed crack growth to that predicted using the Original AFGROW and StressCheck® geometry factors for Lug 1 and Lug 2 specimens

Validation of stress intensity factors for the pin loaded lug

885

1.0 Specimen L3-001

0.9

Specimen L3-002 0.8

Specimen L3-002-Fretting

0.7

Specimen L3-003 Specimen L3-003-Fretting

0.6

c/c max

StressCheck-Primary StressCheck-Fretting

0.5 0.4

W/D 8.0

0.3 0.2 0.1 0.0 0

20000

40000

60000

Cycles

80000

100000

1.0 Specimen L4-002 0.9

Specimen L4-002-Fretting Specimen L4-003

0.8

Specimen L4-003-Fretting 0.7

Specimen L4-004

0.6

c/c max

Specimen L4-004-Fretting StressCheck

0.5

StressCheck-Fretting 0.4 0.3

W/D 12.0

0.2 0.1 0.0 0

50000

100000

150000

200000

250000

Cycles

Figure 13 and Figure 14 – Comparison of observed crack growth to that predicted using the Original AFGROW and StressCheck® geometry factors for Lug 3 and Lug 4 specimens.

886

D. R. Child, N.J. Moyle and A.F. Grandt

DISCUSSION ON THROUGH-CRACKED LUGS Experimental geometry factors for the Lug 1 geometry were greater than predicted values at small crack lengths and as the crack approached the free edge. However, StressCheck® gave a better prediction than the Original AFGROW solution which was conservative for the majority of crack lengths. For the Lug 2 geometry, the StressCheck® solution was generally conservative compared to experimental data yet was a better prediction than the Original AFGROW solution. For Lug 3 and Lug 4 geometries, StressCheck® and AFGROW solutions were similar and closely correlate with experimental geometry factors to a non-dimensional crack length of approximately 0.6. At longer crack lengths StressCheck® geometry factors were more accurate than the Original AFGROW solution, yet in general remained conservative. Predicted fatigue lives using StressCheck® geometry factors also closely correlated to experimental results, as shown in Figure 11 to Figure 14, with predictions either bounding experimental data or conservative by approximately 15% when compared to the closest experimental data set. Based on the results of this research, StressCheck® pin-loaded lug geometry factors for the throughcracked lug have since been incorporated into the AFGROW software.

NUMERICAL MODEL FOR CORNER-CRACKED LUGS StressCheck® Model For the through crack case, parametric StressCheck® FE lug models were created by APES under contract to the ‘Air Force Unitized Metallic Structures Program – External K Solver Project’ to provide K solutions for lug problems. The FE models were meshed using two different, three dimensional element types. The crack tip was surrounded by five sided pentahedron elements which were enclosed by a second layer of six sided hexahedron elements. The remainder of the model was meshed using both pentahedron and six sided hexahedron elements. The geometry parameters were modified during the research to match the geometry of the lug test specimens. Two load and boundary conditions were applied to the FEMs for comparison with the experimental results. For the first set of conditions (Model 1) a remote unit stress was applied to the free end of the lug. The remote stress was reacted at the lug hole surface using the pre-defined ‘Spring’ loading condition as described for the through-crack numerical model. For the second set of conditions (Model 2) the free end was clamped and the StressCheck® default bearing load, being a sinusoidal traction, was applied to the lug hole. Key characteristics of the models are detailed in Table IV.

Validation of stress intensity factors for the pin loaded lug

Model

Model 1 – SC-Spring

Geometry

• •

Loads

Boundary Conditions

•

• •

Material Properties

Elements

Model 2 – SC-Bearing

Varied to meet the two W/D ratios tested.

Unit stress on upper edge Pin load using ‘Spring’ boundary conditions ‘Neat’ fitting pin Core of loading pin fixed •

Analysis

887

•

Sinusoidal traction on hole surface

•

Free end clamped

Ratio of plane stress to plane strain determined by StressCheck®

• •

Steel Pin, E = 206.8 GPa (30Msi), υ = 0.3 Aluminium Lug, E = 68.9GPa (10.5Msi), υ = 0.33

• •

P-type element analysis Pentahedron elements at crack tip surrounded by hexahedron elements Contour path through hexahedron elements

•

Table IV - Properties of the StressCheck® numerical models for the corner-cracked lug.

EXPERIMENTAL PROCEDURE FOR CORNERCRACKED LUGS Marker Banding As corner crack growth in the pin loaded lugs could not be measured visually, an alternate method of measuring crack growth was required. The marker banding technique has been widely applied to measure crack growth and allows crack growth to be measured directly from the fracture surface. Notably, the technique had been successfully applied to 7075-T651 lugs [15], 7050-T7451 lugs [16] and BS2L65 and 7075-T7351 lugs [17]. The marker banding approach adopted in this research was to intersperse blocks of 3000 constant amplitude crack growth cycles with a sequence of three distinct blocks of a lower maximum load as shown at Figure 15. This approach was shown to provide a traceable pattern such that crack growth between blocks could be measured under optical microscope. During testing, the surface crack length was also monitored optically to assist verify the location of the marker bands. The methodology employed assumed that the marker banding spectrum did not significantly affect ΔK calculations when compared to a constant amplitude spectrum. The effect of the marker banding spectrum was checked by analysis and experiment and shown to be negligible for the crack growth rate range observed during lug testing.

D. R. Child, N.J. Moyle and A.F. Grandt

888

Figure 15 – Schematic of the Marker Band Spectrum used for fatigue testing of corner-cracked lug specimens. Specimen Material All test specimens were manufactured from 6.35mm Al2219-T851 aluminium alloy with material properties shown in Table V. Prior to commencing testing of the lug specimens, crack growth properties for the material utilised were determined via cyclical testing of Compact Tension (CT) specimens machined from the same stock. The CT testing was conducted in accordance with ASTM E 647-00. The purpose of the CT specimen tests was to determine crack growth material properties for the 2219-T851 and to show that the affect of the marker banding spectrum on crack growth calculations was negligible. Four CT specimens were tested to determine the crack growth material properties of the Al2219-T851. Two more were tested to observe the effect of the marker banding spectrum. A four segment Paris Law equation was fitted to the stress intensity factor (ΔK) versus crack growth (da/dN) data as shown in Table V.

Material

2219-T851 (6.35-25.4mm Sheet and Plate)

Ultimate Strength

420.6 MPa (61ksi)

Yield Strength

324.1 MPa (47 ksi)

Bearing Strength, e/D = 1.5

655.0 MPa (95 ksi)

Young’s Modulus

72.4 GPa (10.5 Msi)

Poisson’s Ratio

0.33

Validation of stress intensity factors for the pin loaded lug

889

Paris 4.31x10-7 < da/dN < 1.37x10-6 in/cycle C = 9.903x10-10, m = 3.7772 1.37x10-6 < da/dN < 6.00x10-6 in/cycle C = 1.941x10-9, m = 3.5024

Crack Growth Properties [coefficients a function of in/cycle]

6.00x10-6 < da/dN < 6.13x10-5 in/cycle C = 1.056x10-8, m = 2.7619 6.13x10-5 < da/dN < 4.00x10-4 in/cycle C = 1.681x10-5, m = 5.5814

Table V – Material properties for corner-cracked lug specimens [18]. Specimen Geometry A total of seven lug specimens with two different geometries (as shown in Table VI and Figure 16) were tested to failure. These geometries were designated ‘Lug 1’ and ‘Lug 2’ in [4] but for clarity have been designated ‘Lug 5’ and ‘Lug 6’ in this paper. As for through-cracked specimens, all specimens were manufactured of sufficient height to ensure a uniform stress field remote from the lug ligament and were all pre-cracked with an initial corner crack that had a surface length of 1.016-1.194mm (0.040-0.047in). All specimens were polished to assist with surface crack tip detection.

Lug Specimen

Lug 5

Lug 6

NM-DB-1A,

NM-DB-2A, 2B

1B, 1C and 1E

and 2C

W/D Ratio

2.0

2.667

Width [mm]

50.8

50.8

25.4 (-0.00,

19.1 (-0.00,

+0.08)

+0.08)

25.4

19.1

6.10 – 6.38

6.07 – 6.38

Lug IDs

Lug-hole diameter [mm] Pin-diameter [mm] Thickness [mm] Material Quantity Tested

2219-T851 4*

3

Table VI – Geometry details for corner-crack lug specimens. *Two were considered unrepresentative for further analysis.

Figure 16 – Nomenclature for the corner-cracked lug.

890

D. R. Child, N.J. Moyle and A.F. Grandt

The lug specimens were loaded using precision ground dowel pins. Pin clearance for the specimens was 0.0762mm (0.003in) which allowed the loading pin to be inserted into the lug hole with minimal transverse movement once in situ. A small amount of Dow-Corning Molykote lubricant was applied to the first test specimen. However, it was found the lubricant entered the fracture surface making it difficult to record the location of the marker bands. Therefore, the lubricant was not used on further specimens. Fatigue Testing Fatigue testing was undertaken using a modified version of the loading frame developed for the through-crack case. Fatigue testing was undertaken using the marker banding spectrum and in accordance with ASTM E 647-00. Surface crack growth measurements were taken at the conclusion of each marker band cycle for correlation with fracture surface measurements taken after failure. Crack length measurements were taken at 30% of the maximum cyclic load at short crack lengths and at 20% maximum cyclic load to open the crack slightly allowing for easier measurement. At the conclusion of each test, the fracture surface was cut from the specimen using a jeweller’s saw and mounted on a microscope slide for examination.

EXPERIMENTAL RESULTS FOR CORNER-CRACKED LUGS Fatigue Test Results Using an optical microscope at 200-500X magnification, crack fronts were directly measured from the fracture surface and recorded. Subsequently, ellipses were fitted to the recorded crack growth fronts. Utilising the incremental polynomial method detailed in ASTM E 647-00 Appendix X1, crack growth rates were calculated and correlated against the number of cycles at which the marker band was applied (refer Figure 17). Using the material properties derived from the CT tests and the backtracking technique [14], non-dimensional geometry factors around each crack front were then calculated.

Figure 17 – Example of corner crack front propagation. Solid lines and points represent ellipses fitted to marker bands along the fracture surface. Crack growth rates determined along radials of open circles.

Validation of stress intensity factors for the pin loaded lug

891

In total, 36 corner crack fronts (13 for Lug 5 and 23 for Lug 6), and 54 oblique crack front (23 for Lug 5 and 31 for Lug 6) were analysed. The analysis for each crack front is presented in detail in [4] providing the reader with: the fitted crack lengths and coordinates at which the crack growth, ΔK and β were calculated as well as the constants of the fitted ellipses to enable reproduction of the individual crack fronts. An example of the analysis conducted for each crack front is shown in Figure 18. W=50.8mm

x (mm) y (mm) ΔK (Mpa-m1/2) β

Fitted Crack Front #3 d=25.4mm Δσbr=56.09MPa Fitted Ellipse Constants x0 y0 a0 b0 -0.032 0.210 2.856 3.702 Data Point Coordinates (mm) 2.702 2.495 2.185 1.771 1.255 1.327 1.961 2.552 3.073 3.496 9.707 9.714 9.747 9.767 9.762 2.168 2.170 2.177 2.182 2.181

B=6.37mm

2.818 0.000 9.809 2.191

2.811 0.668 9.738 2.175

R=0.083

0.654 3.784 9.704 2.168

0.000 3.899 9.559 2.135

Figure 18 – An example of the crack front data calculated and presented for 36 corner crack fronts and 54 oblique crack fronts. Comparison of Experimental, AFGROW and FEM Geometry Factors To enable comparison between experiment, AFGROW and StressCheck® derived geometry factors, all ΔK values obtained were converted to β using Eqn. 3. This equation uses bearing stress (σbr) as the reference stress and lug thickness (B) as the reference dimension.

β=

ΔK Δσ br B

(3)

Both StressCheck® and AFGROW results assumed that crack fronts are fitted by constrained ellipses. That is, the crack front emanates perpendicularly from the lug face and hole surfaces. The crack fronts observed experimentally did not behave in this manner and were better fitted with unconstrained ellipses. Therefore, in order to compare between experimental and numerical results, the crack front ellipses analyzed in StressCheck® and AFGROW were based on the experimental fitted

D. R. Child, N.J. Moyle and A.F. Grandt

892

surface crack length (c) and the through thickness crack length (a) rather than the unconstrained ellipses fitted to the experimental crack fronts. For the experimental results, ΔK was calculated at nine evenly spaced locations around the corner crack front. AFGROW provided two sets of stress intensity factor solutions on the crack front, ΔKc and ΔKa evaluated at 5 degrees and 80 degrees respectively. In the StressCheck® models, ΔK was evaluated at 25 locations along the crack fronts. Numerical values for the StressCheck® results were taken at approximately 5 and 80 degrees from the lug surface due to then known issues associated with StressCheck’s® ability to calculate ΔK close to free surfaces [19]. This was also in keeping with the AFGROW solution constraints and defined consistent reference locations between specimens. Comparisons of the geometry factor for experimental, AFGROW and StressCheck® models were presented (refer Figure 19 for an example) and a summary of the geometry factors is detailed in Table VII to Table X. These tables summarize geometry factors taken at the lug surface (βc) and hole surface (βa) for the experiment and at approximately 5 and 80 degrees from the two surfaces for the StressCheck® and AFGROW results. The ‘Difference’ columns represent the percentage difference between the numerical and experimental results with low differences highlighted. Comparisons were not made between the experimental oblique crack fronts and the AFGROW or StressCheck® solutions as neither program had an oblique crack model readily available for use. 3.5

3.0

2.506

2.424

2.5 2.215

2.208 1.979

2.0

θ

1.847

β 1.5

1.0 SC - Springs SC - Bearing

0.5 Experimental AFGROW

0.0 0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Normalized Angle measured anti-alockwise around Crack Front (θ)

Figure 19 – Comparison of Non-Dimensional Geometry Factors for one of the Lug 5 specimens (NM-DB-1C, Crack Front #5).

Validation of stress intensity factors for the pin loaded lug

Crack Front B (mm) a (mm) c (mm) Lug 1C CF1 6.10 3.02 1.60 6.38 3.38 2.21 Lug 1A CF1 6.10 3.73 2.06 Lug 1C CF3 6.38 3.91 2.82 Lug 1A CF3 6.10 4.34 2.62 Lug 1C CF5 6.38 5.00 3.78 Lug 1A CF6 Lug 1C CF7 6.10 4.93 3.23

893

Lug 1 - βc a/B Experiment AFGROW Difference SC-Spring Difference SC-Bearing Difference 2.010 1.553 -22.7% 1.654 -17.7% 2.088 0.496 3.9% 2.117 1.733 -18.1% 1.727 -18.4% 2.156 0.530 1.8% 2.101 1.750 -16.7% 1.823 -13.2% 2.301 0.613 9.5% 2.191 0.614 1.886 -13.9% 1.861 -15.1% 2.315 5.7% 0.713 2.208 1.971 -10.7% 1.981 -10.3% 2.502 13.3% 2.270 2.256 2.135 2.666 17.4% 0.785 -0.6% -5.9% 0.808 -2.0% -4.8% 2.248 2.203 2.139 2.706 20.4%

Table VII – Summary of non-dimensional surface crack (c) geometry factors for Lug 5 specimens. Crack Front B (mm) a (mm) c (mm) Lug 1C CF1 6.10 3.02 1.60 6.38 3.38 2.21 Lug 1A CF1 6.10 3.73 2.06 Lug 1C CF3 6.38 3.91 2.82 Lug 1A CF3 6.10 4.34 2.62 Lug 1C CF5 6.38 5.00 3.78 Lug 1A CF6 Lug 1C CF7 6.10 4.93 3.23

Lug 1 - βa a/B Experiment AFGROW Difference SC-Spring Difference SC-Bearing Difference 2.223 1.814 -18.4% 1.477 -33.6% 1.929 0.496 -13.3% 2.118 2.088 1.709 -19.3% 2.234 0.530 -1.4% 5.5% 2.269 2.085 1.650 -27.3% 2.187 0.613 -8.1% -3.6% 0.614 2.135 2.336 9.4% 1.913 -10.4% 2.533 18.6% 0.713 2.215 2.387 7.8% 1.847 -16.6% 2.492 12.5% 2.371 2.817 18.8% 2.180 3.001 26.6% 0.785 -8.0% 0.808 -12.0% 2.330 2.696 15.8% 2.051 2.835 21.7%

Table VIII – Summary of non-dimensional through crack (a) geometry factors for Lug 5 specimens. Crack Front B (mm) a (mm) Lug 2C CF1 6.38 2.92 3.12 Lug 2A CF1 6.38 3.20 Lug 2C CF3 6.38 3.10 Lug 2E CF1 6.07 3.58 Lug 2C CF5 6.38 3.66 Lug 2A CF3 6.38 3.53 Lug 2E CF3 6.07 4.06 Lug 2E CF6 6.07 Lug 2C CF8 6.38 4.37 4.45 Lug 2A CF5 6.38 4.75 Lug 2E CF8 6.07 Lug 2C CF10 6.38 5.00

c (mm) 1.35 1.65 1.68 1.50 2.06 2.24 1.80 2.31 2.74 3.00 2.74 3.23

Lug 2 - βc a/B Experiment AFGROW Difference SC-Spring Difference SC-Bearing Difference 0.458 1.286 0.933 -27.4% 1.047 -18.6% 1.214 -5.5% 1.415 1.017 -28.1% 1.114 -21.3% 1.271 0.490 -10.2% 0.502 1.318 1.019 -22.7% 1.122 -14.9% 1.282 -2.7% 1.231 1.003 -18.5% 1.117 -9.2% 1.284 0.510 4.4% 1.355 1.096 -19.1% 1.178 -13.1% 1.346 0.562 -0.7% 1.477 0.574 1.130 -23.5% 1.191 -19.4% 1.359 -8.0% 1.300 1.076 -17.3% 1.201 -7.7% 1.374 0.582 5.7% 0.669 1.384 1.180 -14.7% 1.280 -7.5% 1.467 6.1% 0.685 1.445 1.209 -16.3% 1.280 -11.4% 1.465 1.4% 1.581 1.247 -21.1% 1.293 -18.2% 1.479 0.697 -6.5% 0.782 1.448 1.259 -13.1% 1.364 -5.8% 1.571 8.4% 0.785 1.485 1.284 -13.5% 1.357 -8.6% 1.671 12.5%

Table IX – Summary of non-dimensional surface crack (c) geometry factors for Lug 6 specimens. Crack Front B (mm) a (mm) 2.92 Lug 2C CF1 6.38 3.12 Lug 2A CF1 6.38 3.20 Lug 2C CF3 6.38 3.10 Lug 2E CF1 6.07 3.58 Lug 2C CF5 6.38 3.66 Lug 2A CF3 6.38 3.53 Lug 2E CF3 6.07 4.06 Lug 2E CF6 6.07 4.37 Lug 2C CF8 6.38 4.45 Lug 2A CF5 6.38 4.75 Lug 2E CF8 6.07 5.00 Lug 2C CF10 6.38

c (mm) 1.35 1.65 1.68 1.50 2.06 2.24 1.80 2.31 2.74 3.00 2.74 3.23

Lug 2 - βa a/B Experiment AFGROW Difference SC-Spring Difference SC-Bearing Difference 0.458 1.257 1.131 -10.0% 0.958 -23.8% 1.108 -11.8% 1.339 1.259 1.065 -20.5% 1.231 0.490 -6.0% -8.1% 1.292 1.263 1.069 -17.3% 1.238 0.502 -2.3% -4.2% 1.283 1.226 1.038 -19.1% 1.200 0.510 -4.4% -6.5% 1.423 1.391 1.170 -17.8% 1.366 0.562 -2.3% -4.0% 1.468 1.449 1.217 -17.1% 1.423 0.574 -1.3% -3.1% 1.382 1.340 1.123 -18.7% 1.309 0.582 -3.0% -5.3% 1.486 1.506 1.249 -16.0% 1.475 0.669 1.3% -0.7% 1.518 1.581 1.311 -13.6% 1.558 0.685 4.2% 2.7% 1.660 1.647 1.366 -17.7% 1.630 0.697 -0.8% -1.8% 1.731 1.629 1.331 -23.1% 1.603 0.782 -5.9% -7.4% 0.785 2.4% 9.2% 1.659 1.699 1.394 -16.0% 1.811

Table X – Summary of non-dimensional through crack (a) geometry factors for Lug 6 specimens.

894

D. R. Child, N.J. Moyle and A.F. Grandt

DISCUSSION ON CORNER-CRACKED LUGS Test Limitations The cracked lug problem is a complex one influenced by lug geometry, relative pin/lug rigidity, pin bearing pressure distribution, the pin to lug fitting conditions, the loading magnitude and direction and the influence of bushings [20] as well as assumptions relating to frictional forces [17] and life improvement techniques such as cold working. During this research there were a number of limitations and assumptions were identified with some of the more important assumptions discussed here. As observed for the through-crack problem, it was found during the tests that the lug fatigue cracks did not propagate perpendicular to the load line. The typical angle of crack propagation was −5 degrees for Lug 5 and –10 degrees for Lug 6. Using finite element analysis in [3], it was determined that a path perpendicular to the principal stresses in the region of the crack had an angle of approximately –14 degrees, suggesting that the experimental fatigue cracks were propagating predominantly under Mode I loading. The negative crack angle has two ramifications when making ΔK comparisons with the finite element models used. The first issue is that experimentally, crack growth has been measured parallel to the fracture surface, not perpendicular to the load line per the finite element models. It was estimated that this would result in a crack length difference of less than 2%. More importantly, by assuming in the FE models that the crack is propagating perpendicular to the load line, potentially a greater amount of Mode II crack tip opening is being introduced than what is occurring experimentally. A preliminary investigation suggested that the Mode II contribution was between 510% of the Mode I value, dependant on the location picked on the crack front, which is consistent with previous research [21]. Two other assumptions relating to the StressCheck® finite element models were the pin fit and the input material properties. Firstly, it was assumed that both the spring constraint and bearing load finite element models provide a surface bearing stress distribution similar to that of a neat fit pin. To ensure a ‘neat fit’ condition was achieved experimentally, lug holes were reamed within -0.000/+0.003 inches of the pin diameter. All lug holes and pin diameters were measured with the greatest difference for any tested specimen being 0.002 inches. Secondly, it was assumed that the 2219-T851 Ex and νxy were 72.4GPa (10.5 Msi) and 0.33 respectively. While no dedicated material testing was carried out during this research a parametric study conducted in StressCheck® indicated that a change of 10% in either parameter had a negligible effect on ΔK. Results for Lug 5 Specimens Tables VII and VIII show that at a through thickness crack length to specimen thickness ratio (a/B) of less than 0.7, Model 1 (SC-Bearing) correlated well with experimental results with an average absolute difference of 5.2% for βc and 7.5%

Validation of stress intensity factors for the pin loaded lug

895

for βa. However, for a/B ratios greater than 0.7, the spring model provided better correlation. The average absolute difference of the spring model at an a/B ratio greater than 0.7 is 7% for βc and 10.0% for βa. This trend suggests that as the crack size grows, the assumptions contained in Model 1 more closely reflect the experimental conditions. However, at short crack sizes, the Model 2 assumptions are more comparable. Results for Lug 6 Specimens Tables IX and X show that at almost all crack sizes up to an a/B ratio of approximately 0.78, the Model 2 assumptions more closely resemble the experimental results with an average absolute difference of 5.1% for βc and 5.4% for βa. The Model 1 assumption began to more closely resemble the experimental results for Lug 6 at an a/B ratio greater than 0.78 for which there were two crack fronts analyzed giving an absolute difference for βc of 7.2%. Common Observations The trend observed in both Lug geometries was that for smaller crack sizes, the bearing load boundary condition (Model 2) provided results that more closely resembled those found experimentally. As crack size increased, the spring constraint boundary condition (Model 1) more closely resembled the experimental results. For both FEMs, the geometry factor at the centre of the crack perimeter was consistently lower than at the edges, especially at larger crack sizes. This trend was not observed by experiment results. Rather, the experimental geometry factor distributions were relatively flat around the crack front with βa becoming noticeably larger than βc, only at larger crack sizes. Except at small crack sizes for the Lug 6 geometry, the results for both FEMs also showed a larger βa than βc with the difference increasing as the crack grew larger. These observations have implications for cracks that may be ‘grown’ using ΔK data from interior locations around the crack fronts. That is, whilst reasonable correlation between the two FEMs and the experimental results were shown over certain crack size ranges, discussion focused on surface or near surface β. Qualitative observation of the whole crack front geometry factor distributions for both lug geometries suggest greater differences at the centre region of the crack front perimeter in both shape and magnitude. Through all the geometry factor comparisons conducted, over the range of crack front sizes analyzed, the two Models produced results between 15% and 30% different to each other. Both of the modelling assumptions would seem ‘reasonable’ to the FEA user and this observation emphasizes the difficulty of the problem at hand. When conducting a lug crack growth analysis a decision will need to be made about which boundary condition assumption (or combination of) should be used or if in fact a different assumption should be considered. For through-cracked lugs, there was good correlation between the spring constraint (Model 1) and experimental results. This conclusion is in keeping with the results

896

D. R. Child, N.J. Moyle and A.F. Grandt

found in this research in that as the corner crack became larger, Model 1 results more closely correlate with experimental results.

CONCLUSION Two independent research programs were conducted at Purdue University to review stress intensity factor solutions for through and corner-cracked, pin-loaded lugs. The purpose of the research was to experimentally validate non-dimensional stress intensity factor solutions (geometry factors) derived from representative StressCheck® finite element models. For the through-cracked lug, two-parameter equations were fitted to geometry factors derived from StressCheck® FE models so that analytical geometry factors could be determined for W/D ratios ranging from 1.5 to 40, and for any crack length through the lug ligament. Fatigue testing was conducted using aluminium alloy lug specimens of four different W/D ratios, and a backtracking method employed to give experimental geometry factors. Despite fretting cracks initiating in lug specimens with large W/D ratios, these were successfully accounted for to enable a comparison of experimental results to the StressCheck® solution and the AFGROW solution in use at the time of the research. Results of this comparison revealed that experimental and StressCheck® geometry factors closely correlated. For any observed differences the StressCheck® solution was generally conservative and was an improvement to the AFGROW solution. For the corner-cracked lug, analytical geometry factors were determined from StressCheck® FE models with two different pin loading conditions. The first model incorporating a Bearing (sinusoidal) load distribution, and the second incorporating a distributed stiffened spring constraint on the hole surface which is modelled to replicate a neat fit steel pin. There was a difference in geometry factor of between 15% and 30% for these two models which confirms that the cracked lug problem is complex given that both modelling assumptions ‘appear reasonable’ to the user. Fatigue testing was conducted using aluminium alloy lug specimens of two different W/D ratios, and a backtracking method employed to give experimental geometry factors for the corner crack case. Verification of the marker banding technique employed in the fatigue test program was also conducted to ensure it did not adversely affect fatigue test results. Comparisons between experimental and StressCheck® derived geometry factors revealed that predicted variations in the geometry factor around the crack front were not observed experimentally. A comparison of StressCheck® pin loading conditions revealed that use of the Bearing loading condition more accurately predicted experimental data up to a non-dimensional lug bore crack length (a/B) of approximately 0.7. At non-dimensional crack lengths greater than 0.7, the Spring loading condition provided closer correlation to experimental results.

Validation of stress intensity factors for the pin loaded lug

897

Based on the results of the research outlined in this paper, StressCheck® pinloaded lug geometry factors for the through-cracked lug have since been incorporated into the AFGROW software. Furthermore, AFGROW now provides an option for the user to select either the Spring or Bearing load condition for the pin-loaded lug, and provides guidance on the situation in which these loading conditions could be used. The AFGROW user manual now states: The default boundary condition in AFGROW is the bearing condition for crack lengths for which (the through thickness crack length to lug thickness ratio) a/t ≤ 0.7. The spring condition is used for a/t ≥ 0.8, and linear superposition is used for intermediate crack lengths. This provided the best correlation to lug test data currently available. Users may change the default condition using the Predict, Preference (Lug Boundary Condition) option. [1]

REFERENCES [1] [2] [3]

[4]

[5]

[6]

[7] [8] [9] [10]

Harter, J. A. (2008), AFGROW Users Guide and Technical Manual, Air Force Research Laboratory. AFGROW Website. (2003). In: AFGROW Website Knowledge Base, http://afgrow.wpafb.af.mil. [Listed Online 11 Jul 2003] Child, D.R (2004), Experimental validation of Mode I stress intensity factors for the single-cracked pin-loaded lug, M.S. Thesis, School of Aeronautics and Astronautics, Purdue University. Moyle, N. (2006), Experimental determination of the Mode I stress intensity factor for a corner-cracked lug using a marker banding technique, M.S. Thesis, School of Aeronautics and Astronautics, Purdue University. Engineering Software Research & Development Inc. Website. (2009), StressCheck Overview, http://www.esrd.com/Software/StressCheck/overview.html. [OnlineAccessed 29 March 2009] Actis, R.L. (2000), Modeling Fastened Structural Connections, Engineering Software Research & Development, Inc. [Online-Accessed 11 March 2003]. Gencoz, O., Goranson, U.G. and Merrill, R.R. (1980). In: International Journal of Fatigue, vol. 2, iss 3, pp. 121-129. Zatz, I.J., Eidinhoff, H.L. and Armen, H. Jr. (1981). In: AIAA 22nd Structures, Structural Dynamics and Materials Conference, pp. 402-415. Brussat, T.R. (1980). In: Engineering Fracture Mechanics, vol. 14, pp. 233-235. Ball, D.L. (1987). In: Engineering Fracture Mechanics, vol. 27, no. 6, pp. 653-681.

898

[11]

[12]

[13]

[14] [15]

[16] [17]

[18] [19]

[20] [21]

D. R. Child, N.J. Moyle and A.F. Grandt

Saunder, T. J. (1991), Fatigue crack growth from coldworked eccentrically located holes in 7075-T651 aluminum, M.S. Thesis, School of Aeronautics and Astronautics, Purdue University. Materials Directorate Wright Laboratory Air Force Material Command (1994), Damage Tolerant Design Handbook – A Compilation of Fracture and Crack Growth Rates for High Strength Alloys, Wright Patterson Air Force Base American Society for Testing and Materials (1994), Annual Book of ASTM Standards, ASTM E647-95a – Standard test method for measurement of fatigue crack growth rates, vol. 03.01. James, L.A. and Anderson, W.E. (1969). In: Engineering Fracture Mechanics, vol. 1, pp. 565-568. Curtin, T., Adey, R., and Brussat, T. (2005), Computational and experimental fracture analysis of a pin-loaded lug, BEASY Software, www.beasy.com [Online-Accessed 20 August 2005]. Kim, J., Lee, S., and Hong, S. (2003). In: Theoretical and Applied Fracture Mechanics vol. 40, pp.135-144. Moon, J. E. (1981). In: Crack growth in pin loaded lugs, Proceedings of the 11th ICAF Symposium – “Aircraft Fatigue in the Eighties”, The Netherlands. Military Handbook 5G (1994), Metallic Materials and Elements for Aerospace Vehicle Structures. Suh, J. (2005), Analysis of fatigue crack growth from countersunk fastener hole, Phd Thesis, School of Aeronautics and Astronautics, Purdue University. Wang, G. S. (1994). In: Journal of Strain Analysis, vol. 29, no. 1, pp. 716. Rigby, R., and Aliabadi, M. H. (1997). In: Engineering Failure Analysis, vol. 4, no. 2, pp. 133-146.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

PROPERTIES OF FATIGUE CRACK PROPAGATION IN FRICTION STIR WELDED 2024-T3 ALUMINUM ALLOY Okada, T.1, Kuwayama, K.2, Fujita, S.2, Asakawa, M.3, Nakamura, T.1 and Machida, S.1 1 2

Japan Aerospace Exploration Agency Graduate School, Waseda University 3 Waseda University

Abstract: Fatigue crack growth rates in an FSW butt joint panel were obtained in order to evaluate the effect of residual stress on crack growth behavior. Two angles between the weld line and loading direction, and two distances between the weld line and the center of the starter notch were employed to investigate the effect of the residual stress field on the crack growth rate and path, and the effect of the initial distance on crack growth acceleration in tensile residual stress field. The test results indicate that the crack growth rate and path on the weld line are merely affected by the angle between the weld line and the loading direction. Furthermore, the acceleration of crack growth was not affected by the distance between the weld line and the initial notch. Under this test condition, the maximum principle stress is not the primary factor determining the direction of crack growth; rather, the direction of the maximum stress range seems to be the primary factor.

INTRODUCTION Friction Stir Welding (FSW) is relatively novel process capable of welding high strength Aluminum alloys. Its application has spread to many structures such as automobiles, trains, rockets, and ships. FSW is expected to be applied to M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 899–908. © Springer Science+Business Media B.V. 2009

T. Okada et al.

900

commercial aircraft primary structures as an alternative to rivet joints to reduce production cost and structural weight. FSW was first applied to a normal category aircraft by Eclipse Aviation in 2006 [1], and its application to a transport category aircraft is planned by other company [2]. The compliance of FSW structures to damage tolerance requirements is thus a subject of intensive research [3–9]. Sato et al compared crack growth behavior in FSW and a riveted panel, and demonstrated the effectiveness of roller burnishing in reducing the tensile residual stress around the welding line to improve the fatigue crack growth behavior in an FSW panel [7]. Lemmen et al. studied the effect of TMAZ and HAZ on crack origins and also the effect of the weld line angle on the origin [8]. Bussu et al. evaluated fatigue crack growth behavior in an FSW butt joint in cases where the weld line was parallel or perpendicular to the loading direction [9]. One of authors has conducted fatigue tests on a riveted joint fuselage scale model [10]. Fuselage bending and cabin pressure were applied to the structure and it was found that the direction of crack growth could incline from the longitudinal direction. On the other hand, the effect of the FSW weld line angle on crack growth behavior has been evaluated by Lim [11] and Lemmen et al. [12], but has not been well understood. This paper describes the results of fatigue crack growth experiments to clarify the properties of fatigue crack growth in an FSW panel. The effects of residual stress on crack growth rate and the redistribution of residual stress during crack growth were investigated. And the effect of an inclined weld line angle on crack growth rate and direction, and the effect of the initial distance on crack growth acceleration in tensile residual stress field are discussed. Additionally, the crack growth rate on a FSW specimen is evaluated using the assumed residual stress distribution and relationship between the crack growth rate on a welded specimen and that on a base material.

EXPERIMENTAL PROCEDURE The specifications of the test specimens made from 2024-T3 Aluminum alloy are shown in figure 1 and table 1. The specimen dimensions are 400 mm width, 1,000 mm length and 2 mm thick. These have an FSW butt joint inclined at an angle of 0 or 30 degrees to the direction of applied stress. The specimens were cut out from large FSW plates prepared by a Japanese aircraft manufacturer, in order to achieve same residual stress field perpendicular to the weld line in all specimens. A starter notch was introduced at the center of the specimens, with a distance of 10 or 40 mm between the weld line and the center of the notch to evaluate the effect of this initial distance on crack growth acceleration in the tensile residual stress field.

Properties of fatigue crack propagation in friction stir welded 2024-T3 aluminium alloy 901

The specimens were subjected to 5 Hz cyclic loading with R = 0.1 and a stress range of 50 MPa. Crack length was measured by a CCD camera and a scale. The residual stress distribution of the FSW plates was measured by the hole drilling method. Residual stress redistribution caused by crack propagation was measured by strain gauges located in front of the notch at 30, 50, 70, 90, 110, 130 and 150 mm away from the center of the weld line on both sides.

RESULTS AND DISCUSSION Residual stress The measured residual stress is shown in figure 2. High tensile residual stress in the longitudinal direction was found on weld line, with peak values on the front and back sides of the specimens of approximately 230 and 310 MPa, respectively. The residual stress outside of the weld line is apparently smaller than that within the weld line, and the residual stress in the transverse direction is smaller than in the longitudinal direction by an order of magnitude. And the distribution shape is nearly symmetric to the weld line. In other references, the residual stress field around the weld line has been found to be complex with peaks around both edges of the weld line [13, 14], and apparent compressive residual stress has also been measured outside of the weld line [13]. Compared to these references, the measured residual stress distribution observed in our experiments is much simpler. It was found that the integration result of the longitudinal residual stress did not become zero. Based on the ASTM standard, residual stress was measured by holes penetrating the specimen. However, the residual stress seems to change along the thickness direction even for the thin plate. This would affect the values of residual stress measured by the hole drilling method and also the value of its integral. Effect of residual stress on crack growth rate Figure 3 shows the da/dN-ΔK curve of the base material and FSW specimens. The stress intensity factor range, ΔK, was calculated by the equation proposed by Ishida [15]: (1) ΔK=Δσ/B (πα/2W sec (πα /2)1/2 α=2a/W where a is the crack length, W is the specimen width, B is the specimen thickness, and Δσ is the stress amplitude. For the FSW specimens, the crack growth rates on each side of the weld line were evaluated separately as it was thought that these would differ due to the effect of residual stress around the weld line. Cracks on right side of the weld line did not pass the weld line, while those on left side ran through the weld line. Figure 3 indicates that regardless of the distance between the weld line and the center of the notch, the growth rate of cracks which did not pass the weld line was almost same as in the base material, while the growth rate of cracks on the other side of the weld line was accelerated by tensile residual stress when the crack tip was located around the weld line. After a crack

T. Okada et al.

902

tip ran through the weld line, its propagation rate gradually decreased toward that in the base material. When the distance between the weld line and the center of starter notch was 40 mm, the growth rate in the FSW specimens was about two times greater than in the base material when the crack tip was in the weld line. When the distance was 10 mm, the ratio of the growth rate in the FSW specimens to that in the base material was almost the same as the ratio for the 40 mm distance case. Figure 4 shows the redistribution of longitudinal stress ahead of the crack tip during crack growth. The value on the longitudinal axis was calculated by subtracting the stress in the base material from that in the FSW specimen for the same crack length. The values on abscissa are the locations of the strain gauges. When the crack tip has not reached the weld line, stress redistribution merely occurs. After the crack tip runs through the weld line, tensile stress is redistributed in front of the crack tip. This tensile stress accelerates the propagation of cracks that cross the weld line. Effect of weld line angle Figure 5 shows crack paths in specimens with weld line angles of 0 and 30 degrees. The result shows that direction of crack growth is not affected by the weld line inclination: cracks grow perpendicular to the loading direction regardless of weld line angle. This feature is observed, irrespective of the distance between the weld line and the center of the notch. For the specimen with weld line angle of 30 degrees, the direction of maximum principle stress is inclined relative to the weld line when the crack tip is near the line, so if maximum principal stress were a dominant factor affecting the direction of crack growth, the crack path would be at an inclined angle In the specimens, however, the crack path was found to be mostly perpendicular to the weld line. This indicates that the maximum principle stress is not the primary factor determining the direction of crack growth; rather, the direction of the maximum stress amplitude seems to be the primary factor. Figures 6 and 7 show da/dN-ΔK for FSW specimens in which the distance between the weld line and the center of the starter notch is 10 and 40 mm, respectively. As these figures show, the crack propagation rate measured in specimens with a weld line inclined at 0 degrees is almost same as that in specimens with a weld line angle of 30 degrees, irrespective of the distance between the weld line and the center of the notch. It is supposed that this is the result of cancellation of crack opening stress in the FSW specimens, because the 268 MPa maximum tensile residual stress is apparently greater than the applied maximum stress of 55.6 MPa. The da/dN-ΔKeff curve of the base material is therefore evaluated using the equation for the ratio of stress intensity range U suggested by Schijve [16]: U = 0.55 + 0.35R + 0.1R2

(2)

Properties of fatigue crack propagation in friction stir welded 2024-T3 aluminium alloy 903

The value of U for the test conditions is calculated as 0.59. The da/dN-ΔKeff curve of the base material is drawn as a solid line in the figures. It appears that da/dN-ΔK of the FSW specimens and da/dN-ΔKeff of the base material do not coincide, even when the crack tip is located on weld line. This implies that the crack opening stress for each weld line angle is not completely canceled by tensile residual stress. The authors measured the crack opening stress of the base material and FSW specimens using the compliance method [17]. The opening stress was measured until the out of plane deformation became significant and the opening stress could not be properly determined. The crack opening stress of the base material was found to be about 30 MPa and the ratio U became 0.51. On the other hand, that for the FSW specimen was found to be about 20 MPa irrespective of the weld line angle. The da/dN-ΔKeff curves for the base material and the FSW specimens calculated from the crack opening stress are shown in figure 8, and these coincide well. Terada [18] and Tada and Paris [19] separately proposed the function of longitudinal residual stress distribution on the butt welded joint and evaluated the stress intensity factor of a crack perpendicular to the weld line. The proposed functions are as follows: σy(ξ) = σ0 ( 1 - ξ 2 ) exp( -0.5 ξ 2 ) σy(ξ) = σ0 ( 1 - ξ 2 ) / ( 1 + ξ 4 )

(in reference [18]) (in reference [19])

(3) (4)

where ξ is the normalized coordinate divided by the distance, c, where the residual stress become zero, σy(ξ) is the longitudinal residual stress, σ0 is the maximum tensile residual stress on the weld line. And Terada [18] also proposed the relationship of fatigue crack growth behavior between the base material and the butt welded panel. (5) (da/dN)welded plate = ( 1 + Fres σ0 / Δσ ) (da/dN)base material where Δσ is the stress range. And Fres is the correction factor obtained by the following equation: (6) Fres = K / (σ0 ( πa ) 0.5) Fatigue crack growth rate on the FSW specimen with a weld line angle of 0 degree is calculated using the crack growth rate for the base material and these equations. Equation (4) is employed to simulate longitudinal residual stress distribution on the FSW panel, because it is closer to the measured stress shown in figure 2 than equation (3). From the figure, the maximum tensile stress, σ0, and the distance, c, are assumed to be 268 MPa and 10 mm, respectively. The calculated fatigue crack growth rates of the FSW specimens for the 10 and 40 mm distance case are also shown in figure 6 and 7 as a dashed line. As shown in figure 6, the calculated fatigue crack growth rate near the weld line is slower than the measured growth

T. Okada et al.

904

rate. The simulated residual stress distribution produces about -50 MPa compressive residual stress near the weld line, while only small residual stress is measured at there. The difference affects the calculated crack growth rate. The calculated crack growth rate on the weld line for 40 mm distance case is similar to the measured growth rate, while that on the weld line for 10 mm distance case is approximately 2.5 times higher than the measured growth rate. When the distance between the weld line and the center of the notch is 10 mm, the value of the correction factor on the weld line is about 0.6, while that on the weld line for 40 mm distance case is about 0.25. The yield stress of the FSW specimen used in this research is about 330 MPa and the maximum residual stress is about 80 % of the yield stress. In reference 18, applicability of equation (5) was confirmed to the fatigue crack growth data for the welded plate [20]. The calculated correction factor is 0.65 or less and is close to the case of this study. The yield stress, the maximum tensile residual stress on the weld line and the stress range were 625, 210 and 107 MPa, respectively. The maximum tensile residual stress is 34 % of the yield stress and is apparently lower than this case. The maximum residual tensile stress and the correction factor affect the stress intensity factor and plastic deformation around the crack tip. Therefore, it is considered that the equation (5) predicts the crack growth rate on the residual stress when the effect of plasticity is not significant.

CONCLUSION Fatigue crack growth tests were conducted in order to evaluate the effect of residual stress field on crack growth behavior. The results indicate that a difference of initial distance between the weld line and the center of the notch merely affects the ratio of the crack growth rate in the FSW specimen to that of the base material. Cracks grew perpendicular to the loading direction irrespective of the weld line angle. Furthermore, the crack growth rate in FSW specimens was almost the same, regardless of the weld line angle. The evaluation of the effective stress intensity factor range using the ratio, U, implies that the crack opening stress for each weld line angle is not completely canceled by tensile residual stress on the weld line. Based on the results, maximum principle stress is not the primary factor determining the direction of crack growth; rather, the direction of the maximum stress amplitude seems to be the primary factor under this test condition.

REFERENCES [1] Eclipse Aviation web site: http://www.eclipseaviation.com/index.html. [2] Dalle-Donne, C. (2005), In: Review of Aeronautical Fatigue Investigations in Germany During the Period March 2003 to May 2005, pp. 36.

Properties of fatigue crack propagation in friction stir welded 2024-T3 aluminium alloy 905

[3] Pacchione, M., Werner, S. and Ohrloff, N. (2007), In: Durability and damage tolerance of aircraft structures: metals vs. composites, Proceedings of the 24th ICAF Symposium, vol. I, pp. 224-240, Lazzeri, L. and Salvetti, A. (Ed.), Pacini Editore, Pisa. [4] Widener, C.A., Tweedy, B.M. and Burford, D.A., (2008), Proceedings of the 7th International FSW Symposium, TWI Ltd. [5] Dalle-Donne, C., Biallas, G., Ghidini, T. and Raimbeaux, G. (2000), Proceedings of the 2nd International FSW Symposium, TWI Ltd. [6] Ghidini, T. and Dalle-Donne, C. (2003), Proceedings of the 4th International FSW Symposium, TWI Ltd. [7] Sato, H., Yamada, Y. and Tanoue, Y. (2007), In: Durability and damage tolerance of aircraft structures: metals vs. composites, Proceedings of the 24th ICAF Symposium, vol. I, pp. 163-171, Lazzeri, L. and Salvetti, A. (Ed.), Pacini Editore, Pisa. [8] Lemmen, H.J.K., Alderliesten, R.C., Homan, J.J. and Benedictus, R. (2007), In: Durability and damage tolerance of aircraft structures: metals vs. composites, Proceedings of the 24th ICAF Symposium, vol. I, pp. 147-162, Lazzeri, L. and Salvetti, A. (Ed.), Pacini Editore, Pisa. [9] Bussu, G. and Irving, P.E. (2001), In: Design for durability in the digital age, Proceedings of the 21st ICAF Symposium, vol. I, pp. 331-350, Rouchon, J. (Ed.), Cépaduès Editions, Toulouse. [10] Terada, H., Okada, T. and Dybskiy, P.A. (2001), In: Design for durability in the digital age, Proceedings of the 21st ICAF Symposium, vol. I, pp. 263-272, Rouchon, J. (Ed.), Cépaduès Editions, Toulouse. [11] Lim, E.I. (2006), presented at ASIP Conference 2006. [12] Lemmen, H.J.K., Alderliesten, R.C., Pieters, R.R.G.M., Benedictus, R. and Pineault, J.A. (2008), Proceedings of the 49th SDM Conference. [13] Dalle-Donne, C., Lima, E., Wegener, J., Pyzalla, A. and Buslaps, T. (2001), Proceedings of the 3rd International FSW Symposium, TWI Ltd. [14] Ali, A., Brown, M.W., Rodopoulos, C.A. and Gardiner, S. (2006), J. of Failure Analysis and Prevention, vol. 6(4), pp. 83-96. [15] Ishida, M. (1966), Trans. of the ASME, vol. 33, pp. 674-675. [16] Schijve, J. (1981), Eng. Fract. Mech., vol. 14, pp. 461-465. [17] Kuwayama, K., Asakawa, M., Okada. T., Nakamura, T., Machida, S. and Fujita, S. (2009), presented 50th SDM Conference. [18] Terada, H. (1987), In: Role of Fracture Mechanics in Modern Technology, pp. 899-910, Shi. G.C., Nishitani, H. and Ishihara, T. (Ed.), Elsevier Science Publishers B.V. [19] Tada, H. and Paris, P.C. (1983), Int. Journ. of Fracture, vol. 21, pp. 279-284. [20] Glinka, G. (1979), In: Fracture Mechanics, Proceedings of the 11th National Symposium on Fracture Mechanics: Part I, ASTM STP 677, pp. 198-214, Smith, C.W. (Ed.), ASTM.

T. Okada et al.

906

Table 1 Specimen specifications Specimen type Specimen number Angle between weld line and loading direction θ (degree) Angle between weld line and center of notch d (mm)

Base material 1 2 N/A

N/A

FSW 1

2 0

40

3

4

5

30

0

30

10

(a) Base material specimen (b) FSW specimen Figure 1 Dimensions of specimens

(a) Longitudinal stress (b) Transverse stress Figure 2 Residual stress distribution perpendicular to weld line

Properties of fatigue crack propagation in friction stir welded 2024-T3 aluminium alloy 907

Figure 3 Crack growth rates in FSW and base material

Figure 4 Relationship between redistribution of longitudinal stress and crack length

T. Okada et al.

908

(a) Angle of weld line, 0 degree (b) Angle of weld line, 30 degree Figure 5 Crack paths in FSW specimens

Figure 6 da/dN-ΔK curve (FSW-2 and -3)

Figure 7 da/dN-ΔK curve (FSW-4 and -5)

Figure 8 da/dN - ΔK, ΔKeff curve

25th ICAF Symposium – Rotterdam, 27–29 May 2009

FATIGUE CRACK GROWTH IN THICK PLATE 7050 ALUMINUM Joel Schubbe United States Naval Academy

Abstract: A study has been accomplished to characterize the fatigue crack growth rates and mechanisms in thin and thick plate commercial 7050-T7451 aluminum plate in the L-S orientation. Crack growth and crack shielding with branching behavior of long, through thickness cracks is examined and compared to L-T and T-L oriented growth data. Compact tension specimens and the compliance method were used to determine crack growth rates. Constant ΔK data showed significant retardation of growth rate curves for the L-S orientation in the range of 10 to 13 MPa√m where branching and splitting parallel to the load axis are dominant growth mechanisms.

BACKGROUND Aerospace alloys are continually being developed, modified, tested and qualified for use in flight vehicles around the world. Extended service life is a prime consideration in every current design and both structural design and material selection is necessary to increase life and ensure safe flight. Every designer is striving to find that optimal mix of properties to include strength, corrosion resistance, fatigue resistance, stiffness, or other desired characteristics [1]. The limitations on use of many available systems are the availability of test data and full qualification of these materials in the environment in which they are intended to survive. Testing, machining, cost of supply, and space requirements drive engineers to use well-characterized materials. But, only small configurational or orientation changes are necessary to make a known material “less known” in any or all of the material selection property areas.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 909–920. © US Government 2009. Created within the capacity of an US governmental employment and therefore public domain. Published by Springer Science+Business Media B.V. Dordrecht.

Joel Schubbe

910

7050-T7451 is purported to be a superior aluminum alloy system, exceeding strength properties of 7075-series alloys and improving upon corrosion characteristics. Historically, the “banding” of properties for selection and design is used to simplify the down select process, narrowing choices to specific manufacturing materials. Testing for the extremes is common practice and helps to reduce the number of tests required the characterization of any specific alloy. Unfortunately, the most modern, tailored alloys tend to be highly anisotropic in nature and must be fully tested with loads representative of those expected in service for the expected orientation to understand the required range of properties for selection. Considerations must be given to the material similar to those for composites for orientation, homogeneity, or durability. The 7050-T7451 alloy is one of these highly anisotropic materials that need full characterization for use. Figure 1 displays the allow morphology. Most strength properties are fully defined but if experimental crack growth rate data and fatigue properties for the L-S orientation are to be used for life estimates, they have not been tested or published at sufficient levels for existing predictive methods. This study was initiated to generate an initial data set for prediction of growth rates in the L-S orientation and to characterize the morphology of the crack progression. In this case, it has been determined that the manner in which the crack progresses may be important to the failure mode for machined parts with multi-axial stress distributions. The following material data is provided for reference from the guaranteed ALCOA material spec [2]: (WT %), Si 0.12, Fe 0.15, Cu 2.0-2.6, Mn 0.10, Mg 1.9-2.6, Zn 5.7-6.7, Zr 0.08-0.115, Ti 0.06, others each 0.5, others total 0.15, Balance Aluminum. KIC values for thin plate (25-51 mm) and thick plate (127-152 mm) were also provided for comparison. Thin plate KIC: L-T 31.9 MPa√m, T-L 27.5 MPa√m, S-L N/A and thick plate KIC : L-T 26.4 MPa√m, T-L 24.2 MPa√m, S-L 23.1 MPa√m were given. In most cases it has been assumed, in the past, that the values for L-S were similar to the L-T orientation. It was witnessed in this study that the growth mechanisms vary dramatically between the two orientations for long cracks or relatively high loads.

Figure 1. Material Morphology

Fatigue crack growth in thick plate 7050 aluminum

911

EXPERIMENTAL METHODOLOGY Specimens tested in this study were cut from 7050-T7451 aluminum plates of parent thicknesses 1.27 cm and 8.3 cm as the configuration shown Figure 2. L-T and T-L oriented specimens were tested to validate the test method against legacy data and identify any differences in growth rates for the thin vs. thick parent material. L-S oriented specimens were tested to generate new da/dN vs. ΔK crack growth data plots and examine the influence of crack length on the mechanisms associated with the material crack morphology [3]. Testing in an approximate range of 5 to 35 MPa√m was accomplished and at load ratios of R=0.05, 0.1, 0.3, and 0.5. It was found that at R=0.7, the ASTM specimen configuration defined for these tests, combined with the required ASTM loading levels, was incompatible for performing this testing even at load levels as low as 5 MPa√m. Crack initiation at the R=0.7 load ratio resulted in immediate splitting parallel to the load application and little or no forward crack progression. Repeat tests for the other load ratios were accomplished to examine repeatability and to compare to L-T and T-L data. All repeat tests were consistent with the discussed trends.

Figure 2. Specimen Configuration

912

Joel Schubbe

A MTS 810 electric servo-hydraulic 22kip test stand was used to pin load the specimens and control was accomplished using an external MTS Flex SE Test controller with MTS PC interface software. Loading parameters included the load ratio, thickness to specimen depth, K-gradients, and maintaining a small scale yielding criteria. Tests were conducted in a tension-tension mode due to specimen type and external Cyctest software was used to regulate the growth loads and gradients of continuous load shedding during test. Cycle by cycle feedback was provided to enable real-time load corrections. In addition, crack tip opening displacement (CTOD) data from the MTS clip gage was recorded and used to determine crack growth rates using the crack closure compliance method. Optical checks were performed to guarantee the crack length accuracy being used for compliance calculations during test. Crack lengths were validated using a Gaertner traveling optical telemicroscope with digital readout. Previous work was accomplished using a generic fatigue spectra and center-hole tensile plate specimens [4]. This work showed a distinct splitting phenomenon parallel to load orientation that was periodic in nature for loading in the L-S orientation. Additionally, L-S growth rates were twice those of the T-L orientation growth for shorter crack lengths despite minor splitting behavior. Post-test analysis showed that the splitting occurred with each application of load above threshold levels in the spectra. It was displayed in this experimental study that these loads initiated longitudinal cracking in the L-S orientation parallel to the load application. This mechanism is distinctly different than the minor crack branching or meandering (θ<120°) that is potentially beneficial for retardation of master (primary) crack growth [5]. Long crack growth or final stage growth in the spectrum loaded specimens is primarily in this longitudinal direction and significantly retards or even arrests forward master crack growth. Figures 3 and 4 show the changes in growth rate over time and with increased crack length (and subsequently increased ΔK values). Figure 5 displays the resulting fracture surface. The severity of splitting in long cracks is troublesome because industry standard crack growth packages do not contain necessary material data (specifically AFGROW or NASGRO) for this orientation and do not accurately reflect the growth rates or failure life if existing 7050-T7451 data is used to predict growth rates or failure for long cracks due to the splitting mechanisms observed. In this light, data was collected for the L-S orientation for use in the validation of analytical models or numerical modeling techniques. Work continues to expand this database and incorporate the results with industry packages as well as micromodels capable of predicting the growth rate mechanisms.

Fatigue crack growth in thick plate 7050 aluminum

913

0.08

Crack Length 2a (m)

0.07 0.06 0.05 0.04 0.03 0.02 0.01 9000

8000

7000

6000

5000

4000

3000

2000

0

1000

0.00

Flight Hours LS Spectrum Load Truncated 5% TL Spectrum Load Truncated 5%

Figure 3. Spectrum Loading Fatigue Progression (L-S/T-L)

4.0E-07

3.5E-07

2.5E-07

2.0E-07

1.5E-07

1.0E-07

5.0E-08

10000

9000

8000

7000

6000

5000

4000

3000

2000

0.0E+00 1000

Growth Rate (m/cycle)

3.0E-07

Flight Hours Spectrum Load Truncated 5% L-S Spectrum Load Truncated 20% T-L

Spectrum Load Truncated 5% T-L Spectrum Load Truncated 30% L-S

Spectrum Load Truncated 20% L-S Spectrum Load Truncated 30% T-L

Figure 4. Spectrum Loading Crack Growth Rates (L-S vs. T-L)

Joel Schubbe

914

Figure 5. Center-hole Tensile Specimen Fracture Surface

DISCUSSION OF RESULTS This heat-treated, rolled plate form of the aerospace aluminum alloy (7050-T7451) is highly anisotropic in nature and has a high degree of lamellar, columnar grains parallel to the L orientation. The difference to be noted between the T and the S orientations is the width and thickness of the pancake shaped grains. S orientation grains are characteristically thin and high in number with the T orientation moderating in number and with a lower aspect ratio of length to width. The initial comparisons of ΔK vs. crack growth rates were performed on specimens originating from thin (1.27 cm) and thick (8.3 cm+) parent material in the L-T and T-L orientations. The results of these tests were compared to legacy data to validate the effectiveness of the test methods employed in this study and to evaluate any differences in the growth morphology for the two orientations. Figure 6 contains growth rate data for the L-T and T-L orientations compared with legacy alloys.

Fatigue crack growth in thick plate 7050 aluminum

915

1.0E-03

Growth Rate (m/cycle)

1.0E-04

1.0E-05

1.0E-06

1.0E-07

1.0E-08

1.0E-09 1

10

100

ΔK (MPa√m) LT R=0.1, 10 Hz LT R=0.1, 10 Hz TL R=0.1, 10 Hz LT Legacy Data (2024-T3 Bare Long Crack)

LT R=0.1, 10 Hz TL R=0.1, 10 Hz LT Legacy Data (T74 Plate) LT 7050-T7451 Legacy Data, R=0.1, 30 Hz

Figure 6. L-T and T-L Growth Rates Comparison It was observed that the crack progression in these two orientations was similar in that they both progressed with minimal crack branching from the horizontal plane, but differed in that the L-T orientation tended to ‘wander’ significantly more with a gentle turning of the crack front. Figures 7-A (L-T), 7-B (T-L), and 7-C (L-S) show the crack progression shapes at a macroscopic level. Figure 8 details the grain structure and the progression of cracking in the specimens on a microscopic scale. It should be noted that the lamellar structure of the L-S oriented specimen has characteristic splitting (90° turn) that follows the grain boundaries and the frequent directional changes of this orientation causes retardation of the master crack progression. These splits become more pronounced, longer, and occur in conjunction with, sometimes exceeding, master crack growth as the stress intensity level increases and the length of the crack increases. The T-L orientation morphology allows the crack to grow in a reasonably straight fashion along the grains oriented with the direction of master crack growth. L-T orientation growth is generally perpendicular to the grain boundaries but tends to ‘wander’. The significant difference in grain width (decreased aspect ratio) for L-T in relation to the L-S orientation allows for significantly more gradual directional changes and more bifurcation or ‘branching’ versus ‘splitting’ as a growth mechanism (Figures 9 and 10).

Joel Schubbe

916

Figure 7. Oriented Crack Growth Progression (A. L-T, B. T-L, C. L-S: Scale=1mm)

Figure 8. Oriented Crack Progression

Fatigue crack growth in thick plate 7050 aluminum

Figure 9. Splitting (Upper) vs. Branching (Lower)

Figure 10. Long Crack Splitting (L-S Orientation)

917

918

Joel Schubbe

The data collected from experiment for the L-T and T-L orientations were in close agreement with legacy data and showed only minimal scatter. For the higher ΔK values the T-L data for the thick plate showed an expected increased rate of growth over the L-T or the thin plate T-L data, due to the lower fracture toughness values in that particular orientation [2]. Figure 11 shows the crack growth data for the LS orientation and selected L-T/T-L data for comparison. Most notable in this data is the transition from an expected sigmoidally shaped fatigue curve to an ever increasingly retarded growth rate curve beginning at ΔK values of greater than 10 to 13 MPa√m. Growth parallel to the load application axis is dominant at ΔK values of greater than 18 to 20 MPa√m. The mode of growth changes from splitting to forward progression with an alternating slowing and increase in growth rate or even more of a divergent sinusoid curve at higher load ratios (R=0.3, 0.5). Referring to the crack progression shown in Figure 11., the trends of growth for the three orientations are consistent with the modes of growth shown. The fastest growth of the master crack is highlighted in the T-L specimens where the growth is only in the master crack direction and is only slightly diverted in direction, with a low level of surface roughness or closure interference. A noticeable level of retardation is observed in the L-T orientation where the crack is progressing in a kinking or bifurcating manner and there is increased interference and surface roughness. The most significant level of retardation or arrest is observed in the L-S orientation at long crack lengths and high stress levels. Multiple cracks, splitting and numerous frequent 90° directional changes can be concluded to have the highest impact on growth rates for this orientation. Surface roughness and closure interference are likely to play a role in the master crack retardation but are overshadowed by the splitting phenomenon.

Fatigue crack growth in thick plate 7050 aluminum

919

Figure 11. Oriented Crack Growth Rates (L-S, T-L, L-T) Below 10 MPa√m it is evident that the growth rates increase with increased load ratio, R. It should also be noted that with the CT testing done, peak loads for a given ΔK also increase. Above 10 MPa√m the trend almost reverses itself with growth rates for R=0.05 eventually being the highest consistent rate of progression for crack growth. Crack turning, meandering, and shielding all have an increased effect on forward progression at the higher load ratios with splitting and longitudinal growth becoming the dominant mode of growth as ΔK continues to increase. All the L-S data curves above ΔK = 10 MPa√m have a characteristic divergent sinusoidal trend of growth where the growth mode alternates between arrested forward growth and splitting or arrested vertical growth and continuing forward crack progression. This behavior was observed in previously examined spectrum loading throughout testing but this study can now determine at what ΔK level the dominant splitting behavior begins [3-4,7]. This behavior has also been shown to be prevalent, to a greater extent, in Al-Li alloy systems and could be beneficial to crack arrest or slowing growth if a means to predict or model the expected branching or splitting behavior is established [8-13].

CONCLUSIONS Findings from both the truncated load spectra study and the investigation of load ratio crack growth show distinct differences of the failure modes exhibited by each of the orientations of 7050-T7451 Plate alloy. Only the thick plate material is likely to be used under the tested stress conditions and where the L-S orientation

Joel Schubbe

920

would occur in a manufactured design. Potential parts might include deep channels, bulkheads, or any part where the part depth is oriented in the plate thickness direction. In these cases, extra consideration must be given to the stress levels, loading conditions, and inspection access to ensure the thresholds shown in this study are not exceeded in design and any aging damage can be detected and managed. Either crack length or loading level can create the condition whereby the ΔK value for crack progression may exceed 10 to 13 MPa√m and initiate a splitting mode of crack growth in the L-S orientation. Further investigative work will include material constituent gradients due to parent plate thickness, stress corrosion influences on initiated splitting, and additional data collection for numerical models.

REFERENCES [1] Hunsicker, H.Y., (1976), In: Philosophical Transactions of the Royal Society of London, Series A, Mathematical and Physical Sciences, Vol 282, No. 1307, pp. 359-376. Rosenhain Centenary Conference: The Contribution of Physical Metallurgy to Engineering Practice. [2] Alcoa – Product Catalog: 7050 Aluminum Alloy Plate and Sheet, 2007 [3] Schubbe, J., (2009) Eng. Fract. Mech. (In press), (DOI Citation: 10.1016/j.engfracmech.2009.01.006 ) [4] Schubbe, J., (2009) Eng. Fail. Analysis, vol 16 pp. 340-349, (DOI Citation: doi:10.1016/j.physletb.2003.10.071) [5] Meggiolaro, M.A., Miranda, A.C.O., Castro, J.T.P., Martha, L.F., (2005) Eng. Fract. Mech. vol 72, pp. 2647-2671. [6] ASM Handbook, Volume 19, Fatigue and Fracture, 1996, pp. 170-176. [7] Schubbe, J., (2008) An Investigation of 7050-T7451 Aluminum Fatigue Crack Growth Results for T-L and L-S Orientations and Truncated Fatigue Life Spectra. In: Proceedings of the Aging Aircraft Conference, Phoenix AZ. [8] Venkateswara Rao, K.T., Yu, W., and Ritchie, R.O., (1988) Metallurgical Transactions A, vol 19A, pp. 549-561. [9] Venkateswara Rao, K.T., Yu, W., and Ritchie, R.O., (1989) Metallurgical Transactions A, vol 20A, pp. 485-497. [10] Venkateswara Rao, and Ritchie, R.O., (1989) In: Seventh International Conference of Fracture (ICF-7), Houston, TX. [11] Crill, M. J. et al, (2006) Materials Science Forum, vols. 519-521, pp. 1323. [12] Philbrook, M.J., Smith, K.P., Crill, M.J., Balmuth, E.S., Amin, J.T., Chellman, D.J., (2006), In: Proceedings of Aeromat Conference, Seattle, Wa. [13] Magnusen, P.E., Bucci, R.J., Hinkle, A.J., Brockenbrough, J.R. and Konish, H.J.,(1997) Int. Journal of Fatigue, vol19, pp. S275-S283.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

FATIGUE LIFE EVALUATION ON TENSION TYPE FITTING Nicolas Baréa and Marie Massé Fatigue & Damage Tolerance department, LATECOERE Company

Abstract: The fatigue life evaluation on tension type fitting is a difficulty in aeronautical structure designing. The goal of this study is to try to define a fatigue life evaluation method on tension type fitting. Finite element analysis shows the difficulty to have consistent results, because of the high influence of parameters like friction and contact, pre-load value precision, way to model fastener and fitting, elements type. This numerical approach is heavy and can be replaced by an easier and faster approach as it is often made for shear joints. More than hundred specimens in several configurations (type of fastener, geometry, loading, pre-load) and real aircraft components have been tested, thus a simple method has been developed and correlated to these fatigue tests. This test program and obtained life prediction method give a good answer for our specific tension type fitting problem and help efficiently and economically assembly design, avoiding heavy complex finite element models. The work summarized here has permitted a better understanding of tension type fitting and its several parameters. Tension fitting life predictions can be obtained very fast with accuracy with a simple method and generic tests.

INTRODUCTION The fatigue life evaluation on tension type fitting is a difficulty in aeronautical structure designing. A multitude of fatigue calculation method and test results exist M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 921–930. © Springer Science+Business Media B.V. 2009

Nicolas Baréa and Marie Massé

922

for shear joints but very few recognized ones for tension fittings. The goal of this study is to try to define a fatigue life evaluation method on tension type fitting.

The easier way to solve this kind of problem is to first think to finite element analysis. But the modelling is not so easy when you start to list the different influence parameters such as: • • • • • • •

Friction. Contact. Pre-load. Way to model friction, contact and pre-load. Way to model fastener and fitting. Type and size of finite elements. …

Because of all these parameters, deviations can appear on finite element model results compared to test results. Finite element analysis needs test correlation for validation, and can be heavy and long when volume models with contact are made. Our idea is to understand the different influence parameters with the help of finite element models and fatigue tests. Then with this better understanding of parameters, we have tried to develop a simple method to evaluate fatigue life on tension fitting. Thus a test program has been launched, with more than one hundred specimens. Then fatigue test and real aircraft components results have been used to develop and correlate a fatigue sizing method on tension fitting assembly.

APPROACH To start on fatigue life of tension fitting, finite element analysis is made using “Promechanica” and “femap/nastran” software. Finite element analysis shows the difficulty to have consistent results, because of the high influence of parameters like friction and contact, pre-load value precision, way to model fastener, fitting, elements type.

Fatigue life evaluation on tension type fitting

923

Different finite element models have been made to determine the critical site and the peak stress associated. The different critical sites found are where peak stresses are. The peak stresses are maximum stresses located at volume surface. According to influence parameters, 3 sites can be critical: • • •

Radius of the fitting, at the junction between fitting flange and wall. Fitting flange near fastener head. Fastener.

These critical sites have been correlated with the failure sites observed by tests. Knowing the peak stress, fatigue life has been estimated by different methods based on SN curves and compared to fatigue test life. Correlation coefficients have been defined to correlate fatigue life estimation based on FEM peak stresses and fatigue test life. These coefficients are depending on fitting type (machined or formed), ratio of stiffness between fitting flange and fastener. The life evaluation results are scattered when they are compared to test results, this is due to high influence on peak stress values of modelling parameters.

Figure 1: Tension fitting and fastener finite element model views.

About this FEM approach, it can be concluded the help given by model to find critical site, but the difficulty to have consistent peak stresses and so fatigue life estimations. Actually this approach shows the necessity to have test results to correlate the method to estimate fatigue life based on peak stress, and especially with FEM peak stress. Also this numerical approach is heavy and that is why we try hereafter to replace it by an easier and faster approach as it is often made for shear joints without finite element model and with only analytical calculation.

Nicolas Baréa and Marie Massé

924

This simple approach is based on FEM and test conclusions. The figure 2 describes this experimental approach and its steps. First of all, the tension fitting assembly geometry must be well known: thickness, radius, length, and distance of fastener to wall. This is essential to have these entry data for the second step of the method, the assembly equilibrium. So in the second step, loads are determined with tension fitting simple equilibrium according to geometry, external loading, prying effect (contact stress), assembly flexibility and fastener pre-load. To do this, the research works of J. Guillot (ref. [1]) on thread screws assembly have been used. The equation (1) has to be solved to find “s” the distance of the point of application of prying effect. Knowing this point of application of prying effect, the equation system is solved, contact and fastener loads determined. C1 s 3 + C 2 s 2 + C 3 s + C = 0

m ⎧ ⎪C1 = 6× E I × S P P ⎪ Φ ⎪ ⎪C2 = q ⎞ ⎛ ⎪⎪ m + 2× EP I P ⎜ q1 + 2 ⎟ m⎠ ⎝ ⎨ ⎪ 2m + 2× EP I P × q1 ⎪C3 = ×Φ q2 ⎞ ⎛ ⎪ m + 2× EP I P ⎜ q1 + ⎟ ⎪ m⎠ ⎝ ⎪ ⎪⎩C4 = −m

(1)

Q SB + SP (1− γ ) ⎧ ⎪Φ = F − S E ⎪ ⎪ q ⎞ ⎛ m + 2× EP I P ⎜ q1 + 2 ⎟ ⎪ m⎠ ⎨F = m × ⎝ ⎪ C s s + 2m + 2× E I q × FE P P 1 ⎪ ⎪FB = FC −F E ⎪ ⎩

With m, the distance between fastener and load introduction point. S=SP+SB, the assembly flexibility is equal to fitting flexibility SP plus fastener flexibility SB. EP, fitting Young modulus. IP, fitting inertia calculated on working bending average length.

γ=

s s max

Q, pretension load. q1 and q2, coefficients depending on boundary condition (wall fitting guided or free). FB, fastener load. FC, contact load. FE, external load.

Fatigue life evaluation on tension type fitting

925

This method permits to have the loads equilibrium in tension fitting assembly, taking into account prying effect, boundary condition on fitting wall (guided or free), assembly flexibility and pre-load in fastener. Knowing the loads equilibrium of the fitting, maximum stresses are calculated for the different critical sites: • • •

for machined fitting, in radius between wall and end pad, tension and bending stresses are considered. for formed fitting, near fastener head and in fastener, only bending stress are considered. for fastener, tension stress are considered.

Test and FEM results have permitted to define failure criterion based on the comparison of fastener and fitting stiffness. According to ratio of stiffness value, configuration type is flexible or stiff. For example a configuration flexible has a critical site in the fitting radius or flange near fastener, but the fastener is not critical. Finally tension fitting fatigue life is estimated using the maximum stress and linear SN fatigue law depending on: • • • • • • •

the fitting configuration (formed or machined fitting, flexible or stiff fitting), pre-load fastener, material type, material protection, scale effect, stress ratio effect, metal fibre orientation.

This fatigue law is correlated by test. Also fatigue life of fastener can be estimated with a method based on ESDU (ref. [2]) and our test specimens.

EXPERIMENTATION The test program described here has been launched to have the experimental data and proof to understand tension fitting assembly behaviour and associated life estimation method. This test program consists of more than hundred specimens in several configurations (type of fastener, geometry, loading, pre-load) and real aircraft components tested.

Nicolas Baréa and Marie Massé

926

ep

Geometry

B

A

R

es

Tension fitting equilibrium/loads calculation

Machined fitting – radius site Sollicitation

Stresses calculation

Tension/Bending

Tension stress

Bending stress

Configuration type determination (flexible, stiff)

Formed fitting – near fastener site

σ c = σ ct ± σ cf

Bending

FE σ ct = ep × W

σ cf =

F × (s + m ) − F C

ep 2 × W

/ B

×m

×6

σ af =

FC × s ep 2 × Da

Stiffness criterion: EI Ef . π .D 4 Kf fastener = = e 64.e 3 EI Ep . Dp.es 3 Kf fitting = = m 12.m

Linear SN law for fatigue life estimation:

Fatigue life estimation

CT.CM ⋅CP. σbase ) σ.Ke ⋅Ksf ⋅ f (R)

p

N =10 .( 5

σ a = σ af

Figure 2: tension fitting method flow

×6

Fatigue life evaluation on tension type fitting

927

Four types of specimens have been fatigue tested: -1 wall formed fitting. -1 wall machined fitting. -2 wall machined fitting. -3 wall machined fitting.

Figure 3: Type of specimens tested For all these fitting type, different parameters have been modified to determine their influence: • • • • • •

Fitting thickness (2mm, 2.4 mm, 3.5 mm, 5 mm for formed fitting, else 2.4 mm, 3.6 mm, 4.8 mm, 6 mm for machined fitting). Fitting material (2024-T42, 7175-T7351). Fitting geometry (distance between wall and fastener). Fastener material (steel, titanium) and diameter (4.2 mm, 4.8 mm, 6.35 mm). Fastener type (hi-lite, tension bolt, screw). Fastener pre-load.

Fatigue tests were made with a stress ratio R=0.1 only. As figure 4 shows, fatigue tests have allowed to define a failure criterion (radius or near fastener of fitting, fastener), based on fitting and fastener stiffness ratio.

928

Nicolas Baréa and Marie Massé

Figure 4: tension fitting failure sites Tests also show the high influence of fastener pre-load, no fastener failure has been observed with pre-load, it has always been a failure in fitting radius or near fastener. Fastener pre-load insures fastener fatigue life. In case of formed fitting, no failure in fitting radius has occurred.

DISCUSSION AND CONCLUSION The exploitation of this test program plus real aircraft test results permits to define correlation coefficients for our life prediction described previously. These correlation coefficients are different if fitting is machined or formed, and if tension assembly is flexible or stiff. Another correlation coefficient tries to quantify conservatively pre-load benefit on fastener behaviour. Thus the method obtained, with these correlated coefficients, gives very satisfactory fatigue life prediction for tension fitting, with: •100% of test results in prediction band width [N/5; 5N] •91.2% of test results in prediction band width [N/3; 3N] For the fastener predictions, results are also satisfactory as can be seen in following graphic. Fastener test results are near the fastener law based on ESDU (ref. [2]), excepted for 3 identical specimens.

Fatigue life evaluation on tension type fitting

929

Tension fitting test results & fatigue life predictions comparisons test results

1000

Latecoere fatigue law N/3

peak stress (Mpa)

N/5 3N 5N Puissance (test results)

100 10 000

100 000

1 000 000

10 000 000

N (cycles)

Fastener test results & life predictions comparisons 1

R2 = 0.6495

life prediction (based on ESDU) Salt/Ftu fastener test results

0.1

Puissance (fastener test results)

TP3

0.01 1000

10000

100000

1000000

10000000

100000000

N (cycles)

Figure 5: Test results and fatigue life predictions comparisons for tension fitting and fastener.

Nicolas Baréa and Marie Massé

930

This test program and obtained life prediction method give a good answer for our specific tension type fitting problem and help efficiently and economically assembly design, avoiding heavy complex finite elements models. Of course our work answers to our specific problems, with the materials, fasteners we are using, our fitting design and our assembly process. But additional tests can be made and enhance the method. Special attention has to be taken when using this method, it has been correlated to fatigue test, which means the peak stress obtained is artificial and has no real signification. This peak stress value suits well to this fatigue life evaluation and cannot be extrapolated to static fitting analysis. The work summarized here has permitted a better understanding of tension type fitting and its several parameters, and so to find explanations to test results or aircraft failures. As it has been shown tension fitting life predictions can be obtained very fast with accuracy with a simple method and generic tests.

REFERENCES [1] Guillot Jean. (1987), In: Assemblages par éléments filetés. technique de l’ingénieur 8 – 1987, B 5 560 – B 5 561 – B 5 562. [2] ESDU (1984). In: Fatigue strength of external and internal steel screw threads under axial loading, ESDU Fatigue - Endurance Data, Vol 4 Stress and Strength, Vol 5 item n°84037. [3] Massé Marie (2008). In: Note fatigue dépliage, FDT technical report N°E0010MM-A, LATECOERE internal note.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

ACCELERATED FATIGUE TESTING ON HYDRAULIC SHAKER M. Fressinet, M. Panis and C. Bordes Toulouse Aeronautical Test Centre (DGA/CEAT), France

Abstract: In term of fatigue initiation, fatigue tests are commonly stopped at 107 cycles assuming that a fatigue limit exists whereas experience feedback on in service aircraft has shown that structural elements could fail after 109 cycles. Thus it is important to explore the gigacycle fatigue domain and an accelerated fatigue testing method which is presented in this paper has been set up. The main idea is to excite a sample, clamped on an electrodynamic shaker, near one of its resonance mode. FEM calculations were needed to define a test specimen which could be solicited at high frequency with the expected stress level. Once the test procedure was set up, test on different samples were carried out to verify the reproducibility of such a test. Besides, before trying to explore the gigacycle fatigue domain, tests were performed near 105-106 cycles to be compared with 4 point bending tests performed with a standard fatigue test device at 2 Hz; the new specimen was specially designed to have the same stress gradient in the studied area. In such a way the influence of test frequency was quantified and the validation of the test procedure was demonstrated. Finally this new fatigue testing methodology showed its capability in exploring the gigacycle domain by increasing the test frequency at more than 800Hz.

INTRODUCTION AND CONTEXT In term of fatigue initiation, fatigue tests are commonly stopped at 107 cycles assuming that a fatigue limit exists. Indeed, under this limit the material is supposed not to be damaged by fatigue and so no fatigue failure should occur. For aluminium alloys, it is known that such a limit does not exist and the Wöhler’s M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 931–938. © Springer Science+Business Media B.V. 2009

M. Fressinet, M. Panis and C. Bordes

932

curve is extended thanks to models like the Haibach ones or thanks to statistical approaches. Unfortunately, despite these precautions, experience feedback on in service aircraft has shown that structural elements could fail after 109 cycles. As a consequence many studies have been carried out since the beginning of the nineties to explore the domain of gigacycle fatigue (108-1010 cycles) and to characterize precisely materials behaviour. Besides, as reaching 109 cycles at 100 Hz lasts more than 16 weeks with traditional fatigue test devices, which is not economically acceptable, many efforts have been undertaken to increase test frequencies and to set up new test processes. To investigate such phenomenon, CEAT developed its own test procedure in a study which is part of a project aiming at understanding the initiation and the growth of some cracks in vibratory environment, phenomena that standard tools are not able to predict satisfactorily. To accelerate the test frequency, the idea was to use hydraulic shakers usually employed to test the robustness of electronic devices in vibratory environment. The philosophy, inspired from [1] and [2], was to excite the test specimen clamped on the shaker near one of its resonant mode. The paper describes in a first step the different stages to elaborate and validate the test procedure and finally the first results obtained thanks to this fatigue testing method.

TEST SPECIMEN GEOMETRY As mentioned previously, the idea was to use the dynamic response of a structure when excited at high frequency. FEM calculations were fully used to choose specimen’s general design and to specify precisely the different sizings necessary to get the right mode shape giving the expected stress level at an high enough frequency. Made of an aluminium alloy, the geometry of the final test specimen and the associated modal analysis results are given hereafter. 125 mm 62.5 mm

Clamping area

12.5 mm

125 mm

62.5 mm

Diameter 5 mm

5 mm th

Mode shape at 875 Hz (4 mode) Thickness 2.5 mm

Figure 1: Test specimen geometry and FEM modal analysis

Accelerated fatigue testing on hydraulic shaker

933

The dynamic response of the sample was simulated too and it was chosen to excite the fourth mode because it gave the required solicitation mode at the expected frequency. It is important to mention that this calculation was performed thanks to the modal superposition method and had needed to be fitted with the first test results which permitted to determine the damping factors which cannot be calculated. The associated stress state is given in the following picture:

Figure 2: Stress state on the fourth resonant mode It can be noticed that it was chosen to work with a stress concentration area. Indeed it is more representative of a real structure and it also enables to reach the expected stress level at low energy. Besides, care was taken to avoid fatigue failure near the clamping area. Gauges were glued to the specimens in order to control the stress level during the test. Some FEM calculations enabled first to verify that the added masses (gauge and wire) were not disturbing the dynamic response of the sample and then to correlate the stress measured by the gauge (σgauge) and the average stress (σaverage) in the working area or more especially at the border of the elongated hole (σmax).

Gauge

1,2

Normalized stress 1

σ max = 2.26 * σ gauge

0,8

0,6

σ average = 1.19 * σ gauge

0,4

0,2

0 0,00

2,00

4,00

6,00

8,00

10

Kt =

σ max = 1.90 σ average

Figure 3: Instrumented test sample and stress variation in the working area As life duration of the gauges does not exceed 107 cycles, a laser was used to measure the displacement of a point on the sample. By this way a correlation with

934

M. Fressinet, M. Panis and C. Bordes

the stress given by the gauge was done which actually enabled to control the test thanks to the laser and so to carry out the test with the greatest accuracy even with a broken gauge. Finally, the test specimen was clamped to the hydraulic shaker thanks to iron manufactured pieces. Special care was taken in the clearance dimensions because misalignment could change drastically the dynamic behaviour of samples. In the same way, particular attention was paid to the gluing of the gauge and the wire.

Figure 4: Test setup

TEST PROCEDURE OVERVIEW Once the specimen clamped, the first task was to achieve a scan at 1g between 100Hz and 1000Hz. It is not damaging for the sample and thanks to its dynamic response, it enabled to verify the test specimen and the gluing of the gauge. After that, the sample was excited around the fourth mode at different accelerations in order to compare the information given by the laser and the gauge. In fact, it enabled to establish the relationship necessary to define the level of displacement or acceleration which gave the right stress level and so the level at which the test should be carried out. The following curve gives the results for different samples, the low scattering in the correlation demonstrating the good reproducibility of the test:

Accelerated fatigue testing on hydraulic shaker

935

140 120

Measured stress (Mpa)

100 80 60 40 20 0 0

0,02

0,04

0,06

0,08

0,1

0,12

0,14

Measured displacement (mm)

Figure 5: Stress-displacement correlation Finally, as driving on a resonance mode is not always stable, it has been chosen to work at a frequency a bit lower than the resonance one and the same correlation has been done. Moreover, for data acquisition reasons, the test has been carried out by scanning a 0.2Hz frequency band. The first tests were controlled by the acceleration. The measured displacement was only used to detect when a crack appears. Actually a crack changes the dynamic response of the structure and thus the displacement generated by a given acceleration. It is illustrated by the following curve which shows that the response of the specimen is constant in time for a given frequency until a brutal change which corresponds to an incipient crack.

0.2Hz band

Variation in the response corresponding to an incipient crack

Crack growth

Several scanning without any variation

f

f

Figure 6: displacement response during cycling The increase of the displacement response during the scanning is due to the fact that for this test, the acceleration was constant while the frequency was evolving in a 0.2Hz band. It should be noted that it generates during the cycling a stress variation of less than 2.5%. This test procedure was used for the tests which were not exceeding 107 cycles. Indeed during the cycles hereafter, the gauge might sometimes disbond and thus

M. Fressinet, M. Panis and C. Bordes

936

change the response of the sample even if no crack has appeared. As a consequence, there was no more way to drive the test far longer. Thus, to drive with the displacement appeared as the only way to reach life duration exceeding 108 cycles. This is what was done for the last tests presented in this paper and what is done now. Moreover contrary to an acceleration driving process, it enables to have a constant stress during the test.

TEST FREQUENCY EFFECT AND TEST PROCEDURE VALIDATION It is commonly admitted that test frequency has no impact on the fatigue life duration if it does not exceed 100Hz. For higher frequency, its effect is related to the temperature and it is assumed that its influence is minimized if the heating is not significant. Thus, as tests were carried out at more than 860Hz, special care was taken to ensure that no consequent damping heating appeared. The temperature of the sample was controlled by a thermal infrared imaging camera (Imager FLUKE© Ti20) and no special heating was noticed. It could be explained by the fact that aluminium alloy is very dissipative and that the working area is very small. Moreover by its solicitation mode itself the sample has its own “cooling system” thanks to a displacement of more than 2mm at more than 860Hz. To establish an accurate correlation, a test specimen was specially designed to have the same stress concentration in the studied area as the previous one. It was solicited on an alternated 4 point bending standard fatigue test device at 2 Hz.

1,2

Normalized stress

1 Normalized stress for shaker sample Normalized stress for bending sample

0,8 0,6 0,4 0,2 0 0,00

2,00

4,00

6,00

8,00

10,00

12,00

Position in the fatigue area

Figure 7: 4 point bending test features

Accelerated fatigue testing on hydraulic shaker

937

Because of the fretting that might appear in such a test setup, the specimen was instrumented by strain gauges to control the loading. Crack wires glued on each faces of the specimen have been used to detect the crack initiation.

Figure 8: test results around 2.105 cycles at 2Hz and more than 870Hz The first results showed a good correlation near 105 cycles and it could be concluded in that case that the test frequency has no influence.

RESULTS Once the test procedure was validated, a test campaign has been launched to establish a Wöhler curve with this accelerated fatigue testing. First results obtained on a 2091T3 aluminium alloy are given in the following curve. Unfortunately because of a lack of time, the curve could not be completed with more test data.

160 Hydrolic shaker - around 870Hz 140

4 bending test - 2 Hz

Average Stress (MPa)

120 100 80 60 40 20 0 1,00E+04

1,00E+05

1,00E+06

1,00E+07

1,00E+08

1,00E+09

Number of Cycles

Figure 9: Wöhler curve obtained with the accelerated fatigue test

938

M. Fressinet, M. Panis and C. Bordes

CONCLUSION AND PERSPECTIVES In this paper, a new accelerated fatigue testing method inspired from [1] is described. Consisting in exciting a sample clamped on an electrodynamic shaker near one of its resonance mode, its efficiency was proved thanks to a correlation with a fully reversed bending fatigue test at 2Hz. While no damping heating was detected during the test, life durations obtained at more than 870Hz were similar to these at 2 Hz. Finally first tests of an ongoing test campaign demonstrated that it was possible to explore the gigacycle fatigue domain with this new test procedure. The next step will be to use this new fatigue testing methodology to study crack initiation on pitting corrosion and to demonstrate that very short cracks could propagate even subjected to a very low stress level.

REFERENCES [1] Methods of testing for endurance of structural elements using simulated acoustic loading, ESDU 93027 [2] Goodman Diagram Via Vibration-Based Fatigue Testing, Tommy J. George, M.H. Herman Shen, Onome Scott-Emuakpor, Theodore Nicholas, Charles J. Cross, Jeffrey Calcaterra, Journal of Engineering Materials and Technology, January 2005, Vol.129, pages 58 to 64

25th ICAF Symposium – Rotterdam, 27–29 May 2009

METHODS FOR FEM ANALYSIS OF RIVETED JOINTS OF THIN-WALLED AIRCRAFT STRUCTURES WITHIN THE IMPERJA PROJECT Jerzy Kaniowski, Wojciech Wronicz1, Jerzy Jachimowicz, Elżbieta Szymczyk2 1

Institute of Aviation, Al. Krakowska 110/114, 02-256 Warsaw [email protected], [email protected] 2 Military University of Technology Faculty of Mechanical Engineering Dep. of Mechanics and Applied Computer Science 2 Kaliskiego Street, 00-908 Warsaw, Poland tel.: +48 022 683 90 39; fax: +48 022 683 94 61 [email protected], [email protected]

Abstract: The paper deals with the modelling of riveted joints in aircraft structures with Finite Element Method. Presented works were carried out within Eureka project No. E!3496 called IMPERJA. The goal of the IMPERJA project is to increase the fatigue life of riveted joints, which will lead to an increase of the aircraft service life, a smaller number of inspections and lower operation costs of an aircraft. The project assumed FEM modelling of the operating aircraft’s structure at three different complexity levels, namely considering the complete structure, a structural detail and a single riveted joint. The paper presents analyses of various rivet models and calculations of a structure and a riveted joint. In the first part examples of various models, at global and local level, were presented and usefulness of them was discussed, influence of the following simplification was analysed; • neglection of rivets in a model (elements are jointed continuously) • rivet as a rigid element (MPC) • neglection of contact phenomenon • neglection of secondary bending • neglection of residual stresses after riveting process The basis of the analysis was the asymmetric butt joint model with 14 rivets. M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 939–967. © Springer Science+Business Media B.V. 2009

Jerzi Kaniowski et al.

940

The model which took into account secondary bending and contact phenomenon was analysed as well. The method of modelling residual stresses with temperature and thermal coefficient was used. In the second part, the example of analysis of riveted joint was demonstrated for a wing of PZL M28 Skytruck aircraft. It’s is a twin-engine, high-wing, cantilever monoplane of all-metal structure with maximum take-off and landing weight 7500 kg. A submodeling technique was used there. At first, part of the wing model, based on a CAD model, was built. It includes 7 ribs and 6 bulkheads between them. Dimensions of the model eliminate stress perturbation, connected with boundary conditions, in the area near the middle rib. It was a shell model. The boundary conditions were taken on a basis of operation data. Presence of rivets wasn't taken into account. Instead of this, parts were connected continuously (nodes were merged). The Linear model of material was used. The purpose of the part of the wing model was to gain accurate boundary conditions for next model of riveted joint on the middle rib. The behaviour of whole model is correct but stress distribution around rivets is not correct. A shell model of riveted joint was build. A boundary conditions were set on a basis of result from previous analysis. Forces, instead of displacements, were used, as boundary conditions, on account of a large stiffness difference between part models (part of wing and riveted joint model). The nonlinear model of material was used. A contact effect and secondary bending were taken into account. Thanks to that, phenomena around rivets were represented considerably better. Results from this analysis could be used as boundary conditions in a detailed calculation of one or few rivets with solid elements. Such a model was considered as well. The presented method allows to analyse phenomena that appear around a rivet in a real structure, during operation. Analyses were performed with MSC PATRAN, NASTRAN and MARC software.

INTRODUCTION Riveting still remains the most popular method of joining metal parts of the aircraft structure. During the riveting process and the operation the severe stress concentrations and the effects such as fretting and secondary bending occur, thus reducing fatigue life. Following factors influence riveting joints quality: Structure and material factors These factors are: type and size of joint, sheet thickness, rivet diameter, pitch between rivet rows and rivet type. All of them should be set in accordance with riveting joint design rules [1] as well as appropriate aircraft design standards.

Methods for FEM analysis of riveted joints of thin walled aircraft structures

941

Different types of rivets are used, depending on the point of application or operational requirements for the individual product. The principal distinctive feature for rivets, except shank diameter, is the shape of the rivet head. In the aircraft structures a significant feature is the shape of the rivet head: protruding or countersunk. Rivets are classified depending on the shape of the rivet head, for instance aircraft rivets are with: flat head [BN-70/1121-01], countersunk head with the angle of 120° [BN-70/1121-05] 90° and 100°, mushroom head [BN-70/1121-06] an others. Besides, there are (according to Russian standards) rivets with compensator: mushroom head rivets [OST 1 34040-79] and countersunk rivets [ANU 0301, OST 1 12020]. Compensator, flattened in dynamic clenching (upsetting) process, should better fill the hole by rivet shank. Rivet type is closely connected with the place of application in the structure. Mushroom rivets can be applied on external surfaces of slow airplanes – because of low velocity, their impact on aerodynamic resistance can be omitted. Countersunk rivets should be used on external surfaces of airplanes when aerodynamic resistance is increased (in the case of velocity growth). Type of material, applied in rivets and joined elements of cover, is an important factor which affects the joint durability. Thin-walled aircraft constructions manufactured in Poland are made of metal sheets (aluminium alloys D16T, 2024T3 1 or titanium alloys) as well as composite materials (based on glass or carbon fabrics). Nevertheless metals are still prevailing in the final product. Rivets are mainly made of aluminium alloys (PA20, PA24 (Russian D18P), PA25, W65 (Russian W65); Russian D16P, D19P, western 2017T4, 2117-T3, 2024-T31, 7050-T73, 2219-T81), steels (15A, 30HGSA, 1H18N9T) or titanium alloys (WT3). Technological factors The internal stresses inducted in materials during technological processes of manufacturing and during fitting of elements exceed yield stress level. Negative stresses generated in the riveting process can be beneficial for the sake of joint durability, but technological processes, especially drilling holes in metal sheets and forming of rivet heads are essential. Method and precision of rivet holes drilling process are selected technological factors. Diameter of rivet hole should be contained in technological clearance limits (i.e. for the rivet with compensator diameter equal 3,5 mm, nominal hole diameter is 3,55 mm, allowable hole deviation +0,12 mm [2] and rivet +0,04 mm [3]). The influence of initial clearance on internal stresses level for steeple head rivet was been analyzed in [4]. A rivet hole should be drilled perpendicularly to the surfaces of joined metal sheets. In the case of curved sheets, hole tolerance is ±5° [2]. The hole correctly drilled is required for the proper and durable joint.

1

T3 state (American standards) corresponds to TN state (Russian standards)

942

Jerzi Kaniowski et al.

Contact surfaces of rivets and sheets should be free of defects. Burrs and microcracks are not allowed. Surface roughness after technological processes should be in the range of 2,5-10 μm [2]. Two methods of rivet upsetting can be distinguished: static and dynamic. Static upsetting of the rivet is performed using riveting press, this method is recommended as more accurate. Uniform compressing of the rivet shank during upsetting process as well as increased productivity are the main advantages of riveting process with the use of press machines. Riveting with the use of pneumatic riveter is an example of dynamic process. Upsetting process can be divided into simple and reverse. Rivet head formed properly (in both static and dynamic manner) should, depending on rivet diameter, get dimensions specified in the riveting manual [2]. These dimensions indirectly inform about the state of stress in the rivet hole – this will be presented in further part of the paper. Mentioned factors have a very strong influence on fatigue life of riveted joints. In the literature there are many works devoted to the riveted joints fatigue life. This area was particularly intensively explored after Aloha Airline Boeing B737 incident in 1988, when approximately 5.49 m feet of the cabin skin and structure aft of the cabin entrance door and above the passenger floorline separated from the airplane during flight, at 7300 m, as a result of multiple site fatigue damage leading to structural failure. E.g. between 1993-1998 in Sweden wide research program focused on joints was carried out. Among other, influence of type of joining elements (rivets and screwed elements) was studied. Fatigue life of specimens varied between 109 217 and 6730 flights, depending on element type [5]. Research program in Netherlands resulted in many PhD theses. E.g. Muller examined three row riveted lap joint [6] and showed that, only by increasing force during riveting process, fatigue life can change from 39 630 cycles (minimal force) through 95 200 (nominal force) to 446 413 cycles (high force). Another works are devoted to influence of coating layers (clad, anode) on fatigue life [7]. Although the literature on the fatigue behaviour of riveted joints is quite abundant, many aspects are still not sufficiently understood and investigated and, therefore, they require a further study. Moreover most of them do not concerns types of rivets and sheets used in Poland. The goal of the Eureka-IMPERJA project is to increase fatigue life. This objective is met by investigating and improving the riveting process as well as improving prediction methods for fatigue life. The project refers to a real structure. One of the IMPERJA consortium partners, PZL Mielec, recommended its product, an airplane M28 Skytruck, as the object of investigations and analyses. In the IMPERJA programme, the study of FEM methodology modelling of the structure at three different complexity levels was planned: • complete structure, • structural detail ,

Methods for FEM analysis of riveted joints of thin walled aircraft structures

943

• single riveted joint. A key issue of the proposed approach will be defining the rules of transferring the excitation (loading and boundary conditions) for the above modelling path. During calculations several material models were used. Fig. 1 presents stress-strain curves for these models. D16AT_lin-wing model is a linear model, used in the part of the wing model (point 3). D16AT-joint model is a smoothed model used in riveted joint analysis (point 4). D16AT-rivet model was used for the skin and the doubler elements in the rivet calculation. PA25-rivet model was used in the same analysis for rivet elements (point 5.1). D16AT properties were assumed according to USSR aerospace industrial branch standard OST 1 90070-92. PA25 properties were taken from [8]. Models used in riveting process simulation were described in point 5.2.

Fig. 1. Material models

THE AIRCRAFT DESCRIPTION The PZL M28 Skytruck (army version called Bryza, figure.2) is a two-engined, braced high-wing monoplane for passenger/cargo transportation purpose with metal structure. The maximum take-off weight is 7500 kg. The aircraft has short take-off and landing (STOL) capability and can operate from short, unprepared runways. It is manufactured by PZL Mielec, and it’s used by e.g. Polish Army, Polish Border Guard as well as customers from USA, Venezuela, Columbia, Vietnam, Indonesia and Nepal.

944

Jerzi Kaniowski et al.

Basic data Length 13,1 m

Height

4,9 m

Wingspa n Maximu m weight Empty weight Maximu m speed Cruising speed Rate of climb Celling

22 m

Range

14001500 km

7500 kg

4090 kg

355 km/h 330 km/h 11 m/s 6200 m

Fig. 2. The PZL M28 Skytruck aircraft [9, 10] It is a semi mono-cook metal structure. The airplane wing consists of a rectangle centre wing and two trapezium outer wing parts. It features the 3º anhedral and 4º geometric twist.

PART OF WING MODEL – GLOBAL MODEL The aim of the work was to analyze low skin riveted joint near rib no 21 on the PZL M28 Skytruck wing. In the case of thin wall consisting of frame (ribs, stringers) and skin, it is assumed (for finite rib stiffness) that force system closes in two bulkheads (between three ribs), on each side. In order to put loads from rest of wing correctly, the model includes three bulkheads on each side of rib no 21, i.e. part between ribs no 18 and 24 (fig. 3). Thanks to that, there is the correct stress distribution on bulkheads near rib no 21.

Methods for FEM analysis of riveted joints of thin walled aircraft structures

945

Fig. 3. PZL M28 wing schema with marked analyzed region [11] The wing is a torsion box construction. The wing torsion is made from spars, ribs and skin panels stiffened with stringers. The spars consist of upper and lower flanges of T-section and webs reinforced with struts. Ribs consist of upper and lower flanges, webs reinforced with struts and connectors. The part of wing FEM model is shown in figure 4. Part of upper skin was hidden to show internal structure.

Fig. 4. The part of wing FEM model The model was prepared in MSC PATRAN software, calculation were made with MSC NASTRAN software. The model was used to gain boundary conditions for more accurate models. On a basis of previous analyses [12,13] rivets weren’t represented. Shell elements (Quad4, Tria3) and linear material models were used.

946

Jerzi Kaniowski et al.

The boundary conditions were taken on a basis of the operational data. All moments were converted into forces and applied to the structure. Forces were applied to each rib, near spars. The model was fixed on one side. Nonlinear analysis was performed (Sol 106). Only geometrical nonlinearity was taken into account. Figure 5. presents equivalent stresses and displacements obtained in the analysis.

Fig. 5. Equivalent stresses and displacements of the global model On the upper skin there are some disturbances connected with local buckling, but there is no significant influence of this phenomenon on the lower skin.

RIVETED JOINT MODEL – I LEVEL LOCAL MODEL Fragment of the bottom skin, near rib no 21 (fig. 6) was chosen for local I level analysis.

Methods for FEM analysis of riveted joints of thin walled aircraft structures

947

Fig. 6. Analysed fragment on bottom skin The Riveted joint FEM model was built for this region. The Presence of the rivet was taken into account, as well as the distance between middle surfaces of jointed parts. The structure of the model is presented in fig. 7. Dimensions of the model are 400 x 150 mm. Shell elements (Quad4, Tria3) were used. For rivets in the middle, where one can see holes, model presented in fig.8 was used. The model is based on previous work [12,13]. It consists of two circular surfaces, with rivet diameter, connected with rigid MPC element. On the surfaces’ edges, on sheets plans there are GAP contact elements. They model interaction between the rivet shank and sheets. Manufactured head and driven head are not represented. Instead of them displacements in axial direction (z) of nodes of sheets and rivet, on hole edges were tied with MPC elements. Outside the middle part, rivets were modelled as a singular rigid MPC element. Boundary conditions (forces) were set based on global model calculation. Previous analyses [12,13] showed that in submodelling technique, in the case of significant flexibility difference between models, it is better to use forces instead of displacements, as boundary conditions. Residual stresses connected with riveted process were modelled using temperature and thermal coefficient [14,15]. Nonlinear analysis was performed with MSC NASTRAN software (Sol 106). Nonlinear material model was used (except of rivets). Figure 9. presents equivalent stresses and displacements obtained in an analysis.

Jerzi Kaniowski et al.

948

Fig. 7. Structure of the riveted joint model

Methods for FEM analysis of riveted joints of thin walled aircraft structures

Fig. 8. Rivet model

Fig. 9. Equivalent stresses and displacements of the riveted joint model

949

950

Jerzi Kaniowski et al.

Fig. 10. Equivalent stresses on skin (top) and doubler (bottom)

Methods for FEM analysis of riveted joints of thin walled aircraft structures

951

On the internal skin (left side) there are higher stresses due to smaller thickness (A skin is locally etched on the internal side, from 1,2 mm to 0,9 mm). Figure 10 presents equivalent stresses on the central part of skins and doubler. The biggest stresses are near rivets connecting stringers with doubler and skins. This work focuses on rivets in the central area, between stringers. On external skin (fig.10, right side) around rivets one can see the influence of asymmetric stringers configuration. On internal skin stress distribution is more symmetrical. Rivet from this area, as far as possible from stringers, (marked with arrow in fig. 10) was chosen for local II level analysis. It connects two layers (the skin and the doubler, without the rib flange) so it is simpler to be modelled. Figure 11. presents equivalent stresses in the area analyzed on the next modelling level.

Fig. 11. Equivalent stresses around the rivet analyzed on the next modelling level.

RIVET MODEL – II LEVEL LOCAL MODEL Simple model The Solid model of the chosen rivet (fig. 10) was built in ANSYS software. It covers the fragment (square of 12 mm side) of skin and doubler with the rivet.

Jerzi Kaniowski et al.

952

Solid (8node Solid 185) and contact elements were used. The Structure of the model is presented in fig. 12. Nonlinear material models were used. Geometry of the rivet was taken according to USSR aerospace industrial branch standard OST 1 34040-79 (shank diameter 3 mm, rivet without compensator) and riveting process instruction used in the PZL Mielec [2]. Loads were applied in two stages. In the first stage residual stresses were inducted. In the second stage external loads were added based on riveted joint model calculation (point 4).

1 ELEMENTS U MAY 16 2008 20:44:09

Z

X Y

Fig. 12. Structure of the rivet model Flexibility of both models (the riveted joint model and the rivet model) is similar. Therefore displacements (not forces, as on riveted joint) were set as boundary conditions. In FEM displacements are calculated with bigger accuracy and it is more convenient to use them. A special tool of ANSYS software for shell-to-solid submodelling was used. Riveting process wasn’t simulated. Residual stresses were induced by setting temperature and thermal coefficient. After riveting process, in a rivet shank, there are compressive radial stresses and tensile axial stresses. Shank rivet elements should expand in radial and shorten in axial direction. It was achieved by defining orthotropic thermal property of rivet material. Thermal coefficient on radial and tangential direction is positive, on axial direction is negative. This method was used among others in [13,14]. Thermal coefficient value and temperature were set in such a way, that average stresses (radial on hole edge and axial on a sheets mating surface) were similar as in [6] (for a ratio of driven head to shank diameter equalling 1,5). The author of [6] performed numerical simulation of riveting process on an axi-symmetric model.

Methods for FEM analysis of riveted joints of thin walled aircraft structures

953

Results of calculations after stage I are presented in fig. 13-14. ¼ of the model was hidden. Fig. 13 shows radial stresses on the whole model and on the doubler. Tangential and axial stresses are presented in fig. 14. One can see that rivet shank diameter increases. It causes compressive radial stresses on the rivet and the sheets. On a tangential direction there are compressive stresses in the shank rivet and tensile stresses in the sheets. The sheet edges were unsupported. Normally they are restricted by the rest of a sheet and then, in sheet, compressive tangential stresses can appear. The rivet clamps the sheets. In axial direction there are tensile stresses in the rivet and compressive stresses in the sheets around the rivet.

1

a)

NODAL SOLUTION STEP=1 SUB =15 TIME=1 SX (AVG) RSYS=5 DMX =.024375 SMN =-270.641 SMX =162.415

b)

1 NODAL SOLUTION

MAY 30 2008 20:41:56

STEP=1 SUB =15 TIME=1 SX (AVG) RSYS=5 DMX =.01085 SMN =-264.454 SMX =1.285

MAY 26 2008 21:16:00

MX

MX MN

X Z Y

Y

X

Z MN

-270.641

-174.406 -222.523

-78.172 -126.289

18.063 -30.054

114.298 66.18

-146.348

-205.401

-264.454

162.415

-234.927

-175.874

-28.241

-87.294 -116.821

1.285

-57.768

Fig. 13. Residual radial stresses; a) in the model, b) in the doubler

1

1 NODAL SOLUTION STEP=1 SUB =15 TIME=1 SY (AVG) RSYS=5 DMX =.024375 SMN =-216.182 SMX =185.436

NODAL SOLUTION

a) MAY 30 2008 20:42:13

STEP=1 SUB =15 TIME=1 SZ (AVG) RSYS=5 DMX =.024375 SMN =-71.008 SMX =89.188

b) MAY 30 2008 20:42:29

X Z

X Z

MX

Y

Y

MX MN

MN

-216.182

-82.31

6.939

-71.008

140.812

51.563

-37.685

-126.934 -171.558

96.188

185.436

-17.609

71.389

35.79

.190476

-35.409 -53.208

17.99

53.589

89.188

Fig. 14. Residual stresses; a) tangential, b) axial. Figures 15-16 presents results from the rivet model (present work - left column) and from [6] (riveting process simulation, axi-symmetric model - right column). Stresses were divided by yield stress. At present work a ratio of driven head to shank diameter is equal to 1,5. Points on the graphs indicate nodes. One can see that mesh of axi-symmetric model was finer.

Jerzi Kaniowski et al.

954

The radial stresses through the thickness on the edge of the hole are presented in figure 15. The author of [6] assumes that mating surface is a plane of symmetry. It means that the driven head is made at both sides (slug rivet). Therefore he presented results only for one sheet. The character of the graphs is similar but in the left graph absolute values are lower. Moreover the difference between the largest and smallest values is smaller in the left graph.

a)

b )

Sr(z) z / t [\] 0

0,5

1

1,5

2

0

-0,2

D/Do=1.5

Sr/ Re [\] -0,4

skin doubler

-0,6

-0,8

-1

Fig. 15. Residual radial stresses through the thickness on the edge of the hole; a) at present work, b) from [6] The residual axial stresses at the mating surface are presented in figure 16. Near the rivet character is different, farther is similar. The maximum absolute value is smaller in the left graph.

a)

Sz(r) r/R [\] 1

1,5

2

2,5

3

3,5

4

0

b)

-0,02

D/Do=1.5

Sz / Re [\] -0,04 -0,06

Sz(r)

-0,08

Driven head

-0,1

Manufactured head

-0,12

Fig. 16. Residual axial stresses at the mating surface; a) in present work, b) from [6] The differences are mainly result of different way of modelling residual stresses. Slightly different are material properties. Generally, results indicate small accuracy

Methods for FEM analysis of riveted joints of thin walled aircraft structures

955

of presented method of residual stresses modelling. It’s not good enough for the detail analysis of riveting process, but could be useful in bigger models including few or even several dozen rivets. Conducting of riveting process simulation on such big models is practically impossible. In the second stage loads (displacements), obtained in riveted joint analysis (point 4), were applied to the model. Figures 17-18 present results. To make them more visible, quarter of model was hidden and displacements were multiplied 10 times.

a)

1 NODAL SOLUTION

b)

1 NODAL SOLUTION

STEP=2 SUB =13 TIME=2 SX (AVG) RSYS=5 DMX =2.742 SMN =-416.652 SMX =339.722

STEP=2 SUB =13 TIME=2 SY (AVG) RSYS=5 DMX =2.742 SMN =-337.164 SMX =272.314

MAY 30 2008 20:47:12

MAY 30 2008 20:47:29

MX

MX MN

MN

-416.652

-248.569 -332.61

-80.485 -164.527

87.598 3.556

-337.164

255.681 171.639

-201.725 -269.444

339.722

-66.285 -134.005

69.155 1.435

204.594 136.874

272.314

Fig. 17. Stresses; a) radial, b) tangential

1 NODAL SOLUTION STEP=2 SUB =13 TIME=2 SZ (AVG) RSYS=5 DMX =2.742 SMN =-172.216 SMX =310.07

1

a)

NODAL SOLUTION

MAY 30 2008 20:47:49

STEP=2 SUB =13 TIME=2 SEQV (AVG) DMX =2.742 SMN =1.528 SMX =288.714

b) MAY 30 2008 20:48:34

MX

MN

MN

MX

-172.216

-65.041

-118.628

42.134 -11.454

149.308

95.721

256.483

202.895

310.07

1.528

97.257

256.804

192.985

129.166

65.347 33.438

161.076

224.895

288.714

Fig. 18. Stresses; a) axial, b) equivalent Fig. 19-20 present radial and equivalent stresses in the sheets. The mating surface is visible.

Jerzi Kaniowski et al.

956 1 1 NODAL SOLUTION STEP=2 SUB =13 TIME=2 SX (AVG) RSYS=5 DMX =2.74 SMN =-420.548 SMX =174.642

NODAL SOLUTION

a)

STEP=2 SUB =13 TIME=2 SEQV (AVG) DMX =2.74 SMN =23.212 SMX =286.63

MAY 30 2008 20:52:34

b)

MAY 30 2008 20:53:01

MX MN MN MX

-156.019

-288.284

-420.548

-222.151

-354.416

-23.754 -89.887

257.361

198.824

286.63

228.093

169.555

111.018

52.481

174.642

140.287

81.749

23.212

108.51

42.378

Fig. 19. Stresses in skin; a) radial, b) equivalent

1 NODAL SOLUTION

1 NODAL SOLUTION

a)

STEP=2 SUB =13 TIME=2 SX (AVG) RSYS=5 DMX =2.742 SMN =-387.68 SMX =129.341

MAY 30 2008 20:55:37

STEP=2 SUB =13 TIME=2 SEQV (AVG) DMX =2.742 SMN =9.118 SMX =287.271

b) MAY 30 2008 20:57:10

MX MN MX

MN

-387.68

-272.787 -330.233

-157.893 -215.34

-43 -100.446

71.894 14.447

129.341

132.742

70.93

9.118 40.024

101.836

194.553

163.647

256.365 225.459

287.271

Fig. 20. Stresses in doubler; a) radial, b) equivalent Stress system around rivets is similar to the results from the shell model of the riveted joint (point 4). On the hole edges there are radial stress concentrations, significantly bigger then in the shell model, but maximum value of equivalent stresses is somewhat smaller. On the edges, on the mating surface, near driven head and manufactured head there are stress concentrations. The presented model is quite simple. More accurate boundary conditions could be obtained by use of contact elements between sheets, near the analyzed rivet, in the shell model. Currently, the sheets interact with each another only through the rivets. In the solid model some details, like filets and chamfers, were neglected. Moreover, material models are simple too. Because this rivet model isn’t suitable for detailed analysis, such action isn’t necessary. An example of the detailed analysis of riveting process, performed in the Military University of Technology, is shown in point 5.2. After riveting process simulation external loads, based on analysis of bigger part (described in point 4), could by applied to the model.

Methods for FEM analysis of riveted joints of thin walled aircraft structures

957

Numerical model of upsetting process (according to [A]) A The analysis was performed for the solid mushroom rivet (shank diameter 3,5 mm) joining two aluminium sheets (thickness 1,2 mm). The three-dimensional numerical model of the neighbourhood (10,5 mm wide) of a single rivet was considered (Fig. 21). Dimensions of the mushroom rivets were taken according to Russian standard [OST 1 34040-79] (head radius R = 4,2 mm, diameter D = 7 mm, height h = 1.88 mm, without compensator). The radius Rp of the rounded tool surface was equal to 4,8 mm.

lower sheet

upper sheet

rivet

rounded tool

Flat tool

Fig. 21. Numerical model of the neighbourhood of the mushroom rivet Rivet and aluminium sheets were described using three-dimensional elements type Hex8 [16]. Elastic-plastic material model was considered. Simplified stress-strain curves for alloys PA25 (rivets) and D16TN (sheets) obtained from uniaxial tension and compression tests are presented in Fig. 22. Real values of stress and strain were used to describe these curves [16]. The analysis of 3D state of stress in presented model was performed using the von Mises yield criterion.

A

Elżbieta Szymczyk, Jerzy Jachimowicz , Grzegorz Sławiński, Agnieszka Derewońko: Modelowanie i analiza mes technologicznego procesu nitowania pokryć lotniczych.

Jerzi Kaniowski et al.

958

σ = where Re – yield stress,

2 d d σ ijσ ij = R e 3

(1)

σ ijd – deviator part of Cauchy stress tensor

500

stress [MPa]

400 300 200 sheet rivet

100 0 0

0,05

0,1

0,15

0,2

strain []

Fig. 22. Nonlinear stress-strain curves Contact with friction was defined between the mating parts of the joint (7 areas in contact). The Coulomb friction model with friction coefficient μ = 0.2 was used. The penalty method was applied to numerically implement the contact constraints. Iterative penetration checking approach allows to change the contact condition within the Newton-Raphson iteration loop. Using this procedure, the iteration process is done simultaneously to satisfy both the contact constraints and global equilibrium using the Newton-Raphson method. This procedure is accurate and stable but may require additional iterations. Numerical analysis The numerical calculations were performed for two cases of upsetting (w1, w2) distinguished by the height of the formed rivet head (Table I, Fig. 21).

Head height h, h/h0 head diameter D, D/d

w1 [mm] 1.82 5.2

[%] 43 148

w2 [mm] 1.57 5.5

Table I. Formed rivet head dimensions

[%] 37 157

Methods for FEM analysis of riveted joints of thin walled aircraft structures

959

Fig. 23. Riveted joint deformations after upsetting Riveted joint deformations are shown in Fig. 23, while rivet deformations for two cases of upsetting are given in Fig. 24. In the first case (w1), the height of formed rivet head has maximum value, when the diameter reaches nominal value [17]. For the case w2, diameter of formed rivet head reaches maximum permissible value according to riveting process manual [2].

a)

b) Fig. 24. Rivet deformations a) w1 b) w2

Jerzi Kaniowski et al.

960

a) σmin = - 408 192

130

b) σmin = - 411 70

6.5

-55

-117 -179 -240 -302 -364 -426 -488

Fig. 25. Radial stresses in rivet after unloading a) w1 b) w2

b)

a)

341

260

182

102

23

-57

-136 -216 -295 -375 -454 -534

Fig. 26. Hoop stresses in rivet after unloading a) w1 b) w2

Rivet upsetting causes metal sheet joining and filling of the rivet hole. Because of increasing fatigue life of the riveted joint, the rivet upsetting and compressive stresses distribution in the sheets around the hole should be uniform and exceed yield stress level [18,19].

Methods for FEM analysis of riveted joints of thin walled aircraft structures

961

Radial and hoop stresses distribution in the rivet after upsetting for two cases of loading is presented in Fig. 25 and Fig. 26 respectively. Compressive stresses are transmitted to sheets material around the rivet hole during the upsetting process. Contact stresses arise on mating surfaces of metal sheets (Fig. 27, 28). During riveting process (using press machine), stresses in rivet hole reach the value of 500 MPa (Fig. 27 – case w1 and w2). The value of contact stresses after unloading is equal 300 MPa (Fig. 28 – w1, w2).

a)

b)

0

50

184

277

369

462

553

645

738

830

922

1015

Fig. 27. Contact stresses during riveting a) w1 b) w2 (scale 1)

1107

Jerzi Kaniowski et al.

962

a)

b)

0

53

106

159

212

265

318

371

424

477

531

584

Fig. 28. Contact stresses after unloading a) w1 b) w2 (scale 1)

Irreversible plastic deformations of sheet material around the rivet hole remain after the riveting process. The plastic deformation region of sheets after upsetting for two cases of loading is shown in Fig. 29. Decreasing of height of the formed rivet head during riveting process causes the increasing of plastically deformed area around the rivet hole (Fig. 29a, b).

Methods for FEM analysis of riveted joints of thin walled aircraft structures

963

a)

b) Fig. 29. Plastic deformations of sheet material a) w1 b) w2

Negative stresses (compressive stress state), considerably exceeding yield stress level, appear in the material of sheets around the rivet hole (Fig. 30). Reduced stress distribution during riveting (pressing) is presented in Fig. 31.

Jerzi Kaniowski et al.

964

CONCLUSIONS Riveting process has a significant impact on the stress state in joined metal sheets, especially in the case of uniform filling of the rivet hole by the rivet shank. A certain method of material transfer from rivet head to rivet shank is the application of rivets with compensator [OST 1 34040-79]. Uniform distribution of residual stresses in the rivet hole increases fatigue life of the rivet joint. Residual stress values should be adjusted to the average level of operating stresses in cover panels of the considered aircraft. In the case of subsonic airplanes and helicopters range of these stresses is equal to 20 – 150 MPa (stress concentration coefficient is not considered).

a)

b)

22

-23

-69

-115 -160 -206 -252 -297 -343 -388 -434 -480

Fig. 30. Radial stresses in sheets after unloading a) w1 b) w2

Methods for FEM analysis of riveted joints of thin walled aircraft structures

965

a)

b)

0

56

115

173

230

288

320

374

420

520

577

635

Fig. 31. Reduced stresses in sheets during riveting (pressing stage) a) w1 b) w2

DISCUSSION AND CONCLUSIONS FOR FURTHER WORKS The paper presents the current state of the work. The riveted joint was analyzed at three levels with submodelling technique. The Model of a single rivet turned out to be too simple and not accurate enough. More precise model (simulation of riveting process) was developed. The next step will be to analyze the model of a rivet with compensator and load it with boundary conditions from riveted joint analysis.

Jerzi Kaniowski et al.

966

REFERENCES [1]

Krukowski A., Szewczyk K.: Basis of Machine Design. Part I Nondetachable joints (in polish). WAT, Warsaw 1987.

[2]

Riveting process manual for rivets with a compensator (in polish), PZL Mielec, no OPI 412-79.

[3]

Mushroom head rivets with compensator. USRR industrial branch standard no OST 1 34040-79.

[4]

Szymczyk E., Jachimowicz J., Sławiński G.: Analysis of technological clearance influence on residual stress fields around the countersunk rivet (in polish), Programy MES w Komputerowym Wspomaganiu Analizy, Projektowania i Wytwarzania, Kazimierz 2007.

[5]

Segerfröjd G., Wang G. S., PalmbergB., Blom A. F.,: Fatigue Behaviour of Mechanical Joints: Critical Experiment and Statistical Analysis. Fatigue in New and Ageing Aircraft. Proceeding of the 19th symposium of the International Committee on Aeronautical Fatigue, 18-20 June 1997, Edinburgh, Scotland, editors R. Cook & P. Poole, EMAS PUBLISHING, pp. 575-598.

[6]

Muller R., An experimental and numerical investigation on the fatigue behaviour of fuselage riveted lap joints, Doctoral Dissertation, Delft University of Technology, 1995.

[7]

Wanhill R.J.H., Effects of cladding and anodising on flight simulation fatigue of 2024-T3 and 7475-T761 aluminium alloys. NLR report no NLR TR 85006 U, Amsterdam 1985.

[8]

Kvasov f.I., Fridlyander I.N. (ed.), Aluminium Alloys (in russian), Moscow 1984.

[9]

http://airnews.pl

[10]

http://www.militarium.net

[11]

PZL M28 Maintenance manual.

[12]

Wronicz W., Methods for global and local FEM analysis of riveted joints on the example of the PZL M28 Skytruck aircraft (in polish), Master’s Dissertation, Warsaw University of Technology, Warsaw 2008.

[13]

Jachimowicz J., Wronicz W., Some problems of FEM modelling of riveted thin-walled aircraft structures (in polish), Przegląd Mechaniczny nr 5'08.

Methods for FEM analysis of riveted joints of thin walled aircraft structures

[14]

967

Fawaz S., Schijve J., de Koning U. A. Fatigue Crack Growth in Riveted Joints, Fatigue in New and Ageing Aircraft. Proceeding of the 19th symposium of the International Committee on Aeronautical Fatigue, 1997, Edinburgh, Scotland, EMAS PUBLISHING 1997.

[15] Jachimowicz J., Kajka R., Problems of FEM modelling of riveted joints (in polish), VIII Konferencja Naukowo-Techniczna; Programy MES w komputerowym wspomaganiu analizy, projektowania i wytwarzania, Rynia, 2003. [16]

MSC Marc Theoretical Manual, MSC Corp. 2004.

[17]

Grigoriew W., Gołdowski P.: Riveting of light-alloys structures (in polish), MON 1954.

[18]

Rijck J.J.M.: Stress Analysis of Fatigue Crack in Mechanically Fastened Joints, Doctoral Dissertation, Delft University of Technology, 2005.

[19]

Szymczyk E., Jachimowicz J., Derewońko A.: Analysis of residual stress fields in riveted joint, Journal of KONES, vol. 14, pp. 465 – 474, Warsaw 2007.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

ON THE CRACK GROWTH BEHAVIOR UNDER SIMPLE COMPRESSIVE LOADS OF 2024-T3 ALUMINIUM ALLOY M. Krkoška1,2, R.C. Alderliesten2, R. Benedictus2 1

2

Materials Innovation Institute, Delft, the Netherlands Aerospace Materials and Structures, Delft University of Technology, the Netherlands

Abstract: In real life, many engineering components and structures are subjected to complex variable load conditions. Sudden changes in loading are known to significantly influence the fatigue behaviour of materials. However, simplified, constant amplitude data are usually altered in different ways and subsequently used in order to predict the variable amplitude fatigue. This practice is known to have its limitations when precision of predictions is concerned. Obviously, there is a great need for an increase in understanding of variable amplitude fatigue, re-evaluation of existing models and development of useful concepts for engineering practice. The results and observations obtained from the investigation of the crack growth behaviour of AA2024-T3 aluminium alloy are outlined in this paper. Specifically, the crack growth lives of samples loaded under simplified variable conditions containing constant amplitude and varying underload cycles are presented. The crack growth lives were shown to be very dependent on the underload magnitude (stress ratio) and on their occurrence (grouping) in the load sequence. Fractographic investigations show that crack growth lives affected by crack extension initiated by applied underloads rather than by crack propagation initiated by constant amplitude loading. Experimental observations on crack path changes and plastic deformation (related to crack wake closure) induced by selective variable amplitude loading is also presented and discussed with respect to crack growth life.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 969–986. © Springer Science+Business Media B.V. 2009

970

M. Krkoška, R.C. Alderliesten and R. Benedictus

INTRODUCTION In real life, many engineering components and structures are subjected to complex variable load conditions. Sudden changes in loading are known to significantly influence the fatigue behaviour of structures [1-16]. Several models were proposed in the past regarding crack propagation during cyclic loading. However, these models mostly assume simplified load conditions; particularly the constant amplitude loading that produce flat crack planes, perpendicularly oriented with respect to the loading direction. Simplified Mode I opening conditions are considered in these models. Although, it took many decades of research to develop these models, they are considered to be inconsistent when variable loading conditions are considered. [17-22] The variable amplitude loading is known to produce progression marks on fatigue crack surfaces. These progression marks are generally local changes in the crack path. These path changes are used to aid crack growth measurement, but they are usually not considered in the prediction of the crack growth behaviour. Consequently, understanding these path changes and their influence on progression of fatigue cracks on microscopic level is essential to fully understand fatigue crack growth propagation macroscopically. Most of the available literature concerning fatigue under variable amplitude loading consists of information regarding overloads and its interaction with constant amplitude cycles loading, both in tension region. Several authors attributed this interaction to different mechanisms; to the state of residual stresses ahead of the crack tip, to different crack closure mechanisms, to material changes ahead of the crack tip, to crack front irregularities and sample geometry, to name a few. However, less literature is available on the effect of underloads, particularly in compressive mode and on its effect on crack propagation and fracture mechanisms. The results of the crack growth behaviour investigation on AA2024-T3 aluminium alloy are presented in this paper. The crack growth lives of samples loaded with test sequences consisting of individual and multiple underloads that varied in magnitude (tensile, compressive, mixture of tensile and compressive) will be discussed in the paper. Also the crack propagation and especially the crack path changes cased by applied underloads as observed with fractography will be presented and discussed.

EXPERIMENTAL Testing procedure A set of load sequences that contained fully reversed underloads in combination with constant amplitude cycles was designed in order to investigate the effect of underload size, spacing and grouping. Figure 1 provides a schematic illustration of all sequences applied in the test program. All tests were performed employing a

On the crack growth behavior under simple compressive loads of 2024-T3

971

100 kN MTS servo-hydraulic machine at a frequency of 10 Hz in ambient environment (air, temperature). The load sequence labelled as CA consisted of constant amplitude (CA) cycles with a maximum load that produced a nominal net section stress of 200MPa and a stress ratio (R=Rmin/Rmax) equal to 0.5 (high R cycles). The load sequences labelled as SU_C (single underload_compressive) and SU_T (single underload_tensile) consisted of single fully reversed underloads separated by varying number of CA cycles (specifically, 10, 20, 30, 50, 80 and 130). Fully reversed compressive loads with stress ratio R=-1.0 were applied in sequence SU_C while reversed underloads with a stress ratio R=0 were applied in sequence SU_T (low R cycles). Sequence MU_C (multiple underload_compressive) and MU_T (multiple underload_tensile) consisted of grouped underloads, increasing subsequently in amount from 1 to 5, separated by 100 CA cycles. A similar logic was adopted for these sequences as for SU_C and SU_T: SU_C contained fully reversed compressive loads with stress ratio R= -1.0, while MU_T contained reversed underloads with a stress ratio R=0. The load sequence SU_VTC (single underload_variable tensile-compressive) consisted of single fully reversed underloads progressively increasing in magnitude (stress ratio R= 0.45, 0.25, 0, -0.25, -0.5, -0.75 and -1.0) separated by 100 CA cycles. This testing sequence was modified by selecting either only positive underloads, as in sequence SU_VT (single underload_variable tensile) (stress ratio R=0.45, 0.25 and 0) or by selecting only negative underloads as in sequence SU_VC (single underload_variable compressive) (stress ratio R=-0.25, -0.5, -0.75 and -1.0) separated by CA cycles. Table I provides the overview of load sequences based on the characteristic loading patterns as they were used in test matrix.

Figure 1. Schematic illustration of all load sequences used in testing program

M. Krkoška, R.C. Alderliesten and R. Benedictus

972 Tensile Underloads

Tensile-Compressive Underloads

Compressive Underloads

SU_VT

SU_VTC

SU_VC

SU_T

-

SU_C

MU_T

-

MU_C

Table I. The overview of load sequences used in testing matrix

Test specimen Flat, dogbone shaped specimens were machined from AA2024-T3 sheets with a length of 177mm, a thickness of 3.2 mm and a width of 40mm. The reduction in width was introduced by a radius of 100mm machined on both sides (to give a net width of 25mm). Additionally, the U-shaped notch of 0.1mm wide and 0.25mm deep with semicircular tip was machined at one of the sample’s side (using electro discharge machining), to propagate a crack in LT direction. This specimen geometry gives an initial stress concentration (Kt) equal to 5.54 [23]. Figure 2 provides all details on the specimen’s geometry. Additionally, anti-buckling plates were attached to the sample prior to testing with openings machined on its sides to allow optical measurements of crack propagation by means of travelling microscope.

Figure 2. Schematic illustration on dog-bone shaped sample used in testing program

Fractography A JOEL JSM7500F scanning electron microscope with secondary electron detector was employed in all the fractographic investigations.

On the crack growth behavior under simple compressive loads of 2024-T3

973

RESULTS AND DISCUSSION Crack growth lives The crack growth life curves obtained for AA2024-T3 alloy tested under load sequences containing single tensile and compressive underloads are plotted in Fig.3. For comparison, also the constant amplitude (CA) load sequence test results are plotted.

crack depth [mm]

100

10

CA

SU_VTC

SU_VT

SU_VC

SU_T

SU_C

1

0.1

0.01 0

20000

40000

60000

80000

100000

120000

140000

crack growth life (cycles) Figure 3. Summary of the crack growth lives obtained from load sequences employing single tensile (▲■) and compressive (Δ □) underloads. The crack growth lives for CA (●) are also plotted.

It is interesting to note the direct reduction in crack growth life (as a function of crack depth and number of loading cycles) from CA loading with the incorporating underload cycles in test sequences. It was observed that tensile underload cycles (filled marks) reduce the crack growth life by approximately 30 %, while compressive underloads (filled marks) reduce the crack growth life up to 50 %. The increase in underload magnitude (decrease in load ratio R) as used in the load sequences could be attributed directly to the reduction in crack growth life. Based on the data in Figure 1, the following order for decreasing crack growth lives could be obtained: CA > variable tensile underloads only (R= 0.45, 0.25 and 0) > tensile underloads only (R= 0) > variable tensile-compressive underloads (R= 0.45, 0.25, 0, -0.25, -0.5, -0.75 and -1.0)> variable compressive underloads only (R= -0.25, 0.5, -0.75 and -1.0)> compressive underloads only (R= -1.0).

M. Krkoška, R.C. Alderliesten and R. Benedictus

974

Data obtained from tests employing multiple underload sequences (tensile: R=0 and compressive: R= -1.0) are plotted in Fig.4. Crack growth lives form CA and single underload sequences are also plotted in Fig.4 for comparison purposes.

crack depth [mm]

100

CA

SU_T

SU_C

MU_T

MU_C

10

1

0.1

0.01 0

20000

40000

60000

80000

100000

120000

140000

crack growth life (cycles) Figure 4. Summary of the crack growth lives obtained from load sequences employing multiple tensile (▲) and multiple compressive (Δ) underloads. The crack growth lives for CA (●) and single tensile and compressive underloads (■ and □), respectively are also plotted.

Similarly to the single underload sequences, introducing the multiple underload sequences were observed to reduce the crack growth life significantly. The 50 % life reduction could be approximately estimated, when compared to CA loading. Three interesting points could be noted from Fig.4. First, approximately the same crack growth lives were observed for both multiple underloads (tensile and compressive) sequences, with just slightly larger reduction in life for compressive underloads (MU_C>MU_T). Second, the reduction in crack growth life caused by multiple underload sequences is larger when compared to reduction caused by single tensile underloads sequence (MU>SU_T), but considerably smaller, when compared to single compressive underload sequence (MU<SU_C). This makes the single compressive load sequence the most damaging sequence from all sequences applied, regardless the amount of underloads! Third, the larger difference between crack growth curves obtained for multiple underload sequences could be recognized at the lower crack depths when compared to the larger cracks. At the larger crack depths, the curves obtained for multiple underload sequences seems to be almost identical.

On the crack growth behavior under simple compressive loads of 2024-T3

975

Figures 5 and 6 summarize the crack growth rates plotted as a function of crack depths for the cases when tensile (Fig.5) and compressive (Fig.6) underloads were used in the load sequences. Crack growth rate data obtained from the CA sequence are plotted as a reference in Figs.5 and 6. 0.01 CA

SU_T

SU_VT

MU_T

da/dn [mm/c]

0.001

0.0001

0.00001

0.000001 0

0.5

1

1.5

2

2.5

1/2

crack debth a [mm ] Figure 5. Summary of the crack growth rates (log scale) plotted as a function of crack depths (mm1/2 scale) for tensile underloads with corresponding trend-lines (— CA, --- Underloads).

Although, the trend-lines in Fig. 5 and 6 are more or less approximated trends, the following observations on crack growth rates could be made. The crack growth rates for all plotted sequences could be divided into two regions. The first region would be represented by steeply inclined trend-lines at the shortest crack depths, followed by gradually levelling region, transiting almost to the linear relationship at larger crack debths. All data obtained from load sequences containing underloads and plotted in Fig.5 shows faster crack growth rates when compared to the CA data. Interestingly, almost no difference was observed in crack growth rates as function of crack depth between the underload sequences. The scatter in data obtained at crack debths bellow 1 mm is evident for all load sequences, including the CA sequence. This observation highlights the scatter at small crack propagation rates. The same method (travelling microscope measurement) was used for determining the crack depth during the whole procedure, and yet, the data falls almost on the same trend-line at larger crack debths. This suggests that observed scatter at small crack growth rates could be the real phenomenon. Similar observations holds for crack growth rates obtained from tests performed by sequences containing compressive underloads, as shown in Fig.6.

M. Krkoška, R.C. Alderliesten and R. Benedictus

976

0.01 CA

SU_C

SU_VC

MU_C

da/dn [mm/c]

0.001

0.0001

0.00001

0.000001 0

0.5

1

1.5

2

2.5

1/2

crack debth a [mm ] Figure 6. Summary of the crack growth rates (log scale) plotted as a function of crack depths (mm1/2 scale) for compressive underloads with corresponding trend-lines (— CA, --Underloads).

Fractographic observations Fracture surface features of AA2024-T3 aluminium alloys as observed postmortem will be discussed in this section. Figure 7a shows the macroscopic fracture surface of the specimen tested with the SU_T load sequence. The starter notch is visible at the left side of the figure from where the crack initiated and subsequently propagated through the specimen. The fracture surfaces of this specimen appear to be levelled and oriented normal to the loading direction (or very close to this direction). The fracture surface located near the notch appeared faceted, revealing cleavage-like, transgranular fracture (point 1 in the Fig.7a with detail in Fig.7c). This brittle-like appearance of fracture surface is gradually changing with increasing distance from the starter notch and almost completely disappearing at the distance of approximately 0.5 mm. Fracture surface with clearly recognizable progression marks at even lover magnification levels is formed at this distance.

On the crack growth behavior under simple compressive loads of 2024-T3

977

(a)

(b)

(c)

(d)

Figure 7. Micrographs of fracture surface; (a) overview of the fracture surface, sample loaded with single tensile underloads (SU_T), starter notch on the left side with crack propagating from the left to the right; (b) closer view of fracture surface as seen at larger crack debths: typical fatigue fracture with markers; (c) closer view of fracture surface as seen at shorter crack debths: cleavage-like fracture; (d) detailed view of the crack front markers at the lowest crack debths (lowest stress intensity levels)

Focusing on the brittle fracture surface (at shortest crack debths) employing higher resolution imaging revealed the occurrence of progression marks (Fig.7d) that correspond to the position of underload blocks, although the cleavage-like fracture was noted originally (Fig.7c). Usually, featureless fracture planes with shallow orientation are reported at short crack depths and stress concentration levels. This difference could be explained as a result of the fairly high stress concentration factor (5.54) obtained with the current specimen geometry. Figure 8a reveals the typical appearance of the fracture surface as it could be observed at larger crack debths (larger stress intensity levels) with the step-like crack path changes occurring at the position of applied underload. The load sequence applied in this case was the SU_T and exact location of particular “steps” could be easily tracked by counting the striations in between the “steps” as they were formed by CA load cycles (10 - 10 - 20 - 30 - 50 ….), illustrated in Fig. 8a.

M. Krkoška, R.C. Alderliesten and R. Benedictus

978

(a)

(b)

(c)

(d)

Figure 8. Micrographs of fracture surface; (a) the step-like crack path changes occurring at the position of applied underload, single tensile underload sequence (SU_T); (b) detail of the ridge; (c) detail of damaged ridge with in-plane crack and small ripples – arrows; (d) plastically deformed fracture surface at the specimen loaded with single compressive underloads (SU_C).

The exact mechanism of ridge formation could be still a matter of speculation. However, as illustrated by Fig. 8a, the CA striations are formed at a certain plane. Subsequently, the general crack path changed at the moment of the underload forming a distinct ridge. The following CA cycles generate striations starting at the top of the ridge on a plane approximately parallel to the original plane until another ridge is formed. Subsequent underloads form therefore step-like features on the fracture surface. Figures 8b shows the detailed view of the ridge on the fracture surface revealing tearing-like fracture. One should by careful though, in jumping into the conclusions based on one observation only, since other mechanisms could be involved here. Figure 8c shows a detailed view of the ridge captured near the position shown in Fig. 8b. The ridge in this image seems to be damaged heavily during the loading. This damage most likely occurs as a consequence of crack wake closure (roughness related closure) after the crack front advanced. There are two features in Figs.8b and 8c supporting this assumption. First, the few CA striations formed right at the top of the ridge are permanently deformed, while subsequently formed CA striations seems intact. This suggests that CA striations were first formed by applied load cycles and then deformed as the crack front

On the crack growth behavior under simple compressive loads of 2024-T3

979

advanced forward. Second, the in-plane damage of the ridge happened most likely in several steps rather than in one step, because the ripples are evidently on different surfaces (indicated by arrows in Fig.8c. Increasing the magnitude of underloads (decrease the load ratio R) from tensile (R= 0) to compressive (R= -1.0) levels can increase the amount of permanent damage on the fracture surfaces and significantly complicate fractographic investigation. Figure 8d shows the micrograph taken from the sample loaded by SU_C load sequence. Post mortem reconstruction of the crack front location in relation to the load history is practically impossible. However, the effect of Mode II loading on the fracture surface appearance is still evident. An example of a fracture surface containing features characteristic for a specimen loaded with the sequence containing succeeding underloads (MU_T load sequence) is shown in Fig.9a. The pattern of marks on the fracture surface corresponds to the underloads in the load sequence, revealing features that appear to be groups of larger striations. Fig.9b provides a closer view at the fracture surface at the location of five larger striations, which are assumed to be induced by five successive underloads. It is evident that the last “striation” differs from the previous four in shape and size; It is believed that all marks were originally created as ridges by compression and unloading similar to the last ridge in the group (fifth in this case). However, they were crushed down in the load sequence by succeeding underload to form large “striations”. The last ridge was not crushed and therefore retained its shape since tensile CA cycles were applied in succession. Additionally, similar evidence of secondary damage as mentioned above (damaged CA striations, in plane damage at the position of the ridge) could be visible in Fig.9b, c and d, suggesting that mechanisms of crack propagation probably took place very similar to the cases where single underloads were applied. Figure 10 shows an overview of typical fracture surface appearance as it was observed for the specimen loaded with the load sequence containing underloads with varying magnitude (load sequence SU_VTC in this case). It could be concluded from fractographic observations that the observed ridge size correspond to the magnitude (inversely related to stress ratio R) of applied underloads. Ill-formed ridges were observed at the position of positive underloads (R= 0.45 and 0.25) while ridges gradually increasing in size were observed at position of negative underloads (R= -0.25, -0.5, -0.75 and -1.0), as marked by arrows in Fig.10. This trend was more pronounced for steeply tilted planes.

M. Krkoška, R.C. Alderliesten and R. Benedictus

980

(a)

(b)

(c)

(d)

Figure 9. Micrographs of fracture surface; (a) overview of the fracture surface, specimen loaded with multiple tensile underload sequence (MU_T); (b) detailed view of five successively formed ridges, first four plastically deformed and the fifth ridge still holding its original shape; (c and d) examples of plastically deformed ridges, specimens loaded with multiple tensile underload sequence.

Sudden amplitude changes in load sequences are known to introduce the interaction effects, which could result in significant changes of the crack growth rates from cycle to cycle. While single overloads and block of overloads are known to induce crack growth retardation (increase the crack growth life), the application of underloads is usually known to increase the crack growth rates (decrease the crack growth life). This effect is strongly influenced by the value of the applied stress ratio R [24, 25]. In contradiction with this general observation, insignificant interaction effects or even underload induced retardation as observed for several materials were reported by other authors [26-29]. The results presented in this paper support the general observation, confirming the reduction of the crack growth life by applying underloads for the applied load sequences and aluminium alloy AA2024-T3.

On the crack growth behavior under simple compressive loads of 2024-T3

981

Figure 10. Micrograph of fracture surface, the specimen loaded with single variable tensilecompressive underloads, the dependence between the size of the ridges and the stress ration of applied underload.

These observations can be explained with the fractographic features observed. No systematic change in striation spacing related to CA loading could be observed at any fracture surface induced by any particular load sequence. However, a direct relationship between the magnitude of a (single) underload (stress ratio) and the ridge size could be established. One should carefully distinguish between interaction effect and crack growth acceleration. The term “interaction effect” suggests that applied overload or underload will have an impact on the crack propagation caused by subsequently applied cycles. This was shown not to be the case; the increase in crack growth rate was attributed directly to the underloads rather than to the CA cycles loading. This would simply imply that no interaction effect was observed, although the decrease in the crack growth life was observed. Applying multiple underloads in succession (sequences MU_T and MU_C) considerably decreased the crack growth life compared to CA loading. However, this happens essentially in a way different to the single underload case. The crack growth life was significantly smaller for the multiple tensile underloads (MU_T) sequence, compared to the single tensile underload (SU_T) sequence. This observation is in agreement with the argument that crack propagation increases due to the underload itself (the same magnitude of underload in these cases). Multiple succeeding underloads increase the crack extension by formation of the ridges. Although, this argument seems not applicable to compressive underloads in the MU_C and SU_C load sequences. Multiple compressive underloads do not

M. Krkoška, R.C. Alderliesten and R. Benedictus

982

decrease the crack growth lives significantly. Despite that the crack growth life of the sample loaded by a single underload sequence decreased considerably, only slight decrease was observed for sample loaded by multiple underload sequence. The explanation might be that if the underloads are applied in succession, the first compressive load in the sequence will lead in formation of the ridge; subsequent underloads must first deform the ridge formed by the previous underload and form another crack path change (the ridge). If more underloads are applied in succession, the whole process is repeated: deformation of previously formed ridge and creation of a new one. This means that one fraction of the energy must be spent on deformation and another fraction on ridge formation, while in the case of a single underload, only a ridge is formed. Crack closure in the wake caused by the Mode II loading may also participate in this process, as it was observed that almost the entire fracture surface of the sample loaded with single compressive underload sequence was deformed (example in Fig.8d) while other samples only exhibit local deformation at ridges. Interestingly, the geometry of the ridge and the direction of the CA striations seem to vary with its local position. In particular the degree of tilting the local fracture plane seems to be an important parameter (detail in Figs.11b and 11c). The ridge formed at the plane normal to the loading direction revealed ductile, round-like appearance with striations growing from its top downwards (trailing face of the “bump”), and then levelling up approximately to the original plane (before ridge). This is in good agreement with observations documented by White et al. [24, 25].

(a)

(b)

(c)

Figure 11. The example of (a) differently shaped ridges with (b) round-like shape and (c) shear-like shape, the specimen loaded with single tensile underload (SU_T) sequence.

On the other hand, the ridge formed at the plane tilted away from the direction normal to the loading direction revealed tearing-like ridges (shear-like based mechanism) with striations formed at its top (disturbed at first) in direction parallel to original plane (before ridge). In this case, no “bumpy” ridge was formed, but rather step-like crack path changes were created. These surface morphologies seem

On the crack growth behavior under simple compressive loads of 2024-T3

983

to differ considerably at first, suggesting that different mechanisms were applied during their creation. However, this might be incorrect. Permanent plastic deformation was observed primarily at the top of the ridge and also some subsequent CA striations. It is possible that the bump-like appearance of the ridge is simply a result of plastic deformation, changing the appearance of shear-like ridge. Similar rounded shaped ridge is shown in Fig.12, supporting outlined view.

Figure 12. Detail of bump-like ridge geometry, showing the signs of plastic deformation, the specimen loaded with single tensile underload (SU_T) sequence

SUMMARY It was shown that increase in (single) underload magnitude could be directly attributed to the reduction in crack growth life. Tensile underload cycles were found to reduce the crack growth life by approximately 30 %, while compressive underloads were found to reduce the crack growth life up to 50 %. Further elaborated, following succession for crack growth lives could be drawn: CA> variable tensile underloads only (R= 0.45, 0.25 and 0) > tensile underloads only (R=0) > variable tensile_compressive underloads (R= 0.45, 0.25, 0, -0.25, -0.5, -0.75 and -1.0)> variable compressive underloads only (R= -0.25, -0.5, -0.75 and -1.0)> compressive underloads only (R= -1.0). The introduction of multiple underload sequences in tests was found to reduce the crack growth life by approximately 50 %, when compared to CA loading. Approximately the same crack growth lives were observed for both multiple underloads (tensile and compressive) sequences, with just slightly larger reduction in life for compressive underloads (MU_C>MU_T).

984

M. Krkoška, R.C. Alderliesten and R. Benedictus

The scatter in crack growth rates data was observed related to the crack depth dimension. The largest scatter of the crack growth rates was observed for the lower crack depths for all test sequences containing underloads. On the other hand, only very limited differences in the crack growth rates data (log-mm1/2 scale) was observed for larger crack debths. It was possible to explain the crack growth life data directly by interpreting the fractographic evidence. No systematic change in striation spacing caused by CA loading could be found at any fracture surface created by applying any load sequence emphasizing different spacing, size, and grouping of underloads. The direct relationship between (single) underload magnitude (stress ratio) and the ridge size formed was shown. The ridges are formed at the positions of applied underload forming step-like features. Detailed microscopic view of the ridge revealed tearing-like mechanism. The geometry of the ridge and direction of CA striations seems to vary depending on its local position; particularly degree of tilting for local fracture plane seems to be important parameter. However, the mechanisms of the ridge formation could be very similar at differently tilted planes despite the differences in appearance. This difference is probably caused by the plastic deformation taking place during crack propagation, creating bump-like ridge. Progression marks could be observed on the whole fracture surface. Although, cleavage-like, brittle fracture surface was observed at the lowest crack debths, employing the higher resolution microscopy technique revealed the occurrence of underload related markers.

ACKNOWLEDGMENT This research was carried out under the project number MC2.06269 in the framework of the research program of the Materials innovation institute M2i.

REFERENCES [1]

McMillan J.C. and Pelloux R.M.N. (1967) ASTM STP 415

[2]

Euw von E.F.J., Hertzberg R.W. and Roberts R. (1972) ASTM STP 513, p. 230

[3]

Schijve J. (1961) In: Advances in Aeronautical Science, vol. 3, p. 387

[4]

Ward-Close C.M., Blom A.F. and Ritchie R.O. (1989) Eng.of Frac. Mech. vol. 32, p. 613

On the crack growth behavior under simple compressive loads of 2024-T3

985

[5]

Vecchio R.S., Hertzberg R.W. and Jacard R. (1984) Fat. and Fract.of Eng. Mat.and Str. vol. 7,p. 181

[6]

Petit J., Tintillier R., Ranganathan N., Aid Abdedain M. and Chalant G. (1988), In: Fatigue Crack Growth under Variable Amplitude Loading, Elsevier Applied Science, London, p.162 Edited by J.Petit, D.L.Davidson, S.Suresh and P.Rabbe (Ed.)

[7]

Shin C.S. and Hsu S.H. (1993) Int. Jour. of Fat., vol. 15, p. 181

[8]

Sehitoglu H., McDiarmid D.L. (1980) Int. Jour. of Fat., vol. 2, p. 55

[9]

Iwasaki T., Katoh A. and Kawahara M. (1982) Naval Arch. and Ocean Eng., vol. 20, p. 194

[10]

Fleck N.A. (1985) Acta Meta., vol. 33, p. 1339

[11]

Ohrloff N., Gysler A., Lütjering G. (1988) In: Fatigue Crack Growth under Variable Amplitude Loading, Elsevier Applied Science, London, pp. 34-34, Edited by J.Petit, D.L.Davidson, S.Suresh and P.Rabbe (Ed.)

[12]

Zhang S., Marissen K., Schulte K., Trautmann K.H., Nowack H. and Schijve J. (1987), Fat. and Frac. of Eng. Mat. and Struct., vol. 10, p. 315

[13]

Stephens R.I., Chen D.K. and Hom B.W. (1976) ASTM STP 595, p. 172

[14]

Henkener J.A., Scheumann T.D. and Grandt A.F. (1990) In: Fatigue Crack Growth Behaviour of a Peak-aged Al- 2.56Li-0.092r Alloy, Proceeding of 4th International Conference Fatigue and Fatigue Thresholds, Honolulu, pp. 957- 962, 15-20 July, MPCE, UK,

[15]

Yu M.T., Topper T.H. and Au P. (1984) In: The Effect of Stress ratio, Compressive load and Underload on the Threshold Behavior of a 2024T351 Aluminium Alloy, Proceeding of Fatigue 84, Proc. 2nd International Conference on Fatigue and Fatigue Thresholds, pp. 179-190, Beevers C.J. (Ed.) Birmingham, UK

[16]

Topper T.H., DuQuesnay D.L. and Pompetzki M.A. (1992) In: Crack Closure, Damage and Short Crack Growth Under Variable Amplitude Loading, Proceeding of Conference Theoretical Concepts and Numerical Analysis of Fatigue, pp. 201-235, Bloom A.F. and Beevers C.J. (Ed.) Birmingham, UK

[17]

Laird C. (1967) ASTM STP 415, p.131

986

M. Krkoška, R.C. Alderliesten and R. Benedictus

[18]

Pelloux R.M.N. (1969) ASM Transactions 62, p.281

[19]

Neumann P. (1974) Acta Met., vol. 22, p. 1155

[20]

Neumann P. (1974) Acta Met., vol. 22, p. 1167

[21]

Tomkins B. and Biggs W.D. (1969) Jour. of Mat. S., vol. 4, No. 6, p. 532

[22]

Wanhill R.J.H. (1975) Metal. Trans. A, Vol. 6A, p.1587

[23]

Broek D. (1989) In: The Practical Use of Fracture Mechanics, Kluwer Academic Publishers, the Netherlands

[24]

White P., Barter S.A. and Molent L. (2008) Int. J. of Fatigue 30, 12671278.

[25]

White P., Barter S.A. and Wright C. (2009) Int. J. of Fatigue, in press

[26]

McMaster F.J. and Smith D.J. (1997) In: The Effect of Load Excursions and Sheet Thickness on Crack Closure Measurements, The 2nd Symposium on Advances in Fatigue Crack Measurements and Analysis, San Diego, California, USA

[27]

Yang R. (1994) Int. Jour. of Fat., vol. 16, p. 397

[28]

Carlson R.L. and Kardomateas G.A. (1994) Fatigue, vol. 16, p. 141

[29]

Buschermohle H., Memhard D. and Vormwald M. (1996) In: Fatigue Crack Acceleration or Retardation due to Compressive Overloads Excursions, Processing of International Conference on Fatigue’96, p. 583, Berlin

25th ICAF Symposium – Rotterdam, 27–29 May 2009

AN ANALYTICAL MODEL FOR LOAD TRANSFER IN A MECHANICALLY FASTENED, DOUBLE-LAP JOINT Ligieja Paletti, Calvin Rans and Rinze Benedictus Delft University of Technology, Delft, 2629 HS, The Netherlands

Abstract: An analytical model for load transfer in a mechanically fastened joint is presented. The model includes the effect of the main load transfer mechanisms: bearing and friction. The influence of the bypass load, secondary bending and fastener fit is neglected. The load splitting between the considered load mechanisms can be evaluated by defining a splitting factor α , which represents the amount of load carried by load transfer mechanisms considered. An iterative analytical procedure is established for the determination of the behavior of the splitting factor. The knowledge of the splitting factor behavior enables the prediction of the predominant load transfer mechanism for the joint configuration of interest.

INTRODUCTION In mechanically fastened joints, fatigue cracks have been observed, both in test and in practice, to nucleate in one of two primary locations. The first location is located close to the hole edge, along the joint net-section. The second location can be found at some distance from the hole edge, causing gross-section failure. The preference for cracks to nucleate at either location is determined by the prevalent load transfer mechanism. An overview of the load transfer mechanisms acting in a mechanically fastened joint is given in Fig. 1. For cracks nucleating at the hole edge, the location can be found on the net section, as for the pin-loaded hole problem described in [1]. When an external load is M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 987–1004. © Springer Science+Business Media B.V. 2009

L. Paletti, C. Rans and R. Benedictus

988

applied, the fastener transfers it to the joint by the application of a pressure on the upper part of the hole. The active load transfer mechanism in this case is bearing. The cracks tend to nucleate at the net section because of the combination of the pressure applied by the fastener and the vicinity of the root of the notch (hole), where the stress concentration factor is higher.

Figure 1:

Load transfer mechanisms acting in a mechanically fastened joint

Cracks which nucleate remotely from the hole edge indicate the occurrence of a damage mechanism known as fretting. Fretting is caused by small relative movements at the interface between two contact bodies, driven by cyclic loading. Evidently fretting and the surface properties of the bodies are closely related to each other, as demonstrated in [2], where the influence of the coefficient of friction on the occurrence of fretting has been studied. It has been demonstrated that an increase of the value of the coefficient of friction causes the failure mode to shift from net-section to gross-section. Therefore, friction is considered to be a key parameter for the shift of the crack location. Despite the evident influence of the coefficient of friction and, consequently, of friction on the crack nucleation location, the frictional contribution is typically neglected in existing models for load transfer in mechanically fastened joints. As a result, fatigue failure analyses based on such load transfer models neglect the possible fretting failure scenario. Thus analyses which exclude friction may lead to wrong or incomplete predictions, whose validity is limited to bearing-dominant cases. By considering now the existing models for frictional load transfer and fretting, similar remarks can be made. In those models, the applied load is assumed to be transferred fully by friction; this assumption is incorrect in a joint, where also the presence of the fastener must be considered. By neglecting the fastener, also its load transfer contribution (bearing) is neglected. Therefore predictions from failure

Load transfer in a mechanically fastened joint

989

analyses based on fretting models are partially incorrect and reliable only if friction is the predominant load transfer mechanism. The predictions from failure models based on the existing load transfer models have been shown to be potentially unreliable, due to inaccuracies intrinsic in the load transfer models. Both bearing and friction mechanisms must be included in order to obtain reliable predictions. If all the load transfer mechanisms are included in a model, the predominant contribution for a certain joint configuration can be determined. The knowledge of the predominant load transfer mechanism indicates the likely failure mode and orients the choice of an appropriate failure model. In this paper a model for the load transfer based on a particular joint configuration is presented. A parameter α representing the splitting factor between the two load mechanisms considered is defined. An iterative procedure is then established for the determination of α . Once the splitting factor behavior is known, the dominance of the two load transfer mechanisms of interest can be predicted, giving indications on the predominant load transfer mechanism acting in the studied joint configuration.

SIMPLIFIED JOINT CONFIGURATION The complex load transfer scenario in a mechanically fastened joint results in numerous mechanisms which can influence the relative contribution of friction and bearing to load transfer. As a result, it is desirable to first simplify the problem, allowing a friction-bearing load transfer model to be developed and verified for a simple case, and subsequently adding to it complexity. Looking at Fig.1, the mechanisms affecting load transfer in a joint can be classified as main load transfer mechanisms and secondary loads. The secondary loads (secondary bending and fastener fit) and the bypass contribution affect the overall load transfer by acting on bearing and friction. The effect of the bypass contribution is to divide the applied load over the number of fastener rows present in the joint. Once the percentage of load transferred by each row is known, bearing and friction have to transfer that percentage of load from one joint member to the other. Secondary bending affects bearing as a result of fastener tilting, which modifies the contact between the fastener itself and the hole edge. The tilting of the fastener affects also friction, because it results in an asymmetric contact pressure distribution and in an alteration of the overall clamping force.

990

L. Paletti, C. Rans and R. Benedictus

Fastener fit affects bearing by modifying the extension of the relative contact between the fastener and the joint members. Regarding friction, the contact stresses originated at the mating surface may be modified by an additional stress contribution due to the fastener fit. In order to initially limit the complexity of the problem being modeled, a doubleshear lap joint with a single row of fasteners and push fit bolts is chosen as the starting configuration. This configuration results in only bearing and friction transferring the load in the joint, as it eliminates secondary bending, bypass load and provides the simplest case of fastener fit. Once the analytical model has been developed and verified for this case, some of the additional effects described above may be introduced into the model.

OVERVIEW OF THE MODELING APPROACH By considering the geometry defined in the previous section, the determination of the splitting factor between bearing and friction requires the study of a contact problem. The region surrounding the fastener hole is modeled as a contact patch, where friction is present; thus the application of an external force to the joint generates a friction force. The friction force can entirely transfer the applied load only until a certain level of load. In fact, the application of the load causes the contact area to split into two concentric regions: the inner one is called the stick region and the outer one the slip region. For each region, the coefficient of friction assumes a specific value: the static coefficient corresponds to the stick region, while the kinetic coefficient is associated with the slip region. The boundary between these two regions is marked by the value of the stick radius. When a slip region is created in the contact area, the frictional force starts to decrease because of the lower value of the kinetic coefficient of friction. Therefore, the amount of load friction can transfer is reduced. Prior to frictional slip, the entire load is transferred by friction. Subsequently, bearing load transfer occurs once the critical load resulting in slip is reached. In order to characterize the behavior of the friction load transfer mechanism, it is necessary to determine the value of the frictional force. This can be done by using the contact mechanics theory. The first step is the determination of the normal contact pressure distribution acting on the contact area. The parameters defining the normal pressure distribution around a fastener hole are the equation of the pressure distribution, the value of the contact radius and the shape of the contact region.

Load transfer in a mechanically fastened joint

991

The equation of the pressure distribution is described by the Hertz contact model. The Hertz model may be used to model the contact of any combination of contact bodies which may potentially lead to fretting. The agreement between fretting stresses and the Hertz model has been verified analytically in [3] and by using finite element modeling in [4-5]. The contact radius is essential in determining the shape of the contact area and consequently the value of the generated friction force. In the Hertz theory the contact radius is dependent on the clamping force; by increasing the clamping force, the contact radius increases and the contact area widens. The knowledge of the contact radius is sufficient to define the size of the contact area because the Hertz theory has been defined as the contact pressure distribution arisen in the contact between two spherical bodies. The domain over which the Hertz normal pressure is distributed has a continuous circular shape. In a joint, the presence of the hole modifies the contact area. The hole introduces a discontinuity in the domain. The circular domain turns into an annular domain, whose inner radius is the fastener hole radius. The normal pressure distribution must change accordingly, modifying its equation such that the resultant still equals the clamping force. When all those parameters are considered, the normal pressure distribution can be used to determine the tangential pressure distribution, whose resultant is the friction force. The two pressure distributions are linked by the friction model, which determines the coefficients of friction for the stick and the slip regions. The friction force depends on the normal pressure distribution, the friction model and the size of the stick-slip regions. To characterize the stick-slip regions the value of the stick radius is necessary. The friction force so-calculated is based on the assumption that the entire applied load is transferred by friction. In a previous section it has been explained why this assumption cannot be valid in a mechanically fastened joint. To take bearing into account, a splitting factor must be defined which represents the amount of load transferred by each contribution. In order to determine the correct values of the splitting factor, an iterative procedure is established. This procedure is based on the assumption that, for the same joint configuration, fixed value of clamping force and same range of applied load, the friction force calculated using the contact mechanics equations and the friction force transferred according the splitting factor must coincide. This condition is verified for a single value of the splitting factor for every value of applied load. By evaluating the splitting factor for the chosen range of applied loads, it is possible to predict which load transfer mechanism is predominant for a certain joint configuration.

992

L. Paletti, C. Rans and R. Benedictus

MODEL DETAILS Clamping pressure distribution In order to characterize a contact problem, the clamping (or normal) pressure distribution must be determined. As explained in the previous section, the normal pressure distribution described by Hertz [7] has been proven to accurately model the clamping pressure between the joint members. The Hertzian pressure distribution is a function of the applied clamping force CF and of the contact radius a , acting over a circular domain in function of the radius r . This pressure distribution can be written as:

3 ⋅ CF ⎛ r 2 ⎞ p (r ) = ⋅ ⎜1 − ⎟ 2 ⋅π ⋅ a2 ⎝ a2 ⎠

1

2

(1)

As explained, due to the presence of the fastener hole, the clamping force is distributed over an annular region, whose outer radius can still be identified with the contact radius a , while the inner radius is represented by the radius of the hole a′ . In order to include the effect of the hole, the Hertzian pressure distribution of Eqn.1 has been modified such that the resultant of the normal pressure (represented by the clamping force) is the same for the continuous and the discontinuous configuration. This requirement has been fulfilled by multiplying the Hertzian distribution for a value K , dependent on the clamping force CF . Including the factor K , the pressure distribution for the hole configuration is described by Eqn. 2:

3 ⋅ CF ⎛ r 2 ⎞ p′ ( r ) = K ⋅ ⋅ ⎜1 − ⎟ 2 ⋅π ⋅ a2 ⎝ a2 ⎠

1

2

(2)

By integrating Eqn.2 over the annular contact area and by setting the result equal to the installation clamping force, the factor K can be determined. This approach has been validated by the determination of the actual normal pressure distribution between the joint members by using a pressure-sensitive film. In Fig.2 the pressure distribution resulting from the installation of a 6 mm steel bolt in an aluminum lap joint is presented. An applied torque moment of 5 Nm resulted in a clamping force of about 8450 N. Pressure sensitive film placed between the joint members during the tightening of the bolt provided a pressure history of the tightening process. After dismantling the joint, the pressure film was analyzed and the resulting pressure distribution plotted as in Fig. 2. In the same graph, the Hertz and the modified (using the factor K ) pressure distributions are also plotted. The agreement between the test data and the modified Hertz distribution is manifest, for a value of the multiplying factor K of about 3.2.

Load transfer in a mechanically fastened joint

993

Hertz pressure distribution Modified Hertz pressure distribution Pressure film test data

30

25

Normal pressure (MPa)

Pressure film threshold 20 Hole edge 15

10

5

0 0

0,5

1

1,5

2

2,5

3

3,5

4

4,5

5

5,5

6

6,5

7

Distance from the center of the hole (mm)

Figure 2: Validation of the modified Hertz pressure distribution using the pressure sensitive film. The contact radius a determines the size of the contact area and can calculated both theoretically and experimentally as a function of the clamping force. In the Hertz theory, the size of the contact area is a function of the applied preload, of the compliances of the contact bodies and their relative curvature [7]:

a3 =

3 ⋅ CF ⋅ A 3 ⋅ CF ⋅ A ⇒a= 3 8⋅ k 8⋅ k

(3)

In the case of a mechanically fastened joint, the contact is established between the two joint plates; consequently the relative curvature cannot be calculated. Nonetheless if the Hertz distribution is actually the pressure distribution at the mating surface, then Eqn. 3 must be verified also for the contact of two plates. An experimental relationship has been calculated using the data shown also in Fig.2. An approximated value of the contact radius a can be identified in Fig.2, for the case the clamping force is of about 8450 N. By collecting contact radius values for different values of the clamping force, a relationship between the applied preload and the contact radius, in the form of Eqn.3, can be determined and is shown by Eqn. 4:

a =C⋅P

1

3

⇒ C 0.334

(4)

L. Paletti, C. Rans and R. Benedictus

994

Friction The relationship between the normal and the tangential pressure distributions depends on the coefficients of friction. For this study, the Coulomb’s friction model is used. Therefore, two coefficients of friction are considered, one per regime: the static value μ s (stick) and the kinetic value μ k (slip). The transition between the two contact regimes happens when the external applied shear load Q reaches a limiting value of the friction force. This value of friction force is given by the product of the static coefficient of friction μ s and the clamping force CF :

Flimit = μ s ⋅ CF

(5)

When the condition Q < Flimit is met, then the contact is in a stick regime, In this case, no movements between the contact bodies can occur. In the region where the stick regime is, the coefficient of friction takes the static value μ s . When the condition Q > Flimit is valid, the contact is in a slip regime. Micro movements may then take place at the interface between the contact bodies, in a region surrounding the stick area. In this slip region the coefficient of friction has the kinetic value μ k . The sizes of those two regions are not fixed, but, for a certain value of clamping force, they vary with the applied shear load. However, in mechanically fastened joints the applied load is generally a normal bulk load, rather than a pure shear load. A correspondence between the shear force Q and a normal bulk load must be evaluated in order to link the application of a bulk load to the generation of a friction force. This problem has been studied in [9], where it is demonstrated that the frictional force, developed at the mating surface, is dependent on an external bulk load through the ratio of the compliances of the contact bodies (Eqn.6):

ΔQ =

C1 ΔPapplied C2

(6)

In case the materials of the contact bodies are the same, the compliances ratio is the unit, and the relationship simply becomes: (7) ΔPapplied = ΔQ A direct proportionality relationship exists between the variations of bulk load and the variations of friction force. This implies that the equations developed for the application of a shear load are valid also to the application of a normal load.

Load transfer in a mechanically fastened joint

995

The difference between stick and slip regions can then be expressed in terms of external applied load. The stick condition is verified when the local value of the clamping force is higher than the applied load. If the local clamping force is lower than the external load, then the slip condition occurs. Tangential pressure distribution By considering the normal pressure distribution and the friction model previously

described, the tangential pressure distribution q ( r ) can be calculated [10]. The tangential pressure distribution can be written as the sum of the stick contribution

q′ ( r ) and the slip contribution q′′ ( r ) [7]:

q ( r ) = q′ ( r ) + q′′ ( r )

(8)

Where: 1 ⎧ ⎛ r2 ⎞ 2 3 CF ⋅ ⎪ q′ ( r ) = μ s ⋅ K ⋅ ⋅ ⎜ 1 − ⎟ ⇒ a′ ≤ r ≤ c 2 ⋅π ⋅ a2 ⎝ a2 ⎠ ⎪ ⎨ 1 ⎪ 3 ⋅ CF ⎛ r 2 ⎞ 2 ⋅ ⎜1 − ⎟ ⇒ c < r ≤ a ⎪ q′ ( r ) = μ d ⋅ K ⋅ 2 ⋅π ⋅ a2 ⎝ a2 ⎠ ⎩ 1 ⎧ ⎛ r2 ⎞ 2 ⋅ 3 CF c ⎛ ⎞ ⋅ ⎜ ⎟ ⋅ ⎜1 − ⎟ ⇒ a′ ≤ r ≤ c ⎪⎪ q′′ ( r ) = − μ s ⋅ K ⋅ 2 ⋅π ⋅ a2 ⎝ a ⎠ ⎝ c2 ⎠ ⎨ ⎪ ⎪⎩ q′′ ( r ) = 0 ⇒ c < r ≤ a

(9)

(10)

Once those contributions are determined, it is possible to calculate the value of the friction force by integrating Eqn. 8 over the correct domain: c

a

a′

c

Q = ∫ q ( r ) ⋅ dA = 2π ⋅ ∫ ( q′ + q′′ ) ⋅ r ⋅ dr + 2π ⋅ ∫ q′ ⋅ r ⋅ dr

(11)

In order to fully define the stick and slip regions, the boundary between the different contact regimes, described by the stick radius c , must be calculated. For a continuous configuration, c can be calculated as a function of the limit friction force and of the applied load by using Eqn. 12:

P ⎛ ⎞ c = a ⋅ ⎜1 − applied ⎟ ⎝ μ s ⋅ CF ⎠

1

3

(12)

Similarly, in a discontinuous configuration, the stick radius c can be determined by solving Eqn. 13:

L. Paletti, C. Rans and R. Benedictus

996 a

Q = 2π ⋅ ∫ q ( r ) ⋅ r ⋅ dr

(13)

a′

Where Q = Papplied and

q ( r ) is given by Eqn. 8-10.

In the hole configuration, the variations of the stick radius are included in the range a′ ≤

c ≤ a . In particular, the borderline cases are determined as follows:

• • •

If no load acts on the joint, the stick area is coincident with the contact area and the stick radius is equal to a . In absence of external load, the stick-slip division would not exist. When the applied load starts increasing, the value of stick radius c decreases, according to Eqn. 12 or 13. If the applied load is less than the limit friction, but the condition

c = a′ is satisfied, then a state of partial slip, named partial reverse •

slip, occurs. If the applied load exceeds the limit friction value, slip occurs everywhere in the contact area and the condition

c = a′ is satisfied.

The value of the friction force calculated by Eqn. 11 will be used to establish a procedure which calculates the percentage of external load effectively transferred by friction, when only friction and bearing are present in a joint. Analytical model for the determination of the load split As described in a previous section, it is reasonable to assume that, in a single-bolt double shear lap joint, only friction and bearing transfer an applied load. Therefore the applied load must be divided between those two contributions as:

F = (1 − α ) ⋅ Papplied B = α ⋅ Papplied

(14)

Where the parameter α represents the fraction of the external applied load transferred by bearing and must satisfy the condition 0 ≤ α ≤ 1 . By considering the friction contribution only, it is immediate to realize that the friction force can be evaluated both from Eqn. 11 and from Eqn. 14. The friction force of Eqn. 11 has been obtained considering the total applied bulk load being transferred by friction, as no bearing load has been considered in the previous analysis. Eqn. 11 must then be rewritten, including the splitting factor dependency, which will affect the determination of the stick radius c (Eqn. 12-13):

F = f ( CF , Papplied , α )

(15)

Load transfer in a mechanically fastened joint

997

By assuming that the geometry considered and the applied bulk load are the same in both cases, then Eqn. 14 and Eqn. 15 must provide the same value for the friction force: FT = FCM (16) The term FT represents the friction force calculated theoretically as a percentage of the applied load (Eqn.14), while the term FCM is the value of friction force calculated using the contact mechanics equations (Eqn.11). The splitting factor can be determined by comparing the two calculated values of the friction force; in order to determine α , the following relationship must be satisfied:

error = Where ε is the desired accuracy.

FCM − FT ⋅100 ≤ ε FT

(17)

In order to determine the correct value of the splitting factor, the following steps must be handled: 1. Fix a value for the clamping force; 2. Out of a range of possible external loads, choose a value for an applied load. 3. Select a tentative splitting factor and determine the bearing and friction loads by using Eqn. 14. 4. Calculate the resultant of the friction stresses by using the equations described in the previous paragraphs. 5. Determine the value of α by satisfying the relationship expressed by Eqn.17. Otherwise choose a different tentative value, and repeat the procedure.

Figure 3: the iterative procedure for the analytical determination of the splitting factor α

L. Paletti, C. Rans and R. Benedictus

998

The iterative approach has been chosen because it was computationally more efficient than the direct calculation of the unknown α . The described procedure can be summarized by a flow chart presented in Fig. 3.

PREDICTIONS The iterative procedure described in the previous section, has been applied to a single bolt, double shear lap joint. The bolt is a 6 mm steel, installed in push-fit configuration. The joint members are in Al 2024-T3. An initial estimate of the static coefficient of friction for the contact between the joint members is found in [6] as μ s 0.65, while the kinetic coefficient is assumed as μ d 0.3. The cross2

section area of the aluminum plates is of 48 mm . The results obtained for this studied case are presented in Fig.4 for a clamping force of 6300 N and an external applied load increasing from 0 to 4800 N. In Fig.4 the variations of the friction load and of the bearing load, as defined in Eqn.14, due to an increasing external applied load are plotted.

By applying a load and gradually increasing it, three regions can be recognized in Fig.4. Those regions can be identified by a certain range of external applied loads:

F≤F.

•

Friction only region, for the values of applied load 0 ≤

•

Friction dominant region, for the values of applied load F < F ≤ FT .

•

Bearing dominant region, for the range of applied load F

> FT .

In the first region, the clamping force CF is able to generate a sufficient amount of friction force to transfer the entire applied load between the members. The bolt transfers no load until a certain value of the applied load

F is reached.

For loads higher than this particular F , the friction force generated is not sufficient to carry the entire applied load. Therefore, the bolt starts carrying part of the load by the bearing mechanism, in a percentage given by Eqn.15. This causes the percentage of load transferred by friction to decrease, still remaining the predominant load transfer mechanism in this region.

Load transfer in a mechanically fastened joint

Friction

999

Bearing

1 P

P P

0,8

Percentage of transferred load

FT

2

0,9

FFFSS

2

0,7 0,6

F

0,5 0,4

d hole = 6 mm

0,3

CF = 6300 N

0,2

Area = 48 mm 2 μ s = 0.65

0,1

μk = 0.3

0 0

400

800

1200 1600 2000 2400 2800 3200 3600 4000 4400 4800 Applie d load (N)

Figure 4: The behavior of the frictional and bearing load contributions for the case CF=6300 Friction is still the predominant load transfer mechanism, but the importance of bearing increases with the increase of the applied load. When a level FT of applied load is reached, bearing and friction shares the applied load equally. For higher values of applied load, bearing becomes the predominant load transfer mechanism. Friction is still transferring part of the load, but the value of the friction force is continuously decreasing until the stick contribution becomes zero, and a full slip contact condition is established. When this value of applied load FFS is reached, the value of the friction force is given by the slip contribution only, whose resultant is represented by: F = μd ⋅ CF (18) Any further increase of the applied load doesn’t cause a further reduction of the value of the friction force; nonetheless the percentage of load transferred by friction diminishes, increasing thus the percentage of bearing load transfer. The loads levels F and

FT appear to be dependent on the preload CF applied to

the bolt. As it can be seen in Fig.5, both values of

F and FT increase as the

clamping force is increased. This is a direct consequence of the direct relationship

L. Paletti, C. Rans and R. Benedictus

1000

between preload (clamping force) and friction force. A higher clamping force can generate a higher friction force which results in transferring the applied load up to a higher percentage. 4000

F 3500

FT

d hole = 6 mm

Transition load (N)

A rea = 48 mm 2 Area 3000

μs = 0.65 μk = 0.3

2500

2000

1500

1000 3000

3500

4000

4500

5000

5500

6000

6500

7000

7500

Clamping force CF (N)

Figure 5: Influence of the clamping force on the transition load levels In Fig.6 an analogous behavior of the transition load levels is presented, caused now by variations of the static coefficient of friction (for example, consequence of different surface treatments). An increase of the static coefficient of friction reflects directly on the friction force; as for the clamping force, a higher generated friction force is able to transfer a higher percentage of the applied load. For this reason, surface treatments on the joint members which reduce the value of the coefficient of friction, also reduce the potential friction force and may promote crack nucleation at the net section. Experimental evidences in [2] have shown that fretting (thus, crack nucleation remote from the hole) does not occur when the coefficient of friction is lowered, for example by using lubricants.

Load transfer in a mechanically fastened joint

1001

4000

F

FT

d hole = 6 mm

3500

Transition load (N)

A rea = 48 mm2 Area

μk = 0.3

3000

CF = 6300 N

2500

2000

1500

1000 0,3

0,35

0,4

0,45

0,5

0,55

0,6

0,65

0,7

0,75

0,8

Static coefficient of friction

Figure 6: Influence of the static coefficient of friction on the transition load levels

VALIDATION In order to validate the proposed analytical model and its predictions, the actual values of the bearing and frictional load must be measured. The direct measure of the frictional load is extremely difficult, particularly because the mating surface, where friction is acting, is not accessible without modifications to the geometry of the joint (e.g. holes or slots for strain gauges and wires). Those modifications will alter the properties of the mating surface itself, consequently affecting the friction distribution, either in terms of contact area or coefficient of friction. In order to keep the frictional properties (and the friction force) undamaged, the preference is given to methods which can measure the bearing contribution. The measure of the bearing load can be realized using an instrumented fastener. In [11], such instrumented fastener has been manufactured by installing four strain gauges in a slot hollowed inside the shank of a 6 mm steel bolt. By placing this instrumented bolt in a bolted joint and applying an external load, the strains have

1002

L. Paletti, C. Rans and R. Benedictus

been measured and plotted in graphs like Fig.7 [11]. The presence of strains in the bolt indicates that the bolt is carrying part of the applied load. The graph of Fig.7 shows the variations in strain due to the application of an external load, for a double shear joint. When the load is applied, the measured strains remain initially zero. The bolt is unloaded and the entire applied load is transferred between the joint members by friction. This situation continues until a certain level of applied load is reached. Beyond this load level, the strains start to gradually increase. The bolt is then carrying part of the load by bearing; friction is not able to sustain the entire load because the application of the external load has modified the contact region, increasing the slip area and decreasing the stick area. The slip area contributes to the friction force with the kinetic friction coefficient, lower than the static coefficient, acting in the stick region. Therefore the friction force decreases with the increase of the external applied load, and after a certain value the friction force becomes lower than the applied load, so the difference must be carried by bearing. Comparison of test data with numerical predictions The situation described in [11] and Fig. 7 was reproduced using the developed analytical model. The behavior depicted in Fig.7 can be compared with the behavior of the splitting factor α presented in Fig.4, determined for the same value of clamping force of about 6300 N. From Fig.4, the load level which marks the transition between a friction-only to a combined friction-bearing load transfer mechanism can be identified as F 1810 N. From Fig.7, it can be seen that the strains in the bolt are zero until the applied load reaches Papplied = 1400 N. This means that the entire applied load is transferred by friction until that particular load level. The load value obtained from the graph is equivalent to the external force at the transition level F determined from the analytical model. The difference between the value presented in [11] and Fig.7 and the value calculated using the analytical procedure is about 20%. Whether the value of the static friction coefficient is decreased to 0.5, the transition level F also decreases. From Fig.6, it can be seen F 1460 N. The difference between the value from Fig.7 and this new calculated value is about 4%.

Load transfer in a mechanically fastened joint

1003

Figure 3: Variations of the strains as function of the applied load for a single-bolt double-lap joint from [11]

CONCLUSION An analytical method for the determination of the amount of load transferred either by friction or by bearing in a single-bolted double-lap joint has been developed. Such joint configuration has been properly selected such that only friction and bearing may be considered active in transferring the load between the joint members. The applied load must divide between these two contributions according to a determined splitting factor α . The behavior of the splitting factor has been determined. Variations of the clamping force and of the static coefficient of friction have the same effect on the splitting factor. Increases of the mentioned parameters correspond to increases of the friction force and, consequently, to increases of the values of the transition loads. The analytical model is able to predict the load levels at which the predominant load transfer mechanism changes. By knowing the prevalent load transfer mechanism, indications on the location where cracks are expected to nucleate may be deduced. In addition to this, appropriate failure models may be selected in order to improve the accuracy of the predictions.

1004

L. Paletti, C. Rans and R. Benedictus

The predictions obtained by the developed analytical model have been validated using test data from [11]. The agreement found between the test data presented in [11] and the predictions presented in this paper is more than satisfactory. The manufacturing of an instrumented bolt is currently on progress and more test data for the validation of the analytical model are expected. The influence of the coefficient of friction will be investigated in more detail. A subsequent step will be the evaluation of the influence of secondary bending on the predictions given by the analytical model.

REFERENCES [1] [2]

[3] [4] [5] [6] [7] [8] [9] [10] [11]

Schijve, J. (2001), Fatigue of structures and materials, Kluwer Academic Publishers, Dordrecht. Dhamari, R.D.A. (2004), The effects of water-displacing corrosion preventatives on the fatigue behavior of mechanically fastened aluminum joints, The University of New South Wales, Sydney. Nowell, D. and Hills, D.A. (1987), Int. J. Mech. Sci., vol. 29, n. 5, p.355. McVeigh, P.A. and Farris, T.N. (1997), Journal of Tribology, vol.119, p.797. Szolwinski, M.P. and Farris, T.N. (1998), Wear, vol.221, p.24. Szolwinski, M.P. (1998), The mechanics and tribology of fretting fatigue with application to riveted lap joints, Purdue University, West Lafayette. Johnson, K.L. (1985), Contact mechanics, Cambridge University Press, Cambridge. Meyer, G. and Strelow, D. (1972), Assembly Engineering, vol. January, p.28. Dominguez, J. (1998), Wear, vol.218, p.43. Hamilton, G.M. and Goodman, L.E. (1966), Journal of Applied mechanics, vol. June, p.371. Palmberg, B. (1991) FFA TN 1991-09, The Aeronautical Research Institute of Sweden, Stockholm.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

SIMULATION OF FATIGUE CRACK GROWTH IN THE HIGH SPEED MACHINED PANEL UNDER THE CONSTANT AMPLITUDE AND SPECTRUM LOADING Petr Augustin Institute of Aerospace Engineering, Faculty of Mechanical Engineering, Brno University of Technology, Czech Republic

Abstract: The paper describes methodology of numerical simulation of fatigue crack growth and its application on integrally stiffened panels made of 2024-T351 aluminium alloy using high speed cutting technique. Presented approach starts by the calculation of stress intensity factor function from finite element results obtained using MSC.Patran/Nastran. Subsequent crack growth analyses were done in NASGRO. For predictions of crack growth under the spectrum loading a sequence representing service loading of the transport airplane wing was prepared. Three crack growth models were applied: non-interaction, Willenborg and Strip Yield. Relatively large experimental program comprising both the constant amplitude and spectrum tests on integral panels and CCT specimens was undertaken in order to acquire crack growth rate data and enable verification of simulations. The work was performed within the scope of the 6th Framework Programme project DaToN.

INTRODUCTION This paper presents results of activities done at the Institute of Aerospace Engineering (IAE) during its participation on the 6th Framework Programme project DaToN - Innovative Fatigue and Damage Tolerance Methods for the Application of New Structural Concepts. Full title of this project indicates that it was focused on the development of damage tolerance assessment tools for M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1005–1018. © Springer Science+Business Media B.V. 2009

Petr Augustin

1006

structures manufactured using newly developed methods. Advanced production techniques under investigation were High Speed Cutting (HSC), Laser Beam Welding and Friction Stir Welding. Their application in the aerospace industry leads to the reduction of manufacturing costs and production time benefits. Structures made by these techniques are typical by their integral character. It leads to the fact that they contain less number of fatigue and corrosion critical locations. However their comparison with differentially stiffened designs from the damage tolerance point of view is less optimistic and in some cases they lack the crackarresting capability that is typical of riveted structures. Example of such a behavior can be, for instance, found in [1]. The methodology of numerical simulation of fatigue crack growth developed by IAE was applied for analysis of 450 mm wide two-stringer integrally stiffened panels made of 2024-T351 aluminium alloy using HSC technique (see Fig. 1). The tests of these panels were carried out under the constant amplitude loading within the common test program of the DaToN project. Moreover, additional testing has been done in the IAE laboratory under the spectrum loading in order to collect the data needed for validation of crack growth models applied by IAE.

METHODOLOGY OF CRACK GROWTH ANALYSIS The approach starts with the calculation of stress intensity factors from results of finite element analysis. Simulation of crack propagation in the panel requires determination of K values for large number of crack configurations and that is why simple FE models built of shell elements were used. It is acceptable time consuming even when there is need of application of geometrical nonlinear solution. Stress intensity factors were calculated from the strain energy release rate determined using the crack closure technique [2]. The method is based on the assumption that the energy released during the extension of the crack from the length a to a+Δa is identical to the energy required to close the crack. The mode I and II components of strain energy release rate can be than derived from this energy as

1 Z1 ⋅ Δw2 2Δa ⋅ b

(1)

1 X 1 ⋅ Δu 2 2 Δa ⋅ b

( 2)

GI = − GII = −

where X1 and Z1 are the shear and opening forces at the crack tip node, Δu2 and Δw2 are the differences in shear and opening nodal displacements at the same

Simulation of FCG in the HSM panel under the CA and spectrum loading

1007

point and b is the width of the elements. This method requires two finite element analyses to calculate the energy. In the first one the forces are determined using the model with the crack closed. In the second step the crack is extended to the length a+Δa which enables determination of displacements. Depending on the mesh density the crack closure technique was able to calculate the stress intensity factor values with the error of about 2%. Crack propagation evaluation has been based on crack growth rate data obtained from the tests of 100 mm wide and 2 mm thick CCT specimens. These specimens were manufactured directly from the two-stringer panels after their testing. The data from the tests performed at stress ratios R of 0.1, 0.5 and 0.8 were subsequently expressed using four different approaches. Two of them were based on the NASGRO equation [3] which is relationship between crack growth rate da/dN and stress intensity factor range ΔK

⎛ ΔK th ⎞ n ⎜1 − ⎟ ⎡⎛ 1 − f ⎞ ⎤ ⎝ da ΔK ⎠ = C ⎢⎜ ⎟ΔK ⎥ q dN ⎣⎝ 1 − R ⎠ ⎦ ⎛ K max ⎞ ⎜⎜1 − ⎟⎟ K C ⎠ ⎝

p

(3)

where ΔKth is the threshold stress intensity factor range and KC the critical stress intensity factor. The crack opening function f = K op K max for plasticity-induced crack closure defined by Newman is a function of stress ratio R and plane stress/strain constraint factor α. Curve fit of crack growth rate data for R = 0.1, 0.5 and 0.8 using the NASGRO equation is shown in Fig. 2. It obvious that the NASGRO equation wasn’t able to closely describe the 2024-T351 data typical by sigmoidal shape with several transitions in slope. Nevertheless this curve fit was used for predictions of crack growth under the spectrum loading using the Willenborg model since there was no alternative in NASGRO. Second approach was applied for simulations of crack growth in the panels under the constant amplitude loading. Thanks to the fact that these calculations start at higher stress intensity factor values, it was possible to use only linear part of the data. Correspondent curve fits based on NASGRO equation are depicted in Fig. 3. The third approach shown in Fig. 4 uses the table lookup input in NASGRO. Polynomial curve fitting was applied to generate the points into the table. This approach was used for predictions under the constant amplitude and noninteraction spectrum loading. The last method is required for calculation of crack propagation using the Strip Yield model. In this model crack growth rates have to be expressed as a function of effective stress intensity factor range ΔKeff. The effective stress intensity factor range was calculated using crack opening function f as

ΔK eff =

1− f ΔK 1− R

( 4)

Petr Augustin

1008

Experimental data for three different stress ratios collapsed into the single band for plane strain constraint factor value of α = 1.8. According to [3] it was supposed that transition from plane strain to plane stress constraint starts at

(ΔK eff )T = 0.5σ 0 B

(5)

where σ0 is the flow stress and B is the material thickness. It corresponds to the crack growth rate of 2.12E-04 mm/cycle. Since the values of ΔKeff above 600 MPa.mm1/2 haven’t been obtained from the tests of CCT specimens, the data in this area were extrapolated. The last two points of da/dN against ΔKeff relationship in Fig. 5 were taken from the work of Yamada at al. [4] analyzing large amount of 2024-T351 data primarily obtained from the NASGRO database. The plain stress constraint factor of 1.2 and the upper limit of transition region at 1.4E-03 mm/cycle were also adopted. A typical feature of integrally stiffened panels is that after the crack in the skin reaches the stiffener we have to consider parallel propagation of two cracks. One of the possibilities how to treat this problem is the use of two-dimensional growth option in NASGRO. Practically it was necessary to calculate stress intensity factors for more or less arbitrary chosen combinations of lengths of cracks in both the skin and stiffener as an input into the NASGRO. Resulting combinations of sizes of skin and stiffener cracks and corresponding K values were than determined during the cycle-by-cycle computational procedure by means of interpolation routine.

TWO-STRINGER PANEL UNDER THE CA LOADING First analyses and verification tests of the two-stringer HSC panels were performed under the constant amplitude loading at maximum nominal stress of 80 MPa with stress ratio R = 0.1 and at maximum nominal stress of 110 MPa with R = 0.5 respectively [5]. The fatigue cracks in the skin started from the central saw cut of 2a = 20 mm. FE model of the panel is shown in Fig. 6. The testing was done without any antibending device able to restrict the displacements perpendicular to the panel plane. From the modelling point of view it is necessary to involve the bending effects acting in the experiments. This is why the application of geometrical nonlinear solution was investigated. Obtained stress intensity factor functions for crack in the skin from both the linear and nonlinear FEM analyses are depicted in Fig. 7. Crack propagation evaluation was based on curve fits of 2024-T351 crack growth rate data shown in Figures 3 and 4. These predictions are in good accordance with

Simulation of FCG in the HSM panel under the CA and spectrum loading

1009

the tests on panels (see Figs. 8 and 9). It is obvious that the best results correspond to stress intensity factor functions obtained from geometrical nonlinear analysis. The reason is in slightly smaller K values for crack approaching the stiffeners. No significant differences in predictions based on curve fits of liner part of data using NASGRO equation and on the table lookup form were found.

TWO-STRINGER PANEL UNDER THE SPECTRUM LOADING For predictions of crack growth and experimental verifications using the spectrum loading a sequence representing service loading of the transport airplane wing was prepared. Following data sources were considered in order to select the flight load spectrum for the tests: database ONERA taken from reference [6], database ACMS [6], measurements on Boeing 737-400 aircraft carried out by FAA within the Airborne Data Monitoring Systems Research Project [7], MD82/83 data related to the same project [8], the FAA dataset collected during the Boeing 767-200ER aircraft operation [9], results of the NASA VGH Program obtained using the Digital Flight Recorder System of Boeing B727 aircraft [10], NASA VGH data of Boeing B747 aircraft [11] and spectrum of the standardized load sequence of transport aircraft wing structures TWIST [12]. Figure 10 provides comparison of flight load spectra considered. All data with exception of the TWIST spectrum are very similar. Substantial differences are in the omission levels and in the truncation of high loads probably related to the extent of the databases. It is obvious that substantial extrapolations would be necessary for many spectra. From this point of view the best alternative seems to be the spectrum of B737 airplane [7] finally selected as a basis for the load sequence. An important question of the development of a load sequence is the definition of the number of flights in the repetitive block defining the magnitude of the highest load included. In accordance with many sources (for instance [12], [13]) the number of flights was established as one tenth of the anticipated lifetime. The B737 design service objective value of 75 000 flights was obtained from [14]. It leads to the repetitive block of 7500 flights. Another decision that has to be made is the omission of small loads. The spectrum of flight loads normalized to 7500 flights was trimmed on the level pertaining to cumulative frequency of 315 000. Provided that the magnitude of 1g nominal flight stress is 70 MPa, the omission stress range is 12.46 MPa. This is a conservative choice in comparison with the recommendations in references [13] and [15]. Continuous flight load spectrum was than divided in ten discrete levels and subsequently distributed into different flight types with various severities. The technique of the definition of flight types in the case of gust dominated spectra of a transport aircraft described in reference [12] was adopted. Consistently with the TWIST load sequence, the number of flight types is the same as the number of load levels. Thus all flight types denoted by letters from A to J can differ in terms of the magnitude of the highest load level, as is shown in Fig. 11. The load history is created on the flight by flight basis and the

Petr Augustin

1010

sequences of loads and flights are random. All flights are closed by insertion of one compression load. Corresponding load factor is the largest load factor selected from number of peaks pertaining to one flight that are randomly selected from the ground load spectrum. Generation of the load history is realized by the in-house computer program. Three crack growth models in NASGRO 4.23 were applied for spectrum loading: non-interaction, Willenborg and Strip Yield model. Simple non-interaction model is fully based on crack growth rates determined from constant amplitude tests hence conservative predictions can be expected. Generalized Willenborg model deals with the crack growth retardation by implementation of the effective stress ratio

Reff =

K min − K R K max − K R

( 6)

that is used instead of the stress ratio R within the crack growth equation. Modified residual stress intensity is defined as 1 ⎛ ⎞ ⎜ OL ⎛ ⎟ Δa ⎞ 2 ⎟⎟ − K max ⎟ K R = φ ⎜ K max ⎜⎜1 − ⎝ Z OL ⎠ ⎜ ⎟ ⎝ ⎠

(7)

where KmaxOL is the maximum stress intensity factor for the overload cycle, Δa stands for the crack growth between the overload cycle and the current cycle and ZOL is the size of the overload plastic zone. The factor Φ is defined as follows

1−

φ=

ΔK th K max − K min ( RSO − 1)

(8)

where RSO is the shut-off value of the stress ratio KmaxOL/Kmax. The second interaction model applied was the Strip Yield model. From two implementations of this model in NASGRO the constant constraint-loss option was selected. The Strip Yield model is based on Dugdale model modified to leave plasticity in the crack wake. It assumes that all plastic deformation is contained within a strip located along the crack line represented by finite-width rigid-perfectly plastic bar elements. They allow to calculate the crack opening stress intensity factor Kop from the contact stresses. Before application on the two-stringer panel, calculations of crack growth under the spectrum loading were performed for simple CCT specimen geometry. Since analytical solution of the stress intensity factor function is known, it was possible to verify the crack growth analyses without the influence of imperfections of FE models. Both the analyses and verification test were performed on 100 mm wide

Simulation of FCG in the HSM panel under the CA and spectrum loading

1011

and 2 mm thick CCT specimens. For all models the range-pair counting technique was applied. Calculated crack growth lives are shown in Fig. 12. The best results were obtained with Willenborg and Strip Yield models that were able to predict the fatigue crack growth life with the error less than 9%. These two interaction models were finally applied also for simulation of crack growth in the two-stringer panel. Since the load sequence includes compression loads, it was decided to perform the testing with the antibending device in order to support the cracked panel in compression. It leads to the reduction of stress intensity factor values for crack in the skin and along with it to the acceleration of crack growth in the stiffener. Obtained predictions of crack propagation in the two-stringer panel under the spectrum loading in comparison with the test carried out at IAE laboratory are presented in Fig. 14. In this case the crack growth lives were predicted with maximum error of about 25%.

CONCLUSION The paper deals with the methodology of numerical simulation of fatigue crack growth and its application on integrally stiffened structures. Results of simulations of crack propagation in two-stringer integrally stiffened panels manufactured using high speed cutting technique were presented. Constant amplitude as well as spectrum loadings were considered. Relatively wide experimental program comprising the tests on panels and CCT specimens was undertaken in order to acquire the crack growth rate data and enable verification of analyses. The work was performed within the scope of the 6th Framework Programme project DaToN - Innovative Fatigue and Damage Tolerance Methods for the Application of New Structural Concepts.

REFERENCES [1] Nesterenko, G.I. (2000), In: Proceedings of the 22nd ICAS Congress, pp. 444.1–444.10, Optimage Ltd., Harrogate. [2] Krueger, R. (2002), The Virtual Crack Closure Technique: History, Approach and Applications, NASA/CR-2002-211628, NASA Langley Research Center, Hampton. [3] NASGRO Reference Manual, version 4.2, NASA Johnson Space Center, Southwest Research Institute, San Antonio. [4] Yamada, Y., Lacy, T., Newman Jr., J., Smith, B.L., Kumar, B. (2007), Int. J. Fatigue, vol. 29, pp. 1503-1509. [5] Lanciotti, A., Lazzeri, L., Polese, C. (2005), Description of the Test Programme, DaToN–WD–WP 3.1–1.0/DIA–UniPi, Pisa. [6] Jonge, J.B. de, Hol, P.A., Gelder, P.A. van (1994), Reanalysis of European Flight Loads Data, DOT/FAA/CT-94/21, FAA, Technical Center, Atlantic City.

Petr Augustin

1012

[7] Rustenberg, J., Skinn, D., Tipps, D.O. (1998), Statistical Loads Data for Boeing 737-400 Aircraft in Commercial Operations, DOT/FAA/AR-98/28, FAA, Office of Aviation Research, Washington. [8] Skinn, D., Tipps, D.O., Rustenberg, J. (1998), Statistical Loads Data for MD-82/83 Aircraft in Commercial Operations, DOT/FAA/AR-98/65, FAA, Office of Aviation Research, Washington. [9] Tipps, D.O., Rustenberg, J., Skinn, D. (2000), Statistical Loads Data for Boeing 767-200ER Aircraft in Commercial Operations, DOT/FAA/AR-00/18, FAA, Office of Aviation Research, Washington. [10] Crabill, N.L. (1989), The NASA Digital VGH Program – Exploration of Methods and Final Results, Volume III – B727 Data 1978-1980: 1765 Hours, NASA CR-181909, Vol. III, NASA Langley Research Center, Hampton. [11] Crabill, N.L. (1989), The NASA Digital VGH Program – Exploration of Methods and Final Results, Volume IV – B747 Data 1978-1980: 1689 Hours, NASA CR-181909, Vol. IV, NASA Langley Research Center, Hampton. [12] Jonge, J.B. de, Schütz, D., Lowak, H., Schijve, J. (1973), A Standardized Load Sequence for Flight Simulation Tests on Transport Aircraft Wing Structures, NLR TR 73029U, NLR, Amsterdam. [13] Lanciotti, A., Lazzeri, L. (1992), Int. J. Fatigue, vol. 14, no. 5, pp. 319-324. [14] Akdenitz, A. (2001), Aircraft Engineering and Aerospace Technology, vol. 73, no. 1, pp. 4-15. [15] Abelkis, P.R (1980), In: Effect of Load Spectrum Variables on Fatigue Crack Initiation and Propagation, ASTM STP 714, pp. 143–169, Bryan, D.F., Potter, J.M. (Eds.), ASTM, San Francisco.

Figure 1. Two-stringer panel at IAE laboratory.

Simulation of FCG in the HSM panel under the CA and spectrum loading

1013

1,0E+00 1,0E-01

da/dN [mm]

1,0E-02 1,0E-03 1,0E-04 1,0E-05 1,0E-06 R = 0.1 R = 0.5 R = 0.8

1,0E-07 1,0E-08 10

100 1000 1/2 ΔK [MPa mm ]

10000

Figure 2. Crack growth rate data for 2024-T351 alloy and curve fit using the NASGRO equation.

1,0E+00 1,0E-01

da/dN [mm]

1,0E-02 1,0E-03 1,0E-04 1,0E-05 R = 0.1 R = 0.5

1,0E-06 1,0E-07 10

100 1000 1/2 ΔK [MPa mm ]

10000

Figure 3. NASGRO equation fitted to the linea part of data.

Petr Augustin

1014 1,0E+00 1,0E-01

da/dN [mm]

1,0E-02 1,0E-03 1,0E-04 1,0E-05 1,0E-06

R = 0.1 R = 0.5 R = 0.8

1,0E-07 1,0E-08 10

100

1000

10000

ΔK [MPa mm ] 1/2

Figure 4. Polynomial curve fit of crack growth rate data.

1,0E+02 1,0E+01 1,0E+00

da/dN [mm]

1,0E-01 1,0E-02 1,0E-03 1,0E-04 1,0E-05 R = 0.1 R = 0.5 R = 0.8

1,0E-06 1,0E-07 1,0E-08 10

100 1000 1/2 ΔKeff [MPa mm ]

10000

Figure 5. Effective stress intensity factor range against crack growth rate.

Simulation of FCG in the HSM panel under the CA and spectrum loading

1015

Figure 6. Stress contour plot for panel with central crack in the skin.

3000 stiff.

K [MPa.mm1/2]

2500 2000 1500 1000 500 nonlinear FEA linear FEA

0 0

20

40 60 80 half crack length a [mm]

100

120

Figure 7. Stress intesity factor function for skin crack in the two-stringer panel.

Petr Augustin

1016 experimental results

simulation for nonlinear FEA

140

simulation for linear FEA

120

a [mm]

100 80 60 40 20 0 0

20000

40000

60000

80000

100000

number of cycles

Figure 8. Example of prediction of crack growth in the skin of the two-stringer panel using NASGRO equation versus result of the test - σmax = 110 MPa, R = 0.5.

experimental results table lookup form

140

NASGRO equation

120

a [mm]

100 80 60 40 20 0 0

20000

40000

60000

80000

100000

number of cycles

Figure 9. Predicted crack growth in the skin of the two-stringer panel versus result of the test (nonlin. FEA) - σmax = 110 MPa, R = 0.5.

Simulation of FCG in the HSM panel under the CA and spectrum loading B737

1,0E+07

ONERA

Exceedance frequency per 1000 flight hours

1,0E+06

ACMS MD82/83

1,0E+05

B767 B727

1,0E+04

B747

1,0E+03

TWIST

1,0E+02 1,0E+01 1,0E+00 1,0E-01 1,0E-02 1,0E-03 1,0E-04 -2

-1

0

1

2

Incremental load factor Δn

Figure 10. Flight load spectra of transport airplanes (TWIST exceedances per 1000 flights).

Figure 11. Examples of typical flights.

Figure 12. Position of the most severe flights in the sequence and magnitudes of largest load factors.

1017

Petr Augustin

1018

experimental data Non-Interaction model, NASGRO equation Non-Interaction model, table lookup form Generalized Willenborg model, NASGRO equation Strip Yield model, table lookup form

50

a [mm]

40

30

20

10

0 0

5000

10000

15000

20000

25000

30000

35000

number of flights

Figure 13. Predicted crack growth in the CCT specimen under the spectrum loading versus result of the test.

experimental data Generalized Willenborg model, NASGRO equation Strip Yield model, table lookup form 140 120

a [mm]

100 80 60 40 20 0 0

10000

20000

30000

40000

50000

number of flights

Figure 14. Predictions of crack growth in the skin of the two-stringer HSC panel under the spectrum loading in comparison with the test.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

BONDED COMPOSITE PATCH TO REPAIR METALLIC STRUCTURES: DISBOND PROPAGATION TESTING AND MODELLING P. Madelpech, S. Juaneda and M. Pradels Centre d’Essais Aéronautiques de Toulouse, Balma - France.

Abstract: Our general objective is to progress in the certification of metallic structure repair process by bonding a composite patch on the damaged part. On military aircraft, some countries already use that repair and its mechanical efficiency is now well established. The composite patch acts as a stress bypass avoiding any supplementary hole drilling and associated stress concentration. Nevertheless the process is not used in France because of a lack of confidence in bonding quality and durability. Besides, a disbond can be detected by classical non destructive techniques. Therefore, disbond propagation can be controlled if a structural inspection program is defined and followed. Thus, damage tolerance requirements may be fulfilled. For this purpose, disbond propagating must be predictable with sufficient confidence. In fact, this paper addresses the problem of predicting the growth of an initiated disbond between the metallic structure and the repair. The modelling of the disbond growth has been performed, based on total energy release rate. The energy release rate related to disbond growth has been determined. Besides, fatigue testing was performed in order to characterize the linear elastic fracture parameters of the adhesive. Using the numerical calculations, the coefficients of the Paris law can be deduced. Finally, it gives a tool to estimate the propagation law of a disbond occurring under a composite patch.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1019–1033. © Springer Science+Business Media B.V. 2009

1020

P.Madelpech, S. Juaneda, M. Pradels

GENERAL CONTEXT Like many others, the French Ministry of Defence is facing the challenge of ageing aircraft in its inventory. Corrosion and fatigue cracking are the main sources of damages in these airframes. Most of the time, a repair to restore the structural integrity is necessary. Most of conventional repair techniques of fatigue damages consist in fastened metallic doublers. They require the drilling of the metallic structure in areas which are already affected by fatigue concerns. New cracks may appear in areas which became consequently difficult to inspect. The concept of a composite patch bonded on the damaged structure has been studied and applied in Australia more than thirty years ago [1]. It offers many advantages such as the possibility to use thin and stiff repairs, its formability that allows low-cost manufacture for fabrication. We can also mention the fact that only one side of the structure needs to be physically accessible and that the temporary use of that repair is possible without irreversible consequences (new hole drilled). However, this repair raises some concerns: the residual stress induced by the differences between metallic and composite thermal dilatation coefficients, the effect of casual in–service temperature on composite and adhesive properties but also the high secondary bending effect. Today the efficiency of the load by-pass and therefore of crack retardation is well demonstrated but the use of this repair technique is limited by the lack of confidence of the bond as single load path for primary and even secondary aeronautical structures. The inability to assess the reliability of a bonding is the main difficulty to overcome in order to let a bonded patch fly. In that context, qualitative and quantitative information on the initiation and the propagation of a disbond between the metal and the composite patch will be helpful for the decisions of the authorities on repair technique validation. Our test centre is performing tests to characterise the fatigue behaviour of the adhesive between a composite patch and a metallic substrate. The idea is to increase our experience but also to be able to transpose to other situation. Basically, the bond is controllable by NDI. The questions are: is a disbond which would be detected dangerous for the structural integrity? Can a periodical inspection of the bond be defined in order to ensure that a sudden unsafe debonding will not occur? A possible approach is to study the debonding in the same manner as a crack with the linear fracture mechanics theory. It can be done based on the common modelling means for composite patch repair [2]. In [3], the authors have concluded that the governing parameter was the global energy release rate.

Bonded composite patch: Disbond propagation testing and modelling

1021

The present study carries out the modelling necessary to determine the energy release rate linked to a given increase of the disbond size. These computation means correlated with test results will enable to determine a threshold value for da n disbond initiation and then a propagation law similar to Paris ones = C (ΔG ) dN for the crack propagation. This will be the first step to fulfil a damage tolerance approach for patch debonding.

GEOMETRY OF THE COUPONS General geometry of the patch The repair consists in a four or eight elliptical Carbon/Epoxy T300/914 plies with various layup. A Glass/Epoxy 644/914 ply is placed between the metallic plate and the C/E patch as this would have been done on an actual aircraft structure to protect against galvanic corrosion. The patch (C/E + G/E plies) is initially co-cured in an autoclave. A Redux 312 adhesive layer between the composite patch and the metallic substrate is polymerised at 120°C. It can be done in an autoclave for in-lab study or using heat blanket on aircraft. Figure 1 illustrates a possible design of the repair.

C/E 0° C/E 90° C/E 90° C/E 0°

G/E ply: to prevent from Galvanic Corrosion

Metallic Substrate

FM73312 Redux

Figure 1: Design of the bonded composite patch Large shear and peel stress peaks usually occur near the edge of a typical patch. To is due to the sudden change of rigidity of the structure. To reduce the severity of the peak, a uniform stepping is used with a step length of 5 mm per lamina ply (see Figure 2). It corresponds to a 3° tapered angle. Thus, the load transfer is smoother.

P.Madelpech, S. Juaneda, M. Pradels

1022 ¾

TaperedTapered-edges: edges: to reduce Peel Stress

angle < 3° chosen

Stepping < 3°

σpeel with square-cut edges

σpeel with tapered edges

Figure 2: stepping to reduce the peel stress Testing coupons Those coupons (see Figure 4) are cut in 1.6 mm thick sheets of aluminium 2024T351. The design in “wasp waist” enables to introduce high loads without breaking the specimen in the load introduction areas. This configuration is representative of a repaired aeronautical structure. Under the patch, the metal is undamaged to avoid any interactions which would be difficult to analyse. The residual stress under the patch is due to the curing temperature of 120 °C and to the difference of thermal dilatation coefficients between the aluminium and the composite. The secondary bending effect is due to the offset of the neutral axis (see Figure 3). This last one is mechanically fixed when the underlying metallic structure is stiffened. Concerning the specimen, the absence of bending can be simulated by applying symmetrically two composite patches.

Figure 3: Residual thermal stress and secondary bending effect The standard patch geometry is one-sided, elliptical shaped with stepping. The first coupons used those kinds of patch (Figure 4a). A bond defect was introduced by replacing locally the adhesive with a circular piece of protective film (white circle on the patch). For those types of patch, no disbond propagation occurs for 10 000 000 load cycles at 175 MPa (constant loading ratio R = 0.1) in the central area. Increasing the load level only caused the specimen break in the metal part (Figure 4b).

Bonded composite patch: Disbond propagation testing and modelling

1023

Therefore the patch definition has been changed. No difference was made by removing the stepping, or double-siding the patch. After that, a static test of traction was performed. A brutal debonding occurs at 405 MPa in the central zone. Almost half of the patch did disbond, but it is far beyond the yield stress of the aluminium (about 340 MPa) and this test result is therefore useless for the modelling. So we decide to change the geometry of the patch to make the configuration more critical. Besides, the disbond growth is controlled periodically by ultrasonic non destructive inspection. This can be done without removing the coupon from the test facilities. The potential disbond growth is measured with a precision of about 1mm. Unfortunately, it was not possible to investigate under the stepping, because of parasitic reflexion of the US signal. It was an additional reason to abandon the stepping.

Figure 4: representative coupons (a, b and c)

MODELLING The chosen approach is to adopt a strategy similar to the linear fracture mechanics theory. More precisely, the disbond propagation will be correlated to the energy release rate variation ΔG as introduced in Irwin theory of linear fracture mechanics. All the computations have been performed using Samcef. Objective and constraints

P.Madelpech, S. Juaneda, M. Pradels

1024

The objective of the modelling is to be able to compute the energy release rates corresponding to observed disbond propagation during tests. For further application, it is important to enable the propagation modelling. All the models will therefore be parameterized to be compatible with an evolving geometry. The adhesive is 0.125 mm thick, which is relatively small in comparison with the specimen width*length (200*600 mm). To have precisely the stress in the adhesive and to avoid a too refined mesh, interface elements have been chosen. As they are only suitable with volume elements, the composite is modelled by orthotropic material. The elements are quadrangles meshed by extrusion. Interface elements in Samcef consist in duplicating the nodes of the considered surface and introducing three unidirectional springs to model the peeling action σzz and two out-of-plane shears σxz and σyz. The debonded area is obtained by first creating the interface elements and then removing them so as to obtain two coincident unconnected nodes. It is suitable as long as the patch do not penetrate the metal during the loading. It is the case for double sided patched coupons. For the other ones, the bond is under compression due to secondary bending. The computations are incoherent but in the same times no debonding occurs during the tests. For further application, it will be necessary to introduce contact elements at this location. The coupons do have thermal residual stress. This stress level is comparable to the one applied mechanically and therefore cannot be neglected. Different computation strategies have been intended. As the material non linearity of the adhesive was not possible to investigate, it was chosen to remain linear in our computation. Geometry, hypothesis and boundary conditions As mentioned above, all the computations are made assuming linear behaviour. The material characteristics are described in the following table 1. Aluminium 2024 T 351 Ex (MPa)

72000

Composite Carbone/epoxy ply 141300

Composite Verre/epoxy ply

Adhesive Redux 312

40000

1800

Ey (MPa)

72000

9300

8000

1800

Ez (MPa)

72000

9300

8000

1800

5500

9000

G (MPa)

ν

0,33

0,3

0,26

0,33

αx (C° -1)

2,3.10-5

-4,5.10-7

3,5.10-5

2,3.10-5

αy (C° -1)

2,3.10-5

2,8.10-5

2,8.10-5

2,3.10-5

αz (C° -1)

2,3.10-5

2,8.10-5

3,5.10-5

2,3.10-5

Table 1: used material properties

Bonded composite patch: Disbond propagation testing and modelling

1025

For symmetry reasons, only an eighth of the coupons is modelled (see Figure 5) when the specimen is patched on both sides and one quarter when only one side is patched. To improve the visibility of the adhesive for post-processing, the composite has been offset so that the disbond as well as the results in the bonded joint can be observed. The disbond front has been slightly refined. The coupons in the model have been loaded with imposed displacement although the tests have been performed with applied loads; In fact, to compute the energy release rate, both approaches are equivalent.

Symmetry

Imposed Imposed displacemen displacement

Figure 5: mesh of representative coupons Stress result Stress results have been used to analyse and understand the test results, especially the premature failure of the coupons in the aluminium part.

P.Madelpech, S. Juaneda, M. Pradels

1026

Figure 6: Von Mises stress for thermo mechanical loading. The stress concentrations on Figure 6 in the metal are at the beginning of the composite patch and also at the extremity of the ‘wasp waist’ part. We can understand the observed failure on Figure 4. The design of the coupons has not enabled to avoid interaction between the patch end and the radius of the metal part. Critical area

Figure 7: peel stress and shear in the adhesive (lozenge shaped patch). The stress results have also been used to control that the stress level in the adhesive has not exceed the yield stress of the Redux 312 (37 MPa). On Figure 7, we can also notice that the maximum peel stress is rather concentrated in the corner of the remaining adhesive. During tests, it has been observed that this area was the main zone for debonding occurrence.

ENERGY RELEASE RATE COMPUTATION Computation method The strain energy release rate is obtained by an energy balance of the global system [4]. With an increase of the debonded area of dS, the creation of a new crack results in an irreversible dissipation of energy. Thus the energy release rate G is given by:

Bonded composite patch: Disbond propagation testing and modelling

G=−

d (U − W ) dS

1027

(E1)

(E1) U is the stored elastic energy and W the external work. Considering the process as isothermal and assuming the linearity we can transform the expression of U by using integral transformation, it becomes:

G=−

⎞⎞ d ⎛⎜ 1 ⎛ ⎜ ∫ (T .ξ d − T d .ξ )da ⎟ ⎟ ⎟⎟ dS ⎜⎝ 2 ⎜⎝ Σ ⎠⎠

(E2)

(E2) ξd and Td being respectively the imposed displacement and load on the external surface Σ of the structure. According to the previous equation, to compute the energy release rate, only the stress and strain at the model border is needed. Indeed, it concerns the reactions at the nodes where displacements are applied, and the displacements of the nodes where the load is applied. To perform the calculation of G, an initial load case is considered corresponding to the state of the coupon at a valley of the experimental spectrum. This load case, cooling from 120°C to 20°C with a low mechanical load level, is taken as the initial state (I). Then two other load cases are computed. The first one (F1) corresponds to the combination of thermal and maximum mechanical loads for a given disbond. The second one (F2) has an increased disbond area and the same thermo-mechanical load. (U – W) is computed between I and F1 and then between the same I and F2 to give two values of energy P1 and P2. It corresponds to the two following possibilities. Either the disbond remains stable or it grows during the load application. G is simply the ratio of the difference between these energies by the increase of the disbond surface. This method enables easily to take into account the initial thermal residual state and to compare both potential energy levels associated to F1 and F2. The sources of errors are multiple. Firstly finite elements method principle is an approximateion of the stress and the strain fields based on the mesh. Thus, changing the mesh will have an impact on the results for computational reasons. In our case, the disbond growth forced to modify the mesh, which will disrupt the results. In this study, it has been verified that refining the mesh has only a slight influence on the results. Secondly, for tiny disbond growth, the difference of energy can be comparable to the incertitude on results. To minimize this problem, the disbond size perturbation has been fitted at 0.25 mm. Finally, from a theoretical point of view, a part of the dissipated energy is stored in plastic deformation at the crack front. In our case, it is acceptable to neglect it since there is no real stress

P.Madelpech, S. Juaneda, M. Pradels

1028

field singularity at the crack tip. Nevertheless, plasticity cannot yet be taken into account for materials compatible with interface elements on Samcef. Energy release rate results For the representative coupons it appears on Figure 9 that the energy release rate increases with the load. More surprisingly the computed energy release, for these coupons, can be negative for some cases. It is due to the thermal residual stress. The specimen being not mechanically loaded, the adhesive is under shear due to the thermal retraction of Aluminium. When the metal is under traction, the shear is progressively reduced, cancelled and re-increased (see Figure 8).

Figure 8 : scheme of the mechanical loading of the adhesive 450 400

energy release rate (J/m²)

350 300 250 5 mm 15 mm 25 mm

200 150 100 50 0 -50

0

50

100

150

200

250

300

350

-100 Applied load (MPa)

Figure 9: G vs. applied load for different disbond length. Otherwise, the energy release rate is also decreasing with the increase of the disbond length (cf. Figure 10). It is due to the fact that the considered disbond is a quarter circle. The critical part is in the sharp edge as seen before and assuming it remains circular is not completely coherent with what happens in the reality. The sharp edge debonds first and the shape of the disbond becomes more or less flat. Thus the transfer zone is larger and the stress concentration in the adhesive is reduced.

Bonded composite patch: Disbond propagation testing and modelling

1029

On Figure 10, small irregularities (for disbond length equal to 15 or 25 mm) can be noticed. They are due to changes in the mesh of the coupons to keep the quality of the elements correct. They are of reasonnable magnitude.

energy release rate (J/m²)

120 100 80 60 40 20 0 0

5

10

15

20

25

30

35

40

45

disbond length

Figure 10: G vs. disbond radius relationship at 225 MPa loading. Contrary to a crack in metallic structures, a disbond does not become more critical when it grows. It is only true as far as it is possible to neglect the interactions betwen different disbonds. It is the case when the typical length of load transfer is smaller than the size of the remaining adhesive. As seen on Figure 7, it is a reasonable hypothesis.

TEST RESULTS AND DISBOND PROPAGATION LAW To estimate the disbond propagation law, we will put all our test results on a dA/dN vs. ΔG diagram (disbond area increase by cycle vs. energy release rate). Ten coupons were tested in fatigue. Over them, only five gave disbond propagation. Among those ones only two gave propagation rates. For the others ones, the evolutions were either too small or just during the ultime cycles so as it was impossible to reject the hypothesis that it appeared during the coupons break. On the following Figure 11, we have summed up the results:

P.Madelpech, S. Juaneda, M. Pradels

1030

log (disbond area per load cycle)

0 2,2

2,22

2,24

2,26

2,28

2,3

2,32

-0,5 -1 -1,5

coupon 5 coupon 6

-2 -2,5 -3 -3,5 log(variation of energy release rate G)

Figure 11: disbond area by loading cycle vs. energy release rate These results are not coherent with what could be expected. Several reasons could explain these observations: • The computation of energy release rate, especially because it is based on an assumption of circular propagation. • The measurement of the disbond, which is performed by ultrasonic non destructive inspection. The precision of the measures is about 1mm only, which is quite poor. • The geometry of the glue joint at the border of the patch, which disrupts the propagation. Consequently a special attention has been paid during the test of coupon 6 especially in measurement. The results are much better for this coupon and a propagation law has been deduced (Figure 11). The results are the following:

dA 5.26 = 1.5.10 −15.(ΔG ) dN

(E3)

(E3) A is the disbond area in mm², N the number of load cycles and ΔG the variation of energy release rate during a load cycle. This law is only based on one coupon. For more accurate values, a new test campaign is about to be launched, taking into account the experience feedback of the first one. In that purpose, the key point will be to use a more efficient disbond measurement process. The geometry has also to be redesigned to avoid early break in the Aluminium.

Bonded composite patch: Disbond propagation testing and modelling

1031

MODELLING THE PROPAGATION Once the disbond propagation law is determined, it is possible to model the propagation. The global objective is to determine the necessary number of load cycle to make the disbond reach a critical size. It is an essential tool to estimate the consequences of an appearing disbond under the composite patch. It could be detrimental for the efficiency of the repair and therefore for the airworthiness. Such a tool has to be applied on realistic conditions. Patches used in service are elliptical with stepping, which is quite different from the tested configuration. Using the previously determined propagation law seems premature regarding the multitude of error sources and the poorness of the experimental database. Consequently, the following section only aims at proposing a strategy without numerical values. Propagation strategy The damage tolerance approach imposes to assume the existence of an initial disbond. Assuming the disbond remains circular, it is possible to simply model the propagation. In that purpose, we have to impose the propagation path in two stages: • First the disbond is growing with a shape of a circle tangential to the previous one. It consists in moving the centre of the circle in the same way as the radius increases (Figure 12). • Then as soon as the centre of the disbond reaches the border of the patch, the disbond propagation enlarges the circle with the same centre (Figure 12).

Tangential evolution

Transition Stage 1

Stage 2

Figure 12: assumed disbond propagation path first and second stage

1032

P.Madelpech, S. Juaneda, M. Pradels

This assumed propagation path seems wise and coherent with the first disbonds observed, in stress concentration areas (Figure 13). Unfortunately, there is no experimental validation for the present time.

Stress concentration area

Figure 13: Von Mises stress in the adhesive. The mesh model is presented Figure 14. Once again the composite patch has been offset for better display. The propagation computations have been performed on circular patch of 4 plies at 0° with stepping. The initial disbond can be observed.

disbond Figure 14: circular patch model Along the propagation path, different steps are defined. For each one, the FE model enables to compute the energy release rate using the same process as previously. The determined propagation law gives allows calculating the number of load cycles need to go from one step to the next one. Eventually, this method leads to the total number of cycles necessary to let the disbond grows from its initial shape to a target size.

CONCLUSIONS AND PERSPECTIVES The objective was to test and model the propagation of a disbond under a composite patch used to repair a metallic structure. It can enable to predict the number of load cycles the structure is able to sustain without dramatic consequences. In that purpose, a test campaign was performed to assess the propagation rate.

Bonded composite patch: Disbond propagation testing and modelling

1033

A simple modelling, using interface elements and a method of computing the energy release rate was defined. It enables to compute the energy release rates correlated to the observed propagation. From that, it was possible to estimate a propagation law similar to Paris one for the propagation of cracks in a metallic structure. Unfortunately, the tests results were not complete enough neither to give a precise and reliable law nor to validate the approach. Therefore, once an additional test program completed, the model will be validated or recalibrated. If the results are satisfying, more model capabilities will be implemented. A first test with elliptical shapes and partial energy release rates has already been performed. Other tests will be led with downgraded bond process, with bad surface preparation to characterize the deterioration of the bond durability. In fact, knockdown factors are necessary to take into account all the environmental conditions which can alter the adhesive properties. It will result in a tool to anticipate the possible disbond of an in-service bonded composite patch. Nevertheless the difficulty to propagate a disbond without applying a stress level which would not be regularly seen by aircraft structures and the stability of the rare observed growth rates are globally reassuring for the use of this repair technique. This could finally to gain in confidence on the reliability of a bonded patch.

REFERENCES [1]

Backer A, Rose F, Jones R (eds.). Advances in the bonded composite repair of metallic aircraft structure, Elsevier, Vol. 1, 2002

[2]

Begue JF, Crassus P, Juaneda S. Bonded composite patch to repair fatigue crack testing and modelling, CEAT, 2008

[3]

Johnson, W.S. and Mall, S., A Fracture Mechanics Approach for Designing Adhesively Bonded Joints, ASTM STP 876, 1985, pp 189199.

[4]

Suquet P, Rupture et Plasticité, Majeure de Mécanique, Ecole Polytechnique, Edition 2004

25th ICAF Symposium – Rotterdam, 27–29 May 2009

THREE DIMINSIONAL CRACK GROWTH PREDICTION Sarah E. Galyon, Saravanan R. Arunachalam, James Greer, Matthew Hammond and Scott A. Fawaz Center for Aircraft Structural Life Extension, United States Air Force Academy

Abstract: To properly model complex aircraft structure an understanding of mixed mode, also known as three dimensional, crack growth is required. The Center for Aircraft Structural Life Extension (CAStLE) at the United States Air Force (USAF) Academy is in the process of completing a project to generate three dimensional (3D) crack growth data for thin sheet (1.6 mm) aluminum alloy (AA) 2024-T351 and predict the measured crack growth rate, crack trajectory and residual strength using state-ofthe-art stress analysis and life prediction tools. To produce the mixed mode crack trajectory and residual strength mechanical testing data ARCAN specimens were used with a modified ARCAN fixture. The ARCAN test fixture allows the ARCAN sample to be rotated to produce different mixtures of mode I and mode II loading (with 0° being pure tension/compression (mode I) and 90° pure shear (mode II)). ARCAN specimens were tested at angles of 0°, 30°, 60° and 75° under constant Keq starting loads followed by a Kincreasing test and a stress ratio (R) of 0.1. A travelling optical microscope was used to track crack trajectory and crack growth. While mechanical testing was being completed, we also developed a crack prediction model of the ARCAN specimen using FRANC3D NG, a solid modeling, mesh generation and fracture mechanics code from Cornell University Fracture Group. FRANC3D NG should be able to predict the cracking behavior observed in the ARCAN tests. A parallel effort was also undertaken to develop an engineering model of mixed mode crack growth where contributions to mode I and mode II growth were M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1035–1068. © US Government 2009. Created within the capacity of an US governmental employment and therefore public domain. Published by Springer Science+Business Media B.V. Dordrecht.

Sarah E. Galyon et al.

1036

accounted for using KI and KII and the appropriate baseline crack growth data. Based on our results some of the mixed mode data can be modeled using the engineering model. The FRANC3D NG model is currently limited by restraints of the analysis tools itself and other factors. Mixed mode crack growth produces data with large amounts of scatter, therefore more data is needed to draw firm conclusions about how mixed mode cracks propagate under fatigue loading.

INTRODUCTION In complex aircraft structure, crack growth rarely propagates in the idealized fashion assumed in durability and damage tolerance analyses (DADTA). Usually the applied loading is not perpendicular to the crack nucleating feature and subsequent crack propagation. This situation is known as mixed mode crack growth or in more general terms, three dimensional (3D) crack growth. Most DADTA’s conducted assume mode I (tension/compression loading) only; thus, engineering judgment is used to estimate the amount of error present in the idealized models. A better understanding of mixed mode fatigue crack growth is needed to design better crack prediction models. Little work has been published in the area of mixed mode fatigue crack growth, hindering the development of newer and more accurate DADTA. The Center for Aircraft Structural Life Extension (CAStLE) at the United States Air Force (USAF) Academy is in the process of completing a project to generate three dimensional crack growth data and predict the measured crack growth rate, crack trajectory and residual strength using stateof-the-art stress analysis and life prediction tools. The ability of current prediction methods to model the complex 3D crack growth generated by mechanical testing will also be evaluated.

TESTING Testing Background The Center for Aircraft Structural Life Extension at the USAF Academy is in the process of completing a program to generate three dimensional (3D) fatigue crack growth data for thin sheet aluminium alloy (AA) 2024-T351 at two material orientations, longitudinal transverse (LT) and transverse longitudinal (TL), and then to attempt to model the results using state-of-the-art stress analysis and life prediction tools. The program also studied the effects of mixed mode loading on the residual strength of AA2024-T351 in the LT and TL material orientation.

Three dimensional crack growth prediction

1037

ARCAN Fixture and specimen There are a few methods that can be used to produce mixed mode (mode I/mode II) loading in a uniaxial servo-hydraulic test frame including 4-point-bending, single edge notch specimens under torsional loading, and multiple test fixtures that rotate to apply a combination of mode I/mode II loading to a specific sample type [1-5]. CAStLE selected the ARCAN test fixture with the ARCAN test specimen to complete the mixed mode mechanical testing as the sample has a comparatively large gauge section for a mixed mode sample with which to study crack trajectory and growth [1-5]. Most of the specimen and loading fixtures mentioned above have been used in stable tearing testing (monotonic loading) of crack growth rather than fatigue crack growth (cyclical loading). Figure 1 shows a diagram of the ARCAN fixture used by CAStLE for testing. It should be noted that a modified ARCAN fixture, which is pinned at three points on each end to prevent rotation during testing, rather than the original ARCAN design which has the fixture pinned into the test frame at only a single point on each end, was used for all mechanical testing in this program [5,6]. CAStLE’s ARCAN fixtures and pin and clevis sets were made from high nickel, medium carbon 4140 steel. The pins were hardened after machining and the rest of the fixture is in the annealed state. The ARCAN test fixture, as originally designed, can only hold samples up to 3.175 mm (0.125 inch) thick, thus it is currently used for thin sheet testing [5,6]. The fixture allows an ARCAN sample to be rotated at 15° increments from 0° to 90° to produce different amounts of mode-mixity during loading; with 0° being pure tension/compression loading (mode I) to 90° being pure shear loading (mode II).

Figure 1: Mechanical drawing of the ARCAN fixture used by CAStLE for all mixed mode testing. All dimensions are in millimeters (mm).

1038

Sarah E. Galyon et al.

The ARCAN specimen, also called a “butterfly specimen”, is used in the ARCAN fixture to complete mixed mode (mode I/mode II) crack growth tests. Figure 2 is a diagram of the ARCAN test specimen used for all of the mixed mode tests. The AA 2024-T351 samples tested had a thickness of 1.6 mm (0.063 inches). The 45° notch on the ARCAN specimen was electric discharge machined (EDMed) into the specimens to ensure a sharp notch for fatigue crack initiation. To align the ARCAN samples in the center of the servo-hydraulic test frame actuator, shims 1.01 mm thick were used during testing.

Figure 2: Mechanical drawing of ARCAN specimen used for all mixed mode testing, all dimensions are in millimeters. Initial Load Determination Very little work has been completed or published in the area of mixed mode fatigue crack growth, particularly using the ARCAN specimen and fixture. The ARCAN system has mainly been used in the evaluation of mixed mode loading on stable tearing [5,6]. It was discovered through finite element models of the ARCAN specimen that controlling the stress levels of the samples during mechanical testing was impossible with the ARCAN specimen; as there is no cross-section through which the stresses are uniform for off-axis loading (see Figure 3). Therefore it was not possible to define a consistent initial stress value to use for all tests. Figure 3, below, shows the finite element (FE) stress contours

Three dimensional crack growth prediction

1039

across the loading axis of the sample. As can be noted, the stress field changes dramatically as the angle of loading is adjusted from 0°, 30°, 60° to 75°.

Figure 3: Finite element stress contours (x, or horizontal) for the ARCAN specimen under four displacement conditions. Instead, each test was started with approximately the same equivalent stress intensity (Keq),:

K eq = K I2 + K II2

(1)

with pre-cracking to be completed at a lower Keq. The equivalent stress intensity takes into account the stress intensity from the mode I portion of loading (KI) and the stress intensity from the mode II portion of loading (KII). The value of Keq selected for pre-cracking would need to be far away enough from the long crack threshold stress intensity factor, Kth of the da/dN vs. ∆K curve that cracking will occur and a stress intensity value low enough that stable fatigue crack growth should occur without tearing during testing at a higher Keq. After starting each test at the same Keq the test continued as an increasing K test.

1040

Sarah E. Galyon et al.

Calculation of Stress Intensities In the finite element analyses, the Virtual Crack Closure Technique (VCCT) was used to derive the stress intensity at the crack tip for the various loading angles [7]. Table I shows some typical results of the analyses for a particular pre-crack length.

Table I: Typical VCCT stress intensity factors at pre-crack tip for five loading angles. The “Applied ftot” column is the resultant magnitude of the constraint forces given an arbitrary 0.254 mm (0.01 in) applied displacement. The KI/ftot and KII/ftot columns are the important results for these analyses, as they give the test machine operator the information needed to determine the load setting required to achieve a certain SIF for a given specimen mounting angle. For these tests, it was decided an equivalent stress intensity would be calculated for the pre-cracked condition, and the loading amplitude would be set such that the starting Keq=7.59 MPa√m (6.9 ksi√in) for each sample. Carpet plots were generated for each loading angle to assist in setting the cyclic loading amplitude. The 75° plot is shown in Figure 4.

Figure 4: Carpet plot for initial load selection for 75° test sample.

Three dimensional crack growth prediction

1041

The figure example indicates a load of approximately 3047 N (685 lbs) is required for a peak Keq of 7.59 MPa√in (6.9 ksi√in) given a pre-crack length of 2.997 mm (0.118 inches) [8]. These loads were based on the FE results using the VCCT technique normalized by the total load calculated from nodal constraint forces at the holes. This normalization assumes the resultant load in the specimen is essentially along the load frame actuator axis. It was decided later in testing that different load values should be used to try and determine if different initial Keq caused variation in the fatigue crack trajectory. Mixed Mode Crack Growth Trajectory Testing One of the project goals was to determine the effect of mode mixity on fatigue crack trajectory. To evaluate different combinations of mode I/mode II loading on fatigue crack growth, four angles were selected to be tested, 0° (pure mode I), 30°, 60° and 75°. During unpublished pilot tests it was noted that cracks would not grow at the 90° orientation for the K levels used in the 3D crack program. Three tests were performed at each angle at each material orientation, longitudinal transverse (LT) or transverse longitudinal (TL). As testing progressed the third set of LT and TL 0° samples were changed to 15° tests. Table II shows the test matrix for all mixed mode crack trajectory testing. Each sample was pre-cracked at 0° with a maximum load (Pmax) of 2669 N (600 lbs), with a stress ratio (R) of 0.1, at a frequency (υ) of 30 Hertz (Hz) The samples were cycled until a pre-crack length of 2.997 mm (0.118 inches) was reached, per ASTM 647 [8]. The front and back crack length was monitored visually with a travelling microscope, with a digital readout with micron resolution.

Table II: Test matrix for crack growth trajectory testing; 24 samples total.

Sarah E. Galyon et al.

1042

Once the proper pre-crack length was achieved, the sample was rotated to the appropriate testing angle and cycled at the maximum load values listed in Table III with R of 0.1, υ equal to 30 Hz. Prior to the start of testing the pre-crack length was measured again, this time at the testing angle, and recorded. The EDM notch tip was the zero point for all crack length measurements. The front and back crack length was checked every 2,500 to 10,000 cycles depending on crack growth rate. A crack length measurement was not recorded until the crack had grown 0.508 mm (0.020 inches) [8]. The fatigue crack was allowed to grow until sample failure by unstable crack growth.

Angle Pmax Set 1 Pmax Set 2 Pmax Set 3

0° 2688 N (600 lbs) 2688 N (600 lbs) ‐

15° ‐ ‐ 2624 N (590 lbs)

30° 2313 N (520 lbs) 2758 N (620 lbs) 2647 N (595 lbs)

60° 2046 N (460 lbs) 2402 N (540 lbs) 2758 N (620 lbs)

75° 2202 N (495 lbs) 2202 N (495 lbs) 3047 N (685 lbs)

Table III: Table of maximum loads used for each set of crack trajectory specimens. Once testing was complete a photograph was taken of the smaller sample portion and placed on a grid using Corel Paint Shop Pro®. The points recorded during testing were used to plot the crack trajectory in coordinates that could then be plotted with Microsoft Excel®. Mixed Mode Crack Growth Residual Strength Testing Another goal of this testing program was to determine the effect of mode mixity on residual strength. The ARCAN fixture and specimen were again used for this part of the program. The angles tested during the residual strength portion of the program were the same as those used in the crack trajectory portion: 0°, 30°, 60° and 75°, with 15° replacing the 0° degree sample in the third set of tests completed. Both material orientations (LT and TL) were tested as well. Table IV show the text matrix for all residual strength testing. All of the residual strength samples were pre-cracked in a servo-hydraulic test frame in the same manner as the crack trajectory specimens expect that the pre-crack lengths for the residual strength specimens were 6.35 mm (0.250 inches) for two sets of tests and 7.62 mm (0.300 inches) for the triplicate test set. The longer pre-crack length was selected based on previous CAStLE work on stable tearing with the ARCAN fixture and specimen. It was noted that if the pre-crack is not long enough then sudden failure can occur before stable tearing takes place [9].

Three dimensional crack growth prediction

1043

Table IV:Test matrix for mixed mode residual strength testing; 24 samples total. Once the pre-crack had reached the proper length the residual strength samples were rotated to the proper testing angle and strained in a servo hydraulic test frame under a constant displacement rate of 0.0254 mm/min (0.001 in/min) [9]. The front and back crack growth was measured using a traveling optical microscope with a digital measurement display. Crack length measurements were taken after at least 0.508 mm (0.02 inch) of crack extension was observed. At each crack length measurement the force on the sample and the current axial displacement were recorded as well. After testing was complete the crack trajectory was plotted using the same method as was used with the crack trajectory specimens. The load versus displacement curve was plotted for each of the residual strength samples tested. Due to time constraints, the residual strength portion was not finished by the time of reporting for this paper, however the work completed to date is reported below.

MECHANICAL TESTING RESULTS Crack growth trajectory testing results Given the large number of mechanical tests completed, only selected representative results are presented below. On average the rotated samples had a shorter fatigue life than the 0° samples. Figure 5 shows the crack trajectory for LT samples fatigued at each angle tested. All of the fatigue test trajectories shown in Figure 5 were started at approximately 7.59 MPa√m and then completed as a K-increasing test. For the LT tests the 60° samples follow a path close to the 30° samples path. Note the triangle and diamond marked crack paths. The 0°, 15° and 75° crack paths are more distinctive.

1044

Sarah E. Galyon et al.

Figure 5: LT crack trajectories for ARCAN samples fatigued at 0°, 15°, 30°, 60° and 75° with a starting Keq value of 7.59 MPa√m (6.9 ksi√in) followed by a K-increasing test.

Figure 6 shows representative TL crack growth trajectory data for 0°, 15°, 30°, 60° and 75°. The shown TL crack trajectories were produced by starting the fatigue tests at approximately Keq equal to 7.59 MPa√m and completing a K-increasing test. On average there is less of a change in the fatigue life of the samples based on mode-mixity for the TL samples compared to the LT trajectory samples. In the TL tests the 60 degree sample trajectory was closer to the 75 degree sample trajectory than the 30 degree sample trajectory. The 0, 15 and 30 degree trajectories are also much closer than the LT trajectories.

Three dimensional crack growth prediction

1045

Figure 6: Representative TL fatigue crack growth trajectory data produced by Kincreasing testing with a starting Keq value of 7.59 MPa√m.

For a better understanding of the data variation at each angle tested a breakdown of the trajectory data by testing angle is presented below. Figure 7 plots the trajectories for all of the 0° samples tested. The 0° samples are all very constistant in their trajectories with no noticable anomolies. The LT samples deviate from the 0° crack path slightly while the TL samples do not. This deviation is most likely caused by the presence of the grain boundaries impeding crack propogation for the LT samples. The average LT fatigue life (372,183 cycles) is very close to the average TL fatigue life (336,422 cycles), this is an unexpected result as it is widely known that material orientation does effect fatigue life for aluminum alloys. Also the LT and TL da/dN curves produced by CAStLE suggest that there should be a differnece between the LT and TL ARCAN fatigue tests at 0°. The TL 0° fatigue samples had a large spread in cycles to failure with one sample taking just over 100,000 cycles and the other exceeding 500,000 cycles prior to failure. It is possible that with only two TL and LT fatigue tests completed an accurate average fatigue life cannot be determined, and that with more testing a difference would be

1046

Sarah E. Galyon et al.

noted between the LT and TL 0° samples. The variation in the cycles to failure for the samples, especially the TL samples suggests, that more testing is needed to fully understand the fatigue life of the ARCAN sample. The samples tested at 30°, shown in Figure 8, have a fairly consistent fatigue crack path with the exception of sample IISA66TL. The reason for the drastic difference in crack path for sample IISA66TL is unknown, none of the other samples removed from the larger panel around that location experienced anomolies in crack path or fatigue life . The sample IISA66TL did have a longer pre-crack than the other samples tested at 30°; it is possible that the longer fatigue crack changed the starting Keq value sufficently to affect the crack trajectory of the sample. The average fatigue life for the LT 30° samples is 319,103 cycles, with the TL average fatigue life being 373,600 cycles. Again the TL samples have a wider spread on cycles to failure, with one sample taking over 800,000 cycles to fail compared to approximately 90,000 cycles and 170,000 cycles. In contrast, the LT samples’ fatigue lives were all within a factor of 2 of the average.

Figure 7: Zero degree crack trajectories for LT and TL ARCAN fatigue test samples.

Three dimensional crack growth prediction

1047

Figure 8: Thirty degree LT and TL crack growth trajectories from ARCAN specimen fatigue testing.

The 60 degree samples, shown in Figure 9, have the widest range of crack trajectories compared to any of the other tested trajectories. Despite the spread in the crack trajectories, the cycles to failure are similar between the LT (181,846 cycles) and TL (137,909 cycles) samples. Also, neither the LT or TL samples have a very wide range of cycles to failure; all of the specimens were well within a factor of two.

1048

Sarah E. Galyon et al.

Figure 9: Sixty degree LT and TL crack trajectories for ARCAN specimen fatigue testing.

The 75 degree samples, shown in Figure 10, have similar crack paths for the LT and TL samples with the exception of sample IISA27LT in which the crack propogated in a path more similar to a 0° sample than a 75° sample. The reason for this unusual crack path is unknown. Again, none of the other ARCAN samples removed from the large panel at near-by locations experienced anomalous results. The sample also did not have an unusually long pre-crack or other factor that can be easily identified as the cause of the unexpected crack path. The cycles to failure for the LT samples is 140,442 cycles, and the fatigue life for the TL samples is 263,750 cycles. Again the TL samples have a larger difference in cycles to failure (57,500 cycles to 300,000 cycles) compared to the LT tests. The 75° samples have the largest difference between the LT and TL average cycles to failure of all the angles tested. It is possible that the larger amount of shear loading on the TL grain orientation makes crack propogation more difficult than the LT orientation. The addition of the predominantly mode II loading does appear to have a determental effect on the average fatigue life of the samples, as both the 60°

Three dimensional crack growth prediction

1049

samples and 75° degree samples have cycles to failure of almost half that of the 0° samples. The mixed mode loading does not appear to affect the 30° specimens’ fatigue life.

Figure 10: Crack growth trajectories for the ARCAN fatigue tests at 75°.

Based on the above results the effect of material orientation on mixed mode crack propagation appears to be limited as far as fatigue cycles to failure for AA2024T351. Only the 75° samples have a greater average fatigue life in the TL material orientation compared to the LT orientation for the samples tested. It does appear that the TL orientation has a greater variation with respect to life cycles. However, these variations do not have a notable effect on the average fatigue crack life. When over 50% of the loading is in shear (mode II) the fatigue life of the samples does decrease compared to 100% mode I loading. In the follow-on program analysis of the crack trajectories in terms of crack growth rates is planned. Based on the variations seen in the crack growth trajectory testing more testing is needed to fully understand how the application of mixed mode loading affects the crack growth mechanisms.

1050

Sarah E. Galyon et al.

Residual strength testing results Another area of interest with mixed mode testing is the effect of mode-mixity on the residual strength of an alloy. Because of the unusual shape of the ARCAN specimen, some discussion is still ongoing regarding how to best choose or calculate cross-sectional area for the purposes of stating a “residual strength” value. Therefore, all residual strength results will be presented in terms of maximum load achieved for each test. Testing is also not complete for this part of the program, however the results for the work completed are presented below. Figure 11 shows the load versus displacement curves for LT and TL ARCAN specimens tested at constant displacement to determine the maximum load achieved. The peak load achieved for both the LT and TL samples is higher for the pure mode I loading (0°) than the samples tested in the mixed mode I/mode II loading conditions. The larger the amount of shear loading, the lower the maximum load achieved. There is a 4% difference between the LT and TL peak load for the 0 degree specimen. The samples loaded at 30 degrees have almost the same peak load value. The 60 degree LT sample has a 4.5% higher peak load than the TL sample. The 75 degree samples again have almost the same peak load achieved. The loading of the microstructure determines how the residual strength is affected by the mixed mode loading.

Figure 11: Load vs. displacement for LT and TL ARCAN specimens. The LT 0° sample, pure mode I loading, produces the highest peak load.

Three dimensional crack growth prediction

1051

MODELING WITH FRAN3D NG Cornell Fracture Group’s (CFG) software, FRANC3D-NG (Next Generation) [10] was used to model two cases (30° and 60° loading) of the ARCAN specimen fatigue crack growth experiments. The modeling effort was an iterative one, with near-real-time assistance being provided by CFG in troubleshooting and improving the software. Multiple software versions were provided by CFG during the course of the modeling effort. The Model The finite element model of the ARCAN specimen was developed using MSC/Patran with the ABAQUS analysis preference. In contrast to the model used in the previously mentioned VCCT analyses, this model used a global/local modeling approach to reduce computational overhead and, more importantly, mesh complexity. By using a local model to capture the cracked area of the specimen, only that local portion had to be remeshed each time a crack growth increment was modeled, see Figure 12. The FRANC3D-NG software preserved the interface nodes, the nodes common to the local and global models, during remeshing. After remeshing, the local model was reinserted into the global model, and the full model was solved using ABAQUS.

Figure 12. The global and local models prior to remeshing of the submodel with FRANC3D-NG.

1052

Sarah E. Galyon et al.

The global model and original local model were modeled using 20-noded hexahedral elements. These were used to provide a compatible element boundary interface to the 10-noded tetrahedral elements used by FRANC3D-NG in the submodel. A typical FRANC3D-NG submodel is shown in Figure 13. Note that the locations of nodes along the boundary of the submodel are preserved to facilitate reinsertion into the global model at each crack growth step.

Figure 13. ARCAN cracked submodel after remeshing by FRANC3D-NG. Node locations along the boundary are preserved for re-insertion into the global model.

Limitations of the Model

At the time these analyses were performed, FRANC3D-NG did not have a nonlinear material capability. Therefore, all material models were linear-elastic. Furthermore, while the software does allow the user to supply orthotropic toughness properties, orthotropic crack growth rate criteria cannot be input into the model—only a single set of rate data (da/dN vs. ∆K) may be used during any one analysis. For these analyses, LT data for thin sheet (1.60 mm thick) AA2024-T351 data were used. These data were generated by CAStLE using the same material from which the ARCAN specimens were fabricated.

Three dimensional crack growth prediction

1053

Modeling Parameters The FRANC3D-NG software has multiple criteria that may be selected for both crack growth direction and crack growth increment. Only one of each was exercised in this effort due to the unforeseen level of effort (i.e., iterating with the code developer) required to complete the analyses. For crack growth direction, the selection “Maximum Stress Criterion (Tensile or Shear Fracture)” was used. This uses the equation:(2)

θ kink = θ ∀ MAX (σ θθ θ

max'

⎛ K Ic ⎜⎜ ⎝ K IIc

2

⎞ ⎛ K ⎟⎟ [σ rθ (θ )]2 + ⎜⎜ Ic ⎠ ⎝ K IIIc

2

⎞ ⎟⎟ [σ θ z (θ )]2 ⎠

)

(2)

for determining the crack growth direction, which evaluates the stresses in the vicinity of the crack tip. The crack is grown in the direction of the larger of the maximum hoop stress and the maximum shear stresses (in an RMS sense). The coefficients (KIc/KIIc) and (KIc/KIIIc) were chosen to be 1.667 and 0 respectively. The (KIc/KIIIc) coefficient was zeroed because, in this thin specimen, it is likely that the effect of through thickness shear is small, and there was no reasonable way to estimate this coefficient. The (KIc/KIIc) coefficient was estimated by using the VCCT method with a Patran/Nastran finite element model of the ARCAN specimen loaded at 90° and some data from AA7079 mixed-mode stable tearing tests[9]. It is not claimed that this coefficient is accurate for AA2024-T351, but estimating it did give the FRANC3D-NG code the opportunity to exercise this portion of the algorithm. The use of this estimation is one potential reason the predicted crack paths differ from the experimental paths. This and other possible sources of error are discussed later herein. For crack growth length increment, FRANC3D-NG allows for the use of usersupplied da/dN vs. ∆K data to determine the crack growth increment for each node, ∆anode i. In this case, the computed stress intensity factor is used, along with a userspecified number of fatigue cycles, to determine the crack growth increment.

(

Δa node i = Δ N f user ΔK node i

)

(3)

where fuser is the user-supplied da/dN vs. ∆K data. For the analyses herein, a LT da/dN vs. ∆K curve produced by CAStLE, per ASTM E647-00 standards, using 15.24 cm (6 inch) wide M(T) specimens were used. If the crack growth increment was too large or too small, modeling issues arose. It was found through trial and error that a satisfactory increment was a number of cycles that yielded between 0.254 and 0.381 mm of crack growth. At the initially low values of ∆K (very early in the simulation), this represented a large number of cycles (as many as 10,000–20,000 cycles for the 30° case). As crack growth

1054

Sarah E. Galyon et al.

proceeded, the number of cycles per increment became shorter; as few as 500 cycles per increment in the latter stages of crack growth. Although the crack growth rate theoretically changes cycle by cycle, and the current algorithm has the effect of averaging the rate over the prescribed number of cycles, the error introduced by this averaging is small due to the small number of cycles used for each crack growth increment. FRANC3D-NG Modeling Results The two cases run were (1) an ARCAN specimen loaded at 30° off from pure mode I with a constant amplitude load of Ppeak = 2647 N, R = 0.1 and (2) an ARCAN specimen loaded at 60° from pure mode I with a constant amplitude load of Ppeak = 2758 N, R = 0.1. The results were compared to those of experiments using similar parameters. In the case of the 30° specimen (see Figure 14), the FRANC3D-NG predicted crack path is shallower than the experimental crack paths and the 30° “waterline” (this line is perpendicular to the loading axis and is shown for reference). The shallowing of the predicted path may be due to load skewness, discussed later. Note that the actual experimental crack paths in this case are steeper than the waterline.

Figure 14: Photo of part of post-test 30° Arcan specimen indicating three experimental crack paths (blue, green, and black) with number of cycles to the end of the crack and the FRANC3D-NG predicted path (red). The yellow reference line indicates 30° from horizontal. The photo orientation matches that of the specimen as mounted in the test frame. The color inset image is the FRANC3D-NG result as depicted in ABAQUS.

Three dimensional crack growth prediction

1055

The 60° case showed similar trends in the FRANC3D-NG model’s predicted crack path: it was shallower than both the experimental paths and the waterline perpendicular to the machine axis (see Figure 15). Again, it is postulated that the shallowing of the predicted path is due to load skewness, described below.

Figure 15: Photo of part of post-test 60° Arcan specimen indicating two experimental crack paths (blue and green) with number of cycles and the FRANC3D-NG predicted path (red). The yellow reference line indicates 60° from horizontal. The photo orientation matches that of the specimen as mounted in the test frame. The color inset image is FRANC3D-NG result as depicted in ABAQUS.

Load Skewness and Results Analysis of the FE model reaction loads at the specimen holes indicates that the resultant load in the specimen is not aligned with the machine load axis, but rather is skewed a few degrees off. Figure 16 shows a free body diagram for the 60° ARCAN, but the effect is also present (though to a lesser degree) in the 30° case. It

1056

Sarah E. Galyon et al.

was also found that the amount of skewness could be on the order of 30° and could vary by as much as 5° over the range of crack lengths analyzed.

Figure 16. Free-body diagram of 60° ARCAN specimen indicating internal forces and moments reacting the hole loads. The angle α indicates the degree of skewness of the resultant load. Units of loading are pounds force; moment is inch-pounds.

The load skewness in the model was corroborated by an analysis of the principal stress directions at the center of the specimen. In the 60° ARCAN, the principal stresses were seen to be approximately 27° off of the loading axis (specimen with pre-crack only) as shown in Figure 17.

Three dimensional crack growth prediction

1057

Figure 17. Difference between loading axis (bold red line) and principal stress directions at specimen center in 60° Arcan specimen model (Nastran). This load skewness may explain why the FRANC3D-NG predicted paths follow the lines seen in the results: if hoop stress (σөө) dominates in the vicinity of the crack tip (see Equation 2), FRANC3D-NG will grow the crack perpendicular to that direction, which is not aligned with the machine axis. Of course, this begs the question of whether the model is correctly representing the actual test specimen, as the experimental crack paths differ from that predicted by the model. Furthermore, they differ in different ways in the two cases: in the 30° case, the crack paths are steeper than the waterline, while in the 60° case, they are shallower. While increasing the model fidelity is the subject of follow-on work, the following are offered now for consideration as to the differences in modeling and mechanical testing crack growth path and rate results: (1) The model assumes a perfectly rigid fixture with no “play” at pins and joints—only the flexibility of the specimen is included. Some FE results

1058

Sarah E. Galyon et al.

indicate that compliance in the fixture (allowing rotation, in particular) reduces the degree of skewness, but increases the in-plane moment. (2) The model assumes isotropic properties, so the predicted crack paths and rates are not affected by any anisotropy present in the actual specimen (which is known to be orthotropic). (3) The model assumes linear elastic material behavior, while the specimen obviously has local plastic behavior near the crack tip. (4) Regarding Equation (2) and the angle of crack growth: further material characterization is needed to accurately determine the correct mode II critical K value. As was mentioned, the assumed value was derived from testing of AA7079 thin sheet. It could be that a higher value of KIc/KIIc is appropriate for this material, which would possibly turn the crack as in plane shear stress begins to dominate. (5) It could also be that none of FRANC3D’s built-in criteria (there are five builtin propagation direction criteria with five additional options for crack tip closure; and there are six built-in crack extension amount criteria) are best for AA2023-T351. Other researchers have developed criteria not included in the above selections (crack opening displacement[6,11] which depends on stress constraint [12] is one possibility). Implementation of new crack growth criteria, some of which would require modeling of plasticity effects, is a lucrative field for future research. Section 10.9.2 of Suresh’s book [13] summarizes some theoretical methods by which the direction of mixed mode crack growth can be predicted, but the complexity of the load path through the ARCAN fixture and specimen make accurate predictions using these methods problematic. Further characterization of the skewed loading phenomenon is the subject of follow-on work, in which the ARCAN fixture will be modeled as a flexible structure. This modeling work, coupled with new and better crack growth criteria, may provide a way forward for more accurate crack path predictions.

ENGINEERING MODEL CRACK GROWTH PREDICTIONS Prediction set-up and logic The final area of interest in the three dimensional crack growth program was to determine if the crack growth rate from mechanical testing could be predicted using other more simple crack growth prediction tools (i.e., tools that do not require an extensive modeling effort of the part of the user). A CAStLE developed crack growth prediction algorithm was used for the three dimensional crack growth predictions. In the interest of time remaining on the project, the algorithm was used to model two 0° tests (one LT and one TL) and two 60° tests (one LT and one TL). The prediction allows for the user to input da/dN vs. ∆K data for use during modeling. For all predictions the CAStLE generated da/dN vs. ∆K curves using the ARCAN specimen AA2024-T351 material were input into an algorithm developed

Three dimensional crack growth prediction

1059

by CAStLE. All da/dN vs. ∆K data were produced per ASTM E647-00 [8]. The 0° LT and TL predictions were developed using the da/dN vs. ∆K curves from testing. For the 0° LT and TL predictions the respective LT or TL da/dN vs. ∆K curve could be input into the algorithm for use with the prediction. However, for the 60° predictions the mode-mixity of the sample loading must be taken into account. Three methods were developed for examining the mode-mixity with the prediction algorithm. The methods are detailed below. Prediction 1The first prediction uses KI and KII (local) tables as a function of crack length, as well as angle of crack trajectory as a function of crack length. The (dK1, da/dN1) material data is the LT da/dN vs.ΔK curve produced by CAStLE as described previously in this paper, and the, (dK2, da/dN2) material data is the TL da/dN vs. ΔK curve produced by CAStLE. The KI and KII local values are treated as vectors, and their components resolved into the 0-degree (mode-I) and 90degree (mode-II) directions, respectively. The KI-global and the (dK1, da/dN1) material data are used to resolve a growth increment. Likewise with KII-global and the (dK2, da/dN2) material data. These two growth increments are added together for a total growth increment per the one cycle as shown by Equation 4:

(4) Checks are performed to stop the simulation if Kmax is achieved in either direction and if the crack length exceeds the maximum length in the crack data table. The crack data should be monotonically increasing, whereas the angle, KI, and KII tables as a function of crack length do not need be. Figure 18 shows the two vectors as described for this prediction.

Figure 18: Diagram of the vectors used in the AFGROW 3 prediction.

1060

Sarah E. Galyon et al.

Prediction 2 The second prediction uses a KI (local) table as a function of crack length, as well as angle of crack trajectory as a function of crack length. The (dK1, da/dN1) material data is the LT da/dN vs.ΔK curve produced by CAStLE as described previously in this paper, and the, (dK2, da/dN2) material data is the TL da/dN vs. ΔK curve produced by CAStLE. The dKI (local) value is used to obtain a crack growth increment from the (dK1, da/dN1) material data set, as well as from the (dK2, da/dN2) material data set. A linear interpolation is then performed between the two crack growth rates with the testing loading angle as the variable shown in Equation 5:

(5)

(For example, at 45-degrees: the dadN_interp= (1/2) [(dK1, da/dN1) @ dKI + (dK2, da/dN2) @ dKI]). Figure 19 shows an explanation of the vectors used in this prediction. Checks are performed to stop the simulation if Kmax is achieved and if the crack length exceeds the maximum length in the crack data table. The crack data should be monotonically increasing; whereas, the loading angle and KI tables as a function of crack length do not need to be.

Three dimensional crack growth prediction

1061

Figure 19: Diagram of the vectors used in the Prediction 2 method. Prediction 3 The third prediction method uses a KI, KII (local) table as a function of crack length, as well as angle of crack trajectory as a function of crack length. The (dK1, da/dN1) material data is the LT da/dN vs.ΔK curve produced by CAStLE as described previously in this paper, and the, (dK2, da/dN2) material data is the TL da/dN vs. ΔK curve produced by CAStLE. The local dKI and dKII values are used to form a dKeff. The dKeff value is used to obtain a crack growth increment from the (dK1, da/dN1) material data set as well as the (dK2, da/dN2) material data set. A linear interpolation is then performed between the two crack growth rates with the angle as the variable as shown with Equation 6:

(6)

Figure 20 shows a diagram of the vectors used in the prediction. (For example, at 45-dgrees: the da/dN_interp= (1/2)[(dK1, da/dN1)@dKeff + (dK2, da/dN2)@dKeff].) Note, Figure 19 shows the location of θ relative to da/dN1 and da/dN2. Checks are performed to stop the simulation if Kmax is achieved and if the crack length exceeds maximum length in the crack data table. The crack data should be monotonically increasing; whereas, the angle and KI tables as a function of crack length do not need to be.

Sarah E. Galyon et al.

1062

Figure 20: Diagram of the vectors used in Prediction 3 method.

Zero Degree AFGROW prediction results Figure 21 shows the predicted 0° LT crack length per cycle curve (circle markers) compared to the crack growth test being predicted (square markers). The other LT 0° test is also shown (triangle markers). The prediction is well within the factor of two spread for the average recorded cycles to failure for both tests, and is within 80,000 cycles of the cycles to failure of the sample being modeled (325,000 cycles predicted compared to 406,737 cycles recorded), overall a good prediction as far as cycles to failure. Another comparison completed between the test data and the predictions was to determine the percentage difference specific data points along each curve. The average percentage difference between the 0 LT prediction and the IISA24LT samples was 26% with the lowest deviation being 19% and the highest being 39%. The predicted final crack length was 2.29 cm (0.9 inches) compared to 2.68 cm (1.06 inches) for IISA24LT sample (a 0.39 cm difference), also a good prediction. Overall the 0° LT prediction was acceptable.

Three dimensional crack growth prediction

1063

Figure 21: This plot shows the 0° LT prediction (325,000 cycles to failure) for the 0° LT samples IISA24LT (406,737 cycles to failure) along with the other 0° LT samples tested (IISA14LT).

The 0° TL AFGROW prediction is shown in Figure 22. The prediction matches the average crack growth rate fairly well. The actual sample (IISA54TL) had a much higher fatigue life (543,175 cycles) than the prediction (225,000 cycles). Also the 0° TL prediction on avarege varied from IISA54TL by 66%, with the least deviation being 58% and the highest being 75%, again, making the prediction less than ideal. However, the 0° TL samples had very different fatigue lives (543,175 cycles compared to 129,669 cycles). The prediction tracks very closely with the average of the two crack growth samples, suggesting that additional testing might have yielded results more comparable to the AFGROW prediction.

1064

Sarah E. Galyon et al.

Figure 22: Plot of the 0° TL prediction for the 0° ARCAN sample IISA54TL. The plot also show the second 0° TL ARCAN test completed.

Sixty Degree AFGROW Predictions For the 60° predictions the results were more diverse. For the LT sample prediction (IISA64LT) all of the predictions are within the acceptable range, with Prediction 3 being the closest to the original data. Prediction 3 under-predicted IISA64LT by 32% on average, with the closest point deviating by about 29% and the largest deviation being 35%. Prediction 1under -predicts the sample by about 67% and Prediction 2 by about 35% on average. Figure 23 shows the 60° LT predictions with the 60° LT testing data. All of the predictions over estimate sample IISA64LT, the mechanical test being modeled, however the other two 60° LT ARCAN tests had longer fatigue lives, so it is possible that sample IISA64LT’s cycles to failure were lower than would typically be seen. With the 60° LT samples’ range of cycles to failure, it is difficult to accurately predict a specific case well. None of the predictions come close to predicting the final crack length. All of the predictions have a final crack length of approximately 2.29 cm (0.9 inches) with the final crack length of sample IISA64LT was 3.34 cm (1.31 inches), over a 30% difference.

Three dimensional crack growth prediction

1065

The 60° TL predictions, shown in Figure 24, are a reasonable approximation of the cycles to failure of sample IISA48TL. Prediction 1 is the closest to the modeled sample (IISA48TL) in cycles to failure. The fatigue life from the Prediction 1 is approximately 120,000 cycles and the actual mechanical testing sample, IISA48TL, failed at approximately 171,000 cycles. Prediction 1 over-predicts the sample by about 30% on average, with the lowest deviation being 22% and the highest being 37 %. Predictions 2 (53 % deviation) and 3 (54% deviation) predictions failed at approximately 90,000 cycles also making them possibly acceptable predictions for sample IISA48TL, however Prediction 1 is substantially better. Compared to the other 60° TL ARCAN tests the predictions are still within the acceptable range. Both of the other 60° TL samples have higher cycles to failure than the sample predicted, suggesting that with more testing Predictions 2 and 3 might be more acceptable. For predicting the final crack length, again all of the AFGROW predictions were shorter (2.29 cm) than the actual test being modeled (3.07 cm), a 25% difference in final crack length.

Figure 23: Plot of 60° LT AFGROW predictions with 60° LT crack growth data.

Sarah E. Galyon et al.

1066

Figure 24: Plot of 60° TL predictions, 60° TL mechanical fatigue test data.

CONCLUSIONS Overall, three dimensional crack growth is a very complex process to test and model. Based on the results of this program it is apparent that mixed mode fatigue crack growth produces a wide variation of fatigue crack lives and crack trajectories. From the fatigue crack growth testing it is apparent that when AA2024-T351 samples are fatigued with over 50% mode II loading, in mode I/II loading, the fatigue life is reduced in both the LT and TL material orientation compared to the pure mode I samples. In residual strength (monotonically increasing displacement) testing, the peak load achieved with mixed mode loading (30°, 60° and 75°) is lower in both the LT and TL AA2024-T351 material orientation than the peak load achieved by the pure mode I tests (0°). Overall the variation in the results on the crack trajectory data suggest that a great deal more testing and analysis is required to fully understand the mechanisms behind mixed mode fatigue crack growth and to make determinations about crack path, fatigue life and crack growth rate. The FRANC3D-NG modeling produced crack trajectories that were not the same as the mechanical data— the 30° model crack trajectory was steeper than the test

Three dimensional crack growth prediction

1067

data and the 60° was shallower. However, it is hypothesized that a number of modeling factors (outlined previously), including load ‘skewness’, could be the cause of these differences. The input and limitations of FRANC3D-NG also need to be examined as the cause of the differences in the modeled crack trajectories and the crack trajectories from mechanical testing. The follow-on program should be able to provide some insight into whether or not these concerns are the cause of the mismatch between the FRANC3D-NG model and the test data. The engineering predictions were able to model the test data with reasonable accuracy except for the 0° TL tests. All of the predictions (Predictions 1, 2 and 3) were able to model the test data suggesting that the different methods for handling the mode mixity can all be considered acceptable for three of the four cases predicted (0° LT, 60° LT and 60° TL) with regard to fatigue life. Further predictions of a larger range of data are needed to make a determination of which method of predicting mode-mixity is best. Again the follow-on program should provide more time and data with which to compare the predictions. None of the 60° degree predictions were able to properly predict the final crack length of the 60° LT and TL samples. The predictions should be refined with the use of more data and other techniques.

ACKNOWLEGEMENTS The authors gratefully acknowledge the sponsorship of Ms. Monica Poelking and Dr. Reji John of the Metals, Ceramics, and Nondestructive Evaluation Division, Materials and Manufacturing Directorate, Air Force Research Laboratory for this work. The support of Dr. Bruce Carter and Dr. Paul "Wash" Wawrzynek of Fracture Analysis Consultants, Inc. was indispensable for the FRANC3D-NG analyses. As well as, Mr. Benjamin Royall and Mr. Ian Pryce of CAStLE for all of the mechanical drawing, machining and other work contributed to the project.

REFERENCES [1]

F.-G. Buchholz, A. Chergui, and H. Richard (2004), Eng. Fract. Mech., vol. 71, pp. 455-468.

[2]

L. Banks-Sills, M. Arcan, and Y. Bortman (1984), Eng. Fract. Mech., vol. 20, no.1, pp.145-157.

[3]

A. Basnal and M. Kumosa (1995), Jorn. Of Comp. Mat., vol. 29, no. 3, pp. 334-358.

1068

Sarah E. Galyon et al.

[4]

F.-G. Buchholz, V. Just, and H. Richard (2005), Fatigue Fract. Eng. Mater. Struct., vol. 28, pp. 127-134.

[5]

B. Amstutz, M. Sutton, D. Dawicke, and J. Newman, Jr. (1995) In: Fracture Mechanics: 26th Volume, ASTM STP 1256 pp.255-271, W. GReuter, J. H. Underwood, and J. C. Newman, Jr. (Eds.), Philadelphia: American Society for Testing and Materials.

[6]

M.A. Sutton, X. Deng, F. Ma, J.C. Newman, Jr. and M. James (2000), Internat. Jour. Of Solids and Struct., vol. 37, pp. 3591-3618.

[7]

R. Krueger, (2002), Contractor Report NASA/CR-2002-211628 "The Virtual Crack Closure Technique: History, Approach and Applications," NASA Langley Research Center, Hampton, VA.

[8]

ASTM Standard E647 (2000) Standard Test Method for Meausurment of Fatigue Crack Growth Rates, ASTM International, West Conshohocken, PA.

[9]

S.R. Shah, M.R. Walters, M.J. Hammond, B.A. Royall, and S.A. Fawaz (2008), Technical Report United States Air Force Academy USAFA TR 2008-07 “The Fracture Toughness and Mixed Mode Failure of 7079-T6 Thin Sheet Aluminium,” United States Air Force Academy, Colorado Springs, CO.

[10]

Wawrzynek, P. and Carter, B., (2007) FRANC3D/NG Reference Manual, Cornell Fracture Group (http://cfg.cornell.edu).

[11]

M.A. Sutton, M.L. Boone, F. Ma, and J.D. Helm (2000), Eng. Fract. Mech., vol. 66, pp.171-185.

[12]

J. Zou, X. Deng, M.A. Sutton, and C.-S. Cheng (2004), “Crack Tunneling: Effect of Stress Constraint,” ASME International Mechanical Engineering Congress, Paper IMECE2004-60700.

[13]

Suresh, Subra, Fatigue of Material, Cambridge University Press, Cambridge, 1998.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

HYBRID STRUCTURES FOR CONCENTRATED LOAD TRANSMISSION IN FIBRE COMPOSITES: INITIAL EXPERIMENTS Pascal Vermeer1,2, René Alderliesten1 and Rinze Benedictus1 1

Faculty of Aerospace Engineering, Delft University of Technology, Netherlands 2 Metallic Technologies & Surface Engineering, EADS Innovation Works, Germany

Abstract: This study can be considered as an element of advanced hybrid structural concepts in composite aircraft design, supporting the development of hybrid titanium composite laminates (HTCL) and its transition region for local–reinforced / fastening–optimised CFRP structures. Focus is on similarities with existing FML, model adaptation and experimental validation. Investigations on fatigue crack initiation and propagation in HTCL show promising intermediate results; Classical laminate theory (CLT) is a suitable method for residual stress calculation as initial experiments show a good correlation with the CLT results. A crack bridging mechanism seems to occur, although the delamination area is not resolvable with the used technique of Xray computer tomography. The visualisation of delamination and in-situ techniques for measurement of crack lengths are special points of attention.

INTRODUCTION The current generation aircraft in development consist of +50wt% (carbon) fibre composites as these materials couple a low density with extremely high directional strength and stiffness. Additionally, fibrous composites perfectly allow in-plane M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1069–1081. © Springer Science+Business Media B.V. 2009

1070

P. Vermeer, R. C. Alderliesten and R. Benedictus

properties to be freely specified; generally by variation of fibre type, orientation and lay-up, so also a quasi-isotropic state can be obtained. However, a concentrated transmission of load strikes on inherent violations of the composite’s basic design; structural coupling by conventional mechanical fastening provokes the composite’s notch sensitivity, its low shear and bearing strengths and the critical dependence on the laminate lay-up-. Although increasing the (quasi-isotropic) joint thickness would make the design uncritical to bearing failure, this is in conflict with the aspiration to structural efficiency. In addition unfavourable effects and increasing complexity are often induced. As fibre metal laminates (FML) are known to be suitable for mechanical fastening [1], adopting the concept of FML seems to be an elegant solution for optimising the joining area of carbon fibre reinforced polymer (CFRP) structures. Hence, a local (inverted) hybrid titanium composite laminate (HTCL) is created by insertion of thin titanium foils in the CFRP laminate, replacing those fibre layers whose orientation is contributing least to load transfer [2] (figure 1).

Figure 1: Schematic lay-up of a hybrid laminate with its corresponding areas HTCL (2×Ti / 3×CFRP) with hole, Transition zone and CFRP (0°/90°)

This requires investigation of the transition zone from the CFRP laminate into the hybrid area and the behaviour of the HTCL under fatigue loading. The latter has a correlation to previous studies on FML for fuselage applications, and it is likely that the different constituents in HTCL affect the laminate’s mechanical behaviour. The current paper discusses initial experiments on the fatigue crack initiation and propagation in HTCL. These investigations are part of a larger research project funded by the German Federal Ministry of Economics and Technology called: development of new materials for fuselage applications (German acronym ENWERUM). The overall scientific objective of this research is to analyse, characterise and model the mechanical behaviour of the HTCL, the joining area and the transition region.

Hybrid titanium composite laminates

1071

NOMENCLATURE α C da/d N E G Kt ΔK(t

coefficients of thermal expansion Paris equation coefficient crack growth rate

Ni Nf R

fatigue initiation life cycles fatigue life to failure cycles stress ratio

modulus of elasticity [MPa] shear modulus [MPa] stress concentration factor (threshold) stress intensity range

σp,Ti,f σ S Sp,Ti,f

stress in a layer, Ti, fibre applied stress stiffness matrix stiffness of a layer, Ti, fibre

stress intensity factor at crack tip SIF far field, bridging

τ ΔT

h)

Ktip Kff,br m

Paris equation exponent

ν

shear stress [MPa] difference curing temperature and RT Poisson’s ratio

MODELLING Autoclave curing at elevated temperatures (commonly 120–180°C) causes a significant residual stress state in the laminate as a result of the constituent’s different coefficients of thermal expansion. The actual stress level in the separate layers is an accumulation of the laminate’s curing stress and the (externally) applied load, to be calculated with the classical laminate theory as exemplary done by Homan [3] and summarised in Eqn. 1 to 4 (assuming no off-axis loading):

⎡σ x ⎤ ⎢ ⎥ σ = ⎢σ y ⎥ ⎢τ xy ⎥ ⎣ ⎦

(1)

the total stress level in layer p is written as the sum of (externally) applied loads and curing stresses: −1 σ p = S p [S htcl σ htcl + ΔT (α htcl − α p )]

the stiffness matrix S:

(2)

P. Vermeer, R. C. Alderliesten and R. Benedictus

1072

⎡ Ex ⎢1 − ν ν xy yx ⎢ ⎢ S=⎢ ⎢ ⎢ ⎢ ⎣

ν xy E y 1 − ν xyν yx Ey 1 − ν xyν yx

⎤ 0 ⎥ ⎥ ⎥ 0 ⎥ ⎥ Gxy ⎥ ⎥ ⎦

(3)

expressed for HTCL with the metal/fibre volume fraction (MVF/FVF):

S htcl = STi MVF + S f , 0 FVF f , 0 + S f ,90 FVF f ,90

(4)

Plotting the applied stress versus the actual stress in metal and fibre layers for the HTCL system (figure 2), the high residual stress (σTi = 107MPa, σf = –87MPa) immediately becomes apparent. As the carbon fibre stiffness is much higher than that of titanium (roughly by a factor 2.5), the actual stress difference convergences and eventually crosses at 411MPa. From this point on, the fibre will take the majority of the load. This is contrary to Glare, where the actual stresses in the constituents divergence due to the glass fibre’s lower Young’s modulus.

Figure 2: Actual stresses in metal and fibre layers due to curing and additional applied stresses. Intersection at 411MPa

Hybrid titanium composite laminates

1073

Fatigue crack initiation Homan [3] and Chang et al. [4] predicted the fatigue crack initiation life at room temperature of Glare and hybrid boron/glass/aluminium FML respectively based on the actual stresses as calculated in the previous section. Considering an applied high cycle fatigue load of 250MPa, the stress levels in the metal and fibre layers of current HTCL configuration are plotted in figure 3.

Figure 3: Stress levels in metal and fibre layers under cyclic loading at σmax = 250MPa, R = 0.1 at a frequency of 12Hz. The curing stress in titanium is approximately 107MPa

Fatigue crack propagation The propagation of cracks in FML under fatigue loading is subject in numerous studies and on different approaches. As the amount of available data on crack propagation within this research is limited, only a rough correlation to monolithic titanium will be made. The crack growth reducing effect is the result of the additional contribution of the fibre bridging stress, expressed by Kbridging on the far field stresses (Kfarfield). The resulting stress intensity factor at the crack tip (Ktip) is shown in eqn. 5, as described by Marissen [5] and later by Alderliesten et al. [6].

K tip = K farfield − K bridging

(5)

P. Vermeer, R. C. Alderliesten and R. Benedictus

1074

With a central crack in an infinite plate and the actual stresses after eqn. 2,

K farfield = σ actual πa

(6)

and the Paris equation as obtained experimentally for titanium,

log ΔK tip = ux + v

(7)

the stress intensity factor for fibre bridging can be calculated by:

K bridging = K farfield −

ΔK tip R

(8)

EXPERIMENTAL The current experimental programme focuses on crack initiation and propagation, and comprises monolithic titanium alloy (Ti-6Al-4V) sheet and the HTCL–2/3 laminate. The laminate is configured of two layers of titanium and three layers of intermediate modulus carbon/epoxy propriety prepreg. The principal orientation of the 0.37mm unidirectional fibre plies is in L-direction of the 0.4mm titanium sheet. In order to obtain a sound adhesive bond, the titanium surface pretreatment is based on a sol-gel process, providing an easy and environmental friendly bonding process. Autoclave curing occurs at 180°C for 2 hours at >7 bar pressure and the laminate product in a controlled vacuum of 800 mbar. Fatigue crack initiation Tests have been performed at different stress levels with a stress ratio R = 0.1 and a frequency of 12Hz on flat 50mm wide titanium specimens with a drilled 5.6mm centre hole giving a stress concentration factor of Kt ≈ 2.7 (figure 4). Crack initiation and propagation in titanium specimens have been monitored by the potential drop method; the signal was picked-up through plugs in small holes above and below the centre hole. Unfortunately, the potential drop method proved practically not applicable to the HTCL specimens due to the complex conductivity of the entire system (i.e. metal and fibres) and additional effects (e.g. heating during testing) influencing the signal. Hence, the initiation of cracks has been monitored by a stereo-macroscope with 40×times magnification on the inside flanks of the centre hole.

Hybrid titanium composite laminates

1075

Figure 4: Ti-6Al-4V and HTCL–2/3 specimens for fatigue crack initiation tests

The measured fatigue crack initiation life Ni of monolithic titanium is typically around 85 to 90% of the total fatigue life to failure Nf. The corresponding Wöhler curve with Ni = 1mm crack is plotted in figure 5. 1000 900 Wöhler line according to Weibull Ti-6Al-4V - applied stress 95% Wöhler line Ti-6Al-4V - Run-out

800

Applied stress [MPa]

700 600 500 400 300 200 100 0 1,00E+00

1,00E+01

1,00E+02

1,00E+03

1,00E+04

1,00E+05

1,00E+06

1,00E+07

Ni [cycles]

Figure 5: Fatigue initiation lives and Wöhler line for 0.4mm Ti-6Al-4V. Stress ratio R = 0.1, frequency 12Hz.

1,00E+08

P. Vermeer, R. C. Alderliesten and R. Benedictus

1076 600 550

Ti-6Al-4V - Wöhler line Ti-6Al-4V - applied stress

500

Ti-6Al-4V - 95% Wöhler line HTCL - applied stress

Applied stress [MPa]

450

HTCL - (calculated) stress in Ti layers 400 350 300 250 200 150 100 1,00E+03

1,00E+04

1,00E+05

1,00E+06

Ni [cycles]

Figure 6: Detailed fatigue initiation lives for HTCL–2/3 laminates tested at applied stress levels of 150, 250 and 300Mpa (open circles). Also shown are stress levels in Ti corrected for curing-stress (closed circles). Stress ratio R = 0.1, frequency 12Hz, data of monolithic titanium added for comparison

Figure 7: X-ray CT image of a HTCL fatigue crack initiation specimen tested 60 000 cycles at 300MPa (σTi = 329MPa). Cracks are difficult to resolve (see arrows and lines), irradiation causes the area next to the hole to appear bloomed

Hybrid titanium composite laminates

1077

As discussed earlier, due to residual (curing) stresses in the laminate, applied loads of 150, 250 and 300MPa imply the titanium layers to be stressed at 218, 292 and 329MPa respectively. Figure 6 shows a detail of the S-N curve with the HTCL laminate’s fatigue initiation live. The open markers represent the applied stress only, the closed ones regard the curing stress and approach the titanium’s 95% Wöhler line. As the initiation crack length was not clearly resolvable with x-ray computer tomography measurements (CT), this remains open for further investigation. An xray CT image of a HTCL laminate tested at 300MPa (σTi = 329MPa) is given in figure 7, the cracks at the flanks are enlarged for more detail. Fatigue crack propagation M(T) specimens of 140mm width, with a 3mm centre crack were made for both titanium and HTCL (figure 8).

Figure 8: Ti-6Al-4V and HTCL–2/3 specimen for fatigue crack propagation tests

A da/dN–ΔK curve for titanium was measured to calibrate the potential drop method (figure 9). However, as discussed previously, this method could not be applied for crack length measurements in HTCL. As the laminate is “inverted” and the fibres remain visibly intact, also no observations can be made by classical methods. The Paris equation was taken from the raw data and the coefficient C and exponent m derived.

P. Vermeer, R. C. Alderliesten and R. Benedictus

1078

1,00E-01

da/dN [mm/Cycle]

1,00E-02

da/dn = 3.68×10-12(ΔK )2.75 (Paris equation) 1,00E-03

1,00E-04 10

20

30

40

50

60

70

80

90

ΔK [MPa√m]

Figure 9: Fatigue crack propagation curve for Ti-6Al-4V. Test stress ratio R = 0.1, frequency 5Hz. Paris equation for region II with the corresponding coefficient and exponent for titanium calculated (eqn. 7).

Three HTCL specimens have been subjected to fatigue testing at different load levels for a certain amount of cycles to obtain at least a qualitative impression of the mechanisms involved. These specimens were subsequently investigated by xray CT for determination of crack length and delamination area. One specimen showed beside a crack from the centre saw cut, also cracks from both holes for potential drop connections. This specimen will not be further discussed. The HTCL laminate’s crack length after a total of 577×103 cycles is shown in figure 10 (loading direction in 0° or L). As this was the first HTCL specimen to be tested, the test started at 200MPa for 134×103 cycles, after which the stress was increased to 300MPa for 152×103 cycles and subsequently to 400MPa (actual stress level 405MPa) for another 291×103 cycles. Although x-ray tomography revealed a crack length of roughly 22.5mm, delamination was not observed. This is in agreement with previous research on Glare at Delft University of Technology in which it was shown that crack could be detected with X-ray CT, but the present delamination not. It has been tried to treat the centre crack with Zinc iodide tracer agents to increase contrast in the assumed delamination area, however without success [7].

Hybrid titanium composite laminates

1079

Figure 10: HTCL–2/3 fatigue crack propagation specimen.

Figure 11 clearly shows the 7.6mm crack length in the HTCL specimen after 166×103 cycles at 200MPa (actual stress level 255MPa), again, no delamination is observed.

Figure 11: HTCL–2/3 fatigue crack propagation specimen.

P. Vermeer, R. C. Alderliesten and R. Benedictus

1080

As solemnly the amount of load cycles and total crack length measured by x-ray CT are known, da/dN could be calculated for the final status only. Kfarfield is calculated by eqn. 6, table I

Kfarfield Ktip Kbridging

2a = 22.5mm 76 MPa√m 13 MPa√m 64 MPa√m

2a = 7.6mm 28 MPa√m 13 MPa√m 15 MPa√m

Table I: Stress intensity factors for HTCL after eqns. 5–8

However, Paris’ law is invalid in da/dN–ΔK region I, and the value for Ktip of approximately 13 MPa√m is well below titanium’s threshold intensity range ΔKth of around 20 MPa√m, figure 9. There have been no damaged fibres observed, neither in the wake of the crack tip nor along the path of the crack.

DISCUSSION & CONCLUSION It has been confirmed that fibres generally have no positive effect on the fatigue crack initiation life; causing the introduction of tensile curing stresses, increasing the actual stress level in metallic layers and therewith result in early initiation of cracks. The classical laminate theory seems to be a reliable and valid calculation method for the prediction of fatigue crack initiation. The mechanisms seem qualitatively similar to Glare, as crack propagation does occur in hybrid titanium composite laminates featuring a fibre/crack bridging effect but without affecting or damaging the fibres. Hence, the occurrence of FMLtypical delamination in an area around the crack is considered to be present, however, due to the applied experimental methods not observed. The near future focus will be on enhancement of the experimental techniques by testing with specific equipment and methods, e.g. digital image correlation, as Rodi et al. [8] proved this to be very suitable for investigating delamination in FML with an inverted lay-up.

Hybrid titanium composite laminates

1081

ACKNOWLEDGMENTS Although it is unusual to express gratefulness in conference publications, but considering the circumstances I do like to thank specific colleagues at Airbus, TUDelft and last but not least EADS Innovation Works for their fair deal of support.

REFERENCES [1]

Van Rooijen, R.G.J. (2006), Bearing Strength Characteristics of Standard and Steel Reinforced GLARE, PhD Dissertation, Delft University of Technology, Faculty of Aerospace Engineering.

[2]

Kolesnikov, B., Herrmann, A. and Pabsch, A. (1999), WO 00/56541.

[3]

Homan, J.J. (2006), Int. Journal of Fatigue, vol. 28, pp. 366–374.

[4]

Chang, P.-Y., Yeh, P.-C, and Yang, J.-M. (2008), Materials Science & Engineering A, vol. 496, pp. 273–280.

[5]

Marissen, R. (1988), Fatigue Crack Growth in ARALL, a Hybrid Aluminium-Aramid Composite Material, Crack Growth Mechanisms and Quantitative Predictions of the Crack Growth Rate, PhD Dissertation, Delft University of Technology, Faculty of Aerospace Engineering.

[6]

Alderliesten, R.C. and Rans, C. (2009), Int. Journal of Fatigue, vol. 31, pp. 213–222.

[7]

Feick, A. (2009), Röntgen-Computertomographische Untersuchung von Schadstellen in CFK mittels der Behandlung mit Kontrastflüssigkeit, MSc Thesis, EADS Innovation Works.

[8]

Rodi, R., Alderliesten, R.C. and Benedictus, R. (2009), In: Bridging the Gap between Theory and Operational Practice, Proceedings of the 25th ICAF Symposium, Rotterdam.

Structural health and structural loads monitoring

25th ICAF Symposium – Rotterdam, 27–29 May 2009

PILATUS PC-21 – A DAMAGE TOLERANT AIRCRAFT Lukas Schmid Pilatus Aircraft Limited

Abstract: The PC-21 Advanced Trainer was designed and certified as a damage tolerant aircraft. The PC-21 is certified to the acrobatic category of FAR 23 Amendment 23-54. Some major certification aspects are addressed in this paper. The design fatigue spectrum was created based on pilot input in a deterministic manner. The distribution of vertical acceleration was shown to result in fatigue damage similar to FALSTAFF. The aircraft inspection intervals were defined by crack growth analyses, whereas results of the FSFT were taken into account. The Full Scale Fatigue Test (FSFT) was performed by considering a life scatter factor of 3. First, two lives of durability testing were performed. After introduction of artificial damages another life of damage tolerant testing was run. Finally, a residual strength test campaign was carried out. The certified Fatigue Monitoring System (FMS) allows monitoring the loading history of major structural assemblies by means of strain sensors. Several Fatigue Indices are calculated from the strain signals on a flight-by-flight basis for Individual Aircraft Tracking purposes (IAT).

AIRCRAFT PROPERTIES The PC-21 was conceived as a 21st century turboprop training solution for training fighter jet pilots. The aircraft is capable of covering primary, advanced, and tactical pilot training missions. The design of aerodynamics and avionics result in jet-like flight behaviour and handling.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1085–1092. © Springer Science+Business Media B.V. 2009

Lukas Schmid

1086

FATIGUE DESIGN Design Philosophy Traditionally, Pilatus designed aircraft to meet safe-life requirements. For the PC21, however, it was decided to adopt damage tolerance as design philosophy. The reasons for this philosophy shift are the following: • • • •

Life-cycle cost: The design cost for damage tolerant aircraft are higher. However, the operational costs are lower as the inspection program is optimised on the basis of test results and analysis. Safety: Damage tolerant aircraft are safer than conventional safe-life aircraft as they are inspected based on worst case initial conditions. Inspectability: As aircraft are getting older more and more inspections are normally implemented by adopting damage tolerance principles. Damage tolerant aircraft are designed for inspectability of critical locations. Material choice: Damage-tolerant alloys reduce the sensitivity to fatigue cracks by allowing longer cracks and extend the inspection intervals.

Certification Aspects The requirements of FAR 23 Amendment 23-54 (§23.571, §23.572, and §23.573) are neither clear in terms of damage tolerance requirements nor did the FAR 23 guidance material such as [1], [2], and [3] provide help in the certification of a damage tolerant aircraft. Therefore, considering the intended use of the aircraft as a military training aircraft, military specifications such as [4], [5], [6], and [7] were reviewed. As a result the following requirements were derived for the PC-21: 1.

2.

3.

4.

Fatigue Design Spectrum: Instead of using normal acceleration (Nz) distributions of [2] for acrobatic category FAR 23 aircraft, a dedicated Master Design Spectrum (MDS) was conceived. Fatigue Analysis: Fatigue analysis was used for risk mitigation purposes in the design phase of the PC-21 and for inspection interval determination upon completion of the Full Scale Fatigue Test. Full Scale Fatigue Test (FSFT): The test scatter factors for the FSFT were derived from military damage tolerance requirements instead of using the factors given for safe-life FAR 23 aircraft. Fatigue Monitoring System (FMS): An FMS was included to enable Individual Aircraft Tracking (IAT) [4].

To each of these four aspects a separate section is dedicated below.

Pilatus PC–21 – A damage tolerant aircraft

1087

FATIGUE DESIGN SPECTRUM The Master Design Spectrum (MDS) was developed on the basis of a mission specification. The MDS was designed according to a bottom-up approach (Figure 1). The manoeuvres were combined into so-called design sorties, i.e. unique flights. The design sorties were assigned to mission type, i.e. either to Primary, Advanced, or Tactical Training. By using discrete values for the mass, centre of gravity, airspeed, and altitude, the number of unique load cases was significantly reduced. This approach was deemed both more flexible and more suitable for the PC-21 than using Nz exceedances provided by the FAA for acrobatic category aircraft [2]. The MDS vertical acceleration (Nz) sequence was compared to the FALSTAFF spectrum [8] by means of a Crack Initiation calculation. It showed that the fatigue damage of the MDS for several stress levels is similar to the fatigue damage of the FALSTAFF.

PC-21 MDS

Missio Primary Training Advanced Training Tactical Training

Design Sortie DS01 DS02 DS03 DS04

Manoeuvr Take Off Climb Stall Loop ...

DS05 DS06 ...

Figure 1: Bottom-up approach for the PC-21 Master Design Spectrum (MDS)

FATIGUE ANALYSIS Fatigue analysis was used early in the development of the PC-21 for risk mitigation, i.e. to demonstrate durability and damage tolerance behaviour of the

Lukas Schmid

1088

critical locations prior to FSFT performance. Post-FSFT analyses mainly covered determination of inspection intervals. Pre-Test Analysis Fatigue analyses for the critical locations were performed before conducting the FSFT in order to prevent early failures and to demonstrate long inspection intervals. Post-Test Analysis After completion of the FSFT the critical locations were re-assessed based on the test results in order to determine inspection intervals. The re-analysis was performed using the so-called pegging procedure [9], i.e. the analytical results were adjusted to the test results.

FULL SCALE FATIGUE TESTING The two main objectives of the Full Scale Fatigue Test (FSFT) were: 1. 2.

To substantiate the service life of the aircraft. To verify the damage tolerance capability of the airframe

The test article configuration included fuselage, engine mount, wing, and vertical tail. The horizontal tail and the landing gears were replaced by dummy structures in order to facilitate load introduction. The test instrumentation consisted of strain gauges and displacement transducers. The loading was applied by push/pull actuators. In order to simulate the pressure distribution in the wing, the actuator loads were distributed by whiffletrees and introduced by bonded pads on the upper wing skin. This setup enabled realistic load introduction and provided access to the critical lower wing locations for inspection (Figure 2). The FSFT consisted of four major phases: 1. 2. 3. 4.

Durability test, two lives [7] Damage tolerance test, one life, after introduction of artificial damages Residual strength test Tear down inspection, focus on critical locations

Pilatus PC–21 – A damage tolerant aircraft

1089

Figure 2: Test setup of the PC-21 FSFT

FATIGUE MONITORING SYSTEM System Overview The Fatigue Monitoring System (FMS) was implemented in a so-called Health and Usage Monitoring System (HUMS). The HUMS consists of an onboard Data Acquisition and Storage Unit (DASU), a laptop computer called Line Test Equipment, and a centralised HUMS Data Analysis System (HDAS), which provides extensive offline data analysis capabilities and network access to the data. The end-to-end process of monitoring fatigue was dubbed FLAME, an acronym for Fatigue Life Appraisal by Monitoring Equipment. This process was implemented into the FLAME software, which calculates Fatigue Indices (FIs) on a flight-by-flight basis. A schematic overview of the system is given in Figure 3. Fatigue Data Acquisition. The fatigue relevant data are recorded off the system bus as well as from sensors connected directly to the DASU. These consist of strain sensors and a dedicated vertical acceleration gauge. In order to use storage capacity economically, the strain/acceleration data are filtered in real time.

Lukas Schmid

1090 DASU • Data Acquisition • Fault detection

HDAS Satellites (Other customers)

HDAS Master (Pilatus) LTE • A/C Data Download and Transfer • Fault Diagnostics • Other Maintenance Functions

HDAS Satellite (Customer) • Fleet Data Analysis and Reporting

• PC-21 Data Analysis and Reporting • Fatigue Data Processing and Reporting

Figure 3: PC-21 HUMS overview

Instrumentation. Locations sensitive to single, predominant loads such as the wing bending moment were down-selected as candidates for applying FMS strain sensors [10]. Several strain sensor locations were defined prior to the FSFT for evaluation purposes. Several locations in the wing, the centre fuselage, and the empennage were chosen as a result of the FSFT measurements. The instrumentation was applied redundantly. The strain sensors are calibrated using both ground and flight measurements. Fatigue Index Results. The FI results are reported to the fleet operators in threemonth intervals. A typical example is given in Figure 4. Such results enable the operators to: 1) Distribute fatigue usage evenly across the fleet 2) Take action to improve the maintenance practises if data have been lost 3) Review the fleet usage in terms of damaging missions if applicable No direct connection of FI results with aircraft inspection intervals is established. Large deviations from the design spectrum are dealt with on a case-by-case basis.

Fatigue Index

Pilatus PC–21 – A damage tolerant aircraft

1091

Aircraft A Aircraft B Aircraft C Aircraft D Accumulated FH

Figure 4: Typical FI results

Certification Aspects Certifying an FMS under FAR 23 was found to be a novelty, as neither requirements nor guidance existed. Therefore, guidance on HUMS systems for normal category rotorcraft was considered [11]. Also, additional documentation provided by the FAA was consulted [12], [13], [14]. For certification, an alternative approach based on qualitative assessment was chosen [15]. The FMS was designed such that its components are either redundant or monitored independently. In addition, procedures were put into place to ensure integrity of the FI results.

CONCLUSION The PC-21 was certified as a damage tolerance aircraft to FAR 23 requirements. The most noteworthy aspects of the certification approach include the following: 1) Development of a deterministic fatigue design spectrum similar to FALSTAFF 2) Full-scale test concept using push-pull actuators supported below the wing but loading the wing from top and thus providing access to the lower wing 3) Fatigue analysis (CI/CG) in early design in order to improve damage tolerance behaviour and after full-scale testing to adjust inspection intervals based on test results

Lukas Schmid

1092

4) Implementation of a strain-based fatigue monitoring system for Individual Aircraft Tracking (IAT) purposes with a significant level of redundancy

REFERENCES [1] FAA (1993), AC 23-13, Fatigue, Fail-Safe, and Damage Tolerance Evaluation of Metallic Structure for Normal, Utility, Acrobatic, and Commuter Category Airplanes [2] FAA (1973), AFS-120-73-2, Fatigue Evaluation of Wing and Associated Structure on Small Airplanes [3] FAA (1994), ACE-100-01, Fatigue Evaluation of Empennage, Forward Wing, and Winglets/Tip Fins on Part 23 Airplanes [4] USAF (1975), MIL-STD-1530A, Aircraft Structural Integrity Program, Airplane Requirements [5] USAF (1979), MIL-A-83444, Guidelines for the Analysis and Design of Damage Tolerant Aircraft [6] USN (1987), MIL-A-8867C, Airplane Strength and Rigidity Ground Test [7] USN (1987), MIL-A-8866C, Airplane Strength and Rigidity Reliability Requirements, Repeated Loads, Fatigue and Damage Tolerance [8] van Dijk, G.M. and de Jonge, J.B. (1975). In: Problems with Fatigue in Aircraft, Proceedings of the 8th ICAF Symposium, pp. 3.61/1–3.61/39, Branger, J. and Berger, F. (Eds.), Swiss Federal Aircraft Establishment, Emmen [9] M. Gottier, A. Kuo, S. McCord (1993), Application of FSFT Results to Structural Modifications, Proceedings of the 17th ICAF Symposium [10] Molent, L. (1998), A Unified Approach to Fatigue Usage Monitoring of Fighter Aircraft Based on F/A-18 Experience, In: Proceedings of ICAS 1998, Melbourne, Australia [11] FAA (2003), AC 27-1B, Chg 1, MG 15, Airworthiness Approval of Rotorcraft Health Usage Monitoring Systems (HUMS) [12] U.S. Department of Transportation (2004), DOT/FAA/AR-04/3, Assessment of Helicopter Structural Usage Monitoring System Requirements [13] U.S. Department of Transportation (2004), DOT/FAA/AR-04/6, Continued Evaluation and Spectrum Development of a Health and Usage Monitoring System [14] U.S. Department of Transportation (2004), DOT/FAA/AR-04/19, Hazard Assessment for Usage Credits on Helicopters Using Health and Usage Monitoring System [15] FAA (1999), AC 23.1309-1C, Equipment, Systems, and Installations in Part 23 Airplanes

25th ICAF Symposium – Rotterdam, 27–29 May 2009

INTRODUCTION TO SERVICE OF AN ARTIFICIAL NEURAL NETWORK BASED FATIGUE MONITORING SYSTEM Steve Reed1, Brian McCoubrey2 and Andy Mountfort3 1

Defence Science and Technology Laboratory, UK MOD 2 Bombardier Aerospace, Belfast 3 QinetiQ Ltd, Farnborough

Abstract: Many legacy military aircraft have relatively simple fleet-wide aircraft fatigue monitoring systems, often based upon exceedance counts from a normal accelerometer (Nz). These data are coupled with additional information, such as aircraft mass and store configuration and assumptions of usage, such as point-inthe-sky dynamic pressures. This information is combined within fatigue meter formulae to give an indication of fatigue consumption, for Nz-driven critical features, usually on a sortieby-sortie basis. Substantiation of these systems using Operational Loads Measurement (OLM) data is usually based upon average performance of the fatigue meter formulae in comparison with the OLM strain-based data. It is usual for corrections to be applied to these formulae as a result of this process; often these corrections can be very significant. Sufficiently accurate monitoring of inService fatigue life is essential from both a safety and cost-ofownership perspective. However, obtaining funding for improvements to equipment in individual aircraft fatigue monitoring systems for legacy aircraft is often difficult due to higher-priority funding requirements. This paper describes an alternative, cost-effective approach to obtaining a significant increase in individual aircraft fatigue monitoring accuracy, using legacy equipment and ride-along data from an existing OLM programme. In this process, artificial neural networks are used to determine the relationships between input parameters from the legacy equipment, such as Nz counts and the flight-by-flight M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1093–1119. © Springer Science+Business Media B.V. 2009

1094

Steve Reed, Brian McCoubrey and Andy Mountfort

fatigue damage calculated from strain data captured during the OLM programme. Within this paper, the development, verification and validation of the Structural Health and Usage Neural Network (SHAUNN) fatigue meter formulae for 2 critical wing features for the Tucano TMk1 military trainer is described and processes for the ongoing monitoring of the continued performance of the SHAUNN formula in service are recommended.

INTRODUCTION Fleet-wide Monitoring Systems (FMS) are used to track the fatigue life consumption of individual aircraft, and form an integral part of the fatigue substantiation process of UK military aircraft as defined in Defence Standard 00970 [1]. Typically a FMS will use the output from instrumentation installed on an aircraft (e.g. accelerometers or strain gauges) combined with additional usage information, and calculate the flight-by-flight consumed life relative to the fatigue life clearances provided by fatigue testing and supporting analysis. The FMS will return an output in Fatigue Index (FI), which is a measure of the total consumed life relative to the safe life of the aircraft. Ideally, the safe life of the aircraft will equate to 100FI, with FI providing a measure of the percentage of total fatigue life consumed. The FMS will only be able to track the consumed fatigue life of structural components for which the instrumentation and usage information allow the loading history to be calculated. Structural components tracked by the FMS are termed “monitored”, and FI will be used for fatigue management. The remaining structure is termed “unmonitored”, and will have higher fatigue scatter factors applied as part of the safe life substantiation process to account for the additional uncertainties of not being monitored on individual aircraft. The fatigue life of unmonitored structure will be tracked using the measurands most appropriate to that component e.g. flying hours or landings.

RAF TUCANO TMK1 FLEET-WIDE MONITORING SYSTEM The Royal Air Force (RAF) Tucano TMk1 (Figure 1) FMS uses a Fatigue Meter and a Fatigue Meter Formula (FMF) to calculate FI. The Fatigue Meter registers aircraft normal acceleration (Nz) exceedances at a number of pre-set g levels covering the design range of the aircraft. The g exceedances for each sortie are downloaded into a database and the FI for the sortie calculated using the fatigue formula.

Neural network fatigue monitor

1095

Figure 1. RAF Tucano TMk1 The current FMF, developed at the introduction to service, calculates the fatigue damage on the wing rear spar lower boom at station 545 (Rib 3). This location had been identified by analysis as the critical structural location. The basic steps in the formula are that the Nz exceedances and other captured Flight and Fatigue Data recorded (i.e. mass data and sortie profile code (SPC)) are combined with design assumptions to calculate the wing root bending moment for each exceedance [2]. The bending moments are then converted to stresses using a relationship derived from the wing static test results. Thereafter, the exceedances are manipulated and matched together into range-mean-pairs representing stress cycles. Then the range-mean-pair stress cycles are applied to a S-N curve and a Miner’s sum damage is calculated. The S-N curve in the formula reflects the wing spar material, aluminium alloy 7075-T7351, with a Kt = 4.0. A life factor of 5 is applied in the damage calculation and the formula returns an FI value for each sortie. A number of minor changes have been made to the current FMF since its introduction to service, mainly to reflect the introduction of new SPCs and an increase in Max Take-off Weight (MTOW). The wing is the only major component with loading proportional to Nz. Consequently, the wing is considered monitored, with the remainder of the structure including fuselage, tailplane, fin and landing gear is considered unmonitored. The specified design life of the Tucano T Mk1 is 12000 flying hours. The associated load spectrum, deduced from sortie profile information and usage assumptions, was applied to the fatigue formula resulting in an equivalent fatigue index of 96 FI. This assumed spectrum was applied to an early RAF production standard airframe Full Scale Fatigue Test, termed the 1st FSFT. This resulted in a 1st test-based life clearance of 4800 hours and 38.4 FI.

1096

Steve Reed, Brian McCoubrey and Andy Mountfort

SUBSTANTIATION USING OPERATIONAL LOADS MEASUREMENT The effectiveness of the fatigue meter formula was assessed during the Operational Loads Measurement (OLM) programme [2]. The basic conclusion was that the current fatigue formula underestimated fatigue consumption by a typical factor of 2½ (dependant upon the particular feature). This was attributed to 3 main sources: the differences between the design assumed and OLM-derived Nz to bending moment relationships, the differences between the strains measured on the 1st FSFT and the OLM aircraft and the change in loading environment between 1st test spectrum and actual usage, particularly increases in number of small cycles. The different S-N curves employed in the fatigue formula and OLM analysis would have had different sensitivities to this change in loading environment. Simplistically it should be possible to account for the discrepancy between the OLM damages and the fatigue formula calculation by either applying a factor to the calculated FI or reducing the published life clearances. The latter approach has been applied to the Tucano in the interim; however these life clearances are based on the OLM flying distribution and it was demonstrated that their validity was dependant on individual aircraft flying reflecting this distribution. This arises because, as already mentioned, the fatigue formula and OLM analysis use different S-N curves. Therefore the ratio of damage between the fatigue formula and the OLM features would vary as the severity of the flight varies. Therefore, the recommendations of the OLM final report included a revision to the FMF to revise the Nz to wing bending moment relationships and to use 2 of the S-N curves used for the OLM analysis within the formula, to reflect the different types of features found in the wing structure.

REVISIONS INCLUDED IN THE FATIGUE METER FORMULA Fatigue Critical Features A complex wing structure like the Tucano contains different types of structural features with varying sensitivities to changes in loading spectrum. Following a post-OLM review of the wing structure and early results from the 2nd FSFT, the 2 features selected to act as monitors in the FMF were the wing centreline buttstrap and the wing lower spar boom. Wing Centreline Buttstrap (WCLB) The wing main spar centreline buttstrap could be considered as the primary fatigue critical component on the aircraft and had a life limitation imposed as a result of the OLM analysis. In the OLM analysis the feature was represented by a High Load Transfer (HLT) S-N curve reflecting the high load transfer across the joint and an anticipated failure location at the 1st bolthole outboard of the centreline.

Neural network fatigue monitor

1097

This was the initial approach for the development of the FMF. However during the development of the revised FMF (timed to coincide with expected failure point on test) damage was detected on the centreline buttstrap of the 2nd FSFT article. The damage was at the centreline of the strap and a detailed review of this damage concluded that the wing centreline joint would be better represented as a notch feature and hence an appropriate S-N curve from the OLM analysis was used. Wing Spar Lower Boom (WSLB) It is necessary to have a feature to be used as a general monitor of the consumed life of the wing. The predominant features on the wing are the skin spar joints and spar centre-line joints. As the centre-line joint has a specific feature it was considered that the wing general monitor should be represented by a feature representative of a typical spar lower boom using a Low-Load Transfer (LLT) S-N curve. The wing spar lower boom at rib 3 (wing stn 545) was assessed during the OLM programme and this location was used within the FMF. Fatigue Scatter Factors The FI calculations in the original FMF used a fixed life factor of 5 applied to the life calculated from the mean S-N curve. However, the revised FMF employs Safe S-N curves, established using factors on life and stress, as defined in [1].

STUCTURAL HEALTH AND USAGE NEURAL NETWORK Regulations Recent changes in UK MOD regulations [3] provided a framework within which qualification evidence for the use of non-adaptive prediction methods, such as artificial neural networks, could be used to provide a clearance for their use in aircraft fatigue monitoring systems. A detailed explanation of this framework is described in [4]. Much of the work published to date in this area of development concentrates on predicting stress, strain or load time histories using flight parameter time history data, effectively producing virtual strain gauge outputs. However, for legacy systems such as the Tucano, the approach needed was to predict fatigue damage directly from the data already captured in Service. Supervised Learning Neural Network The basis of the Structural Health And Usage Neural Network (SHAUNN) is a relatively simple multi-layer preceptron (MLP) artificial neural network (ANN) [5, 6]. MLPs are supervised networks in that they are given examples of input parameters and corresponding output parameters and the ANN learns the relationship between the inputs and outputs by adjusting the initially random weights and biases within the network. This is basically a regression fitting technique. A supervised learning schematic is illustrated in Figure 2.

1098

Steve Reed, Brian McCoubrey and Andy Mountfort

Figure 2. Supervised learning schematic for a supervised ANN Weights and Biases A simplified version of the insides of the MLP is illustrated schematically in Figure 3. The normalized inputs are each linked to the hidden-layer neurons by a weight vector (w) and a bias vector (b) is applied to each neuron. These weights and biases can be thought of as analogous to the equation used in linear regression (y = mx+c). The weight can be considered as the gradient (m) and the bias as the intercept (c). The network in Figure 3 has been illustrated with one output node, as is used in the application here but many output nodes can be used if required.

Figure 3. Schematic of a simplified MLP ANN Hidden Layer Activation Functions A MLP is a feed-forward network and the hidden layer contains a number of neurons that receive weighted data from each input node and pass it to the next layer node, via the activation function. The function of each of these non-linear

Neural network fatigue monitor

1099

activation functions is akin to a mechanical non-linear gearbox. The relationship between the input and output is non-linear but constant. In the network illustrated in Figure 3 and used in this application, the logistic sigmoid functions are used. Other shaped functions can be used as long as they are continuous differentiable monotonically non-decreasing functions. Being able to differentiate the activation function is the key element which separates a MLP from previous developments and the back-propagation of the error (solving the credit assignment problem) is dependent upon this facet. Consequently hard-limit functions cannot be used in MLP. Output Layer Data from each neuron in the final hidden layer are passed to the output layer. Here the output layer is a logistic sigmoid function (range 0-1), again with a bias. However, simple linear summation, again with a bias, is often used for an output without range restriction. The outputs of the 2 layer network, illustrated in Figure 3, and used in this application, with both layers using a logistic sigmoid activation functions would be calculated by using Eqn. 1: y = logistic[∑ {w2 * ( logistic { ∑ (w1*x + b1)}+b2}]

(1)

where, y is the output matrix, logistic describes the logistic sigmoid activation function, x is the input matrix, w1 is the first layer weight matrix, w2 is the second layer weight matrix, b1 is the first layer bias vector and b2 is the second layer bias vector. Training of the network can be considered as an error minimisation process between the desired output and the predicted output but being wary of overfitting the solution and obtaining poor generalization. This is where the network, when given new data, is unable to match the previous level of prediction. A pictorial illustration of the training process along the error surface is illustrated in Figure 4.

Steve Reed, Brian McCoubrey and Andy Mountfort

1100

Figure 4. Simplified schematic of training process on error surface To ensure successful training, data are divided into training, test and validation sets. Training data are used to optimize the weights and biases within the model. The statistically similar test data are used to ensure the solution is not over fitting to the training data (good generalization). Thereafter, the validation data sets are used for independent assessment of the performance of the model and to identify solution confidence limits.

PRELIMINARY MODEL OPTIMISATION Input data A total of 867 sorties of data captured from the 3 instrumented OLM aircraft were used to train, test and validate the revised FMF. The data set is identified in Table I.

Tail Number

Sorties Provided

ZF212 ZF406 ZF512

246 330 291

Total

867

Range of Sortie Numbers 2120 Æ 2498 1926 Æ 2298 1961 Æ 2276

Parameters g Counts (-4g Æ 7g), Sortie Profile Code, Sortie duration, Roller and Braked Landings, Mass (Basic, Fuel Start and End, Crew, Bag), Spins Left and Right WCLS and WSLB safe damage values from loads calibrated strain data

Table I. Input data set

Neural network fatigue monitor

1101

Input optimisation There is always a significant burden of optimisation in developing an ANN model. As well as optimising the model architecture and training parameters (discussed later), optimisation of the input data is also required and domain knowledge is an essential element in this process. The first level of optimisation undertaking was elimination of input parameters that were not significant contributors to the mapping solution. Use of preliminary models for optimisation In the following sections, reference is made to the use of “preliminary” models as a course method of assessing the contribution of input parameters. These preliminary models were single hidden layer 25 hidden unit models, with a learning rate of 0.1 (i.e use 10% of the error to back propagate) and a momentum coefficient of 0.5 (based upon previous experience of modelling wing strains from flight parameters. Half the data set was used as training data and half as test data. Typically, models were trained for between 1000 and 3000 epochs (passes of the data), depending on model convergence and were trained sequentially. This means the error was back propagated after each data point (434,000 – 1,302,000 times during training) and the data order was randomised after each epoch. Where repeatable results were obtained, 3 models were trained for each condition; if results were not so repeatable 5 or more models were trained and outliers eliminated. The optimisation of the architecture of the final models is discussed in later sections. These preliminary architectures had been found generally to produce adequate models and were considered acceptable for initial optimisation work. Sortie duration It was considered that g counts, mass and possibly landings were likely to be the major input parameters and hence preliminary models with and without sortie duration were run. Sortie duration, as expected did not improve the accuracy of these preliminary models as fatigue damage accrual for the wing of an aircraft such as Tucano is not driven significantly by sortie duration and hence was rejected from the data set. Sortie profile code Sortie profile code (SPC) is not a scalar quantity (i.e. SPC8 is not twice or half as damaging as SPC4 or even necessarily more or less damaging) and hence to be used in these models, data would have to be segregated by SPC and SPC-specific models generated. This was considered a highly unattractive option as it would reduce the training and testing data sets in some cases to single figures. Hence, preliminary models were run and the performance of differing SPCs investigated. It was concluded that it was not necessary to segregate the data by SPC and hence this parameter was removed from the data set.

1102

Steve Reed, Brian McCoubrey and Andy Mountfort

Spins The number of spins (left and right) was also considered unlikely to rank highly as input parameters in wing-feature models, although this might not be the case for a fin model for example. Again, this position was confirmed with several preliminary models. Landings The remaining input parameters were g counts, mass information and landings. It was decided to group landings as a single parameter as the sum of braked and roller landings as there was considerable variation in the number of landings within the data set (1Æ13) and this would provide a relatively good scalar parameter, particularly for some of the benign sorties with very few g counts. Mass The representation of mass was considered in greater detail. Following discussions, it was decided to represent mass in 3 ways as either a single mid-sortie mass parameter, as a 2-parameter input with a fuel mass value and a fixed mass value and finally to exclude mass entirely from the inputs. Over 12 models trained the single parameter mid-sortie mass models had the better performance. Although the 2 parameter method is probably the most correct in engineering terms, the effect of effectively separating out the more highly variable fuel mass from the more fixed masses (basic mass + crew) means that the fixed mass occupied only a handful of states throughout the data set and hence was a poor scalar parameter. Based upon these results, it was decide to use a single mid-sortie mass parameter in the optimised models. g counts The Tucano is fitted with a ‘synthesised’ g meter and exceedances of normal acceleration of -4g, -3g, -2g, -1g, 0g, 2g, 3g, 4g, 5g, 6g, and 7g are recorded. Ideally, use of each of these data would have been made; however, the 867 sortie data set did not contain any counts in the -4g, -3g and the 7g windows. Hence, these data were removed from the training parameters. Additionally, the maximum value in the 6g window was one count and only 6 sorties in the data set had registered a count. This would risk producing a poor parameter of only 2 states which is to be avoided if possible. Hence the 6g data were also removed from the data inputs. The -2g record had 22 non-zero entries with values up to 3 counts and hence this parameter was retained in the data set. All the remaining g windows were well populated in the range from zero counts to the maximum number of counts. Therefore, the g-count data inputs used in the models were: -2g, -1g, 0g, 2g, 3g, 4g and 5g. Methods of accounting for entries outside of these values are discussed later in this report.

Neural network fatigue monitor

1103

Input Parameter Summary The following input parameters were used for the development of the WSLB and WCLS SHAUNN models: • • • • • • • • •

Total landings -2g counts -1g counts 0g counts 2g counts 3g counts 4g counts 5g counts Mid-sortie mass

TRAINING, TEST, INDEPENDENT VALIDATION SET AND FINAL CHECK DATA Data sets The preliminary models described in the previous section were trained using half the data set and tested with the other half of the data. However, for model optimisation and validation it is necessary to divide the available data into more than just a training and test set. It is essential that any prediction is matched by an indication of the confidence in that prediction. The background to the use of confidence limits is explained in the following section. How the data were divided into [Train], [Test], [IVS] and [FCheck] data sets is explained later on in this section. The following section describes the use to which the independent validation set (IVS) data are put. Confidence Limits The difference in performance in predicting the test-set of several models can be relatively small and it would be unwise to use test-set performance alone to select models. Although the test set is independent of the training set, a further degree of confidence results in assessing model performance using a totally, although statistically similar, independent data set. The procedure developed for using an IVS, to determine expected prediction values, using linear regularisation, and defining confidence limits, was developed from a method described by Jepson et al. [7]. Confidence in the prediction made by a neural-network based model is probably the greatest impediment to their use for in-service safety related or safety critical applications. Therefore, establishing a process to define confidence limits in the predictions made by the models was considered an essential development. Each model is presented with the IVS data set and the Pearson correlation coefficient (R) and the best-fit line, using the least squares method, for the

1104

Steve Reed, Brian McCoubrey and Andy Mountfort

predicted and target IVS values were calculated. The best-fit line between predicted and target was then used to calculate an expected value for the model, where the expected value was governed by the following simple equation:

Expected Value = m * Predicted Value + c

(2)

Where, m is the gradient of the least squares regression line and c is the intercept term. Hence, a linear relationship between the output of the network (prediction) and the expected value, from the IVS, was identified. If it is assumed that the errors around the regression line are normally distributed, it is possible to generate a percentage confidence limit around this line. These limits are in effect a pair of parabolic lines with the limits being closer to the best-fit regression line in the area where the bulk of the data lie (i.e. near mid range) and moving outwards as the regression line is extrapolated in each direction. This is illustrated in Figure 5.

Figure 5. Regression Line and Confidence Limits By using this procedure, rather than relying upon a single estimate produced by the model, the performance of the model from the IVS could be used to correct the output and produce an expected value. Furthermore, and most importantly, it was possible to say with a 95% confidence that the expected value should lie between the upper and lower confidence limits. This was further verified by predictions using the 4th Final Check data set. Therefore, for each of the models generated, values were produced for predicted damage, expected damage and upper and lower confidence limit damage. Selection of Input Data for Training, Test, IVS and Final Check Data Sets Selection of data for training and various test sets from a limited data set poses several conflicts and, inevitably, some compromises have to be made. Where

Neural network fatigue monitor

1105

possible, ANN models should be used to interpolate within their training experience. However, it is unrealistic to assume that 867 sorties of OLM data will contain sufficient examples from throughout the flight envelope of all input parameters to ensure the model is always interpolating rather than extrapolating. Several methods of extracting training and test from the provided data set were experimented with. The favoured method ensured that the extremes of each input parameter were contained within the training set and the remainder of the data were divided between the data sets in a semi random method (by flight number). This ensured that the maximum values in each input parameter were in the training set and that the data sets had broadly similar statistics. Additionally, each data set contained a different number of points to reduce the risk of selecting the wrong data set in error. In all, 294 sorties were used for the training sets, 192 sorties for the test sets, 191 sorties for the IVS and 190 sorties for the FCheck data set. The convention for sortie numbering is for example, sortie 2122200 is from aircraft ZF212 sortie 2200.

MODEL DEVELOPMENT Model Optimisation Optimisation of the models to ensure the best possible solution, while avoiding unnecessary complexity, requires a degree of iteration. There are also many elements within the architecture and training regime that can be optimised and therefore dramatically increase the number of models produced. Preliminary models illustrated that single hidden layer architecture would suffice and hence the principal architectural optimisation was the number of hidden units, or neurons, in the hidden layer. Too few could fail to capture the non-linearity in the hyper-plane mapping between input parameters and target values; too many would introduce unnecessary complexity and greater risk of hard-coding error. Obviously this cannot be visualised in 9 dimensions (9 input parameters) but if one thinks of the problem in 2 dimensions, the ANN can be thought of as trying to describe the equation of a curved line. The more complex the shape, the potentially higher order polynomial is required. Equally, describing a straight line using a multiorder polynomial is equally challenging. Hence, around 150 models were run with between 5 and 25 hidden units. As well as the architecture of the model, its length of training, in terms of epochs or passes of the data needed to be optimised. The aim is reach the minimum error solution without the risk of over fitting the model to the training set. Over fitting is characterised by a reducing error in the training data set whilst and increasing error in the test data set occurs. Training Environment A model training environment was constructed, using the core Matlab® programming language. The sequential training algorithm was used to update the weights and biases within the model after every presentation of a training data

1106

Steve Reed, Brian McCoubrey and Andy Mountfort

example (data point). With a training set of 294 sorties the weights in the model would have been updated 735,000 times during the training (2500 {epochs} x 294 {data lines}), with the order of sorties being randomly arranged after each pass (epoch) of the data. Typically, this model would take approximately 200 seconds to train on a PC with Pentium M 2.10GHz processor and 2GB of RAM. Although weight updates were carried out at each presentation of a training example, summary data were presented to screen only after the completion of each epoch during training. At the end of the model training, a summary of training and test set errors was presented to screen and a graphical representation of the training was produced (Figure 6):

Figure 6. Example training environment plot In Figure 6, plot ‘a’ provides a history of the change in mean squared error for the training and test data sets, updated at the end of each epoch. Divergence of the test set error and training set error would indicate a high risk of developing an overfitted solution. Plots ‘b’ and ‘c’ provide correlation data for the trained algorithm, using the training set (plot ‘b’) and the test set (plot ‘c’). For plots ‘b’ and ‘c’ the correlations for the entire training and test set between target values and those predicted by the model were plotted, with the correlation coefficient and the bestfit line appended to the chart. Again, large variations between training and test set values would indicate a risk of an over-fitted solution. Additionally, these correlation plots provided a rapid indication of the likely performance of the model. Plot ‘d’ provides an illustration of the final values for the weights and

Neural network fatigue monitor

1107

biases in the trained model. This plot provides an extremely useful indication of the health of the network. A normal-type distribution, centred close to zero, as illustrated in this plot, would suggest a healthy network. Finally, plot ‘e’ provides a comparison of cumulative damage from the test target data and the corresponding model prediction. This plot could be zoomed in on to provide a more detailed view of specific areas of the time history, for investigation of prediction errors. Model Selection As already discussed, the training and model variables optimised during the development process were the number of hidden units (neurons) and the number of epochs, or passes, of the training data. Additionally, several criteria we used in combination to select models for further assessment. These criteria are outlined below and the relevant plot from Figure 6 is identified: • • • • • • • • • • •

Cumulative damage ratio (total test damage / total target damage) (Plot ‘e’). Progressive damage ration (Plot ‘e’) Pearson Correlation Coefficient for the training set (Plot ‘b’). Pearson Correlation Coefficient for the test set (Plot ‘c’). Comparison between above Correlation Coefficients. Best line fit for the training set (Plot ‘b’). Best line fit for the test set (Plot ‘c’). Comparison between above best line fits. Training and test error performance (Plot ‘a’). Error weight distribution (Plot ‘d’). Mean squared error for the test data set.

Analysis of the more than 150 models generated illustrated that models with 5 neurons in the hidden layer were sufficiently complex to capture the mapping relationships within the data for both the spar lower boom (WSLB) and the centre line strap (WCLS) feature. Furthermore, no significant improvement in performance was achieved by increasing the model complexity. Additionally, optimum models were achieved by training the WSLB feature models over 1500 epochs, without risk of over fitting; whereas, 2500 epochs appeared to be more appropriate for the WCLS feature. Based upon these results, 5 models for each feature were selected for further testing. These were models WSLB046Æ WSLB050 and models WCLS052 Æ WCLS056.

MODEL TESTING IVS Assessment Each of the 10 selected models was then assessed using the IVS. As already described, this process allowed a completely independent assessment of the

1108

Steve Reed, Brian McCoubrey and Andy Mountfort

performance of the models; additionally, a linear regularisation correction and confidence limits were generated. For each model, the damage ratio (sum of the predicted damage / sum of the target damage), correlation coefficient (R) and best fit line between the predicted and target values were calculated. Then the linear regularisation equation was calculated (effectively converts the best fit line into y = x). This regularisation was then checked against the corrected prediction (termed the expected values). Additionally, the upper and lower 95% confidence limits were calculated and this was cross checked by identifying the percentage of expected values lying outside of the upper and lower confidence limits. The performance for all 10 models is summarised in Table II.

Table II. IVS performance WSLB and WCLS The IVS assessment process is illustrated in Figures 7 and 8, which are example plots for models WSLB050 and WCLS055 respectively.

Neural network fatigue monitor

Figure 7. Model WSLB050 IVS performance

Figure 8. Model WSLS055 IVS performance

1109

1110

Steve Reed, Brian McCoubrey and Andy Mountfort

The outlier point evident in both the WSLB050 and WCLS055 cross plots (and all the other 8 plots) was investigated and identified as Sortie 4062239 (Figure 9). When compared with other sorties of similar target safe damage, this sortie was seen to have significantly higher numbers of g counts. The g counts seen were more typical of the expected values predicted by the model (see Table III). This would suggest that the target damage values for this sortie may be erroneous and it was recommended that this sortie should be investigated before being used in any further work.

Figure 9. Suspect data Sortie 4062239

Table III. Suspect data (4062239) compared with other target and expected values FCheck Assessment The models with linear regularisation were assessed further against the FCheck data set. This allows further confidence in the models’ performance and the linear regularisation and 95% confidence limit settings. The performance of the 10 models is summarised in Table IV and 2 examples are illustrated in Figures 9 and

Neural network fatigue monitor

1111

10. The performances have been identified before and after the application of linear regularisation and the percentage of expected values outside of the 95% confidence limits are identified. This value should be around 5%. It is also relevant to note that the expected values outside of the 95% confidence limits are evenly spread between over and under predictions.

Table IV. FCheck performance WSLB and WCLS

Figure 9. Model WSLB050 FCheck performance

1112

Steve Reed, Brian McCoubrey and Andy Mountfort

Figure 10. Model WCLS055 FCheck performance FCheck ‘Working Range’ Assessment The FCheck data sets were re-run through all 10 models but just using data in the range of 4FI/1000Hr to 12FI/1000Hr (8FI/1000Hr +/-4FI/1000Hr) to give an indication of performance for damaging but reasonably typical sorties. Obviously, this reduced the size of the data set further; for WSLB 60 sorties from the FCheck data set fell within this band and 49 sorties from the WCLS data were included. The damage ratios for the models are reproduced in Table V. All models returned predictions within 5% of the target values. Assessment Against Artificial ‘Extrapolated’ Sorties Although the training set covered a range close to 90FI/1000Hr there were relatively few very severe flights within the data. Furthermore, it is understood that the Tucano is now generally being operated more severely than was seen during the OLM data capture phase. Therefore, an assessment of the performance of the models was undertaken using artificially created sorties by concatenating g counts, landings and safe damage values for sorties of similar mid-sortie masses. It is accepted that this is not a precise method but the aim was to ensure that the models did not ‘fall over’ when presented with such data. Twenty five sorties were ‘created’ in this way and these ranged from 37 Æ 87FI/1000Hr for WSLB and 44 Æ 98FI for WCLS features. The damage ratios achieved for these data sets are reproduced in Table VI. These results show that the models continued to perform

Neural network fatigue monitor

1113

well even from these very severe artificial sorties with the range of predictions lying between +8.5% and -6.2%.

1 2 3 4 5 6 7 8 9 10

Model WSLB046 WSLB047 WSLB048 WSLB049 WSLB050 WCLS052 WCLS053 WCLS054 WCLS055 WCLS056

FCheck (Working) Damage Ratio 0.976 0.971 0.971 0.957 0.966 0.960 0.960 0.955 0.975 0.950

Sorties 60 60 60 60 60 49 49 49 49 49

Table V. FCheck working range assessment WSLB and WCLS

Model

1 2 3 4 5 6 7 8 9 10

WSLB046 WSLB047 WSLB048 WSLB049 WSLB050 WCLS052 WCLS053 WCLS054 WCLS055 WCLS056

Extrapolated Check (Artificial) Damage Ratio 1.052 1.047 1.058 1.085 1.060 0.979 0.970 1.006 0.978 0.938

Sorties

25 25 25 25 25 25 25 25 25 25

Range (FI/1000Hr) 37Æ87 37Æ87 37Æ87 37Æ87 37Æ87 44Æ98 44Æ98 44Æ98 44Æ98 44Æ98

Table VI. Extrapolated Check with artificial data assessment WSLB and WCLS Model Selection It was clear from the results presented in the various tests above that there was little to choose between the models. However, models WSLB050 and WCLS055 were selected for further analysis as they considered the best performing models overall.

Steve Reed, Brian McCoubrey and Andy Mountfort

1114

Model Sensitivities In addition to checks on the performance of the models, an indication of the sensitivity of the model to variations of the individual inputs was determined. The method used was to increment and decrement each individual input, using the FCheck data set and note the effect on the damage sum for the WSLB and WCLS features. The approach taken was to set the increment and decrement to credible values (i.e. one count for g inputs) but without taking the physics into account. Therefore, a count increment of the 5g window did not attract counts in the 4g, 3g and 2g windows. Additionally attention was paid to avoid incredible decrements; only sorties with 2 or more landings were decremented and only sorties with 2 or more g counts were decremented. Using this approach, the following increment and decrements were applied (Table VII). The only deviation from this approach was for the -1g increment and decrement for Model WSLB050 where the effect of the increment / decrement was very small in isolation and was slightly counter intuitive on average (i.e. a single count decrement caused a 1% increase in damage on average). As a cross check all the 5 short listed WSLB models was undertaken and the same effect was noted. Incrementing and decrementing by 2 counts showed a clear ‘intuitive’ effect however. Hence, to reflect reality the -1g increment and decrement was accompanied by a 0g increment and decrement for the WSLB model only (marked as *). It should also be noted that -2g count decrement was calculated from only 2 points but they produced very similar values in both models. The resultant damage ratios for all increment and decrement effects (compared with the FCheck data) are reproduced in Table VIII.

1 2 3 4 5 6 7 8 9

Input Total Landings -2g Counts -1g Counts 0g Counts 2g Counts 3g Counts 4g Counts 5g Counts Mid Sortie Mass

Increment +1 Count

Sorties 190

Decrement -1 (sorties 2 counts or more)

Sorties 92

+1 Count +1Count +1 Count +1 Count +1 Count +1 Count +1 Count +25 kg

190 190 190 190 190 190 190 190

-1 (sorties 2 counts or more) -1 (sorties 2 counts or more) -1 (sorties 2 counts or more) -1 (sorties 2 counts or more) -1 (sorties 2 counts or more) -1 (sorties 2 counts or more) -1 (sorties 2 counts or more) -25kg

2 51 107 180 148 108 35 190

Table VII. Sensitivity increment and decrement values (FCheck data)

Neural network fatigue monitor Model WSLB050 Increment / Damage Ratio Decrement (Inc or Dec / Original) Landings+1 1.008 Landings-1 0.993 -2g Plus 1 1.299 -2g Minus 1 0.793 -1g Plus 1* 1.008 -1g Minus 1* 0.995 0g Plus 1 1.036 0g Minus 1 0.972 2g Plus 1 1.016 2g Minus 1 0.985 3g Plus 1 1.015 3g Minus 1 0.988 4g Plus 1 1.089 4g Minus 1 0.936 5g Plus 1 1.186 5g Minus 1 0.898 MSM+25kg 1.018 MSM-25kg 0.982

1115 Model WCLS055 Increment / Damage Ratio Decrement (Inc or Dec / Original) Landings+1 1.009 Landings-1 0.991 -2g Plus 1 1.286 -2g Minus 1 0.824 -1g Plus 1 1.215 -1g Minus 1 0.971 0g Plus 1 1.031 0g Minus 1 0.977 2g Plus 1 1.009 2g Minus 1 0.992 3g Plus 1 1.028 3g Minus 1 0.977 4g Plus 1 1.091 4g Minus 1 0.936 5g Plus 1 1.150 5g Minus 1 0.919 MSM+25kg 1.022 MSM-25kg 0.977

Table VIII. Sensitivity increment and decrement damage effects (FCheck data) With the exception of the -1g isolated effects for the WSLB models, the remaining input showed similar and intuitive effects across the models. Generalised Solutions - Single Aircraft Models The models selected for the FMF replacement were derived using data from all 3 OLM aircraft in order to produce the most representative models. Ideally data from further aircraft would be available to check the generalised solutions produced. However, this is not the case and removing one or more aircraft’s data from the training set would run the risk of reducing the generalisation of the solution in a drive to verify this generalisation, which would not be acceptable. Hence, the approach taken was to verify generalisation by similarity. Models were generated using the same architecture and training parameters as the selected models but data from one aircraft in turn was used as the training set and the other 2 aircraft were used as the test set. This generated 6 models. All of which produced damage ratios within 10% of the target values. This was despite data from ZF512 having significantly less range of severity than the other 2 aircraft. An example plot from model WCLS_ZF406 is illustrated in Figure 11 and the results from all the models are reproduced in Table IX. Although not direct evidence of generalised models, this approach was considered to provide strong ‘circumstantial’ or in-direct evidence.

Steve Reed, Brian McCoubrey and Andy Mountfort

1116 Model WSLB_ZF212 WSLB_ZF406 WSLB_ZF512 WCLS_ZF212 WCLS_ZF406 WCLS_ZF512

Hidden Units 5 5 5 5 5 5

Epochs 1500 1500 1500 2500 2500 2500

Damage Ratio 1.031 1.032 0.910 1.026 1.052 0.901

R Train 0.981 0.985 0.976 0.987 0.989 0.986

R Test 0.967 0.969 0.960 0.978 0.980 0.972

Best Fit Line (Test) 1.030P-7.6E-3 0.939P+7.9E-3 1.140P-3.4E-2 0.982P+1.3E-3 0.956P+4.0E-3 1.190P-2.4E-2

Table IX. Summary of One-aircraft model performance

Figure 11. Example one-aircraft model (WCLS_ZF406) training performance Selected Model Performance on Entire Data Set As a final check and to provide data for validation and verification of the FORTRAN 77 formula code, the entire 867 sortie data set was run through the WSLB050 and WCLS055 models. The results are presented in Table X and graphically in Figures 12 and 13. It should be remembered that 294 of the 867 sorties in this data set were used to train the models. The average 95% confidence limit factor is the sum of the upper 95% confidence limit values divided by the expected prediction values. On average the ‘factor’ applied to data outside of the experience of the training set will be close to 1.5 – the current unmonitored factor used.

Neural network fatigue monitor Model

Damage Ratio

Error

WSLB050

0.994

-0.6%

WCLS055

1.001

+0.1%

1117 Correlation Coefficient (R) 0.978 0.986

Data outside 95% Conf Lim

Average 95% CL Factor

5.88%

1.49

5.88%

1.41

Table X. Summary of model performance on complete data set

Figure 12. Model WSLB050 – performance with 867 sortie data

1118

Steve Reed, Brian McCoubrey and Andy Mountfort

Figure 13. Model WCLS055 – performance with 867 sortie data set Testing Philosophy The WSLB050 and WCLS055 models were coded into a SHAUNN FORTRAN 77 subroutine to replace seamlessly the existing FMF. These models were developed within the training environment described earlier in this report. This environment was developed with the facility to output all interim calculations within the process. Therefore, training environment interim calculations and hand calculations were used to check each functional step in the SHAUNN subroutine. A range of test harnesses were used to check modules within the SHAUNN. Thereafter, end-to-end checks and independent checks against system requirements were carried out and documented.

PERFORMANCE AND IN-SERVICE ONGOING ASSESSMENT On average this SHAUNN model predicts fatigue damage accrual with around 1% of the target value for data not used to train the models. Although currently are no accuracy requirements for monitoring systems, a revision to Def Stan 00-970 [8] identifies that the monitor shall not underestimate the fatigue damage accumulation of a representative service usage spectrum by greater then 10%. The accuracy of the Tucano SHAUNN system falls well within this criteria. Furthermore, any inputs outside of the experience of the training data automatically attract an upper 95% confidence value. The additional factor of

Neural network fatigue monitor

1119

safety is variable but on the average of the test data set is greater than 1.4. This is very close to traditional unmonitored factors of 1.5. The SHAUNN is now entering service and ongoing monitoring of performance of the SHAUNN models is essential to provide continued confidence in the system. Hence a further review of the model performance will be undertaken once the most recently captured OLM data are available for comparison. Input data outside of the training experience are flagged automatically within the model and periodic reviews of the extent of data outside of the training experience will be undertaken. Performance warning levels will be set once the latest OLM data have been processed and the model performance under latest usage is determined.

ACKNOWLEDGEMENT The authors wish to acknowledge the foresight and pivotal role played by the MOD Training Aircraft Integrated Project Team and the MOD Aircraft Structural Integrity Branch in the development of this fatigue monitoring system.

REFERENCES [1] UK Ministry of Defence (2007), Defence Standard 00-970 - Design and Airworthiness Requirements for Service Aircraft, Part 1 Section 3 Issue 5. [2] McCoubrey, B. (2008), Tucano TMk1 Fatigue Meter Formula Revision, Bombardier Aerospace Report S312-FA-104, Issue 1. [3] UK Ministry of Defence (2007), Defence Standard 00-970 - Design and Airworthiness Requirements for Service Aircraft, Part 1 Section 3 Issue 5 Chapter 3.2.29 – 3.2.55 and Leaflet 42 - Structural Monitoring Systems Using NonAdaptive Prediction Methods. [4] Reed S C, (2007), Development of a Parametric-Based Indirect Aircraft Structural Usage Monitoring System using Artificial Neural Networks, The Aeronautical Journal of the Royal Aeronautical Society, April 2007. [5] Rumelhart D E, McClelland J L. (1986) Parallel Distributed Processing, Exploration in the Microstructure of Cognition, Volume 1: Foundations and Volume 2: Psychological and Biological Models. The MIT Press, Cambridge, Massachusetts, ISBN 0-262-18120-7 and 0-262-13218-4. [6] Beale R, Jackson T. (1990) Neural Computing: An Introduction, Institute of Physics Publishing, Bristol, UK, ISBN 0-85274-262-2. [7] Jepson B, Collins A and Evans A, (1993) Post-Neural Network Procedure to Determine Expected Prediction Values and their Confidence Limits, Neural Computing and Applications, Volume 1 Number 3, pp 224-228, Springer-Verlag London [8] UK Ministry of Defence (2008), Joint Airworthiness Committee JAC Paper No. 1337 Issue 1 – Revision of Defence Standard 00-970, Part 1, Section 3 and Leaflet 38 Fatigue – Service Monitoring.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

TOWARDS AUTOMATED FLIGHTMANEUVER-SPECIFIC FATIGUE ANALYSIS Juha Jylhä1, Marja Ruotsalainen1, Tuomo Salonen2, Harri Janhunen3, Tomi Viitanen3, Juho Vihonen1, and Ari Visa1 1

Tampere University of Technology Department of Signal Processing Korkeakoulunkatu 1, FIN-33720 Tampere, Finland Email: [email protected], [email protected], [email protected], [email protected] 2 Patria Aviation Oy Naulakatu 3, FIN-33100 Tampere, Finland Email: [email protected] 3 Technical Research Centre of Finland Vuorimiehentie 3, Espoo, Finland P.O. Box 1000, FIN-02044 VTT Email: [email protected], [email protected]

Abstract: Structural health monitoring of aircraft is a necessary topic especially with aging fleets. In general, it means tracking the structural integrity and keeping the risk of missing the detection of hazardous cracks low. Traditionally the fatigue analyses from the collected data have been carried out in a flight-by-flight basis. However, simultaneous good-quality recordings of the most relevant flight parameters with the calculated fatigue damage synchronized with the recordings enable more thorough analysis. Using an adequate collection of time stamped flight parameter signals, this paper aims at analyzing actions inside flight missions. We introduce a procedure for flight maneuver identification which allows maneuver-specific fatigue assessment when timestamped fatigue information is available. The presented damage distributions for F-18 aircraft demonstrate the new ability to consider the influence of various maneuvers on the fatigue of different structural details. M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1121–1134. © Springer Science+Business Media B.V. 2009

Juha Jylhä et al.

1122

INTRODUCTION Modern multi-role military aircraft experience a wide range of loadings during their service. Operational usage of a single aircraft can be today––or in the future– –significantly different from that assumed in the design phase. Because the knowledge of operational usage’s effects on the structure is an essential prerequisite for aircraft structural integrity management, at least a basic fatigue monitoring system fitted into an individual military aircraft is a de facto standard. There exist several approaches which operate with different accuracy and scope. The system may rely on a simple “g-counter”, or it may be realized as a comprehensive operational loads measurement [12] program. Our study relates to the F-18 Hornet aircraft structural integrity program of the Finnish Air Force (FINAF). A general overview of the program is given in [10]. Previously, a parameter based fatigue life analysis scheme has been developed in Finland [2], [11]. It provides location and flight specific fatigue life expenditure (FLE) estimates to every aircraft in the fleet supplementing the fatigue monitoring system of the original equipment manufacturer (OEM). Recent investigations aim to the determination of structural damage caused by individual flight maneuvers [7]. Incorporated with the fatigue life expenditure assessed in temporal windows of a few seconds, an ability to identify maneuvers allows fatigue tracking inside flight missions of an individual aircraft. However, manual flight maneuver identification (FMI) is an exhausting task [9]. Flight parameter recordings from a few hundreds of flights constitute an extensive database implying the use of data mining methods. Data mining methods have earlier proved to be useful, for example, in classification of helicopter maneuvers [6]. Expanding the earlier work in Finland [2], [11], and [9], this paper introduces a semi-automatic procedure for maneuverspecific FLE allocation. The procedure rests on data mining using a novel pattern identification algorithm [8]. Its good applicability to different maneuvers is verified by comparing the results of our automated procedure with those identified manually by an experienced analyst with pilot background. This work is aimed to support pilot training as well as mission planning. Fundamentally, any life expenditure which is not justified by operational or training objectives needs to be excluded. Presented maneuver-specific FLE distributions illustrate the potential of our procedure. Next we shortly outline the paper. The following section summarizes the F-18 fatigue monitoring efforts of the FINAF as a relevant background for discussed FLE analysis. Then we consider different approaches for FMI and introduce our procedure. At the end, the experiments are presented and discussed.

Towards automated flight-maneuver-specific fatigue

1123

SUPPLEMENTING F-18 FATIGUE ANALYSIS Since 2006 the FINAF has routinely been running the Hornet Operational Loads Measurement (HOLM) program. The goal is to quantify the effects of flights flown by the FINAF on the structure of the F-18 aircraft. The program is facilitating fleet management by the determination of fatigue damage rates of major structural assemblies in different flight training syllabi. The HOLM program employs two Boeing F-18C Hornet aircraft with identical onboard data acquisition system and instrumentation: 36 strain gauges have been fitted on globally and locally significant structural locations. The optimized sampling rates of the strains vary from 1280 Hz in the highly vibrating locations to 640 Hz elsewhere. In addition to strain signals, over 200 aircraft parameters from MIL-1553 bus are being recorded. Sampling rate of the flight parameters in the HOLM jets is appropriately 20 Hz. Thus besides the capability of reliable fatigue tracking, the HOLM system also provides an accurate aircraft state (of movement etc.). The HOLM instrumentation will be in routine use in the FINAF until the planned withdrawal date of the aircraft. Over a period of three years, more than 400 operational flights have been recorded and analyzed using the HOLM system. It has become evident that OEM’s fatigue tracking system (SAFE) is capable to provide sufficient fatigue information only in the vicinity of the center of gravity (CoG). Therefore SAFE’s left wing root FLE is indicator for structural life consumption of the FINAF Hornet fleet. The following results are based on that FLE value. This is illustrated in Figure 1 in which HOLM FLE versus SAFE FLE has been plotted. The linear correlation between the two is very good for bulkhead Y488.000. On the other extreme, vertical tail stub (VTStub) shows almost no correlation at all due to inadequate sampling rate of the SAFE strain gauge for this specific structure. Recent analyses have also been exposed that some of the fatigue-prone structural locations will not have adequate life due to present-day flying. Among these locations are VTStub and wing fold. The extensive strain gauge instrumentation of the HOLM system makes it indispensable, especially with certain structures not close to CoG such as empennage. Since the HOLM program contains only the two Hornets, methods to exploit the F-18’s standard memory unit (MU) data have been developed. For example, the fatigue analysis system [11] based on neural networks employs the F18 standard (by OEM) instrumentation. For this parameter based system, the HOLM is applied only to the neural network training and verification. The system is able to assess the flight-specific and aircraft-specific FLE at adequate level. At present, it has been implemented for several structural details otherwise infeasible to be monitored through the whole fleet.

Juha Jylhä et al.

1124

INNER WING AFT SPAR R2 = 0,535 6 000

HOLM_FLE

5 000

4 000

3 000

2 000

1 000

0 0,0000

0,0002

0,0004

0,0006

0,0008

0,0010

0,0012

0,0014

0,0012

0,0014

SAFE_FLE

VERTICAL TAIL STUB

INNER WING FORWARD SPAR

BULKHEAD Y488.000

2 800

1 400

1 400

R2 = 0,027

R2 = 0,760

R2 = 0,915 1 200

1 000

1 000

HOLM_FLE

1 600

800

1 200

600

800

400

400

200

0 0,0000

HOLM_FLE

1 200

2 000

HOLM_FLE

2 400

0,0002

0,0004

0,0006

0,0008

SAFE_FLE

0,0010

0,0012

0,0014

0 0,0000

800

600

400

200

0,0002

0,0004

0,0006

0,0008

SAFE_FLE

0,0010

0,0012

0,0014

0 0,0000

0,0002

0,0004

0,0006

0,0008

0,0010

SAFE_FLE

Figure 1. The linear correlation (shown as dashed line) between HOLM FLE and SAFE FLE in the selected locations of F-18 aircraft (each point indicates an individual flight). The further the measuring point is located from the center of gravity, the worse the correlation. Bulkhead Y488.000 shows good linear correlation (R2 = 0.915) whereas vertical tail stub shows hardly any correlation at all (R2 = 0.027). The solid line implies whether the OEM’s fatigue tracking system is adequately monitoring the location in question. The results above the solid line indicate that SAFE is not fully protecting that location and vice versa. To sum up, the HOLM program provides appropriate time-stamped strain data also for the structures under insufficient fatigue monitoring otherwise. Their fatigue tracking is enabled by the above-discussed system [11]. However, the inadequate life of certain problematic structural details requires investigations. Next we will discuss a procedure which focuses on the analysis of the reasons behind their FLE.

DATA MINING FOR MANEUVER-SPECIFIC ANALYSIS The accurate, time stamped strain gauge signals of the HOLM system combined with its good quality flight parameter recordings enable flight maneuver identification and thus fatigue analysis inside flights. Note that MU data expands the FMI to the every aircraft in the fleet, but without HOLM strain data, there should be some other way to produce time-stamped, accurate fatigue information. Flight parameter data has many characteristics that set special requirements for the FMI procedure. Firstly, flight parameter signals constitute a time series so temporal behavior of the signals must be considered. Secondly, the HOLM system is a

Towards automated flight-maneuver-specific fatigue

1125

multi-sensor instrumentation with a wide range of different sensors and thus the data consists of signals whose operational range and statistical properties vary. In addition, many signals are closely interconnected while representing the performed maneuvers together. Therefore flight parameter signals cannot be handled separately when identifying flight maneuvers. The FMI can be stated as recognition of patterns in multi-sensor time series data. The procedure of extracting hidden patterns from large amounts of data, such as the HOLM database, is called data mining. A comprehensive treatment of pattern recognition and data mining are given in the books [1], [3]. Next we will discuss three potential approaches for flight maneuver identification. After that, we will introduce our FMI procedure. At the end of the section, we will show illustrative examples of extracted flight maneuvers. Approaches for flight maneuver identification From the information engineering point of view, there exist three potential approaches that are applicable in FMI. These approaches could be called pattern detection, classification, and clustering. We will introduce them in more detail next. In the first approach, pattern detection, the analyst detects an interesting pattern (flight maneuver) in the flight data and wants to find all the maneuvers similar to it from the flight database. The pattern detection procedure must automatically adapt to new various flight maneuvers the analyst wants to find. This is a great challenge because one example pattern does not indicate itself how the similar enough patterns can deviate from it. One solution would be to use template-based pattern recognition as, for example, in [4]. This allows us to incorporate human expertise into the templates and thus get the information we cannot learn from a single example pattern. If the analyst wants to find multiple flight maneuvers simultaneously, a library of templates can be collected. In classification approach, the analyst collects, for example manually, a notable amount of instances of the certain flight maneuvers. A group of the instances of the same flight maneuver then represent a class. The most important thing to consider here is feature selection. So the analyst must find features that separate the maneuvers of the different classes. These features are then given, as training data, for some classification algorithm, which learns to separate the classes. This kind of approach has been successfully implemented with (car) driving events identification using hidden Markov models [5]. However, collecting the training set from flight parameter data denotes an excessive manual labeling work. Another drawback of this approach is that the classifier has to be tuned (or even taught again) every time when the analyst wants to find maneuvers belonging to a previously untaught class. In the third approach, clustering, the analyst has a group of maneuvers or flight segments and he lets a clustering algorithm to decide, which maneuvers are similar

1126

Juha Jylhä et al.

to each other. This approach is very challenging because the result depends strongly on how features (describing the relevant flight parameters and their behavior) and the distance measure (used to measure the similarity between maneuvers) are chosen. If designed well, this approach could, however, save a lot of manual work and afford new knowledge of maneuvers. Our flight maneuver identification procedure We have adopted a template-based pattern recognition approach for FMI. Our identification procedure comprises three steps: choosing, modeling, and identifying maneuvers, as Figure 2 shows. For every maneuver to be identified we build a template that we compare with flight parameter data––in other words, we measure the similarity between the template and patterns that exist in the data. We discuss the applied pattern identification algorithm more detail in [8], but next we describe its fundamentals concerning FMI. As the first step of the procedure, a representative instance of the maneuver is chosen from the flight parameter recordings by an experienced analyst. When choosing the maneuvers for modeling, the analyst must consider the time scale of the maneuvers as well as their sensibility and significance especially from the fatigue analysis point of view. An exemplary maneuver could be for example a purely flown loop. These kinds of questions require knowledge about the use, behavior, and the structures of the aircraft.

Figure 2. The block diagram of the flight maneuver identification procedure. At first, a representative instance of the maneuver is chosen. Then it is modeled; a template is generated. These two steps require manual adjustment by an analyst. The flight maneuver templates are used for extracting the same kind of maneuvers from flight parameter recordings in an automated manner. At the modeling step, so-called templates are built based on the chosen representative maneuvers. Because the amount of available flight parameters is high and only a few of them are relevant in modeling a specific maneuver,

Towards automated flight-maneuver-specific fatigue

1127

expertise is required in choosing the most descriptive flight parameters. The chosen parameters are quantized to three levels in such a way that after the quantization the parameters have only values -1, 0, and 1. The two boundaries that separate the three levels are manually adjusted. The quantization enables a consistent treatment of different parameters by allowing a convenient comparison of parameters’ values from a single time instant. In addition, it helps to overcome “shape” variations of the flight parameter signals during a specific maneuver. In order to “catch” the characteristic features of the certain maneuvers, quantization boundaries for the flight parameter signals are set template-specifically. The quantized, chosen flight parameters within the representative maneuver constitute the template. The last step of the procedure is fully automatic flight maneuver extraction from the flight parameter recording database. In this step, an identification algorithm based on the built templates is used to detect template-like maneuvers from the unanalyzed recordings. In reality, the same maneuvers can be performed in slightly different ways, and their duration can vary. To cope with this, the dynamic time warping (DTW) algorithm is used to handle the temporal variations. Because the DTW warps the flight parameter signals nonlinearly, we are able to find, for example, a flight maneuver (the pattern) that is performed two times faster at the beginning and three times slower at the end than the representative maneuver (the template). A DTW matrix is calculated between the template and the whole flight using the quantized flight parameters that were chosen at the modeling step. Beginnings and ends of the maneuvers (the patterns) are detected from the DTW matrix which provides similarity values for the patterns. Because multiple patterns are detected from a single DTW matrix, the identification algorithm is computationally efficient.

Juha Jylhä et al.

1128

(a) Flight parameters of the template

(b) Flight parameters of pattern I

(c) Template

(d) Pattern I, difference 0

(e) Pattern II, difference 1.6

(f) Pattern III, difference 9.8

Figure 3. The naive split-s template (c) is produced using one exemplary instance with three recorded F-18 flight parameters (a): Mach number (IAMACH), pitch angle (INPCTH), and roll rate (ICARRT). The pitch signal has two quantization boundaries which are indicated with horizontal lines in (a) and (b). Notice the correspondence between the parameters and the patterns. Gray scale indicates quantization of the flight parameter signals: white = 1, gray = 0, and black = -1. Three patterns I–III show the robustness of the identification algorithm. Because of its high difference value, pattern III would not be identified as split-s. Note the significance of the lacking roll movement at the beginning of the maneuver. [8]

Towards automated flight-maneuver-specific fatigue

1129

Illustration As an example, Figure 3 illustrates identification of a split-s maneuver using a simple and illustrative but still functional template. Notice that this split-s template is not the same as the one used in subsequent section. In the beginning of split-s there may be left or right roll movement as seen in Figure 3 (a) and (b). The Mach number is included in the template because it is reasonable to separate the maneuvers with different flow regimes regarding structural loads. Figure 3 (c) shows the template and (d), (e), and (f) three illustrative patterns denoted by I, II, and III. Time variations can clearly be seen. The pattern I represents a clean split-s, i.e. similar to the template regarding the FMI procedure. In the case of the pattern II, there are some minor disturbances in the pitch signal caused by the measurement system. The disturbances cause the difference growing to 1.6 which is low and the pattern is still identified as split-s. The FMI procedure is robust to minor disturbances in the signals. Maneuver III is actually not a split-s because of a missing roll action at the beginning of the maneuver. Its difference value 9.8 is high, and it is not identified as split-s. [8]

RESULTS Our automated FMI makes it feasible to extract maneuvers from a large amount of flight parameter data. Using good-quality, time-stamped FLE estimates based on HOLM strain data, we can derive maneuver-specific FLE distributions. The FMI method has been tested to extract maneuvers from 298 HOLM flights of the FINAF F-18 aircraft. In the first test, two maneuvers, fast roll and split-s, were searched and their fatigue damage was calculated for two critical locations: wing fold and vertical tail stub. In the second experiment, a certain highly damaging maneuvers were indentified. The flight maneuver specific damages, presented in this paper, are calculated in two comparable ways. It’s estimated using stress spectrum during the maneuver and feeding the spectrum to crack initiation software or monitoring cumulative damage of opening and closing cycles during the maneuver––subsequent results with the former method and the consideration after that with the latter method. Split-s and fast roll maneuvers For split-s and fast roll, the representative maneuvers were chosen based on their high FLE for the wing fold and the VTStub. Templates were well-adjusted and all the maneuver identifications seemed to be correct. Possible missed detections were not considered. All purely flown, 78 roll instances and 61 split-s instances were extracted from 298 flights. The damage scatter and damage histograms are illustrated in Figure 4 and Figure 5, respectively. Note that both these maneuvers represent rigorously defined type of split-s and roll. One may find huge amount of, for example, loosely defined roll maneuvers from 298 flights.

1130

Juha Jylhä et al.

Maneuver-specific FLE distributions that have high variance are interesting because it might be possible to avoid performing the related maneuvers in the way that consumes FLE the most. Particularly, savings may be substantial when reducing the average FLE of certain frequent maneuvers. But on the other hand, the maneuvers that cause severe damage in general, may be considered to be avoided or at least to be performed less frequently if allowed regarding training and operational aspects. The result in Figure 5 shows that the split-s causes diverse damage depending on the flying style. Hence, the pilot can be conferred to perform the split-s maneuvers in a fatigue-friendly way. For this specific maneuver, the angle of attack is the key factor behind the damage. Highly damaging split-s maneuver As shown in Figure 1, the vertical tail stub is one of the most problematic structural details in the F-18 as its lifetime can not be anticipated to be sufficient for the whole life cycle of the aircraft. Thus the VTStub is an inherent choice for a profound experiment. Here the representative maneuver is searched and the template is build from different point of view than discussed earlier. The investigation initiated by screening the most damaging flights from the HOLM database. The maneuver that caused the highest FLE of the VTStub was identified and verified by the flight visualization tool to be a split-s. This was selected as a representative maneuver––to be converted into a template for the pattern recognition. The aim was to extract, not all the split-s maneuvers, but instead only the corresponding ones producing high damage. The template parametrization required analyst experience and engineering judgment, but with the ready template, the rest of the FMI procedure was automated and fast. Notice that this template for split-s is not the same that was considered in Figure 3. The first batch run for 298 flights took about 50 minutes with a desktop computer and resulted 162 indications out of 77 flights. The flight visualization tool exposed that five of the indications altogether were not split-s maneuvers. Next, the template was fine-tuned such that the maneuvers of low damage rate were to be excluded. Also the difference threshold was increased. This time the automated FMI tool found 107 indications out of 51 flights. Although building the template in some cases can be a lengthy task, the automated identification routine pays for itself in the long run.

Towards automated flight-maneuver-specific fatigue

1131

Figure 4. Damage scatter of fast roll and split-s maneuvers. Wing fold damage on the top figure and vertical tail stub damage on the bottom. Damage is normalized with the maneuver instance having the maximum damage. See corresponding histograms in Figure 5.

Figure 5. Damage histograms of fast-roll and split-s maneuvers. Wing fold histogram on the top figure and VTStub histogram on the bottom. Histograms are normalized with the number of instances. Damage is normalized with the maneuver instance having the maximum damage. See corresponding damage scatter plot in Figure 4. The roll maneuver produces more deviation in the FLE of the both structural details. The split-s maneuver is clearly more damaging than the roll, expecially for the VTStub.

Juha Jylhä et al.

1132

This profound experiment is at the moment in its preliminary phase, and so we do not present corresponding distributions as above. However, the first observations prove that our FMI procedure can be applied to identifying differently damaging maneuvers. The split-s maneuvers found with the latter template has notably lower damage expectation value compared with the former experiment, but the shape of the distributions is the same. Consequently, with sophisticated template fitting, the procedure has even been able to roughly separate the most damaging, “hard flying” split-s incidents from nominally very similar incidents of lower damage rate.

DISCUSSION Important properties of the presented data mining of the flight parameter recordings include the robust and consistent handling of multisensor data and computationally efficient method for comparison of templates and patterns despite their temporal variations. They make the FMI procedure flexible, reliable, and fast. FMI was tested with different maneuver types. The extraction of maneuvers from the HOLM database turned out to be an iterative process because the results are (naturally) highly dependent on the quantization boundaries of the descriptive flight parameters. The tighter the boundaries, the more ideal maneuvers were found (closely similar to the representative one). In addition, the engineering judgment was required for selecting the appropriate cost threshold value. See [8] for more details about the procedure parametrization. The demonstrated flight-maneuver-specific fatigue allocation is semi-automatic as depicted in Figure 2. Although the procedure requires manual adjustment of the template for each maneuver-type to be analyzed, it has been found convenient for analyzing the F-18 fatigue critical structural details using large amounts of flight recordings. Importantly, the procedure has been applicable for all the tested maneuver types. This in-depth fatigue allocation is applicable only for structural details whose fatigue can be tracked appropriately within the maneuvers, i.e. in temporal windows of a few seconds. For the present, this limits the use of the procedure to HOLM data regarding the most of the critical structural details. Nevertheless, also the quality of the MU data has proven adequate for the FMI purpose (not for indepth fatigue tracking). Thus all the F-18 flights can be segmented into maneuver categories for various flight analysis purposes. Future efforts are focused on determining the reasons behind the scatter of damage with certain maneuvers. It implies reasoning of the connection between flight parameters’ temporal behavior and fatigue life expended.

Towards automated flight-maneuver-specific fatigue

1133

CONCLUSION Previously, the fatigue monitoring has contributed fleet management in flight-byflight basis or even in a more general level. There have not been appropriate tools for inside-mission fatigue analysis. The workload in manual flight maneuver identification from flight parameter recordings has been excessive. This paper presented an automated data mining procedure for extracting maneuvers from the recordings. Combined with proper fatigue tracking ability, the procedure allows flight-maneuver-specific structural damage allocation. This new analysis capability supports fleet management by providing detailed information about flying, for example, for adjusting flight training syllabi.

ACKNOWLEDGMENT The authors would like to thank the Finnish Air Force and the Nokia Foundation for funding and support.

REFERENCES [1]

Duda, R., Hart, P., and Stork, D, (2001). Pattern classification, Second edition, Wiley, New York.

[2]

Jylhä, J., Vihonen, J., Ala-Kleemola, T., Kerminen, R., Tikka, J., and Visa, A. (2007). An article (“Modeling structural vibrations for automated aircraft fatigue monitoring”), in: Durability and Damage Tolerance of Aircraft Structures: Metals vs. Composites, Proceedings of the 24th ICAF Symposium, vol. II, pp. 1031–1042, Lazzeri, L. and Salvetti, A. (Eds.), Naples.

[3]

Han, J. and Kamber, M. (2006). Data mining: Concepts and techniques, Second edition, Morgan Kaufmann, San Francisco.

[4]

Harada, L. (2004). An article (“Detection of complex temporal patterns over data streams”), Information Systems, vol. 29, n. 6, pp. 439–459.

[5]

Mitrovic, D. (2005). An article (“Reliable method for driving events recognition”), IEEE Transactions on Intelligent Transportation Systems, vol. 6, n. 2, pp. 198–205.

[6]

Oza, N. C., Tumer, K., Tumer, I. Y., and Huff, E. M. (2003). An article (“Classification of Aircraft Maneuvers for Fault Detection”), in book: Multiple Classifier Systems, vol. 2709/2003, p. 160, Springer, Berlin.

1134

Juha Jylhä et al.

[7]

RTO-TR-045 (2002). Design Loads for Future Aircraft, Sections 3.2.2– 3.2.4, RTO/NATO, France.

[8]

Ruotsalainen, M., Jylhä, J., Vihonen, J., and Visa, A. (2009). An article (“A novel algorithm for identifying patterns from multisensor time series”), in Proceedings of the 2009 World Congress on Computer Science and Information Engineering (CSIE 2009), IEEE, Los Angeles.

[9]

Siljander, A. (2007). A report (“A Review of aeronautical fatigue investigations in Finland May 2005 – April 2007 (A. Siljander, Ed.). Presented at the 30th Conference of the International Committee on Aeronautical Fatigue (ICAF), Naples, ICAF Doc. 2410.

[10]

Siljander, A. (2009). A report (“A Review of aeronautical fatigue investigations in Finland May 2007 – April 2009 (A. Siljander, Ed.). Presented at the 31st Conference of the International Committee on Aeronautical Fatigue (ICAF), Rotterdam, ICAF Doc. 2418.

[11]

Tikka, J. and Salonen, T. (2007). An article (“Parameter based fatigue life analysis for F-18 aircraft”), in: Durability and Damage Tolerance of Aircraft Structures: Metals vs. Composites, Proceedings of the 24th ICAF Symposium, vol. I, pp. 412–426, Lazzeri, L. and Salvetti, A. (Eds.), Naples.

[12]

Viitanen, T., Koski, K., Bäckström, M., Voutilainen, E., Lahtinen, R., and Siljander, A. (2005). An article (“The OLM database as a tool to sort particular data sets from a bulk of data”), in: Proceedings of the 23th ICAF Symposium, vol. II, pp. 513–540, EMAS Ltd., Hamburg.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

SERVICE HISTORY ANALYSIS AND TEARDOWN EVIDENCE – KEY ELEMENTS FOR STRUCTURAL USAGE MONITORING OF AN AGEING FLEET K. A. Lucas and M. J. Duffield QinetiQ, Farnborough, UK

Abstract: Various elements are combined in this paper to demonstrate a credible basis for structural usage monitoring of the mixed fleet of ageing VC10 aircraft. Practical teardown findings are used to substantiate the theoretical assessment of the aircraft in terms of historic and current flight profiles. Problems with limited or incomplete records for early usage of the VC10 aircraft are overcome to account for both commercial and military service. The results of all of these elements in combination with fatigue and damage tolerance calculations are presented. An approach developed using NzW principles is shown to provide a reliable basis for structural usage monitoring of the VC10 fleet. From the combined results, a Limit of Validity (LOV) is derived for the critical wing root joint on each aircraft. This represents the point at which a programme of inspection and bolt replacements would be necessary to satisfy damage tolerance requirements. Consequently it is possible to determine the remaining margins with respect to the LOV and ensure that they are sufficient to enable the expected retirement dates to be achieved within those limits.

INTRODUCTION The VC10 has been in continuous service with the Royal Air Force (RAF) for over 40 years. It is required to remain in service for several years to come until the planned replacement by a new fleet of modern tanker/transport aircraft. This requirement dictates that structural usage monitoring of the fleet is essential, which M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1135–1153. © Springer Science+Business Media B.V. 2009

1136

K.A. Lucas and M.J. Duffield

itself relies on a proper understanding of the histories of the individual aircraft since entry into service. Therefore, the basis for structural usage monitoring is a comprehensive record of flying data for each aircraft. These records may be used to generate estimates of fatigue loads spectra which can then be analysed in terms of fatigue and damage tolerance criteria. There are currently three versions of the VC10 in service with the RAF and each version has a distinctly different typical history. The CMk1 aircraft were built as multi-role transport aircraft for the RAF and were upgraded to an additional tanker capability and re-designated CMk1K during the mid-1990s; the KMk3 aircraft (Figure 1) served with a commercial operator during the 1970s before being converted to 3-point Air-to-Air Refuelling (AAR) tankers for the RAF and entering RAF service during the early 1980s; the KMk4 aircraft served with another commercial operator for many years and were eventually converted to 3-point AAR tankers for the RAF, albeit to a different overall configuration to the KMk3, during the mid-1990s. Whilst the service records from 1991 onwards of all aircraft are complete and available in an easily interrogated database, the earlier service histories are limited to basic and, usually, incomplete information. This means that it is not possible to easily or directly determine the total value of the service histories of the aircraft in terms of structural usage.

Figure 1. VC10 KMk3 in flight As reported in the UK National Review at the 2007 ICAF Conference [1] a structural monitor using the NzW method (where Nz = normal acceleration or ‘g’ and W = nominal aircraft mass) was developed for the CMk1K. This methodology can also be applied to the KMk3 and KMk4 aircraft. However, in order to quantify the historic usage of the aircraft, additional work is needed to generate a baseline for the monitoring programme. This involves a comprehensive analysis of the available information for the different variants, including the KMk2 aircraft which had been retired from service during the late 1990s. The KMk2 aircraft were of particular interest because they had been used to provide a large number of samples taken from the wing structure for the purposes of teardown inspection – as described in the UK National Reviews at the 2001 and 2003 ICAF conferences [2-3]. The results from those inspections were combined with the analysis of the service histories of the donor airframes in order to validate the fatigue calculations for the other variants of the aircraft remaining in service.

Service history analysis and teardown

1137

The purpose of this paper is to demonstrate how the various elements of the theoretical and practical assessments of the aircraft in terms of historic and current flight profiles, analysis of ‘g’ spectra, teardown findings and fatigue and damage tolerance calculations were combined to provide a credible basis for structural usage monitoring of a varied fleet of large, ageing aircraft. It will be shown how the various assumptions underpinning the assessments were validated and how the results have been used to demonstrate that the structural integrity of the critical joint features on the VC10 wing can be assured until the planned retirement from service at an average age of over 40 years.

THE CRITICAL WING JOINT In order to bring the VC10 structural maintenance philosophy in-line with modern criteria used to assess ageing jet transport structures in the civil world, the aircraft was re-qualified on the basis of damage tolerance. Numerous difficulties are associated with the systematic application of such criteria to a design based on the fail-safe principle of the late 1950s and early 1960s and several instances arose where the damage tolerance analyses proved incapable of providing a rational structural inspection programme. This was largely due to the absence of any reliable benchmark by which to qualify the fundamental assumptions contained within the analyses. Structural elements were also considered with respect to multiple site damage (MSD). The wing lower surface skin joint at the root, or Rib 0 as it is commonly known, is a typical example of a design that may be susceptible to MSD. This joint comprises abutted inboard and outboard skin panel tongues connected by a series of inner and outer joint plates. The assembly is secured by a staggered double row of steel taper bolts either side of the joint and a typical section is shown in Figure 2.

UP ↑

OUTBOARD → Figure 2. VC10 lower wing skin joint at Rib 0 – typical section

K.A. Lucas and M.J. Duffield

1138

This joint configuration is very efficient because the fasteners are working in double shear and there are no off-sets to generate undesirable bending forces; however, the inner and outer joint plates render the critical section of the skin panels uninspectable except by directed NDT techniques. With similar stress levels across the entire chord, the joint is a classic example of a design feature where the threat of MSD must be considered. The fatigue and damage tolerance analysis of the Rib 0 joint resulted in calculated critical crack lengths that were very small. The necessary inspections would require the bolts to be removed and for high frequency eddy current probes to be used to search for evidence of fatigue cracking in the skin tongues. In total, there are 344 bolts in each joint, or 688 per aircraft. Clearly, to undertake such an inspection on a regular basis would be a major task with significant effects on maintenance costs and downtime. The extraction of the taper bolts is likely to cause damage to the hole bores leading to a requirement for reworking of the holes and the provision of an equivalent number of non-standard replacement bolts. Recovery of the joints after inspection is also likely to engender problems in fuel tank sealing; experience has shown that the joints tend to spring apart when adjacent bolts are removed and such movement may be difficult to recover in the un-jigged configuration of the aircraft during maintenance. Furthermore, the entire KMk4 fleet had exceeded the calculated inspection threshold life whilst the KMk3s were close to doing so. Hence, the inspection requirements, if applied rigorously, threatened to compromise the availability of a significant proportion of the RAF AAR fleet. The analytical results were necessarily conservative because there was no test evidence available to qualify the assumptions contained in the fatigue and damage tolerance calculations. However, all of the KMk2 aircraft had also exceeded the calculated threshold inspection life by a margin in excess of 50% with total recorded flying hours (FH) of between 50,768 and 52,818 across the five airframes. Consequently, it was argued that, if the joints on these aircraft could be demonstrated as free from any significant defects via teardown analysis, then this margin could be interpreted as an alleviation factor on the calculated lives. As the structure is identical on all variants and the fatigue calculations for all variants used exactly the same analysis models and equivalent loads and stress data, it was also concluded that this factor could be applied with confidence to the entire VC10 fleet.

TEARDOWN The decision to retire the five fleet leading VC10 KMk2 aircraft in 2000 came at an opportune moment and sections of the chordwise Rib 0 joint were recovered from three of the redundant airframes. It was not considered necessary to dismantle

Service history analysis and teardown

1139

the entire joint from each wing of the three aircraft but to undertake the examination on a sampling basis both to save time and to ensure that the scope of the investigation encompassed as many aircraft as possible. The selection of the sample sections ensured that at least two examples of every bolt location would be examined in detail with up to six examples in the areas of particular interest, such as the intermediate spar run-out locations. The Rib 0 teardown work has been previously reported to the ICAF Conference in 2001 [2] and 2003 [3] but is summarised here for convenience. Each sample underwent systematic dismantling and examination by visual, NDT techniques and, ultimately, laboratory analysis by sectioning. The initial inspection techniques applied were those that would be used in service; a total of 1,106 locations were subject to bolt extraction, assisted visual assessment of the hole bores and inspection using rotating head high frequency eddy current (RHFEC) probes. Then, with the joints dismantled, the initial results (including nil findings) were corroborated by further visual and NDT examinations, including the use of ultrasonic and dye penetrant techniques. Finally, any defects or defect indications resulting from these phased inspections were subject to laboratory analysis by sectioning and microscopy techniques, both optical and using the scanning electron microscope (SEM).

Figure 3. Typical section of Rib 0 joint with the outer joint plates disassembled Only one hole was found to contain a defect that could not be attributed to damage during installation or removal of the bolts. This defect was found in the outer joint

K.A. Lucas and M.J. Duffield

1140

plate at the bolt head end of the hole. The analysis of this defect confirmed that it was a crack that had propagated under the influence of fatigue loading, having reached a size of 1mm x 2mm (Figure 4). A similar crack (albeit much smaller and undetected by RHFEC) was discovered in an adjacent hole. In both cases the initiation of the cracks may have been influenced by the presence of corrosion. Furthermore, both holes revealed evidence of very small multiple cracks within the bores and, again, these appeared to be associated with surface corrosion effects.

(a)

(b)

20mm

0.5mm

Figure 4. (a) Cracking found in the outer joint plate – strained open and (b) detail of the fracture surface obtained by SEM The general condition of the affected joint plate was no different from any of the other samples obtained from the three aircraft. Consequently, it was suggested that the type of cracks evident in the hole bores might be indicative of a widespread phenomenon. To investigate this possibility, several more holes were examined for evidence of similar deterioration, especially within the main skin planks. This additional work established that there was no evidence of the type of cracking observed in the joint plate at the critical section of the skin panels. In summary, over 1100 of the critical bolt holes were examined in samples from the three airframes with no significant findings of fatigue. This demonstrated that the original fatigue threshold calculations were indeed conservative by a

Service history analysis and teardown

1141

significant degree. Taking this into account, the typical achieved lives of about 52000 FH for the KMk2 teardown airframes showed that the true threshold points were unlikely to be reached by any of the remaining aircraft. This meant that plans for the precautionary refurbishment of the joints by replacing the bolts with oversize equivalents in reworked holes (688 steel taper bolts per aircraft) were not immediately necessary and a considerable maintenance burden could be postponed. It was now necessary to understand these results with respect to the airframe’s historic and current usage. This would enable the teardown evidence to be used to validate an analysis process for monitoring fatigue consumption of the remaining in-service variants against the benchmark provided by the KMk2 fleet.

THE NzW APPROACH FOR RIB 0 Fatigue monitoring of the VC10 KMk3/KMk4 fleet is accomplished by use of the well-proven Fatigue Meter Formula (FMF) methodology. This empirical formula is used to process flight data and is based on the Rib 19 location for the VC10 wing; it is not applicable to the Rib 0 joint. In addition, there is no FMF available for the CMk1K fleet and, historically, structural usage was simply monitored on the basis of flying hours and landings accrued. It has been recognised that, given the evolving RAF usage patterns, the overall age of the VC10 fleet and the adoption of damage tolerance as the primary structural qualification criterion, this approach may not be adequate to assure fully structural airworthiness up to the expected retirement date. Consequently, an alternative approach for structural usage monitoring specific to the Rib 0 location was required to make use of the information routinely recorded on a flight-by-flight basis. The approach combines the key parameters of aircraft mass, sortie profile and ‘g’ spectra with the results from the finite element models used for the damage tolerance qualification of the primary structure. By working in close cooperation with the Design Organisation for the VC10, an analysis process based on NzW principles was generated, ensuring that no further data capture requirements were introduced. The analysis can be undertaken at any convenient time according to the demands of the Airworthiness Authority for the aircraft. The principal elements of the analysis process are the sortie profile code (SPC), the total mass of the aircraft at start (typically referred to as the Ramp Weight (W)) and the maximum ‘g’ level (or normal acceleration, Nz) recorded for the flight. For this reason, the method is generally referred to as NzW. The SPC describes the type of flying undertaken and provides the link back to the finite element model results which are the source of a defined set of stresses. One set of stresses was provided for each of the reference SPCs forming the basis of the corresponding fatigue calculations. These calculations combine the product of Nz times W with

1142

K.A. Lucas and M.J. Duffield

the reference stresses and an appropriate fatigue endurance (s-N) curve to generate an estimate of fatigue damage, flight-by-flight. The main element of the NzW calculation is the damage due to the Ground to Air Cycle (GTAC) and various corrections are applied to account for additional sources of damage such as the gust and manoeuvre spectrum and intermediate / roller landings. There are also some sortie types which are outside the reference SPCs and these are accounted for by a simple adjustment according to the number of flights. Whilst a little crude, the number of such sorties is typically very small and has only a minor effect on the overall estimate of fatigue damage. In addition, a set of factors is applied to account for a number of sources of uncertainty in the analysis, i.e. - The test pieces used to generate the s-N curve were made from 2024T351 whilst the VC10 wing joints are made from 24ST4. - The test pieces were assembled with titanium taper bolts whilst the VC10 uses steel taper bolts. - There was no test evidence to validate the s-N curve (i.e. aircraft structure behaviour versus test coupons). Finally, the conventional factors accounting for the likely scatter in fatigue performance were applied to ensure conservatism, i.e. - 3⅓ scatter factor on fatigue life of nominally identical aluminium alloy wing structures subject to similar load spectra. - 1½ for variation in actual loads versus the assumptions used for the analysis of sortie profiles. Having developed this NzW approach, it is possible to monitor the rate of fatigue damage accrual at the Rib 0 joint on VC10 aeroplanes using the flight information now routinely recorded for each sortie. In order to make use of the teardown evidence it is necessary to understand the historic usage of the fleet and apply the same NzW principles to estimate the total level of damage accrued.

SERVICE HISTORY ANALYSIS Each variant of the VC10 has a different service history. Whilst the CMk1(K) aircraft have been in RAF service since their conception, the KMk2, KMk3 and KMk4 variants were converted for RAF use following commercial service with different airlines (BOAC/BA/Gulf Air, East African Airways (EAA) and BOAC/BA, respectively).

Service history analysis and teardown

1143

Flight data for all variants have been fully recorded in a database since 1991, but prior to that, varying levels of information are available regarding the usage of each fleet. In particular, the data recording the commercial use of the KMk2/KMk3/KMk4 airframes are limited to the numbers of flying hours and landings achieved. The first VC10 CMk1 entered service with the RAF in 1966, so, whilst there are extensive records of the most recent usage, the earliest 25 years of history are poorly recorded. Limited data on the numbers of landings and flying hours accrued by the fleet prior to 1991 have been obtained from a review of all the historic information held in the available contemporaneous records. However, no information on typical operating weights, sortie profiles or ‘g’ spectra has been found. For UK military aircraft such information is nowadays published within the Statement of Operating Intent (SOI) which was introduced as an important element of structural integrity management during the 1980s. The SOI provides a detailed description of the way in which an aircraft type or variant is expected to be typically operated by the RAF. It is not a prescriptive document as it does not limit the operation of the aircraft in any way; however, the information provided allows the designer to assess the typical usage in terms of structural fatigue consumption versus the baseline fatigue life. The SOI allows that baseline, which is determined at the design stage and verified by analysis and testing, to be interpreted and exploited in terms of actual usage. The earliest versions of the SOI for the VC10 were released in 1987 following the introduction of the KMk2 and KMk3 AAR tanker variants. This was in recognition of the very different usage profiles of the tankers versus the air transport primary role of the CMk1s which was fundamentally the same as the commercial airlines. Since then, the SOI has been regularly revised to reflect evolving usage patterns, especially with the conversion of the CMk1 aircraft to the CMk1K standard. The document is now known at the Statement of Operating Intent and Usage (SOIU) as it includes statistical data on previous actual usage as well as details of the expected future usage. The current SOIU contains twenty different SPCs grouped into four categories; SPC 1 – 4, Crew Training (CT); SPC 11 – 14, Route Flying or Air Transport (AT); SPC 21 – 29, Air-to-Air Refuelling Operations (AAR) and SPC 31 – 33, Miscellaneous. SPC1, 11, 12, 13, 21, 22 and 23 are the reference SPCs previously referred to; typically within the NzW analysis, these account for in excess of 90% of the aircraft usage. Details of the process used to model the historic usage of the CMk1/CMk1K fleet have previously been described in some detail [4]. The same concept has been applied to the AAR tanker airframes with the added challenge of having to account

K.A. Lucas and M.J. Duffield

1144

for the commercial usage prior to conversion to the tanker configuration for the RAF. The following summary of the necessary analysis is presented for the specific case of the ex-BOAC/BA Super VC10 aircraft which became the VC10 KMk4 in RAF service after tanker conversion. The same approach was also applied to the exBOAC/BA/Gulf Air Standard VC10s (subsequently KMk2s in RAF service, i.e. the teardown airframes). In both cases the available usage data were analysed using NzW principles to estimate the corresponding fatigue damage accrual. These estimates provide the basis for quantifying the evidence from the KMk2 teardown programme such that NzW can be used to assure structural integrity up to the planned retirement date of each KMk4 airframe. Commercial usage For both the Standard and Super VC10 variants only very limited information is available regarding the routes flown by the commercial operators. However, for all airframes, the total numbers of FH and landings are known, with landings being divided between full-stop landings (FSL) and “rollers”, otherwise known as “touch-and-go” landings. These data provide sufficient information upon which to build a reasonable description of the commercial usage of the aircraft which can be assessed by NzW principles to estimate the value of that usage in terms of accrued fatigue damage. The starting point for this assessment is the presumption that the commercial usage of the VC10 can be modelled using the results from the fatigue analysis of the AT and CT SPCs for the CMk1 variant in RAF service. The wing root structure is the same for all variants and it is assumed that the operating weights routinely flown by BOAC/BA for the Super VC10 were similar to those achieved by the RAF CMk1 in the AT role. This assumption is reasonable due to the marginal differences in design weights between the two variants. Details of the typical sortie durations, numbers of FSL and roller landings and the proportion of flying in each of the relevant SPCs were obtained from the SOI and are shown in Table I. SPC CT (Crew Training) Subtotal AT (Air Transport) Other

1 2 3 11 12 13

% of Flying Hours 8.4 1.7 0.7 10.8 43.8 29.5 15.5 0.4

Duration (mins.)

Sorties / 1000FH

195 240 175 460 280 95 -

25.8 4.3 2.4 32.5 57.1 63.2 97.9 1.7

Rollers / sortie 12 6 0 -

FSL / sortie 2 2 1 -

Rollers / 1000FH 310.2 25.5 0 336 -

Table I. RAF aircraft usage patterns from the CMk1 SOI

FSL / 1000FH 51.7 8.5 2.4 63 -

Service history analysis and teardown

1145

Overall flying hours and landing data for each variant were used in combination with the typical SOI profile data (Table I) in order to estimate the usage mix in terms of CT and AT sorties. The five BOAC/BA Super VC10 aircraft that were subsequently converted to KMk4 AAR tankers accrued a combined total of 71,186 landings, comprising 62208 FSL and 8,978 rollers, over a commercial service life of 235,718 FH. It was assumed that all of the rollers flown in commercial service were associated with training flights. Using the average number of rollers per CT sortie derived from the SOI, the number of training sorties in airline service was estimated along with the associated number of full stop landings. The remainder of full stop landings was assumed to correspond to the number of commercial revenue flights. Referring to Table I and using the combined totals of landings and hours for the five subject aircraft, the estimated number of commercial training flights was calculated: 8,978 x 32.5 / 336 = 868 commercial training flights (1) The corresponding number of FSLs for commercial training flights is: 868 x 63 / 32.5 = 1683 FSL

(2)

And the FH for commercial training flights is estimated as: (10.8 /100) x 868 x (1000 / 32.5) = 2,884 FH

(3)

Subtracting these estimates from the previously mentioned combined totals yields the following statistics describing the revenue flights achieved in commercial service by the five aircraft: Revenue FH = 235,718 – 2,884 = 232,834 (4) Revenue FSL = 62,208 – 1,683 = 60,525 (5) Assuming one FSL per revenue flight, the typical duration for revenue flights is estimated as: 232,834 / 60,525 = 3.85 Hr. = 230.8 mins. (6) As part of the damage tolerance calculations at the Rib 0 location, fatigue stress data appropriate to each reference SPC in the 1993 SOI were generated from the structural finite element models for the CMk1. These reference stresses were used to calculate a nominal fatigue life for the joint. When adjusted for all of the necessary allowances previously described the results are expressed as a Limit of Validity (LOV) for each SPC; the calculated LOVs are those that would apply for 100% usage of the aircraft in that particular SPC and represent the basis for the interpretation of the teardown evidence (as will be shown in a later section of this paper.)

1146

K.A. Lucas and M.J. Duffield

The reference stresses for the CMK1 are considered generally applicable to the Standard and Super VC10 variants in airline service because the overall configurations were the same and it is assumed that commercial flights were similar to the RAF AT SPCs in terms of weight and duration. In keeping with the principles of the NzW methodology, adjustments for small variations in weight can be accommodated by simply factoring the reference stresses according to the ratio of ramp weight for any particular flight type. Revenue flights were associated with the AT SPCs and an approximate mix of these SPCs was derived for the BOAC/BA Super VC10 aircraft. The equivalence with the RAF CMk1 SPCs was apparent from the descriptions of the three reference AT SPCs, namely: SPC 11 Route Flying – Long Range Task (> 6 Hr.) SPC 12 Route Flying – Medium Range Task (3 – 6 Hr.) SPC 13 Route Flying – Short Range Task (< 3 Hr.) For example, BOAC/BA flying patterns were presumed to have consisted largely of trans-Atlantic flights. Although these flights might be expected to fall into the long-range SPC 11 profile, given the other data available this is not consistent with the aircraft history. In particular, the mean revenue flight duration of 3.85 FH suggested that there were a significant number of medium-range flights at the top end of the SPC 12 duration limits and, similarly, a considerable number of shortrange SPC 13 flights at the top end of their range. A simple numerical analysis of the combined totals of revenue flights and hours was used to investigate the significance of this observation; this sensitivity analysis demonstrated that the mix of long-, medium- and short-range flights actually would have little effect on the total fatigue damage in commercial airline service (Table II). With reference to Table II, direct interpolation by flight duration of the LOVs for the three reference SPCs results in a damage estimate of 5.560. This result confirms that the conclusion from the sensitivity analysis is reasonable. For the purposes of the subsequent analysis, and in the absence of any other evidence, the typical usage of the BOAC/BA Super VC10 aircraft was taken to be represented by a mix of 5% long-range flights, 63.5% medium-range flights and 31.5% short-range flights. Using the same analysis process as for the combined service history totals, the number of training sorties was estimated from the commercial airline’s recorded number of roller landings for each individual aeroplane (Table III). Having accounted for the commercial training flights, the number of revenue flights for each aircraft were derived, as before – see Table IV. The results show a

Service history analysis and teardown

1147

SPC11

SPC12

SPC13

a (%) 0% 1% 2% 3% 4% 5% 10% 12% 15% 20% 25% 30% 50% 55% 60%

b (%) 73.4% 71.4% 69.5% 67.5% 65.5% 63.5% 53.7% 49.7% 43.8% 34.0% 24.1% 14.2% -25.2% -35.1% -45.0%

c (%) 26.6% 27.6% 28.5% 29.5% 30.5% 31.5% 36.3% 38.3% 41.2% 46.0% 50.9% 55.8% 75.2% 80.1% 85.0%

Damage Flying Sorties Hours 5.531 5.531 5.529 5.529 5.528 5.528 5.527 5.527 5.525 5.525 5.524 5.523 5.516 5.516 5.513 5.513 5.509 5.509 5.502 5.502 5.494 5.494 5.487 5.487 5.458 5.458 5.450 5.450 5.443 5.443

1.6%

If "a", "b" and "c" are the corresponding proportions of long-, medium- and short-range flights comprising the total number of revenue flights: a + b + c = 100% Assuming that the SPC 11, 12 and 13 sortie durations (Table I) apply to the equivalent commercial airline flights: (a x 460) + (b x 280) + (c x 95) = 100% x 230.8

Table II. Sensitivity analysis of “damage” estimates based on fatigue analyses of long-, medium- and short-range Air Transport SPCs for the CMk1. The approximate revenue usage mix for the BOAC/BA aeroplanes is highlighted

Landings (Pre-conversion) Aircraft No. 851 857 862 863 866 Σ Mean

FH

Roller

Braked

53046 46899 46879 45682 43212 235718 47144

2046 2606 1374 1312 1640 8978 1796

14325 12955 12158 11654 11116 62208 12442

Commercial training flights Number of Training Sorties 198 252 133 127 159 868 174

Braked Landings (Training) 384 489 258 246 308 1683 337

FH (Training) 658 838 442 422 527 2886 577

Table III. Analysis of the recorded histories of the BOAC/BA airline service – training flights

K.A. Lucas and M.J. Duffield

1148

small variation in typical flight times of about 15 minutes in just less than 4 hours; taking cognisance of the considerable range of total service lives, this variation does not seem unreasonable. Aircraft No.

No. Revenue Flights

FH (Revenue)

851 857 862 863 866

13941 12466 11900 11408 10809 60525 12105

52388 46061 46437 45260 42685 232832 46566

Σ Mean

Mean Revenue Flight Time Hrs Mins 3.76 225.47 3.69 221.69 3.90 234.13 3.97 238.04 3.95 236.95 19.27 1156.28 3.85 231.26

Table IV. Analysis of the recorded histories of the BOAC/BA airline service – revenue flights The assumed revenue flight type mix from Table II was then applied to the commercial revenue flights (Table IV). Dividing the number of flights accrued by each aeroplane in each range category by the appropriate LOV for the corresponding RAF CMk1 SPC, the individual increments of damage were estimated. Summing the damage for each category gives an estimate of the total damage accrued by each aeroplane during their commercial revenue service. The results of these analyses are shown in Table V.

Aircraft No. 851 857 862 863 866

Number of flights Long Medium Short Range Range Range 697 8859 4384 623 7922 3921 595 7562 3742 570 7249 3588 540 6868 3399

Total damage (Revenue) 1.272 1.138 1.086 1.041 0.986

Table V. Estimated damage for commercial revenue flights An additional allowance for roller landings and intermediate braked landings is made in the NzW analysis. For commercial flying it has been assumed that all rollers are associated with training sorties, and that the associated damage per roller is 1/25 of the damage due to the GTAC as obtained from the LOV for the CT SPC 01. Intermediate braked landings are conservatively assumed to generate the same level of damage as the primary GTAC. The damage calculations for the commercial training flights are summarised in Table VI.

Service history analysis and teardown

Aircraft No.

No. Training Sorties

851 857 862 863 866

198 252 133 127 159

1149 Number of landings Roller Intermediate braked 2046 186 2606 237 1374 125 1312 119 1640 149

Total damage (Training) 0.0030 0.0038 0.0020 0.0019 0.0024

Table VI. Estimated damage for commercial training flights The total damage estimated for each of the five Super VC10s whilst in commercial service with BOAC/BA is the sum of the damages calculated in Tables V and VI; the totals are shown in Table VII.

Aircraft No. 851 857 862 863 866

Estimated Fatigue Damage Revenue Training Total flights flights 1.272 0.003 1.275 1.138 0.004 1.141 1.086 0.002 1.088 1.041 0.002 1.043 0.986 0.002 0.989

Table VII. Estimated total damage for the five Super VC10 aeroplanes under BOAC/BA commercial service The same analysis process was applied to the five Standard VC10 aircraft which had been operated commercially by BOAC/BA and Gulf Air prior to conversion to the KMk2 configuration for the RAF. This information was needed because it was these aircraft that provided the teardown evidence previously described. The results are shown in Table VIII.

Aircraft No. 806 809 811 813 814

Estimated Fatigue Damage Revenue Training Total flights flights 0.962 0.001 0.963 1.011 0.001 1.012 0.999 0.001 1.000 0.933 0.002 0.935 1.027 0.001 1.028

Table VIII. Estimated total damage for the five Standard VC10 aeroplanes under BOAC/BA/Gulf Air commercial service

K.A. Lucas and M.J. Duffield

1150

Having accounted for the commercial usage of the subject aircraft, it is now necessary to apply the same methodology to the RAF usage after conversion of these airframes to the KMk4 and KMk2 AAR tanker configurations. RAF usage Across the fleet of all variants there are generally two phases of RAF usage that must be considered for analysis by the NzW method; these comprise the periods prior to and subsequent to the starting point of the flight records database in 1991. Fortunately, for the KMk4 aircraft, the entire military service lives are captured within the database and so the NzW calculations can be conducted on a flight-byflight basis directly; the results are summarised in Table IX.

KMk 4 Aircraft No. 851 857 862 863 866

NzW Damage Increment (See Note) 0.030 (See Note) 0.072 0.082

Note: Calculations not completed for Aircraft 851 and 862 as these airframes were retired from service while the analysis method was being developed.

Table IX. NzW damage increments for KMk4 aircraft in RAF service Conversely, the KMk2 had entered service with the RAF during the 1980s and so a considerable proportion of the total service life of each of these aircraft is not contained within the database. The NzW method was used to analyse the service history between 1991 and the retirement date to generate the required estimate of fatigue damage for this period. The resulting damage rates per SPC were then superimposed upon the known total FH for the period preceding the start of the database on the assumption that the usage patterns for the KMk2 had remained fundamentally the same for the entire RAF service life. As an example, the database for KMk2 aircraft 809 contained records for 1,211 sorties and 3,766 FH; the earlier period encompassed 3,118 FH. The NzW calculations yielded an estimate of 0.041 as the damage increment for the database sorties and this translated to an equivalent estimate for the earlier period of 0.048. The second figure is higher because it contains an additional allowance for the uncertainty where no details of individual sorties are known. Consequently, the total damage estimate for aircraft 809 comprises THREE parts, namely (i) the increment due to commercial operations (1.012); (ii) the estimated

Service history analysis and teardown

1151

increment for the period of RAF usage prior to the commencement of database records (0.048); and (iii) the increment calculated directly by NzW from the database records up to the retirement date (0.041). The total damage estimate for aircraft 809 is therefore 1.101. Thus the NzW damage has been calculated for each phase of flying from commercial usage up to the most recent sorties available in the database. The damage calculations are in terms of “damage units” which can be used to derive a relative measure of fatigue life consumed versus an allowable life. It should be noted that the values obtained by the NzW method are approximately 1 for all of the KMk2 aircraft which is not only convenient but also shows that the original indications that these aircraft had exceeded the theoretical inspection threshold for the Rib 0 joint by 50% had been unreliable. However, this was solely due to the conservative choice of s-N curve in the earlier work undertaken before the refined curve used for the NzW calculations became available. Teardown comparison The use of NzW principles accounts for the actual usage of each aeroplane, not only in terms of SPC mix, but also the actual weights and ‘g’ spectra. There have been changes in the aircraft usage patterns so the balance between flying hours and sorties has evolved; this is also accommodated in the NzW analysis. Finally, intermediate landings (both full-stops and rollers) are accounted for in terms of estimated damage increments, although these contributions to the total damage are relatively small. In order to provide a baseline against which to compare the remaining aircraft in the VC10 fleet, the damage accrued at the point of retirement for the teardown aircraft was calculated as outlined earlier in this paper. There was no significant fatigue damage found during the teardown inspection of the Rib 0 joints taken from any of the sample aircraft. The scope of the teardown inspections, the similar service histories of the teardown airframes and the inclusion of all of the appropriate factors in the fatigue calculations allow the cumulative fatigue damage at the point of retirement to be taken as the effective LOV for the remaining inservice VC10 fleet. In Table X, the calculated fatigue damage for KMk4 aircraft 866 is compared with that for the teardown KMk2 aircraft 809 because the Rib 0 joints on these two aircraft were structurally identical. Various modification/repair standards applicable to other aircraft were such that different teardown results must be used for the other KMk4 aircraft; the conclusions from those comparisons were essentially the same as for aircraft 866.

K.A. Lucas and M.J. Duffield

1152

Aircraft No. 809 (KMk2 teardown airframe) 866 (KMk4)

Commercial 1.012 0.989

Damage accrued Military Military (pre-database) (database) 0.048 0.041 0.082

Total 1.101 1.071

Table X. Examples of damage accrued for comparable teardown and inservice airframes Given the results from the various damage calculations it is possible to project the remaining usage available for individual aeroplanes before they reach the revised LOV; at that point the inspections and bolt replacement programme necessary to sustain the required airworthiness standard would need to be invoked. From Table X the margin of LOV available for aircraft 866 is (1.101 – 1.071 =) 0.030; at first sight this appears to be almost inconsequential but it can be exploited in terms of the equivalent numbers of FH and/or sorties by relating it to current usage. In this case, the increment due to military flying to date is 0.082 which was achieved during 2064 sorties and 6639 FH of RAF usage. The remaining margins with respect to the LOV are therefore equivalent to about 755 sorties or 2430 FH. For the expected utilisation rates of the KMk4 aircraft, these margins are likely to be sufficient to enable the expected retirement date to be achieved within the teardown-derived LOV provided that typical utilisation remains similar to past experience and that usage is monitored by the NzW method upon which the LOV depends.

CONCLUSIONS The problems of limited or incomplete records for the early periods of service usage for the VC10 aircraft operated by the RAF have been overcome, including accounting for the commercial use of the AAR tanker airframes prior to conversion. This has been achieved by taking advantage of the data generated for the purposes of the fatigue and damage tolerance qualification of the airframe and a review of all available usage statistics. The results have been combined with the evidence obtained from the structural teardown of a number of samples of the critical wing root joint at Rib 0 to provide a baseline against which to monitor ongoing usage of the VC10 as operated by the RAF in its CMk1K, KMk3 and KMk4 variants. That baseline is expressed in terms of a Limit of Validity representing the point at which a programme of inspections and bolt replacements would be necessary to satisfy the damage tolerance requirements. The monitoring process relies upon the NzW method to account for variations in operating weights, ‘g’ spectra and usage patterns and this is achieved by reference to a set of sortie profiles which are described in terms of these parameters plus a set of stress data derived from the finite element model of the Rib 0 joint. The

Service history analysis and teardown

1153

associated fatigue damage calculations have been performed using an s-N curve optimised for the purpose by the incorporation of suitable factors to account for the differences in construction between the subject joint and the fatigue test pieces. This work has been necessary due to the design of the joint which was qualified according to the fail-safe principle of the late 1950s and early 1960s; that principle led to a design that is theoretically susceptible to multiple-site damage. Furthermore, the damage tolerance analysis of the joint resulted in very small critical crack sizes being identified, leading to the requirement for invasive NDT inspections. The results obtained from the combination of the usage review, teardown evidence and fatigue damage calculations have been presented for the specific example of the VC10 KMk4 AAR tanker. It has been shown that the NzW method provides reliable basis for achieving the projected retirement date within the derived Limit of Validity, thereby allowing a considerable maintenance burden to be deferred whilst sustaining the required standard of airworthiness. The conclusions in this paper relate only to the Rib 0 wing root joint and form part of the continued activities that are used to collectively assure the structural integrity of the VC10 fleet.

ACKNOWLEDGEMENTS The work presented in this paper was undertaken by the Airworthiness & Structural Integrity Group of QinetiQ. However, thanks are due to the support provided by the Design Organisation for the VC10, namely BAE SYSTEMS at Chadderton, Manchester, without whose contributions of technical information and advice the analysis would not have been practicable.

REFERENCES [1] Duffield, M.J., Lucas K.A. (2007). “RAF VC10 fatigue monitoring by NzW analysis”. In: Minutes of the 30th ICAF Conference. [2] Duffield, M.J. (2001). “Structural teardown examination of the VC10 aircraft”. In: Minutes of the 27th ICAF Conference, pp.2/25-2/26, Rouchon, J. (Ed.). [3] Duffield, M.J. (2003). “Teardown examination of the VC10 tanker/transport airframe”. In: Minutes of the 28th ICAF Conference. [4] Duffield, M.J. (2008). “Predicting the Past – An Exercise in Re-building History”. In Ageing Structures – Growing Old Gracefully, Institution of Mechanical Engineers.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

DEVELOPMENT OF FATIGUE LIFE MONITORING OF RMAF FIGHTER AIRPLANES Wahyu Kuntjoro1, M Suhaimi Ashari 2, M Yazid Ahmad3 Assanah M Mydin4 1

Professor, Universiti Teknologi MARA, Malaysia Colonel – Director of CAESE RMAF, Malaysia 3 Principal Researcher – STRIDE, Malaysia 4 Director – CAIDMARK, Malaysia

2

Abstract: The Royal Malaysian Airforce (RMAF) operates fighter squadrons of F/A-18D, MiG-29, and Sukhoi-30 which were designed on Safe Life principle. RMAF has been conducting an intensive fatigue life monitoring program to the RMAF F/A-18D and is developing fatigue life monitoring (FLM) program to other fighters in its inventory. For this purpose, RMAF MiG-29 became the focus of the development and research. The fatigue life of MiG-29 was decided to be based on the wingfuselage lug joint structure. Low Cycle Fatigue (LCF) approach was adopted. The stress spectra of the wing-fuselage lug joint, was derived through mapping of g-spectra to the 1-g stress level of the lug. The g-history was obtained from the accelerator installed in the airplane. The 1-g stress level was obtained by finite element modeling of the wing structure and lug joints using NASTRAN. Cycle counting was obtained by using rainflow procedure. Hardness and chemical composition tests combined with empirical approach were then employed for developing strain-cycle diagrams. Notched effect was taken into account using Neuber theory. Mean stress effect was dealt with using Smith-WatsonTopper formula. Miner’s rule was used to calculate the fatigue damage accumulation. A fatigue life prediction software which incorporates the above concepts had been developed. The software is ready and can be used for the FLM of the RMAF MiG-29. The methodology used in the development of the fatigue life monitoring is also applicable to other Safe Life designed airplanes M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1155–1164. © Springer Science+Business Media B.V. 2009

1156

Kuntjoro, Ashari, Ahmad and Mydin

and becomes part of the RMAF effort in its Aircraft Structural Integrity Program (ASIP).

INTRODUCTION The Royal Malaysian Airforce (RMAF) operates fighter squadrons of F/A-18D, MiG-29, and Sukhoi-30 which were designed on Safe Life principle. RMAF has been conducting an intensive fatigue life monitoring program to the RMAF F/A18D and is developing fatigue life monitoring (FLM) program to other fighters in its inventory. In the development of FLM capability, RMAF MiG-29 became the focus of the development and research. When the MiG-29 airplanes were purchased, no fatigue life monitoring program existed for this airplane. After the experience with the F/A-18D, it was decided to conduct the FLM to the MiG-29 based on the similar approach used for the F/A-18D. The traditional total life method makes no differentiation between crack initiation and crack growth. Since the development of the total life method, fatigue has been re-defined as a process of initiating a crack followed by an investigation of how that crack propagates through the structure (Figure 1). This paper deals only with the crack initiation (Safe-Life) portion, which is adopted in the fatigue design of MiG-29.

Figure 1: Three main fatigue analysis methods Safe Life is a concept with a fundamental objective of having a structure, which is not going to fail during the life it is designed for [1]. Safe Life concept is closely related to the fatigue of structure. While static loading is associated with high magnitude type of loading, fatigue failure is often related with low magnitude cyclic load. However, it is important to note that fatigue failure is not always due to the low magnitude of cyclic loading; there are cases of high fatigue loading encountered during service. The fatigue problem, which is associated with high

Development of fatigue life monitoring of RMAF fighter airplanes

1157

magnitude fatigue loading, is called Low Cycle Fatigue (LCF), while the low magnitude fatigue loading is called High Cycle Fatigue (HCF). The fatigue life can be represented as the fatigue life expended (FLE). The value of FLE is between zero and one. FLE value below one means the component is safe. For example FLE equals to 0.5 can be interpreted as the component comes to half of its fatigue life. FLE equals to one means the component has come to the end of its life. In the case of FLE equal to one, either the aircraft goes to retirement or it can undergo a Service Life Extension Program (SLEP). In the RMAF MiG-29 Fatigue Life Monitoring Program, the fatigue life is based on the wing-fuselage lug joint structure, and Low Cycle Fatigue (LCF) approach is adopted [2]. The stress spectra of this component, is derived through mapping of gspectra to the 1-g stress level of the lug. The g-history is obtained from the accelerator installed in the airplane, while the 1-g stress level is obtained by finite element modeling of the wing structure and lug joints. Rainflow cycle counting procedure is then applied. The fatigue characteristics (strain-life) of the lug material was obtained from the laboratory test, using the lug material sample, combined with the empirical formula of strain-life diagram. Notched effect is taken into account using Neuber theory. Mean stress effect is dealt with using SmithWatson-Topper formula. Miner’s rule is used to calculate the fatigue damage accumulation. A fatigue life prediction software for RMAF MiG-29, called MiGSLA (MiG-Service Life Assessment), which incorporates the above concepts had been developed.

MIG-29 SAFE-LIFE ANALYSIS PROCEDURE The general procedure of the RMAF MiG-29 Safe-Life Analysis as used in the fatigue life monitoring program is to be desribed in this section. The crack initiation approach requires estimation of stress history. This needs input data of service load, lug maximum stress, and material properties. After the stress and strain at the critical location on the lug are estimated, ‘Rainflow’ cycle counting method is then used to reduce load time history into a number of events and sequence. When the load history is obtained, it is then multiplied with the stress obtained from the finite element analysis. This stress factor was a loading based on 1-g level flight. Then the strain-life methods that incorporate mean-stress effects are employed for predicting structural fatigue life. Following this, the linear damage hypothesis proposed by Palmgren and Miner is used to accumulate the fatigue damage. In strain-life approach, fatigue resistance of metals can be characterized by a strain-life curve. The relationship between total strain amplitude, Δε / 2 , and reversal to failure 2Nf, can be expressed through the following form [3,4]:

Kuntjoro, Ashari, Ahmad and Mydin

1158

Δε σ f (2 N f = 2 E '

where

ε

' f

σ 'f

)

b

+ ε 'f (2 N f

)

c

(1)

is the fatigue strength coefficient; b is the fatigue strength exponent;

is the fatigue ductility coefficient; c is the fatigue ductility exponent.

For a cyclic loading, stress-strain curve can be expressed through the following form:

ε total

⎡σ ⎤ = +⎢ '⎥ E ⎣K ⎦

σ

1

n'

(2)

where K’ is the cyclic strength coefficient and n’ is cyclic strain hardening exponent (results of b divided by c). When constructing a strain-time history from stress-time data, hysterisis loops have to be considered using cyclic stress-strain curve. Using Masing’s Hypothesis, which assumes that the line describing a stress-strain hysteresis loop is geometrically similar to the cyclic stress-strain curve Eqn. 2 but numerically twice its size [3], it is obtained,

⎡ Δσ ⎤ Δσ Δε = + 2⎢ ' ⎥ E ⎣ 2K ⎦

1

n'

(3)

Neuber’s rule [4-6] is used to take into account the notch effect. The relation between the notch geometry and the stress-strain can be expressed through the following form:

( K t Δσ ) 2 = ΔσΔε E

(4)

where Kt is the stress concentration factor of the wing lug structure. Eqns. 3 and 4 need to be solved sequentially, and as it requires the repeated try and error calculation to find the root of curve, Newton’s iteration method is applied [7]. The strain-life equation need to be modified to account for mean stress effects. Smith, Watson and Topper (SWT) [8] have proposed an equation to represent the mean stress effects.

σ max where

σ max =

Δσ +σ0 2

' 2 Δε (σ f ) (2N f )2b + σ 'f ε 'f (2N f )b+c = 2 E

(5)

Development of fatigue life monitoring of RMAF fighter airplanes

1159

Palmgren-Miner’s cumulative damage summation rule [9,10] was proposed as a way to sum damage from different fatigue events or cycles, first by the Swedish engineer A. Palmgren and resurrected 20 years later by M.A. Miner:

∑

n n n n = 1 + 2 + .... + n = 1.0 N N1 N 2 Nn

(6)

where n is the number of cycles of a certain stress amplitude and N is the number of life of certain stress amplitude. The ratio of n/N representing the damage of the stress given.

LOAD HISTORY DEVELOPMENT The RMAF MiG-29 Load In many air forces of advanced countries, there is a practice to fit a Flight Data Recorder (FDR) in their aircrafts to capture in-flight data including aircraft parameters and strain reading for the purpose of fatigue usage monitoring. The digital data captured are used as the loading history. In the case of the RMAF MiG-29 aircraft, the FDR fitted in the aircraft captures various data during flight, which include g-force history. At this stage a reliable method to collect and extract the related structural data was introduced. The available data for fatigue analysis is in g-history, Figure 2. What needed is to convert this loading spectrum from greading to strain form, and restructure the cycles event so that it is in the structured order.

Figure 2: Example of g-history [2]

For the purpose of converting the load spectrum form, finite element analysis was conducted. The main objective of the FEA is to find the lug stress at the

Kuntjoro, Ashari, Ahmad and Mydin

1160

symmetrical level flight of 1-g condition. This stress value is then multiplied to the loading history for every single peak and valley. Cycle Counting The rainflow method [11] has been used for cycle counting. Reducing the measured history into a series of cycles and half cycles consistent with basic material behavior is critical, and ‘Rainflow’ cycle counting is established herein as the soundest technique for achieving such reduction. At this stage, it is also important to note that load truncated procedures need to be applied. In most of loading history, there is an existence of small number of ‘passed cycles’- cycles that do not cause any effect to the fatigue life. These cycles are removed from the load sequence. Fatigue life then can be predicted by combining the results of the cycle count with relevant basic data using the linear cumulative damage hypothesis.

FATIGUE MATERIAL PROPERTIES To determine the material characteristic of the lug, a specimen was prepared for material testing. Hardness test and chemical composition test were conducted [12]. It was found that the material was a titanium based alloy. In the strain-life analysis, the cyclic material data is needed. Here, approach proposed by Jun-Hyub Park and Ji-Ho Song [13] was used. Using extensive experimental strain-life curve data on 116 steels, 16 aluminium alloys and six titanium alloys, nearly all methods currently available for estimation of fatigue properties from simple tensile data are discussed in detail. They proposed that for the purpose of estimating the properties of titanium alloy, the uniform material law by Baumel and Seegar was considered the most accurate, and this approach was used in this project.

σ Δε = 1 . 67 B (2 N 2 E

σB

)

− 0 . 095

f

+ 0 . 35 (2 N

)

− 0 . 69

f

(7)

is the tensile yield strength.

FINITE ELEMENT ANALYSIS The purpose of conducting the FEA is to get the maximum stress either at fuselage joint or wing lug joint based on the symmetrical level flight condition at 1-g. This stress will become the reference stress of other g-reading position in load spectrum.

Development of fatigue life monitoring of RMAF fighter airplanes

Figure 3: Model of The MiG-29 Wing

Figure 4: Wing And Fuselage Side Lug

1161

1162

Kuntjoro, Ashari, Ahmad and Mydin

Finite element analysis was performed on the Mig-29 Wing with the wing-fuselage lug joints regarded as wing supports [14]. The reaction forces at the support can be found and to be applied as external forces to the lugs of the wing and fuselage of the aircraft. The maximum value of stresses occurring at the lugs at 1-g flight condition were obtained. Figure 3 shows the finite element model of the wing [14], while the lug model is shown in Figure 4.

SOFTWARE DEVELOPMENT The software was developed in C++ language environment with flat file and binary database. In this way, it facilitates ease of installation in every RMAF base (it does not need the third party database and tools such as SQL, Oracle and Crystal Report). The software will read flight data from TOPAZ and can perform life prediction analysis. The software is able to find out the mission severity for each mission performed by the airplane. The fatigue index at any stage of operation can be obtained. Through its excellent graphical presentation, the software can display the mission profile for various flight data. Figure 5 below shows the interface of the software system. Currently the MiG-SLA has been in operation to conduct fatigue life monitoring to the RMAF MiG-29. The RMAF F/A-18 reporting format [15] is used as basis to produce the RMAF MiG-29 Fatigue Usage report. The data can then be used to assist the fatigue management of the RMAF MiG-29 squadron.

Figure 5: MiG-SLA software interface

Development of fatigue life monitoring of RMAF fighter airplanes

1163

CONCLUSION This paper has reported the development of fatigue life monitoring program of RMAF MiG-29. The fatigue life is based on the wing-fuselage lug joint structure, and Low Cycle Fatigue (LCF) approach is adopted. An algorithm for performing the fatigue safe life analysis was proposed. The fatigue characteristics (strain-life) of the lug material was obtained from the laboratory test, using the lug material sample, combined with the empirical formula of strain-life diagram. Notched effect is taken into account using Neuber theory. Mean stress effect is dealt with using Smith-Watson-Topper formula. Miner’s rule is used to calculate the fatigue damage accumulation. Loading spectrum development also was highlighted. The stress spectra of this component, is derived through mapping of g-spectra to the 1-g stress level of the lug. The g-history is obtained from the accelerator installed in the airplane, while the 1-g stress level is obtained by finite element modeling of the wing structure and lug joints. Rainflow cycle counting procedure is then applied. A fatigue life prediction software for RMAF MiG-29, called MiG-SLA (MiGService Life Assessment), which incorporates the above concepts had been developed. The software is ready and can be used for the FLM of the RMAF MiG29. The methodology used in the development of the fatigue life monitoring is also applicable to other Safe Life designed airplanes and becomes part of the RMAF effort in its Aircraft Structural Integrity Program (ASIP).

REFERENCE [1]

Abd Ghani, R. and Kuntjoro, W et.al. (2003). In: The Royal Malaysian Air Force Fatigue Life Monitoring Program of F/A-18 and Hawk. The Proceeding of International Seminar on Aerospace Technology 2003. Jogyakarta, Indonesia.

[2]

Idrus, D. and Kuntjoro, W. (2004). Fatigue Life Prediction of The RoyalMalaysian Air Force (RMAF)MiG-29 Fighter Aircraft. TR/006 /ASI/ CM/2004, CAIDMARK, Malaysia.

[3]

Bishop, N.W.M. and Sherratt, F. (2000). Finite Element Based Fatigue Calculation. NAFEMS. Farnham, UK.

[4]

Bentachfine S. and Pluvinage, G. (1999). Notch Effect In Low Cycle Fatigue. International Journal of Fatigue, Vol. 21 pp. 421-30. Elsevier Publication.

[5]

Bannantine, J.A., Comer, J.J. and Handrock, J.L. (1990). Fundamentals of Metal Fatigue Analysis. Prentice-Hall Inc. England.

1164

Kuntjoro, Ashari, Ahmad and Mydin

[6]

Fuchs, H.O. and Stephens, R.I. (1980) Metal Fatigue In Engineering. 1st Edition. John Wiley & Sons Publishers, USA.

[7]

Kelley, C.T. (1995) “Iterative Methods For Linear And Non-Linear Equation”. Frontiers In Applied Mathematics, Volume 16. Publisher: Society For Industrial And Applied Mathematics.

[8]

Smith, K.N., Watson, P. and Topper T.H. (1970). A stress-strain Function for the Fatigue of Materials. Journal of Material, Vol. 5, No.6.

[9]

C.Y. Niu, M.C.Y. (1988). ”Aircraft Structural Design”, Conmilit Press Limited, Hong Kong.

[10]

Schijve, J. (2001). Fatigue of Structures And Materials”. Kluwer Academic Publishers.

[11]

Downing, S.D. and Socie, D.F. (1982). Simplified Rainflow Cycle Counting Algorithm. International Journal of Fatigue, Vol. 4, No.1. Elsevier Science Publishers.

[12]

Fathi, M. (2004). Hardness and Chemical Composition Testing On MiG29 Wing-Lug Material. TR/005 /ASI/ CM/2004, CAIDMARK, Malaysia.

[13]

Park, J. and Song, J. (1994). Detailed Evaluation of Methods for Estimation of Fatigue Properties. Department of Automation and Design Engineering, Korea Advance Institute of Science and Technology, Korea.

[14]

Hussein A.M. (2004). Royal Malaysian Air Force MiG-29 Aircraft Structural Integrity Report: Identification of The Maximum Stress Value That Occur At The Wing-Fuselage Joints At 1-G Symmetrical Level Flight Condition of The RMAF MiG-29. TR/005/ASI/CM/2004. CAIDMARK, Malaysia.

[15]

Idrus, D. and Abd. Malek, A.M. (2004). RMAF F/A-18D Fatigue Usage Report. TR/003/ASI/CM/2004, CAIDMARK, Malaysia.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

RESEARCH AND DEVELOPMENT OF IMPACT DAMAGE DETECTION SYSTEM FOR AIRFRAME STRUCTURES USING OPTICAL FIBER SENSORS Hiroaki Tsutsui1, Noriyoshi Hirano1, Junichi Kimoto1, Takahiko Akatsuka1, Hirofumi Sashikuma1, Nobuo Takeda2 and Naoyuki Tajima3 1

Aerospace Company, Kawasaki Heavy Industries, Ltd., 1 Kawasaki-cho, Kakamigahara, Gifu 504-8710, Japan 2 Graduate School of Frontier Sciences, The University of Tokyo, Mail Box 302, 5-1-5 Kashiwanoha, Kashiwa, Chiba 277-8561, Japan 3

R&D Institute of Metals and Composites for Future Industries, 25-2, Toranomon 3-Chome, Minato-ku, Tokyo 105-0001, Japan (Present address: Kawasaki Heavy Industries, Ltd.)

Abstract: The basic technologies of the impact damage detection system (IDDS) of composite structures were developed and demonstrated using a composite structure with embedded smalldiameter optical fiber sensors in FY2002. In our current R&D, the IDDS consisting of optical fiber sensors installed in a composite structure and an interrogation unit is developed for practical airframe application. Under our R&D plan to proceed towards product, cyclic loading tests using different types of composite specimens are conducted to verify the effect of IDDS. In that test using coupon specimens, how much those optical fibers are involved in the mechanical properties of the composites, and the effect of optical signals are investigated. IDDS capability is also evaluated by an impact loading test using a stiffened panel subjected to a cyclic loading.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1165–1176. © Springer Science+Business Media B.V. 2009

H.Tsutsui et al.

1166

INTRODUCTION The application area of carbon fiber reinforced plastic (CFRP) composites to aircraft structures has become much wider with an expectation for weight reduction of airframe structures due to high specific strength and stiffness of the material. However, impact damage in a composite structure caused by tool dropping, bird strike, hailstone, etc. is likely to be significant damage for aircraft. Even though the impact energy is low, internal damage such as delamination and matrix crack which are difficult to find from the surface by visual inspection, can reduce strength of composite parts. Moreover, the strength of compression-afterimpact (CAI) of subcomponent is regarded as an important design criterion for damage resistance/tolerance requirements [1]. For an operational aircraft, using a visual inspection as well as conventional non-destructive inspection (NDI) methods, e.g., A-scan, C-scan and X-ray, an airframe composite structure may be examined at great expense in both labor and downtime. So, if the system being able to diagnose in real-time or part-time is installed in a composite structure, it may be possible to obtain impact events in the structure directly. Structural health monitoring (SHM) technologies have potential to be helpful technologies for the composite structures to reduce inspection time, operating cost. We have developed basic technologies of detecting impact damage by using a small-diameter optical fiber [2] from FY1998 to FY2002 [3, 4]. Diameter of the optical fiber is enough small not to effect for serious strength degradation of a composite structure. We are now developing the impact damage detection system (IDDS) with the accumulated fundamental knowledge [5]. In establishing the IDDS that applies to the practical aircraft, it is necessary to develop a system with high durability. In this paper, the durability of the IDDS is validated through a cyclic loading test and/or an impact loading test by which barely visible impact damage (BVID) is induced in a quasi-isotropic composite specimen with embedded small-diameter optical fiber sensors. From the test result, the ability of damage detection of the IDDS is evaluated.

IDDS TECHNOLOGIES IDDS concept The IDDS consists of small-diameter optical fiber sensors installed in a composite structure and a diagnostic system. The damage detecting method is achieved with optical loss in optical fiber sensors and the strain response by Fiber Bragg Grating (FBG) sensors made of small-diameter optical fibers respectively. A newly developed small-diameter optical fiber with the diameter of 40μm cladding and 52μm polyimide coating respectively [2] has no serious effect on the mechanical properties of composites. It was also confirmed that a multi-mode optical fiber had more advantage than a single-mode optical fiber for detecting damage because of the sensitivity for bending loss. It is possible to construct a simple damage

R&D of impact damage detection system

1167

detection system for measuring the optical loss in a multi-mode fiber before and after an impact event. The first damage detection method by using the optical loss in optical fiber sensors is attained by measuring the optical intensity in multi-mode type optical fibers, as shown in Fig.1 (a). Incident light entered from one end of an optical fiber passes through a damaged area and the intensity of the light attenuated by the damage is detected at the other end. The optical loss is caused by impact damage such as matrix crack and delamination, and it is also possible to increase the optical loss by local deformation. Damage initiation can be judged from the degree of the optical loss. This method does not need to measure an optical intensity in real-time. The other method is to use single-mode type FBG sensors. The sensor can measure the strain induced by an impact event as illustrated in Fig.1 (b). Damage initiation can be also estimated from the change in strain responses, and the damage position can be detected using the differences of arrival times of the strain at each sensor. Schematic diagram of the IDDS is shown in Fig.2.

Impact Optical fiber

Impact

FBG sensor

Delamination, Matrix crack

Bended fiber Damaged fiber

Strain Impact response

(a) Optical loss measurement

(b)Strain measurement

Figure 1. Damage detection method of the IDDS. INSTALLED OPTICAL FIBER SENSOR IMPACT DATA RECORDER

COMPOSITE STRUCTURE

ANALYSIS／ VISUALIZATION SYSTEM LIGHT SOURCE PHOTO DETECTOR

TRIMMABLE CONNECTOR

OPTICAL INTENSITY

FBG SENSOR SYSTEM

Damage

STRAIN

Figure 2. Schematic diagram of impact damage detection system (IDDS).

H.Tsutsui et al.

1168

Basic technologies and demonstration test We have conducted the R&D for SHM since FY1998 collaborating with the University of Tokyo as a part of the “R&D for Smart Material/Structure System” project within the Academic Institutions Centered Program supported by “R&D Institute of Metals and Composites for Future Industries (RIMCOF)” and “New Energy and Industrial Technology Development Organization (NEDO),” Japan [3]. As a demonstrator, the cylindrical composite airframe sub-structure with 3.0 m in length and 1.5 m in diameter (approximately 1/3-scale size of a small class jetliner) [3], was manufactured. Several elemental technologies, such as 1) installation technique of optical fibers to a stiffened panel with the optimum position and number of the optical fibers, 2) damage detection technique including algorism of the detecting method and necessary specification of measurement equipment and 3) connecting technique by using a newly developed optical connector "timmable connector", were applied to the upper panel of the sub-structure. The demonstration of our health monitoring technique by the demonstrator with embedded small-diameter optical fiber sensors was carried out as shown in Fig.3. It was confirmed from the demonstration test that the barely visible impact damage in a skin, stringer and interface between skin and stringer could be detected by the IDDS [4].

Figure 3. Demonstration test.

R&D of impact damage detection system

1169

Important issues For the application of IDDS to a practical airframe, it is important to evaluate mainly following items: 1. Durability/Reliability 2. Small sized interrogation unit in IDDS 3. Operation Method 4. Repair Method Assessment of the durability and reliability of the IDD system is necessary for its practical use. And small sized interrogation unit in IDDS should be developed for installing the system in an aircraft. In addition, cost-effective operation method for the IDDS both on the ground and in flight situations should be investigated. Repair method for composite structures including optical fiber sensors should be also established.

CYCLIC LOADING TEST Getting the prospect of aircraft application of the IDDS has been a target of our R&D since FY 2006. The R&D is conducted in collaboration with the University of Tokyo as a part of the project, “ Civil Aviation Fundamental Technology Program-Advanced Materials & Process Development for Next-Generation Aircraft Structures ” under the contract with RIMCOF. In this paper, we focus attention on the durability of the IDDS in the important issues as mention above. As the first step, a cyclic loading test is conducted using several coupon level composite specimens with embedded small-diameter optical fibers, to investigate the effect of the embedment on the mechanical properties of composites and the optical intensity in the optical fibers. Test specimen Ten quasi-isotropic CFRP specimens [+45/0/-45/90]2S were manufactured. The specimen with 265 mm in length, 25 mm in width and 3 mm in nominal thickness, was molded at 180ºC by using an autoclave. Several polyimide coated optical fibers with 40 μm in cladding diameter, were embedded in the interlaminar resin between 90º plies, by an equal space of 5 mm as shown in Fig.4. In addition, the specimens without an optical fiber were prepared for checking the effect of the inside fibers on fatigue strength. Test method and results Tension-tension cyclic loading test in RT / DRY was conducted at a stress ratio of 0.3 at maximum stress level of 143 MPa. The cycle number of the constant amplitude load is 104,000. The load cycle is 2 Hz. Figure 5 indicates an overall appearance of test system.

H.Tsutsui et al.

1170

It was confirmed that all specimens could be loaded without any fracture in composites judged from visual inspection. In this test, the load history as well as the optical intensity in the optical fiber embedded in F0 specimen was measured in order to evaluate the possibility that an optical fiber sensor can be applied to a composite without serious problem on durability. It was measured by using a light source and a photo detector. The relationship between the change of the optical intensity and the cycle number is shown in Fig.6. The optical intensity in the optical fiber installed to the specimen was normal during cyclic loading test. It could be found that the durability (fatigue) of the optical fiber used in the IDDS was in sight. It is because there was no serious effect on the strength of any specimens as well as the optical intensity in all optical fibers from the test result.

F0 (3EA)

Small-diameter optical fibers (3EA)

F90 (4EA)

Small-diameter optical fibers (25EA)

Figure 4. Composite coupon specimen.

Unit:mm

R&D of impact damage detection system

1171

Testing machine Specimen

Data recorder

Light source Photo detector

CHANGE OF OPTICAL INTENSITY , dB

Figure 5. Appearance of test system.

20 0 - 20 Specimen S/N : F0- 1 Specimen S/N : F0- 2 Specimen S/N : F0- 3

- 40 - 60 - 80 1

10

100

1,000 10,000 CYCLE NUMBER

Figure 6. Change of optical intensity.

100,000

1,000,000

H.Tsutsui et al.

1172

SYSTEM EVALUATION TEST A system evaluation test using a stiffened panel with embedded small-diameter optical fiber sensors is conducted to evaluate the durability of the IDDS. Test specimen CFRP stiffened panel with 1,800 mm in length and 500 mm in width has a skin and 3 blade types of stringers as shown in Fig. 7. The space of aluminium frames is 500 mm. And nine optical fibers and four FBG sensors, which have small-diameter, are embedded in the skin, stringer and interface of the skin/stringer. Trimmable connectors are also embedded into both edges of the panel and it is possible to trim these parts after curing. So, the developed connector improves handling of the optical fiber sensors and reduces the risk of unexpected cutting of hung-optical fiber.

Material CFRP skin / stringer panel (16PLY)

Embedded trimmable connectors 1800 500

Optical Fiber Sensors Small-diameter FBG sensors (4EA) Small-diameter Optical Fibers (9EA)

Embedded trimmable connectors

Stringer(3EA)

Frame(2EA) 500

Unit: mm

Skin

Note: White line indicates embedded small-diameter optical fiber sensors. Figure 7. Composite stiffened panel.

Test method and results The panel is subjected to several impact loadings before and after a compressive cyclic loading test. Figure 8 shows a configuration of system evaluation. In step1, an impact loading test as shown in Fig.9, was carried out to make sure that IDDS functioned properly before a cyclic loading test. The panel was subjected to several impact loads of approximately 1 to 2 J by changing the dropping height of the impactor with 55 N in weight. Embedded optical fiber sensors are connected with optical fiber cables by embedded trimmable connectors.

R&D of impact damage detection system

1173

In the test, an impact load, a strain and an optical intensity were measured using a measurement system. It could be predicted that there was no damage in the panel from no reduction of the optical intensity. It was also judged from an ultrasonic Ascan. Strain responses measured by FBG sensors and bonded foil strain gauges were well correlated. From those results, it was confirmed that there was no trouble with the IDDS function. Figure 10 shows a cyclic loading test in step 2 after an impact loading test in step1. The panel was subjected to a cyclic compressive load using an actuator. Maximum and minimum compressive loads are 177 kN and 53 kN, respectively. Cycle number is 104,000. In the test, the load measured by a load cell, the optical intensity in an optical fiber, and the strain measured by FBG sensors and foil strain gauges were monitored as shown in Figure 11. Figure 12 indicates the strain data measured by a FBG sensor in a strain survey which was conducted before and after a cyclic loading test. Both strain data were well correlated. So it was found that there was no serious problem in those optical signals during the test. In step 3, an impact loading test after a cyclic loading test in step 2 was carried out for the purpose of IDDS function check. Impact testing system is the same as that of step1. Impact points are on a mid-bay skin, a stringer flange and a stringer center. Figure 13 shows the result in case of a stringer flange impact. BVID with a dent depth 0.15mm, occurred in the interface between a skin and a middle stringer. The damage shape could be found by the inspection using an ultrasonic A-scan. That damage part was also judged from the reduction of the normalized optical intensity in the optical fiber sensor, which was embedded in the same interface with BVID. Impact damage point estimated by the strain responses measured by FBG sensors could be predicted at the near impact location.

Stiffened Panel Test Equipment

Impactor

step 1 step 2

Impact Loading

Cyclic Loading

step 3

Stiffened Panel

Measurement system

IMPACT LOADING TEST

Actuator

Measurement system

CYCLIC LOADING TEST

Figure 8. Configuration of a system evaluation test

H.Tsutsui et al.

1174

Impactor

Stiffened panel

Measurement system

Figure 9. Set-up configuration of an impact loading test.

Stiffened panel

Cyclic loading

Figure 10. Set-up configuration of a cyclic loading test.

R&D of impact damage detection system Compressive loadLoad Compressive Strain (strain gauge) Strain (Strain Gauge)

[kN] 250

0

200 150 100

0.20

50 0

FBG4（Before cyclic loading test）

0.18

[%]

FBG4（After cyclic loading test）

0.16

Compressive strain, , %% Compressive strain

- 0.1

- 0.2 0

Norm. optical Strain intensity sensor) Strain (FBG Sensor) Norm. Optical Intensity (FBG

1175

[%]

- 0.1

- 0.2 2

0.14 0.12 0.10 0.08 0.06 0.04

1.5

0.02

1 0.5

0.00

0 992

993

994

995

996

997

998

0

999

50 100 150 CCompressive ompressive Loadload , , kN kN

Number CycleCycle number

Figure 11. Typical time histories. STR CL

STR CL

Figure 12. FBG sensor's strain.

STR CL

Damaged Area (Estimated)

STR CL

Damage shape (A scan)

Impact location

View from impact loading side

CL

STR CL

STR CL

STR CL

250

FBG2

FBG1

200 150 100 50

0 -250 -200 -150 -100 -50 0 -50

50

100 150 200 250

View from stringer side

-100 -150

FBG3

-200

Impact location FBG sensor Estimated point

FBG4

-250

Figure 13. Test result of a system evaluation test.

200

H.Tsutsui et al.

1176

CONCLUSIONS Several composite coupon specimens with embedded small-diameter optical fibers were subjected to a tensile cyclic loading. As the results, it could be found that the durability (fatigue) of the embedded optical fiber and the composite was in sight because there was no serious effect on the strength of any specimens as well as the optical intensity in all optical fibers. Integrity of the IDDS function was also evaluated whether the IDDS could detect impact damage before and after a compressive cyclic loading. It was confirmed that the IDDS kept ability to detect a damaged part and an impact point after a cyclic loading. After this, it is necessary to confirm that an installed (i.e. embedded and/or attached) optical fiber sensor and a composite with the sensor have enough fatigue strength for the aircraft application of IDD system by acquiring the additional data under the various test conditions assumed. The reliability, environment resistance, and operation and maintenance procedure of the IDDS should be also investigated.

ACKNOWDGMENT This work was conducted as a part of the project, “ Civil Aviation Fundamental Technology Program-Advanced Materials & Process Development for NextGeneration Aircraft Structures ” under the contract with RIMCOF, founded by Ministry of Economy, Trade and Industry (METI) of Japan.

REFERENCES [1] [2]

[3] [4] [5]

MIL-HDBK-17-3F, Composite Materials Handbook, Volume 3, Polymer Matrix Composites Materials Usage, Design, and Analysis. Satori, K., Fukuchi, K., Kurosawa, Y., Hongo, A. and Takeda, N. (2001), In: Smart Structures and Materials 2001: Sensory Phenomena and Measurement Instrumentation for Smart Structures and Materials, Proceedings of SPIE, vol.4328, pp.285-294, E.Udd and D. Inaudi (Eds.), SPIE, Bellingham, WA. Tajima N., Sakurai T., Sasajima M., Takeda N. and Kishi T. (2004), Adv. Composite Mater., vol. 13, n. 1, pp. 3-15. Tsutsui H., Kawamata A., Kimoto J., Isoe A., Hirose Y., Sanda T. and Takeda N. (2004), Adv. Composite Mater., vol. 13, n. 1, pp. 43-55. Hirano N., Tsutsui H., Kimoto J., Akatsuka T., Sashikuma H., Takeda N. and Koshioka Y. (2009), In: Smart Sensor Phenomena, Technology, Networks, and Systems 2009, vol. 7293, N. G. Meyendorf, K.J.Peters and W. Ecke (Eds.), SPIE, Bellingham, WA. (in press)

25th ICAF Symposium – Rotterdam, 27–29 May 2009

LOAD FACTOR (NZ) / ROLL RATE: FIRST COMPARISONS BETWEEN DESIGN AND IN-FLIGHT RECORDED DATA ON EUROFIGHTER TYPHOON ITALIAN FLEET

Tommaso Giacobbe and Fabio Sardo Alenia Aeronautica

Abstract: In order to monitor the Eurofighter Typhoon fatigue life and allow an efficient fleet management, the Structural Health Monitoring (SHM) system is currently being used to register a large quantity of flight parameters that influence life consumption, or simply characterize the usage of every single aircraft. Furthermore, synthetic fatigue parameters are also calculated by the system in order to monitor fatigue life associated to any single flight. This amount of flight and fatigue data is being collected and stored in a data base developed for the scope. As the number of monitored flights increases, data stored within the database become more and more able to represent the effective usage of the fleet. Actually, increasing the number of aircrafts operated by Italian Air Force, the amount of data available in the database is becoming significant to perform first studies. A further feature of the SHM system, in conjunction with the data base, is to allow focusing on the analysis of some specific aspects of aircraft usage, such Load Factor (Nz) occurrences, altitude cycles, airbrake usage, etc. Hence, also considering the increasing amount of data being managed by the database, it’s becoming interesting to perform some early analyses to compare “design” and “in flight recorded” data. The aim of this paper is to present a comparative analysis between design and in-flight registered data, performed on the Load Factor (Nz) coupled with Roll Rate data. During every single monitored flight, the on-board Structure M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1177–1186. © Springer Science+Business Media B.V. 2009

T. Giacobbe and F. Sardo

1178

Health Monitoring (SHM) system registers the number of occurrences of pre-established combinations of Load Factor (Nz) and Roll Rate values. Occurrences of these values belonging to the already analyzed monitored flights have been collected, analyzed and compared to the design values. In this way it is possible to analyze the in-flight behaviour in terms of symmetric and asymmetric manoeuvres. Specific investigations have been conducted upon roll rates distribution and asymmetric manoeuvre segments.

Nz AND ROLL RATE DESIGN ASSUMPTIONS REFERENCE In order to design the overall manoeuvre spectrum, the subsonic and supersonic asymmetric manoeuvre spectra had been combined. Assumptions on the distribution of the overall design g-spectrum into symmetric and asymmetric manoeuvres have been made. This subdivision was performed considering air-toair and ground-attack missions. Symmetric and asymmetric manoeuvres

The design g spectrum was used to establish the total number of manoeuvres with significant Nz increment. The Nz increment was assumed to be significant whether either positive above 2g or negative below 0g. These manoeuvres are then allocated to Symmetric and Asymmetric. Asymmetric manoeuvres are assumed significant if Roll Rate is higher than 30 deg/s. Two types (see fig. 1) of asymmetric manoeuvres was defined and called Type A and Type B. The type A consists of a roll to bank at Nz=1g, pull to required Nz at null Roll Rate, return to Nz=1g and finally roll out of bank at Nz=1g. Hence the Type A asymmetric manoeuvre is characterized by non concurrent Nz and roll rate. On the other side, type B asymmetric manoeuvre consists of a roll to bank at Nz=1g, starting to pull to Nz>1 before peak roll rate and reaching the required Nz null roll rate; after the required Nz>1 phase, starting to roll out at peak Nz>1 already, reach required roll rate at Nz=1 and finally rolling back to straight level. Hence the type B asymmetric manoeuvre is characterized by concurrent Nz and roll rate. Design number of Asymmetric Manoeuvres. In order to calculate the number of asymmetric manoeuvres, the 90% of the Air to Air and the 10% of the Ground Attack manoeuvres were assumed to be asymmetric (see table I), just as it had been done for Tornado on the basis of F-14 flight data [2].

Load factor (Nz) / roll rate: Design vs in flight data

1179

Fig.1: Symmetric and asymmetric manoeuvres definition ACTIVITY

SYMMETRIC

AIR-TO-AIR GROUND-ATTACK

10% 90%

ASYMMETRIC TYPE A 80% 10%

ASYMMETRIC TYPE B 10% 0%

Table I: Allocation of manoeuvres. Hence, from the design overall g spectrum (distributed into air to air and ground attack spectra), the number of manoeuvres either exceeding 2g level and less than 0g were calculated to be: Exceeding 2g Air to Air Ground Attack Less than 0g Air to Air Ground Attack

23000 Manoeuvres / 1000 hours overall flying 8000 manoeuvres / 1000 hours overall flying 1000 Manoeuvres / 1000 hours overall flying 500 manoeuvres / 1000 hours overall flying

1180

T. Giacobbe and F. Sardo

The design distribution of manoeuvres for 1000 flight hours is synthetically detailed within table II. On the basis of the manoeuvre distribution and occurrences per 1000 flight hours, the number of roll command per flight hour is easily calculated: (23000 x 0.9) + (8000 x 0.1) + (1000 x 0.9) + (500 x 0.1) = =22459 Manoeuvres / 1000 hours overall flying = = 22.45 ≈ 22 Manoeuvres / hour overall flying Hence the number of roll commands per hour is 44. Moreover, on the basis of [3], 30 per flight hour asymmetric manoeuvres without significant Nz were assumed for design. These manoeuvres are taken into account among the initiating roll in both type A and B manoeuvres and by the recovery roll for type A. ASYMM. ASYMM. ALL SYMMETRIC TYPES TYPE A TYPE B 2300 18400 2300 23000 >2g (71%) (7%) (57%) (7%) Air-to100 800 100 1000 Air <0g (3%) (0.3%) (2%) (0.3%) all 2400 19200 2400 24000 (74%) g<>1 (7%) (59%) (7%) 7200 800 8000 >2g 0 (25%) (22%) (2%) Ground 450 50 500 attack <0g 0 (2%) (1%) (0.15%) all 7650 850 8500 0 (26%) g<>1 (24%) (3%) 9500 19200 2300 31000 >2g (29%) (59%) (7%) (95%) 550 850 100 1500 Overall <0g (2%) (3%) (0.3%) (5%) all 10050 20050 2400 32500 g<>1 (31%) (62%) (7%) (100%) Table II: Number of manoeuvres (percentage with respect to overall manoeuvres) per 1000 flight hours.

SHM SYSTEM TO MONITOR ASYMMETRIC MANOEUVRES Structure Health Monitoring system During flight SHM system registers a great quantity of parameters, either related to fatigue life consumption or simply to in-flight aircraft behaviour. Among them the Nz occurrences, the Nz at Maximum Roll Rate and the Roll Rate at maximum Nz (for example) are continuously registered and made available via SHM files.

Load factor (Nz) / roll rate: Design vs in flight data

1181

G-Meter. In the past a g-meter incorporating an accelerometer, that measured the centre of gravity load factor Nz exceedences over predetermined thresholds, has been used to calculate fatigue damage accumulation. On Eurofighter Typhoon the mechanical g-meter has been replaced by a software based system which retains the method of operation of a g-meter as used on-board the Tornado aircraft. Hence the prime and count technique is employed, whereby a count is recorded when one of the recording g levels is exceeded, providing that the level has first been primed at a lower preset level. Asymmetric manoeuvres monitoring. In the past, various assumptions relating the number of such manoeuvres and g-meter data have traditionally been used to assess the usage of structure not directly related to the centre of gravity accelerations (i.e. Fin, Leading Edge Devices). To perform fatigue calculations relating to asymmetrical usage, the SHM system records the occurrences of Roll Rate at Maximum G and the G at Maximum Roll Rate, which give a measure of the asymmetrical manoeuvres effectively performed. In particular, a “maximum roll rate” is identified when both rise and fall either side of a peak turning point is greater than a predefined threshold of 30 deg/s (maximum roll rates which lie within the range -30 deg/s to +30 deg/s are not recorded). When a maximum roll rate is identified by the system, it is recorder together with the correspondent Nz level. Vice versa a “maximum g” is identified when both rise and fall either side of a peak turning point is greater than a pre-defined threshold of 1.0 g (“maxima g” which lie within the range 0 to 2.0 g are not recorded). When a maximum g is identified by the system, it is recorded together with the correspondent Roll Rate level. Detailed description of Data representation in matrices Organizing the in-flight recorded data into matrix form (Fig. 2), it is possible to read and distinguish the occurrences belonging to each type of manoeuvre or manoeuvre segment. One matrix is dedicated to “Nz at Maximum Roll Rate” and the other is dedicated to “Roll Rate at Maximum Nz”. Within both these matrices, Nz and Roll Rates bands are reported within lines and columns respectively. In this way, each area within matrices can be associated with different types or segments of manoeuvres. The pure Nz occurrences (with concurrent Roll Rate less than 30 deg/s), are representative of the number of Symmetric Manoeuvres plus the symmetric parts of Asymmetric Type A manoeuvres. The pure roll (with Nz above 0g but less than 2g) occurrences are representative of the number of Asymmetric type A manoeuvres. Both OS and SO parts of asymmetric type B manoeuvres are represented in dedicated parts of the matrices. Two lines of Roll Rate at Max G matrix and two columns of G at MAX Roll Rate matrix have been left blank: the data that should have been contained therein are automatically considered not relevant by the SHM system, being out of the design assumption to identify a manoeuvre. These data are respectively the Roll Rate occurred in correspondence to a Maximum Nz value between 0 and 2g and the Nz in correspondence of the maximum Roll Rate between -30 and + 30 deg/s.

1182

T. Giacobbe and F. Sardo

Fig. 2: Matrix representation of Nz and Roll Rates in-flight recorded by SHM

IN-FLIGHT REGISTERED DATA After first Eurofighter Typhoon delivery to Italian Air Force, the monitoring activity started by means of data recorded by the SHM system. Nowadays a significant amount of flight has been flown and monitored to allow first studies on aircraft usage and analyses upon Italian Air Force activity against design requirements. For the present study a set of 7 aircrafts (6 single seat and 1 twin seat) has been considered. A total amount of 700 flights flown by these 7 aircrafts has been analyzed, leading to the results below exposed. G-spectrum According to the design g-spectrum 31000 manoeuvres with Nz>2g and 1500 with manoeuvres less than 0g should be performed every 1000 flight hours. In-flight Nz occurrences data, registered by the SHM dedicated utility, result in a spectrum (fig. 3) with similar shape to the design g-spectrum, but significantly below the design values. This circumstance finds confirmation in the low fatigue life consumption (with respect to design) showed by examined aircraft.

Load factor (Nz) / roll rate: Design vs in flight data

1183

Fig. 3: In-flight registered g-spectrum versus design g-spectrum. Symmetric and Asymmetric Manoeuvres Aiming to a better data representation, unless otherwise stated all the in-flight registered data reported herein have been recalculated as the total number of manoeuvres in all conditions was 100000.

Fig. 4: In flight registered Nz/Roll Rate pairs

T. Giacobbe and F. Sardo

1184

Roll commands. SHM registered an average of 34 roll command per flight hour, significantly below the design assumptions (about 72%). This is a further confirmation that the use of the aircraft is definitely below design, not only in terms of load factor (Nz) but in terms of asymmetric manoeuvres also. Manoeuvres distribution. Within table IV the in-flight compared to the design distribution of manoeuvre is reported. The totally symmetric manoeuvres are a little more, but substantially in line, with the number assumed for design. Consequently the number of asymmetric manoeuvres results lower than expected. Nevertheless it has to be underlined that, among asymmetric manoeuvres, while the type A result in number largely lower than design predicted, the asymmetric type B manoeuvres result definitely more than expected.

SYMMETRIC ASYMM TYPE A ASYMM TYPE B

IN-FLIGHT REGISTERED 35238 45524 19238

DESIGN 30.9 % 61.7 % 7.4 %

Table IV: Manoeuvre types distribution comparison between in-flight and design (percentage of total manoeuvres) This aspect becomes even more evident comparing the number of Aymmetric Manoeuvres type A and B with respect to the total number of asymmetric manoeuvres. In fact the type B are expected (per design) to be the 9% only of the number of asymmetric manoeuvre, while the effective ratio is about 30%. Also due to the resultant excess of type B manoeuvres, it becomes interesting to evaluate in detail the type B manoeuvre segments (OS and SO parts). Asymmetric type B segments: OS versus SO parts. Concerning the asymmetric type B manoeuvres, among all registered data it is also possible to concentrate the analysis to the 2 segments of this type of manoeuvre. The number of OS segments (initial roll of the type B) is determined by the pair Nz at Maximum Roll Rate occurrences, while the number of SO segments is determined via the number of Roll Rate at Maximum Nz occurrences. The number of OS parts (both side rolls) performed are 19238 while the SO parts flown are 2190 only. The ratio between these two values is about 1/10. This condition was definitely not expected and denotes a very aggressive attitude while entering the asymmetric manoeuvre: the pilot’s stick pull (for Nz increasing) begins very often when the rolling is still in the first phase (first half). On the opposite side, a much less aggressive attitude is observed at the moment of coming out from an asymmetric manoeuvre: a great number of asymmetric type B manoeuvres starts as type B but end as type A. Only about a 10% of asymmetric manoeuvres started aggressive (type B) are concluded in an aggressive (type B) manner too, i.e. beginning the roll back action when the Nz is still at the maximum reached in that particular manoeuvre. The remainder

Load factor (Nz) / roll rate: Design vs in flight data

1185

(about 90%) of the asymmetric type B manoeuvres is concluded in a less severe way, starting the roll back only when the Nz is already lower than maximum. OS with high concurrent Nz. Furthermore a dedicated analysis has been performed on OS segments with concurrent Nz higher than 4g. It came out the number of these occurrences is extremely high with respect to design data, obtained from gspectrum. This because an OS segment with concurrent Nz at Max Roll Rate over 4g was associated (per design) only with manoeuvres that have a Nz peak of 9g (in SA segment, fig. 1). This assumption is not confirmed by the registered high number of occurrences of high over 4g) Nz at Max Roll Rate in absence of the correspondent manoeuvres with Nz=9g peak. Hence it should be concluded that “aggressive” roll commands are often performed also in manoeuvres definitely below the Nz=9g limit. By a similar analysis on the Nz at Maximum Roll rates with concurrent Nz>2, the number of occurrences is high but in line with the excess of type B manoeuvres with respect to design. Therefore it can be concluded that, limited to the OS parts, the roll rates commands are sensibly more aggressive than foreseen by design. This is even more interesting considered the average of roll commands number and intensity among all manoeuvres is instead lower than expected, as below exposed. Roll Rate probability spectrum. Further analyses has been performed on the roll rate distribution among all the roll commands (hence via the Nz at Maximum Roll Rate), considering the registered value of Roll Rates Maxima. This data have been organized to the scope and then compared to the roll rate probability spectrum defined at design.

Max Roll Rates >50 Max Roll Rates>130 Max Roll Rates>150

DESIGN

IN-FLIGHT

53.3% 5.4% 3.4%

29.7% 0.9% 0.2%

Table V: roll rate probability spectrum, design versus in-flight It emerges that the effective distribution of roll rates is less severe than design. This datum can be considered in joint with the number of roll commands per hour, observing that both the number and the intensity of the roll commands are less severe than design spectra. Moreover it has to be considered that, for design assumption, the severity of roll rates probability spectra is not different for type A and B asymmetric manoeuvres. From these data it is confirmed an average “action” (in terms of roll commands) less aggressive than expected. However this trend has to be compared to the opposite attitude of performing hybrid manoeuvres, with “aggressive” type B entries with concurrent Nz and roll rate but with less aggressive type A exits from the manoeuvre. It can be hence concluded

T. Giacobbe and F. Sardo

1186

that type B entrances are usually more aggressive and even more frequent than expected, but being type B manoeuvres less frequent than other manoeuvres type, the overall average roll rate distribution results still below design.

CONCLUSIONS Symmetric and asymmetric manoeuvres have been analyzed on the basis of SHM data, registered on a set of 7 Eurofighter Typhoon aircrafts. Nz spectrum has been found with the similar shape but significantly less severe than design. The number of roll commands and their overall average “intensity” is below design assumptions. Type B manoeuvres (considering the OS segments) resulted instead in higher number than assumed for design, but it has to be underlined that the number of OS segments is about 10 times the number of SO parts: this means that the great part of asymmetric type B manoeuvres starts only in “type B”, but ends in a “Type A” way. Moreover the number of roll commands has been analyzed with respect to the concurrent Nz values, resulting that the most aggressive entrances (roll rates with concurrent Nz>4g) are flown not only in manoeuvres with peak of Nz=9g (as it was assumed for design), but actually in less severe manoeuvres also. Finally, analyzing the Roll Rate design probability spectrum it has been found the confirmation that the actual in-flight distribution is less severe: for every level the number of roll rates countered results below he design level. Hence the most “aggressive” rolls are deemed to be related to the Type B manoeuvre entrance, which often are become type A manoeuvres when the exit rolls are flown.

REFERENCES [1]

Holpp, J.E. and Landy, M.A. (1976), “The Development of Fatigue/Crack Growth Analysis Loading Spectra”, AGARD Report No. 640.

[2]

Lauridia, R.R. (1979), “Statistical Analysis of Aircraft Manoeuvring Data”, 20th AIAA/ASME/ASCE Structures, Structural Dynamics and Material Conference.

[3]

Taylor, J. (1965), “Manual on Aircraft Loads”, AGARDograph No. 83.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

DEVELOPMENT OF A STRUCTURAL HEALTH MONITORING SYSTEM TO EVALUATE DE-BONDING IN COMPOSITE ADHESIVE STRUCTURES Hideki Soejima1, Noritsugu Nakamura1, Toshimichi Ogisu1, Hiroshi Wakai1, Yoji Okabe2, Nobuo Takeda3, Yoshihiro Koshioka4 1

Fuji Heavy Industries Ltd., Aerospace Company Department of Mechanical and Biofunctional Systems, Institute of Industrial Sciences, The University of Tokyo 3 Graduate School of Frontier Sciences, The University of Tokyo, c/o Department of Aeronautics and Astronautics, School of Engineering, The University of Tokyo 4 R&D Institute of Metal and Composites for Future Industries (RIMCOF) 2

Abstract: We have developed a novel structural health monitoring (SHM) system that can evaluate the debonding of adhesive layers and delamination in carbon-fiber-reinforced plastic (CFRP) structures, such as stiffened skin panels with bonded stringers in wing structures. Our SHM system consists of fiber Bragg grating (FBG) optical fiber sensors and piezoelectric (PZT) transducers as sensors and actuators, respectively. In order to evaluate the damage detection capability of our system, we compared the results derived from an analysis of Lamb waves detected by the system with results identified by conventional nondestructive inspection (NDI), A-scan. The test article that was used in the investigation simulated CFRP-bonded structures where debondings and delaminations might initiate and grow as a result of operational loads on an aircraft. In order to evaluate damage quantitatively with our system, we introduced a correlation coefficient as a damage index that could evaluate the initiation and growth of damage. Our investigation confirmed that our damage index correlated reasonably well with the debonding M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1187–1197. © Springer Science+Business Media B.V. 2009

H. Soejima et al.

1188

area identified by A-scan. Therefore, we verified that our damage index can be used to evaluate the initiation and growth of debonding in CFRP structures, and that the structural integrity of CFRP structures in aircraft can be diagnosed by means of our SHM system. However, more data must be acquired, and more detailed investigations must be conducted.

INTRODUCTION Among the best properties of CFRP are its high specific strength ratio and its high specific modulus ratio, which can contribute to reducing the structural weight of aircraft. Therefore, CFRP has been applied recently not only to the secondary structures but also to the primary structures of aircraft. Furthermore, the excellent corrosion resistance and fatigue resistance properties of CFRP enable one to improve the comfort of passengers by making it possible to increase humidity and pressure in the cabin compared to what is possible with conventional metallic airframes. Consequently, it is expected that CFRP will be applied increasingly to aircraft structures. However, because of the laminated construction of CFRP, this material is susceptible to delamination and debonding, which are not easily detected by means of simple external visual inspection. Such damage in the CFRP might cause extreme fractures that imperil the airworthiness of the aircraft. Therefore, when CFRP is used in aircraft structures, one must apply a concept of “no growth” in designing those structures. With a “no growth” design concept, one must provide a sufficient margin of safety so that any damage that cannot be detected during the manufacturing process will never increase throughout the service life of the aircraft. For example, values for compression after impact (CAI) strength or openhole compression (OHC) strength are used to calculate the design allowables of CFRP structures, with typical values being approximately one-quarter of the tensile strength of the CFRP lamina. In order to apply CFRP effectively and reduce structural weight further, a concept of “damage growth” that limits the growth of damage between aircraft maintenance events is absolutely necessary, just as would be the case with conventional metallic structures. SHM technologies show promise in improving the safety and reliability of aircraft. SHM systems can monitor strains, damage, corrosion, and so on by means of sensors that are permanently installed in structures; one can then analyze the data monitored by the sensors to diagnose structural integrity. When SHM systems are realized to diagnose the structural integrity of CFRP structures, one can overcome the above-mentioned structural problems by applying to CFRP structures the damage-tolerant designs that are used in metallic structures. This results in a reduction in weight and an increase in safety and reliability. Moreover, by

Development of a structural health monitoring system

1189

replacing schedule-based maintenance with condition-based maintenance, one can reduce operational costs and optimize fleet operation. Accordingly, we have developed a novel damage monitoring system that can detect the initiation and growth of delaminations and debondings in CFRP structures. To evaluate the damage detection capability of an SHM system that we developed, we compared the results of our method with those obtained by means of A-scan, using test articles that simulated the critical sections of CFRP bonded structures.

THE DEVELOPED DAMAGE MONITORING SYSTEM Figure 1 illustrates the scheme of the SHM system we developed to monitor damage. The system consists of PZT transducers, FBG optical fiber sensors, a function generator, a high-speed optical wavelength-detection device [1], and a damage evaluation system.

High speed FBG High-speed optical demodulator wavelength detection device

Light source

Generator

AWG-type filter

PZT film actuator PZT actuator

FBG sensor

Processor

Figure 1 Scheme of the developed structural health monitoring system PZT transducers A PZT transdcer is a component that can convert voltages into forces and vice versa. PZT transducers are frequently used in sensors, actuators, and energyharvesting devices. In our SHM system, we used PZT transducers as actuators that generate Lamb waves into CFRP structures. Figure 2 shows a typical input signal of the PZT actuator as used in our tests. FBG optical fiber sensors A FBG optical fiber sensors has a portion where the refractive index in the core of the single mode optical fiber is changed periodically. As the intervals of the grating change in response to environmental changes, the grating reflects light with a specific wavelength that corresponds to the change of the grating intervals [2].

H. Soejima et al.

1190

Voltage

Therefore, these sensors can be used to detect changes in strain, temperature, and so on. In our SHM system, we used FBG optical fiber sensors to detect Lamb waves propagating through CFRP structures. Moreover, because the FBG sensors we used have a smaller diameter [3] than that of conventional optical fibers, such sensors can be embedded into CFRP structures without degrading the mechanical properties of the CFRP material.

Time

Figure 2 Typical input signal of the PZT actuator High-speed optical wavelength detection device Our high-speed optical wavelength detection device uses an arrayed waveguide grating (AWG) as an optical filter that converts shifts in the center wavelength of the light reflected from the FBG into changes in the output voltage. Therefore, we can detect changes in strains in CFRP structures by measuring changes in output voltages. Because the device does not use mechanical moving parts, the device can detect strains in the CFRP even though the frequency of the Lamb waves that are propagated through the structures is extremely high. Damage evaluation system Lamb waves generated by a PZT actuator propagate through a CFRP structure and reach an FBG sensor installed in the structure. Because the Lamb waves correspond to the structure through which they propagate, changes in the structure such as the initiation and growth of damage result in changes in the Lamb waves. By analyzing changes in the Lamb waves, we can use this system to diagnose structural integrity.

TEST PROCEDURES Figure 3 shows the test article, which simulated a stiffened skin panel with bonded stringers as found in large aircraft. Note the placement of the FBG optical fiber sensors and PZT actuators. To manufacture the skin and stringers, we cured CFRP prepregs (pre-impregnated composite fibers) in an autoclave. We installed FBG optical fiber sensors by means of adhesive into the bond lines between the skin and the stringers, and then integrated them by means of a secondary bonding process.

Development of a structural health monitoring system

1191

Optic fibers

CFRP skin/stringer Dimensions of the skin/stringer panel, 900 W × 1500 L (mm)

Details of installed positions of FBG sensors and PZT actuators

Figure 3: Test article simulating a stiffened skin panel with stringers of CFRP We used fasteners to assemble the skin panel, ribs, and spars made of aluminum. Finally, we bonded the PZT actuators and FBG optical fibers onto the surface of the skin panel by means of a cyanoacrylate-type adhesive. Figure 4 illustrates the evaluation procedure, with the following sequence:

1st step

2nd step

3rd step

4th step

Measurement FL1 of healthy state

Measurement FL1 after 1st impact

Measurement FL1 after 2nd impact

Measurement FL1 after 3rd impact

1st impact loading

2nd impact loading

3rd impact loading

4th impact loading NDI by A-scan NDI by A-scan

NDI by A-scan

NDI by A-scan

Measurement FL1 after 4th impact

Measurement results of the system were compared with the results of the A-scan. FL1: Measured by the SHM system

Figure 4 Procedure for evaluating the developed SHM system

H. Soejima et al.

1192

1. We used our system to measure the Lamb waves, and we used A-scan to inspect the bonding condition in the healthy state. 2. In order to initiate damage in the bond line between the skin and the stringers, we applied impact loading to the bond line by means of a dropped weight type impactor.

3. We used our system to measure the Lamb waves once again, and we used Ascan to identify the damage initiated by the first impact. 4. We repeated the above series of procedures four times, in order to simulate the initiation and growth of damage in the CFRP structures. Table I lists the energy levels of the impact. The locations of the impact loadings were the same in all cases.

5. Finally, in order to evaluate the damage detection capability of our developed SHM system, we compared the results obtained by this system with the extent of damage that was identified by A-scan. Table Ι Energy levels of impact loadings Number of Initial height Cumulative energy impacts (mm) (J) 0 0 0 1 150 0.7 2 180 3.9 3 180 7.2 4 220 9.4

RESULTS AND DISCUSSIONS Figure 5 shows a picture of a damaged area that was identified by A-scan. The targets of the impact loadings were the edges of the bond lines between the skin and the stringers. Although the damage included not only delaminations but also debondings, the main damage was the debonding of the bond line, taking into account the energy level. Figure 6 shows the relationship between the damaged area and the cumulative impact energies. (Impact energies were derived by multiplying each initial height and weight of the impactor.) Note how the size of the damaged area increased with an increase in the impact energy. Figure 7 and 8 show the Lamb waves measured by the FBG optical fiber sensors that were bonded onto the surface of the stringer, and the sensors that were

Development of a structural health monitoring system

1193

integrated into the bond line between the skin and the stringers, respectively. By comparing the waveforms obtained by the bonded FBG optical fiber sensors with those obtained by the integrated sensors, one can see that the amplitude of the Lamb waves measured by the integrated sensors is smaller than that obtained by the bonded sensors. This phenomenon results from the optical loss that was caused by a core crash or a microbending of the fiber that was integrated into the bond line.

Figure 5 Results of the inspection by A-scan

Damage area / mm2

600 400 200 0 0

2

4

6

8

10

Cumulative impact energy / J

Figure 6 Relationship between damaged area and cumulative energies However, our SHM system can solve this problem by increasing the gain of optical power of the light source. To distinguish the difference of Lamb waves easily, the changing parts of the Lamb waves were indicated with arrows and other markings in each figure of Figure 7 and 8. These figures confirm that our system, whether by means of bonded or integrated FBG optical fibers, can measure Lamb waves and detect changes in them that result from changes in the damaged area. The changes in the Lamb waves are caused by changes in the propagation path of the Lamb waves as a result of the damage, which causes reflection of the Lamb waves and a decrease in their amplitude. In order to evaluate the extent of the damaged area by measuring the Lamb waves quantitatively, we introduced a correlation coefficient, C.C., that represents the similarity of two waveforms. This C.C. as a damage index is calculated by Eqn. 1,

H. Soejima et al.

1194 t = t max

Σ f(t) g(t)

C.C. =

t=0 t = t max

t = t max

t=0

t=0

Σ f ( t )2

(1)

Σ g ( t )2

where f(t) is the waveform measured under a healthy state, and g(t) is the waveform measured under a damaged condition. Figure 9 shows the relationship between the correlation coefficient calculated by Eqn. 1 and the damaged area identified by A-scan. This figure confirms that the C.C. that corresponds to both the bonded and the integrated FBG optical fiber sensors decrease gradually with an increase in the damaged area. Moreover, both C.C. indicate the same tendency. Therefore, this confirms that the C.C. is effective in evaluating damage quantitatively. Although additional data and further detailed investigations are required, we confirmed that our SHM system can diagnose the structural integrity of CFRP structures in aircraft. If application of SHM systems to aircraft can be achieved, the design philosophy that is tolerant of damage growth, similar to that used for metallic structures, can be used to design CFRP structures for aircraft.

CONCLUSIONS We developed an SHM system and evaluated its damage detection capability by comparing the results of an analysis of Lamb waves detected by this system with results obtained by conventional nondestructive inspection, A-scan. The test article we used simulated CFRP bonded structures, in which debondings and delaminations might initiate and grow as a result of operational loads in an aircraft. Our evaluation is summarized and confirmed as follows: 1. The system in which FBG optical fiber sensors, PZT actuators, and AWG optical filters are used can detect the Lamb waves propagated through CFRP bonded structures. 2. The system can detect changes in the Lamb waves when such changes result from the initiation and the growth of damage. 3. In order to evaluate damage quantitatively, we introduced a correlation coefficient as a damage index. The index decreased in proportion to the growth of damage, and can be used to evaluate the initiation and growth of damage in CFRP structures. 4. Our system can diagnose the structural integrity of CFRP structures in aircraft. However, more data and detailed investigations are required.

Development of a structural health monitoring system Healthy state

0.3

After 3rd impact

0.3 0.2 Voltage / V

0.2 Voltage / V

1195

0.1 0 - 0.1

0.1 0 -0.1 -0.2

- 0.2

-0.3

- 0.3 0

30

60

90

0

120

30

60

Time / s.

after 1st impac t

0.2 Voltage / V

Voltage / V

After 4th impact

0.3

0.2 0.1 0 -0.1

0.1 0 -0.1 -0.2

-0.2

-0.3

-0.3 0

30

60

90

120

0

30

60

90

Time / s .

Time / s.

(b) After 1st impact

(e) After 4 th impact

after 2nd impact

0.3 0.2 Voltage / V

120

(d) After 3rd impact

(a) At the healthy state 0.3

90

Time / s.

0.1 0 -0.1 -0.2 -0.3 0

30

60

90

120

Time / s.

(c) After 2 nd impact

Figure 7 Lamb waves detected by the bonded FBG optical fiber sensor

120

H. Soejima et al. Healthy state

0.15 0.1 0.05 0 -0.05 -0.1 -0.15

After 3rd impact

0.15 voltage / V

voltage / V

1196

0.1 0.05 0 -0.05 -0.1 -0.15

0

30

60

90

120

0

30

After 1st impact

120

0 -0.05

After 4th impact

0.15 voltage / V

voltage / V

0.1 0.05

0.1 0.05 0 -0.05 -0.1

-0.1

-0.15

-0.15 0

30

60

90

120

time / μs.

0

30

60

90

time / μs. (e) After 4th impact

(b) After 1st impact After 2nd impact

0.15 voltage / V

90

(d) After 3rd impact

(a) At the healthy state

0.15

60 time / μs.

time / μs.

0.1 0.05 0 -0.05 -0.1 -0.15 0

30

60

90

120

time / μs. (c) After 2nd impact

Figure 8 Lamb waves detected by the integrated FBG optical fiber sensor

120

Development of a structural health monitoring system

1197

Correlation coefficient

1 0.9 0.8 0.7

●：Integrated sensor

0.6

■：Bonded sensor 0.5 0

100

200

300

400

500

600

Damaged area, mm2

Figure 9 Relationship between correlation coefficient and damaged area

ACKNOWLEDGEMENTS This study was carried out as a part of the “Civil Aviation Fundamental Technology Program–Advanced Materials & Process Development for NextGeneration Structures” project, under contract with the R&D Institute of Metals and Composites for Future Industries (RIMCOF) founded by the Ministry of Economy, Trade, and Industry (METI) of Japan. We thank everyone related to this project.

REFERENCES [1]

S. Kojims, K. Satori, K. Fukuchi, and A. Hongo, (2002), Proc. SPIE, Vol. 4694, p 202.

[2]

A. Othonos and K. Kalli: Fiber Bragg gratings: Fundamentals and applications in telecommunications and sensing, Artech House (1999), Chapter 5.

[3]

A. Hongo, S. Kojima, and S. Komatsuzaki, (2005), Structural Control and Health Monitoring, Vol. 12, p 269.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

OLM: A HANDS-ON APPROACH Stephen Willis ACRA CONTROL Ltd

Abstract: With increasing costs and shrinking budgets, there is considerable economic pressure to get the most out of existing aircraft extending their in-service life if possible and to do so in a manner that does not compromise flight safety. One of the methods to achieve this is Operational Loads Monitoring (OLM). This paper presents an overview of military aircraft OLM programs, the typical parameters they measure and how they store and analyse the data. The research utilizes case studies of four aircraft for which ACRA CONTROL provided the data acquisition systems. The paper focuses on the data acquisition technology used to measure the flight parameters, the data processing and storage and how this data is processed on the ground. Finally, some of the technology and other improvements that have allowed newer OLM systems to provide a better fatigue profile of the aircraft are described.

INTRODUCTION Operational Loads Monitoring (OLM) shares similar data acquisition requirements and goals with other methodologies including: Health and Usage Monitoring Systems (HUMS), Condition Based Maintenance (CBM), Flight Parameter Recording (FPR), Operational Data Recording (ODR), Integrated Vehicle Health Management (IVHM), and Loads/Environmental Spectral Surveys (L/ESS). Benefits can include • •

Maintenance: reduced mission aborts, fewer aircraft on ground (AOG), simplified logistics for fleet deployment Cost: “maintain as you fly” maintenance flights are not required. Performing repairs when the damage is minor increases the aircraft mean

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1199–1214. © Springer Science+Business Media B.V. 2009

Stephen Willis

1200

• •

Time Before Failure (MTBF) and decreases the Mean Time to Repair (MTTR). Operational: improved flight safety, mission reliability and effectiveness Performance: improved aircraft performance and reduced fuel consumption

There have been several incidents involving military aircraft in recent years that have highlighted the importance of structural analysis for military aircraft e.g. F-15 Eagle (5 crashes since 2007). The US Navy issued a safety bulletin requiring inspections for all 636 F/A-18A-Ds [ i ] late last year after discovering fatigue cracks on outboard aileron hinges. The safety issue came as the navy seeks to extend the life of its oldest fighters to limit the impact of a projected strike aircraft shortfall across the next decade. This will apply to virtually all Air Forces around the world. USN officials issued the bulletin after finding cracks in the aluminium outer wing components of 15 F/A-18s. The USN has already extended the life of the F/A-18A-D fleet from 6,000 flight hours to 8,000 flight hours, with the oldest aircraft now approaching 7,500 flight hours. The USN is now seeking to make structural improvements to keep the fleet in service for 10,000 flight hours. Monitoring the fatigue of individual military aircraft is necessary to quantify the loads experienced by each aircraft as these can show great variation. Once the total fatigue experienced by the airframe can be determined, an informed decision can be made regarding safety and maintenance. Initial design calculations and predicted flight fatigue can be compared with the actual fatigue to ensure that the aircraft can be flown safely, and with adequate flight clearance for the airframe. It is the aircraft’s use that compromises fatigue life, not so much the number of hours in use. For aircraft the fatigue life will depend on several factor including • • • • • •

Flight role – e.g. Fighter, Trainer, Recon, Carrier etc. Number of take-offs and landings Payload weight and distribution Environment – hot, cold, wet, dry, salty, quality of runway Different pilot flying practices Servicing intervals

TRADITIONAL SYSTEMS Traditionally, OLM systems have been tailored for each aircraft type based on what specific measurements and analysis tools are perceived to be required at the time. In most cases the specifications are either vague (leaving the potential suppliers to make calculated guesses as to what might actually be required) or so specific that there is little flexibility allowed in the system design. The final result typically ends in a ‘black box’ system with bespoke hardware and software that might only ever be used on one specific aircraft type. Qualification of both

OLM: A hands-on approach

1201

hardware and software are expensive, flexibility is low and maintenance costs are high which often result in a poor return on investment. The time taken to debug and qualify the system means that many systems become obsolete before even getting into service.

COTS SYSTEMS Increasingly OLM users are looking towards Commercial-Off-The-Shelf (COTS) solutions in order to take advantage of new acquisition, processing and data storage technologies. Fielding [ ii ] states these benefits include • Cost – can easily be a factor lower • Future Proof –new technologies and upgrades can be integrated at low risk • I/O – a wide range of avionics interfaces available; support for future standards • Flexibility – expandable and programmable, as the needs or flight profiles change the system configuration can be changed to suit • Cost of Ownership – spares are standard product; production equipment lead times are lower • Environmental qualifications: Mil-Std-810/461 are already a baseline • High quality software that is freely available Adaptability can be particularly important. There is no guarantee that an aircrafts’ role may remain the same in the future and it may be necessary to alter the system. The development of COTS OLMS solutions follows the work in other aerospace fields. Flight test is always the most challenging due to the heavy demand for highspeed sensor, databus and video data. The same equipment concept is easily adapted to other functions (ODR, FDAU etc.) – in the majority of cases the building blocks are identical, only the application is different. Table I presents the differences between some aerospace data acquisition fields. Flight Test OLM ODR FDAU Thousands Thousands Hundreds Tens

Number of Parameters 1 – 100 sps Data Rates 50K to >2M sps Onboard Onboard Data and Storage telemetered Real-Time Post flight Data Processing and Post flight

>10Ksps

HUMS Tens

Onboard

1– 10sps – 10Ksps 100sps Onboard Onboard

Post flight

Post flight

Post flight (Realtime may reduce data)

Table I: The Data Acquisition Requirements of Aerospace Applications

Stephen Willis

1202

EMERGING TECHNOLOGIES New technologies, such as higher density micro-electronic chip fabrication, have allowed the size of electronic parts to shrink, while at the same time enabling vastly increased functionality, examples being computer processors, solid-state memory devices and Field-Programmable Gate Array logic devices. These new chips have allowed the size of DAUs to shrink from a system that occupied half a rack of equipment to a DAU that occupies a fraction of the space, while at the same time improving performance, reliability and functionality. Many common systems in use today quote typical analog accuracies of 0.25% to 0.5%. Next generation signal conditioning and acquisition hardware is available that offer greatly improved accuracies of 0.1% over the entire industrial temperature range (-40°C to +85°C). Time is an important factor for the correct synchronization of data from busses, sensors and other DAUs to be correlated in a distributed system. Now Global Positioning Systems (GPS) facilitate very precise time information that can be ‘stamped’ onto data to ensure data coherency for later analysis – even between different craft in a fleet. Today’s solid-state cameras are small and rugged enough to be placed in restricted space areas or outside the aircraft with little modification to the aircraft structure. A picture provides so much more information than an open/closed status switch, or a flap angle setting switch. The necessary PC processing tools for this type of video compression are readily available, in many cases distributed through the public domain – hence basically free. It’s now clear that the availability of large high speed memory chips mean that data as well as video can now be stored simultaneously, over many hours of flight, at an affordable cost. Back in the 40s and 50s, using a camera to record the instrument readings periodically was often one of the only sources of instrumented information available to the maintenance crews on the ground [ iii ]. It is ironic that it has taken this long for digital techniques to make it feasible again.

OPEN STANDARDS In traditional FTI systems, the Master is responsible for synchronizing the Slave DAUs and gathering data from them for storage and transmission as PCM. The Master-Slave configuration and the PCM data format are highly deterministic so it is possible to know in advance exactly how long it will take for a sampled parameter to reach its destination. However, this renders the system inflexible. It is difficult to make configuration changes to a slave DAU without those changes having ramifications and potential conflicts with the master DAU. Changing even a single configuration sometimes requires stopping the data acquisition process and restarting the device with the configuration change. Furthermore, the technologies

OLM: A hands-on approach

1203

and protocols used in the Master-Slave topology are not fully standardized – creating integration and deployment challenges for multi-vendor systems. The benefits of open standards include a broader user base, which in turn leads to better support and on-going development, and being freed from proprietary hardware and software. In recent years, there has been a shift from proprietary and closed solutions towards more open standards-based systems using Ethernet technology in particular. These provide a standardized open common technology platform enabling fast data transfer, data mining rates, greater interoperability between devices and ease of connectivity. Since Ethernet is the most important and common data transmission technology in the world, it is future proof as a core technology as standardized protocols and technologies based on Ethernet are in a continuous state of development. Networked device autonomy results in a lower cost of ownership (e.g. common spares and commonality across aircraft types) and lower maintenance costs. The IEEE-1588 Precision Time Protocol (PTP) is one of the more important standard network technologies. Before its development, the time synchronization accuracy of the standard Network Time Protocol (NTP) was insufficient to meet the timing requirements of networked FTI. By using the PTP protocol, all distributed DAUs in a networked FTI system may be synchronized to within 100ns. Networking tools such as Ping, trace route, and packet sniffers may be used to debug and analyze the system performance without interrupting data acquisition. Asynchronous and ad hoc protocols may be used to dynamically control and interact with the various networked devices. For example, the Simple Network Management Protocol (SNMP) may be used to dynamically query and configure settings in the DAU without requiring the device to be restarted or rebooted. COTS software is generally highly programmable and configurable to suit a wide and varied market. Like the airborne hardware manufacturers, software suppliers maintain a continuous development plan, using state of the art programming techniques and languages. The software/hardware interface requires a common meta-data standard through which all vendors describe their hardware configuration and setup for multi-vendor deployments. In order for a meta-data standard to be useful it must meet a minimum set of criteria i.e. it must model the domain effectively, be adaptable, vendor independent, extendable and based on XML [ iv ]. One such meta-data standard is XidML – an open meta-data standard that describes the entire FTI domain from the configuration of acquisition hardware to ground station analysis. Storage Formats Rapidly advancing technologies in solid state media can permit storage of high bandwidth data in real time at an affordable cost. This precludes the use for complex data reduction processes in flight thus improving system reliability and qualification costs. Another advantage of this approach is that it preserves all raw

Stephen Willis

1204

data enabling detailed analysis to be performed, post flight, by the many commercially available statistical and structural data analysis programs.

Figure 1: A CF Ethernet Recorder and A Recording Module For The KAM-500 Solid-state recorders are now widely used in the industry as their size goes up and their cost down due to demand in commercial electronics. There are no moving or mechanical parts, making them far less susceptible to vibration and/or failure. The media, particularly in the case of the CF cards, is very small (almost too small!), allowing the recorder to be small as well, reducing weight and space requirement. For example, the ruggedized Pretec 100GB card supports a write speed of 35MBps and is housed in a metal enclosure making it 10 times more durable and resistant to impact and shock over a standard CF card [ v ]. By combining the use of CF technology with an efficient file format, like PCAP, months of uninterrupted, continuous recording is possible. For example, Figure 2 shows that a 64GB CF card can hold approximately 1086 hours of data of 8000 samples per second (i.e. 500 samples per packet with 16 packets recorded per second).

Figure 2: Hours of Recorded Packetized Data

OLM: A hands-on approach

1205

AIRCRAFT CASE STUDIES The cross-industry adoption of new technologies offers many benefits for military, and civilian, aircraft over their life time. The follow section illustrates some of these with four case studies conducted by ACRA CONTROL. The planes in question are the Shorts Tucano, the E-3D Sentry, the C-130J and the T-38 trainer. Shorts Tucano The Shorts Tucano is a trainer for the Royal Air Force that underwent a program to extend the useful and safe life of the aircraft as economically possible. The trainer aircraft were originally fitted with a ‘g’ counter, which could only provide a very course view of the forces being experienced by the aircraft. Loading on individual components could only be inferred. The first aim of the program was to provide more detailed data for actual flights that could be used to validate, and if necessary modify the flight clearances provided by the first fatigue test. The second aim of the program was to enable a follow-on ground fatigue test to mimic the actual fatigue being experienced by the flying trainer aircraft. This would help better understand the maintenance requirements of the aircraft and to help extend the flight clearance of the airframe – possibly extending its life in the long term. It was decided that a total of three aircraft from the fleet would be instrumented for the purposes of the program. Aircraft Instrumentation and DAS. Each aircraft was fitted with a DAS utilizing KAM-500 hardware. Each DAS consisted of two chassis and a number of plug-in user configurable modules (up to 16 in certain configurations). The DAS was fitted in an instrumentation bay located behind the cockpit. The aircraft were instrumented with a number of sensors to measure the desired parameters, as shown in Table II. Strains 12 on the wings

Aircraft Parameters Airspeed

6 on the tail-plane 3 on the fin

Altitude Elevator Aileron, Rudder Angles Tail, Port and Normal G, acceleration Flap position Compass heading Engine Torque Fuel

4 on the fuselage 9 on landing gear

Discrete Signals Undercarriage Position Air-brake Weight on Wheels Radio Transmit Compass Flag Pilot Event Switch Start/Run/Stop Switch Engine Speed Trend Switch

Table II: Parameters Acquired by the Trainer Aircraft DAS

1206

Stephen Willis

Each strain gauge was calibrated using a static loading mechanism. The set of calibration coefficients that were generated were used later during the analysis to convert the combined strains of several gages directly into the bending/torsion/shear forces being experienced by the airframe. The DAS provided excitation for all the sensors, including the strain gages, accelerometers, and synchro sensors. Signals were also acquired directly from the aircraft itself in the form of voltages, discrete signals and from the avionics bus. The sampled data was then stored on a solid-state flash memory module. The DAS also included a battery backed real-time clock to provide time stamping of the data. Some of the sensors used in the program can be seen in Figure 3.

Figure 3: The Distribution of Data Collection in the DAS Data Gathering and Analysis. The data was downloaded to a PC after every flight (as the requisite personnel were always on-hand to do so) and then sent for processing. Over the course of the program the DAS proved itself to be reliable and robust, exceeding the targets set for the system reliability. Only a small percentage of the sorties resulted in unusable data due to a problem with the DAS or human error. The achieved reliability was of the order of 96-97%. Over a two year period approximately 1,200 instrumented flights were carried out making this one of the most extensive structural test programs carried out on in-service aircraft in the UK in terms of the number of aircraft, number of sorties, data gathering and analysis performed. The data from each flight was put through a set of custom routines that had been developed and validated over a period of several months by the design authority in order to automate the data processing task. The analysis calculated the loading of specific structures within the airframe during a flight. This loading information was processed using data reduction methods such as S-N curves and rain-flow algorithms to derive fatigue spectra for the undercarriage, tail, wings and the fuselage for the given sortie type. These spectra served two purposes. The fatigue spectra from the flights were combined to create a fatigue index for the aircraft. Second, the spectra formed the basis for the test stimuli to be used during the ground based fatigue test.

OLM: A hands-on approach

1207

A production airframe had been set aside, and this was instrumented in a similar manner to the aircraft and placed into a fatigue rig. The rig consisted of forty-two hydraulic jacks attached to the airframe, and instrumented with linear variable differential transformers (LVDTs) to monitor their displacement, making sure safe limits were not exceeded, and to ensure the jacks are operated in phase so that the airframe is not shaken apart. When the fatigue rig is ‘flying’ each of the hydraulic jacks is driven according to a specific program derived from the combined fatigue spectra in order to replicate the loads seen during the different sorties. Several months were spent refining the jack programs and aircraft setup to ensure that the jacks were applying the correct loads to the different elements of the airframe. About 1000 sorties can be ‘flown’ over about five weeks of continuously running the rig with minor inspections after every month of operation and major inspections every third month. Due to the significant time taken to perform inspections, more time is spent inspecting the airframe than flying the fatigue rig. Different scatter values were applied to elements of the airframe, typically 5, meaning that the fatigue rig must perform 5000 sorties for it to be considered representative of 1000 flown sorties. There was good confidence that the testing done was representative of the fatigue being experienced by the flying aircraft. For example, a crack found in the airframe during one of the inspections corresponded to a known problem that had been observed in the flying fleet. In theory, future crack propagation and component life predictions can be made. The monitoring systems are still fitted to the aircraft, and maintaining the instrumentation on the aircraft has had an ongoing beneficial side effect. The DAS are still in occasional use today when there is a need to provide test data for design modifications made to the landing gear or tail for example. As the DAS and instrumentation are already installed and calibrated it has meant that it is comparatively quick, easy and inexpensive to evaluate the impact of any design changes. A lot of time and money has been spent with the intent and purpose of ensuring that the trainer aircraft are flying safely and are maintained in a flight worthy condition through a well managed maintenance program. From a flight safety point of view the benefits are simple, providing actual flight data that can be compared against the design time predictions and assumptions, allowing a check to be made that aircraft is being flown within its design limits. E-3D Sentry L/ESS The Boeing E-3D Sentry is an Airborne Warning and Control System (AWACS) aircraft that provides all-weather surveillance, command, control and communications to local aircraft. The fleet of aircraft are fitted with a Loads and Environmental Spectra Survey (L/ESS) recording system. This system recorded airframe stress and other aircraft operational parameters on every flight. Data from all the aircraft are replayed into a ground processing system where the data is scaled and formatted before being sent for analysis in the US. Here the data from

Stephen Willis

1208

many other aircraft in use around the world is collected, correlated and analyzed thus forming a unique history of the all the aircraft’s fatigue and usage parameters over the fleet’s operational life. The analysis of such trend data readily indicates if there are some potential fatigue or maintenance problems on the horizon well before they become critical. The original system installed consisted of a Data Acquisition system and recorder that was no longer supported when the company changed its focus, thus a replacement was sought. ACRA CONTROL proposed a commercial off the shelf (COTS) solution using current hardware. The UK fleet are all flying with this updated system today, illustrated in Figure 4. The location and wiring of the sensors can be seen in Figure 5.

Figure 4: The Old and the new Systems Installed in an Equipment Tray

Figure 5: The Strain Gauge Locations

OLM: A hands-on approach

1209

The software was also legacy, thus it was upgraded to the latest Windows environment. This had the benefit of allowing the quality of the raw data to be assessed using the new Quicklook software before reformatting. A simple piece of conversion software was produced to transfer the data stored in flash memory from the KAM-500 to a suitable format for further analysis. The program was carried out successfully with some significant benefits resulting over the previous system including • • • • • • •

Weight – KAM-500 15Kg lighter Power – 50% lower Space – < one quarter Maintainability – standard product Supportability – spares, new modules available Future Proof – continuous development of technology Commonality – other aircraft use the same basic system building blocks

T-38 HUMS A HUMS program was undertaken on the Northrop Grumman T-38 – the worlds first and most produced supersonic trainer that remains in service today around the globe. The program had maintenance cost and safety aims including • • • • • •

Extending replacement interval for Critical Parts Extending/eliminating inspection intervals Extending Time Between Overhauls (TBO) Safety benefits Replacement of critical parts before they fail Detection of impending failures by HUMS (e.g. engine, etc.)

The program structure had two phases. The first phase was a six-month trial with the aim of proving that the KAM-500 system could acquire the data reliably. The second phase added direct structural monitoring. Phase one used three systems, each measuring Air Speed, Altitude, Vertical & Lateral Acceleration and Time Reference to flight time. There was onboard storage that was changed out every few months to the concept or measuring and recoding the data worked as expected. Once phase one was complete, an extra system was added and 24 strain gauges were instrumented to each of the four aircraft. In addition to the phase one parameters, a real time clock and an 8 channel RS-422/485 universal asynchronous receiver were added to each system. Of particular interest was the installation of a ‘Built in-Test’ card that could be used to examine the system so the engineers would know if everything was working as planned without needing to examine the system itself.

1210

Stephen Willis

The data was milked from the memory cards at regular intervals and transferred to MatLab for Analysis. The piecewise nature of this program is a good example of how a partial system could be installed to prove a concept, with the knowledge that expanding the range of sensors and other parameters was as easy as adding some extra plug-in modules. Lockheed C-130J The Lockheed-Martin C-130J is a military transport aircraft that is the main tactical airlifter for many military forces worldwide with over 40 models and variants. This is a good example of an aircraft that operates in several different roles in many different environments. The project undertaken was tasked with measuring strains from over 200 channels around a large airframe. A distributed system with a PCM based intercommunication protocol was used, although ACRA CONTROL would advice-using Ethernet for new projects. The system from ACRA CONTROL comprised six KAM-500s installed on the aircraft. The master was installed on the ‘centre wing box’ with a removable memory module. This was used to record the system’s data. A KAM-500 module provided this functionality allowing easy access to the data for transfer to the analysis machines.

Figure 6: The KAM-500 System Installed on the C-130J Centre Wing Box The system monitored an extensive network of strain gauges as well as analogue and digital signals and the aircraft bus as seen in Table III.

OLM: A hands-on approach

Airspeed Pressure Altitude Fine Radio altitude (Low Range) Aileron (Left Hand) Rudder Position Fuel Flow Rate (4 Parameters) Engine Torque Number 2 Flap Position (Left Hand) Pitch Attitude Pitch Rate

1211

Cabin Differential Pressure Fuselage Total Fuel Weight Mach Number Climb/Descent Rate Pressure Altitude Coarse Radio Altitude (High Range) Elevator Position Aileron (Right Hand) Flap Position (Right Hand)

Engine Torque Number 1 Condition Lever Position Pitch Trim Deflection Roll Attitude Roll Rate Normal Acceleration at Centre Lateral Acceleration Angle of Attack Ground Speed

Table III: The Parameters Measured by the C-130J DAS

Another C-130 HUMS application ACRA CONTROL undertook was on the HC130H – a US Coast Guard variant that is used for applications such as search and rescue. The Coast Guard required a data acquisition system to help validate baseline spectra and stress/load ratios and to record environmental data during flight operations for future comparisons e.g. correlation to environmentally accelerated cracking and general corrosion. Both digital and analog signals were fed into an ACRA CONTROL data acquisition system where A/D conversion and signal conditioning were performed. This data was then saved to a removable CF card for transfer into the client’s suite of programs for analysis. The program was successful in validating their concepts and the value of the process.

INTO THE FUTURE New ‘smart sensors’ are becoming available that perform the sensing and measurement of the signal at the sensor location, typically providing a serial output allowing large numbers of sensors to be digitally read with only a few wires – vastly reduce wiring requirements [ vi ]. Wireless, self powered sensors could simplify this further [ vii ] while nanotechnology, microelectromechanical systems (MEMS) and computational advances like artificial intelligence [ viii ] and distributed computing could eventually lead to self repairing and ageless structures [ ix ]. However, it is essential for their continual evolution that open system architectures are used [ x ].

1212

Stephen Willis

There has been much research in recent years in the development of open standard, off-the-shelf mobile wireless IP technologies. These technologies are advancing at a rapid rate allowing for higher data rates, greater coverage distances and higher mobile velocity support. For example, the Worldwide Interoperability for Microwave Access (WiMax, IEEE 802.16E standard released in 2005) is a bidirectional IP-centric technology that allows for coverage distances of up to 50km or data rates of 70Mbps [ xi ]. Similarly the Mobile Wireless Broadband Access (MBWA, IEEE 802.20 standard released in 2008) allows for low latency (<20ms), lower rates at 1Mbps and ranges less than 15km. In the short to medium term, PCM/FM links may be replaced with IP-centric wireless technologies. There has been much development of bi-directional IP centric mobile wireless technologies that should provide greater bandwidth over the air-to-ground links and more interactivity with the onboard network. In the future, a fully end-to-end networked solution will open the gateway to recording and analysis systems where data can be analyzed anytime and from anywhere. By taking advantage of the emerging wireless technologies, a new generation of network based telecontrol services is possible. The Real-Time Streaming Protocol (RTSP) is a network protocol designed to facilitate data streaming [ xii ]. RTSP is analogous to a TV remote control with standard PLAY, PAUSE, and time-based requests that allow the user to request and play back the desired data stream for any time window of interest if the data exists in the recorder. The RTSP protocol only describes the request-response mechanism between the client analysis station and the on-board RTSP server. RTSP is completely independent of the protocols used to transmit the requested data to the client. For example, consider a networked multi-flight data acquisition system with a packetized stream called 0xAAAA in which each packet contains N time-stamped samples of the measured left-wing vibration. RTSP can be used to request this data from the IP recorder for any time interval of interest including the current time. Moreover, RTSP allows for multiple requests to be transmitted in parallel since there is an RTSP server inside the IP recorder that handles all these. Using an RTSP server embedded in the recorder ensures the read-while-write functions are available. Read-while-write is the ability of the recorder to continue recording and writing data whilst in parallel it can satisfy ad hoc asynchronous data mining read requests. In addition to mining data, it is possible using standard network protocols to remotely configure and manage networked DAUs. This may be beneficial for multi-flight recording systems whereby the recording requirements may change over time. The Simple Network Management Protocol (SNMP) is a simple but powerful utility that can be used to telecontrol or configure (SET) and query (GET) devices for key information on-demand [ xiii ]. In addition, a device can be configured to set a trap for key events (TRAP), for example, send an SNMP event message when the temperature of the device exceeds some set threshold. This

OLM: A hands-on approach

1213

trapped event can then be directed to an analysis station monitoring such events or equally it could be recorded on the IP recorder for post-processing analysis.

CONCLUSIONS The use of common configurable COTS hardware and software technologies available today can drastically reduce the cost of purchase, cost of ownership, support, reliability and maintainability of OLM requirements. They also offer the possibility to alter the systems application by virtue of their modular nature and the commonality of hardware. Despite being smaller and lighter, they can contain more computational power and storage capacity. Open standards have many advantages over proprietary systems due to increased flexibility, longer lifetimes, reduced cost and future multi-vendor support. The case studies highlight some of these advantages, in particular how systems can stay in use for long periods of time and be adapted, but only when they can be maintained and/or upgraded. The future prospects of fully integrated, smart systems are exciting with the potential for more granular data with positive implications for flight safety and maintenance programs. However, these may be some time off and are likely only in new airframes. In particular the aircraft industry is suffering at the moment, like most sectors, from the current global recession [ xiv ]. Many aircraft operators, military and civilian, are keen to extend the existing lifetime of their current fleet. COTS solutions offer both current affordable ways to do this and the ability to be upgraded or integrated with next generation systems as they are adopted. Integrated system will still use line replaceable units (LRUs) and likely over an Ethernet bus as these are being introduced on next generation airliners like the Boeing 787[ xv ].

REFERENCES i

Virginian-Pilot, (2008), "Inspections Ordered for F/A-18 Hornets", Online http://www.military.com/news/article/inspections-ordered-for-fa18-hornets.html ii Fielding R., (2005), “COTS Approach To Dynamic Flight Data Acquisition For Hums/Olm Applications”, In proceedings of Eleventh Australian International Aerospace Congress, Melbourne, Australia iii Strääf, K., (2008), "History of FlightTest Instrumentation at Saab Flight Test and Verification", Online, http://www.aerospacetesting.com/files/straaf.pdf iv Cooke A., (2008), “Integrating Heterogeneous Systems in an FTI Environment”, Proceedings of the International Telemetry Conference, 2008, San Diego, USA v Pretect, (2007), “Product Press Release”, Online: 27th January 2009, http://www.pretec.com/epages/Store.storefront/?ObjectPath=/Shops/Store.Pretec/P roducts/%22news-September%2023%2C2008%22

1214

Stephen Willis

vi Bradley M. and Maguire K., (2004), “Operational Loads Measurement: What’s been done and where it’s going”, European Telemetry Conference Congress and Aircraft Integrated Monitoring Systems, May 24 - 27, 2004, Heidelberg, Germany vii Lieven, N.A.J., Escamilla Ambrosio, Burrow P.J, S. and Clare L., (2007), “Strategies for Wireless Intelligent Sensing Devices (WISDs)”, AIAC-12 Twelfth Australian International Aerospace Congress Fifth DSTO International Conference on Health & Usage Monitoring – HUMS2007, Melbourne, Australia viii Boakes S., (2007), “Expanding Applications, Data Management Technologies and Benefits of HUMS”, In Proceedings of the Twelfth Australian International Aerospace Congress Fifth DSTO International Conference on Health & Usage Monitoring, Melbourne, Australia ix Price D. C., Scott D. A., Edwards G. C., Batten A., Farmer A. J., Hedley M., Johnson M. E., Lewis C. J., Poulton G. T., Prokopenko M., Valencia P. and Wang P., (2003), "An Integrated Health Monitoring System for an Ageless Aerospace Vehicle", 4th International Workshop on Structural Health Monitoring, Stanford University, USA x Parekh D. and Sinha A., (2007), "Recent Applications of Health and Usage Monitoring Systems to Rotorcraft – A Survey", AIAC12 – Twelfth Australian International Aerospace Congress, Australia xi IEEE, IEEE Std. 802.16e-2005, (2005), “IEEE Standard for Local and metropolitan area networks Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems”, Approved December 2005 xii IETF, (2009a), RFC 2326, “Real Time Streaming Protocol”, http://www.rfceditor.org/rfc/pdfrfc/rfc2326.txt.pdf Online: February 2009 xiii IETF, (2009b), RFC 3416, “Version 2 of the Protocol Operations for the Simple Network Management Protocol”, http://www.rfceditor.org/rfc/pdfrfc/rfc3416.txt.pdf Online: February 2009 xiv Commercial Aviation Online, (2009), “Fleet Watch: Year 2008”, Online, Available from: http://www.flightglobal.com xv Flight International, (2006), “Boeing 787s to be fitted with enhanced data recorders”, Online, Available from: http://www.flightglobal.com

25th ICAF Symposium – Rotterdam, 27–29 May 2009

F-35 JOINT STRIKE FIGHTER STRUCTURAL PROGNOSIS AND HEALTH MANAGEMENT AN OVERVIEW Fallon, Timothy1; Mahal, Devinder2; and Hebden, Iain3 1

JSF Program Office, 2 JSF Program Office, 3 BAE Systems

Abstract: The three variants of the Joint Strike Fighter aircraft pose unique problems in designing and implementing and operating a structural health management system that will ensure structural integrity of the aircraft while allowing it to reach its design goals for service life. Airworthiness requirements from the US Air Force, NAVAIR and our partner countries have to be considered when designing the system. The JSF will have a state of the art onboard and off board system to facilitate structural prognostics and health management of the JSF fleets. This presentation will provide an overview of the system and will discuss how the design challenges are being addressed. Structural health management of a fleet is the embodiment of bridging the gap between theory and operational practice. New onboard sensors, more data collection than ever before, latest fatigue damage and crack growth algorithms, corrosion monitoring and prediction, unprecedented interface with operational and maintenance data - what does it all mean to the maintenance officer trying to keep his jets in the air, or to the fleet manager in planning future deployment of the "best" jets (from a structural life perspective). This presentation will discuss what the JSF program is doing to address structural health management and from these perspectives, and how prognostics will aid a Performance Based Logistics contract. Due to unforeseen circumstances, this paper could not be included in these proceedings. To obtain a copy of the presentation, please contact the first author on [email protected]. M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1215. © US Government 2009. Created within the capacity of an US governmental employment and therefore public domain. Published by Springer Science+Business Media B.V. Dordrecht.

Fatigue life enhancement methods and repair solutions

25th ICAF Symposium – Rotterdam, 27–29 May 2009

BONDED REPAIR FOR FUSELAGE DAMAGE: AN OVERALL BENEFIT TO COMMERCIAL AVIATION D. Furfari1, H.J.M. Woerden2, R. Benedictus3, A. Kwakernaak4 1

Airbus – Structural Analysis Stress Methods, Research and Technology 2 Shell Chemicals Europe – Engineering Integrity Department 3 Delft University of Technology – Aerospace Materials Department 4 Delft University of Technology – Adhesion Institute

Abstract: Adhesively bonded repairs of aircraft structures are frequently used in military applications, especially at the Australian and United States Air Forces. Although bonded repairs have the capacity to provide a maintenance-free repair solution to a variety of structural damages when compared to conventional mechanically fastened repairs, commercial operators still only use them on a very limited scale. In this paper the fatigue behaviour of bonded repairs to aluminium fuselage skin structures are discussed. A fatigue test campaign with increasing complexity of the test specimen (“pyramidal approach”) was carried out. Fatigue test results of cold and hot bonded repairs were compared to those of conventional riveted repairs. The results of the coupon specimens were used as input for design of the next specimen type in the “pyramidal approach” by selecting the combination of the parameters showing the most promising and most interesting results. It was found that the amount of cycles to initiate a fatigue crack in bonded repair solutions on the coupon specimens strongly depends on the adhesive type used. The fatigue lives of cold bonded coupon specimens were not longer than that of the reference riveted repair. The hot bonded coupon specimens showed a significant enhancement in fatigue performance with fatigue lives up to 6 times longer than the conventional riveted repair. M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1219–1243. © Springer Science+Business Media B.V. 2009

1220

D. Furfari, H.J.M. Woerden, R. Benedictus and A. Kwakernaak

INTRODUCTION The operational costs to maintain in service aircraft, with the required level of safety, increase with the increasing of the time that these aircraft remain in service [1]. In-service repairs on a primary structure of an aircraft are usually performed by the airlines according to the fuselage skin repair solutions provided in the Structural Repair Manual (SRM). Existing repair solutions, contemplated in the SRM of each in-service aircraft require the use of the same material for repair patches (here after called doublers) and the fuselage skin to be repaired. These solutions could be improved and developed to extend the fatigue crack initiation and the crack growth lives (i.e. reduce the operation costs). Advanced repair solutions, including adhesively bonded repairs, have the capacity to provide a maintenance-free repair solution when compared to conventional mechanically fastened repair according to the current SRM. In the frame of the EU project IARCAS, fatigue behaviour of advanced repair solutions, both mechanically fastened and adhesively bonded, have been investigated [2]. The experimental activity of the advanced fastened repairs is described in detail in paper [3]. Adhesively bonded repair of aircraft structures is frequently used in military applications, especially with the Australian and United States Air Forces [4]. Although bonded repairs have the capacity to provide a maintenance-free repair solution to a variety of structural damages when compared to conventional mechanically fastened repairs, commercial operators still only use them on a very limited scale. In this paper the fatigue behaviour of bonded repairs to aluminium fuselage skin structures are discussed.

EXPERIMENTAL WORK Bonded repair doubler manufacturing steps Two types of environmental conditions during application of bonded repairs have been considered: the “production” environment meaning a bonded doubler with surface pre-treatments and bond line curing conditions as will be used in production environment (Grit-blast Silane, autoclave); and the “repair” environment as the one typical when a repair is applied to an aircraft in-field (SolGel® surface pre-treatment [5], vacuum bag and heat blanket). The bonded repair solutions for the coupon level, small flat panel level, and curved pressurized panel level test programme all share a more or less common manufacturing process of three distinct phases: surface treatment (normal production), surface pretreatment (preparation for bonding), and repair doubler bonding. The steps below are taken for both the skin panel(s) and the doubler on the sides and for the area to be bonded, except in the case of Glare® doublers where only a pre-bonding solvent cleaning step is taken (outer surface of Glare® is primed). The three phases in the manufacturing process were: surface treatment (for both specimen types),

Bonded repair for fuselage damage

1221

composed of solvent cleaning, pickling in tank, phosphoric acid anodizing in tank, primer application and curing; surface pre-treatment (for repair type specimen), composed of removal of primer layers (Figure 1a) and Sol-Gel application (Figure 1b); doubler bonding (for repair type specimen), composed of film adhesive application (Figure 1c), heat – blanket application (Figure 1d),vacuum bag buildup and vacuum bag – heat blanket cure (Figure 1e).

Figure 1 Preparation of bonded large flat panels, repair type specimen For the coupon specimens two adhesive systems have been investigated: the herein called cold bonded solution using Vantico® 1590 adhesive and the hot bonded solution using Cytec® FM73M.06 adhesive. For all the remaining types of specimens (small and large flat panels and the curved stiffened panels) the used bonded repair system was the hot bonded one, due to the unsatisfying performance of the cold bonded repairs in coupon specimen tests, described in the next section.

1222

D. Furfari, H.J.M. Woerden, R. Benedictus and A. Kwakernaak

Fatigue test programme The fatigue testing programme is based on the test pyramid approach shown in Figure 2. The complexity of the test specimens increases as the number of test reduces.

Figure 2 Test pyramid Tests ranged from simple coupon specimens (butt-joints with structural details) to aircraft structural components (curved stiffened fuselage panels), in which both a conventional riveted repair and the hot bonded repair were embodied. Flat unstiffened panels with increasing cutout size were also tested using the different bonded repair solutions (i.e. small and large flat panels). All specimens were subjected to constant amplitude fatigue loading at stress ratio R=0.1. The applied loading conditions of the curved stiffened panels were internal pressure and longitudinal load, simulating in-service loading conditions of an aircraft fuselage. The fatigue life (until complete failure of the specimens) of the conventional riveted repair (herein called baseline) was used as reference to estimate the fatigue

Bonded repair for fuselage damage

1223

life enhancement for the bonded repairs. The influence on fatigue life of surface pre-treatment, environment (“production” or “repair”) as well as adhesive type (cold or hot bonding) was studied. The results of the coupon specimens were used as input for design of the next specimen type in the “pyramidal approach” by selecting the combination of the parameters showing the most promising and most interesting results. In viewing of the testing pyramidal approach, the fatigue test results, in terms of fatigue life until complete failure of the specimens, were compared to each other and the best performing solutions were selected to be implemented in the more complex specimen geometry. Coupon Specimens. Static testing. Static testing has been performed on one sample of the specimen series. Since the dimensional configuration of all adhesively bonded coupon specimens is identical and since no influence is to be expected from the different surface pre-treatment methods, the different epoxy adhesives used and the different cure methods used, only two of the bonded specimen designations were equipped with strain gauges (Figure 3).

Figure 3 Strain gauge locations on coupon specimen Fatigue testing. All bonded coupon specimens were fatigue tested (until final failure) in the Netherlands by Delft University of Technology at different stress

D. Furfari, H.J.M. Woerden, R. Benedictus and A. Kwakernaak

1224

levels ranging from 100 MPa c.a. to 170 MPa c.a. as maximum nominal stress (far field stress). Small flat panel specimens. Static testing. Strain gauge measurements have been performed for stepwise static load increases on the bonded small flat panel specimens. These tests give a clear picture of the load distribution over the specimen in question. Since the dimensional configuration of all adhesively bonded small flat panel specimens is identical and since no influence is to be expected from different surface pretreatment methods, only two of the bonded specimen designations were equipped with strain gauges. The strain gauge locations can be seen in Figure 4.

Figure 4 Default strain gauge locations for bonded small flat panel coupon Fatigue testing. Constant amplitude fatigue tests at different stress levels (from 90 MPa to 120 MPa as maximum nominal stress) were carried out to compare the fatigue initiation and crack growth life of the different bonded small flat panel configurations. Although the detection of crack initiation proved to be possible, it was still difficult due to fact that the fatigue cracks always initiate in the basic material right next to or just under the adhesive fillet at the doubler edges. However, since the thickness of the small flat panels (1.6 mm) is smaller than the coupons (2.0 mm), and since the small flat panels are much wider, cracks initiation detection (when missed on the doubler side of the panel) was at least possible after the crack had grown through the thickness. Large Flat Panel. The specimen had overall dimension of 3000 mm x 1300 mm (longitudinal x transversal) and 1.6 mm thick made in Al-2024 T3 Al clad. The skin contained a central cutout (140 mm high and 396 mm width, cutout corner radius: 10 mm) where a bonded repair has been applied. The repair doubler (533 mm width, 276 mm high and 1.8 mm thick), also made in Al-2024 T3 Al clad, was hot bonded

Bonded repair for fuselage damage

1225

(repair environment). The repair has been performed by Delft University of Technology – Faculty of Aerospace Engineering and the specimen was sent to the Structural and Material Laboratory of Aerospace Engineering Department in Pisa University in the final condition prior fatigue test. The maximum nominal stress applied was 100 MPa. Specimen was subjected to periodical inspections to detect fatigue damage, in areas of stress concentration and to monitor the growth of the discovered cracks. Curve Stiffened Panels. Two curved stiffened panels, representing a real fuselage structure, were tested. Each panel contained four different repair solutions (shown in Figure 5): a typical SRM repair (baseline repair), two advanced fastened repairs (support doubler and driven rivet, presented in ICAF 2007, [3]) and a hot bonded repair, of which the results are presented in this paper.

Figure 5 Curved panel with the four repair solutions. The phases of preparing the bonded repair solution are given in Figure 6.

1226

D. Furfari, H.J.M. Woerden, R. Benedictus and A. Kwakernaak

Figure 6 Curve panel - Preparation of bonded repair

Bonded repair for fuselage damage

1227

The repair locations in the panel were designed to have always both full load transfer conditions (repair within a frame bay) and the repair corner in the middle of a frame bay satisfied for each repair solution. The dimensions of the bonded repair were limited by the size of the heat blanket, but even if the bonded patch is smaller than other solutions, it can still be considered comparable to others, since care was taken to ensure equally loading of different patches as demonstrated by the strain gauge readings. Static testing. The tests were carried in IMA GmbH Dresden, Germany [13, 14], the test rig used for this investigation is shown in Figure 7. Details of the load introduction areas (longitudinal and circumferential direction) are also shown. After the fuselage panel was prepared and installed into the test rig three static tests with different load cases were carried out. •Static test 1: stress dependent on internal pressure (FL = 30 kN) •Static test 2: stress dependent on longitudinal load (pi = 0 mbar) •Static test 3: stress dependent on longitudinal load (pi = 593 mbar) The strain gauge measurements performed during the static test were used as finetuning of the load setting (internal pressure, longitudinal load and frame force) for the fatigue test. Both panels were equipped with a large numbers of strain gauges: 1 rosette, 15 XY strain gauges; 76 single strain gauges (most of them installed back to back). Each repair solution had strain gauges in the surrounding area of the skin. Also frames and stringers had strain gauges at several locations in the panel. Fatigue testing. After the strain gauge readings the fatigue test was carried out.. The applied fatigue loading was internal pressure and longitudinal loading, simulating a similar in service loading conditions of a fuselage of an aircraft. Sinusoidal pressure load in phase with longitudinal load and frame forces were applied. The fatigue test loading conditions of the first panel were: pmax = 593 mbar, Flong-max = 250 kN and Fframe = 12 kN at R = 0,1. After 5 x DSG (Design Service Goal of the aircraft of which the panels represent the fuselage structure) no fatigue cracks were detected, then the loads were increased by 10% to decrease the testing time. The new load settings were: pmax = 650 mbar, Flong-max = 275 kN and Fframe-max = 12 kN at R = 0,1. The second panel tested was subjected to fatigue loads of pmax = 650 mbar, Flong-max = 275 kN and Fframe-max = 12 kN at R = 0,1 from the beginning of the test. The fatigue test of both panels was split into 2 phases. In the first phase the panel was tested till fatigue cracks occurred in the baseline repair and the support doubler solutions (more details in [3]). At the end of phase I it was necessary to repair (with large external doublers riveted to the skin as described in [3]) the “cracked” base line and the support doubler solutions to continue the test. The loading conditions after the large repairs have been embodied remained the same (pmax = 650 mbar, Flong-max = 275 kN and Fframe-max = 12 kN) as confirmed by additional strain gauge readings.

1228

D. Furfari, H.J.M. Woerden, R. Benedictus and A. Kwakernaak

Figure 7 IMA GmbH Dresden, Germany test rig, load introduction lug attachments detail also shown

Bonded repair for fuselage damage

1229

RESULTS AND DISCUSSION Coupon Specimens The strain gauge test results for the two statically tested adhesively bonded specimens (cold bonded and hot bonded) are shown in Figuure 8. From the figure it can be seen that there is hardly any influence between two different bonded conditions (cold and hot) concerning strain distribution. When looking at the strain gauges 1 and 5, as well as strain gauges 4 and 8 (strain gauge map shown in Figure 3), it becomes clear that both in the skin plates as well as in the doubler the amount of secondary bending is significantly higher than in the baseline riveted coupon. This shows that the epoxy adhesive connection created is even stiffer than the fastened connections. The rest of the strain gauges on the bonded coupons show the general load introduction from skin panel into repair doubler (strain gauges 2, 6, 3 and 7).

120

140

SG1 (full symbol) SG5 (empty

Remote applied stress (MPa)

Remote applied stress (MPa)

140

100 80 60 40

Hot bonded repair

20

Cold bonded repair

SG2 (full symbol) SG6 (empty symbol)

120 100 80 60 40

Hot bonded repair

20

Cold bonded repair

Baseline repair 0

0

-0,05

0,00

0,05

0,10 Strain (%)

0,15

0,20

0,25

0,00

140 Remote applied stress (MPa)

Remote applied stress (MPa)

SG3 (full symbol) SG7 (empty symbol)

100 80 60 40 Hot bonded repair

20

Cold bonded repair

0 -0,02

0,10

0,15

Strain (%)

140 120

0,05

SG4 (full symbol) SG8 (empty symbol)

120 100 80 60 40 Hot bonded repair Cold bonded repair Baseline repair

20 0

0,00

0,02

0,04 Strain (%)

0,06

0,08

0,10

-0,05

0,00

0,05

0,10 Strain (%)

0,15

0,20

0,25

Figure 8 Strain form strain gauges during static test on adhesively bonded but-joint specimens To show the difference between the baseline solution and the bonded solutions, all fatigue life results are assembled in Figure 9. For ease of comparison, the fatigue life results of all coupon specimens are translated to a fatigue life factor. The fatigue life factor of the baseline riveted coupon is set to unity. This permits to see if improved solutions show enhancement of fatigue life on the coupon level

D. Furfari, H.J.M. Woerden, R. Benedictus and A. Kwakernaak

1230

compared to baseline riveted solution. Figure 9 shows clearly that bonded solutions are indeed improvements in terms of fatigue life compared to the riveted baseline solution. It can be seen that using a bonded solution can even result in an increase of up to 4.5 times the riveted baseline fatigue life, again depending on the fatigue load level during the experiment. There is also no significant difference in fatigue behaviour between production type (bonding directly on primer layer) or repair type (removal of coating, primer and small percentage of anodizing layer) manufacturing, for cold bonded specimens.

IF=interference fit, FP=flap peening, CW=cold working, HB=hot bonded, CB=cold bonded, P=production, R=repair 47,5

σamplitude

(MPa)

Riveted Reference, Al-Al Hi-Lok,IF Hi-Lok & Shim,IF Hi-Lok,FP,CW,IF Hi-Lok,CW,IF PAA,BR127,Gritblast-Silane,HB,R PAA,BR127,HB,P CAA,BR127,CB,P CAA,BR127,HB,R CAA,BR127,CB,R CAA,BR127,HB,P

57

76

0,0

1,0

2,0

3,0

4,0

5,0

Fatigue Life Factor (normalized to riveted reference) ..

Figure 9 Fatigue life factors of all improved mechanically fastened and bonded coupon solutions in comparison to the riveted aluminium baseline coupon (normalised results) A striking result in Figure 9 is the behaviour of the hot bonded in-field repair specimen with CAA pre-treatment. For the high stress levels the fatigue life of this coupon is similar to the poorly performing cold bonded specimens. For the lower stress levels its fatigue life is longer than that of the cold bonded specimens, and even approaching the fatigue life of its hot bonded counterpart. This behaviour can be explained by the repair surface pre-treatment (removal of coating, primer and some of the anodizing layer). It seems likely that more or the entire stable oxide layer formed during anodizing was sanded away, resulting in a poor surface for bonding and thus premature interfacial failure of the bond line. A last conclusion that can be drawn from Figure 9 is that the more environmentally friendly phosphoric acid anodising pre-treatment (PAA) performs somewhat better than the chromic acid anodising (CAA) pre-treatment.

Bonded repair for fuselage damage

1231

The failure mode and locus of each bonded specimen can fully explain the differences in fatigue lives between the bonded coupons, because of the fact that the strain gauge results were more or less identical for all bonded coupons (Figure 8). Therefore the explanation for the differences in fatigue life found cannot lie in the latter. Figure 10 gives an overview of the three different failure modes/locations found during the bonded coupon tests. The common failure location of the best performing bonded coupons was metal failure in one of the skin panels of the butt-joint specimen (left picture in Figure 10). Some coupons did fail in one of those skin panels, but prematurely due to positioning tape that was left behind before the bonding operation (middle picture in Figure 10), resulting in a larger than anticipated skin stresses at the doubler runout. Finally, those bonded coupons that show no fatigue life improvement compared to the riveted baseline (e.g. cold bonded coupons) all failed adhesively due to fatigue delamination growth along the pre-treated aluminium-adhesive interface.

Figure 10 Different failure modes and locations in bonded coupons; from left to right: skin plate failure adjacent to adhesive fillet, skin plate failure under adhesive failure, and bond line failure

Figure 11 shows that there is a considerable difference of failure modes between differently bonded coupons. The cold bonded coupons show significantly shorter fatigue lives compared to their hot bonded counterparts (lives comparable to baseline repair). The shorter fatigue lives of cold bonded coupons seem to be predominantly caused by the absence of a correctly bonded adhesive fillet between doubler and the skin plates, leading to high skin and adhesive stress at both doubler ends.

1232

D. Furfari, H.J.M. Woerden, R. Benedictus and A. Kwakernaak

80 HOT BONDED REPAIR COLD BONDED REPAIR

Stress Amplitude (MPa)

70

Base Line

60

50

40

30

20 Cycles to Failure

Figure 11 Fatigue life comparison between cold bonded, hot bonded and baseline repair

FEM studies have been carried out (using StressCheck® software) and it was found that the adhesive thickness and the adhesive phase length (length of adhesive fillet squeezed out after curing in hot bonded repair) are the main parameters that influence the stress distribution at the critical location (doubler run out). Figure 12 shows the non linear analysis results in terms of maximum principal stress distribution. The stress applied (far field stress based on skin gross area) was 100 MPa. The adhesive thickness and the adhesive phase length were 0.1 mm and 0.3 mm respectively. A detailed study was performed to determine the local stress at the skin for increasing values of the adhesive phase length. Two values of adhesive thickness were selected for this analysis (0.1 mm and 0.2 mm). Based on the experimental evidence that fatigue failure of most of the coupon specimens (the best performing in fatigue) occurred at the skin the analysis was focused on the stress distribution at the skin.

Bonded repair for fuselage damage

1233

Figure 12 Maximum principal stress distribution of coupon model with both skin and doubler in aluminum. Far field stress of 100 MPa, non linear analysis, adhesive thickness of 0.1 mm, adhesive phase length of 0.3 mm.

Figure 13 shows the maximum principal stress distribution at the skin of the bonded coupon specimen plotted versus the adhesive phase length. Both adhesive thickness models (0.1mm and 0.2mm) have shown a minimum local stress above a certain value of the adhesive phase length. The local stress is reducing considerably up a fixed value of the adhesive phase length and above this value there is not further reduction of the stress. This value resulted approximately 0.15 mm and 0.4 mm for the adhesive thickness of 0.1 mm and 0.2 mm respectively. The reduction of stress achievable with optimised adhesive phase length is up to 16% of the maximum stress for the 0.1 mm adhesive thickness model and up to 20% of the maximum stress for the 0.2 mm adhesive thickness model.

1234

D. Furfari, H.J.M. Woerden, R. Benedictus and A. Kwakernaak

Figure 13 Maximum principal stress distributions at the skin of bonded coupon specimen model for increasing values of adhesive phase length.

Small flat panel The most important strain gauge results for the hot bonded in repair environment specimens are shown in Figure 14 and compared to the baseline riveted small flat panel. Figure 14 shows that far away form the doubler there is almost no secondary bending, but more striking is the very low amount of secondary bending in the doubler. The by-pass loading (stresses flowing around the doubler) was again much higher for the baseline riveted solution. Furthermore, it can be seen that (except for the doubler strains) there is no difference in panel strains between the bonded solutions, thus the assumption that the surface pre-treatment has no influence on the stress and strain distribution in the skin panel was correct. A small difference (and a small amount of secondary bending in both bonded solutions) becomes apparent in the doubler strains, however, because of the two different doubler materials used (aluminium and Glare® 3-4/3-0.3). The solution with the aluminium doubler (triangular symbol in Figure 14) showed somewhat more secondary bending, while the overall strains in the solution with the Glare® doubler (round symbol in Figure 14) were slightly higher. Overall, the doubler strains of both bonded solutions were higher than those of the baseline riveted solution.

Bonded repair for fuselage damage SG1 (full symbol) SG2 (empty symbol)

120 Remote applied load (MPa)

Remote applied load (MPa)

120

1235 SG4 (full symbol) SG5 (empty symbol)

100

100 80 60 40

Scotch-Brite, Sol-Gel, Al Doubler

Grit blast, Silane, Glare Doubler

20

80 60 40 Scotch-Brite, Sol-Gel, Al Doubler

20

Grit blast, Silane, Glare Doubler

Baseline repair

Baseline repair

0

0

Remote applied load (MPa)

120 100

0,05

0,10 Strain (%)

0,15

0,00

0,20

120

SG6 (full symbol) SG7 (empty symbol)

80 60 40

Scotch-Brite, Sol-Gel, Al Doubler Grit blast, Silane, Glare Doubler

20

0,05

0,15

SG9 (full symbol) SG10 (empty symbol)

100 80 60 40 Scotch-Brite, Sol-Gel, Al Doubler

20

Grit blast, Silane, Glare Doubler Baseline repair

Baseline repair

0

0 0,00

0,10 Strain (%)

Remote applied load (MPa)

0,00

0,02

0,04

0,06

0,08

Strain (%)

0,10

0,12

0,14

0,00

0,05

0,10 Strain (%)

0,15

0,20

Figure 14 Remote, patch and by-pass strains from strain gauges during static tests on small flat panel specimens; comparison between two bonded specimen designations and baseline riveted specimen Crack initiation times and fatigue crack growth lives for the bonded small flat panel specimens in comparison to the riveted baseline are shown in Figure 15. It can be immediately seen from the figure that there seems no real consistency in the results, which makes it difficult to draw conclusions about the bonded solutions in comparison to the baseline riveted solution. However, some conclusions can be drawn, although care has to be taken in doing so based on so few specimens. It has to be kept in mind also that the results for the bonded specimens depend largely on its direction of initial bending (convex or concave) after the high temperature cure cycle. This initial bending direction is also shown for each bonded solution in Figure 15. Concave bending (hollow on doubler side) in this respect had detrimental effect on fatigue initiation and growth; convex bending (hollow on panel backside) had a positive effect on fatigue initiation and growth. Convex or concave bending, however, was caused during the curing cycle phase and was totally arbitrary. In general it can be said that a bonded Glare® or aluminium doubler does seem to provide an improvement in fatigue crack initiation and growth lives compared to the riveted baseline. For both remote applied loads the aluminium baseline shows crack initiation before the majority of its bonded counterparts and in case of the highest applied loading, failure also occurs before it does in the bonded counterparts. A Glare® doubler in this respect was not performing as well as an aluminium doubler, even though the opposite was true for the fastened doublers on the small flat panels. This behaviour is expected in this

D. Furfari, H.J.M. Woerden, R. Benedictus and A. Kwakernaak

1236

case, because the advantage of the lower extensional stiffness of the Glare® doubler and the slightly higher bending stiffness is overruled by the lower coefficient of thermal expansion of the Glare® doubler compared to the aluminium doubler. This will introduce detrimental thermal residual tensile stresses in the small flat panel skin under the repair doubler, where no thermal residual stresses will develop in case of the aluminium doubler. Adding tapering in the loading direction to the edges of the Al doubler should theoretically have a beneficial effect on fatigue initiation and crack growth by gradually introducing the neutral line jump from skin panel to doubler and thereby lowering the amount of secondary bending and thus the stress intensity factor at the doubler runout.

convex convex Scotch-Brite/SolGel,HB, R, Al-Al

concave

Scotch-Brite/SolGel,HB, R, Al-Al

concave

Riveted Reference, Al-Al Scotch-Brite/SolGel,HB, R,Al-Al

convex

Gritblast/Silane,HB, R, Al-Al tapered Gritblast/Silane,HB, R, Al-Gl

σamplitude

(MPa)

54

40,5

concave

RUN OUT

convex concave Fatigue Life ..

Figure 15 Fatigue crack initiation (solid parts of bars) and fatigue crack growth of the bonded small flat panels. Baseline solution also shown

To show the differences between the baseline solution and the bonded solutions, all fatigue life (until final failure of the specimens) results are assembled in Figure 16. For ease of comparison, the fatigue life results of all small flat panel specimens are translated to fatigue life factors. The fatigue life factor of the baseline riveted coupon is set to unity. This permits to see which of the solutions show improvement in fatigue life compared to baseline riveted solution. Overall, the bonded repairs showed the amount of improvement of factor 2 or more.

Bonded repair for fuselage damage

1237

63

σamplitude

(MPa)

IF=interference fit, FP=flap peening, SD=Support Doubler, HB=hot bonded, R=repair

54

Riveted Reference, Al-Al Hi-Lok,FP, IF, Al-Al Hi-Lok,IF, Al-Gl Hi-Lok,FP, IF, Al-Gl Hi-Lok,IF, SD, Al-Al-Al Scotch-Brite/SolGel,HB, R, Al-Al Gritblast/Silane,HB, R, Al-Al tapered Gritblast/Silane,HB, R, Al-Gl Scotch-Brite/SolGel,HB, R, Al-Al Scotch-Brite/SolGel, HB, R, Al-Al

45 RUN OUT

40,5

0,0

0,5

1,0

1,5

2,0

2,5

3,0

Fatigue Life Factor (normalized to Baseline) ..

Figure 16 Fatigue life factor of the bonded small flat panel solutions and the fastened small flat panel solutions in comparison to the baseline riveted small flat panel solution (normalized)

Large flat panel The hot bonded doubler solution in repair environment, which was the best performing in the coupon test program (fatigue life improvement more than 4.5 times than the baseline solution), when applied to the large flat panel has shown a fatigue life more than 2 times longer than the fatigue life of the baseline repair. Figure 17 gives an overview of all the fatigue test results for the large flat panel including baseline, bonded and advanced fastened repair solutions. Where BL=BaseLine; BL+FP=BaseLine+Flap Peening; SD=Support Doubler; SD+FP=Support Doubler+Flap Peening; CB=CounterBore; DR=DrivenRivet and more details can be found in [3]. For the large flat panels some of the improved fastened solutions (e.g. flap peened and driven rivet) seem to show much better fatigue lives than their bonded counterpart. It should be taken into account that the bonded repair was simply a design copy of the fastened doublers. No effort has been put into designing the bonded doubler from scratch to optimise its performance such as adding edge tapering, non-rectangular design, overlap length and width, bonded repair of a crack instead of a cut-out, etc. In fact, as demonstrated with the small flat panels, tapering the runout of the repair doubler introduces favourable effect on fatigue life (Figure 15).

D. Furfari, H.J.M. Woerden, R. Benedictus and A. Kwakernaak

1238

Fatigue Life (Cycles)

RUN OUT

Crack Growth Life First Detection Crack started from corner cut-out and stopped at first rivet; No further crack detected; test suspended.

BL

BL

BL+FP

SD

SD

SD+FP

Hi-Lok

Hi-Lok

CB

DR

DR

Bonding

Figure 17 Fatigue test results of large flat panels. Figure 18 shows the average stresses (arithmetic mean between stresses measured on the external and the internal surface of the panel, strain gauges installed back to back) around the doubler of the baseline repair solution with 100MPa far field stress applied. An increase of the stress was monitored at the strain gauge position 13-14 (by-pass stress) and decrease of the stresses was observed in the doubler (lower stresses than the one introduced in the panel - far field stress).

Figure 18 Mean stress distributions around the doubler of the baseline solution repair with 100MPa far field stress applied.

Bonded repair for fuselage damage

1239

Figure 19 shows how the stresses flow around the repair and through it for the bonded repair solution. For this repair solution the by-pass stress is comparable to the stress introduced in the panel at the far field area (remote stress). The highest average stress was monitored at the skin of the doubler corner (position 9/10). This justifies the failure mode for the bonded repair with fatigue crack occurring as shown in Figure 20. This type of failure was typical for all the bonded repair solutions embodied in the small flat panels.

Figure 19 Mean stress distributions around the doubler of the bonded repair with 100MPa far field stress applied. Based on small flat panel tests, the repair corners (doubler and cut-out) have been identified as the most critical for fatigue crack initiation. The crack originated in the skin, at the run-out of the doubler, almost at the lower right angle (looking the panel front side). Indeed, at the very beginning, it ran parallel to the adhesive squeeze-out and therefore it was more clearly detectable from the rear side of the panel, Figure 20.

Figure 20 Bonded large flat panels crack start (back and front)

1240

D. Furfari, H.J.M. Woerden, R. Benedictus and A. Kwakernaak

Curved stiffened panels It has been already observed in small and large flat panel results that, the adhesive connection is stiffer than the fastened connections, resulting with an increase of the load flowing through the doubler for the bonded repair. Since the curved stiffened panels have stiffer structure (influence of stringers and frames) than the other evaluated specimens, this effect is not so pronounced, as shown by Figure 21 and Figure 22.

Figure 21 Mean stress distributions around the doubler of the baseline repair in curved stiffened panels

Figure 22 Mean stress distributions around the doubler of the bonded repair in curved stiffened panels

Bonded repair for fuselage damage

1241

Figure 23 Results of the curve stiffened panel tests are summarized in the table (left); Damage locations detected during the test of one of the two panel tested are shown with lines (right); the sequence of the findings are also shown (from [3]) The fatigue test results of the two curved stiffened panels were already reported in [3], where the fatigue cracks detected at baseline and advanced repair solutions as well as fatigue crack growth lives were described in detail (Figure 23 from [3]). In both panels tested no fatigue crack was detected at the bonded doubler repair. A teardown inspection was carried out in both panels and the results of this investigation have been reported in [3]. C-scan inspection has been carried out at the bonded repair of both panels tested to detect possible debonding areas at the bonded repair solution. The result of this investigation for one of these panels is shown in Figure 24. No major debonding

Figure 24 C-scan inspection of the bonded repair at the curved stiffened panel

1242

D. Furfari, H.J.M. Woerden, R. Benedictus and A. Kwakernaak

areas that could cause premature failure have been identified and only small debonding area at the frame to skin attachment has been detected. Adhesive not uniformly distributed (grey shading areas) was also detected but this did not have an influence on the overall fatigue behaviour of the panels.

CONCLUSIONS The following conclusions can be drawn, split in parts dealings with coupon specimens, flat panels and curve stiffened panels respectively. Coupon Specimens The bonded repair solution, tested on coupon specimens, showed an increase of fatigue life more than 4.5 times in respect to the baseline repair. Small Flat Panels The bonded small flat panel solutions showed the amount of improvement of factor 2 or more, compared to the baseline solution Large Flat Panels The fatigue life improvement of this solution on large flat panel tests has shown an improvement of more than 2 times the baseline repair. It seems that the benefit in fatigue life achievable with the bonded solutions embodied in the coupon specimens (fatigue life longer than 4.5 times of the baseline solution) cannot be obtained when similar repair solutions are applied in the small and large flat panels. These results can be explained with the different load transfer conditions between coupon specimens and the other specimen geometry where we have an influence of the cutout size. The most important difference between coupon specimens, the small and large flat panels are the load which flows through the doubler. The coupon specimens are full load transfer specimens, that is, the whole applied load will be transferred in the repair doubler while in the small and large flat panels the by-pass stresses depends of the repair solutions adopted. The by-pass stresses of the baseline repair solution were higher than the other advanced repair solutions. Curved Stiffened Panels In both curve panels tested after a number of cycles correspondent to more than 2 times the fatigue life of the first fatigue crack detected in the baseline repair no fatigue crack was detected in the bonded repair.

ACKNOWLEDGEMENTS The authors would like to thank the European Committee for partial sponsorship of this work and a PhD student Goran Ivetic of the University of Pisa, Italy, for his contribution in preparation of this article.

Bonded repair for fuselage damage

1243

REFERENCES [1] [2]

[3]

[4]

[5]

[6]

[7]

Armstrong A. (2003), In: Future structures supportability strategy, Proceedings DSTL Structures Contextual Review. `Fifth Framework European Project (2000), IARCAS Improve and Assess Repair Capability of Aircraft Structures, Annex I: Description of the Work, Contract n0 G4RD-CT2000-00401, Project n0 GRD1-2000-25182, coordinated by Airbus. Furfari D., Meyer C., Lafly A.L., Pramono A. (2007), Advanced Repair Design Principles to Improve Fatigue and Damage Tolerance Behavior of Fastened Repairs, Proc. of 24th ICAF Symposium, Naples, Italy. Wieland D., Stargel D. (2006), In: Bonded Repair of a F-16 bulkhead Flange, Proceedings of the 9th Joint DoD/FAA/NASA Conference on Aging Aircraft, Atlanta, Georgia, USA. `Mazza J., Gaskin G., DePiero B. and Blohowiak K. (2000), In: Faster Durable Bonded Repairs using Sol-Gel Surface Treatments, Proceedings of the 4th Joint DoD/FAA/NASA Conference on Aging Aircraft, St. Louis, MO, USA. Fleisher T., Sachse M., Semsch M. (2005), In: Engineering Approach and Results of Pressurized Fuselage Panel Testing, Proceedings of the 23rd ICAF Symposium, vol. I, p.219-227, Dalle Donne C. (Ed.), Hamburg, Germany. Best R., Fleischer T., Hanel W., Sachse M., Semsch M. (2007), In: Metals vs. Composites and Metal Meets Composite – The Influence of the Material on Test Concepts, Test Equipment and Test Programs, Proceedings of the 24th ICAF Symposium, to be published, Salvetti A. (Ed), Naples, Italy.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

FATIGUE & DAMAGE TOLERANCE OF HYBRID MATERIALS & STRUCTURES – SOME MYTHS, FACTS & FAIRYTALES R.C. Alderliesten Aerospace Materials & Structures, Delft University of Technology, Delft, the Netherlands

Abstract: This paper discusses briefly several aspects related to fatigue and damage tolerance of hybrid materials and structures based upon the reported applied and academic research performed over the last two decades on. The most important aspects observed in both fields are highlighted. Subsequently, the lessons learned from both the academic and applied research fields are being addressed. These lessons are then used to discuss the (recent) developments on hybrid material concept technology in the context of optimizing and designing to increase the fatigue performance and damage tolerance of structures.

INTRODUCTION In the past decades, the Fibre Metal Laminate (FML) material concept has been developed with success for aeronautical applications. Although originally aimed at lower wing skin panel applications, with FMLs containing aramid fibres (Arall), the concept has been brought to a technology readiness level for fuselage skin applications with FMLs containing glass fibres (Glare). Inspired by this successful concept, many researchers all over the world have pursued research on various FML types to investigate the behaviour of these materials under mechanical-, thermal- and environmental loading. As a consequence, the list of internationally refereed journal papers concerning FMLs has become impressive. However, there seems to be an enormous gap between the M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1245–1260. © Springer Science+Business Media B.V. 2009

R.C. Alderliesten

1246

work reported in these journal papers and the work being done by institutes and companies on actual FML applications. A lot of academic research reported in the literature describes a selection of experiments or (numerical) analyses on a particular combination of metallic and fibre reinforced polymer constituents. The research topic is approached with a specific scientific question or objective in mind. Often this leads to detailed understanding of a particular problem. However, despite claims and conclusions, the investigated combination of materials never leads to actual applications, because of the lack of a thorough development process towards applications. Applied research on the other hand, often approaches the material concept from the perspective of particular design criteria or problem. In an attempt to obtain a specific behaviour, the nature of the hybrid laminates seem to be neglected and the scope of research limited. As a consequence, the possible design options remain limited. This paper provides a brief discussion of applied and academic research reported in the last two decades on fatigue and damage tolerance of hybrid materials. The most important aspects observed in both fields will be highlighted. Subsequently, the lessons learned from both the academic and applied research fields are being addressed. These lessons are then used to discuss the (recent) developments on hybrid material concept technology in the context of optimizing and designing to increase the fatigue performance and damage tolerance of structures.

DEVELOPMENT OF FMLS Focus on Damage Tolerance The history of the development of FMLs has been described in detail by Vlot [1] and Vogelesang [2]. Originating from the poor man’s solution to build up aluminium structures using bonding technology, rather than using expensive milling equipment, the major driver in the development of FMLs has been to develop a material concept that has inherently higher resistance to (fatigue) cracking. High fatigue crack growth resistance results in slow crack growth and thus increased inspection intervals, while high fracture toughness increases the critical crack length, further increasing the damage tolerance of structures. The first patented FML concept combined aramid fibres with aluminium sheet materials, initially developed for a F27 wing application. This concept has been thoroughly investigated especially with respect to its fatigue characteristics [3]. Developing the concept for fuselage applications, however, revealed poor fatigue resistance due to compressive cycle induced fibre failure. Subsequent development led to the FML based on glass fibres, Glare [4]. Within the development of Glare, additional beneficial properties were identified that supported its application to primary fuselages structures; high impact

F& DT of hybrid materials & structures-Some myths, fact and fairytales

1247

resistance and tolerance, high burn through resistance which could potentially increase evacuation time (safety aspect), and improved corrosion resistance (durability aspect) derived from its layered nature. Despite all the effort within the Glare development programme [5] to achieve technology readiness of Glare with respect to all relevant aspects, the prime focus remained on the fatigue and damage tolerance characteristics. The literature on this programme illustrates the process from generic understanding of fatigue initiation and crack propagation, towards more detailed understanding on fatigue resistance of mechanically fastened joints, impacts, external and interlaminar doubler run-outs, edge notches and rows of holes. The latter topic is considered important to address the resistance against widespread fatigue damage. Although clearly explained in the literature, damage tolerance as a development driver has not always been well understood. Especially, academia, not related to institutes involved in the development of Glare, have presented research that, although being interesting, has never contributed to the development of FML applications, despite claims otherwise. For example, developing structural health monitoring concepts for FMLs without accounting for the inherent high resistance against impact and (fatigue) cracking [6] makes the developed concepts obsolete or inapplicable for many reasons, especially the economic reasons. The need for FML structures originates from the desire to develop damage tolerant structures that are care free (i.e. low inspection and maintenance burden). The application of a health monitoring system seems to be contradictory to the application of such care free FML structures. On the other hand, one should realize that such monitoring techniques applied in laboratory experiments may add to the understanding of certain mechanisms. This is illustrated by Austin et al. [7], who utilized fibre bragg grating sensors to quantify fibre stresses in fatigue crack growth tests; information that previously could only be derived implicitly from other parameters measured. Inspired by the development of FML as combination of metallic sheet material and fibre reinforced polymers composites, many combinations have been presented in the literature. None of these combinations ever led to actual applications. For example, combining magnesium and carbon fibre reinforced polymer layers as reported by Cortes and Cantwell [8] is inapplicable to actual structures for reasons not addressed in their work. The low stiffness of the magnesium layers and the poor fatigue (see Figure 1) and corrosion resistance of these materials in comparison with aluminium do not compensate for the lower density, which seems the prime driver for this study [9]. This also yields for FML combinations based on thermoplastic matrix systems for which curing temperatures need to be applied that exceed thermal stability levels of widely applied aerospace alloys [10,11]. As a result, only particular applications may be considered for such FMLs that often exhibit poor fatigue performance, due

R.C. Alderliesten

1248

to the severe tensile curing stresses in the metallic layers. Here the application of thermoplastic matrix systems should be investigated together with relevant aerospace alloys that can sustain the high curing temperatures. 30

AZ31B(H24)-S2/FM94-FML3-3/2-0.4

2024-T3-S2/FM94-FML3-3/2-0.4

2024-T3

25

a [mm]

20

15

10

Smax = 120 MPa, R = 0.1 5

Smax = 100 MPa, R = 0.1

0 0

50000

100000

150000

200000

250000

300000

N [cycles]

Figure 1: Predicted fatigue crack growth curves of Glare3-3/2-0.4 and a similar FML configuration based on the magnesium AZ31B alloy for two maximum stress levels. Monolithic 2024-T3 curves have been added for comparison [9] Constituents, codes and standardisation FMLs have been treated as a material family, consisting of thin metal layers with in-between fibres in an adhesive system. From the introduction of the first concept Arall onward, trade names have been filed to describe the laminates developed. However, the introduction of trade names for particular configurations seems to add to confusion. For example, Glare has been defined by Roebroeks in [12] as the FML based on uni-directional S2-glass fibres embedded in FM94 adhesive (manufactured by Cytec). This is in principle another material system as originally introduced by him in [4] based on the AF163-2 adhesive (manufactured by 3M). The AF163-2 system was selected based upon the observed high resistance against delamination obtained in an optimization study. Nevertheless, the FM94 adhesive system has been adopted for the FML application on the Airbus A380. Indifferent of the reasoning for selecting FM94, it should be noted here that the trade name Glare, nor the grading (Glare1, Glare 2, etc) changed when the other adhesive system was adopted. Technically, the question rises whether the proposed coding is sufficient to describe the material composition. In fact, the aluminium

F& DT of hybrid materials & structures-Some myths, fact and fairytales

1249

alloy, the fibre type and the adhesive system are only implicitly considered in the name and grading, but are not explicitly defined. It seems that the freedom of tailoring the FML concept similar to the tailoring of fibre reinforced polymer composites conflicts with the approach adopted in the metal world to code each material. To illustrate the issue, one may look for instance closer to Glare. As the name indicates, Glare is a laminate consisting of aluminium reinforced by glass fibres. In general, this could be any combination of glass fibres, aluminium alloys and adhesive systems, as it has not been defined explicitly. However, within the community involved in development of Glare for the A380, the application of unidirectional S2-gass fibre/FM94 epoxy prepreg has always been considered in combination with aluminium 2024-T3. Recently, a study has been published on the aging effect on fatigue of Glare [13], were instead of uni-directional prepreg, woven fabric was applied. It may still be considered Glare, but the observed behaviour might not directly relate to experimental observations for Glare based on uni-directional prepreg. For example, the delamination behaviour at the metal/fibre interfaces for uni-directional prepreg and fabric are known to be different [14].

Figure 2: Cross sections of Glare laminates based on woven fabric [13] (left) and based on uni-directional prepreg (right)

FATIGUE INITIATION Probably the clearest illustration of the gap between the academic and engineering understanding is given by the assessment of fatigue initiation in FMLs. In the characterisation of the fatigue behaviour of FMLs a distinction has been made in the past between the initiation life and the crack growth life. Initiation was defined as the fatigue life up until an initial crack of 1 mm length has been formed. The crack growth life was then considered as the life from that initial length or greater up to critical crack lengths.

1250

R.C. Alderliesten

The misperception about the fatigue initiation behaviour is related to the following two aspects: First, the selection of 1 mm as the transition crack length between initiation and crack growth; and second, the meaning of the obtained initiation life for structural design. Transition from initiation to crack growth Traditionally, the transition from initiation to crack growth in monolithic metals has been related to the transition from nucleation and formation of micro cracks towards propagation of macro cracks [15]. The initiation phase is predominantly affected by the surface conditions, whereas the crack growth phase is determined by the bulk material properties. Indeed, because of the argument that the initiation behaviour in the metal layers of the FML are determined solely by the metal layer [16], there seems no scientific reason for taking a different definition for fatigue initiation in FMLs. For the thin sheets in FMLs (commonly between 0.3 – 0.5 mm), it can be argued that the transition length thus would be in the same order of magnitude as the sheet thickness. However, from an engineering perspective a different transition length is preferred. Homan [17] has shown that the initiation behaviour is predominantly described by the actual nominal stress cycle and the peak stress cycle at the notch root. Using the Classical Laminate Theory he derived the scientific theory that the nominal stress cycle and the correct Kt enable the application of an S-Ni (fatigue life until initiation) curve obtained for the monolithic metal to the metal layer in the FML. However, the available fatigue data for metallic materials consist either of fatigue life until failure data (S-N) or of crack resistance data (da/dN versus ΔK). This means that either S-Ni curves need to be determined by testing a statistically significant amount of monolithic metal samples, or another engineering approach should be adopted to use the available fatigue life data. Here, the hypothesis is that if the transition length is selected appropriately, the vast amount of experimental SN data, available in the open literature, can be used to predict the initiation behaviour of FMLs. Figure 3 illustrates the evolution of cracks length as percentage of the life until failure for monolithic aluminium 2024-T3. From this figure it is evident that once a transition crack length of ai = 1 mm is taken; the error with the total fatigue life is rather small, in the order of a few percentages. However, if initiation would be defined up to a transition crack length of about 0.1 mm the error would be significant, up to 60%. Based upon this figure, it is suggested to take a transition crack length of at least 1 mm, but preferable greater to further limit the error with the S-N data in the literature. However, Figure 3 also illustrates the crack growth for two Glare grades for crack lengths up to 5 mm. From this figure it can be observed that the fibres start having

F& DT of hybrid materials & structures-Some myths, fact and fairytales

1251

an effect on the crack growth once the crack is still in the order of a tenth of a mm. In other words: this graph suggests that a transition length should be defined as small as possible, preferable less than 1 mm in order to exclude any effect of fibre bridging on the initiation life.

Figure 3: Length of small cracks as percentage of the fatigue life for unnotched Al2024-T3 [15,18] (upper) and the growth of small cracks in Glare [15,19] (lower)

1252

R.C. Alderliesten

In the past, when discussing the transition from initiation to crack growth, the selected transition length of 1 mm was often justified with the argument that after 1 mm the fibre bridging became effective. In other words, fatigue initiation was related to fibres being ineffective, while crack growth was related to effective fibre bridging. This distinction seems to be supported by the illustration in Figure 4, presented by Homan [17]. However, in the text of his paper Homan explains that the transition length of initiation to crack growth is not related to the crack length at which fibre bridging becomes effective. Figure 3 illustrates that the fibres start influencing the crack growth immediately, but based upon other work [16] it is known that the length at which fibre bridging becomes fully effective might range between tenth of millimetres up to several millimetres depending on the laminate, notch and loading. To conclude, the transition crack length of 1 mm is not directly related to any physical phenomenon. It is a somewhat arbitrary but convenient selection to enable application of S-N data to the prediction of fatigue initiation in FMLs. However, when discussing initiation, one always has to specify the selected transition length.

Figure 4: Different phases of the fatigue life in FMLs together with relevant factors [17] The meaning of initiation life for structural design It requires no argument that the fatigue initiation behaviour of metal structures relates to the inspection threshold defined in the structural design. Early initiation requires first inspection at an earlier phase in the operational life. As a consequence of the difference in stiffness and coefficient of thermal expansion between the constituents in an FML, both the stress amplitude and the mean stress in the metal layers of the FML are higher than the applied stress cycle. This implies that the fatigue initiation life (up to 1 mm) for the metal layer in an FML is shorter than the initiation life for a monolithic metal at equivalent applied stress cycle. The fact that the initiation life of FMLs is shorter than for monolithic metals has attributed to the confusion about the benefits of this material concept for structural design. At the 30th ICAF conference in Naples, for example, the question was raised how FMLs such as Glare could be considered beneficial if the initiation life is shorter than for monolithic aluminium. This and similar questions assume that the threshold inspection is solely influenced by the fatigue initiation life. However,

F& DT of hybrid materials & structures-Some myths, fact and fairytales

1253

taking a closer look at the common procedure to define an inspection threshold, it becomes clear that also other factors play quite a significant role. As explained by Schmidt [20] in his Plantema lecture, all airframe manufacturers perform fatigue life (S-N) analyses to determine the inspection threshold for certain structures. Similar to the schematic presented by Beumler [21], Figure 5 illustrates schematically for a single crack the approach to define the inspection threshold for metallic structures based upon a factor of the total fatigue life. The figure correlates the approach with the equivalent approach for FMLs. UL LL NTH,FML ac,FML ac,al

NTH,al

adet ‘aini’ Ni,FML Ni,al

NLL,al

NLL,FML

NUL,al NUL,FML

Figure 5: Illustration of determining the inspection threshold based on fatigue life evaluation for monolithic aluminium and FML (not to scale) Here, it is clear that the fibre bridging induced slow crack growth and the larger critical crack lengths resulting from high fracture toughness, give a significant increase in life. This will not only translate into longer inspection intervals, but because of the longer crack lengths that can be detected potentially during inspection, it will also shift the inspection threshold further in time. In fact, the capability to tune the laminate lay-up to some extent enables to tune the inspection interval and threshold.

THRESHOLD FATIGUE Despite similarity in metallic fracture phenomena, the experiments to assess certain fatigue aspects in FMLs could be different from monolithic alloys. An example to illustrate that is given by the threshold fatigue of FMLs. The fatigue threshold is often used to define omission levels in (full-scale) tests, below which cyclic loads can be disregarded. For monolithic alloys there are several experiments developed

R.C. Alderliesten

1254

to determine the fatigue threshold; each of them having their own drawbacks. Even the standards leave room for discussion. However, any of these experiments would be inappropriate to determine fatigue threshold levels in FMLs. The link between stress intensity factor for FMLs and the applied stress cycle is less straightforward as result of a fibre bridging component. This fibre bridging component is related to delamination shape and size, and the limit on loads influencing delamination growth may not be the same as for crack growth in the metal layers. As a result, the lower limit, or threshold, for crack growth may be less clear for macro cracks in FMLs compared to monolithic alloys [22]. Figure 6 compares the predicted fatigue crack growth rate reduction obtained by load reduction tests in Glare, similar to monolithic aluminium threshold fatigue experiments. These results are compared to three constant amplitude fatuigue tests for Glare at 30, 40, and 50 MPa. In the load reduction tests, the larger delaminations created by the previous higher loads induce larger crack growth rates at stresses equal to the constant amplitude results. The link between delamination shape and size, and crack tip stress intensity factor produce an inherent load-history effect in FMLs, making such load reduction tests for determining FML thresholds inapplicable. 1.E-04 Smax = 120 MPa, R = 0.05

Smax = 95 MPa, R = 0.05

1.E-05

da/dN [mm/cycle]

Smax = 65 MPa, R = 0.05 Smax = 60 MPa, R = 0.05 Smax = 55 MPa, R = 0.052 Smax = 50 MPa, R = 0.05 Smax = 45 MPa, R = 0.068 1.E-06 Smax = 40 MPa, R = 0.065 Smax = 37.5 MPa, R = 0.068 Smax = 35 MPa, R = 0.063 Smax = 32.5 MPa, R = 0.065 Smax = 30 MPa, R = 0.065 1.E-07 2.5

4.5

6.5

8.5

10.5

12.5

a [mm]

Figure 6: Calculated crack growth rates using the loading sequence applied in the test, compared with three cases of CA amplitude loading at Smax = 30, 40 and 50 MPa [22]

F& DT of hybrid materials & structures-Some myths, fact and fairytales

1255

In addition, if threshold can not be observed for delamination growth, this implies that for a stress intensity factor below the threshold of the metallic layer, the delamination might continue to propagate. As the bridging component of the stress intensity factor would change as a result of delamination growth, it may be possible for the crack tip stress intensity factor to exceed the threshold after a given period of delamination growth. In other words, the level below which no damage growth can be observed is related not only to metallic fatigue behaviour, but also to the delamination resistance. Although the fatigue threshold may seem less relevant for FMLs, because of characteristic slow crack propagation, this example illustrates that the mechanisms should be fully understood before appropriate tests can be defined to assess fatigue behaviour. In this example, the fatigue threshold should be investigated separately for crack growth in metallic sheets and for delamination growth at interfaces. The combination of these two lower limits defines the lower limit for the FML.

NON-DESTRUCTIVE INSPECTION The application of new inspection techniques is related to the discussion on fatigue initiation and inspection threshold. In academia, a lot of research is being performed on the development of inspection techniques, especially on in-situ measurements. Although worthwhile, one has to consider in which way these developments play a role in the application of hybrid material technologies. Ultrasonic inspection. An example that may be considered illustrative in this respect is the research on ultrasonic inspection techniques for FMLs [23]. This research has increased the detection capabilities (increased accuracy) as well as the evaluation procedure. On the other hand, in the development programme for FMLs [5], effect of defect studies have been performed on the maximum flaw sizes to be allowed in the as manufactured FML panels. The discussion on the allowable flaw sizes should have been related to the performance in operational life of FML panels containing these flaws, i.e. whether or not the flaws grow or impair the structural integrity during operation. However, this discussion seems to be influenced by the developed ultrasonic inspection techniques that allowed detection of significant smaller sizes. Rather than accepting a large flaw size, it was decided to set smaller flaw size requirements. Aside from the enormous impact on the scanning time and thus manufacturing costs, this also implies an additional safety factor redundant to the other safety factors applied. In-situ health monitoring. A lot of research on non-destructive techniques seems to focus on in-situ measurement for health monitoring. To a great extent this effort is

R.C. Alderliesten

1256

related to the development of CFRP primary structures, which may benefit from the detection capabilities in life. Nevertheless, although the inspection is an important aspect of damage tolerance assessment, one has to keep in mind that true damage tolerance aims to minimize the inspection burden [24]. This objective relates to both the amount of scheduled inspections as well as the required techniques during inspection. As a consequence of the airworthiness requirements, the application of sensors in primary structures will never replace the scheduled inspection (at least not the half life inspection to which inspection should be limited); they will only be complementary. This has several risks that one should consider. First, the risk of incorporating sensors in primary structures that may no longer work at the time they are really needed, i.e. near the end of life. This may be an important aspect, because the first customer is often no longer operating the airframe in this phase of the life, especially for commercial aviation. In addition, the risk of false indications may imply more downtime rather than less. Studies presented in the literature still report inaccurate detection and false indications. In the end, it could be argued that the reasoning behind health monitoring seems to conflict with the hybrid technology intention to obtain care free structures. Manufacturing. The above issues related to non-destructive inspection seem to indicate that the best road here should be to develop the techniques for in-situ inspection during manufacturing. With the allowable flaw sizes as large as possible, because of the true damage tolerant nature of FML, the only objective is to prevent large flaws to occur by early detection. This will minimize scrapping components, reducing manufacturing costs to an absolute minimum.

FUTURE DEVELOPMENTS Material design or structural design When discussing the early developments of the FMLs it has been mentioned that coming from actual structural needs (F27 wing and A380 fuselage) FMLs have been developed as a material. This is illustrated by the use of trade names and the formulation of patents being filed [25] Despite all academic understanding developed in the past decades, the true composite nature is not yet acknowledged and industry (especially the metallic) keeps developing hybrid materials as if they are alloys. This is illustrated by the substantiation of new FML or hybrid materials; crack growth curves are compared and curves with increased crack growth life implies better material. Although the prime focus in the Glare development programme [5] has been fatigue and damage tolerance, one thing has become clear in the past years: it requires much more to develop such concept to an actual application. In fact, that

F& DT of hybrid materials & structures-Some myths, fact and fairytales

1257

is the reason why for example Magnesium/CFRP laminates [8] have no future in primary aircraft structures [9]. Rather than inventing a different lay-up made of metallic and fibre reinforced polymer constituents, which is considered a straightforward exercise in the composite world, one should consider tailoring hybrids to actual structural applications. That is, once an application is selected and requirements have been defined, the application of individual constituents can be tailored for that particular structure appropriately. This does not result in a new material concept, but in a new structural concept. In the end, similar to composites, the material is considered to be formed together with the structure, instead of prior to manufacturing and assembling the structure. Such an approach avoids the risk of proposing a material concept that needs further modification once requirements from an actual application are defined that were not considered initially. An example for this case could be the requirement coming from a care free philosophy that damage should always start at the outer visible side of the structure [24]. If a material concept has been developed initially that, despite slow crack growth, exhibits initiation and propagation of cracks at the nonvisible side, further development is required, probably changing the material. A structural approach offers a more flexible approach that enables design changes and modifications based on an understanding of the combination of hybrid constituents rather than an understanding of one particular combination, as in a material design approach. This comment implies, however, initiative taken by OEMs rather than material suppliers and academia, which has been the case traditionally for FMLs. Considering future developments on hybrid materials, it seems evident that there is hardly any room left for innovative patents on plain material concepts [25]. The fibre types and the metallic materials that can be selected to be combined in a hybrid panel are more or less known. Development of new synthetic fibres and metallic alloys will take place and will lead to new patents, but introducing these new materials into the similar hybrid concepts can hardly be considered an innovation. The hybrid concept has become a well documented technology that may be considered a carrier for new materials to be developed.

THE WAY FORWARD Based upon the topics discussed in this paper, it seems evident that in order to develop FMLs and hybrids further towards applications, several aspects must be considered. Although fundamental understanding is raised at material level and therefore no doubt necessary, the development of concepts should be performed at a structural

R.C. Alderliesten

1258

level. Here, perhaps assisted by academia, the OEMs need to identify the structures that benefit from the hybrid technology. However, they must define the requirements that are needed to develop the concepts. In fact, these requirements should be the dominant drivers in the concept development rather than preferences coming from constituent materials being developed by suppliers. Here, the fundamental understanding and analysis capability generated in the past decades will act as an aid. One should note that this implies a close collaboration between OEMs, suppliers and academia. Design what you can model or model what you can design The OEM may consider two approaches. The first approach is a low risk approach: developing other applications using similar hybrid material technology previously developed. One may think here of developing for example a wing structure based on already qualified and certified FML concepts initially developed for fuselages. A general benefit to this approach is the availability of prediction tools, developed for the first application and its certification. Instead of thorough research to quantify behaviour and development of prediction models based on that data, the non-specific tests may be limited to a minimum being covered by available models. The second approach has a higher risk: developing structural concepts from scratch. Based upon full system requirements specific combinations of constituent materials and related processes could be developed. Depending on the predictive capability of previously developed analysis tools, this requires extensive development programmes including material qualification and certification of the concepts developed. Here, the generic tests may need to be performed to generate a reliable data set to support method development. However, even for this case it is recommended that the applicability and potential extension of available methods is studied before such an extensive programme is performed.

CONCLUSIONS Industry should acknowledge the nature of FMLs as being a combination of metallic and composite materials, rather than an advanced metallic concept. This means that specific observations (such as fatigue initiation) does not automatically relate to similar effects in monolithic metals when structural performance is being considered. One may have to consider alternative design philosophies, but will only be able to do so, if the mechanisms are fully understood. Here, more attention should be given to the understanding developed within academia. Academia on the other hand, should pay attention to the needs and requirements that are the dominant drivers in development of new technologies. This may force academia to focus more on the understanding necessary to develop the hybrid

F& DT of hybrid materials & structures-Some myths, fact and fairytales

1259

technology towards actual applications, rather than studying various (but often for industry irrelevant) phenomena. Industry should implement all the predictive capability of analysis tools developed by academia. This will not only support certification of FML structures, but also help identification of potential other applications and future model extensions. Industry and academia should no longer develop material concepts, but consider development of structural concepts. Here, the industry may consider to use existing technology (low risk) or to develop new hybrid structural concepts (high risk). In any way forward, industry should consider closer collaboration with academia to exploit the knowledge and analysis tools developed for their own purposes. The need to do so will significantly increase if academia in return is willing to focus more on the relevant aspects related to the hybrid technology.

REFERENCES [1] [2]

[3] [4] [5]

[6] [7] [8] [9] [10] [11]

[12]

Vlot, A., (2001), Glare – history of the development of a new aircraft material, Kluwer Academic Publishers, Dordrecht. Vogelesang, L.B. (2003), In: Fatigue of Aeronautical Structures as an Engineering Challenge, Proceedings of the 22rd ICAF Symposium, vol. I, p. 1–41, Editor, G. (Ed.), EMAS publishing, UK. Marissen, R., (1988), PhD dissertation, TU Delft. Roebroeks, G.H.J.J., (1992), PhD dissertation, TU Delft. Vlot, A. (2000), Towards Technology Readiness of Fibre Metal Laminates – GLARE Technology Development 1997-2000, Proceedings of the 22nd International Congress of Aeronautical Sciences, Harrogate, United Kingdom, pp. 1-15. Kuang, K.S.C., Kenny, R., Whelan, M.P., Cantwell, W.J., Chalker, P.R. (2001), Smart Mater. Struct, vol. 10, p. 338. Austin, T.S.P., Singh, M.M., Gregson, P.J., Powell, P.M. (2008), Composites Science and Technology, vol. 68, p. 1399. Cortes, P., Cantwell, W.J. (2006), Composites: Part B 37, p. 163. Alderliesten, R., Rans, C, Benedictus, R. (2008), Composites Science and Technology, vol. 68, issue 14, p. 2983-2993. G. Reyes V., W. J. Cantwell, Composites Science and Technology, Volume 60, Issue 7, May 2000, Pages 1085-1094 J.G. Carrillo, W.J. Cantwell, Mechanical properties of a novel fiber-metal laminate based on a polypropylene composite, Mechanics of Materials, In press, 2009. Vlot, A., Gunnink, J.W. (2001), Fibre Metal Laminates – an introduction, Kluwer Academic Publishers, Dordrecht.

1260

R.C. Alderliesten

[13]

Da Silva, D.A., Botelho, E.C., Rezende, M.C. (2008), J. Reinf. Plast. & Comp, In press. Khan, S.U., Alderliesten, R.C., Benedictus, R., Delamination in Fibre Metal Laminates under variable amplitude loading, to be published. Schijve, J. (2009), Fatigue of Structures and Materials, Springer Science+Business Media B.V. Alderliesten, R.C. (2007), Int. J. Fatigue, vol. 29, p. 628. Homan, J.J. (2006), Int. J. Fatigue, vol. 28, p. 366. Schijve, J. (1967), Fatigue Crack propagation, ASTM-STP 415, p. 415. Papakyriacou, M., Schijve, J. and Stanzl-Tschegg, S.E. (1997), Fatigue Fract. Engrg. Mater. Struct., vol. 20, p. 1573. Schmidt, H.J. (2005), In: Structural Integrity of Advanced Aircraft and Life Extension for Current Fleets – Lessons Learned in 50 Years after the Comet Accidents, Proceedings of the 23rd ICAF Symposium, vol. I, p. 1–41, Dalle Donne, C. (Ed.), DGLR, Germany. Beumler, Th. (2008), MoC for A380 hybrid structure, Proceedings of the 2008 ASIP conference, San Antonio, USA. Alderliesten, R.C., Rans, C.D., (2009), Int. J. Fatigue, vol. 31 (2), p. 213. Coenen, R.A.M., (1998), PhD dissertation, TU Delft. Fredell, R. (2007), “Carefree” Hybrid Wing Structures for Aging USAF Transports, Proceedings of the first conference on Damage Tolerance of Aircraft Structures, Delft. Alderliesten, R.C. (2009), Recent Patents on Engineering, vol. 3 (1), p. 25.

[14] [15] [16] [17] [18] [19] [20]

[21] [22] [23] [24]

[25]

25th ICAF Symposium – Rotterdam, 27–29 May 2009

HIGH CYCLE FATIGUE OF LASER BEAM DEPOSITED Ti-6Al-4V AND INCONEL 718 E. Amsterdam and G.A. Kool National Aerospace Laboratory NLR, The Netherlands.

Abstract: Laser Beam Deposition (LBD), or also called Laser Engineered Net Shaping (LENS), is used for fabrication and repair of metallic aerospace components. Basic research and process development have already been done on Ti-6Al-4V and the interest is now expanding to other aerospace alloys. For application of the technology on critical aircraft and engine parts, the status of the technology has to be assessed. Two identical components have been defined by NLR and fabricated at Optomec by the LENS™ technique. The components have been made with standard process parameters. One component was made of Ti-6Al-4V and the other one of Inconel 718. The components were given a standard heat treatment prior to cut-up for machining the test samples. Test samples were cut from the laser track and build directions to examine the tensile properties and the microstructure for both orientations. High cycle fatigue (HCF) tests were performed on round samples to create a stress vs. cycles curve for both alloys. Both components contained many macro and micro defects, such as large cavities, lack of bonding between layers and porosity. The fatigue performance of the laser deposited Ti-6Al-4V is better than of wrought material. Although the presence of pores decreases the performance, making the fatigue performance similar to wrought Ti-6Al-4V. The fatigue performance of Inconel 718 is similar to wrought material, even though the material contains a lot of small pores. Lack of bonding between layers decreases the fatigue performance of Inconel 718 accordingly.

M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1261–1274. © Springer Science+Business Media B.V. 2009

E. Amsterdam and G.A. Kool

1262

INTRODUCTION Laser Engineered Net Shaping (LENS) is a rapid deposition technique for metals. The LENS process is known by various names, such as: Laser additive manufacturing (LAM), Direct Metal Deposition (DMD™), 3D Laser Cladding, Direct Laser Deposition (DLD), Laser Powder Deposition (LPD) and Laser Beam Deposition (LBD). The technique applies a laser to create a melt pool on a substrate; metal is deposited by injecting metal powder through a nozzle into the melt pool. The melt pool is shielded from air by inert gas. Rapid cooling solidifies the deposit, while the substrate is rastered back and forth according to a 3D-CAD pattern. A complex 3D component can be created by adding multiple layers. The use of Laser Engineered Net Shaping (LENS) techniques for fabrication and repair of metallic aerospace components has increased with increasing process development [1]. For application of the technology on more critical aircraft structures and gas turbine engine parts, the status of the technology has to be assessed by determining the mechanical properties of aerospace alloys that are frequently used for gas turbines, such as Ti-6Al-4V and Inconel 718. Process development and basic research has been already done on Ti-6Al-4V and the interest is now expanding to other aerospace alloys, such as Inconel 718. For an assessment of the current LENS technology status two identical components have been fabricated by the LENS technique. One component was made of Ti-6Al-4V and the other of Inconel 718. The shape of the components is chosen such that it incorporates a thin wall, a thick wall and different angles at which the walls meet (see Fig. 1). Microstructure, tensile and fatigue samples were cut from the components to asses the microstructure and the mechanical properties (see Fig. 2). The results for Ti-6Al-4V are compared with the results from (LENS-) literature, and owing to the same component geometry and process conditions the results on Inconel 718 can be compared with those of Ti-6Al-4V and wrought Inconel 718.

EXPERIMENTAL SET-UP Two identical components have been fabricated by the LENS™ technique at Optomec in Albuquerque (NM), USA. One component is made of Ti-6Al-4V and the other one of Inconel 718 (see Fig. 1). After receiving the components, they are given a standard heat treatment; for Ti-6Al-4V this is a solution heat treatment at 970°C for 1hr, water quench and age at 538°C for 4hr, air cool [2]. For Inconel 718, a post weld heat treatment was chosen; 1093°C for 1 hour, cool to 718°C with 84°C per hour, 718°C for 4 hours, cool at 56°C per hour to 620°C, 620°C for 16 hours, cool at 139°C per hour to below 371°C, air cool [3]. Round fatigue samples have been cut from wall A (see Fig. 1) and are polished to a roughness Ra< 0.2 μm. The fatigue samples have been tested at different stress levels on an electro-magnetic resonance machine Amsler HFP 422 with a frequency of 110 and 140 Hz for the Ti-6Al-4V and Inconel 718 samples

HCF of laser beam deposited Ti-6Al-4V and Inconel 718

1263

respectively. The frequency depends on the material stiffness. The length of the samples is parallel to the build direction (BD). Flat tensile samples are cut from the thin wall B and round tensile samples are cut from wall C, according to ASTM E8 testing standard. The length of the samples is parallel (BD) and perpendicular (LD) to the build direction. The gage length of the flat and round tensile samples is 32 mm and, respectively, 36 mm. Instron test machines, with a 100 kN and 300 kN loadcell, are used for the flat and round samples respectively. The tests have been performed at room temperature and the displacement speed of the crosshead was 1 mm/min (quasistatic). A MTS clipgauge with a gage length of 25 mm was used to measure the strain. Microstructure samples have been cut from the top of the build and at the root of the build (including the substrate) of wall A and B. For each location crosssections are made parallel (half way the thickness) and perpendicular to the wall. Optical- and scanning electron microscopy (FEG-SEM) have been used to characterise the microstructure. The SEM, equipped with energy dispersive analysis of X-rays (EDX), was used to determine the chemical composition. The porosity was measured by optical microscopy and the macrohardness by an Instron 930 testor using a 5 kg load. Both measurements have been performed before etching. Ti-6Al-4V was etched with Keller’s reagent and Inconel 718 with Beraha ІІ to reveal the microstructures.

RESULTS Components After a visual inspection of the Ti-6Al-4V component, the following can be remarked (see Fig. 1a): the 7 mm thick substrate is curved. The surface of the component is rough and the thickness of all the walls is 2.5 mm larger than specified. Due to the curved substrate the height is less than specified. There are some defects present in the component that results from droplets of liquid metal that fall from the powder nozzle onto the build. This is encircled in black on the component, see Fig. 1a. After a visual inspection of the Inconel 718 component, the following can be remarked (see Fig 1b): the 7 mm thick substrate is also curved. The surface of the component is rough (large droplets are attached to the walls) and the thickness of all the walls is 1.1-1.5 mm larger than specified. Due to the curved substrate the height is less than specified. At the top of wall C, the wall no longer has the thickness of the root. Large grooves are present from partially melting of the wall. This occurred since the system was not equipped with a melt pool detector. Microstructure The chemical composition of both components is measured at different locations. The results are given in Table 1. None of the alloying elements are less than

1264

E. Amsterdam and G.A. Kool

specified. However, for Inconel 718 some elements are slightly higher than specified. The grain structure of the thin wall is given in Figure 3. The grains are smaller near the interface with the substrate and the grain size increases with the build-up height. This is typical for laser beam deposition [4]. The grains stretch across multiple tracks in the horizontal and vertical (build-up) direction. Some pores are present at the interface with the substrate. The microstructure of the Ti-6Al-4V component consists of a Widmanstätten structure and represents a normal casting structure. The microstructure of the Inconel 718 component consists also of a grain structure with small grains at the root of the build-up. The grain size increases for increasing build-up height. The long axis of large stretched grains ranges from 100 to 250 microns and the short axis of the same grains from 50-100 microns. Carbides are present in the microstructure (white areas, see Fig. 4a) and very small pores have been observed (small black circles). After etching, the γ/γ´ distribution was made visible; cuboids and spheroids can be distinguished. The cuboid size is about 600 nanometer. The delta phase was also observed in cross-sections from the top of the build. Vickers macrohardness (HV5kg) measurements performed on a crosssection of the thin wall (B) showed a very large scatter (HV 359-429) within a few millimeters distance. The porosity was determined with optical microscopy for some cross-sections and is given in Table 2. The porosity of the Ti-6Al-4V cross-sections seems to increase towards the root of the build-up for the thin and thick wall (see also Fig. 3a). The Inconel 718 samples contain many pores with a wide range of diameters. Figure 4a shows that very small pores, with a size < 1 μm, are present. The measurement of the porosity, with diameters more than 4 μm, has been done by image analysis on optical microscope images. The porosity in the Inconel 718 component is much higher than in the Ti-6Al-4V component. Tensile properties Ti-6Al-4V. Table 3 shows the tensile properties of the round and flat Ti-6Al-4V samples. For both, the round and the flat samples, there is one sample (Ti-C6 and Ti-B2) that failed under a low stress due to insufficient bonding between the buildup layers. For sample Ti-C6, very large cavities (1-2 mm) and lack of bonding are present. The other round tensile samples all have the same tensile properties in terms of yield stress, ultimate tensile strength (UTS) and elongation to failure. The flat samples show a higher yield stress and UTS, but with a large range of elongations to failure. No effect of the orientation of the samples can be observed. The fracture surfaces of the samples Ti-C1-5 and Ti-B1,3-10 show features of ductile failure.

HCF of laser beam deposited Ti-6Al-4V and Inconel 718

1265

The Vickers macrohardness (HV5kg) of wall B (thin wall) and C (thick wall) is HV 390 ± 2 and HV 363 ± 6 respectively, corresponding to the flat and round tensile samples. Herewith the hardness measurements confirm that the strength of the material in wall B is 7.5% larger than that of wall C. Inconel 718. Table 4 shows the tensile properties of the round and flat Inconel 718 samples. From the six round samples, three failed outside the gauge length and the fracture surfaces showed the absence of bonding between the layers for several tracks (and are therefore not included in Table 4). Sample IN-C4 failed within the gauge length, but the fracture surfaces also show an area of lack of bonding between the layers. Sample IN-C5 failed outside the gauge length, but did not show signs of lack of bonding. For the flat samples, five of the eleven samples showed strange behaviour and are not included in Table 4. These samples failed, at low elongations, near the radius and exhibited a lower Young’s modulus. These characteristics indicate the presence of a bending mode. However, special attention was paid to the clamping after the first sample failed in this manner. This only occurred to the Inconel 718 samples and not to the similar Ti-6Al-4V samples. There is a large scatter in the elongation to failure for the flat samples. For samples with an elongation to failure < 10%, the UTS increases with increasing yield stress. The fracture surfaces of these flat Inconel 718 samples are rather flat and perpendicular to the load direction. For the samples with a higher elongation to failure, the fracture surface looks more jagged and is slanted with respect to the loading direction. All fracture surfaces of the round and flat samples showed numerous pores with different sizes. Fatigue Figure 5 shows the S-N curves of Ti-6Al-4V and Inconel 718. Figure 5a contains fewer points, because some samples with a diameter of 4.5 mm failed in the thread. Therefore the diameter of the remaining samples is reduced to 4 mm. The sample tested at 800 MPa has a diameter of 4.5 mm and the other samples have a diameter of 4.0 mm. Two samples of each material were stopped at 5·107 cycles and were tested again (indicated by the squares). One of the Ti-6Al-4V samples that was stopped at 5·107 cycles, fractured in the thread when tested at a higher stress. Since this sample, with a smaller diameter of 3.96 mm, failed in the thread, the stress concentration factor is close to the ratio of the area of the reduced section and the area at the thread. The area of the thread has a diameter of 9.70 mm; therefore the ratio of the areas is 6.0. Fatigue fracture surfaces The fatigue fracture surfaces of the Ti-6Al-4V samples show crack initiation from pores. The sample tested at 800 MPa showed nucleation from a large round pore (see Fig. 6). The diameter of this sample is 4.46 mm and is the same as for the other samples that fractured in the thread. Since this sample did not fail in the threat, the stress concentration factor of the pore is >6.0. For the sample tested at

E. Amsterdam and G.A. Kool

1266

700 MPa (with the lowest number of cycles), the crack initiated in a pore just beneath the surface. For the run out (RO) sample, tested again at 700 MPa, the crack initiated from a very small pore exactly on the centreline of the test sample. The fatigue fracture surfaces of the Inconel 718 samples show crack initiation from the surface. Near the initiation point, the fracture surface is facetted. Most area of the fracture surfaces are by overload and much porosity was observed on these areas for all the samples. None of the samples fractured in the thread and therefore Inconel 718 is not as notch sensitive as Ti-6Al-4V. From literature it is known that notch behaviour is a concern for Ti-6Al-4V [2]. The fracture surface of the sample tested at 800 MPa showed tracks of lack of bonding between layers.

DISCUSSION The yield stress and UTS values of the Ti-6Al-4V samples are similar to literature values for solution heat treated and aged material [2]. Both the yield stress and the UTS of the flat Ti-6Al-4V samples are larger than that of the round samples and the range of elongation to failure values is lower. Also the hardness of the material in wall B (thin) is equally larger (7.5%) than that of wall A (thick). This indicates that there is an intrinsic difference in strength between the two. This can originate from a difference in the microstructure, either created during the build-up with the laser due to potentially different wall temperatures or during the heat treatment from a possible difference in cooling rates. The Ti-6Al-4V component is water quenched during the heat treatment and therefore there are differences in cooling rates for the thin and thick walls. The high cooling rates are not present in the heat treatment of Inconel 718 and the difference in strength between the flat and round samples is also not observed. Since the fatigue samples are cut from a thick wall (A) the yield stress and the UTS are equal to that of the round tensile samples (also indicated by the hardness measurements). The yield stress and UTS of the Inconel 718 samples are also similar to the literature values for solution heat treated and aged material [3]. However, there is a very large difference in the elongation to failure compared with literature. To a large extent, this is due to macro defects such as lack of bonding between the layers (for the elongation to failure ≤2.4%). The sample with an elongation to failure >2.4% did not show signs of macro defects, but showed many pores. Even the fracture surfaces of the flat and round samples with an elongation to failure ≥14.2% showed many pores. The fatigue performance of Ti-6Al-4V is better compared with literature data on wrought annealed samples with a similar yield (925 MPa) and tensile strength (1000 MPa) [2]. However, when large pores, or subsurface pores, are present the fatigue performance is similar to wrought annealed samples. The reference data points in Fig. 5a correspond to samples of Ti-6Al-4V produced by the same

HCF of laser beam deposited Ti-6Al-4V and Inconel 718

1267

technique and producer [5]. Since Ti-6Al-4V is notch sensitive, the presence of a pore decreases the number of cycles to failure significantly (see Fig. 5a). The fatigue performance of the Inconel 718 samples is similar to annealed and aged sheet material [3], in spite of the many small pores. The presence of macro defects does have a detrimental effect on the fatigue performance, as can be seen for the 800 MPa sample.

CONCLUSION Two identical components of Ti-6Al-4V and Inconel 718 are fabricated by Optomec using the LENS™ technique. Both components contained many macro and micro defects, such as large cavities, lack of bonding between layers and porosity. The yield stress and ultimate tensile strength of both materials are similar to wrought material, however due to the defects the elongation to failure is lower and shows large scatter. The fatigue performance of Ti-6Al-4V is better than wrought material. However, the presence of pores decreases the performance significantly, making the fatigue performance similar to wrought Ti-6Al-4V. The fatigue performance of Inconel 718 is similar to wrought material, even though the material contains many small pores. Lack of bonding between layers decreases the fatigue performance of Inconel 718 accordingly. The laser powder deposition technique demonstrated to be capable of producing the same material properties as wrought material for component shaped geometries with both, thick and thin walls. However, further development work and process parameter optimization should be done to decrease the number of defects and increase the reliability and reproducibility.

REFERENCES [1] Hedges, M. and Calder, N. (2006), Meeting Proceedings RTO-MP-AVT139, Paper 13, Neuilly-sur-Seine, France [2] Donachie, M.J. (2000). Titanium – A Technical Guide (2nd Ed.), ASM int., Ohio. [3] Aerospace Structural Metals Handbook, 1997 ed. (1997), Purdue University, West Lafayette. [4] Oliveira, U.O.B de (2007), Laser treatment of alloys: processing, microstructure and structural properties, Doctoral thesis, University of Groningen, Groningen. [5] Grylls, R. (2006) LENS process white paper: fatigue testing of LENS Ti-6-4, Optomec internal report, www.optomec.com/downloads/EADS_Fatigue_Testing_ Technical_Brief_2006.pdf

E. Amsterdam and G.A. Kool

1268

Table 1 Chemical composition (wt%) of the components and the nominal composition of Ti-6Al-4V and Inconel 718 [3]. (a) Ti-6Al-4V Ti 90.21 Balance

Measured (avg. of 5 measurements) Nominal composition

Al 5.91 Min. 5.50 Max. 6.75

V 3.87 Min. 3.50 Max. 4.50

(b) Inconel 718 Measured (avg. of 5 meas.) Nominal Min. compositi Max. on

Ni 50.51

Cr 19.90

Nb 5.70

Mo 3.44

Ti 1.11

Al 0.71

Fe 18.64

50 55

17 21

4.75 5.50

2.80 3.30

0.65 1.15

0.20 0.80

Balan ce

Table 2 Porosity of (a) the Ti-6Al-4V and (b) the Inconel 718 component. (a) Ti-6Al-4V Section (wall, location, plane)

Section size (mm2)

Largest pore (μm)

>2 μm

Thick, top, 208 32 2 parallel Thick, top, perp. 224 25 6 Thin, top, parallel 196 25 1 Thin, top, perp. 29 Thick, bottom, 95 70 6 parallel Thick, bottom, 64 86 8 perp. Thin, bottom, 19 56 2 parallel Thin, bottom, 99 16 3 perp. The largest diameter is given as the size of the pore.

Number of pores 25-50 50-100 larger μm μm than 100 μm 1

-

-

1 1 -

-

-

4

3

-

3

3

-

2

1

-

-

-

-

HCF of laser beam deposited Ti-6Al-4V and Inconel 718

1269

(b) Inconel 718 Section (wall, location, plane)

Section size (mm2)

Largest pore (μm)

Pore density (pore/mm2) >2 25-50 50-100 larger μm μm μm than 100 μm

Thick, top, 22.1 112 11.4 1.0 0.2 parallel Thick, top, perp. 22.1 25 10.0 0.9 0.2 The largest diameter is given as the size of the pore. Note: the number of pores is given in (a) and the pore density in (b).

0.05 -

Table 3 Tensile properties of (a) round and (b) flat Ti-6Al-4V samples. The samples are orders from the lowest to the highest elongation to failure.

(a) Round samples Sample Orientation Yield stress UTS Elongation (MPa) (MPa) to failure (%) Ti-C6 LD 191 0.3 Ti-C4 LD 989 1089 9.2 Ti-C1 BD 943 1073 9.8 Ti-C5 LD 957 1057 10.9 Ti-C2 BD 935 1073 11.1 Ti-C3 BD 947 1051 13.3

(b) Flat samples Sample Orientation Yield stress UTS Elongation (MPa) (MPa) to failure (%) Ti-B2 BD 263 0.2 Ti-B5 LD 1025 1138 3.4 Ti-B9 LD 1076 1162 3.4 Ti-B8 LD 1085 1168 3.7 Ti-B4 LD 1014 1162 4.0 Ti-B6 LD 1063 1135 6.6 Ti-B7 LD 1065 1162 7.8 Ti-B10 LD 1034 1145 7.9 Ti-B3 BD 1062 1151 8.4 Ti-B1 BD 1028 1130 10.0

E. Amsterdam and G.A. Kool

1270

Table 4 Tensile properties of (a) round and (b) flat Inconel 718 samples.

(a) Round samples Sample IN-C4 IN-C5 IN-C6

(b) Flat samples A Sample IN-B5 IN-B7 IN-B8 IN-B10 IN-B11 IN-B9

Orientation Yield stress UTS Elongation (MPa) (MPa) to failure (%) LD 661 966 2.4 LD 998 1322 LD 1014 1350 18.4

C

Orientation Yield stress UTS Elongation to (MPa) (MPa) failure (%) LD B 635 958 2.6 LD 1002 1271 5.5 LD 1049 1296 6.5 S1057 LD 1362 LD 1107 1415 14.2 LD 1060 1369 15.3

For the samples IN-C5 and -B10, the elongation to failure is missing, since the sample failed outside the extensiometer area.

C A B S Inconel 718

Ti-6Al-4V

Figure 1 Photographs of the components fabricated with laser powder deposition. The letters indicate wall A, B, C and the substrate (S). The black circle on the Ti6Al-4V sample indicates a building defect.

HCF of laser beam deposited Ti-6Al-4V and Inconel 718

1271

Figure 2 Samples cut from the components. Left: fatigue sample, middle: round tensile sample, right: flat tensile sample, left top corner: microstructure sample (cross-section of the substrate and the thick wall).

a

b

Figure 3 (a) Grain structure of thin wall and the substrate of the Ti-6Al-4V component. Some pores can be seen on the boundary with the substrate (indicated by the white arrow). (b) Widmanstätten microstructure of the Ti-6Al-4V component.

E. Amsterdam and G.A. Kool

1272

a

b

Figure 4 Microstructure of the Inconel 718 component; (a) before etching, showing different grains, carbides (white) and porosity (black dots), (b) after etching showing the γ/γ´ distribution. Cuboids and spheroids can be observed.

HCF of laser beam deposited Ti-6Al-4V and Inconel 718

1273

900 R=0.1

max. stress (MPa)

800 700 600 500 400 300 200 100

(a)

0 1.0E+04

Ti-6Al-4V RO sample Ref. 5 1.0E+05

1.0E+06

1.0E+07

1.0E+08

1.0E+09

cycles

900 R=0.1

max. stress (MPa)

800 700 600 500 400 300 200 100 0 1.0E+04

(b)

Inconel 718 RO samples Ref. 3 1.0E+05

1.0E+06

1.0E+07

1.0E+08

1.0E+09

cycles

Figure 5 S-N curve of the (a) Ti-6Al-4V and (b) Inconel 718 samples fabricated with the LENS technique. Two samples of each material were stopped at 5·107 cycles (indicated by the arrows) and tested again at a higher stress level (run out (RO) samples, indicated by the squares). Orientation of the samples is in the buildup direction.

1274

E. Amsterdam and G.A. Kool

Figure 6 Fracture surface of the Ti-6Al-4V HCF sample that is tested at 800 MPa. A circular crack initiated and grew from a pore with a diameter of 102 μm.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

THE VARIABLE EXPANSION PROCESS – A NEW COST EFFICIENT METHOD FOR COLD WORKING FASTENER HOLES IN ALUMINIUM AIRCRAFT STRUCTURES Eggert D. Reese 1, Anthony L. Dowson2, Timothy G. B. Jones3 1

EADS, EADS Innovation Works, 81663 München, Germany 2 AIRBUS, Department of Material and Processes, New Filton House, Filton, Bristol BS99 7AR, United Kingdom 3 AIRBUS, Department of Material and Processes, Site de Saint Martin du Touch, 316 route de Bayonne, 31060 Toulouse Cedex 9, France

Abstract: Cold expansion of holes (‘cold working’) is a fatigue enhancement technique which is used extensively on metallic aircraft components to enhance the fatigue performance of a structural assembly. This paper reports on a new cold expansion tool, developed by EADS Innovation Works. The tool provides a number of potential benefits over the conventionally employed Split Sleeve or Split Mandrel tools. The specific design of the tool enables the control of variable expansion levels which in turn can compensate for hole tolerances and wear entailed changes in tooling diameter thus ensuring the same level of cold expansion for each hole is maintained. Furthermore, the tool generates a more homogeneous strain distribution around the hole which can be used for an extended exploitation of a material’s deformation behaviour. This may lead to an optimised and improved cold working of high strength anisotropic Aluminium alloys (e.g. 7xxx series alloys) using higher levels of applied expansion with a reduced risk of crack formation. There are also additional benefits associated with the tool compared with conventional cold working methods including potential reduced manufacturing costs and also the ability to cold work hybrid stacks in-situ (e.g. M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1275–1284. © Springer Science+Business Media B.V. 2009

1276

E.D. Reese, A.L. Dowson and T.G.B. Jones

aluminium/cfrp) whereby only a portion of the hole requires cold expansion for fatigue life improvement (e.g. cold expand aluminium part of stack only).

INTRODUCTION In service, aircraft structures are subjected to severe static and dynamic loads. In particular, fatigue performance of the metallic components is an important parameter in aircraft design. Fatigue vulnerable areas are often concentrated around fastener or open hole locations. Therefore, fatigue enhancement methods are applied to these critical locations prior to joint assembly in order to retard the potential for in-service crack initiation and growth. In addition, fatigue enhancement of critical joint regions is required to not only increase the fatigue performance of joint systems, but also to reduce weight and cost. During the 1960's, the aerospace technical community was presented a solution to the problem of metal fatigue at fastener holes. The process was simple, yet effective. Referred to as mandrelizing, it was utilized by the leading aircraft manufacturers, and at that time it entailed pulling an oversized tapered mandrel through a fastener hole in a two-sided operation. The hole increased in size, plastically deforming surrounding material which resulted in high compressive residual stresses around the hole. The zone of compressive residual stresses extends approximately one radius from the edge of the hole, depending upon variables such as material, hole diameter, and applied expansion levels as well as the effectiveness of the selected process. The resultant residual stresses counteract tensile forces which are acting on the hole. The zone is effective as a barrier to crack growth, reducing both the stress intensity factor acting on cracks originating from the fastener hole and the local stress ratio. Hence, crack formation and crack growth emanating from the hole are retarded or even suppressed thereby increasing the overall life of the component. Fatigue life enhancement is generally increased by a factor of from two to seven. In addition to significant fatigue life enhancement, cold working offers several other benefits. In the early 1970s, a pre-lubricated split sleeve was developed by the Boeing Company, which allowed for a one-sided operation and greater fatigue life improvement. This method is referred to as cold working and has become a common method to enhance the fatigue behaviour of fastener holes. Widely used methods are the “split sleeve process” by Fatigue Technology Inc. (FTI) and the “split mandrel process” by West Coast Industries (WCI). Drawbacks of current expansion processes Although the current processes provide reasonable results in fatigue enhancement of Aluminium alloys, they exhibit some disadvantages like inhomogeneous strain distribution around the hole, risk of crack formation near ridges, occasional tool

Variable expansion - process

1277

seizure due to high frictional loads between tool and surface of the bore, and – for the split sleeve process - a large number of sleeves which to dispose of is costly. Therefore, research for alternative processes is mainly driven by criteria like: further enhancement of fatigue resistance, possibly by a more homogeneous strain field surrounding the hole, application of cold working process to new alloy systems (which may not be cold workable with conventional expansion methods) – especially since cold working has become a material qualification issue, improvement of process parameters (e.g. reduction of friction and wear), reduction of process costs (e.g. avoid need for sleeves).

THE VARIABLE EXPANSION PROCESS The Principle A new process, named “Variable in-situ Expansion” process was established at the EADS Innovation Works laboratory in Munich, Germany, in collaboration with Airbus U.K.. Based on the notion that the presence of compressive residual stresses surrounding a fastener hole, introduced by plastic deformation, may significantly improve the fatigue performance of a mechanical joining, a tool was developed which allows the cold expansion of fastener holes in a variable manner. The Variable Expansion tool consists basically of two parts, a mandrel of conical shape and a sectioned bushing of corresponding geometry (Figure 1).

Mandrel

ess kn thic ck a t S

l ve Le

le Ho ter me dia

of

d plie ap

ion ns pa ex

Work piece

Split bushing

Metal

Figure 1: Principle of the variable in-situ expansion process.

For the expansion process, the bushing is placed into the hole to be cold worked. The tapered mandrel is inserted into the bushing. It is then pushed in an axial direction into the bushing thereby forcing the sections apart in a radial direction (Figure 1). The level of applied expansion is controlled by the displacement of the

1278

E.D. Reese, A.L. Dowson and T.G.B. Jones

mandrel, and the geometry of the bushing corresponds to the hole diameter and stack thickness to be cold worked. Finite element analysis of the Variable Expansion process The Variable Expansion process has been analysed by the finite element method. It could be shown that by a small modification of the tool a significant change in the stress/strain field around the hole from an inhomogeneous cloverleaf-like appearance to a more homogeneous circle-like geometry (Figure 2) could be obtained. Analysing the deformation limit of the material, it was noted that even highly anisotropic materials like 7xxx-series aluminium alloys which are difficult to cold work may now be cold expanded with a significantly reduced risk of crack formation. Figure 2 (right) also shows the strain concentrations located next to the pips, but the strain level achieved is still below the predicted point of crack initiation. This observation is contributed to the more homogeneous deformation characteristics generated by the Variable Expansion tool and was confirmed experimentally.

Figure 2: Strain distribution around a hole in a high strength aluminium alloy after 4 % applied expansion. Left: Variable Expansion – initial design. – Right: Variable Expansion – optimised design.

Experimental set-up The functionality of the Variable Expansion tool was tested using plates of various hole diameters (6 – 14 mm) and stack thicknesses (5 – 15 mm) employing different high strength Aluminium alloy systems (2xxx-, 7xxx-series). The expansion trials were conducted on a universal testing machine in displacement control mode. The Aluminium plate with the hole to be expanded was mounted to the cross-head. On the one (upper) surface, strain gauges were attached near to the hole to record the strains in both radial and hoop directions. In addition, the optical strain measurement system ARAMIS was employed. The Variable Expansion tool (in this set-up) was inserted from beneath the plate, the direction of the mandrel movement was therefore upwards. On the tip of the mandrel (at the upper surface of the Al-plate), an inductive displacement

Variable expansion - process

1279

transducer (Linear Variable Differential Transformer - LVDT) was mounted to record the mandrel displacement. Figure 3 provides on overview of the experimental set-up showing the upper side.

Figure 3: Experimental set-up for the Variable Expansion tool test.

The cross head speeds – representing the expansion rate - were selected to be between 2 - 200 mm/min, respectively. The tests were carried out at room LVDT temperature in a laboratory environment. Both, the mandrel and the bushing were lubricated with a commercially available lubricant paste. Cold working with the Variable Expansion tool Figure 4 shows the load-displacement curve obtained by cold working a 14 mm thick plate with a hole diameter of 14 mm. The displacement of the mandrel (cross head) may be seen as an indirect measure of the hole expansion. 20000

AA7xxx series alloys 18000

Plate thickness: 14 mm Nominal hole diameter: 14.3 mm (9/16 ")

16000

(Axial) Load [N]

14000 12000 10000 Variable in-situ expansion (VarEx) Cx applied: nominal: 4 % Cross head displacement: 2 mm /min; 20 and 100 mm/min

8000 6000

7085 - C6-1 7085 - C6-3 7085 - C6-5 7040 - I3-4 7040 - I3-5 - 20mm/min 7040 - I3-3 - 100mm/min

4000 2000 0 0

2

4

6

8

10

12

14

16

18

20

22

Displacement (cross head) [mm]

Figure 4: Load-displacement curve from cold working a hole in an Aluminium plate at various displacement rates.

E.D. Reese, A.L. Dowson and T.G.B. Jones

1280

The strain measurements were conducted via standard strain gauges and via the optical strain measurement system ARAMIS; in addition, the strains were estimated/converted from the LVDT-signal. Figure 5 displays the strain evolution during the expansion process comparing the strains on both sides (top and bottom) of the aluminium plate. The hoop strains were positive indicating tensile loading, where the larger values represent the deformation near the edge of the hole, the lower ones indicate the strains at a further distance from the hole edge. The radial strains are negative, i.e. compressive loading; again, the larger values were taken near the hole edge, the lower values at a further distance.

3,00

Applied strain [%]

2,00

1234

1,00

0,00 Hoop strain at P1-upper side

-1,00

Hoop strain at P1-lower side Radial strain at P2-upper side Radial strain at P2-lower side Hoop strain at P3-upper side

-2,00

Hoop strain at P3-lower side Radial strain at P4-upper side Radial strain at P4-lower side

-3,00 0

10

20

30

40

50

60

70

80

90

100

Time [s]

Figure 5: Hoop and radial strains determined via strain gauges.

The strain measurement results obtained at the top and the bottom surface of the Aluminium plate coincide very well indicating a simultaneous and homogeneous deformation pattern on both surfaces. This could readily be confirmed when evaluating the specimen surface after the expansion trials. Figure 6 shows both sides of an Aluminium plate with a cold worked hole. A deformation band around the hole can be observed, and the location of the splits of the bushing is clearly marked. A small region of material upset (“volcano effect”) can be detected at the hole edge. There is no significant difference in material upset between the front and the rear of the plate. Comparing the deformation behaviour of the split sleeve process with the Variable Expansion process, it is noticeable that the material upset is significantly larger for the split sleeve process and there is also a pronounced difference between entry and exit side.

Variable expansion - process

1281

Variable Expansion 3%

Split sleeve 3%

front

“entry”

rear

“exit”

Figure 6: Material upset at the surface after 3 % applied expansion. Split sleeve process (left), Variable Expansion process (right).

Fatigue behaviour of samples with cold worked holes An important – if not the most important – question to be answered is the one addressing the fatigue enhancement capability of the new process. 300

Stress [MPa]

250

VarEx, 4 % applied

200 VarEx, 3 % applied Split sleeve, 3 % applied

(data by EADS)

150

7085 (AIMS spec)

100 1000

10000

100000

1000000

10000000

Number of cycles to failure

Figure 7: S-N-curves (open hole) for different expansion methods and levels.

E.D. Reese, A.L. Dowson and T.G.B. Jones

1282

In order to investigate the fatigue properties of specimens containing holes that have been expanded using the Variable Expansion process, fatigue tests were conducted utilising a servo-hydraulic test machine and standardised open-hole fatigue specimens with Kt = 2.43-value. The tests were performed at room temperature in a laboratory environment. Figure 7 displays the S-N-curves for (a) specimens which had been cold worked at 3 % expansion by the Variable Expansion method and the Split Sleeve process, and (b) specimens that had been cold worked at 4 % expansion by the Variable Expansion method. The fatigue life of all specimens was significantly enhanced by both expansion methods when compared with non-cold worked specimens as given by the material specification. Both the Variable Expansion and Split Sleeve methods were yielding the same fatigue life enhancement. More interesting though is the fact that, for the alloy tested, even 4 % applied expansion was successfully applied leading to an even greater fatigue enhancement whereas cracks at the hole edges were observed (and therefore could not be tested) when this alloy was cold expanded utilizing the Split Sleeve process. The fatigue specimens were then analysed with regard to the fracture appearance. Figure 8 depicts two specimens which were cold worked at approximately 3 % by the conventional Split Sleeve method and by the Variable Expansion process respectively. They were then fatigued at a stress level of approximately 230 MPa at an R-value of R = 0.1; failure occurred after approximately 300.000 cycles. Variable Expansion

Nf

300.000 cycles

Split sleeve

Nf

300.000 cycles

Entry side

Figure 8: Comparison of fracture appearance between Variable Expansionand Split Sleeve expanded holes.

Variable expansion - process

1283

Although the fracture occurred after a comparable number of cycles, the appearance of the fracture surfaces reveals pronounced differences. For the Split Sleeve specimen, the fracture surface shows a “triangular” pattern for the region of stable fatigue crack growth indicating that the fatigue crack initiated at the entry side. The fracture surface of the Variable Expansion specimen displays a thumbnail shape as it is known from standard fatigue crack growth studies. The crack nucleated at the centre of the plate and propagated evenly to either surface. This supports the conclusion of a homogeneous strain field being generated by the Variable Expansion process.

DISCUSSION The variable in-situ expansion tool represents an alternative to conventional cold expansion tools. It is based on a mechanical principle similar to e.g. the split sleeve and the split mandrel process. The mandrel is of conical shape and together with a bushing of corresponding geometry, the axial load applied to the mandrel is transformed into a radial load which forces the initial hole diameter to expand to a predefined final size. Thus, compressive residual stresses are induced which counteract e.g. tensile forces retarding fatigue crack growth and so significantly enhance the fatigue life of a structural component. Main differences to the “split sleeve” process and other similar methods are based on the fact that the expansion process is applied to the entire plate thickness in “at once” whereas the expansion of the split sleeve (or split mandrel) process occurs gradually throughout the thickness as the mandrel is pulled through. The Variable Expansion process offers several potential benefits. With regard to ‘process control’ the benefits are (a) access of the hole from one side only (one-sided operation), and (b) variable in-situ expansion level allowing for compensation of hole tolerances and of wear entailed changes in tooling diameter which in turn may enable extended / prolonged use of the tooling. The expansion of the hole across the entire thickness of the component simultaneously results in a homogeneous expansion of the plate material, thus (a) not only avoiding the differences between ‘entry’ and ‘exit’ side – as known from the Split Sleeve and Split Mandrel process -, but (b) also allowing for an increased exploitation of the material’s deformation potential i.e. expansion to higher expansion levels, e.g. 5 to 6 %, without the risk of crack formation, as could be shown experimentally and was also confirmed by the Finite Element study. This is advantageous enabling the application of these with an option of weight and cost savings, in particular, for new generation high strength Aluminium alloys which cannot be cold worked using conventional methods.

E.D. Reese, A.L. Dowson and T.G.B. Jones

1284

Further, normal forces acting on the component surfaces will be avoided, i.e. no bending moments emerge. In addition, there is no relative motion between the bushing and the surface of the bore (as for the Split Mandrel process) eliminating the danger of tool seizure. There is a significant cost savings potential because there is no need for ‘sleeves’ (‘Split Sleeve process’) and no need to dispose of these sleeves. ‘No need for taperlocks’ and significant savings in manual labour may also reduce costs. There may be an ease of inspection: for the split sleeve cold expansion, the cracks emanate from the entry side, e.g. the inner side of the wing structure which cannot be inspected. For Variable Expansion process, cracks emanate from the centre of the hole and surface on both sides of the component. The Variable Expansion process, in conjunction with FE simulation, may be used for estimation of expansion level in repair situations. There is a significant potential for joining dissimilar materials (e.g. application to hybrid stacks) in a one-step assembly process since the metal component may be cold worked without affecting the composite part.

SUMMARY A new cold expansion process, named “Variable in-situ Expansion” was developed at the Munich based EADS Innovation Works laboratory in collaboration with Airbus U.K.. The Variable Expansion tool consists of a tapered mandrel and segmented bushing. It allows the continuous cold working of a hole in the range from (theoretically) zero to several percent applied expansion. The tool/process offers a number of benefits in the areas of process control, exploitation of a material’s deformation potential, repair, joining dissimilar materials, costs savings and allows to cold work even difficult to expand high strength Aluminium alloys without crack formation. A patent is pending, WO2007121932.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

STRESS MEASUREMENTS WITH X-RAY DIFFRACTOMETRY ON ALUMINUM ALLOYS. DETERMINATION OF THE MOST OPTIMIZED PARAMETERS OF THE MEASUREMENT Elżbieta Gadalińska, Jerzy Kaniowski 1, and Andrzej Wojtas2 1

Institute of Aviation Al. Krakowska 110/114, 02-256 Warsaw, Poland 2 METLAB ul. Odcinek 19, 51-522 Wroclaw, Poland

Abstract: X-ray diffractometry is one of the basic methods of stress measuring. This method was used to measure stress distributions around rivets as described further in this paper. There were two types of riveted samples, nine types of samples made of rivet wire (after different types of treatment), the aluminium sheet sample with three measurement areas: plate with both cladding and anodized layer, plate after removing the anodized layer and plate after removing both cladding and anodized layer and the alminum alloy plate sample with chemically milled area which was anodized after that. Riveted samples were prepared to measure the stress distribution around the rivets and the samples of wire and the plate with three areas were prepared to check the effect of different types of treatment on stress state. The sample with the chemically milled cladding was prepared to choose the optimized measuremet parameters for stress measurements on alminum.

INTRODUCTION X-ray diffractometry serves as a one of a number of methods for residual stress evaluation. The essential advantage of this method is that it is the accurate and M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1285–1304. © Springer Science+Business Media B.V. 2009

1286

Elżbieta Gadalińska, Jerzy Kaniowski and Andrzej Wojtas

absolute. The diffraction of X-rays on a crystal lattice of a copper sulphate was detected by Max von Laue in 1912. The mathematical formula for this phenomenon was expressed by William Bragg and it is as follows: nλ = 2d sin Θ

(1)

where: n - is an integer determined by the order given, λ - is the wavelength of diffracting X-ray, d - is the spacing between the planes in the atomic lattice, sinΘ - is the angle between the incident ray and the scattering planes. The formula shows that the diffraction will occur only for unique Bragg’s angles. Since its publication in 1913 the Bragg’s law has been a powerful tool for studying the crystals’ structure. One of its applications is the residual stress measurement. The formula for stress calculation in a crystal is as follows: ⎛ E ⎞ 1 ⎛ ∂dϕψ ⎜⎜ ⎟ 2 ⎝ 1 + ν ⎠ hkl d 0 ⎝ ∂ sin ψ

σϕ = ⎜

⎞ ⎟⎟ ⎠

(2)

Fig. 1: X-ray diffraction parameters. where: φ - is the angle between the projection onto a plane of a specimen and the direction of σ11 ψ - is the angle between the incident beam and the perpendicular to the sample surface, σφ- is the surface stress in direction determined by angle, E – is the Young modulus, ν - is the Poisson ratio, d0– is the stress-free lattice spacing, dφψ - is the lattice spacing in the direction determined by the angles φ, ψ, The stress values are obtained from the graph of the function of dφψ depending on sin2ψ.

Stress measurements with X-ray diffractometry on aluminum alloys.

1287

Fig. 2: The example of stress measurement results obtained with Xstress3000 All the stress measurement results presented here were obtained using a Stresstech Oy built X-ray diffractometer model Xstress3000. It is a mobile device capable of performing measurements in two modes: ψ and Ω, with Φ rotation and Ψ and Φ oscillation. There is a set of collimators from the 0,5mm diameter to 5mm. This equipment is fully automated with a software controlled X-Y table for sequential measurements at a programmable set of points with given X-Y coordinates. The purpose of this study was to determine the residual stress distribution around the rivets and compare it with the FEM analysis performed by R.P.G. Müller in his Ph.D thesis [1].

SPECIMENS Four types of specimen were used: 1. Large specimens – metal sheets made of D16CzATV alloy, 2mm thick; rivets made according to the standard OST 1 34040 – mushroom head rivets with compensator.

1288

Elżbieta Gadalińska, Jerzy Kaniowski and Andrzej Wojtas

Fig. 3: Large specimens - photo

Fig. 4: Large specimens - drawing 2. Small specimens – made of D16AT alloy, 1.2mm thick sheet; rivets made of W65 alloy, according to the Russian standard ANU 0301, adhered by PZL Mielec – countersunk rivets with compensator. 3.

Fig. 5 Small specimens - photo

Stress measurements with X-ray diffractometry on aluminum alloys.

1289

Fig. 6: Small specimens - drawing 4. Specimens made of rivet wire (two types: 3,5mm and 5mm diameter) after different treatment (first set): • 1: polished with abrasive paper grade 100 • 2: polished with abrasive paper grade 100, and then 280 • 3: polished as 2, and then grade 600 • 4: polished as 3, and then grade 1000 • 5: polished as 4 and polished with diamond paste, grain size 3μm • 6: polished as 5, and then polished electrolytically for 15 sec.

Fig. 7: Specimen made of wire for rivets 5. Specimens made of rivet wire (5mm, made of PA25 alloy) after different treatment (second set):

Elżbieta Gadalińska, Jerzy Kaniowski and Andrzej Wojtas

1290

• • •

1: crumpled with a force of 3 tones, polished with abrasive paper grade 800 and then polished electrolytically, 2: cut with fretsaw and polished with abrasive paper grade 120, 500, 800 and then polished with diamond paste, grain size 3μm, 3: polished as 2 and annealed in 500°C for ½h.

Fig. 8: Specimen made of wire for rivets 6. Specimen of a clad and anodized plate with three measurement areas: (1) original plate, (2) after removing the anodized layer with diamond paste, (3) after removing the anodized layer and cladding with abrasive paper grade from 280 to 800 and diamond paste, grain 3 μm.

Fig. 9: Specimen with three measurement areas 7. Plate with area of chemical milled cladding.

Stress measurements with X-ray diffractometry on aluminum alloys.

1291

Fig. 10: Plate with area of chemical milled cladding

FEM ANALYSIS RESULTS The first series of results obtained from FEM analysis by R.P.G Müller predicted the stress distribution on the mating surface, for rivets with a driven head on both sides (for radial and tangential stresses) and for countersunk rivets (for tangential stresses only). The results are presented below in figures 11, 12, 13 (on the scale normalized to the yield point):

Fig. 11: Residual radial stresses along the mating surface for rivets with a driven head on both sides [1]

1292

Elżbieta Gadalińska, Jerzy Kaniowski and Andrzej Wojtas

Fig. 12: Residual tangential stresses along the mating surface for rivets with a driven formed head on both sides [1]

Fig. 13 Residual tangential stresses along the mating surface for countersunk rivet [1] The second series of results predicted the stress distribution through the thickness at the edge of the hole for rivets with a driven head on both sides (radial stresses only) and for countersunk rivets (radial and tangential stresses). Results of modelling are presented below on figures 14, 15, 16:

Stress measurements with X-ray diffractometry on aluminum alloys.

1293

Fig. 14 Residual radial stresses through the thickness at the edge of the hole for rivets with a driven formed head on both sides [1]

Fig. 15 Residual radial stresses through the thickness at the edge of the hole for countersunk rivets [1]

1294

Elżbieta Gadalińska, Jerzy Kaniowski and Andrzej Wojtas

Fig. 16 Residual tangential stresses through the thickness at the edge of the hole for countersunk rivets [1]

Measurements were performed at Stresstech Oy on the surface of the specimens so the comparison of the measurement results and the FEM analysis results is limited. For every specimen the appropriate value D/D0 is 1,5.

RESULTS OF X-RAY DIFFRACTION MEASUREMENTS The X-ray measurements were performed on the surface of the specimen, on the side with machined rivet head. The measurements were performed in radial (phi=0) and tangential directions (phi=90°). The anodized layer and the cladding were not removed before measurement.

Stress measurements with X-ray diffractometry on aluminum alloys.

Fig. 17: Large specimen during the measurement. Directions of measured stresses

Fig. 18: Small specimens. Directions of measured stresses

1295

Elżbieta Gadalińska, Jerzy Kaniowski and Andrzej Wojtas

1296

Results are presented below:

0,6

0,4

ΔσΘ/σy

0,2

0,0

-0,2

-0,4

-0,6 1,0

1,5

2,0

2,5

3,0

3,5

4,0

4,5

5,0

5,5

6,0

6,5

7,0

r/R

Fig. 19: The tangential stresses increment ΔσΘ on the surface of the specimen on the side of it with machined rivet head (large specimen) 0,2

0,0

Δσr/σy

-0,2

-0,4

-0,6

-0,8

-1,0 1,0

1,5

2,0

2,5

3,0

3,5

4,0

4,5

r/R

Fig. 20 The radial stresses increment Δσr on the surface of the specimen on the side of it with machined rivet head (large specimen)

5,0

Stress measurements with X-ray diffractometry on aluminum alloys.

1297

0,6

0,4

ΔσΘ/σy

0,2

0,0

-0,2

-0,4

-0,6 1,0

1,5

2,0

2,5

3,0

3,5

4,0

4,5

5,0

5,5

6,0

6,5

7,0

r/R

Fig. 21 The tangential stresses increment ΔσΘ on the surface of the specimen on the side of it with machined rivet head (small specimen riveted manually: M3) 0,0

-0,2

Δσr/σy

-0,4

-0,6

-0,8

-1,0 1,0

1,5

2,0

2,5

3,0

3,5

4,0

4,5

5,0

r/R

Fig. 22 The radial stresses increment Δσr on the surface of the specimen on the side of it with machined rivet head (small specimen riveted manually: M3)

Elżbieta Gadalińska, Jerzy Kaniowski and Andrzej Wojtas

1298 0,6

0,4

ΔσΘ/σy

0,2

0,0

-0,2

-0,4

-0,6 1,0

1,5

2,0

2,5

3,0

3,5

4,0

4,5

5,0

5,5

6,0

6,5

7,0

r/R

Fig. 23 The tangential stresses increment ΔσΘ on the surface of the specimen on the side of it with machined rivet head (small specimen riveted with press: P1)

0,0

-0,2

Δσr/σy

-0,4

-0,6

-0,8

-1,0 1,0

1,5

2,0

2,5

3,0

3,5

4,0

4,5

5,0

r/R

Fig. 24 The radial stress increment Δσr on the surface of the specimen on the side with machined rivet head (small specimen riveted with press: P1)

Stress measurements with X-ray diffractometry on aluminum alloys.

1299

Comparison of the graphs presented above and the results obtained by R.P.G. Müller leads to the conclusion that the character of graphs obtained during the residual stresses measurement is inconsistent in each case.

Large specimens The character of the graph on fig. 12 is inconsistent with the character of graph on fig. 19 and the relative value of tangential stresses in point r/R=1 wasn’t the same as that determined by FEM analysis. The graph in the fig. 11 is inconsistent with the graph on the fig. 20: the residual radial stresses predicted by FEM analysis were positive and measured values were negative. Additionally the relative value of radial stresses for r/R=1 is totally different than that obtained in FEM analysis.

Small specimen riveted manually The character of the graph presented in fig. 13 is inconsistent with the character of the graph given on fig. 21 The relative value of the residual tangential stresses in point r/R=1 is the same as obtained in FEM analysis (fig. 16). The FEM analysis for radial stresses for countersunk rivets weren’t performed. The relative value of residual stresses in r/R=1 is the same as in the FEM analysis (fig. 15).

Small specimen riveted with press The character of the graph presented in fig. 13 is inconsistent with the character of graph given in fig. 23 The relative value of the residual tangential stresses in point r/R=1 is the same as obtained in FEM analysis (fig. 16). The FEM analysis for radial stresses for countersunk rivets weren’t performed. The relative value of residual stresses in r/R=1 is the same as in the FEM analysis (fig. 15). The lack of consistency between the results of stress measurements and the FEM analysis forced the authors to analyze the measurement method and to improve it. It was stated that the reason for the error in stress measurement was the anodized layer and the cladding so it was necessary to work out the method of removing those layers to obtain more reliable results. The additional aim of preparing this type of specimen was to work out the method of removing the formed rivet head. To check if the applied treatment hasn’t changed the stress state measurements were performed on rivet wire specimens, after different kinds of treatments. The results of stress measurements on those samples are listed below:

Elżbieta Gadalińska, Jerzy Kaniowski and Andrzej Wojtas

1300

Stress phi=0° Tangential direction MPa ±MPa -63 4 -69 3 -28 3 -38 3 -39 5 -3 3

Rivet no 1 2 3 4 5 6

Stress phi=90° Axial direction MPa ±MPa -50 5 -51 4 -41 3 -36 5 -44 2 -36 5

FWHM phi=0° ° 1.43 1.48 1.49 1.47 1.43 1.42

FWHM phi=90° ° 1.46 1.47 1.48 1.45 1.43 1.45

Table I: Results of the measurements on 3.5 mm diameter rivets, phi=0°corresponds to tangential direction and phi=90° to axial direction (first set) Rivet

Stress phi=0° Tangential direction MPa ±MPa -56 4 -76 5 -68 2 -52 10 15 3 -6 4

no 1 2 3 4 5 6

Stress phi=90° Axial direction MPa ±MPa -49 6 -68 3 -61 5 -58 6 -15 5 -11 8

FWHM phi=0° ° 1.44 1.46 1.47 1.43 1.47 1.41

FWHM phi=90° ° 1.45 1.46 1.47 1.44 1.45 1.41

Table II: Results of the measurements on 5 mm diameter rivets, phi=0°corresponds to tangential direction and phi=90° to axial direction (first set) 100

50

Stress / MPa

0

-50

-100

-150

-200 1.0

2.0

3.0

4.0

5.0

[index] / rivet Diameter_3.5mm_Phi: 0.0° tangential direc Diameter_3.5mm_Phi: -90.0° axial direction Diameter_5mm_Phi: 0.0° tangential directi Diameter_5mm_Phi: -90.0° axial direction

Fig. 25 Residual stresses on different rivets

6.0

Stress measurements with X-ray diffractometry on aluminum alloys.

1301

2.0

1.8

FWHM

1.6

1.4

1.2

1.0 1.0

2.0

3.0

4.0

5.0

6.0

[index] / rivet Diameter_3.5mm_Phi: 0.0° tangential direc Diameter_3.5mm_Phi: -90.0° axial direction Diameter_5mm_Phi: 0.0° tangential directi Diameter_5mm_Phi: -90.0° axial direction

Fig. 26 FWHM values on different rivets. Rivet

no 1 2 3

Stress phi=0° Tangential direction MPa ±MPa -101 8 -67 2 -40 7

Stress phi=90° Axial direction MPa -113 -75 -32

±MPa 5 4 2

FWHM phi=0°

FWHM phi=90°

° 1.41 1.43 1.20

° 1.44 1.44 1.21

Table III: Results of the measurements on 5 mm diameter rivets, phi=0°corresponds to tangential direction and phi=90° to axial direction (second set). The surface layers of the specimens were removed with 6 different types of treatment. Removed layers were thick enough and the treatment of step 5 and 6 gentle enough to uncover the core material of wire without changing its stress state. The results obtained for the second set of specimens showed that the X-ray diffractometry stress measurements are reliable. Specimen of plate with three areas The measurements on those three areas were performed before the application of any additional treatment (depth=0,000) and after removing a few microns of the

Elżbieta Gadalińska, Jerzy Kaniowski and Andrzej Wojtas

1302

surface layer by electropolishing. The aim of applying the additional treatment was to remove contamination of the specimen surface

Fig. 27: Plate with three different areas. Directions of measured stresses.

Plate Depth 0,000 0,006

Plate Depth 0,000 0,005 0,008

Plate Depth 0,000 0,005

Stress phi=0°

Stress phi=90°

FWHM phi=0° MPa ±MPa MPa ±MPa ° -24 7 6 10 1,30 -55 5 -25 10 1,34 Table IV: Residual stress values on point 1

FWHM phi=90° ° 1,31 1,34

Stress phi=0°

Stress phi=90°

FWHM phi=90° ° 1,32 1,34 1,35

MPa -35 -35 -53

FWHM phi=0° ±MPa MPa ±MPa ° 5 -68 5 1,31 5 -54 5 1,34 7 -61 8 1,34 Table V: Residual stress values on point 2

Stress phi=0° MPa -53 -45

±MPa 4 3

Stress phi=90° MPa -54 -60

±MPa 3 5

FWHM phi=0° ° 1,33 1,31

FWHM phi=90° ° 1,33 1,31

Table VI: Residual stress values on point 3 The difference between the results obtained for area no. 1 before applying the electropolishing and after removing of 6 microns of surface layer was caused

Stress measurements with X-ray diffractometry on aluminum alloys.

1303

by different number of grains of aluminium participating in diffraction. The X-ray penetration depth is about 13 microns and the anodized layer thickness is about 7 microns. After removing of 6 microns of the surface layer the measurement was performed almost only in cladding layer. Plate with area of chemical milled cladding.

Fig. 28: Stress measurement result for the specimen of plate with chemically milled area. Stress phi=0° MPa ±MPa -8,1 4,8

Stress phi=90° MPa ±MPa -38,1 6,9

The measurement was performed with 3mm collimator and with the φ-oscillations. This kind of measurement on an anodized plate with chemically milled area is possible. The next step is to find optimal measurement parameters for 0,5mm to enable stress gradient measurements.

CONCLUSIONS The stress measurements around the rivets were not satisfactory enough. It was necessary to work out a method of removing the anodized layer and cladding to

1304

Elżbieta Gadalińska, Jerzy Kaniowski and Andrzej Wojtas

allow the stress measurements in the core material and compare them with the FEM analysis results.

REFERENCES [1]

Muller, R. P. G., An Experimental and Analytical Investigation on the Fatigue Behavior of Fuselage Riveted Lap Joints, Ph. D. Thesis, Delft University of Technology, The Netherlands, 1995, p.161.

[2]

Hakanen M., Institute of Aviation – XRD measurements on riveted samples, non published report No #699, Stresstech Oy, Vaajakoski 2007

[3]

Hakanen M., Institute of Aviation – Residual stress on rivets, non published report No #744, Stresstech Oy, Vaajakoski 200.

[4]

Cullity B.D. Podstawy dyfrakcji promieni rentgenowskich, Państwowe Wydawnictwo Naukowe, wyd. I, Warszawa 1964.

[5]

ASM Handbook, Volume 9: Metallography and Microstructures 2004.

[6]

ASM Handbook, Volume 2:0 Properties and selection: Nonferrous Alloys and Special-Purpose Materials.

[7]

Industrial standard of Soviet Union, Mushroom head rivets with compensator, No. OST 1 34040-79.

[6]

Internat standard of PZL Mielec, Countersunk rivet, No. ANU 0301.

25th ICAF Symposium – Rotterdam, 27–29 May 2009

DELAMINATION GROWTH AT INTERFACES IN HYBRID MATERIALS AND STRUCTURES UNDER VARIOUS OPENING MODES G. Delgrange, R.C. Alderliesten and R. Benedictus Aerospace Materials & Structures, Delft University of Technology, Delft, the Netherlands

Abstract: This paper describes the investigation on the contribution of Mode I and II to the fatigue failure in bonded metallic structures and hybrid metallic materials. Double cantilever beam opening tests (pure mode I) and end notched flexure bending tests (pure mode II) have been performed on specimens with aluminium & glass fibre/epoxy interfaces. Both static and fatigue loading have been applied. In case of fatigue loading, the delamination growth rate has been recorded. The test data has been evaluated using models that link the strain energy release rate to the delamination growth with a Paris-type equation. The purpose of this evaluation is to develop a model to predict the delamination for an arbitrary geometry and loading. Finally, the potential of the developed model, for example the application to finite element models and the extension from basic specimen geometries to more global structures, is discussed.

INTRODUCTION For aircraft components made from typical aerospace aluminium alloys and fastened together by rivets or bolts, there are two natural limits which determine the fatigue life and the damage tolerance. The first limit is the length of time (the ‘fatigue life’) that aluminium remains intact under cyclic load before cracks appear at stress concentrations caused by fastener holes. The second limit is the relatively fixed ability of the microstructure of aluminium alloys to retard the speed of propagating cracks (a measure of ‘damage tolerance’). M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1305–1319. © Springer Science+Business Media B.V. 2009

1306

G. Delgrange, R.C. Alderliesten and R. Benedictus

There are several ways to extend the fatigue life and increase the damage tolerance of such component. Two of the more promising solutions involve firstly adhesively bonding of aluminium parts to remove holes and improve the fatigue life, and secondly through bonding additional reinforcing layers of other materials, such as glass-fibres, to the aluminium structure (thus creating a ‘hybrid’ material) to reduce the crack opening normally experienced in monolithic aluminium alloys under load. The objective of this research is to better understand the fatigue and damage tolerance behaviour of bonded metallic hybrid metallic structures to account for it in a true damage tolerance evaluation. The typical failure associated with bonded structures is known as delamination. So far, methods to predict simple in-plane crack opening (Mode II) for onedimensional delamination under shear in thin bonded and hybrid materials exists [1]. However, a tensile mode of failure (mode I) is naturally induced if material thicknesses increase for example by bonding additional material layers, or in case of bonded joints between components,. The understanding of the contribution of both these modes of failure to delamination will greatly increase the ability of a designer to specify materials and structures that will either prevent delamination from occurring or limit its growth within scheduled inspection intervals.

EXPERIMENTAL PROCEDURE Material and specimens The specimens that were used for both mode I and mode II tests were cut from a GLARE panel. This GLARE panel was made of two Al 2024-T3 layers of 0.4 mm thickness with in-between two uni-directional fibre layers. It is coded as Glare 2A2/2-0.4, which refers respectively to the GLARE grade, the lay-up and the aluminium thickness. More details on other glare configuration can be found elsewhere [ref?]. In both type of tests, such a thin specimen would be too flexible and would plastify too easily. In order to prevent an unwanted plasticisation of the aluminium layers, two external doublers made of Al 7075-T6 of 4.1mm thickness were bonded at each side. The doublers were bonded using a FM94 resin. The full lay-up is illustrated in Figure 1.

Delamination growth at interfaces in hybrid materials and structures

1307

Figure 1: The lay-up of the specimens The GLARE panel was fabricated at Delft University of Technology with a length of 915 mm and a width of 350 mm. Specimen of 320 mm length and 25 mm width were cut from this panel. In order to initiate a pre-crack, a 0.02 mm thickness Teflon layer was introduced between the two fibre layers. The initial crack length was about 50 mm. Holes were drilled in each specimen to enable screwing hinges for the mode I tests. To complement these tests, a few specimens made of 24 plies of 0 degree glass fibres were manufactured. The geometry of those specimens was the same as the GLARE specimens previously described. The purpose of testing pure glass fibres specimen is twofold. First, the transparency will enable to observe the delamination front. Second, the data will be used in the future to compare the delamination behaviour with the GLARE specimens (same delamination interface but different internal structure). Test set-up and procedure Despite the differences in mode I and mode II testing, the same procedure was followed to perform the tests. All tests were conducted at a constant displacement to prevent any unstable behaviour (load control could lead to a hazardous response in displacement). Fatigue tests were performed at a frequency of 5Hz. First, a series of static tests were performed at different delamination lengths in order to get critical displacements δcrit. Then fatigue tests were performed at different amplitudes below the critical displacement to get a range of values for the energy release rate. On each specimen manufactured, a pre-crack was initiated by a static loading. I avoid to make measurement error due to an artificial straight front interface or some residual resin at the interface. Tests were performed on a 10kN fatigue machine. Force and displacement were continuously recorded: for each cycle the maximum and minimum value of the force was saved. Due to the filtering process of the acquisition software, the force recorded is not the exact maximum (or minimum) force, but a value very close to that. To determine the exact maximum and minimum force afterwards, the exact position of the actuator is recorded respectively with the force. The correction then consists of linearly relating force with displacement. The delamination growth was

G. Delgrange, R.C. Alderliesten and R. Benedictus

1308

monitored with a digital camera. Pictures were regularly taken for a later precise measurement. Mode I tests. The test selected for mode I was the standard test according ASTM: the double cantilever beam specimen (DCB) [2]. As mentioned before, holes were drilled at the pre-delaminated side of the specimen and piano hinges were screwed on the outer doublers (see Figure 2).

Hinge

Mode I Figure 2: the double cantilever beams specimen (Mode I) The standard way to localise the crack tip consists of using some correction fluid to paint the edge of the specimen. As the delamination grows, the brittle correction fluid cracks along with the delamination enabling localisation of the delamination tip. For different reasons which have not yet been indentified, this method gave unfortunately bad results, and the delamination tip could hardly be localised (a lot of scatter; delamination tip not clearly defined). In order to solve this problem, another method based on digital image correlation (DIC) was used (see Figure 3). The edge of the specimen was painted with a random black and white speckle pattern and a programme developed at TU Delft was used to calculate the strain in the x and y-directions based on the deformation of the speckle pattern. After calculation of the strains, the location where the strain in y-direction becomes zero corresponds to the delamination tip. This method gave more accurate results than the correction fluid.

Delamination growth at interfaces in hybrid materials and structures

1309

Figure 3: Implementation of the DIC for mode I tests. The most surprising observation during the DCB tests is that the delamination did not progress at the pre-crack between the two fibre layers, where it was expected (in the pre-delaminated plane). The delamination jumped to the interface between fibres and aluminium. It seems that in mode I the fibre/aluminium interface is weaker than the fibre/fibre interface. Moreover, in some specimens the delamination occurred at both aluminium/fibre interfaces at the same time. The number and location of delamination fronts could not be estimated a priori. The fact that the delamination ‘jumps’ to an unexpected interface may be one of the reasons of the problems encountered with the correction fluid method. An unexpected consequence of this behaviour is that due to residual stresses, the fibres that are free at both sides wave, inducing compression in the specimen at small amplitude opening. This closure phenomenon had to be taken into account in the data reduction as explained later. Mode II tests. The end-notched flexure (ENF) specimen, which is essentially a three-point bending specimen with a delamination at one extremity (see figure 4), was selected for the mode II characterisation. ENF specimens have been used

G. Delgrange, R.C. Alderliesten and R. Benedictus

1310

several times before to characterise mode II delamination [3, 4]. This test is known to be very instable: the delamination growth can increase quickly and the displacement response to a load control can be hazardous. R2

R1

R1 Figure 4: the end notched flexure specimen (Mode II)

The specimen dimensions of the ENF tests were (see Figure 4) - specimen half-length L = 50 mm - lower radius R1=5 mm - upper radius R2=12.5 mm Here, the traditional method employing correction fluid worked properly and was thus used. The delamination occurs at the same interface as the pre-crack: between the two fibre layers. Test matrix The tests performed are summarised in table I. Not all static tests are, because a lot of tests were needed to fully characterize the behaviour. The complete list would not be relevant. Mode I

II

Nr St 1 2 3 4 5 St 1 2 3 4 5 6 7 8

δmax Up to 6.7 2.7 1.11 5.36 5.36 2 Up to 1.14 0.5 0.5 0.5 0.5 0.4 0.35 0.45 0.3

Stress ratio NA 0.1 0.1 0.1 0.1 0.1 NA 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1

Comments 10 static tests with different a0 Delaminated at the upper interface Delaminated at the lower interface Delaminated at both interfaces Delaminated at both interfaces Delaminated at the upper interface 10 static tests with different a0 Delaminated at the glass fibre/glass fibre interface.

No delamination occurred

Table I: test matrix of performed tests

Delamination growth at interfaces in hybrid materials and structures

1311

Data evaluation and scatter reduction The purpose of the performed tests is to find a correlation between the delamination growth rates (da/dN) and the strain energy release rate SERR (G) applied during the fatigue loading, similar to previous studies on other materials. A Paris-type law similar to the one that is used for crack growth (see figure 5), has been proposed in the past [5].

Figure 5: qualitative correlation between crack growth curve and delamination curve In Eqn. 1 Cd and nd are parameters to be determined with experiments

∂a = Cd ⋅ ΔG nd ∂N

(1)

The range ΔG is related to the difference between the maximum and minimum SERR levels of the specimen during the fatigue cycle (Eqn. 2).

ΔG =

(

Gmax − Gmin

)

2

(2)

This range is the actual parameter in the equation rather than the maximum level of the SERR, as adopted in previous studies. Because all tests have been performed at a stress ratio R=0.1, any of the earlier proposed relations for ΔG will be sufficient (no stress ratio effect). In this paper the formulation for the range as given by Eq. 2 will be used. There are a lot of different formulations proposed in the literature to calculate the SERR for the applied test and specimen geometry [6-8]. Each formulation is related to the specific test performed and the approximations that have been made.

G. Delgrange, R.C. Alderliesten and R. Benedictus

1312

Nevertheless all the formulations are derived from a generic relation given by Eqn. 3, where C is the compliance (C = δ/F), δ the imposed displacement, F the reaction force, and w the width of the specimen.

G=

F 2 ∂C 2 w ∂a

(3)

This generic formulation was used in the present study. So it was necessary to obtain the evolution of the compliance with respect to the delamination length from the experiments. In the experiments that were performed, there are several sources for the scatter in the data, i.e. the measurement of the delamination length and the sensitivity of the load sensor. When these measurements are multiplied with each other, they may lead to inconsistent results. To obtain a good fit scatter reduction techniques had to be applied to the data. Calibration of the compliance. The first step consists of finding for both mode I and II tests the correlation between the compliance C and the delamination length a. It has been explained before that the waviness of fibres in the mode I test caused compression at small displacements. To avoid the influence of this closure phenomenon on the compliance measurement, only the upper range of the fatigue cycle has been considered. For mode II both upper and lower positions were considered for a better correlation. Even though a similar trend for all specimens was obtained, some variations from one specimen to another were observed. It was decided to perform the calibrations for each specimen individually with its own curve instead of a single generic curve that would be based on all tests together. The relation between C and a was used to calculate the derivate of C in the SERR equation (Eqn. 3), but also to filter the discrete measurements of the delamination length. As the force measurement is continuously recorded, a continuous variation of C could be computed. From the continuous variation of C, a continuous variation of the delamination length a has been deduced. This means that the value of a is then obtained indirectly. The main advantages of using an indirect measurement of a are: - The possibility to get a value for the delamination length at a moment when no picture has been taken, in case of high increments in delamination growth for example.

Delamination growth at interfaces in hybrid materials and structures

-

1313

To reduce the amount of scatter that might induce problems in calculating the delamination growth ∂a ∂N

Calculation of the strain energy release rate. The approximation made to calculate the SERR was to consider that the compression phenomenon observed in mode I have no influence on the delamination growth. It was thus decided to avoid it in the calculation of the strain energy release rate. Considering the fact that the displacement was constant, Eqn. 3 can also be written as:

⎛δ ⎞ ⎜ ⎟ ∂C C G=⎝ ⎠ 2 w ∂a 2

(4)

Because the compliance has been calculated such that it does not take the compressive effects into account, this formulation of G also excludes these effects. For the cases where the delamination occurred at two different interfaces at the same time, the SERR was divided by the number of interfaces to obtain the value for G per interface. Calculation of the delamination growth. The calculation of the delamination growth was based on the indirect measurement of a to avoid irregularities. The method chosen for calculating the derivate was the standard 2nd order polynomial best fit as described in [9]. The number of points considered for the fit was 5. Thus, the second order polynomial method requires 5 subsequent data points to determine the derivative at a specific location. The selection of the 5 data points in each set for calculating the derivative has been driven by the attempt to cover the largest range in da/dN

RESULT AND DISCUSSION Compliance calibration (mode I) Figure 6 shows the compliance measurement obtained from the performed tests.

G. Delgrange, R.C. Alderliesten and R. Benedictus

1314

0,07

0,06 specimen 1 specimen 2 0,05 specimen 3

2,6694

y = 2E-07x 2 R = 0,9836

C (m/N)

specimen 4 0,04

specimen 5

0,03

0,02

0,01

0 0

20

40

60

80

100

120

a (m)

Figure 6 : Compliance measurement for all Mode I specimens To check the validity of this measurement, it has been compared to a model for DCB specimens based on the beam theory [8]. This model uses the Young’s modulus E and the second moment of area I. This commonly used model leads to the relation given by Eqn. 5.

C=

δ F

=

2a 3 3EI

(5)

The average exponent for the performed tests is 2.65 which is close to the exponent 3 in Eq. 5. The difference can be explained with the fact that the model is not perfectly representing the applied specimen geometry and that the delamination did not occur at the symmetry plane. Compliance calibration (mode II) Also for the ENF specimen a beam theory based model exists [3]. It can be expressed as given by Eqn. 6.

C=

2 L3 + 3a 3 8 E1wh3

(6)

However, instead of using this model for the correlation with the measurements, a modified beam theory model has been adopted. This modified beam theory model

Delamination growth at interfaces in hybrid materials and structures

1315

is also frequently used in the literature [4]. A linear best fit has been made to correlate C to a3 to find a relation like Eqn. 7.

⎛a⎞ C = C0 + m ⎜ ⎟ ⎝L⎠

3

(7)

The result is shown in figure 7.

6,0E-04

5,0E-04

C (mm/N)

4,0E-04

3,0E-04 Specimen 1 Specimen 2 Specimen 3

2,0E-04

Specimen 4 Specimen 5 Specimen 6 Specimen 7

1,0E-04

static tests

0,0E+00 0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

(a/L)^3 (no unit)

Figure 7: the calibration of the compliance for the ENF specimens

Correlation between the direct and indirect measurement The extrapolation described earlier to get an indirect measurement of the delamination length gives a good correlation as illustrated by figure 8 for one example. This figure also shows the amount of scatter in the measurements and the way it has been reduced.

G. Delgrange, R.C. Alderliesten and R. Benedictus

1316

80

75

a (mm)

70

65 measurement

60 extrapolation

55

50 0

2000

4000

6000

8000

10000

12000

14000

16000

18000

N (cycles)

Figure 8: delamination measurement and interpolation for mode I - specimen 5 Paris diagrams The Paris-type graph in Figure 9 shows a systematic correlation for the trends between da/dN and SERR. Because of the applied specimen configuration, the SERR decreases with the propagating delamination. As can be concluded from Figure 9, the delamination growth rate decreased during the tests as well. 1,E-01

specimen 1 specimen 2 da/dN (mm/cycle)

1,E-02

specimen 3 specimen 4 specimen 5

1,E-03

1,E-04 1,0E+01

1,0E+02

1,0E+03

ΔG (N/m)

Figure 9: comparison of the Paris-type relation for each mode I specimen In the mode I tests performed, a lower threshold in fatigue life could not be observed. To identify whether or not a lower threshold on delamination growth is present, more dedicated tests will be needed.

Delamination growth at interfaces in hybrid materials and structures

1317

The ENF specimen for mode II results in very small ranges of the SERR. As a consequence, the Paris-type relation does not show a clear trend similar to mode I. Instead there are several series of data corresponding each to one particular test, as illustrated by Figure 10. Some tests were performed at a SERR level apparently too low to obtain any delamination growth. Based on these tests, some threshold values were added in Figure 10. A critical SERR level measured during the static tests has also been added. 1,0E-01

1,0E-02

da/dN (mm/cycles)

1,0E-03

Specimen 1 Specimen 2 Specimen 3

1,0E-04

Specimen 4 Specimen 5 Specimen 6 Specimen 7 G crit G no delamination

1,0E-05

1,0E-06

1,0E-07 100

1000 ΔG (N/m)

Figure 10: The Paris type relation for mode II tests Discussion of test setup To define the fatigue delamination resistance, the DCB specimen appeared to be valuable. The SERR range in each test is significant and clear trends can be obtained as presented in Figure 9. The ENF test, on the other hand, did not provide good results. A single test covers only a very limited range of the SERR. This is due to the fact that at a constant displacement the SERR level first increases and then decreases along with the propagating delamination. As a consequence, a large amount of tests is necessary to characterize the delamination behaviour over the full range of relevant SERR levels. In these conditions, a test other than the ENF test, allowing a constant SERR level during the test might be considered to reduce scatter. Moreover, a shear stress singularity may not have been captured due to applied measurement technique or the use of the modified beam theory. A finite element

1318

G. Delgrange, R.C. Alderliesten and R. Benedictus

model to calculate accurately the crack length dependence of C and G for this particular specimen configuration would be useful. In addition, some more tests should be performed to verify whether the delamination measured represents the complete delamination front in specimen width direction, related to the actual degradation of the material. Such tests have been performed recently for mode II with glass fibre/epoxy specimens. The transparency in such specimen can be used to compare the delamination length measured with the correction fluid to the actual delamination front. Fig. 11 presents the correlation that seems to be very good for mode II.

Figure 11: the correlation between measurement and actual degradation It should be noted that Fig 11 shows that for these specimens the delamination front is a straight line (no curvature in the profile). With these glass fibre/epoxy specimens it can also be determined whether for mode I the aluminium/glass fibre interface is weaker than the glass fibre/glass fibre interface. By comparing the obtained delamination resistance in these tests with the first tests on GLARE specimens, it can be determined whether the delamination resistance only dependents on the adhesive system at that interface or also on the aluminium layers.

Delamination growth at interfaces in hybrid materials and structures

1319

CONCLUSION AND FUTURE WORK The presented research is a first step towards understanding the fatigue delamination resistance at laminate interfaces. Both mode I and mode II were analysed using respectively the double cantilever beam and the end notched flexure specimens. Based on the presented tests, a relation between the delamination growth and the strain energy release rate was defined for mode I. Because of the peculiar correlation for mode II obtained with the ENF test, such relation could not be determined reliably. It is proposed to perform additional tests, including different geometry and test to determine the relation. Future work will also include similar tests on pure glass fibre/epoxy specimen to verify whether the previously defined relation for the glass fibre/glass fibre interface only depends on the delaminated interface itself. Such specimens, used in mode I, would also allow verifying that the delamination measured on the side is the actual degradation of the specimen. Some mixed-mode test will also be performed to define the contribution of each mode in a rule of mixture. Finally, it is the objective to develop a finite element model utilizing the defined Paris-type relations and rules of mixture to extend the current research to more complex structures and to establish design criteria.

ACKNOWLEDGEMENT The work presented in this paper was supported in part by Alcan Centre de Recherche de Voreppe. The authors are grateful to J.C. Ehrström for his support.

REFERENCES [1] Alderliesten R. (2007), Int. J. of fatigue, vol 29, n X, p. 628-646 [2] ASTM standard D6115-97 (reapproved 2004) [3] Trethewey B.R. Jr and Gillespie J.W. Jr (1988), J of Comp. materials, vol 22, p 459-483 [4] Carlson L.A., Gillespie J.W. and Trethewey B.R. (1986), J. of reinforced plastics and composites, vol 5 , n. X, p. 170-197 [5] Alderliesten R., Schijve J. and van der Zwaag S. (2006), Eng. Frac. Mech., vol 73, p 697-709 [6] Pirondi A., Nicoletto G. (2004), Eng. Frac. Mech, vol 71, p. 859-871 [7] Raghu Prasad B.K. and Pavan Kumar D.V.T.G. (2008), Thin-walled structures, vol 46, N X, p. 676-688 [8] Williams J.G. (1988), Int. J. of fracture, vol 36, n X, p 101-119 [9] ASTM standard E647-00

25th ICAF Symposium – Rotterdam, 27–29 May 2009

DEVELOPMENTS IN METAL BONDING Jarkko J. Aakkula1 , Kari Lumppio1, Olli Saarela1 and Tapani Haikola2 1

Helsinki University of Technology, Aeronautical Engineering 2 DIARC-Technology OY

Abstract: The difficulties in developing durable and robust surface preparation methods for metals have limited the use of bonded joints in highly loaded areas in aircraft applications. The grit blast silane (GBS) and AC-130 Sol-Gel methods can develop durable bond to aluminium and with primer to titanium. No suitable method has been available for structural steel bonding. In the DIARC surface preparation process the metal part is plasma treated in a vacuum chamber at a low temperature. Ions with enough kinetic energy form a thin nanostructed film on the surface. The film is dense, stable, hard, corrosion resistant, has very low coefficient of friction and is also bondable to epoxy without any primer. Testing has been performed with the GBS, Sol-Gel and DIARC methods for metal bonding applications. Bare and clad 7075-T76 aluminium, Ti 6Al-4V titanium, AISI 304 and AISI 4130N steels were bonded with the FM300-2 adhesive. Wedge test specimens were tested at hot/wet and hot fresh and salt water immersion. The GBS and Sol-Gel treatments provided acceptable results at hot/wet with bare and clad aluminium without primer and with titanium with the BR 6747-1 primer. Sol-Gel provided unacceptable results with steels at hot/wet and with 7075 aluminiums in hot water immersion. The DIARC plasma coating provided good results with all metals at hot/wet and in immersion. The results were as good as the best wedge test results achieved in aluminium bonding. M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1321–1341. © Springer Science+Business Media B.V. 2009

1322

J.J. Aakkula, K.Lumppio, O.Saarela and T.Haikola

INTRODUCTION Bonded joints in highly loaded metal to metal and metal to composite structural joints could offer several advantages in aircraft and other structural applications. Difficulties have been encountered, however, in developing durable and robust surface preparation techniques that could be used in industry or field level. The phosporic acid anodizong (PAA) and γ-GPS silane based techniques have been used for decades and they provide acceptable bond durability when used in aluminium bonding. The γ-GPS based techniques, grit blast silane (GBS) and AC130 Sol-Gel, can be also used with titanium, but a durable bond usually requires additional coating with primer. No satisfactory method has been available for steel in creating a strong and durable bond in highly loaded structural applications. The objective of this investigation was to compare the epoxy bonding durability of a new low temperature plasma coating method DIARC with the silane based methods grit blast silane (GBS) and AC-130 Sol-Gel. Typical aircraft grade aluminium and titanium along with stainless and high strength steels were selected for testing.

MATERIALS AND PROCESSES The first tested materials in this investigation were Ti 6Al-4V titanium and AISI 304 stainless steel. Later 7075-T76 bare and clad aluminiums and AISI 4130N high strength steel were added to the test matrix. In the clad aluminium the clad layer was not removed, i.e. these specimens surface treatment and bonding were performed on clad surface. Silane methods are based on γ-glycidoxypropyltrimethoxysilane, which creates a chemical bond with the fresh and still active aluminium oxide layer. In the other end of the hydrocarbon chain the silane has an epoxy group which is bondable to epoxy adhesives or primers. In the grit blast silane (GBS) method Dow Corning Z6040 silane is used as a 1% aqueous water solution. In the Sol-Gel AC-130 method small amounts of glacial acetic acid (GAA) and zirconium n-propaxine (TPOZ) are added to a 1% silane-water solution. These chemicals lower pH level of the solution from approximately 5.3 to 3.5. Additional chemicals make the use of the silane solution more robust and should improve the durability when compared to the grit-blast silane method. AC-130 is manufactured in 4-component kits and in 2-component kits. Older results presented in this paper (year 2007 and earlier) were achieved using the 4-component kit. In the most recent tests described in this paper 2-component kits AC-130-2 were used. Both kit types should provide the same performance [1].

Developments in metal bonding

1323

In the silane based processes the metal surface must be cleaned with a solvent and grit-blasted immediately before applying the silane or the AC-130 Sol-Gel solution. After the treatment the surface must be bonded or covered with primer within 12 hours. White pure 180/220 grit aluminium oxide was used in grit blasting. The blowing agent was clean air. The blasting pressures used for aluminium, titanium and steels were 3, 4.5 and 6 bars, respectively. Primer was not generally used in these tests. An exception was titanium AC-130 Sol-Gel specimens, which were tested without primer and with the BR 6747-1 primer at the hot/wet environment. In the DIARC process the metal part is plasma treated in a vacuum chamber at a low temperature. Ions with enough kinetic energy form a thin (from nanometers to microns) well adherent amorphous and dense nanostructured layer when they hit the surface. The process is environmentally safe and economical. The surface layer is dense, hard, corrosion resistant, has very low coefficient of friction, and is also bondable to epoxy. In addition to epoxy bonding, the DIARC process can be used for coating tools, sliding parts and molds for lower friction and wear resistance, and for replacing other environmentally hazardous corrosion protection coatings on metals (e.g. epoxy primers containing chromates or cadmium often used with high strength steels). For epoxy bonding the treated metal part does not need any additional primer. It also stays stable and is robust to the bonding process. The DIARC pre-treatment thus decreases the workload in the final bonding and also decreases the requirements for the surface treatment stage of the process. Since the DIARC was a new method for structural bonding, slightly different variations of pre- and post-tertaments were tested with the stainless steel and titanium specimens. First method was a simple DIARC coating, in which the bonding surface was only solvent cleaned before applying the DIARC bond coating in a vacuum chamber. In the second variation the surface to be treated and bonded was grit blasted a couple of days before the DIARC treatment. The grit blasting techniques described earlier were used. In the third variant the surface was grit blasted, DIARC coated and additionally treated with the AC-130 Sol-Gel solution. The idea in the last variant was to create a chemical bond between the silane and epoxy and, hopefully, to create a bond between the chemical and the DIARC coating. In this investigation only surfaces to be bonded were treated with the DIARC coating. An exception was a second test of high strength steel in the salt water immersion, where the individual specimens were coated with DIARC on all surfaces before they were bonded together one by one.

J.J. Aakkula, K.Lumppio, O.Saarela and T.Haikola

1324

All specimens were bonded with the Cytec FM300-2 epoxy adhesive having areal weight of 490 g/m2. The adhesive film was cured 2 hours at 120°C under 0.7 bar vacuum pressure.

TESTING METHOD ASTM D3762 wedge testing standard was used to measure the bonding durability. Good durability is indicated by a short initial crack length, gradual crack growth rate and cohesive failure mode. In this investigation the following acceptance values were used with the FM300-2 film: initial crack length less than 45 mm, crack growth in 48 hours less than 6.5 mm and failure mode more than 80% cohesive. ASTM D3762 advises to use 3.2 mm thick aluminium plates with 3.2 thick wedges. With other materials the wedge thicknesses should be modified according to the Young’s modulus. Crack opening energy GI for specimens can be calculated using Equation (1) [2], in which Y is the thickness of the wedge, h is the thickness of the plate, E is the Young’s modulus of the material and a is the crack length.

GI =

[

Y 2 Eh3 3(a + 0.6h ) + h 2

[

2

16 (a + 0.6h ) + ah 3

]

2 2

]

(1)

Wedge thicknessess for 2.0 mm thick titanium and steel plates were based on fracture toughness calculations. The fracture toughness GI for aluminium specimens was calculated using constant crack length of 40 mm, see Table I. The wedge thickness for steel and titanium specimens were selected to have a reasonably close match with GI value of the aluminium specimens. Calculations using the constant 40 mm crack length are presented in Table I. It can be seen that the crack opening energy with steel specimens is lower than with aluminium specimens. That might result in shorter initial crack length and slower crack growth. It will not, however, affect too much to the failure mode, which is also very important indicator of the durability.

Developments in metal bonding

1325

Table I. Crack opening energy for the test specimens having equal crack length.

Wedge testing was started using a typical accelerated hot/wet aging environment (60°C/98%RH). After hot/wet testing a more demanding exposure was considered. It was easily created by immersing the specimens in hot (60°C) tap water. Even more demanding exposure was created using salt water. The North Baltic Sea water from close to the Finnish coastline was used. It had 0.5% salt content. The pH in the fresh water tank was 9.9 and in the salt water tank 8.9. The Redoxpotentials in fresh water and salt water tanks were respectively -150 mV and -108 mV. Since the Redox-potentials of aluminiums and high strength steels differ from these values [3], oxidation will happen during the immersion in alkaline water.

RESULTS Tests at hot/wet environment The results of the tests at the hot/wet environment are shown in Table II and in Figures 1 – 8. The AC-130 Sol-Gel treatment provided acceptable results with clad aluminium without primer. With the bare aluminium without primer acceptable results were achieved with the GBS treatment. The crack growth rates were gradual and the failure modes were over 95% cohesive, see Table II and Figures 1, 2 and 7. During these tests the AC-130 Sol-Gel provided unacceptable results with the bare aluminium specimens without primer. The DIARC treatment on aluminium provided acceptable results. The crack growth rates were gradual and the failure modes were 80-90% cohesive, see Figure 2. Alclad specimens had average 90% cohesive failures. With bare aluminium specimens the average remained at 80% and there were more variations. Some specimens had 70% cohesive failures, while the best had 95% cohesive failures. With titanium the AC-130 Sol-Gel provided unacceptable results without primer. The BR 6747-1 primer improved the durability of the AC-130 Sol-Gel treatment to an acceptable level, see Table II and Figures 3, 4 and 7.

1326

J.J. Aakkula, K.Lumppio, O.Saarela and T.Haikola

The testing of the DIARC plasma coating with titanium provided good results. The crack growth measured in the wedge test was less than 5 mm in two days. The failure modes with titanium were 90-100% cohesive. However, using Sol-Gel solution after DIARC coating degraded the results, see Table II and Figures 3, 4 and 7. AC-130 Sol-Gel did not provide acceptable results with the AISI 304 stainless steel (SS), see Table II and Figures 5, 6 and 7. The crack growth rate was too high and failure modes were totally adhesive. The testing of the DIARC plasma coating with steels provided very good results. The crack growth measured in the wedge test was less than 5 mm in two days. The failure modes of the stainless steel (SS) specimens were 95% cohesive. Using SolGel solution after DIARC coating again decreased performance, see Table II and Figures 5, 6 and 7. With the DIARC treated AISI 4130N high strength steel specimens the crack growth rate was as low as with stainless steel specimens. Variation in the failure mode between the specimens was higher. Three specimens had 95-100% cohesive failure modes while two other specimens had only 50% cohesive failures. The best DIARC results within these series were, anyhow, as good as the best wedge test results achieved in aluminium bonding with other methods. Initial crack lengths and crack growth after 1000 hour exposure of specimens tested at hot/wet environment are shown from the shortest to the longest in Figure 7. All tested specimens had an initial crack length less than 45 mm. Steel and titanium specimens had short initial cracks, as expected according to calculations shown in Table I. Figure 8 shows the crack opening energy GI of the specimens tested at the hot/wet environment. The energy has been calculated using Equation (1) for the actual initial crack length and for the final crack length of the specimens. The order of the materials and processes is the same as in Figure 7. Table I indicated that crack opening energies for the steel and titanium specimens are lower than for aluminium specimens with the same crack length. Figure 8, however, shows that the actual crack opening energy in steel and titanium specimens was very high. This comparison shows, as well as Figure 7, that steels had very good durability in these tests.

Developments in metal bonding

Table II. Wedge test results, the hot/wet (60°C/95%RH) exposure.

1327

J.J. Aakkula, K.Lumppio, O.Saarela and T.Haikola

1328 48 hours Crack growth in [mm]

1000 hours

55 50

Al-SG-HW2

45

Al-GBS-HW

40

Al-DIARC-HW 35

Ac-SG-HW 30

Ac-DIARC-HW 48 hours crack growth acceptance limit

25 20 15 10 5 0

0

5

10

15

20

25

30

35

Exposure time SQRT(hours)

Figure 1. Crack growth of bare (Al) and clad (Ac) 7075-T76 aluminium specimens at hot/wet (60°C/95%RH), grit-blast silane (GBS), AC-130 Sol-Gel (SG) and DIARC surface treatments.

Figure 2. Typical failure modes of bare (Al) and clad (Ac) aluminium specimens with GBS, SG and DIARC treatments after 1000 hours exposure at hot/wet (60°C/95%RH) environment.

Developments in metal bonding

1329

48 hours

1000 hours

Crack 40 growth in [mm] 35 30

Ti-SG-HW

25

Ti-SGP-HW 20

48 hours crack growth acceptance limit

Ti-DIARC-HW Ti-G-DIARC-HW

15

Ti-G-DIARC-SG-HW 10 5 0 0

5

10

15 20 25 Exposure time SQRT(hours)

30

35

Figure 3. Crack growth of Ti 6Al-4V titanium specimens at hot/wet (60°C/95%RH), AC-130 Sol-Gel (SG), Sol-Gel with primer (SGP) and different variations of DIARC treatments.

Figure 4. Typical failure modes of titanium specimens prepared with SG, SGP and DIARC treatments after 1000 hours exposure at hot/wet (60°C/95%RH) environment.

J.J. Aakkula, K.Lumppio, O.Saarela and T.Haikola

1330

48 hours

1000 hours

Crack 40 growth in [mm] 35

SS-SG-HW SS-DIARC-HW

30

SS-G-DIARC-HW SS-G-DIARC-SG-HW

25

SH-DIARC-HW 48 hours crack growth acceptance limit

20

15

10

5

0 0

5

10

15

20

25

30

35

Exposure time SQRT(hours)

Figure 5. Crack growth of AISI 304 (SS) and AISI 4130N (SH) steel specimens at hot/wet (60°C/95%RH), AC-130 Sol-Gel (SG) and different variations of DIARC treatments.

Figure 6. Typical failure modes of AISI 304 (SS) and AISI 4130N (SH) steel specimens prepared with SG and DIARC treatments after 1000 hours exposure at hot/wet (60°C/95%RH) environment.

Developments in metal bonding Crack length in [mm]

1331

Crack length at hot/wet exposure

100 90

Crack growth after 1000 h

80

Initial crack length

70 60

Intial crack length limit

50 40 30 20 10

Al-SG-HW2

Ti-SG-HW

Ti-G-DIARC-SG-HW

Ti-G-DIARC-HW

SS-G-DIARC-SG-HW

Al-DIARC-HW

Ac-DIARC-HW

Ti-DIARC-HW

Ti-SGP-HW

Al-GBS-HW

S-SG-HW

SH-DIARC-HW

Ac-SG-HW

SS-DIARC-HW

SS-G-DIARC-HW

0

Figure 7. Initial crack length and crack growth in 1000 hours of the specimens tested at hot/wet (60°C/95%RH) environment. Crack opening energy at hot/wet exposure

GI in [J/m2]

GI of initial crack length

3500

GI of final crack length after 1000 h

3000 2500 2000 1500 1000 500

Al-SG-HW2

Ti-SG-HW

Ti-G-DIARC-SG-HW

Ti-G-DIARC-HW

SS-G-DIARC-SG-HW

Al-DIARC-HW

Ac-DIARC-HW

Ti-DIARC-HW

Ti-SGP-HW

Al-GBS-HW

S-SG-HW

SH-DIARC-HW

Ac-SG-HW

SS-G-DIARC-HW

SS-DIARC-HW

0

Figure 8. Crack opening energies of the specimens GI with the initial crack length and with the final crack length after 1000 hours at hot/wet (60°C/95%RH) environment.

1332

J.J. Aakkula, K.Lumppio, O.Saarela and T.Haikola

Immersion test results The test results in hot water immersion are shown in Table III and in Figures 9 – 14. In immersion tests the AC-130 Sol-Gel treatment was used only for aluminium, since it did not provide good results with stainless steel or titanium without primer at the hot/wet environment. The tests indicated that durability of the bond was unacceptable. Some of the tests were repeated in order to confirm that there was no process error during the first surface preparation or bonding. All aluminium AC130 Sol-Gel results in immersion remained, however, unacceptable, see Table III and Figures 9 and 10. Based on the hot/wet steel and titanium test results no grit blasting was required with the DIARC coating. Therefore it was not used with the immersion test specimens. In immersion the DIARC coated aluminium specimens provided acceptable crack growth rate in all cases, see Table III and Figures 9, 10 and 13. The failure modes of bare aluminium specimens were 90% cohesive. The clad aluminium specimens, however, did not pass the failure mode criterion having only 30% cohesive failure mode, see Figure 10. The DIARC coating with titanium provided excellent results in immersion. The crack lengths were short and the failure modes 100% cohesive, see Table III and Figures 11, 12 and 13. The DIARC coated stainless steel provided the slowest crack growth rate in immersion, see Table III and Figures 11, 12 and 13. According to the very slow crack growth, cohesive failure mode was expected. The opened surfaces had cohesive failures in the very short humid crack growth area. The percent value of the cohesive failure mode was difficult to measure, since the opening of the specimens resulted in new adhesive failures on the surfaces in very close vicinity of the area in interest, see Figure 12. Cohesive failure modes of 70% and 20% were estimated for fresh water specimens and salt water specimens, respectively, see Table III. The AISI 4130N high strength steel was highly corrodible resulting in corrosion failure between the base material and the DIARC coating when immersed in hot water, see Figures 11 and 12. The crack growth rate in salt water immersion was high. The failure mode was adhesive between the base material and the DIARC coating. In fresh water immersion the crack growth rate was slow and the failure mode in the humid crack growth area was mainly cohesive. Adhesive failure mode on the interface of the base material and the DIARC coating was also detected, see Figure 12.

Developments in metal bonding

1333

In the second trial the individual specimens were first grit blasted for removing all corrosion stains and then totally covered on all surfaces with the DIARC coating. The specimens were bonded together one by one. Thys, there were no untreated surfaces in the specimens. Also, the edges of the specimens were not machined before testing. However, the coating of the edges was slightly damaged during removal of the adhesive fillet from the bondlines, from which the crack growth was measured. These techniques provided a remarkable improvement in durability, see Figures 11 and 13. Light corrosion was still detected on edges due to the damaged coating. Failure mode was cohesive. Adhesive was present on both surfaces of the opened specimens. Figure 13 shows the initial crack lengths and the crack growth after 1000 hour exposure of specimens tested in immersion tanks. The initial crack lengths exceeded 45 mm only with the DIARC coated bare aluminium specimens (AlDIARC-FW), suggesting that some problems have been present during the bonding process. Most DIARC specimens showed good performance. An exception was AISI 4130N high strength steel in salt water (SH-DIARC-SW), where the specimens suffered from the base material and the DIARC coating corrosion problem. The repetition of the high strength steel specimens testing using the grit blasting and the complete DIARC coating (SH-G-TC-DIARC-SW) improved performance. The crack growth rates of the AC-130 Sol-Gel treated aluminium specimens were high, much higher than the aaceptable values. Figure 14 shows the crack opening energy GI of the specimens tested in immersion. The order of the specimens is same as in Figure 13. Figure 14, again, shows that with the measured crack lengths the crack opening energies in the steel and titanium specimens are very high even after 1000 hours exposure, which also indicates good bonding durability performance.

J.J. Aakkula, K.Lumppio, O.Saarela and T.Haikola

1334

Table III. Wedge test results from hot (60°C) fresh water (FW) and salt water (SW) immersion. 1000 hours

48 hours Crack growth in [mm]

60

Al-SG-FW2 Al-SG-SW Ac-SG-FW2

50

Ac-SG-SW2 Al-DIARC-FW 40

Al-DIARC-SW Ac-DIARC-FW

30

Ac-DIARC-SW 48 hours crack growth acceptance limit

20

10

0 0

5

10

15 20 25 Exposure time SQRT(hours)

30

35

Figure 9. Crack growth of bare (Al) and clad (Ac) 7075-T76 aluminium specimens in hot (60°C) fresh water (FW) and hot salt water (SW) immersion, AC-130 SolGel (SG) and DIARC treatments.

Developments in metal bonding

1335

Figure 10. Typical failure modes of bare and clad aluminium specimens with SG (above) and DIARC (below) treatments after 1000 hours exposure in hot (60°C) fresh water (FW) and hot salt water (SW) immersion.

J.J. Aakkula, K.Lumppio, O.Saarela and T.Haikola

1336

48 hours

1000 hours

40

Ti-DIARC-FW

Crack growth 35 in [mm]

Ti-DIARC-SW SS-DIARC-FW

30

SS-DIARC-SW 25

SH-DIARC-FW

48 hours crack growth acceptance limit

SH-DIARC-SW 20

SH-G-TC-DIARC-SW 15 10 5 0 0

5

10

15

20

Exposure time SQRT(hours)

25

30

35 hours

Figure 11. Crack Growth of DIARC surface treated Ti 6Al-4V titanium, AISI 304 and AISI 4130N steel specimens in hot (60°C) fresh water (FW) and hot salt water (SW) immersion.

Developments in metal bonding

1337

Figure 12. Typical failure modes of Ti 6Al-4V titanium and AISI 304 stainless steel (above) and AISI 4130N high strength steel (below) specimens with DIARC treatment after 1000 hours exposure in hot (60°C) fresh water (FW) and hot salt water (SW) immersion.

J.J. Aakkula, K.Lumppio, O.Saarela and T.Haikola

1338

Crack length in [mm]

Crack growth after 1000 h Initial crack length

Al-SG-SW

Ac-SG-SW2

Al-SG-FW2

SH-DIARC-SW

Ac-SG-FW2

Al-DIARC-FW

SH-DIARC-FW

Al-DIARC-SW

Ac-DIARC-SW

Ti-DIARC-FW

SH-G-TC-DIARC-SW

Ac-DIARC-FW

Ti-DIARC-SW

SS-DIARC-SW

Intial crack length limit

SS-DIARC-FW

100 90 80 70 60 50 40 30 20 10 0

Crack length in hot fresh water and hot salt water immersion

Figure 13. Initial crack length and crack growth in 1000 hours of the specimens tested in hot (60°C) fresh water (FW) and hot salt water (SW) immersion.

Crack opening energy in hot fresh water and hot salt water immersion

GI in [J/m2] 3000 2500

GI of initial crack length

2000

GI of final crack length after 1000 h

1500 1000 500

Al-SG-SW

Ac-SG-SW2

Al-SG-FW2

SH-DIARC-SW

Ac-SG-FW2

Al-DIARC-FW

SH-DIARC-FW

Al-DIARC-SW

Ac-DIARC-SW

Ti-DIARC-FW

SH-G-TC-DIARC-SW

Ac-DIARC-FW

Ti-DIARC-SW

SS-DIARC-SW

SS-DIARC-FW

0

Figure 14. Crack opening energy GI with initial crack length and final crack length after 1000 hours of the specimens tested in hot (60°C) fresh water (FW) and hot salt water (SW) immersion.

Developments in metal bonding

1339

DISCUSSION AND CONCLUSIONS Steel bonding durability The testing of the DIARC plasma coating with steels in epoxy bonding provided very good results at the hot/wet environment and in the hot fresh water and in the hot salt water immersion. The results were as good as the best wedge test results achieved in aluminium bonding with best possible surface treatment methods at the hot/wet environment. The DIARC bonding surface treatment increases the durability of bonded steel joints to a level of composite to composite joints and opens a new possibility to design structural steel joints. Good stainless steel bonding durability was the most interesting result in this investigation, since a durable bond has not been achieved with any other method. AC-130 Sol-Gel has been reported to give some acceptable results in steel bonding [4], but those results could not be repeated during this study. Here AC-130 Sol-Gel with the AISI 304 stainless steel provided a reasonable crack growth but an unacceptable failure mode at hot/wet conditions. Due to the results, there was no further reason to repeat the test in hot water immersion tank. The DIARC coating performed well with stainless steel. No grit blasting is necessary. Also, no additional AC-130 Sol-Gel should be used after the DIARC coating. In hot water immersion stainless steel bonding provided the best results. The steel plate and the wedge, as such, produced lower crack opening force than the thicker aluminium plates, but the actual crack was so short that the crack opening energy was in these specimens higher than in aluminium or titanium specimens. The failure modes at the hot/wet were cohesive. The failure modes after immersion could not be reliably measured from all stainless steel specimens since the surfaces were slightly damaged during the opening of the specimens. AISI 4130N is a more problematic material than stainless steel since it is very corrodible. First three sets of the AISI 4130N specimens were coated with the DIARC coating without any pre-treatment. There were evident corrosion stains on the surface before the coating. That resulted in a high scatter in the crack growth rate during immersion tests. In hot salt water immersion the high strength steel failure mode was unique, i.e. cohesive failure between the base material and the DIARC coating. However, the AISI 4130N results in the hot/wet environment were acceptable and results in the hot fresh water immersion were almost acceptable even with this unsophisticated approach. In the second trial with the high strength steel all surfaces were DIARC coated in order to prevent corrosion intrusion between the base material an the coating. These individually bonded specimens were tested in hot salt water immersion. The results were as good as with the other DIARC coated specimens. The failure mode was cohesive.

1340

J.J. Aakkula, K.Lumppio, O.Saarela and T.Haikola

Titanium bonding durability There are some acceptable surface treatment methods for titanium bonding. One method is to use AC-130 Sol-Gel with BR 6747-1 primer. The results of the hot/wet tests showed that omitting primer in titanium bonding with AC-130 SolGel makes the durability conclusively deficient. Another acceptable surface treatment method for titanium was found in these investigations to be the DIARC coating, which worked well at the hot/wet exposure and in hot fresh water and hot salt water immersion. Titanium was tested in immersion only after simple a DIARC treatment. The results indicated that titanium does not require any grit blasting before the DIARC coating. The failure modes were cohesive in titanium specimens tested at hot/wet as well as in hot water immersion. No additional AC-130 Sol-Gel should be used on on DIARC surface treated titanium either. It will result in a faster crack growth and adhesive failure mode. Aluminium bonding durability Durable aluminium bonding can be achieved with GBS, AC-130 Sol-Gel and DIARC treatments when exposed to hot/wet environment. Good results have been achieved for clad and bare aluminium without any clad removal and without primer. However, bare aluminium with AC-130 Sol-Gel did not provide in these tests as good results as expected. Earlier testing (2005-2007) with the GBS method gave acceptable results also for bare aluminium without primer. The bare aluminium AC-130 Sol-Gel testing was repeated, but no improvement was found. In neutral hot water and in salt hot water immersion, which is considered to be very demanding exposure condition for hygroscopic silane chemicals, AC-130 SolGel did not provide any acceptable results with 7075 aluminiums. Crack growth was fast and failure modes were always totally adhesive. The DIARC coating provided good results for 7075 aluminiums at hot/wet and in immersion. The scatter of the test results was, anyhow, higher than with stainless steel or titainium. This may come from varying thickness of the stable aluminium oxide layer, since in these tests the aluminium surfaces did not have any pretreatment. It has been recommended by DIARC-Technology Oy that aluminium surfaces should be grit blasted for removing the stable oxide layer no longer than 5 days before the DIARC coating. The grit blasted parts shall be kept in dry conditions or otherwise protected against exposure to heat or moisture. The opening of some of the specimens was done several months after the end of the exposure.The adhesive film had time to dry and regain its RT/dry strength, which is greater than the hot/wet value. It is suggested that it might be one reason why some opened surfaces in the figures look very fractured. The adhesion fractures in the specimen bondlines caused by the opening may have caused some

Developments in metal bonding

1341

extra damage also to close proximity of the humid crack growth area, which is the area of interest in these specimens. Future work The testing with the DIARC surface treatment will continue with the wedge tests 7075-T76 bare and clad aluminiums, Ti 6Al-4V titanium, AISI 304 staineless steel and AISI 4130N high strength steel specimens at salt fog chamber. Also, some aluminium and steel specimens will be retested in the hot salt water immersion to confirm reported results. This time grit blasting is used with the aluminium and high strength steel specimens before the the DIARC treatment. The AC-130 SolGel surface treatment will be tested with a marine aluminium in the hot salt water immersion.

ACKNOWLEDGEMENTS This research was sponsored by the Finnish Air Force. The authors acknowledge FINAF Air Materiel Command for the possibility to accomplish this work. The support of the Patria Aviation and DIARC Technology Oy is highly appreciated.

REFERENCES [1]

Blohowiak, K.Y., Grob, J., Grace, W.B., Cejka, N. and Berg, D., (2007), In: From Art to Science: Advancing Materials and Process Engineering, Improvements on Sol-Gel Surface preparation Methods for Metal Bonding Applications, Cincinnati, OH, October 29 – November 1, Society for the advancement of Materials and processes Engineering. CD-ROM. 15 pp.

[2]

Rider, A., (2002), The Durability of Epoxy Adhesive Bonds formed with Titanium Alloy. Report No: DSTO-TR-1333. Defence Science and Technology Organization, Platform Science Laboratory, Melbourne, Australia. 35 p.

[3]

Fourrier, T. and Horst, B. (1986) Galvanic corrosion between carbon fiber reinforced plastics and metallic aeronautical materials. Forschungsbericht, DFVLR Insitut für Werkstoff-Forschung, Köln DFVL-FB 86-16, 1986, 83 p.

[4]

Mazza, J.J, Gaskin, B.G., De Piero, W.S. and Blohowiak, K.Y. (2004), SolGel Technology for Low-Voc, Nonchromated Adhesive Bonding Applications. SERD Project PP-1113, Task 1. AFRL-ML-WP-TR-20044063. 163 p.

Environmental effects

25th ICAF Symposium – Rotterdam, 27–29 May 2009

ON SEPARATING THE EFFECT OF CORROSION ON INTER-LAMELLAR FATIGUE OF THIN SHEET AA7079-T6 Sandeep R. Shah and S.A. Fawaz Center for Aircraft Structural Life Extension, HQ USAFA/DFEM, 2354 Fairchild Dr., Ste. 2J2A, USAF Academy, CO – 80840, USA

Abstract: AA7079-T6 was widely used in the 1960s for manufacturing aircraft due to its high mechanical strength. Subsequently, this alloy was discontinued due to very poor fracture toughness, crack growth resistance, and corrosion resistance. A large strategic cargo carrier has aft upper fuselage (crown) skin made of this alloy. Various cracks have been identified in this area which has led to significant inspection of the aircraft. Visual inspection of the complete crown skin is done at least three times a year to detect cracks. A detailed failure analysis of several cracks from a few replaced crown skins was conducted to understand the root cause of these cracks. The failure analysis revealed that initial propagation was along the thickness of the skin which then diverged and propagated along mid thickness parallel to the surface of the skin. This crack propagation was accompanied by branching along the grain boundaries in the propagation path. Such crack morphologies are typical of stress corrosion cracking (SCC). However, study of the fracture surfaces revealed the presence of fatigue striations. Comprehensive analysis of the loading condition and operating environment does confirm that fatigue and corrosion occurs separately for these cracks. The fatigue crack propagation occurs while the aircraft is in flight and corrosion occurs while the aircraft is on ground. This corrosion weakens the material preferentially along the grain boundary leading to intergranular fatigue crack growth in flight. For rolled thin sheet product, this intergranular crack appears as an inter-lamellar crack as the grains are oriented in one line. M. Bos (ed.), ICAF 2009, Bridging the Gap between Theory and Operational Practice, 1345–1364. © US Government 2009. Created within the capacity of an US governmental employment and therefore public domain. Published by Springer Science+Business Media B.V. Dordrecht.

S.R. Shah and S.A. Fawaz

1346

This hypothesis was also verified with controlled corrosionfatigue experiments in the laboratory which replicated the failure surface obtained in service. Experiments were also conducted to determine the threshold for crack growth in fatigue as well as SCC. The fatigue crack growth threshold was observed at 2.2 MPa√m and the SCC threshold was found to be 8.5 MPa√m. The crown skin area where the cracks are located is loaded only during flight when the environment is benign for SCC to occur. On the ground where the environment is amenable to corrosion, there is very little load on the crown skin area where cracks are located. This results in very low stress intensity values and no crack propagation occurs by SCC. The cracks are therefore believed to have propagated during flight by fatigue loading, which is also supported by failure analysis results.

INTRODUCTION Legacy 7XXX series aluminum alloys are prone to environmental attack resulting in intergranular stress – corrosion cracking. Most legacy alloys from the 1950s and 1960s were used in the peak aged condition of T6 temper. This temper is highly susceptible to stress corrosion cracking (SCC)[1]. Many aircraft in service still have structural elements with legacy alloys in the peak aged temper. Most often the in-service failure of these legacy alloy components is due to SCC. Since there are no reliable models to predict the SCC, such failures in service lead to increased inspection and maintenance burdens at the depot. Apart from affecting the stress corrosion cracking behavior, the amenability of the material to environment also results in significantly reduced fatigue life, as crack growth rates in the corrosion-fatigue regime are much higher than that observed for pure fatigue. A large strategic United States Air Force (USAF) cargo carrier has upper aft fuselage (crown) skins made up of AA7079-T6. Numerous cracks have been found in this crown skin which has led to significant inspections on these aircraft. Whenever cracks are found, the crown skin is replaced with newer generation alloys which are resistant to SCC growth. A few cracked skins which were replaced during depot maintenance were sent to our lab for evaluation and analysis. Four different cracks were analyzed in detail. This paper describes the detailed metallurgical failure analysis of one of the cracks on one skin panel. Based on the observations of the failure analysis, simulated lab experiments were performed to replicate the observed failure in the laboratory. Following sections describe the failure analysis and laboratory experiments along with results and discussions.

On separating the effect of corrosion on the inter-lamellar fatigue of AA7079-T6

1347

FAILURE ANALYSIS The skin panel came from the right side of the upper aft fuselage. This skin was marked with possible locations of the cracks determined by some Non Destructive Inspection (NDI). All the fasteners, stringers and other frame components were already dismantled from the skin. Figure 1 shows the photo of the actual panel with NDI indications. A small piece was cut from the skin, keeping the cracks at the center of the piece, for ease of handling during metallurgical examinations. The sealant and paint from the panels were remove using acetone and rubber scrapper.

40

39

40

-

UP FWD

Crown Skin 37

Right Side

1530 mm

35 34

33

32

36

31

30

29

1905 mm

Figure 1: As received panel showing the locations of the cracks on the forward and aft edge of the panel. Crack 29 located in thebottom right corner of the pnael is also shown in an enlarged view.

S.R. Shah and S.A. Fawaz

1348

Macroscopic Analysis A photo of the as received skin is shown in Fig. 1. For ease of readability, the markings on the sheet are annotated on the photo. Most of the cracks can be seen located on the forward and the aft edge of the skin. The forward and aft directions refer to the aircraft orientation. Crack 29 was selected for the present analysis and its enlarged view is shown in Fig. 1. Crack 29 was located on the forward edge of the skin and it extended from the skin edge to the countersink hole nearby. A small specimen was cut around this crack such that the crack was at least 1” away from the cut edges. This specimen was then observed under stereomicroscope to locate the possible sites of nucleation and growth of this crack and to measure the dimensions of the crack. Figure 2 shows the crack in the LT plane and the view of the crack from the edge in the LS plane as seen from stereo microscope. The inset shows the propagation of the crack in the countersink hole. FWD Edge LS-Plane LS-Plane

Crack

Polished Area

Nucleation LS-Plane Nucleation Site Site Nucleation

(c)

Site

FWD Edge View

LT-Plane Residual Paint

(a)

(b) 2 mm

Crack in Countersink Hole

Figure 2: Stereo-Micrographs of the crack 29. (a) Planar view in LT Plane showing the extension of crack from the edge to the countersink hole. (b) Shows the propagation of crack in the countersink hole. (c) Side view of the crack from the forward edge, showing possible crack nucleation site. Observation with the naked eye shows the presence of some residual paint on the surface of the skin, as well as the crack extending between the skin edge and the countersink hole. Next to the crack is a polished area, probably polished with a coarse sand paper to ascertain the position of the crack below the paint after initial NDI. The stereo micrographs in the Fig. 2 clearly reveal these in detail. Furthermore, looking from the edge of the specimen, a probable nucleation site is

On separating the effect of corrosion on the inter-lamellar fatigue of AA7079-T6

1349

seen at the edge as indicated in Fig. 2. There was no obvious nucleation site in the countersink hole. The edge view in Fig. 2(c) also shows a lot of scrapping marks with a shiny surface indicating the severity of the sealant stripping from the edge of the sheet. The crack in LT plane is almost 12 mm long and in LS plane it extends about 0.8 mm before deviating on the planes parallel to the LT surface. The primary crack extends about 1.5 mm in the up direction of the skin and a secondary branch extends in the down direction of the skin for about 5.7 mm. The schematic of the crack is shown in Fig. 3.

C

B A

B C

A

Figure 3: Schematic of the crack and various sections along which the specimen was cut to reveal the microscopic features of the crack. Figure 3 also shows the sections along which the specimen was cut for microscopy and fractography. Section AA was made about 1.5 mm from the edge. The specimen taken along this axis was used for fractography. Another section was made at BB again 1.5 mm from previous section (AA). This was used for microstructural observation. A third section was made at CC, tangential to the countersink edge to study the crack extension next to the countersink. The crack was located perpendicular to the forward edge of the right upper aft crown skin. The origin of the crack is below the surface on the edge of the skin as seen in Fig. 2. The crack has propagated in LS plane initially, before deflecting onto the LT plane parallel to the surface. The mode of crack propagation suggests the load acting on the skin is in the up/down direction of the aircraft. Since this crack has originated and propagated in the region between the edge of the panel and the first row of fastener holes, it does not carry any longitudinal loads such as bending due to tail load. The only load acting on the skin at this location is hoop stress due to pressurization during flight. Although fit up stress may be present in longitudinal as well as up/directions, they cannot be quantified and thus are not considered. The deflection of the crack in the planes parallel to the surface is quite

S.R. Shah and S.A. Fawaz

1350

interesting. Considering only the hoop stress acting on the skin at the edge, one would expect the crack to grow through the LS plane. The presence of crack on the skin starting from surface to inside prevents the outer layers of skin from carrying any hoop stress. Since the hoop stress is primarily carried by the subsurface area of the skin, it induces shear stress in skin along the layers parallel to the surface of the skin. The crack propagating along LS plane is therefore deviated parallel to the surface due to the presence of the shear stress. The orientation of the grains parallel to the surface during rolling of the skin also facilitates the growth of the crack in the planes parallel to the surface due to intergranular corrosion commonly seen in AA7079-T6. Presence of corrosion along the crack surface is evident from the Energy Dispersive X-ray Spectroscopy (EDXS) and microstructural observations presented in the following sections. Energy Dispersive X-Ray Spectroscopy Analysis (EDXS) Energy Dispersive X-ray Spectroscopy was carried out on the opened crack surface from the specimen cut along section AA in Fig. 3. After the sample was cut along section AA, small slits were made using a diamond wafering blade just below the area where the cracks terminated as shown in Fig. 4. The cracks were opened by bending the sample using locking pliers as shown in Fig. 4. This opened up two cracks, one, the primary crack and the other the secondary crack. The opened crack surfaces were analyzed in JEOL-JSM 6480 LV SEM with EDXS analyzer.

Secondary Crack

Primary Crack

Slits

Figure 4: Schematic drawing illustrating the crack opening for the primary and secondary crack from the specimen cut along section AA in Fig. 4.

On separating the effect of corrosion on the inter-lamellar fatigue of AA7079-T6

1351

The specimen, after opening the crack, was cleaned in acetone in an ultrasonic bath to remove all loose debris as well as to remove the cutting oil, which might have contaminated the parts while making the slits. The opened crack surface of the primary crack was inspected in the SEM. The fracture surface exhibited extensive electron charging in the SEM suggesting presence of corrosion products. The initiation site, characterized by extremely corroded surface and unevenness was selected for EDXS analysis. Figure 5 shows the EDXS spectrum obtained from the corroded region on the fracture surface. The EDXS spectrum shows presence of various different elements. EDXS revealed the presence of elements such as calcium, potassium, sulfur, phosphorous and oxygen apart from aluminum, magnesium, zinc and copper commonly found in AA7079 alloy. The corrosion product on the fracture surface should be comprised of the compounds of these elements with the elements of the constituent alloy phase. Sodium, calcium, potassium, chlorine, bromine and silicon are commonly found in the corrosion products.

Figure 5: EDXS spectrum of the surface deposits on opened primary crack surface described in Fig. 4. Table 1 gives the quantitative analysis of the elements seen in EDXS spectrum in Fig. 5. It can be seen that the amount of sulfur and phosphorous are quite significant compared to other elements commonly found in corrosion products. The sample surface has been thoroughly cleaned in acetone in the ultrasonic bath so these elements cannot be from external dirt or debris, but they are part of the corrosion scale. Sulfates are commonly found in the environment, but phosphates are not so common. However, both these elements can be found in the exhaust of the aircraft engines. The amounts of other common corrosion products such as Na, Cl, Ca and K are very low and probable reason for this could be that they have been washed away while cleaning the sample. Presence of sulfates and phosphates indicates the corrosion of the crack surface was more likely when the aircraft was

1352

S.R. Shah and S.A. Fawaz

on the ground in neighborhood of the exhaust of other aircraft, rather than in flight, since the concentration and activity of these elements at the cruising altitude, temperature and pressure will be very minimal. Table I: Quantitative elemental analysis of the corrosion product on the fracture surface as shown in Fig. 6. Element Weight (%) Atomic (%) O 49.5 63.4 Al 39.3 29.8 Mg 1.1 0.9 Si 3.1 2.2 P 2.5 1.6 S 1.5 0.9 K 0.2 0.1 Ca 0.3 0.1 Cr 0.5 0.2 Fe 0.6 0.3 Zn 1.5 0.5 Total 100 100 Microstructural Analysis Microstructural analysis was performed on both the fractured surface as well as polished cross section revealing the path of the crack and its surrounding. Both optical and scanning electron microscopies were performed to study the microstructural features in detail. Optical Microscopy

Figure 6: Composite micrograph of the specimen cut along section BB Figure 3 and polished. The micrograph shows the primary and secondary cracks. The sections from the specimen cut at BB and CC in Fig. 3 were carefully polished to remove all the cutting marks and were observed under the optical microscope for the propagation of the crack. Figure 6 shows the composite optical micrograph of the crack for the specimen cut along section BB. The primary crack is characterized by extreme corrosion. This crack initially propagates in LS plane but

On separating the effect of corrosion on the inter-lamellar fatigue of AA7079-T6

1353

then deviates to the right (up direction of the aircraft) along a plane parallel to the surface. The length of the crack in this plane is about 1.6 mm. Whereas the secondary crack propagating on the left side is much longer and is about 5.7 mm long. To understand the effect of the microstructure on the propagation of the crack, the polished sample was etched with Keller’s Reagent (1% HF, 1.5% HCl and 2.5% HNO3 by vol. in DI Water) for 20 seconds. The etching reveals the grain structure of the skin. The micrographs of the etched specimen near crack are shown in Fig. 7. The micrographs reveal that the crack propagation is primarily along the grain boundaries, since this alloy is prone to intergranular corrosion

Figure 7: Optical microgrpahs of the polished and etched cross section at section BB (Fig. 4). (a) shows the propagation of the secondary crack from the primary crack along the grain boundaries, (b) shows the region between the primary and the secondary crack, where a number of finer cracks are running along the grain boundaries. (c) shows the secondary crack far away from the initiation site and (d) shows the primary crack deviating along the grain boundaries. Figure 7(a) shows the primary crack and also the initiation of the secondary crack from the primary initiation site. The top aluminum clad layer is visible in this micrograph due to differential etching between pure aluminum and AA7079.

1354

S.R. Shah and S.A. Fawaz

Extensive corrosion of the material along the grain boundaries can be seen in Figs. 7(b) and 7(d). The growth of the secondary crack is in the LT plane and between the grain boundaries is shown in Fig. 7(c). This is an interesting growth mechanism since the only stress acting on the skin in this region is hoop stress and the growth of the crack is parallel to the direction of the hoop stress profile. This mode of crack opening is possible only if there is shear between the layers of the skin due to applied hoop stress. A couple of scenarios may provide shear between the layers. 1) Initiation of crack near the surface of the skin changes the loading at the surface and inside of the skin by hoop stress leading to the shear between the layers and 2) the riveted joint between the skin and doublers may induce some residual surface compression stress on the inside of the skin which changes the stress profile from the surface to the inner edge of the skin. Detailed SEM investigation presented in the following section shows striations along the primary and secondary crack surfaces, suggesting some kind of cyclic loading along this crack path. The hoop stress experienced by the skin is cyclic in nature and is the most likely source of the crack growth.

Figure 8: Composite micrographs of the crack 29 at section CC near the countersink (Fig. 3). (a) Micrographs of the polished specimen and (b) micrographs of polished and etched specimen showing the deviations of the crack along the grain boundaries. Figure 8 shows the optical micrographs of the crack at section CC in Fig. 3, tangential to the countersink hole. The origin of this crack is on the forward edge of the panel and the crack extends till the countersink hole. However, as seen in Fig. 8, only the primary crack propagates to the countersink hole from the edge. Even the growth of the primary crack in this micrograph is predominantly in the LS plane with deviations along the grain boundaries.

On separating the effect of corrosion on the inter-lamellar fatigue of AA7079-T6

1355

SEM Observation

50 μm

10 μm

Primary Crack Surface

50 μm

10 μm

Secondary Crack Surface

Figure 9: The fracture surface of as opened crack showing the corrosion products on both primary and secondary crack surfaces. The bright regions in the micrographs are due to the charging of electrons on non conducting corrosion products. The SEM observation was done primarily on the fracture surface after opening the cracks for the specimen cut along section AA in Fig. 3. The samples were observed in JEOL-JSM 6480 LV Scanning Electron Microscope. Figure 9 shows the micrographs of the opened fracture surfaces. The surfaces appear smeared with a layer of corrosion products. The corrosion layer is also characterized with cracking due to the brittle nature of the products. The EDXS analysis of the surface presented earlier shows the presence of many environmental elements responsible for the corrosion of the skin. At higher magnifications, the micrographs also reveal some delaminated layers (leaves). To understand the micro-mechanisms of the failure experienced by this skin, it was necessary to wash off these corrosion products from the surface to reveal the microstructure of the base material. Figure 10 shows the SEM micrograph of the fractured surface showing both the primary and secondary crack surfaces. The specimen was washed in 15 gm Alconox™ in 300 ml water solution, preheated to 95ºC, for 20 minutes in an ultrasonic bath. The washing was done to remove the surface corrosion product which tends to charge the electrons on the surface as seen on the pristine surfaces of the primary and secondary crack sides in Fig. 10.

S.R. Shah and S.A. Fawaz

1356

400 μm Crack Initiation

Primary Crack Side

Secondary Crack Side

50 μm

Figure 10: SEM micrograph showing the fracture surface after cleaning with Alconox™ solution near the forward edge of the skin. The ridge separating the primary and the secondary crack is also shown.

On separating the effect of corrosion on the inter-lamellar fatigue of AA7079-T6

1357

Studies in our laboratory have confirmed that washing with AlconoxTM does not introduce any artificial marks on the fracture surface. Specifically, specimens failed in fatigue and SCC in our lab were washed in same way and no artificial marks were seen on the fracture surface. Figure 10 reveals the corrosion at the initiation site and a number of cracks branching out, including a couple propagating to the secondary crack side. Corroded regions appear dark in the micrograph making it easy to locate the nucleation site for the cracks. This crack seems to have nucleated by pitting corrosion followed by extension due to fatigue and corrosion, which is evident from the corrosion scale on the fracture surfaces shown in Fig. 9 and from the intergranular cracks and striations seen at higher magnifications in Fig. 11.

Residual Corrosion Product

10 μm 0.4 mm from the initiation site

10 μm 1.5 mm from the initiation site

c c c

c

c 10 μm

1 mm from the initiation site

10 μm 3.5 mm from the initiation site

Secondary Crack Surface Primary Crack Surface Figure 11: Higher magnification micrographs of the cleaned fracture surface on the primary and secondary crack side. Intergranular corrosion can be easily seen in all the micrographs. The places near the initiation site are characterized by higher corrosion product deposits, whereas far away locations exhibit fatigue striations. The primary crack at 1mm from the initiation site is also characterized by quasi cleavage planes denoted by c indicating brittle failure. Average striation spacing is bigger on primary crack (1.5 μm as compared to 0.5 μm for secondary crack).

1358

S.R. Shah and S.A. Fawaz

Intergranular corrosion is characterized by the cracks along the grain boundaries which is clearly visible in Fig. 11. Primary crack surface shows more corrosion than the secondary crack surface indicating that it opened up much earlier than the secondary crack and the opened surface has seen the corrosion environment for a longer time. Both the primary and the secondary crack surface exhibit striation like features. Lynch et al. have shown that SCC crack growth can also exhibit such marks. However, as described earlier, this edge of the skin panel which is beyond the last row of fastener, does not see any static loading while parked on ground and most aircraft structures do not show SCC growth in flight, unless there are occluded regions filled with electrolyte next to the structural element. Residual and fit up stresses may result in SCC, however, in these cases, when the crack grows, the stress gets relieved and the spacing between these striation like marks should reduce with crack growth, whereas in present study we find that the striation spacing increase with the crack growth. We therefore believe that these striations indicate that the skin edge has undergone fatigue crack growth. Again the only possible load applied to the edge of the skin beyond the rivet holes is the hoop stress. The hoop stress in fuselage is due to pressurization of the fuselage during flight. This loading is cyclic in nature and is most likely responsible for the striations on the crack surface. The average striation spacing for the primary crack is 1.5 micrometer at about 1 mm from the initiation site as compared to 0.5 micrometer at about 3.5 mm from the initiation site, observed in secondary crack. This indicates that crack opening per cycle of loading and correspondingly ΔK (stress intensity factor per cycle) is higher for the primary crack. The orientation of primary crack is predominantly in the LS plane, whereas that of the secondary crack is in LT plane. LS plane is oriented perpendicular to the acting hoop stress, whereas the LT plane is parallel to the direction of hoop stress. The primary crack opens in mode I whereas the secondary crack is most likely opening in mode II. The mode II opening is due to the shear between the layers of the skin due to the presence of the crack. This shear stress should be much smaller than the tensile stress responsible for opening the primary crack and therefore the ΔK is smaller for the secondary crack (even at larger lengths). The primary crack surface also shows some cleaved planes. Cleavage is representative of brittle failure due to corrosion. This indicates that there is corrosion interaction with fatigue crack propagation.

LABORATORY EXPERIMENT Laboratory experiments were conducted to verify if in fact the crack can propagate in inter-lamellar regions as seen for the secondary crack during the failure analysis. Our hypothesis was that the crack initially propagates perpendicular to the applied hoop stress in the fuselage structure. Then as the propagating crack is exposed to the environment the rolled grain boundaries of the thin sheet structure gets weakened and the crack propagates along this boundary. The driving force for this crack propagation is mode II, which occurs due to the hoop stress when the inner layers carry the load and the outer layers do not carry any load due to the presence

On separating the effect of corrosion on the inter-lamellar fatigue of AA7079-T6

1359

of the crack. In laboratory, such a scenario was created by loading the specimen in flexure. A four point bend fixture was used. The corrosion environment was introduced by building a tiny reservoir on the top of the specimen using silicone as shown in Fig. 12. The reservoir was filled with saturated salt solution and water was replenished over time.

Figure 12: Schematic showing the laboratroy tests run on the specimen from the aircraft skin to reproduce inter-lamellar fatigue in laboratory.

For these experiments, the specimen was cut from the pristine part of the original skin panel. The specimen was 203 mm (8”) long and 51 mm (2”) wide with 1.58 mm (0.062”) thickness. A 0.127 mm (0.005”) deep notch was machined on the center of the specimen using a wire EDM machine. Liquid silicone rubber was applied around at 12.7 mm (0.5”) around the notch as shown in the Fig. 12. The specimen was loaded in a four point bend fixture with 150 mm (6”) outer span and 50 mm (2”) inside span. This ensured that whole reservoir area was under constant tensile stress. Only about 110 N of load was required to generate a mean stress intensity factor Kmean of about 4.4 MPa√m. Superimposed ΔK was 2.2 MPa√m. For this purpose a 90 KN Satec servo-hydraulic frame fitted with a 250 N load cell was used. The load could be controlled to within +/- 5 N. To introduce the effect of corrosion during the flexure fatigue, it was necessary to perform the experiments at low frequency. The lowest frequency possible in our test frame, 0.01 Hz, was used. Trial experiments at 0.1 Hz frequency did not show any inter-lamellar corrosion and the specimen failed in about 10,000 cycles with the crack propagating from the notch along the thickness. In flight, the hoop stress cycling can occur over 3 hours, which is about 0.0001 Hz. We could not run tests that slow due to machine limitations and time required to complete the test. The laboratory specimen lasted for 38,842 cycles (about 45 days). At his point the specimen had split in two pieces.

1360

S.R. Shah and S.A. Fawaz

Figure 13: SEM micrograph of the specimen failed in laboratory test, showing inter-lamelar crack branching into primary and secondary crack as was observed in the crack found during service. Birght spots indicates the corrosion product deposits.

The failed specimen was inspected in optical and scanning electron microscopes. Figure 13 shows the SEM micrograph of the specimen failed in the laboratory test. It can be seen that the crack starts below the notch and propagates along the thickness direction and then parallel to the surface along the mid thickness. The primary crack shows one more branching parallel to the surface before final fracture. These deviations of the crack path parallel to the surface results in interlamellar failure and this failure occurs only during very slow cycle flexural fatigue. We have run long term corrosion fatigue experiments in pure tension, but none of them show the inter-lamellar fracture like this. We therefore believe that such failure can occur only due to very slow cycle flexural fatigue loading which exists on the fuselage crown skin due to cyclic hoop stress as a result of pressurization. Very slow cycling is necessary for degradation of the material in the corrosive environment. Specifically, there should be enough corrosion along the grain boundaries to weaken them so that the crack can propagate at very low mode II loading along the grain boundaries.

On separating the effect of corrosion on the inter-lamellar fatigue of AA7079-T6

1361

Figure 14: High magnification SEM micrograph of the primary crack step surface of the specimen shown in Fig. 13.

Figure 14 shows the high magnification SEM micrograph of the primary crack, step surface shown in Fig. 14. This micrograph reveals the striation like features similar to those seen on the in-service crack specimen, confirming our hypothesis of the crack growth in slow cycle fatigue. If we compare the micrograph shown in figure 14, to striation like crack arrest marks observed by Lynch et al. [2] in SCC crack growth, there are two clear differences. First of all in their study Lynch et al. found that the crack arrest marks are about 100-250 nm apart consistently. Whereas in our observations we see the striation spacing varying from 200-1500 nm as a function of distance from crack tip and secondly they observed crack arrest marks only in case of high humidity air and they were not able to generate these marks in saturated salt solution. There are other instances where some crack arrest marks were seen for 3.5 mol% salt solution [3], but the features are not as consistent as in the present case. Again none of these studies show the fracture along inter-lamellar regions as in present case. The loading geometry may also be responsible for these differences.

S.R. Shah and S.A. Fawaz

1362

DISCUSSION SCC and fatigue are competing processes for crack growth. In a given condition, the rate controlling process depends upon various factors. In the present case where the cracks occur in skin panels along the forward and aft edge, there is no mechanical load on the edges of the panel, when the aircraft is on ground. This is primarily because all load is transferred to the frame at the last row of fasteners and the edges carry no structural load. At this location, however, hoop stress can occur. The cyclic stress occurs only during ground-flight-ground operation. Residual or fit up stresses may exist on these edges throughout. For SCC to occur three elements are necessary, susceptible material, environment and sustained mechanical stress. During flight, low temperature, pressure and almost negligible humidity, prevents corrosion. SCC can therefore occur only when the aircraft is on the ground. If significant residual or fit up stresses are present, SCC can occur at this location. In fact, the initiation of the crack is due to pitting corrosion at the edge of the panel where the clad layer meets the AA7079-T6 alloy, as shown in Fig. 2. Once the defect sites are generated, the crack growth is controlled by the stress intensity factor. Both SCC and fatigue crack growth exhibit threshold stress intensity factors below which no crack growth occurs. Figure 15 shows the threshold stress intensity values for fatigue and SCC for AA7079-T6 alloy measured in our lab. These specimens were also obtained from the panels removed from in-service aircraft. The details of the experiments can be found elsewhere[4]. Figure 15 also shows the data provided by the OEM of the aircraft. It can be seen that there is some discrepancy between OEM and our data, but no conditions were specified for OEM data so the comparison is difficult.

Figure 15: SCC and fatigue growth rates for AA7079-T6 alloys specimens obtained from an in-service aircraft skin panel under various experimental conditions. Threhold stress intensity values can easily be identified from these plots.

On separating the effect of corrosion on the inter-lamellar fatigue of AA7079-T6

1363

From Fig. 15 the threshold stress intensity value for SCC is about 7.7 MPa√m and that for fatigue is about 2.2 MPa√m. This suggests that fatigue can occur at much lower stress intensites as compared to SCC. However, if the stress intensity increases beyond 10-12 MPa√m, SCC may dominate since the growth rate is much faster for SCC provided fatigue cycle frequency is less than 1000 Hz. At typical cruising altitude, the hoop stress in the fuselage is about 120 MPa (18 ksi). At this load level for a single edge notch specimen would need at least a 10 mm long crack for SCC to dominate. The total crack length from the edge to the countersink hole as shown in Fig. 2 is about 12 mm and near the countersink hole, the dominant crack growth mechanism was in fact found to be SCC. We therefore believe that the crack initiated by pitting corrosion grew by fatigue to a length of about 10 mm and then grew by SCC to the fastener hole. The branching of the crack at mid thickness is due to weakening of the material by corrosion and interlamellar crack growth is again controlled by fatigue loading. Our arguments to support these claims are: 1) Extreme pitting and corrosion was found near the initiation sites at the intersection of aluminum clad layer and AA7079-T6 alloy substrate. 2) Surface corrosion products revealed the presence of various common corrosion elements along with sulfur and phosphorous. Sulfur and phosphorous are common to exhaust gages and commonly found when the aircraft is on ground. 3) The cracks are located on the forward and the aft edges of the skin panel, beyond the last fastener lines. Any longitudinal aircraft load, will not be seen by these edges. Furthermore, the cracks run in the forward/aft direction, suggesting the load is perpendicular to the longitudinal axis. 4) The crack orientation suggests the applied load runs in the up/down direction of the aircraft. The only load acting in this direction will be the pressurization hoop stress of the fuselage. 5) This stress occurs only when the aircraft is in flight. During flight temperature and pressure are low and humidity is almost zero suggesting absence of any corrosive reaction. Presence of any electrolyte in the occluded region may lead to some corrosion; however, at cruising altitudes the temperature can be as low as -60ºC, resulting in extremely low corrosion activity. 6) SCC can therefore not occur during flight and the only stress responsible for SCC on the ground is residual or fit up stress. 7) SCC growth by residual or fit up stress can not result in inter-lamellar crack growth. This can occur only due to flexural stress which arises due to hoop stress. 8) The striation like marking on inter-lamellar crack vary in spacing as it grows away from the initiation site. In our failure analysis the striation spacing changes from 200 nm to 1500 nm over the distance of 5 mm. Crack arrest marks seen during SCC are about 100-250 nm in spacing consistently [2].

S.R. Shah and S.A. Fawaz

1364

9) Residual stresses get released as the crack grows, leading to slower crack growth and may be crack arrest. However, no such signs were observed on the fracture surface. 10) The inter-lamellar crack is therefore due to fatigue, as a result of cyclic hoop stress only. 11) The crack growth in longitudinal direction changes from fatigue to SCC after about 10 mm of crack growth when SCC growth rate is much faster than fatigue crack growth rate.

CONCLUSION In-service cracks were found in upper aft fuselage skin of a large strategic cargo aircraft. These cracks were located on the forward and aft edges of the skin panel. All cracks exhibited similar morphology. One of the cracks was studied in detail for the present study. Detailed metallurgical failure analysis revealed severe pitting at the initiation site at the forward edge of the skin panel. Significant corrosion deposits were also found on the crack surface. The crack showed branching along the mid thickness plane with cracks growing preferentially along the grain boundary. These observations suggested the crack to be a stress corrosion crack; however, more detailed microscopy revealed the presence of striation like marks on the crack surface. Furthermore, an analysis of mechanical loading on the skin panel suggested that the crack should have grown as a result of hoop stress acting along the skin. This hoop stress occurs only during flight, when environmental conditions are non-conducive to corrosion activity. Detailed failure analysis suggested that the crack growth has to be due to fatigue. To confirm this hypothesis, laboratory experiments were conducted under very slow cycle flexural stress in saturated slat solution environment. This test replicated the failure surface observed in-service. Static load SCC tests could not produce the failure surface observed in service, confirming our hypothesis of crack growth by fatigue. However, when the crack grew beyond 10 mm, the crack growth was due to SCC since the applied stress intensity at that point resulted in faster crack growth kinetics by SCC as compared to fatigue. The present study suggests that careful mechanical stress analysis and detailed failure surface analysis are necessary to differentiate between failure due to SCC and fatigue.

REFERENCES [1] [2] [3] [4]

Speidel M. O., (1975), Metall. Trans. A,., vol. 6A, p. 631. Lynch S. P., Knight S. P., Birbilis N. and Muddle B. C.,. (2008). Presented at the International Hydrogen Conference, Wyoming, September 7-10. Martin P., Dickson J. L., Bailon J.-P., (1985) MaterSci and Eng.,, vol.69, n. 1, p. L9. Walters M. R., Shah S. R. and Fawaz S. A., (2006), USAFA-TR-2006-05 United States Air Force Academy Technical Report.

AUTHORS INDEX Aakkula, Jarkko J................... 1321 Adey, R.A.A. ........................... 789 Akatsuka, Takahiko ............... 1165 Alderliesten, R.C.73, 493, 539, 619 839, 969, 1069, 1245, 1261, 1305 Andersson, B............................ 707 Arunachalam, Saravanan R.... 1035 Asakawa, M. ............................ 899 Attia, Mohamed A.A............... 529 Augustin, Petr ........................ 1005

Delgrange, D.......................... 1305 Dixon, B. ................................. 123 Dowson, Anthony L. ............. 1275 Dragan, Krzysztof.................... 263 Ducher, D. .............................. 475 Duffield, M.J. ....................... 1135

Backman, D.S. ......................... 301 Bakuckas Jr, John G. ....... 109, 407 Baréa, Nicolas .......................... 921 Barter, S.A. ............... 15, 123, 643 Basov, Valentin N. .................. 661 Baynham, J.M.W. .................... 789 Bellinger, Nick......................... 811 Benedictus, R. 493, 619, 839, 969 987, 1069, 1219, 1305 Ben-Simon, M............................ 55 Best, R.......................................... 3 Beumler, Thomas ............ 375, 765 Bhugaloo, H. ............................ 475 Blom, A.F................................. 707 Bode, Michel ........................... 109 Bombardier, Yan...................... 811 Bonnand, V. ............................. 743 Bordes, C. ................................ 931 Borgonje, Bob .......................... 765 Bosch, A.F. .............................. 145 Brot, A. .................................... 237 Burchill, M............................... 279

Falk, U. .................................... 707 Fallon, Timothy ..................... 1215 Fawaz, Scott A......187, 1035, 1345 Fleischer, Thomas................ 3, 559 Fressinet, M. ............................ 931 Fujita, S. .................................. 899 Furfari, D. .............................. 1219

Eastin, Robert G. ..................... 169 Evans, R................................... 279 Elyahu, D................................... 55

Gadalińska, Elżbieta .............. 1285 Galyon, Sarah E..................... 1035 Gary, M. Interator...................... 55 Ghilai, G. ................................... 55 Giacobbe, Tomamaso ............ 1177 Giglio, M. ................................ 859 Gomez, P. ................................ 743 Götze, M. ..................................... 3 Grandt, A.F. ............................ 871 Greer, James .......................... 1035 Guagliano, M. ......................... 859 Haan, P.H.de............................ 145 Hackel Lloyd ........................... 355 Haikola, Tapani ..................... 1321 Hammond, Matthew .............. 1035 Harrison, Jim ........................... 355 Hattenberg, T........................... 145 Havar, Tamas........................... 365 Hebden, Iain .......................... 1215 Heller, M.................................. 279 Hengel, Cees van ..................... 601 Herrmann, C. .......................... 225

Chaboche, J.L........................... 743 Cherouali, H............................. 743 Cheung, Terence ..................... 811 Child, D.R. ............................... 871 Das, Girindra K.......................... 61 Daverschot, Derk ..................... 375 David, T. .................................... 55 1365

1366

Hilfer, G. .................................. 225 Hirano, Noriyoshi..................... 165 Hirose, Yasuo .......................... 209 Hiscocks, R.J............................ 301 Hsu, Ching .............................. 407 Hu, W....................................... 643 Huber, Cyril ............................. 245 Hoang Ngoc, C.T. .................... 475 Irving, Phil ............................... 387 Jachimowicz, Jerzy .................. 939 Janhunen, Harri ...................... 1131 Johansson, S.A.H. .................... 585 Jones, Timothy G.B. .............. 1275 Juaneda, S. ............................. 1019 Jylhä, Juha.............................. 1121 Kaniowski, Jerzy ........... 939, 1285 Kanouté, P................................ 743 Kaye, R. ................................... 279 Khan, S..................................... 569 Khan, S.U................................. 839 Kimoto, Junichi...................... 1165 Klimaszewski, Sławomir ......... 263 Kool, G.A............................... 1261 Korbel, A. ................................ 449 Kortbeek, Peter ....................... 601 Koshioka, Yoshihiro .............. 1187 Koski, K. .................................. 707 Kressel, I. .......................... 55, 237 Krkoška, M. ............................. 969 Kuntjoro, Wahyu.................... 1155 Kurdelski, Marcin .................... 263 Kuroki, Hiroshi ........................ 685 Kuwayama, K. ......................... 899 Kwakernaak, A. ..................... 1219 Lahtinen, R............................... 707 Leap, Mike ............................... 355 Lemmen, H.J.K. .............. 539, 619 Leski, Andrzej.......................... 263 Liao, Min ......................... 245, 811 Linna, J..................................... 707 Liu, Q. ...................................... 543

Author Index

Llopart Prieto, Ll. .................. .513 Lucas, K.A............................. 1135 Lumppio, Kari ....................... 1321 Ma, Ye E.................................. 387 Machida, S.............................. 899 Machniewicz, T. ...................... 449 Madelpech, P. ........................ 1019 Mahal, Devinder .................... 1215 Main, B. ................................... 123 Mariani, U................................ 859 Marley, Serena......................... 355 Massé, Marie ........................... 921 Maxfield, K.............................. 123 McCoubrey, Brian ................. 1093 McDonald, M. ......................... 279 McIver, Keith .......................... 407 Mellings, S.C. .......................... 789 Meneghin, Ivan........................ 427 Miettinen, A............................. 707 Miller, Matthew ......................... 61 Mills, T. ................................... 123 Minakuchi, Shu........................ 209 Molent, L. .......................... 15, 123 Moore, Gary ............................ 387 Mosler, J. ................................. 569 Mountfort, Andy.................... 1093 Mowery, Jon B. ....................... 169 Moyle, N.J. .............................. 871 Mydin, Assanah ..................... 1155 Nachshhon, Y. ........................... 55 Nakamura, Noritsugu............. 1187 Nakamura, T. ........................... 899 Nemeth, Joe ............................ 355 Nesterenko, Boris .................... 661 Nesterenko, Grirory I............... 661 Ogisu, Toshimichi ................. 1187 Okabe, Yoji............................ 1187 Okada, T. ................................. 899 Ostoja-Kuczynski, E................ 743 Pacchione, Marco ............ 387, 427 Pacou, D. ................................. 743

Author Index

1367

Paletti, Ligieja .......................... 987 Panis, M. .................................. 931 Paroissien, E............................. 475 Paulmier, P............................... 743 Peled, A...................................... 55 Peleg-Wolfin, Y. ...................... 237 Peters, C. .................................. 225 Platenkamp, D.J. ...................... 145 Plokker, Matthijs...................... 375 Pradels, M. ............................. 1019

Stuible, Eckart ........................ 365 Suhaimi Ashari, M................. 1155 Swanton, G. ............................. 123 Swift, Steve................................ 91

Rankin, Jon .............................. 355 Rans, C.D. .......................... 73, 987 Ratti, G..................................... 859 Reed, Steve ............................ 1093 Reese, Eggert D. .................... 1275 Renaud, Guillaume........... 245, 811 Rodi, Riccardo ................. 493, 539 Rößler, N.................................. 225 Ruotsalainen, Marja ............... 1121 Rutledge, R.S .......................... .301

Uebersax, Andrea .................... 245 Ueda, Yusuke ......................... 685 Vaaraniemi, K.......................... 707 Vermeer, Pascal ............. 427, 1069 Vihonen, Juho........................ 1131 Viitanen, Tomi....................... 1131 Visa, Ari ................................ 1121 Vogel, F. .................................. 743 Vyshnevskyy, A. ..................... 569

Saarela, Olli............................ 1321 Sachse, M..................................... 3 Salonen, Tuoma ..................... 1121 Sandeep R. ............................. 1345 Santos, Jorge dos...................... 387 Sardo, Fabio .......................... 1177 Sashikuma, Hirofumi ............. 1165 Schmid, Lukas ....................... 1085 Schubbe, Joele.......................... 909 Schulze, Frank.......................... 559 Semsch, M.................................... 3 Servetti, Guido ......................... 387 Shinozaki, Masaharu................ 685 Shoales, Gregory A. ......... 187, 329 Siljander, A. ............................. 707 Simon, A. ................................... 55 Sinke, J............................. 585, 765 Skorupa, A. .............................. 449 Skorupa, M............................... 449 Soejima, Hideki...................... 1187 Sparr, Holger............................ 559 Spenninger, G. ......................... 513 Szymczyk, Elżbieta ................. 939

Tajima, Naoyuki .................... 1165 Takeda, Nobuo .....209, 1165, 1187 Tsutsui, Hiroaki ..................... 1165 Tusch, O. ................................. 225

Wagner, H................................ 513 Wakai, Hiroshi....................... 1187 Waldman, W............................ 279 Walters, Molly R. .................... 187 Wang, G.S. .............................. 707 Wanhill, R.J.H. .................. 15, 145 Wenke, Karsten ....................... 559 Westcott, R. ............................. 279 White, P. .................................. 123 Williams, Stewart .................... 387 Willis, Stephen ...................... 1199 Wilson, Greg............................ 539 Woerden, H.J.M. ................... 1219 Wojtas, Andrzej ..................... 1285 Wrona, Mirosław ..................... 263 Wronicz, Wojciech .................. 939 Yamashitam, Yoichi ................ 685 Yamauchi, Ippei....................... 209 Yazid Ahmad, M. .................. 1155 Yosef, Z. .................................. 237 Zhang, Xiang ........................... 387 Ziegenhorn, Matthias............... 559

PROCEEDINGS OF ICAF SYMPOSIA 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21.

ICAF-Doc. 175, Amsterdam, 1959, Ed. F.J. Plantema and J. Schijve, Publ. Pergamon Press ICAF-Doc. 228, Paris, 1961, Ed. W. Barrois and E.L. Ripley, Publ. Pergamon Press ICAF-Doc. 300, Rome, 1963, Ed. J. Schijve, J.R. Heath-Smith and E.R. Welbourne, Publ. Pergamon Press ICAF-Doc. 487, Munich, 1965, Ed. E. Gassner and W. Schütz, Publ. Pergamon Press ICAF-Doc. 595, Melbourne, 1967, Ed. J.Y. Mann and I.S. Milligan, Publ. Pergamon Press ICAF-Doc. 609, Miami, 1971, Ed. H.F. Hardrath and J.R. Davidson, Publ. NASA (SP-309) ICAF-Doc. 705, London, 1973, Ed. A.M. Stagg, Publ. RAE (TR 73183) ICAF-Doc. 801, Lausanne, 1975, Ed. J. Branger and F. Berger, Publ. F+W ICAF-Doc. 960, Darmstadt, 1977, Ed. O. Buxbaum and D. Schütz, Publ. LBF (TR-136) ICAF-Doc. 1169, Brussels, 1979, Ed. A. Maenhaut, Publ. Aeronautics Administration ICAF-Doc. 1216, Noordwijkerhout, 1981, Ed. J.B. de Jonge and H.H. van der Linden, Publ. NLR ICAF-Doc. 1336, Toulouse, 1983, Ed. R. Labourdette and D. Deviller, Publ. CEAT ICAF-Doc. 1490, Pisa, 1985, Ed. A. Salvetti and G. Cavallini, Publ. EMAS ICAF-Doc. 1601, Ottawa, 1987, Ed. D.L. Simpson, Publ. EMAS ICAF-Doc. 1680, Jerusalem, 1989, Ed. A. Berkovits, Publ. EMAS ICAF-Doc. 1822, Tokyo, 1991, Ed. A. Kobayashi, Publ. EMAS ICAF-Doc. 1951, Stockholm, 1993, Ed. A.F. Blom (2 vols), Publ. EMAS ICAF-Doc. 2055, Melbourne, 1995, Ed. J.M. Grandage and G.S. Jost (2 vols), Publ. EMAS ICAF-Doc. 2177, Edingburgh, 1997, Ed. R. Cook and P. Poole (2 vols), Publ. EMAS ICAF-Doc. 2269, Seattle, 1999, Ed. J.L. Rudd and R.M. Bader (2 vols), Publ. EPIC ICAF-Doc. 2289, Toulouse, 2001, Ed. J. Rouchon (2 vols), Publ. Cépaduès-Editions

22. ICAF-Doc. 2311, Lucerne, 2003, Ed. M. Guillaume (2 vols), Publ. EMAS 23. ICAF 2005 Proceedings, Hamburg, 2005, Ed. C. Dalle Donne (2 vols), Publ. DGLR 24. ICAF-Doc. 2417, Naples, 2007, Ed. L. Lazzeri and A. Salvetti (2 vols), Publ. Pacini 25. ICAF-Doc. 2420, Rotterdam, 2009, Ed. M.J. Bos, Publ. Springer

Intelligent Tinkering: Bridging the Gap between Science and Practice

Information and Communication Technologies in Tourism 2009: Proceedings of the International Conference in Amsterdam, The Netherlands, 2009

STACS 2009: Proceedings of the 26th International Symposium on Theoretical Aspects of Computer Science, Freiburg, Germany, 26-28 February 2009 (Leibniz International Proceedings in Informatics - LIPIcs)

NeuroAnalysis: Bridging the Gap between Neuroscience, Psychoanalysis and Psychiatry

Handbook of Evidence-Based Psychodynamic Psychotherapy: Bridging the Gap Between Science and Practice (Current Clinical Psychiatry)

Proceedings of the Ninth International Symposium on Cyclodextrins

Seventh IUTAM Symposium on Laminar-Turbulent Transition: Proceedings of the Seventh IUTAM Symposium on Laminar-Turbulent Transition, Stockholm, Sweden, 2009

ICAF 2009, Bridging the Gap between Theory and Operational Practice: Proceedings of the 25th Symposium of the International Committee on Aeronautical Fatigue, ... Rotterdam, The Netherlands, 27-29 May 2009

Intelligent Tinkering: Bridging the Gap between Science and Practice

Bridging the Gap

Bridging the Gap

Bridging the Gap

Information and Communication Technologies in Tourism 2009: Proceedings of the International Conference in Amsterdam, The Netherlands, 2009

Bridging the gap

Bridging the Gap

Sinope, The Results of Fifteen Years of Research. Proceedings of the International Symposium, 7-9 May 2009

Bridging the Gap between Theory and Practice in Educational Research: Methods at the Margins

Contemporary Ergonomics 2009: Proceedings of the International Conference on Contemporary Ergonomics 2009

Quantum theory and symmetries. Proceedings of the 3rd international symposium

Proceedings of International Science Education Conference 2009

Treatment of Eating Disorders: Bridging the research-practice gap

STACS 2009: Proceedings of the 26th International Symposium on Theoretical Aspects of Computer Science, Freiburg, Germany, 26-28 February 2009 (Leibniz International Proceedings in Informatics - LIPIcs)

NeuroAnalysis: Bridging the Gap between Neuroscience, Psychoanalysis and Psychiatry

Handbook of Evidence-Based Psychodynamic Psychotherapy: Bridging the Gap Between Science and Practice (Current Clinical Psychiatry)

Proceedings of the Ninth International Symposium on Cyclodextrins

FM 2009 : formal methods : second world congress, Eindhoven, the Netherlands, November 2-6, 2009 : proceedings

Proceedings of the Third Japan-USSR Symposium on Probability Theory

Intelligence and Personality: Bridging the Gap in Theory and Measurement

Proceedings of the Second Japan-USSR Symposium on Probability Theory

Professional Photographer May 2009

Architectural Record - May 2009

Chess Life - May 2009

Fundamentals of Piano Practice (2009)

Symposium on the Theory of Numerical Analysis

I. D. International Design Magazine May 2009

Symposium On The Theory Of Numerical Analysis

The Heritage of Egypt - Issue 5 - May 2009

Seventh IUTAM Symposium on Laminar-Turbulent Transition: Proceedings of the Seventh IUTAM Symposium on Laminar-Turbulent Transition, Stockholm, Sweden, 2009

ICAF 2009, Bridging the Gap between Theory and Operational Practice: Proceedings of the 25th Symposium of the International Committee on Aeronautical Fatigue, ... Rotterdam, The Netherlands, 27-29 May 2009

Intelligent Tinkering: Bridging the Gap between Science and Practice

Bridging the Gap

Bridging the Gap

Bridging the Gap

Information and Communication Technologies in Tourism 2009: Proceedings of the International Conference in Amsterdam, The Netherlands, 2009

Bridging the gap

Bridging the Gap

Sinope, The Results of Fifteen Years of Research. Proceedings of the International Symposium, 7-9 May 2009

Bridging the Gap between Theory and Practice in Educational Research: Methods at the Margins

Contemporary Ergonomics 2009: Proceedings of the International Conference on Contemporary Ergonomics 2009

Quantum theory and symmetries. Proceedings of the 3rd international symposium

Proceedings of International Science Education Conference 2009

Treatment of Eating Disorders: Bridging the research-practice gap

STACS 2009: Proceedings of the 26th International Symposium on Theoretical Aspects of Computer Science, Freiburg, Germany, 26-28 February 2009 (Leibniz International Proceedings in Informatics - LIPIcs)

NeuroAnalysis: Bridging the Gap between Neuroscience, Psychoanalysis and Psychiatry

Handbook of Evidence-Based Psychodynamic Psychotherapy: Bridging the Gap Between Science and Practice (Current Clinical Psychiatry)

Proceedings of the Ninth International Symposium on Cyclodextrins

FM 2009 : formal methods : second world congress, Eindhoven, the Netherlands, November 2-6, 2009 : proceedings

Proceedings of the Third Japan-USSR Symposium on Probability Theory

Intelligence and Personality: Bridging the Gap in Theory and Measurement

Proceedings of the Second Japan-USSR Symposium on Probability Theory

Professional Photographer May 2009

Architectural Record - May 2009

Chess Life - May 2009

Fundamentals of Piano Practice (2009)

Symposium on the Theory of Numerical Analysis

I. D. International Design Magazine May 2009

Symposium On The Theory Of Numerical Analysis

The Heritage of Egypt - Issue 5 - May 2009

Seventh IUTAM Symposium on Laminar-Turbulent Transition: Proceedings of the Seventh IUTAM Symposium on Laminar-Turbulent Transition, Stockholm, Sweden, 2009

Recommend Documents