Vous êtes sur la page 1sur 326

A dissertation presented for the degree of

Doctor of Philosophy

October 2009

School of Mechanical and Manufacturing Engineering,


University of New South Wales, Sydney, NSW 2052, Australia
I hereby declare that this submission is my own work and to the best of my knowledge
it contains no materials previously published or written by another person, or
substantial portions of material which have been accepted for the award of any other
degree or diploma at UNSW or any other educational institution, except where due
acknowledgement is made in the thesis. Any contribution made to the research by
others, with whom I have worked at UNSW or elsewhere, is explicitly acknowledged in
the thesis. I also declare that the intellectual content of this thesis is the product of my
own work, except to the extent that assistance from others in the projects design and
conception or in style, presentation and linguistic expression is acknowledged.

Signed ...................................................

Date ...................................................

i
An investigation is presented into the behaviour of carbon fibre composite joints
subjected to dynamic loading rates in the range of 0.1 m/s to 10 m/s. The research is
focused on the response of single fastener joints and more complex structural
arrangements involving multiple fasteners and complex loads.

Fasteners play a crucial role in the joining of aerospace components due to their ease
of installation and inspection and their resistance to creep and environmental
degradation. A consequence of the operating environment of aircraft is that many
critical load cases involve impact and crash. These loading events are characterised by
high loading rates, high kinetic energy and possibly loads well above the static design
case. The properties of composite materials change with loading rate, so it is likely that
the behaviour of bolted composite joints may also vary significantly. Dynamic
behaviour of bolted joints is an area of research that has been given little attention to
date. The few available papers on the topic are limited to the investigation of ideal
bearing loads and include some contradictory results.

The research developed a detailed understanding of the behaviour of bolted joints in


composite structures through a combined numerical and experimental investigation. A
set of quasi-static and dynamic single fastener joint tests was conducted to develop an
understanding of the complex failure mechanisms present in bolted composite joints.
Simple structural tests were developed to investigate the interaction of multiple bolts
in a joint. High speed camera footage, full-field strain measurement and CT scanning
techniques were all used to develop an understanding of the changes in the failure
process with increased loading rate.

Finite element analyses used implicit and explicit dynamic algorithms to model the
tests. The finite element analysis contributed to the understanding of the experimental
results as well as providing a predictive tool to minimise the need for further testing. A

ii
method of incorporating detailed information about bolt failure into large scale
structural models was investigated and developed.

This thesis is part of the research program of the Cooperative Research Centre for
Advanced Composite Structures and contributed to collaboration between the CRC-
ACS and the German Aerospace Centre (DLR). A period of 6 months was spent by the
author in Stuttgart to gain access to high rate testing equipment.

The original contributions of this thesis involve novel dynamic joint testing including
dynamic pull-through and structural tests. CT Scanning was utilised in a novel way to
investigate the complex failure modes within a bolted joint. Novel finite element
techniques were developed for modelling bolted joints at both a detailed level and a
simplified level for structural analyses. These contributions significantly improve the
current understanding of bolted joint failure, both quasi-statically and dynamically,
and will allow for more efficient design of bolted composite structures for crash and
impact loads.

iii
For Hilary.

iv
First and foremost I would like to acknowledge the financial support offered by the
Australian Government through the Australian Postgraduate Award. The Australian
Government also provided funding through International Science Linkages Grant
CG100184 that facilitated my secondment to the German Aerospace Centre (DLR).

I would like to thank the Cooperative Research Centre for Advanced Composite
Structures (CRC-ACS) for ongoing financial and in-kind support throughout my
candidature. The CRC-ACS provided a top-up scholarship, access to materials and
fabrication facilities and facilitated collaboration of research between myself and the
DLR. Special thanks go to David Elder who has helped with the research throughout.
Andrew Gunnion and Michael Marelli have also provided assistance along the way.

I would like to thank the DLR for providing access to test facilities that enabled all of
the dynamic testing to be conducted. The research would have not been possible
without this assistance. Special thanks to Alastair Johnson whose support in Germany
and since has been invaluable. Thanks to Harald Kraft for technical assistance.

ESI Group (through Pacific ESI) provided software and support for numerical modelling.
Thanks to Damian McGuckin and Allen Chhor in particular for their expertise.

Special thanks to my supervisors; Don Kelly and Rodney Thomson, for their guidance
and scrutiny of the research program.

Thank you to others who have helped out along the way; Luke Djukic, Mathew
Joosten, Zoltan Mikulik, Rob Wootton, Ian Watson and many more.

Last, and by no means least, thanks go to my family, friends and in particular my


fiance Hilary, for love and support during the epic challenge of the last few years.

v
Originality Statement ................................................................................................ i

Abstract.................................................................................................................... ii

Acknowledgements .................................................................................................. v

Table of Contents .................................................................................................... vi

List of Figures ........................................................................................................ xiii

List of Symbols and Abbreviations....................................................................... xxviii

Chapter 1 : Introduction ........................................................................................... 1

1 Background ............................................................................................................ 2

2 Goals of the Research ............................................................................................ 4

3 Thesis Outline ......................................................................................................... 6

3.1 Literature Review ............................................................................................ 6

3.2 Single Fastener Joint Testing .......................................................................... 6

3.3 Detailed Joint Modelling ................................................................................. 6

3.4 Structural Testing ............................................................................................ 7

3.5 Structural Modelling ....................................................................................... 7

3.6 Conclusions and Future Work......................................................................... 7

3.7 Appendices...................................................................................................... 7

4 Intellectual Property Generated ............................................................................ 8

4.1 Conferences Proceedings ............................................................................... 8

4.2 Journal Papers................................................................................................. 8

4.3 Technical Reports............................................................................................ 8

Chapter 2 : Literature Review ................................................................................... 9

vi
1 Introduction ......................................................................................................... 10

2 Composite Failure Modes .................................................................................... 10

2.1 Fibre Failure .................................................................................................. 10

2.2 Matrix Failure ................................................................................................ 16

2.3 Delamination................................................................................................. 20

3 Dynamic Effects in Isotropic and Composite Materials ....................................... 22

3.1 Fibre Properties ............................................................................................ 23

3.2 Matrix Properties .......................................................................................... 24

3.3 Laminate Properties...................................................................................... 25

3.4 Delamination................................................................................................. 25

3.5 Titanium Properties ...................................................................................... 26

3.6 Effects of Heat Generation on the Dynamic Failure of Composites ............. 27

4 Pin Loaded Holes .................................................................................................. 29

5 Bolted Joints ......................................................................................................... 33

5.1 Bearing Loaded Joints ................................................................................... 34

5.2 Pull-through Loaded Joints ........................................................................... 37

6 Dynamically Loaded Bolted Joints ....................................................................... 39

7 Composite Failure Onset Theories ....................................................................... 42

7.1 Maximum Value Failure Theories ................................................................. 42

7.2 Interactive Failure Theories .......................................................................... 45

7.3 Mechanistic Failure Theories ........................................................................ 48

7.4 Failure Mode Concept Failure Theories........................................................ 52

7.5 Delamination................................................................................................. 54

8 Damage Modelling ............................................................................................... 57

8.1 Intralaminar Damage .................................................................................... 57

8.2 Interlaminar (Delamination) Damage ........................................................... 60

vii
9 Pin Loaded Hole Modelling .................................................................................. 62

9.1 Analytical Solutions....................................................................................... 62

9.2 Critical Distance Approaches ........................................................................ 63

9.3 2-D Pin Loaded Hole Modelling .................................................................... 64

9.4 3-D Pin Loaded Hole Modelling .................................................................... 65

10 Bolted Joint Modelling ..................................................................................... 66

10.1 Bearing Modelling ..................................................................................... 67

10.2 Pull-through Modelling ............................................................................. 67

10.3 Simplified Modelling for Explicit Analysis ................................................. 68

11 Conclusion ........................................................................................................ 69

Chapter 3 : Single Fastener Joint Tests .................................................................... 71

1 Introduction ......................................................................................................... 72

1.1 Hi-Lok Bolts ................................................................................................... 72

1.2 Pull-through Tests ......................................................................................... 72

1.3 Bearing Tests ................................................................................................. 73

1.4 Quasi-static Tests .......................................................................................... 74

1.5 Dynamic Tests ............................................................................................... 75

1.6 Video Capture ............................................................................................... 76

1.7 CT Scanning ................................................................................................... 76

2 Pull-Through Tests................................................................................................ 77

2.1 Quasi-static Pull-Through Tests .................................................................... 77

2.2 Dynamic Pull-Through Tests ......................................................................... 82

2.3 Quasi-static versus Dynamic Pull-through Test Comparison ....................... 92

3 Bearing Tests ........................................................................................................ 93

3.1 Quasi-static Bearing Tests ............................................................................. 93

3.2 Dynamic Bearing Tests .................................................................................. 98

viii
4 Discussion ........................................................................................................... 117

4.1 Pull-through Rate Dependence .................................................................. 118

4.2 Bearing Rate Dependence .......................................................................... 118

4.3 Interaction of Failure Modes ...................................................................... 118

4.4 Novel Rate Dependent Joint Design ........................................................... 123

5 Conclusion .......................................................................................................... 123

Chapter 4 : Detailed Joint Modelling .................................................................... 125

1 Introduction ....................................................................................................... 126

1.1 Modelling Philosophy ................................................................................. 126

1.2 Mesh Generation Tool ................................................................................ 127

1.3 Non-linear Implicit Finite Element Analysis ................................................ 130

1.4 Explicit Finite Element Analysis .................................................................. 130

2 Implicit Finite Element Modelling ...................................................................... 133

2.1 Implicit Modelling Philosophy .................................................................... 133

2.2 Composite Material Model ......................................................................... 133

2.3 Delamination............................................................................................... 135

2.4 Countersunk Bearing Model ....................................................................... 136

3 Explicit Finite Element Modelling ...................................................................... 148

3.1 Explicit Modelling Philosophy ..................................................................... 148

3.2 Sublaminate (Stacked Shell) Modelling ...................................................... 150

3.3 Composite Material Model ......................................................................... 151

3.4 Cohesive Zone Interface Model .................................................................. 153

3.5 Contact ........................................................................................................ 157

3.6 Countersunk Pull-through Model ............................................................... 157

3.7 Countersunk Bearing Model ....................................................................... 176

3.8 General Comments on Stacked Shell Modelling ........................................ 186

ix
4 Conclusions ........................................................................................................ 187

Chapter 5 : Structural Testing ............................................................................... 189

1 Introduction ....................................................................................................... 190

2 Structural Geometry .......................................................................................... 191

3 Line Loaded Test Arrangement .......................................................................... 193

3.1 Quasi-static Testing..................................................................................... 193

3.2 High-rate Testing ........................................................................................ 202

4 Point Loaded Test Arrangement ........................................................................ 209

4.1 High-rate Testing ........................................................................................ 209

5 Discussion ........................................................................................................... 216

6 Conclusion .......................................................................................................... 218

Chapter 6 : Structural Modelling .......................................................................... 219

1 Introduction ....................................................................................................... 220

2 Simplified Bolt Modelling Techniques ............................................................... 220

3 Finite Element Model ......................................................................................... 221

3.1 Composite Material Model ......................................................................... 222

3.2 Point Link Elements .................................................................................... 222

4 Line Loaded Structure ........................................................................................ 230

4.1 Model .......................................................................................................... 230

4.2 Quasi-static Model Results ......................................................................... 232

4.3 Dynamic Model Results .............................................................................. 234

4.4 Discussion ................................................................................................... 236

5 Point Loaded Structure ...................................................................................... 237

5.1 Model .......................................................................................................... 237

5.2 Results ......................................................................................................... 239

5.3 Discussion ................................................................................................... 241

x
6 Conclusion .......................................................................................................... 242

7 PLINK Recommendations ................................................................................... 243

7.1 PLINK Modelling Guidelines ........................................................................ 243

7.2 PLINK Rupture Model Improvements ......................................................... 244

Chapter 7 : Conclusion ......................................................................................... 247

1 Overview ............................................................................................................ 248

2 Key Findings and Advancements ....................................................................... 248

2.1 Dynamic Bolted Joint Testing ..................................................................... 248

2.2 Dynamic Testing of Bolted Joints in Simplified Composite Structures ....... 250

2.3 Novel Application of CT Scanning to the Investigation of Bolted Joint Failure


in Composite Materials ......................................................................................... 250

2.4 Advancement of FE Modelling Techniques for Bolted Composite Joints... 251

3 Future Work ....................................................................................................... 253

3.1 Arcan Testing .............................................................................................. 253

3.2 CT Scanning ................................................................................................. 254

3.3 Solid Element Modelling ............................................................................. 255

3.4 Interdependence of Failure Mode and Rate Effects .................................. 256

3.5 Effects of Heat Generation and Friction ..................................................... 256

References ........................................................................................................... 257

Appendix A : Strain Invariant Failure Theory (SIFT) ................................................A-1

1 Background ........................................................................................................ A-1

2 Invariants............................................................................................................ A-1

3 Micromechanical Enhancement ........................................................................ A-2

4 Implementation ................................................................................................. A-6

5 Application of SIFT to prediction of inter-laminar splitting failure mode in a


composite lug ............................................................................................................ A-7

xi
Appendix B : CT Scanning ...................................................................................... B-1

1 X-ray Computer Tomography (CT) ..................................................................... B-1

Appendix C : Hi-Lok Fasteners ............................................................................... C-1

1 Hi-Lok Fastener Details .......................................................................................C-1

Appendix D : Marc FORTRAN Subroutines ............................................................ D-1

1 Structure............................................................................................................. D-1

2 Failure Indices .................................................................................................... D-1

2.1 Modified Hashin Fabric ............................................................................... D-1

2.2 Max Stress Fabric ........................................................................................ D-2

3 UFAIL .................................................................................................................. D-3

4 UPROGFAIL ......................................................................................................... D-4

Appendix E : PLINK Property Investigation ............................................................ E-1

1 PLINK Properties.................................................................................................. E-1

1.1 Comprehensive Element Behaviour ............................................................ E-1

1.2 Stiffness from Curves ................................................................................... E-5

1.3 Multiple PLINKS............................................................................................ E-6

xii
Figure 1-1: Bolted joint failure modes .............................................................................. 3

Figure 2-1: Distribution of initial fibre damage............................................................... 11

Figure 2-2: Fibre tensile failure surface .......................................................................... 11

Figure 2-3: Fibre tensile failure including fibre pull-out ................................................. 12

Figure 2-4: Simplified compressive response of composite lamina ............................... 13

Figure 2-5: Compressive kink-band failure ..................................................................... 13

Figure 2-6: Flexural fibre failure during kink-band formation ........................................ 14

Figure 2-7: Fibre compressive failure in shear ................................................................ 14

Figure 2-8: Broughtman test schematic .......................................................................... 15

Figure 2-9: Complimentary kink-bands in pin bearing.................................................... 16

Figure 2-10: Gross matrix tensile failure ......................................................................... 17

Figure 2-11: Matrix tensile micro-cracking ..................................................................... 18

Figure 2-12: Transverse compressive shear crack .......................................................... 18

Figure 2-13: Matrix non-linearity during cyclic shear testing of 45 braided composite
......................................................................................................................................... 19

Figure 2-14: Localised discrete matrix damage leading to non-linear shear response .. 19

Figure 2-15: Tensile delamination of epoxy composite .................................................. 20

Figure 2-16: Shear delamination of composite parallel to the fibres ............................21

xiii
Figure 2-17: Matrix shear failure modes in a fibre composite material ......................... 21

Figure 2-18: CT scan of failed pin loaded bearing specimen .......................................... 22

Figure 2-19: Compressive response of AS4/3501-6 unidirectional composite .............. 23

Figure 2-20: Matrix damage at comparison between quasi-static and dynamic tests
loaded to 4.9% strain ...................................................................................................... 25

Figure 2-21: Fracture surfaces for quasi-static and dynamic crack growth.................... 26

Figure 2-22: Temperature effects in neat polymer resins .............................................. 28

Figure 2-23: Schematic of a pin loaded bearing test ...................................................... 29

Figure 2-24: Simplified FBD of reaction with a section at the net tension plane ........... 30

Figure 2-25: Load paths for an elastic isotropic plate loaded in bearing ....................... 30

Figure 2-26: Optical micrograph of failed bearing specimen ......................................... 31

Figure 2-27: Effect of clamping the laminate through the thickness on the bearing
strength of a pin-loaded hole.......................................................................................... 32

Figure 2-28: Progressive pin bearing in a quasi-isotropic fabric laminate...................... 32

Figure 2-29: Joint failure modes ..................................................................................... 33

Figure 2-30: Simplified FBD for a double-lap bolted joint .............................................. 34

Figure 2-31: Simplified FBD for a single-lap bolted joint without equilibrium ............... 34

Figure 2-32: Load alignment and secondary bending ..................................................... 35

Figure 2-33: Bolt hole coordinate system ....................................................................... 35

Figure 2-34: Laminate stresses for a single-lap bolted joint connecting homogeneous
orthotropic plates ........................................................................................................... 36

Figure 2-35: Bearing damage in single lap bolted joints ................................................. 37

xiv
Figure 2-36: Conical damage zone during fastener pull-through ................................... 38

Figure 2-37: Typical load-displacement curve for a pull-through test ........................... 38

Figure 2-38: Bending and shear stresses in pull-through tests....................................... 39

Figure 2-39: Dynamic-static comparison for a single lap hybrid bolted joint................. 40

Figure 2-40: Comparison between quasi-static and dynamic single lap bearing tests .. 41

Figure 2-41: Maximum stress theory failure surface in biaxial stress space .................. 44

Figure 2-42: Tsai-Wu theory failure surface ................................................................... 47

Figure 2-43: Hashin-Rotem theory failure surface in transverse-shear stress space ..... 49

Figure 2-44: Matrix failure envelope for Puck and Schrmann theory .......................... 51

Figure 2-45: Fracture mechanics delamination modes .................................................. 56

Figure 2-46: Damage modelling using the element failure method ............................... 59

Figure 2-47: Stress-displacement function for a typical cohesive zone model .............. 61

Figure 2-48: Cohesive zone connections ........................................................................ 62

Figure 2-49: Characteristic curve for bearing failure ...................................................... 63

Figure 2-50: Bolted joint under bending load ................................................................. 68

Figure 3-1: Hi-Lok bolt cross-section ............................................................................... 72

Figure 3-2: Basic schematic of pull-through test specimen assembly ............................ 73

Figure 3-3: Loading of pull-through specimen ................................................................ 73

Figure 3-4: Basic schematic of bearing test specimen assembly .................................... 74

Figure 3-5: Loading of bearing specimen ........................................................................ 74

Figure 3-6: Pull-through test fixture ............................................................................... 77

xv
Figure 3-7: Close-up of pull-through test fixture ............................................................ 78

Figure 3-8: Quasi-static pull-through test results ........................................................... 79

Figure 3-9: Displacement measurements for pull-through test ..................................... 79

Figure 3-10: Bolt assembly stiffness calibration ............................................................. 80

Figure 3-11: Effect of local damage on pull-through LVDT results ................................. 81

Figure 3-12: LVDT output ................................................................................................ 81

Figure 3-13: Comparison between different bolt heads for pull-through failure .......... 82

Figure 3-14: Dynamic pull-through test fixture .............................................................. 83

Figure 3-15: Different boundary conditions for pull-through tests ................................ 84

Figure 3-16: Instrumented bar used for pull-through tests ............................................ 85

Figure 3-17: Dynamic pull-through test results .............................................................. 85

Figure 3-18: 0.1 m/s pull-through test failure progression ............................................ 87

Figure 3-19: Typical failure mode for pull-through specimens....................................... 88

Figure 3-20: Unique failure mode for one pull-through specimen tested at 1 m/s ....... 88

Figure 3-21: Full-field strain measurement results for pull-through test ....................... 89

Figure 3-22: 2D Cross-section of failed specimen ........................................................... 90

Figure 3-23: 3D constructed image of failed pull-through specimen ............................. 90

Figure 3-24: Negative image of failed pull-through specimen ....................................... 90

Figure 3-25: Two through thickness failure modes present in pull-through specimens 91

Figure 3-26: Progression of the through thickness damage in a pull-through test ........ 91

Figure 3-27: Low rate versus high rate pull-through test data ....................................... 92

xvi
Figure 3-28: ASTM D 5961 bearing test fixture............................................................... 93

Figure 3-29: Bearing test specimen dimensions ............................................................. 94

Figure 3-30: Face mounted extensometer ..................................................................... 94

Figure 3-31: Face mounted extensometer mounted on specimen ................................ 95

Figure 3-32: Comparison between protruding head and countersunk fastener joints .. 96

Figure 3-33: The effect of load-unload-reload cycle on bearing test with countersunk
fastener ........................................................................................................................... 96

Figure 3-34: Damaged quasi-static bearing specimens .................................................. 97

Figure 3-35: Dynamic bearing test fixture ...................................................................... 99

Figure 3-36: Strain gauge location used for calculating joint load ................................. 99

Figure 3-37: Assembled protruding head fastener test specimen ................................. 99

Figure 3-38: Protruding head fastener tests at 0.001 m/s ........................................... 100

Figure 3-39: Dynamic protruding head fastener tests at 0.1 m/s ................................ 100

Figure 3-40: Dynamic protruding head fastener tests at 1 m/s.................................... 101

Figure 3-41: Dynamic protruding head fastener tests at 10 m/s ................................. 101

Figure 3-42: Failure modes for dynamic bearing specimens with protruding head
fasteners........................................................................................................................ 102

Figure 3-43: Comparison between protruding head fastener joint tests at different
loading rates .................................................................................................................. 103

Figure 3-44: Dynamic-static comparison for a single lap hybrid bolted joint ............... 104

Figure 3-45: Energy absorption for the dynamic tests ................................................. 104

Figure 3-46: Net tension failure of specimen 12 .......................................................... 105

xvii
Figure 3-47: Bearing-cleavage failure of specimen 6 .................................................... 106

Figure 3-48: Dynamic bearing test results for countersunk fastener joints ................. 107

Figure 3-49: Low rate (0.1 m/s) bearing test failure progression ................................. 108

Figure 3-50: High rate (10 m/s) bearing test failure progression ................................. 109

Figure 3-51: Penetration of the bolt head into the laminate ....................................... 110

Figure 3-52: Position and rotation of the bolt tail at failure ......................................... 110

Figure 3-53: Bolt failure mode for low rate dynamic bearing specimens with
countersunk fasteners .................................................................................................. 111

Figure 3-54: Progressive bearing failure mode for high rate dynamic bearing specimens
with countersunk fasteners .......................................................................................... 111

Figure 3-55: Full-field strain measurement results for bearing test ............................. 112

Figure 3-56: Limit of full-field strain measurement due to surface damage ................ 112

Figure 3-57: CT scan reconstruction of the tail side of bolt failure specimen ............. 113

Figure 3-58: CT cross-sections through the width of the bolt failure joint .................. 114

Figure 3-59: Shape of scanned progressive bearing failure specimen ......................... 115

Figure 3-60: CT cross-sections through the width of the progressive bearing failure joint
....................................................................................................................................... 116

Figure 3-61: Difference between load-displacement measurements from the quasi-


static and low-rate (0.1 m/s) dynamic tests ................................................................. 117

Figure 3-62: Range of joint geometries at which changes in failure mode will occur .. 120

Figure 3-63: Possible loading rate behaviour for protruding head bearing tests......... 122

Figure 3-64: Possible loading rate behaviour for countersunk head bearing tests ...... 122

xviii
Figure 4-1: In-plane geometry parameters ................................................................... 128

Figure 4-2: Through-thickness geometry parameters .................................................. 128

Figure 4-3: Bolt, collar and hole geometry parameters ................................................ 128

Figure 4-4: Pull-through model mesh example (solid plate) ........................................ 129

Figure 4-5: Bearing model mesh example (stacked shell plate) ................................... 129

Figure 4-6: Stress-strain curve for progressive damage ............................................... 135

Figure 4-7: Implicit FE bearing model mesh with countersunk bolt ............................. 136

Figure 4-8: Bearing countersunk head bolt mesh ......................................................... 137

Figure 4-9: Bearing countersunk head laminate mesh ................................................. 138

Figure 4-10: Through-thickness compressive stress distribution ................................. 139

Figure 4-11: Through-thickness stress at the interface between the two laminates
(MPa) ............................................................................................................................. 140

Figure 4-12: Stress in the loading direction after bolt pre-load ................................... 140

Figure 4-13: Loading direction for countersunk bearing model ................................... 141

Figure 4-14: Through-thickness stresses under applied bearing load .......................... 141

Figure 4-15: Stress in the loading direction prior to failure initiation .......................... 142

Figure 4-16: Comparison between experimental results and default simulation ........ 142

Figure 4-17: Effect of the stiffness reduction term for compressive matrix damage .. 143

Figure 4-18: Simple stiffness degradation leading to infinitely strong material .......... 143

Figure 4-19: Progressive damage simulation with included element elimination ....... 144

Figure 4-20: Through-thickness shear crack prediction................................................ 145

xix
Figure 4-21: Comparison between stiffness degradation and damage mechanics...... 145

Figure 4-22: Comparison between experimental results and default simulation ........ 146

Figure 4-23: Progressive damage using progressive stiffness reduction ...................... 147

Figure 4-24: Two representations of composite laminate in a pull-through model .... 148

Figure 4-25: Composite laminate represented two equivalent ways........................... 150

Figure 4-26: 2-D cantilevered plate .............................................................................. 150

Figure 4-27: Stacked shell verification for a built-in plate ............................................ 151

Figure 4-28: Strain based damage evolution ................................................................ 152

Figure 4-29: Strain based modulus reduction ............................................................... 153

Figure 4-30: Stress as a function of strain ..................................................................... 153

Figure 4-31: Mode I stress-opening function for Pickett model in PAM-CRASH .......... 154

Figure 4-32: Relationship between G0 and Gu and the damage parameters................ 155

Figure 4-33: Finite element models used to calibrate interface properties ................. 156

Figure 4-34: Delamination material card ...................................................................... 156

Figure 4-35: Pull-through model mesh with countersunk bolt .................................... 158

Figure 4-36: Pull-through countersunk bolt mesh ........................................................ 158

Figure 4-37: Pull-through bushing mesh ....................................................................... 158

Figure 4-38: Pull-through loading plate mesh .............................................................. 158

Figure 4-39: Countersunk sublaminate arrangement relative to the laminate geometry


....................................................................................................................................... 159

Figure 4-40: Pull-through stacked shell laminate mesh ............................................... 160

xx
Figure 4-41: Sublaminate to solid contact for the countersunk head bolt ................... 161

Figure 4-42: Pull-through model boundary conditions................................................. 161

Figure 4-43: Failure sequence for the pull-through simulation .................................... 163

Figure 4-44: Default model results compared to experimental results ....................... 164

Figure 4-45: Delamination growth within bottom interface ........................................ 165

Figure 4-46: Variation of Mode I initiation stress ......................................................... 166

Figure 4-47: Variation of Mode I energy release rate ................................................... 166

Figure 4-48: Variation of Mode I propagation stress .................................................... 167

Figure 4-49: Variation of Mode II initiation stress ........................................................ 167

Figure 4-50: Variation of Mode II energy release rate.................................................. 168

Figure 4-51: Variation of Mode II propagation stress ................................................... 168

Figure 4-52: Variation of bolt preload displacement .................................................... 169

Figure 4-53: Stacked shell laminate under preload by a countersunk fastener ........... 170

Figure 4-54: Hou delamination model .......................................................................... 171

Figure 4-55: Location of extra delamination interface links ......................................... 171

Figure 4-56: Comparison between the default pull-through load response with and
without extra interface links ......................................................................................... 172

Figure 4-57: Comparison between the default pull-through damage a) with and b)
without extra interface links ......................................................................................... 173

Figure 4-58: Comparison between the experimental and numerical pull-through load
response ........................................................................................................................ 173

xxi
Figure 4-59: Comparison between experimental and numerically predicted external
damage .......................................................................................................................... 174

Figure 4-60: Comparison between experimental and numerically predicted


delamination damage ................................................................................................... 174

Figure 4-61: Effect of increasing experimental loading rate or decreasing GII interface
properties ...................................................................................................................... 176

Figure 4-62: Bearing model mesh with countersunk bolt ............................................ 177

Figure 4-63: Bearing countersunk bolt mesh ................................................................ 178

Figure 4-64: Bearing collar mesh .................................................................................. 178

Figure 4-65: Bearing top laminate mesh for countersunk bolt .................................... 178

Figure 4-66: Bearing bottom laminate mesh ................................................................ 179

Figure 4-67: Bearing test support mesh........................................................................ 179

Figure 4-68: Sublaminate to solid contact .................................................................... 180

Figure 4-69: Stiffness correlation between the quasi-static tests and the numerical
predictions .................................................................................................................... 181

Figure 4-70: Comparison between experimental load and numerical prediction for the
countersunk bearing model .......................................................................................... 182

Figure 4-71: Failure sequence for the countersunk bearing simulation....................... 183

Figure 4-72: Bearing failure around the bolt head ....................................................... 184

Figure 4-73: Delamination failure around the bolt collar ............................................. 184

Figure 4-74: Bending of stacked shell laminates .......................................................... 186

Figure 4-75: Failure sequence for through-thickness shear damage in stacked shell
laminates ....................................................................................................................... 187

xxii
Figure 5-1: Load alignment and secondary bending ..................................................... 190

Figure 5-2: Combined loading of bolted joint ............................................................... 191

Figure 5-3: Structural geometry (all dimensions in mm) .............................................. 192

Figure 5-4: Test rig and loading direction ..................................................................... 194

Figure 5-5: Line loaded test rig ..................................................................................... 194

Figure 5-6: Load-displacement curves for quasi-static line-loaded tests ..................... 195

Figure 5-7: Brittle composite failure mode of the five fastener test ............................ 195

Figure 5-8: Test sequence for the quasi-static line loaded test with three fasteners .. 196

Figure 5-9: Failure mode for the three fastener test .................................................... 197

Figure 5-10: Failure mode for the two fastener test .................................................... 198

Figure 5-11: Angular displacement of bolt ................................................................... 198

Figure 5-12: Friction effect in the line loaded test rig .................................................. 199

Figure 5-13: Joint separation during loading ................................................................ 200

Figure 5-14: Prying loads on bolted joints .................................................................... 200

Figure 5-15: Similarities between the observed damage and pull-through damage ... 201

Figure 5-16: Test fixture for line loaded high-rate tests ............................................... 202

Figure 5-17: High-rate line load test fixture.................................................................. 203

Figure 5-18: Load-displacement curves for dynamic line loaded tests ........................ 203

Figure 5-19: Test sequence for the 1 m/s line loaded test ........................................... 204

Figure 5-20: Test sequence for 0.1 m/s line loaded test .............................................. 205

Figure 5-21: Bending wave propagation in 10 m/s test ................................................ 206

xxiii
Figure 5-22: Failure mode of unprotected composite plates ....................................... 207

Figure 5-23: Evidence of bearing damage in 0.1 m/s test ............................................ 207

Figure 5-24: Failure mode of plate supported 1 m/s test ............................................. 208

Figure 5-25: Quasi-static and High Rate Test Comparison ........................................... 209

Figure 5-26: Test fixture for point loaded high-rate tests ............................................ 210

Figure 5-27: Tab locations for point loaded test........................................................... 210

Figure 5-28: Specimen in test rig .................................................................................. 210

Figure 5-29: Dynamic ball loaded test comparison ...................................................... 211

Figure 5-30: Test sequence for the 1 m/s ball loaded test ........................................... 212

Figure 5-31: Representative panel damage to the ball loaded test ............................. 213

Figure 5-32: Overview of scanned ball loaded test specimen ...................................... 214

Figure 5-33: Low and Intermediate rate ball loaded test data ..................................... 215

Figure 5-34: Dynamic pull-through test results ............................................................ 215

Figure 5-35: Joint loads during line loaded structure tests .......................................... 217

Figure 5-36: Load alignment via secondary bending .................................................... 217

Figure 6-1: Load-displacement curves simplified joint models .................................... 221

Figure 6-2: Base finite element model of the composite structure ............................. 221

Figure 6-3: Effect of penalising different DOFs ............................................................. 223

Figure 6-4: PLINK element schematic ........................................................................... 224

Figure 6-5: Total element force () ............................................................................... 225

Figure 6-6: Displacement based softening function () ............................................... 225

xxiv
Figure 6-7: PLINK load-displacement response ............................................................ 226

Figure 6-8: Single fastener joint test data and PLINK calibration strategies showing .. 227

Figure 6-9: Corresponding pull-through PLINK calibration for the displacement method
....................................................................................................................................... 227

Figure 6-10: Pull-through stiffness calibration using PAMCRASH model ..................... 228

Figure 6-11:Pull-through stiffness calibration using LVDT calibration ......................... 229

Figure 6-12: Calibration of PLINK elements for dynamic simulation ............................ 229

Figure 6-13: Line loaded test fixture ............................................................................. 230

Figure 6-14: Geometry of line-loaded half symmetry FE model .................................. 230

Figure 6-15: Boundary conditions for line loaded model ............................................. 231

Figure 6-16: Axial load-displacement response of support bars .................................. 231

Figure 6-17: Comparison between line loaded experiment and FE models ................. 232

Figure 6-18: Comparison between experimental and numerical results ..................... 233

Figure 6-19: Comparison between the experimental results and other joint modelling
methods ........................................................................................................................ 234

Figure 6-20: High-rate line load test fixture.................................................................. 234

Figure 6-21:Comparison between dynamic experiment and FE model with dynamic


PLINK calibration ........................................................................................................... 235

Figure 6-22: Ball loaded test specimen in fixture ......................................................... 237

Figure 6-23: Geometry of point-loaded finite element model ..................................... 238

Figure 6-24: Experimental load-deflection response compared to numerical analyses


....................................................................................................................................... 239

xxv
Figure 6-25: Damage comparison for different stacking arrangements ...................... 240

Figure 6-26: Failure surfaces in displacement space showing three loading paths ..... 244

Figure 6-27: Displacement-based model calibrated to B32 test data .......................... 246

Figure 7-1: Combined loading of bolted joint ............................................................... 253

Figure 7-2: Arcan test fixture ........................................................................................ 254

Figure 7-3: Countersunk bearing test with key points of interest ................................ 255

Figure A-1: Fibre matrix packing arrangements ........................................................... A-3

Figure A-2: Square array unit cell with single applied strain ........................................ A-4

Figure A-3: Thermal load applied to square array unit cell .......................................... A-5

Figure A-4: Schematic representation of SIFT subroutine ............................................ A-6

Figure A-5: Splitting mode of lug with TFP layers ......................................................... A-8

Figure A-6: Fibre orientation in tailored fibre pattern (TFP) ........................................ A-8

Figure A-7: Lug finite element model ........................................................................... A-9

Figure A-8: Preliminary SIFT results .............................................................................. A-9

Figure B-1: Schematic of X-ray CT arrangement ........................................................... B-1

Figure C-1: Hi-Lok bolt cross-section..............................................................................C-1

Figure C-2: HL523 ...........................................................................................................C-2

Figure C-3: HL1012 .........................................................................................................C-3

Figure C-4: HL97 .............................................................................................................C-4

Figure C-5: HL1087 .........................................................................................................C-5

Figure D-1: Schematic of subroutine structure ............................................................. D-1

xxvi
Figure D-2: Schematic of UFAIL subroutine .................................................................. D-3

Figure D-3: Schematic of UPROGFAIL subroutine ......................................................... D-4

Figure E-1: Vector breakdown of PLINK element .......................................................... E-2

Figure E-2: Total element force () ................................................................................ E-3

Figure E-3: Displacement based softening function () ................................................ E-3

Figure E-4: Irregular loading condition .......................................................................... E-4

Figure E-5: Multiple PLINK connectivity ......................................................................... E-6

Figure E-6: PLINK test mesh ........................................................................................... E-7

Figure E-7: Satellite PLINK arrangement ........................................................................ E-7

Figure E-8: Effect of multiple PLINKs.............................................................................. E-8

xxvii
This section summarises commonly used symbols and abbreviations throughout the
thesis. The list of symbols is not comprehensive; variations exist within the text but are
defined explicitly whenever they are used.

Stress
i or ii Direct stress in i direction
ij or ij Shear stress in ij plane
Xt, Yt, Zt Allowable tensile stress in the X, Y and Z directions
Xc, Yc, Zc Allowable compressive stress in the X, Y and Z directions
Sij Allowable shear stress in the ij plane
S Allowable in-plane shear stress
ST Allowable transverse shear stress

Strain
i or ii Direct strain in i direction
ij Tensor shear strain in ij plane
ij Engineering shear strain in ij plane
1, 2, 3 Principal strains
it Allowable tensile strain in the i direction
ic Allowable compressive strain in the i direction
ij Allowable (tensor) shear strain in the ij direction

E Young's Modulus
Eii Young's Modulus in the i direction
Et Tangent modulus
Gij Shear modulus in the ij plane
ij Poisson's ratio in the ij plane

xxviii
K Linear spring stiffness
Dij Stiffness degradation factors for stiffness term ij

GIc or G1c Mode I strain energy release rate


GIIc or G2c Mode II strain energy release rate
s Mode I crack initiation stress
s Mode II crack initiation stress
p Mode I crack propagation stress
p Mode II crack propagation stress

d Bolt diameter
w Joint width
e Joint edge distance
t Joint laminate thickness

X First time derivative of variable X


X Second time derivative of variable X

BOJCAS Bolted Joints in Composite Aircraft Structures


CDM Continuum Damage Mechanics
CFRP Carbon Fibre Reinforced Polymer
CRC-ACS Cooperative Research Centre for Advanced Composite Structures
CSA Cross-sectional Area
CT Computer Tomography
DCB Double Cantilever Beam
DLR German Aerospace Centre
EFM Element Failure Method
ENF End Notch Flexure
ERT Expanding Ring Test
FBD Free Body Diagram
FE Finite Element
FF Fibre Failure

xxix
FMC Failure Mode Concept
IFF Inter-Fibre Failure
IT Impact Test
LVDT Linear Variable Differential Transformer
MME Micro-Mechanically Enhanced
NCN Non-Coincident Node
PDM Progressive Damage Modelling
PLINK Point LINK
RMIT Royal Melbourne Institute of Technology
SERR Strain Energy Release Rate
SEM Scanning Electron Microscopy
SHPB Split Hopkinson Pressure Bar
SIFT Strain Invariant Failure Theory
SPH Smooth Particle Hydrodynamics
TFP Tailored Fibre Pattern
VCCT Virtual Crack Closure Technique

xxx
The Airbus A380 is one of a range of new generation aircraft that utilise
a high percentage of composite materials within the structure

Reproduced courtesy of the GNU public license from WikiCommons


Chapter 1: Introduction

Composite materials are becoming more commonly used with every generation of
aircraft. The Boeing 787 is a prime example, which includes 50% composite material by
weight [1]. Increased understanding of the material properties and failure mechanisms
of composites are allowing them to be used more efficiently and with greater
confidence. Above and beyond the composite material properties, a detailed
understanding of the assembly and joining techniques of composite materials is crucial
to the design and repair of composite structures. Joining materials can create stress
concentrations and load discontinuities that dictate the overall strength of the
structure.

In general the joining of composites is broken into two methods, mechanical fastening
and bonding. The low efficiency of fastened joints in composite materials compared
with metal structures makes bolts or rivets a less than ideal choice for joining
composite parts. Bonded joints offer much greater joining efficiency than bolted joints,
but suffer some drawbacks which have limited their use in large scale composite
structures. The cost of producing, testing and maintaining a bonded joint is much
greater than for the equivalent bolted joint. Limitations of NDI techniques mean that
bonded structures are difficult to inspect, and there may be no way to determine that
the load carrying capacity of a joint has been compromised. Bolted joints, on the other
hand, are a robust positive connection that does not creep and can connect very thick
structures without the need for joint scarfing. Bolted joints may also be disassembled
throughout the life of the part if there is a requirement to remove and inspect
components. It is for these reasons that bolted joints are still employed heavily within
composite airframes for joining primary and secondary structural components.

Utilising bolted joints in composite materials to their full potential requires an


understanding of all the failure mechanisms that can occur within a bolted structure.
The most common failure modes considered are shown in Figure 1-1, although
combinations of these failure modes are also possible, as well as failure within the
bolt. Predicting these failures in composite materials is a complex problem. Joint
strengths are dependent on material properties of the bolt and laminate, geometry of

2
Chapter 1: Introduction

the joint, laminate thickness, stacking sequence and bolt pre-tightening. Joints can
often carry loads well in excess of the load at which damage initiates, which must be
understood if the full ultimate design strength of bolted joints is to be realised.

Net-Tension Shear Out Cleavage Bearing Pull-through

Figure 1-1: Bolted joint failure modes

It is an unfortunate consequence of the operating environment of aircraft structures


that they may occasionally be subjected to highly dynamic loading events, such as
gusts, impacts and crash landings. These loading events are characterised by high
loading rates, high kinetic energy and possibly loads well in excess of the static design
strengths. Modern aircraft must be designed to survive many of these dynamic load
cases, such as bird strike and gust loads. The integrity of the structure relies in the
integrity of the joints, so an appreciation of the effects of dynamic loading on bolted
composite joints is critical to gauge airframe survivability under dynamic loading.

The properties of carbon fibre composites change with strain rate, so it is likely that
the properties of bolted carbon fibre composite joints change also. Many studies have
been conducted on the behaviour of composite structures when loaded dynamically, a
typical case being bird strike of a leading edge structure. The specifics of the load
transfer at the bolted joints is generally overlooked in these studies, even though it has
been shown [2] that the bolted (or riveted) connections have a profound effect on the
behaviour of the structure. A few authors have addressed dynamic effects in bolted or

3
Chapter 1: Introduction

riveted joints but these studies have been purely experimental and limited to a select
few joint configurations. The two main studies involving composite bolted joints also
conflict with each other on key points regarding the behaviour of bolted joints in
composite structures, as discussed in Chapter 2, Section 6.

There is still a great deal of experimental investigation and numerical modelling


required to understand the behaviour of dynamically loaded bolted joints. The aims of
this thesis are to develop the knowledge of dynamic effects in bolted joints using
experimentation and numerical analysis.

A research program was defined by the Cooperative Research Centre for Advanced
Composite Structures (CRC-ACS), in conjunction with the German Aerospace Centre
(DLR), Pacific ESI, the Royal Melbourne Institute of Technology (RMIT) and UNSW. The
project aimed to investigate the behaviour of composite structures under impact and
improve the available modelling techniques for impact analysis. The project entitled,
Improved Analytical Techniques for High Strain Rate Impact Behaviour, included three
key objectives. The objectives were;

- To characterise through experiment the behaviour to failure of various


composite joint types (rivets, bonded, welded, scarf, etc.) under different
strain rates.

- To develop validated procedures for modelling composite joints to failure


in implicit and explicit finite element codes.

- To establish through experiment and analysis the impact behaviour of


various joint types with and without preload.

This thesis contributes a significant proportion of the research for the project, and as
such the goals of this thesis align with those of the research project. The two main
goals of the thesis research are to

Experimentally investigate the failure of bolted joints


in composite structures under dynamic loads

4
Chapter 1: Introduction

and to

Identify and develop modelling techniques for bolted joints that can
minimise the requirement for further experimental investigation.

In order to fulfil these goals, the following steps were undertaken;

- A comprehensive review of the available literature was conducted. State-


of-the-art experimental and numerical investigations of bolted joints were
examined and a research path was planned.

- The rate dependence of single fastener bolted joints over a range of


loading rates between quasi-static and 10 m/s was experimentally
investigated.

- Finite element modelling techniques for bolted composite joints were


identified and developed to a level capable of predicting the complex loads
and failure modes present in the experimental results.

- The sources of the discovered loading rate sensitivities were indentified


and explained with a combination of post-experimental analysis and finite
element modelling.

- Bolted joints in a simple composite structure under dynamic loading were


investigated and the relationship between the single fastener joint test
results and the multi-fastener structural test results was determined.

- The effectiveness of simplified efficient joint models for modelling bolted


joints in complex multi-fastener problems was investigated.

5
Chapter 1: Introduction

The relationship between the main chapters of this thesis is shown in Table 1-1.
Combined, the chapters present a broad study of bolted joint failures both
experimental and numerically, as well as at two different complexities.

Detailed General
Single Fastener Structural
Experimental
Joint Tests Testing
Detailed Joint Structural
Numerical
Modelling Modelling
Table 1-1: Relationship between thesis sections

A brief summary of each chapter is presented in the following sections.

The literature review summarises the state-of-the-art research that has been
published on testing of fastened joints to identify rate dependence and finite element
modelling of bolted joints in composite structures, which provides a solid basis for this
thesis.

A variety of tests were conducted to develop an understanding of the failure modes of


bolted joints both quasi-statically and dynamically. A set of test results are reported
that investigate the failure of a specific bolt-laminate combination over a range of test
speeds and loading configurations. Significant post-failure analysis of the specimens
was conducted including non-destructive CT-scanning of the failed parts.

This chapter presents detailed FE modelling of bolted joints. The models use 3-D
contact simulations to capture the localised failure mechanisms within a bolted joint in
a composite material. Different simulations are attempted for each series of tests
reported in the previous chapter. Implicit and explicit FE methods are attempted and
benefits and drawbacks of each are explained.

6
Chapter 1: Introduction

A test structure was developed and tested under quasi-static and dynamic loads. The
test structure included multiple fasteners and more complex loads than the two
simplified loading configurations used in the single fastener tests. The dynamic results
for the test structure shed light on the scalability of the single fastener test results. The
results also indicate the importance of the boundary conditions when testing bolted
joints.

The structures tested in the previous chapter were modelled using explicit FE analysis.
Simplified elements were used to represent the bolted joints, as would be required for
larger scale modelling of bolted composite structures. Calibration strategies for these
simplified joint models are discussed as well as some possible improvements to the
element formulations.

As with any research, some important questions have been answered but many more
have presented themselves. The main conclusions of the thesis are summarised here
with a forward view to remaining unresolved issues that could provide the basis for
future research.

Appendices are attached which include extra detail regarding:

- Strain Invariant Failure Theory (SIFT)

- CT scanning

- Hi-Lok fasteners

- Marc subroutines

- PLINK element investigation.

7
Chapter 1: Introduction

G.M. Pearce, A.F. Johnson, R.S. Thomson, and D.W. Kelly, High Strain-Rate Response of
Fastened Carbon Fibre Composite Joints in Composite Structures, in ECCOMAS
Thematic Conference on Mechanical Response of Composites. 2007. Porto, Portugal.

G.M. Pearce, A.F. Johnson, R.S. Thomson, and D.W. Kelly. Experimental Investigation of
a Simple Composite Structure including Loading Rate Effects. in CRC-ACS/Composites
Australia Conference and Exhibition. 2008. Melbourne.

A.F. Johnson, G.M. Pearce, R.S. Thomson, and D.W. Kelly. Experimental Techniques and
Data Acquisition for High-Rate Loading of Composite Joints and Structures. in
CompTest 2008: Composites Testing and Model Identification. 2008. Dayton, OH.

G.M. Pearce, L.P. Djukic, R.J. Wootton, and D.W. Kelly, Application of SIFT to micro-
cracking and lug failure, in 13th Australian International Aerospace Congress. 2009.
Melbourne.

G.M. Pearce, A.F. Johnson, R.S. Thomson, and D.W. Kelly, Experimental Investigation of
Dynamically Loaded Bolted Joints in Carbon Fibre Composite Structures. Applied
Composite Materials, 2009. Submitted.

G.M. Pearce, A.F. Johnson, R.S. Thomson, and D.W. Kelly, Numerical Investigation of
Dynamically Loaded Bolted Joints in Carbon Fibre Composite Structures. Applied
Composite Materials, 2009. Submitted.

G.M. Pearce, Modelling Riveted Joints in Composite Structures. DLR-IB 435 - 2008/4,
DLR - Institute for Structures and Design, 2008.

8
We are like dwarves sitting on the shoulders of giants. We see more, and things that
are more distant, than they did, not because our sight is superior or because we are
taller than they, but because they raise us up, and by their great stature add to ours.

- John of Salisbury
Chapter 2: Literature Review

This chapter will summarise the state of the art literature regarding bolted joints in
composite structures. Specific attention will be paid to research that promotes the
understanding of bolted joint failure mechanisms and modelling techniques which can
capture these failure mechanisms, both for quasi-static failure and dynamic failure.

The first half of the literature review considers only experimental observations, while
the second half discusses modelling of the observed phenomena. Separation of
subsections is based on the complexity of the system being discussed (pristine
composite > pin loaded hole > bolted joint) and the rate at which the tests were
conducted (quasi-static > dynamic).

Effective utilisation of bolted joints requires a thorough understanding of their failure


loads and failure modes. Most bolted joint failures occur within the composite material
phase, as this material is often locally much weaker than the metallic fastener, due to
the stress concentrations generated by drilling the bolt hole. Understanding bolt
failure modes starts with understanding composite failure modes. This section will
address the most common failure modes of composite materials and the importance
of these failure modes to bolted joint failure. Delamination is a matrix failure, but it
will be addressed in a separate section because it plays an important role in the failure
of bolted joints.

The carbon fibres are the primary load carrying phase in any well designed carbon fibre
composite laminate. To understand failure of the composite it is therefore crucial to
understand the failure modes of the carbon fibres.

Carbon fibres loaded in tension are highly elastic up until the point of failure. If the
fibre-matrix interface is strong, initial fibre failure will not lead to catastrophic failure
of the composite, as the load will be redistributed to the adjacent fibres. Due to the
statistical distribution of defects in the carbon fibres and the load redistribution effect,

10
Chapter 2: Literature Review

it is possible for local fibre failures to be distributed throughout the composite, as


shown in Figure 2-1.

Figure 2-1: Distribution of initial fibre damage [3]

Gross tensile failure occurs when the damage across one section coalesces to the point
where the section can no longer sustain the applied load (or strain) and the fibre
failures cascade and the structure collapses. For tough, well bonded matrices, the
failure surface is generally very well defined and crack paths may even be determined
by microscopic analysis, as shown in Figure 2-2.

Figure 2-2: Fibre tensile failure surface [4]

11
Chapter 2: Literature Review

If the fibre-matrix interface is weaker, significant fibre pull-out can occur, as shown in
Figure 2-3.

Figure 2-3: Fibre tensile failure including fibre pull-out [4]

Tensile fibre failures in bolted joints generally occur tangential to the bolt hole, in the
regions of maximum tensile stress. Tensile fibre failure drives catastrophic failure
modes such as net tension and cleavage modes. Fortunately the tensile failure of
carbon fibres is well understood.

Fibre compressive failure is a much more complicated problem than tensile failure. In
general, carbon fibres do not fail due to direct compression of the fibre, but rather
from local instabilities such as micro-buckling, kink-band formation and delamination.

Micro-buckling is local buckling of the fibre reinforcement. Micro-buckling has been


identified as a compressive failure initiation mechanism in unidirectional laminates [5-
7]. A simplified representation of compressive micro-buckling is shown in Figure 2-4a.

12
Chapter 2: Literature Review

a) b)

Figure 2-4: Simplified compressive response of composite lamina


showing a) micro-buckling and b) kink-band formation

With increased load, the fibres and surrounding matrix begin to fail and a kink-band
begins to form. The width of the kink band is the half wave length of the micro-buckle
[5]. An SEM image of a kink-band failure is shown in Figure 2-5.

Figure 2-5: Compressive kink-band failure [6]

During kink-band formation, the carbon fibres fail in nearly pure bending, and the clear
distinction between the tensile and compressive sides of the flexural failure was
observed by Purslow [4], and is shown in Figure 2-6.

13
Chapter 2: Literature Review

a)

b)

Figure 2-6: Flexural fibre failure during kink-band formation


showing a) tensile failure and b) compressive failure [4]

In some fibre-matrix systems, compressive shear failure is also possible, as shown in


Figure 2-7.

Figure 2-7: Fibre compressive failure in shear [8]

Compressive shear failure can also lead to kink-band formation [9], and there is debate
in the literature as to the primary initiation mechanism of compressive failure. Kink-
bands have also been observed with very few fibre failures and large matrix shear
deformation [10]. It is difficult to isolate the initiation of kink-band formation from
damage resulting from the presence of the kink-band itself.

Kink-bands generally propagate with a small angle relative to the fibre normal plane.
Narayanan and Schadler [11] showed that this kink angle is related to the interface
strength between the fibre and matrix and the degree of yielding that can occur. They
suggest that for a perfectly bonded, perfectly elastic matrix material, the kink band
would be normal to the fibres. If the matrix is allowed to yield slightly but is still
bonded to the fibre, then the kink band angle increases to a maximum of 45 and the

14
Chapter 2: Literature Review

kinking resembles a matrix transverse shear crack. If the interface between fibre and
matrix weak however, the kink band angle can exceed 45 and approach 90, which
could be considered a delamination failure.

Similar results were found by Ageorges, Friedrich and Ye [12] when investigating the
compressive strength of carbon fibres. Ageorges et al. used the Broughtman test,
where a single fibre is imbedded in matrix. The fibre-matrix specimen is then loaded in
compression in the direction of the fibre until failure. The intention is that fibre is
loaded in direct compression, but the Poisson's ratio mismatch between the fibre and
matrix will generate tensile normal forces at the fibre-matrix interface, as shown in
Figure 2-8.

Transverse
tension load

Front View Side View


Figure 2-8: Broughtman test schematic

Ageorges et al. found that two competing failure mechanisms occur in this test
depending on the strength of the fibre matrix interface. For well bonded interfaces the
fibre will fail in a compressive shear mode. For a weakly bonded interface the
transverse tensile forces will debond the matrix from the fibre.

The compressive strength of carbon fibre composites is one of the determining factors
in the bearing failure of bolted joints. In well controlled bearing tests conducted by Wu

15
Chapter 2: Literature Review

and Sun [13], two complimentary kink-bands were found between the central plane of
the laminate and the outer edges of the laminate at approximately 45 to the loading
direction.

Bearing Surface

Figure 2-9: Complimentary kink-bands in pin bearing [13]

Compressive fibre failures in bolted joints generally occur normal to the bolt hole, in
the regions of maximum compressive stress.

In summary, the compressive failure of fibres is significantly more complicated than


the tensile failure. Carbon fibres generally do not fail in direct compression, but lose
integrity due to local instabilities brought about by local defects, micro-buckling, shear
failure of the fibre and eventual kink band formation. This process is dependent on
many different factors, and is still not very well understood for all but the most
controlled conditions. Compressive fibre failure is a critical mechanism for
characterising bearing failure of composite laminates.

The matrix in a composite material is generally intended to transfer load effectively


between fibres and protect the brittle fibres from damage. Well designed laminates
should avoid having a heavily loaded matrix phase as the matrix is generally much
weaker than the fibres. Under the complex loading conditions experienced within a
bolted joint however, the matrix can experience significant loads, regardless of the
layup configuration.

In a composite, the matrix behaves quite differently than in its neat form, due to both
the restraint provided by the fibres and the high thermal residual strains that are

16
Chapter 2: Literature Review

imposed during the cooling process after cure. Matrix failure in composites must be
understood in situ and not based on failures of neat resin tests.

In bolted joints, matrix failure contributes to, if not dominates, many of the primary
failure modes. An understanding of matrix failures is therefore critical to
understanding bolted joint failures.

The reader is reminded that for the coming descriptions of matrix failure, a clear
distinction between tensile, compressive and shear failures is not possible as the
matrix is an isotropic or near isotropic material. Intermediate failure modes between
tension, shear and compression failures are all possible.

Matrix materials in carbon fibre composites are generally two to three orders of
magnitude weaker in tension than the fibres that they surround. Gross tensile matrix
cracking generally only occurs in a direction in which the fibre reinforcement has
already failed or was not there to begin with. Failure in the transverse direction will
occur parallel to the fibres, but may change plane occasionally by propagating through
a few fibres, as shown in Figure 2-10.

Figure 2-10: Gross matrix tensile failure [14]

A much more common although less catastrophic tensile matrix failure mode is micro-
cracking which can occur in plies with fibres oriented transverse to the main loading
direction. Most epoxy matrix materials have a much higher strain to failure than
carbon fibres, however the thermal residual strains in the matrix can cause the matrix
to fail well before the load carrying fibres do. A diagram of matrix micro-cracking is
shown in Figure 2-11.

17
Chapter 2: Literature Review

Figure 2-11: Matrix tensile micro-cracking

Most bolted joints use layups that are very close to quasi-isotropic, and strain to failure
of the matrix is greater than the strain to failure of the fibres. Catastrophic tensile
matrix cracking therefore generally only occurs after the main fibre phase has already
failed. The most important form of tensile matrix failure for bolted joints is
delamination, which will be discussed in a separate section.

Homogenous non-porous materials have exceeding high strengths under pure tri-axial
hydrostatic compression. When loaded in uniaxial or biaxial compression, polymer
matrices will fail via transverse shear failures, as described by Puck and Schrmann
[15].

Figure 2-12: Transverse compressive shear crack

Compressive shear matrix failure is an important failure mode for bolted joints. The
strength of the matrix in compressive shear plays an important role in the overall
compressive response of the composite, which determines the bearing strength of the
joint.

18
Chapter 2: Literature Review

Matrix shear failure is generally not as 'brittle' as other composite failure mechanisms.
Significant matrix nonlinearity has been observed for non crimp fabric [16], woven
fabric [17] and braided [18] thermoset matrix composites. One example of this non-
linearity for a cyclic shear test on a braided composite is shown in Figure 2-13.

Figure 2-13: Matrix non-linearity during cyclic shear testing of 45 braided composite [18]

For thermoset matrix composites, this non-linearity is generally attributed to


development of discrete shear damage within the matrix and delamination rather than
matrix plasticity. The mechanism of this damage formation is shown in Figure 2-14.

Figure 2-14: Localised discrete matrix damage leading to non-linear shear response [16]

Matrix shear failures are an important mechanism for failure in bolted joints. The load
distribution around the hole is a complex mix of tensile and compressive loads both in-
plane and through the thickness. As a result there are many regions where shear
failure is the dominant failure mechanism.

19
Chapter 2: Literature Review

Delamination is matrix failure which leads to internal separation of plies within the
composite laminate. Delamination can arise from a number of sources including;
manufacturing defects, edge effects, impact of foreign bodies with the composite and
through thickness stresses in thick curved structures. Delamination is a significant
failure mode in composite materials, and one of the most difficult to characterise.
Delamination is an insidious failure mode as it is often embedded within the composite
laminate and not detectable by visual inspection.

Physically, there are two modes of delamination, tensile (opening) and shear (sliding)
delamination, although three modes are defined for fracture mechanics
characterisation of delamination. In brittle composite matrices there is generally a
clear distinction between tensile and shear delamination failure surfaces.

Tensile delamination is characterised by regions of smooth tensile matrix failure and


river or chevron matrix markings. These two features can be seen in Figure 2-15.

Figure 2-15: Tensile delamination of epoxy composite showing


smooth matrix fracture and river markings [19]

Shear delamination is characterised by matrix cusps or hackles, which are small


sections of matrix that have failed ahead of the crack tip and are peeled up from the
surface as the crack front passes. These cusps can be seen clearly in Figure 2-16.

20
Chapter 2: Literature Review

Figure 2-16: Shear delamination of composite parallel


to the fibres showing typical shear cusps [14]

Shear delamination in a composite can occur via two general modes. The first mode is
parallel to the fibres and the second mode is normal to the fibres, as shown in Figure
2-17. Failure may also occur as a combination of the two modes. The two modes have
slightly different failure surfaces due to the depth of free matrix available to form the
cusps.

a)

b)

Figure 2-17: Matrix shear failure modes in a fibre composite material


showing a) fibre parallel and b) fibre normal modes

Experimental results [20-22] have shown that delamination failure plays a significant
role in the failure of composite bolted joints. Toth-Antal [23] observed significant
delamination failure within a composite laminate failed under bearing load and imaged
the internal damage with non-destructive CT scans. A section plane through a failed
pin loaded specimen is shown in Figure 2-18. The section shows significant
delamination failure between many internal plies.

21
Chapter 2: Literature Review

Section Plane

Figure 2-18: CT scan of failed pin loaded bearing specimen [23]

It has been found that bolt or washer clamping pressure significantly improves the
bearing strength of pin-loaded composite laminates [22, 24-25], which has been
primarily attributed to the delaying of initial delamination failures.

Many materials exhibit significant changes when loaded at high rates. The primary goal
of this thesis is to understand the failure of bolted joints when loaded at high rates, so
the dynamic behaviour of the composite materials involved is critical.

Sierakowski [26] conducted an extensive review into the effects of high rate loading on
composite materials. It was found that, in general, most materials exhibit a critical
strain rate at which a transition between rate insensitive and rate sensitive behaviour
occurs. Above this critical rate, material strengths increase linearly on a log(strain rate)
scale. The critical strain rate and the slope of the linear region in stress/log-strain space
determine the extent to which a material can be considered rate sensitive. Sierakowski
also stressed the difficulties in generating a constitutive model of any material over a
complete range of strain rates due to incompatibilities between the testing
methodologies at different strain rates.

This section will address observed changes in the material properties of carbon fibre
composite materials when loaded dynamically. It will also include a section describing
rate effects in Ti-6Al-4V, the metal used for the HiLok fasteners used in this testing.

22
Chapter 2: Literature Review

Carbon fibres are generally considered rate insensitive over the range of strain rates
that are important to this research. Daniel et al. (cited in [26]) found that the tensile
strengths of AS carbon fibre composites were almost insensitive to strain rates of up to
300/s using an expanding ring test (ERT). The Young's Modulus increased by only 30%
over the same range of loading rates. Zhou, Jiang and Xia [27] tested T300 fibre
bundles in tension using an impact test (IT) and found no rate sensitivity at strain rates
up to 1300/s.

Compressive failure of carbon fibres does show some significant rate dependence
according to some authors, but considering compressive fibre failure is a matrix
dominated phenomenon, this is most likely due to changes in the matrix with strain
rate. There is also disagreement in the literature as to the effect of strain rate on the
compressive fibre failure of carbon fibre composites. Weeks and Sun [28] found no
evidence of rate dependence in the compressive failure of AS4/PEEK composite at
strain rates of up to 100/s using a split Hopkinson pressure bar (SHPB). Hosur et al. [29]
used a SHPB to test unidirectional carbon/epoxy composite (PANEX33/DA4518U) in
compression. They found that the composite exhibited a slight decrease in stiffness
and an 18% increase in strength over a range of compressive strain rates between 82/s
and 817/s. Newill and Vinson found that the compressive strength of unidirectional
carbon/epoxy (AS4/3501-6) increased slightly over a range of strain rates between
250/s and 600/s, as shown in Figure 2-19.

0.8
Load (Gpa)

0.6

0.4
250/s
0.2 400/s
600/s
0
0 0.01 0.02 0.03 0.04 0.05
Strain
Figure 2-19: Compressive response of AS4/3501-6 unidirectional composite [30]

23
Chapter 2: Literature Review

In general the response of carbon fibres in epoxy resin loaded in compression could be
considered rate sensitive. The rate sensitivities seem to depend heavily of the type of
test used specifically the way the load is transferred between the test rig and the
specimen [31]. Overall, the rate sensitivity is quite mild, and only moderate changes in
strength and stiffness were found over many orders of magnitude of strain rates.

Composite matrix materials vary significantly in their chemistry and physical


properties, but in general are very sensitive to loading rate. Strain rate effects in matrix
materials are sometimes measured from neat resin [32], but are generally inferred
from transverse loading of unidirectional laminates or direct loading of cross-ply
laminates.

Sierakowksi [26] reports expanding ring tests of AS/SP288 carbon/epoxy composite


loading transverse to the fibres. It was found that the transverse modulus increased by
up to 400% while the transverse strength increased by up to 300% with strain rates of
up to 300/s. Similar results were found for compression loading of the same material,
but the compressive strength only increased by 50%. Bing and Sun [31] found a
decrease in plastic strain and an increase in strength for off-axis compression tests of
AS4/3501-6 carbon/epoxy using a SHPB with strain rates up to 250/s. Staab and Gilat
[33] tested high angle glass-epoxy laminates in compression and found that the
strength increased by around 80% when loaded at a strain rate of approximately
1000/s. Hosur et al. [29] loaded UD carbon/epoxy specimens transversely with a SHPB
at strain rates of up to 817/s. They found that the modulus increased significantly with
strain rate and the strength increased by around 30% over static values.

Investigation of failed tensile specimens has shown that matrix damage is much more
localised for dynamic loading rates. Fitoussi et al. [34] developed a means to stop
dynamic tests at a fixed strain level. SEM images were taken of matrix damage of two
different specimens; one tested quasi-statically and one tested dynamically at 25/s.
The results can be seen in Figure 2-20. It was found that the matrix damage in the
quasi-static tests is quite diffuse, while in the dynamic test, localised cracking is clearly
visible.

24
Chapter 2: Literature Review

a) b)

Figure 2-20: Matrix damage at comparison between quasi-static and dynamic tests
loaded to 4.9% strain showing a) quasi-static test and b) dynamic test

It is commonly accepted within the literature that epoxy matrix materials are rate
sensitive. Most papers report an increase in strength and modulus with strain rate for
transversely loaded UD specimens of high angle cross-ply laminates.

In general the dynamic response of carbon fibre/epoxy laminates will depend on the
percentage of off-axis plies. The more plies with fibres oriented away from the loading
direction and the higher the angle of orientation, the more rate sensitive the laminate
will be, for both tensile and compression loadings.

There are conflicting reports in the literature as to the effect of crack speed or loading
rate on the Mode I and II fracture toughness of the epoxy matrix. Due to the large
number of epoxy matrix systems available, it is difficult to find specific data about
Cycom 970 used in this research, but a number of researchers have investigated the
rate sensitivity of epoxy matrix and adhesive materials. Elder, Dorsamy and Rheinfurth
[35] investigated the fracture toughness of FM300 epoxy adhesive in thin bonded
joints and found that the fracture toughness drops with increased loading rate. Sun et
al. [36] observed a transition from high toughness to low toughness crack behaviour in
an epoxy bonded joint at a crack speed of 1 m/s. The two failure modes had
significantly different failure surfaces, as shown in Figure 2-21.

25
Chapter 2: Literature Review

Figure 2-21: Fracture surfaces for quasi-static and dynamic crack growth [36]

Huang et al. [37] and Dwivedi and Espinosa [38] discuss a critical crack speed at which
the Mode II fracture propagation exceeds the Raleigh wave speed of the material and
creates a sharp change in the fracture toughness at this speed.

Conversely, Sun and Han [39] found that the fracture toughness for Cycom 977-3 was
virtually unaffected by loading rate up to crack speeds of 1000 m/s.

It has been observed by many authors over a range of different crack velocities and
test speeds that delamination is dependent or loading rate or crack speed.
Unfortunately there is significant disagreement about what the loading rate sensitivity
actually is. As there is no specific agreement in the literature as to the effect of loading
rate on delamination, the effect of loading rate in this research will only be conducted
via numerical parameter studies, inferring possible dynamic delamination effects from
trends that agree with experimental evidence.

The titanium used for the bolts in this research is an alloy with 6% aluminium and 4%
vanadium (Ti-6Al-4V). This is a very common titanium alloy and there is a great deal of
research into the machining and cutting of this material. Machining simulations require
complex material models which include loading rate effects, thermal effects and other
effects such as strain softening or hardening. One of the most commonly accepted
models for metals under high rate loading is the Johnson-Cook model [40]. The
Johnson-Cook law is an empirical relationship relating von Mises flow stress to strain,
strain rate and temperature. The Johnson-Cook model has been modified by many

26
Chapter 2: Literature Review

authors to account for various phenomenological effects, but the rate sensitive part is
generally written as

 
   A  B n  1  C ln
...(2.1)
0

where

  von Mises flow stress


 von Mises strain
 Strain rate
0  Reference strain rate
A,B,C,n  Experimentally determined constants

The Johnson-Cook model agrees with Sierakowski in that the material strength
increases linearly in stress-log(strain rate) space. Calamaz, Coupard and Girot [41] used
a modified Johnson-Cook law very successfully to model chip formation in the
machining of Ti-6Al-4V.

All physical processes that involve energy transfer have associated inefficiencies that
generate heat. Friction between surfaces can also generate significant thermal energy.
In quasi-static failure tests the rate at which this energy is generated is generally much
lower than the rate at which it can be dissipated from the specimen. This type of test is
known as isothermal, because the temperature of the test specimen remains nearly
constant, controlled by the temperature of the surrounding environment.

If a specimen is loaded dynamically, the rate at which heat is generated within the
specimen can be much higher than the rate at which it can be dissipated. In this case
the test is known as adiabatic, which means that for the duration of the test there is
no thermal energy lost to the environment and all thermal energy generated is
trapped within the specimen. Adiabatic test conditions can lead to extremely high local
temperatures around critically loaded regions of the test specimen.

27
Chapter 2: Literature Review

The transition between isothermal behaviour and adiabatic behaviour depends on the
materials used, specimen geometry and many other factors. As a general rule, if
materials are loaded above a strain rate of 1/s, then adiabatic heating must be
considered [26].

In general, polymer material properties degrade with increased temperature [42-43].


Matrix dominated properties of composites made with F593 [44], Cycom 970 [45] and
Cycom 977-3 [46] reduce with temperature. Representative temperature behaviour for
a neat polymer resin is shown in Figure 2-22.
Equivalent Stress

Temperature

Equivalent Strain

Figure 2-22: Temperature effects in neat polymer resins

If increased loading rate leads to adiabatic test conditions then temperature


dependence is linked to rate dependence. It is shown in Figure 2-19 that increasing
loading rate increases the compressive strength of a carbon-epoxy composite.
Temperature has the opposite effect; compressive strength of a carbon-epoxy
composite reduces with increasing temperature.

Therefore, there are competing mechanisms at work when a composite material is


loaded dynamically. For dynamic compression strength, increasing the loading rate
should increase the material strength, as per Figure 2-19. However, if adiabatic heating
occurs at the crack front due to friction or other physical effects, then the associated
temperature increase acts to reduce the compressive strength. The competition
between these two effects is dependent on specimen geometry and friction

28
Chapter 2: Literature Review

coefficients. The two competing mechanisms begin to explain why papers examining
the rate sensitivity of very similar composite materials can come to quite different
conclusions.

Temperature was not measured directly for the experiments reported in this thesis as
the capability to do so was not available. In some cases the temperature effects could
be inferred from the failure loads and damage mechanisms, and will be discussed
where appropriate.

Load transfer in a bolted joint is a complicated process. Failure can occur by a number
of different mechanisms. The simplest first approximation to a bolted joint is a plate
with a hole loaded by a stiff pin. This is often referred to as a pin loaded bearing test. A
simplified schematic of a pin loaded bearing test is shown in Figure 2-23.

P P/2 P/2

Figure 2-23: Schematic of a pin loaded bearing test

Load applied by the pin is reacted by a compression load in the composite, and to a
lesser degree, friction load between the pin and the composite. The load is carried
back past the hole as a tension load, as shown in the simplified FBD of the load
reaction in Figure 2-24.

29
Chapter 2: Literature Review

P
P/2 P/2
Figure 2-24: Simplified FBD of reaction with a section at the net tension plane

The load distribution within the plate can be visualised as load paths [47], which are
contours analogous to stream lines in fluid flow. The principal stress trajectories for an
isotropic elastic plate loaded in bearing are shown in Figure 2-25. Within a typical pin
loaded bearing specimen there are regions of high compression, high tension and high
shear stresses in both the in-plane and through thickness directions.

Figure 2-25: Load paths for an elastic isotropic plate loaded in bearing
showing principal compression (red) and principal tension (blue) paths [47]

As there is such a variation in the local load distribution around the hole, the layup is
generally chosen to be quasi-isotropic [3] so that fibres are oriented in enough
directions to carry the load effectively.

Three primary failure mechanisms exist for pin-loaded holes; net tension failure, shear
out failure and bearing failure. Shear out failure can be avoided by selecting a large
enough edge to diameter ratio (e/d) and so it is generally not studied in great depth.
Net tension failures can be avoided by selecting a large width to diameter ratio (w/d),
but this comes at the cost of structural efficiency because less fasteners are available
per unit length of the joint to carry the load. Net tension failures generally lead to the
highest joint efficiencies [3], but are a catastrophic failure mode, so are avoided in

30
Chapter 2: Literature Review

many applications. Net tension failures are also quite well understood because the
primary failure mechanism within the composite is tensile fibre failure. Bearing failures
are the most difficult failures to characterise, as they involve complex failure
mechanisms within the composite, such as compressive fibre failure, delamination and
through thickness shear failure.

Many composite failure mechanisms can contribute to bearing failure in composites.


The three main failure modes; compressive fibre failure, compressive matrix shear
cracking and delamination can all be seen in Figure 2-26. The failure mechanism that
initiates failure depends the laminate, the matrix shear strength, through thickness
restraint and the amount of friction between the pin and the laminate.

Matrix shear Bearing Compressive


cracking surface fibre failure

Delamination
Figure 2-26: Optical micrograph of failed bearing specimen [23]

The understanding of failure initiation is difficult because many factors that affect the
failure of pin-loaded holes are test related rather than material property related.
Prabhakaran and Naik [48] showed that increasing the friction coefficient between the
pin and the hole increased the contact angle significantly, which provides more area
for carrying the bearing load. Kelly and Hallstrm [49] showed that small clearances
between the pin and the hole could have a significant effect on the failure initiation
load for unclamped pin loaded specimens but had little influence on the ultimate loads
or the failure loads of specimens that were clamped with a washer. Many authors [22,
24-25, 49] have shown that through thickness restraint around the hole significantly
improves the bearing strength of the specimens, as shown in Figure 2-27.

31
Chapter 2: Literature Review

Figure 2-27: Effect of clamping the laminate through the thickness on the
bearing strength of a pin-loaded hole [49]

Bearing failure is generally a very stable failure mechanism that can absorb a great deal
of energy. Camanho and Matthews [50] present experimental results showing the
peak failure load is maintained during progressive crushing of the laminate for at least
twice the displacement to ultimate load. Mikulik, Dutton and Thomson [51] conducted
controlled progressive bearing tests well beyond the initiation of failure and found that
the laminate still carries significant load, as shown in Figure 2-28. In some cases this
progressive bearing load was similar in magnitude to the peak bearing load of the
specimen.

a)

b) 25
Q45-1
20
Q45-2
Load (kN)

15
10
5
0
0 5 10 15 20
Displacement (mm)

Figure 2-28: Progressive pin bearing in a quasi-isotropic fabric laminate


showing a) specimen showing large bearing damage and b) load results [51]

32
Chapter 2: Literature Review

Pin loaded holes do not capture all of the failure mechanisms of bolted joints. The pin
loaded hole test is effectively a two body problem, where the pin is part of the test
fixture, and the displacement and load at the pin is well known.

A real bolted joint test is a multi-body problem, where the load transmission between
the two (or more) composite parts is not a well controlled as in the pin bearing case.
Bolts in bolted joints carry not only bearing (shear) loads, but pull-through (tension)
loads as well as moments, especially in single lap joint configurations.

Failure in bolted joints is more complex than for pin-loaded holes. The general set of
failures for pin-loaded holes is expanded to include bolt-specific failures such as bolt
pull-through and bolt failure, as shown in Figure 2-29. Joints can fail in a combination
of these modes as well. Almost all single lap bolted joint failures will have some degree
of bearing damage due to the high contact pressures that exist at the shear plane.

Shear-out Failure Pull-Through Failure


Tension Failure

Cleavage Failure Bearing Failure Bolt Failure

Figure 2-29: Joint failure modes [52]

This section will address experimental studies that have been conducted into the
failure of bolted composite joints. The section will be broken into two subsections; the
first discussing failure of bolted joints when loaded in the shear (bearing) direction and
the other discussing failure when the joints are loaded in the tensile (normal)

33
Chapter 2: Literature Review

direction. As yet a paper has not been found by the author which addresses a
combined state of loading for a bolted joint.

A double-lap bolted joint can be approximated with some degree of accuracy to a pin-
loaded hole. A simplified exploded free body diagram (FBD) for a double-lap bolted
joint is shown in Figure 2-30. The FBD ignores friction and some higher order bending
effects, but is reasonably accurate as a first approximation.

Figure 2-30: Simplified FBD for a double-lap bolted joint

If a similar approach is taken for a single-lap bolted joint then the bolt is no longer in
equilibrium, and is experiencing a moment that is proportional to the distance
between the two mid-planes of the joined laminates, as shown in Figure 2-31.

Figure 2-31: Simplified FBD for a single-lap bolted joint without equilibrium

In practice, the unbalanced loads are reacted by contact pressures at the bolt head and
nut and the through-thickness stress profile changes so that the contact pressures are
highest toward the central shear plane of the joint. Unless restrained, the bolt will
generally experience significant secondary bending, where the joined plates will bend
out of plane, as shown in Figure 2-32.

34
Chapter 2: Literature Review

Figure 2-32: Load alignment and secondary bending

For the description of contact pressures on the hole surface and stresses in the
composite material, a bolt based cylindrical coordinate system is used. The origin for a
single lap joint is normally the intercept of the bolt axis with the plane between the
two composite plates. The z-axis is coincident with the bolt axis, the radius is measured
from the central axis of the bolt and direction is measured anti-clockwise from the
bearing plane, as shown in Figure 2-33.

B B

S S

Figure 2-33: Bolt hole coordinate system showing


bearing plane (B-B) and shear plane (S-S)

McCarthy and McCarthy [53] studied the stresses around the bolt hole with FE analysis
with an elastic bolt and homogenised laminate properties. It can be seen in Figure 2-34
that the stresses in the laminate are much higher at the central shear plane of the
joint. The outer plies are carrying almost no load in the radial or tangential direction.
Initial failure in a single-lap joint is initially much more localised than in the equivalent
double-lap joint, because of the unequally shared load through the thickness of the
laminate.

35
Chapter 2: Literature Review

Figure 2-34: Laminate stresses for a single-lap bolted joint connecting homogeneous
orthotropic plates showing a) radial stress and b) tangential stress [53]

An study of bolted joints in composites was conducted within the European research
program Bolted Joints in Composite Aircraft Structures (BOJCAS), which was
coordinated by the University of Limerick. Experimental papers [54-58] have been
published by authors at Limerick which detail many aspects of bolted joint failure
including the influence of bolt-hole clearance.

Friction plays an important role in bolted joints if the bolts are tightened to provide
clamping pressure. In a single- or double-lap bolted joint, the connected plates are
moving relative to one-another when load is applied to the joint. Friction between the
plates can account for a significant proportion of the total load carried by the joint [3],
especially in the double-lap case where there are two surfaces to carry the friction
load.

A typical failure in a single-lap composite bolted joint includes features of many of the
failure modes presented in Figure 2-29. Failures will generally have some initial bearing
damage in the laminates at the intersection of the shear and bearing planes where the
contact stresses are highest. The progression of the failure however is very dependent
on laminate lay-up, fastener type and the geometry of the joint. A bearing-dominated
failure mode is shown in Figure 2-35. Note the high amount of bolt rotation and
damage associated with the fastener head.

36
Chapter 2: Literature Review

Figure 2-35: Bearing damage in single lap bolted joints


showing a) universal head fastener test
and b) countersunk head fastener test [59]

Pull-through failure of bolted composite joints is a critical failure mechanism that is


often overlooked. Understanding pull-through failure in composites is critical, as the
weak through-thickness properties of composite materials makes them especially
susceptible to out-of-plane failure modes. The push toward thin skinned post-buckling
structures means that bolted joints now, more than ever, are being used to carry
significant tensile loads.

There has only recently been a standard testing methodology adopted by the ASTM
[60], so any available test data on the subject is using slightly varied methodologies.
Banbury and Kelly [21] conducted the most significant test program and failure
analysis of pull-through failure for carbon fibre composites found to date. They found
that pull-through failure was less dependent on geometry than bearing failure, but
there were still some variations in failure mode with different thickness materials.
Thick specimens were found to fail via through-thickness shear cracks, radiating from
the fastener head in a conical zone, as shown in Figure 2-36. This through thickness
shear damage was accompanied by significant, but localised, delamination within the
composite. For thinner specimens, failure generally initiated as tensile fibre failure in
the rear face of the composite.

37
Chapter 2: Literature Review

Figure 2-36: Conical damage zone during fastener pull-through [21]

A typical load-displacement curve for the pull-through loading is shown in Figure 2-37.
Load

Displacement
Figure 2-37: Typical load-displacement curve for a pull-through test [21]

Gunnion et al. [59] and Krber [61] conducted a series of pull-through tests for bolts
pulling through woven carbon fibre fabric. These authors found similar results to
Banbury but none of the tests experienced the load drop after initial failure observed
by Banbury. All of the specimens reported in Krber initially failed by fibre tensile
failure rather than matrix through thickness shear failure.

One important experimental parallel that was not recognised by these authors was the
similarity between a pull-through test and a three-point bend test or short beam shear
test. Assuming that the area under the bolt is well clamped, a pull-through test is much
like an axi-symmetric three point bend test. For the thin specimens, bending loads are
critical, as shown in Figure 2-38a, which promotes fibre tensile failure on the bottom
surface. For thick specimens, shear loads are critical, as shown in Figure 2-38b, which
promotes through thickness shear cracking and Mode II delamination.

38
Chapter 2: Literature Review

a)

Bending Shear
(Critical)

b)

Bending Shear
(Critical)
Figure 2-38: Bending and shear stresses in pull-through tests
showing a) thin and b) thick tests

Most studies of bolted joints have focused on the quasi-static behaviour of the joint. It
is an unfortunate consequence of the operating environment of aircraft that many
critical load cases involve impact and crash. These loading events are characterised by
high loading rates, high kinetic energy and possibly loads well in excess of the static
design strengths. The properties of many materials change with loading rate, as does
the frictional interaction between surfaces. It is therefore likely that the behaviour of
bolted composite joints varies significantly with varied loading rates.

Ger, Kawata and Itabashi [62] tested a number of hybrid carbon-kevlar and carbon
composite joints dynamically (6-7 m/s) and quasi-statically under pin-loaded, single-lap
and double-lap joint configurations. It was found that for all joint configurations the
stiffness of the joints increased significantly with loading rate. Contrastingly, the total
energy absorption of the joint decreased significantly in the dynamic tests. The
variation of peak load carrying capacity with loading rate depended on the joint type.
The pin loaded joints were significantly weaker at higher loading rates while the single
lap-joints were not significantly affected. Conversely, the double lap joint

39
Chapter 2: Literature Review

configurations carried more load at high loading rates. No explicit test curves were
given for carbon composite laminates but a comparison between static and dynamic
test curves for a single lap bolted joint with a hybrid laminate is shown in Figure 2-39.

Figure 2-39: Dynamic-static comparison for a single lap hybrid bolted joint [62]

Li, Mines and Birch [63] tested a number of different carbon fibre composite joint
configurations subjected to bearing load at rates between quasi-static and 8 m/s, and
found results that contradicted Ger et al. It was found that for the majority of
specimens tested that the stiffness and strength of the joint only increased slightly
with loading rate but there was a significant change in failure mode at higher rates (4-8
m/s) which generally resulted in increased energy absorption. This effect was most
pronounced for the protruding head bolts. This is in direct contradiction to Ger et al,
who found a sharp drop in energy absorption with loading rate. A comparison between
quasi-static and dynamic joint tests is given in Figure 2-40 for a single lap bolted joint.
It can be seen that the peak loads are similar however the energy for the dynamic tests
is larger, especially for the protruding head test.

40
Chapter 2: Literature Review

a)

b)

Quasi-static Tests Dynamic Tests

Figure 2-40: Comparison between quasi-static and dynamic single lap bearing tests
showing a) protruding head fasteners and b) countersunk head fasteners [63]

Ger et al. and Li et al. directly conflict with one another on some key aspects of the
high rate behaviour of carbon fibre composite bolted joints. No other papers could be
found that specifically addressed this issue for bolted composite materials. Birch and
Alves [64] conducted similar dynamic testing for two types of metallic joints;
spotwelded and bolted joints. The spotwelded metallic joints showed significant rate
sensitivity in the failure modes but the increase in energy absorption with loading rate
was small. The bolted joints on the other hand experienced no change in failure mode
with increased loading rate, but experienced significant changes in energy absorption
and peak load. The authors of this paper suggest that changes in failure mode may
hide the underlying strain rate sensitivity of the joint materials. In bolted joints in
composite materials, different failure modes are driven by geometric effects as well as

41
Chapter 2: Literature Review

different materials strengths. This suggests that rate sensitivity is likely to be a function
of both material properties and geometric effects.

In all cases above, the loading was limited to an ideal bearing load, carried in the shear
plane of the joint. Loading of joints is not always ideal and in-plane. Impact of foreign
objects or propagation of bending waves through the structure can force the joint to
carry significant pull-out loads in addition to the usual bearing loads. No
investigation can be found into the effects of dynamic pull-through loads or mixed
mode loading.

The preceding sections have all dealt with experimental observations of failure in
composites and bolted composite joints. The following sections will deal with
modelling these failures. Effective modelling of bolted joint failure can reduce the
requirements for extensive joint testing and aid in the design of novel bolted joint
configurations.

The understanding of experimental joint failures relies primarily on understanding


failure of composite materials. Consequently, the effective modelling of bolted joint
failure relies primarily on effectively modelling failure of composite materials. This
section will discuss composite failure onset theories that are commonly used in the
literature, or those that have been identified as promising failure theories that deserve
more investigation.

The section will be broken into five main subsections that each address a broad class of
failure theory. The first four relate to intralaminar failure modes while the fifth section
addresses those failure theories that have been proposed specifically for interlaminar
(delamination) failure.

Maximum value failure theories decouple all failures into separate modes, generally by
comparing each component of a tensor to a previously determined maximum
allowable value for that component. Separate allowables are often defined for tensile
and compressive directions.

42
Chapter 2: Literature Review

The general form of a maximum value failure theory is shown in Equation 2.2.

x x x x x x 
f  max 1t , 1c , 2t , 2c ,..., nt , nc
...(2.2)
X X X X X 3 X 3

1 1 2 2

where

xi = Component of tensor x in the i direction


Xit = Allowable tensile component of tensor x in the i direction
Xic = Allowable compressive component of tensor x in the i direction

The maximum stress theory states that failure has occurred when any of the
components of the stress tensor (ij) within a ply exceed the maximum allowable value
in that direction. The stress tensor is generally oriented such that the 1 direction is
parallel to the fibres while the 2 direction is normal to the fibres and in the plane of
the laminate. Failure occurs when Equation 2.3 is satisfied.

         
max 11 , 22 , 33 , 11 , 22 , 33 , 12 , 23 , 31
X Y Z X Y

 1 ...(2.3)
t t t c c Zc S12 S23 S31

where

Xt = Tensile failure stress in the X direction


Yt = Tensile failure stress in the Y direction
Zt = Tensile failure stress in the Z direction
Xc = Compressive failure stress in the X direction
Yc = Compressive failure stress in the Y direction
Zc = Compressive failure stress in the Z direction
Sij = Shear failure stress in the ij plane

It is assumed that failures in the 1 direction are fibre related and all other failures are
matrix related.

It is common to plot failure theories on a biaxial 1-2 stress space. A locus of all failure
points defines a failure surface. For all points on or outside the failure surface the

43
Chapter 2: Literature Review

composite is assumed failed. The failure surface for maximum stress theory is shown in
Figure 2-41.

22

Yt

Xc Xt 11

Yc

Figure 2-41: Maximum stress theory failure surface in biaxial stress space

Maximum strain theory has the same form as maximum stress theory except that the
components of the strain tensor (ij) are used instead of the stress tensor. Failure
occurs when Equation 2.4 is satisfied.

 
max 11 , 22 , 33 , 11 , 22 , 33 , 12 , 23 , 31
   

 1 ...(2.4)
1t 2t 3t 1c  2c  3c 12  23  31

where

1t = Tensile failure strain in the X direction


2t = Tensile failure strain in the Y direction
3t = Tensile failure strain in the Z direction
1c = Compressive failure strain in the X direction
2c = Compressive failure strain in the Y direction
3c = Compressive failure strain in the Z direction
ij = Shear failure strain in the ij plane

44
Chapter 2: Literature Review

Maximum value failure theories assume that failure in each direction is only
dependent on loads/strains in that direction. This assumption does not account for any
interaction between the loads in different directions, which becomes especially
erroneous for complex load states involving combinations of transverse and shear
loads. Despite the problems with this type of theory, they are very easy to implement
in finite element codes and still used as a reasonable first approximation to failure.

Interactive failure theories account for interaction between loads in different


directions by combining all the components of the stress tensor and allowables into a
single equation. These equations can be linear, quadratic or both, with a general form
given by Equation 2.5.

6 6 6
f 2   Fi  i   Fij  i  j ...(2.5)
i 1 j 1 i 1

where

Fi = Weighting coefficient for stress component i


Fij = Weighting coefficient for tensor stress product ij

The weighting coefficients Fi and Fij are related to ultimate strengths measured from
uniaxial tests and shear tests. The Fi terms have the units [1/Stress] and the Fij terms
have the units [1/Stress2] such that Equation 2.5 is a sum of fractional failure terms. A
consequence of the interactive nature of these failure theories is that all loads
interact, so that load in any direction influences failure in all other directions. The two
most common interactive failure theories are Hill-Tsai and Tsai-Wu theories, but many
others have been proposed.

Tsai-Hill theory is an extension of Hill's theory of anisotropic materials to


heterogeneous composite materials. Failure occurs when Equation 2.6 is satisfied.

45
Chapter 2: Literature Review

2 2 2
 11    22   11   22   12 
X
 Y
 X
X
 S
1 ...(2.6)

where

X = Tensile or compressive failure stress in the X direction


Y = Tensile or compressive failure stress in the Y direction
S = In-plane shear failure stress

where the choice of tensile or compressive failure stresses in the X and Y directions
depends on the nature of the respective stresses.

The Tsai-Wu failure theory [65] is another general interactive theory of the form
shown in Equation 2.7.

6 6 6

 Fi i   Fij i j  1
i 1 j 1 i 1
...(2.7)

with all Fij = 0 except,

1 1
F1  
Xt Xc
1 1
F2  
Yt Yc
1
F11  
Xt Xc
1
F22  
YtYc
1
F66 
S2

where

Xt = Tensile failure stress in the X direction


Yt = Tensile failure stress in the Y direction
Xc = Compressive failure stress in the X direction
Yc = Compressive failure stress in the Y direction

46
Chapter 2: Literature Review

S = In-plane shear failure stress

The failure theory defines a simple elliptical failure surface in 1-2 space which crosses
the axes at the four points corresponding to strengths X t, Xc, Yt, Yc. The ellispse can be
rotated by defining a value for F12, which has the effect of increasing the biaxial
strength of the material. F12 can only be determined from biaxial testing. A plot of a
Tsai-Wu failure surface in biaxial stress space is shown in Figure 2-42. The effect of
varying F12 is also shown.

22 F12 = 0
F12 > 0

Yt

Xc Xt 11

Yc

Figure 2-42: Tsai-Wu theory failure surface

Tsai-Wu failure theory does not explicitly distinguish between tensile and compressive
failures, as all coefficients are symmetric to a tensile-compressive reflection.

Interactive failure theories have been utilised for homogenous materials for many
decades with great success. Smooth continuous yield envelopes work for metals
because the materials are homogenous, and the mechanism of yielding is the same
regardless of the combination of stresses applied to the material [66] or any
anisotropy that may occur due to process induced grain elongation. Composite
materials are both anisotropic and heterogeneous, so failure does not occur by the
same mechanism in each direction. In composites, interactive failure theories
superficially seem to account for interaction between stresses in different directions.
Despite this, they have been widely discredited within the scientific literature [66-68],
due to a lack of physical basis and significant physical inconsistencies that arise when

47
Chapter 2: Literature Review

strength parameters are varied. The most famous counter-example to this type of
failure theory is Hart-Smith's figure 'Improved' composite material for submarine hulls
by decreasing transverse tension strength of unidirectional lamina presented in [66].

Mechanistic (or phenomenological) failure theories attempt to capture the true


mechanisms of composite failure and assign a separate failure equation to each
mechanism. Separation of the failure modes allows each material strength term to be
varied and not influence failures that are caused by independent mechanisms,
avoiding Hart-Smith's submarine inconsistency. A few of the more common (or more
promising) mechanistic failure criteria are described below.

The Hashin-Rotem failure criteria [69] was one of the first failure criteria to account for
separate fibre and non-fibre failure modes and the interaction between transverse
stress and shear strain. It is quoted as a 2-D failure, and it does not seem that any
attempts to generate a 3-D version have ever been made, as more advanced failure
criteria were available by the time 3-D criteria were required.

Fibre Tensile Failure (11 0)

 11
1 ...(2.8)
Xt

Fibre Compressive Failure (11 < 0)

11
1 ...(2.9)
Xc

Matrix Tensile Failure (22 0)

2 2
  22   12 


1 ...(2.10)
Yt S

Matrix Compressive Failure (22 < 0)

48
Chapter 2: Literature Review

2 2
  22   12 


1 ...(2.11)
Yc S

where

Xt = Tensile failure stress in the X direction


Yt = Tensile failure stress in the Y direction
Xc = Compressive failure stress in the X direction
Yc = Compressive failure stress in the Y direction
S = In-plane shear failure stress

Hashin-Rotem failure theory is effectively a separation of Hill-Tsai failure criteria into


two independent criteria. It maintains some of the benefits of the Hill-Tsai formula
while removing a lot of the problems of interactive failure criteria.

In biaxial stress space, the Hashin-Rotem failure surface is the same as the Maximum
Stress failure surface in Figure 2-41. If a plot of the failure surface is made in 2-12
space then the interactive nature of the matrix failures becomes apparent, as shown in
Figure 2-43.

12

Yc Yt 22

Figure 2-43: Hashin-Rotem theory failure surface in transverse-shear stress space

Hashin failure theory [70] extends Hashin-Rotem failure theory to account for more
experimental observed interactions between the different stress terms. Hashin theory

49
Chapter 2: Literature Review

can be easily cast into a 3-D form for solid element simulations. The core equations of
the 3-D Hashin theory are given in Equations 2.12 to 2.15

Fibre Tensile Failure (11 0)

2
 11 

 2 12  13   1
1 2 2
...(2.12)
Xt S

Fibre Compressive Failure (11 < 0)

11
1 ...(2.13)
Xc

Matrix Tensile Failure (22 + 33 0)

1
Yt 2  22
1
   33   2  23
2

ST
2
 1
  22 33  2 12
S
2
 13
2
1   ...(2.14)

Matrix Compressive Failure (22 + 33 < 0)

1  Y 2  1
 c
 1  22   33   2  22
   33 
2

Yc  2ST  4ST
...(2.15)
1
ST

 2  23
2 1
  22 33  2  12
S
2

  13
2
1 

where

Xt = Tensile failure stress in the X direction


Yt = Tensile failure stress in the Y direction
Xc = Compressive failure stress in the X direction
Yc = Compressive failure stress in the Y direction
S = In-plane shear failure stress (assuming S12 = S13)
ST = Through thickness shear failure stress (S23)

Hashin is probably the most widely used failure theory in the general scientific
composites literature. 3-D Hashin failure criteria is now included in many commercial
finite element packages.

50
Chapter 2: Literature Review

Puck and Schrmann [15, 71] is a novel failure theory that separates failure in
composites into two broad categories, fibre failure (FF) and inter-fibre failure (IFF). The
theory uses Mohr's and Hashin's considerations on brittle fracture. IFF is separated
into three main modes which are determined by the relative proportion of tensile,
compressive and shear loads on a unit cell of unidirectional material.

Figure 2-44: Matrix failure envelope for Puck and Schrmann theory [15]

One advantage of this theory is that it claims to be able to predict the angle of matrix
crack between the fibres and also the risk that delamination will develop in the
compressive half of the loading regime.

There are numerous other failure criteria presented in the literature which fall under
the banner of mechanistic failure criteria. Most are based loosely on the ground work
of Hashin and Rotem. Yamada and Sun [72] and Chang and Chang [73] have made
notable additions to the field.

The choice of a specific failure theory within this class is more likely decided by what is
available within a given finite element code than any fine distinction between the pros
and cons of each theory.

51
Chapter 2: Literature Review

Mechanistic failure theories have been the most widely adopted over recent years,
due to the somewhat sound physical basis on which they are constructed and the
relative ease in which they can be implement in FE simulations. The general trend in
composite failure theories has been to try and identify the minimum set of failure
modes, and develop a complete understanding of what stresses (or strains) lead to
that type of failure mode. Paris [68] identified at least five failure modes which must
be considered to fully describe the in-plane failure of composite material. Those
modes are; fibre breakage in tension, fibre breakage in compression, fibre matrix
shearing, matrix in tension and matrix in compression. Fully understanding these five
failure modes defines a complete upper limit on the strength of the composite loaded
in any direction. Other failure modes, if identified, simply truncate parts of the stress
or strain space, but the composite cannot exceed the strengths defined by the five
main failure modes under any loading condition.

Failure mode concept (FMC) failure theories are a stronger version of mechanistic
failure theories that attempt to reduce all failures to independent failure modes, each
with a unique equivalent strength. FMC failure theories use invariants of stress or
strain to reduce the load/deformation at a point to a scalar potential that can be
compared to the material strength value for that failure mode. FMC failure theories
remain almost exclusively within the domain of scientific literature, but interest in
them is increasing rapidly.

Strain Invariant Failure Theory (SIFT) was first proposed by Gosse and Christensen [74]
and a closely related companion paper by Hart-Smith [75]. SIFT is a strain-based,
mechanistic failure criteria. SIFT supposes that failure of composite materials at the
micro-mechanical level can only occur via a small set of constituent failure
mechanisms. There are three failure mechanisms considered by Gosse; critical volume
change (dilatation) of the matrix material, critical dilatation-free angle change
(distortion) of the matrix material and fibre failure. Mechanistically, dilatation failure
corresponds to crazing and micro-cavitation while distortion failure corresponds to

52
Chapter 2: Literature Review

shear-yielding. Fibre failure can be controlled by any preferred mechanism, however


investigations by Hart-Smith [75] and Buchanan et al. [76] suggest compatible fibre
failure mechanisms.

Dilatational failure is related to the first strain invariant (J1) shown in Equation 2.16
while distortional failure is related to the second deviatoric strain invariant (J2') shown
in Equation 2.17.

J1  x  y  z
(2.16)
 1  2  3

J12
J2'   J2
3


  x  y    y  z    z  x     xy 2   yz 2   zx 2
1 1

2 2 2
  (2.17)
6  4
1
   1  2    2  3    3  1  
2 2 2

6  

where

x,y,z = Direct strains in the x, y and z directions


ij = Engineering shear strains in the ij plane
1,2,3 = Principal strains

where all strains are micromechanically enhanced to account for the true local strain
field in the heterogeneous composite.

SIFT has been used to predict matrix failure in curved composite sections and co-cured
composite T-sections by Li, Kelly and Ness [77]. SIFT has been used to predict
progressive failure in a composite 3-point bend test by Tay et al. [78].

SIFT will be discussed in more detail in Appendix A.

Cuntze and Freund [79] expanded the work of Puck and Schrmann [15, 71] to include
stress invariants and reduce all the failure mechanisms described in [15, 71] to a
smaller set of specific failure modes. The two failure theories differ on a few key points
but in general the predictions of this theory are similar to those of Puck.

53
Chapter 2: Literature Review

FMC failure theories are still very difficult to implement in FE analyses and require a
deep level of understanding of the failure modes of composites to even measure any
of the characteristic strengths.

Scientifically, FMC failure theories seem to be on a sound basis and this can only be to
their benefit in the long run. Increased research and development of these theories
may see them become more common place outside the composites research
literature.

Stress based delamination onset criteria are based on the same principles as
intralaminar failure criteria. A stress (or combination of stresses) is compared to an
allowable value of stress to determine whether delamination has occurred.

The simplest stress based delamination failure theory is

 33
1 (2.18)
Zt

where

Zt = Through-thickness tensile strength


which assumes that all delamination is a result of through thickness tension.

Ye [80] proposed a delamination criteria that includes an interaction between normal


and shear through thickness stresses. Delamination occurs when either Equation 2.19
or 2.20 is satisfied.

Tensile delamination (33 0)

2 2 2
  33   13    23 



1 (2.19)
Zt S ST

Compressive delamination (33 < 0)

54
Chapter 2: Literature Review

2 2
  13    23 
S
 S
1 (2.20)
T

where

Zt = Through-thickness tensile strength


S = In-plane shear strength
ST = Transverse shear strength

Zhang [81] separated the interactive Ye criteria into two modes, a tensile mode and a
shear mode.

 33
1 (2.21)
Zt

 13
2
  23
2


1 (2.22)
ST

where

Zt = Through-thickness tensile strength


ST = Interlaminar shear strength

Hou, Petrinic and Ruiz [82] have shown that even small compressive stresses have a
large restraining influence on the growth of delamination. The Hou model penalises
compressive normal loading heavily and even forbids delamination if the compressive
normal load exceeds a certain critical cut-off curve.

Tensile delamination (33 0)

2
  33   13
2
  23
2




1 (2.23)
Zt ST

Compressive delamination (-  2
13 
 223 /8 33 < 0)

2
  33   13
2
  23
2
 8 33
2




1 (2.24)
Zt ST

55
Chapter 2: Literature Review

No delamination for (33 < -  2


13 
 223 /8 )

This addition makes a significant improvement to the theory for modelling


delamination damage in the presence of compressive through-thickness loads, such as
impact on a composite plate. Stress based delamination onset theories have been
shown to predict delamination onset well, but require extreme mesh refinement
through the thickness of the laminate to account for the complex stress field that
exists in the through thickness direction.

While not truly an onset theory, fracture mechanics can be applied to composite
materials to monitor crack growth within the composite. Fracture mechanics relates
the rate at which strain energy is released during defect propagation to a previously
determined critical strain energy release rate (SERR) for the material. The strain energy
release rate is generally separated into three components, with each component
related to loading in a different orientation relative to the crack propagation direction.

Figure 2-45: Fracture mechanics delamination modes [83]

Fracture mechanics is not easily applicable to the problems that will be modelled in
this research, as at any given time there may be multiple crack fronts, intralaminar
degradation around or ahead of the crack and highly varied load distributions. Fracture
mechanics can be used indirectly via cohesive elements which can track delamination
damage approximately without many of the difficulties involved with implementing a

56
Chapter 2: Literature Review

fracture mechanics approach such as the Virtual Crack Closure Technique (VCCT).
Cohesive elements will be discussed in a later section.

Attempts have been made to use micromechanically enhanced failure theories such as
SIFT to predict delamination onset [77], but efforts are quite novel and have not been
fully developed yet. There is promise in the use of micromechanically enhanced failure
theories for delamination onset prediction which can account for different thermal
residual strains between plies of different orientation.

Bolted joints in composite structures still have significant strength well after damage
initiation, as most of the failure processes involved are not catastrophic. Correctly
modelling the material degradation that occurs when composite material failure is
detected is often as important as predicting the onset of failure. Damage modelling is
far from a closed area of research. There has been a recent review in the area by
Orifici, Herszberg and Thomson [84] which has served to highlight the wealth of
research still continuing in this area and the lack of general consensus among
researchers.

This section will discuss common methods of modelling damage within composite
materials. This section is far from comprehensive, as there are far too many methods
in the literature to include here. The section will be broken into two main subsections,
the first will discuss damage modelling within the composite plies (intralaminar
damage) while the second section will discuss damage modelling between the
composite plies (interlaminar damage). For further information in the area please refer
to Orifici et al. [84].

Modelling intralaminar damage in composite has been attempted many ways. This
section will discuss the more common techniques for modelling damage within the
composite ply.

57
Chapter 2: Literature Review

Element elimination is the simplest method of modelling damage within the


composite. The element elimination method removes any element that exceeds a
given failure criterion. The removed element cannot carry any load, as it ceases to be
included in any stiffness calculations.

Element elimination has some significant drawbacks. Damage within composite


materials can occur within either material phase or at the interface. Damage of one
phase does not necessarily preclude the other phase from carrying load, so eliminating
an element after any failure is detected can lead to overly conservative failure load
estimates.

Element elimination requires that the stiffness matrix be rebuilt after each elimination.
This requires small time steps for accurate failure modelling, and can also lead to
convergence problems. Element elimination also leads to many more discrete corners
in the mesh, increasing singularity problems.

Progressive damage modelling (PDM) attempts to rectify some of the issues with
element elimination. PDM reduces the stiffness of the failed element, rather than
completely eliminating it. When used in conjunction with a mechanistic failure theory,
the stiffness reduction can be chosen such that it is specific to the type of failure that
has occurred.

PDM still requires a change to the stiffness matrix, but the stiffness drop is generally
not as abrupt, which can improve convergence. There are also no new edges or
surfaces created, so the singularities that are created are weaker.

Camanho and Matthews [50] have used PDM to successfully model damage for a pin-
loaded composite hole. PDM is commonly used in other pin loaded hole and bolted
joint simulations to model damage.

The element failure method (EFM) was first used for metallic structures for modelling
dynamic crack growth by Beissel, Johnson and Popelar [85], although earlier uses exist

58
Chapter 2: Literature Review

in other fields. EFM involves reducing the external reaction forces at the nodes of a
failed element to zero in a given direction. In this way surrounding elements effectively
experience a free surface rather than an element.

Figure 2-46: Damage modelling using the element failure method [78]

Tay et al. [78] successfully used EFM with the strain invariant failure theory for
modelling damage in a composite three point bend test. An undamaged element
carrying a load in both directions is shown in Figure 2-46a. If matrix failure is detected,
the nodal reactions in the transverse direction are zeroed such that the surrounding
elements are not loaded in that direction, as shown in Figure 2-46b. If the fibre and
matrix are both determined to have failed then the reaction forces can be cancelled in
all directions, as shown in Figure 2-46c. Note that the failed element still exists with
the same stiffness and carries significant loads. Extra forces are added to the
simulation to remove the influence of the failed element on the surrounding elements.

The benefit of EFM is that no change to the stiffness matrix is required throughout the
simulation, regardless of the extent of failure. Convergence usually takes a few
iterations after each failure, but is guaranteed because the stiffness matrix remains
unchanged.

Continuum damage mechanics (CDM) is similar to progressive damage modelling.


Damage is spread over an entire element by reducing the stiffness properties. In
general, a damage parameter will be defined as a function of the applied strain on an
element and the stiffness will be modified as shown by Equation 2.25.

Ei  Ei 0 1  di   ...(2.25)

where

59
Chapter 2: Literature Review

Ei = Damaged stiffness in the i direction


Ei0 = Initial stiffness in the i direction
di = Damage parameter associated with the i direction
= Strain tensor

The damage function controls the stiffness reduction. Different damage parameters
and damage functions can be defined for different failure mechanisms. For instance,
the PAMCRASH fabric global ply model [86] based on a work by Ladevze et al. [87]
defines five damage parameters. One damage parameter represents matrix damage
while the other four represent tensile and compressive damage in the two orthotropic
fibre directions.

Many authors [16-18, 86, 88-92] have successfully used CDM to successfully model
intralaminar damage under tensile, compressive and shear loads. CDM is applicable to
both implicit and explicit FE analysis and has been incorporated into various
commercially available FE packages. One benefit of CDM over PDM is that the damage
function can be tailored in such a way as to incorporate the energy absorption
associated with damage propagation through a composite.

Bolted joints can experience significant delamination damage before failing, and in
some cases more than half the ultimate load is carried post-delamination. Modelling
delamination damage accurately is critical to modelling the post-bearing response of
bolted joints and pin loaded holes.

Delamination damage is a discrete phenomenon that poses a significant challenge for


traditional continuum FE analysis. This section will discuss commonly used methods for
capturing delamination damage within a composite material simulation.

PDM for delamination uses the same principle as for intralaminar damage. The
element properties are degraded in the through-thickness direction once a given
delamination failure criteria is reached. PDM has been used to model delamination
damage in a pin-loaded bearing test [93].

60
Chapter 2: Literature Review

PDM has the benefit that no discreet interface needs to be modelled, so there is no
significant increase in computational cost over a traditional PDM simulation. The
problem with this methodology is that a discrete phenomenon is being modelled with
continuum mechanics, so a fine mesh through the thickness of the composite is
required to achieve reasonable results.

Interface elements are separate elements inserted into the FE model to discretely
model delamination damage between plies. Interface elements take a number of
forms, from nodal spring connections [94] to full 3-D solid element formulations [95].
Interface elements define the traction between the two connected substructures as a
function of the relative displacements between the two parts.

Cohesive elements are a special case of interface elements that include some elements
of a fracture mechanics approach. The traction-displacement relation is defined in such
a way that the work done by the interface element to failure is equal to the strain
energy release rate for the material. A typical linear softening cohesive zone traction-
displacement relation is shown in Figure 2-47 for a Mode I opening.

GIc

max

ucrit umax u

Figure 2-47: Stress-displacement function for a typical cohesive zone model

Cohesive elements are less mesh dependent that a general interface element because
the total energy absorbed to create a unit area of surface is always equal to the critical
strain energy release rate. However, the cohesive zone must spread across a few

61
Chapter 2: Literature Review

elements for a cohesive formulation to converge to a correct solution, as shown is


Figure 2-48a. This can lead to prohibitively fine meshes for large simulations.

a)

Cohesive zone

b)

Figure 2-48: Cohesive zone connections with a) fine mesh and b) coarse mesh

Cohesive elements have been used effectively to model delamination damage in


composites by many authors [17, 89, 91, 96-100].

Fracture mechanics approaches such as VCCT were not considered in great detail for
this research as many of the assumptions required for accurate predictions are not
applicable to the analyses required for this project, such as self-similar cracks and the
definition of a pre-existing cracked region.

Most modelling of bolted joints to date has been conducted for pin-loaded holes,
which are a first approximation to the failure of double-lap bolted joints. Modelling of
failure in pin-loaded holes has been conducted in a number of different ways, from
pure analytical solutions through to complicated detailed 3-D models. Each method
will be discussed briefly in the following sections.

Analytical solutions to stress fields in engineering structures are attractive because


they offer the opportunity to determine critical regions and failure loads with minimal
computational effort. Exact analytic solutions only exist for a small set of problems, but
approximate solutions to many problems can be made with careful assumptions and
approximations.

62
Chapter 2: Literature Review

Kradinov et al. [101] created a powerful general analytical solution to the contact
stresses in pinned joints made from elastic isotropic materials. Echavarra, Haller and
Salenikovich [102] created an analytical solution to the stress concentration factors
around a hole for elastic orthotropic plates.

Analytical solutions for pin-loaded holes are useful tools to determine the likely
locations of failure initiation, for relatively simple load cases, but still rely on the
accuracy of the chosen failure criteria. Analytical solutions are not applicable for a
complex problem such as pin bearing once damage has initiated.

Critical distance approaches use a technique first proposed by Chang, Scott and
Springer [103]. The technique assumes that failure in the pin loaded hole will occur
when the stresses in any ply at any point along a characteristic curve reach a given
failure criteria. The curve is described as a radius from the centre of the bearing hole
as a function of the angle from the bearing plane and is given by Equation 2.26 and
shown in Figure 2-49.

d
rc     ROT   ROC  ROT  cos ...(2.26)
2

where d is the diameter of the hole and ROT and ROC are experimental determined
critical distances from notched tension and compression tests.

ROC
Characteristic
curve

Bearing
plane

ROT

Figure 2-49: Characteristic curve for bearing failure

63
Chapter 2: Literature Review

Critical distance approaches use 2-D FE analysis with an assumed pin contact pressure
to determine a stress field in the material which is used to evaluate the stresses along
the characteristic curve.

The technique uses a critical distance for failure because it was recognised at the time
that stresses predicted at the hole surface were not very accurate due to many
modelling approximations. The method is purely empirical and only works because
calibrations have been conducted on notched tension and compression specimens of
the same material. This approach was very popular for many years after it was first
proposed, but is now only seen in the literature in conjunction with analytical solutions
[104].

The first attempts to model the complete failure of pin loaded holes with the FE
method used 2-D layered shell elements. This was likely due to the large
computational cost of modelling a layered composite with 3-D elements. There was
also a lack of accepted 3-D failure criteria and a difficulty in measuring out-of-plane
composite material properties.

Many authors have attempted to predict the failure of composite pin loaded hole
specimens with 2-D FE analysis [105-114]. In general the models used in these papers
provide reasonable predictions of failure initiation loads when the geometric ratios e/d
and w/d are large, as edge effects in these cases are minimal. Progressive modelling in
2-D models can provide some insight into the post-failure behaviour, but is much
better for net tension type failures than for bearing failures.

Two dimensional models provide reasonable predictions of the failure loads of some
pin loaded hole configurations but there are significant deficiencies in the 2-D
modelling method for pin loaded holes. Firstly, experimental evidence has shown the
large influence of stacking sequence [25] and through-thickness restraint [24-25] on
the behaviour of pin loaded holes. 2-D models cannot account for these effects.
Secondly, delamination and through thickness shear failure have been identified as

64
Chapter 2: Literature Review

major failure mechanisms in bearing tests, which clearly cannot be accounted for in
2-D models.

Although reasonable strength predictions can be achieved with two dimensional


models, they will not be considered for this research as they fail to capture many of
the primary failure mechanisms of pin loaded holes or bolted joints, so cannot be
considered adequate for a general bolted joint.

Many authors have observed through-thickness failure modes in pin-loaded holes, yet
to date few have tried to model the problem with 3-D models.

Kelly and Hallstrm [49] conducted a 3-D stress analysis of a pin-loaded hole to
investigate the effect of bolt hole clearance. Strain results were compared with
experimental determined values and reasonably good correlation was achieved. No
failure analysis was conducted in this paper.

Marshall et al. [115] conducted another 3-D stress analysis to investigate the effect of
through thickness clamping with a washer on the through thickness tensile stress
under the washer. The authors identified through thickness tensile stress as the
initiator of delamination within the laminate and showed that clamping the laminate
can significantly reduce the through thickness stress.

Oh, Kim and Lee [116] used a 3-D stress analysis with an assumed contact pressure to
predict failure of hybrid composite pin-loaded holes. It was assumed that the primary
failure initiation mechanism was delamination, and the Ye delamination criteria [80]
was used effectively to predict failure onset. As no in-plane failure was included, no
ultimate load predictions could be made. Chen, Lee and Yeh [117] performed a similar
analysis with good results.

Camanho and Matthews published two papers investigating failure of composite pin-
loaded holes. The first paper [50] investigated failure via intralaminar modes only. A
progressive damage algorithm was used with 3-D Hashin failure criteria. Good
prediction of failure modes was achieved and initial non-linearity of the load-

65
Chapter 2: Literature Review

displacement response was well predicted. The model was not capable of predicting
the significant load carrying capacity of the composite after initial failure.

The second paper [93] was concerned with delamination onset prediction. The finite
element model was modified to include a cubic spline approximation for the through
thickness and interlaminar shear stresses in the model. The model was capable of
predicting different delamination patterns for different laminates with the same
overall percentage of plies in each direction. This result is crucial if a model is to be
robust when used to predict failure onset loads in bolted joints. No intralaminar
damage was considered for this model.

Modelling of pin-loaded holes is still an open field of research. Although many authors
have made good predictions of various failure mechanisms within a pin-loaded hole,
there has not been a convincing model in the literature that can describe all the failure
mechanisms thoroughly and robustly. Most researchers now have moved on to
modelling 3-D bolted joints which include two or more laminates and a bolt within the
FE model. Modelling a 3-D bolted joint has the same challenges as modelling a pin-
loaded hole, with the addition of a few extra problems. There is still great scope to
improve modelling of bearing failure in pin-loaded holes, especially in the post-bearing
regime, with increased computing power and more powerful numerical models which
are becoming available.

A bolted joint is a very complex system. Some additional complications arise over and
above the difficulties of modelling pin-loaded holes, such as bolt rotation, bolt pre-
stress, secondary bending and singular stress fields arising from sharp corners around
the bolt head and nut. Some attempts have been made to model bolted joints semi-
analytically [118] or with complex higher-order B-spline approximations [119], but in
general the FE method is used [53, 120-127] to account for the many complex
phenomena that are taking place during loading.

66
Chapter 2: Literature Review

Ireman [120] and Lin and Jen [121] built full 3-D models of single-lap composite joints
that included secondary bending, bolt rotations and bolt pre-stress. The analyses were
only for stress and deflection analysis however, and no failure predictions were made.
Kelly [126] used a 3-D finite element model for stress analysis of a bolted and bonded
composite joint.

Tserpes et al. [123] used a 3-D progressive damage model to model failure in a hybrid
aluminium-composite joint. The model was capable of predicting in plane damage
quite reasonably, but the failure mode of the joint test was quite brittle, so the
progressive damage formulation was not tested to large displacements.

The most extensive study of bearing modelling of composite bolted joints was
conducted within the European research program BOJCAS. BOJCAS was coordinated by
the University of Limerick. A few papers have been published [53, 122, 124-125] which
detail finite element investigations into failure of bolted joints. The main focus of these
finite element models is investigating the effect of bolt-hole clearance on the load
distribution of multi-row bolted joints. Some preliminary failure analyses were
conducted using Hashin failure criteria.

A very recent study by Hhne [127] has outlined the importance of stiffness
degradation models for composite material damage. It was shown that much of the
behaviour of bolted joint models is determined by the rules governing element
damage after failure rather than the particular theory used to initiate failure. This
result was independently verified with numerical modelling in this thesis.

There have been few attempts in the literature to model the pull-through failure of
bolted joints in composites. Banbury, Kelly and Jain [128] used an axi-symmetric 2.5-D
model to predict pull-through failure. The model gave good predictions of matrix shear
failures but could not well account for the discrete delamination failures or in-plane
fibre failures. Elder, Verdaasdonk and Thomson [129] extended this modelling
methodology to an explicit FE model which included cohesive zone elements between

67
Chapter 2: Literature Review

plies which allowed for the discrete modelling of delamination damage. This model
was also axi-symmetric and could not account for the in-plane fibre damage observed
in some pull-through tests. The nature of axi-symmetric models also means that the
orthotropic fibre directions are not modelled correctly, so stiffness and strength
predictions may be inaccurate. Moscardo [130] at the DLR, recently used a stacked
shell modelling approach to model both delamination damage and in-plane composite
damage. Moscardo's approach captures most of the failure mechanisms involved in
pull-through failure, however it lacks the ability to explicitly separate through thickness
shear failure and delamination failure.

Chen and Lee [131] used a full 3-D solid model to predict failure in a bolted composite
joint under what they refer to as bending loads. The test they were using for validation
was effectively a 3-point bend test of a composite laminate with the load applied via a
countersunk fastener, as shown in Figure 2-50. Reasonable correlation between the
experiment and the numerical models was achieved, although prediction of loads after
failure initiation was not good because the model could only account for delamination
via total element failure. Chen and Lee also stressed the importance of accurately
modelling friction between the bolt and the laminate. A numerical parameter study
was used to show that quite different behaviour resulted from small variations in the
friction coefficient.

Figure 2-50: Bolted joint under bending load [131]

In large scale modelling of composite structures it is not feasible to model each


individual fastener with complex contact models. It is particularly problematic to use
small elements for explicit FE analysis because the time step of the simulation is
governed by the size of the smallest element. There exists a need to model fastened
joints with simple connection elements that do not impact dramatically on the

68
Chapter 2: Literature Review

computational cost of the simulation. McCarthy et al. [2] showed that the response of
the fasteners can have a significant impact on the simulation of impact on a leading
edge structure.

Many engineering approximations need to be made in order to program a simple


element formulation that captures the behaviour of bolted joints loaded to failure.
One element formulation that has been shown to accurately capture much of the
behaviour of bolted joints [59] is the PAM-CRASH point link (PLINK) element [86, 88].
The behaviour of PLINK elements will be investigated in this thesis under complex
loading conditions of a bolted structure under impact.

A significant survey of available scientific literature was conducted. A summary of the


key findings was presented in this chapter.

Composite materials consist of at least two distinct constituents which are combined
to achieve material properties superior to those of each constituent. The multitude of
ways that fibre reinforce plastics can be combined presents significant challenges for
general materials characterisation. Generally materials characterisation tests only see
a small cross-section of the true material behaviour.

The dynamic behaviour of carbon fibre reinforced plastics is particularly difficult to


characterise because one constituent is generally highly rate sensitive while the other
is generally considered to be rate insensitive. Additionally, high rate test results are
very dependent on specimen geometry, boundary conditions and the means by which
load is transferred into the specimen. Temperature also plays a significant role, as in
general the effects of higher temperatures on composites are opposite to the effects
of dynamic loading, and so the two phenomena compete against each other during
any dynamic test. As a result it has been very difficult to find reliable standardised tests
for high rate material properties. In general, fibre dominated properties, such as in-
plane tension strength, are considered rate insensitive, while matrix dominated
properties, such as shear or compression strength, are considered rate sensitive. The

69
Chapter 2: Literature Review

specifics of the rate dependence of these properties is still the subject of vigorous
debate.

Bolted joints play a crucial role in joining composite airframe structures. There has
been a large amount of testing conducted quasi-statically into the strengths and failure
modes of pin loaded holes and bolted joints. There is a scarcity of knowledge however,
when it comes to the dynamic behaviour of composite bolted joints. The few papers
that are presented on the topic contain some critical contradictions. There is a need
for further research in this area as dynamic load cases such as bird strike, belly landing
and gust loading are critical to the effective design of composite airframe structures.

Modelling of failure in composites or composite structures also presents a significant


challenge to researchers. Accurate modelling of bolted joints is of particular
significance for aircraft design, because the large number of parameters that describe
a bolted joint make comprehensive test matrices very expensive. FE modelling of pin-
loaded holes and bolted joints in composite materials has been reported many times in
the scientific literature however some significant problems have yet to be addressed.
Critically, the modelling of compression failure in composite materials has still not
been consistently achieved within the literature. Modelling the effect of failure on
composite materials is also crucial for accurate bolted joint modelling, as often the
majority of the load is transferred through elements that have already experienced
failure.

Finally, there exists a need for improved simplified joint models for aircraft structural
modelling. Even simple aircraft structure may contain hundreds, if not thousands of
bolted connections. Efficiently and accurately modelling these bolted connections is
critical for structural designers. Detailed joint models are computationally very
expensive due to the fine mesh size and the complex non-linear contact algorithms
required. Simplified approaches are often used in place of detailed models, such as
nodal tying or beam elements, which do not capture much of the behaviour of bolted
joints. Improved fastener modelling for structural analysis is an area where significant
scientific contribution can be made.

70
These images show the damage progression in two high rate bearing tests filmed from different sides.
The large amount of degradation of the composite absorbs a great deal of energy. Understanding this
highly dynamic phenomenon is the prime goal of this thesis.
Chapter 3: Single Fastener Joint Tests

Failure of bolted joints is a complex phenomenon and in order to further understand


the failure process, a comprehensive set of baseline tests needed to be conducted. The
tests were designed to characterise the stiffness, strength and failure behaviour of a
range of single-fastener composite joints. The edge and width distances for all tests
were slightly larger than the most efficient design to promote bearing failures rather
than geometry dependent failure modes such as shear out or net tension failure. The
composite material used for all tests was T300/CYCOM970 CFRP plain weave.

Hi-Lok bolts are aerospace grade fasteners commonly used for bolting composite
structures. Hi-Lok bolts feature an automatic preload feature which controls the
tightening torque that can be applied. When the specified level of torque is reached,
based on the collar selection, then the hexagonal tightening section breaks away from
the collar, preventing further tightening. A cross-section of a Hi-Lok bolt is shown in
Figure 3-1.

Figure 3-1: Hi-Lok bolt cross-section [132]

The fasteners used for this research were made from Ti-6Al-4V, a low corrosion, high
strength titanium alloy commonly used in fasteners when in contact with carbon fibre
composite materials. The manufacturer specified minimum shear strength for this
material is 95 ksi (655 MPa). Details of the specific bolt and collar types used for this
research can be found in Appendix C.

A pull-through test is designed to measure the strength of the bolted joint under
normal (pull-out) loads. A standard test method has only recently been introduced,

72
Chapter 3: Single Fastener Joint Tests

ASTM D 7332 [60], so much of the testing that has been done in the past uses slightly
different methodology. In general, a bolt is used to connect a square or circular piece
of composite plate to a metallic bushing with an external thread. A cross-section of a
typical pull-through test specimen assembly is shown in Figure 3-2.

Bolt

Bushing
(Threaded)

Composite Plate
Nut

Figure 3-2: Basic schematic of pull-through test specimen assembly

The composite plate is then restrained by a clamping mechanism while the bushing is
loaded which pulls the bolt through the composite material. This is shown in Figure
3-3. For the rest of the discussion, the term entry face will refer to the top surface and
exit face will refer to the bottom surface of the composite material in Figure 3-3.

Entry Face

Composite
Exit Face panel restrained

Load applied
to bushing
Figure 3-3: Loading of pull-through specimen

A bearing test measures the strength of a bolted joint under shear (bearing) loading. A
bolt is used to join two composite laminates. This can be seen schematically in Figure
3-4. A test standard has been developed for bearing testing of bolted joints, ASTM D
5961 [133]. The region of composite around the bolt is restrained from out of plane
deformation by two large steel guides on either side. The composite plates are then
loaded in tension perpendicular to the bolt direction, as shown in Figure 3-5. Tabs are
used to keep the load aligned. The standard test fixture is shown later in Figure 3-28.

73
Chapter 3: Single Fastener Joint Tests

Bolt

Composite Plates Nut

Figure 3-4: Basic schematic of bearing test specimen assembly

Load applied
to composite

Out of plane
motion
restrained

Figure 3-5: Loading of bearing specimen

The quasi-static tests reported in this section were conducted by the CRC-ACS as part of
research program P1.3.12. These results have been presented before as part of student
research by Hannes Krber [61] and in a paper by Gunnion et al. [59]. The research in
this thesis contributes to P1.3.12 and so this testing is included here for completeness.
Significant analysis was conducted by the author of this thesis.

To build up knowledge of the failure mechanisms in bolted joints it was necessary to


conduct a series of quasi-static tests so that failure could be tracked in progress and to
provide a point of reference with which to compare the dynamic tests. The quasi-static
tests are much more controllable and observable than the dynamic tests and help
build up a better picture of the failure mechanisms present. The bolt and laminate
combinations used for the testing are shown in Table 3-1.

Bolt Diameter Head Type Thickness Laminate Layup


(in) (mm)
5/32 Protruding 2.42 [0/45/0/45/0/ 45 ]s
3/16 Protruding 2.42 [0/45/0/45/0/ 45 ]s
3/16 Protruding 3.52 [(45/0)4]s
1/4 Protruding 3.52 [(45/0)4]s
3/16 Countersunk 2.42 [0/45/0/45/0/ 45 ]s
3/16 Countersunk 3.52 [(45/0)4]s
1/4 Countersunk 3.52 [(45/0)4]s

Table 3-1: Joint combinations used for quasi-static testing

74
Chapter 3: Single Fastener Joint Tests

The dynamic tests reported here were conducted by the DLR as part of research
cooperation with the CRC-ACS. The pull-through tests and counter-sunk fastener
bearing tests were conducted by Thomas Bornshegel [134] with the assistance of
Alastair Johnson and Harald Kraft. This testing falls within the scope of CRC-ACS
research program P1.3.12. The protruding head fastener tests were planned and
manufactured by the author of this thesis and conducted by the DLR. Most of the post-
failure analysis of the specimens was conducted by the author.

There has been little research conducted into the dynamic response of composite
joints, and the little that does exist is contradictory, as discussed in Chapter 2, Section
6. To understand the loading-rate sensitivities that exist in composite joints, a series of
dynamic tests were conducted. Due to the need to test the joints at a range of
velocities, it would have been expensive, time consuming and impractical to test every
joint configuration from the quasi-static tests under dynamic loading conditions. For
this reason one bolt-laminate configuration was selected for testing at a number of
loading rates.

The bolt-laminate combination chosen was a bolt joining 16 ply quasi-isotropic


laminates, as highlighted in Table 3-2.

Bolt Diameter Head Type Thickness Laminate Layup


(in) (mm)
5/32 Protruding 2.42 [0/45/0/45/0/ 45 ]s
3/16 Protruding 2.42 [0/45/0/45/0/ 45 ]s
3/16 Protruding 3.52 [(45/0)4]s
1/4 Protruding 3.52 [(45/0)4]s
3/16 Countersunk 2.42 [0/45/0/45/0/ 45 ]s
3/16 Countersunk 3.52 [(45/0)4]s
1/4 Countersunk 3.52 [(45/0)4]s

Table 3-2: Joint combinations chosen for dynamic tests

Pull-through and bearing joint configurations were tested, with both protruding head
and countersunk fasteners used for the bearing tests. The joints were tested at
constant loading velocities between 0.1 m/s and 10 m/s. A high rate Instron VHS
100/20 (100 kN max / 20 m/s max) was used to conduct the tests. All tests results

75
Chapter 3: Single Fastener Joint Tests

reported for quasi-static and dynamic tests are from joints using this bolt-laminate
combination.

The use of loading-rate is preferred in this research over strain-rate. Strain fields are
not easily measured in the complex 3-D environment around a loaded bolt. Many
approximations to strain in bolted joints such as normalising by a gauge length or
multiplying by a strain concentration factor are techniques based on known strain
distributions for quasi-statically loaded joints in equilibrium and, at best, the technique
predicts the maximum strain. These distributions are not as accurate for dynamically
loaded joints that are not in equilibrium. Also, knowing the maximum strain (or strain-
rate) does not confer a great deal of information about the response of the joint as a
whole, even if the strain-rate dependent behaviour of the bulk composite is well
understood.

Loading-rate is a metric that is more difficult to relate to material properties, but is


more closely related to the physical phenomena which are likely to cause the dynamic
loading in the structure. It is much more practical, for instance, to calculate the
loading-rate experienced by a piece of structure in a crash than it is to calculate the
strain-rate.

All dynamic tests were filmed with a high speed camera with a maximum 3000 fps at
full resolution (1024x1024) or up to 250000 fps with split resolution. Each test was
filmed at a rate that was appropriate for the duration of the test. The specimens were
given a speckle coating using a white background and black spray paint to allow full
field strain measurement post-processing. High intensity lighting was used to allow
enough light to illuminate the cameras CMOS sensor at high frame rates.

X-Ray Computer Tomography (CT) scans were conducted on a representative selection


of failed high-rate test specimens. The CT scans were capable of imaging internal
damage to the specimens with a high resolution. See Appendix C for further
information regarding the capturing of the CT images.

76
Chapter 3: Single Fastener Joint Tests

The pull-through test fixture shown in Figure 3-6 was designed by Krber [61] at the
CRC-ACS as part of project P1.3.12 to test the behaviour and strength of bolted joints
when loaded normal to the laminate.

Figure 3-6: Pull-through test fixture [59]

The composite specimen used it this rig was square with external dimensions 65 mm x
65 mm and a central bolt hole. The laminate thicknesses and layups used, along with
the bolt types and dimensions are summarised in Table 3-1. The specimen was
clamped between two steel rings with 55 mm internal diameter.

A close-up of the functional components of the test rig is shown in Figure 3-7. The rig
included two linear variable differential transformers (LVDTs) which were used to
measure the displacement of the fastener head and the displacement of the
composite material nearby. The use of two closely positioned LVDTs allowed the onset

77
Chapter 3: Single Fastener Joint Tests

of pull-through damage to be established by observing when the LVDT measurements


diverged.

Figure 3-7: Close-up of pull-through test fixture [59]

Characteristic load-deflection curves for the pull-through tests are shown in Figure 3-8,
and can be characterised by the following:

- Initial linear elastic section.

- Distinct damage initiation point followed by a near linear inelastic plateau.

- Near elastic unloading and reloading with minimal hysteresis losses.

- Material failure with large strength and stiffness reductions.

The influence of a load-unload-reload cycle on each bolt-laminate combination was


also investigated by Krber. It can be seen that the unloading has no impact on the
damage propagation and eventual failure of the specimen. This was the case with all
pull-through specimens tested. There was minimal hysteresis in the unload-reload
response of all the pull-through tests. The mean unload-reload slope was lower than
the initial elastic slope.

78
Chapter 3: Single Fastener Joint Tests

6
Constant
Loading
5 Loading with
Unload Cycle
Cross-head Load (kN)
4

0
0 1 2 3 4 5
Cross-head Displacement (mm)

Figure 3-8: Quasi-static pull-through test results

A schematic of the different displacement measurement techniques used for the pull-
through tests is shown in Figure 3-9. Examining the difference between the three
measurements sheds light on many aspects of the failure process.

Bolt LVDT

Laminate
LVDT

X-head

Figure 3-9: Displacement measurements for pull-through test

Firstly, the difference between the cross-head displacement output and the bolt LVDT
output reveals the overall stiffness of the bolt-nut combination. If this difference is
scaled and plotted beside the cross-head load, the results are remarkably similar, as
can be seen in Figure 3-10. The scaling factor used, with units [Force/Length], yields a
number which is the equivalent spring stiffness of the bolt and nut when loaded

79
Chapter 3: Single Fastener Joint Tests

normal to the laminate plane. The scaling value of 24 kN/mm is extremely useful when
calibrating the stiffness of PLINK elements, which will be discussed later in Chapter 6.

Cross-head Load
Bolt LVDT - Cross-head (Scaled)
Load

Cross-head Displacement

Figure 3-10: Bolt assembly stiffness calibration

Secondly, the difference between the laminate LVDT output and the bolt LVDT output
gives an indication of the damage to the composite material in the region local to the
bolt. Internal damage and delamination lowers the local bending stiffness of the
laminate, which in turn leads to less restraint of the laminate under the bolt head,
increasing the difference between the bolt LVDT and the laminate LVDT. This effect is
illustrated in Figure 3-11. In Figure 3-11a, the composite is undamaged and restrained
by the bolt head, so the laminate LVDT displaces by approximately the same amount
as the bolt LVDT. In Figure 3-11b however, local damage has decreased the bending
stiffness of the laminate which allows the laminate to bend more, increasing the
displacement output of the laminate LVDT relative to the bolt LVDT.

This effect was observed in the experiment. A plot of LVDT results, with units of
[Length], from one test is shown in Figure 3-12. The cross-head load from the test is
included (on a different scale) so that changes in the measured displacement can be
compared to changes in the load. It can be seen that the LVDT curves are initially
similar but begin to diverge when the laminate becomes damaged, ie when the load
deviates from linear elastic behaviour.

80
Chapter 3: Single Fastener Joint Tests

a)
Bolt LVDT
Laminate
LVDT

b)
More
Displacement

Figure 3-11: Effect of local damage on pull-through LVDT results showing


a) well restrained laminate b) locally damaged laminate

Cross-head Load
Bolt Head LVDT
Laminate LVDT
Load or Displacement

Crosshead Displacement

Figure 3-12: LVDT output showing the divergence of the LVDT output as the laminate fails

Thirdly, when the bolt begins to pull into the surface of the laminate, the bolt and
laminate LVDT outputs diverge rapidly, as shown in Figure 3-12.

81
Chapter 3: Single Fastener Joint Tests

The specimens all failed in a similar manner, with only slight differences between the
failure modes for specimens with protruding or countersunk head fasteners. The
protruding head fasteners produced a larger damage region relative to the
countersunk head fastener specimens. A comparison between the damage area of
protruding head and countersunk fastener joints is shown in Figure 3-13. In both cases
the damage on the entry face was minimal and confined to an area very close to the
size of the fastener head.

a-i) a-ii)

b-i) b-ii)

Figure 3-13: Comparison between different bolt heads for pull-through failure
showing a) protruding head and b) countersunk head fastener joints
and i) exit face and ii) entry face

The commonly used fixtures for conducting pull-through testing, such as the one
shown in Figure 3-6, had a number of features that limited their ability to be used for
high-rate testing. Firstly, a small but significant displacement was required for the
Instron cross-head to accelerate from rest to the desired constant test speed.
Secondly, the mass of the test rigs is large, so their inertia is very significant when they

82
Chapter 3: Single Fastener Joint Tests

are accelerated, which can swamp the load cell results. Thirdly, the test rig is very
enclosed, which limits the ejection of the high velocity debris created during the test.
Finally, it was desirable to film the composite material during the test, as full-field
strain measurement was possible with post-processing software. This was not possible
with the standard test rig design. A new pull-through test fixture was designed by the
DLR that avoided most of the above issues, which is shown in Figure 3-14.

Calibrated
loadcell bar
Fixed
upper part
Test specimen

Loading plate

Mirror showing
Moving reflection of
lower part specimen

Moving
cross-head

Figure 3-14: Dynamic pull-through test fixture

The bolt was connected to an instrumented bar using the same fitting used in the
quasi-static tests. A loading plate with a 55 mm hole moved downward to strike the
specimen and pull it from the fixture. A mirror was mounted at 45 to axis of the test
rig so that the experiment could be filmed from two perspectives with the one camera.
The rig had a small gap between the specimen and the loading plate which allowed the
cross-head to accelerate to the constant test velocity before striking the specimen.
This gap unfortunately meant that the boundary conditions were not identical to the
boundary conditions from the quasi-static tests, as depicted in Figure 3-15. Therefore

83
Chapter 3: Single Fastener Joint Tests

comparison between the quasi-static and dynamic test results is more complicated,
although many conclusions can still be drawn.

a)

Fully clamped
edge condition

b)

Simply supported
edge condition

Figure 3-15: Different boundary conditions for pull-through tests


showing a) quasi-static condition and b) dynamic condition

The test specimens used were the same dimensions as those used in the quasi-static
tests, which were 65 mm x 65 mm square plates with a central hole. The tests were
conducted at speeds of 0.1, 1, 5 and 10 m/s. As the tests were conducted dynamically,
the test specimens were not in equilibrium. The force measured at the cross-head of
the test machine was not the same as the force closer to the test specimen. Also, the
natural vibration frequencies of each component of the test rig were very important,
because the experimental signal generally had a great deal of oscillatory noise
modulating the true load pulse. If the period of oscillation of the test fixtures was close
to duration of the test, all useful information could be swamped by noise.

An instrumented bar was developed to overcome both these issues. A steel bar with
appropriate fittings was machined and instrumented with a strain gauge very close to
the test specimen. The bar was calibrated as a load cell by recording the output strain
at the gauge for a given load. In this way the bar approximated a load cell or
Hopkinson bar [135]. An image of the bar is shown in Figure 3-16.

84
Chapter 3: Single Fastener Joint Tests

Figure 3-16: Instrumented bar used for pull-through tests

The bar removed much of the equilibrium problem by placing the gauge very close to
the specimen. The vibration modes of the bar were easily calculated from first
principles and were accounted for in the tests. The load-time pulse was recorded
during the tests at 200 kHz and converted to a load displacement pulse by using the
cross-head velocity. The cross-head was very stiff compared to the specimen so the
cross-head velocity was a good approximation of the velocity of the loading plate.

Only countersunk head fasteners were used for the dynamic pull-through testing. A
comparison of the load-deflection responses of the tests conducted at different speeds
is shown in Figure 3-17. One selected result from each loading rate was used. The raw
data has had a 50 point (2.5x10-4 s) moving average filter applied to it to filter out the
very high frequency random noise. The oscillation that remains in the data had a
frequency at least an order of magnitude lower than the filtered noise. This oscillation
is a real effect and is related to the vibration of the test specimen and test rig.

6
0.1 m/s
1 m/s
5
5 m/s
Gauge Load (kN)

10 m/s
4

0
0 2 4 6 8
Deflection (mm)

Figure 3-17: Dynamic pull-through test results

85
Chapter 3: Single Fastener Joint Tests

It can be seen from Figure 3-17 that there were some minor differences between the
load-displacement pulses of the tests. In general, the higher the loading rate, the
higher the peak load carried by the specimen, although the oscillation of the load with
time makes it difficult to discern if this was a real effect. Also, the faster the loading
rate, the shorter the load plateau and the earlier load began to drop off. The overall
shape of the load-displacement pulse is similar to the quasi-static test, shown in Figure
3-8, even though the boundary conditions for the two tests differed slightly.

A sequence of still images from the test footage is shown in Figure 3-18. It can be seen
that the composite panel experiences significant of out-of-plane deformation before
the bolt eventually pulls from the laminate. If Figure 3-18b and c are compared, it can
be seen that the entry face of the laminate returned to nearly the undeformed state
while the bulk panel still had some curvature. This suggests that large delamination
had started occurring within the laminate.

All tests but one exhibited a cruciform splitting pattern on the exit face with 4 radiating
cracks from the bolt hole and 4 raised petals. The raised sections were the result of
extensive delamination over an area much larger than the bolt head, an example of
which is shown in Figure 3-19. The angle at which the cracks radiated from the hole
varied slightly, however the cracks generally aligned themselves with the 45
directions. One test exhibited a similar damage pattern but the splitting was slightly
changed, which is shown in Figure 3-20. In all cases the entry face appeared
undamaged, with a neat hole the size of the fastener head.

86
Chapter 3: Single Fastener Joint Tests

Test Specimen and Reflection


a)

50 mm

b)

c)

d)

e)

Figure 3-18: 0.1 m/s pull-through test failure progression

87
Chapter 3: Single Fastener Joint Tests

Figure 3-19: Typical failure mode for pull-through specimens

Figure 3-20: Unique failure mode for one pull-through specimen tested at 1 m/s

A third party video post-processing algorithm, known as GOM Aramis was applied to
the test footage. The algorithm calculated the 2-D in-plane strain field from relative
locations of the black dots on the surface of the test specimen. The software also had
the capability to create 3-D strain fields from stereoscopic video footage but two high-
speed cameras were not available for this testing. The strain field measured by the
software close to joint failure is shown in Figure 3-21a. The overlaid blue and red lines
in Figure 3-21a correspond to the blue and red curves in Figure 3-21b. The strain-
position curves for these two lines are shown in Figure 3-21b.

88
Chapter 3: Single Fastener Joint Tests

a) b) 6

Von Mises Strain %


4

Distance along/across specimen

Figure 3-21: Full-field strain measurement results for pull-through test

Unfortunately the test specimen experienced a great deal of out-of-plane


displacement during the test which could not be accounted for in the strain
calculation. Therefore the strains calculated are significantly affected by the net panel
deformation rather than localised strains. In this case the measured strains were only
accurate very early in the test when the out-of-plane deformation was minimal.

The representative failed pull-through specimen in Figure 3-19 was scanned in a CT


scanner. The uniquely damaged specimen in Figure 3-20 was also scanned to attempt
to explain the different failure mode. There was no significant dependence on the
loading rate for the pull-through tests so it was not considered necessary to test a
specimen from each loading rate. For more information about the CT scanning process
refer to Appendix B. An example of constructed 2D cross-section images is shown in
Figure 3-22.

The 2D cross-sections were then stitched together to form a 3D volume representation


of the failed specimen. The 3D volume representation adds insight into the internal
damage within the part without the need to section or damage the specimen in any
way. A section of the specimen volume is shown in Figure 3-23. The software also
allows manipulation of the greyscale values which allows the negative space (cracks
and voids) to be viewed in the positive sense. An example of this is shown in Figure
3-24. It shows a portion of the delaminated region around the bolt. It also shows a few
voids that existed within the composite specimen.

89
Chapter 3: Single Fastener Joint Tests

A-A
a) b)

B-B B-B

A-A
c)

Figure 3-22: 2D Cross-section of failed specimen showing a) the sectioning planes relative to the
specimen, b) section through line A-A and c) section through line B-B (not to scale)

Figure 3-23: 3D constructed image of failed pull-through specimen

Figure 3-24: Negative image of failed pull-through specimen


showing a 3D representation of the damage

90
Chapter 3: Single Fastener Joint Tests

The pull through specimens exhibited two types of through thickness damage, which
are both clearly shown in Figure 3-25. The first type, delamination was clearly visible in
the failed specimens to the naked eye. The extent of the delamination was determined
by the CT scans. The second type, not as clearly visible externally, was vertical through-
thickness cracks, either caused by matrix shear failure or bending failures of the
sublaminates.

b)

a)

Figure 3-25: Two through thickness failure modes present in pull-through specimens,
a) delamination and b) through thickness cracks

The co-existence of both these failure modes suggests a damage progression


mechanism that will later be supported by numerical modelling. The damage
progression mechanisms are represented in Figure 3-26.

a)

b)

Mode II
delamination

c)

Bending or shear
failures

Figure 3-26: Progression of the through thickness damage in a pull-through test

The initial damage occurred via Mode II delamination near the mid-plane of the part,
the delamination was not unstable due to the 2-D nature of the crack front and the

91
Chapter 3: Single Fastener Joint Tests

constraint provided by the collar clamping pressure. As the displacement increased,


more delamination growth began between other plies. Progressive stable
delamination growth continued until the local bending strengths in the detached
sublaminates could no longer sustain the applied load and failed in bending. This does
not fully explain the damage propagation because in-plane failures were also occurring
during the test. The relationship between the in-plane and through-thickness failure
mechanisms will be investigated later in Chapter 4.

The load-displacement responses of the pull-through tests were quite similar,


regardless of loading rate. The boundary conditions and the bolt head type had a much
greater influence on the load than the loading rate. There was no immediately obvious
change to the failure mode of the specimens when the loading rate was varied
between quasi-static and 10 m/s. There was one small but important difference
between the low rate (0.1 m/s) and high rate (5 m/s) tests, which can be seen in Figure
3-27. The large load drop in the 3-4 mm displacement range initiates earlier in the high
rate test and the slope of the load relaxation is much lower than for the low rate test.

5
0.1 m/s
5 m/s
4
Gauge Load (kN)

0
0 2 4 6 8
Deflection (mm)

Figure 3-27: Low rate versus high rate pull-through test data

92
Chapter 3: Single Fastener Joint Tests

A bearing test measures the resistance of the bolted joint to in-plane shear loading.
The standard bearing test rig assembly is detailed in ASTM D 5961 [133] and is shown
in Figure 3-28.

Figure 3-28: ASTM D 5961 bearing test fixture

The test rig consists of two long grips which restrain the out-of-plane motion of the
joint and two short grips which clamp the other side of the specimen tabs to keep
the load aligned through the centre of the joint. Additionally there are two support
plates which support the long grips and add further restraint to the out-of-plane
bending. Slots are cut from both long grips to allow room for the fastener head and tail
and also allow room to mount a clip gauge to the specimen to measure strain during
the test.

The specimen geometry used for all the tests is shown in Figure 3-29. The specimen
geometry includes some parameters that were varied throughout the tests to
investigate their effect on the overall joint behaviour. Firstly the thickness, t, was
determined by the laminate layup. The hole diameter, d, was determined by the Hi-lok
bolt specifications, as was the counter-sink angle and depth (for the counter-sunk
fasteners only). All other parameters were constant throughout the tests to provide
controlled conditions for comparison between different tests. Geometric ratios w/d
and e/d varied between tests but were always large enough to avoid net composite

93
Chapter 3: Single Fastener Joint Tests

failure. Also, t/d and h/d varied, but were kept small to avoid laminate thickness
related problems.

Figure 3-29: Bearing test specimen dimensions [61]

Three measurements were taken during the testing; cross-head load, cross-head
displacement and a face mounted extensometer measurement. The extensometer
used is shown in Figure 3-30 and the attachment to the sample is shown in Figure 3-31.

Figure 3-30: Face mounted extensometer [61]

94
Chapter 3: Single Fastener Joint Tests

Figure 3-31: Face mounted extensometer mounted on specimen [61]

Unlike the pull-through tests, the quasi-static bearing tests showed different behaviour
depending on what fastener type was used. In general, there was a significant
difference between the protruding head and countersunk head joints for a given
laminate thickness and bolt diameter. One such pair, a bolt in a 16 ply laminate, is
shown in Figure 3-32. It can be seen that the protruding head fastener joints have a
distinct linear response that deviates as damage starts to accumulate. The load quickly
reaches a peak after which the load remains relatively constant for a large
displacement. The countersunk fastener joints on the other hand have a shorter
discernable linear region, due to the early onset of damage. The load continues to
increase at a lower stiffness and the peak load does not occur until a large
displacement has been reached. Note that in most cases the peak load was similar to,
yet slightly lower than the equivalent protruding head fastener joint, although
significantly more damage precedes the peak load.

95
Chapter 3: Single Fastener Joint Tests

18

16

14

12
Force (kN)

10

4
Protruding Head
2 Countersunk
0
0 1 2 3 4 5
Displacement (mm)

Figure 3-32: Comparison between protruding head and countersunk fastener joints

The influence of a load-unload-reload cycle was investigated. The results of one test
with a countersunk fastener are shown in Figure 3-33. It can be seen that the unload
cycle had no impact on the response of the joint. Hysteresis was present in the unload-
reload response of all bearing tests. The mean unload-reload slope was approximately
equal to the loading slope.

16

14

12

10
Load (kN)

4
Constant Loading
2
Loading with Unload Cycle
0
0 1 2 3 4 5
Displacement (mm)

Figure 3-33: The effect of load-unload-reload cycle on bearing test with countersunk fastener

96
Chapter 3: Single Fastener Joint Tests

Some examples of failed quasi-static bearing specimens are shown in Figure 3-34. It
can be seen that the protruding head and countersunk fasteners generate similar
external damage. For the same overall extension of the joint the countersunk fastener
experienced slightly more rotation. This was due to lower restraint of the fastener
head by the surrounding composite material.

a) d)

b) e)

c) f)

Figure 3-34: Damaged quasi-static bearing specimens showing


a),b),c) protruding head fasteners and d),e),f) countersunk fasteners [61]

Comparing Figure 3-34c) and f), it can be seen that the external damage from the
protruding head fasteners is more localised than that from countersunk fasteners.
These observations do not yield any information about the internal damage however,
and this damage could be much more extensive than any visible damage.

97
Chapter 3: Single Fastener Joint Tests

Similar to the pull-through test apparatus, the ASTM standard bearing test fixture
shown in Figure 3-28 had some limitations which made it unsuitable for dynamic
testing. Firstly, the rig was much heavier than the test specimen, and the inertia of the
rig would distort the results. Secondly, the cross-head could not be accelerated prior to
loading of the specimen. Thirdly, it was very enclosed, which did not allow the ejection
of the high velocity debris created during the test. Finally, the specimen could not be
properly filmed due to the size and obstruction of the rig.

A new test rig was developed by the DLR, shown in Figure 3-35, which alleviated these
problems while maintaining the majority of the features of the ASTM rig. The main
feature of the standard test rig is the out-of-plane restraint provided to the specimen.
The new fixture provided this support via two metal guides which remained stationary
during the tests, so did not impact on the inertia of the specimen. A strike plate and
stop plate were added to allow an initial displacement for crosshead acceleration. The
specimens were still tabbed to align the load correctly, so the overall load transfer and
net joint deflection was very similar when compared to the ASTM fixture. One major
difference between the restraint offered by the two fixtures was the relative motion of
the guide and the composite material near it. In the ASTM fixture, the guides are
always stationary relative to the composite because the guide for each plate is fixed in
the same clamp. In the new rig both sides of the guide are fixed in the top clamp, so
there was relative motion between the composite material on one side of the joint and
the guide, which may have led to some load being transferred by friction.

The tests were conducted at 0.1, 1 and 10 m/s. Cross-head load and displacement
were measured during the test, although due to the dynamic nature of the experiment
the load measured at the cross-head was not as accurate as a load measurement taken
closer to the specimen. For this reason a strain gauge was added to all the specimens
near the hole and the strain-load relationship was calibrated on a quasi-static test. The
location of this strain gauge is shown in Figure 3-36. The black speckle pattern on the
white background was used for calculating full field strain data with the video post-
processing application.

98
Chapter 3: Single Fastener Joint Tests

Grip

Restraining
Guide

Upper Half of
Specimen

Bolt

Lower Half of
Specimen

Grip

Loading Bar

Strike Plate
Stop Plate

Figure 3-35: Dynamic bearing test fixture

Figure 3-36: Strain gauge location used for calculating joint load

For the sake of this discussion, low-rate will refer to the 0.1 m/s and slower tests while
high-rate will refer to the 1 m/s and faster tests.

An assembled protruding head fastener specimen is shown in Figure 3-37.

Figure 3-37: Assembled protruding head fastener test specimen

99
Chapter 3: Single Fastener Joint Tests

Three tests using the dynamic test fixture were completed on a screw driven Instron
tensile test machine at a loading rate of 0.001 m/s as a comparison to the dynamic
tests. The load-displacement curves for these tests are shown in Figure 3-38. The
dynamic test results for the 0.1, 1 and 10 m/s tests are shown in Figure 3-39 to Figure
3-41 respectively.

20
Sample 20
Sample 18
Sample 17
15
Load (kN)

10

0
0 5 10 15
Cross-head Displacement (mm)

Figure 3-38: Protruding head fastener tests at 0.001 m/s

20
Sample 12
Sample 13
Sample 15
15
Load (kN)

10

0
0 5 10 15
Cross-head Displacement (mm)

Figure 3-39: Dynamic protruding head fastener tests at 0.1 m/s

100
Chapter 3: Single Fastener Joint Tests

20
Sample 9
Sample 11
Sample 14
15
Load (kN)

10

0
0 5 10 15
Cross-head Displacement (mm)

Figure 3-40: Dynamic protruding head fastener tests at 1 m/s

20
Sample 6
Sample 7
Sample 10
15
Load (kN)

10

0
0 5 10 15
Cross-head Displacement (mm)

Figure 3-41: Dynamic protruding head fastener tests at 10 m/s

The load measurements shown for the 0.001, 0.1 and 1 m/s tests use the load cell in
the Instron cross-head as this was the most direct measurement technique that did
not rely on previous calibration. The difference between the calibrated strain gauge
load and the cross-head load was minimal initially but did diverge slightly as the
displacement became large.

101
Chapter 3: Single Fastener Joint Tests

The 10 m/s data was dominated by the oscillation of the test rig and the load pulse
lagged the video data by a significant amount due to the time it took the tension wave
to reach the cross-head. The calibrated strain gauges measurements were used for the
10 m/s data because the measurement was taken much closer to the specimen and is
not as affected by test rig oscillation.

There was significant variability within the tests. The two most common failure modes
were cleavage failure and net tension failure, as shown in Figure 3-42. The mode
shown in Figure 3-42a is a classic net-tension failure mode, where the composite
material has failed across the width of the specimen. Figure 3-42b is another failure
mode, which was a mixture of a pure tension failure mode and a cleavage failure
mode.

a) b)

Figure 3-42: Failure modes for dynamic bearing specimens with protruding head fasteners showing a)
pure tension mode and b) tension-cleavage mixture

It was discovered that the tests were on the cusp of the two different failure modes
and this would explain the variability within the tests. The variability was not present in
the specimens of the same geometry when tested using the ASTM quasi-static test
fixture, so an investigation into the possible experimental factors that led to the
difference in failure mode is required. In some tests the failure occurred in the

102
Chapter 3: Single Fastener Joint Tests

laminate on the head side of the bolt, but in most cases the failure occurred in the
laminate on the tail side. The HiLok bolt was nearly symmetric with regards to the head
and tail, so a distribution of failures between head and tail side was not unexpected.

The tests conducted at 10 m/s all failed via the cleavage mechanism shown in Figure
3-42b, while the slower tests shared failure modes. On further investigation of the
specimens and video evidence it was found that the cleavage failures for the 10 m/s
tests were preceded by significant bearing damage prior to final failure that was not
present at the other loading rates.

To compare the load-displacement response of the tests at different loading rates,


loading curves were selected for specimens that all failed via the bearing-cleavage
failure mechanism, as this was the only mode common to all test speeds. The load-
displacement curves for tests at the three dynamic test speeds are shown in Figure
3-43. The curve for the 0.0001 m/s test was not included as it was very similar to the
0.1 m/s test. Note that the load measurement for the 10 m/s test was acquired from a
different sensor than the other loads, so the magnitude may not be directly
comparable. The results compare well with those found by Ger et al. [62], as shown in
Figure 2-39, reproduced here as Figure 3-44.

20
0.1 m/s
1 m/s
10 m/s
15
Load (kN)

10

0
0 2 4 6 8 10
Cross-head Displacement (mm)

Figure 3-43: Comparison between protruding head fastener joint tests at different loading rates

103
Chapter 3: Single Fastener Joint Tests

Figure 3-44: Dynamic-static comparison for a single lap hybrid bolted joint [62]

The energy absorption to failure of each test is shown in Figure 3-45 and it can be seen
that the total energy absorption drops significantly with loading rate for these tests.

140

120
Energy Absorption (J)

100

80

60

40

20

0
0.1 m/s 1 m/s 10 m/s
Test Speed
Figure 3-45: Energy absorption for the dynamic tests

Still images from videos of two distinct failure modes are shown in Figure 3-46 and
Figure 3-47. The video in Figure 3-46 shows specimen 12, tested at 0.1 m/s, which
failed very abruptly from a net-tension failure with almost no prior bearing damage.
The three middle images span a time period of 0.0004 s, which shows how abrupt and
catastrophic this failure mode can be. The energy absorption for this test was very low,
as most of the internal elastic energy was still present at failure. The bolt can be seen
to return to its original position after failure. The video in Figure 3-47 shows a
progressive bearing failure mode from specimen 6, tested at 10 m/s. This failure mode
was present in all three tests at 10 m/s. A large amount of bearing damage precedes a
final cleavage-tension failure.

104
Chapter 3: Single Fastener Joint Tests

Displacement Bolt Tail

0 mm

5.9 mm -

5.9 mm

5.9 mm +

~ 10 mm

Figure 3-46: Net tension failure of specimen 12 tested at 0.1 m/s

105
Chapter 3: Single Fastener Joint Tests

Displacement Bolt Tail

0 mm

~ 1 mm

~ 8 mm

~ 10 mm

~ 12 mm

Figure 3-47: Bearing-cleavage failure of specimen 6 tested at 10 m/s

106
Chapter 3: Single Fastener Joint Tests

The countersunk fastener test results are shown in Figure 3-48. Unlike the protruding
head fastener tests, the failure modes were very consistent between tests at the same
velocity. One representative curve is shown from each test velocity. It is clear that
there is a significant difference between the response of the low-rate and high-rate
tests. A step change in failure mode occurred somewhere between 0.1 and 1 m/s.

16
0.1 m/s
14 1 m/s
10 m/s
12

10
Load (kN)

0
0 2 4 6 8 10 12 14 16 18 20
Cross-head Displacement (mm)

Figure 3-48: Dynamic bearing test results for countersunk fastener joints

Raw test footage yielded a great deal of information about the failure progression of
the dynamic bearing tests with countersunk fasteners. Still images from the test
footage are shown for the low rate test in Figure 3-49 and the high rate test in Figure
3-50 with the corresponding displacement calibrated from the test.

107
Chapter 3: Single Fastener Joint Tests

Displacement Bolt Tail Bolt Head

0.0 mm

2.5 mm

4.0 mm

5.0 mm

5.8 mm

Figure 3-49: Low rate (0.1 m/s) bearing test failure progression

108
Chapter 3: Single Fastener Joint Tests

Displacement Bolt Tail Bolt Head

0.0 mm

0.6 mm

5.0 mm

10.6 mm

17.5 mm

Figure 3-50: High rate (10 m/s) bearing test failure progression

109
Chapter 3: Single Fastener Joint Tests

The low rate test footage shown in Figure 3-49 shows the external damage progression
in the low rate tests. The two columns of images are videos from two different tests
that have been synchronised so that the failure of both sides can be seen together.

Similarly, the high rate test footage is shown in Figure 3-50. It should be noted that at a
displacement of 5 mm, the low rate test has completely failed at the bolt whilst the
high rate test has initiated a progressive bearing failure that continues for a much
longer displacement. It was observed that the bearing damage sustained by the
composite and the amount of bolt rotation at failure was dependent on loading rate.
The final position of the bolts in the joint is shown in Figure 3-51 and Figure 3-52.

a) b) c)

Figure 3-51: Penetration of the bolt head into the laminate


for a) 0.1 m/s, b) 1 m/s and c) 10 m/s tests

a) b) c)

Figure 3-52: Position and rotation of the bolt tail at failure


for a) 0.1 m/s, b) 1 m/s and c) 10 m/s tests

The dynamic bearing specimens with countersunk fasteners had two distinct failure
modes, shown in Figure 3-53 and Figure 3-54. The first, which will be referred to as the
bolt failure mode, involved the bolt fracturing at the beginning of the threaded section,

110
Chapter 3: Single Fastener Joint Tests

at the first stress concentration. This failure mode was not observed in the quasi-static
tests, either because the quasi-static tests were stopped at or just before 5 mm
displacement, or because the boundary conditions for the tests were slightly different.
There is also visible external bearing damage to all load bearing surfaces.

a) b)

Figure 3-53: Bolt failure mode for low rate dynamic bearing specimens with
countersunk fasteners showing a) bolt head side and b) bolt tail side

a) b)

Figure 3-54: Progressive bearing failure mode for high rate dynamic bearing specimens
with countersunk fasteners showing a) bolt tail side and b) bolt head side

The second failure mode, which will be referred to as the progressive bearing failure
mode, involved gross bearing damage to the composite panel. An eventual cleavage
failure occurred when the remaining composite material could no longer resist the
load. It can clearly be seen in Figure 3-54a that the bolt hole has been elongated by
approximately 200% on the bolt head side and that there is extensive damage to the
composite on the tail side. The bolt failure mode was seen exclusively in the low-rate
tests while the progressive bearing failure mode was seen exclusively in the high-rate
tests.

111
Chapter 3: Single Fastener Joint Tests

The full-field strain measurement for the countersunk head bearing tests is shown in
Figure 3-55. These measurements also suffered from the same problem that the pull-
through tests suffered from. The rotation of the bolt is a 3-D effect that is not
interpreted correctly by the software, and the strain-free motion appears as very high
2-D strains. Theoretically this 3-D effect could have been resolved by using
stereoscopic cameras, but another issue was discovered during this test which was an
inherent problem with using surface strain measurement. A frame from a post-
processed results video after bearing damage had initiated is shown in Figure 3-56. It
can be seen that the surface damage removed the paint pattern required for strain
measurement, rendering the technique useless in these areas. The debris from the
damage also obscured the video in other regions. This technique is therefore limited to
use during the early phases of loading, even when stereoscopic cameras are used.

a) b)
20
Von Mises Strain %

10

Distance along/across specimen

Figure 3-55: Full-field strain measurement results for bearing test

Figure 3-56: Limit of full-field strain measurement due to surface damage

112
Chapter 3: Single Fastener Joint Tests

One failed test specimen of the bolt failure type and one specimen of the progressive
bearing type were scanned in a CT scanner, in the same manner as described for the
pull-through tests. For the bolt failure specimen, the laminate from the tail side of the
bolt failure specimen was scanned. A 3-D reconstruction of the CT scan is shown in
Figure 3-57. The bearing direction is shown with a red arrow. A series of reconstructed
CT sections through the width of the specimen, parallel to the bearing plane, are
shown in Figure 3-58. The top of the images is the collar (tail) side and the bottom of
the images is the mid-plane of the joint. The distance shown is the distance of the
plane from the bearing plane of the joint.

Figure 3-57: CT scan reconstruction of the tail side of bolt


failure specimen showing the bearing direction

113
Chapter 3: Single Fastener Joint Tests

-4.2 mm

-3.2 mm

-2.1 mm

-1.1 mm

Bearing
direction
0 mm

1.1 mm

2.1 mm

3.2 mm

4.2 mm

Figure 3-58: CT cross-sections through the width of the bolt failure joint
showing the distance from the bearing plane

114
Chapter 3: Single Fastener Joint Tests

The progressive bearing type specimens were more difficult to scan with a CT scanner
due to the presence of the titanium bolt. The large density mismatch between the
titanium bolt and the carbon based composite material meant that significant beam
hardening would occur in the immediate vicinity of the bolt. It was decided that
removing the bolt before scanning would cause too much damage to the specimen, so
the part was scanned with the bolt in situ. The shape of the specimen scanned is
shown in Figure 3-59.

Figure 3-59: Shape of scanned progressive bearing failure specimen

A reconstructed CT scan is visualised in Figure 3-60 as a number of cross-sections


through the thickness of the joint parallel to the bearing plane. The number on the left
represents the parallel distance between the bearing plane and the cross-sections
shown. The presence of the titanium bolt clearly has a large affect on the clarity of the
images, and significant beam hardening is present. Beam hardening is seen in these
images as grey smudges around the bolt, both in-plane and through the thickness of
the part. There are also other artefacts in the images related to the density mismatch
between the carbon and the titanium, including two parallel white lines which look like
cracks in the top six images, but are actually related to the sharp edges at the head and
collar of the bolt.

115
Chapter 3: Single Fastener Joint Tests

10mm

7.5mm

5.0mm

4.1mm

3.2mm

1.6mm

0mm

10 mm

Figure 3-60: CT cross-sections through the width of the progressive bearing failure joint
showing the distance from the bearing plane

116
Chapter 3: Single Fastener Joint Tests

It should be noted here that there was significant difference between the measured
load-displacement curves for the quasi-static tests conducted by Krber [61] and the
low-rate dynamic tests, as shown in Figure 3-61. The main difference is during the
elastic region, which is significantly stiffer for the quasi-static tests. The two test rigs
were not identical, however the small differences in the restraint of the specimen
should not have caused such a large difference in stiffness. It is believed that the
clamping of the specimens in the dynamic tests may have not been sufficient and that
some slippage in the grips may have occurred.

16
Quasi-static
14 Dynamic

12

10
Load (kN)

0
0 2 4 6 8
Displacement (mm)

Figure 3-61: Difference between load-displacement measurements


from the quasi-static and low-rate (0.1 m/s) dynamic tests

It was discussed in the Chapter 2 that matrix materials are generally very sensitive to
loading rate. Composite material characteristics that derive from the matrix properties
are also likely to have some degree of rate dependence. Shear stiffness, shear
strength, compression strength, delamination initiation stress and delamination strain
energy release rate are all material properties that depend of the properties of the
matrix. These properties also play a large role in the failure of bolted composite joints,
specifically bearing failure and pull-through failure.

117
Chapter 3: Single Fastener Joint Tests

The results of the pull-through tests suggest that the total energy absorption of a
specimen drops slightly with increased loading rate. The failure mode and rate
sensitivity of these tests will be investigated in more detail with parameter studies in
Chapter 4.

The results of both the protruding head and countersunk fastener tests suggest that
the bearing strength of quasi-isotropic T300/CYCOM970 fabric decreases with
increased loading rate, but the extent to which this occurs is difficult to gauge because
the bearing failure mode was not completely isolated.

Composite compression strength, one of the key factors in determining bearing


strength, has been shown to be rate dependent by some authors, but rate
independent by others. Most papers report modest increases in compressive strength
with loading rate. It has also been shown, however, that high temperatures
significantly decrease composite compressive strength. Due to the adiabatic nature of
dynamic loading it is often not possible to separate high loading rates from high
temperatures.

If temperature has a negative effect on the compressive strength of carbon fibre


composite then it almost certainly has a negative effect on the bearing strength of the
composite. The temperatures did increase with the high rate bearing specimens. No
temperatures were monitored during the bearing tests however in the structural test,
shown later in Chapter 5, the heat generated due to failure of the bolted connection at
1 m/s was enough to ignite the composite debris falling from the test article. The
dramatic temperature rise gives one compelling explanation as to the reason that
progressive failure loads for the bearing specimens decreased with loading rate.

This set of test results highlight a number of important factors regarding the
dependence of fastener composite joints and, for that matter, any fastened composite
structure. The failure of bolted joints in composites is a complicated problem quasi-

118
Chapter 3: Single Fastener Joint Tests

statically, and the addition of loading rate adds a further significant complication.
Loading rate can influence the failure mode of composite joints as well as failure load,
and a distinct separation of these two concepts is required for understanding the test
results and for designing future testing.

Within one failure mode, for instance bearing failure, there may be significant rate
sensitivity, as seen in the protruding head fastener bearing tests. This change can
influence the stiffness, failure load and post-failure energy absorption of the joint.

Also, varying loading rate in composites bolted joints affects each failure mode
differently. As a result, failure modes in a bolted joint that may occur at one test speed
may be suppressed at another test speed. For instance, net tension failures are
dominated by tensile fibre failure and should be relatively unaffected by loading rate.
Bearing failures however are dominated by compressive fibre failures, which are
matrix dependent, and can vary dramatically with loading rate. A joint that may have
failed by bearing failure in a quasi-static tests may switch to net tension failure for a
dynamic test or vice versa. In the countersunk fastener tests, bolt failure at low rates
switched to bearing failure at higher rates. In the protruding head fastener bearing
tests, the quasi-static failure modes were shared between net-tension and cleavage
failure modes. At high loading rates the failures all occurred via a bearing-cleavage
failure mechanism.

Unfortunately, well designed joints are generally near the cusp of two failure modes,
as this gives the best joint efficiencies. Separating the rate dependence of one
particular failure mode from the variation of failure mode with loading rate is not a
simple task, and requires two different testing methodologies that were not fully
appreciated before this testing was undertaken.

Capturing one particular failure mode at different loading rates requires the exclusion
of all others by choosing unrealistic joint geometries that force one particular failure
mode. If bearing failure is desired, then large edge-to-diameter and width-to-diameter
ratios are required as well as careful selection of bolts to avoid bolt failures. Similarly, if
net-tension failures are desired then a small width to diameter ratio should be chosen.

119
Chapter 3: Single Fastener Joint Tests

This would allow the rate dependence of each failure mode to be investigated
independently.

Investigating the change in failure mode with loading rate requires careful selection of
joint geometries such that the joint is near the cusp of two failure modes, and a prior
knowledge of what affect loading rate has on each particular failure mechanism.
Consider a typical design curve for a bolted joint with a constant diameter bolt and
varying joint width [3], as shown in Figure 3-62. Assume, simplistically, that the bearing
strength of the joint drops by 50% at a given dynamic loading rate, while the net-
tension strength is unaffected. Three different behaviours would be observed
depending on the joint width. For the case of small joint widths, Region A in Figure
3-62, no change in behaviour would be observed at all; the failure mode and failure
load would not be affected by loading rate. For the case of wide joints, Region C in
Figure 3-62, the failure mode would still be bearing failure but the failure load would
decrease by 50%. For intermediate widths, Region B in Figure 3-62, both the failure
mode and failure load would change with loading rate, and the percentage of peak
load drop would depend on the exact joint geometry.

Region A Region B Region C

Bearing
failures
Load

Tension
failures

Quasi-static
Dynamic

0 2 4 6 8 10
w/d

Figure 3-62: Range of joint geometries at which changes in failure mode will occur

The best joint efficiencies for bolted joints measured in terms of strength of the joint
per unit weight in carbon fibre composites are achieved for joint widths near the cusp

120
Chapter 3: Single Fastener Joint Tests

of tension and bearing failures [3]. Therefore, well designed joints for quasi-static joint
efficiencies will generally be within this third region, where both failure load and
failure mode change with loading rate.

This simplistic example only considers two failure modes and the experimental
program to fully characterise the behaviour is already large. If additional failure modes
are also considered then the test matrix grows rapidly. The contradictions observed in
the literature as to the effect of loading rate on bolted joints are most likely related to
the different joint configurations tested.

It is quite clear that there is rate sensitivity in all the joint configurations tested for this
research. The rate dependence is more pronounced for the bearing test. Based on the
joint geometries used and the results observed, it is assumed that the protruding head
fastener tests straddled the border between regions b and c in Figure 3-62. At lower
rates the failures were a mix of net tension and bearing-cleavage. At higher rates the
failures were all bearing-cleavage failures.

A similar comparison can be made for the countersunk bearing tests. If it is assumed
that the titanium material strength increases logarithmically with loading rate [41]
then a transition between bolt failure and bearing failure would occur if the initial
strengths of the two modes were similar.

Qualitatively the results for the bearing loaded joints can be plotted on a strength-
loading rate graph as shown for the protruding head bearing specimens in Figure 3-63
and for the countersunk head bearing specimens in Figure 3-64. It is assumed that the
rate dependence appears at some critical strain rate [26] and the strengths increase or
decrease logarithmically with strain rate. Note that these graphs show just one
hypothetical description of very complicated behaviour.

121
Chapter 3: Single Fastener Joint Tests

Bearing Failure
Net-Tension Failure

Failure Load

0.001 0.01 0.1 1 10 100


Loading Rate (m/s)

Figure 3-63: Possible loading rate behaviour for protruding head bearing tests
showing rate dependence that would explain test results

Bearing Failure
Bolt Failure
Failure Load

0.001 0.01 0.1 1 10 100


Loading Rate (m/s)

Figure 3-64: Possible loading rate behaviour for countersunk head bearing tests
showing rate dependence that would explain test results

122
Chapter 3: Single Fastener Joint Tests

The reduction of progressive bearing load at dynamic loading rates allows for an
interesting approach to joint design. Total energy absorption is a function of both load
and displacement. The progressive bearing displacement can be extended by
increasing the edge to diameter ratio of the joint, effectively increasing the total
absorbing capability of the joint, with minimal structural weight penalty. If the
observed interaction between net tension failure and bearing failure can be shown to
hold more generally, then joints could be designed to fail in the efficient net tension
mode [3] quasi-statically, but fail in the energy absorbing progressive bearing mode
dynamically.

The behaviour of single fastener bolted joints in carbon fibre composite material has
been investigated experimentally. A test matrix of various bolt and laminate
combinations has been completed quasi-statically. One bolt diameter and laminate
combination was used for dynamic testing at loading rates between 0.1 and 10 m/s
with a high rate Instron tensile test machine, as shown in Table 3-2.

A wide range of methods were used to capture information during the high-rate tests.
Load was measured using a load cell in the Instron cross-head and by the use of
calibrated strain gauges placed strategically on the specimens or fixtures. Visual
information about the failure modes was captured using a high speed camera filming
at rates up to 6000 fps. Failed specimens were examined using non-destructive CT
scanning, and vast amounts of information about the structure of the internal damage
within the specimen were gathered.

It was found that some rate dependence was present in all tests, which was expressed
in a number of ways which are discussed in the previous section. This dependence was
very difficult to characterise precisely within the limitations of a reasonable test
matrix. Future test programs, which have been discussed, could be planned to capture
some aspects of the rate sensitivity.

123
Chapter 3: Single Fastener Joint Tests

Within the pull-through tests only minor variations in the total energy absorption of
the specimens were found with increased loading rate across the range of test speeds
considered. No overall change in specimen failure mode was observed when
examining the high-speed film or the failed specimen CT images.

Within the bearing specimens, variations in failure mode, failure load and energy
absorption were identified with increased loading rate. The bearing specimens
exhibited at least three distinct failure modes; progressive bearing failure, net tension
failure and bolt failure. There was a strong correlation between the test speed and the
failure mode observed. In general, it was found that the resistance of the composite to
bearing damage was reduced with increased loading rate, especially above 1 m/s.
There was a general propensity for all shear tests of bolted joints to fail via progressive
bearing at high loading rates and the extent of the bearing damage also increased with
increased loading rate.

124
A comparison between the experimentally observed delamination of a
pull-through specimen and the prediction of a stacked shell model in
PAM-CRASH using cohesive zone interface elements
Chapter 4: Detailed Joint Modelling

The process of generating and testing large experimental programmes is both time
consuming and expensive. Composite properties are very sensitive to laminate layup
and therefore it may not be possible to define non-dimensional quantities such as
bearing strength that scale to different laminates or bolt dimensions, especially in the
single-lap bolt configuration. To generate joint stiffness and failure loads for new bolt
and laminate combinations, a completely new set of tests need to be conducted.
During the design phase of a composite structure, the laminate and bolt geometry may
change many times with each design iteration. A new experimental program for each
new joint type would be extremely costly and slow down the design process.

It is therefore highly preferable to have validated finite element models that can be
used to assess basic properties of a new joint configuration such as stiffness, failure
load and failure mode. The bolted joint tests conducted in the previous chapter
provide an excellent database with which to validate numerical modelling. This chapter
will investigate different modelling methodologies for bolted joints and assess the
capabilities and limitations of each.

The failure of bolted composite joints is a complex combination of failure mechanisms


that rely on both the composite material properties and the specific load to which the
joint is subjected. A selection of failure mechanisms and their causes within a joint is
shown in Table 4-1. Any tool for modelling composite joints must be able to capture all
of these failure mechanisms, or at least the dominant ones for the specific test. The
tool must also be capable of predicting progressive failure, as for many bolted joint
configurations, a significant amount of residual stiffness, strength and energy
absorption are still available to the joint after the initial onset of damage.

Finite element (FE) analysis is the most commonly used tool for analysing structural
components for stiffness and strength. Most commercially available FE packages have
the ability to impose complex failure criteria upon the composite material which can
capture some, if not all, of the mechanisms shown in the table. The specific advantages
of the FE codes used will be discussed in each section.

126
Chapter 4: Detailed Joint Modelling

Causes and Location of Different Failure Mechanisms


 Tension load on 0 fibres normal to the net tension plane
Tensile Fibre
 Tension load on 45 fibres normal to the 45 planes
Failure
 Tension load on 90 fibres normal to bearing plane
 Generally follows fibre failure in the same locations
Tensile Matrix
Failure  Matrix micro-cracks in plies transverse to the principal loading
direction
Compressive  Compressive load on 0 fibres on bearing plane
Fibre Failure  Compressive load on 45 fibres on the 45 planes
Compressive  Compressive load on hole surface between 45 planes
Matrix Failure  Through-thickness crushing
Matrix Shear  Shear load on 45 planes
Failure  Through-thickness shear load due to bolt rotation
 Mixture of Mode I and II delamination in many locations on
Delamination
the bearing side of the net tension plane
 Axial or bending loads at stress concentrating corners or
Bolt Failure
threads
Table 4-1: Bolted joint failure mechanisms and their locations

Generating the complex mesh for a composite bolted joint model is a time consuming
task. The complex geometry requires careful domain decomposition to maintain a high
quality structured mesh. A parametric bolt modelling tool has been developed using
ANSYS APDL language. The tool allows for fast the generation of complex mesh
structures for composite bolted joints. A comprehensive parameter set allows fine
control over both geometric and nodal dimensions.

The tool generates a bolt-laminate assembly with controllable structured mesh for the
composite laminate, bolt, collar and any restraints that are necessary to accurately
represent the experimental conditions. The tool accounts for a user-defined number of
symmetry planes and is equally capable of modelling pull-through and bearing models.
The model is controlled through simple parameters and flags input through an ANSYS
command file.

127
Chapter 4: Detailed Joint Modelling

Figure 4-1 through to Figure 4-3 show most of the geometry parameters that can be
controlled by the ANSYS input file. A list of the parameters and their descriptions is
given below.

Length e

Figure 4-1: In-plane geometry parameters

Nply Nsub

Tply Tsub

Composite Solid Stacked


laminate model shell model

Figure 4-2: Through-thickness geometry parameters

CSA A
H
H

Grip D

Acoll

Protruding
Countersunk
Head
Figure 4-3: Bolt, collar and hole geometry parameters

w = Joint width
e = Joint edge distance (same as w for pull-through)
Length = Joint tab length (bearing joint only)
Nply = Number of composite plies
Tply = Thickness of each ply
Nsub = Number of sublaminates (stacked shell model)
Tsub = Thickness of each sublaminate (stacked shell model)

128
Chapter 4: Detailed Joint Modelling

D = Diameter of bolt (or hole)


H = Depth of bolt head
CSA = Countersink angle (countersunk head)
A = Head diameter (protruding head)
Grip = Grip length of bolt (twice the laminate thickness)
L = Length of collar
Acoll = Collar clamp diameter

The input file generates a mesh on all parts of the bolt, laminates and grips that is
completely structured and controllable with parameters. All necessary contact areas
have a parametric tolerance that allows the surfaces to be offset slightly if required.
Two representative meshes created with the tool are shown in Figure 4-4 and Figure
4-5.

Figure 4-4: Pull-through model mesh example (solid plate)

Figure 4-5: Bearing model mesh example (stacked shell plate)

129
Chapter 4: Detailed Joint Modelling

For the dynamic equation of motion [136],

M Dn  C Dn  K Dn  R ext n (4.1)

implicit finite element analysis uses a difference expression of the general form

Dn1  f Dn1 ,Dn1 ,Dn ,Dn ,Dn ,Dn1 ,... (4.2)

where the condition of the system { D } at time tn+1 depends on the current state time
derivatives { D } and { D } at tn+1.

Modelling bolted joints is a highly non-linear problem. Non-linearity exists in the


material behaviour, the contact definitions and the geometry. For the structural
equilibrium equation [136],

K D  R (4.3)

both the coefficient matrix [ K ] and the load vector { R } may dependent on the
deformation state of the system { D }. To solve for the deformations for a given external
force, a non-linear solution scheme such as the Newton-Raphson method is employed.

Implicit non-linear analysis is excellent for modelling many non-linear problems but
problems still arise for highly non-linear materials or discrete damage regions which
may cause the stiffness matrix to become ill-conditioned or singular.

The non-linear implicit FE solver used for this research was MSC.Marc. Marc is a
powerful non-linear solver that is highly customisable with the option of user defined
subroutines for many functions.

For the dynamic equation of motion [136],

M Dn  C Dn  K Dn  R ext n (4.4)

130
Chapter 4: Detailed Joint Modelling

explicit finite element analysis uses a difference expression of the general form

Dn1  f Dn ,Dn ,Dn ,Dn 1 ,... (4.5)

where the condition of the system { D } at time tn+1 does not depend on the current
state time derivatives { D } and { D } at tn+1. Explicit FEA does not require inversion of
the stiffness matrix in order to solve for { D } and hence is a much simpler calculation,
especially as the number of degrees of freedom in the simulation increases. For a
single time step, explicit FEA can be many orders of magnitude faster than implicit FEA,
because large implicit finite element analyses require the inversion of large broad-
band stiffness matrices. Explicit finite element analysis scales far more favourably with
mesh size. The severe limitation that hampers explicit FEA is that to maintain
numerical stability, the time step, t must meet the CFL [137] condition

t tc (4.6)

L
tc (4.7)
c

where

L = Smallest dimension of an element in the finite element mesh


c = Speed of sound in the material

Note that different materials will have different speeds of sound, so a minimum t c
needs to be found across the material domain. In general, the critical time step is very
small for finite element analyses of engineering components. This makes explicit FEA
unsuitable for quasistatic or low speed structural dynamics problems, as the number of
time steps required to cover the simulation domain becomes prohibitively large.

Explicit FEA can handle non-linearity of geometry and material properties with a
minimum of computational effort and no need for convergence calculations. Heavy
material degradation and element elimination do not cause the stability problems that
they do in implicit FEA.

131
Chapter 4: Detailed Joint Modelling

Explicit FEA
 Short (dynamic) simulations
 Large DOF solutions
Suitable for  Highly non-linear problems
 Element elimination and collapse
 Coarse mesh sizes
 Quasi-static and low speed analyses
Not suitable for
 Fine mesh sizes
Table 4-2: Explicit FEA suitability

The explicit FEA code used was PAM-CRASH, developed by ESI, which is heavily used in
the automobile industry for crash studies. The code has many features which make it
an attractive option for modelling the high rate bolted joint failures.

132
Chapter 4: Detailed Joint Modelling

It has been shown that bolt bearing failure is a 3-D phenomenon. Studies that have
attempted to capture the failure load of pin-loaded holes or bolted joints with 2-D
shell element models have failed to capture many of the key failure modes. The
modelling in this thesis will use a full 3-D contact model with fully integrated 3-D
elements in the vicinity of the bolt. Solid shell elements with reduce integration
through the thickness will be used for regions away from the high gradient stress fields
near the bolt.

The laminates were made from a plane weave fabric. The composite fabric was
modelled with a 3-D orthotropic material. The model had a full set of 3-D stiffness
terms as shown in Table 4-3. These stiffness values were taken from material data
sheets where available and private CRC-ACS communications.

E11 55800 MPa


E22 55800 MPa
E33 5560 MPa
G12 3650 MPa
G23 3650 MPa
G31 3650 MPa
12 0.06
23 0.38
13 0.38
Table 4-3: T300/Cycom970 fabric orthotropic stiffness properties

Failure was predicted using modified 3-D Hashin fabric failure criteria. The allowables
for the criteria will be discussed later. The use of SIFT was also evaluated on a bolted
joint model and lug model, but the code is still under development and requires
further refinement. These results are presented in Appendix A.

The strength values used for the Hashin failure routine are shown in Table 4-4.
Accurate strength values were difficult to obtain, especially for compressive and
through-thickness properties. When specific data was not available from material data
sheets, approximations were made.

133
Chapter 4: Detailed Joint Modelling

Tensile Compressive
X 630 MPa 550 MPa
Y 630 MPa 550 MPa
Z 20 MPa 100 MPa
XY 100 MPa 100 MPa
YZ 30 MPa 50 MPa
ZX 30 MPa 50 MPa
Table 4-4: T300/Cycom970 fabric strength properties

After failure, a progressive damage method was used to degrade the material stiffness
to the appropriate value. The degradation factors are expressed as a fraction of the
initial stiffness of the material. Undamaged material has degradation factors of 1. The
relationship between the damaged stiffness, the initial stiffness and the degradation
factors is shown in Equation 4.8.

 E11  D11 . . . . 0   E11 


E   . D22 . .  E22 
 22  
E33   . D33 . .  E33 
    ...(4.8)
G12   . . D12 .  G12 
G23   . . D23 .  G23 
    
G31   0 . . . . D31  G31  0

where

Eij = Degraded stiffness


Dij = Degradation factor
Eij0 = Initial stiffness

The degradation factors for each failure mode were adapted from Camanho and
Matthews [50] to be applicable to composite fabrics. The degradation takes into
account whether the predicted failure is within the fibre or matrix phase and whether
the failure is of the tensile or compressive type. The factors are summarised in Table
4-5. An investigation of the affect of these parameters on the simulation was
undertaken. Fortran subroutines were written to apply the failure criteria and the
progressive damage algorithm to the model. These are discussed in Appendix D.

134
Chapter 4: Detailed Joint Modelling

Tensile Compressive
Fibre D11 or D22 = 0.07 D11 or D22 = 0.14
D12 = 0.2 D12 = 0.4
Matrix D23 = 0.2 D23 = 0.4
D31 = 0.2 D31 = 0.4
Table 4-5: Stiffness reduction factors for progressive damage analysis

The form of the stress-strain curve for a simple tension or compression load on a single
element is shown in Figure 4-6. The initial stiffness, E0, is used until a failure stress (or
strain) is reached, max. The stiffness is immediately dropped to the degraded stiffness,
Ed, after which the loading continues. Final element elimination was imposed when the
element became numerically unstable, ie high distortion or inversion. Once failure has
occurred, element unloading occurs at the damaged stiffness.

max
Stress

E0
Ed

Strain

Figure 4-6: Stress-strain curve for progressive damage

No explicit delamination prediction was conducted for the implicit modelling in


MSC.Marc. Delamination onset was predicted with the Hashin matrix failure envelope.
The damage was modelled with element stiffness reduction via the progressive
damage subroutine.

Marc has recently implemented a cohesive zone delamination model that can be used
with this model. Unfortunately the implementation is not currently compatible with
element elimination so was not able to be used.

135
Chapter 4: Detailed Joint Modelling

For details of the experimental program and results please refer to Chapter 3.

The bearing model with a countersunk fastener is shown in Figure 4-7. The model is
assembled from six independent sub-parts; the bolt, collar, top laminate, bottom
laminate, top grip and bottom grip. All parts were meshed with a structured mesh of
eight node hex elements. The model utilises one symmetry plane to improve
computational efficiency. The symmetry plane was justified as the laminate was
symmetric about this plane and the damage observed in the experiments was
symmetric also.

The mesh used for the bolt and collar is shown in Figure 4-8. The mesh used for the
steel grips is shown in Figure 4-7. The bolt and collar were given elastic-linear
hardening properties of Ti-6Al-4V while the grips were given elastic properties of mild
steel. The properties are summarised in Table 4-6.

Side View

Isometric View

Figure 4-7: Implicit FE bearing model mesh with countersunk bolt

136
Chapter 4: Detailed Joint Modelling

Side View
Isometric View

Figure 4-8: Bearing countersunk head bolt mesh

E = 115 GPa
= 0.33
Ti-6Al-4V
yield = 900 MPa
Et = 10 GPa
E = 210 GPa
Mild Steel
v = 0.33
Table 4-6: Isotropic material properties

The composite laminates were meshed with a structured radial mesh around the bolt
hole. The mesh used for the laminates in the region around the bolt hole is shown in
Figure 4-9. The laminates were modelled with one layer of elements per ply, so that no
homogenised laminate elements were required. The ply properties were linear
orthotropic with a progressive damage formulation.

The mesh was biased such that the density was highest close the bolt hole. The
average in-plane element size at the edge of the hole was 0.3 mm. The constant
through-thickness mesh size of 0.22 mm created elements of very high aspect ratio in
regions away from the bolt. A reduced integration element formulation was used in
these areas to improve the element stiffness behaviour [138].

Penalty based contact conditions were used in the analysis [138]. All friction
coefficients were 0.2. There was conflicting reports in the literature as to the
coefficient of friction for composite-composite and composite-metal interactions. A
coefficient of 0.2 was considered within the range of values most commonly quoted
although values as high as 0.5 and as low as 0.03 have been quoted in the literature for
similar material combinations.

137
Chapter 4: Detailed Joint Modelling

Side View

Isometric View

Figure 4-9: Bearing countersunk head laminate mesh

The load was applied in three load steps. Firstly, the bolt and collar were preloaded
while the rest of the model remained unrestrained. The preload was applied as a fixed
displacement to the external surface of the bolt shaft and the internal surface of the
collar. The second load step was a null load step where the preloaded bolt and collar
were held in place while a glued contact condition was applied to fix the bolt and collar
relative to one another in the region where they came into contact, i.e. the threaded
section. The third load step was the applied shear load where one end of the specimen
was fixed while the other was loaded with a ramped displacement. Each displacement
increment was 0.05 mm initially and shorter when large amounts of damage caused
convergence issues.

A second model with a significantly more refined mesh was run without any failure
criteria for a pure stress analysis. This model took over 48 hours to run even on the
fastest available CPU, so was only used to investigate stress fields rather than failure.
All the stress plots in the results section that do not have a mesh overlaid upon the
stress data were generated with this model. This model was intended primarily as an
indicator of stress distributions and critical locations and unfortunately stress contour
values were not recorded for all these results images.

138
Chapter 4: Detailed Joint Modelling

The clamping stress was dependent on the relative stiffness of the bolt shaft, collar
and laminate. A smoothed stress plot of the clamp-up load for a 0.1 mm tightening
displacement is shown in Figure 4-10. As the clamp-up was a small displacement
relative to laminate thickness, the clamp-up stress was linearly related to the clamp-up
displacement. It can be seen that there are two singular regions where the stresses
and stress gradients are very high. These two regions are at edge of the countersunk
region and at the edge of the collar.

Side View
Isometric View

Figure 4-10: Through-thickness compressive stress distribution showing


high stress (blue) through to low stress (dark yellow) regions

The resulting through-thickness stress at the interface between the top and bottom
laminates is shown in Figure 4-11. The average stress in the region is approximately 25-
30 MPa however the maximum stress is approximately 80 MPa. The standard
approximation by other authors who have studied bolt clamp-up [24-25] is to calculate
an average clamping stress based on the bolt load normalised by the area under the
bolt head or washer. This pre-stressing analysis has shown that the average stress is
not a useful indicator for the actual stresses at the central contact plane.

The pre-load had a number of effects on the overall joint performance that should be
noted. The first and most important benefit of the pre-load was to increase the friction
force between the two laminates such that a significant proportion of the load could
be redistributed away from the bearing surfaces.

139
Chapter 4: Detailed Joint Modelling

Figure 4-11: Through-thickness stress at the interface between the two laminates (MPa)

The less obvious effect of bolt pre-tension is the pre-loading of the matrix and fibres,
especially for the countersunk fastener joint. The preload in the 0/90 plies in the
loading direction is shown in Figure 4-12. If the singular regions around the
countersunk edge and bolt collar are ignored, the maximum compressive stress in the
fibres due to the 0.1 mm preload is 49 MPa, which implies that the fibres are at
approximately 10% of their failure load before the bearing load is applied.

Figure 4-12: Stress in the loading direction after bolt pre-load

Other benefits or drawbacks of the clamping load must be interpreted in conjunction


with a good failure theory, as failure is not well predicted in the clamped region by

140
Chapter 4: Detailed Joint Modelling

simple failure theories such as maximum stress theory. The compressive stress
through-the-thickness provides strength benefits for some theories. Within SIFT for
instance, a compressive through-thickness matrix load would minimise J1 types of
failure while promoting some J2' type failures.

The bearing load was applied as shown in Figure 4-13.

Figure 4-13: Loading direction for countersunk bearing model

The interaction between the preload and the bearing load was complex. The
countersunk model was not symmetric through the thickness of the laminate so the
bolt rotation caused by the bearing load did not affect the top and bottom laminates
equally. The through-thickness compression load was significantly larger in the bottom
laminate than the top. A simulation was conducted without pre-load and the through-
thickness stress after a small number of load steps is shown in Figure 4-14a. The
equivalent state for the simulation with pre-load is shown in Figure 4-14b, with the
same stress scale.

a) b)

Figure 4-14: Through-thickness stresses under applied bearing load


showing a) with no bolt pre-tension and b) with pre-tension

As found by McCarthy et al. [53], the bearing load was not reacted evenly through the
thickness of the laminate. The load carried by the plies nearest the shear plane was
significantly higher than the load carried by those furthest from the shear plane.

141
Chapter 4: Detailed Joint Modelling

Figure 4-15: Stress in the loading direction prior to failure initiation


showing the high stresses near the shear plane

Failure was predicted with modified 3-D Hashin fabric failure criteria. The default
stiffness degradation factors were used, and the load-displacement results are shown
in Figure 4-16. The simulation did not include element elimination, so failed elements
were retained in the simulation. The simulation predicted the onset of damage well
and the stiffness of the joint was captured extremely well. It is also apparent that the
progressive damage behaviour is not captured well at all, and the simulation quickly
diverges from the experiment after a small displacement.

20
18
16
14
Load (kN)

12
10
8
6
4 Experimental
2 Default simulation
0
0 1 2 3 4 5
Displacement (mm)

Figure 4-16: Comparison between experimental results and default simulation

An investigation into the effect of the stiffness reduction factors was conducted and it
was found that the two most significant terms were the two indices related to
compressive damage in the fibres and matrix. Both terms affected the slope of the

142
Chapter 4: Detailed Joint Modelling

load-displacement curve after damage onset. The results for the compressive matrix
damage simulations are shown in Figure 4-17.

20
18
16
14
Load (kN)

12
10
8
Experimental
6
Dmc = 0.6
4 Dmc = 0.4
2 Dmc = 0.2
0
0 1 2 3 4 5
Displacement (mm)

Figure 4-17: Effect of the stiffness reduction term for compressive matrix damage

The progressive damage algorithm suffers from a significant problem for this type of
modelling. After the stiffness is reduced by the initial damage, the material can still
carry loads in excess of the initial failure load if the strain is increased, as shown in
Figure 4-18. The simple stiffness degradation does not accurately capture material
degradation in compression.
Stress

Strain
Figure 4-18: Simple stiffness degradation leading to infinitely strong material

To improve the behaviour of the simulation, element elimination was added to remove
the highly strained elements. Elements were eliminated when the Von Mises

143
Chapter 4: Detailed Joint Modelling

equivalent strain exported from MSC.Marc [138] reached a critical limit of 20%. The
element elimination stopped the failed elements carrying continually higher loads. The
load-displacement curve for the simulation with element elimination included is shown
in Figure 4-19. It can be seen that the numerical curve does tend toward the
experimental results however the curve includes a step drop in load which is due to a
row of elements eliminating.

20
18
16
14
Load (kN)

12
10
8
6 Experimental
4 Dmc = 0.6 with elimination

2 Dmc = 0.6

0
0 1 2 3 4 5
Displacement (mm)

Figure 4-19: Progressive damage simulation with included element elimination

The introduction of element elimination caused significant convergence problems


because the contact surfaces were changing with each eliminated element and the
removal of an element caused large load discontinuities in the surrounding elements.
Attempts were made to reduce the time step and other methods were used to try and
force convergence however the simulations would not converge with element
elimination beyond ~2 mm displacement.

Although some numerical problems resulted in poor load-displacement results,


prediction of failure modes was still possible. All the primary failure mechanisms were
predicted to some extent, including the through-thickness shear crack which was
found using a CT scan of a failed joint, as shown in Figure 4-20. The blue regions
represent matrix failures, which include shear cracks and delamination. The yellow
regions represent matrix and fibre failures, which is material debris that would not

144
Chapter 4: Detailed Joint Modelling

remain attached to the laminate after the restraint of the bolt was removed. The
analysis predicted both the shear crack and debris shape quite well.

Figure 4-20: Through-thickness shear crack prediction

A fundamental problem exists with modelling compressive damage with simple


element stiffness reduction. The damaged material, although much more compliant,
can continue to carry load indefinitely and effectively becomes an infinitely strong
elastic material. This is in direct contrast to a damage mechanics type formulation
where the stiffness of the element is slowly degraded over a finite strain. A schematic
of the two methods is shown in Figure 4-21. It should be noted that the gradient of the
curves after failure depends heavily on the failure mode. Matrix failures and
compressive fibre failures generally carry more load after damage initiation whilst
tensile fibre failures result in a sharper load drop.

Stiffness
degradation
Stress

Damage
mechanics

Strain
Figure 4-21: Comparison between stiffness degradation and damage mechanics

Based on the experience gained it is concluded that the damage mechanics approach
would much more accurately predict the load-displacement response of the bolted

145
Chapter 4: Detailed Joint Modelling

joint. When comparing the simulation results to the experimental results in Figure
4-22, it is clear that the model under-predicts the stiffness of the experiment after
damage initiation but eventually over-predicts the stiffness of the model. If this is
compared to Figure 4-21, the same relative stiffness result is observed, with the simple
stiffness degradation method initially having lower stiffness than the damage
mechanics approach, but increasing instead of decreasing beyond this point.

20
18
Stiffness over-predicted
16
14
Load (kN)

12
10
8
Stiffness under-predicted
6
4 Experimental
2 Default simulation
0
0 1 2 3 4 5
Displacement (mm)

Figure 4-22: Comparison between experimental results and default simulation

In general, damage mechanics alters the material stiffness, and requires damage
parameters as a function of strain, rather than stress. A damage mechanics approach
with gradual stiffness degradation was not able to be implemented in MSC.Marc with
the authors stress-based user defined subroutine without wholesale changes to the
way that failure indices were recorded.

MSC.Marc does however include a gradual stiffness reduction algorithm [138] within
the code for use with the default failure criteria. The gradual stiffness reduction was
implemented with the maximum stress failure theory and the default gradual stiffness
reduction algorithm within Marc. The load-displacement response of this simulation is
shown in Figure 4-23. It is clear that the shape of this load-displacement curve is much
closer to that of the experiment.

146
Chapter 4: Detailed Joint Modelling

16

14

12

10
Load (kN)

4
Experiment
2
Progressive stiffness reduction
0
0 1 2 3 4 5
Displacement (mm)

Figure 4-23: Progressive damage using progressive stiffness reduction

This analysis has shown that the propagation of damage throughout the joint is as
important for predicting ultimate loads in countersunk fastener joints as the prediction
of initial damage. In countersunk fastener joints, a significant proportion of the load at
failure is carried by previously damaged material. Capturing the affect that material
damage has on the material properties is vitally important in predicting the peak load
carrying capacity of the joints. This finding has been echoed by a very recently
available paper by Hhne et al. [127].

Implicit FE analysis is extremely effective at investigating stress and strain within a


bolted joint. Initial failure, although dependent on the choice of failure criteria, is also
very well predicted with implicit FE analysis. Total energy absorption of the joints,
which is important for dynamic loading, is much more difficult to predict with implicit
FE analysis. Firstly, the load after initial failure is heavily dependent on material
degradation factors that are still not very well understood within the literature.
Secondly, and more importantly, implicit FE analysis attempts to force convergence of
loads or displacements. This convergence becomes extremely difficult and
computationally expensive when the material becomes highly distorted and damaged.
As much of the energy absorbing capability of the joints is due to material damage well
beyond failure initiation, predicting energy absorption of bolted joints with implicit FE
analysis would be an extremely challenging task.
147
Chapter 4: Detailed Joint Modelling

Two distinct methods for modelling laminated composites in detail are possible within
PAM-CRASH. The two methods differ in the type of elements used to represent the
composite material. One method, which is common to many implicit finite element
analyses of bolted joints [117, 120, 122], is to build each ply separately as a layer of
solid elements with orthotropic material properties applied. This technique is used for
the implicit finite element models presented earlier. The other method involves using
stacked shell modelling of the composite material. The composite plate is represented
by discrete sublaminates which are plates of 2-D composite layered shell elements.
The sublaminates can include any number of plies or fraction of a ply if desired. The
sublaminates are then tied together using nodal ties with special material properties.
Stacked shell approaches have been used in the past to model composite laminates
under impact but no results have been published in the general literature which use
stacked shell models for detailed contact analyses such as bolted joints. An example of
the difference between a solid model and stacked shell model for a bearing test is
shown in Figure 4-24.

a)

b)

Figure 4-24: Two representations of composite laminate in a pull-through model


showing a) solid element representation and b) stacked shell representation

Both modelling approaches have advantages and disadvantages. Solid models give
good predictions of the full 3-D stress and strain field, however they are

148
Chapter 4: Detailed Joint Modelling

computationally expensive because every ply must be added as a separate element


through the thickness. The through-thickness dimension is often the smallest within
the elements, so the time step becomes prohibitively small for most models. The solid
elements in PAM-CRASH also have only one integration point, so suffer from zero
energy hour-glassing problems [86] when subjected to contact loads. The prediction of
delamination damage is also more difficult for the solid elements.

Stacked shell models on the other hand have a discontinuity between through-
thickness loads and in-plane loads because they are carried by different elements.
They are much cheaper computationally because many plies can be added to one
sublaminate; so the number of elements required for the solution can be cut down
significantly. Another advantage of stacked shell models over solid models is the time
step. The elements which carry the through-thickness loads have a time step which is
independent of size, so the small distances between plies can be represented without
compromising the time step calculation. Stacked shell models also have the intrinsic
ability to predicted delamination damage by using cohesive zone models to tie the
shells together. Some papers [17, 99] have shown that stacked shell models with
cohesive zone interface properties predict damage and delamination in other
problems.

A summary of the benefits and drawbacks of the two methods is shown in Table 4-7. It
was decided for this research to follow a stacked shell modelling approach for the
explicit FE analyses. The slight increase in accuracy of the solid model is outweighed by
the major increase in solution time brought about by the small time step.

Solid Model Stacked Shell Model


 Good overall stress prediction  In-plane and through-thickness
 Computationally expensive properties separated
 Hour-glassing problems  Computationally cheap
 Delamination prediction is difficult  Stress field is not easily interpreted
 Intrinsic delamination prediction
Table 4-7: Comparison between solid and stacked shell modelling

149
Chapter 4: Detailed Joint Modelling

With a stacked shell model, each sublaminate is modelled by a layered shell element
which only includes the properties of the plies that belong to the sublaminate. The
sublaminates are connected by a contact-like interface to which various material
properties can be applied. A schematic representation of this is shown in Figure 4-25.

a) b)

Figure 4-25: Composite laminate represented two equivalent ways


by a) a single layered shell and b) n stacked shells with n-1 interfaces

Greve and Pickett [99] showed that the stacked shell method produces the same
elastic behaviour for a cantilever beam as single shell laminated composite elements. A
similar analysis was conducted by this author for a 2-D plate supported along two
edges, as shown in Figure 4-26.

Figure 4-26: 2-D cantilevered plate

It was shown for two or four sublaminates that the stacked shell modelling approach
provided the same elastic response as classical laminate theory (CLT), as shown in
Figure 4-27.

150
Chapter 4: Detailed Joint Modelling

Single Laminated
Composite Shell
2 Sublaminates Moment
Reaction Force or Moment
4 Sublaminates

Shear Force

Tip Displacement

Figure 4-27: Stacked shell verification for a built-in plate

The composite material model used for all FE models was a bi-phase fabric formulation
(PAM-CRASH, Material 132, Ply Type 6) [86, 88]. The bi-phase formulation considers
separate contributions from the fibre and matrix to the overall properties of the
composite. The stiffness properties are a rule-of-mixtures combination of two
orthotropic fibre phases with an equivalent matrix phase. Calibration strategies for the
elastic moduli are given in the documentation.

Each phase can accumulate damage separately. The fibre phases have a damage
function that is dependent on the strain in the fibre direction while the matrix phase
has two separate damage functions based on the equivalent shear strain and the
volumetric strain. The damage functions of each phase are defined separately but all
have the same form. The damage function and stiffness degradation used are shown in
Equation 4.9 to Equation 4.11 [86] and shown graphically in Figure 4-28 to Figure 4-30.

151
Chapter 4: Detailed Joint Modelling

!  i
# d , i " " l
l  i l
#
#  l
d    $  du  d l   d l , l " " u ...(4.9)
# u  l
#
# 1  1  du  u , u " " %
&

E    1  d    E0 ...(4.10)

    E  
...(4.11)
 1  d    E0

where

= Strain (fibre, equivalent or volumetric)


d() = Damage
E() = Youngs modulus (or shear modulus)
() = Stress
i = Initial threshold strain
l = Intermediate strain
u = Ultimate strain
dl = Intermediate damage
du = Ultimate damage
E0 = Initial elastic modulus

d()

du

dl

i l u
Figure 4-28: Strain based damage evolution [86]

152
Chapter 4: Detailed Joint Modelling

E()

E0

El

Eu

i l u
Figure 4-29: Strain based modulus reduction [86]

()

E()
u

i l u
Figure 4-30: Stress as a function of strain [86]

Layered shell elements were used to generate the laminate (or sublaminate) level
material properties. The shell elements had at least one integration point per ply, so
the local strain at each ply could be calculated accurately. Damage was accumulated at
a ply level and elements were only eliminated after all plies reached the ultimate
damage level.

Material properties were calibrated from material data sheets where available and
calibration strategies described by the CRC-ACS [139].

A linear softening cohesive zone model based on work by Johnson et al. [17] and Greve
and Pickett [99] (referred to as the Pickett Model) has been implemented in PAM-
CRASH. The stress-opening curve for Mode I fracture is shown in Figure 4-31.

153
Chapter 4: Detailed Joint Modelling

G0
max Gu

ucrit u
Figure 4-31: Mode I stress-opening function for Pickett model in PAM-CRASH [86]

The model defines two critical energy absorption modes as shown in Equations 4.12
and 4.13.

u
G I  (  (u )du
0
ucrit
G0I  (  (u )du ...(4.12)
0

'  ma
2
x

v
G II  ( ) (v )dv
0
v crit
G0II  ( ) (v )dv ...(4.13)
0

' ) max
2

u  x.n
...(4.14)
v  x  x.n

where

GI = Mode I energy absorption


GII = Mode II energy absorption
G0X = Critical energy absorption
 = Normal (opening) stress for Mode I
) = Shear (sliding) stress for Mode I
u = Mode I opening displacement
v = Mode II sliding displacement

154
Chapter 4: Detailed Joint Modelling

x = Displacement of the slave node relative to the master segment


n = Master segment normal

Note that Mode II in this model is not identical to the conventional Mode II fracture. In
this model the conventional Mode II (in-plane shear) and Mode III (anti-plane shear)
are combined vectorially into a generic shear mode. Failure initiates when

GiI GiII
 1 (4.15)
G0I G0II

The relationship between critical and ultimate energy absorption in both Mode I and II
is shown in Figure 4-32.

nd
II
Gu
II
GA
max
d
II
GB
II
G0 v
 nd
II
GB I
GA
II max
G d
I
GB
II
GA u

I I I I I
GA G0 G GB Gu

Figure 4-32: Relationship between G0 and Gu and the damage parameters [86]

The delamination material model contains many parameters which cannot easily be
determined independently from experiment. To find appropriate values for these
parameters, experimental results are used to calibrate representative finite element
models. The PAM-CRASH models used are shown in Figure 4-33.

155
Chapter 4: Detailed Joint Modelling

Process Zone

a) b)

Figure 4-33: Finite element models used to calibrate interface properties


showing a) Mode I fracture (DCB) and b) Mode II fracture (ENF)

The delamination material card is shown in Figure 4-34.

$# IDMAT MATYP RHO ISINT ISHG ISTRAT IFROZ


MATER / 1 303 0. 0 0
$# BLANK AUXVAR1 AUXVAR2 AUXVAR3 AUXVAR4 AUXVAR5 AUXVAR6 QVM THERMAL IDMPD
0 0 0 0 0 0 1. 0
$# TITLE
NAME Delamination Interface 0.11
$# SDMP1 SLFACM IDEABEN
0.1 0.1 0
$# I3DOF IDMOD
0 0
$# hcont E0 G0 D p Nfilt
0.66 9.7 3.95 0. 0.
$# SIGprpg GAMAprpg Gu' Gu" SIGstrt GAMAstrt
0.003 0.03 0.000625 0.00325 0.1 0.2
$# Ncycle Gcont
20 0

Figure 4-34: Delamination material card

The parameters from the delamination card are summarised in Table 4-8.

Default
Code Symbol Explanation
Value
I3DOF
hcont hcont Contact thickness of interface
E0 E0 Interface normal stiffness 9.7 GPa
G0 G0 Interface shear stiffness 3.95 GPa
SIGprpg p Mode I propagation stress for delamination 0.03 GPa
GAMAprpg p Mode II propagation stress for delamination 0.03 GPa
GIc
Gu' Mode I strain energy release rate 0.68 kJ/m2
(or G1c)
GIIc
Gu'' Mode II strain energy release rate 3.25 kJ/m2
(or G2c)
SIGstrt s Mode I initiation stress for delamination 0.03 GPa
GAMAstrt s Mode II initiation stress for delamination 0.1 GPa
Table 4-8: Delamination material card parameters

156
Chapter 4: Detailed Joint Modelling

PAM-CRASH uses penalty based contact. A penalty method uses reaction spring forces
to penalise penetration into a given contact volume. Friction forces can be included
with a friction coefficient or a more complex friction model. Further information about
the contact formulations can be found in the software manuals [86, 88].

For details of the experimental program and results please refer to Chapter 3.

The pull-through model used for the following analyses is shown in Figure 4-35. The
model is assembled from four independent sub-parts; the bolt, bushing, laminate and
loading plate. All parts are meshed with a structured mesh. The model utilised two
symmetry planes to improve computational efficiency. This approximation was
justified because damage evolution in the specimen was generally symmetric and the
composite laminate was a [45/0]4s fabric laminate. This laminate is identical under
reflection in planes normal to the 0 and 90 directions, so imposing symmetry
conditions on these planes does not create any mathematical discontinuities.

The bolt, bushing and loading plate were all meshed with a regular eight node hex
mesh, as shown in Figure 4-36 to Figure 4-38.

The loading plate was defined as a rigid body, with one free degree-of-freedom in the
loading direction. The plate was given a constant velocity throughout the simulation.
Unless otherwise specified this loading velocity was 5 m/s, as this velocity gave the
best computational efficiency with minimal inertial effects.

The bolt was given the properties of Ti-6Al-4V which is the aerospace grade titanium
alloy used for HiLok fasteners. The alloy was considered elastic-plastic with a linear
hardening slope. The bushing was given the properties of mild steel, which was
considered elastic as the loads involved were well below yield. The properties for the
two materials are shown in Table 4-9.

157
Chapter 4: Detailed Joint Modelling

Isometric View

Plan View

Side View

Figure 4-35: Pull-through model mesh with countersunk bolt

Side View Plan View

Figure 4-36: Pull-through countersunk bolt mesh

Figure 4-37: Pull-through bushing mesh

Figure 4-38: Pull-through loading plate mesh

158
Chapter 4: Detailed Joint Modelling

E = 115 GPa
= 0.33
Ti-6Al-4V
yield = 900 MPa
Et = 10 GPa
E = 210 GPa
Mild Steel
v = 0.33
Table 4-9: Isotropic material properties

By default the solid elements in PAM-CRASH are single integration point elements.
Significant hour-glassing was observed using these elements in contact, so a selective
reduced integration scheme was used instead, which increased the computation time
but removed any zero energy deformation modes and improved the contact pressures.

The composite laminate was modelled as a stack of laminated shell elements joined by
cohesive zone delamination interfaces, or stacked shells. The laminates were
organised according to the geometry of the laminate, hole and the locations of
expected delamination damage. A number of different shell-interface arrangements
were tested, but the best balance of accuracy to computational efficiency was found
with a five shell (four interfaces) arrangement. The arrangement of the sublaminates
relative to the ply stack and the countersunk bolt hole is shown in Figure 4-39.

45 Laminate free edge


0/90 Sublaminates
45 Virtual sublaminate
0/90 extent
45
0/90
45
0/90
0/90
45
0/90
45
0/90
45
0/90
45
Figure 4-39: Countersunk sublaminate arrangement relative to the laminate geometry

The laminate was meshed with a structured mesh of four node quad elements. The
mesh size varied with location but the average mesh size near the hole was 0.4 mm.

159
Chapter 4: Detailed Joint Modelling

The final mesh can be seen in Figure 4-40. The material properties for the composite
were as described earlier in the chapter.

Figure 4-40: Pull-through stacked shell laminate mesh

Load was transferred between all parts via penalty based contact conditions. The
specifics of the contact are shown in Table 4-10.

Contact Pair Contact Thickness Friction Coefficient


Bolt - Laminate 0.1 mm 0.18
Plate - Laminate 0.55 mm 0.18
Bushing - Laminate 0.55 mm 0.18
Ply - Ply Ply gap - 10% 0.2
Table 4-10: Contact conditions for pull-through model

The stacked shell contact extends beyond the edge of the sublaminates by a distance
equal to the contact thickness, as shown in Figure 4-41. In the simulations the bolt
diameter is reduced by 0.2 mm so that interpenetration does not occur.

160
Chapter 4: Detailed Joint Modelling

Sublaminate
contact extent

Figure 4-41: Sublaminate to solid contact for the countersunk head bolt

The tail of the bushing was held fixed while the tail of the bolt was displaced by a small
preload displacement (0.1 mm by default) and then held fixed. The loading plate was
offset from the laminate slightly such that the loading plate struck the laminate at the
same time that the preload was completed. The displacement conditions as a function
of time are shown diagrammatically in Figure 4-42.

dz

Time
dpreload

Figure 4-42: Pull-through model boundary conditions

Crosshead load was measured from the reaction force imposed by the bolt and
bushing boundary conditions. Crosshead displacement was considered to be the same
as the loading plate displacement.

A pull-through simulation was conducted with the default composite and interface
parameters. A sequence of images showing the failure progression of the simulation is
shown in Figure 4-43.

161
Chapter 4: Detailed Joint Modelling

Initial state

First fibre
failure

Delamination
initiation

162
Chapter 4: Detailed Joint Modelling

Unstable
delamination
growth

Final state

Figure 4-43: Failure sequence for the pull-through simulation

The load-displacement curve for the simulation is shown in Figure 4-44. Note that
numerical results have been filtered by a frequency filter (CFC180_10kHz) to remove
high frequency noise on the signal that is common in explicit FE analyses.

163
Chapter 4: Detailed Joint Modelling

6
Experiment 1 m/s
Experiment 5 m/s
5
Default Model

4
Load (kN)

0
0 1 2 3 4 5 6
Displacement (mm)

Figure 4-44: Default model results compared to experimental results

The stiffness of the specimen, the peak load and the general shape of the load
displacement graph are all predicted well. The initiation of final failure occurred at a
lower displacement than in the experiment and as a result the overall energy
absorption of the simulation was slightly low.

Two key failure mechanisms drive the behaviour of the specimen. The first mechanism,
which dominates the initial damage and determines the peak load carrying capacity, is
ply failure via fibre breakage. The second mechanism, which is dependent on the
extent of the in-plane damage, is Mode II delamination between plies. This
delamination determines the extent of the linear inelastic load plateau after initial ply
damage.

Damage initiates in the outer 45 ply due to bending of the laminate. The first fibres
to break are those tangential to the bolt hole on the outer face, as these fibres are
carrying the majority of the bending load. The crack quickly grows radially away from
the bolt, as shown in Figure 4-43, as well as penetrating into the laminate in the
thickness direction.

After the initial unstable crack growth, delamination begins to grow around the crack
in a typical lobe shaped region [140], as shown in Figure 4-45. Most of this

164
Chapter 4: Detailed Joint Modelling

delamination is in the two interfaces closest to the exit face of the laminate where the
crack initiated. The length of the crack is therefore critical in determining the extent of
delamination damage and energy absorption.

Crack extent No delamination (outside


coloured bands)
Delaminated area
(inside coloured bands)
Figure 4-45: Delamination growth within bottom interface

The specimen became unstable and lost integrity when the delamination areas
reached the boundary of the model. This corresponds with the load drop at around 3 -
4 mm displacement.

The influence of the delamination interface material parameters was investigated. The
parameters relating to Mode I delamination (s, GIc and p) were varied over a small
range and the sensitivity to each parameter was studied. The results are shown in
Figure 4-46 to Figure 4-48. It is clear from Figure 4-46 that initiation stress for Mode I
failure did not influence the solution. No significant change was noticed in the failure
behaviour. The energy release rate for Mode I failure also did not affect the simulation,
as shown in Figure 4-47. The Mode I delamination propagation stress had a minor
influence on the solution, but only when it was a very small value (5 MPa). It is evident
from the results in Figure 4-46 to Figure 4-48 that Mode I delamination did not play a
significant role in the failure of the pull-through simulations.

165
Chapter 4: Detailed Joint Modelling

6
s = 0.03 (default)
s = 0.05
5
s = 0.1

4
Load (kN)

0
0 1 2 3 4 5 6
Displacement (mm)

Figure 4-46: Variation of Mode I initiation stress (GPa)

6
G1c = 0.00048

5 G1c = 0.00058

G1c = 0.00068 (default)


4
G1c = 0.00078
Load (kN)

0
0 1 2 3 4 5 6
Displacement (mm)
2
Figure 4-47: Variation of Mode I energy release rate (kJ/m )

166
Chapter 4: Detailed Joint Modelling

6
p = 0.005
p = 0.01
5
p = 0.02
p = 0.03 (default)
4 p = 0.05
Load (kN)

p = 0.1
3

0
0 1 2 3 4 5 6
Displacement (mm)

Figure 4-48: Variation of Mode I propagation stress (GPa)

The parameters relating to Mode II delamination (s, GIIc and p) were varied over a
small range and the sensitivity of the simulation to each parameter was investigated.
The results are shown in Figure 4-49 to Figure 4-51.

6
s = 0.05
s = 0.1 (default)
5
s = 0.15

4
Load (kN)

0
0 1 2 3 4 5 6
Displacement (mm)

Figure 4-49: Variation of Mode II initiation stress (GPa)

It is clear from Figure 4-49 that the Mode II initiation stress did not play a significant
role in the simulation. Variation of the Mode II energy release rate did have a

167
Chapter 4: Detailed Joint Modelling

significant impact on the simulation. Increasing GIIc changed the onset point of the
delamination failure in the simulation significantly without changing the initial stiffness
or peak load. This effect can be clearly seen in Figure 4-50.

6
G2c = 0.00225
G2c = 0.00275
5 G2c = 0.00325 (default)
G2c = 0.00375
4 G2c = 0.00425
Load (kN)

Increasing G2c
3

0
0 1 2 3 4 5 6
Displacement (mm)

2
Figure 4-50: Variation of Mode II energy release rate (kJ/m )

6
p = 0.01
p = 0.03 (default)
5 p = 0.05
p = 0.1
4
Load (kN)

Increasing p
3

0
0 1 2 3 4 5 6
Displacement (mm)

Figure 4-51: Variation of Mode II propagation stress (GPa)

Varying the Mode II propagation stress also had a significant effect on the simulation.
It can be seen in Figure 4-51 that increasing the propagation stress by an order of

168
Chapter 4: Detailed Joint Modelling

magnitude significantly changed total energy absorption of the specimen, even though
the interface strain energy release rate remained unchanged. The displacement to
failure increased considerably with increasing p. Peak load carrying capacity of the
specimen increased slightly with increasing p but appears to converge to a constant
value for p above 0.05 GPa.

The results in Figure 4-50 and Figure 4-51 suggest that Mode II delamination plays a
significant role in the pull-through failure of bolted joints. More significantly, for this
laminate and boundary conditions, the initial stiffness of the specimen and the damage
initiation are relatively unaffected by the interface parameters, but the failure
initiation and total energy absorption are significantly affected by the interface
parameters.

The effect of the bolt pretension was investigated. The nut tightening displacement
was varied between 0 mm (finger tight) and 0.3 mm (the approximate preload
displacement for the current configuration). The result is shown in Figure 4-52.

6
Preload 0
Preload 0.1
5
Preload 0.2
Preload 0.3
4
Load (kN)

0
0 1 2 3 4 5 6
Displacement (mm)

Figure 4-52: Variation of bolt preload displacement (mm)

It can be seen from Figure 4-52 that the preload did affect the result of the simulation.
For a small preload (0.1 mm), the response of the simulation was very similar up until
the initiation of delamination. The preload caused the delamination to initiate slightly

169
Chapter 4: Detailed Joint Modelling

earlier. For larger preloads (0.3 mm), the preload initiated damage in the sublaminates
prematurely because the preload was large enough to cause damage around the bolt.

The observation that preload actually weakens the model rather than strengthening it
does appear to disagree with literature that claims that preload strengthens bolted
joints [24-25]. These papers do not address pull-through loads, and it is believed that
pre-tightening, although it has a positive effect on bearing strength, may have a
weakening effect on the pull-through failure of bolts, because the tightening of the
bolt imposes very similar loads on the material around the hole as those experienced
in the pull-through tests.

The true effect of preload was not possible to gauge from these simulations. Stacked
shell models do not account for through-thickness damage well, because the through-
thickness strains are only approximated from in-plane and bending deformations.
Preload in these simulations has two independent effects that in reality are not
independent. A schematic of bolt preload for a stacked shell laminate is shown in
Figure 4-53. The pressure of the bolt on the laminates produces a load in the
composite approximately normal to the surface of the bolt head. In a stacked shell
laminate this load is reacted by an in-plane compressive load in the composite material
and a through thickness compressive and shear load in the interface. This modelling
methodology cannot accurately account for the interaction between the in-plane and
interface loads.

Figure 4-53: Stacked shell laminate under preload by a countersunk fastener

The interface model is not currently programmed to account for any beneficial effect
that compressive load may have on the shear strength of the interface. The benefit of
this compressive stress could be very significant. Hou, Petrinic and Ruiz [82] have
shown that even small compressive stresses have a large restraining influence on the
growth of delamination. The Hou model penalises compressive normal loading heavily

170
Chapter 4: Detailed Joint Modelling

and even forbids delamination if the compressive normal load exceeds a certain critical
cut-off curve. The failure surface for the Hou model is shown in Figure 4-54.

Shear Load Failure Surface


Compressive Delamination Cutoff

Normal Load
Figure 4-54: Hou delamination model

The interface model could not account for the restraint provided by the preload, a
separate set of interfaces were created in the vicinity of the bolt with artificially
enhanced properties, as an attempt to simulate the effect identified by Hou. The extra
interfaces were added within the compressive cone that the bolt pretension provided.
The size of this region was found by running a preload analysis on the default model
and selecting the links which had compressive loads of greater than 20% of the
maximum compressive load. The extra links are shown in Figure 4-55.

Figure 4-55: Location of extra delamination interface links

171
Chapter 4: Detailed Joint Modelling

The extra links had the twice the strength for a given opening or sliding displacement,
effectively tripling the strength of the interface near the bolt. This was considered
reasonable as the normal loads were generally significantly above those where
delamination was considered impossible.

The extra interface strength representing the preload effect influenced the response of
the model. The load displacement curves were not significantly changed apart from
the onset and slope of the final failure, as shown in Figure 4-56.

6
No extra links
Extra links
5

4
Load (kN)

0
0 1 2 3 4 5 6
Displacement (mm)

Figure 4-56: Comparison between the default pull-through load response


with and without extra interface links

The damage pattern near the hole changed significantly. The delamination which
started around the main crack did not reach the hole surface for most plies. Instead, a
cone of undamaged material split from the rest of the laminate. A comparison
between this damage mode and that of the default model is shown in Figure 4-57.

172
Chapter 4: Detailed Joint Modelling

a)

b)

Figure 4-57: Comparison between the default pull-through damage


a) with and b) without extra interface links

A comparison between the experimental results at two different loading rates and the
numerical predictions with extra interface links is shown in Figure 4-58.

6
Experiment 1 m/s
Experiment 5 m/s
5
Default Model
Numerical with extra links
4
Load (kN)

0
0 1 2 3 4 5 6 7
Displacement (mm)

Figure 4-58: Comparison between the experimental and numerical pull-through load response

The external damage to the specimen in the simulation is very similar to that of the
real specimen, as shown in Figure 4-59. The extent of the delamination damage and
the location of the through-thickness cracks agreed well with the experimental results
obtained from a CT scan, as shown in Figure 4-60.

173
Chapter 4: Detailed Joint Modelling

a) b)

Figure 4-59: Comparison between experimental and numerically predicted external damage
showing a) failed specimen and b) numerical prediction

a)

b)

Figure 4-60: Comparison between experimental and numerically predicted delamination damage
showing a) CT scan of failed specimen and b) numerical prediction

It was found that there were two main failure drivers for pull-through failure in the
tested specimens. The first damage that occurred was due to fibre breakage on the
exit face of the specimen. The fibre breakage led to a crack propagating radially away
from the bolt and penetrating into the laminate considerably. The second damage
mechanism was Mode II delamination, which initially propagated steadily within the
laminate until the damage reached a critical limit when the remaining bending
strength of the delaminated plies could not support the load being carried. Banbury et

174
Chapter 4: Detailed Joint Modelling

al. [21] found similar failure mechanisms, although in the test geometries used in those
experiments, fibre failure was not the initial damage mechanism.

It was found that accounting for bolt pretension was important to achieve the correct
failure mode. The current cohesive zone interface models in PAM-CRASH do not have
any compensation for compressive normal loads so the beneficial effects of bolt
tightening cannot be accounted for directly. Extra interface elements were added to
the regions of the laminate which were found to be under compression during a
conventional pre-stress analysis. These extra elements improved the damage
predictions, however they required the assumption that the compressive zone did not
change throughout the simulation. This assumption was not entirely valid, and a much
better approach would be to implement compressive behaviour into the cohesive zone
interface model.

Significantly, it was found that Mode I interface parameters had very little influence on
the simulation, while the Mode II parameters did have a significant influence,
especially the SERR and the propagation stress. It was found that the trend of the
experiments with increasing loading rate agreed with the trend of simulations with
reducing Mode II interface parameters. This suggests that the Mode II fracture
toughness of the material dropped with increased loading rate. A diagram of the effect
on the simulations is shown in Figure 4-61.

The stacked shell method does have some drawbacks over regular 3-D element
methods. The stacked shell models do not explicitly model through-thickness shear
failures such as those seen by Banbury et al. [21]. Instead, the failures are accounted
for by interface shear damage and sublaminate bending damage. This method would
be less suited to thick pull-through specimens as the failure mechanisms are more
dominated by through-thickness shear cracks.

175
Chapter 4: Detailed Joint Modelling

Load

Displacement
Figure 4-61: Effect of increasing experimental loading rate or decreasing G II interface properties

For details of the experimental program and results please refer to Chapter 3.

The finite element model for the countersunk head bearing model is shown in Figure
4-62. The model was assembled from five independent sub-parts; the bolt, collar, top
laminate, bottom laminate and support rig. All parts were meshed using a structured
mesh. The model used a single plane of symmetry to reduce computational cost. The
mesh applied to each sub-part is shown in the following pages.

The countersunk bolt and collar were meshed with a regular eight node hex mesh. The
bolt mesh is shown in Figure 4-63. The collar mesh is shown in Figure 4-64. Both the
bolt and collar were given the properties of Ti-6Al-4V as shown earlier in Table 4-9.
The mesh used for the top and bottom laminates is shown in Figure 4-65 and Figure
4-66 respectively. The mesh had two distinct sections. Close to the bolt the mesh was
dense and not biased in the radial direction so that all the elements in this region had
the same radial dimension. Outside the high density region the mesh was biased so
that large elements were used away from the hole to reduced computational effort.

The stacked shell elements that composed the top plate were adjusted to allow for the
countersunk bolt geometry. The through-thickness location of each sublaminate in the
stack was the same as that used for the pull-through model, as shown earlier in Figure

176
Chapter 4: Detailed Joint Modelling

4-39. The number of element seeds used radially to generate the localised mesh
refinement around the bolt was the same as for the bottom laminate. The average
element length near the bolt was 0.4 mm while the largest element length was
approximately 6 mm.

A rigid support was added to represent the test fixture used for the tests, as shown in
Figure 4-67.

Plan View

Side View

Isometric View

Figure 4-62: Bearing model mesh with countersunk bolt

177
Chapter 4: Detailed Joint Modelling

Side View Isometric View

Figure 4-63: Bearing countersunk bolt mesh

Plan View

Isometric View

Figure 4-64: Bearing collar mesh

Plan View

Detailed Isometric View

Figure 4-65: Bearing top laminate mesh for countersunk bolt

178
Chapter 4: Detailed Joint Modelling

Plan View

Isometric View

Figure 4-66: Bearing bottom laminate mesh

Plan View

Isometric View

Figure 4-67: Bearing test support mesh

Contact was used to transfer load between the separate sub-parts. The specifics of
each individual contact are shown in Table 4-11.

179
Chapter 4: Detailed Joint Modelling

Contact Pair Contact Thickness Friction Coefficient


Bolt - Top 0.1 mm 0.5
Bolt - Bottom 0.1 mm 0.5
Collar - Bottom 0.33 mm 0.15
Top - Bottom 0.88 mm 0.2
Top - Support 0.22 mm 0.15
Bottom - Support 0.33 mm 0.15
Table 4-11: Contact conditions for protruding head bearing model

Contact with stacked shell laminates is not as simple as for solid-solid contact. Contact
can occur parallel or normal to the sublaminate plane as shown in Figure 4-68. It was
found that larger coefficients of friction were required for the normal contact to
achieve reasonable results, most likely due to a difficulty in calculating the correct
contact area for this type of contact.

Normal Parallel
contact contact

Sublaminate
Sublaminates
contact extent

Figure 4-68: Sublaminate to solid contact

The load was applied in two steps. Firstly a preload was applied by displacing the collar
relative to the bolt. The bolt and collar were then fixed together and the load was
applied to the free end of the joint. The clamped end and the support were held fixed
throughout the simulations.

180
Chapter 4: Detailed Joint Modelling

The stacked shell modelling approach gave reasonable predictions of the failure
progression and failure load for the countersunk bearing test. As failure in the bolt was
not being explicitly modelled, the prediction of the two different failure modes was not
possible, however the progressive bearing failure mode was predicted quite well
during the initial phase of failure. As the failure progressed and the effect of material
debris became more important the simulation was less able to predict the correct
failure progression.

The initial stiffness of the explicit FE model agrees well with the quasi-static tests but
poorly with the dynamic tests, as shown in Figure 4-69. A load-displacement curve for
the stacked shell countersunk bearing model is shown in Figure 4-70. Due to the
unresolved stiffness mismatch between the dynamic tests and the quasi-static tests,
the dynamic experimental results have been scaled in the elastic region to match the
quasi-static test data. After the point of damage initiation the displacement remains
unscaled.

12

10

8
Load (kN)

4
Quasi-static test
2 Low rate dynamic test
Numerical Simulation
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Displacement (mm)

Figure 4-69: Stiffness correlation between the quasi-static tests and the numerical predictions

181
Chapter 4: Detailed Joint Modelling

16
Experimental (10 m/s)
14 Numerical

12

10
Load (kN)

0
0 5 10 15 20 25
Displacement (mm)

Figure 4-70: Comparison between experimental load and


numerical prediction for the countersunk bearing model

A failure progression for the simulation is shown in Figure 4-71. The failure mode
followed a very similar progression to that of the test. The first damage occurred as
compressive intralaminar damage around the bolt head, shown in Figure 4-72,
followed by through-thickness damage initiating near the edge of the collar, as shown
in Figure 4-73. Once the progressive bearing failure had triggered in the tail side of the
laminate, the analysis soon diverged from the experimental results.

182
Chapter 4: Detailed Joint Modelling

Figure 4-71: Failure sequence for the countersunk bearing simulation

183
Chapter 4: Detailed Joint Modelling

a) b)

Figure 4-72: Bearing failure around the bolt head showing


a) experimental result and b) numerical prediction

a) b)
Delamination

Delamination

Figure 4-73: Delamination failure around the bolt collar showing


a) experimental result and b) numerical prediction

The stacked shell modelling method provided a good prediction of the failure
progression of the countersunk head fastener bearing tests. Failure mechanisms were
quite well predicted and the overall load-displacement response was quite good for
the beginning of the simulation.

Two factors drove the divergence between the experimental and numerical results
during this phase of the analysis. Firstly, the size of the elements that the bolt was
damaging became too large for the cohesive zone to work effectively, hence increasing
the progressive bearing load. Secondly, all the elements that were eliminated are not
acting effectively as debris as they do in the experiments, so the final failure mode was
not correctly predicted.

184
Chapter 4: Detailed Joint Modelling

Another drawback of this modelling approach is that the through-thickness damage is


constrained to occur within the interface elements. The shell elements which
represent the sublaminates do not calculate through-thickness shear or direct stress
accurately. Failure modes which depend on these loads are not well captured, such as
the penetration of the bolt collar into the laminate.

An internal self contact was used to model material debris from crushed elements. The
simulation was quite sensitive to the parameters of this contact. If the contact zone
was set too large, then progressive bearing damage was very limited. If the contact
zone was set too small then the elements often became unstable and were eliminated,
over-predicting bearing damage. A better capability to model material debris is
required to accurately predict both the initial stages when the debris is constrained,
and the later stages when the debris is free to escape. A possible solution to this
problem is coupling smooth particle hydrodynamics (SPH) to the FE method, as
proposed by Aktay and Johnson [141]. This method was not able to be implemented
for this research.

Overall the stacked shell modelling method provided a good tool to simulate damage
in bolted composite joints. If the overall energy absorption of the specimen was to be
accurately predicted then the mesh would have to be more carefully constructed so
that the element size throughout the bearing region was kept small enough that the
cohesive zone formulation converged.

185
Chapter 4: Detailed Joint Modelling

Stacked shell models offer a computationally efficient solution to solving problems


involving significant material delamination. Elastically the stacked shell laminates
behave identically to single shell elements formulated with CLT. The stacked shell
model has the intrinsic ability to model delamination damage through the cohesive
zone elements which are used to connect the sublaminates together. When a
significant number of sublaminates are used, ie four or more, then the majority of the
bending loads carried by the sublaminate stack are reacted by direct loads at the shell
nodes rather than local bending moments, as shown in Figure 4-74. Consequently the
laminate behaves more like a solid element model. It was found that stacked shell
models actually have slightly poorer bending performance when two sublaminates are
used because both the local bending moments and the direct loads have a significant
influence.

a)

b)

c)

Figure 4-74: Bending of stacked shell laminates showing a) real internal


bending load reactions, b) bending load reaction of shell elements
and c) bending load reaction in stacked shell models.

It was found that stacked shell models, as expected, do not predict through-thickness
shear failures as well as solid models. Failures generally initiate as local Mode II
interface failures after which the unsupported sublaminates fail in bending, as shown

186
Chapter 4: Detailed Joint Modelling

in Figure 4-75. During this study, this more complicated failure process over-predicted
the failure load and energy absorption of this failure mechanism. Through thickness
shear failures of solid elements would be expected to absorb significantly less energy.

Figure 4-75: Failure sequence for through-thickness shear damage in stacked shell laminates

The implementation of stacked shell modelling in PAM-CRASH has a few problems that
limit the possible applications of the modelling technique. The most significant issue
relates to the behaviour of a stacked shell laminate under compressive through-
thickness loads.

The link elements have a normal stiffness which can be defined by the user, but the
sublaminates also have a contact for which the stiffness cannot be directly user
controlled. The contact stiffness is usually at least an order of magnitude larger than
the link stiffness. This leads to significant instabilities when a compressively loaded
stacked laminate delaminates in shear, as would happen, for instance, during an
impact simulation or a bolt pull-through simulation. There are many options for
controlling the size of the contact region and when it is active however all of them
have significant drawbacks. The author suggests a new contact formulation is required
for stacked shell modelling that has definable contact stiffness.

Detailed modelling of bolted joints in composite materials under pull-through and


bearing loading was conducted with implicit and explicit FE codes. The simulations

187
Chapter 4: Detailed Joint Modelling

used 3-D and 2.5-D models including complex material properties and contact
interactions.

Implicit FE analysis was used to model bearing failure in a countersunk fastener joint. It
was shown that implicit FE modelling is excellent for understanding the stress and
strain fields within a bolted joint, including the effects of pre-load. With the selection
of a good set of failure criteria, implicit FE modelling is a good tool to predict the onset
of failure in a bolted joint. The prediction of the ultimate strengths of bolted joints or
total energy absorption is significantly more difficult with implicit FE analysis. The high
distortion of elements, large stiffness changes and the elimination of elements cause
significant problems for numerical convergence, and often the load steps need to be
prohibitively small to attain convergence.

Explicit FE analysis was used to model bearing failure and pull-through in a


countersunk fastener joint. The explicit FE analysis used a stacked shell modelling
approach to represent the composite material. Stacked shell models are built from
multiple layers of shell elements called sublaminates which are connected using tied
interface connections. Stacked shell models have the intrinsic ability to model
delamination damage due to the cohesive zone formulation applied to the tied
interface elements.

Prediction of failure onset, damage progression and total energy absorption for the
pull-through models was excellent. Some deficiencies in the compressive behaviour of
the tied interface element formulation were overcome by including extra interface
links within the preload region of the laminate.

It was shown that stacked shell models could also be used to predict bearing failure in
composite joints. The models were able to predict the onset of damage, and failure
modes, as well as the ultimate load of the joint. The model used was not able to
accurately predict the progressive bearing damage that was present in the
experiments. It was postulated that improved debris models were required to
accurately predict this damage progression.

188
An image from a structural test loaded at 1 m/s. The damage shows the importance of
joint modelling to the failure prediction of composite structures. The flash indicates
debris being ejected has been heated by frictional effects and is spontaneously igniting.
Chapter 5: Structural Testing

Single fastener joint tests offer a great deal of information about the properties of
bolted joints. One drawback of a single-fastener test is that the boundary conditions
are highly idealised to comply with testing standards and maintain consistency among
all tests conducted globally. These idealised boundary conditions, although useful for
characterising the base joint behaviour, do not accurately represent the true
conditions imposed on a bolted joint in a composite structure. Single-lap bearing tests,
for instance, are conducted using a guide that restrains the composite plates from out-
of-plane deformation.

In real structures however, a single-lap joint experiences significant out-of-plane


bending, due to load realignment, and the failure modes can be much more complex.
The through-thickness offset between the two plates creates a bending moment which
must be counteracted by a bending moment carried by the bolt. This secondary
bending has an impact on the failure loads and modes for the overall structure. An
illustration of this effect is shown in Figure 5-1.

Figure 5-1: Load alignment and secondary bending

Another limitation of the idealised boundary conditions is that they do not account for
any interaction between normal and shear loads. A bearing test and a pull-through test
for the same laminate and bolt combination can yield an understanding of the joint
strength in the limiting load cases, but no information about the intermediate
combined loading condition is obtained as represented by Figure 5-2. Some single
fastener joint testing has been conducted for metallic bolted joints to investigate
'mixed mode' loading using a modified Arcan-type test rig [142], however no research
in this area has been found that involves composite materials.

190
Chapter 5: Structural Testing

a)
c)

b)

Figure 5-2: Combined loading of bolted joint showing a) pure normal loading
b) pure shear loading and c) unknown effect of combined loading [61]

Structural testing extends beyond the limits of single-fastener joint testing because it
allows the bolted joints to behave as they would in a real structure. Boundary
conditions still play a role, and must be carefully considered, but the behaviour of the
bolted joint can be observed in a condition which is further removed from the
influence of the boundary. The corollary of this is that the behaviour of each individual
fastener is more difficult to observe, and experimental results tend to be a combined
response of the entire structure rather than of the individual fastener properties.

A number of structural tests were conducted to investigate the behaviour of fasteners


when not loaded under ideal conditions. The test geometry and loading conditions
were constructed in such a way that the bolts experienced a combination of normal
and shear loading, as well as being free to deform out-of-plane. The observed failure
modes will shed some light onto the behaviour of bolted joints under these complex
loading conditions.

Many factors influenced the design and final geometry of the structural test
specimens. The key factors that drove the final geometry are discussed briefly in Table
5-1.

191
Chapter 5: Structural Testing

To investigate more complex joints to determine whether loading rate


Multiple dependencies are evident in non-idealised joints. It was also
Fasteners important that the joints were loaded in mix of bearing and pull-
through loadings.
Consistent The structural tests maintained the same joint parameters as the
Joint high-rate single-fastener tests (Bolt type and diameter, laminate
Geometry layup, edge distance etc.)
Flat composite panels are much easier and cheaper to manufacture
Flat Panels than more complex shapes. They can also be reliably modelled with
shell (or stacked shell) elements.
Flexibility At least two test configurations were desired for this test series.
of Loading Flexible loading conditions allow the specimen geometry to be used
Conditions for different tests in the future if the need arises.
Table 5-1: Key drivers for panel design

The eventual geometry selected consisted of two rows of single lap joints joining three
flat composite plates. There were also two tabs added to the specimen for one of the
test configurations. Extensive pre-test simulations were conducted to determine the
best dimensions for the panel. The final geometry is shown in Figure 5-3.

76
38

190

38

76

19

160

19 19

Tab
100 19 19 Tab
160

Figure 5-3: Structural geometry (all dimensions in mm)

The nominal thickness of each plate was 3.52 mm (16 ply T300/Cycom970 Plain
Weave). The lay-up was identical to that used for the single fastener tests.

192
Chapter 5: Structural Testing

Nine tests were conducted in total. A test matrix is shown in Table 5-2. The tests were
conducted with two different loading geometries, a line load configuration and a ball
loaded configuration. The quasi-static tests were conducted with three different bolt
configurations. A protective aluminium plate was used in some tests to force failure at
the bolted joints.

Loading Method Number of Bolts Loading Rate (m/s) Protective Plate


Line 5 Quasi No
Line 3 Quasi Yes
Line 2 Quasi Yes
Line 3 0.1 No
Line 3 1 Yes
Line 3 10 Yes
Ball 3 0.1 N/A
Ball 3 1 N/A
Ball 3 10 N/A
Table 5-2: Test matrix for structural specimen

Please note that all structural testing for this chapter was completed at DLR, Stuttgart.
Dynamic tests were completed using the same Instron VHS 100/20 used for the single
fastener tests.

A test rig was designed to experimentally replicate a rotationally free boundary


condition. Two opposite edges of the specimen were clamped in special brackets
which included regions of circular cross-section that allowed the edges to rotate
without having translational freedom. The load for the quasi-static tests was applied
via a cylindrical section that loaded the plates along a single line perpendicular to their
longitudinal direction. This arrangement is shown in Figure 5-4. A picture of the rig
prior to and during testing is shown in Figure 5-5. The panels were loaded at 10
mm/min for all quasi-static tests.

193
Chapter 5: Structural Testing

Figure 5-4: Test rig and loading direction

a) b)

Figure 5-5: Line loaded test rig a) prior to and b) during testing

Three line loaded tests were conducted quasi-statically, with three different joint
configurations. The load and displacement for each test were measured at the Instron
cross-head, and are shown in Figure 5-6.

194
Chapter 5: Structural Testing

30
5 Bolts
3 Bolts
25
2 Bolts
Crosshead Load (kN)
20

15

10

0
0 10 20 30 40
Crosshead Displacement (mm)

Figure 5-6: Load-displacement curves for quasi-static line-loaded tests

Each test had a distinct failure mode. Initial simulations suggested that a joint of five
bolts would be strong enough to resist failure, as the panel under the loading cylinder
would fail prior to the bolted connection. This prediction was verified by the five bolt
test, and the failed specimen is shown in Figure 5-7. The failure mode is a brittle
composite bending failure.

Figure 5-7: Brittle composite failure mode of the five fastener test

A sequence of images from the three fastener test is shown in Figure 5-8. The final
image was captured just prior to catastrophic specimen failure.

195
Chapter 5: Structural Testing

a)

b)

c)

d)

Figure 5-8: Test sequence for the quasi-static line loaded test with three fasteners

196
Chapter 5: Structural Testing

To ensure that the three bolt test failed at the bolted connections, a 5 mm thick
protective aluminium plate was added to the test configuration under the loading
cylinder. This plate ensured that no bending failures occurred in the composite plate.

The failed test specimen for the three bolt test is shown in Figure 5-9.

a)

b) c)

Figure 5-9: Failure mode for the three fastener test

The failure mode for the two bolt test is shown in Figure 5-10. Both joint failure modes
were characterised by initial bearing damage and bolt rotation brought about by the
high shear forces in the joint. This bolt rotation is visible in the final angular set of the
bolts in the two bolt test, as shown in Figure 5-11.

197
Chapter 5: Structural Testing

a)

b) c)

Figure 5-10: Failure mode for the two fastener test

Figure 5-11: Angular displacement of bolt

198
Chapter 5: Structural Testing

Distinct load drops were observed prior to final failure. On revision of the photographic
evidence post-experiment, it was revealed that these were not due to any failure of
the panel, simply a deficiency of the test rig. The friction between the cylindrical rollers
and the supports was high enough that the rotation caused the panel to climb the
support slightly. This effect can be observed in Figure 5-12a, which clearly shows the
roller climbing the grip.

a) b)

Friction Clearance No Clearance


Figure 5-12: Friction effect in the line loaded test rig
showing a) gripped case and b) adjusted case

The effect of this climbing is that the cross-head load is artificially increased relative to
the ideal situation. The load drops on the curves in Figure 5-6 are simply the points
where the extra displacement is released and the part drops down to its intended
position as seen in Figure 5-12b. This tarnishes the data slightly but in the two tests
which included bolt failure, it occurred well prior to final failure of the panel. The
friction existed throughout the test, so it is likely that the load at the cross-head is
always slightly inflated for a given cross-head displacement, and a curve through the
local minima of the test curves is likely to be much more representative of the desired
load-displacement curve.

Under high loads, the bolt rotation allowed the top and bottom plates to separate
significantly, which is visible in Figure 5-13.

199
Chapter 5: Structural Testing

Large positional and


rotational offset

Figure 5-13: Joint separation during loading

It is clearly visible that the fasteners experienced deflection and load both in the shear
direction and normal direction, which meets the desired outcome of the boundary
condition selection. It was also observed that the combined shear, normal and bending
moments carried by the fasteners subjected the joint to 'prying' loads not often
accounted for in failure characterisation of bolted joints. A diagram of how prying
loads are reacted in a bolted joint is shown in Figure 5-14.

Contact

Bending moment reacted


by a tension on the bolt
and contact forces
between the plates

Figure 5-14: Prying loads on bolted joints

It is difficult to standardise a test for prying failure, because the loads involved depend
heavily on the geometric parameters of the joint. The bending moment is being
reacted by the laminates away from the bolt as well as the bolted region. The prying
loads magnify the pull-through loading on the joint.

Prying is not the only significant effect that arises under the loading conditions that
were present in the test. Joint rotation and separation shown in Figure 5-13 changes

200
Chapter 5: Structural Testing

the effect of friction within the joint. Friction has a positive effect on both the strength
and stiffness of the joint [143], and is aided by appropriate clamping forces on the
bolts [25, 144]. The quasi-static test results indicate that the joint was loaded such that
the friction force was removed causing in a significant reduction in the joint
performance.

As the load was increased, the tensile forces carried by the joint became significant,
and the failure progressed into a mode that more closely resembled pull-through
failure. If a section could taken through the bearing plane of a pull-through test and
the line-loaded structural test, very similar delamination damage would be present, as
shown in Figure 5-15. This suggests that the normal load is contributing significantly to
the failure mode.

a)

b)

Figure 5-15: Similarities between the observed damage and pull-through damage
showing section through a) line-loaded test and b) pull-through test

Joints on both sides of the structure, not just the failed side, exhibited evidence of
bearing failure. This supports the idea that the joints could carry load well after the
onset of damage, as a catastrophic failure mode would have not allowed damage to
grow in both joints simultaneously.

201
Chapter 5: Structural Testing

To investigate the behaviour of the joints under high strain-rate loading, a number of
tests were completed at varying loading rates up to 10 m/s. All critical dimensions of
the test rig were identical to the quasi-static rig. The rotational constraints were
modified to allow the rig to be used upside down, as was necessary for the high-rate
Instron testing machine. A schematic of the rig is shown in Figure 5-16. Due to the high
cost of manufacturing each panel, only one test was conducted at each speed, as
shown in Table 5-2.

Figure 5-16: Test fixture for line loaded high-rate tests

The load was applied via a cylindrical impactor and was measured through the load cell
in the driving piston of the Instron test machine. The test was filmed with a high-speed
camera at 6000 fps. In the case of the 1 m/s test, failure was forced to occur at the
bolted connection by protecting the composite panel with a 5mm thick aluminium
plate, as was used in the quasi-static test. The 0.1 and 10 m/s tests were loaded
directly by the cylinder onto the panel. The set-up of the test rig in the high-rate
Instron is shown in Figure 5-17.

202
Chapter 5: Structural Testing

Figure 5-17: High-rate line load test fixture

The load-displacement results for the tests at three different loading rates are shown
in Figure 5-18. It should be noted that the 0.1 m/s test and the 10 m/s test can be
directly compared, however the 1 m/s test included the protective aluminium plate so
the load application to the composite plate differed significantly from the other two
tests.

20
0.1 m/s Test
1 m/s Test
10 m/s Test
15
Crosshead Load (kN)

10

0
0 10 20 30 40
Crosshead Displacement (mm)

Figure 5-18: Load-displacement curves for dynamic line loaded tests

A series of still images taken from the video describe the failure process for the 1 m/s
test, as shown in Figure 5-19.

203
Chapter 5: Structural Testing

a)

b)

c)

d)

e)

f)

Figure 5-19: Test sequence for the 1 m/s line loaded test

204
Chapter 5: Structural Testing

Images from the 0.1 m/s test are shown in Figure 5-20. The images only show the
failure process, as the deflection during loading was similar to the 1 m/s test.

a)

b)

c)

Figure 5-20: Test sequence for 0.1 m/s line loaded test

Some images from the 10 m/s test are shown in Figure 5-21. The load-displacement
data for the 10 m/s test is clearly very noisy and not easily filtered for analysis,
although the data does follow the same general trend as the lower speed tests. The
large scale oscillation superimposed on the data is due to bending wave propagation in
the specimen after impact. The three images in Figure 5-21 are snapshots over a small
time difference equating to a cross-head displacement of approximately 5 mm. From
inspection of the curvature of the panels in the image and the relative rotation of the
panels in the overlap region, a bending wave could be seen propagating from left to
right in the structure. This bending wave was the cause of the oscillating load-
displacement signal and also the premature failure of the central plate at a lower
displacement. The observed peak-to-peak oscillation period was 0.5 m/s.

205
Chapter 5: Structural Testing

a)

b)

c)

Figure 5-21: Bending wave propagation in 10 m/s test

The two unprotected tests exhibited no difference in failure mode, and both failed in
the central composite panel as shown in Figure 5-22. The 0.1 m/s test also showed
significant evidence of bearing damage in the bolted connections as shown in Figure
5-23, even though the main failure mode was in the composite panel. The initiation of
bearing damage suggests that the failure load would not have been significantly higher
in this test if bolt failure had been forced to occur by the use of the protective
aluminium plate.

206
Chapter 5: Structural Testing

a)

b)

Figure 5-22: Failure mode of unprotected composite plates showing


a) 0.1 m/s test specimen and b) close up of failure site

a) b)

c) d)

Figure 5-23: Evidence of bearing damage in 0.1 m/s test


a) permanent offset of top and bottom laminates
b), c) and d) localised bearing damage around holes

207
Chapter 5: Structural Testing

The failure mode for the 1 m/s test was at the bolted connection and a large piece of
material delaminated and fractured away from the central plate, as shown in Figure
5-24.

Figure 5-24: Failure mode of plate supported 1 m/s test

The two tests conducted dynamically, at 0.1 m/s and 10 m/s, were conducted without
the protective aluminium plate. In this configuration the load on the composite plate
was more concentrated and preliminary modelling suggested that the central panel
would fail just prior to the bolted connection. Elastically however, the response of the
panel was very similar to the protected configuration, as shown in Figure 5-18. The
failure modes were very similar, however the 10 m/s test failed prematurely due to
bending waves in the specimen. This test highlighted one of the difficulties of dynamic
testing; separating the intended applied load from the large scale vibration response of
the structure.

The load deflection responses of the quasi-static and 1 m/s test are shown in Figure
5-25. The responses of the two tests were very similar, despite the four orders of
magnitude difference between the loading rates of the two tests. The peak load and
deflection at failure for the two tests only differed by a small margin. The quasi-static
response generally has a higher load throughout the test, but it is believed that this
was due to the friction effects of the test rig described earlier. A plot through the local
minima of the quasi-static test would agree much more closely with the high-rate test.

208
Chapter 5: Structural Testing

20
Quasi-Static
18 Experiment
16 Dynamic Experiment
Crosshead Load (kN) 14
12
10
8
6
4
2
0
0 10 20 30 40
Crosshead Displacement (mm)

Figure 5-25: Quasi-static and High Rate Test Comparison

If Figure 5-9 is compared with Figure 5-24, it is evident that the failure modes of the
quasi-static test and the 1 m/s test are very similar. When the load-deflection
response, the video evidence and post-failure examination of the specimens was taken
into account, it was concluded that there was no discernable rate effect present for
this structure. This is a significant result, because the loading rates across the joints in
the line loaded structure were well in excess of the cross-head speed. If the dynamic
single fastener joint tests were used as an indication of the failure mode for this test,
then rate sensitivity would have been expected. None of the bolts failed in the high
energy bearing failure mode witnessed in the single fastener tests.

The line loaded rig produced relatively uniform loading in each of the fasteners and
there were minimal edge effects. To investigate the effect of a more complex loading
on a structure, a spherical impactor was used to impact the plate to promote double
curvature in the panel. The more complex loading gave an insight into the behaviour of
the joint when a non-uniform load existed across the bolted connection.

209
Chapter 5: Structural Testing

The test rig used was a simply supported frame with the two transverse edges securely
clamped. The arrangement is shown in Figure 5-26.

Figure 5-26: Test fixture for point loaded high-rate tests

To maintain a flat base to fit the test fixture, the point loaded specimens had 50mm
tabs bonded to the underside of the top panels, as shown in Figure 5-27. A 50 mm
diameter ball was used to impact the specimens at 0.1, 1, and 10 m/s. A high speed
camera filming at 6000 fps was used to collect visual data and the impact load-
displacement pulse was recorded directly from the testing machine.

Tabs

Figure 5-27: Tab locations for point loaded test

The specimen in the test rig is shown in Figure 5-28. All surfaces have been coated in
white spray chalk to enhance the light levels for the high rate video recording. The
back side of the specimen was filmed during the test as most of the failure was
expected to happen on the opposite side to the initial point of impact.

Figure 5-28: Specimen in test rig

210
Chapter 5: Structural Testing

The load-displacement curves for the three tests at different loading rates are shown
in Figure 5-29. It is clear that the 0.1 m/s test and the 1 m/s test are similar throughout
the majority of the failure regime. The 10 m/s test had a large amplitude oscillation
over the load signal. This was due to bending wave propagation in the specimen.

14
0.1 m/s
1 m/s
12
10 m/s
Crosshead Load (kN)

10

0
0 10 20 30 40
Crosshead Displacement (mm)

Figure 5-29: Dynamic ball loaded test comparison

A series of images taken from the test footage for the 1 m/s test is shown in Figure
5-30. Upon initial impact the whole structure was placed under double curvature. The
first external failure was a splitting of the fibres in tension on the top face of the
impacted panel. This failure event corresponded with the image in Figure 5-30c. The
failure quickly spread to a large central crack in the longitudinal direction of the
specimen through all plies and smaller cracks running radially from the impact location
on the rear (top) face of the laminate. Large delamination regions spread throughout
the panel and the rear sublaminate of the panel lifted and separated from the front
(bottom) sublaminate significantly. The rear pieces, unable to support the bending
loads then failed in bending and the load on the impactor dropped to nearly zero. The
structure responded in essentially the same manner for the 0.1 m/s test. The 10 m/s
test was dominated by bending waves as was the 10 m/s line loaded test. A failed
specimen is shown in Figure 5-31.

211
Chapter 5: Structural Testing

a)

b)

c)

d)

e)

f)

Figure 5-30: Test sequence for the 1 m/s ball loaded test

212
Chapter 5: Structural Testing

a)

b)

Figure 5-31: Representative panel damage to the ball loaded test


showing a) front (bottom) side and b) back (top) side

As in most impact events, the damage was significantly more pronounced on the
opposite side of the laminate to the impactor. One key feature to note from Figure
5-31b is that the large central crack did not pass through the bolt hole, which would
have provided the greatest stress concentration. It is likely that the bolt clamping
forces provided enough compressive through-thickness restraint that the damage
propagated away from the fastener.

213
Chapter 5: Structural Testing

A CT scan of a failed ball impact test was conducted to determine the extent of the
damage within the composite material. A zoomed out view of the failed panel can be
seen in Figure 5-32.

Figure 5-32: Overview of scanned ball loaded test specimen

Many artefacts are present in the scan, due to the presence of the high atomic mass
titanium bolts beside the low atomic mass composite material. The scan was very good
for tracking the gross damage in the composite panel, including the large delamination
and bending crack. Very little information regarding bearing damage around the bolt
could be obtained due to beam hardening effects.

The 0.1 m/s and 1 m/s test data is presented separately in Figure 5-33. It is clear that
the load displacement response of the two tests was very similar, as well as the total
energy absorption.

214
Chapter 5: Structural Testing

8
0.1 m/s
1 m/s

6
Crosshead Force (kN)

0
0 10 20 30 40
Crosshead Displacement (mm)

Figure 5-33: Low and Intermediate rate ball loaded test data

The general shape and response of the ball loaded structure is very similar to a pull-
through test. If the curve is compared to the dynamic pull-through test results in Figure
5-34, it becomes clear that there are some strong similarities between the two test
configurations.

5
1 m/s
5 m/s
4
Gauge Load (kN)

0
0 2 4 6 8
Crosshead Displacement (mm)

Figure 5-34: Dynamic pull-through test results

215
Chapter 5: Structural Testing

The load in both cases is a transverse load applied to a constrained plate. Both load-
displacement responses are essentially the same form. The pattern of damage in
Figure 5-31 closely resembles the damage in the pull-through specimens, except on a
larger scale. Both tests exhibit a similar sensitivity in regards to loading rate. The large
load drop toward the end of each test is related to the spread of delamination damage
in the composite. In both cases the initiation of delamination damage and the rate at
which it spread was related to loading rate. The higher loading rate test in both cases
experienced an earlier initiation of delamination damage and a shallower unloading
slope.

There was no discernable loading rate sensitivity in the load-displacement response or


damage for the line loaded structural test. There was a minor loading rate sensitivity in
the load-displacement response of the ball loaded structural test, although the
damage propagation seemed insensitive to loading rate. The ball loaded structural test
had many similarities with the pull-through test reported in Chapter 3. The load in both
tests was applied transversely to a constrained composite plate. The damage pattern
in both tests was similar, as was the load displacement response. The effect of loading
rate was also similar, with the final material failure initiating earlier in the faster tests.

The lack of observed loading rate sensitivity for the line-loaded test can be explained
by examining the loads that were applied to the joint and the mode of failure that
occured. In this case, the primary load is carried in bearing, but with significant
contributions from pull-through and coupling forces. The approximate free body
diagram for the joints during the line loaded structure tests is shown in Figure 5-35.

When the bolt began to fail in bearing, the increased hole size allowed the bolt
increased freedom to rotate. When a significant amount of hole damage had been
reached, the remaining resistance to pull-through was reduced and the bolt pulled
from the laminate, delaminating the material on the bearing side of the hole only.

216
Chapter 5: Structural Testing

Contact

Bending moment reacted


by a tension on the bolt
and contact forces
between the plates

Figure 5-35: Joint loads during line loaded structure tests

With this damage progression, it was expected that the failure loading rate sensitivity
would show the characteristics of bearing failure initially and pull-through failure
would dictate the final behaviour. If this was the case, the loading rate dependence
would be extremely hard to detect. The initial bearing damage was a rate independent
phenomena, while the final pull-through damage, while displaying some rate
dependence, was limited to a small displacement range during the tests, and all tests
absorbed similar amounts of energy, regardless of the loading rate. Any difference in
the failure mode with loading rate would be absorbed within other scatter between
the tests.

It is assumed that most unconstrained countersunk fasteners would fail in a similar


manner, because load alignment and secondary bending produce significant tensile
loads in the bolt as shown in Figure 5-36. The small initial bearing damage zone for
countersunk fasteners allows for more hole elongation and, therefore, reduced
resistance to pull-through failure. This would explain why Li et al. [63] did not detect a
loading rate dependence for countersunk fasteners.

Figure 5-36: Load alignment via secondary bending

217
Chapter 5: Structural Testing

A simple composite structure with two parallel rows of single shear bolted joints was
tested in two different loading configurations. Although the structure tested was not
representative of a particular aircraft component, the loading conditions experienced
by the bolted joints would be present in many impact or crash scenarios experienced
by aircraft. The load was applied at varying loading rates between quasi-static (5
mm/min) through to 10 m/s. The range of test speeds overlapped those of the single
fastener joint tests presented in Chapter 3.

It was found that, over the range of test speeds, loading rate had no effect on the
failure mode of the bolted connections, only on the failure of the composite material
away from the joints. Any loading rate dependence that was present in the single
fastener joint tests did not express itself in the structural tests. It was determined that
the failure modes of the bolted joints present in the structural tests were not rate
sensitive.

218
Modelling of a simple composite structure in PAM-CRASH using
PLINK elements to represent the bolted connection
Chapter 6: Structural Modelling

A consequence of the operating environment of aircraft is that many critical load cases
involve impact and crash. These loading events are characterised by high loading rates,
high kinetic energy and possibly loads well above the static design strengths. It has
been shown earlier in this thesis that the properties of bolted joints can change with
loading rate. Modelling of structural level impact phenomena such bird-strike or crash
landing requires large explicit FE models which can take many hours through to many
weeks to solve. It is impractical to model each individual fastener in detail in such FE
models, as some structures have many thousands of bolted connections. Simplified
approximations to bolted joints are generally employed to reduce the computational
cost. Fasteners can be approximated in a number of ways, but one method that has
been shown to give an excellent balance between accurate joint behaviour and
computational cost is the PAM-CRASH Point Link (PLINK) element.

An investigation into the application of PLINK elements for impact simulation is


presented in this chapter. This section will address the calibration, suitability and
limitations of the PLINK element for representing bolted joints in quasi-static and
dynamic structural applications.

Two common simplified bolt modelling techniques will be used as a comparison to the
behaviour of the PLINK. The two methods will be referred to as the tied nodes
approach and the beam element approach.

A tied nodes approach joins the two halves of a joint together with a fixed constraint.
The connection has an infinite stiffness and relies on the stiffness of the surrounding
composite elements. The loads across the fixed constraint are measured and failure of
the element occurs when the combined shear and normal loads meet given failure
criteria.

A beam element approach joins the two halves of the joint together with and elastic
beam element. The stiffness of the joint is a combination of the stiffness of the beam

220
Chapter 6: Structural Modelling

element and the surrounding composite elements. Failure of the beam element occurs
in the same manner as for the tied nodes approach.

Behaviour of the two methods is shown in Figure 6-1.

a) b)
Fmax
Load

Displacement
Figure 6-1: Load-displacement curves simplified joint models
showing a) tied nodes and b) beam element approach

The structural tests conducted in Chapter 5 will form the basis for the modelling
investigation.

A skeleton finite element model of the composite structure was created that could be
used in all models. Boundary conditions and extra detail could then be added as
necessary. As all experiments were nearly symmetric up until just prior to complete
failure, the model utilised a single symmetry plane that significantly improved solution
time. The base finite element model is shown in Figure 6-2.

Figure 6-2: Base finite element model of the composite structure

221
Chapter 6: Structural Modelling

The composite plates were composed of 2-D four node, reduced integration, quad
elements (Material 131 or 132) which were joined together using 1-D PLINK elements
(Material 302) to represent the bolted joints. The mesh was a regular structured mesh
with an element size of approximately 2 mm. Further information regarding the
element formulations can be found in the PAMCRASH documentation [86, 88].

A large investigation into the composite material properties was conducted in earlier
chapters and so will not be investigated in this section. Material properties for the
composite were the same as those used for the detailed finite element modelling.

Aircraft structures may include many thousands of fasteners. Even the simplest bolted
structural component will generally have tens or hundreds of fasteners. Modelling
each of these fasteners individually in detail is not feasible with current computing
technology, however the behaviour of these fasteners under load can be critical to the
response of the structure. As an example, one detailed joint model with a coarse mesh
might include 10,000 nodes, each with multiple degrees of freedom. The time step of
this model would be determined by the smallest element, generally in the sub-
millimetre length scale. One of these models would take minutes, if not hours to solve
on the best available desktop computing hardware.

A numerically efficient approximation to a bolted joint or rivet is required to allow the


accurate modelling of composite structures. PAM-CRASH includes a 1-D element called
a PLINK element which offers many advantages for very little computational cost.

A summary of the PLINK properties will be included here. For further detail please
refer to Breitweg [145] and the software documentation [86, 88]. Extensive
investigation has also been conducted by this author into the behaviour of the PLINK
element. Refer to Appendix E for more information regarding the PLINK element
properties. Most of the element behaviour described here is provided in the software

222
Chapter 6: Structural Modelling

documentation, however some is reverse engineered from extensive single element


tests.

The PLINK is a penalty based mass-less contact element with user definable linear and
non-linear element stiffness and strength capabilities. The material density, , and
length do not affect the element stiffness, mass, strength properties, or model time
step and the element length can be zero. The PLINK can be connected to nodes, shells
or solid element faces. Therefore the link element achieves a degree of mesh
independence because it can be connected exactly where required, not just at given
nodes. Breitweg [145] studied the element in detail and determined that the most
robust results are achieved when a non-coincident nodal geometry is used, especially
when a dummy node is used to attach the PLINK between two shell elements. This
arrangement has been used throughout. Further improvement to the mesh
independence can be achieved by using a multi-PLINK option. The multi-PLINK option
introduces a number of PLINK elements at a small distance from control link which
smooths out the response of the link over a number of connection points. These
attributes can result in considerable efficiencies in the modelling of complex structures
with a large number of fasteners.

The PLINK element can be used to penalise three DOFs or six DOFs, known as the
"rivet" and "spotweld" models respectively, as shown in Figure 6-3. The loads
transmitted by the link in the three degrees of freedom case are described in Figure
6-4.

Figure 6-3: Effect of penalising different DOFs [86]

223
Chapter 6: Structural Modelling

Initial shape Deformed shape

Figure 6-4: PLINK element schematic

If the user defined stiffness formulation is used, then behaviour of the PLINK prior to
failure (F < 1), is given by Equations 6.1. Note that equilibrium is not strictly conserved
in the case where Kn Ks.

NE  K n d sin  
SE  K s d cos  
RE  NE2  SE2 ...(6.1)
an as
 N   S 
F  E
 E

Nmax Smax

where

NE = Normal (elastic) force transmitted by the link


SE = Shear (elastic) force transmitted by the link
RE = Magnitude of the (elastic) force transmitted by the link
d = Magnitude of the penalized gap vector
F = Failure index
Nmax = Normal failure strength
Smax = Shear failure strength

If at any time step F 1 then failure is identified and two new constants are defined
from current state values,

RF  RE
...(6.2)
dF  d

224
Chapter 6: Structural Modelling

where

RF = Magnitude of the failure force


dF = Magnitude of the failure displacement

After failure is initiated, two new functions are defined, and , which are plotted in
Figure 6-5 and Figure 6-6. The function presented in Figure 6-6 is a function of two user
defined parameters D1 and D2.


1.0

RF
RE
Figure 6-5: Total element force ()

Loading Unloading
1.0

dF dF+D1 dF+D2
d
Figure 6-6: Displacement based softening function ()

The element forces, N and S, are then given by the Equations 6.3.

NE *
N
 ...(6.3)
S*
S E


General load-displacement curves for the normal and shear loading directions are
shown in Figure 6-7. After initial failure, the link experiences a period of constant load
extension, followed by a softening region. These energy absorbing mechanisms are

225
Chapter 6: Structural Modelling

controlled by two critical displacements, common to both the normal and shear
loading directions, which give the element controllable energy absorption.

D2
D1
Load Smax

Ks
D2
D1
Nmax

Kn

Displacement
Figure 6-7: PLINK load-displacement response

Gunnion et al. [59] and Breitweg [145] showed that PLINK elements can be used
effectively to model single-fastener bolted joints in composite structures. In general,
good agreement between experiment and simulation can be achieved if care is taken
with the element attachment and calibrated stiffness and strength properties are
used.

The PLINK elements can be calibrated from experimental data. A characteristic load-
deflection response for a bolted joint is shown in Figure 6-8a. The load-displacement
response of the single fastener tests did not have an identical form to the PLINK
element, so approximations need to be made when calibrating the elements. Three
different calibration strategies were considered for the PLINK element properties. All
the methods were calibrated against the bearing test and maintained the same energy
absorption under the test curve. The three methods varied the peak load,
displacement to failure and element stiffness respectively to achieve the desired
energy absorption. The three PLINK curves are shown in Figure 6-8c-d. The
corresponding PLINK displacement method calibration for the pull-through direction is
shown in Figure 6-9. The failure displacements could not be calibrated exactly because
the PLINK element only has one set of displacement parameters.

226
Chapter 6: Structural Modelling

a) 15 b) 15
Pull-through PLINK
Bearing

10 10
Load (kN)

Load (kN)
5 5

0 0
0 2 4 6 0 2 4 6
Deflection (mm) Deflection (mm)
c) 15 d) 15
PLINK PLINK

10 10
Load (kN)
Load (kN)

5 5

0 0
0 2 4 6 0 2 4 6
Deflection (mm) Deflection (mm)
Figure 6-8: Single fastener joint test data and PLINK calibration strategies showing
a) experimental data, b) load method, c) displacement method and d) stiffness method

6
Pull-through
5 PLINK

4
Load (kN)

0
0 2 4 6
Deflection (mm)
Figure 6-9: Corresponding pull-through PLINK calibration for the displacement method

The pull-through stiffness of the PLINK cannot be calibrated directly from experimental
data because the stiffness is mainly driven by the bending stiffness of the laminate,
rather than the link itself. Gunnion et al. [59] and Breitweg [145] calibrated the PLINK

227
Chapter 6: Structural Modelling

pull-through stiffness by modelling a simple pull-through test in PAM-CRASH and


adjusting the stiffness of the link until agreement was reached with experiment, as
shown in Figure 6-10.

Figure 6-10: Pull-through stiffness calibration using PAMCRASH model [59]

Another method which achieves similar results with much less computational effort is
to use the LVDT data collected during the quasi-static testing. The quasi-static tests
conducted by the CRC-ACS included an LVDT which measured the displacement of the
head of the fastener during a pull-through test. The difference between the LVDT
measured displacement and the cross-head displacement was approximately
proportional to the load applied to the bolt.

KN + LVDT  +Crosshead  , PN ...(6.4)

where

Kn = Constant of proportionality (Stiffness in kN/mm)


LVDT = LVDT measured displacement of bolt head
Crosshead = Crosshead displacement
PN = Applied load

The constant of proportionality was taken as the stiffness of the fastener. A plot of the
scaled displacements versus the applied load is shown in Figure 6-11. The stiffness
calculated using this method for a " countersunk titanium HiLok fastener was 24
kN/mm as opposed to 28.7 kN/mm calibrated from the finite element model. The FE
model technique required knowledge about the boundary conditions of the test, the
material properties of the composite and a detailed FE model was required. The

228
Chapter 6: Structural Modelling

stiffness calibration using the LVDT data produced a similar result without the
requirement for any modelling or knowledge of material parameters.

Cross-head Load
Bolt LVDT - Cross-head (Scaled)

Load or Displacement

Cross-head Displacement
Figure 6-11:Pull-through stiffness calibration using LVDT calibration

The PLINK element does not have any strain-rate capability built in, but if the loading
rate is known a priori then the element can be calibrated to the correct failure curve. A
dynamic PLINK calibration was performed, based on the 10 m/s loading rate test data.
The energy absorbing displacements were calibrated against the high rate bearing test
data. It can be seen in Figure 6-12 that the calibration method cannot capture both the
pull-through and bearing behaviour simultaneously.

a) b) 6
14 10 m/s Bearing 10 m/s Pull-through
PLINK Calibration 5 PLINK Calibration
12
10 4
Load (kN)
Load (kN)

8 3
6
2
4
2 1

0 0
0 10 20 0 10 20
Crosshead Displacement (mm) Crosshead Displacement (mm)
Figure 6-12: Calibration of PLINK elements for dynamic simulation showing
a) bearing (shear) calibration and b) pull-through (normal) calibration

229
Chapter 6: Structural Modelling

The line loaded structure was modelled using the base model shown in Figure 6-2 with
the addition of some extra features. The quasi-static test rig with the specimen in situ
is shown in Figure 6-13. The PAMCRASH model geometry is shown in Figure 6-14. A
symmetry plane was used to improve computational efficiency.

Figure 6-13: Line loaded test fixture

Cylindrical
Protector impactor
plate Beam
representing test
rig supports
PLINK locations

Rigid plate Middle plate


representing
test rig grip Edge plate

Figure 6-14: Geometry of line-loaded half symmetry FE model

A cylindrical impactor was added as a rigid body to represent the loading cylinder. An
aluminium plate was added between the impactor and the composite for those tests in
which it was used. A section at the end of the composite plate was assumed to be a

230
Chapter 6: Structural Modelling

rigid body and boundary conditions were applied which replicated the conditions of
the experimental setup used. The boundary conditions are shown in Figure 6-15.

Roller Y
a) b)
Grip
Support X Rotation Free

X restraint provided Y displacement


Frame by calibrated beam fixed

Figure 6-15: Boundary conditions for line loaded model showing


a) experimental condition and b) model boundary condition

The beam element used to restrain the X direction of the model displacement in Figure
6-15b was calibrated by measuring the stiffness of the supports in the test rig by
loading them with an Instron test machine. The measured stiffness was used in the
PAMCRASH model for the beam elements. The overall behaviour of the model
depends strongly on this stiffness, so it was very important to represent it correctly in
the model. The calibration curves used to measure these stiffness values are shown in
Figure 6-16.

30

25
Axial Force (kN)

20

15

10
Bar 1
5 Bar 2
0
0 0.5 1
Displacement (mm)
Figure 6-16: Axial load-displacement response of support bars

The cylindrical impactor was loaded at 1 m/s. The force required to drive the impactor
corresponded to the cross-head load from the experiment. The displacement of the
impactor corresponded with the cross-head displacement from the experiment.

231
Chapter 6: Structural Modelling

The results of the line loaded FE model with the three different PLINK calibration
methods are shown in Figure 6-17a-c. The PLINKs in these simulations penalise six
degrees of freedom. The effect of the number of penalised degrees of freedom for the
PLINK elements also was investigated. It was not clear before the simulation which
formulation would be more physically accurate. It was observed experimentally that
the bolts rotated quite significantly during the tests, suggesting that bending moments
may not have been effectively transmitted through the joint. The result of the three
DOF simulation, using the displacement calibration method, is shown in Figure 6-17d.

a) 20 b) 20
Quasi-static Quasi-static
Experiment Experiment

Impactor Force (kN)


Impactor Force (kN)

15 Load 15 Displacement

10 10

5 5

0 0
0 10 20 30 40 0 10 20 30 40
Impactor Displacement (mm) Impactor Displacement (mm)

c) 20 d) 20
Quasi-static Quasi-static
Experiment Experiment
Impactor Force (kN)
Impactor Force (kN)

15 Stiffness 15 3 DOF

10 10

5 5

0 0
0 10 20 30 40 0 10 20 30 40
Impactor Displacement (mm) Impactor Displacement (mm)
Figure 6-17: Comparison between line loaded experiment and FE models
showing a) load method b) displacement method c) stiffness method
and d) displacement method with 3 DOF penalised

The experimental result shown in Figure 6-17 includes two sharp load drops at 25-30%
of the peak load, which were not associated with failure in the composite, but were a

232
Chapter 6: Structural Modelling

problem with the test rig, so may be disregarded. For the six DOF penalty case, the
three calibration methods all closely approximated the experimental results, with each
having certain advantages. The load method predicted the peak load of the
experiment most accurately, the displacement method predicted the displacement to
failure most accurately, and the stiffness method predicted the overall stiffness of the
structure most accurately. The closest approximation to the overall energy absorption
of the panel was the displacement method. This calibration strategy was used for the
modelling described subsequently in this chapter.

It is clear from Figure 6-17d that the three DOF solution did not accurately capture the
behaviour of the experiment. In this case, the bending moments were not transferred
through the joint and were therefore carried by a force couple which overloaded the
bolt in tension.

A comparison between the experimental result and the FE model is shown in Figure
6-18, which shows a good agreement between the deformed shape of the structure
under load.

a)

b)

Figure 6-18: Comparison between experimental and numerical results showing


a) experiment just prior to failure and b) FE model at the equivalent
displacement with plot of average stress in the longitudinal direction

The value of the PLINK element formulation over other simplified bolt modelling
techniques was investigated by removing the key features of the PLINK, namely user
defined stiffness and post failure energy absorption. The first test included PLINK

233
Chapter 6: Structural Modelling

elements with user defined stiffness but no post failure energy absorption, which
approximated elastic beams or springs. The second test included PLINK elements that
had very large stiffness values, which approximated a tied nodes approach.

a) 20 b) 20
Quasi-static Quasi-static
Experiment Experiment
Impactor Force (kN)

Impactor Force (kN)


15 Elastic 15 Rigid

10 10

5 5

0 0
0 10 20 30 40 0 10 20 30 40
Impactor Displacement (mm) Impactor Displacement (mm)
Figure 6-19: Comparison between the experimental results and other joint modelling methods
a) elastic beam PLINK and b) rigid PLINK

It can be clearly seen that the PLINK elements predicted the behaviour of the structure
more accurately than the other two methods. The rigid link approach was overly stiff
and therefore the shear load became high before the structure could deform
significantly. The beam method, which had the same elastic behaviour as the PLINK
failed to capture a significant part of the energy absorption of the structure because
the beam model did not have post-failure energy absorption.

The dynamic test rig with the specimen in situ is shown in Figure 6-20.

Figure 6-20: High-rate line load test fixture

234
Chapter 6: Structural Modelling

The rig was similar to the quasi-static rig, but there were a few significant differences
that created modelling challenges. The rig support bars were not acting in the direct
line between the cylindrical grips, so there was a tendency for the rig to deform in
bending when the load was applied. This increased compliance was accounted for by
reducing the stiffness of the support bars in the finite element model. Unlike the quasi-
static rig however, the stiffness for the dynamic rig was unknown. The stiffness of the
support bars was calibrated such that the initial stiffness of the structure matched the
stiffness of the dynamic test results. The failure behaviour of the model with the PLINK
elements included could then be compared with the experimental results. The
dynamic PLINK calibration was used. The result is shown in Figure 6-21.

25
1 m/s Experiment

20 Dynamic PLINK
calibration
Load (kN)

15

10

0
0 10 20 30 40 50 60
Crosshead Displacement (mm)

Figure 6-21:Comparison between dynamic experiment and FE model


with dynamic PLINK calibration

It is clear from Figure 6-21 that the FE model did not agree well with the experimental
results. This is due to the large failure displacement of the PLINK elements. In the
dynamic bearing test, the laminates were constrained to remain in-plane, and the bolt
continued to progressively tear through the composite, absorbing a great deal of
energy over a large displacement. When this property was programmed into the
PLINK, the structure remained intact for much longer and absorbed much more energy
than in the experiment. In reality, once a small amount of damage had occurred
around the hole, the bolt pulled through the laminate and the structure failed.

235
Chapter 6: Structural Modelling

It is clear from the quasi-static line-loaded structural model that the PLINK element is
an effective way of modelling bolted composite joints in composite structures.
Although some care has to be taken with connectivity and choosing the correct
element stiffness, the overall response of the composite structure with PLINK elements
included is far more accurate than a structure modelled with simple bar elements or
tied nodes. When the minimal increase in computational cost is considered, then the
PLINK approach becomes even more attractive.

The three calibration methods evaluated for the PLINK elements all achieved far better
prediction of the experimental load, stiffness and displacement to failure than the
other two simplified methods tested. Overall the best agreement between experiment
and FE model was achieved using the displacement calibration method, although
boundary condition approximations in the FE model may have outweighed any minor
differences between the three calibration techniques.

The model which included the PLINK elements predicted the response of the structure
more accurately than the other methods for two reasons. Firstly, the element had two
independent orthogonal stiffness parameters. When compared with the tied nodes,
which relies on the parent material stiffness, the PLINKs can be calibrated to achieve
the specific properties of real composite bolted joints. Secondly, the post-failure
behaviour of the PLINK element can be tailored to some degree to represent the actual
load-displacement response of composite joints. The tied nodes approach has no built
in failure behaviour, while the beam element approach only accounts for a fixed failure
load, without the controlled "softening" that can be programmed into the PLINK.

The PLINK elements did not perform well in the dynamic model. Calibrating the PLINK
elements for the dynamic case was problematic. The post-failure energy absorbing
displacements for the dynamic bearing and pull-through tests were significantly
different, so calibrating the PLINK against the bearing test did not allow accurate
calibration against the pull-through test. It was found that a failure displacement
calibrated against the bearing test did not agree with experiment. This observation is
consistent with the physical reality of the experiment, because the joint failed in a

236
Chapter 6: Structural Modelling

mode that was more closely related to a pull-through failure than a bearing failure.
Care must be taken when calibrating the PLINK elements that the real physical failure
modes are being taken into account. The conservative approach is to calibrate the
PLINK displacements to the smallest observed displacement-to-failure from the tests.

Further research is required into the mixed mode loading of bolted joints before the
extra energy absorption of the bearing failure can be utilised. Some suggestions for
changes to the PLINK element that may improve this behaviour are included at the end
of this chapter.

The point loaded structure was modelled using the base model shown in Figure 6-2
with the addition of some extra features. The test rig for the ball loaded structure is
shown in Figure 6-22 with the specimen in situ. The PAMCRASH model geometry is
shown in Figure 6-23.

A spherical impactor rigid body was added to represent the ball impactor from the
experiments. A rigid body was added to represent the support frame and the clamps
that held the panel in place. A tab section was added with greater thickness that
represented the tabs added to the panel.

Figure 6-22: Ball loaded test specimen in fixture

237
Chapter 6: Structural Modelling

Middle plate

Tab section
Spherical
impactor

Sublaminates
Support frame Edge plate

Figure 6-23: Geometry of point-loaded finite element model

The main difference between the ball loaded model and the default model was the
addition of sublaminates to the middle plate of the model to capture the delamination
damage that occurs under the impactor. The experimental evidence showed that the
delamination damage was extensive, and without the stacked shell modelling
approach it would be impossible to achieve correlation with the experiments. Two
different stacking sequences were investigated, as well as a single shell model (with no
delamination interface) as a comparison. The three different arrangements are shown
in Table 6-1.

Stack arrangement and


Number of Number of
Stack name sublaminate layup
Sublaminates Interfaces
(from the impactor side)

Single Shell 1 0 [45,0/90]4s

Double Sublaminate 1: [45,0/90]4


2 1
Shell Sublaminate 2: [0/90,45]4

Quadruple Sublaminate 1,2: [45,0/90]2


4 3
Shell Sublaminate 3,4: [0/90,45]2

Table 6-1: Sublaminate stacks used for ball loaded model

The composite material properties were the same as those used in other models
throughout the thesis. The interface parameters were the same as those calculated in
Chapter 4. The mesh size used for the interface calibration models was very similar to
that used for the structural models in this chapter.

238
Chapter 6: Structural Modelling

The load-deflection responses for the three laminate stacks are shown in Figure 6-24a-
c. The effect of the delamination interface was investigated by increasing the fracture
toughness (GIc and GIIc) properties by 20%. The result of this simulation is shown in
Figure 6-24d. Please note that the numerical results have had a five point moving
average applied to them to filter out the high frequency noise from the explicit FE
analysis. The noise is spurious high frequency vibration that is common in with explicit
time integration schemes, which is not related to any real effect.

a) 8 b) 8
1 m/s 1 m/s
Single Shell Double Shell

Impactor Force (kN)


Impactor Force (kN)

6 6

4 4

2 2

0 0
0 10 20 30 40 50 0 10 20 30 40 50
Impactor Displacement (mm) Impactor Displacement (mm)
c) 8 d) 8
1 m/s 1 m/s
High Gc
Impactor Force (kN)

Impactor Force (kN)

Four
6 6
Shells

4 4

2 2

0 0
0 10 20 30 40 50 0 10 20 30 40 50
Impactor Displacement (mm) Impactor Displacement (mm)
Figure 6-24: Experimental load-deflection response compared to numerical analyses
showing a) single shell model, b) double shell model, c) quadruple shell model
and d) double shell model with GIc and GIIc increased by 20%

A comparison between the final damaged specimen and the experimental results is
shown in Figure 6-25. The highlighted areas show the extent of delamination damage
for the experiment (taken from CT scan data) compared with the predicted
delamination from the FEA code. It can be seen clearly from Figure 6-24a and Figure

239
Chapter 6: Structural Modelling

6-25b that the single shell approach does not agree well with the experiment. No
prediction of delamination damage is possible and the load-deflection response under-
predicts the energy absorption of the panel.

The double shell and four shell arrangements agree quite well with experiment,
although in regards to failure mode and failure load-deflection response, the double
shell model performed slightly better. The predicted delamination area shown in
Figure 6-25c is much closer to the experimental results. In the four shell simulation, the
interfaces absorbed too much energy, and as a result the delamination area was
significantly smaller. Further investigation is required to understand what properties of
the interface model are intrinsic and which are extrinsic, so that when the model is
scaled by adding more interfaces, the same overall energy absorption is maintained.

a) b)

c) d)

Figure 6-25: Damage comparison for different stacking arrangements showing


a) Impacted specimen, b) Single Shell model,
c) Double shell model and d) Four Shells model

240
Chapter 6: Structural Modelling

Modelling the impact of the spherical ball with composite structure attempted with
three different stacked shell configurations. The double shell and quadruple shell
arrangements agree quite well with experiment, although in regards to failure mode
and failure load-deflection response, the double shell model performed slightly better.
The predicted delamination area in Figure 6-25c is much closer to the experimental
results. In the four shell simulation, the interfaces absorbed too much energy, and as a
result the delamination area was significantly smaller. Further investigation is required
to understand what properties of the interface model are intrinsic and which are
extrinsic, so that when the model is scaled by adding more interfaces, the correct
overall energy absorption is maintained.

By comparing the high speed video footage from the tests, experimental curves [146]
and numerical results a qualitative understanding of the failure progression can be
achieved. Delamination damage for this specimen was initially quite localised, and
initial Mode II delamination under the impactor nearly coincided with the first ply
failures at the back face of the laminate. The modelling and test footage suggest that
the fibre failure occurred slightly before the delamination failure. After initial ply
failure and localised delamination damage, the structure experienced a region of near
linear inelastic energy absorption. This period of damage was a mixture of in-plane ply
damage and localised delamination damage. Large delamination damage did not occur
until the final stages of impact event, after the impactor had significantly penetrated
the composite panel. The period of greatest delamination damage could be
determined by comparing the simulations in Figure 6-24b and d. The only difference
between the two simulations was the delamination fracture toughness, so periods of
the test when delamination was propagating have higher load for a given
displacement.

The use of PLINK elements for the ball loaded model did not offer significant
improvements over other joint modelling techniques because neither the stiffness of
the panel nor the dominant failure modes depended on the bolted connections.
During the course of the impact however, one of the PLINK elements did reach 70% of

241
Chapter 6: Structural Modelling

the maximum allowable normal force, so it is possible that a change in impactor


location may have induced a failure in the bolted connection.

Explicit FE modelling of quasi-static and dynamic structural tests has been conducted in
the commercial FE code PAM-CRASH. Two different test configurations were
considered, a line loaded structure and a point loaded structure. Two FE models were
created that reflected the different loads and boundary conditions of each test. Bolted
joints in the structures were modelled with simplified 1-D PLINK elements which were
calibrated from existing test data. Three calibration strategies for the PLINK element
were investigated.

It was shown that the response of the line loaded structure was well predicted by the
PLINK elements, even though the loading was a complicated mix of bearing and pull-
through loads. It was also shown that simpler methods, such as approximating the
bolted joint with linear elastic beams or rigid links, were a far less accurate method of
predicting the structural behaviour.

The ball loaded structural failure was dominated by composite ply failures and
delamination failure. It was shown that good agreement could be achieved between
experimental results and numerical results as long as the composite panel was
modelled using a stacked shell approach that allowed explicit delamination damage to
be captured. Further investigation into the mesh dependence of the interface and
composite material properties would be required to improve the simulations further. It
was found that the PLINK elements performed similarly to other simple joint modelling
methods.

242
Chapter 6: Structural Modelling

Please refer to Breitweg [145] and Gunnion [59] initial guidelines for calibrating and
using the PLINK. This section will only discuss features that were investigated within
the thesis.

Breitweg investigated many different scenarios for the connectivity between PLINK
elements and the surrounding laminates. It was found that the specifics of the
connectivity influence the behaviour of the element significantly. The most robust and
accurate results were found when a non-coincident node (NCN) geometry was used. In
order to remove a degree of mesh dependence, the PLINK can now be used with what
is known as a multi PLINK formulation. When using this formulation, the code adds an
extra set of PLINK elements at a user defined radius from the initial link. This
formulation attempts to smooth the behaviour of the PLINK over the cluster of links. If
a radius is set that is of the same size or bigger than the local element size, then a
multitude of different connection points can be found and the element becomes much
less mesh dependent. In PAM-CRASH 2007, there is a problem with the element
stiffness calibration when using this method. Previously calibrated user stiffness
properties need to be divided by the total number of links before entry into the
material card.

It was found that the 6 DOF PLINK, known as the spotweld model, was far superior to
the 3 DOF rivet model when modelling bolted joints in composite structures.

Three different calibration strategies were considered for calibrating the PLINK against
quasi-static joint test results. All methods produced similar results, suggesting that the
model is quite tolerant of minor errors. The displacement method was shown to
perform marginally better than the other calibration schemes, see Section 3.2.3 of this
chapter.

A significant difficulty was discovered when attempting to calibrate the PLINK element
against dynamic test data. The countersunk fastener bearing tests were rate
dependent, and absorbed significantly more energy at higher loading rates. Attempts

243
Chapter 6: Structural Modelling

to program in a greater failure displacement, to account for the energy difference,


conflicted with the dynamic pull-through data. The overall performance of the PLINK
elements calibrated against the dynamic bearing results was very poor. Modifications
to the numerical PLINK formulation have been suggested that would remedy this
problem.

Some deficiencies in the PLINK element behaviour were identified while modelling
bolted composite structures for this chapter. The rupture model of the PLINK, for
instance, was only controlled by a single set of displacements. It was shown in this
chapter that this rupture model does not allow both bearing and pull-through failure
modes to be calibrated separately. This section details possible changes to the PLINK
element that would provide much more flexibility to adjust the failure of the element
in the shear and normal directions.

A hypothetical displacement based failure profile is shown in Figure 6-26. Three


different loading paths have been included to illustrate the behaviour of the model in
load-displacement space.

a)
d3n c)
Normal Displacement

d2n

d1n

b)
d1s d2s d3s

Shear Displacement

Figure 6-26: Failure surfaces in displacement space showing three loading paths a), b), and c)

244
Chapter 6: Structural Modelling

In the current formulation, six values are entered to define the rupture model of the
PLINK; two loads, Nmax and Smax; two displacements, d1 and d2; and two exponents a1
and a2. For the model proposed, six values are also required; three normal
displacements, d1n, d2n and d3n; and three shear displacements, d1s, d2s and d3s. The
subscripts 1, 2 and 3 represent the displacements at which initial failure, softening and
element elimination occur respectively. Users who prefer using Nmax and Smax can
calculate d1n and d1s using the stiffness values KS and KN. The basic formulation would
initially be identical to the force based method as shown in the Equations 6.5. Three
failure criteria would be calculated instead of one as shown in Equations 6.6.

NE  K n d n
SE  K s ds
...(6.5)
RE  NE2  SE2
d  d n2  ds2
2 2
d  d 
F1  n
 s

d1n d1s
2 2
d  d 
F2  n
 s
...(6.6)
d 2 n d2s
2 2
d  d 
F3  n
 s

d 3 n d3s

Just as for the current formulation, two functions, and , would be calculated. These
are given in Equations 6.7.

  max 1, F1 

*  min 1, F2
 F3  1  ...(6.7)

 F3  F2 

If F3 1 then the element is eliminated. The element forces are then given by Equation
6.8, as for the current element properties.

NE *
N
 ...(6.8)
S*
S E


245
Chapter 6: Structural Modelling

This formulation was calibrated using load and displacement data from quasi-static
fastener tests presented in Chapter 3. In this case a countersunk bolt with diameter
6.35 mm and laminate thickness 3.52 mm were chosen, corresponding to test series
B32.

Path a) Path b) Path c)


PT test Bearing Test
15
Load (kN)

10

0
0 1 2 3 4 5 6
Displacement (mm)

Figure 6-27: Displacement-based model calibrated to B32 test data

The results for a), b), c) and the actual test data are shown in Figure 6-27. Note that
the normal stiffness has been lowered to represent the deflection of the composite
panel as well.

It can be seen that the model would provide much greater flexibility in terms of
tailoring specific normal and shear failure properties. At any loading condition in
between shear and normal loads the failure would be a combination of properties. The
model calculates the failure indexes on-the-fly and so does not suffer from the history
dependence present in the current formulation.

An unloading formulation is being considered to allow the element to absorb energy


by unloading along a separate curve to loading. One possibility would be to neglect the
function and use a degraded stiffness parameter instead.

246
The two mains goals of this research were to,

Experimentally investigate the failure of bolted joints


in composite structures under dynamic loads

and to

Indentify and develop modelling techniques for bolted joints that can
minimise the requirement for further experimental investigation.
Chapter 7: Conclusion

Many studies have been conducted on the behaviour of composite structures when
loaded dynamically, a typical case being bird strike of a leading edge structure. The
specifics of the load transfer at the bolted joints is generally overlooked in these
studies, even though it has been shown [2] that the bolted (or riveted) connections
have a profound effect on the behaviour of the structure. A few authors have
addressed dynamic effects in bolted or riveted joints but these studies have been
purely experimental and limited to a select few joint configurations. The two main
studies involving composite bolted joints also conflict with each other on key points
regarding the behaviour of bolted joints in composite structures.

In response to the scarcity of available knowledge in this area, an investigation into the
dynamic properties of bolted joints in composite structures was conducted. The
investigation was extended above and beyond what was currently available in the
published scientific literature. The dynamic experimental test regime in this thesis
included load cases that were not considered previously, such as pull-through loading.
The dynamic experiments also considered bolted joints in simplified composite
structures to investigate the effects of complicated loading conditions on the rate
sensitivity of bolted joints. Numerical modelling of dynamically loaded composite
bolted joints, both at the detailed joint level and at the structural level was also
investigated in this thesis.

The key contributions of this thesis will be outlined in the following sub-sections.

The rate dependence of single fastener bolted joints over a range of loading rates
between quasi-static and 10 m/s was experimentally investigated. New test fixtures
were developed which captured the key features of the ASTM standard fixtures but
were compatible with high-rate testing. Dynamic pull-through testing was conducted
for the first time with bolted composite joints.

248
Chapter 7: Conclusion

The dynamic pull-through tests exhibited weak loading rate dependence. The stiffness
and failure load of the dynamic tests was reasonably constant through the spectrum of
loading rates tested. The total energy absorption of the specimens decreased slightly
with loading rate, as delamination failure initiated at a lower cross-head displacement.
Numerical studies of pull-through FE models with cohesive zone interface elements
suggested that the shear delamination (Mode II) properties of the composite material
were critical in determining the failure behaviour of the specimen. It was found that a
decrease in Mode II SERR and propagation stress with loading rate fit with the
observed experimental data.

The dynamic bearing tests exhibited a strong rate dependence of the range of loading
rates tested. The protruding head fastener tests exhibited two different failure modes
when tested quasi-statically, net tension failure and bearing failure. When loaded at 10
m/s, only the bearing failure mode was observed, and the extent of bearing damage
was greater. Within the bearing failure mode, distinct loading rate dependence was
observed. The peak load achieved by each test was relatively constant with test speed
however the progressive bearing load of the tests dropped with increasing loading
rate. The stiffness of the initial bearing response was also higher for the dynamic tests.
The observed behaviour of the protruding head fastener specimens agrees well with
the results presented by Ger et al. [62] but disagrees with a few critical features of Li et
al. [63].

The countersunk head fasteners exhibit two distinct failure modes. The tests
conducted at 0.1 m/s all failed via a bolted failure mode. The tests conducted at 1.0
m/s and above all failed in a progressive bearing failure mode, absorbing significantly
more energy.

One key observation of the dynamic tests is the interaction between rate sensitivity of
a particular failure mode and the failure mode of the joint. It was clear that the bearing
failure mode, which is mainly controlled by matrix properties, was rate sensitive. Net
tension failure was not observed to be rate sensitive, which is reasonable given that it
is controlled by fibre tensile properties. Given these two behaviours, it is easy to devise
joint geometries that will fail in one failure mode quasi-statically while failing in

249
Chapter 7: Conclusion

another failure mode dynamically. Future investigations into dynamically loaded


bolted joints or pin loaded holes need to consider this effect carefully.

Quasi-static and dynamic testing was conducted with a novel composite test article.
The test article consisted of two rows of single lap bolted joints connecting three flat
composite plates. The bolt type, layup and joint geometry were identical to the
countersunk head bearing joints tested dynamically in Chapter 3.

A line-loaded configuration was tested at loading rates between quasi-static and 10


m/s. No loading rate sensitivity was observed over the range of loading rates. This
result is significant because the single fastener joint tests experienced significant
changes in failure mode over the same range of loading rates. It was identified that
there were mechanisms which allowed the joints to fail in a pull-through mode
regardless of the load applied to the joint.

A ball-loaded configuration was tested at loading rates between 0.1 and 10 m/s. Minor
rate sensitivity was observed in the failure of the composite laminate but no rate
sensitivity was observed at the bolted connection.

Many of the key failure modes of bolted joints, such as delamination and through-
thickness shear cracks are not easily detected within a failed specimen. Previous
investigations into these failure modes can be broken into two main categories. Non-
destructive techniques such as ultrasonic scanning and X-ray imaging do not interfere
with the specimen but only give 2-D information about the damage location.
Destructive techniques such as optical microscopy require careful specimen
preparation and successive polishing-imaging iterations if 3-D damage information is to
be attained.

X-ray Computer Tomography (CT) scanning is a 3-D non-destructive imaging technique


that is common in the medical sciences. CT scanning has only recently started receiving

250
Chapter 7: Conclusion

attention from engineering disciplines. CT scanning can generate fully 3-D


reconstructions of engineering test specimens and is a great deal faster to implement
than optical microscopy. CT scanning has been employed previously to investigate
simplified pin bearing failure [23].

In this thesis CT scanning has been employed to investigate the failure modes of bolted
joints. This novel application of existing technology has allowed the failure modes of
bolted bearing and pull-through specimens to be investigated in excellent detail.

Implicit finite element modelling of bearing failure in a composite bolted joint has
been undertaken. Progressive damage modelling of composite bolted joints is still a
rich area of research. At the time of writing, the most recently available publication on
the topic [127], used very similar modelling methodology to that presented in this
thesis. The conclusions of this research, reached independently from the paper, echo
similar key findings. The most significant finding is that the material degradation rules
imposed on failed elements are as important, if not more important, than the criteria
used to predict failure.

Stacked shell modelling utilises a stack of shell elements tied together by interface
elements to represent the overall behaviour of a laminate. The sublaminate stack is
capable of predicting more failure modes than a conventional layered shell element.
Delamination prediction is possible by applying cohesive zone formulations to the
interface elements connecting each sublaminate. The stacked shell approach offers an
intermediate level of complexity and computational effort between layered shell
elements and ply-by-ply solid element models.

Stacked shell modelling has previously been used to predict impact damage in
composites. The application of stacked shell laminates to bolted joint modelling has
not previously been attempted. Significant investigation has been conducted into the
capabilities of the stacked shell modelling approach, specifically the application to

251
Chapter 7: Conclusion

bolted joint modelling. It was found that the stacked shell models handled the complex
stress and strain fields around a bolted joint quite well, and the technique was applied
to model and investigate failure in both pull-through tests and bearing tests. Some
recommendations were made to improve the behaviour of the stacked shell cohesive
zone interface model when loaded in compression, based on the work of Hou et al.
[82]. The improvements were demonstrated on a pull-through model by using extra
interface elements in the compressive zone to artificially enhance the strength of the
interface.

In general, pull-through failure was modelled very well with a stacked shell model.
Failure load, failure mode, delamination and total energy absorption were all modelled
well. Bearing failure was modelled very well with stacked shell modelling during the
initial failure progression. Some difficulties modelling damaged composite material
debris prevented the accurate prediction of the progressive bearing failure mode with
stacked shell modelling.

Modelling fasteners in complicated dynamically loaded structures, such as a wing


leading edge subjected to a bird impact, requires simplified joint models that still
adequately predict the behaviour of real bolted joints. Many such simplified joint
models exist but in general they lack the ability to predict critical features of the
response of bolted joints. Novel efficient joint models in PAM-CRASH, PLINKs, were
investigated for modelling bolted joints in the multi-fastener structural tests and
compared to other common methods for representing fasteners in large explicit
models. It was shown that a model which used PLINK elements for the fastened
connection predicted the response of the structure extremely well, far exceeding the
response of the comparison methods.

Extra guidelines for the use of PLINK elements have been include in this thesis to add
to those developed by Breitweg [145]. Recommendations have been included for a
change to the PLINK numerical model that would significantly improve the failure
behaviour in mixed mode loading conditions. These changes have been demonstrated
for a single bolt and laminate combination used commonly throughout the thesis.

252
Chapter 7: Conclusion

As with any research program, investigating the failure of bolted joints has answered
many questions but also raised many more. This section will discuss some of the key
areas of bolted joint characterisation in which research is still required. It is believed
that all the areas discussed could, and should, have much more research effort
devoted to them if bolted joints in composite materials are to be fully understood.

There exists a gap in the understanding of bolted joint failure in composite materials.
Many tests have been conducted to characterise the bearing failure of bolted joints
and a few papers have conducted tests to characterise the pull-through failure of
bolted joints. Within the field of composite materials however, no publicly available
material addresses the issue of 'mixed mode' loadings, as shown in Figure 7-1.

a)

c)

b)

Figure 7-1: Combined loading of bolted joint showing a) pure normal loading
b) pure shear loading and c) unknown effect of combined loading [61]

The Arcan test [147], was developed as a method to test the shear properties of plastic
and was eventually adapted to investigate fibrous composite materials. The test fixture
is designed so that specific combinations of normal and shear loads can be applied to
the specimen cross-section. A schematic of the test fixture is shown in Figure 7-2.

253
Chapter 7: Conclusion

Figure 7-2: Arcan test fixture

A modified Arcan test fixture has been used by ONERA to investigate the failure of
metal riveted joints under mixed mode loading conditions [142].

Investigating the failure of composite bolted joints under mixed mode loading
conditions would improve the understanding of joint failure modes as well as highlight
any critical load cases that may not have been previously identified. Understanding the
spectrum of failures between pull-through failure and bearing failure would also
provide much more accurate data for calibrating simplified joint elements such as the
PLINK.

It was shown in Chapter 3 that CT scanning was capable of imaging failure modes in
bolted joints non-destructively, and with a high resolution. Unfortunately, at the time,
more testing was not possible as the equipment is in Stuttgart and was accessed
during a 6 month secondment to the DLR laboratory. Based on the experience gained
it is concluded that CT scanning could provide an invaluable tool for studying bolted
joint failure in composite materials, and could be used in conjunction with any other
test regime.

By stopping a number of bearing tests at various points throughout the failure process
and scanning the specimens, an understanding of the failure progression could be

254
Chapter 7: Conclusion

attained. Some key points of interest are shown in Figure 7-3. The CT scanner would
allow detailed information to be gathered about failure modes which are not
externally visible, such as delamination and through thickness shear cracks.

16

14

12

10
Force (kN)

0
0 1 2 3 4 5
Displacement (mm)

Figure 7-3: Countersunk bearing test with key points of interest

It has been shown that stacked shell models are a useful tool in predicting the failure
behaviour of bolted joints in composite materials. However, it was found that there
were some failure modes that cannot be captured accurately with stacked shell
models, such as through-thickness shear cracks. Certain joint geometries promote
these failure modes, so stacked shell models do not provide a predictive model of bolt
failure that is universally applicable.

Conducting explicit FE analyses of bolted joints with solid elements was not attempted
in this thesis due to the high computational cost. Mass scaling can be used to lower the
computational cost of solid element models, but extreme care must be taken not to
introduce spurious inertial forces. With the seemingly perpetual increase in
computational power available to the common FE user, it is possible that this
modelling technique will become practical in the near future.

255
Chapter 7: Conclusion

One joint geometry that has been identified as most suited to solid element modelling
is pull-through failure of thick laminates. The predominant failure mode in this test is
through-thickness matrix shear failure, which is not well predicted by stacked shell
models. The numerical models can also be highly simplified by the use of symmetry.

A numerical study, investigating the effect of laminate thickness and modelling


approach, could be compared to the results of Banbury [21] and those presented in
this thesis. Limits could then be placed on the capabilities of stacked shell modelling
and when it is necessary to adopt solid element modelling.

One of the key findings of Chapter 3 is that there is interdependence between the rate
effects within a specific failure mode and the failure modes that occur. To fully
understand the rate dependence of bolted joints in composite materials, the rate
dependence of each failure mode needs to be investigated separately, as each failure
mode relies on different composite failure mechanisms. The testing would require
joints with specially selected geometries that strongly promote one failure mode over
all others. This was discussed in depth in Chapter 3.

It was discussed in the literature review that heat generation within a failure test,
generated from friction or other physical effects, can have a significant impact on the
failure of composite materials or metals.

Temperature and friction effects were not addressed in great detail within the body of
the thesis because the scope of the research would have become too broad and
experimental characterisation of these effects would have required the completion of
additional expensive testing regimes.

It is the author's opinion that these effects are of significant importance for failure of
not just composite bolted joints, but the failure of any dynamically loaded system.
Research into the effects of material heating via frictional or other means is a rich area
of possible research.

256
1. Badders, D. Dreamliner 101: All about the Boeing 787. 2007 [cited 2008 May
23rd]; Available from:
http://seattlepi.nwsource.com/boeing/787/787primer.asp.
2. McCarthy, M.A., et al., Modelling of Bird Strike on an Aircraft Wing Leading
Edge Made from Fibre Metal Laminates Part 2: Modelling of Impact with SPH
Bird Model. Applied Composite Materials, 2004. 11(5): p. 317-340.
3. Baker, A., S. Dutton, and D.W. Kelly, Composite Materials for Aircraft
Structures. 2nd ed. 2004: American Institute of Aeronautics and Astronautics.
4. Purslow, D., Fractography of fibre-reinforced thermoplastics, Part 3. Tensile,
compressive and flexural failures. Composites, 1988. 19(5): p. 358-366.
5. Berbinau, P., C. Soutis, and I.A. Guz, Compressive failure of 0 unidirectional
carbon-fibre-reinforced plastic (CFRP) laminates by fibre microbuckling.
Composites Science and Technology, 1999. 59(9): p. 1451-1455.
6. Soutis, C., Measurement of the static compressive strength of carbon-
fibre/epoxy laminates. Composites Science and Technology, 1991. 42(4): p.
373-392.
7. Wu, P.S. and C.T. Sun, Modeling bearing failure initiation in pin-contact of
composite laminates. Mechanics of Materials, 1998. 29(3-4): p. 325-335.
8. Goutianos, S., T. Peijs, and C. Galiotis, Mechanisms of stress transfer and
interface integrity in carbon/epoxy composites under compression loading: Part
I: Experimental investigation. International Journal of Solids and Structures,
2002. 39(12): p. 3217-3231.
9. Bai, J. and S.L. Phoenix, Compressive failure model for fiber composites by kink
band initiation from obliquely aligned, shear-dislocated fiber breaks.
International Journal of Solids and Structures, 2005. 42(7): p. 2089-2128.
10. Vogler, T.J. and S. Kyriakides, On the initiation and growth of kink bands in fiber
composites: Part I. experiments. International Journal of Solids and Structures,
2001. 38(15): p. 2639-2651.
11. Narayanan, S. and L.S. Schadler, Mechanisms of kink-band formation in
graphite/epoxy composites: a micromechanical experimental study. Composites
Science and Technology, 1999. 59(15): p. 2201-2213.
12. Ageorges, C., K. Friedrich, and L. Ye, Experiments to relate carbon-fibre surface
treatments to composite mechanical properties. Composites Science and
Technology, 1999. 59(14): p. 2101-2113.
13. Wu, P.S. and C.T. Sun, Bearing Failure in Pin Contact of Composite Laminates.
AIAA Journal, 1998. 36(11): p. 2124-2129.
14. Purslow, D., Matrix fractography of fibre-reinforced epoxy composites.
Composites, 1986. 17(4): p. 289-303.

257
15. Puck, A. and H. Schrmann, FAILURE ANALYSIS OF FRP LAMINATES BY MEANS
OF PHYSICALLY BASED PHENOMENOLOGICAL MODELS. Composites Science and
Technology, 1998. 58(7): p. 1045-1067.
16. Greve, L. and A.K. Pickett, Modelling damage and failure in carbon/epoxy non-
crimp fabric composites including effects of fabric pre-shear. Composites Part A:
Applied Science and Manufacturing, 2006. 37(11): p. 1983-2001.
17. Johnson, A.F., A.K. Pickett, and P. Rozycki, Computational methods for
predicting impact damage in composite structures. Composites Science and
Technology, 2001. 61(15): p. 2183-2192.
18. Pickett, A.K. and M.R.C. Fouinneteau, Material characterisation and calibration
of a meso-mechanical damage model for braid reinforced composites.
Composites Part A: Applied Science and Manufacturing, 2006. 37(2): p. 368-
377.
19. Bascom, W., D. Boll, B. Fuller, and P. Phillips, Fractography of the interlaminar
fracture of carbon-fibre epoxy composites. Journal of Materials Science, 1985.
20(9): p. 3184-3190.
20. Camanho, P.P., S. Bowron, and F.L. Matthews, Failure Mechanisms in Bolted
CFRP. Journal of Reinforced Plastics and Composites, 1998. 17(3): p. 205-233.
21. Banbury, A. and D.W. Kelly, A study of fastener pull-through failure of
composite laminates. Part 1: Experimental. Composite Structures, 1999. 45(4):
p. 241-254.
22. Yan, Y., W.D. Wen, F.K. Chang, and P. Shyprykevich, Experimental study on
clamping effects on the tensile strength of composite plates with a bolt-filled
hole. Composites Part A: Applied Science and Manufacturing, 1999. 30(10): p.
1215-1229.
23. Toth-Antal, B., Damage Detection in Carbon Fibre Composites. 2004, University
of New South Wales: Sydney, Australia.
24. Khashaba, U.A., H.E.M. Sallam, A.E. Al-Shorbagy, and M.A. Seif, Effect of washer
size and tightening torque on the performance of bolted joints in composite
structures. Composite Structures, 2006. 73(3): p. 310-317.
25. Park, H.-J., Effects of stacking sequence and clamping force on the bearing
strengths of mechanically fastened joints in composite laminates. Composite
Structures, 2001. 53(2): p. 213-221.
26. Sierakowski, R.L., Strain rate effects in composites. Applied Mechanics Reviews,
1997. 50(12/Pt 1): p. 741-761.
27. Zhou, Y., D. Jiang, and Y. Xia, Tensile mechanical behavior of T300 and M40J
fiber bundles at different strain rate. Journal of Materials Science, 2001. 36(4):
p. 919-922.
28. Weeks, C.A. and C.T. Sun, Modeling non-linear rate-dependent behavior in
fiber-reinforced composites. Composites Science and Technology, 1998. 58(3-
4): p. 603-611.
29. Hosur, M.V., J. Alexander, U.K. Vaidya, and S. Jeelani, High strain rate
compression response of carbon/epoxy laminate composites. Composite
Structures, 2001. 52(3-4): p. 405-417.
30. Newill, J.F. and J.R. Vinson. Some High Strain Rate Effects on Composite
Materials. in 9th ICCM. 1993. University of Zaragoza: Woodhead Publishing.

258
31. Bing, Q. and C.T. Sun, Modeling and testing strain rate-dependent compressive
strength of carbon/epoxy composites. Composites Science and Technology,
2005. 65(15-16): p. 2481-2491.
32. Gilat, A., R.K. Goldberg, and G.D. Roberts, Strain Rate Sensitivity of Epoxy Resin
in Tensile and Shear Loading. NASA/TM2005-213595 Glenn Research Centre,
NASA, 2005.
33. Staab, G. and A. Gilat. High Strain Rate Characterization of Angle-Ply
Glass/Epoxy Laminates. in 9th ICCM. 1993. University of Zaragoza: Woodhead
Publishing.
34. Fitoussi, J., et al., Experimental methodology for high strain-rates tensile
behaviour analysis of polymer matrix composites. Composites Science and
Technology, 2005. 65(14): p. 2174-2188.
35. Elder, D., Y. Dorsamy, and M. Rheinfurth, Failure of composite bonded joints at
elevated strain rates. 2009: Melbourne.
36. Sun, C., et al., Ductile-brittle transitions in the fracture of plastically-deforming,
adhesively-bonded structures. Part I: Experimental studies. International Journal
of Solids and Structures, 2008. 45(10): p. 3059-3073.
37. Huang, Y., W. Wang, C. Liu, and A.J. Rosakis, Analysis of intersonic crack growth
in unidirectional fiber-reinforced composites. Journal of the Mechanics and
Physics of Solids, 1999. 47(9): p. 1893-1916.
38. Dwivedi, S.K. and H.D. Espinosa, Modeling dynamic crack propagation in fiber
reinforced composites including frictional effects. Mechanics of Materials, 2003.
35(3-6): p. 481-509.
39. Sun, C.T. and C. Han, A method for testing interlaminar dynamic fracture
toughness of polymeric composites. Composites Part B: Engineering, 2004.
35(6-8): p. 647-655.
40. Johnson, G.R. and W.H. Cook. A constitutive model and data for metals
subjected to large strains, high strain rates and high temperatures. in 7th
International Symposium on Ballistics. 1983. The Hague, Netherlands.
41. Calamaz, M., D. Coupard, and F. Girot, A new material model for 2D numerical
simulation of serrated chip formation when machining titanium alloy Ti-6Al-4V.
International Journal of Machine Tools and Manufacture, 2008. 48(3-4): p. 275-
288.
42. Richeton, J., et al., Influence of temperature and strain rate on the mechanical
behavior of three amorphous polymers: Characterization and modeling of the
compressive yield stress. International Journal of Solids and Structures, 2006.
43(7-8): p. 2318-2335.
43. Khan, A.S., O. Lopez-Pamies, and R. Kazmi, Thermo-mechanical large
deformation response and constitutive modeling of viscoelastic polymers over a
wide range of strain rates and temperatures. International Journal of Plasticity,
2006. 22(4): p. 581-601.
44. HexPly F593 Data Sheet. 2005, Hexcel.
45. Cycom 970 Data Sheet, Cycom.
46. Cycom 977-3 Data Sheet, Cycom.
47. Kelly, D.W. and M.W. Tosh, Interpreting load paths and stress trajectories in
elasticity. Engineering Computations, 2000. 17(2): p. 117-135.

259
48. Prabhakaran, R. and R.A. Naik, A fiber optic technique to assess the influence of
interfacial infraction in a pin-loaded hole. ISA Transactions, 1989. 28(2): p. 35-
39.
49. Kelly, G. and S. Hallstrm, Bearing strength of carbon fibre/epoxy laminates:
effects of bolt-hole clearance. Composites Part B: Engineering, 2004. 35(4): p.
331-343.
50. Camanho, P.P. and F.L. Matthews, A Progressive Damage Model for
Mechanically Fastened Joints in Composite Laminates. Journal of Composite
Materials, 1999. 33(24): p. 2248-2280.
51. Mikulik, Z., S.E. Dutton, and R.S. Thomson, The Effect of Stacking Sequence on
Strength and Energy Absorption of a Composite Pin Joint under Progressive
Bearing Failure. A. J. M. Ferreira, Editor. 2009: Porto, Portugal.
52. Hart-Smith, L.J., Mechanically-Fastened Joints for Advanced Composites.
Fibrous Composite in Structural Design, ed. E. M. Lenoe, Oplinger, D. W., and
Burke, J. J. 1980, New York: Plenum Press.
53. McCarthy, C.T. and M.A. McCarthy, Three-dimensional finite element analysis of
single-bolt, single-lap composite bolted joints: Part II--effects of bolt-hole
clearance. Composite Structures, 2005. 71(2): p. 159-175.
54. Lawlor, V.P., W.F. Stanley, and M.A. McCarthy, Characterisation of Damage
Development in Single-Shear Bolted Composite Joints. Journal of Plastics,
Rubber and Composites, 2002. 31(3): p. 126-133.
55. McCarthy, M.A., V.P. Lawlor, W.F. Stanley, and C.T. McCarthy, Bolt-Hole
clearance effects and strength criteria in single-bolt, single-lap, composite
bolted joints. Composites Science and Technology, 2002. 62: p. 1415-1431.
56. Lawlor, V.P., M.A. McCarthy, and W.F. Stanley, Experimental study on effects of
clearance on single bolt, single shear, composite bolted joints. Plastics, Rubber
and Composites, 2002. 31: p. 405-411.
57. Lawlor, V.P., M.A. McCarthy, and W.F. Stanley, An experimental study of bolt-
hole clearance effects in double-lap, multi-bolt composite joints. Composite
Structures, 2005. 71(2): p. 176-190.
58. Frizzell, R.M., C.T. McCarthy, and M.A. McCarthy, An experimental investigation
into the progression of damage in pin-loaded fibre metal laminates. Composites
Part B: Engineering, 2008. 39(6): p. 907-925.
59. Gunnion, A., H. Koerber, D. Elder, and R. Thomson, Development of Fastener
Models for Impact Simulation of Composite Structures. 2006: Hamburg,
Germany.
60. Standard Test Method for Measuring the Fastener Pull-Through Resistance of a
Fiber-Reinforced Polymer Matrix Composite. ASTM D 7332 / D 7332M - 07,
ASTM International, 2007.
61. Koerber, H., Pull-Out and Shear Failure of Bolted Single Lap Joints in Composite
Laminates. 2006, Stuttgart University: Stuttgart, Germany.
62. Ger, G.S., K. Kawata, and M. Itabashi, Dynamic tensile strength of composite
laminate joints fastened mechanically. Theoretical and Applied Fracture
Mechanics, 1996. 24(2): p. 147-155.
63. Li, Q.M., R.A.W. Mines, and R.S. Birch, Static and dynamic behaviour of
composite riveted joints in tension. International Journal of Mechanical
Sciences, 2001. 43(7): p. 1591-1610.

260
64. Birch, R.S. and M. Alves, Dynamic failure of structural joint systems. Thin-
Walled Structures, 2000. 36(2): p. 137-154.
65. Tsai, S.W. and E.M. Wu, A General Theory of Strength for Anisotropic Materials.
Journal of Composite Materials, 1971. 5(1): p. 58-80.
66. Hart-Smith, L.J., Should fibrous composite failure modes be interacted or
superimposed? Composites, 1993. 24(1): p. 53-55.
67. Sun, C.T., B.J. Quinn, J. Tao, and D.W. Oplinger, Comparative Evaluation of
Failure Analysis Methods for Composite Laminates, DOT/FAA/AR, 1996.
68. Paris, F., A Study of Failure Criteria of Fibrous Composite Materials, Langley
Research Centre, 2001.
69. Hashin, Z. and A. Rotem, A Fatigue Criterion for Fiber Reinforced Materials.
Journal of Composite Materials, 1973. 7: p. 448-464.
70. Hashin, Z., Failure Criteria for Unidirectional Fiber Composites. Journal of
Applied Mechanics, 1980. 47: p. 329-334.
71. Puck, A. and H. Schrmann, Failure analysis of FRP laminates by means of
physically based phenomenological models. Composites Science and
Technology, 2002. 62(12-13): p. 1633-1662.
72. Yamada, S.E. and C.T. Sun, Analysis of Laminate Strength and Its Distribution.
Journal of Composite Materials, 1978. 12(3): p. 275-284.
73. Chang, F.-K. and K.-Y. Chang, A Progressive Damage Model for Laminated
Composites Containing Stress Concentrations. Journal of Composite Materials,
1987. 21(9): p. 834-855.
74. Gosse, J.H. and S. Christensen, Strain Invariant Failure Criteria for Polymers in
Composite Materials. 2001: Seattle.
75. Hart-Smith, L.J., Mechanistic Failure Criteria for Carbon and Glass Fibers
Embedded in Polymers in Polymer Matrices. 2001: Seattle.
76. Buchanan, D.L., et al., Micromechanical enhancement of the macroscopic strain
state for advanced composite materials. Composites Science and Technology,
2009. 69(11-12): p. 1974-1978.
77. Li, R., D. Kelly, and R. Ness, Application of a First Invariant Strain Criterion for
Matrix Failure in Composite Materials. Journal of Composite Materials, 2003.
37(22): p. 1977-2000.
78. Tay, T.E., S.H.N. Tan, V.B.C. Tan, and J.H. Gosse, Damage progression by the
element-failure method (EFM) and strain invariant failure theory (SIFT).
Composites Science and Technology, 2005. 65(6): p. 935-944.
79. Cuntze, R.G. and A. Freund, The predictive capability of failure mode concept-
based strength criteria for multidirectional laminates. Composites Science and
Technology, 2004. 64(3-4): p. 343-377.
80. Ye, L., Role of matrix resin in delamination onset and growth in composite
laminates. Composites Science and Technology, 1988. 33(4): p. 257-277.
81. Zhang, X., Impact damage in composite aircraft structures-experimental testing
and numerical simulation. Proceedings of the Institution of Mechanical
Engineers, Part G: Journal of Aerospace Engineering, 1998. 212(4): p. 245-259.
82. Hou, J.P., N. Petrinic, and C. Ruiz, A delamination criterion for laminated
composites under low-velocity impact. Composites Science and Technology,
2001. 61(14): p. 2069-2074.

261
83. Tay, T.-E., Characterization and analysis of delamination fracture in composites:
An overview of developments from 1990 to 2001. Applied Mechanics Review,
2003(56): p. 1-31.
84. Orifici, A.C., I. Herszberg, and R.S. Thomson, Review of methodologies for
composite material modelling incorporating failure. Composite Structures,
2008. 86(1-3): p. 194-210.
85. Beissel, S.R., G.R. Johnson, and C.H. Popelar, An element-failure algorithm for
dynamic crack propagation in general directions. Engineering Fracture
Mechanics, 1998. 61(3-4): p. 407-425.
86. Iannucci, L., Progressive failure modelling of woven carbon composite under
impact. International Journal of Impact Engineering, 2006. 32(6): p. 1013-1043.
87. Ladevze, P., O. Allix, J.-F. De, and D. Lvque, A mesomodel for localisation
and damage computation in laminates. Computer Methods in Applied
Mechanics and Engineering, 2000. 183(1-2): p. 105-122.
88. Iannucci, L. and J. Ankersen, An energy based damage model for thin laminated
composites. Composites Science and Technology, 2006. 66(7-8): p. 934-951.
89. Iannucci, L. and M.L. Willows, An energy based damage mechanics approach to
modelling impact onto woven composite materials--Part I: Numerical models.
Composites Part A: Applied Science and Manufacturing, 2006. 37(11): p. 2041-
2056.
90. Boutaous, A., B. Peseux, L. Gornet, and A. Belaidi, A new modeling of plasticity
coupled with the damage and identification for carbon fibre composite
laminates. Composite Structures, 2006. 74(1): p. 1-9.
91. Iannucci, L. and M.L. Willows, An energy based damage mechanics approach to
modelling impact onto woven composite materials: Part II. Experimental and
numerical results. Composites Part A: Applied Science and Manufacturing,
2007. 38(2): p. 540-554.
92. Xiao, X., C. McGregor, R. Vaziri, and A. Poursartip, Progress in braided
composite tube crush simulation. International Journal of Impact Engineering,
2009. 36(5): p. 711-719.
93. Camanho, P.P., C.G. Davila, and D.R. Ambur, Numerical Simulation of
Delamination Growth in Composite Materials, Langley Research Centre, 2001.
94. Petrossian, Z. and M.R. Wisnom, Prediction of delamination initiation and
growth from discontinuous plies using interface elements. Composites Part A:
Applied Science and Manufacturing, 1998. 29(5-6): p. 503-515.
95. Goyal, V.K., N.R. Jaunky, E.R. Johnson, and D.R. Ambur, Intralaminar and
interlaminar progressive failure analyses of composite panels with circular
cutouts. Composite Structures, 2004. 64(1): p. 91-105.
96. Johnson, A.F. and M. Holzapfel, Influence of delamination on impact damage in
composite structures. Composites Science and Technology, 2006. 66(6): p. 807-
815.
97. Guimard, J.M., O. Allix, N. Pechnik, and P. Thevenet, Energetic analysis of
fragmentation mechanisms and dynamic delamination modelling in CFRP
composites. Computers & Structures. In Press, Corrected Proof.
98. Meo, M. and E. Thieulot, Delamination modelling in a double cantilever beam.
Composite Structures, 2005. 71(3-4): p. 429-434.

262
99. Greve, L. and A.K. Pickett, Delamination testing and modelling for composite
crash simulation. Composites Science and Technology, 2006. 66(6): p. 816-826.
100. Aiello, D., Delamination Damage Modelling of Advanced Aerospace Composite
Stiffened Structures Under Low-Velocity Impact. 2004, Royal Melbourne
Institute of Technology: Melbourne.
101. Kradinov, V., A. Barut, E. Madenci, and D.R. Ambur, Bolted double-lap
composite joints under mechanical and thermal loading. International Journal
of Solids and Structures, 2001. 38(44-45): p. 7801-7837.
102. Echavarra, C., P. Haller, and A. Salenikovich, Analytical study of a pin-loaded
hole in elastic orthotropic plates. Composite Structures, 2007. 79(1): p. 107-
112.
103. Chang, F.-K., R.A. Scott, and G.S. Springer, Failure of Composite Laminates
Containing Pin Loaded Holes--Method of Solution. Journal of Composite
Materials, 1984. 18(3): p. 255-278.
104. Whitworth, H.A., M. Othieno, and O. Barton, Failure analysis of composite pin
loaded joints. Composite Structures, 2003. 59(2): p. 261-266.
105. Conti, P., Influence of geometric parameters on the stress distribution around a
pin-loaded hole in a composite laminate. Composites Science and Technology,
1986. 25(2): p. 83-101.
106. Lin, H.J. and C.C. Tsai, Failure analysis of bolted connections of composites with
drilled and moulded-in hole. Composite Structures, 1995. 30(2): p. 159-168.
107. Wu, T.J. and H.T. Hahn, The Bearing strength of e-glass/vinyl-ester composites
fabricated by vartm. Composites Science and Technology, 1998. 58(9): p. 1519-
1529.
108. Dano, M.-L., G. Gendron, and A. Picard, Stress and failure analysis of
mechanically fastened joints in composite laminates. Composite Structures,
2000. 50(3): p. 287-296.
109. Icten, B.M. and R. Karakuzu, Progressive failure analysis of pin-loaded carbon-
epoxy woven composite plates. Composites Science and Technology, 2002.
62(9): p. 1259-1271.
110. Okutan, B., The effects of geometric parameters on the failure strength for pin-
loaded multi-directional fiber-glass reinforced epoxy laminate. Composites Part
B: Engineering, 2002. 33(8): p. 567-578.
111. Okutan, B. and R. Karakuzu, The strength of pinned joints in laminated
composites. Composites Science and Technology, 2003. 63(6): p. 893-905.
112. Karakuzu, R., T. Gulem, and B.M. Icten, Failure analysis of woven laminated
glass-vinylester composites with pin-loaded hole. Composite Structures, 2006.
72(1): p. 27-32.
113. Icten, B.M., R. Karakuzu, and M.E. Toygar, Failure analysis of woven kevlar fiber
reinforced epoxy composites pinned joints. Composite Structures, 2006. 73(4):
p. 443-450.
114. Dano, M.-L., E. Kamal, and G. Gendron, Analysis of bolted joints in composite
laminates: Strains and bearing stiffness predictions. Composite Structures,
2007. 79(4): p. 562-570.
115. Marshall, I.H., W.S. Arnold, J. Wood, and R.F. Mousley, Observations on bolted
connections in composite structures. Composite Structures, 1989. 13(2): p. 133-
151.

263
116. Oh, J.H., Y.G. Kim, and D.G. Lee, Optimum bolted joints for hybrid composite
materials. Composite Structures, 1997. 38(1-4): p. 329-341.
117. Chen, W.-H., S.-S. Lee, and J.-T. Yeh, Three-dimensional contact stress analysis
of a composite laminate with bolted joint. Composite Structures, 1995. 30(3): p.
287-297.
118. Kradinov, V., E. Madenci, and D.R. Ambur, Combined in-plane and through-the-
thickness analysis for failure prediction of bolted composite joints. Composite
Structures, 2007. 77(2): p. 127-147.
119. Whitney, T.J., E.V. Iarve, and R.A. Brockman, Singular stress fields near contact
boundaries in a composite bolted joint. International Journal of Solids and
Structures, 2004. 41(7): p. 1893-1909.
120. Ireman, T., Three-dimensional stress analysis of bolted single-lap composite
joints. Composite Structures, 1998. 43(3): p. 195-216.
121. Lin, W.-H. and M.-H.R. Jen, The Strength of Bolted and Bonded Single-Lapped
Composite Joints in Tension. Journal of Composite Materials, 1999. 33(7): p.
640-666.
122. Padhi, G.S., M.A. McCarthy, and C.T. McCarthy, BOLJAT: a tool for designing
composite bolted joints using three-dimensional finite element analysis.
Composites Part A: Applied Science and Manufacturing, 2002. 33(11): p. 1573-
1584.
123. Tserpes, K.I., G. Labeas, P. Papanikos, and T. Kermanidis, Strength prediction of
bolted joints in graphite/epoxy composite laminates. Composites Part B:
Engineering, 2002. 33(7): p. 521-529.
124. McCarthy, M.A. and C.T. McCarthy, Finite element analysis of effects of
clearance on single shear composite bolted joints. Plastics, Rubber and
Composites, 2003. 32: p. 65-70.
125. McCarthy, M.A., C.T. McCarthy, V.P. Lawlor, and W.F. Stanley, Three-
dimensional finite element analysis of single-bolt, single-lap composite bolted
joints: part I--model development and validation. Composite Structures, 2005.
71(2): p. 140-158.
126. Kelly, G., Quasi-static strength and fatigue life of hybrid (bonded/bolted)
composite single-lap joints. Composite Structures, 2006. 72(1): p. 119-129.
127. Hhne, C., et al., Progressive damage analysis of composite bolted joints with
liquid shim layers using constant and continuous degradation models.
Composite Structures. In Press, Corrected Proof.
128. Banbury, A., D.W. Kelly, and L.K. Jain, A study of fastener pull-through failure of
composite laminates. Part 2: Failure prediction. Composite Structures, 1999.
45(4): p. 255-270.
129. Elder, D.J., A.H. Verdaasdonk, and R.S. Thomson, Fastener pull-through in a
carbon fibre epoxy composite joint. Composite Structures, 2008. 86(1-3): p.
291-298.
130. Moscardo, M., Simulation of fastened composites subject to pull-through and
shear loads. DLR-IB 435-2009/I, German Aerospace Center, 2009.
131. Chen, W.-H. and S.-S. Lee, Numerical and Experimental Failure Analysis of
Composite Laminates with Bolted Joints under Bending Loads. Journal of
Composite Materials, 1995. 29(1): p. 15-36.

264
132. Corporation, H.-S. http://www.hi-shear.com/fastener_hl_stds.htm. 2007 [cited
2009 20-05-2009].
133. Standard Test Method for Bearing Response of Polymer Matrix Composite
Laminates. ASTM D 5961 / D 5961M - 05, ASTM International, 2005.
134. Bornschegel, T., Experimentelle Untersuchung von Nieten in Verbundstrukturen
unter Hochgeschwindigkeitsbelastungen. In preparation, 2007.
135. Hopkinson, B., A Method of Measuring the Pressure Produced in the Detonation
of High Explosives or by the Impact of Bullets. Philosophical Transactions of the
Royal Society of London. Series A, Containing Papers of a Mathematical or
Physical Character, 1914. 213: p. 437-456.
136. Cook, R.D., D.S. Malkus, M.E. Plesha, and R.J. Witt, Concepts and Applications of
Finite Element Analysis. 4th ed. 2002: John Wiley & Sons.
137. Courant, R., K. Friedrichs, and H. Lewy, ber die partiellen
Differenzengleichungen der mathematischen Physik. Mathematische Annalen,
1928. 100(1): p. 32-74.
138. Marc 2007r1 Manual, Volume A: Theory and User Information. 2007, MSC
Software.
139. Mikkor, K., Method for Setting Bi-Phase Material Input Properties. 2005, CRC-
ACS.
140. Aymerich, F., F. Dore, and P. Priolo, Prediction of impact-induced delamination
in cross-ply composite laminates using cohesive interface elements. Composites
Science and Technology, 2008. 68(12): p. 2383-2390.
141. Aktay, L. and A.F. Johnson, FEM/SPH Coupling Technique for High Velocity
Impact Simulations, in Advances in Meshfree Techniques. 2007. p. 147-167.
142. Patronelli, L., et al., Experimental Procedure for Riveted Joint Design - From
Material Law Until Dynamic Strength. 2000: London, UK.
143. Crews, J.H., Jr., Bolt-Bearing Fatigue of a Carbon/Epoxy Laminate, in Joining of
Composite Materials, K. T. Kedward, Editor. 1981, American Society for Testing
and Materials. p. 131-144.
144. Collins, T.A., The Strength of Bolted Joints in Multidirectional CFRP Laminates.
Composites, 1977. 8: p. 43-55.
145. Breitweg, T., Simulation of fastened joints in composite structures to failure.
2006, Stuttgart University: Germany.
146. Pearce, G.M., A.F. Johnson, R.S. Thomson, and D.W. Kelly, Experimental
Investigation of Dynamically Loaded Bolted Joints in Carbon Fibre Composite
Structures. Applied Composite Materials, 2009. Accepted - In Press.
147. Arcan, M., Z. Hashin, and A. Voloshin, A Method to Produce Uniform Plane
Stress States with Applications to Fibre-Reinforced Materials. Experimental
Mechanics, 1978. 18(14): p. 141-146.
148. Pearce, G.M., L.P. Djukic, R.J. Wootton, and D.W. Kelly, Application of SIFT to
micro-cracking and lug failure. 2009: Melbourne.
149. Li, R., et al., Improving the efficiency of fiber steered composite joints using load
path trajectories. Journal of Composite Materials, 2006. 40(18): p. 1645-1658.
150. Bollen, L., Application of SIFT to failure in composite laminates. 2008, UNSW:
Sydney.
151. Lai, W.K., Composite Lug Design and Analysis. 2008, UNSW: Sydney.

265
152. Ketcham, R.A. and W.D. Carlson, Acquisition, optimization and interpretation of
X-ray computed tomographic imagery: applications to the geosciences.
Computers & Geosciences, 2001. 27(4): p. 381-400.
153. Djukic, L.P., et al., Contrast enhancement in visualisation of woven composite
architecture using a MicroCT Scanner. Part 2: Tow and preform coatings.
Composites Part A: Applied Science and Manufacturing. In Press, Corrected
Proof.
154. Djukic, L.P., et al., Characterisation of Woven Composites using Tow
Architecture from MicroCT Scan. 2009: Melbourne, Australia.
155. Geandier, G., et al., Microstructural analysis of alumina chromium composites
by X-ray tomography and 3-D finite element simulation of thermal stresses.
Scripta Materialia, 2003. 48(8): p. 1219-1224.
156. Schilling, P.J., et al., X-ray computed microtomography of internal damage in
fiber reinforced polymer matrix composites. Composites Science and
Technology, 2005. 65(14): p. 2071-2078.
157. Bayraktar, E., S. Antolonovich, and C. Bathias, Multiscale study of fatigue
behaviour of composite materials by [chi]-rays computed tomography.
International Journal of Fatigue, 2006. 28(10): p. 1322-1333.
158. Bayraktar, E., M.A. Gonzalez Garza, and C. Bathias, A comprehensive study of
plastic deformation mechanism of the metallic materials by 'X-ray' computed
tomography ('X-ray' CT). Journal of Materials Processing Technology, 2008.
200(1-3): p. 133-145.
159. Pearce, G.M., A.F. Johnson, R.S. Thomson, and D.W. Kelly, High Strain-Rate
Response of Fastened Carbon Fibre Composite Joints in Composite Structures.
2007: Porto, Portugal.

266
Appendix A: Strain Invariant Failure Theory

During the research program it was not clear whether the difficulties with the implicit
finite element approach in modelling progression of delamination could be offset by
improved failure prediction. As the research progressed it was discovered that the best
approach was to change to an explicit finite element algorithm. The work completed
on the SIFT then become redundant to the main research objective. However, SIFT is a
new failure theory with limited publication. A significant result was achieved in
understanding development of a delamination and the failure of a complex lug and a
conference paper was presented on the topic [148]. This work will be reported here.

Strain Invariant Failure Theory (SIFT) was first proposed by Gosse and Christensen [74]
and a closely related companion paper by Hart-Smith [75]. SIFT is a strain-based,
mechanistic failure criteria. SIFT supposes that failure of composite materials at the
micro-mechanical level can only occur via a small set of constituent failure
mechanisms. There are three failure mechanisms considered by Gosse; critical volume
change (dilatation) of the matrix material, critical dilation-free angle change
(distortion) of the matrix material and fibre failure. Mechanistically, dilatation failure
corresponds to crazing and micro-cavitation while distortion failure corresponds to
shear-yielding. Fibre failure can be controlled by any preferred mechanism, however
investigations by Hart-Smith [75] and Buchanan et al. [76] suggest compatible fibre
failure mechanisms.

Volume expansion failure generally arises due to deformation of the highly constrained
resin pockets located between the fibres. Mathematically it is related to the first strain

A-1
Appendix A: Strain Invariant Failure Theory

invariant (J1). Strain invariants are rotationally invariant terms extracted from the
strain tensor. The first strain invariant corresponding to volume change is given by
Equation A.1.

J1  x  y  z
(A.1)
 1  2  3

Distortional failure generally occurs at the fibre-matrix interface or in the interstitial


locations between fibres. Distortional failure is related to the modified second strain
invariant (J2). The modified strain invariant is given by Equation A.2.

J12
J 
'
2  J2
3


  x  y    y  z    z  x     xy 2   yz 2   zx 2
1 1

2 2 2
  (A.2)
6  4
1
   1  2    2  3    3  1  
2 2 2

6 

Generally a von Mises equivalent strain is used for convenience because it is of the
same order as J1.

vm  3J2' (A.3)

SIFT failure mechanisms can only be applied at the constituent level. Accurate
constituent modelling of even the most basic composite structures is prohibitively
expensive computationally, so a means to generate constituent level strains from
lamina level strains was required.

Strain fields at the constituent level are very complex, and are dependent on not only
the applied strains but the relative stiffness parameters of each constituent, as well as
the location relative to each constituent. In a normal composite material, the fibres
and resin are cured at high temperatures and have different coefficients of thermal
expansion. When the composite is cooled down to room temperature the constituents
retain a great deal of thermal residual strain as a result. Therefore the true strains
within the constituents are a function of many variables, as shown in Equation A.4.

A-2
Appendix A: Strain Invariant Failure Theory


ij  f , Ei ,- ij ,. i ,T ,location  ...(A.4)

To account for the local variations in strain due to heterogeneous material, a


micromechanical modelling approach is adopted, known as Micromechanical
Enhancement (MME). MME involves modelling representative unit cells including the
fibre and matrix material as separate phases and constructing a relationship between
the macroscopically applied strains and the microscopic strains in the constituents.
Although an infinite number of unit cells exist for an arbitrary packing of fibres within a
matrix material, two unit cells have been chosen that represent the most efficient and
inefficient packing arrangements within an ordered packing array. A diagram of the
two fibre-matrix unit cells is shown in Figure A-1, taken from Buchanan et al. [76]. Vf is
the volume fraction of the fibre.

a) z b) z

y
y

x x
Figure A-1: Fibre matrix packing arrangements
a) square array and b) hexagonal array

The application of a unit strain in any direction will generate a strain field within the
unit cell. At any one point the strain tensor will have at least one non-zero term, and
due to Poissons ratio and elastic modulus mismatches, at most points the strain
tensor will have more than one non-zero term. At any given point we can construct a
6x6 matrix where each column corresponds to the strain tensor at the point
determined by finite element analysis from a unit strain applied in one direction. For
instance, if a strain field of

z = za, x = 0, y = 0, xy = 0, yz = 0, zx = 0 (A.5)

A-3
Appendix A: Strain Invariant Failure Theory

is applied to the square array unit cell, then the resulting strain field (component in the
loading direction) is shown in Figure A-2.

By applying each set of boundary conditions in turn, a relationship can be constructed


between the applied strain tensor and the micromechanically enhanced strain tensor
at any given point within a given unit cell model, as shown in Equation A.6.

za

 x   A13 
  A 
 y  23 
 z   A33 
     za
 xy   A43 
 yz   A53 
   
 zx  m  A63 

za

Figure A-2: Square array unit cell with single applied strain

m a
 x   A11 A12 A13 A14 A15 A16   x 
  A A22 A23 A24 A25 A26   y 
 y  21
 z   A31 A32 A33 A34 A35 A36   z 
     ...(A.6)
 xy   A41 A42 A43 A44 A45 A46   xy 
 yz   A51 A52 A53 A54 A55 A56   yz 
    
 zx   A61 A62 A63 A64 A65 A66   zx 

The approach so far does not account for the effects of thermal contraction mismatch
that make up a large component of the micromechanical strain field. The thermal load
affects the composite laminate in two separate ways. The classical influence of thermal
contraction is at the laminate or interlaminar level. Each ply has orthotropic
coefficients of thermal expansion. When the plies are laminated at various angles
relative to one another and subjected to a thermal load (i.e. cooling from cure
temperature to room temperature), then thermal residual strain is developed in each
ply. This strain acts just like mechanical strain for the purposes of micro-mechanical
enhancement, and is included in the applied strain tensor for Equation A.6.

A-4
Appendix A: Strain Invariant Failure Theory

The other effect that the thermal contraction mismatch has on the composite laminate
is at the intralaminar level. Assuming that the laminate is constrained in all directions,
the interlaminar thermal strains remain zero. At the unit-cell level, this condition is
equivalent to the boundary conditions

z = 0, x = 0, y = 0, xy = 0, yz = 0, zx = 0. (A.7)

There is however still a significant thermal strain present, which is due to the
contraction of the fibre and matrix within the unit cell. This effect can be measured by
applying

z = 0, x = 0, y = 0, xy = 0, yz = 0, zx = 0, T = Ta (A.8)

to the unit cell. The resulting strain field (component up the page) for the square array
unit cell is shown in Figure A-3.

 x   B1 
  B 
 y  2
 z   B3 
     T
 xy   B4 
 yz   B5 
   
 zx  m  B6 

Figure A-3: Thermal load applied to square array unit cell

Equation A.6 can be augmented to include the intralaminar thermal strains as shown in
Equation A.9.

m a
 x   A11 A12 A13 A14 A15 A16   x   B1 
  A A22 A23 A24 A25  
A26   y   B 
 y  21  2
 z   A31 A32 A33 A34 A35 A36   z  B3 
       T   (A.9)
 xy   A41 A42 A43 A44 A45 A46   xy  B4 
 yz   A51 A52 A53 A54 A55 A56   yz  B5 
      
 zx   A61 A62 A63 A64 A65 A66   zx  B6 

A-5
Appendix A: Strain Invariant Failure Theory

Without prior knowledge of the strain to which a unit cell will be subjected, the
location of failure is unknown. Within each unit cell a number of points of interest are
selected, at which it is assumed the strain tensor will be critical for some loading case.
Choosing more points increases the likelihood that one of the points will correspond to
the most highly loaded location within the unit cell, although it increases the
computational effort at solution time.

An implementation of the failure theory has been generated for the implicit finite
element software MSC.Marc. The failure theory is included in an MSC.Marc analysis via
a Fortran subroutine. A schematic representation of the operation of the subroutine is
shown in Figure A-4.

J1crit
Strain Tensor vm crit
Temperature Difference

SAF(point, Calculate modified strain


array) tensor at given point J1max>J1crit,
No
or
vmmax> vmcrit?
Loop Calculate J1 and vm from
over all modified strain tensor
positions Yes
and
arrays Degrade element
stiffness matrix
Is J1 > J1max
No J1max
or
vm max
vm > vm max? End

Yes
Update Maximums
Figure A-4: Schematic representation of SIFT subroutine

The subroutine runs for every element and integration point for every load increment.
The subroutine takes a macroscopic mechanical strain tensor and temperature
difference as inputs from the FEA code. The modified strain tensor is calculated at one
point in one array using Equation A.9. The J1 and vm values for ( m) are calculated
using Equations A.1 to A.3. The record of the highest J1 and vm is kept while the
subroutine loops over all the points of interest within each array. The eventual

A-6
Appendix A: Strain Invariant Failure Theory

maximum values are compared against the critical values obtained from experiment
[78]. If J1 or vm exceed the critical values then failure has occurred, if not then the
subroutine ends and MSC.Marc increments the load step. If failure occurs and the user
is running a progressive failure analysis, then the subroutine calls another built-in
MSC.Marc subroutine to degrade the element stiffness matrix according to what
failure has occurred. The degradation factors used are taken from Camanho and
Matthews [50].

This section has been taken from a conference paper submitted by this author and
others at UNSW [148]. Lugs are used as fittings in aerospace structures to form
connections between structural components. Composite materials provide a
lightweight design solution but are subject to a wider range of failure modes than
fittings made from metals. In addition the lugs are highly loaded and attempts have
been made to create fibre architectures in which the fibres are optimally aligned to
increase the stiffness and strength. In one such research program an unexpected
failure mode was encountered when the thickness of the lug was increased. Strength
improvements were obtained for 6mm thick lugs when steered fibres were included in
the laminate via a set of tailored fibre pattern (TFP) layers. The improvement was
erased in 12mm thick lugs by the appearance of a failure mode that involved splitting
of the laminate between the steered layer of the laminate and the neighbouring plies.
This failure mode can be seen clearly in Figure A-5. This splitting mode was not present
in the composite lugs without TFP layers.

The orientation of the fibres in the TFP layer is shown in Figure A-6. The axial fibres
around the hole are intended to carry the bearing load while the tangential fibres are
intended to carry the hoop load around the outside of the lug. The steered layers
account for 25% of the laminate a restriction recommended in Li et al. [149] to give a
quasi-isotropic laminate designed to resist inter-laminar failure in thin laminates.

A study was initiated involving two undergraduate thesis projects Bollen [150] and Lai
[151] to try to investigate the failure. The first applied a new SIFT algorithm developed

A-7
Appendix A: Strain Invariant Failure Theory

by the authors to the progressive failure of a 3 point bend test published by Tay et al.
[78] to validate the SIFT methodology. The second developed a finite element model of
the lug including multi-body contact between the pin and the lug and modelling of the
individual plies in the laminate including the steered layer. In the undergraduate
research, failure in the lug was predicted using a simple maximum stress criterion.

Figure A-5: Splitting mode of lug with TFP layers


(Figure reproduced with permission of CRC-ACS)

Figure A-6: Fibre orientation in tailored fibre pattern (TFP)


(Figure reproduced with permission of CRC-ACS)

A SIFT analysis, using a subroutine developed by this author and colleagues, validated
by Bollen [150], was applied to the same finite element model used by Lai [151]. The
finite element model of the lug is shown in Figure A-7. The model utilises two planes of
symmetry to reduce computational cost. The preliminary results of the SIFT analysis
are shown in Figure A-8. A stiffness degradation method was used to introduce
damage into the elements after SIFT predicted failure. The region shown is the first ply
above the central TFP layer. The dark locations show regions of damage predicted by
the SIFT subroutine. It can be clearly seen that the drop-off of the axial steered fibres is
initiating a new failure mechanism that is separate from the initial bearing damage.

A-8
Appendix A: Strain Invariant Failure Theory

Figure A-7: Lug finite element model

a)

b)

Figure A-8: Preliminary SIFT results showing a) Bearing failure region and
b) Splitting initiated at axial fibre termination

Due to the arrangement of the TFP, there is a large stiffness mismatch in the loaded
direction at the interface between the axial and hoop fibres. At this interface, the load
carried by the stiff fibres is passed into the neighbouring plies (with intermediate
stiffness) as inter-laminar shear stress. This shear initiates the splitting failure mode
shown in Figure A-5.

A question arises as to why this failure is triggered in the thick laminates and not in the
thin laminates. A characteristic of bearing from a pin in composites is the formation of
a wedge at the bearing surface extending from the surface of the laminate to the mid-
ply at an angle of approximately 45o. For the laminates considered here the wedge for
the thicker laminate extends to the vicinity of the ply drop and could be the cause of
the premature failure. If this analysis is correct re-design of the fibre placement could
eliminate the new failure mode and restore the strength improvement expected for
the specimens with fibre placement.

A-9
Appendix B: CT Scanning

X-ray computed tomography (CT) is a non-destructive inspection technique that uses


high energy radiation (generally x-rays) to visualise the interior of objects. CT images
can be viewed as 2-D slices of the object or 3-D reconstructions made from volume
pixels or voxels.

Detector

X-ray source

Specimen Specimen Image

Figure B-1: Schematic of X-ray CT arrangement

The basic configuration of an engineering CT scanner is shown in Figure B-1. The


specimen lies between the source and the detector, as in conventional x-ray imaging.
The detector measures x-ray intensities over a sampling time and generates a single x-
ray image of the part in a given orientation. The intensity of the x-rays detected
depends on the attenuation of the beam within the specimen. The attenuation of
mono-energetic x-rays in a homogenous material is described by Beer's Law [152],

I  I0 e / x ...(B.1)

B-1
Appendix B: CT Scanning

where

I = Intensity of x-rays at distance x into the material


I0 = Incident x-ray intensity
= Linear attenuation coefficient
x = Distance within the material along the x-ray path

The equation can be expressed in a piecewise form for a number (n) of different
materials along the x-ray path

 n 
I  I0 exp   /i xi  ...(B.2)
 i 1 

In a CT scan 3-D information about the specimen is gathered by rotating the part
through 180 or 360. A number of x-ray images are taken as the specimen is rotated
relative to the source and detector. If a discretised 3-D domain is assumed within the
specimen then Equation B.2 can be used with the x-ray intensity information to solve
for the linear attenuation coefficient of each element of the 3-D domain (voxel). Since
the attenuation is related to material properties such as density and elemental
composition, contrast between different materials can be achieved with a resolution
related to the voxel size. The voxel size is related to the size of the pixels of the
detector and the relative distance between the source, specimen and detector.
Magnification is achieved by moving the specimen closer to the source, so the angle
subtended by the specimen relative to the source is increased, increasing the size of
the image on the detector.

There are many artefacts that emerge in CT images, however the most important two
are beam hardening and ring artefacts. Beam hardening appears as bright regions at
the edges of objects and dark regions in the centre, even for homogenous materials.
Beam hardening is due to preferential attenuation of low-energy x-rays [152], and can
cause difficulties in comparing material composition throughout the specimen. Ring
artefacts appear as full or partial circles centred on the rotational axis of the specimen.
Ring artefacts are due to changes in the way the pixels of the detector respond to
changes in the testing conditions. Ring artefacts are reasonably easy to remove with

B-2
Appendix B: CT Scanning

software corrections, however some real features of the specimen may be interpreted
as artefacts if they lie parallel to the artefact rings.

CT scanning has had medical applications for a long period of time, but is only recently
become adopted by engineering disciplines. CT imaging of composite microstructures
and mesostructures has become an active area of research in recent years [153-158].
Although CT is a lengthy process, it has become a very valuable tool in understanding
damage mechanisms in composites.

The CT scanner used for these results is a Phoenix Nanotom 180NF. It has a 180 kV x-
ray source and a 12-bit 5.2 megapixel detector. The volume pixel (voxel) size of the
final images is 47 m.

B-3
Appendix C: Hi-Lok Fasteners

Hi-Lok bolts are aerospace grade fasteners commonly used for bolted composite
structures. Hi-Lok bolts feature an automatic preload feature which controls the
tightening torque that can be applied to them. When the specified level of torque is
reached, based on the collar selection, then the hexagonal tightening section breaks
away from the collar, preventing further tightening. A cross-section of a Hi-Lok bolt is
shown in Figure C-1.

Failure plane

Figure C-1: Hi-Lok bolt cross-section [132]

The fasteners used for this research were made from Ti-6Al-4V, a low corrosion, high
strength titanium alloy commonly used in fasteners when in contact with carbon fibre
composite materials. The manufacturer specified a minimum shear strength of 95 ksi
(655 MPa) for this material.

The following Hi-Lok fastener specifications are the property of Hi-Shear Corporation
and have been reproduced without significant detail due to size restrictions. Please
consult the official Hi-Lok specifications for unabridged versions.

C-1
Appendix C: Hi-Lok Fasteners

Figure C-2: HL523

C-2
Appendix C: Hi-Lok Fasteners

Figure C-3: HL1012

C-3
Appendix C: Hi-Lok Fasteners

Figure C-4: HL97

C-4
Appendix C: Hi-Lok Fasteners

Figure C-5: HL1087

C-5
Appendix D: Marc FORTRAN Subroutines

Checking for element failure in MSC.Marc was achieved via a user programmed
subroutine. Every iteration the subroutine checks for failure. Three user subroutines;
UFAIL, UPROGFAIL and a failure calculation routine function together to check for
failure at each integration point. A schematic of the subroutine structure is shown in
Figure D-1.

Loop each
integration Stress Failure
point
Routine
Failure
Indices
Marc UFAIL

Failure
UPROGFAIL

Figure D-1: Schematic of subroutine structure

2 2 2
     
F1t  11
 12
 13
...(D.1)
X t S ST

2
 
F1c  11
...(D.2)
Xc

2 2 2
     
F2t  22
 12
 23
...(D.3)
X t S ST

D-1
Appendix D: Marc FORTRAN Subroutines

2
 
F2c  22
...(D.4)
Xc

2 2 2 2
       
Fmt  33
 12
 23
 13
...(D.5)
Zt S ST ST

2 2 2 2
       
Fmc  33
 12
 23
 13
...(D.6)
Zc S ST ST

11
F1t  ...(D.7)
Xt

11
F1c  ...(D.8)
Xc

 22
F2t  ...(D.9)
Xt

 22
F2c  ...(D.10)
Xc

 33
F3t  ...(D.11)
Zt

 33
F3c  ...(D.12)
Zc

12
F12  ...(D.13)
S

 23
F23  ...(D.14)
ST

13
F13  ...(D.15)
ST

D-2
Appendix D: Marc FORTRAN Subroutines

A schematic of the UFAIL subroutine is shown in Figure D-2. The failure routine
considers failure in three separate phases; two orthogonal fibre phases and one matrix
phase. The two fibre phases have the same failure criteria but can be failed separately.

Stress tensor

Has element Failure indices


Yes previously (9 per
failed in all integration
three phases? point)

No

Calculate stress state in


each phase (tension or
compression)

Call failure routine

Has any Yes Update failure


phase failed? indices

No

Return
(to Marc)

Call
UPROGFAIL

Figure D-2: Schematic of UFAIL subroutine

D-3
Appendix D: Marc FORTRAN Subroutines

A schematic of the UPROGFAIL subroutine is shown in Figure D-3. The subroutine is


called whenever UFAIL predicts failure at an integration point. Stiffness reduction is
imposed on each phase separately depending on the failure index and the state
(tension or compression) of the applied stress.

Stress tensor
(from UFAIL)

Calculate stress state in Failure


each phase (tension or indices
compression) (9)

Has first fibre Yes Reduce


direction stiffness
failed (t/c)?

No

Reduce Yes Has second


stiffness fibre direction
failed (t/c)?

No

Has matrix Yes Reduce


failed (t/c)?
No stiffness

No

Return
(to Marc)

Figure D-3: Schematic of UPROGFAIL subroutine

D-4
Appendix E: PLINK Property Investigation

The PLINK is a connection element implemented in PAM-CRASH to model mechanically


fastened joints. The links are a highly versatile method of modelling connections that
are a feature of the PAM-CRASH commercial code. They can be deployed between
dissimilar meshes where nodes do not match and can be employed in full nonlinear
transient crash simulations.

The PAM-CRASH Solver Manuals [86, 88] do not present a very detailed description of
the PLINK behaviour, particularly in the regime between initial failure and final
element elimination. Previous reports and papers [59, 61, 145, 159] have addressed
various aspects of the element behaviour. Advancements in the code have made
minor changes to the PLINK element, and so an updated version of the element
behaviour will be presented here. All PLINK data presented in this appendix was
current in PAM-CRASH version 2007. This work was presented in the author's final
report presented to DLR in December 2007.

The true behaviour of the element was checked using a single element model
connecting between two rigid plates. The end displacements were varied without
introducing a moment into the link. The formulations presented here for the element
behaviour are reverse engineered or interpreted by testing the response of PAM-
CRASH on simple problems, so may not be implemented mathematically exactly as
described here. The formulations presented here do however match every benchmark
test yet conducted.

E-1
Appendix E: PLINK Property Investigation

At any point in the solution, a penalised gap vector can be created between the ideal
vector R and the real vector R. This penalised vector can be broken down into normal
and tangential components as shown in Figure E-1.

Figure E-1: Vector breakdown of PLINK element [86]

Given the normal displacement (dn) the tangential or shear displacement (d s), and the
user defined stiffness (K), strength (AFAIL) and power law exponents (a), a complete
picture of the ideal element behaviour can be established.

At any given time step the following values can be defined:

NE  K n d n
SE  K s ds
RE  NE2  SE2 ...(E.1)
d  d n2  ds2
a as
 NE   SE 
n

F 


AFAILN AFAILS

where

N = Normal force transmitted by the link


S = Shear force transmitted by the link
R = Magnitude of the force transmitted by the link
d = Magnitude of the penalized gap vector
F = Failure index.

E-2
Appendix E: PLINK Property Investigation

Until F 1 the forces transmitted by the link, N and S, are given by their elastic values
(NE and SE). If at any time step F 1 for the first time then,

RF  RE
...(E.2)
dF  d

where subscript F signifies failure. Once the failure values are set for each PLINK they
are permanent are not changed within the model, which leads to some very strange
behaviour under certain loading conditions. One complicating factor is that one failure
mechanism is defined by force and the other by displacement, which are not
interchangeable units when user defined stiffness values are defined.

After failure is initiated, two new functions are defined, and , which are plotted in
Figure E-2 and Figure E-3.

1.0

RF RE

Figure E-2: Total element force ()


Loading Unloading

1.0

dF dF+d1 dF+d2 d

Figure E-3: Displacement based softening function ()

The element forces are then given by the Equations E.3.

E-3
Appendix E: PLINK Property Investigation

NE *
N
 ...(E.3)
S*
S E


A few important observations must be made about this set of equations. Firstly and
are measured against different variables which are only interchangeable if KS = KN.
Therefore situations can arise when the failure index (F) is less that 1 but the element
is eliminated due to displacement increase. This may occur if the loading direction
changes during the simulation.

Similarly, once the failure is triggered, all information about strengths (AFAILN and
AFAILS) is lost to the element. This can result in forces much larger than AFAILN being
carried in the normal direction. For instance, in an impact simulation where the initial
shear load is replaced by normal load as the tension and bending waves impinge on
the joint. In this situation, pictured in Figure E-4, the initial failure happens at a load
level of RF = AFAILS. If the loading direction gets changed the joint is capable of
carrying a load in the normal direction which is independent of AFAILN.

AFAILS.KN
KS

AFAILN

AFAILS S

Figure E-4: Irregular loading condition

This is an extreme case which is unlikely to be replicated exactly in a solution, but it is


quite common for it to occur to a lesser degree in many simulations, especially when

E-4
Appendix E: PLINK Property Investigation

there is an initial high shear displacement. A solution could be to normalise the forces
with the failure index function (F) so that for any combination of N and S the load
cannot exceed the failure surface. This is a numerically concise solution but does still
not address the Force-Displacement incongruence created by the user defined
stiffness values. A much more consistent approach would be to move all the failure
prediction into the displacement space, so that user defined stiffness values only
appear in the load calculation rather than the failure calculation. Three critical
displacements could be set for both the normal and shear directions, which would
allow a much more flexible and consistent set of behaviour for the link. A
mathematical treatment of this approach is given later.

The stiffness of the PLINK can be defined by user defined curves. The stiffness this
refers to is not stiffness at all, but load carrying capacity. The load-deflection curve is
entered for the normal and tangential components of loading. With this option the
load deflection response of the PLINK can be very accurately modelled for loading in
pure shear or pure normal force.

It has two major drawbacks however. Firstly there is no displacement based failure
that can be imposed without first failing via a force method, so although a complex
curve can be used, the element will not fail and if the load is reversed it will unload
along exactly the same curve. This is similar to the way the PLINK behaves normally so
it is not a major problem.

The other, more important issue with this formulation is that there is no compensation
when load is not applied along one of the primary directions. If the applied
displacement combines normal and shear components then the link can hold the full
shear load and normal load from the equivalent curves. Therefore the link is much
stronger when loaded at 45 degrees than it is when loaded in shear or normal
directions. It is for this reason that the curves method is unsuitable for use with the
PLINK.

E-5
Appendix E: PLINK Property Investigation

Breitweg [145] showed that the performance of the PLINK element is very dependent
on the connectivity between the two parts. One suggested remedy to this presented in
the PAMCRASH documentation is the multiple PLINK approach. The program attempts
to smooth the properties of one PLINK out among a user specified number of satellite
PLINKs. In this way, it ensures that there is a mixture of connectivity conditions and in
principle the response should be similar regardless of the positioning of the initial
node.

Along with the normal PLINK definition the user must specify a radius and number of
satellite PLINKs to be created (at least 2). During the initial phase of the solution, the
specified number of PLINKs is created at equal angular spacing around a circle of the
specified radius, as shown in Figure E-5.

Figure E-5: Multiple PLINK connectivity

The satellite connections can connect any sections (dark grey) that share a node with
the initial section of connectivity (light grey).

E-6
Appendix E: PLINK Property Investigation

Figure E-6: PLINK test mesh

The initial test was conducted with a central PLINK the centre of the middle element
with NCN connection. The PLINK was then moved to an adjacent edge and another test
was run with the same loading conditions. Finally, to gauge the effect of the multiple
PLINKs, the edge connected PLINK had 8 satellite elements added at a radius of 0.99
mm (element edge length = 1 mm). The results for the centrally connected element
were then compared to the two variations of the edge connected element. The
satellite PLINK arrangement is shown in Figure E-7. The loading in all cases was a
simple shear displacement to failure.

Figure E-7: Satellite PLINK arrangement

The edge connected PLINK, as expected, was significantly different from the centrally
loaded link. When the satellite links were added, there was a significant improvement
to the strength and post-failure response, but the stiffness had increased dramatically.
The stiffness was increased by a factor of the number of links connected, which
suggests that this option has not been properly coded to handle user defined stiffness.
The results are shown in Figure E-8.

E-7
Appendix E: PLINK Property Investigation

Shear Force Central Link


Shear Force Edge Link
Shear Force Multi-link
Load

Displacement

Figure E-8: Effect of multiple PLINKs

E-8

Vous aimerez peut-être aussi