The Numerical Prediction of Panel

Dent Resistance Incorporating Panel

Forming Strains

by

Dylan Thomas

A thesis
presented to the University of Waterloo
in fulfillment of the
thesis requirement for the degree of
Master of Applied Science
in
Mechanical Engineering

Waterloo, Ontario, 2001

Dylan
c Thomas, 2001
I hereby declare that I am the sole author of this thesis. This is a true
copy of the thesis, including any required final revisions, as accepted by my
examiners.

I understand that my thesis may be made electronically available to the

public.

ii
 Acknowledgements

I would like to thank Mark Finn, Larry Sutak and Kevin Gong of AL-
CAN for their help and guidance during this research. Mark, your modelling
expertise helped immensely, thanks for letting me use your computing re-
sources. Larry, your knowledgeable guidance and help during dent testing
was much appreciated.
Thanks to Michael Worswick for his guidance during this research. I
owe Nathan Dwyer a debt of gratitude for his help with dent testing and
submitting models. Thanks also go out to Steve Truttmann, Young Lee and
Xavier Lalbin for their contributions during the research.
I would like to acknowledge the funding recieved by ALCAN, OGS and
NSERC without which this research would not have been possible.
Finally, thanks Tasha for your patience, love and support during this
research.

iii
 Dedication

To Mom, Dad
and Tasha.
Thanks for all the love
and support.

iv
 Abstract
This thesis presents a numerical method of predicting both static and
dynamic denting phenomena in automotive body panels. The finite element
method is used as a predictive tool to assess panel performance prior to
production of tooling. A custom software package has been developed to
transform existing finite element forming models into “ready-to-run” finite
element denting models, minimising the effort required to perform dent sim-
ulations.
Over 50 multi-step finite element models were performed. Each of these
models simulated the forming, springback and subsequent denting of either
1.05mm thick AA5754, or 0.81mm, 0.93mm or 1.00mm thick AA6111 alu-
minum sheet
Experimental validation of dent predictions using this method has shown
that the trends in both static and dynamic dent resistance have been cap-
tured quite well. These validation studies demonstrated the sensitivity of
the results to various parameters such as panel thickness, pre-strain, curva-
ture and thickness, as well as numerical formulation parameters. It has been
determined that it is particularly important to use forming data within the
denting models for accurate results to be obtained.

v
Contents

1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Automotive Body Panels . . . . . . . . . . . . . . . . . . . . . 2
1.2.1 Automotive Panel Stiffness . . . . . . . . . . . . . . . . 3
1.2.2 Panel Dent Resistance . . . . . . . . . . . . . . . . . . 5
1.2.3 Panel Stiffness and Dent Resistance Testing . . . . . . 11
1.3 Finite Element Method Applied to Dent Resistance Predictions 13
1.3.1 Forming Models . . . . . . . . . . . . . . . . . . . . . . 13
1.3.2 Springback Models . . . . . . . . . . . . . . . . . . . . 15
1.3.3 Static Dent Models . . . . . . . . . . . . . . . . . . . . 16
1.3.4 Dynamic Dent Models . . . . . . . . . . . . . . . . . . 17
1.4 Current Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2 D-Mesh Software 21
2.1 Software Overview . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.1 Node Class . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.2 Shell Element Class . . . . . . . . . . . . . . . . . . . . 24
2.2.3 Solid Element Class . . . . . . . . . . . . . . . . . . . . 24
2.3 External Libraries . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.1 Standard Template Library . . . . . . . . . . . . . . . 26
2.3.2 MOTIF Library . . . . . . . . . . . . . . . . . . . . . . 26
2.3.3 Tab Library . . . . . . . . . . . . . . . . . . . . . . . . 27

vi
 2.4 Program Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4.1 Input Routines . . . . . . . . . . . . . . . . . . . . . . 27
2.4.2 Element Refinement and Stress Remapping . . . . . . . 28
2.4.3 Panel Curvature Approximation . . . . . . . . . . . . . 33
2.4.4 Removal of Mid-Edge Nodes . . . . . . . . . . . . . . . 38
2.4.5 Indentor Positioning . . . . . . . . . . . . . . . . . . . 38
2.4.6 Output Routines . . . . . . . . . . . . . . . . . . . . . 39

3 Materials Tested 44
3.1 Material Selection . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.2 Material Characterisation . . . . . . . . . . . . . . . . . . . . 45

4 Experimental Procedures 56
4.1 Panel Stamping . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.1.1 Specimen Preparation . . . . . . . . . . . . . . . . . . 61
4.1.2 Tooling Setup . . . . . . . . . . . . . . . . . . . . . . . 62
4.2 Paint Bake Cycle . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.3 Static Dent and Panel Stiffness Tests . . . . . . . . . . . . . . 63
4.4 Dynamic Dent Tests . . . . . . . . . . . . . . . . . . . . . . . 68

5 Finite Element Models 74

5.1 Forming Models . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.1.1 Finite Element Mesh . . . . . . . . . . . . . . . . . . . 76
5.1.2 Element Formulation . . . . . . . . . . . . . . . . . . . 80
5.1.3 Contact . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.1.4 Friction . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.1.5 Drawbeads . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.1.6 Displacement Boundary Conditions . . . . . . . . . . . 83
5.1.7 Anisotropy . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.1.8 Forming Model Output . . . . . . . . . . . . . . . . . . 86
5.2 Springback Models . . . . . . . . . . . . . . . . . . . . . . . . 86
5.2.1 Finite Element Mesh . . . . . . . . . . . . . . . . . . . 86

vii
 5.2.2 Boundary Conditions . . . . . . . . . . . . . . . . . . . 87
5.2.3 Element Formulation and Material Models . . . . . . . 87
5.2.4 Springback Solution Control Parameters . . . . . . . . 88
5.2.5 Springback Model Output . . . . . . . . . . . . . . . . 88
5.3 Remeshing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.4 Denting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.4.1 Panel Stiffness and Static Denting . . . . . . . . . . . . 91
5.4.2 Element Formulation and Material Properties . . . . . 91
5.4.3 SIGINI and HARDINI User Subroutines . . . . . . . . 92
5.4.4 Contact . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.4.5 Boundary Conditions . . . . . . . . . . . . . . . . . . . 93
5.4.6 Static Denting Model Output . . . . . . . . . . . . . . 94
5.5 Dynamic Denting and Final Springback . . . . . . . . . . . . . 94
5.5.1 Element Formulation and Material Properties . . . . . 94
5.5.2 Contact . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.5.3 Boundary Conditions . . . . . . . . . . . . . . . . . . . 95
5.5.4 Dynamic Denting Springback . . . . . . . . . . . . . . 95
5.5.5 Dynamic Denting Model Output . . . . . . . . . . . . 95

6 Experimental and Numerical Results 96

6.1 Panel Forming and Springback . . . . . . . . . . . . . . . . . . 96
6.1.1 Blank Draw-In and Final Shape . . . . . . . . . . . . . 96
6.1.2 Panel Thickness . . . . . . . . . . . . . . . . . . . . . . 99
6.1.3 Forming Strains . . . . . . . . . . . . . . . . . . . . . . 100
6.2 Stiffness and Static Dent Tests . . . . . . . . . . . . . . . . . . 103
6.2.1 AA5754 Panel Load-Displacement Responses . . . . . . 104
6.2.2 AA6111 Panel Load-Displacement Responses . . . . . . 108
6.2.3 Panel Stiffness . . . . . . . . . . . . . . . . . . . . . . . 114
6.2.4 Residual Dent Depths for the AA5754 panels. . . . . . 115
6.2.5 Residual Dent Depths for the AA6111 Panels. . . . . . 117
6.3 Dynamic Dent Tests . . . . . . . . . . . . . . . . . . . . . . . 126
6.3.1 Dynamic Dent Response of the AA5754 panels. . . . . 126

viii
 6.3.2 Dynamic Dent Response of the AA6111 panels. . . . . 127
6.3.3 Sensitivity of the Dynamic Denting Results to Finite
Element Model Parameters. . . . . . . . . . . . . . . . 132

7 Discussion and Conclusion 137

7.1 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
7.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
7.3 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . 141

ix
List of Tables

3.1 Summary of Tensile Tests Performed . . . . . . . . . . . . . . 48

3.2 Results of Tensile Tests . . . . . . . . . . . . . . . . . . . . . . 50
3.3 Aluminum Yield Strength Curve Coefficients . . . . . . . . . . 51

4.1 Summary of Dynamic Dent Tests Performed . . . . . . . . . . 70

5.1 Number of Finite Elements in Each Blank . . . . . . . . . . . 79

6.1 AA5754 Panel Draw-in Summary. . . . . . . . . . . . . . . . . 98

6.2 AA6111 Panel Draw-in Summary. . . . . . . . . . . . . . . . . 98
6.3 Panel Thickness after Forming. . . . . . . . . . . . . . . . . . 99
6.4 Initial Panel Stiffness Summary. . . . . . . . . . . . . . . . . . 115
6.5 Secondary Panel Stiffness Summary. . . . . . . . . . . . . . . . 116
6.6 Final Panel Stiffness Summary. . . . . . . . . . . . . . . . . . 116

x
List of Figures

1.1 A Typical Autobody Panel Stiffness Plot . . . . . . . . . . . . 3

1.2 Static Dent Energy for Two Panels of Different Stiffness . . . 7
1.3 Dynamic Dent Depth Predictions of Worswick et al. . . . . . . 10
1.4 Static Dent Depth Predictions of Worswick et al. . . . . . . . 11
1.5 Steps of a Finite Element Denting Prediction . . . . . . . . . . 14

2.1 D-Mesh Input Tab Window . . . . . . . . . . . . . . . . . . . 23

2.2 Node Class Properties . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Element mid-edge and Vertex Neighbours . . . . . . . . . . . . 25
2.4 Element Class Properties . . . . . . . . . . . . . . . . . . . . . 25
2.5 Solid Element Class Properties . . . . . . . . . . . . . . . . . . 25
2.6 D-Mesh Remeshing Tab Window . . . . . . . . . . . . . . . . 29
2.7 Refinement of Shell Elements . . . . . . . . . . . . . . . . . . 29
2.8 Remeshing Parameters . . . . . . . . . . . . . . . . . . . . . . 31
2.9 Area Remeshing Function . . . . . . . . . . . . . . . . . . . . 31
2.10 Area Based Remeshing of A Finite Element Forming Model . 32
2.11 Shell Element Parametric Coordinate System . . . . . . . . . 35
2.12 Shell Element Parametric Coordinate System . . . . . . . . . 35
2.13 Shell Element Normal Calculation . . . . . . . . . . . . . . . . 36
2.14 Surface Tangent Calculation For a Quadrilateral Shell . . . . . 36
2.15 Surface Tangent Approximation for a Triangular Shell . . . . . 37
2.16 Curvature Approximation During Refinement . . . . . . . . . 38
2.17 Mid-Edge Node Removal Rules . . . . . . . . . . . . . . . . . 40
2.18 D-Mesh Indentor Tab Window . . . . . . . . . . . . . . . . . . 40

xi
2.19 Positioning of an Indentor . . . . . . . . . . . . . . . . . . . . 41
2.20 D-Mesh Output Tab Window . . . . . . . . . . . . . . . . . . 42

3.1 Automated Tensile Testing Robot . . . . . . . . . . . . . . . . 46

3.2 Tensile Test Extensometers . . . . . . . . . . . . . . . . . . . . 47
3.3 Tensile Test Specimen . . . . . . . . . . . . . . . . . . . . . . 49
3.4 AA5754-O True Stress - Effective Plastic Strain Curve . . . . 52
3.5 AA6111 0.81mm True Stress - Effective Plastic Strain Curve . 53
3.6 AA6111 0.93mm True Stress - Effective Plastic Strain Curve . 54
3.7 AA6111 1.0mm True Stress - Effective Plastic Strain Curve . . 55

4.1 Sketch of IRDI Panel . . . . . . . . . . . . . . . . . . . . . . . 57

4.2 Curvature and Dent locations of the IRDI Panel . . . . . . . . 57
4.3 IRDI Press . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.4 IRDI Die Set . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.5 IRDI Binder Configuration . . . . . . . . . . . . . . . . . . . . 61
4.6 Wrinkling of the IRDI Pan . . . . . . . . . . . . . . . . . . . . 64
4.7 Static Dent Experimental Setup . . . . . . . . . . . . . . . . . 64
4.8 Closeup of a Static Dent Test . . . . . . . . . . . . . . . . . . 65
4.9 Bubble-Type Level . . . . . . . . . . . . . . . . . . . . . . . . 66
4.10 Digital Gauge Measurement of Displacement . . . . . . . . . . 67
4.11 Dynamic Dent Experimental Setup . . . . . . . . . . . . . . . 69
4.12 Measurement of Panel Location . . . . . . . . . . . . . . . . . 72
4.13 Plumb-Bob and Magnetic Coil . . . . . . . . . . . . . . . . . . 73

5.1 Finite Element Mesh of the IRDI Tooling . . . . . . . . . . . . 78

5.2 Adapted Finite Element Mesh . . . . . . . . . . . . . . . . . . 79
5.3 Schematic of a Drawbead . . . . . . . . . . . . . . . . . . . . . 82
5.4 IRDI Drawbead Restraining Forces . . . . . . . . . . . . . . . 84
5.5 IRDI Drawbead Normal Forces . . . . . . . . . . . . . . . . . 84
5.6 Die and Binder Velocity Boundary Condition. . . . . . . . . . 85
5.7 A Refined Finite Element Mesh . . . . . . . . . . . . . . . . . 90

xii
6.1 Draw-In of the AA5754 Panel . . . . . . . . . . . . . . . . . . 97
6.2 Predicted AA5754 Panel Strains - X-Direction . . . . . . . . . 101
6.3 Predicted AA5754 Panel Strains - Y-Direction . . . . . . . . . 101
6.4 Predicted AA6111 1.0mm Panel Strains - X-Direction . . . . . 102
6.5 Predicted AA6111 1.0mm Panel Strains - Y-Direction . . . . . 102
6.6 Predicted AA6111 1.0mm Panel EPS Strains . . . . . . . . . . 104
6.7 Static Load vs Displacement of indentor on 1.05mm AA5754
at Point A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.8 Static Load vs Displacement of indentor on 1.05mm AA5754
at Point B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.9 Static Load vs Displacement of indentor on 1.05mm AA5754
at Point C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.10 Static Load vs Displacement of indentor on 1.05mm AA5754
at Point D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.11 Static Load vs Displacement of indentor on 1.05mm AA5754
at Point E . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.12 Static Load vs Displacement of indentor on 0.81mm AA6111
at Point A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.13 Static Load vs Displacement of indentor on 0.81mm AA6111
at Point B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.14 Static Load vs Displacement of indentor on 0.81mm AA6111
at Point C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.15 Static Load vs Displacement of indentor on 0.93mm AA6111
at Point A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.16 Static Load vs Displacement of indentor on 0.93mm AA6111
at Point B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.17 Static Load vs Displacement of indentor on 0.93mm AA6111
at Point C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.18 Static Load vs Displacement of indentor on 1.00mm AA6111
at Point A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

xiii
6.19 Static Load vs Displacement of indentor on 1.00mm AA6111
at Point B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.20 Static Load vs Displacement of indentor on 1.00mm AA6111
at Point C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.21 Residual Dent Depth vs Peak Load on 1.05mm AA5754 at
Point A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.22 Residual Dent Depth vs Peak Load on 1.05mm AA5754 at
Point B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.23 Residual Dent Depth vs Peak Load on 1.05mm AA5754 at
Point C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.24 Residual Dent Depth vs Peak Load on 1.05mm AA5754 at
Point D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.25 Residual Dent Depth vs Peak Load on 1.05mm AA5754 at
Point E . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.26 Residual Dent Depth vs Peak Load on 0.81mm AA6111 at
Point A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.27 Residual Dent Depth vs Peak Load on 0.81mm AA6111 at
Point B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.28 Residual Dent Depth vs Peak Load on 0.81mm AA6111 at
Point C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.29 Residual Dent Depth vs Peak Load on 0.93mm AA6111 at
Point A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.30 Residual Dent Depth vs Peak Load on 0.93mm AA6111 at
Point B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.31 Residual Dent Depth vs Peak Load on 0.93mm AA6111 at
Point C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.32 Residual Dent Depth vs Peak Load on 1.00mm AA6111 at
Point A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.33 Residual Dent Depth vs Peak Load on 1.00mm AA6111 at
Point B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

xiv
6.34 Residual Dent Depth vs Peak Load on 1.00mm AA6111 at
Point C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
6.35 Residual Dent Depth vs Peak Load for three thicknesses of
AA6111 at Point A . . . . . . . . . . . . . . . . . . . . . . . . 125
6.36 Dynamic Dent Test Results - 1.05mm AA5754 Panels . . . . . 126
6.37 Dynamic Dent Test Results - 0.81mm AA6111 Panels . . . . . 129
6.38 Dynamic Dent Test Results - 0.93mm AA6111 Panels . . . . . 129
6.39 Dynamic Dent Test Results - 1.00mm AA6111 Panels . . . . . 130
6.40 Effect of AA6111 Panel Thickness at Point A . . . . . . . . . 130
6.41 Effect of AA6111 Panel Thickness at Point B . . . . . . . . . . 131
6.42 Effect of AA6111 Panel Thickness at Point C . . . . . . . . . 131
6.43 Sensitivity of Dent Depth to Finite Element Parameters. . . . 133

7.1 Yield Function Comparison . . . . . . . . . . . . . . . . . . . 139

xv
Chapter 1

Introduction

1.1 Motivation
A key performance indicator limiting the potential usage of aluminum
alloy sheet in automotive body panels is dent resistance. This material is
attractive in automotive body applications due to the weight savings and
resulting fuel economy improvements over current steel autobodies that are
necessary to meet current and anticipated legislative fuel consumption stan-
dards. The cost to replace steel body panels with aluminum alloy sheet is
roughly $1.00 per pound of steel removed which currently represents a bar-
rier to the widespread use of aluminum. A significant component of this cost
penalty is introduced through the need to utilise thicker gauges of aluminum
sheet in order to meet panel stiffness and dent resistance requirements. The
opportunity to reduce panel thickness exists, but a fundamental understand-
ing of the mechanical and material parameters controlling the denting pro-
cess and the interaction between forming and denting are needed to allow an
engineered approach to panel optimisation.

1
1.2 Automotive Body Panels
Historically, automotive body panels have been manufactured from steel
sheet. The sheet is stamped into the required shape using tooling which
is mounted in a large mechanical press. Two stampings are required for
each automotive panel, known as the “inner” and the “outer”. The outer
or closure panel embodies the “shape” of the vehicle and is also subject to
external loads due to wind loading, palm-prints and hail, for example. The
inner panel is shaped to provide structural support (stiffness) for the closure
panel and is attached to the car structure. The inner is joined to the outer
using a hemming process that bends the edges of the outer panel around
the inner. In addition, a heat-curable adhesive known as mastic is used to
bond inner regions of the panels together. The panels are then joined to the
frame of the automobile and can either be stressed members, as in a unibody
construction, or non-stressed members, as in a space-frame construction.
Recent advances in vehicle construction have seen the replacement of
conventional steel sheet with higher strength, lighter gauge steels and in some
cases aluminum. Typically the use of aluminum panels has been limited to
those that can be bolted on, such as hoods, liftgates and decklids. Advances
in material processing has led to the development of heat-treatable steels and
aluminum alloys with good formability. These materials typically have a low
yield strength to allow reasonable formability, but when heated for a period
of time exhibit a yield strength increase. The hardening kinetics of materials
used in closure panels have been optimised to harden during the paint bake
cycle of the manufacturing process. Examples of such materials are North
American steels 210B BA and 210 CA, European steel 220B, and aluminum
alloys such as AA6111 and AA6016.
All automotive body panels require several desirable characteristics in
their final form. Panel stiffness has to be sufficiently high so that the outer
panels can resist externally applied loads and fluttering due to wind loading
at speed. Panel dent resistance is also desirable so that damage does not
occur during manufacturing, shipping and end use. Customer awareness of

2
these two panel characteristics has been increased through recent marketing
strategies of Saturn and Volkswagen, for example. The factors affecting
panel stiffness and dent resistance in automotive panels are discussed in the
following sections.

1.2.1 Automotive Panel Stiffness

Automotive panel compliance is defined as the rate at which the panel
displaces with respect to a normally applied load. Conversely, stiffness is the
inverse of compliance. A typical curve of applied load versus panel displace-
ment is shown in Figure 1.1. This figure highlights three distinct stiffness
regions observed in many curved automotive panels. These stiffness regions
are known as the initial (1), secondary (2) and final stiffnesses (3).

Load Versus Displacement

2
Increasing Load

Oil-canning
(Soft)
1

Increasing Displacement

Figure 1.1: A typical autobody panel stiffness plot showing the

three stiffness regions in a curved panel and the point
at which the oil-canning response occurs.

The initial panel stiffness is known to be a function of elastic material

properties such as Young’s modulus, material thickness, curvature and sup-

3
porting conditions (van Veldhuizen et al. [1]). In the initial stiffness region,
the applied load is supported primarily by bending and compressive mem-
brane stresses in the panel. As the applied load is increased, automotive
panels exhibit a geometric softening response known as oil-canning. This
panel softening occurs as the local curvature of the panel changes from a
convex shape to a concave shape. The name “oil-canning” was given to this
response as the base of old-style oil cans exhibited the same softening be-
haviour when pressed during use. This oil-canning response has two types,
hard and soft. In a soft oil-canning response, the automotive panel remains
in contact with the indentor, while in a hard oil-canning response, the panel
can snap through, moving away from the indentor. In Figure 1.1, a soft
oil-canning response is shown at the transition between initial and secondary
stiffnesses. During the secondary stiffness region, the load is carried by a
combination of bending and tensile membrane stresses. The panel stiffness
increases again during the final stage as more of the load is carried by mem-
brane tension. Often, the panel supporting conditions affect the stiffness in
this region as well.
Several studies have been undertaken in order to understand the factors
affecting panel stiffness. Mahmood [2] and Vadhavkar et al. [3] have stated
that the initial stiffness of an unsupported autobody panel can be estimated
using the following equation that was derived from spherical shell theory:

9.237Et2 hπ 2
S= √ (1.1)
kL1 L2 1 − ν 2

where S is the panel stiffness, E is Young’s modulus, t is the material thick-

ness, ν is Poisson’s ratio and L1 ,L2 are the unsupported panel lengths. The
value of k is 7.5 for 4 ≤ ht ≤ 15 and 6.3 for 15 ≤ ht ≤ 60. The panel crown
height (h) can be estimated using:

L21 L2
h= + 2 (1.2)
8R1 8R2

4
where R1 ,R2 are the principal radii of curvature, corresponding to L1 ,L2 .
Unfortunately, as shown by Alaniz et al. [4], the applicability of Equation
(1.1) is limited to the initial stiffness region shown in Figure 1.1. There
are no known analytical solutions to calculate stiffness in the secondary and
final regions, within which the deformation required to produce a visible dent
normally occurs.

1.2.2 Panel Dent Resistance

A key distinction that is sometimes lacking in dent resistance studies is
the nature of the dent. There are two types of denting that are of equal
importance, static and dynamic denting. Static denting refers to a gradually
applied load over a small area typified by a hand pushing on a hood. Dy-
namic denting occurs under impact loading typified by a hail stone. The key
difference is the nature of load application, a static dent indicates a slowly
applied force while a dynamic dent is driven by inertia and impact energy.

1.2.2.1 Static Dent Resistance

One of the first studies of static dent resistance was by DiCello and George
[5]. They concluded that the energy required to produce a visible static dent
on a panel could be represented by the equation:

σy2 t4
W =K (1.3)
S
where W is energy input during the loading of the panel via an indentor,
K is a constant, σy is the panel yield strength, t is the panel thickness and
S is the stiffness of the panel. DiCello and George described panels that
could absorb high levels of input energy without producing a visible dent as
having good static dent resistance. This definition of static dent resistance
is different from that suggested by Yutori et al. [6], who described a panel
with good static dent resistance as one that could support a large load prior
to permanent indentation. From experiments performed on steel cylindrical

5
shells of varying radii, Yutori et al. [6] formulated the following equation that
describes the static load required to initiate a dent on a curved panel:

Pd = Ktn σy (1.4)
In Equation (1.4), Pd is the minimum load (in N) required to initiate a dent,
K is a panel dependent constant, t is the sheet thickness (in mm) and n was
experimentally determined to lie between 2.3 ∼ 2.4 for their steel panels.
The yield strength, (σy ) had units of MPa.
Interestingly, DiCello and George [5] and Yutori et al. [6] adopted differ-
ent definitions of panel dent resistance and consequently arrived at conflicting
conclusions regarding the effect that panel stiffness has on dent resistance.
DiCello and George concluded that a less stiff panel was good for dent resis-
tance whereas Yutori et al. concluded the opposite, that stiffer panels were
more dent resistant.
Figure 1.2 illustrates how both conclusions regarding the effect of stiffness
on panel dent resistance are true when the different definitions of static dent
resistance is taken into account. The figure shows the load and unload curves
for two panels of different stiffnesses, but equal thickness and yield strengths.
For illustration purposes, it is assumed that the two panels experience the
same residual dent depth (point 3 in the figure), but that the load to produce
this dent is higher for the stiffer panel (point 1), compared to the more
compliant panel (point 2). DiCello and George defined the static denting
energy to be the work done on each panel; that is, the area under the loading
curve. In Figure 1.2, the lower stiffness panel requires more work to produce
a visible dent (point 3) even though the peak force at point 2 is less than at
point 1. DiCello and George reasoned that this was due to the stiffer panels
exhibiting less elastic deflection and would conclude that the more compliant
panel was more dent resistant. In contrast, Yutori et al. [6] would conclude
that the panel supporting the highest load for a given dent depth, in this case
the stiffer panel, had the superior dent resistance. In light of these conflicting
definitions of static dent resistance, for the purposes of this work, the static

6
 Load Versus Displacement

1 Loading
2
Increasing Load Unloading

3 Static Denting Energy

Increasing Displacement

Figure 1.2: Hypothetical static dent loading and unloading curves

with dent energy for two theoretical panels of equal
thicknesses and yield strength, but of different curva-
tures. The stiffness of curve 1 is higher due to a tighter
radius of panel curvature. Both panels were loaded to
different forces that resulted in an identical, barely vis-
ible dent depth.

dent resistance of a panel will be taken as the load the panel supports prior
to permanent indentation.
Mahmood [7] expanded on the work of DiCello and George, by incorpo-
rating the estimate of stiffness given by Equation (1.1) into DiCello’s Dent
Energy Equation (1.3). Mahmood states that the stiffness of either a shallow
spherical shell [8] or an unsupported rectangular area can be used, resulting
in the following expression for the energy required to statically dent a panel:
√
C2 ka 1 − ν 2 σy2 t2
W = (1.5)
Eh
C2 is a constant that should not be calculated, as this equation is meant

7
for relative comparisons of the dent resistance of similar panels. The term k
equals 7.5 for 4 ≤ ht ≤ 15 and equals 6.3 for 15 ≤ ht ≤ 60. The area of the
panels, a, can be calculated using either the area of a circular or rectangular
unsupported area, whichever fits the dent location best. E, ν, and σy are
Young’s modulus, Poisson’s ratio and the material yield strength. h is the
crown height of the panel and t is the material thickness. Mahmood used
these equations to design liftgate and decklid slam areas that were resistant
to dents.
More recently, van Veldhuizen et al. [1] used the same definition of static
dent resistance as that of Yutori et al. [6] and concluded that a power law
equation similar to Equation (1.4) could be used to determine the load re-
quired to produce small dents. This equation has the form:

α
β
γ
δ σy t
Fδ = C (1.6)
δ0 σ0 t0
where: C is a constant in Newtons, Fδ is the dent force in Newtons, δ is the
dent depth in millimeters, σy is the yield strength in MPa and t is the material
thickness. The denominators of δ0 , σ0 and t0 are factors that ensure the
dimensional integrity of the equation and have magnitudes of 1mm, 1N/mm2
and 1mm respectively. The results of their work on aluminum specimens
yielded the following simplification of Equation (1.6) that could be used to
calculate the load required to produce a barely visible dent of 0.1mm:

F0.1 = 0.557σy1.1 t1.5 (1.7)

where the units are consistent with Equation (1.6), above. It is worthwhile
to note that results of tests performed on steel samples had a similar form,
but showed a dependence on curvature as well. This was confirmed by tests
performed by McCormick et al. [9].

8
1.2.2.2 Dynamic Dent Resistance

There have been several studies published on dynamic denting; one of

the first was by Johnson and Schaffnit [10]. Their experiments consisted of
dropping an indentor onto rigidly constrained, flat, low-carbon steel plates
of different thicknesses and yield strengths. For their work, they defined
dynamic dent resistance as the following ratio:

impact energy
Dent Resistance (Dr ) = (1.8)
dent depth
which they used to compare They were able to establish that the dent resis-
tance of their samples followed the following relationship:

˙ 2
Dr ∝ σy (")t (1.9)
where Dr is the dent resistance of the sample, σy is the sample yield strength
and t is the sheet thickness. They stressed that, for steels, it was imperative
that the yield strength be corrected for the strain rate at which the dent
took place. Unfortunately their experiment was designed so that little or no
rebound of the indentor was noted. This implies that the sheet was absorbing
most of the impact energy through plastic work. It is well known that if most
of the impact energy can be absorbed elastically, then there will be little or
no plastic work on the sheet. This results in better dynamic dent resistance.
This has lead to the conclusion that panels of lower stiffness are better for
dynamic dent resistance as shown by Worswick et al. [11].
In another dynamic dent resistance study, Burley et al. [12] showed that
there exists a linear relationship between impact velocity and dent depth.
Moreover, there exists a finite velocity below which no residual dent will be
observed. These results have been confirmed in many subsequent studies
[12–17]. More importantly, Burley et al. [12] also showed that dynamic dent
resistance was not merely a function of yield strength and thickness, as re-
ported by Johnson and Schaffnit [10], but was affected by factors such as
panel density, modulus of elasticity and curvature.

9
 The work of Nomura et al. [18] concluded that the dent resistance of
curved panels did not always increase with sheet thickness, as was the case
with flat panels. Their work also showed that the flatter panels tended to
have better dynamic dent resistance. Worswick et al. [11] added to this by
examining a full range of curvatures, from flat to highly curved. They showed
that dynamic dent resistance for a panel will reach a maximum at one level of
curvature, but will decrease for both flatter and more curved configurations,
as shown in Figure 1.3. Interestingly, at the curvature for which a minimum
dynamic dent occurred, a maximum static dent occurred. This opposing
behaviour is evident when Figure 1.3 is compared to Figure 1.4.

1.6

1.4

1.2
Dent Depth (mm)

1.0

0.8

0.6 T4, L=200

T4, L=600
T8X, e=.02, L=200
0.4
T8X, e=.02, L=600
T8XP, e=.02, L=200
0.2
T8XP, e=.02, L=600

0.0
0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.01
1/Radius of Curvature (1/mm)

Figure 1.3: Dynamic dent depth predictions of Worswick et al. [11]

for 1.0mm AA6111 sheet of varying prestrain (e), cur-
vature (R), support span (L) and temper. Predictions
are for a 25.4mm ball bearing dropped from 1219.2mm
above the panel.

10
 0.18

L=200, T=1.0
0.16
L=600, T=1.0

0.14 L=200, T=0.9

L=600, T=0.9

0.12 L=200, T=0.8

Dent Depth (mm)

L=600, T=0.8

0.10

0.08

0.06

0.04

0.02

0.00
0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.01
1/Radius of Curvature (1/mm)

Figure 1.4: Static dent depth predictions of Worswick et al. [11] for
AA6111 panels of varying thickness (T), curvature (R),
and support span (L). Predictions are for a 25.4mm ball
bearing dropped from 1219.2mm above the panel.

1.2.3 Panel Stiffness and Dent Resistance Testing

Even though there has been considerable work on panel stiffness [4, 19–
21], static dent resistance [1, 3–7, 9, 16, 19, 20, 22–38] and dynamic dent
resistance [3, 5, 10, 12–16, 19, 34, 37, 39–44], the automotive industry is just
starting to adopt standardised test methods. A recent report published by
the Standardized Dent Resistance Test Task Force (an Auto/Steel Partner-
ship) [34], contains guidelines for performing stiffness and dent tests. Key
points of this report include specification of indentor types, static load ap-
plication methods, indentor velocities for dynamic denting and specification

11
of the depth of a barely visible dent.
Although numerous types of indentors have been used in prior works,
only two types of indentors are recommended for static denting. These are
a steel hemispherical indentor and a flat rubber disk. The steel hemispher-
ical indentor is 25.4mm in diameter and is meant to simulate elbow marks,
luggage rack damage, thumb prints, etc. This indentor is recommended for
use on laboratory prepared specimens, automotive outer body panels, or full
vehicle testing. The flat rubber disk is 50.8mm in diameter and is meant to
simulate palm, knee, hip and large object panel denting on automotive outer
body panels or full vehicles.
For dynamic denting, the Standardized Dent Resistance Test Task Force
recommends use of a 25.4mm steel ball. Experience shows that this type of
indentor can produce sufficient impact energy to generate visible dents when
dropped from heights up to 5 meters.
There has been considerable debate over what constitutes a “barely visible
dent” in the literature. DiCello and George [5] used a dent depth of 0.0254mm
(0.001 inches), which was the smallest dent that could be measured. Seel [17]
determined that on average, a dent first becomes visible at 0.08mm. Other
studies by Nomura et al. [18] and McCormick et al. [9], have designated dent
depths of 0.1mm as barely visible.
The experimental program for the current study was derived from rec-
ommendations based on experience of ALCAN personnel [45] and the test
procedures set out in the Auto/Steel Partnership report [34]. The actual
testing methods used in this work are described in Chapter 4.
It has been mentioned that dent resistance is a function of many coupled
influences such as panel stiffness, curvature, support conditions and material
strength. In addition, the effects of prior work, that is, thickness reduc-
tion and work hardening during panel stamping must be considered when
optimising panels for weight. In an effort to include more of the factors
contributing to panel dent resistance, several researchers have used the fi-
nite element method (FEM) as a means of predicting panel stiffnesses and

12
both static and dynamic dent resistance. An overview of application of the
finite element method to predict panel stiffness and denting characteristics
is discussed next.

1.3 Finite Element Method Applied to Dent

Resistance Predictions
A principle goal of this research is to produce software tools and method-
ologies which can be used with the finite element method (FEM) to accurately
predict panel stiffness and dent resistance. A flow chart of the complete pro-
cess is shown in Figure 1.5. The prediction of a dent requires several steps,
the first being a sheet metal forming analysis. This step is followed by a
springback prediction where the stamped panel is removed from the tooling
and allowed to deform due to the residual stresses in the sheet from the form-
ing stage. Once the finite element model has been allowed to springback, the
mesh in the model must often be locally refined so that an accurate denting
prediction can be performed. This denting prediction can be either static or
dynamic; if a dynamic dent is being simulated, then an additional springback
simulation step is required. Complicating the issue of dent resistance pre-
diction is the number of software programs required to perform this type of
analysis. The individual stages of dent and stiffness prediction are discussed
in the following sections.

1.3.1 Forming Models

An accurate prediction of a dent begins with the initial finite element
forming model. This model must capture the panel stamping process accu-
rately to predict the panel prestrain, shape and thickness changes. In indus-
try, finite element software codes such as LS-DYNA[46], PAMSTAMP[47],
OPTRIS[48] and AUTOFORM[49] are routinely used to predict the forma-
bility of autobody panels. The particular software used for this research was

13
 FORMING

SPRINGBACK

REFINEMENT

STATIC DENT DYNAMIC DENT

SPRINGBACK

Figure 1.5: The steps required to accurately predict the static or

dynamic dent of a panel using the finite element method.
Each bubble requires a finite element simulation and
often different software programs.

14
LS-DYNA[46], an explicit dynamic code that is ideally suited to solving large
deformation, non-linear problems such as sheet metal forming.
Considerable work on finite element sheet metal forming prediction has
been undertaken recently. One only has to search the literature to find that in
recent years many organizations have adopted finite element analysis (FEA)
as part of their design cycle to predict the formability of parts prior to pro-
ducing tooling. In the work by Harpell [50], FEA was used to predict the
limiting draw ratio of aluminum in cylindrical cup drawing. Lamontagne [51]
has incorporated plastic anisotropy into sheet metal forming simulations. Re-
cent benchmark studies as part of the Numisheet conference series [52] have
examined forming of automotive panels as well as other pressing operations.
The focus of this thesis is not on the finite element method, nor on sheet
metal forming predictions using the finite element method. Both of these
topics have been covered extensively in prior works, so only the aspects of
FEA particular to this research will be discussed, in Chapter 5.

1.3.2 Springback Models

Although explicit time integration codes tend to handle large deformation
quite well, they introduce dynamic effects such as oscillation of the panel.
These oscillations must be damped out prior to denting using a quasi-static
or implicit time integration code such as LS-DYNA version 950 or ABAQUS
[53]. This step is known as springback, a procedure used in finite element
analysis to allow the residual stresses in the formed part to find static equilib-
rium without any externally applied loads. The residual stresses may arise
from constraints imposed by tooling, or dynamic effects from the forming
models. Numerical prediction of springback has been covered in a recent
thesis by Zhou [54].

15
1.3.3 Static Dent Models
The first known example of FEM simulation of static denting was in 1980
by Yutori et al. [6] who used FEA to look at strain levels underneath the
indentor as the panel was displaced. Their model used only 30 axi-symmetric
elements and the indentor was described as a point load.
Following this study, the work of Sakai et al. [26] compared finite element
stiffness values to those obtained from laboratory panels. They were able
to show good correlation between panel stiffnesses in both the primary and
secondary stiffness regions; however, convergence issues prevented them from
obtaining a residual dent depth after unloading.
As computing power became more available, the size of the models used
in static dent resistance computations became larger and more complex. In
1990, Alaniz et al. [4] showed the applicability of the finite element method
to predict panel stiffness in the secondary and final stiffness regions, but did
not incorporate panel forming data in their analysis. The material properties
they used were in the as-received condition, which is known to affect accuracy
in predicting the secondary and final panel stiffnesses.
Finite element codes have progressed as well. In 1993, Werner [24] com-
pared the results of static dent tests obtained using three different finite
element codes, MARC [55], ABAQUS [53] and LS-DYNA [46]. Interestingly,
this was the first time a meshed indentor was used instead of a point load at
the dent site. It was also the first time a residual dent depth was obtained
after unloading of the indentor. Unfortunately, poor correlation with the ex-
perimental results was obtained, which may have been due to an insufficiently
refined mesh.
In 1994, Montgomery and Brooks [23] performed several finite element
analyses on laboratory type panels of varying curvature. They were able
to show that for stiffer panels, significant dents can occur at loads slightly
greater than oil-canning load shown in Figure 1.1; however, for less stiff
panels, the dent initiation load was farther along the secondary stiffness
curve.

16
 Sabbagh et al. [22] were the first to use FEA to predict static dent depths
in actual automotive panels. Two doors were modelled, both considering
several dent locations. One of the doors was a complete assembly, including
the inner and outer panel, with the mastic modelled as a linear spring. Their
work was also the first to include local refinement of the finite element mesh in
the denting region in order to gain sufficient resolution. Material stress-strain
data for the finite element simulations was obtained from tensile sections cut
from existing doors. They were able to establish that FEA was a viable
technique to predict the denting characteristics of automotive panels.
Chavali and Song [38] continued the work of Sabbagh et al. by using a
forming simulation instead of tensile data to account for the initial forming of
the door panel. This was the first example of incorporating forming strains
from a finite element analysis directly into a subsequent denting analysis.
The work of Raghavan and Arwashan [36] again used forming strains
from a forming analysis performed using LS-DYNA. The results of this anal-
ysis was subsequently used in a static denting analysis using the LS-NIKE
software code. No local mesh refinement was used between the forming and
denting steps, as the samples tested were small and could be refined prior
to the forming simulation. Interestingly, they concluded that there was a
need for better interfaces between implicit and explicit finite element codes
to solve complex multi-stage forming problems such as denting.

1.3.4 Dynamic Dent Models

There has been very limited use of the finite element method to predict
the dynamic denting characteristics of panels. The first known example was
by Ni [40] in 1976. This study considered a finite element model to predict
the dynamic denting of a flat rectangular plate subjected to an impulsive
load. The predicted dent depths were higher than those measured, which
was attributed to the one-point contact used in the model. No springback
was modelled; however, it was accounted for by running the simulation long
enough to note the range of panel oscillation after impact. This allowed

17
prediction of the dent depth by using the average location of the contact site
during oscillation.
Following this work, no examples of dynamic dent prediction using FEA
was reported until 1995 when Vreede et al. [41] used the method to predict
dynamic denting in small, flat laboratory specimens. These specimens were
modelled using a simple axi-symmetric element formulation that included
strain rate effects for steel. Springback was handled using a technique known
as dynamic relaxation. This method applies a carefully selected damping
matrix to the system of explicit finite element equations, which eventually
allows the system to come to rest. This method has not yet gained acceptance
as an accurate way of predicting springback.
The most recent work on dynamic dent prediction using FEA was by
Worswick et al. [11] in 1997. This study included both static and dynamic
dent predictions of panels of varying curvature, size and material. Spring-
back of the dynamic dents was handled using the implicit LS-NIKE software
code; however, no comparison to experimental results was made. This study
did, however, show that both static and dynamic dent depths have a strong
dependence on material strength, thickness, panel size, curvature, support
conditions and loading. Most interestingly was the conclusion that static
and dynamic denting mechanisms are of a competing nature, such that pan-
els that perform well in static dent tests perform poorly in dynamic dent tests
(Figure 1.3 and 1.4). They concluded that high local panel stiffness, which
favours static dent performance increases contact forces generated during a
dynamic dent to the detriment of dynamic dent performance. Conversely,
the same stiffness allowed a higher static load to be carried by the panel
prior to a static dent occurring.

1.4 Current Work

The use of initial forming models for subsequent dent predictions is a new
concept. Historically, researchers have omitted this step and made simplifica-

18
tions regarding material properties in their finite element dent models. Typ-
ically, in order to capture panel pre-strain, a tensile test would be performed
on a formed panel; however, this is not possible unless a panel has already
been produced. This method does not allow for optimisation of a panel
early-on in the design cycle. Only recently have coupled explicit/implicit
finite element codes become available, such as LS-DYNA and ABAQUS Ex-
plicit/Standard. These codes can simplify the process of transferring data
between the various simulations outlined in Figure 1.5; however, there is still
a requirement for considerable user intervention between each stage.
The finite element models can capture the forming and springback pro-
cesses quite well and could be used for a subsequent denting analysis; how-
ever, the amount of pre-processing work required to convert one of these
models into a subsequent denting model has been prohibitive. This pre-
processing typically includes refinement of the finite element mesh in the
region to be dented and re-mapping the stress and strain values to the new
refined mesh. Positioning of an indentor over the panel finite element mesh
is also necessary, as is setting up the initial conditions for the simulation to
run. For this reason, one of the goals of this research was the development
and validation of a software tool, described in Chapter 2, that could help
manage the preprocessing work.
The second major goal of this research was to perform forming experi-
ments followed by static and dynamic denting that could be used to assess
or validate the finite element predictions. Chapter 3 describes the material
characteristics of the AA5754 and AA6111 aluminum sheet used in these
studies. The forming and subsequent denting of these panels is discussed in
Chapter 4. Finite element modelling of the complete forming, springback and
denting process was undertaken using the commercial finite element codes
LS-DYNA and ABAQUS. These finite element models were used to predict
panel displacement and dent depths as a function of static loading. Dynamic
dent depths were also predicted as a function of indentor velocity. The fi-
nite element models are presented in Chapter 5, while the experimental and

19
numerical results are presented in Chapter 6. Discussions of the experimen-
tal results and numeric predictions are given in Chapter 7 along with the
conclusions drawn from this research.

20
Chapter 2

D-Mesh Software

2.1 Software Overview

Finite element analysis of the autobody panel stamping process is rou-
tinely performed using software such as LS-DYNA [56]. Such forming models
can be used in subsequent dent predictions; however, they typically have in-
sufficient refinement in regions where dent resistance studies are performed.
Forming models also lack the supporting conditions presented by inner pan-
els which, together with the closure panel make up the panel assembly. The
interaction between the inner and closure panel greatly affects the dent re-
sistance characteristics of the autobody closure panel [57].
As a major component of this research, a software program that can lo-
cally refine an existing finite element mesh and remap the plastic strain, and
stress components to the new mesh has been created. This software program
has been entitled D-Mesh. The main input window of D-Mesh is shown in
Figure 2.1. Functions were added to the software to minimise the prepro-
cessing work required to set up a finite element dent analysis model. These
functions include the ability to place an indentor above the dent location and
also to replace adaptive-type constraints with combinations of quadrilateral
and triangular meshes.
D-Mesh was programmed using an object oriented approach and the C++

21
programming language. The Standard Template Library [58] was used to
store and manage memory for large amounts of data. The MOTIF [59]
libraries were used to provide a standard graphical user interface, that could
be ported to any of the UNIX platforms also supporting MOTIF.

2.2 Classes
In object oriented programming, objects are created as a way to group
program data with functions that operate on data. This has the effect of
encapsulating code, so that it can be used in other programs. It also provides
an efficient way to handle large amounts of associated data. In C++, a class
refers to a certain type of object that has been defined. This definition
includes declaration of the data associated with that type of object and the
functions that provide the interface to the data. In the context of C++
programming, data is considered to belong to the class that it was defined
in and is known as a property of that class. C++ functions can also belong
to a class. The ones that do are known as methods. These methods can
only operate on the properties of the class, a typical method known as a set
type is used to set the value of a certain property. Conversely, a get type
method is used to return the current value of a property. These two types of
methods were written for each of the properties defined in classes of the D-
Mesh program. The properties of these classes are outlined in the following
sections.

2.2.1 Node Class

The node class is used to define storage space for the node number, coor-
dinate location, boundary conditions and neighbouring element information
for a given node in the finite element mesh. Figure 2.2 shows the properties
stored within each node object in D-Mesh.
In finite element remeshing software, an element that uses a certain node
is known as a neighbour to that node. If the element uses the node as a

22
 Figure 2.1: D-Mesh user interface showing the input window.

node
node_num = 4
x_loc = 0.5
y_loc = 0.5
z_loc = 0.0
vertex_neighbours = 2,3
midedge_neighbours = 1

Figure 2.2: D-Mesh Node Class showing the class properties stored
within a typical node.

23
corner node, then the element is a vertex neighbour. If the element uses the
node along an edge, then the element is a mid-edge neighbour to the node.
Figure 2.3 shows a finite element mesh containing both vertex and mid-edge
neighbours.
Although not necessary, the storage of neighbouring element information
with each node greatly speeds up the time taken to search for associated
element data when remeshing. As such it was decided to include the element
neighbour data along with each node, at the expense of additional memory.

2.2.2 Shell Element Class

Figure 2.4 shows the properties of each shell element in D-Mesh. These
properties include the element number, the number of nodes the element
contains, the node numbers, thicknesses at each node, an element type iden-
tifier and the section number that the element belongs to. Also included
is the centroid of the element, number of stress components stored at each
integration point, as well as the number of through-thickness points and the
number of in-plane integration points. Any number of stress values can be
stored for each element. Memory is only allocated for the required amount.

2.2.3 Solid Element Class

Figure 2.5 shows the properties of each solid element read into D-Mesh.
Stored within each element is the element number, node connectivity and the
section number that the solid belongs to. The use of this class is currently
limited to storage of solid indentor elements. As such, no gauss point data
is stored with each of these elements, nor can the solid elements be refined.

2.3 External Libraries

When necessary, external programming libraries were used to speed up
the code development process. These libraries can be considered to lie outside

24
 9 6 3 2
5 2
10 7 4 1
4 3
11 8 5 1

Figure 2.3: Connectivity of some elements showing how element 1

is a vertex neighbour to node 1,2,3 and 5, but also a
mid-edge neighbour to node 4.

element
elem_num = 1
num_nodes = 5
node_numbers = 1,2,3,5,4
thicknesses = 1.0,1.1,1.2,0.9
element_type = 1
section_id = 1
centroid = 10.0,5.0,0.0
stress_data = sig11,sig22,sig33,
sig12,sig13,sig23,
eps.....
num_stress_components = 7
num_in_plane_points = 4
num_thickness_points = 5

Figure 2.4: D-Mesh Element Class showing the class properties

stored within a typical element.

element_sol
elem_num = 1
node_numbers = 1,2,3,4,5,6,7,8
section_id = 1

Figure 2.5: D-Mesh Solid Element Class showing the class proper-
ties stored within a typical element.

25
of a standard C++ installation and are highlighted here.

2.3.1 Standard Template Library

The Standard Template Library [58], or STL, is a library written in C++
for object oriented programming. This library provides a group of classes
called containers that are used to hold other objects. An analogy of a very
simple container class would be of a FORTRAN array, which can hold only
one value in each location of the array, such as integer numbers. The con-
tainer classes defined in the STL can hold any type of object and automat-
ically manage memory when these objects are inserted and deleted. The
obvious benefit of this is the ability to access all of the information associ-
ated with an object through the use of only one index. Depending on the
type of container used, these objects can be inserted or deleted at any point
within the container. Within the context of the STL library, a vector is
defined as an STL container that supports random insertion and deletion
of objects, as well as automatic memory management. For this reason, the
STL vector was chosen as the container class for any objects within D-Mesh
that may increase or decrease in number. These objects include the elements
and nodes among others. The STL library also provides basic methods and
iterators for each of its container classes. The methods provide a means to
add and delete objects from within the containers while the iterators can be
considered pointers that reference the different objects.

2.3.2 MOTIF Library

MOTIF [59] is an X Windows toolkit that provides the necessary pro-
gramming tools (called widgets) to easily program a graphical user interface.
These widgets include objects such as text boxes, forms, radio buttons, etc.
This toolkit was chosen over others as it enables platform-independent cod-
ing of the user interface. To date, D-Mesh has been compiled on Silicon
Graphics and Linux operating systems using the same code.

26
2.3.3 Tab Library
The Tab Widget library [60] was downloaded from the internet. It is an
addition to the standard set of MOTIF programming widgets that provides
the tabbed window look of D-Mesh. This software was made available for
free use by Pralay Dakua.

2.4 Program Flow

The D-Mesh program has been structured to allow the user to try differ-
ent remeshing parameters on the existing finite element mesh without having
to re-input data. This is achieved through the use of the MOTIF interface,
which retains all of the user supplied data while a given mesh is being pro-
cessed. After the program has been run once, the output can be examined
to determine if the resulting finite element mesh is sufficiently refined. If
further refinement is necessary, individual values can be changed to fine-tune
the results. The execution process consists of reading in the input file, refine-
ment and remapping stresses within the finite element mesh, approximating
original panel curvature, removal of any adaptive constraints, positioning of
the indentor and output of the new data files. These sections of the program
are discussed further here.

2.4.1 Input Routines

There are three different input file formats supported in D-Mesh. These
are LS-NIKE short, LS-NIKE long and LS-DYNA Keyword. The appropri-
ate input file format is selected using the File Type radio buttons shown in
Figure 2.1. The input file is specified using the Input File Name text box.
The routine that processes this file only reads in the nodes, shell elements,
any adaptive constraints and stress data. This data is stored into two STL
vectors, one for the nodes and one for the shell elements. Each section of
the input file is parsed using the string data entered in the text boxes below

27
the Input File Name text box. The lines before the start of each section of
the file must be unique. The default values are for a LS-DYNA Keyword
format file. Reading in the stress and adaptive constraint data is activated
or deactivated through the use of their associated option buttons.

2.4.2 Element Refinement and Stress Remapping

Figure 2.6 shows the remeshing window of D-Mesh. To refine the mesh,
three parameters must be entered. These parameters are the refinement
location, refinement radius and the number of refinements.
Figure 2.7 shows how individual elements are refined. One refinement
for a quadrilateral shell element consists of subdivision into four smaller
quadrilateral shell elements, while one refinement of a triangular element
results in three smaller quadrilateral elements. The location of any new mid-
edge nodes is calculated to lie halfway between the two vertex nodes lying
on the same edge. This is consistent with the position of nodes used in
LS-DYNA adaptive constraints. The location of the new node that lies in
the center of the old element is calculated as the average of the vertex node
positions.
There are two element paving methods used in D-Mesh. The first is shown
in Figure 2.8. The refinement location specifies the centroid of a sphere that
will encompass all of the elements being refined, while the refinement radius
specifies the radius of this sphere. The refinement location can either be
given as a node number, or a geometric location in X,Y,Z coordinates. If
the centroid of an element lies within the radius of the sphere, the number
of refinements it will receive decreases linearly from the center of the sphere
to one refinement at the periphery. The value at the center of the sphere
is specified using the Number of Refinements text box. If the finite element
mesh consists of several sections (as in a structural model), individual sec-
tions can be omitted from refinement by entering their section identification
number in the Sections to Avoid text box.
The second element paving method available in D-Mesh uses a target

28
 Figure 2.6: D-Mesh remeshing tab window.

a a/2 a/2

b/2
b
b/2

h/2
h
h/2

Figure 2.7: Refinement of quadrilateral and triangular shell ele-

ments by one level.

29
area as a means of choosing whether an element will be refined. This option
is activated by entering a negative value in the Number of Refinements text
box. If this value is negative, the magnitude corresponds to a target element
area at the refinement location. Elements that are larger than this size
will not be refined further. Smooth paving of the element size to the outer
refinement radius is achieved through a target area function. This function
has two regions, the first is a straight line, the second is a parabolic function.
This target area function is plotted in Figure 2.9, the parabolic section is as
follows:
Al − At 2 2z (Al − At ) Al z 2 − 2At (1 − z)
Alt = r − r + (2.1)
R2 (z 2 − 2z + 1) R (z − 1)2 (z − 1)2

where Alt is the local element target area at any distance (r) from the refine-
ment location. R is the user defined refinement radius and At is the target
element area. Al is the area of the largest element that existed in the refine-
ment area prior to refinement. The length of the straight line shown in Figure
2.9 has been set to 0.55 times the refinement radius, 0.55 is also the value
of z in the target area equation above. This constant enables a wide refined
region that is necessary to capture oil-canning and secondary stiffness during
denting. Use of this function has been found to give nicely refined meshes
when adaptive elements have been used in the original forming model. An
example of remeshing using this function is shown in Figure 2.10.
Remapping of Gauss point data is handled differently depending on the
number of shell element in-plane Gauss points used in the input deck. For
reduced integration elements, such as in the Belytschko Lin Tsay element [61],
[62], there is only one in-plane Gauss point. In these cases, the stress and
effective plastic strain values prescribed to each new element are not modified.
In non-reduced integration elements, such as those using the Hughes-Liu
element formulation [63], there exists four in-plane Gauss points. In this case,
each of the in-plane Gauss point data variables is averaged before prescribing
the result to the new shell elements.

30
 No. of 4
Refinements
Refinement 3
Location 2
1 Refinement
Radius

4 Refinements
Refinement Radius = 50mm 50mm

Figure 2.8: Finite element mesh before and after remeshing. 4 re-
finements were specified within a radius of 50mm.

Area Al

Elements with an area In

this region will be refined

At
Elements with an area In this region will not be refined

0
0 0.55R R
Distance From Refinement Location

Figure 2.9: This figure shows the area based remeshing function
used to determine whether an element will be refined.

31
 A

Figure 2.10: A finite element forming model before (a) and after
(b) remeshing using the area-based element paving al-
gorithm. Note that the original subdivided elements do
not hinder the final mesh when this approach is used.

32
2.4.3 Panel Curvature Approximation
As local panel curvature is thought to greatly influence panel stiffness and
dent resistance, a method of approximating panel curvature during refine-
ment has been implemented in D-Mesh. This method uses bi-cubic hermite
f-patches [64, pp 169 - 183] to represent the curved surface of each element
as it is refined. The conventional geometric form of this type of surface patch
is given in matrix notation as:

p(u, w) = UMF BMTF WT (2.2)

where:
u, w ∈ [0, 1]
U = [ u3 u2 u 1 ] (2.3)
W = [ w3 w2 w 1 ] (2.4)
 
2 −2 1 1
 −3 3 −2 −1 
 
MF =   (2.5)
 0 0 1 0 
1 0 0 0
and  
p00 p01 pw w
00 p01
 p11 pw w 
 p10 10 p11 
B=  (2.6)
 pu00 pu01 0 0 
pu10 pu11 0 0
Note that B is actually a 4x3x3 array, with the third dimension having
different values for each Cartesian component of (x, y, z). p(u, w) gives a
position vector in cartesian coordinates to a point lying on the patch surface
at parametric coordinates (u, w). For a general 4-noded shell element, the
local directions of u and w are chosen so that u = 0 lies along the element
edge pointing from node 1 to 4, while u = 1 lies along the edge from node 2
to 3. Similarly, w = 0 lies along the element edge pointing from node 1 to

33
2, while w = 1 lies along the edge from node 4 to 3. This local coordinate
system is shown in Figure 2.11.
Figure 2.12 shows each component of the B array. The vectors p00 , p01 ,
p10 and p11 point from the origin to the corner nodes of each element. The
subscripts of 0 or 1 refer to the local parametric coordinates of u and w.
As an example, p01 would mean the position vector pointing to the node at
u = 0 and w = 1. The superscripted vectors of pu00 , pu01 , pu10 and pu11 are
the surface tangent vectors in the local u direction at each of the four corner
nodes, while pw w w w
00 , p01 , p10 and p11 are the surface tangent vectors in the local
w direction. The length of these surface tangent vectors (pαij ) determines the
amount of curvature to be approximated during refinement.
Calculation of the surface tangent vectors pαij begins with an approxima-
tion of the average surface normal at each node. In D-Mesh, this is calculated
as:

Elneighb.

n̂i
¯= i=1
n̂ (2.7)
Elneighb.
where Elneighb. is the number of element neighbours for the node and n̂i is
the unit normal calculated at the node using the element neighbour. Figure
2.13 shows how the normal ni is calculated for both vertex and mid-edge
neighbouring elements. In all calculations, the vectors V1 and V2 are ordered
so that the resulting cross product, V1 × V2 , gives an outward normal vector
ni whose direction is consistent with the node connectivity right hand rule.
The unit normal for each element is then calculated as:
ni
n̂i = (2.8)
|ni |
Once the average unit normal n̂ ¯ has been calculated, the surface tangent
vectors pαij are calculated as in Figure 2.14 using an intermediate vector a at
each node:

¯
a = eα × n̂ (2.9)

34
 4
u=0 w=1
w
3
1 u
u=1
w=0
2

Figure 2.11: Shell element parametric coordinate system conven-

tion.

p01w
p01u
p00w p11w
w
p00u
u p11u
p10w
p10u
p00 p01
p10
p11

x,y,z(global)

Figure 2.12: P-Vectors that are used to generate the bi-cubic her-
mite f-patch B matrix for each element.

35
 ni
ni
V2 ni
V1 V2

V2
V1
V1
At Vertex At Midedge

Figure 2.13: Shell element normal calculation at a vertex node and

a mid-edge node.

where eα is a vector lying along the element edge that coincides with the
surface tangent vector being calculated. In this example, the pu00 vector is
calculated using:

¯×a
pu00 = n̂ (2.10)

ni

w
p00u
u
a
eu

Figure 2.14: Surface tangent calculation at the corner node of a

quadrilateral shell.

Triangular elements require special treatment for the surface tangent vec-
tors of pu01 and pu11 . In this case there is no edge vector eα since by convention

36
nodes 3 and 4 of the shell element are coincident. For this case only, pu01 and
pu11 are equal and approximated using a vector created between two points
lying 1/10th of the distance along the element edges as shown in Figure 2.15.
This allows equations [2.2] through [2.6] to be used for 3 points instead of
the usual four.

ni

Surface Tangent
Approximating p01u = p11u
Vector w
u

Figure 2.15: Surface tangent of a triangular shell at a corner node

Required input for this function is a value for the curvature scale factor,
λ which is entered on the remeshing tab (Figure 2.6). This value can range
from zero for no curvature to the default value of one. This value scales the
length of the surface in-plane tangent vectors pαij from zero to the length of
the element edge that the in-plane tangent vector corresponds to (eα ). Values
greater than 1.0 should be avoided, as an erroneous surface will result.
A typical result of the bi-cubic hermite approximation is shown in Figure
2.16. The advantage of this method is that the original nodes lie on the
surface and the surface in-plane and normal vectors are also matched at
each original node. This has the effect of creating a smooth and continuous
surface within the region being refined. If curvature is being approximated,
the position of any new nodes created during the refinement step are moved
to lie on the surface.

37
 Before After

Figure 2.16: Finite element mesh before and after curvature approx-
imation using Bi-Cubic Hermite F-Patches.

2.4.4 Removal of Mid-Edge Nodes

In Figure 2.3, node 4 is a mid-edge node. Adaptive constraints are used
during sheet metal forming simulations to constrain these mid-edge nodes to
remain at the element mid-edge during the forming simulation. The pres-
ence of these adaptive constraints was thought to hinder convergence of both
springback and static dent simulations in LS-NIKE. This prompted the addi-
tion of a function that would take an existing finite element mesh with mid-
edge nodes and replace them with combinations of smaller triangular and
quadrilateral shell elements. This function is activated by selecting the Re-
move Adaptivity option button on the Other Options tab. Figure 2.17 shows
the resulting elements created to replace any original elements containing
mid-edge nodes. Any elements that contain these nodes are subdivided in
such a way that the use of triangular elements is minimised.

2.4.5 Indentor Positioning

Currently there are three types of indentors commonly used in the au-
tomotive industry for panel dent resistance and stiffness tests. As an aid in
streamlining the setup of denting simulations, a method of inserting these
indentors into the output mesh has been implemented.

38
 Figure 2.18 shows the D-Mesh tab window related to indentor input and
positioning. This window allows the user to specify the indentor and how
it will be positioned within the output mesh. As a separate input file is
used, any shape of indentor may be used. The only restrictions are that the
file must be in LS-DYNA Keyword format and solid brick elements must be
used. Required input is the indentor file name, as well as strings to parse
out the nodes and solid elements from the input file. The New Indentor
Location text box allows the user to specify a node that the indentor will be
positioned above. The indentor is rotated and translated so that its local
z axis lies along the average outward normal (n̂) ¯ of the surface calculated
at the node. This normal is calculated as in Equation (2.7). The local x-
direction of the indentor is rotated to lie in the plane created by the outward
normal n̂¯ and an element edge vector e. The element chosen for the edge
vector is the lowest numbered element, while the edge is chosen by the right
handed node numbering rule. The distance away from the surface is specified
using the Distance From Node text box, to which half of the shell thickness
can be added. The loading direction can be reversed using the Flip Indentor
Side option button. The nodes of the indentor can be renumbered using the
Renumber Indentor Nodes option. This option requires that the new starting
number for the nodes be entered in the New Starting Number text box. If
the value entered is less than the maximum node number after remeshing,
consecutively renumbered nodes will be output. An example of indentor
positioning is shown in Figure 2.19.

2.4.6 Output Routines

Two output files are created during refinement. The first is a text file
that is used as input for subsequent denting analysis. One of four different
output file formats can be chosen, they are LS-NIKE short, LS-NIKE long,
LS-DYNA Keyword and ABAQUS Standard. The appropriate file format is
selected using the File Type radio buttons shown in Figure 2.20. The Output
File Name text box can be used to change the default output filename.

39
Figure 2.17: Shell element mid-edge node removal rules for each pos-
sible combination of nodes.

Figure 2.18: D-Mesh indentor tab window.

40
 Figure 2.19: Positioning of a 25.4mm solid element indentor normal
to the surface to be dented.

The second output file is a LS-TAURUS Data Base file [65] that contains
data that can only be viewed using commercial viewers such as LS-POST
[66]. This file is written in a proprietary format used mainly by the LS-
DYNA software. The name of this file can be changed by modifying the
D3Plot File Name text box. This binary database file should be used in
conjunction with d3plot file viewers such as LS-POST to verify that the
results of mesh refinement are acceptable.
In the current work LS-DYNA Keyword is the preferred output method
for a dynamic dent analysis while ABAQUS Standard is preferred for a static
dent. For an ABAQUS Standard output file, element stress values are written
to a separate file entitled “stresses.inp”. This is necessary as current versions
of ABAQUS are limited to allow initialisation of shell element stress values at
only one location if the standard input file method is used. For shell elements
this is insufficient as bending stresses cause significant differences in stress
values through the shell thickness. In order to initialise all of the stress
values, the ABAQUS user subroutines SIGINI and HARDINI [67, 25.2.11,
25.2.17] must be used. These input routines require that the initial stress
values be in a local coordinate system for each shell element, so the element
stress components are first rotated from the global to the local coordinate

41
 Figure 2.20: D-Mesh output tab window.

system using:

σuw(local) = aui awj σij(global) (2.11)

where:

u, w & i, j = (1, 2, 3)

In Equation (2.11), aij represents a rotation tensor from the global Carte-
sian coordinate system to the local system. The individual components can
be obtained from a unit triad located at each shell element centroid. This
triad is formed using the same convention as in ABAQUS [53, 3.6.3], whereby
the shell normal acts as the 3rd local coordinate direction, while the 2nd local
coordinate is calculated using a cross product of the shell normal and the
global x-axis. The exception to this rule is that if the x component of the
shell normal lies within 0.1◦ of the global x-axis then the 2nd local coordinate
is calculated using a cross product of the normal and the global z-axis. The

42
1st coordinate direction is then calculated using the cross product of the 2nd
with the 3rd .
Once the local unit triad is known, the individual components of a tensor
that rotates from a local to a global coordinate system is obtained from the
direction cosines of the local unit triad. ie:
 
a11 a12 a13
 
aij =  a21 a22 a23  (2.12)
a31 a32 a33
in which the terms aij represent the direction cosines of the ith principal
directions.
The results of using this stress rotation have shown good agreement with
the thin shell element stress condition that σzz = 0 when stress values were
rotated from an LS-DYNA input deck to the local coordinate system used
within ABAQUS. This was tested on the three main element formulations
used for sheet metal forming, namely the Belytschko Lin Tsay element for-
mulation, Hughes-Liu element formulation and element 16 [68] formulation.
Slight errors can occur in the output for warped elements; however, so care
should be taken to ensure that a well refined mesh is being used.

43
Chapter 3

Materials Tested

3.1 Material Selection

There are several aluminum alloys that have been developed for use in au-
tomotive body panels. Of these, AA6111, AA6016, AA5754 and AA5182 are
commonly used. Candidate materials for outer or closure panels are AA6111
and AA6016, which are heat-treatable alloys with higher yield strengths for
dent resistance. Typically, these alloys are formed in the T4 or unaged temper
and then harden to a T8X condition during the paint bake cycle. Inner pan-
els are fabricated from either AA5754 or AA5182, which are work-hardening
alloys. AA5754 tends to have lower formability than AA5182, but is less
expensive.
A typical automotive paint bake process consists of three heat cycles,
to cure the electro-galvanizing coat, primer paint and top/clear coat paint
applications. These cycles last 15-25 minutes each at 170 - 185◦C, 160 -
170◦C and 130 - 150◦C, respectively. Lloyd [69] has stated that in laboratory
tests considering AA6111, these three cycles can be condensed into a single
representative 32 minute heat treatment cycle at 177◦C. This single heat
treatment cycle achieves the same yield strength increase in AA6111 because
the panels do not have to heat up and cool three times as in the automotive
paint bake process. Several variants of AA6111 undergo pre-treatments to

44
obtain the maximum possible yield strength increase during an automotive
paint bake process [69]. This pre-treatment does not affect the yield strength
in the T4 condition which is good for formability, but does promote higher
hardening rates during the paint bake cycle.
Two aluminum alloys, AA6111 and AA5754, were made available by
ALCAN International Limited for use in this study. Three thicknesses of
AA6111 were supplied, 0.81mm, 0.93mm and 1.00mm. The sheets were
unaged, but pre-treated and were in the T4PD condition (a proprietary T4-
variant with an optimised paint-bake response). Two gauges of 1.0mm and
1.6mm thick AA5754 were also supplied. The AA5754 was unstrained, or in
the O-temper annealed condition. The 1.6mm thick AA5754 sheet was not
used; however, as initial dent tests revealed that the panels were too strong
to obtain dents at the loads considered. The 1.0mm AA5754 sheet used for
this study was sheared from ALCAN coil number 12918117, the coil numbers
of the AA6111 sheets were unknown.

3.2 Material Characterisation

The mechanical properties of the two aluminum alloys (AA6111 and
AA5754) were determined at the ALCAN International Kingston Research
and Development Center (KRDC). An Instron 4200 Series IX robotic ten-
sile testing machine was used to test each sample, as shown in Figure 3.1.
This tensile testing machine was configured to take three initial thickness
and width measurements of each tensile test specimen. Each specimen was
then mounted in the Instron apparatus and two extensometers were affixed
as shown in Figure 3.2. The extensometers measured the axial and transverse
extensions as the tensile test was performed. The load was measured during
the tensile test using a 97 KN capacity Instron load cell.
Rolled sheet metal typically exhibits anisotropy due to the plastic work
done on the sheet. During rolling, the individual grains within the metal
tend to become elongated in the rolling direction and the distribution of

45
 Figure 3.1: Automated tensile testing robot used at ALCAN
KRDC.

slip plane orientations within each grain becomes less random. This texture
creates differences in material yield strength and ductility when measured at
various angles relative to the rolling direction (See Schey [70, p260]). This
texture is often manifest in directional differences in the Lankford Coefficient,
a measure of sheet anisotropy attributed to Lankford et al. [71] and given
by:

"w
R= (3.1)
"t
where R is the Lankford Coefficient, commonly referred to as the R-value.
The width and thickness strains ("w , "t ) are obtained from the uniaxial ten-
sion tests.

46
 Transverse
Extensometer

Axial
Extensometer

Specimen

Figure 3.2: Axial and transverse extensometers affixed to a tensile

specimen prior to testing

In order to quantify the yield strengths and Lankford Coefficients of each

material, the tensile tests summarised in Table 3.1 were performed. A total
of 39 tensile specimens were cut for each sheet thickness to the specifications
shown in Figure 3.3. Twelve specimens were cut at each angle of 0◦, 45◦
and 90◦ to the rolling direction. Six samples at each angle were then tested
prior to heat treatment and six were tested following heat treatment. In
addition, three AA6111 specimens were cut from a formed and paint baked
panel for each gauge. This provided measured values of the yield strength
increase achieved through forming and paint baking for comparison with the

47
finite element forming model predictions. Tensile tests were also performed
on the 1.0mm AA5754 sheet in both the unstrained and strained condition.
For this work, the value of Young’s Modulus that was used for the aluminum
was 70GPa, while Poisson’s ratio was 0.33, as taken from handbook data
[72]. The Lankford Coefficients (R00, R45 and R90) were calculated at ten
percent true strain. The stress-strain curves and R-values were used in the
FEM models as described in Chapter 5.

Table 3.1: A listing of the tensile tests performed on the aluminum

sheet. PB - Paint Baked, RD - Rolling Direction, F -
Strained During Panel Forming
Material Condition Gauge [mm] at 0◦ at 45◦ at 90◦
5754 O 1.05 6 6 6
5754 F 1.05 3
6111 T4 0.81 6 6 6
6111 T4 0.93 6 6 6
6111 T4 1.00 6 6 6
6111 PB 0.81 6 6 6
6111 PB 0.93 6 6 6
6111 PB 1.00 6 6 6
6111 PB,F 0.81 3
6111 PB,F 0.93 3
6111 PB,F 1.00 3

The results of the tensile tests are summarised in Table 3.2, with the
true yield stress versus effective plastic strain curves shown in Figures 3.4 to
3.7. The material stress-strain curves that were used during forming simula-
tions were all taken from regression analyses that fit the following fifth order
polynomial to the rolling direction true yield strength curves.

σ̄ = A¯"5 + B¯"4 + C¯"3 + D¯"2 + E¯" + F (3.2)

In Equation 3.2, σ̄ is the effective true yield stress, "¯ is the effective plastic
strain and A, B, C, D, E and F are coefficients determined from the regression

48
 BILL OF MATERIAL
ITEM QTY MAT'L DESCRIPTION SUPPLY BY

7.50
= =

2.25
= =

1 1
.005
2.14 2.14
= 2.000 =
GAUGE LENGTH

.010
.75 .500

=
C
POINT A POINT B

1.00 RAD TYP

IMPORTANT NOTE 4 NOTE MODIFIED A.C OCT 98

2 ALL SAMPLES TO BE CNC MILLED
3 NOTE ADDED A.C APR 98
4 NO CUTTER DWELL ON TRANSITION RADIUS
3 FINISH CUTS ARE REQUIRED 2 NOTE ADDED A.C MAY 97
3rd LAST CUT ..010" MAT'L TO BE REMOVED 1 1.00 RAD WAS .50 2.14 WAS 2.30 A.C JUNE 96
2nd LAST CUT .005" MAT'L TO BE REMOVED DATE
REV DESCRIPTION DRAWN
LAST CUT .002" MAT'L TO BE REMOVED

EDGE ROUGHNESS TO BE 0.5 um Ra Alcan International Limited

3
BETWEEN POINTS A & B BOTH SIDES KINGSTON LABORATORIES, P.O. Box 8400, Kingston, Ontario

DESIGNED

UNSPECIFIED TOLERANCES DRAWN

ON FRACTIONAL
DIMENSIONS
ALANCAIN RECTANGULAR TENSION
CHECKED
+
- 1/64

ON
UNSPECIFIED TOLERANCES
DECIMAL DIMENSIONS APPROVED TEST SPECIMEN
.X = +
- .010
.XX = +
- .001
.005 DATE
.XXX = +
- 22 NOV 95
.XXXX = +
- SCALE DWG NO.

8.01.04
DO NOT SCALE DRAWING

IF IN DOUBT ASK
NTS
C- 6294 -KRC 4

Figure 3.3: Drawing of tensile test specimen required for use with
the robotic tensile tests at KRDC.

49
analysis. The values of these coefficients are given in Table 3.3, while the
resulting curves are shown in Figures 3.4 to 3.7. During forming operations,
strains can reach values considerably higher than those measurable during
a tensile test. To approximate the material behaviour at the higher strains,
a constant work hardening rate was assumed. This allowed the final work
hardening rates for each material to be approximated as the average rate
during the final 2% strain recorded for each tensile test.

Table 3.2: Results of the tensile tests performed on the aluminum

sheet. PB - Paint Baked, RD - Rolling Direction, F -
Strained During Panel Forming
Material Condition Gauge [mm] σy [MPa] R00 R45 R90
5754 O 1.05 101 0.64 0.75 0.77
5754 F 1.05 152 0.77
6111 T4 0.81 140 0.62 0.66 0.86
6111 T4 0.93 158 0.68 0.54 0.54
6111 T4 1.00 140 0.60 0.71 0.62
6111 PB 0.81 222 0.61 0.66 0.87
6111 PB 0.93 215 0.68 0.55 0.55
6111 PB 1.00 215 0.59 0.68 0.63
6111 PB,F 0.81 258 0.65
6111 PB,F 0.93 253 0.69
6111 PB,F 1.00 252 0.62

Table 3.2 highlights the differences in final yield strengths of the AA5754
and AA6111 sheet. The AA5754 sheet has an initial yield strength of 101MPa
in the O-temper annealed condition. Work hardening this material by only
2-3% strain, which is typical of strain levels in closure panels, increases the
yield strength to 150MPa. This is a 50% increase in strength. The initial
yield strength of 140MPa for the AA6111 sheet is considerably higher than
the initial AA5754 yield strength. By paint baking the AA6111 sheet, a
further increase to 215MPa is realised. The combination of pre-strain and
paint bake leads to a final yield strength of 253 MPa. Thus, the AA6111 sheet

50
 Table 3.3: Listing of the coefficients that resulted from the regres-
sion analysis of the tensile material data. PB - Paint
Baked, RD - Rolling Direction, F - Strained During Panel
Forming
Yield Strength Coefficient
Material
A B C D E F
5754-O, 1.05mm 1113511 -674531 172952 -26398 2903 101
6111-T4, 0.81mm 603652 -376164 103805 -18477 2596 140
6111-T4, 0.93mm 664284 -424787 114182 -18923 2568 158
6111-T4, 1.00mm 706940 -440602 118062 -19734 2618 140
6111-PB, 0.81mm 2072015 -981266 188695 -22164 2401 222
6111-PB, 0.93mm 2116174 -991991 188173 -21884 2407 215
6111-PB, 1.00mm 2848011 -1331566 248236 -26801 2545 215
6111-PB&F, 0.81mm 6008427 -2431145 379395 -32429 2471 258

has the largest final yield strength, which provides greater dent resistant than
AA5754.

51
 450

400

350
0
True Yield Stress [Mpa]

90
300
F
250

200
45 Modelling Data
150
O Tensile Data
100

50

0
0.00 0.05 0.10 0.15 0.20 0.25 0.30
Effective Plastic Strain [-]

Figure 3.4: AA5754-O true stress versus effective plastic strain

curves showing the slight differences in yield strength
when tensile specimens are pulled at in-plane orienta-
tions of 0◦ 45◦ and 90◦ to the rolling direction. The data
points used within the finite element models are shown
as circles, these points correspond to the rolling direc-
tion. The curve from the formed panels is also shown.

52
 450

400
PB,F PB
350
True Yield Stress [Mpa]

300

250
T4
200
Modelling Data
150
Tensile Data
100

50

0
0.00 0.05 0.10 0.15 0.20 0.25 0.30
Effective Plastic Strain [-]

Figure 3.5: AA6111 0.81mm true stress versus effective plastic

strain curves showing the yield strength increase due
to paint bake (PB) heat treatment over the as-received
(T4) condition. Each curve represents an average of the
tensile tests performed at 0◦ 45◦ and 90◦ to the rolling
direction. The data points used within the finite element
models are shown as circles, these points correspond to
the rolling direction. The curve from the formed and
paint baked (PB,F) panels is also shown.

53
 450

400 PB
PB,F
350
True Yield Stress [Mpa]

300

250 T4

200
Modelling Data
150
Tensile Data
100

50

0
0.00 0.05 0.10 0.15 0.20 0.25 0.30
Effective Plastic Strain [-]

Figure 3.6: AA6111 0.93mm true stress versus effective plastic

strain curves showing the yield strength increase due
to paint bake (PB) heat treatment over the as-received
(T4) condition. Each curve represents an average of the
tensile tests performed at 0◦ 45◦ and 90◦ to the rolling
direction. The data points used within the finite element
models are shown as circles, these points correspond to
the rolling direction. The curve from the formed and
paint baked (PB,F) panels is also shown.

54
 450

400
PB,F PB
350
True Yield Stress [Mpa]

300

250 T4

200
Modelling Data
150
Tensile Data
100

50

0
0.00 0.05 0.10 0.15 0.20 0.25 0.30
Effective Plastic Strain [-]

Figure 3.7: AA6111 1.0mm true stress versus effective plastic strain
curves showing the yield strength increase due to paint
bake (PB) heat treatment over the as-received (T4) con-
dition. Each curve represents an average of the tensile
tests performed at 0◦ 45◦ and 90◦ to the rolling direc-
tion. The data points used within the finite element
models are shown as circles, these points correspond to
the rolling direction. The curve from the formed and
paint baked (PB,F) panels is also shown.

55
Chapter 4

Experimental Procedures

In order to perform dent resistance experiments, a tooling set of sufficient

size to represent commercial closure panels was chosen. The formed panel,
shown in Figure 4.1, was fabricated using tooling available at the Industrial
Research and Development Institute (IRDI) in Midland, Ontario. The pan-
els were stamped using both AA5754-O and AA6111-T4 sheet. The top or
crown of these panels has a swept surface which affords differences in curva-
ture and local stiffness from one location to the next. The curvature of the
panel at different cross sections was estimated using calculations based on the
original CAD surfaces of the tooling and is shown in Figure 4.2. Using these
curvatures as a guide, both static and dynamic dent tests were performed at
locations that would provide a range in local panel stiffness. The AA5754-O
panels were statically dented at locations A through E, shown in Figure 4.2
and dynamically dented at locations A through C only. The AA6111 panels
were statically and dynamically dented at locations A through C.

4.1 Panel Stamping

The deep drawing press used to stamp the panels was located at the
Industrial Research and Development Institute (IRDI) in Midland, Ontario.
This press, shown in Figure 4.3, has a moving platen and stationary punch.

56
 100mm

570mm
785mm

Figure 4.1: Sketch of the IRDI panel adopted in the current re-
search, showing the draw depth and initial length and
width.

A
D
E
B
C 1600mm
800mm
800mm
500mm
1600mm
800mm

Figure 4.2: Radius of curvature of the IRDI panel at six different

cross sections. These values were estimated by fitting
a circular arc to three points along each of the dot-
ted paths shown. Dent locations A through E are also
shown.

57
 Platen

Punch

Figure 4.3: 1333 kN IRDI Press that was used to stamp the panels
of AA5754-O and AA6111-T4 sheet.

58
The capacity of the press is 1333 kN acting on the main platen, while the
blank holder can provide 667 kN of clamping force. These loads are controlled
by hydraulic pressure; the main platen is actuated by a 457mm diameter (18”
dia.) hydraulic piston, while the binder is actuated by four - 152mm diameter
(6” dia.) hydraulic pistons. Experimental data is recorded using a PC-based
data acquisition system controlled by the LabView software system.
The tooling used to stamp the panels, shown in Figure 4.4, was man-
ufactured as part of an unrelated study on lubrication also sponsored by
ALCAN. The design of this tooling is intended to simulate a scaled down
hood of a typical automobile. Drawbeads are incorporated in the IRDI tool-
ing, as shown in Figures 4.4 and 4.5. These drawbeads maintain tension in
the sheet during stamping, by forcing the sheet to bend over the drawbead
as the sheet flows through the drawbead region. The tension provided by
the drawbead helps resist the onset of sheet wrinkling, caused when com-
pressive in-plane stresses develop in the sheet. The mathematical modelling
of drawbeads will be covered in further detail in section 5.1.5; however it
is important to note that there exists two drawbead regions on the IRDI
tooling, as shown in Figure 4.5. The drawbead geometry varies somewhat
within each region as a hand grinder has been used to achieve their present
shape. In the shallow region, the female drawbead depth is 3 - 4mm, while
in the deeper region it is 5 - 6mm. The male drawbead is part of the binder,
shown in Figure 4.4, the height of the male drawbead ranges from 5 - 6mm
in both regions. The binder is relatively thin at 50mm (2”) and flexed during
the forming operations.
Also shown in Figures 4.4 and 4.5 are several stop-blocks that were bolted
to the binder. The height of the stop-blocks can be varied to prevent the
binder from fully closing on the die and also to control sheet draw-in during
panel stamping. The minimum clearance between the die and binder was
1.4mm, a limitation that exists because of drawbead geometry and the flex
in the binder. If the die is allowed to come closer than 1.4mm, then the sheet
would be pinched between the drawbeads, resulting in tearing of the blank.

59
 Female
Drawbeads

Die
Cavity

Binder / Stamped Panel Male

Blankholder Drawbeads

Punch Stop-Blocks

Figure 4.4: IRDI Die Set. The top half of the figure is rotated to
show the die cavity and surrounding female drawbeads,
while the lower half shows the punch, the blank holder,
the stop-blocks and male drawbeads. The visible stop-
blocks are circled.

60
 aaaa aaaa aaaa
aaaa
aaaa
aaaa aaaa aaaa
aaaa
aaaa
aaaa
aaaa
aaaa
aaaa Stop-Block
aaaa aaaa
aaaa aaaa
aaaa aaaa ~3-4mm deep
drawbead cavity
~5-6mm deep
y drawbead cavity
aaaa aaaa
(drawbead was ~5-6mm tall)
aaaa
aaaa
aaaa
x aaaa
aaaa
aaaa

aaaa aaaa
aaaa
aaaa aaaa
aaaa
aaaa aaaa
aaaa aaaa
aaaa
aaaa aaaa
aaaa
aaaa aaaa

Figure 4.5: IRDI Binder Configuration. This figure shows the lo-
cation of the binder stop-blocks. The blank is shaded
to show placement prior to panel stamping and the two
drawbead regions are shown.

4.1.1 Specimen Preparation

The 1.0mm AA5754 sheet used for this study was sheared from ALCAN
coil number 12918117. Each blank was shipped in a pre-lubricated condition
with an oil based lubricant, Parker MP404. The blanks were sheared to the
dimensions of 785mm x 570mm. The sheet rolling direction was parallel to
the 785mm dimension of each blank. 50.3mm diameter circles were scribed
onto the surfaces of several undeformed blanks. During the forming opera-
tions these circles deform into ellipses that can be measured to provide an
estimate of strain in each panel. Relatively large diameter circles of 50.3mm
were used, as the strain levels in the panel only reached 2-3 percent strain
which was too low for the resolution of more conventional techniques such as
circle-grid analysis [73].
The 0.8mm, 0.93mm and 1.0mm AA6111-T4PD sheet provided by AL-
CAN for this study was sheared to the dimensions of 785mm by 570mm. In

61
order to save material, the sheet rolling direction was chosen to lie parallel to
570mm dimension of each blank, which is perpendicular to the orientation of
the AA5754 sheet. Parker MP404 lubricant was applied to each side of the
AA6111 sheets prior to stamping using a paint roller.

4.1.2 Tooling Setup

The IRDI tooling was set up so that the stop-blocks shown in Figure 4.4
stood 1.4mm proud of the binder. This gap results in an initial overclosure
of 0.6 - 1.6mm in the shallow drawbead cavities, but due to the flex in the
binder still produced the best part. This configuration was kept constant for
all of the panels. For both the AA5754 and AA6111 sheet, the main ram
pressure was 6,067kPa (880psi), which results in a force of 996kN acting on
the die. The average pressure acting on each of the four binder posts was
7,670kPa (1,113psi) for the AA5754 sheet, but for the AA6111 sheet, this
was increased to an average of 12,404kPa (1,800psi). This results in a force
of 249kN acting on the binder for the AA5754 and 402kN for the AA6111.
All of the panels were drawn to a depth of 100mm as shown in Figure 4.1.
During forming, it was found that the four stop-blocks shown in the top of
Figure 4.5 prevented the binder from applying sufficient tension in the 0.8mm
and 0.93mm AA6111 sheet. This caused wrinkling in the stamped panels as
shown in Figure 4.6. As was mentioned earlier, the stop-blocks could not be
removed without causing the drawbeads to tear the sheet during stamping
of the panel. As the wrinkling was outside of the regions subjected to dent
loading, it was decided that the lower thickness AA6111 panels could still be
used for subsequent denting analysis.

4.2 Paint Bake Cycle

All of the AA6111 panels underwent a single heat treatment cycle of 32
minutes at 177◦C prior to dent testing. As was discussed in Chapter 3, this
single heat treatment cycle is representative of an automotive paint bake

62
process. A heat treatment oven, approximately 10 cubic meters in volume
was used for this step. The oven was manufactured by Despatch Limited
and had a closed loop temperature control system. The temperatures of
the oven, the stamped panels and tensile specimens were monitored during
heat treatment, using a PC data acquisition system running the LABTECH
software. Thermo-couples were suspended in the oven and were taped to
several of the panels and tensile specimens during heat treatment to ensure
that differences in temperature were small.

4.3 Static Dent and Panel Stiffness Tests

Static dent tests were performed on both the AA5754 and AA6111 panels.
The AA5754 panels were statically dented at the five locations (A,B,C,D,E)
shown in Figure 4.2, while the AA6111 panels were dented at locations A, B
and C. A schematic of the static dent test apparatus is shown in Figure 4.7,
while a closeup of a panel being tested is shown in Figure 4.8. A 25.4mm (1
inch) steel ball was used as the indentor for both the static dent and stiffness
tests.
It was mentioned in Chapter 1 that the experimental programme was
derived from recommendations of ALCAN personnel [45] and the Standard-
ized Dent Resistance Task Force report [34]. The static dent test procedure
developed for the current work was as follows:

1. The IRDI panel was clamped to the adjustable platform. If necessary,

the panel was trimmed to fit inside of the clamps shown in Figures 4.6
and 4.8.

2. A fiber-glass template was used to mark the location of each dent. This
template was fabricated by moulding fiber-glass over one of the panels,
after which holes were drilled at each dent location.

3. The bubble-type level shown in Figure 4.9 was used to insure that the
panel was level at the dent location.

63
Figure 4.6: Wrinkles that were present in the 0.8mm and 0.93mm
AA6111 sheet after forming.

Digital Loading
Gauge Arm

Indentor
Panel Loading Counter
Weights Weights

Adjustable
Platform

Figure 4.7: Experimental setup for a static dent test. Each panel
is clamped in an adjustable platform that can be po-
sitioned so that the loading direction is normal to the
surface of the panel. The loading arm is leveled and
balanced prior to each test.

64
 Figure 4.8: Closeup of a static dent test showing the clamped panel
under load. The digital gauge measures panel deflection
relative to the base of the platform.

4. The loading arm was positioned so that the indentor would contact the
panel at the dent location.

5. The platform that held the weights was positioned halfway between the
indentor arm and the loading arm pivot. For each weight added, a load
of 1/2 the weight would be applied through the indentor to the panel.

6. The loading arm was levelled and balanced with counter-weights to en-
sure that the indentor was balanced at, or just above, the dent location.

7. A 4.45N (1-lb) weight was loaded onto the weight platform to pre-load
the system.

8. As shown in Figure 4.10, the digital gauge was affixed to the adjustable

65
 Figure 4.9: Bubble-type level used to level each panel at the dent
site.

platform with a magnetic clamp, aligned to act in the vertical direction

at the dent location and was zeroed. As the panel was leveled at the
dent location, this ensured that the normal deflection of the panel would
be measured.

9. A 44.48N (10-lb) weight was loaded onto the weight platform and 10
seconds was allowed to pass to allow any settling of the panel surface.

10. The displacement of the indentor was measured.

11. The 44.48N (10-lb) weight was removed and 10 seconds was allowed to
pass.

12. The displacement of the indentor was measured.

13. The method described in the above four steps was repeated for total
applied loads of 88.96, 133.44, 177.92, 222.40, 266.88, 311.36, 355.84,
400.32, 444.80 and 489.28N (20, 30, 40, 50, 60, 70, 80, 90, 100 and 110-
lbs). As each weight was added or removed from the weight platform,
ten seconds was allowed to pass before recording a displacement.

14. In order to save time during some tests, the ten seconds was only al-
lowed to pass prior to recording the displacement at peak load, and

66
Figure 4.10: Close-up of the digital gauge used to measure normal
displacement of the IRDI panel during static denting.

67
 after complete unloading for each cycle. During these tests, no dis-
placement data was recorded at the intermediate loads for each load
cycle. For example, if the peak load for the cycle was 50-lbs, then the
full 50-lbs would be loaded onto the weight platform, ten seconds would
be allowed to pass and the displacement was then measured. Following
this, the full 50-lbs would be removed, ten seconds would be allowed
to pass and the displacement was then measured.

15. The residual dents for each load was the displacement of the indentor
after unloading to the 1-lb preload weight.

16. Initial panel stiffness was calculated for each panel using the static dent
load-displacement data. Each stiffness value was calculated as 44.5N
divided by the displacement of the indentor at the 44.5N load, which
is in accordance with the Auto/Steel Partnership report [34].

4.4 Dynamic Dent Tests

A schematic of the test apparatus used to perform the dynamic dents
is shown in Figure 4.11. This setup uses kinetic energy gained from a free-
falling 25.4mm diameter indentor to dent the panel. Prior to its release, the
indentor is held by an electromagnet above the panel, the height of which
can be varied. The impact velocities of the indentor ranged from 1.98 to 4.2
meters per second, calculated based on drop heights of 200mm to 1200mm,
and neglecting drag. Each type of panel was dented at locations A, B and C
in Figure 4.2; however, the number performed depended on the repeatability
of each test, as well as the dent depths that were achieved in prior tests. As
the number of tests performed on each type of panel varied, a summary of
all the dynamic dent tests performed is given in Table 4.1.

68
 Drop Tower
(Adjustable Height)

Indentor or
Plumb-Bob Magnetic
Holder Coil

Plumb-Bob

Panel

Adjustable
Platform

Figure 4.11: Experimental setup for a dynamic dent test. Each

panel is clamped to an adjustable platform that is po-
sitioned so that the vertical direction is normal to the
surface of the panel. The electromagnetic coil is used
to hold the indentor, or plumb-bob directly above the
dent site and is de-activated with a remote switch. The
height of the drop tower is adjustable so that sufficient
kinetic energy can be gained during the indentor’s free-
fall.

69
 Table 4.1: A listing of the number of dynamic dent tests performed
on each type of stamped panel.
Panel Type and Location
Drop
5754 6111
Height
1.0mm 0.8mm 0.93mm 1.0mm
[mm]
A B C A B C A B C A B C
200 3 1 1 2
300 1 1 1 1
400 1 1 3
500 1 1 1 3 3 3 3 3 3 3 3 3
600 3 1 1
700 1 1 1 2
800 1 1 2 1 1 1 1 1 1 1
900 1 2 1
1000 2
1100
1200 4 3 4 3 3 3 3 3 3

The procedure for each dynamic dent test was again derived from recom-
mendations of ALCAN personnel [45] and the Standardized Dent Resistance
Task Force report [34] and was as follows:

1. Each panel was trimmed and clamped to the adjustable platform as in

the static dent tests.

2. The dent location was marked and levelled using the adjustable plat-
form and a level.

3. A dent measurement fixture, shown in Figure 4.12(a), was positioned

over the dent location and levelled. This fixture consists of a flat plat-
form and three sharpened pins for legs. The pins enabled small indenta-
tions to be made in the panel surface, which allowed the measurement
fixture to be repositioned in the same location.

70
4. The initial position of the panel surface relative to the measuring plat-
form was taken as shown in Figure 4.12(b). This reading was taken
three times for each test and between each reading, the measurement
fixture was removed and re-positioned. The panel height prior to dent-
ing was taken as the average of these three readings

5. The electro-magnet on the drop tower was aligned above the dent loca-
tion at the drop height using a plumb-bob as shown in Figure 4.13(a).

6. The indentor (a 25.4mm steel ball bearing) was held by the electro-
magnet as shown in Figure 4.13(b).

7. The indentor was allowed to fall onto the dent site and was caught
manually as it rebounded from the panel.

8. Three measurements of the final position of the panel surface relative

to the measuring platform were taken as in Steps 3 and 4 (above). An
average value was calculated.

9. The dynamic dent depth for each test was the magnitude of the differ-
ence between initial and final positions of the panel surface.

71
A B

Figure 4.12: (a) Flat measuring platform that is positioned over the
dent site, note the use of a bubble-type level to insure
level positioning. (b) Measurement of panel location
relative to the platform at the dent site.

72
A B

Figure 4.13: (a) Plumb-Bob that was used to position the drop
tower. (b) Electro-magnetic coil holding the indentor
prior to dynamic denting.

73
Chapter 5

Finite Element Models

A principal goal of this research was to produce software tools and method-
ologies to facilitate use of the finite element method (FEM) for accurate
predictions of panel stiffness and dent resistance. Figure 1.5 portrays the
steps (forming, springback, mesh refinement, static or dynamic denting and
springback) that must be simulated in order to predict panel stiffness and
dent resistance. The numerical models and simulation techniques developed
as part of this research are discussed in this chapter.
The numerical simulations were performed on three different UNIX com-
puting platforms. These were an 18 cpu Silicon Graphics Origin 2000, a 4
cpu Origin 200 and an 8 node Linux cluster. The Silicon Graphics comput-
ers ran the IRIX 6.5 operating system, while the Linux cluster ran RedHat
version 6.2. The Origin 2000 was configured with 9 gigabytes of memory, the
Origin 200 had 2 gigabytes and the cluster had 256 megabytes on each node.
The forming, springback and dynamic denting simulations were solved using
the LS-DYNA version 950-d [46] finite element software. The static denting
simulations were solved using ABAQUS version 5.8.18 [53].
Both LS-DYNA and ABAQUS have implicit and explicit solution capa-
bility. In general, implicit solvers are used to solve a system of equations
having the general form:

74
 F̄i = Kx̄i (5.1)
where F̄i and x̄i are nodal force and displacement vectors, respectively, and
K is the global stiffness matrix. Usually the forces are specified as boundary
conditions, whereas the displacements are unknown. This makes it necessary
to invert the stiffness matrix, (K) which for large problems can be quite time
consuming and require large amounts of memory. This matrix inversion is
subject to numerical round-off errors. Also, certain classes of problems such
as sheet metal springback compound the issue due to the ill-conditioned na-
ture of the stiffness matrix. When material non-linearities are considered, the
initial estimate of stiffness can be significantly different from the actual stiff-
ness. This requires iterative techniques to be used in order to solve Equation
5.1 for static equilibrium.
Conversely, explicit solvers use forward integration to solve the equations
of motion:

F̄i = Kx̄i + Bv̄i + Māi (5.2)

where F̄i , x̄i , v̄i , āi are nodal force, displacement, velocity and accelera-
tion vectors, respectively and K, B, M are the stiffness, damping and mass
matrices. Usually the damping matrix is either neglected or treated in a
simplified fashion in explicit dynamic finite element codes so that the system
of equations reduces to:

F̄i = Kx̄i + Māi

The approach to solving these explicit equations is that at time t=0, the
displacements and forces are known, but the acceleration vector is unknown,
so:


āi = M−1 F̄i − Kx̄i

can be used as an estimate of the nodal accelerations. Inversion of the mass

matrix (M) is trivial since in explicit formulations a diagonal lumped mass

75
matrix is typically used. Acceleration is the second derivative of displace-
ment, so multiplication of āi by the time step twice is an estimate of nodal
displacement for the time step. This process is repeated over many time steps
until the specified problem time duration is reached. When this method is
used the time step must be kept quite small to avoid instabilities. The time
step required for solution and stability is governed by the Courant criterion.
Computationally, the time is spent doing many simple calculations, whereas
in an implicit calculation, the time is spent doing fewer more difficult calcula-
tions (matrix inversion). These differences in solution technique makes each
form of solver more attractive to a certain class of problem. In problems such
as sheet metal forming, many solution steps are required to accurately model
the deformation and intermittent tooling-workpiece contact. This makes the
explicit solution technique attractive. An implicit technique would require
too many matrix inversions to be practicable. Springback problems gener-
ally require less solution steps as well as a requirement of attaining static
equilibrium. Thus, they are normally solved using implicit techniques.

5.1 Forming Models

In recent years, the explicit dynamic finite element method has been
shown to model forming of sheet metal reasonably well with good compu-
tational efficiency [50–52]. These simulations must capture tooling contact,
friction, drawbeads and sheet anisotropy for accurate prediction of panel
strains and final shape. A brief discussion as to how these problems were
modelled in the current work is discussed in the following sections.

5.1.1 Finite Element Mesh

Discretisation of the IRDI tooling was provided by ALCAN in the form
of a finite element mesh. This model was modified for the current work to
simulate both the AA6111 and AA5754 forming operations. An exploded
view of this finite element mesh is shown in Figure 5.1. The tooling surfaces

76
for each forming model were modelled using rigid four-node quadrilateral
shell elements. Since each forming model was to be used in subsequent
denting analysis, the symmetry of the panel was not used to reduce the
problem size.
One rather useful feature of LS-DYNA is automatic or “adaptive” mesh
refinement of the blank during the simulation. This allows the coarsely
meshed blank shown in Figure 5.1 to be refined only as necessary during
the simulation, minimizing the computational effort. Within LS-DYNA, ele-
ment refinement occurs when the curvature of approaching tooling will cause
a user specified angle between adjacent blank elements to be exceeded. The
element subdivision method is similar to that used in the D-Mesh program,
(See Figure 2.7); however, the original curvature of the panel is not ap-
proximated. Any new midedge nodes that are added during refinement are
constrained to lie halfway between the corner nodes of the original element.
Automatic mesh refinement was used for the blank in each forming model.
This increased the problem size significantly during the simulation. The
refined finite element mesh of the AA5754 blank after forming is shown in
Figure 5.2. Two levels of refinement were specified for the 1.0mm AA5754
and AA6111 models, while one was allowed for the 0.93 and 0.81mm AA6111
models. It was found that at two levels of refinement the problem sizes were
becoming large with the increased number of nodes and elements. The larger
problem sizes required significant computer memory and CPU time to solve
the subsequent springback and denting analysis and this mandated a decrease
in problem size. It should be noted that no appreciable difference in forming
strains were visible in trial runs of the same models with different levels of
adaptivity. A summary of the number of elements present in each blank
before and after forming is given in Table 5.1, along with the solution time
while running on one SGI Origin 200 (180 Megahertz) cpu.

77
 Die

Draw-Beads
Blank

Z
Y Binder
X

Punch

Figure 5.1: Exploded view of the finite element mesh used for the
IRDI forming model. The simulated drawbead regions
are shown attached to the binder.

78
 Z
Y
X

Figure 5.2: The finite element mesh of the blank after adaptive
meshing has refined the sidewalls of the panel during
the forming operation. This mesh shows two levels of
adaptive refinement in the sidewalls.

Table 5.1: A listing of the number of elements initially present in

the blank at the start of each forming simulation and
the final number after automatic mesh refinement. Also
shown is the time required to solve the problem on one
SGI Origin 200 cpu running at 180 Megahertz.
No. of Elements in Blank Solution
Forming Model
Initial After Refinement Time
1.00mm AA5754 5785 27622 104.6 hours
1.00mm AA6111 5785 36563 99.6 hours
0.93mm AA6111 5785 13235 36.2 hours
0.81mm AA6111 5785 12061 30.2 hours

79
5.1.2 Element Formulation
Until recently, the Belytschko-Lin-Tsay element formulation [62] has been
a popular choice for sheet metal forming simulations within LS-DYNA. This
element formulation had provided the best balance between computational
efficiency and the accuracy provided by a fully integrated shell element. It
was found that this element formulation caused convergence difficulties in
subsequent springback analysis, because very accurate residual stresses from
the forming simulation are required. For this reason, the blank in each form-
ing model was discretised with a new shell element available within LS-DYNA
that is based on the work of Simo and Armero [74]. Within LS-DYNA this
element is simply known as element number 16. This is a fully integrated
shell element with four (2x2) integration points in-plane and seven through
thickness integration points. Gaussian integration is employed for both the
in-plane and through thickness directions and thickness changes are captured
based on the in-plane strains. Use of a 2x2 integration scheme in the forming
model eliminates the need for hourglass control and properly matches the
integration scheme used in subsequent springback analyses. A higher com-
putational cost (4 times) is incurred compared to a single point (in-plane)
integration scheme as used by the Belytschko-Lin-Tsay element.

5.1.3 Contact
Within LS-DYNA, there are three types of contact algorithms. The one
used for sheet metal forming simulations is known as the penalty method
(see Hallquist [46, p23.2]). In this method, contact occurs if the nodes of one
body penetrate the surface of another. When this condition occurs during
the simulation, normal interface springs are placed between the penetrating
nodes and contact surface. The interface spring stiffness, ki of the penetrated
element i is given by:

fsi Ki Ai
ki = (5.3)
li

80
where fsi is the user defined interface stiffness scaling factor, Ki is the bulk
modulus of the tooling, Ai is the surface area of the penetrated element, and
li is the maximum shell diagonal of element i. For the current work, the
default scale factor of fsi =0.1 was used, this value is dimensionless.
Several contact types within LS-DYNA use the penalty method, some of
which have been formulated specifically for use in sheet metal forming sim-
ulations. These special formulations are required to account for variations
of sheet thickness in the blank during forming. The particular contact al-
gorithm used for the current work was the “Forming One Way Surface to
Surface” algorithm which checks for penetration of the blank through the
tooling, but not of the tooling through the blank. This one way method is
faster than a two way algorithm and works best when the mesh of the blank
is smaller than the tooling mesh.

5.1.4 Friction
The frictional characteristics of sheet metals can vary considerably from
region to region within the tooling during a forming operation. There are
many reasons for this, some of which are the quantity and type of lubri-
cant in the local contact region, rolling direction of the sheet metal relative
to material drawing direction, high contact pressure between the blank and
tooling, temperature gradients within the tooling and degree of sliding along
the tooling. Characterisation of different friction values within a forming
simulation generally is not done; however, a Coulomb friction model is avail-
able within LS-DYNA [46]. This friction model allows for differences in static
and dynamic coefficients of friction and has the form:


F̄f = µd + (µs − µd ) e−c|v| · F̄n (5.4)
where F̄f ,F̄n are the friction and normal contact forces, µs , µd are the static
and dynamic coefficients of friction, c is a decay constant and v is the relative
velocity between the two parts. Examination of this equation reveals that

81
the static coefficient is used until relative motion exists between the blank
and tooling, after which the value will transition to the dynamic dynamic
coefficient based upon the sliding velocity and decay constant. For the cur-
rent work, values of 0.125 and 0.120 were used for the static and dynamic
coefficients of friction, a value of 1.0 s
m was used for c. These values were
selected based on past experience at ALCAN and the University of Waterloo
with aluminum and the oil-based lubricant that was used, Parker MP404.

5.1.5 Drawbeads
The cross section of a typical drawbead is shown in Figure 5.3. Drawbeads
are used in sheet metal stamping processes to produce tension in normally
compressive stress regions of the sheet in order to prevent the formation of
wrinkles.

Normal Force

Restraining Force Tooling

Sheet
Movement
Sheet

Tooling

Figure 5.3: Schematic of a drawbead showing the motion of the

sheet and the forces exerted on the sheet by the tooling.

Drawbeads can be modelled in two ways within a finite element simula-

tion. The first way is to directly incorporate (mesh) the drawbead geometry
within the simulation. This method is generally not done as the amount of
refinement required to accurately model the drawbeads would result in an
exceedingly small time step size. The second method is to use an analytical

82
drawbead treatment that introduces the tension produced by the drawbead
as a restraining force on the nodes of the blank as they pass through the
drawbead region. A normal force is also applied to the nodes as the tooling
closes. This method was used for all of the forming models in this study; the
drawbead regions are shown attached to the die in Figure 5.1. Note that the
tooling and the sheet remain flat, as the geometry of the drawbead is not
modelled. The tension and normal forces can be estimated for the simulation
from prior experiments using a drawbead simulator [75] or from the results
of separate finite element simulations that explicitly model of the drawbead
geometry. For the current work, small finite element models of the drawbeads
were simulated using the method of Maker [76], which predicted the sheet
restraining forces and normal forces as a function of drawbead penetration as
shown in Figures 5.4 and 5.5, respectively. Note how the shallow drawbead
sections discussed in Section 4.1 create larger forces as the sheet is pinched
between the die and binder.

5.1.6 Displacement Boundary Conditions

As was mentioned in Chapter 4, the IRDI press had a moving platen
and stationary punch. The die, shown in Figures 4.4 and 5.1, was bolted to
the platen, while the binder was forced against the moving die by hydraulic
pistons. In order to model these motions during the forming operation, the
following boundary conditions were enforced.
The downwards (-z direction) motion of the die was enforced with a
cosine-hold velocity profile as shown in Figure 5.6. In order to model the
gap created by the stop blocks discussed in Section 4.1.2, the motion of the
binder also had to be enforced with the same velocity profile. The binder and
die were additionally constrained from movement in the x and y directions,
while the punch was fully constrained.

83
 0.0100

1.0mm 5754 sheet, Deep Drawbead Section

0.0090
1.0mm 5754 sheet, Shallow Drawbead Section
0.0080 1.0mm 6111 sheet, Deep Drawbead Section
Restraining Force [kN/mm]

1.0mm 6111 sheet, Shallow Drawbead Section

0.0070 0.8mm 6111 sheet, Deep Drawbead Section
0.8mm 6111 sheet, Deep Drawbead Section
0.0060

0.0050

0.0040

0.0030

0.0020

0.0010

0.0000
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Drawbead Penetration [mm]

Figure 5.4: Estimation of IRDI drawbead restraining forces for

AA5754 and AA6111 sheets. The drawbead force is nor-
malised per unit length (1mm) of drawbead.

0.0300

1.0mm 5754 sheet, Deep Drawbead Section

1.0mm 5754 sheet, Shallow Drawbead Section
0.0250
1.0mm 6111 sheet, Deep Drawbead Section
1.0mm 6111 sheet, Shallow Drawbead Section
Closing Force [kN/mm]

0.8mm 6111 sheet, Deep Drawbead Section

0.0200
0.8mm 6111 sheet, Shallow Drawbead Section

0.0150

0.0100

0.0050

0.0000
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Drawbead Penetration [mm]

Figure 5.5: Estimation of IRDI drawbead normal forces for AA5754

and AA6111 sheets. The drawbead force is normalised
per unit length (1mm) of drawbead.

84
 6 120
Velocity [mm/msec]

Displacement [mm]
5 100

4 80
3 60
2 40
1 20
0 0
0 5 10 15 20 25 30
Time [msec]

Figure 5.6: Prescribed cosine-hold velocity boundary condition for

the die and binder. This type of boundary condition
results in a smooth displacement curve which is also
shown.

5.1.7 Anisotropy
As was discussed in Chapter 3, the aluminum sheet is an anisotropic mate-
rial that has different physical properties in different directions. Anisotropic
materials have been modelled using FEA, as in Lamontagne [51] who looked
at the effect of using different plastic yield functions on the formability pre-
dictions of aluminum sheet. One of these yield functions was created by
Barlat and Lian [77] and is available in LS-DYNA as the 3-Parameter Bar-
lat material model. This material model is considered state of the art for
capturing anisotropy and material thinning in sheet metal during forming
simulations using shell elements, so it was used in this work to predict the
sheet behaviour during the forming.

85
5.1.8 Forming Model Output
The results of the forming model are output in the form of computer
files. One of these is the so-called “dynain” file, which contains the deformed
finite element mesh, as well as the resulting stresses and strains at each
integration point. This file is used as the input for the next stage of the
process, springback.

5.2 Springback Models

The use of the implicit finite element method to predict springback in
sheet metal after a forming simulation has been covered in several prior
studies. Two of these are the Numisheet benchmark on springback of a U-
shaped channel [52] and more recently by Zhou [54] on springback of U-
channel formed tailor-welded blanks. Springback prediction for stamped
parts remains an art due to the ill-conditioned nature of the equations.
Ill-conditioning of any stiffness matrix (K in Equation 5.1) arises when a
large percentage of both high and low eigenvalues are present after inversion.
In sheet metal, large eigenvalues result from the stiff in-plane deformation
modes, while the less stiff out-of-plane deformation modes produce small
eigenvalues. The size of the finite element mesh also contributes to the prob-
lem, as round-off errors accumulate during the matrix inversion. Due to
these problems, the analyst must typically coax the solution to converge by
using numerical tricks such as artificial stabilisation and solution step size
control. A discussion of the springback models and the techniques used to
obtain convergent solutions are discussed next.

5.2.1 Finite Element Mesh

The finite element mesh for each springback model was taken from the
last stage of the forming prediction. These models are written by LS-DYNA
for the purpose of springback prediction and include the sheet thickness,

86
residual stresses and plastic work accumulated by the elements. A typical
mesh after forming was shown previously in Figure 5.2.

5.2.2 Boundary Conditions

During the experiments, each stamped panel was clamped to a flat ad-
justable base as shown in Figures 4.8 and 4.11. To enforce this condition for
the subsequent denting analyses, the flat outer flange on each blank was fully
restrained from movement. This flange is visible in Figure 5.2 and can best
be described as the material that hasn’t drawn into the die cavity during the
forming simulation.

5.2.3 Element Formulation and Material Models

The proper choice of element formulation during a springback analysis is
critical to obtaining a convergent solution. It was mentioned previously that
LS-DYNA shell element number 16 is a fully integrated element, which aids
convergence of springback problems. This element was found to give the best
chance to obtain a convergent springback solution, so it was used for each
springback problem. As in the forming simulations, a seven point Gaussian
through thickness integration rule was used.
The choice of which material model to use for the springback solution
depended on whether the model was to be used for a static or dynamic dent
simulation. ABAQUS does not support the more advanced anisotropic 3-
parameter Barlat material model [77], so for any models that were to be
statically dented an isotropic von-Mises model [53] had to be used. The
models which were to be used for dynamic denting retained the 3-parameter
Barlat material description. The process of paint baking the AA6111 sheet
was also included in the springback models by using the stress-strain curves
taken from the tensile tests performed on paint baked panels as discussed in
Chapter 3.

87
5.2.4 Springback Solution Control Parameters
Typically, each springback simulation was broken down into a target of
four load steps, allowing non-linearities in the springback to be spread out
over the four steps. LS-DYNA’s capability to automatically decrease the
step size in the event of non-convergence was employed for more problematic
springback solutions. Several of the springback problems required many iter-
ations, so the maximum allowable iterations per solution step was increased
to 300 before LS-DYNA would decrease the step size and re-try. The default
quasi-Newton BFGS non-linear equation solver [46] was used.
There are two criteria used by LS-DYNA to determine whether the solu-
tion has converged. These are the conditions placed on displacement:

∆ūi 
< "d (5.5)
ūmax
and energy:
 t 
∆ūi Q̄t 
 
∆ūt0 Q̄0  < "e (5.6)

where ∆ūi are the increments in nodal displacement and ūmax is the max-
imum nodal displacement. Q̄0 is the initial residual force at each node and
Q̄t is the final residual force at each node. The parameters "d and "e are user
specified constants. For the current work they were set to 0.001 and 0.01,
respectively.

5.2.5 Springback Model Output

LS-DYNA writes the results of the springback simulation to a series of
computer files. The so-called “dynain” file is also written and is used as the
input to the next stage, remeshing.

88
5.3 Remeshing
Once a springback solution was obtained, the resulting finite element
mesh had to be refined at the location to be dented. This was necessary to
insure sufficient resolution to capture the very small displacements and sharp
gradients in strain adjacent to the dent site. Initial static denting models on
the AA5754 pans used the linear paving algorithm discussed in Section 2.4.2.
These models had three levels of refinement within a 30mm radius. Any mid-
edge nodes present from the forming simulation were removed as discussed in
Section 2.4.4. Although these models performed well, it was felt that panel
softening would be better captured by a larger radius of refinement. As such,
all denting models were refined using the target area paving algorithm also
discussed in Section 2.4.2. The target area for each element was 3mm2 , while
the refinement radius was 100mm. It was found that these parameters re-
sulted in a final element size of slightly less than 1mm2 beneath the indentor.
Figure 5.7 shows a model refined at location C prior to denting. The other
locations had similar refinements.
As discussed in Section 2.4.3, the curvature in each panel was estimated
from the starting mesh and the newly created nodes were projected to this
curved surface. For this, the default curvature scale factor of 1.0 was used.
The indentor was positioned with its z-axis normal to the surface of each
panel. ABAQUS does not account for shell thickness during implicit contact,
so for models that were to be statically dented, the indentor was positioned
right on the refined mesh, without accounting for shell thickness. This is
necessary to ensure that contact will be properly established between the
indentor and the blank at the start of the static denting simulation. Within
LS-DYNA, shell thickness can be accounted for during contact simulations,
so for dynamic denting models, each indentor was positioned 0.2mm away
from the outer panel surface.

89
 Z
Y
X

Figure 5.7: A refined finite element mesh prior to static denting at

location C.

5.4 Denting
The finite element models output from D-Mesh consists of input files that
require little modification to run. The content of these files is discussed in
the following sections.

90
5.4.1 Panel Stiffness and Static Denting
The panel stiffness and static denting models were solved with ABAQUS
Standard, version 5.8.18. The size of these models became quite large, re-
quiring close to 900 Megabytes of RAM to run. File output from these simu-
lations was considerable as well. It was common for a static dent simulation
to output three Gigabytes of data to disk, which resulted in a more than 40
Gigabytes of data written to disk for the 14 static dent simulations. Fortu-
nately, most of this simulation data was not required during post-processing,
thereby allowing more reasonable disk space usage once the simulation had
been completed.

5.4.2 Element Formulation and Material Properties

The IRDI panel was modelled using ABAQUS shell element type S4.
This is a non-reduced integration element, similar to the element formulation
adopted within LS-DYNA. Both of the elements use four in-plane integration
points; however, the ordering convention of these points is different within
each code. This fact is accounted for when stresses were initialised by the
SIGINI and HARDINI subroutines, to be discussed next. The through thick-
ness integration rule of each element was a seven point Gaussian scheme, the
same used within LS-DYNA.
The indentor was discretised using the C3D8R element formulation. This
is a reduced-integration, eight noded brick element. Although a rigid shell
indentor could have been used, it was easier to use the same mesh in both
the static and dynamic denting simulations. In the dynamic simulations, it
was easiest to let LS-DYNA determine the inertial properties of the indentor,
which required the use of a solid mesh. A limitation of the D-Mesh software
also prevented the use of shell elements, as the indentor had to be modelled
using brick elements.
The yield stress curves after paint baking were used for the AA6111 pan-
els; these properties, along with the AA5754 sheet properties, were discussed

91
in Chapter 3. An isotropic von-Mises hardening rule was used to model the
sheet material yield surface within ABAQUS. The indentor was modelled us-
ing the elastic steel properties of E=207GPA and ν=0.29 with ρ=7850kg/m3 .

5.4.3 SIGINI and HARDINI User Subroutines

It is well known that work hardening effects must be included for any
dent simulation to be accurate. By default, ABAQUS will only read in three
local stresses (σxx , σyy , σxy ) and one effective plastic strain value for each
shell element in the input deck. This means that any differences throughout
the element thickness would be lost if this default method of stress initialisa-
tion were used. ABAQUS also allows specification of user subroutines that
are compiled and linked to ABAQUS at run time. The interface to these
subroutines is described in the ABAQUS user manual [67, 25.2.11, 25.2.17].
If the SIGINI and HARDINI user subroutines are used, the stress and effec-
tive plastic strain values can be initialised at each integration point within
the finite element mesh. For this work, both SIGINI and HARDINI were
used. These were written in FORTRAN 77. These subroutines read the
“stresses.inp” file written by D-Mesh (see Section 2.4.6) into memory and
then assign the plastic strain and local values of stress for each integration
point when called. In this manner, no element history data is lost when
transferring results from LS-DYNA to ABAQUS.

5.4.4 Contact
Surface-to-surface-with-gap contact is used to model the interaction be-
tween the indentor and the panel during denting. This type of contact allows
the surfaces of the indentor and panel to slide and open or close relative to
one another. A friction coefficient of µ=0.15 was specified for the contact.
The region that can contact with the indentor had to be kept small to min-
imise memory requirements. By default, D-Mesh includes any element on
the panel that lies within 0.6 times the refinement radius as part of the con-

92
tact definition. All of the elements in the indentor are included within the
contact definition. Closed contact is initialised by ABAQUS between any
node of the indentor that lies within 0.0005 mm of the panel surface. The
careful placement of the indentor described in Section 5.3 resulted in closed
contact only at the node closest to the refinement point.
The importance of this last point cannot be understated. Without this
initial closed contact between the indentor and panel, the static dent simula-
tion would be unable to converge at the first solution step. Alternatively, if
more than one node were in contact, ABAQUS moves the nodes on the panel
(without causing strain) to lie on the indentor surface. This would deform
the surface of the blank prior to any applied load.

5.4.5 Boundary Conditions

The D-Mesh software writes a series of loading and unloading cycles for
each static dent simulation to the output deck. Each cycle starts with a
preload of one Newton on the indentor. This preload level was found through
trial and error as the minimum amount that could be used without conver-
gence issues becoming a problem. This load is increased to a maximum load
in 22.2 Newton (5lbf) increments and then decreased back to 1 Newton by
22.2 Newton increments. The peak load is increased each cycle by 22.2 New-
tons until a maximum load of 244.7 Newtons (55lbf) is reached. Recall that
this is the same loading used in the experiments discussed in Chapter 4. Sec-
tion 2.4.5 discussed how the indentor is positioned a specified distance above
the panel surface. The application of the load is by a concentrated force to
the node on the indentor that is closest to the node on the panel specified
by the user. This concentrated force acts in a normal direction to the panel
surface. The indentor is also constrained to move only in a direction normal
to the panel surface. The panel is constrained from movement using the same
constraints specified during springback, i.e., the outer flange of the panel is
fully constrained.

93
5.4.6 Static Denting Model Output
ABAQUS can write abundant data to disk during the simulation. Only
one file from each simulation is required to post process the resulting load
versus displacement curves. This file has the extension of ‘.sta’ and is know
as the “status file”. D-Mesh generates input control data specifying that the
z-displacement of a node on the indentor to be written to this file. Note
that the z-axis of the indentor’s reference coordinate system is aligned with
the outward normal of the panel at the dent location. This makes it easy to
post-process the denting results.

5.5 Dynamic Denting and Final Springback

The dynamic denting models were solved with LS-DYNA version 950-
d. On average, each denting model required 1 day to solve on one cpu of an
Origin 2000 computer running at 300MHz. The input file written by D-Mesh
for each dynamic denting simulation requires no modification to run. The
contents of each file is discussed below.

5.5.1 Element Formulation and Material Properties

To retain continuity between the initial forming, springback and dynamic
denting models, the element formulation used for each dynamic denting
model was element number 16. As before, a seven point Gaussian through
thickness integration rule was adopted.
The paint baked yield stress curves were used for the AA6111 panels,
while the AA5754 panels used the AA5754-O yield stress curve. The material
hardening rule for each panel was the 3-Parameter Barlat model. Initial
stresses and effective plastic strains resulting from the forming and springback
stages were also included. As in the static denting simulations, the indentor
was modelled using the elastic steel properties of E=207GPA and ν=0.29
with ρ=7850kg/m3 .

94
5.5.2 Contact
A surface-to-surface-with-gap contact formulation was used to model the
interaction between the indentor and the sheet during denting. In order to
minimise the computing resources required, D-Mesh creates a list of nodes
and shell elements that can interact during the simulation. The contact
definition includes all nodes of the indentor; however, only shell elements
lying within 0.6 times the refinement radius are considered during contact.

5.5.3 Boundary Conditions

Initial velocities of 2 m
s, 3 m
s and 4 m
s were prescribed for the indentor
at each dent location, for each panel. These velocities were oriented to strike
the panel normal to impact surface. This allowed dent depth versus impact
velocity to be simulated as in the experimental results. Each panel was
constrained from movement using the same constraints as in the springback
simulation, i.e., the outer flange of the panel was fully constrained.

5.5.4 Dynamic Denting Springback

Following each dynamic denting simulation, each panel had to undergo
one final springback step. These models were run with the same element
formulation, boundary conditions, material models and solution control pa-
rameters as in Section 5.2.4

5.5.5 Dynamic Denting Model Output

Output from each dynamic denting simulation was in the form of a “dy-
nain” file. The final location of the node under the indentor was compared
to its original location prior to denting. The dent depth for every simulation
was calculated as the displacement of this node.

95
Chapter 6

Experimental and Numerical

Results

This chapter presents a comparison of the panel forming, springback,

and dent predictions with the experimental results. Comparisons of form-
ing strains, blank draw-in and sheet thinning are presented after forming
and springback. Load displacement history, panel stiffness and residual dent
depth are compared for the static dent tests. Lastly, predictions of dynamic
dent depth are compared with measured values for the range of indentor
impact velocity considered.

6.1 Panel Forming and Springback

6.1.1 Blank Draw-In and Final Shape
Blank draw-in can be described as the amount of material that is drawn
into the die during the forming operation. Typically this is measured as the
displacement of the outer edge of the blank during the forming operation. For
the current work, blank draw-in was measured for the AA5754 and AA6111
panels and compared to values taken from the forming models. The difference
in draw-in between individual gauges of AA6111 was insignificant, so only

96
a comparison of the values for 1.0mm AA6111 is given here. Figure 6.1
shows the dimensions that were measured on both the forming models and
stamped parts. A comparison of the predicted and measured values of these
dimensions is given in Tables 6.1 and 6.2.

A
E

D H F B

G
C
Figure 6.1: Dimensions used to compare the predicted and mea-
sured values of draw-in on the 1.0mm AA5754 and
1.0mm AA6111 panels after forming and springback.
The initial shape of the blank is shown as a dotted line.

The values given in Tables 6.1 and 6.2 show that although there are some
differences between the predicted and measured values of draw-in, the final
shape of the panel has been predicted well. Small differences are expected to
occur as the drawbead geometry has not been discretised in the finite element
models. In the current work, it is important for the forming models to capture
the draw-in, but it is much more important for the forming models to capture

97
Table 6.1: Predicted and measured dimensions of the 1.05mm
AA5754 IRDI panel after forming and springback. All
dimension values are in mm.
Error w.r.t. Measured
Dimension Predicted Measured
Absolute Percentage
A 747.0 750.0 -3.0 -0.48%
B 550.5 550.0 0.5 0.01%
C 760.5 760.0 0.5 0.07%
D 550.9 550.0 0.9 0.16%
E 33.5 33.0 0.5 1.5%
F 28.4 26.0 2.4 9.2%
G 45.8 45.0 0.8 1.8%
H 29.6 26.0 3.6 13.9%

Table 6.2: Predicted and measured dimensions of the 1.0mm

AA6111 IRDI panel after forming and springback. All
dimension values are in mm.
Error w.r.t. Measured
Dimension Predicted Measured
Absolute Percentage
A 743.4 747.5 -4.1 -0.55%
B 545.6 547.0 -1.4 -0.26%
C 753.1 752.5 0.6 0.08%
D 545.4 548.0 -2.6 0.47%
E 37.5 30.0 7.5 25.0%
F 32.0 30.5 1.5 4.9%
G 51.3 46.8 4.5 9.6%
H 33.3 30.5 2.8 9.2%

98
the thickness changes and strains present in the crown of the deformed panel.
These values are compared in the following sections.

6.1.2 Panel Thickness

Tensile specimens were cut from the crown (the center) of each type of
panel after stamping. Data from these specimens provided insight into the
work-harding of each panel (Section 6.1.3, below) and also allowed measured
panel thickness to be compared with the forming model predictions. Com-
parisons of the average predicted and measured thicknesses at the crown of
each panel type are presented in Table 6.3. The AA5754 finite element form-
ing models were run before the material was received, so nominal gauges had
to be used in those finite element models. This error created the largest dis-
crepancy, as the initial thickness of the AA5754 sheet was actually 1.050mm,
(+5%) but a value of 1.000mm was used in the finite element model. The
final measured thickness of the AA5754 panel was 1.030mm, while the pre-
dicted thickness was 0.975mm, the majority of the error (90%) due to the
error initial thickness. Unfortunately, this model was never updated to re-
flect the actual thickness of the as-received sheet. It is doubtful whether this
affected the subsequent static denting results significantly, as will be shown
in Section 6.2; however, this may have affected the dynamic dent results as
shown in Section 6.3.

Table 6.3: Predicted and measured panel thicknesses at the center

of the IRDI panel after forming. All dimension values
are in mm.
Initial Final Thickness Error w.r.t. Measured
Panels
Thickness Predicted Measured Absolute Percentage
AA5754 1.050 0.975 1.030 -0.055 -5.3%
1.002 0.979 0.971 0.008 0.8%
AA6111 0.931 0.909 0.904 0.005 0.6%
0.809 0.785 0.785 0.000 0.0%

99
6.1.3 Forming Strains
Measurements of engineering strain in the crown of the AA5754 panels
revealed strains of approximately 2% in the x-direction and 1% in the y-
direction. These values were calculated from measured elongations of 1.0mm
and 0.5mm; respectively, of 50.3mm circles scribed in the panel surfaces. The
accuracy of these values was estimated to be within 0.3mm as a steel rule was
used for measurement. The predictions of strain obtained from the AA5754
forming model are shown in Figure 6.2 for the x-direction and Figure 6.3 for
the y-direction.
Note that in Figures 6.2 and 6.3 the range of contour values was set to
-3.0% to 3.0%, even though the panels underwent strains as high as 25% in
the sidewalls. This setting was used to show the variation of strain in the
panel crown, which is the region of interest for denting. Both the measured
and predicted strains are comparable, the predicted strains in the x-direction
ranged from 1.5% to 2.7%, while strains in the y-direction ranged from 0.3%
to 1.8%.
The predicted forming strains for the 1.0mm AA6111 panel are shown
in Figure 6.4 for the x-direction and Figure 6.5 for the y-direction. The
predicted strains in the 6111 panel had a similar pattern to those in the
AA5754 panel, but were slightly less in magnitude. This is expected since
AA6111 has a higher yield strength and is able to pull more material into the
die prior to plastic yielding. The other gauges of AA6111 had similar strain
distributions and are not shown here.
Instead of comparing strain values measured from the surface of each
AA6111 panel, it is useful to compare the effective plastic strain values pre-
dicted by the forming simulation to those estimated from tensile tests per-
formed on the IRDI panels after forming. This comparison is important
because the adopted constitutive formulations utilise a strain hardening hy-
pothesis in which strength is a function of effective plastic strain. Thus, the
strength of the tensile specimens taken from the formed and paint baked
panels can be used to determine the effective plastic strain for comparison

100
Figure 6.2: Predicted X-direction strains in the 1.000mm thick
AA5754 IRDI panel after forming.

Figure 6.3: Predicted Y-direction strains in the 1.000mm thick

AA5754 IRDI panel after forming.

101
Figure 6.4: Predicted X-direction strains in the 1.0mm AA6111
IRDI panel after forming.

Figure 6.5: Predicted Y-direction strains in the 1.0mm AA6111

IRDI panel after forming.

102
with model predictions. In the current work, no tensile tests were performed
on the formed AA6111 panels prior to paint baking, so the tensile tests per-
formed on the panels after paint baking will be used. Note that it is generally
considered that the yield strength increase achieved during a paint bake pro-
cess is directly additive to the strength increase during work hardening [11].
Thus the yield strength curves shown in Figures 3.5, 3.6 and 3.7 can be used
to determine the effective plastic strain in the panels after forming.
Comparison of initial yield strengths for each of the formed and paint
baked tensile specimens with the yield strength curve for the paint baked
specimen shows a strength increase due to approximately 2.0 to 2.5% effective
plastic strain. This value is consistent for each of the AA6111 samples. Figure
6.6 shows the effective plastic strain predicted in the 1.0mm AA6111 panel
after forming. At the center of the panel, the effective plastic strain ranges
from 2.1% to 2.5%, which compares well with the tensile results. The values
of effective plastic strain for the 0.81 and 0.93mm gauges of AA6111 were
also in this range. An interesting feature of the AA6111 forming models
was that the wrinkling of the sheet evident in Figure 6.6 occurred in the
same region as in the experiments (See Figure 4.6). From these results it can
be concluded that the forming models are capturing the strain distributions
present in the crown of each of the IRDI panels.

6.2 Stiffness and Static Dent Tests

In this section, a series of plots of load versus displacement for each of the
static dent tests is presented. The results for the AA5754 panels are presented
first, followed by those for the AA6111 panels. For comparison purposes,
each of the plots contain both the predicted and measured displacement
data. Following the load versus displacement plots, tabulated values of panel
stiffness are presented. For dent tests that exhibited more than one stiffness
region (See Figure 1.1), data for the secondary and final stiffness regions are
also presented. Plots of residual dent depth as a function of applied load

103
 Figure 6.6: Predicted effective plastic strains in the 1.0mm AA6111
IRDI panel after forming. Note the pattern produced
by wrinkling on the top of the panel.

are presented last, again with the AA5754 test data presented before the
AA6111 test data. For comparison purposes, each type of plot is displayed
with identical ranges on the axes.

6.2.1 AA5754 Panel Load-Displacement Responses

The indentor load versus displacement response of various sites on the
AA5754 panels are shown in Figures 6.7 through 6.11. These graphs show
remarkably good agreement between the numerical predictions of panel stiff-
ness and their associated experimental results. Figure 6.8 shows how the
trends of initial, secondary and final panel stiffnesses are captured well using
the finite element method, while Figure 6.11 highlights the good agreement
for the first two stiffness regions particularly well.
The tests performed on panel locations A,C,D and E (shown in Figures
6.7,6.9,6.10 and 6.11, respectively) showed only the first two panel stiffness

104
regions, while the test performed on location B (Figure 6.8) showed all three.
Recall that location B was chosen to lie at the point on the panel with the
least curvature, which results in lower initial panel stiffness. During loading,
this low initial stiffness was observed to allow a larger deformation zone in
the panel relative to the stiffer locations. This deformation zone spread out
with increased load, an action which causes bending stresses in the panel
to first increase due to the extra moment created, but then decrease due to
the larger cross sectional area in bending. As the deformation zone spreads
out, the load applied by the indentor is carried initially by bending stresses,
followed by membrane tension.
The difficulties in obtaining convergent solutions for these rather poorly
conditioned models are highlighted in Figures 6.10. Although this panel is
relatively stiff, during the final unloading step (from a load of 22.2 N to 1.0
N) the solver (ABAQUS) had to reduce the step size by 1/5th to obtain
convergence. This is evidenced by the tighter clustering of data points on
the final unloading step.
300

Predicted
250 Measured

200
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.7: Predicted and measured load displacement curves for

the 1.05mm AA5754 panel at location A.

105
 300
Final Stiffness
Predicted
250 Measured

Secondary Stiffness
200

Initial Stiffness
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.8: Predicted and measured load displacement curves for

the 1.05mm AA5754 panel at location B.

300

Predicted
250 Measured

200
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.9: Predicted and measured load displacement curves for

the 1.05mm AA5754 panel at location C.

106
 300

Predicted
250 Measured

200
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.10: Predicted and measured load displacement curves for

the 1.05mm AA5754 panel at location D.

300

Predicted
250 Measured

200
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.11: Predicted and measured load displacement curves for

the 1.05mm AA5754 panel at location E.

107
6.2.2 AA6111 Panel Load-Displacement Responses
The indentor load versus displacement responses of the AA6111 panels are
shown in Figures 6.12 through 6.20. The panels that were dented at locations
A and B were tested without recording the data for the intermediate load
and unload points, (as discussed in Section 4.3), although the finite element
models included these points. During the tests, this made it possible to place
or remove all of the weights from the loading arm at once, without waiting 10
seconds after the placement or removal of each weight. This was necessary
to save time during the experiments, as each recorded load or unload point
takes approximately 45 seconds to record. A test with complete loading
and unloading history was estimated to take 2 to 3 hours to complete, so
a significant time savings was achieved. It was determined that this did
not affect the test results, as some test configurations were tested in both
manners to confirm repeatability.
A comparison of the predicted and measured static dent test results for
the AA6111 panels shows that the trends of initial, secondary and final panel
stiffness were captured by the models. The predictions for the thinner gauge
sheet (0.81mm) were particularly close to the measured values, while the
thicker gauges (0.93 and 1.0mm AA6111) were not as close. Upon detailed
inspection, it appears that, for each model, the oil-canning load at the tran-
sition between initial and secondary stiffness regions has not been predicted
well.
Close scrutiny of Figures 6.12 through 6.20 shows that the initial stiffness
of each test has been predicted quite well. This suggests that the panel
sheet thickness, curvature and the modulus of elasticity were not in error,
as these properties are known to affect the initial stiffness of a panel as
stated by van Veldhuizen et al. [1]. Once oil-canning in the panel occurs, the
deformation becomes a complex function of panel stiffness, curvature and
material yield strength. At this point it is interesting to note that for each of
the load displacement curves shown in Figures 6.12 through 6.20, the slope
of each secondary stiffness region has also been captured. This suggests that

108
accurately predicting the onset of the transition between initial and secondary
stiffness regions is paramount to obtaining good static dent results.
At this stage in the research, it was decided not to pursue more accu-
rate prediction of the onset of oil-canning. Numerical predictions of such
instabilities are notoriously sensitive to meshing and other numerical param-
eters. The approach taken was to use the “best available” material data and
forming modelling results and avoid “numerical tuning” of the predictions.
Improvements to the predictions of oil-canning are left for future studies.
300

Predicted
250 Measured

200
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.12: Predicted and measured load displacement curves for

the 0.81mm AA6111 panel at location A.

109
 300

Predicted
250 Measured

200
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.13: Predicted and measured load displacement curves for

the 0.81mm AA6111 panel at location B.

300

Predicted
250 Measured

200
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.14: Predicted and measured load displacement curves for

the 0.81mm AA6111 panel at location C.

110
 300

Predicted
250 Measured

200
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.15: Predicted and measured load displacement curves for

the 0.93mm AA6111 panel at location A.

300

Predicted
250 Measured

200
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.16: Predicted and measured load displacement curves for

the 0.93mm AA6111 panel at location B.

111
 300

Predicted
250 Measured

200
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.17: Predicted and measured load displacement curves for

the 0.93mm AA6111 panel at location C.

300

Predicted
250 Measured

200
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.18: Predicted and measured load displacement curves for

the 1.00mm AA6111 panel at location A.

112
 300

Predicted
250 Measured

200
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.19: Predicted and measured load displacement curves for

the 1.00mm AA6111 panel at location B.

300

Predicted
250 Measured

200
Load [N]

150

100

50

0
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Displacement [mm]

Figure 6.20: Predicted and measured load displacement curves for

the 1.00mm AA6111 panel at location C.

113
6.2.3 Panel Stiffness
Tabulated values of measured and predicted initial panel stiffness are
given in Table 6.4. The stiffness values in this table are calculated from the
load-displacement plots discussed in the Sections 6.2.1 and 6.2.2. In all cases,
the initial stiffness is over predicted by the finite element models. The mean
over-prediction of stiffness was 26N/mm, which corresponds to an average
error of 14%. One expects the panels to be numerically stiff when compared
to the experiments because during the experiments, as each weight is loaded
onto the loading arm, a small impulsive load is applied through the loading
arm to the indentor when the weight is released. This impulsive load is in
addition to the weight being applied and causes the panel to settle an extra
amount. Careful handling of the weights minimised this effect; however, a
displacement of 0.03mm is sufficient to cause this error. Considering the
size of the panel, it is plausible that the placement of the weights could
have caused this small error. In addition, the finite element simulation uses
numerous nodal constraints to model clamping of the panel to the loading
platform. The toggle clamps that were used during the experiments are
marginally less rigid than the nodal constraints and so allow the panel to
settle slightly during loading.
For the test sites that exhibited a clear secondary stiffness region, com-
parisons of the predicted and measured secondary stiffness values are given
in Table 6.5. The 1.0mm panel was quite strong and a well-defined secondary
stiffness region was not noted for locations A and C, the stiffest locations on
the panel. Overall, the secondary stiffnesses were well predicted by the finite
element models, the average error on secondary stiffness was only 2.1N/mm,
or 5%. Only three of the test sites exhibited a final panel stiffness region.
Comparisons of the predicted and measured final stiffness values are given
in Table 6.6. The final stiffness values were also well predicted, the average
error was 2.5N/mm, or 6%.

114
 Table 6.4: Predicted and measured values of initial panel stiffness
at the different locations tested on the IRDI panels.
Initial Stiffness Error w.r.t.
Panel Loc. [N/mm] Measured Values
Predicted Measured [N/mm] [%]
A 292 259 33 13
B 182 148 34 23
1.05mm
C 276 253 23 9
AA5754
D 264 232 32 14
E 216 193 23 12
A 182 157 24 16
0.81mm
B 113 98 15 16
AA6111
C 172 156 16 10
A 253 220 33 15
0.93mm
B 170 143 26 18
AA6111
C 243 233 10 4
A 290 249 42 17
1.00mm
B 198 168 30 18
AA6111
C 272 254 18 7

6.2.4 Residual Dent Depths for the AA5754 panels.

Using the static dent test data, graphs of residual dent depth (after un-
loading) versus peak applied load can be obtained. Figures 6.21 through
6.25 show a comparison of the predicted and measured residual dent depths
for the AA5754 panels at locations A through E. Again, there is remarkably
good agreement between the predicted and measured values. Considering
that each panel measures approximately 700mm by 500mm, the agreement
of the predicted dent depths to within 0.03mm of the measured values can
be considered quite good.
Recall that a commonly accepted measurement of static dent resistance
is the ability of a panel to support a load prior to a dent of 0.1mm occurring.
With this definition, it can be said that location C on the panel had the
highest dent resistance. This location was able to support a load of 163N

115
Table 6.5: Predicted and measured values of secondary panel stiff-
ness at the different locations tested on the IRDI panels.
Secondary Stiffness Error w.r.t.
Panel Loc. [N/mm] Measured Values
Predicted Measured [N/mm] [%]
A 49.0 44.8 4.1 9
B 30.7 33.7 -3.0 -9
1.05mm
C 51.1 51.3 -0.2 0
AA5754
D 40.0 42.9 -2.9 -7
E 39.3 40.7 -1.4 -3
A 26.2 26.3 -0.1 -1
0.81mm
B 25.7 25.0 0.7 3
AA6111
C 35.3 35.7 -0.4 -1
A 36.8 33.8 3.0 9
0.93mm
B 29.1 29.0 0.1 0
AA6111
C 54.2 46.3 7.9 17
1.00mm
B 34.8 33.9 0.9 3
AA6111

Table 6.6: Predicted and measured values of final panel stiffness at

location B on the IRDI panels.
Final Stiffness Error w.r.t.
Panel Loc. [N/mm] Measured Values
Predicted Measured [N/mm] [%]
1.05mm
B 40.9 44.3 3.4 8
AA5754
0.81mm
B 38.4 38.5 0.2 -
AA6111
0.93mm
B 43.0 39.0 4.0 10
AA6111

116
prior to a dent depth of 0.1mm occurring. Location B had the lowest dent
resistance, with a permanent dent occurring at 139N. It is interesting to note
that the locations that had the highest static dent resistance (Locations A
and C) were also the stiffest locations on the panel. This trend is in agreement
with results from Yutori et al. [6] and van Veldhuizen et al. [1].

6.2.5 Residual Dent Depths for the AA6111 Panels.

Comparisons of the predicted and measured residual dent depths versus
peak applied load for the AA6111 panels are shown in Figures 6.26 through
6.34. The finite element models consistently underpredict the dent depth in
each case. This is undoubtably due to over prediction of the oil-canning load
for each of the panels. The magnitude of error is at most 0.15mm, which is
quite small compared to the size of the panels (700mm by 500mm).
Figure 6.35 shows the residual dent depths obtained from the experiments
performed on the 0.81, 0.93 and 1.00mm AA6111 panels at location A. This
figure shows how an automotive panel can be tailored to pass a required per-
formance specification. In all cases, increasing the panel thickness resulted
in better static dent performance of the panel; however, if the panel was re-
quired to support a 150N load prior to a permanent dent of 0.1mm occurring,
then the 0.81mm gauge sheet would be too thin.

117
 1.00

0.90
Measured
Modelled
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.21: Predicted and measured residual dent depths for the
1.05mm AA5754 panel at location A.

1.00

0.90
Measured
Modelled
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.22: Predicted and measured residual dent depths for the
1.05mm AA5754 panel at location B.

118
 1.00

0.90
Measured
Modelled
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.23: Predicted and measured residual dent depths for the
1.05mm AA5754 panel at location C.

1.00

0.90
Measured
Modelled
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.24: Predicted and measured residual dent depths for the
1.05mm AA5754 panel at location D.

119
 1.00

0.90
Measured
Modelled
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.25: Predicted and measured residual dent depths for the
1.05mm AA5754 panel at location E.

1.00

0.90
Measured
Predicted
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.26: Predicted and measured residual dent depths for the
0.81mm AA6111 panel at location A.

120
 1.00

0.90
Measured
Predicted
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.27: Predicted and measured residual dent depths for the
0.81mm AA6111 panel at location B.

1.00

0.90
Measured
Predicted
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.28: Predicted and measured residual dent depths for the
0.81mm AA6111 panel at location C.

121
 1.00

0.90
Measured
Predicted
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.29: Predicted and measured residual dent depths for the
0.93mm AA6111 panel at location A.

1.00

0.90
Measured
Predicted
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.30: Predicted and measured residual dent depths for the
0.93mm AA6111 panel at location B.

122
 1.00

0.90
Measured
Predicted
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.31: Predicted and measured residual dent depths for the
0.93mm AA6111 panel at location C.

1.00

0.90
Measured
Predicted
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.32: Predicted and measured residual dent depths for the
1.00mm AA6111 panel at location A.

123
 1.00

0.90
Measured
Predicted
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.33: Predicted and measured residual dent depths for the
1.00mm AA6111 panel at location B.

1.00

0.90
Measured
Predicted
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.34: Predicted and measured residual dent depths for the
1.00mm AA6111 panel at location C.

124
 1.00
Measured - 0.81mm
0.90 Measured - 0.93mm
Measured - 1.00mm
0.80

0.70
Dent Depth [mm]

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0 50 100 150 200 250
Load [N]

Figure 6.35: Measured residual dent depths for the 0.81, 0.93 and
1.00mm AA6111 panels at location A.

125
6.3 Dynamic Dent Tests
6.3.1 Dynamic Dent Response of the AA5754 panels.
The predicted and measured dynamic dent test results for the 1.05mm
AA5754 panels are shown in Figure 6.36. The trend of dent depth increasing
linearly with impact velocity is well captured with each of the simulations.
This is a known trend that has been recorded by several previous studies
[12–17]. The relative difference between dents performed at different panel
locations has also been captured. Dent depths were lowest at location B,
as this was the least stiff panel location. Locations A and C had quite
similar initial stiffness regions as shown in Figures 6.7 and 6.9; however, the
dents produced at location C were visibly smaller. This is a result of the
slightly higher plastic straining during forming and resultant higher strength
at location C than A.
0.90
Measured, Pt. A
0.80 Measured, Pt. B
Measured, Pt. C
0.70 Predicted, Pt. A
Predicted, Pt. B
0.60 Predicted, Pt. C
Dent Depth [mm]

0.50

0.40

0.30

0.20

0.10

0.00
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Impact Velocity [mm/sec]

Figure 6.36: Predicted and measured dynamic dent depths for the
1.05mm AA5754 panel at locations A,B and C.

In general, the measured dynamic dent depths were over-predicted by

about 0.1mm. This discrepancy is attributed to the 0.050mm difference

126
in sheet thickness between the value of 1.000mm assumed in the forming
simulation and the actual value of 1.050mm. Although this seems like a
small difference, prior studies [10] have stated that the dynamic dent depth
produced in a panel is proportional to thickness squared. A scaling law can
be developed combining equations 1.8 and 1.9, that can be used to compare
dynamic dents produced on panels of equal curvature, stiffness and yield
strength but differing thicknesses. This scaling law is:

D1 t21
D2 = (6.1)
t22
Here D1 is the dent depth produced on a panel of thickness t1 and D2 is
the dent produced on a panel of thickness t2 . Consider, for example, the
measured dent depth of 0.707mm obtained from a 4 m/s dynamic dent on
the 1.05mm AA5754 panel at location A (see Figure 6.36). The final thickness
of the 1.050mm AA5754 panel after forming was actually 1.030mm; however,
the denting simulation used 0.975mm from the forming model. Substituting
t1 = 1.030mm, t2 = 0.975mm and D1 = 0.865mm gives a corrected dent
depth (D2 ) of 0.775mm. This corrected value is considerably closer to the
measured dent depth (0.707mm) than the original prediction of 0.865mm.
Thus, while it was decided not to rerun the calculations for the 1.05mm
AA5754 panels, these results serve to emphasise the importance of accounting
for panel thickness changes during forming and to use actual panel thicknesses
rather than nominal values.

6.3.2 Dynamic Dent Response of the AA6111 panels.

The predicted and measured dynamic dent test results for the 0.81, 0.93
and 1.00mm AA6111 panels are shown in Figures 6.37, 6.38 and 6.39, re-
spectively. The dynamic dent predictions for the AA6111 panels are much
closer to experiment than those predicted for AA5754. This is attributed
in part to better prediction of panel thickness and strain during the form-
ing simulations. Overall, the AA6111 panels exhibited much better dynamic

127
dent resistance than the AA5754 panels, due to the higher strength of the
AA6111.
Figure 6.37 for the 0.81mm AA6111 panel shows an interesting change in
slope for the predicted dent depths at location B. This change occurred be-
cause the 2 m/s denting simulation performed at location B did not converge
during the final springback solution for the tolerances imposed for the other
simulations. The convergence criterion used by LS-DYNA (given in equation
5.5) does not allow convergence in springback problems with very small dis-
placements. Convergence will never occur when the Euclidean norm of the
displacements (∆ūi ) is less than the maximum displacement (ūmax ) multi-
plied by a convergence tolerance ("d ). This situation occurred at location B,
so the tolerance on displacement ("d ) had to be increased from 0.001 to 0.005,
which resulted in a less precise solution. Note how this combination of panel
location (B) and thickness (0.81mm) was the least stiff of all the tested con-
figurations (see Tables 6.4 and 6.5). This combination produced the lowest
dynamic dent depths, but relatively large static dent depths, which highlights
the competitive role that stiffness plays in static and dynamic denting.
Figures 6.40, 6.41 and 6.42 show the effect of increasing sheet thickness
at each of the dent locations. General trends are observable for the stiff
panel locations (Locations A and C), but not at the less stiff location (B).
For the stiffer panels, at lower impact velocities, increased thickness aids
dynamic dent resistance, but at higher impact velocities, increased sheet
thickness hinders dynamic dent resistance. This trend is in agreement with
results of Nomura et al. [18] who concluded that increased sheet thickness
did not necessarily improve dynamic dent resistance. Thus when designing
automotive panels for dynamic dent resistance, increasing sheet thickness
may not have the desired effect, depending on the indentor velocity and
sheet curvature. It also may have minimal effect if the panel is not very stiff
as shown by Figure 6.41.

128
 0.90
Measured, Pt. A
0.80 Measured, Pt. B
Measured, Pt. C
0.70 Predicted, Pt. A
Predicted, Pt. B
0.60 Predicted, Pt. C
Dent Depth [mm]

0.50

0.40

0.30

0.20

0.10

0.00
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Impact Velocity [mm/sec]

Figure 6.37: Predicted and measured dynamic dent depths for the
0.81mm AA6111 panel at locations A,B and C.

0.90
Measured, Pt. A
0.80 Measured, Pt. B
Measured, Pt. C
0.70 Predicted, Pt. A
Predicted, Pt. B
0.60 Predicted, Pt. C
Dent Depth [mm]

0.50

0.40

0.30

0.20

0.10

0.00
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Impact Velocity [mm/sec]

Figure 6.38: Predicted and measured dynamic dent depths for the
0.93mm AA6111 panel at locations A,B and C.

129
 0.90
Measured, Pt. A
0.80 Measured, Pt. B
Measured, Pt. C
0.70 Predicted, Pt. A
Predicted, Pt. B
0.60 Predicted, Pt. C
Dent Depth [mm]

0.50

0.40

0.30

0.20

0.10

0.00
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Impact Velocity [mm/sec]

Figure 6.39: Predicted and measured dynamic dent depths for the
1.00mm AA6111 panel at locations A,B and C.

0.90
Measured, 0.81mm
0.80
Measured, 0.93mm
0.70
Measured, 1.00mm
0.60
Dent Depth [mm]

0.50

0.40

0.30

0.20

0.10

0.00
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Impact Velocity [mm/sec]

Figure 6.40: Measured dynamic dent depths for the 0.81, 0.93 and
1.00mm AA6111 panel at location A.

130
 0.90
Measured, 0.81mm
0.80
Measured, 0.93mm
0.70
Measured, 1.00mm
0.60
Dent Depth [mm]

0.50

0.40

0.30

0.20

0.10

0.00
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Impact Velocity [mm/sec]

Figure 6.41: Measured dynamic dent depths for the 0.81, 0.93 and
1.00mm AA6111 panel at location B.

0.90
Measured, 0.81mm
0.80
Measured, 0.93mm
0.70
Measured, 1.00mm
0.60
Dent Depth [mm]

0.50

0.40

0.30

0.20

0.10

0.00
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Impact Velocity [mm/sec]

Figure 6.42: Measured dynamic dent depths for the 0.81, 0.93 and
1.00mm AA6111 panel at location C.

131
6.3.3 Sensitivity of the Dynamic Denting Results to
Finite Element Model Parameters.
A series of finite element models were run to gain an understanding of
how sensitive the results were to modelling parameters. It has already been
shown that material thickness plays a very important role in obtaining good
dynamic dent predictions. Many other factors such as element formulation,
material yield function, mesh size and refinement radius, indentor position
and residual stresses can also have an effect and were considered in this
sensitivity study. Figure 6.43 shows the effect that each of these parameters
had on the predicted dent depth for a 1.05mm AA5754 panel dented at
location A for an initial velocity of 4 m/s. For comparison purposes, the
measured dent depth for this case was 0.707mm.

6.3.3.1 Sensitivity to Element Formulation.

The choice of element formulation used in the finite element model had
the largest effect on predicted dent depth. Use of the reduced integration
Belytschko - Tsay shell element [61] resulted in a dent depth of 1.110mm,
which represents an over prediction by 0.403mm, as shown by the result for
finite element model number 1 in Figure 6.43. This element formulation is
widely used in sheet metal forming simulations due to its numeric efficiency;
however, it was not sufficiently accurate for the current work. This led to
a trial of the non-reduced integration element formulated by Hughes and
Liu [63]. This element formulation also overpredicted dent depth (0.919mm
for model number 2 in Figure 6.43). The number 16 element formulation
available within LS-DYNA [68] was recommended for use during this study
by Maker [78]. This element produced dent predictions consistently closer to
the measured value than the other shell element choices, as shown by models
labelled 3 to 8 in Figure 6.43.

132
 1.20
1.110
Measured Value = 0.707
1.00
0.919
0.865 0.850 0.835
0.797 0.795
0.80 0.755
Dent Depth [mm]

0.707

0.60

0.40

0.20

0.00
1 2 3 4 5 6 7 8
Finite Element Model Number
Model Element Hardening Mesh Size and Position of Residual
Number Type Rule Refinement Radius Indentor Stresses
1 Belytschko - Tsay 3-Paramater Barlat 1mm, 30 mm 0.2 mm above Yes
2 Hughes - Liu 3-Paramater Barlat 1mm, 30 mm 0.2 mm above Yes
3 Element 16 3-Paramater Barlat 1mm, 100 mm 0.2 mm above Yes
4 Element 16 von - Mises 1mm, 100 mm 0.2 mm above Yes
5 Element 16 von - Mises 1mm, 30 mm 0.2 mm above Yes
6 Element 16 3-Paramater Barlat 1mm, 30 mm 0.2 mm above Yes
7 Element 16 3-Paramater Barlat 1mm, 30 mm 0.7 mm above Yes
8 Element 16 3-Paramater Barlat 1mm, 30 mm 0.2 mm above No

Figure 6.43: This figure shows the sensitivity of the dynamic denting
models to element formulation, material yield function,
mesh size, location of the indentor above the panel and
use of residual stresses. These results were obtained
from 4 m/s dents simulated at point A on the 1.05mm
AA5754 panel.

133
6.3.3.2 Sensitivity to Material Yield Function and Mesh Size.

Two material yield functions were considered for use in the finite element
models. These were an isotropic von-Mises yield function and the anisotropic
3-Parameter Barlat yield function. Both of these yield functions are discussed
in Chapter 5. Dent depths of 0.850mm and 0.835mm were obtained using the
von-Mises Yield function (see models 4 and 5 in Figure 6.43). The result for
model number 4 was obtained using a small refinement radius of 30mm, while
the result from model number 5 used a larger refinement radius of 100mm.
All other models in Figure 6.43 used the 3-Parameter Barlat yield function.
Comparison of the yield functions can be made by either model 4 to model 3
or model 5 to model 6. In the first comparison (models 3 and 4), the use of
the von-Mises yield criteria resulted in a dent depth prediction that is closer
to the measured value, while in the second comparison the opposite is true.
The lack of a general trend is likely a combined effect of mesh refinement and
choice of material yield function.
Models 4 and 5 required switching material yield functions after the first
springback step. Without considering the dent, switching the yield function
after the first springback step could result in slightly different equilibrium
position for the panel obtained during the final springback solution. Mesh
refinement compounds this issue by introducing an averaged stress distri-
bution in the refined elements, which also causes a slight difference in the
equilibrium position of the panel prior to denting.
Even though model 3 in Figure 6.43 shows that the combination of a
3-Parameter Barlat yield function and wide radius of refinement did not
produce a predicted dent in better accord with the measured values, these
modelling parameters were adopted in the current study as they represent
“best practice”. The 3-Parameter Barlat yield function was chosen to help
match the material thinning during forming, which was deemed very impor-
tant. The larger refinement radius was chosen for consistency with the static
dent models and to avoid mesh sensitivity issues, even though it produced a
larger dent depth in this particular case.

134
6.3.3.3 Sensitivity to Position of the Indentor.

One would generally consider that the relative position of the indentor
above the panel should not affect the dynamic dent prediction. In a dy-
namic dent simulation, this parameter should only affect how long it takes
the indentor to strike the panel, as the velocity of the indentor is constant
prior to denting in the absence of gravity. However, during the dynamic dent
simulations it was also observed that the panels tended to oscillate after the
springback stage, prior to the indentor striking the panel. These oscillations
produced negligible displacements, but velocities of the same order of magni-
tude as the indentor. One would expect the panel to be stationary following
a springback step, so questions were drawn regarding the validity of the equi-
librium state. It was determined by Maker [78] that these oscillations were
very high frequency stress waves traversing the panel, an artifact of the ex-
plicit finite element formulation in LS-DYNA. These stress waves produced
high velocities, but only small displacements at the nodes. As a method
of determining whether these oscillations caused differences in dynamic dent
predictions, two simulations were run so that the panel was struck at different
times. Any differences in dent depth caused by local panel momentum effects
would thus be visible. The dent depth predictions obtained from models 6
and 7, shown in Figure 6.43, are nearly identical, so these stress waves did
not have an effect.

6.3.3.4 Sensitivity to Residual Stresses.

Residual stresses are present in the panel after the forming stage. The
springback step allows these stresses to find an equilibrium state, which in
some cases can cause large deflections. If these residual stresses did not
contribute to the final dent depth, the dent simulation process could be
simplified for stiff panels by foregoing springback steps. The result for model
number 8 in Figure 6.43 was simulated without the residual stresses from the
forming simulation, but utilised the predicted effective plastic strains (work
hardening). The dent that was predicted without the use of residual stresses

135
was 0.755mm, while the dent predicted using residual stresses was 0.797mm.
This difference shows that residual stresses do have an effect on dent depth
predictions and should be included in the denting simulation.

136
Chapter 7

Discussion and Conclusion

7.1 Discussion
A numerical method of predicting both static and dynamic dents in au-
tomotive panels has been developed. This method is centered around an
integrated suite of commercial and custom software programs. This mod-
elling system uses the finite element method as a predictive tool which en-
ables automotive panel designers to assess performance prior to production
of tooling. Experimental validation of dent predictions using this method
has shown that the trends of both static and dynamic dent resistance have
been captured quite well. These validations also showed the sensitivity of the
results to various parameters such as panel thickness, pre-strain, curvature
and thickness as well as numerical formulation parameters. It is particularly
important to use forming data within the denting models for accurate results
to be obtained. In this regard, the software created as part of this research
is able to both refine and remap data from existing forming models for use
in subsequent dent tests. This software has been particularly successful in
transforming the forming models into “ready-to-run” finite element denting
models, thereby minimising the effort required to perform dent simulations.
The current models have captured static and dynamic denting trends
very well. In some experiments, such as the AA5754 static dents and AA6111
dynamic dents, values were predicted that were very close to those measured.

137
In other tests, such as the dynamic dents performed on the AA5754 panels,
the predicted dent depths were larger than those measured, while in the
AA6111 static tests, the dent depths were under-predicted. These errors are
not unexpected considering the small size of the dent depth compared to
the panel size and the numerous sources of uncertainty in the experiments.
Thus, it is advisable at present to treat dent prediction as a qualitative
comparison tool. Such conservatism would still permit comparison of the
relative performance of different panel designs in assessments of changes in
panel thickness or alloy, for example.
The resolution required in a dent test exacerbates the issue. In general,
dents are determined to be visible when their depth is greater than 0.1mm.
Therefore, for this method to be applicable for panel certification purposes,
repeated resolution to values of less than 0.01mm are necessary. For this
level of resolution, it may be necessary to consider other finite element dis-
cretisation techniques, such as several layers of brick elements in the denting
region. Considering the large size of most automotive panels, the computa-
tional costs associated with such an approach are likely prohibitive. Models
that were run during this work were almost unmanageable due to their size
and computer hardware requirements.
Owing to the sensitivity of the results to material data, it may be neces-
sary to obtain more accurate tensile data of the initial yield behaviour of the
aluminum. The tensile data given in Chapter 3 was adequate for the form-
ing simulation; however, the strains present in a dent are small compared
to those produced during forming. The prediction of the initial yield point
requires very accurate low strain data, which may have been obscured during
the curve fitting technique used during the current work. This could help
explain the over-prediction of the oil-canning loads in the AA6111 panels.
The choice of material yield criterion was limited by the finite element
codes. Whereas LS-DYNA has several anisotropic material models, such
as the 3-Parameter Barlat model, that are formulated specifically for sheet
metal forming, ABAQUS does not. This necessitated the use of an isotropic

138
von-Mises yield function for the static dent tests. Use of this isotropic yield
function could have contributed to some of the discrepancy within the static
dent models since a different yield locii are predicted by the two models.
These yield locii are shown in Figure 7.1 for the 0.81mm AA6111-T4 material
data (R00=0.62, R45=0.66, R90=0.86 and σy =140MPa).

200

180 3-Parameter Barlat

160
Local Y Stress [MPa]

140
von-Mises
120

100

80

60

40

20

0
0 20 40 60 80 100 120 140 160 180 200
Local X Stress [MPa]
Figure 7.1: Yield locii predicted by the 3-Parameter Barlat and
von-Mises yield functions for the 0.81mm AA6111-T4
material data (R00=0.62, R45=0.66, R90=0.86 and
σy =140MPa).

Adoption of the 3-Parameter Barlat model has a further disadvantage.

Although Harpell [50] has shown that the 3-Parameter Barlat material model
captures sheet thinning well, Figure 7.1 highlights a deficiency of the model.

139
The 3-Parameter Barlat model matches the three R-values (R00, R45 and
R90) and also the yield strength in the rolling direction. It does not, how-
ever, match the yield strength in material directions other than the rolling
direction, such as the transverse direction. This limitation can cause under
or over-prediction of the yield strength in the transverse direction, as shown
when the transverse yield strength is compared to that measured during the
tensile tests. In Figure 7.1 the transverse yield strength is 154Mpa, while the
measured value shown in Figure 3.5 is 135Mpa. Newer material models that
address this issue have been proposed by Barlat [79]; however, these models
remain proprietary and are not yet available.

7.2 Conclusions
The following conclusions are drawn from this research.

• A software system which enables the automotive panel designer to

quickly post-process an existing finite element forming model to gener-
ate a static or dynamic denting model has been created. This software
tool set can be used to qualitatively predict both dynamic and static
denting behaviour. Quantitative prediction requires greater caution
and “calibration” with available experimental data.

• Both static and dynamic dents have been predicted for AA5754 and
AA6111 panels. The AA6111 panels performed much better than the
AA5754 panels due to their higher yield strength.

• Increasing sheet thickness will help static dent resistance, but in cer-
tain situations will hinder dynamic dent resistance. This is due to the
effect that thickness has on panel stiffness. Whereas increased panel
stiffness allows a higher static load to be carried by the panel prior
to oil-canning, it creates larger contact forces during dynamic denting.
These high contact forces cause localised deformation during a dynamic
dent.

140
 • The fully integrated shell element (number 16) was necessary to obtain
repeatable convergent solutions within the springback simulations. It
was also necessary to obtain reasonable dynamic dent predictions.

• During the current work it was found that the implicit LS-DYNA for-
mulation was not robust enough to repeatedly gain convergence during
static dent simulations. Moreover it did not have required boundary
conditions that would enable its use for general automotive assemblies.
To this end ABAQUS was used for the static dent tests. It was found
that convergence of the static dent tests was repeatably achieved, with-
out the requirement for user intervention.

7.3 Recommendations
Future work should include a mesh refinement study to determine if the
accuracy of the results can be improved. This refinement should include
other discretisation techniques such as solid elements. The refined denting
region could be modelled with a different yield stress curve, one that only
considers the strain range present in the denting region.
An improved yield surface formulation is required to accurately model the
material yield strength in directions other than the rolling direction. This
formulation should be used in both the implicit and explicit calculations for
consistency.
This work should be extended to include complete automotive panel as-
semblies. In this manner, the support provided by inner panels can be in-
cluded to assess the denting characteristics of the complete panel assembly.
Automating the static test apparatus would reduce the error caused by
manual placement of the weights on the loading arm and would decrease the
time required to perform a static dent test. Automation should follow the
guidelines published in the Auto/Steel Partnership Report [34].

141
Bibliography

[1] Bart van Veldhuizen, Winfried Kranendonk, and Rob Ruifrok. The re-
lation between the curvature of horizontal automotive panels, the panel
stiffness and the static dent resistance. In Materials and Body Testing
IBEC’95, pages 62–70. IBEC, 1995.

[2] Hikmat F. Mahmood. Combined shallow spherical shell and finite ele-
ment method of panel stiffness. Technical Memorandum 8031, Chrysler
Corporation, September 1978.

[3] A.V. Vadhavkar, M.G. Fecek, V.C. Shah, and W.E. Swenson. Panel
optimization program (pop). Paper 810230, SAE, 1981.

[4] C.L. Alaniz, L.E. Brookes, and T.N. Seel. Investigation of body panel
stiffness as predicted by finite element analysis. Paper 900716, SAE,
1990.

[5] J.A. DiCello and R.A. George. Design criteria for the dent resistance of
auto body panels. Paper 740081, SAE, 1974.

[6] Y. Yutori, S. Nomura, I. Kokubo, and H. Ishigaki. Studies on the static

dent resistance. Memoires Scientifiques Revue Metallurgie, 77(4):561–
569, April 1980.

[7] Hikmat F. Mahmood. Dent resistance of surface panel and slam area.
Paper 810099, SAE, 1981.

142
 [8] E. Reissner. Stresses and small displacement of shallow spherical shells.
Math and Physics, 25(24), 1947.

[9] M.A. McCormick, D.J. Meuleman J.R. Fekete, and M.F. Shi. Effect of
steel strengthening mechanisms on dent resistance of automotive body
panels. Paper 980960, SAE, 1998.

[10] T. E Johnson, Jr and W.O. Schaffnit. Dent resistance of cold-rolled

low-carbon steel sheet. Paper 730528, SAE, 1973.

[11] M.J. Worswick, M.J. Finn, and J.J. Thorburn. A numerical study of the
geometric and material parameters affecting static and dynamic dent-
ing. Internal Report KR-97/025, Alcan International Limited, Kingston
Research and Development Centre, November 1997.

[12] C.E. Burley, B. A. Niemeier, and G. P. Koch. Dynamic denting of

autobody panels. Paper 760165, SAE, 1976.

[13] C. E. Burley and B. A. Niemeier. Denting properties of aluminum au-

tobody components. Paper 770199, SAE, 1977.

[14] R. L. Rolf, M. L. Sharp, and H. H. Stroebel. Structural characteristics

of aluminum body sheet. Paper 770200, SAE, 1977.

[15] Bernard A. Niemeier and Carlton E. Burley. Hailstone response of body

panels - real and simulated. Paper 780398, SAE, 1978.

[16] D. C. Chang and R. P. Khetan. Surface damage of steel, aluminum,

and chopped-fiber composite panels due to projectile impact. Journal
of Reinforced Plastics and Composites, 3:193–203, July 1984.

[17] Thomas N. Seel. Bake hardening steel application study - key factors of
dent resistance improvement. Paper 910291, SAE, 1991.

[18] S. Nomura, Y. Yutori, J. Iwaya, M. Miyahara, and I. Kokub. A study of

the dynamic dent resistance. In IDDRG Conference Proceedings, pages
394–402. IDDRG, 1984.

143
[19] Herbert J. Thorburn. Comparative tests of stiffness and dent resistance
on aluminum and steel fenders. In Automotive Body Materials IBEC’94,
pages 105–112. IBEC, 1994.

[20] Gunnar Ekstrand and Nader Asnafi. On testing of the stiffness and the
dent resistance of autobody panels. Materials and Desigh, (19):145–156,
1998.

[21] Nader Asnafi. On strength, stiffness and dent resistance of car body
panels. Journal of Materials Processing Technology, (49):13–31, 1995.

[22] Marcus A. Sabbagh, Ravi N. Chavali, and Jay S. Montgomery. Quasi-

static dent depth simulation using non-linear fea. In IBEC ’95 Materials
and Body Testing, pages 87–90. IBEC, 1995.

[23] Jay Montgomery and Alan Brooks. Influence of geometry on outer panel
dent resistance as demonstrated through fea. In IBEC ’94 Automotive
Body Materials, pages 85–89. IBEC, 1994.

[24] Mark F. Werner. Finite element simulation of steel body panel perfor-
mance for quasi-static dent resistance. In IBEC ’93 Automotive Body
Materials, pages 87–90. IBEC, 1993.

[25] Kuo-Kuang Chen and Paul A. Salamie. A mathematical model for cal-
culating the dent initiation loads at the door centers. Paper 841201,
SAE, 1984.

[26] H. Sakai, K. Saito, and H. Tsukada. Stiffness and dent characteristics

of body outer surface panel - finite element analysis and experiment.
Inernational Journal of Vehicle Design, 4(1):13–22, 1983.

[27] Chung-Hsin Chen, Prabhat K. Rastogi, and Curt D. Horvath. Effects

of steel thickness and mechanical properties on vehicle outer panel per-
formance: Stiffness, oil canning load and dent resistance. In Automotive
Body Materials IBEC’93. IBEC, 1993.

144
[28] C. L. Alaniz and J. E. Borchelt. Study of parameters that affect body
panel performance predictions. Paper 910289, SAE, 1991.

[29] Ming F. Shi, David J. Meuleman, Christine L. Alaniz, and Stephen J.

Zurdosky. Prestrain effects on static dent resistance of automotive steels.
Paper 910288, SAE, 1991.

[30] F. Gatto and D. Morri. Forming properties of some aluminium alloys

sheets for car - body use. In 12th Biennal Congress International Deep
Drawing Research Group. IDDRG, May 1982.

[31] A. Daglish and D.J. Cunnane. The static dent resistance of autobody
alloys as a function of strain. Technical Report BR-81/77, ALCAN,
1981.

[32] Ming F. Shi and J. S. Wilczynski. Potential weight savings versus dent
resistance performance of high strength steel body panels. In Automotive
Body Materials IBEC’93. IBEC, 1993.

[33] Ronald P. Krupitzer and Rodney P. Harris. Improvements in the dent

resistance of steel body panels. Paper 920243, SAE, 1992.

[34] Daniel J. Schaeffler. Procedures for dent resistance evaluations of auto-

motive outer body panels and assemblies under quasi-static and dynamic
indenting conditions. Technical Report 1.0, Auto / Steel Parnership
Standardized Dent Resistance Test Task Force, January 1997.

[35] Aleksy A. Konieczny, Yishan Qiu, and Balbir Singh. A practical program
set for stiffness / dent resistance calculation of autobody outer panels.
In Automotive Body Materials IBEC’97, pages 54–58. IBEC, 1997.

[36] K.S. Raghavan and Naji Arwashan. Analysis of quasi-static denting

behavior of automotive sheet steels. Paper 970984, SAE, 1997.

145
[37] M. F. Shi, J. A. Brindza, P. F. Michel, P. Bucklin, P.J. Belanger, and
Jr. J. M. Prencipe. Static and dynamic dent resistance performance of
automotive steel body panels. Paper 970158, SAE, 1997.

[38] Ravi Chavali and Wenhui Song. Coupling forming and denting sim-
ulation for automotive closure panels. Unknown Origin, 1996. Paper
Number 96NM098.

[39] J.W. Phillips and H.H Calvit. Impact of a rigid sphere on a viscoelastic
plate. Journal of Applied Mechanics, pages 873–878, December 1967.

[40] Chi-Mou Ni. Analysis of panel dent resistance. Internal Report N77
10290, General Motors, 1976.

[41] Peter T. Vreede, Paul J. Tamis, and Mary E. Roelofsen. The influence
of material properties and geometry on dynamic dent resistance: Exper-
iments and simulations. In Materials and Body Testing IBEC’95, pages
79–86. IBEC, 1995.

[42] Ming F. Shi, David J. Meuleman, Christine L. Alaniz, and Stephen J.

Zurdosky. An evaluation of the dynamic dent resistance of automotive
steels. Paper 910287, SAE, 1991.

[43] New dent resistance data. Dow Metal Products News, December 1960.

[44] Sidney H. Greenfeld. Hail resistance of roofing products. Paper, United

States Department of Commerce, National Bureau of Standards, August
1969. CODEN: BSSNB.

[45] Larry Sutak. ALCAN, Private communication.

[46] John O. Hallquist. LS-DYNA THEORETICAL MANUAL. Livermore

Software Technology Corporation, 1998.

[47] Pamstamp software. ESI Group Headquarters, 6, rue Hamelin, BP 2008-

16 75761, Paris Cedex 16 France.

146
[48] Optris virtual press software, 2000. Dynamic Software, Vincent FER-
RAGU 5, Parc Club du Golf, 13856 Aix-en-Provence Cedex 3, France.

[49] Autoform 2.1 - finite element software, 2000. AutoForm Engineering

GmbH, Technoparkstrasse 1, CH-8005 Zurich Switzerland.

[50] Eric T. Harpell. Numerical prediction of limiting draw ratio in aluminum

cylindrical cup drawing. M.eng, Carleton University, Ottawa, Ontario,
May 1997.

[51] Chantal G. Lamontagne. Numerical simulation of aluminum sheet form-

ing incorporating plastic anisotropy. M.eng, Carleton University, Ot-
tawa, Ontario, May 1996.

[52] J. C. Gelin and P. Picart, editors. Proceedings of NUMISHEET ’99,

volume 2, Beasancon - France, September 1999. Benchmark A - Forming
of a Front Door Panel.

[53] ABAQUS/Standard Theory Manual. Hibbit, Karlsson & Sorensen, 1080

Main Street Pawntucket, RI 02860-4847 USA, 1997. Version 5.7.

[54] Xiao Zhou. Numerical prediction of springback in u-channel forming of

aluminum tailor welded blanks. M.eng, Carleton University, Ottawa,
Ontario, October 1999.

[55] MARC Nonlinear Finite Element Solver. MARC Analysis Research

Coorporation.

[56] LS-DYNA USERS’S MANUAL (Nonlinear Dynamic Analysis of Struc-

tures in Three Dimensions), Version 940. Livermore Software Technol-
ogy Corporation, 1997.

[57] B. Hodgins, M. Worswick, D. Thomas, M. Finn, and K. Gong. Dent

resistance of medium scale aluminum structural assemblies. Paper 01B-
119, SAE, 2001.

147
[58] Standard Template Library Programmer’s Guide. Silicon Graphics Com-
puter Systems, Inc., 1996.

[59] CDE 2.1/Motif 2.1 User’s Guide. The Open Group, 29B Montvale Ave,
Woburn MA 01801 USA, 1997.

[60] Pralay Dakua. Free source code downloaded from the internet, July
1997. ftp://ftp.x.org/contrib/widgets/motif/Tab.

[61] T. Belytschko and C.S. Tsay. Explicit algorithms for nonlinear dynamics
of shells. AMD, 48:209–231, 1981.

[62] T. Belytschko, J.Lin, and C.S. Tsay. Explicit algorithms for nonlinear
dynamics of shells. Computer Methods in Applied Mechanical Engineer-
ing, 42:225–251, 1984.

[63] T.J.R. Hughes and W.K. Liu. Nonlinear finite element analysis of shells:
Part ii. three dimensional shells. Computer Methods in Applied Mechan-
ics, 27:331–362, 1981.

[64] Michael E. Mortenson. Geometric Modeling. John Wiley and Sons, Inc.,
New York, 2 edition, 1997.

[65] LS-TAURUS Data Base. Livermore Software Technology Corporation,

1998.

[66] LS-POST User’s Manual. Livermore Software Technology Corporation,

1998.

[67] ABAQUS/Standard User’s Manual. Hibbit, Karlsson & Sorensen, 1080

Main Street Pawntucket, RI 02860-4847 USA, 1997. Version 5.7.

[68] LSTC. LS-DYNA USERS’S MANUAL (Nonlinear Dynamic Analysis

of Structures in Three Dimensions), Version 950. Livermore Software
Technology Corporation, 1999.

148
[69] D. J. Lloyd. Aluminum alloys used in automotive skin sheet. In D.S.
Wilkinson, W.J. Poole, and A. Alpas, editors, Advances in Industrial
Materials, pages 3–17. The Metallurgical Society of CIM, 1998. ISBN:
0919086-79-9.

[70] John A. Schey. Introduction to Manufacturing Processes. Mc Graw Hill,

3 edition, 1999.

[71] W.T. Lankford, S.C. Snyder, and J. A. Bauscher. Transactions of the

ASM, 42:1197, 1950.

[72] Theodore Baumeister, Eugene A. Avallone, and Theodore Baumeis-

ter III, editors. Marks’ Standard Handbook for Mechanical Engineers.
McGraw-Hill Book Company, 8 edition, 1978.

[73] R. Sowerby and J. L. Duncan. Failure in sheet metal in biaxial tension.

International Journal of Mechanical Science, 13:217–229, 1971.

[74] J. C. Simo and F. Armero. Geometrically non-linear enganced strain

mixed methods and the method of incompatible modes. International
Journal for Numerical Methods in Engineering, 33:1413–1449, 1992.

[75] William F. Hosford and Robert M. Caddell. Metal Forming - Mechanics

and Metallurgy. Prentice Hall, 2 edition, 1993.

[76] B. Maker. On drawbeads in sheet metal forming. Paper 2000-01-0776,

SAE, 2000.

[77] Fred Barlat and J. Lian. Plastic behavior and stretchability of sheet
metals. part 1: A yield function for orthotropic sheets under plane stress
conditions. International Journal of Plasticity, 5:51–66, 1989.

[78] B. Maker, 1999. Livermre Software Technology Corporation.

[79] Fred Barlat. Private Communication, 2000.

149