Tjlahey 2002

Modelling Hysteresis in the Bending of Fabrics
by
Timothy John Lahey
A thesis
presented to the University of Waterloo
in fulfilment of the
thesis requirement for the degree of
Master of Applied Science
in
Systems Design Engineering
Waterloo, Ontario, Canada, 2002
Timothy
c John Lahey 2002
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
Abstract
This thesis presents a model of fabric bending hysteresis. The hysteresis model is de-
signed to reproduce the fabric bending measurements taken by the Kawabata Evalua-
tion System (KES) and the model parameters can be derived directly from these prop-
erty measurements. The advantage to using this technique is that it provides the ability
to simulate a continuum of property curves. Results of the model and its components
are compared and constrasted with experimental results for fabrics composed of differ-
ent weaves and yarn types.
An attempt to incorporate the bending model as part of a fabric drape simulation is
also made.
iii
Acknowledgements
Over the duration of this thesis I have received a great deal of advice, assistance, and
encouragement. I would first like to thank Shekar for his encouragement in pursuing
a graduate degree. Without him, I would not have considered it. My sincere thanks go
to to Rebecca Kaufman without whose support I would have not have completed this
thesis.
The critiques offered by members of CGL, the ConStruct group, and members of the
Systems Modelling and Simulation lab helped me refine the concepts and my under-
standing. I would especially like to thank Ajay, Ivan, and Hank for their help, encour-
agement, and distractions during this thesis. My thanks go to Dr. Jeffrey Eischen for
providing me with a copy of Shigan Deng’s thesis which contained the necessary data
for this thesis work.
I received a great deal of help from my former supervisor Richard Bartels even if it
was only asking the right question. Lastly, I would like to thank my supervisor, Glenn
Heppler for his patience, help, and encouragement.
iv
Contents
1 Introduction 1
1.1 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Research in Computer Graphics . . . . . . . . . . . . . . . . . . . 2
1.1.2 Research in Textile Engineering . . . . . . . . . . . . . . . . . . . . 3
1.2 Layout of this document . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Fabric Mechanical Properties 5

2.1 Woven Fabric Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Tension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Shear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Bending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5 Kawabata Evaluation System . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5.1 Tension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5.2 Shear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5.3 Bending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3 Previous Work on Hysteresis 22

3.1 Early Mechanical Material Hysteresis Models . . . . . . . . . . . . . . . . 22
3.2 General Hysteresis Models . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
v
3.3 A Differential Model of Frictional Hysteresis . . . . . . . . . . . . . . . . 25
3.3.1 Physical Parameters and Their Model Equivalent . . . . . . . . . 27
3.3.2 Other Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4 A Model for Fabric Hysteresis 30

4.1 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2 Determining Model Parameters from Experiments . . . . . . . . . . . . . 32
4.2.1 Modelling the KES Bending Test Position, Velocity, and Accelera-
tion Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.2.2 Fitting the Model Parameters to Experiments . . . . . . . . . . . . 39
4.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3.1 Fabric 1 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.3.2 Fabric 2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5 The Fabric Drape Simulation Model 80

5.1 The B-spline Field Approximation Method . . . . . . . . . . . . . . . . . 81
5.1.1 Properties of the B-spline discretisation . . . . . . . . . . . . . . . 81
5.2 A Linear Approximation: Pure Bending of a Beam . . . . . . . . . . . . . 82
5.2.1 Formation of the Mass Matrix . . . . . . . . . . . . . . . . . . . . . 82
5.2.2 Formation of the Stiffness Matrix . . . . . . . . . . . . . . . . . . . 84
5.2.3 Formation of the Consistent Force Vector . . . . . . . . . . . . . . 86
5.2.4 Application of Boundary Conditions . . . . . . . . . . . . . . . . . 87
5.2.5 Solution of the System of Equations . . . . . . . . . . . . . . . . . 88
5.3 A Linear Approximation: Pure Bending of a Plate . . . . . . . . . . . . . 89
5.4 Modelling the Non-linear Elastic Component . . . . . . . . . . . . . . . . 90
5.5 Modelling the Frictional Component . . . . . . . . . . . . . . . . . . . . . 93
5.6 Implementation Details and Problems . . . . . . . . . . . . . . . . . . . . 93
vi
6 Summary 95
6.1 Discussion of the Current Model . . . . . . . . . . . . . . . . . . . . . . . 95
6.2 Discussion of Modelling for Simulation Purposes . . . . . . . . . . . . . 96
A Glossary 97
B Further Information 100

B.1 Equations of an Elastica . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
B.2 Derivation of Bliman and Sorine results . . . . . . . . . . . . . . . . . . . 106
B.3 KES Bending Test Profile Equations . . . . . . . . . . . . . . . . . . . . . . 108
B.3.1 Unloading/Reverse Bending Path (Regions 4-6) . . . . . . . . . . 108
B.3.2 Re-Loading Bending Path (Regions 7-9) . . . . . . . . . . . . . . . 110
B.4 Polynomial Fit to Experimental Data . . . . . . . . . . . . . . . . . . . . . 110
Bibliography 114
vii
List of Tables
2.1 Classes of Fabric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
4.1 Details of Fabrics Tested [Den94] . . . . . . . . . . . . . . . . . . . . . . . 40

4.2 Fabric Properties [Den94] . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.3 Directional Slope Comparsion . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.4 Fabric 1 - Warp Direction: Model Parameters . . . . . . . . . . . . . . . . 44
4.5 Fabric 1 - Warp Direction: Supplementary Friction Results . . . . . . . . 44
4.6 Fabric 1 - Warp Direction: Model Comparision . . . . . . . . . . . . . . . 46
4.7 Fabric 1 - Weft Direction: Model Parameters . . . . . . . . . . . . . . . . . 55
4.8 Fabric 1 - Weft Direction: Supplementary Friction Results . . . . . . . . . 55
4.9 Fabric 1 - Weft Direction: Model Comparision . . . . . . . . . . . . . . . . 56
4.10 Fabric 2 - Warp Direction: Model Parameters . . . . . . . . . . . . . . . . 64
4.11 Fabric 2 - Warp Direction: Supplementary Friction Results . . . . . . . . 64
4.12 Fabric 2 - Warp Direction: Model Comparision . . . . . . . . . . . . . . . 65
4.13 Fabric 2 - Weft Direction: Model Parameters . . . . . . . . . . . . . . . . . 72
4.14 Fabric 2 - Weft Direction: Supplementary Friction Results . . . . . . . . . 72
4.15 Fabric 2 - Weft Direction: Model Comparision . . . . . . . . . . . . . . . . 73
B.1 Fabric 1 - Warp Direction: Loading Polynomial Coefficients . . . . . . . . 110
viii
B.2 Fabric 1 - Warp Direction: Unloading/Reverse Loading Polynomial Co-
efficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
B.3 Fabric 1 - Warp Direction: Reverse Unloading Polynomial Coefficients . 111
B.4 Fabric 1 - Weft Direction: Loading Polynomial Coefficients . . . . . . . . 111
B.5 Fabric 1 - Weft Direction: Unloading/Reverse Loading Polynomial Co-
efficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
B.6 Fabric 1 - Weft Direction: Reverse Unloading Polynomial Coefficients . . 111
B.7 Fabric 2 - Warp Direction: Loading Polynomial Coefficients . . . . . . . . 112
B.8 Fabric 2 - Warp Direction: Unloading/Reverse Loading Polynomial Co-
efficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
B.9 Fabric 2 - Warp Direction: Reverse Unloading Polynomial Coefficients . 112
B.10 Fabric 2 - Weft Direction: Loading Polynomial Coefficients . . . . . . . . 112
B.11 Fabric 2 - Weft Direction: Unloading/Reverse Loading Polynomial Co-
efficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
B.12 Fabric 2 - Weft Direction: Reverse Unloading Polynomial Coefficients . . 113
ix
List of Figures
2.1 Pierce’s Geometrical Model [HGB69] . . . . . . . . . . . . . . . . . . . . . 7

2.2 Race-Track Cross-section [HGB69] . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Forces on Olofsson’s elastica [HGB69] . . . . . . . . . . . . . . . . . . . . 11
2.4 Generalised Tensile Load-Extension Curve [HGB69] . . . . . . . . . . . . 13
2.5 Grosberg’s micromechanical model [Gro66a] . . . . . . . . . . . . . . . . 13
2.6 Zhou and Ghosh Bending Model [ZG99] . . . . . . . . . . . . . . . . . . . 16
2.7 Shi et al. Bending Model [SHY00] . . . . . . . . . . . . . . . . . . . . . . . 16
2.8 KES: Application of Tension Force [Kaw80] . . . . . . . . . . . . . . . . . 17
2.9 KES: Tension Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.10 KES: Shear Process [Kaw80] . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.11 KES: Shear Measurements [Kaw80] . . . . . . . . . . . . . . . . . . . . . . 20
2.12 KES: Bending Measurements [Kaw80] . . . . . . . . . . . . . . . . . . . . 21
3.1 Fatigue Mechanical Model [MTS71] . . . . . . . . . . . . . . . . . . . . . 23

3.2 Piecewise Linear Approximation to Stress-Strain Curve [MTS71] . . . . . 23
3.3 Ordinary Play [KP89] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.4 Ordinary Stop [KP89] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.5 Bliman and Sorine 2nd order model . . . . . . . . . . . . . . . . . . . . . . 28
4.1 Model of Fabric Bending . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
x
4.2 KES Bending Test Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 Acceleration Profile (T=1) . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.4 Velocity Profile (T=1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.5 Portion of Fabric 1 Weft Direction Objective Surface . . . . . . . . . . . . 41
4.6 Warp-float Satin (4x1) Weave Cross-section . . . . . . . . . . . . . . . . . 43
4.7 Warp-float Satin (4x1) Weave Pattern . . . . . . . . . . . . . . . . . . . . . 44
4.8 Fabric 1 - Warp Direction Full Model . . . . . . . . . . . . . . . . . . . . . 47
4.9 Fabric 1 - Warp Direction (No Slowing at End) . . . . . . . . . . . . . . . 48
4.10 Fabric 1 - Warp Direction (No Friction Term) . . . . . . . . . . . . . . . . 49
4.11 Fabric 1 - Warp Direction (No Cubic Term) . . . . . . . . . . . . . . . . . 50
4.12 Fabric 1 - Warp Direction (No Cubic or Friction Term) . . . . . . . . . . . 51
4.13 Fabric 1 Warp - Frictional Contribution . . . . . . . . . . . . . . . . . . . . 53
4.14 Fabric 1 Warp - Non-linear Visco-elastic Contribution . . . . . . . . . . . 54
4.15 Fabric 1 - Weft Direction Full Model . . . . . . . . . . . . . . . . . . . . . 57
4.16 Fabric 1 - Weft Direction (No Friction Term) . . . . . . . . . . . . . . . . . 58
4.17 Fabric 1 - Weft Direction (No Cubic Term) . . . . . . . . . . . . . . . . . . 59
4.18 Fabric 1 - Weft Direction (No Cubic or Friction Term) . . . . . . . . . . . 60
4.19 Fabric 1 Weft - Frictional Contribution . . . . . . . . . . . . . . . . . . . . 61
4.20 Fabric 1 Weft - Non-linear Visco-elastic Contribution . . . . . . . . . . . . 62
4.21 Twill (1x2) Weave Cross-section . . . . . . . . . . . . . . . . . . . . . . . . 63
4.22 Twill (1x2) Weave Pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.23 Fabric 2 - Warp Direction Full Model . . . . . . . . . . . . . . . . . . . . . 66
4.24 Fabric 2 - Warp Direction (No Friction Term) . . . . . . . . . . . . . . . . 67
4.25 Fabric 2 - Warp Direction (No Cubic Term) . . . . . . . . . . . . . . . . . 68
4.26 Fabric 2 - Warp Direction (No Cubic or Friction Term) . . . . . . . . . . . 69
4.27 Fabric 2 Warp - Frictional Contribution . . . . . . . . . . . . . . . . . . . . 70
4.28 Fabric 2 Warp - Non-linear Visco-elastic Contribution . . . . . . . . . . . 71
xi
4.29 Fabric 2 - Weft Direction Full Model . . . . . . . . . . . . . . . . . . . . . 74
4.30 Fabric 2 - Weft Direction (No Friction Term) . . . . . . . . . . . . . . . . . 75
4.31 Fabric 2 - Weft Direction (No Cubic Term) . . . . . . . . . . . . . . . . . . 76
4.32 Fabric 2 - Weft Direction (No Cubic or Friction Term) . . . . . . . . . . . 77
4.33 Fabric 2 Weft - Frictional Contribution . . . . . . . . . . . . . . . . . . . . 78
4.34 Fabric 2 Weft - Non-linear Visco-elastic Contribution . . . . . . . . . . . . 79
xii
Chapter 1
Introduction
Do not seek to follow in the footsteps of the men of old; seek what they sought.
– Basho
Textiles have become more prominent in engineering applications in recent years,

with uses such as airbags and domes. Since the 1930s, there has been research reported
in the textile engineering literature regarding the mechanical properties of fabric, its
behaviour and, recently, on how to model its behaviour in computer graphics applica-
tions.
The initial goal of this thesis was to develop a model of fabric mechanics from prop-
erty curves to be employed as part of a computer simulation of fabric drape. Efforts
to model the shear and tensile behaviour were abandoned in favour of focusing on the
bending model. The decision to focus on accurately modelling the bending behaviour,
including hysteresis, was taken because bending rigidity and hysteresis were found to
be important in determining fabric drape [HC98, JP98].
1
1.1. PREVIOUS WORK 2
1.1 Previous Work
Work on models of fabric behaviour has been done in two fields, computer graphics and
textile engineering with little overlap. The work done in computer graphics is focused
upon a good visual approximation while the work in textile engineering focuses on a
model that is physically justified.
1.1.1 Research in Computer Graphics
The first paper on modelling fabric drape in the computer graphics field was done
by Weil [Wei86] at SIGGRAPH ’86. This model was purely geometric and as a result,
only provided a static representation. In that same year Nisselson gave a talk [Nis86]
on the relationship of computer graphics and fashion. This lead to a panel discussion
[NWGC87] on the same topic at SIGGRAPH ’87.
Over the years a large number of lumped parameter models (called particle models
in the literature) have been presented in the computer graphics literature for the dy-
namic simulation of fabric behaviour. Considerable work was done by N. Magnenat
Thalmann, D. Thalmann and their research groups using a linear elastic model for cloth-
ing on virtual actors [CYMTT92, MT98, VCMT95, VMT94, VMT95, VMTJT96, VT98].
The work of Dias et al. in the textile engineering literature appears very similar
[DGR00]. Also of note is the work of Breen et al. [BHG92, BHW94, HB98] whose model
is a grid of particles where energy potential functions (based upon results from the
Kawabata Evaluation System (KES) [Kaw80]) dictate the particle interactions. How-
ever, there is no physical justification given for the energy potential equations and only
the loading portion of the KES data is used (see Chapters 2 and 4 for information on
the KES). A similar approach is taken by Eberhardt et al. [EW97, EW99, EWS96, SE98]
who also used energy potentials but include both the loading and unloading KES data.
The difficulty with their approach is that polynomial fits to the loading and unloading
1.1. PREVIOUS WORK 3
curves are used and there no is discussion on how the transition from loading to un-
loading is handled. The work of Baraff and Witkin [BW98a, BW98b] is the basis for the
model in the Alias-Wavefront Maya Cloth software and for likely the basis for Pixar’s
cloth model used in Monsters Inc. [Rob01]. Their particle model gives no physical jus-
tification for the energy functions between neighbouring particles. In addition to the
particle models mentioned above, there are a number of other particle models in the
literature [Eis98a, LDG96, LKC96, LPC95, NGA95, Pro95, Pro96, VOVL92].
In addition to the work done in the computer graphics literature, work on computer-
aided design (CAD) with textiles has also been done. Work on garment design was done
by Hinds and McCartney [HM90, HM91], and Okabe et al. [OITN92] but their work is
strictly geometric. That is, their model does not account for differences due to fabric
properties. Aono et al. presented a useful framework [ABW93] which was fleshed out
with techniques for mapping fabric onto curved surfaces [ABW94] and handling darts
in fabric [ADBW96]. These papers do not present a mechanical model of fabric.
1.1.2 Research in Textile Engineering
There have been a number of fabric simulation models developed in the textile engi-
neering literature. The majority of the models are based upon standard engineering
approximation methods such as finite elements or finite volumes. A notable exception
is Postle and Postle’s model [PP99] which used techniques from non-linear dynamics to
analytically solve their equations. Note that this model does not include internal fric-
tion as part of the fabric model. Another exception is the model of Chen et al. [CSWY99]
which uses a multigrid approach to solve the energy minimisation. This model neglects
bending rigidity but adds a term they call ”wrinkling energy” to compensate.
One early model was that of Moskowitz et al. [MDS66] which considered uniform
lateral loads acting on two sets of plane, parallel elements superimposed at 90◦ to each
1.2. LAYOUT OF THIS DOCUMENT 4
other. This was done to represent the weave of the fabric. A finite difference approxima-
tion was used to solve the equations derived through the theory of minimum potential
energy. Later, Shanahan et al. [SLH78] presented a discussion of how the theory of
plates and shells could be applied to fabrics.
The majority of the models [CG95, GLS95, HC00a] found in the literature use an
orthotropic, linear elastic assumption for material behaviour. Where these models differ
is in their element choice. Gan et al. [GLS95] use a non-linear shell model as do Chen
and Govindaraj [CG95] while Hu and Chen [HC00a] use a finite volume method.
There have been some models that include non-linear effects. Bias-Singh et al.
[BSBG98] developed a non-linear finite element model for non-woven fabrics that in-
cludes a bi-linear tension model. Shigan Deng [Den94, EDC96, Eis98b] developed a
non-linear, orthotropic shell model that includes a polynomial fit to the KES bending
data. However, this model does not include bending hysteresis.
1.2 Layout of this document
The remaining portion of this thesis is divided into three main parts. The first part
(Chapters 2 and 3) details the background research done in developing the model. This
research includes background on fabric structure, fabric properties, and models of hys-
teresis. The second part presents the bending model (Chapter 4) and an effort to incor-
porate it into a fabric drape simulation (Chapter 5). The final chapter summarizes the
thesis. In addition, due to the number of unfamiliar terms used in this thesis a glossary
is provided in Appendix A.
Chapter 2
Fabric Mechanical Properties
Fabrics vary in type, fibre, geometric structure and their basic mechanical properties.
Using the classification system given by Hearle et al. [HGB69], fabrics and laminae can
be composed of material from five different classes. This system is described in Table
2.1.
Class Title Description

A Interlaced Yarns The traditional textiles. Once fibres with particular
properties are selected for the yarn, the arrangement
of fibres within the yarn and the arrangement of in-
terlaced yarns determine the properties of the fabric.
This class includes textiles such as: woven cloth, knit-
ted cloth, lace, and crochet.
B Non-interlaced Yarns This class is composed of bonded yarn sheets. A yarn
sheet is laid down without interlacing and a bonding
material is applied to the yarns, providing a coarse
network structure. This class includes a wide variety
of composite material (e.g., laminated composites).
5
2.1. WOVEN FABRIC STRUCTURE 6
Class Title Description

C Fibre Assemblies These are materials where individual fibres are laid
down in a web. This class includes textiles such as felt,
bonded fibre fabric, and spun-bonded filament fabric.
D Fibrous Sheets This class of materials has fibres but they are not eas-
ily identifiable as separate entities, instead the whole
material is fibrous in form. Examples of this class of
materials are leather and paper.
E Non-fibrous Sheets This class of materials has no discernible fibres, and
includes materials such as plastic film, rubber sheets,
metal foil, and foamed plastic.
Table 2.1: Classes of Fabric
A fabric or laminae can be made of material from a single class or they can be also
composed of material from a combination of classes. For example, fur is composed of
materials from classes C and D while synthetic leather is made of materials from classes
C and E. While the results of this thesis can likely be applied to all classes of fabrics,
due to the use of property curves instead of a micromechanical model, only fabrics from
class A are considered herein.
2.1 Woven Fabric Structure
The geometric structure of woven fabric plays a significant role in determining the
mechanical properties of the fabric and therefore some discussion of the structure of
fabric is necessary. Initially, a purely geometrical model was done by Pierce (described
in Hearle et al. [HGB69]) which is illustrated in Figure 2.1 which shows the geometrical
model in cross-section. The main components in Pierce’s model are, crimp height h,
L/2
h/2
d
D θ
Figure 2.1: Pierce’s Geometrical Model [HGB69]
yarn length l, yarn spacing p, yarn diameter d, crimp angle θ , and the sum of the two
yarn diameters, D. With the exception of D, each of the model components can have
different values in each of the weft (1) and warp (2) directions. The basic equations for
the weave are:
p = (l − Dθ ) cos θ + D sin θ (2.1)
h = (l − Dθ ) sin θ + D(l − cos θ ) (2.2)
D = h1 + h2 (2.3)
The basic model parameters are: h1 , h2 , l1 , l2 , p1 , p2 , and D (crimp angles are considered
to be dependent variables). If we have four of the parameters, the other three can be
calculated (and optionally the crimp angles).
Pierce’s model relies upon the assumptions that the bending resistance of the yarns
is negligible and that the yarns are circular in cross-section. These assumptions are
mutually consistent because if the yarn bending resistance is not negligible, then the
yarn cross-section will deform, producing a flatter cross-section. Alternative yarn
cross-sections have been proposed to solve this problem (see Hearle et al. pp 325-326
[HGB69]). The most popular alternative is a race-track cross-section as shown in Figure
2.2 which was proposed by Kemp in 1958. This cross-section is used in the calculation
of the “jamming” condition of fabrics.
Figure 2.2: Race-Track Cross-section [HGB69]
Both of these models suffer from the problem that they ignore the internal forces in
the fabric and the deformation of the cross-section that results. To compensate for this,
Olofsson developed a model that calculates the yarn geometry from the point loads
acting at the intersections (see Hearle et al. pp. 326-330 [HGB69]). This approach gives
a better approximation to observed yarn geometry than Pierce’s model. If V is the point
load at the cross-section (see Figure 2.3) then the bending moment at any point is given
by:
Mx = −Vx (2.4)
and the radius of curvature of the bent yarn (ρ) is given by:
m
ρ=− (2.5)
Vx
where m is the bending modulus of the yarn. We also have that:
ρdψ = ds (2.6)
dx = cos ψ ds (2.7)
where ds is the infinitesimal arc length along a yarn. Therefore, the relationship between
the x coordinate and the yarn angle is:
m cos ψ dψ = −Vxdx (2.8)
Due to symmetry considerations it is assumed that the boundary condition at x = 0 is

ρ = 0. Therefore, integrating equation (2.8) we get:
2m
V= (sin θ − sin ψ ) (2.9)
x2
which gives a relationship between the force and the yarn geometry. At the yarn inter-
section, x = p/2, and ψ = 0, therefore:
8m sin θ
V= (2.10)
p2
To obtain the shape of the elastica, we also need the relationship between y and x:
dy = tan ψ dx (2.11)
The shape is determined by V, m, and θ . Equations for x, y, and s still need to be

derived. An equation for x can be obtained by rearranging equation (2.9):

2m
x= sin θ − sin ψ (2.12)
V
Equations for y and s (arclength) are derived in Appendix B.1. The final results are:

m π π
y= (F( ) − 2E( ) − F(φ0 ) + 2E(φ0 )) (2.13)
V 2 2

m π
s= (F( ) − F(φ0 )) (2.14)
V 2
where
v
E(v) = 1 − k2 sin2 φdφ (2.15)
0
v
1
F(v) = dφ (2.16)
0
1 − k2 sin2 φ
θ π
k = sin( + ) (2.17)
2 4
ψ
sin( 2 + π4 )
sin φ = (2.18)
k
with φ = φ0 when ψ = 0 so that sin φ0 = √1

2k
Note that this model was extended by Leaf and Anandjiwala [LA85] to include a bi-
linear bending model for the yarn.
ψ
s V
θ
x
Figure 2.3: Forces on Olofsson’s elastica [HGB69]

2.2. TENSION 12
2.2 Tension
Consideration of the tensile properties of woven fabrics involves a number of different

problems because the cloth is orthotropic and its modulus varies considerably with
strain. Not only does the modulus vary in the two principal directions (warp and weft)
but extension in the 45 degree direction (the bias) involves a different mechanism than
extension in the principal directions. In addition, extension along the bias is usually
of a higher order of magnitude. The mechanism for extension in the bias direction is
determined by the shear behaviour.
The main mechanisms for extension in the principal directions [HGB69] are: crimp
redistribution (for the initial extension), fiber extension and yarn compression (after
decrimping). Generally, there are three stages to the extension mechanism (see Figure
2.4). First, a high initial modulus is observed until the frictional resistance to the yarn
bending is overcome. Next, a low modulus occurs while the yarns unbend. As the
crimp is decreased, the force needed to cause fibre extension increases. In the final stage,
the load extension behaviour of the fabric is governed by the extension properties of
the yarns.
One of the earliest models for fabric tension was done by Grosberg [Gro66a] who
developed a micromechanical model similar to Olofsson’s [HGB69]. Grosberg included
a tensile force U in addition to the crossing yarn force V (see Figure 2.5) and calculated
the change in yarn spacing for a given extension.
Another micromechanical model is that of Realff et al. [RBB97] which uses a simple
unit cell and derives its model from the constitutive yarn properties. Sun et al. [SSG97]
also developed a micromechanical model but their unit cell geometry allows for more
complex weaves than that of Realff et al., but uses a simpler model of the yarn mechan-
ics.
An alternative approach for the modelling of fabric tension was taken by Sinoimeri
2.2. TENSION 13
Load
Yarn
Extension
Decrimping
Inter-fibre
friction
Extension
Figure 2.4: Generalised Tensile Load-Extension Curve [HGB69]
θ V
V
Figure 2.5: Grosberg’s micromechanical model [Gro66a]

2.3. SHEAR 14
and Dréan [SD97] using the energy methods of de Jong and Postle [DJP78] where con-
strained minimisation of the strain energy of the fabric structure is performed using
optimal control theory.
None of the the research discussed previously in this section has tried to include the
recovery process or examine how their models behave during the recovery process. In
addition, the micromechanical models are limited in their application to general fabric
mechanical behaviour (e.g., drape) since they do not consider interactions between fab-
ric properties. Energy methods are difficult to apply to complex deformations because
of they require new derivations for each loading condition.
2.3 Shear
Modes of deformation involve several forms depending on the degree of shear imposed
upon the fabric. These modes are (from Grosberg and Park [GP66]): Deformation due
to rigid intersections, when the shear cannot overcome the friction. Once the friction is
overcome, the yarns begin to slip at the intersection. Next, an elastic deformation of the
yarns occurs. Finally, jamming occurs in the fabric structure.
The mechanics of shear are highly dependent upon the geometric construction of the
fabric. For example, tightly woven fabric close to the jammed condition and will behave
elastically, while loosely woven fabric behaviour is more dependent on the frictional
resistance between the yarns (see Lindberg et al. [LBD61])
It is generally assumed that the hysteresis produced during shear is the result of
the frictional forces at the yarn intersections [GP66, LBD61, Ske76]. All of the models
in these papers assume a frictional resistance to the deformation proportional to the
normal force acting on the intersection.
2.4. BENDING 15
2.4 Bending
It has been difficult to characterise the mechanisms of fabric bending. Abbott [Abb51]
compares five different labratory tests for bending stiffness. These are: cantilever test,
heart loop test, Schiefer Flexometer, Planoflex, and the M.I.T. Drapemeter to subjective
measurements of stiffness. The cantilever test was preferred over the other methods
due to the simplicity of the test and the high correlation to the subjective measurements.
Grosberg and Abbott [GA66a] discuss the apparatus of Livesey and Owen that bends
the fabric in almost constant curvature. They also discuss [GA66b] the importance of
friction during the bending process and note that large errors are present if a linear
bending approximation is used.
One of the initial models of fabric bending was done by Grosberg and Swani
[Gro66b, GS66] which modelled bending with an initial frictional restraint Mo that must
be first overcome then a linear moment-curvature relationship. The Mo represents fric-
tional resistance to bending at the yarn intersections. Once this frictional resistance is
overcome, the yarns can be bent, hence the linear relationship once the initial resistance
is overcome.
One of the more developed models of fabric bending was done by Zhou and Ghosh
[ZG99] which represented the bending behaviour with a piecewise linear model that
includes hysteresis (see Figure 2.6). This allowed the authors to include cyclic bending
behaviour in their measurement system.
The work of Shi et al. [SHY00] used a rheological model of fabric bending that
includes a spring and frictional elements that are strictly functions of the curvature (see
Figure 2.7).
Other investigations of bending include the work of Hu and Chung [HC00b] which
examined the effect of vertical seams on bending stiffness (using a linear bending
model) and the work of Hu et al. [HLL00] which examined the bending hysteresis
2.4. BENDING 16
Moment, M (gf.cm/cm)
Mo
O
Curvature, K (1/cm)
Figure 2.6: Zhou and Ghosh Bending Model [ZG99]
M(κ) Mf (κ)
Figure 2.7: Shi et al. Bending Model [SHY00]

2.5. KAWABATA EVALUATION SYSTEM 17
in various directions.
2.5 Kawabata Evaluation System
The Kawabata Evaluation System (KES) [Kaw80] is a widely used system of fabric me-
chanical tests that are designed to provide an objective measurement of fabric hand.
Unfortunately, the characterisation of hand does not yield data which is easily exploited
for an engineering characterisation of the material. However, since this system is used
for quality control of fabrics the KES is common. The various KES properties and the
associated testing procedures are described in the following sections.
2.5.1 Tension
The tensile force is applied to a 5 cm by 20 cm fabric specimen as shown in Figure

2.8. The standard [Kaw80] states that the test is conducted to a maximum loading
level of 500 gf/cm . It also states that the test is conducted at a constant strain rate of
4.0 × 10−3 /sec.
ε
20 cm εT = 0
5 cm
Figure 2.8: KES: Application of Tension Force [Kaw80]
The system measures both the loading and unloading responses. Three values are
used to characterise the tension response. Linearity (LT), Tensile Energy per unit area
(WT) (see Figure 2.9), and Resilience (RT) where they are defined as:
m
WT = Fd (2.19)
0
LT = 2WT /(Fm m ) (2.20)
RT = WT /WT (2.21)
where WT is the WT value integrated over the unloading response as shown in Figure
2.9.
500 Fm 500
Fm
F, gf/cm
F, gf/cm
WT
WT‘
ε εm ε εm
Figure 2.9: KES: Tension Measurements

2.5.2 Shear
Shear tests are conducted with a constant tension W applied in the direction orthogonal
to the test (see Figure 2.10). The test is conducted at a velocity of 0.417 mm/s which gives
a rate of shear strain of approximately 8.34 × 10−3 s−1 . In addition, the characteristic
values of G (shear stiffness), 2HG (hysteresis at φ = 0.5◦ ), and 2HG5 (hysteresis at
φ = 5◦ ) are calculated. See Figure 2.11 for the relationship between the recorded curve
and the characteristic values. G is defined as:
Fs
G= (2.22)
φ
which is the slope of the recorded curve. This is measured in the region from 0.5◦ to
5◦ . The standard also states that in the event that the curve is not linear, the mean slope
should be used. Note that this definition is not the same as shear modulus, and as
a result, Hu and Zhang [HZ97] performed finite element analysis of the process and
determined a conversion from the G given by the KES to a shear modulus.
W = const
Fs
φ
20 cm
5 cm
Figure 2.10: KES: Shear Process [Kaw80]

G
Fs , gf/cm
2HG5
2HG
0.5 5
φ, degrees
Figure 2.11: KES: Shear Measurements [Kaw80]
2.5.3 Bending
The bending moment is measured as a sample of fabric is bent through a range of cur-
vatures between 2.5 and −2.5 cm−1 . Kawabata [Kaw80] states that the KES bending test
is conducted at a constant rate of 0.5 cm−1 /s. However, acceleration transition regions
are necessary to achieve that speed from rest and back. No discussion of these transi-
tion regions is made in Kawabata [Kaw80]. The test sample is mounted vertically to
prevent the effect of gravity influencing the experiment. The system measures forward
and backwards bending as shown in Figure 2.12. The characteristic values recorded
by the system are: B, the bending rigidity per unit length, and 2HB, the moment of
hysteresis per unit length. Note that in addition to measuring these values for the weft
and warp directions, they are also measured in the forward and backward directions
as shown in Figure 2.12.
M, gf.cm/cm
Bf
2HBf
-2.5 0 0.5 1.5 2.5

2HBb
K, cm-1
Bb
Figure 2.12: KES: Bending Measurements [Kaw80]

Chapter 3
Previous Work on Hysteresis
In order to represent the hysteresis exhibited during fabric mechanical tests, it is first
necessary to understand hysteresis and how to model it. To that end, this chapter dis-
cusses a number of different models of hysteresis, some specific to mechanical systems,
others more general. Finally, one model of friction is discussed in-depth in Section 3.3.
3.1 Early Mechanical Material Hysteresis Models
Some of the earliest work in this area was done by Ramberg and Osgood [RO43], who
developed a three parameter model of stress-strain behaviour. The results are similar
to those of Dahl [Dah76] (who modelled dynamic friction) and Martin et al. [MTS71]
(whose model is used in fatigue research).
Martin, Topper, and Sinclair’s paper [MTS71] describes a hysteresis model used in
fatigue research. This paper details a simple, extensible model involving only springs
and frictional elements (Figure 3.1) which can provide a good approximation to the ex-
perimental stress-strain curve (see Figure 3.2). Each of the slider elements are set with
a yield value of σi+1 > σi such that the corresponding springs are added in sequence.
22
3.1. EARLY MECHANICAL MATERIAL HYSTERESIS MODELS 23
Upon unloading each of the frictional elements lock up, and retain a residual stress.
However, in order for the residual stress to be retained, σmax (the maximum stress ex-
perienced during the loading) must be less than twice the stress value at which the
slider is activated. While this is true in the case of metals where the model is typically
used, this is not true of fabrics.
En E2 E1 E
... F
σn σ2 σ1
Figure 3.1: Fatigue Mechanical Model [MTS71]
δε*i
σ i* δσi*
ith segment
ε*i ε
Figure 3.2: Piecewise Linear Approximation to Stress-Strain Curve [MTS71]

3.2. GENERAL HYSTERESIS MODELS 24
3.2 General Hysteresis Models
A general mathematical framework for modelling hysteresis was presented by Kras-

nosel’skii and Pokrovskii [KP89]. They also describe lumped parameter models for
systems composed of plays (Figure 3.3), stops (Figure 3.4), and relays (Note that Mo
in Figures 3.3 and 3.4 indicates the initial condition of each system respectively). The
combination of these elements can represent a wide variety of hysteretic systems. In
addition, in the English translation [KP89] a section detailing graph-theoretic represen-
tations of the lumped parameter models is presented.
Mo
x
h
x+
F=
x
F=
Figure 3.3: Ordinary Play [KP89]
Another type of hysteresis model is the differential automata model proposed by

Tavernini [Tav93]. This type of model discretises the continuous problem into a set
of continuous subproblems with possibly discontinuous transitions. The transitions
3.3. A DIFFERENTIAL MODEL OF FRICTIONAL HYSTERESIS 25
h
Mo
x
-h
Figure 3.4: Ordinary Stop [KP89]
between the subproblems are described as a finite state machine where each of the
subproblems is represented by a state.
3.3 A Differential Model of Frictional Hysteresis
Bliman and Sorine [BS93a, BS93b, BS95] detailed a differential equation model for hys-
teresis arising from friction. Their model is capable of producing a subset of Dahl’s
friction models [Dah76] but is also able to model other types of friction as well. Note
that this model is built upon the hysteresis framework of Krasnosel’skii and Pokrovskii
[KP89].
The general form of their model is given in Equation (3.1) where x is a state vector
and u is the displacement from rest.
ẋ = |u̇| Ax + Bu̇, x(0) = 0 (3.1)
F(u)(t) = Cx(t) (3.2)
For a form that reproduces a subset of Dahl’s model they use the following substi-
tutions [BS95] for A,B, and C. Note: This corrects an omission in their earlier paper
[BS93a] and is different than the parameter values used in [BS93b].
1 f1
A=− , B= f , C=1 (3.3)
f
However, this model does not include static friction. To include both static and
dynamic friction we need a second order model. For this model we perform the sub-
stitutions given in Equation (3.4). Note: These substitutions also come from [BS95] and
are also different than those given in [BS93b].
   
1
1  η 0
f1
 η
, C = 1 1
A=− , B= 1
f (3.4)
f 0 1 − f2
If we make the change of variables ds = |u̇|dt in Equation (3.1) we get the Linear
Space Invariant (LSI) system:
dxs
ds = Axs + Bus , x(0) = 0 (3.5)
ys = Cxs (3.6)
From this point on we will only concern ourselves with the combined dynamic and
static friction model.
3.3.1 Physical Parameters and Their Model Equivalent
For the above model, Bliman and Sorine [BS95] defined the following relationships
between physical parameters and the model (see Figure 3.5). The dynamic friction f k is
defined as the friction asymptote when u̇ > 0:
f k = lim F(u)(t) (3.7)

u̇>0,u(t)
and s p is the displacement above which F(u)(t) is within 5% of f k . For the systems of
concern, the static friction is given as:
f s = sup F(u)(t) (3.8)

u , t >0
and se is the displacement value where this value of f s is reached. The minimal and
maximal slopes of the F versus u curves are given as:
F(u)(t)
k F ± = sup ± (3.9)
u , t >0 u̇(t)
These parameters, as well as the other parameters in Figure 3.5, are given in terms of
the model parameters in [BS95]. However, in re-deriving these results (see Appendix
B.2) some errors were discovered in the original paper. The new results are:
fk = f1 − f2 (3.10)

1−η
η
η f2
fs = fk + f2 (1 − η ) (3.11)
f1

f η f1
se = log (3.12)
1−η η f2
sp = 3 f (3.13)

η
f 2 η 2 f 2 1−η
kF− = (1 − η ) (3.14)
f f1
f − f2 η
kF+ = 1 (3.15)
η f
Note: The k F + result given in the original paper (with the factor of 2 removed from the
appropriate equations) corresponds to the difference between the actual k F + and k F −
given here. Also of note is that the k F + given here occurs at t = 0.
F
fs
.
u>0
fk
se sp u
.
u<0
Figure 3.5: Bliman and Sorine 2nd order model

3.3.2 Other Results
Bliman and Sorine [BS95] also give a few other results (also derived in Appendix B.2).
If u and u̇ are integrable, F(u) is given by:
t t
|u̇(τ )|dτ
F(u)(t) = Ce A t Bu̇(t )dt (3.16)
0
If s(0) = 0, making the substitutions for A, B, and C we can reduce the above to (see
Appendix B.2:

− s(t)
− s(t)η
F(u)(t) = sign(u̇(t))( f 1 − f 2 ) + sign(u̇(0)) e f f2 − e f f1 (3.17)
So, the only necessary components are the total distance covered s(t), the sign of the
displacement rate at the current time and at the start of the interval.
Finally, in order to ensure that the system is dissipative Bliman and Sorine [BS95]
give the following conditions:
f1 > f2 ≥0 (3.18)
f > 0 and 0 < η < 1 (3.19)

Chapter 4
A Model for Fabric Hysteresis
From the discussion of fabric bending in Chapter 2, it is clear that there are a number of
mechanisms that influence the bending behaviour. From Kawabata [Kaw80] hysteresis
is exhibited in the results from the KES. So, any model that purports to accurately repro-
duce the bending behaviour should contain elements that can model hysteretic effects.
There has been little work in modelling the hysteresis shown in fabric mechanics. As
a result, the current techniques used in fabric simulations are unable to capture all (or
even most) of the characteristics of the mechanical property curves. To remedy this it is
necessary to borrow modelling techniques from other fields.
This chapter focuses on a modification to the friction modelling technique by Bli-
man and Sorine [BS93a, BS93b, BS95] to model the bending behaviour of fabrics. The
specifics of the model as well as investigations in fitting parameters to experiments are
detailed in this chapter.
30
4.1. THE MODEL 31
k1κ +k 3κ 3
Mext
d
εf ,η,f1,f2
Figure 4.1: Model of Fabric Bending
4.1 The Model
A simple model of linear elastic bending is a spring element. In order to capture rate-
dependent effects, a damper element is also necessary. Lindberg et al.[LBD61] intro-
duced both friction and damper elements in their load-deformation model of fabric.
The shape of the initial bending curve indicates that there is likely more than a linear
viscoelastic behaviour so we will also include a non-linear cubic spring element in the
model. Finally, we include a frictional element to model the hysteresis exhibited in the
experiment. This model is depicted in Figure 4.1. The equation for the model is given
in Equation (4.1) below. In Section 4.3, we investigate the effect of the different model
components.
mκ̈ = k1 κ + dκ̇ + k3 κ3 + M f (κ) + Mext (4.1)
To model the frictional hysteresis produced during bending we use the second order
model of Bliman and Sorine [BS95] described in Section 3.3. We have two forms of
the frictional model that we can use. First, we have the differential equation form as
represented in Equation (3.1). Second, we have the integral form as shown in Equation
4.2. DETERMINING MODEL PARAMETERS FROM EXPERIMENTS 32
(3.16). The first form is useful if the displacement history is to be calculated, the second
is useful if the displacement history is known. If we assume (or know) that there is
no initial displacement from rest, we can use Equation (3.17) to model the frictional
bending restraint. This form is useful for our purposes because the displacement is
controlled during the KES-FB test. Converting the notation into the moment-curvature
form we get:

s(t)
− − s(t)
M f (κ)(t) = ( f 1 − f 2 )sign(κ̇(t)) + sign(κ̇(0)) e f f 2 − e f η f 1 (4.2)
where s(t) is the total curvature that the fabric has experienced at time t as given in
Equation (4.3):
t
s(t) = |κ̇|dt (4.3)
0
4.2 Determining Model Parameters from Experiments
Now that the model has been determined, it is necessary to fit the model to experimen-
tal results. If we are trying reproduce the KES bending test we know the curvature
and its corresponding derivatives at each point in time and we are trying to calculate
the correlated external moment. Therefore, looking at a different problem to that in
Equation (4.1), and we can write the new problem as:
Mext = mκ̈ − k1 κ − dκ̇ − k3 κ3 − M f (κ) (4.4)
where M f (κ) is determined by Equation (4.2). As was mentioned in Chapter 2, accel-

eration transition regions are needed between the points of rest and the constant rate
of 0.5 cm−1 /s mentioned in Kawabata [Kaw80]. However, since they are not discussed
in Kawabata, they will have to be estimated. Given that the above model is dependent
upon the acceleration, velocity, and position these transition regions will be important.
4.2.1 Modelling the KES Bending Test Position, Velocity, and

Acceleration Profiles
The KES Bending Test consists of three phases:
1. Bending to a maximum positive curvature K+ ;
2. Bending to a maximum negative curvature K− ; and
3. Bending back to zero curvature.
Each of these phases consists of an initial acceleration to constant velocity, a period of

constant velocity, and deceleration to rest. Thus, we must define the position, velocity
and acceleration profiles for these nine regions. These are shown in Figure 4.2.
Nomenclature
In addition to the nine regions in Figure 4.2, we can also divide the curves into other
categories:
1. Initial Loading Path. Consists of the first three regions.
2. Unloading Path. Consists of region 4 and the portion of region 5 until the mo-
ment axis.
3. Reverse Loading Path. Consists of the remaining portion of region 5 and all of
region 6.
4. Reverse Unloading Path. Consists of the final three regions.
In the following subsections, displacement is used to mean curvature, velocity to

mean curvature rate, and acceleration to mean the second derivative of curvature with
respect to time.
0.03
0.02
2 4
Bending Moment (gf-cm/cm)
0.01
1
9
0
8
5
-0.01
-0.02
6
-0.03
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
Curvature (cm-1 )
Figure 4.2: KES Bending Test Regions

The Initial Acceleration Region (1)
In order to achieve a constant velocity, it is necessary to accelerate up to some maximum

value and then reduce the acceleration back to zero with the integral over that time
frame equal to the desired constant velocity value. However, there are a wide range of
acceleration profiles that acheive that task. The basic constraints are given below:
v(0) = 0 (4.5)
v(T) = 0.5 cm−1 /s (4.6)
a(0) = 0 (4.7)
a(T) = 0 (4.8)
where T is the total time taken to reach constant velocity. The simplest equation that
can meet these constraints is a quadatric polynomial. The chosen polynomial is given
in Equation (4.9). A plot of the acceleration profile with a time period of one second
is given in Figure 4.3. Note that this does not give zero jerk at zero or at the transition
time T. While it may be possible to acheive the above constraints with zero jerk at the
transition points with a spline curve, polynomials of higher degree than quadratic must
have some negative acceleration region in order to meet the zero jerk requirement.
t(t − T)
a1 (t) = −3 (4.9)
T3
The resulting velocity profile that arises is given in Equation (4.10). A plot of this
profile with a time period of one second is given in Figure 4.4.
1 t2 (2t − 3T)
v1 (t) = − (4.10)
2 T3
Acceleration Profile (T=1)

0.8
0.7
0.6
Curvature Acceleration (cm /s )
−1 2
0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)
Figure 4.3: Acceleration Profile (T=1)

Velocity Profile (T=1)

0.5
0.45
0.4
0.35
Curvature Rate (cm−1/s)
0.3
0.25
0.2
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)
Figure 4.4: Velocity Profile (T=1)

The displacement in this region is given by the integral of the velocity profile. The
displacement equation is shown in Equation (4.11).
1 t3 (t − 2T)
d1 (t) = − (4.11)
4 T3
The total displacement in this region is T4 .
The Constant Velocity Region (2)
In this region, we know that the acceleration is zero and that the velocity is 0.5 cm−1 /s.
The displacement equation for this region (including the displacement from the previ-
ous region) is:
t T
d2 (t) = − (4.12)
2 4
The Deceleration Region (3)
The acceleration profile used in this region is the negative of that given in Equation (4.9)
with the substitution of t = t − t2 where t2 is the time where deceleration begins. This
results in Equation (4.13).
(t − t2 )(t − t2 − T)
a3 (t) = 3 (4.13)
T3
Since we know the final displacement in this region from experiment, we can de-
termine the total time for the loading path T f l . Knowing that the deceleration starts at
time T f l − T we can make this substitution for t2 in Equation (4.13) and get:
(t − T f l + T)(t − T f l )
a3 (t) = 3 (4.14)
T3
We know that the initial velocity is Vd and that the change in velocity is given by:
vc (t) = −v1 (t − t2 ) (4.15)
where v1 (t) is the velocity profile in the acceleration region. Therefore, the velocity
profile in the deceleration region is:
1 (t − T f l + T)2 (2t − 2T f l − T)
v3 (t) = + (4.16)
2 2T 3
Using the same approach, the acceleration, velocity, and displacement equations
can be derived for the remaining sections of the test curve. For complete details see
Appendix B.3.
4.2.2 Fitting the Model Parameters to Experiments
Optimisation techiques are used in order to determine the values of the model param-
eters that best approximate the experimental results. To facilitate the optimisation, the
position, velocity, and acceleration for the experiment were pre-calculated. Thus, one
can vectorise the solution for Mext in Equation (4.4). The design variables are f 1 , f 2 , f ,
η , k1 , d, and k3 . It is assumed that T (the acceleration time) is a constant for the test
equipment and will be set a priori.
Since the equipment to perform the experimental tests is not available at this uni-
versity, Deng’s experimental data [Den94] was used. Model fits were attempted for two
types of fabric with varying weaves and yarn types. The types of fabric are given in
Table 4.1. The basic fabric properties are given in Table 4.2.
Quantifying the Fit
In order to quantify the quality of the fit, a least squares criterion was used. The ob-
jective function to minimise is given in Equation (4.17). However, since Mexp is not
Fabric No. Colour Weave Description

Fabric 1 White Satin 100% Polyester. Bleached, dyed, and
pre-shrunk in a resin finish.
Fabric 2 White Twill 100% Cotton. Bleached, dyed, and
pre-shrunk in a pure finish.
Table 4.1: Details of Fabrics Tested [Den94]
Fabric No. Thickness (cm) Weight Density (gm f /cm2 )

Fabric 1 0.0254 0.01715
Fabric 2 0.0489 0.02665
Table 4.2: Fabric Properties [Den94]
defined as a continuous function, fourth order polynomials were fit to the loading, un-
load/reverse loading, and re-loading paths. These were substituted for Mexp for each
of the paths. While the total number of points varied depending upon the experimental
results, the model and polynomial fits to experiment were sampled at intervals of one
millisecond.
R(κ) = ∑(Mmodel (κ) − Mexp (κ))2 (4.17)
A portion of the objective surface for the weft direction of Fabric 1 is shown in Figure
4.5. We can see that this hyperplane has a line of optimal solutions. While this portion
of the surface could be optimised through SQP methods, there may be difficulties with
other parts of the objective surface.
Optimisation Techniques
Initially, the optimisation was attempted using a genetic algorithm implemented in

MATLAB by Ajay Seth [Set00]. A genetic algorithm was used because it is a global
search technique that has the possibility of finding the global optimum. Using this
Portion of Fabric 1 Weft Objective Surface
0.03
0.028
Objective Function Value
0.026
0.024
0.022
0.02
0.018
1
0.9
0.8
0.7
0.2185 0.219
ε 0.6 0.2175 0.218
f 0.2165 0.217
0.216
f1
Figure 4.5: Portion of Fabric 1 Weft Direction Objective Surface

4.3. SIMULATION RESULTS 42
implementation, some difficulties were found. There was considerable difficulty in

describing what constitutes a converged solution. As a result, the genetic algorithm
never converged to the same solution.
Because of the above difficulties, MATLAB’s non-linear least squares solver,
LSQCURVEFIT (part of the Optimization toolbox) was used to find the optimal solu-
tion. Because this algorithm requires an initial guess, the best solution from the genetic
algorithm tests were used as a seed to the solver. In all cases, the results from the
genetic algorithm were improved.
Because of the difficulties in converging to a globally optimal solution with the
earlier methods, a MATLAB interface [Sak00] to publically available simulated an-
nealling code [Ing00] (described in [Ing89],[Ing92],[Ing93], [Ing96]) was also tried.
This approach produced better results than with either the genetic algorithm or
GA/LSQCURVEFIT combination. As such, the optimisation results in the following
section were produced with this technique.
4.3 Simulation Results
To understand some of the difficulties in capturing the behaviour of these fabrics we

present the +κ and −κ slopes and their differences for each fabric and direction. They
are shown in Table 4.3. In all cases, the slope of the moment-curvature graph shows
large differences between the bending directions. In addition, one direction showed a
increase in slope rather than a decrease.
The following cases were considered for each of the fabric types and directions.
1. Complete model including friction and a cubic spring;
2. As above, assuming no slowing to rest at the end of test;
3. Complete model but excluding friction;

Fabric No. and Direction +κ slope −κ slope % Difference

Fabric 1 - Warp 0.322 0.264 18.0 %
Fabric 1 - Weft 0.199 0.125 37.2 %
Fabric 2 - Warp 0.598 0.477 20.2 %
Fabric 2 - Weft 0.145 0.196 -26.0 %
Table 4.3: Directional Slope Comparsion
4. Complete model but excluding the cubic spring;
5. A linear model that excludes both friction and the cubic spring.
4.3.1 Fabric 1 Results
Warp yarn
Weft yarns
Figure 4.6: Warp-float Satin (4x1) Weave Cross-section
Figure 4.6 shows a cross-section of a 4x1 warp-float satin weave while Figure 4.7
shows the pattern. The warp is represented in black while the weft is represented in
white.
Figure 4.7: Warp-float Satin (4x1) Weave Pattern
Parameter Value
f1 6.184 × 10−1 gm f − cm
f2 6.164 × 10−1 gm f − cm
η 9.998 × 10−1
f 7.302 × 10−1 gm f − cm/cm
d 5.793 × 10−3 gm f − cm/s
k1 6.380 × 10−3 gm f − cm
k3 8.656 × 10−4 gm f − cm3
Table 4.4: Fabric 1 - Warp Direction: Model Parameters
Parameter Value
se 10.92 cm−1
sp 2.19 cm−1
fk 0.198 × 10−2 gm f − cm
fs 0.198 × 10−2 gm f − cm
Table 4.5: Fabric 1 - Warp Direction: Supplementary Friction Results

Warp Direction
The final parameters determined through the optimisation procedure are given in Table
4.4. Examining Table 4.5 we see some interesting results. First, the curvature value se
required for reaching the maximum static friction value f s is greater than the value
s p required to reach the dynamic friction threshold f k . Second, there is no difference
between the static and dynamic friction values. These two observations combined,
indicate that only dynamic friction is present.
In Table 4.6 we can quickly compare the quality of the model fit for each of the dif-
ferent test cases. The complete model provides the best fit, as measured by the residual,
out of all the test cases and we can see that the cubic term is very important to the model
as the quality of the fit decreases significantly in the cases where it is not included.
Most figures in the remainder of this chapter show comparisons of three differ-
ent curves. The dash-dotted line is the digitised version of experimental results from
Deng’s thesis [Den94]. The dotted line is constructed from three separate fourth order
polynomial fits to the digitised data. One polynomial is fit to the loading path, another
to the unloading/reverse loading path, and the third to the reverse unloading path.
The solid line shows the model simulation results.
Examining Figure 4.8, we can see that the model provides an excellent fit throughout
most of the experiment. In the initial loading stage, the model provides a better fit
than the fourth order polynomial. The greatest disagreement in the model appears in
the final reloading path. The model result curves inwards due to the slowing of the
curvature rate.
Figure 4.9 shows the result if we assume that there is no final deceleration region.
While the final inward curvature from Figure 4.8 is not presenet, the path continues to
diverge from the experimental results, but nearly completes a cycle. This implies that
this model does not exhibit any hardening or softening behaviour. This behaviour is
also shown with the weft direction and in both directions for the other fabric. As a
result, the plots of those results are not shown.
If we remove the friction term from the model and then try to fit the model to the
experimental data as is shown in Figure 4.10, it is not possible to acheive the final Mext
value. Instead, the model returns to zero external moment. In addition, the model is
unable to fit parts of the initial loading and final reloading paths as closely as in the full
model.
If we remove the cubic term from the model the best result is shown in Figure 4.11.
Without this term, the path followed is generally straight with the curves resulting from
the transition to and from a constant curvature rate.
Lastly, if we consider a linear viscoelastic model, that is a model without the cubic
or the friction term, we get the results shown in Figure 4.12 which is a very poor fit,
with a residual nearly 9x that of the full model.
Test Case Residual

Full Model 1.104 × 10−2
No Deceleration at End 1.545 × 10−2
No Friction Term 2.523 × 10−2
No Cubic Term 5.792 × 10−2
No Friction or Cubic Term 8.977 × 10−2
Table 4.6: Fabric 1 - Warp Direction: Model Comparision
To understand the relative contribution of the frictional component of the model in

Figure 4.8 we will separate the friction component from the other components of the
model resulting in Figures 4.13 and 4.14. From Figure 4.13 we can see that the model ex-
hibits first-order behaviour. That is, no static friction is observed, only dynamic friction.
Also of note is the fact that the maximum frictional moment is two orders of magnitude
Kawabata Bending Test and Simulation for Fabric 1 − Warp

0.03
Model Fit
Piecewise Polynomial Fit
Experimental Data
0.02
Bending Moment (gf−cm/cm)
0.01
−0.01
−0.02
−0.03
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.8: Fabric 1 - Warp Direction Full Model


0.03
Model Fit
Experimental Data
0.02
0.01
−0.01
−0.02
−0.03
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.9: Fabric 1 - Warp Direction (No Slowing at End)


0.03
Model Fit
Experimental Data
0.02
0.01
−0.01
−0.02
−0.03
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.10: Fabric 1 - Warp Direction (No Friction Term)


0.03
Model Fit
Experimental Data
0.02
0.01
−0.01
−0.02
−0.03
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.11: Fabric 1 - Warp Direction (No Cubic Term)


0.03
Model Fit
Experimental Data
0.02
0.01
−0.01
−0.02
−0.03
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.12: Fabric 1 - Warp Direction (No Cubic or Friction Term)

less than the external moment exhibited in Figure 4.14. Yet, the friction plays a key role
in acheiving a good overall fit.
Initially, the inertial term dominates the behaviour of the model. This can be seen
from the similarities in the initial portion of the loading curve (Figures 4.8-4.12), even
without the cubic and friction terms. It is believed that the friction arises due to the
slipping of the yarns with respect to each other. In this case, we have the warp yarn
slipping over the weft yarns which act like ball bearings. This is consistent with the
frictional response given in Figure 4.13. The cubic term appears to arise from the jam-
ming of the yarns and their subsequent compression, although some of the response
may be due to a non-linear yarn bending response.
−4 Kawabata Bending Test and Friction Simulation for Fabric 1 − Warp

x 10
20
15
10
−5
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.13: Fabric 1 Warp - Frictional Contribution

Kawabata Bending Test and Viscoelastic Simulation for Fabric 1 − Warp

0.03
Viscoelastic Model Fit
Experimental Data
0.02
0.01
−0.01
−0.02
−0.03
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.14: Fabric 1 Warp - Non-linear Visco-elastic Contribution

Weft Direction
In contrast to the warp direction, the weft direction for this fabric exhibits more compli-
cated behaviour. First, the maximum bending moment measured is approximately half
that of the warp direction. Second, the reverse bending portion (the negative half the of
graph) has a smaller loop width and slope. This would indicate that the fabric has dif-
ferent parameters for forward and reverse bending. This behaviour is not implemented
in the current model so it cannot accurately capture this behaviour.
Parameter Value
f1 2.173 × 10−1 gm f − cm
f2 2.161 × 10−1 gm f − cm
η 9.876 × 10−1
f 7.316 × 10−1 gm f − cm/cm
d 2.141 × 10−3 gm f − cm/s
k1 4.548 × 10−3 gm f − cm
k3 1.619 × 10−4 gm f − cm3
Table 4.7: Fabric 1 - Weft Direction: Model Parameters
Parameter Value
se 1.07 cm−1
sp 2.19 cm−1
fk 0.127 × 10−2 gm f − cm
fs 0.189 × 10−2 gm f − cm
Table 4.8: Fabric 1 - Weft Direction: Supplementary Friction Results
From Table 4.8 we can see that the maximum frictional moment f s occurs at 1.07
cm−1 (se ) which is approximately halfway through the initial loading path. We only
approach the s p value at the end of the initial loading path. In addition, the static
frictional moment f s is about 6 × 10−4 gm f − cm greater than the dynamic frictional

moment f k .
While none of the models provide an excellent fit, the full model still provides the
best fit. As with the warp direction, the model fits better than the polynomial for the
initial acceleration region. With respect to Figure 4.15, at a curvature of 1 cm−1 the
model underestimates the external moment. On the reverse bending portion of the
experiment, the model overestimates the external moment.
The magnitudes of the linear model parameters (d and k1 ) are roughly half to three-
quarters that of the warp direction (see Table 4.7). However, the magnitude of the
frictional restraint (Table 4.8) is close to that in the warp direction and the non-linear
stiffness k3 is considerably smaller than that in the warp direction. If the non-linear
stiffness is due to jamming of the yarns this behaviour could be due to a greater spacing
between yarns in the warp direction than the weft.
Test Case Residual

Full Model 1.974 × 10−2
No Deceleration at End 2.250 × 10−2
Table 4.9: Fabric 1 - Weft Direction: Model Comparision
As shown in Figure 4.16, if we remove the friction term, we get an excellent fit for the
initial acceleration region, but a poorer overall fit. For the weft direction, the addition
of the friction term contributes more to the overall fit than the cubic term as is shown
in Table 4.9. As expected there is no external moment at the completion of the test.
If the cubic term is omitted from the model, there is little difference in the overall
quality of the fit. This could be expected from examining the parameter k3 parameter
Kawabata Bending Test and Simulation for Fabric 1 − Weft

0.02
Model Fit
Experimental Data
0.015
0.01
0.005
−0.005
−0.01
−0.015
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.15: Fabric 1 - Weft Direction Full Model


0.02
Model Fit
Experimental Data
0.015
0.01
0.005
−0.005
−0.01
−0.015
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.16: Fabric 1 - Weft Direction (No Friction Term)

value in Table 4.7 since it is an order of magnitude smaller than the other parameters.

0.02
Model Fit
Experimental Data
0.015
0.01
0.005
−0.005
−0.01
−0.015
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.17: Fabric 1 - Weft Direction (No Cubic Term)
The difference without the friction and cubic terms for the weft direction is not as
dramatic as in the warp direction case. However, as is shown in Figure 4.18, with
these effects absent the model is unable to capture the subtle curve exhibited during
the unloading portion of the experiment.
Unlike the warp direction, second order friction effects are observed. Examining
Figure 4.19, we see that during initial loading static friction is overcome, but does not

0.02
Model Fit
Experimental Data
0.015
0.01
0.005
−0.005
−0.01
−0.015
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.18: Fabric 1 - Weft Direction (No Cubic or Friction Term)

reach the dynamic friction value during the decay from the static friction peak. Using
the results from Chapter 3, we see that the value of se is 1.071 cm−1 and a value of 2.195
cm−1 for s p . We therefore see the decay to the dynamic friction during the reverse path.
In Figure 4.20 we can see that the loops are roughly half the width of the full model.
In fact, the size of the loop for reverse bending is roughly the size of the experimental
loop. This may indicate that there is little frictional restraint for reverse bending. Al-
ternatively, this may be due to the characteristics of the test as only dynamic friction is
present during the reverse loading path.
−3 Kawabata Bending Test and Friction Simulation for Fabric 1 − Weft

x 10
2.5
2
1.5
0.5
−0.5
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
−1
Curvature (cm )
Figure 4.19: Fabric 1 Weft - Frictional Contribution

Kawabata Bending Test and Viscoelastic Simulation for Fabric 1 − Weft

0.02
Experimental Data
0.015
0.01
0.005
−0.005
−0.01
−0.015
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.20: Fabric 1 Weft - Non-linear Visco-elastic Contribution

4.3.2 Fabric 2 Results
Weft yarn
Warp yarns
Figure 4.21: Twill (1x2) Weave Cross-section
Figure 4.22: Twill (1x2) Weave Pattern

As stated in Table 4.1, this fabric is a twill fabric which means that there is a diagonal
pattern of alternating weft and warp floats. Figure 4.21 shows a cross-section of a 1x2
twill weave while Figure 4.22 shows the pattern. The warp is represented in black
while the weft is represented in white. In addition, this fabric is a 100 % cotton fabric.
From Meredith [Mer59], we know that cotton fibres during tensile tests show hysteresis
and that under cyclic loading conditions the slope of the curve (the modulus) becomes
progressively steeper from one step to the next. These observations combined indicate
that there are more factors than in the previous fabric, to consider in understanding the
fabric’s behaviour.
Warp Direction
Parameter Value
f1 7.836 × 10−1 gm f − cm
f2 7.810 × 10−1 gm f − cm
η 9.949 × 10−1
f 7.207 × 10−1 gm f − cm/cm
d 1.474 × 10−2 gm f − cm/s
k1 1.154 × 10−2 gm f − cm
k3 1.092 × 10−3 gm f − cm3
Table 4.10: Fabric 2 - Warp Direction: Model Parameters
Parameter Value
se 1.17 cm−1
sp 2.16 cm−1
fk 0.255 × 10−2 gm f − cm
fs 0.334 × 10−2 gm f − cm
Table 4.11: Fabric 2 - Warp Direction: Supplementary Friction Results

As with the weft direction of fabric 1, from Table 4.11 the maximum frictional mo-
ment occurs about half way through the initial loading path. In addition, the static and
dynamic frictional moments are much greater than those exhibited by fabric 1.
Test Case Residual

Full Model 1.067 × 10−1
No Slow Down at End 1.200 × 10−1
Table 4.12: Fabric 2 - Warp Direction: Model Comparision
Figure 4.23 shows that the model slightly underestimates the maximum loading
moment. In addition, the loading slope of the model is slightly less than the loading
slope of the experiment. However, the model does show a good fit for the the reverse
loading portion from 0 cm−1 through −1.6 cm−1 . There is also a good fit for the region
from −2.25 cm−1 to −1.5 cm−1 on the reverse unloading portion of the test.
As with the other direction (and fabric), when we omit the friction term, as shown
in Figure 4.24, the model does not exhibit a final moment since both κ and κ̇ go to
zero. Therefore, only the friction term contributes to the final moment. The model has
a poorer fit on the initial loading curve and the final deceleration region.
If we omit the cubic term in the model (Figure 4.25), we have an excellent fit through
the reverse loading/unloading portion of the test. However, the model overestimates
the moment throughout most of the initial loading and unloading. If both of the fric-
tional and cubic terms are removed (Figure 4.26), the model is unable to provide a good
fit except during the initial loading regions.
The experimental curve is linear for most of the loading path which may result
from the close packing acting more like a continuous material. However, the cubic

0.05
Model Fit
0.04 Experimental Data
0.03
0.02
0.01
−0.01
−0.02
−0.03
−0.04
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.23: Fabric 2 - Warp Direction Full Model


0.05
Model Fit
0.03
0.02
0.01
−0.01
−0.02
−0.03
−0.04
−0.05
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.24: Fabric 2 - Warp Direction (No Friction Term)

term appears to play an important role during the unloading path. The reasons for this
are not clear, but it is likely due to the recovery behaviour of the cotton yarns.

0.05
Model Fit
0.03
0.02
0.01
−0.01
−0.02
−0.03
−0.04
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.25: Fabric 2 - Warp Direction (No Cubic Term)
Like the weft direction of Fabric 1, we see the participation of both static and dy-
namic friction in the model. Figure 4.27 shows the distinct static friction peak ( f s in
Table 4.11) followed by the decay to the dynamic friction value ( f k in the same table).
However, as mentioned earlier, we see larger values for static and dynamic friction
moments. This is not suprising as the experimental moments are greater than that of

0.05
Model Fit
0.03
0.02
0.01
−0.01
−0.02
−0.03
−0.04
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.26: Fabric 2 - Warp Direction (No Cubic or Friction Term)

Fabric 1. In addition, there is a larger separation between the static and dynamic friction
values.
−3 Kawabata Bending Test and Friction Simulation for Fabric 2 − Warp

x 10
4
3.5
3
2.5
1.5
0.5
−0.5
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.27: Fabric 2 Warp - Frictional Contribution

Kawabata Bending Test and Viscoelastic Simulation for Fabric 2 − Warp

0.05
0.03
0.02
0.01
−0.01
−0.02
−0.03
−0.04
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.28: Fabric 2 Warp - Non-linear Visco-elastic Contribution

Weft Direction
Examining Table 4.14, we can see that we have very little dynamic friction, and very
small curvatures to reach the maximum static friction se and to get within 5% of the
dynamic friction value. This indicates that the model is approximating Coloumb static
friction. This is verified by examing the friction contribution to the model in Figure
4.33. This would lead one to conclude that static friction does not play a large role in
the overall fit of the model. From Table 4.15 it appears that alone it does not, but it does
appear to couple with the cubic term in providing a better fit.
Parameter Value
f1 1.600 × 10−1 gm f − cm
f2 1.600 × 10−1 gm f − cm
η 9.932 × 10−1
f 7.646 × 10−4 gm f − cm/cm
d 5.637 × 10−3 gm f − cm/s
k1 3.722 × 10−3 gm f − cm
k3 5.436 × 10−4 gm f − cm3
Table 4.13: Fabric 2 - Weft Direction: Model Parameters
Parameter Value
se 7.62 × 10−2 cm−1
sp 2.29 × 10−1 cm−1
fk 2.120 × 10−6 gm f − cm
fs 4.045 × 10−2 gm f − cm
Table 4.14: Fabric 2 - Weft Direction: Supplementary Friction Results
Figure 4.29 shows the kind of compromises the model fit makes. Although the fit
is poor on the initial loading path, the model provides a good fit through parts of the
unloading/reverse loading path. It also predicts the correct shape of this path as well
Test Case Residual

Full Model 4.479 × 10−2
No Slow Down at End 4.486 × 10−2
Table 4.15: Fabric 2 - Weft Direction: Model Comparision
as the right shape for most of the reverse unloading path. Note that from Table 4.3, this
is the only direction for the two fabrics that has a increase in slope from +κ to −κ.
Without the friction term in the model (Figure 4.30), there is very little change in
the overall fit which is confirmed by the residual values in Table 4.15. Tight packing of
the warp threads may be preventing the movement of the yarns with respect to each
other. Without the cubic term (Figure 4.31) we can see noticeable differences in the
shape of the curve. As with the other fabric and direction, removing the cubic term
eliminates the inward sloping curve on the unloading/reverse loading path. Without
both the cubic and frictional terms (Figure 4.32) we have a poor fit throughout most of
the experiment.
Although the type of twill weave is not mentioned in Deng’s thesis [Den94] the large
asymmetry in the material behaviour suggests that this is a 1x2 twill weave. On load-
ing, the 2 warp threads on the back of the fabric move apart while only 1 is compressed.
On reverse loading, the 2 warp threads will be pressed together causing a jammed con-
dition. This would account for the large non-linearity in the reverse loading direction.

0.015
Model Fit
Experimental Data
0.01
0.005
−0.005
−0.01
−0.015
−0.02
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.29: Fabric 2 - Weft Direction Full Model


0.015
Model Fit
Experimental Data
0.01
0.005
−0.005
−0.01
−0.015
−0.02
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.30: Fabric 2 - Weft Direction (No Friction Term)


0.015
Model Fit
Experimental Data
0.01
0.005
−0.005
−0.01
−0.015
−0.02
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.31: Fabric 2 - Weft Direction (No Cubic Term)


0.015
Model Fit
Experimental Data
0.01
0.005
−0.005
−0.01
−0.015
−0.02
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.32: Fabric 2 - Weft Direction (No Cubic or Friction Term)

−4 Kawabata Bending Test and Friction Simulation for Fabric 2 − Weft

x 10
5
2
−1
−2
−3
−4
−5
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.33: Fabric 2 Weft - Frictional Contribution

Kawabata Bending Test and Viscoelastic Simulation for Fabric 2 − Weft

0.015
Experimental Data
0.01
0.005
−0.005
−0.01
−0.015
−0.02
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Curvature (cm−1)
Figure 4.34: Fabric 2 Weft - Non-linear Visco-elastic Contribution

Chapter 5
The Fabric Drape Simulation Model
To demonstrate the utility of the constitutive model developed in Chapter 4, it will be

used in a simulation of fabric drape. The model from the previous chapter defined the
bending of a fabric using the equation:
mκ̈ = k1 κ + dκ̇ + k3 κ3 + f (κ̇) + Mext (5.1)
where f (κ̇) is the friction function and Mext is the external moment exerted on the body.
It is necessary to put this constitutive model into a form that can be discretised by the
FEM (or some other Ritz-based method) such that the deflections can be obtained.
The layout of this chapter is as follows: Section 5.1 of this chapter presents the dis-
cretisation method; Section 5.2 presents a model with the linear term of Equation (5.1)
in one dimension; Section 5.3 extends the model to a plate; Section 5.4 adds the cubic
term to the simulation model; Section 5.5 includes the frictional term in the simulation
model; and lastly, Section 5.6 the describes the problems in implementing the model.
80
5.1. THE B-SPLINE FIELD APPROXIMATION METHOD 81
5.1 The B-spline Field Approximation Method
The B-spline field approximation model (FAM) used in this thesis is a technique de-
veloped by Allan Vermeulen in his PhD thesis [Ver95]. In the B-spline FAM, the field
variables u are discretised using B-splines, rather than the more traditional Lagrange
polynomials.
The governing variational principle for structural dynamics is Hamilton’s Principle:
t2
δ (T(u̇) − U(u) + W(u))dt = 0 (5.2)
t1
where T(u̇) is the kinetic energy of the structure, U(u) is the strain energy, and W(u) is
the work due to conservative forces. Non-conservative forces must be handled sepa-
rately as variational methods cannot be used for non-conservative forces. If we discre-
tise the field variables using B-splines we get:
n
û = ∑ Ni ui (5.3)
i=1
where Ni are the basis functions and ui are the control vertices (CVs) for the discretisa-
tion. Substituting this into Equation (5.2) gives:
t2
δ ˙ − U(û) + W(û))dt = 0
(T(û) (5.4)
t1
˙ gives rise to the mass matrix M, U(û) leads to the stiffness

From this equation, T(û)
matrix K, and W(û) the consistent force vector F. During the simulation model devel-
opment, in subsequent sections, each of the above will be handled separately.
5.1.1 Properties of the B-spline discretisation
The B-spline discretisation has a number of useful properties:

5.2. A LINEAR APPROXIMATION: PURE BENDING OF A BEAM 82
1. Convex Hull Property: The degree m spline function is bounded by m + 1 CVs at

each point. Therefore, the CVs approximate the curve and provide a minimum
and maximum for the curve at each point.
2. Variation Diminishing Property. The spline crosses the line y = c for some con-
stant c, no more often than the polyline formed by the CVs.
3. Local Continuity Control. It is possible to control continuity at each point. This

allows one to have regions (or points) of different continuity than the rest of the
domain.
4. Numerically Stable. B-splines have a evaluation scheme that only requires convex
combinations of the CVs which has little computational error.
5. Completeness. B-splines are a complete basis for the space of all polynomials of
degree m.
5.2 A Linear Approximation: Pure Bending of a Beam
5.2.1 Formation of the Mass Matrix
The kinetic energy of a structure is given by:

1
T= ρḋT ḋdV (5.5)
2 V
where ḋ is given by:

 
u(x, y, z, t)
∂ 


ḋ =  v(x, y, z, t)  (5.6)
∂t  
w(x, y, z, t)
where u, v, and w the displacements in the x, y, z directions respectively. In the pure

bending case, we are assuming u = 0, v = 0 and w(x, y, z, t) = w(x, t), so in this case ḋ is
∂
ḋ = w(x, t) (5.7)
∂t
Using a separation of variables,
w(x, t) = NT (x)w(t) (5.8)
where N(x) are the basis functions for the discretisation and w(t) are the undetermined
coefficients. Substituting into Equation (5.7) we get:
d
ḋ = NT w (5.9)
dt
Substituting this result into Equation (5.5):

1
T= ρ(ẇT N)(NT ẇ)dV (5.10)
2 V
since ẇ has no spatial dependence, those entries can be moved outside the integral to
get:

1
T = ẇT ρNNT dV ẇ (5.11)
2 V
or,
1
T = ẇT Mẇ (5.12)
2
where,

M=ρ NNT dV (5.13)
V
Taking the variation of T over ẇ leads to a system of equations (with consistent force
vector):
Mẅ = F (5.14)
5.2.2 Formation of the Stiffness Matrix
In general, the strain in the system is of the form:
= 0 + z κ (5.15)
for pure bending, we assume 0 = 0 and with a small deflection assumption,
d2 w
κ=−
dx2
Given that,
h
2
M= zσ dz = k1 κ (5.16)
− 2h
σ is of the form:
σ = E1
= zE1 κ (5.17)
where we need to find E1 in terms of k1 . From M compute
h /2
M = z2 E1 κdz
− h /2
h3
= E1 κ (5.18)
12
For a rectagular cross section (which is what we have) recall
bh3
I=
12
so
E1 I
k1 = (5.19)
b
or
k1 b
E1 = (5.20)
I
Now, the strain energy density is

1
U= σ d = E1 2
0 2
1
= E1 z2 κ2 (5.21)
2
and integration of U over the cross section area (dydz) yields
h /2 b /2
1
Uyz = U dydz = E1 I κ2
− h /2 −b /2 2
1
= k1 b κ2 (5.22)
2
and therefore the total strain energy is
L L
1
U= Uyz dx = (bk1 κ2 )dx (5.23)
0 2 0
For the discretisation introduce a trial function of the form (as was used in Equation
(5.8)):
w(x) = NT (x)w (5.24)
where N is the matrix of basis functions and w is the vector of undetermined parame-
ters. From the definition of κ we find that
κ(x) = N (x)w

T
(5.25)
Substitution into (5.23) yields
L
1
bk1 wT N N wdx
T
U= (5.26)
2 0
Taking the variation over w yields
L
k1 δ wT (N N )wdx
T
δU = b
0
L
k1 (N N )wdx
T
= δ wT b
0
L
k1 (N N )dxw
T
= δ wT b (5.27)
0
This will lead to a system of equations of the form (includes a consistent force vector):
K1 w = F (5.28)
where
L
N N dx
T
K1 = k 1 b (5.29)
0
5.2.3 Formation of the Consistent Force Vector
In order to form the consistent force vector F, we must first calculate the work done by
the applied surface tractions and the body forces. The formula for this is:

W= B ddV + ∑
T
SsT ddA (5.30)
V s Ss
where B is the vector of body forces (e.g., gravity), Ss the vector of surface tractions,
and d a vector of displacements.
Ignoring in-plane displacements and assuming only gravity is acting on the body
(and substituting Equation (5.24)) we get:

W = ρg NT wdV (5.31)
V
Taking the variation of W with respect to w gives us the consistent force vector,

F = ρg NdV (5.32)
V
5.2.4 Application of Boundary Conditions
To apply a forced boundary condition, there are two main techniques:
1. Removal of the degrees of freedom from the system;
2. Modify the system of equations to enforce the constraint.
While the first is more computationally efficient, the second is easier to implement so
the second approach is used. Since the system of equations represents a deflection
from the original configuration, we will only consider the zero deflection boundary
conditions. The process for setting a zero deflection constraint is:
1. Ensure that a CV is set for the location of the boundary condition on the mesh;
2. Set the knot multiplicity for the point equal to the degree of the curve;
3. Zero the row and column corresponding to the CV;
4. Set the diagonal entry equal to -1.0;
5. Set the consistent force vector row equal to zero.

5.2.5 Solution of the System of Equations
From subsections 5.2.1 through 5.2.3 we have developed the system of equations:
Mẅ + K1 w = F (5.33)
we incorporate damping as proportional to the stiffness matrix K1 (Rayleigh damping)

so that the damping matrix D is:
L
N N dx
T
D = db (5.34)
0
where d is the damping coefficient from Equation (5.1). We add this to the system of
equations to get:
Mẅ + Dẇ + K1 w = F (5.35)
To facilitate solution of this system of equations, we reduce the system to first order
form using the technique described by Foss [Fos58]. Equation (5.35) can be written in
reduced form as:
R Ż + Kr Z = Fr (5.36)
where
   
[0] M −M [0]
R=  , Kr =  
M D [0] K1
   
ẇ [0]
Z =   , Fr =  
w F
This is now a system of linear first-order differential equations that can be solved using
standard ODE methods.
5.3. A LINEAR APPROXIMATION: PURE BENDING OF A PLATE 89
5.3 A Linear Approximation: Pure Bending of a Plate
The main difference between the beam and plate models is that we define the discreti-
sation as
w(x, y) = NT (x, y)w (5.37)
where,

N(x, y) = N1x (x)N1y (y) N2x (x)N1y (y) ... Nnx (x)N1y (y) ... Nnx (x)Nny (y)
that is, w(x, y) is discretised using a tensor product of the B-spline basis functions in
the x and y directions. The development of the mass matrix M proceeds in the same
manner as Section 5.2.1 with the only difference being the definition of N. For curvature
κ we are now concerned with curvature in the principal directions κx and κ y . They can
be obtained by taking the second partial derivative with respect to x and y respectively.

Nxx = (x)N (y)
N1x 1y
(x)N (y)
N2x 1y ... (x)N (y)
Nnx 1y ... (x)N (y)
Nnx ny

N yy = (y)
N1x (x)N1y (y)
N2x (x)N1y ... (y)
Nnx (x)N1y ... (y)
Nnx (x)Nny
κx = Nxx w (5.38)
κ y = N yy w (5.39)
If there is no coupling between bending in the prinicipal directions, the stiffness matrix
calculations decouple resulting in,
K1 = K1x + K1y (5.40)

5.4. MODELLING THE NON-LINEAR ELASTIC COMPONENT 90
where,

Nxx Nxx dA
T
K1x = k1x (5.41)
A

Nyy Nyy dA
T
K1y = k1y (5.42)
A
with the damping matrix D calculated in the same manner. We therefore have the
following equations:

M = ρt NNT dA (5.43)
A

Nxx Nxx dA + k1y Nyy Nyy dA
T T
K1 = k1x (5.44)
A A
Nxx Nxx dA + d y Nyy Nyy dA
T T
D = dx (5.45)
A A

F = ρ gt NdA (5.46)
A
The reduction of the system of equations into first order form as done in Section 5.2.5
proceeds in the same manner.
5.4 Modelling the Non-linear Elastic Component
Since we assume separation of the bending behaviour in the principal directions, this
section will develop the behaviour for one direction only, with the recognition that the
same procedure will also apply for the other direction. Now, we have developed a
moment-curvature relation of the form:
M = k1 κ + k3 κ3 (5.47)
Now, the moment can be determined from
h
2
M= zσ dz = k1 κ + k3 κ3 (5.48)
− 2h
We can define σ as an odd polynomial of the form:
σ = E1 + E3 3
= zE1 κ + z3 E3 κ3 (5.49)
Now, since we have handled the κ term in earlier analyses, we shall consider only the
κ3 term for the moment. From M compute:
h
2
M3 = z4 E3 κ3 dz
− 2h
h5
= E3 κ3 (5.50)
80
Using the moment of inertia relationship, we get
20k3 b
E3 = (5.51)
3Ih3
Now, the strain energy density is:

1
U= σ d = E3 4
0 4
1
= E3 z4 κ4 (5.52)
4
and integration of U over the thickness yields:
h /2
1 3E3 Ih2 4
Uz = U dz = κ
− h /2 4 20b
1
= k3 κ4 (5.53)
4
and therefore, the total strain energy for the cubic term is,

1
U3 = Uz dA = k3 κ4 dA (5.54)
A 4 A
Substituting the definition of κ from equation 5.25 we have:

1
Uz dA = k3 (wT N N w)2 dA
T
U3 = (5.55)
A 4 A
Variation over w yields:

k3 δ wT N N w(wT N N w)dx
T T
δ U3 =
A

k3 (N N )(wwT )(N N )wdx
T T
= δ wT
A

k3 (N NT )(wwT )(N N )dxw
T
= δw T
(5.56)
A
which leads to a system of equations of the form:
K1 w + K3 (w)w = F (5.57)
where K1 is the linear stiffness matrix computed earlier and K3 (w) is the non-linear
stiffness matrix.
5.5. MODELLING THE FRICTIONAL COMPONENT 93
5.5 Modelling the Frictional Component
It is necessary to use the integral form of the frictional model from Chapter 4 and not
the differential equation form since it is necessary to be able to integrate over the area.
The virtual work done by the moment M f (κ̇) can be defined as

dW = SsT dκdA (5.58)
Ss
with Ss a vector of M f (κ̇) at each of the CVs. The frictional term is defined as (from
equation 4.2):

− s(t)
− s(t)η
Ss = ( f 1 − f 2 )sign[κ̇(t)] + sign[κ̇(0)] e f f2 − e f f1 (5.59)
where s(t) is the time integral of κ. Substitution of equation 5.25 yields:

s(t)
T T − − s(t)η
f 1 N wdAdw
T
dW = ( f 1 − f 2 )sign(N ẇ(t)) + sign(N ẇ(0)) e f f2 − e f
Ss
(5.60)
In an implementation integral s(t) can be accumulated at each of the CVs, then the
integral over the area can be performed.
5.6 Implementation Details and Problems
The above model was implemented in Fortran 95 using the B-spline software from the
SLATEC library [Law01]. NAG F77 stiff ODE routines [Num01] were used to solve the
system of ODEs.
The following results were found: without in-plane strains, the fabric deformation
was unrealistic; and, the inclusion of the frictional term sigificantly slowed the calcula-
tion time. On a Pentium III 600 Mhz machine running Linux, one millisecond of simu-
5.6. IMPLEMENTATION DETAILS AND PROBLEMS 94
lation time takes approximately five minutes of wall clock time. Without the frictional
term, a full second of simulation time can be completed in five minutes of wall clock
time.
There was considerable difficulty in finding a method of coupling in-plane and out-
of-plane deformation using this bending model. No satisfactory method was found
during the course of this research. As a result, we were unable to accurately model the
drape of fabric and therefore, no results of the simulations are reported.
Chapter 6
Summary
Most of the previous work on fabric mechanics has either used linear models or relied
upon micromechanical models. In Chapter 4 the problems of using linear models for
modelling bending have been explored. Micromechanical models use a unit cell to
represent the behaviour of the fabric which causes difficulty due to the idealisation of
the structure of the fabric. These idealisations do not account for the finish applied to
the fabric which is important in the overall properties [WKL+ 00], or variations in the
weave, either due to imperfections or by design.
6.1 Discussion of the Current Model
From Chapter 4 it is clear that the additional cubic term and frictional term have a sig-
nificant impact on the quality of the model (as measured by the fit to experiment). The
frictional term models the slip of the yarns with respect to each other while the cubic
term models the jamming behaviour and the non-linear yarn bending properties. How-
ever, to accurately determine the model parameters, it is important to have a detailed
understanding of the curvature rate profile throughout the experiment. This will assist
95
6.2. DISCUSSION OF MODELLING FOR SIMULATION PURPOSES 96
in improving the model. From experimental results (Table 4.3) it appears that includ-
ing differing behaviour for forward and reverse bending would assist in improving the
model as well.
6.2 Discussion of Modelling for Simulation Purposes
The model presented herein relates moments and curvature. However, in order to in-
clude it as part of a simulation model, it is necessary to have a displacement relation.
If the experimental results were in terms of displacement, it would faciliate the mod-
elling for simulation purposes. In addition, there is currently no knowledge on how
tension, shear, and bending are coupled. Further research is needed to characterise this
coupling in order to include the model as part of a FE model.
Appendix A
Glossary
Fabric Hand The concept of fabric ”hand” is ill-defined. Fabric ”hand”

is examined mainly by the sense of touch. Fabrics with a
different ”hand” are used for different purposes and may
be judged to have a different quality. Kawabata [Kaw80]
tried to convert this subjective measurement to an objective
measurement. The qualities he considered are: smoothness,
crispness, stiffness, spread (or anti-drape), fullness, softness,
and the appearance of the surface.
Grey Fabric A term used to refer to fabrics that have just left the loom or
knitting machine (from Collier [Col70]). These fabrics usu-
ally undergo a finishing process before being used.
Jammed Condition A geometrical state of the fabric where there is no space be-
tween the yarns.
Micromechanical Model A model where the overall properties of the material are de-
termined by models of the micro-scale interactions.
97
APPENDIX A. GLOSSARY 98
Relaxed Fabric A fabric which has released the stresses and strains occurred
during manufacture.
Satin Satin is a type of weave characterised by long floats (or runs)
on the face of the fabric. These floats are caught under cross
yarns as far apart as possible. Also of note is that at no time
do parallel yarns come in contact with each other. In satin
weaves, the warp yarns float. If the weft yarns float, the
weave is known as sateen. [Jos77]
Thread From Joseph (pp. 218) [Jos77], the term thread indicates “a
product used to join pieces of fabric together to create textile
products. ”
Twill Twill is a type of weave characterised by a diagonal line on
the face and possibly the back of the fabric. The warp yarn
floats (goes over) two weft yarns and under one in a 2/1
twill. In a regular twill, each succeeding float begins one
weft yarn higher or lower than the adjacent float. This cre-
ates the diagonal pattern on the surface of the fabric. Of note
is that the twill weave permits close packing and produces
strong, durable fabrics. Examples of common fabrics wth a
twill weave are denim and flannel.
Warp Yarns that run parallel to the selvage or to the longer dimen-
sion of a bolt of fabric are called warp yarns or ends.
Weft Yarns that run perpendicular to the long direction of the bolt
of fabric. Also yarns in this direction may be referred to as
the woof yarns, filling yarns or picks.
APPENDIX A. GLOSSARY 99
Yarn From Joseph (pp. 203) [Jos77], “a generic term for a contin-
uous strand of textile fibres, filaments or material in a form
suitable for knitting, weaving, or otherwise intertwining to
form a textile fabric. Yarn occurs in the following forms:
• a number of fibres twisted together
• a number of filaments laid together without twist
• a number of filaments laid together with more or less

twist
• a single filament ... monofilament
• One or more strips made by the lengthwise divisions

of a sheet of material such as a natural or synthetic
polymer, a paper, a metal foil, used with or without
twist in a textile construction.”
Appendix B
Further Information
B.1 Equations of an Elastica
Rewriting Equations (2.6) and (2.5) respectively we get:
ds
ρ= (B.1)
dψ
m
= −Vx (B.2)
ρ
Also, we have an equation for x (Equation (2.12)) repeated here for clarity:

2m
x= sin θ − sin ψ (B.3)
V
Substituting Equation (B.3) into Equation (B.2) we get:

m m V 1
ρ=− =− [sin θ − sin ψ ]− 2
Vx V 2m

1 m 1
= −√ [sin θ − sin ψ ]− 2 (B.4)
2 V
100
B.1. EQUATIONS OF AN ELASTICA 101
From Equation (B.1) we have:

ds = ρdψ (B.5)
Using Equation (B.4) in Equation (B.5):

1 m 1
ds = − √ [sin θ − sin ψ ]− 2 dψ (B.6)
2 V

1 m 1
ds = − √ [(1 + sin θ ) − (1 + sin ψ )]− 2 dψ (B.7)
2 V
From trigonometry:
θ π θ π θ π
sin( + ) = sin cos + cos sin
2 2 2 4 2 4
1 θ θ
= √ (sin + cos ) (B.8)
2 2 2
Also, from Esbach’s Handbook (pp. 2-69) [Tap89]:
θ 1√ 1√
sin = 1 + sin θ − 1 − sin θ (B.9)
2 2 2
θ 1√ 1√
cos = 1 + sin θ + 1 − sin θ (B.10)
2 2 2
Combining Equations (B.8), (B.9), and (B.10):
θ π 1√
sin( + ) = 1 + sin θ (B.11)
2 2 2
Defining k = sin( θ2 + π2 ) we have:
2k2 = 1 + sin θ (B.12)
Similarly,
ψ π
1 + sin ψ = 2 sin2 ( + ) (B.13)
2 4
Using Equations (B.12) and (B.13) in B.7:

1 m 2 ψ π 1
ds = − √ [2k − 2 sin2 ( + )]− 2 dψ
2 V 2 4

1 m 1 1 ψ π 1
= −√ √ (1 − 2 sin2 ( + )]− 2 dψ (B.14)
2 V 2k k 2 4
Also, define:
1 ψ π
sin φ = sin( + ) (B.15)
k 2 4
Use Equation (B.15) in (B.14):

1 m1 1
ds = − √ [1 − sin2 φ]− 2 dψ
2 Vk

1 m 1
= −√ dψ (B.16)
2 V k cos φ
Implicitly differentiating Equation (B.15):
1 ψ π
cos φdφ = cos( + )dψ
2k 2 4
dψ 2k
= dφ (B.17)
cos φ cos( ψ2 + π4 )
Similar to the simplifications that were done in Equations (B.8)-(B.10) we can simplify
cos( ψ2 + π4 ):
ψ π ψ ψ
cos( + ) = cos cos π 4 − sin sin π 4
2 4 2 2
1 ψ ψ
= √ (cos − sin )
2 2 2
1
=√ 1 − sin ψ (B.18)
2
From Equation (B.13):

ψ π
sin ψ = sin2 ( + )−1 (B.19)
2 4
and from Equation (B.15), Equation (B.19) can be written as:
sin ψ = 2k2 sin2 φ − 1 (B.20)
Using Equation (B.20) in Equation (B.18) to get

ψ π 1
cos( + ) = √ 1 + 1 − 2k2 sin2 φ
2 4 2
√
2
=√ 1 − k2 sin2 φ
2

= 1 − k2 sin2 φ (B.21)
Using Equation (B.21) in Equation (B.17) such that
dψ 2k
= (B.22)
cosφ
1 − k2 sin2 φ
Using Equation (B.22) in Equation (B.16) produces

1 m1 2kdφ
ds = −
2 Vk
1 − k2 sin2 φ

m dφ
=− (B.23)
V
1 − k2 sin2 φ
Integrating Equation (B.23) such that
s
m φ0 dφ
ds = − (B.24)
0 V π2
1 − k2 sin2 φ
at ψ = θ , from Equation (B.15)
1 θ π
sin φ = sin( + )
k 2 4
1
= k=1
k
π
∴φ=
2
From Equation (B.24)

m π
s= [F( ) − F(φ0 )] (B.25)
V 2
Now deriving the elliptic integral for y, from Equation (2.11) we have:
dy = tanψ dx (B.26)
From Using Equations (2.7) and (B.5) in Equation (B.26):
dy = sin ψ ds = ρ sin ψ dψ (B.27)
Using ρ from Equation (B.4) in Equation (B.27):

1 m 1
dy = − √ [sin θ − sin ψ ]− 2 sin ψ dψ
2 V

1 m 1
= −√ [(1 + sin θ ) − (1 + sin ψ )]− 2 sin ψ dψ (B.28)
2 V
Recall that,
from Equation (B.12) 1 + sin θ = 2k2 (B.29)

ψ π
from Equation (B.13) 1 + sin ψ = 2 sin2 ( + ) (B.30)
2 2
2k cos φ
from Equation (B.22) dψ = dφ (B.31)
1 − k2 sin2 φ
from Equation (B.20) sin φ = 2k2 sin2 φ − 1 (B.32)
Using Equations (B.29) through (B.32) in Equation (B.28):

1 m 2 ψ π 1 2k cos φdφ
dy = − √ [2k − 2 sin2 ( + )]− 2 (2k2 sin2 φ − 1)
2 V 2 4
1 − k2 sin2 φ

m 1 1 k cos φ
=− ( )[1 − sin2 φ]− 2 (2k2 sin2 φ − 1) dφ
V k
1 − k2 sin2 φ

m 1 (1 − 2k2 sin2 φ) cos φ
= dφ
V cos φ
1 − k2 sin2 φ

m 1 − 2k2 sin2 φ
= dφ
V (1 − k2 sin2 φ) 12

m 2(1 − k2 sin2 φ) − 1
= dφ
V 2
1 − k sin φ
2

m m 1
dy = 2 1 − k2 sin2 φdφ − dφ (B.33)
V V
1 − k2 sin2 φ
Integrate Equation (B.33) to get
y φ0 φ0
m m 1
= π
2 1 − k2 sin2 φdφ − π
dφ
0 V V
2 2 1 − k2 sin2 φ
B.2. DERIVATION OF BLIMAN AND SORINE RESULTS 106

m π π
y= [(2E(φ0 ) − 2E( )) + (F( ) − F(φ0 ))] (B.34)
V 2 2
B.2 Derivation of Bliman and Sorine results
This section provides derivations for the Equations (3.10) through (3.12) and Equation
(3.14). We start with the integral form of the system:

+s(t)

− s(0)η − s(0)+ s(t)
F(u)(t) = f 1 1 − e − f2 1 − e (B.35)
Our definition of f k from Equation (3.7) is restated below:
f k = lim F(u)(t) (B.36)

u̇>0,u(t)
which when applied to Equation (B.35) gives:
fk = f1 − f2 (B.37)
The definition of f s is given in Equation (3.8) is:
f s = sup F(u)(t) (B.38)

u , t >0
Differentiate F(u)(t) with respect to time as the first step to finding the maximum:

s(0)+s(t)


− η − s(0)+s(t)
∂ F(u)(t) ∂ f e f
f e f

= s(t)  1 − 2  (B.39)
∂t ∂t f η f
Since s(t) will always be positive, the second term must go to zero at the maximum.
Solving for the value of s(t) that results in that term going to zero (and assuming s(0) =
B.2. DERIVATION OF BLIMAN AND SORINE RESULTS 107
0), results in se :
f η f
se = log 1 (B.40)
1−η η f2
subtituting this into F(u)(t) and simplifying gives:

1−η
η
η f2
fs = fk + f2 (1 − η ) (B.41)
f1
Determining k F ± is slightly more complicated. Starting with k F − we have the following

definition:
F(u)(t)
k F − = sup − (B.42)
u , t >0 u̇(t)
First, we form the ratio:

s(0)−s(t) s(0)−s(t)
f η f
F(u)(t) f e − f2 e η
− =− 1 (B.43)
u̇(t) f η
Differentiating with respect to time and factoring the result:

− s(t)η − s(t)

f1 e f
− f2 e f
η2
∂ F(u)(t) ∂
− = s(t) (B.44)
∂ t u̇(t) ∂t η2 f 2
Which results in the following equation for the maximum:

− s(t)η − s(t)

f1 e f
− f2 e f
η2
=0 (B.45)
η2 f 2
Solving for the value of s(t):

f1
log f2 η2
f η
s− = (B.46)
1−η
B.3. KES BENDING TEST PROFILE EQUATIONS 108
Substituting this result into Equation (B.43) and simplifying:

1−η
η
− f η2 f2
kF = 2 (1 − η ) (B.47)
f f1
B.3 KES Bending Test Profile Equations
This section continues the derivation of the equations given Section 4.2.1 for the remain-
ing six regions of the curve in Figure 4.2.
B.3.1 Unloading/Reverse Bending Path (Regions 4-6)
This path consists of reversing the direction of bending after first unbending the fabric.
It is assumed that the test continues at a constant velocity through zero curvature.
Negative Acceleration (Region 4)
Initially, we have a total displacement d3 f , zero velocity, and zero acceleration. The ac-
celeration profile is given in Equation (B.48) and can be determined by taking the nega-
tive of the acceleration profile given in Equation (4.9) and shifting the time through the
substitution of t = t − T f l where T f l is the time at which the initial loading is complete.
(t − T f l )(t − T f l − T)
a4 (t) = 3 (B.48)
T3
As with region 3, the change in velocity is given by the negative of the velocity
profile in region 1 v0 (t), but in this case timeshifted as above to T f l , resulting in a velocity
as given in Equation (B.49).
(t − T f l )2 (2t − 2T f l − 3T)
v4 (t) = 3 (B.49)
2T 3
B.3. KES BENDING TEST PROFILE EQUATIONS 109
The displacement profile over this region is:
(t − T f l )3 (t − T f l − 2T)
d4 (t) = d3 f + (B.50)
4T 3
Constant Negative Velocity (Region 5)
Since the acceleration and velocity are constant, the displacement profile is easily de-
termined to be:
Tf l + T − t
d5 (t) = d4 f + (B.51)
2
where d4 f is the displacement at the end of region 4.
Deceleration (Region 6)
Given that we are decelerating from a negative velocity, the acceleration and velocity
profiles from region 1 are correct if we timeshift the profile with t = t − (T f u − T) where
T f u is the time where we end this path. Performing this, we get the following equations:
(t − T f u + T)(t − T f u )
a6 (t) = −3 (B.52)
T3
1 (t − T f u + T)2 (2t − 2T f u − T)
v6 (t) = − − 3 (B.53)
2 2T 3
The displacement over this region is given by:
T t (t − T f u + T)3 (t − T f u − T)
d6 (t) = T f l − − − (B.54)
4 2 4T 3
B.4. POLYNOMIAL FIT TO EXPERIMENTAL DATA 110
B.3.2 Re-Loading Bending Path (Regions 7-9)
The procedure for this region is the same as regions 1-3 except the profiles are
timeshifted with t = t − T f u .
B.4 Polynomial Fit to Experimental Data
Power Value
x4 -1.034 ×10−18
x3 -2.744 ×10−5
x2 3.898 ×10−5
x1 1.212 ×10−2
x0 1.716 ×10−18
Table B.1: Fabric 1 - Warp Direction: Loading Polynomial Coefficients
Power Value
x4 2.956 ×10−4
x3 1.022 ×10−3
x2 -7.164 ×10−4
x1 6.131 ×10−3
x0 2.950 ×10−3
Table B.2: Fabric 1 - Warp Direction: Unloading/Reverse Loading Polynomial Coefficients

Power Value
x4 -2.287 ×10−3
x3 -6.208 ×10−3
x2 -6.022 ×10−3
x1 3.605 ×10−3
x0 1.923 ×10−3
Table B.3: Fabric 1 - Warp Direction: Reverse Unloading Polynomial Coefficients
Power Value
x4 -9.201 ×10−19
x3 -2.028 ×10−4
x2 3.779 ×10−18
x1 8.059 ×10−3
x0 -3.219 ×10−18
Table B.4: Fabric 1 - Weft Direction: Loading Polynomial Coefficients
Power Value
x4 2.240 ×10−4
x3 4.910 ×10−4
x2 -4.607 ×10−4
x1 3.165 ×10−3
x0 -1.368 ×10−3
Table B.5: Fabric 1 - Weft Direction: Unloading/Reverse Loading Polynomial Coefficients
Power Value
x4 -4.661 ×10−4
x3 -1.096 ×10−3
x2 -4.812 ×10−4
x1 3.982 ×10−3
x0 8.17 ×10−4
Table B.6: Fabric 1 - Weft Direction: Reverse Unloading Polynomial Coefficients

Power Value
x4 4.697×10−18
x3 -1.114 ×10−3
x2 -9.870 ×10−18
x1 2.558 ×10−2
x0 9.126 ×10−21
Table B.7: Fabric 2 - Warp Direction: Loading Polynomial Coefficients
Power Value
x4 1.193 ×10−3
x3 1.706 ×10−3
x2 -3.808 ×10−3
x1 9.244 ×10−3
x0 -6.849 ×10−3
Table B.8: Fabric 2 - Warp Direction: Unloading/Reverse Loading Polynomial Coefficients
Power Value
x4 -5.485 ×10−3
x3 -1.789 ×10−2
x2 -1.911 ×10−2
x1 3.703 ×10−3
x0 5.482 ×10−3
Table B.9: Fabric 2 - Warp Direction: Reverse Unloading Polynomial Coefficients
Power Value
x4 -7.839 ×10−19
x3 -2.793 ×10−4
x2 2.155 ×10−18
x1 6.759 ×10−3
x0 -9.275 ×10−19
Table B.10: Fabric 2 - Weft Direction: Loading Polynomial Coefficients

Power Value
x4 1.996 ×10−4
x3 4.785 ×10−4
x2 -1.003 ×10−3
x1 4.132 ×10−3
x0 -3.115 ×10−3
Table B.11: Fabric 2 - Weft Direction: Unloading/Reverse Loading Polynomial Coefficients
Power Value
x4 -2.112 ×10−3
x3 -6.842 ×10−3
x2 -8.550 ×10−3
x1 -3.615 ×10−4
x0 1.329 ×10−3
Table B.12: Fabric 2 - Weft Direction: Reverse Unloading Polynomial Coefficients

Bibliography
[Abb51] N.J. Abbott. The Measurement of Stiffness in Textile Fabrics. Textile Re-
search Journal, 21:435–444, June 1951.
[ABW93] M. Aono, D. Breen, and M. Wozny. A Computer Aided Broadcloth Composite
Layout Design System, pages 223–250. North-Holland, 1993.
[ABW94] M. Aono, D. Breen, and M. Wozny. Fitting a Woven Cloth Model to a
Curved Surface: Mapping Algorithms. Computer Aided Design, 26(4):278–
292, April 1994.
[ADBW96] M. Aono, P. Denti, D. Breen, and M. Wozny. Fitting a Woven Cloth to a
Curved Surface: Dart Insertion. IEEE Computer Graphics and Applications,
16(4):60–70, September 1996.
[BHG92] D. Breen, D. House, and P. Getto. A Physically-Based Particle Model of
Woven Cloth. The Visual Computer, 8(5-6):264–277, 1992.
[BHW94] D. Breen, D. House, and M. Wozny. A Particle-based Model for Simulating
the Draping Behaviour of Woven Cloth. Textile Research Journal, 64:663–
685, November 1994.
[BS93a] P.A. Bliman and M. Sorine. Friction modeling by hysteresis operators.
Application to Dahl, stickton, and Stribeck effects. In Models of Hysteresis.
Longman Scientific and Technical, 1993.
[BS93b] P.A. Bliman and M. Sorine. A system-theoretic approach of systems with
hysteresis. application to friction modelling and compensation. In Pro-
ceedings of the 2nd European Control Conference, pages 1844–1849, Gronigen,
Netherlands, 1993.
[BS95] P.A. Bliman and M. Sorine. Easy-to-use realistic dry friction models for
automatic control. In Proceedings of the 3rd European Control Conference,
pages 3788–3794, Rome, Italy, September 1995.
114
BIBLIOGRAPHY 115
[BSBG98] Smita Bais-Singh, Sherrill B. Biggers, Jr., and Bhuvenesh C. Goswami. Fi-
nite Element Modeling of the Nonuniform Deformation of Spun-Bonded
Nonwovens. Textile Research Journal, 68(5):327–342, May 1998.
[BW98a] David Baraff and Andrew Witkin. Faster cloth dynamics. In Cloth and
Clothing in Computer Graphics, number 31 in SIGGRAPH 98 Course Notes,
pages G–1, Orlando, Florida, July 19-24 1998. ACM SIGGRAPH.
[BW98b] David Baraff and Andrew Witkin. Large steps in cloth simulation. In
Proceedings of SIGGRAPH 98, Annual Conference Series, pages 43–54, Or-
lando, Florida, July 19-24 1998. ACM SIGGRAPH.
[CG95] B. Chen and M. Govindaraj. A Physically Based Model of Fabric Drape
Using Flexible Shell Theory. Textile Research Journal, 65:324–330, June 1995.
[Col70] Ann M. Collier. A Handbook of Textiles. Pergamon Press, Oxford, 1970.
[CSWY99] M.X. Chen, Q.P. Sun, Z. Wu, and M.M.F. Yuen. A Discretized Linear Elastic
Model for Cloth Buckling and Drape. Journal of Manufacturing Science and
Engineering, 121:695–700, November 1999.
[CYMTT92] M. Carignan, Y. Yang, N. Magnenat-Thalmann, and D. Thalmann. Dress-
ing Animated Synthetic Actors with Complex Deformable Clothes. Com-
puter Graphics, 26(2):99–104, July 1992.
[Dah76] Phillip R. Dahl. Solid friction damping of mechanical vibrations. AIAA
Journal, 14(12):1675–1682, December 1976.
[Den94] Shigan Deng. Nonlinear Fabric Mechanics Including Material Nonlinearity,
Contact, and an Adaptive Global Solution Algorithm. PhD thesis, North Car-
olina State University, 1994.
[DGR00] José Miguel S. Dias, Manuel N. Gamito, and José M. Rebordão. A Dis-
cretized Linear Elastic Model for Cloth Buckling and Drape. Textile Re-
search Journal, 70(4):285–297, April 2000.
[DJP78] S. De Jong and R. Postle. A General Energy Analysis of Fabric Mechanics
Using Optimal Control Theory. Textile Research Journal, 48:127–135, March
1978.
[EDC96] J. Eischen, S. Deng, and T. Clapp. Finite-Element Modeling and Control of
Flexible Fabric Parts. IEEE Computer Graphics and Applications, 16(4):71–80,
September 1996.
[Eis98a] Jeffrey W. Eischen. Drape Modeling of Cloth. In Cloth and Clothing in
Computer Graphics, number 31 in SIGGRAPH 98 Course Notes, pages E–1
through E–23, Orlando, Florida, July 19-24 1998. ACM SIGGRAPH.
BIBLIOGRAPHY 116
[Eis98b] Jeffrey W. Eischen. Finite Element Modeling of Cloth - Implementation.

In Cloth and Clothing in Computer Graphics, number 31 in SIGGRAPH 98
Course Notes, pages E–24 through E–50, Orlando, Florida, July 19-24 1998.
ACM SIGGRAPH.
[EW97] B. Eberhardt and A. Weber. Modeling the Draping Behaviour of Woven
Cloth. MapleTech, 4(2):25–31, 1997.
[EW99] B. Eberhardt and A. Weber. A particle system approach to knitted textiles.
Computers and Graphics, 23:599–606, 1999.
[EWS96] B. Eberhardt, A. Weber, and W. Strasser. A Fast, Flexible Particle Sys-
tem Model for Cloth Draping. IEEE Computer Graphics and Applications,
16(4):52–59, September 1996.
[Fos58] K.A. Foss. Co-ordinates which uncouple the equations of motion of
damped linear dynamic systems. ASME Journal of Applied Mechanics,
pages 361–364, September 1958.
[GA66a] P. Grosberg and G.M. Abbott. Measurement of Fabric Stiffness and Hys-
teresis in Bending. Textile Research Journal, 36:928–930, October 1966.
[GA66b] P. Grosberg and G.M. Abbott. The Fabric Cantilever. Textile Research Jour-
nal, 36:930–932, October 1966.
[GLS95] L. Gan, N.G. Ly, and G.P. Steven. A Study of Fabric Deformation Using
Nonlinear Finite Elements. Textile Research Journal, 65:660–668, November
1995.
[GP66] P. Grosberg and B.J. Park. The Mechanical Properties of Woven Fabric
- Part V: The Initial Modulus and the Frictional Restraint in Shearing of
Plain Weave Fabrics. Textile Research Journal, 36:420–431, May 1966.
[Gro66a] P. Grosberg. The Mechanical Properties of Woven Fabric - Part I: The Initial
Load Extension Modulus of Woven Fabrics. Textile Research Journal, 36:71–
79, January 1966.
[Gro66b] P. Grosberg. The Mechanical Properties of Woven Fabric - Part II: The
Bending of Woven Fabrics. Textile Research Journal, 36:205–211, March
1966.
[GS66] P. Grosberg and N.M. Swani. The Mechanical Properties of Woven Fabric
- Part IV: The Determination of the Bending Rigidity and Frictional Re-
straint in Woven Fabrics. Textile Research Journal, 36:338–345, May 1966.
[HB98] Donald H. House and David E. Breen. Representation of Woven Fabrics.
In Cloth and Clothing in Computer Graphics, number 31 in SIGGRAPH 98
BIBLIOGRAPHY 117
Course Notes, pages B–1 through B–17, Orlando, Florida, July 19-24 1998.
ACM SIGGRAPH.
[HC98] Jinlian Hu and Yuk-Fung Chan. Effect of Fabric Mechanical Properties on
Drape. Textile Research Journal, 68(1):57–64, January 1998.
[HC00a] Jinlian Hu and Shui-Fu Chen. Numerical Drape Behaviour of Circular
Fabric Sheets Over Circular Pedestals. Textile Research Journal, 70(7):593–
603, July 2000.
[HC00b] Jinlian Hu and Siuping Chung. Bending Behaviour of Woven Fabrics with
Vertical Seams. Textile Research Journal, 70(2):148–153, February 2000.
[HGB69] J.W.S. Hearle, P. Grosberg, and S. Backer. Structural Mechanics of Fibres,
Yarns, and Fabrics (Volume 1). Wiley-Interscience, New York, 1969.
[HLL00] J.L. Hu, W.M. Lo, and M.T. Lo. Bending Hysteresis of Plain Woven Fabrics
in Various Directions. Textile Research Journal, 70(3):237–242, March 2000.
[HM90] B.K Hinds and J. McCartney. Interactive Garment Design. The Visual Com-
puter, 6:53–61, 1990.
[HM91] B.K Hinds and J. McCartney. Pattern Development for 3D Surfaces. Com-
puter Aided Design, 23(8):583–592, August 1991.
[HZ97] Jin-Lian Hu and Yi-Tong Zhang. The KES Shear Test for Fabrics. Textile
Research Journal, 67:654–664, September 1997.
[Ing89] Lester Ingber. Very fast simulated re-annealing. Mathematical Computer
Modelling, 12(8):967–973, 1989.
[Ing92] Lester Ingber. Genetic algorithms and very fast simulated reannealing: A
comparison. Mathematical Computer Modelling, 16(11):87–100, 1992.
[Ing93] Lester Ingber. Simulated annealing: Practice versus theory. Mathematical
Computer Modelling, 18(11):29–57, 1993.
[Ing96] Lester Ingber. Adaptive Simulated Annealing (ASA): Lessons learned.
Control and Cybernetics, 25(1):33–54, 1996.
[Ing00] Lester Ingber. Adaptive Simulated Annealing (ASA).
http://www.ingber.com/ASA.tar.gz, August 2000. Version 21.3.
[Jos77] Marjory L. Joseph. Introductory Textile Science, 3rd Edition. Holt, Rinehart,
and Winston, New York, 1977.
[JP98] Y.J. Jeong and D.G. Phillips. A Study of Fabric-drape Behaviour with Im-
age Analysis Part II: The Effects of Fabric Structure and Mechanical Prop-
erties on Fabric Drape. Journal of the Textile Institute, 89(1):70–79, 1998.
BIBLIOGRAPHY 118
[Kaw80] Sueo Kawabata. The Standardization and Analysis of Hand Evaluation. The
Textile Machinery Society of Japan, Osaka, 2nd edition, 1980.
[KP89] Mark A. Krasnosel’skii and Aleksei V. Pokrovskii. Systems with Hysteresis.
Springer-Verlag, New York, 1989.
[LA85] G.A.V. Leaf and R.D. Anandjiwala. A Generalized Model of Plain Woven
Fabric. Textile Research Journal, 52:92–99, February 1985.
[Law01] Lawrence Livermore and Sandia National Labs. SLATEC Common Math-
ematical Library. http://www.netlib.org/slatec/, May 2001. Version 4.1.
[LBD61] J. Lindberg, B. Behre, and B. Dahlberg. Mechanical Properties of Textile
Fabrics - Part III: Shearing and Buckling of Various Commerical Fabrics.
Textile Research Journal, 31:99–122, February 1961.
[LDG96] L. Ling, M. Damodaran, and R.K.L. Gay. Aerodynamic force models for
animating cloth motion in air flow. The Visual Computer, 12:84–104, 1996.
[LKC96] J.-D. Liu, M.-T. Ko, and R.-C. Chang. Collision Avoidance in Cloth Ani-
mation. The Visual Computer, 12:234–243, 1996.
[LPC95] J. Louchet, X. Provot, and D. Crochemore. Evolutionary Identification of
Cloth Animation Models. Proceedings of the 6th Eurographics Workshop on
Animation and Simulation, pages 30–43, 1995.
[MDS66] D. Moskowitz, Gordon, J. H. Dillon, and Edward W. Suppiger. Large Lat-
eral Deflection of Textile Fabric Structures. Textile Research Journal, 36:770–
786, September 1966.
[Mer59] Reginald Meredith, editor. The Mechanical Properties of Textile Fibres. North-
Holland Publishing Company, Amsterdam, 1959.
[MT98] Nadia Magnenat Thalmann. Clothing Virtual Actors. In Cloth and Clothing
in Computer Graphics, number 31 in SIGGRAPH 98 Course Notes, pages
D–1 through D–26, Orlando, Florida, July 19-24 1998. ACM SIGGRAPH.
[MTS71] J.F. Martin, T. Topper, and G.M. Sinclair. Computer Based Simulation of
Cyclic Stress-Strain Behaviour with Applications to Fatigue. Materials Re-
search and Standards, 11:23, 1971.
[NGA95] H.N Ng, R.L. Grimsdale, and W.G. Allen. A System for Modelling and Vi-
sualization of Cloth Material. Computers and Graphics, 19(3):423–430, 1995.
[Nis86] J. Nisselson. Computer Graphics and the Fashion Industry. Proceedings
Graphics Interface ’86, 1986.
BIBLIOGRAPHY 119
[Num01] Numerical Algorithms Group. NAG F77 Library.

http://www.nag.com/asliR5/, May 2001. ASLI Mark 5.
[NWGC87] J. Nisselson, J. Weil, R. Gordon, and J. Charles. Computer Graphics and
Fashion. Computer Graphics, 21(4):333–334, July 1987.
[OITN92] H. Okabe, H. Imoaka, T. Tomiha, and H. Niwaya. Three Dimensional
Apparel CAD System. Computer Graphics, 26(2):105–110, July 1992.
[PP99] Jacqueline R. Postle and Ron Postle. The Dynamics of Fabric Drape. Textile
Research Journal, 69(9):623–629, September 1999.
[Pro95] X. Provot. Deformation Constraints in a Mass-Spring Model to Describe
Rigid Cloth Behaviour. Proceedings of Graphics Interface ’95, April 1995.
[Pro96] X. Provot. Collision and Self-Collision Handling in a Cloth Model Dedi-
cated to Design Garments. Proceedings of Graphics Interface ’96, 1996.
[RBB97] M.L. Realff, M.C. Boyce, and S. Backer. A Micromechanical Model of the
Tensile Behaviour of Woven Fabric. Textile Research Journal, 67(6):445–459,
June 1997.
[RO43] W. Ramberg and W.R. Osgood. Technical Note No. 902 - Description of
Stress-Strain Curves by Three Parameters. Technical report, National Ad-
visory Committee For Aeronautics, 1943.
[Rob01] Barbara Robertson. Monster Mash. Computer Graphics World, October
2001.
[Sak00] Shinichi Sakata. ASAMIN.
http://www.econ.lsa.umich.edu/∼ssakata/software/, August 2000. Ver-
sion 1.7.
[SD97] A. Sinoimeri and J.Y. Dréan. Mechanical Behaviour of the Plain Weave
Structure Using Energy Methods. Textile Research Journal, 67(5):370–378,
May 1997.
[SE98] W. Strasser and B. Eberhardt. Representation of Knit Fabrics or the art of
knitted fabrics. In Cloth and Clothing in Computer Graphics, number 31 in
SIGGRAPH 98 Course Notes, pages F–1 through F–20, Orlando, Florida,
July 19-24 1998. ACM SIGGRAPH.
[Set00] Ajay Seth. A Predictive Control Method For Human Upper-Limb Motion:
Graph-Theoretic Modelling, Dynamic, Optimization, and Experimental
Investigations. Master’s thesis, University of Waterloo, 2000.
[SHY00] Fengjun Shi, Jinlian Hu, and Tongxi Yu. Modeling the Creasing Properties
of Woven Fabrics. Textile Research Journal, 70(3):247–255, March 2000.
BIBLIOGRAPHY 120
[Ske76] J. Skelton. The Fundamentals of Fabric Shear. Textile Research Journal,

46:862–869, September 1976.
[SLH78] W. Shanahan, D.W. Lloyd, and J.W.S. Hearle. Characterizing the Elastic
Behaviour of Textile Fabrics in Complex Deformations. Textile Research
Journal, 48:495–505, September 1978.
[SSG97] Fangning Sun, Abedelfattah M. Seyam, and B. S. Gupta. A Generalized
Model for Predicting Load-Extension Properties of Woven Fabrics. Textile
Research Journal, 67(12):866–874, December 1997.
[Tap89] Byron D. Tapley, editor. Esbach’s Handbook of Engineering Fundamentals.
John Wiley and Sons, New York, 4th edition, 1989.
[Tav93] L. Tavernini. Differential automata models of hysteresis models. In Models
of Hysteresis. Longman Scientific and Technical, 1993.
[VCMT95] P. Volino, M. Courchesne, and N. Magnenat-Thalmann. Versatile and Ef-
ficient Techniques for Simulating Cloth and Other Deformable Surfaces.
Proceedings of SIGGRAPH ’95, 1995.
[Ver95] Allan H. Vermeulen. The B-Spline Field Approximation Method in Solid Me-
chanics. PhD thesis, University of Waterloo, 1995.
[VMT94] P. Volino and N. Magnenat-Thalmann. Efficient self-collision detection on
smoothly discretized surface animations using geometrical shape regular-
ity. Proceedings of Eurographics ’94, 1994.
[VMT95] P. Volino and N. Magnenat-Thalmann. Collision and Self-Collision De-
tection: Efficient and Robust Solutions for Highly Deformable Surfaces.
Proceedings of Eurographics ’95, 1995.
[VMTJT96] P. Volino, N. Magnenat-Thalmann, S. Jianhua, and D. Thalmann. An
Evolving System for Simulating Clothes on Virtual Actors. IEEE Computer
Graphics and Applications, 16(4):42–51, September 1996.
[VOVL92] C.W.A.M. Van Overveld and E. Van Loon. Hanging Cloths and Dangling
Rods: a Unified Approach to Constraints in Computer Animation. The
Journal of Visualization and Computer Animation, 3:45–59, 1992.
[VT98] Pascal Volino and Nadia Magnenat Thalmann. Developing Simulation
Techniques for an Interactive Clothing System. In Cloth and Clothing in
Computer Graphics, number 31 in SIGGRAPH 98 Course Notes, pages D–
51 through D–63, Orlando, Florida, July 19-24 1998. ACM SIGGRAPH.
[Wei86] J. Weil. The Synthesis of Cloth Objects. Computer Graphics, 20(4):49–54,
August 1986.
BIBLIOGRAPHY 121
[WKL+ 00] Tomiji Wakida, Kyoko Kida, Muncheul Lee, Soyeung Bae, Hiromu Yosh-
ioka, and Yuichi Yanai. Dyeing and Mechanical Properties of Cotton Fab-
rics Treated with Sodium Hydroxide/Liquid Ammonia and Liquid Am-
monia/Sodium Hydroxide. Textile Research Journal, 70(4):328–332, April
2000.
[ZG99] Naiyue Zhou and Tushar K. Ghosh. On-Line Measurement of Fabric
Bending Behaviour - Part III: Dynamic Considerations and Experimental
Implementation. Textile Research Journal, 69(3):176–184, March 1999.

Tjlahey 2002

Uploaded by

Copyright:

Available Formats

Tjlahey 2002

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Tjlahey 2002

Uploaded by

Copyright:

Available Formats

Modelling Hysteresis in the Bending of Fabrics

Timothy John Lahey

Waterloo, Ontario, Canada, 2002

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

2 Fabric Mechanical Properties 5

3 Previous Work on Hysteresis 22

4 A Model for Fabric Hysteresis 30

5 The Fabric Drape Simulation Model 80

B Further Information 100

2.1 Classes of Fabric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

4.1 Details of Fabrics Tested [Den94] . . . . . . . . . . . . . . . . . . . . . . . 40

B.1 Fabric 1 - Warp Direction: Loading Polynomial Coefficients . . . . . . . . 110

2.1 Pierce’s Geometrical Model [HGB69] . . . . . . . . . . . . . . . . . . . . . 7

3.1 Fatigue Mechanical Model [MTS71] . . . . . . . . . . . . . . . . . . . . . 23

4.1 Model of Fabric Bending . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

Textiles have become more prominent in engineering applications in recent years,

1.1 Previous Work

1.1.1 Research in Computer Graphics

1.1.2 Research in Textile Engineering

1.2 Layout of this document

Fabric Mechanical Properties

Class Title Description

Class Title Description

2.1 Woven Fabric Structure

Figure 2.1: Pierce’s Geometrical Model [HGB69]

p = (l − Dθ ) cos θ + D sin θ (2.1)

h = (l − Dθ ) sin θ + D(l − cos θ ) (2.2)

Figure 2.2: Race-Track Cross-section [HGB69]

where m is the bending modulus of the yarn. We also have that:

m cos ψ dψ = −Vxdx (2.8)

Due to symmetry considerations it is assumed that the boundary condition at x = 0 is

The shape is determined by V, m, and θ . Equations for x, y, and s still need to be

with φ = φ0 when ψ = 0 so that sin φ0 = √1

Figure 2.3: Forces on Olofsson’s elastica [HGB69]

Consideration of the tensile properties of woven fabrics involves a number of different

Figure 2.4: Generalised Tensile Load-Extension Curve [HGB69]

Figure 2.5: Grosberg’s micromechanical model [Gro66a]

Figure 2.6: Zhou and Ghosh Bending Model [ZG99]

Figure 2.7: Shi et al. Bending Model [SHY00]

2.5 Kawabata Evaluation System

The tensile force is applied to a 5 cm by 20 cm fabric specimen as shown in Figure

Figure 2.8: KES: Application of Tension Force [Kaw80]

Figure 2.9: KES: Tension Measurements

Figure 2.10: KES: Shear Process [Kaw80]

-2.5 0 0.5 1.5 2.5

Figure 2.12: KES: Bending Measurements [Kaw80]

Previous Work on Hysteresis

3.1 Early Mechanical Material Hysteresis Models

Figure 3.1: Fatigue Mechanical Model [MTS71]

Figure 3.2: Piecewise Linear Approximation to Stress-Strain Curve [MTS71]

3.2 General Hysteresis Models

A general mathematical framework for modelling hysteresis was presented by Kras-

Figure 3.3: Ordinary Play [KP89]

Another type of hysteresis model is the differential automata model proposed by

Figure 3.4: Ordinary Stop [KP89]

3.3 A Differential Model of Frictional Hysteresis

ẋ = |u̇| Ax + Bu̇, x(0) = 0 (3.1)

F(u)(t) = Cx(t) (3.2)

3.3.1 Physical Parameters and Their Model Equivalent

f > 0 and 0 < η < 1 (3.19)