Mathematics of Crimping

Mathematics of
Crimping
Warrick Cooke
St Catherine’s College
University of Oxford
A thesis submitted for the degree of

Doctor of Philosophy
June 2000
To my parents, who showed me the importance of love
Acknowledgements
Thanks to my supervisors, John and Hilary Ockendon, for the many

worthwhile discussions on the subject of crimping. I would also like to
give thanks to V. Saward, K. Gillow and J. Lawry for help with the layout
of this document. I have also appreciated the assistance of Philip Baggot
and Ifor Griffiths for helpful insights to the operation of the stuffer box.
Finally I would like to acknowledge the funding given by the EPSRC and
Courtaulds PLC.
Abstract
The aim of this thesis is to investigate the mathematics and modelling of

the industrial crimper, perhaps one of the least well understood processes
that occurs in the manufacture of artificial fibre.
We begin by modelling the process by which the fibre is deformed as it is

forced into the industrial crimper. This we investigate by presuming the
fibre to behave as an ideal elastica confined in a two dimensional channel.
We consider how the arrangement of the fibre changes as more fibre is
introduced, and the forces that are required to confine it. Later, we apply
the same methods to a fibre confined to a three dimensional channel.
After the fibre has undergone a preliminary deformation, a second process

known as secondary crimp can occur. This involves the “zig-zagged” ma-
terial folding over. We model this process in two ways. First as a series
of rigid rods joined by elastic hinges, and then as an elastic with a highly
oscillatory natural configuration compressed by thrusts at each end. We
observe that both models can be expressed in a very similar manner, and
both predict that a buckle can occur from a nearly straight initial condi-
tion to an arched formation. We also compare the results to experiments
performed on the crimped fibre.
Throughout much of the process, the configuration of the fibre does not
alter. This part of the process we call the block, and model the material in
this region in two ways : as a series of springs; and as an isotropic elastic
material. We discuss the coupling between the different regions and the
process that occurs in the block, and consider both the steady state and
stability of the system.
Contents
1 Introduction 1
1.1 1997 : The Year of the Man-made fibre . . . . . . . . . . . . . . . . . 1
1.2 The fibre manufacturing process . . . . . . . . . . . . . . . . . . . . . 2
1.2.1 The stuffer box crimper . . . . . . . . . . . . . . . . . . . . . 3
1.2.2 Terminology and length scales of the stuffer box crimper . . . 5
1.3 Structure of this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Statement of originality . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5 Experimental Observations and Results . . . . . . . . . . . . . . . . . 8
1.5.1 Observations of crimp formation . . . . . . . . . . . . . . . . . 8
1.5.2 Quantitative results . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5.3 General observations on the fibre crimper . . . . . . . . . . . . 9
2 Primary Crimp 13
2.1 Previous work on related contact problems . . . . . . . . . . . . . . . 15
2.2 2D Confined Fibre . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2.1 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.2 Non-dimensionalisation . . . . . . . . . . . . . . . . . . . . . . 30
2.2.3 Linearisation . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2.3.1 Linearisation of the boundary and free boundary con-
ditions . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2.3.2 Summary of Linearised equations . . . . . . . . . . . 33
2.2.4 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.5 Example 1a : The rectangular channel . . . . . . . . . . . . . 35
2.2.5.1 Section Types . . . . . . . . . . . . . . . . . . . . . . 35
2.2.5.2 Types of Solution . . . . . . . . . . . . . . . . . . . . 41
2.2.5.3 Choice of Solution . . . . . . . . . . . . . . . . . . . 45
2.2.5.4 Degeneracy . . . . . . . . . . . . . . . . . . . . . . . 50
2.2.6 Example 1b : Rectangular channel fed midway . . . . . . . . . 51
i
2.2.6.1 Section types . . . . . . . . . . . . . . . . . . . . . . 52
2.2.6.2 Summary of the types of Solution . . . . . . . . . . . 55
2.2.6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.2.7 Example 2: Fibre confined in a narrow wedge . . . . . . . . . 60
2.2.8 Section Types . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.2.9 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.3 3D Confined Fibre . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.3.1 The equations of an elastica in three dimensions . . . . . . . . 65
2.3.2 Relationship with the co-ordinates in fixed space . . . . . . . . 69
2.3.3 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . 69
2.3.4 Non-dimensionalisation . . . . . . . . . . . . . . . . . . . . . . 69
2.3.5 Linearisation . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.3.6 Boundary Condition at s = 1 + 2t . . . . . . . . . . . . . . . . 72
2.3.7 Summary of Linearized Equations . . . . . . . . . . . . . . . . 73
2.3.8 3D Confined Fibre : Results . . . . . . . . . . . . . . . . . . . 74
2.3.8.1 Normal contact φ = 0 . . . . . . . . . . . . . . . . . 75
2.3.8.2 Energy in three dimensions . . . . . . . . . . . . . . 76
2.3.8.3 General Fibre Orientation - Single contact Point . . 76
2.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.4.1 Two dimensional fibre . . . . . . . . . . . . . . . . . . . . . . 84
2.4.2 Three dimensional crimp . . . . . . . . . . . . . . . . . . . . . 85
3 Secondary Crimp 86
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.2 Hinged Rods : Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.2.1 Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.2.2 Non-dimensionalization . . . . . . . . . . . . . . . . . . . . . . 91
3.2.3 Asymptotics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.2.3.1 O(1) . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.2.3.2 O(κ) . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.2.3.3 O(κ2 ) . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.2.4 A reformulation of the problem . . . . . . . . . . . . . . . . . 94
3.2.5 Results of the asymptotic analysis . . . . . . . . . . . . . . . . 99
3.2.6 Numerical solution of the hinged rods model . . . . . . . . . . 100
3.2.6.1 Comparisons and Observations . . . . . . . . . . . . 100
3.3 Hinged Rods : Fitting Experimental Curves . . . . . . . . . . . . . . 108
ii
3.3.1 Fitting Experimental Data . . . . . . . . . . . . . . . . . . . . 109
3.3.1.1 Typical Results . . . . . . . . . . . . . . . . . . . . . 110
3.3.2 Extensible, Unbending rods . . . . . . . . . . . . . . . . . . . 111
3.3.3 Fitting the data . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.3.3.1 Large Tension . . . . . . . . . . . . . . . . . . . . . . 112
3.3.3.2 Small Tension (θ = π/2 − φ/2 − ) . . . . . . . . . . 113
3.3.3.3 Numerical Fitting . . . . . . . . . . . . . . . . . . . 113
3.4 Naturally-curved Fibres . . . . . . . . . . . . . . . . . . . . . . . . . 115
3.4.1 Multiple scales analysis . . . . . . . . . . . . . . . . . . . . . . 116
3.4.2 Hinged Rods revisited . . . . . . . . . . . . . . . . . . . . . . 117
4 Modelling the Block Motion 120

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.2 Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.2.1 Models for the amount of material entering the block . . . . . 125
4.2.2 Models for the micro-structure of the block . . . . . . . . . . . 127
4.2.2.1 Loops of Elastica . . . . . . . . . . . . . . . . . . . . 127
4.2.2.2 Loop touching others above and below . . . . . . . . 131
4.2.2.3 Loop touching others above, below and to the sides . 131
4.2.2.4 A folded Elastica touching others above and below . 133
4.2.3 Stress, strain and nondimensionalisation . . . . . . . . . . . . 136
4.2.4 The primary crimp length and the force applied to the block . 137
4.3 First model of the Block : Vertical Springs . . . . . . . . . . . . . . . 138
4.3.1 Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.3.2 Time varying Parameters . . . . . . . . . . . . . . . . . . . . . 140
4.3.3 Non-dimensionalization . . . . . . . . . . . . . . . . . . . . . . 140
4.3.4 Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.3.5 Closing the model . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.3.6 Uniform equilibrium states and their stability . . . . . . . . . 143
4.3.6.1 Finding the roots . . . . . . . . . . . . . . . . . . . . 145
4.3.7 Predicting the equilibrium state . . . . . . . . . . . . . . . . . 149
4.3.7.1 Steady State Solutions . . . . . . . . . . . . . . . . . 150
4.3.7.2 A new model for secondary crimp . . . . . . . . . . . 154
4.3.7.3 Nondimensionalisation . . . . . . . . . . . . . . . . . 156
4.4 The Block as an Isotropic Elastic Material . . . . . . . . . . . . . . . 157
4.4.1 Constitutive relations . . . . . . . . . . . . . . . . . . . . . . . 157
iii
4.4.1.1 Nondimensionalisation . . . . . . . . . . . . . . . . . 158
4.4.1.2 Steady State Solution . . . . . . . . . . . . . . . . . 161
4.4.1.3 Relationship between micro-structure and elastic prop-
erties . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.4.2 Stability of the stuffer box . . . . . . . . . . . . . . . . . . . . 164
4.4.2.1 Stability of the stuffer box with a nearly horizontal
upper plate . . . . . . . . . . . . . . . . . . . . . . . 165
4.4.2.2 Stability of the Alternative Stuffer box design . . . . 167
4.4.3 Springs model revisited . . . . . . . . . . . . . . . . . . . . . . 171
4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5 Conclusions 174
5.1 Primary Crimp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
5.1.1 Crimp in three dimensions . . . . . . . . . . . . . . . . . . . . 175
5.2 Secondary Crimp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
5.2.1 The hinged rods model . . . . . . . . . . . . . . . . . . . . . . 177
5.2.2 Curved fibre model . . . . . . . . . . . . . . . . . . . . . . . . 178
5.3 The Block . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
5.3.1 Stability of the stuffer box . . . . . . . . . . . . . . . . . . . . 180
5.4 Open Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
Bibliography 184
iv
List of Figures
1.1 Crimped fibres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 The flat plated stuffer box . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Schematic of the Hinged Plate Crimper . . . . . . . . . . . . . . . . . 4
1.4 Photograph of the hinged plate paper crimper . . . . . . . . . . . . . 5
1.5 Video still of a section of the block. . . . . . . . . . . . . . . . . . . . 8
1.6 Secondary crimp occuring in the paper crimper . . . . . . . . . . . . 9
1.7 Time Series showing how the angle made by the top plate changes with
time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.8 Variation in Crimp Length along the crimped paper strip . . . . . . . 10
1.9 Angle Against Applied Moment . . . . . . . . . . . . . . . . . . . . . 11
1.10 Crimp Length against Applied Moment . . . . . . . . . . . . . . . . . 11
2.1 Model problem for the industrial drier [1] . . . . . . . . . . . . . . . . 16

2.2 Folding of paper fed vertically onto a table [11] . . . . . . . . . . . . 17
2.3 Infinite elastica confined to a channel [15] . . . . . . . . . . . . . . . . 18
2.4 Buckling in a wide channel [3] . . . . . . . . . . . . . . . . . . . . . . 19
2.5 Buckling in a curved channel [13] . . . . . . . . . . . . . . . . . . . . 20
2.6 Contact problem for an elastica [16] . . . . . . . . . . . . . . . . . . 22
2.7 Contact problem for a freely supported elastica [9] . . . . . . . . . . 23
2.8 Section types for constrained Euler buckling [8] . . . . . . . . . . . . 25
2.9 A general geometry for a fibre confined in a 2d box. . . . . . . . . . . 27
2.10 Example 1a : The rectangualr box . . . . . . . . . . . . . . . . . . . 35
2.11 The unconfined fibre . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.12 The starting sections . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.13 The ending sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.14 The Middle sections . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.15 Dimple . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.16 Key for figures 2.17 and 2.18 . . . . . . . . . . . . . . . . . . . . . . . 42
2.17 t̄ vs. λ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
v
2.18 Energy vs. t̄ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.19 Ai vs. t̄ for the first dimple. Transition from solid to dashed line indi-
cates transition from physical to unphysical (outside the box) solution. 49
2.20 Three possible solutions that are indistinguishable energetically . . . 50
2.21 Example 1b: The rectangular box fed midway . . . . . . . . . . . . . 51
2.22 Unconfined fibre being fed into a rectangular box midway up the left-
hand wall. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.23 The fibre confined only by the bottom wall . . . . . . . . . . . . . . . 53
2.24 The starting section for fibre fed into a rectangular box midway up the
lefthand wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.25 Key to figures 2.26 and 2.27 showing the types of solutions found . . 56
2.26 t̄ vs. λ for a fibre entering midway . . . . . . . . . . . . . . . . . . . . 58
2.27 Energy vs. t̄ for a fibre entering midway . . . . . . . . . . . . . . . . 59
2.28 Geometry for a fibre in a wedge . . . . . . . . . . . . . . . . . . . . . 60
2.29 The unconfined fibre in a wedge . . . . . . . . . . . . . . . . . . . . . 61
2.30 The fibre confined by the bottom surface of a wedge . . . . . . . . . . 61
2.31 Starting section for a fibre in a wedge . . . . . . . . . . . . . . . . . . 62
2.32 The mid section for a wedge . . . . . . . . . . . . . . . . . . . . . . . 62
2.33 The ending section for a wedge . . . . . . . . . . . . . . . . . . . . . 63
2.34 λ against t̄ for a fibre in a wedge . . . . . . . . . . . . . . . . . . . . 64
2.35 Observed 3d crimp formations . . . . . . . . . . . . . . . . . . . . . . 66
2.36 A three dimensional elastica confined in a narrow channel . . . . . . . 67
√
2.37 λ vs. t̄ for a 3d elastica, σ2 = 2/5 . . . . . . . . . . . . . . . . . . . 77
√
2.38 Energy vs. t̄ for a 3d elastica, σ2 = 2/5 . . . . . . . . . . . . . . . 78
2.39 Typical path for a 3d crimped fibre, σ2 = 0.25 . . . . . . . . . . . . . 79
2.40 3d configurations for a confined fibre with increasing λ. φ = π/4,
σ2 = 0.25 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.1 The two ways of seeing secondary crimp . . . . . . . . . . . . . . . . 87

3.2 Hinged Rod configuration for an even number of rods (N ) . . . . . . 90
3.3 Comparison of asymptotics and Numerical Solution for N = 20 and
N = 21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.4 Comparison of asymptotics and Numerical Solution for N = 100 and
N = 101 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.5 Σ0 against κ in the case N = 20, φ = 0.9 . . . . . . . . . . . . . . . . 104
3.6 Key to Figure 3.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
vi
3.7 Σ0 against κ in the case N = 21, φ = 0.9 . . . . . . . . . . . . . . . . 106
3.8 Key to Figure 3.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.9 Fitting Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.10 A comparison of a theoretical and experimental curve . . . . . . . . . 114
3.11 A smoothly curved fibre . . . . . . . . . . . . . . . . . . . . . . . . . 116
3.12 s and s0 for the hinged rods model . . . . . . . . . . . . . . . . . . . . 118
4.1 Video still of a section of the block. . . . . . . . . . . . . . . . . . . . 122

4.2 Diagram of the stuffer box, showing movement of material from A to A0 124
4.3 Formation of the block through secondary crimp . . . . . . . . . . . . 125
4.4 Two possible fibre arrangements . . . . . . . . . . . . . . . . . . . . . 128
4.5 Comparison of loop microstructure with the material from the block . 130
4.6 Strain vs. stress for a loop of elastica . . . . . . . . . . . . . . . . . . 130
4.7 Strain vs. stress for a loop of elastica touching similar loops above and
below . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.8 A loop touching others at four points . . . . . . . . . . . . . . . . . . 133
4.9 Strain vs. stress for a loop of elastica touching similar loops on four sides134
4.10 Structure of the folded Elastica . . . . . . . . . . . . . . . . . . . . . 135
4.11 Strain vs. Stress for a folded elastica touching similar loops above and
below . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.12 An idealized view of the crimped material in the stuffer box . . . . . 138
4.13 Forces acting on an elemental section of the block . . . . . . . . . . . 139
4.14 Closing the springs model . . . . . . . . . . . . . . . . . . . . . . . . 142
4.15 Regions of stability (black) and instability (white) for the springs model,
with x0 = 1.0 and 0.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
4.16 Regions of stability (black) and instability (white) for the springs model,
with x0 = 0.8 and 0.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
4.17 Closing the springs model . . . . . . . . . . . . . . . . . . . . . . . . 149
4.18 Geometry for the new model for secondary crimp . . . . . . . . . . . 154
4.19 The nip as a wedge . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
4.20 The block composed of an isotropic elastic solid . . . . . . . . . . . . 157
4.21 An alternative design of stuffer box . . . . . . . . . . . . . . . . . . . 165
4.22 Stability contours for the stuffer box crimper with isotropic elastic
block and nearly flat top plate . . . . . . . . . . . . . . . . . . . . . . 168
4.23 Stability contours for the stuffer box crimper with isotropic elastic
block and nearly flat top plate . . . . . . . . . . . . . . . . . . . . . . 169
vii
4.24 The different models considered in this chapter . . . . . . . . . . . . . 173
5.1 Tow being removed from a box . . . . . . . . . . . . . . . . . . . . . 182
viii
Chapter 1
Introduction
1.1 1997 : The Year of the Man-made fibre

Over the last century there has been a silent revolution in the textile industry. Up
until the end of the 19th century, fabrics were limited to the small number of natu-
rally occuring fibres. However, over the next few decades, spurred on in part by the
wartime economies, chemists began to perfect the techniques of manufacturing arti-
ficial fibres. Large scale production brought novel fabrics into the lives of the many.
For many years, the emphasis was on producing low cost replacements for naturally
occurring fibres such as wool, cotton and silk, which artificial filaments could achieve
with only limited success.
The technology of weaving had not stood still over these years. Fundamental re-
search into how fibres behaved when woven into fabrics meant that textile engineers
could at last begin to tailor the properties of the cloth to the use. These so called “de-
signer fabrics” required fibres with particular properties, and artificial filaments were
soon produced to fulfil this need. Outdoor pursuits were one of the first beneficiaries,
where new micropore jackets gave superior warmth and wind protection, whilst al-
lowing the body to lose moisture and hence maintain comfort. Fashion, always to the
forefront of textile revolution, began to experiment with novel materials, that allowed
designers to work with a greater range of textures and styles than was possible before.
Artificial fibres have a great impact in areas outside the textile industry. New
fibre chemistry and coatings have improved the efficiencies of filters. In addition,
filaments are often used as precursors to other industrial products (for example to
make the carbon fibre brake pads for large jets). In all these applications, the ability
to control the properties of the fibres is crucial. Much work had been done ensuring
1
Figure 1.1: Crimped fibres
the correct concentrations in the polymers from which the fibre is formed, and careful
control of the spinning process ensures an even thickness of filament. However the
three-dimensional structure of the fibre (its crimp- see figure 1.1) has proved much
more problematic. Indeed, the design of the stuffer box crimper (the part of the
production line that turns straight filaments into crimped filaments) has always been
regarded as a black art. It is this subject that this thesis will attempt to investigate.
1.2 The fibre manufacturing process

To understand the design of the stuffer box crimper, one must first understand the
process by which fibres are manufactured. Artificial fibres begin as polymer melts,
liquids formed from large chain polymers and solvents. This solution is extruded (ei-
ther into water or air) through plates containing thousands of holes1 to form the tow2 .
Once in the air or water, the polymer begins to lose solvent, and hence solidify. At
this point, the filaments are much thicker than required, and so the next process is to
stretch them, thus making them thinner. Two techniques are typically used. Fibres
extruded into water are taken across a pair of rollers, running at different speeds.
This provides the stretch. Air spun filaments are blown with high speed gas jets as
1
The shape of the holes determines the cross-section of the fibres, typical examples include cir-
cular, square or triangular.
2
The tow is the name given to the collection of filaments that is being manufactured. It is
typically 1000 filaments wide by 20 deep.
2
Crimped
Tow Sprung
Door

Tow
Figure 1.2: The flat plated stuffer box
they leave the extruding plate, which both dries and tensions the fibres.
The fibres are now washed and dried, to remove all traces of solvents. Depend-
ing on the eventual destination, they may also be dyed to the appropriate hue. Up
until now the filaments have remained straight3 . The final process before packaging
is to introduce the permanent crimp. Conceptually, it is easy to describe how this
happens. The tow is fed into a small box, and must fold up many times in order to
fit. This fold becomes permanent to form crimp. Implementing this idea as part of a
continuous production line is where the fun begins.
1.2.1 The stuffer box crimper

The main problem to overcome is having a design that forces the fibre to buckle
while still forming part of a continuous process. There are two main methods used to
achieve this. The easiest to understand is the flat plated stuffer box (figure 1.2). This
forces the fibres into a rectangular tube, the far end of which is blocked by a sprung
door. Fibre fills the tube until the force required to add more fibre is sufficient to
force open the door, allowing crimped fibre to escape.
The more common design of stuffer box, on which we shall concentrate in this
thesis, uses a hinged top plate rather than a sprung door. The plate supports a large
3
Except for a limited type of polymer which crimps during solidification due to instabilities
3
W
Crimped
Tow
Tow
Figure 1.3: Schematic of the Hinged Plate Crimper
weight, and as more fibre is forced in, the buckled tow lifts the plate, allowing tow
to escape. Figure 1.3 shows a schematic of the hinged plate crimper, whilst 1.4 is
a photograph of a modified model that crimps paper (see section on experimental
work 1.5).
Construction of a stuffer box crimper is a difficult undertaking. The individual
filaments are very thin, so construction tolerances must be high lest the fibre escape
and jam the machine. Unfortunately, as we wish to make as much fibre as possible,
we require the crimper to operate at high speeds and under large loads. This makes
the engineering of such a device an expensive job. When a new fibre comes into
production, it will need a modified crimper, in order to cope with its particular
properties. Hence the motivation behind this thesis is to predict how properties of
the crimped fibre depend on the parameters of the machine and polymer.
At this point, it is worth discussing the history of the modelling of the crimping
process. Courtaulds first brought the crimping problem to the 26th European Study
Group with Industry, in 1993 [2]. The participants modelled the primary crimp
formation as an Euler strut, but the model contained many unknown parameters,
and hence was of little use in making predictions. This thesis, which was partly
funded by Courtaulds (as a Case award) aims to improve and extend those early
ideas to give a unified model for the stuffer box crimper.
4
Figure 1.4: Photograph of the hinged plate paper crimper
1.2.2 Terminology and length scales of the stuffer box crimper

When we observe the stuffer box crimper in action4 , and examine closely the output,
we see structures on a number of length scales. The smallest (∼ 2mm) is referred to
as primary crimp. This is the main permanent deformation that occurs. When the
fibre exits the stuffer box, it is folded a second time (∼ 15mm), in the arrangement
we call secondary crimp. For comparison, the stuffer box is about 200mm long, and
has rollers 100mm in diameter. The region between the rollers we will refer to as the
nip.
1.3 Structure of this thesis

In section 1.5 we shall introduce experimental observations made using the paper
crimper. Although this does not exhibit all the features of the fibre crimper, there is
sufficient similarity to gain an understanding of the basic processes involved. These
observations will help focus our investigation into the mathematics of crimp. Included
in this section will be details of the limited quantitative results obtained from exper-
4
See section 1.5 on experimental observations
5
iments, and a brief discussion of the difficulties in carrying out experiments.
Chapter 2 will discuss the formation of primary crimp. We begin with a brief re-
view of the literature on contact problems for two dimensional elastic rods, and then
proceed to describe a technique to solve the problem of a elastica confined within a
narrow channel, which is applicable to a range of different geometries. In particular,
we present results for three shapes of channel: two for comparison with another paper
on a similar subject (that used a different methodology), and one which is new (the
fibre in a wedge).
The remainder of chapter 2 extends these techniques to the problem of a three-

dimensional elastica confined to a channel. The aim is to gain understanding as to
if/when the fibres of elliptic cross-section adopt a helical form. We find that for small
deflections, the deformations in each of the principal directions of the fibre uncouple.
This allows us to solve the problem using results from the early part of the chapter.
Unfortunately, although the fibre adopts a three-dimensional shape, we do not predict
any helical crimp.
Secondary crimp involves the deformation of highly “concertinaed” fibres. In

chapter 3, we will describe two models for such a system (in two dimensions). The
first models the crimped fibre as a series of rigid rods coupled by elastic hinges. We
first derive the difference equation for the orientation of the rods, then attempt to
solve it asymptotically for small applied force. The solution we find is not valid when
the number of rods is large, and we therefore have to reformulate the problem in
terms of rapidly and slowly varying terms. We then seek a solution that is uniformly
valid for all numbers of rods. This predicts a critical value of the force for which there
is a change in behaviour from a nearly flat configuration to an arched shape. The
equation for this force is observed to be different depending if there are an even or
odd number of rods present. This feature is observed numerically. We then discuss
experiments to determine values of the parameters in the model.
Our second model for secondary crimp treats the crimped fibre as an elastica with
a rapidly varying shape. We carry out a multiple scales analysis for the problem, and
find that the fibre behaves as an Euler strut of lower stiffness than the fibre from
which it is composed. We then make comparisons between this result and that for
6
the hinged rods model.
Chapter 4 looks at the stuffer box crimper as a complete system. We present two
models for the material between the plates, treating it as either a system of springs
or as an isotropic elastic material. We derive equations in each case that describe
the force needed to move the material in terms of the current properties of the ma-
terial. In particular, we relate elastic constants to the microstructure of the block,
and hence to the primary crimp length. We also relate the secondary crimp length
model described in chapter 3 to the height of the material between the plates. We
then discuss methods of closing the model to describe how the global behaviour of the
crimping machine changes with time. As well as the models discussed in chapter 3,
we also discuss a second model for the coupling between the force required to move
the material and the properties of the incoming fibre. The advantage of this model
is that it takes into account the geometry of the region between the rollers.
As well as discussing steady states, we also examine the stability of these states to
small perturbations. We find that the “springs” model predicts that the stuffer box
is almost always stable. In order to examine the stability of the “elastic material”
model, we choose to study a simpler system, and find that the system can go unstable
if there is much friction between fibre and plates, and the equilibrium height of the
material is large.
To compare the stability of the two models, we also look at the much simpler hor-
izontal plate stuffer box design. We observe that this is less stable than the hinged
plate version, and that the springs model is more stable in general than the elastic
model.
In the final chapter, we bring these results together, and also present some open
problems.
1.4 Statement of originality

In chapter 2, the results of the work carried out on a fibre confined to 3 dimensional
channel is original work. All of chapter 3 is original, as are the models and results for
the stuffer box discussed in chapter 4.
7
Figure 1.5: Video still of a section of the block.
1.5 Experimental Observations and Results

In order to improve their understanding of the crimping process, Courtaulds built a
paper crimper. This works the same way as the hinged plate stuffer box crimper, but
instead of a tow, uses a continuous sheet of paper as its input. As the paper cannot
escape from the sides, the walls of the crimper can be made of perspex, and hence
it is easy to observe the crimp occuring. It is also much smaller (as the fibre comes
from a roll, not a production line!) and therefore allows experiments to be carried
out much more easily.
1.5.1 Observations of crimp formation

• Primary Crimp
Even when using the paper crimper, primary crimp is difficult to observe, as it
happens in the small gap between the rollers. When running the crimper very
slowly, we observe that contact occurs first at one surface, then the other.
• Secondary Crimp
Secondary crimp occurs relatively infrequently. Primary crimped fibre builds
up until it suddenly folds, and moves towards the plates (see figure 1.6).
• The material between the plates

The material between the plates does not move continuously, but rather in jerks
as more fibre is added through secondary crimp. The structure of the material
is very complicated (see figure 1.5), with many gaps. This makes modelling the
properties difficult. Thankfully, as it moves along the gap, the orientation of
8
Figure 1.6: Secondary crimp occuring in the paper crimper
the fibre does not change much. We shall make use of this fact in chapter 4
when we model the fibre in this area.
Under certain circumstances, especially when the applied load is large, the block
may buckle, lifting the angle of the top plate. This is a feature of the paper
crimper, and is not observed in the industrial process.
1.5.2 Quantitative results

Quantitative results were hard to come by with the paper crimper. Over a period of
a week, numerous experimental runs were performed with different loadings on the
upper plate of the crimper. The angle of the plate was recorded at regular intervals,
and after the experiment was concluded, the primary crimp length was measured at
regular points along the paper. Figures 1.7 and 1.8 show typical data sets. Notice the
large variations in both quantities. The graphs 1.9 and 1.10 show how the arithmetic
means of these properties changed with the applied load. Note that “trends” are
difficult to observe.
1.5.3 General observations on the fibre crimper

Observation of the industrial crimper shows a number of differences to the paper ver-
sion studied. First is the relative stability of the fibre crimper. Unless the plates are
close to requiring replacement, the crimper is relatively stable, and exhibits none of
the large variations in angle oberverved with paper. Secondly, the crimp of the fibre
9
Angle (2) Chart 1
4
Angle
0
0 200 400 600 800 1000 1200 1400 1600 1800
Time (s)
Page 1
Figure 1.7: Time Series showing how the angle made by the top plate changes with
time
2.5
2
Crimp Length (mm)
1.5
0.5
0
0 500 1000 1500 2000 2500 3000 3500
X (mm)
Figure 1.8: Variation in Crimp Length along the crimped paper strip
10
Average Angle
6.00
5.00
4.00
Angle
3.00
2.00
1.00
0.00
0.14 0.15 0.16 0.17 0.18
Moment (Nm)
Page 1
Figure 1.9: Angle Against Applied Moment
Average Freq
2.5
2
Av. Crimp Length (mm)
1.5
0.5
0
0.145 0.15 0.155 0.16 0.165 0.17 0.175 0.18
Moment
Page 1
Figure 1.10: Crimp Length against Applied Moment
11
emerging from the industrial process is much more uniform (these observations may
well be related).
The fibre crimper also has some features that arise only due the nature of the
material being fed in. Unlike a sheet of paper, the fibre can easily deform in three
dimensions. This is an important feature of the industrial process that cannot be
understood through studying the paper crimper alone.
12
Chapter 2
Primary Crimp
In this chapter we will discuss how the primary crimp forms as the fibre enters the
crimping machine between the rollers. We believe that the processes that occur at
formation and the length scale of the crimp are two of the major factors that deter-
mine the properties of the fibre. To gain a better understanding of what is happening
when the fibre first enters the machine, we will model it as an elastica1 being forced
into a long, thin box. The fibre must bend to allow more to enter, and so we expect
it to touch the walls at many points. This gives rise to a contact problem. Let us
begin our discussion with a brief outline of other appropriate contact problems that
can be found in the literature.
Consider a fibre confined to two dimensions. The full problem is nonlinear and it
is possible, but difficult, to make some progress in this case. However, for the sake
of simplicity we will concentrate on the linearised model for the elastica. To do so,
we will have to confine ourselves to small displacements of the fibre. Due to this
linearisation we find that the equations can be solved exactly in the regions between
the contact points. These segments can then be joined together by applying suit-
able conditions at the contact points. Finally, we can apply conditions on the fibre
as a whole to find a solution to the full problem. In the best traditions of applied
mathematics in OCIAM we will start by applying this to the simplest problem of a
thin, rectangular box. This will illustrate both the advantages and limitations of the
method. We will end the discussion of the two-dimensional fibre by briefly looking at
the results for a fibre in a wedge.
1
Elastic beam
13
Armed with the knowledge and experience of the confined two-dimensional fibre,
we will then begin to look at the effect of giving the fibre a third direction into which to
deform. Our hope is to explain the observations about the three-dimensional nature
of crimp. The equations for a three-dimensional elastica are numerous and nonlinear,
so we will again make a start by limiting ourselves to small deflections and hence lin-
earising the equations. Under these circumstances, we will see that the deformations
in each of the principle axes of inertia of the cross-section are independent. This will
allow us to exploit many of the techniques we developed for the two dimensional fibre.
Once again, we consider only the simplest geometries.
14
2.1 Previous work on related contact problems
Contact problems between elasticas and surfaces are very common in industrial prob-
lems. One particularily rich source has been modelling the handling of sheets of
material (be they paper or tow). At the 22nd Study Group with Industry [1], Cour-
taulds brought a problem they encountered when fibre was being transfered between
the rollers on an industrial drier. Under certain operating conditions, the tow would
buckle as it was being laid down on the second cylinder. The model problem consid-
ered was that of feeding paper onto a table at twice the rate that the feeder moved
along the table (see figure 2.1). This resulted in an excess of fibre, and hence a com-
pressive force. The rate of motion and feeding was assumed sufficiently small that
inertia effects could be ignored. A brief outline of the equations and some details of
the solution can be found in Ockendon [6]. Starting from the steady state equations
of an elastica acted on by both a compressive force (due to the overfeeding) and a nor-
mal body force (from gravity), the linearized equations give a fourth order equation
relating the height of the fibre (y) as a function of position (x),
y (4) + λy 00 + γ = 0, d < x < t.
where λ is the nondimensional compressive force and γ the nondimensional gravity

term. At the contact point with the table (x = d) the boundary conditions are
y = y 0 = y 00 = 0
and at the feeder (x = t)

y=1 y0 = 0
(which is fixing the height and direction of the fibre). The overfeed condition is that
Total length − Distance moved along table = Excess Fibre
which can be rewritten as Z t 2

dy
dx = 2t.
0 dx
Solving this free boundary problem involved solving a pair of trancendental equations,
and there are multiple possible shapes the fibres could take up. In [1] use was made
of energy arguments to determine which was most likely. The result was believed
not to be valid for large times due to steepening effects making the nonlinear terms
ignored in the derivation important.
15
2
y
1 Height=1
x
x=d x=t
Figure 2.1: Model problem for the industrial drier [1]
Certain problems will always require one to consider nonlinear effects. Mahadevan
and Keller [11] [12] have modelled the problem of thin sheets and ropes being fed ver-
tically onto a horizontal table (see figure 2.2). During the evolution of a fold the angle
must change by π/2; it will never be possible to ignore the geometrical nonlinearity.
Mahadevan and Keller began with the dynamic equations for an elastica acted on by
a body force (gravity). They then focussed their attention on situations when the
inertia terms were negligible. At the feeder they specified both the position and angle
made by the sheet. The boundary conditions made at the other end depended on
which configuration they expected to see. Experimental observations suggested that
the folding began with no contact with the surface, followed by touching line contact.
As more sheet is added, the angle between the sheet and the table decreases towards
zero. After this is reached, a second contact line forms, which moves as more fibre
is added. Eventually, the weight of fibre causes it to form a third contact line (fig-
ure 2.2). Mahadevan and Keller solved the problem numerically using a “homotopy
method.” Their solution agreed qualitatively with experimental results. The authors
then considered a free standing fold of paper, and applied the same techniques to pre-
dict the width and height. These were found to agree well with experimental results
made for different paper types.
Vaillette and Adams [15] looked at the problem of paper crumpling in photo-
copiers. They modelled this as an infinite, periodic elastica confined within a channel
between two rigid walls. The only forces acting on the elastica came from a compres-
sive thrust delivered along the undeformed length, and reaction forces at the contact
points (see figure 2.3a). They made the assumption that, through symmetry, each
section was identical, and that contact could only occur at points, not over a con-
tinuous region. This reduced the problem to the equivalent situation of a cantilever
(as shown in figure 2.3b). The equations for this are well known, and because of
the simple geometry, they were able to solve the full nonlinear problem using elliptic
16
Figure 2.2: Folding of paper fed vertically onto a table [11]
17
(a)
(b)
Figure 2.3: Infinite elastica confined to a channel [15]
integrals. This gave multiple solutions, and the authors resorted to energy arguments
to determine which were stable for a given applied force. They found that regular so-
lutions were only stable for a limited range of forces, above which the elastica “folded
over” and was considered unstable.
In a follow-up paper, Adams and Benson [3] discussed the postbuckling of a finite
elastica confined between two parallel plates. This time the length of the elastica was
finite, although once again the only forces acting were the equal and opposite thrusts
applied to each end and reaction forces with the walls. The aspect ratio of the channel
was O(1), and hence it was necessary for them to solve the full, nonlinear model for
the elastica. At the feeding end, they fixed the position and angle of the elastica, and
at the other end the position was fixed and it was assumed that there was no applied
moment. Their method was to divide the beam into a series of sections, and solve the
problem in each region exactly using elliptic integrals. This required them to make
assumptions as to what was occuring between the contact points. They proposed a
sequence of events that began with the buckled fibre, which then made contact at
the upper surface. The curvature at the contact point decreased as more fibre was
added until a flat section (where the fibre is in contact over a region) formed. This
grew until a critical condition was met when the fibre folded over (see figure 2.4).
Clearly the nature of the contact set is crucial to the modelling and analysis, and we
will discuss this from a mathematical viewpoint shortly.
Soong and Choi [13] looked at the problem of an elastica being fed round a curved
channel (see figure 2.5). They applied a very general technique to solve the problem
that would be applicable to channels where the curvature was discontinuous. Be-
ginning with the nonlinear equations for the elastica, they assumed the number of
18
Figure 2.4: Buckling in a wide channel [3]
19
Figure 2.5: Buckling in a curved channel [13]
contact points present. They then attempted to solve the problem by a shooting
method, where they used numerical integration to find the shape of the elastica and
to improve on initial guesses to unknown parameters (such as forces and position of
contact points). They then attempted to improve the accuracy by adding another
contact point. The close proximity of many contact points sugguested that continuous
contact was possible and we shall make use of this idea of trying different numbers
of contact points in the next section.
Variational techniques are very popular to solve contact problems for elasticas.
Westbrook [17] [16] looked at beams and plates loaded non-uniformly. He considered
a beam of stiffness B(x) held a distance δ above a flat surface. The beam was acted
on by a body force f (x) which is normal to the undeformed beam and such that f > 0
imples a force lifting the beam away from the surface (see figure 2.6). Starting from
the linearised equations of the elastica, and letting u(x) be the displacement from the
surface, Westbrook looked for solutions to the problem
u(x) ≥ 0,
00
(B(x)u00 (x)) − f ≥ 0,
00
u(x) (B(x)u00 (x)) − f = 0
−1 ≤ x ≤ 1
20
with boundary conditions
u(±1) = δ B(±1)u00 (±1) = M,
where M is the moment applied to the ends. He then rewrote it in the weak formu-
lation as a variational inequality. Defining a convex cone of functions

K = u|u − δ ∈ H02 (−1, −1), u ≥ 0
and
Z 1
a(u, v) = B(x)u00 · v 00 dx,
−1
Z 1
l(w) = f wdx + M (w0 (1) − w0 (−1)),
−1
then u is the unique solution of
u ∈ K, a(w − u, u) − l(w − u) ≥ 0 ∀w ∈ K.
He proceeded to find this solution using a finite element method and to compare for
simple cases with exact solutions. The examples given in [16] are confined to the
special cases of
B(x) = 1 and f (x) = A + Cx2
for constant A and C. For C = 0, four types of contact are observed to occur:
1. No contact.
2. Point contact at x = 0.
3. Line contact in the region −ρ ≤ x ≤ ρ.
4. Point contact at x = ±ρ.
with the type of contact observed depending on the relative sizes and signs of A and
M.
Hodges and Bless [9] approached the problem of a loaded beam constrained by a
rigid boundary using techniques from optimal control (see figure 2.7). Their notation
is quite different2 from that presented elsewhere in this chapter, and so we shall
go through the derivation in more detail. They used u(x) and v(x) to denote the
displacment of the elastica from its initial line. The orientation of the cross-section
2
But does reduce to ours under the appropriate assumptions
21
f(x)
M M
δ
Figure 2.6: Contact problem for an elastica [16]
of the elastica was written θ(x). The three strain measures associated with the
deformation of the beam were (x) (the stretching of the reference line), γ(x) (the
transverse shear strain measure) and κ(x) (the bending strain measure), and these
took the part of the “control parameters” in the formulation of the problem. They
used the equations
u0 = (1 + ) cos θ − γ sin θ − 1
v 0 = (1 + ) sin θ + γ cos θ
θ0 = κ
as the equations of state. The constraint that the elastica remain above the boundary
was written as
S = −(v + h) ≤ 0 ∀x.
They first made the assumption that the ends of the elastica were simply supported
such that the reference line was unstretched ( = 0). The optimisation involved
minimising the total of the strain and potential energy, which in nondimensional
terms is Z
1 1 2
J= κ + gγ 2 + 2µv dx
2 0
where g and µ are given by
GJl2
g =
EI
ql3
µ =
EI
22
-v(x)
u(x)
Figure 2.7: Contact problem for a freely supported elastica [9]
and EI is the bending rigidity of the beam, GK the transverse shear rigidity, l the
length of the beam and q the body force per unit length. g therefore determines
the relative importance of bending to shear, whilst µ indicates the balance between
gravity and bending stiffness.
The solution of this problem was performed numerically using commercially avail-
able software. Two important cases were considered. The first, labelled in the paper
the Timoshenko model, had general values of g and µ. The authors found that the
shear force acting on the fibre was continuous, and that the fibre made contact over
a region. This was in contrast to the second model they examined, labelled Euler-
Bernoulli, where in addition to assuming = 0 they also chose γ = 03 . In this case,
they found that the shear force acting was discontinuous, which the authors consid-
ered unphysical. However, for the purposes of this thesis we shall ignore this question
and continue to use the Euler-Bernoulli equations, under the assumption that the
transverse shear rigidity is sufficiently large that these effects are minimised.
3
This is equivalent to setting g infinite, i.e. a beam with infinite transverse shear rigidity
23
Domokos, Holmes and Boyce [8] have studied the problem of constrained Euler
buckling as it applies to the stuffer box crimper. This work is the most relavent to
the model of primary crimp that we shall proceed to develop, and was first reported
in a simple form in [4]4 . They worked with the nonlinear equations of a two dimen-
sional elastica, and assumed that inertial and gravity terms were not significant. The
technique they used to examine the problem was very similar to that employed by [3]
in that they divided the fibre into regions between contact points and solved the
equations exactly in this region using elliptic integrals. Their solutions encompassed
the following contact/boundary conditions (see figure 2.8):
1. Pinned-pinned (no contact)
2. Clamped-pinned without contact
3. Pinned-contact
4. Clamped-contact
5. Contact-contact (same wall)5
They then used continuity arguments to form full solutions, and plotted load against
displacement of the end. They observed that flat solutions could occur (where the
elastica was in continuous contact with the walls, but forces only acted at the ends
of the fibre).
In later parts of the paper, Domokos, Holmes and Boyce also approached the prob-
lem using both numerical and experimental techniques. The experimental technique
duplicated where possible the conditions modelled in the theoretical part. The chan-
nel walls were made as smooth as possible to reduce the amount of friction present,
the displacement of the end was carefully controlled, and as uniform a beam as pos-
sible was used. Numerically, a shooting technique was used to locate values of the
unknown parameters, and the resulting plots of load against displacement compared
with the experimentally obtained curves.
The technique we shall discuss in the next section has some similarities and many
differences to the important pieces of work we have summarised above. The ge-
ometries we shall consider include the two parallel plates discussed by Vaillette and
4
Note that the paper [8] was published at the same time that the work we present was being
undertaken.
5
The authors did not consider the case of contact-contact which went between the walls.
24
(1)
(2)
(3)
(4)
(5)
Figure 2.8: Section types for constrained Euler buckling [8]
25
Adams [15], Adams and Benson [3] and Domokos, Holmes and Boyce [8]. Unlike these
authors, we shall limit our discussion to situations where the aspect ratio is small,
and hence where we can use the linearised equations for the elastica. Despite this, we
shall observe a similar bifurcation diagram to that reported for the wide channel [8].
We shall also expand our methodology to describe a stuffer box in the shape of a thin
wedge.
Unlike the cases considered by Westbrook [17] [16] and Hodges and Bless [9],
there will be no advantage in using variational techniques to solve the problem. This
is because of the potential existence of multiple solutions when a compressive force is
added. In their discussion of the variational technique for moving boundary problems,
Elliott and Ockendon [5] show that a unique solution exists if the energy operator
is coercive. This requirement is not satisfied if we apply a compressive force to the
elastica. Hence we lose the advantages of formulating the problem as a variational
inequality.
The technique we will adopt is most similar to that employed by Soong and
Choi [13] where we try different numbers of contact points until we find a solution.
Note that we extend their method to cases when there are two surfaces to which we
can make contact. We also have the advantage that we can write down an exact
solution to the linearised elastica equations in the region between contact points, and
hence do not need to use numerical integration to solve the problem.
26
YU (x)
y Fy
x hXb
Fx
YL (x)
Xb
Figure 2.9: A general geometry for a fibre confined in a 2d box.
2.2 2D Confined Fibre

Let us begin our discussion of the problem by setting out a typical configuration we
may expect to see. Figure 2.9 shows a fibre confined between two general surfaces. In
all the examples we shall consider, the wall geometry will be much simpler than that
shown here, but the methods we shall discuss will be applicable to many possible
geometries. As we expect the weight of the fibre to be unimportant, we will not
consider the effect of allowing any body forces to act on the fibre. Instead, we will
suppose that the fibre is acted on by reaction forces whenever it touches the walls, and
by a thrust along its undeformed direction, applied at the ends of the fibre. Before
we begin to discuss the equations that govern the shape, let us begin by introducing
the notation. Throughout, the distance along the fibre will be denoted by s, where
0 ≤ s ≤ L.
We will write x(s) and y(s) to describe the position of the fibre, and θ(s) to denote its
orientation and say that the undeformed fibre lies along the line y(s) = constant. The
walls of the box are given by x ∈ (0, Xb ), y = YU (x) and y = YL (x), where the sub-
scripts U and L refer to upper and lower boundaries. Later, we will need a quantity
h that gives a typical height of the box. The internal forces that act on the fibre are
written as Fx (s) and Fy (s) where the subscript defines the direction in which they act.
Throughout this thesis, we shall limit ourselves to the case when there are no
body forces acting on the fibre. If this is the case then we have when the fibre is not
touching the walls
dFy
= 0, (2.1)
ds
27
dFx
= 0, (2.2)
ds
and θ(s), x(s) and y(s) unknown functions. Alternatively, if the fibre is in contact
with the walls, θ(s), x(s) and y(s) are known, and the forces exerted by the walls on
the fibre have to be found.6
In both contact and non-contact cases, balancing the moments acting on the fibre
gives
d2 θ
EI + Fx sin θ − Fy cos θ = 0. (2.3)
ds2
If the fibre is in contact with the walls, this equation describes the relationship be-
tween Fx (s) and Fy (s), without determining either. In addition, if the walls are flat
(as we will limit ourselves in this thesis), this equation can be interpreted as requiring
that the internal force in a fibre in the contact region acts along the boundary. If
we further assume that the walls are smooth, and hence able to exert forces only
perpendicular to themselves, we can also deduce that, when the fibre is in contact,
Fx and Fy are constant. From equation (2.3) the forces can only change when θ 00 (s)
changes discontinuously. This can occur when
• the fibre changes from being in contact with the surface to being free of the
surface; or
• the fibre touches the surface at a point.
We shall call these points boundary points denote them as si (i = 1, · · · N + 1). We

shall mostly be interested in the behaviour of the fibre in between these points, and
for that reason s1 , the first boundary point, is always at the leftmost end of the fibre
s = 0 and the last sN +1 at s = L. The reaction forces that act in the x and y
directions at these points are given by fx,i and fy,i respectively, and change the forces
acting on the fibre as
XN
dFy
= fy,i δ(s − si ), (2.4)
ds i=1
X N
dFx
= − fx,i δ(s − si ). (2.5)
ds i=1
6
We shall discuss the boundary conditions necessary when the fibre goes from contact to free
shortly
28
We also have from the geometry of the fibre in the channel
dy
= sin θ, (2.6)
ds
dx
= cos θ. (2.7)
ds
The free boundary conditions between equations (2.4) and (2.5) will depend on
the model of contact that we use. We will only assume a frictionless contact.7
Here we assume that the walls of the box are smooth. If this is the case then the
total reaction force, which has the direction

fx,i
fy,i
must be normal to the surface. In particular, if the wall is at an angle ψ to the

horizontal then
fx,i cos ψ
· = 0. (2.8)
fy,i sin ψ
Equation (2.8) is an example of a free boundary condition. We will make use of
this and others like it when joining together solutions between the contact points.
2.2.1 Boundary Conditions

There are two types of place where we will need to apply boundary conditions. Firstly,
we must model the ends of the fibre. Depending on the geometries involved, we may
need to specify position, direction of the fibre and/or and moment (or lack thereof)
applied. Note that prescribing the moment is equivalent to prescribing θ 0 . We shall
need to apply free boundary conditions at the boundary points. Most of these will
be self evident : as the contact point is on a surface, we know a relationship between
x(si ) and y(si ); the fibre must also be tangent to the surface at the contact point,
and hence θ is determined; the fibre cannot penetrate the surface so we can also say
for certain that θ 0 ≤ 0 at the upper surface, and conversely θ 0 ≥ 0 at the lower.8 It is
important to realise what effect the boundary points have on the fibre. Firstly, the
fibre is always continuous, hence
[x(si )]+ +
− = [y(si )]− = 0.
7
Frictional contact
A second assumption is to have friction acting at the contact points. This makes solution of the
problem considerably more complex, and we will not discuss this option further.
8
In the case of non-flat walls, these become θ 0 ≤ ψ 0 and θ 0 ≥ ψ 0 respectively
29
Secondly, we assume that the boundary points do not introduce any kinks in the fibre,
so
[θ(si )]+
− = 0.
Finally, we assume the boundary points do not apply a moment to the fibre, so
+
[θ0 (si )]− = 0
Note that, as yet, the position of the boundary points is not known, nor the magnitude
of the contact forces fh,i and fv,i .
2.2.2 Non-dimensionalisation
We shall now choose suitable scalings for our equations. We are interested in the
behaviour of the fibre as it fills the channel, so it is natural to non-dimensionalise all
lengths with Xb . Unfortunately, we have no prior knowledge of the magnitude of the
forces involved. Thankfully, there is a natural force scaling that arises from (2.3):
EI
.
Xb2
These lead us to the following variables
x = x 0 Xb
y = y 0 Xb
s = s 0 Xb
si = s0i Xb
L = Xb (1 + 2t)
EI
Fy = λi p i
Xb2
EI
Fx = λi
Xb2
EI 0
fx,i = f
Xb2 x,i
EI 0
fy,i = f
Xb2 y,i
At this point, we should note that Fy and Fx are both constant between boundary
points, and so we have written them in terms of quantities (λi and pi λi ) that are
constant for each section. These subscripts are written such that pi is the quantity
30
that corresponds to the section from boundary point i to i + 1. Inserting into our
equations (2.4), (2.5) and (2.3) and dropping the primes (’) gives
[pi λi ]ssi+
i−
= fv,i , (2.9)
[λi ]ssi+
i−
= fh,i , (2.10)
d2 θ
+ λi sin θ = pi λi cos θ. (2.11)
ds2
where [ ]ssi+
i−
denotes the change across the boundary point si .
It is perfectly possible to begin solving these equations as written. The results

are difficult to interpret as they involve many elliptic integrals. Instead, we shall now
seek a suitable linearisation.
2.2.3 Linearisation
Looking at the typical geometry we are interested in, we realise that the channel is
much longer than it is wide (i.e. h, the aspect ratio 1). Let us attempt to make
use of this in linearising the equations. Firstly, let us scale all vertical distances with
this parameter.
y = h ȳ
Yu = hXb Y¯u
Yl = hXb Ȳl
Equation (2.6) becomes

hȳ 0 = sin θ
and in order to balance, we will need either a rapid variation as we move along the
fibre, or
θ = h θ̄.
It is this latter case that we shall consider here. Bearing this in mind, and inserting
the scalings into (2.11) requires that
pi = h p̄i
in order that the term does not dominate. Finally, for reasons that will become clear
later, we need to write that
t = h2 t̄.
31
Inserting these into (2.11), gives
d2 θ̄
2
+ λi θ̄ = p̄i λi + O(h2 ) (2.12)
ds
In addition, the equations for the position become
dȳ
= θ̄ + O(h2 ) (2.13)
ds
dx
= 1 + O(h2 )
ds
2.2.3.1 Linearisation of the boundary and free boundary conditions
We shall now see the effect of these scalings on the boundary conditions we apply
both at the ends of the fibre and the boundary points. Firstly, we will assume that
the right hand end is at x = 1:
Z 1+2h2 t̄
2
x(1 + 2h t̄) = cos hθ̄ds = 1
0
and hence one condition is Z 1

θ̄2 = 4t̄. (2.14)
0
At the boundary points, we will still require that θ̄, θ̄0 , x and y are all continuous.
The condition that the fibre be tangent to the wall can be expressed, in dimensional
variables, as
dYu
tan θ =
dx
Making the scalings for these quantities, this becomes
dY¯u
θ̄ = . (2.15)
dx
Once again, for the upper surface θ̄0 ≤ 0.
Finally, we must consider the changes in λi and λi p̄i that occur across the boundary
point. Remembering our continuity condition (2.8) and changing variables gives
s
[λi ]si+
i− cos ψ
s · = 0,
h[p̄i λi ]si+
i− sin ψ
where ¯
−1 d Yu −1 dȲl
ψ = tan h or tan h
dx dx
32
as required. Expanding this to O(1) gives,
[λi ]ssi+
i−
= 0. (2.16)
This is an important result, as it means that the each section will feel the same thrust,
λ. For this reason, we can treat λ as a global parameter rather than determining it
for each section in turn.
2.2.3.2 Summary of Linearised equations
The linearised equations for the poisition of the fibre within the channel are
ȳ 0 (s) = θ̄
x0 (s) = 1 s ∈ 0, 1
θ̄00 (s) + λi θ̄ = p̄i λi .
At the extreme ends of the fibre we fix the position, and either apply no moment (in
which case θ̄0 = 0 or fix the angle (θ̄ fixed). The length of fibre within the box leads
to the global boundary condition
Z 1
θ̄2 = 4t̄.
0
At the boundary points, we assume no friction, and hence
[λi ]ssii +
− = 0.
In addition, we assume that x, ȳ, θ̄ and θ̄0 are all continuous.
2.2.4 Methodology
We shall now attempt to make use of the fact that equation (2.12) is simple to solve
analytically in the regions where p̄i is fixed. For si < s < si+1 it has the general
solution
√ √
θ̄ = Ai sin λs + Bi cos λs + p̄i . (2.17)
Problems arise because the position and number of the boundary points, as well as
the changes they make to p̄i are not specified as part of the problem. Below we will
discuss a methodology that can partially overcome these problems by making use of
the symmetries present in the solutions of (2.12).
33
Our method will be to construct candidate solutions to the problem. These will
be built out of a series of exact solutions in the regions between boundary points. By
requiring that all the necessary quantities are continuous, we can join these sections
together. Finally, we can look at global parameters to ensure that the solution solves
the problem as a whole. Let us begin by noting that each part of the elastica is
defined by five constants, Ai , Bi , λ,p̄i and di , which is the length of the section and
is given by
di = (si+1 − si ).
The values of these constants are fixed by the requirements that θ̄ and θ̄0 are contin-
uous, and that each end of the section should be a boundary point. This gives four
boundary conditions for four unknowns in each section. Also λ is a global parameter,
and we have the global boundary condition that
X
di = 1 (2.18)
i
To find a solution,we select a particular number and type9 of boundary points, and
start by choosing trial values for λ and the free initial condition at the start (either θ̄
or θ̄0 ). The remaining parameters are fixed by requiring that the section satisfies the
condition at the first boundary point. The continuity conditions are then applied, and
the next section generated so that it goes to the next point. This continues until the
end of the fibre is reached. We then check to see if a solution has been found, based
on the global condition (2.18). If not, new values of our shooting parameters are tried.
We also need to check that the solution is physically acceptable, namely that at
each contact point (1) θ̄0 has the correct sign, and (2) the change in the force λp̄i
has the correct sign10 . Given that this is confirmed, we can then calculate the length
of fibre that is contained in the channel from equation (2.14). It is important to
note that the method we have just described will generate solutions to the problem
as formulated. It does not distinguish between solutions that are possible and those
that actually occur in practice. For this, we need to discuss a selection criteria
such as seeking solutions that minimise the elastic potential energy. This point will
be discussed in section 2.2.5.3 for the example of a fibre confined to a rectangular
channel.
9
The type of boundary point refers to its position at either the top or bottom boundary.
10
For the top boundary, this force must act inwards, hence [λp¯i ]+
− ≥ 0. Equivalently, we can say
that [θ̄ 00 ]+
− ≥ 0
34
Figure 2.10: Example 1a : The rectangualr box
2.2.5 Example 1a : The rectangular channel

Let us now put these ideas into practice. We will start with a very simple geometry
for the channel, a rectangle (with boundaries at y = 0 and y = h). In addition, we
will apply the same boundary conditions of no applied moment (θ̄0 = 0) and y = 0
at each end of the fibre. Our first task will be to identify what can happen to the
fibre between the contact points. The solutions of these “prototype” sections will
then be written down, and any undetermined parameters identified. These will be
fixed when we come to join the sections together. The solutions to the full problem
will be classified depending on the types of section they contain. We will then plot a
diagram showing how λ varies with the amount of fibre in the channel (t̄) for different
classes of solution. These will show that there can be more than one solution for a
particular value of t̄. Hence we will need to introduce criteria to select which solution
is observed in practice.
We will now introduce two points of notation. Firstly, as the position along the
channel makes no difference to the boundary conditions (because the height of the
channel is constant), we will write down all of the solutions in terms of the distance
from the contact point ri where
ri = s − s i .
¯ i,
Secondly, we will define the quantity δx
Z di
¯i= θ̄i2
δx dri .
0 2
Hence the global boundary condition (2.14) becomes
X
¯ i = 2t̄.
δx
i
2.2.5.1 Section Types
We now consider the solution of the problem for all possible “sections” between two
contact points. By joining these sections together, we can construct the complete
solution.
35
Figure 2.11: The unconfined fibre
1. Unconfined Fibre (figure 2.11)

The fibre only has contact points at the start and end, and does not touch the
top boundary at all. The boundary conditions for this type of section are:
θ̄0 = 0 at ri = 0, 1
p̄i = 0
max ȳ(r) < 1 (2.19)
min ȳ(r) ≥ 0. (2.20)
Applying these requires that

√ √
θ̄ = Bi cos λri and λ = mπ (2.21)
for some integer m. The amplitude of the fibre, Bi , is determined by the bound-
ary conditions on the length of the fibre in the channel (2.14).
¯ i.
In addition, we can also calculate the additional quantities y(ri ) and δx
Bi
ȳ(ri ) = sin mπri
mπ
2
¯ i = Bi
δx
4
If we now specify that y(0) = 0, it becomes clear that in order for the equa-
tion (2.20) to hold, sin mπri must remain positive. This implies that m = 1.
We can now summarise that, for the unconfined fibre,
θ̄(r) = Bi cos πri

Bi
ȳ(r) = sin πri
π
where Bi2 = 8t̄. In order to satisfy (2.19), Bi < π, which implies that the total
length of the fibre in the box is less than
h2 π 2
1+
4
36
Figure 2.12: The starting sections
for this type of solution to be possible.
2. Starting Section (figure 2.12)

The starting section goes from the base to the first contact point at ȳ = 1 The
boundary conditions we apply are:
θ̄0 (0) = 0,
θ̄(di ) = 0,
θ̄0 (di ) ≤ 0,
p̄i < 0,
ȳ(0) = 0,
ȳ(di ) = 1. (2.22)
These can then be applied to solution (2.17) to solve for the coefficients Ai and
Bi . Hence
√ !
cosλr
θ̄(ri ) = p̄i 1− √ i (2.23)
cos λdi
√ !
p̄i √ sin ri λ
ȳ(ri ) = √ ri λ − √ (2.24)
λ cos λdi
√ √ √ !
p̄i 2 √ 8 sin λdi 2 λdi + sin 2 λdi
¯i
δx = √ 4 λdi − √ + √ . (2.25)
8 λ cos λdi cos2 λdi
The condition that this section goes from the base to the top plate allows us to
fix p̄i through use of equations (2.22) and (2.24), if we assume that di is known
√
λ
p̄i = − √ √ . (2.26)
di λ − tan λdi
The requirements that p̄i ≤ 0 and θ0 (di ) ≤ 0 mean that
π √
< λdi ≤ π. (2.27)
2
37
Figure 2.13: The ending sections
At this point we introduce the variable zi where

√
zi = λdi
the importance of which will become apparent later.
3. End Section (figure 2.13) The end section is the mirror image of the start section,
and has the boundary conditions
θ̄0 (di ) = 0
θ̄(0) = 0
θ̄0 (0) ≤ 0
p̄i > 0
ȳ(0) = 1
ȳ(di ) = 0.
Having applied these we find that

√ !
cos λ(di − ri )
θ̄(ri ) = p̄i 1 − √ . (2.28)
cos λdi
Once again, we can integrate this to obtain the position within the channel
√ √ !
p̄i √ sin λ(di − ri ) − sin λdi
ȳ(ri ) = 1 + √ λri + √ (2.29)
λ cos λdi
√ √ √ !
p̄ i
2 √ 8 sin λd i 2 λd i + sin 2 λdi
¯i = √
δx 4 λdi − √ + √ . (2.30)
8 λ cos λdi cos2 λdi
As the section joins one plate to the other, we can use equation (2.29) and
√
ȳ(di ) = 0 to calculate p̄i in terms of di and λ
√
λ
p̄i = √ √ . (2.31)
λdi − tan λdi
38
Figure 2.14: The Middle sections
The requirement that p̄i ≥ 0 means that

π
≤ zi ≤ π (2.32)
2
√
where again zi = λdi . The value of zi will be determined by applying the
continuity conditions for the fibre.
4. Middle Section (figure 2.14)

The middle section goes from top plate to bottom plate (or vice versa). The
boundary conditions are
θ̄(0) = 0
θ̄(di ) = 0
ȳ(0) = 0 or 1
|ȳ(di ) − ȳ(0)| = 1. (2.33)
Applying these gives

√ √
θ̄(ri ) = p̄i 1 − cos λri − F (zi ) sin λri , (2.34)
√
where once again zi = λdi and
1 − cos zi
F (zi ) = . (2.35)
sin zi
39
Figure 2.15: Dimple
We can now proceed to integrate this to find the coordinates of the fibre in the
box
p̄i √ √ √
ȳ(ri ) = ȳ(0) − √ F (zi )(1 − cos λri ) − (ri λ − sin λri ) (2.36)
λ
2
¯ i = p̄√i (6zi − 6F (zi ) + 8F (zi ) cos zi − 2F (zi ) cos 2zi +
δx
8 λ
2zi F 2 (zi ) − 8 sin zi + sin 2zi − F 2 (zi ) sin 2zi ). (2.37)
Applying the boundary condition on ȳ(di ) − ȳ(0) (2.33) and making use of half
angle identities we arrive at the following equation for p̄i
√
λ
p̄i = Gi (2.38)
zi − 2 tan z2i
where Gi is 1 if the fibre goes from base to top and -1 if it goes from top to base.
As we wish p̄i to be positive as it goes from base to top, and that θ̄0 (di ) ≤ 0 we
also require that
π ≤ zi ≤ 2π (2.39)
Once again, zi must be determined by use of the continuity conditions
5. Dimple (figure 2.15)

The dimple is one of two solutions that go from one plate to the same plate
(the other being a flat section). The boundary conditions are
θ̄(0) = 0
θ̄(di ) = 0
p̄i = 0
max |(ȳ(ri ) − ȳ(0))| < 1
y(0) = y(di ) = 0 and θ0 (0) = θ0 (di ) ≥ 0
or y(0) = y(di ) = 1 and θ0 (0) = θ0 (di ) ≤ 0.
40
Application of these leads to
√
θ̄(ri ) = Ai sin λri (2.40)
mπ
di = √
λ
for some integer m. If we require that y(0) = y(di ) then we can deduce that
m must be even. Note that Ai has still not been determined. Instead, this will
need to be calculated from the continuity of θ̄0 . The position of the fibre within
the channel can be found simply by integration to give
Ai √
ȳ(ri ) = ȳ(0) + √ (1 − cos λri )
λ
2
¯ i = Ai di .
δx (2.41)
4
The limit on the height of the dimple requires that
|ȳ(ri ) − ȳ(0)| ≤ 1
√
λ
and |Ai | ≤ .
2
6. Flat section
We also have the possibility of a flat section, where
θ̄ = 0.
By multiplying (2.12) by θ 0 and integrating, we can readily see that the mag-
nitude of θ0 is directly related to θ, and hence if theta0 is 0 at one end of the
section, it must be 0 at the other (and hence theta0 = 0 at all contact points
where theta = 0. We will discuss the special treatment required for these solu-
tions shortly.
2.2.5.2 Types of Solution
We can now begin to combine the above sections to form solutions to the full problem.
First let us note that the free boundary conditions are dependent on the parameter
√ √
λ only through the quantity zi = λdi . Hence we can make use of the global
conditions on di (2.18) to write11
√ X
λ= zi . (2.43)
i
11
This is true if there are no flat sections present. If this is not the case then
√ 1 X
λ= zi (2.42)
1−σ i
where σ < 1 is the total length of flat fibre.
41
(a)
(b) (g)
(c) (h)
(d) (i)
(j)
(e)
(f) (k)
Figure 2.16: Key for figures 2.17 and 2.18
Secondly, in order to satisfy the boundary conditions, we require that :
Either The only section present is the unconfined fibre

Or The solution begins with a starting section and finishes
with an ending section.
Or The solution has θ̄0 = 0 at each end of all the sections
The third of these is the only way the fibre can contain a flat section.
The method outlined above produces many different solutions (see figure 2.16).
However they can be classified into a small number of different categories depending
on how they are constructed.
1. Unconfined Fibre (a)
The fibre has not yet touched the top plate, and hence encounters no vertical
42
force. As shown in section 1 this will occur for Bi < π and for a constant force.
Explicitly we have that
√
λ = π
Bi2
t̄ = .
4
When Bi = π the fibre just touches the top and a new category of solution is
needed.
2. Point Contact (b), (e) and (h)

These solutions occur after a dimple or unconfined fibre just touches one of the
plates. They always begin with a starting section, finish with an ending section,
and may have an even number of middle sections.
If there are only the two sections present, then the symmetry of the start and
end types requires that z1 = z2 and hence
√
λ = 2z1
¯ 1 (z1 ).
t̄ = δx
In the presence of middle sections, we can still say that, by symmetry
z1 = z N
zi = z j i, j 6= 1, N
and in addition the continuity of θ̄0 requires that (from (2.23), (2.26) , (2.34),
and (2.38)
tan z1 tan z22

= (2.44)
z1 − tan z1 z2 − 2 tan z22
π
< z1 < π
2
π < z2 < 2π.
If the number of middle sections is 2M then

√
λ = z1 + M z2
¯ 1 + M δx
t̄ = δx ¯2
¯ 1 is the δx
where δx ¯ for sections 1 and N , and δx
¯ 2 is the δx
¯ for the middle
sections. As z1 approaches π the curvature at the contact point tends to zero,
at which point a flat section is possible.
43
3. Flat Sections (c), (f ) and (i)
These have the same sections present as before, but with the additional option
√
of flat sections. The presence of these greatly simplifies the calculation of λ
and t̄. As the θ̄0 = 0 at the contact points, we require that
z1 = z N = π
zi = 0 if section is flat
zi = 2π otherwise
(This arises from the symmetry of the different sections, in particular that the
curvature at one end of a middle section is the opposite to that at the other
¯ i = 0 for the flat section, and again writing 2M to be
end.) In addition as δx
the number of middle sections we can see from (2.25), (2.26), (2.37) and (2.38)
that
3 √
t̄ =(2 + M ) λ (2.45)
8π
The growth of the flat section is limited only by the requirement that the total
flat length (σ) is less than 1 for the linearised case. As will be seen later, this
limit is never reached in a stable configuration.
4. Flat sections (j) and (k)

It is not necessary for flat sections to begin with a starting section and end with
an ending section. Instead these can be replaced by equivalent middle sections.
Most of the above discussion holds true, except the values for z1 and zi are
altered, and if P is the number of start or end sections then
3 √
t̄ = (P + 2M + 2) λ
16π
which reduces to (2.45) for the case P = 2.
44
5. Dimple Solutions (d) and (g)
Dimple solutions are similar to the point contact described above except that
they also include a dimple section. Once again, symmetry simplifies greatly the
√
problem of determining the properties λ and t̄, and we can immediately write
that :
zN = z 1
zdimple = 2π
zi is found from the solution of (2.44),
where zdimple refers to the dimple section and zi applies to all the middle-type
sections. We can now apply the continuity condition for θ̄0 and use the equations
for θ̄ ((2.23), (2.26) and (2.40)) to determine the value for Ai , the amplitude of
the dimple :
√
λ tan z1
Ai = ± (2.46)
z1 − tan z1
√
λ
and |Ai | ≤ .
2
Note that the ambiguity over the sign of Ai arises because of the possibility of
the dimple forming at the top or the bottom. It is now trivial to insert these
quantities in order to calculate t̄ and λ.
Figure 2.17 shows the calculated values of t̄ and λ for the configurations discussed
above and drawn in figure 2.16. Note the presence of kinks in the curves. These
can be classified into two types. The first occur when a dimple begins to form. This
is a bifurcation. The second occur when a dimple touches the base and begins to
flatten. These are not bifurcations. For each of these curves, all valid values of z1
were used. Looking at this graph we see immediately the problem of non-uniqueness
of the solutions if we specify t̄ or λ alone. We must now apply energy minimisation
arguments to determine which configuration is stable for each of these cases.
2.2.5.3 Choice of Solution
The problem we are solving can be attacked in a number of different ways. The
approach used so far has been based on equilibrium conditions for elemental sections
of the fibre. However an alternative method is to seek configurations that minimise
the energy of the system. Explicitly, we can write that the solution minimises :
Strain Energy − Work Done by External Forces
45
(f) (g) (h) (i)
2000
(k)
(j) 1500
Figure 2.17: t̄ vs. λ
46
λ 1000
500
0
0 5 10 15 20 25
t
(a) (b) (c) (d) (e)
where the strain energy is given by
Z l
Curvature2
Strain Energy = EI ds
0 2
In the non-dimensional, linearised system this becomes
Z 1 02
θ̄
ds. (2.47)
0 2
The work done depends on the particular boundary conditions we choose. If the
force is controlled (equivalent to prescribing λ) then it acts as an external part of the
system and work is done
Z l
Work Done = Fh (1 − cos θ)ds
0
or (in linear, non-dimensional terms)

Z 1
θ̄2
Work Done = λ ds + O(h2 )
0 2
= 2λt̄ + O(h2 ). (2.48)
For the situation where the length is specified (t̄ is prescribed) there is no external
work done to the system, hence the only contribution to the energy is due to strain
in the fibre.
We now apply the latter condition for the case of increasing amounts of fibre being
fed into the box. Plotting log(Energy) vs the control variable t̄ we obtain figure 2.18
From this we can see the progression of the solution as more fibre is fed into the box:
The fibre begins to buckle as soon as t̄ is positive (a). The amplitude increases
steadily until the fibre just touches the top. It then begins to flatten (b). When
the curvature at the contact point reaches zero, the flat solution begins to grow (c).
This it does until it is energetically feasable for a dimple (d) to form. Note that the
amplitude of the dimple is not zero. The dimple grows until it reaches the base, at
which point a second flattening occurs (e). This then forms a flat solution (f) which
again forms a dimple when it is able. The cycle of dimple to touching to flat to dimple
continues indefinitely.
The dimple in this arrangement is favoured energetically. However, the transition
from flat to dimple is not straightforward. Looking at figure 2.17 we see that for each
√
dimple, the plot of t̄ against λ curves back on itself. This is also visible on the
47
(f) (g) (h) (i) (j) (k)
14
(e)
12
(d)
Figure 2.18: Energy vs. t̄
10
Log(Energy)
(c) 8
48
6
(b)
2
(a)
0
0 5 10 15 20 25
t
0.4
Ai
0.2
-0.2
-0.4
2.98 3 3.02 3.04 3.06
t
Figure 2.19: Ai vs. t̄ for the first dimple. Transition from solid to dashed line indicates
transition from physical to unphysical (outside the box) solution.
energy - t̄ plot (figure 2.18). The effect is clearer if we examine the variation of the
dimple amplitude (Ai ) with t̄.
The case where Ai is zero marks the intersection of the flat and dimple curves.
When this happens, we must have z1 = π If we therefore write that for the first
dimple solution (d)
z1 = π +
2
1X ¯
t̄ = δxi
2 i=0
√ tan z1
Ai = λ
z1 − tan z1
and expand in powers of we get that
3
t̄ = 3 + + 22 + O(3 )
2π
2
Ai = 4 + 2 + O(3 )
π
Note that for physical solutions, < 0 This can be plotted to give figure 2.19
49
(i)
(ii)
(iii)
Figure 2.20: Three possible solutions that are indistinguishable energetically
Notice that the gradient is positive at Ai = 0, and hence the dimple exists locally
for t̄ less than the critical value. We would therefore expect that there is a fold in the
curve for a smaller value of t̄, which is the energetically favourable branch. This is in
contrast to the case for an unconfined, non-linear Euler strut.
2.2.5.4 Degeneracy
Our method is successful in that it predicts solutions for different values of the control
parameters, λ or t̄. However in many circumstances there exist solutions with the
same energy . Consider the situation where the fibre begins with a starting section,
has two middle sections, a flat section and an end section. The process by which λ
and t̄ are calculated involves the sum of various quantities, and hence does not take
into account the exact order of the sections. Hence there are three different versions
of the fibre described above that are indistinguishable in terms of t̄, λ and Energy
(see figure 2.20)
This occurs as there is no energy penalty for moving the points of contact. If
we include a frictional force this would change, and the arrangement would become
dependent on the past history of the fibre, in such a way as to minimise the movement
of the contact points.
50
Figure 2.21: Example 1b: The rectangular box fed midway
2.2.6 Example 1b : Rectangular channel fed midway

The simple rectangular box we have just considered showed how our method can
work in practice. Unfortunately, the boundary conditions on the fibre are a long way
from those we would expect in the industrial crimper. Let us therefore move one step
closer by feeding fibre into the box at one end, whilst leaving the other free to rotate
(see figure 2.21). Mathematically, we will specify that at s = 0
1
θ̄ = 0 and y= ,
2
and at s = 1
θ̄0 = 0.
It will be natural to then progress to a fibre confined to a wedge.

Once again, we begin by considering the types of section that will make up our final
solution. Thankfully, we have only changed the boundary conditions at s = 0. So,
except for the starting section and unconfined fibre, we can make use of the results
from example 1a. Having discussed the changes that must be made, we will join
√
the sections together in the same way as before, and hence produce our graph of λ
against t̄. Energy minimisation arguments will again be needed to select the solutions
that will be observed.
51
Figure 2.22: Unconfined fibre being fed into a rectangular box midway up the lefthand
wall.
2.2.6.1 Section types
1. Unconfined Fibre (1) (figure 2.22)

Once again we will solve equation (2.12) but now we consider a fibre which
touches neither the top nor the bottom wall (figure 2.22). Physically, we are
treating the right-hand end as free to move vertically as more fibre enters, until
it reaches the lower wall. This means that there can be no vertical forces on
the fibre and so p̄i = 0. Hence the boundary conditions we will apply are
θ̄(0) = 0
θ̄0 (1) 0 =
1
ȳ(0) =
2
0 ≤ ȳ(ri ) ≤ 1 for 0 ≤ ri ≤ 1 (2.49)
p̄i = 0.
Applying these gives

√ √ π
θ̄ = Ai sin λri and λ=
2
Solving (2.13) for ȳ we find
1 2Ai √
ȳ = + 1 − cos λri
2 π
and hence in order to satisfy (2.49) we place limits on Ai such that
−π π
≤ Ai ≤
4 4
When Ai = −π/4, the fibre is touching the bottom righthand corner of the box
A2
and12 t̄ is trivially found to be 8i . Thus Ai is determined by the amount of
fibre that has been fed into the box.
12
and when Ai = π/4 the fibre touches the top right hand corner
52
Figure 2.23: The fibre confined only by the bottom wall
2. Fibre confined by the bottom wall (figure 2.23)

This section has the fibre confined only by the bottom wall (see figure 2.23).
Now we have another parameter p̄i to determine, but there is an extra boundary
condition at ri = 1 The boundary conditions are
θ̄(0) = 0
θ̄0 (1) = 0
1
ȳ(0) =
2
ȳ(1) = 0
with 0 ≤ ȳ(ri ) ≤ 1 for 0 ≤ ri ≤ 1. (2.50)
Solving (2.12) for θ̄ we find

√ √ √
θ̄ = −p̄i (tan λ · sin λri + cos λri − 1)
√
for λ ≥ π2 . Integrating up, we find
1 p̄i √ √ √ √
ȳ = − √ tan λ(1 − cos λri ) + sin λri − λri .
2 λ
The boundary conditions on ȳ(1) fix the value of p̄i to be
√
1 λ
− √ √
2 λ − tan λ
whilst the requirement that the fibre remain in the box (2.50) places an upper
bound on λ.13 Finally, we can calculate t̄ as
√ √ √ √ √
λ 4 λ + 2 λ cos 2 λ − 3 sin 2 λ
√ √ √
4 16( λ cos λ − sin λ)2
so that in this case λ is determined by the length of fibre in the box.
13
This comes from solving
ȳ(rmax ) = 1
where 0 ≤ rmax ≤ 1 solves θ̄(rmax ) = 0
53
(a)
(b)
Figure 2.24: The starting section for fibre fed into a rectangular box midway up the
lefthand wall
3. Starting section (figure 2.24)

In order to solve a starting section of the type shown in figure 2.24 we apply
the boundary conditions
θ̄(0) = 0
θ̄(di ) = 0
ȳ(0) = 0.5
ȳ(di ) = 0 or 1
0 ≤ ȳ(ri ) ≤ 1 0 ≤ r i ≤ di .
The solution depends on how many zeros θ̄ has between the contact points. If
none, then the solution is equivalent to a midsection of half the height, analysed
as on page 37, so that

√ 1 − cos zi √
θ̄ = p̄i 1 − cos λri − sin λri ,
sin zi
where p̄i 14 is √ !
1 λ
±
2 zi − 2 tan z2i
14
The positive value correspnds to the solution going to the bottom plate.
54
√
as zi = λdi . If one now allows the fibre to go downwards before going upwards
as shown in (2.24b) (or vice versa), another solution is possible which is
√ √ √
θ̄ = p̄i (1 − cos λr) − α λ sin λr (2.51)
zi = 2π
where α is an unknown constant (α ≥ 0). We can integrate to find ȳ

1 p̄i √ √ √
ȳ(ri ) = + √ λri − sin λri − α(1 − cos λri ) (2.52)
2 λ
¯
and δx √
Z di 2
¯ θ̄ λ 3 2
δx = dri = + πα
0 2 2 16π
The requirement that the fibre remains within the box places limits on the
values of α allowed. In addition, the equation for ȳ allow us to determin p̄i as
√
λ
p̄i = .
4π
The value of α will be determined by the global boundary conditions.
2.2.6.2 Summary of the types of Solution
Once again, we can form solutions by piecing together the different sections discussed
above and in section 2.2.5.1. They are shown in figure 2.25.
1. Unconfined Fibre (a)

The fibre has touched neither plate, and we have that
√ π
λ = ,
2
A2i
t̄ = .
8
This is valid for
π
− ≤ Ai ≤ 0.
4
2. Fibre confined at the base (b)
The fibre is touching only the lower plate. Varying λ gives various values of t̄,
with
π √ √
≤ λ ≤ λmax
2
√
where λmax satisfies the equation
3√ √
π= λmax − tan λmax .
2
55
(a)
(b) (c)
(d) (e)
(f) (g)
(h) (i)
(j) (k)
Figure 2.25: Key to figures 2.26 and 2.27 showing the types of solutions found
56
3. Fibre with point contact (c), (f ) and (i)
The fibre is in point contact with the surfaces, and in the first section θ̄ has no
zeros between the contact points. If one or more middle sections are present,
then
1 tan z21 tan zmid
2
= .
2 z1 − 2 tan z21 zmid − 2 tan zmid
2
The continuity condition at the end is
1 tan z21 tan zN

z1 = .
2 z1 − 2 tan 2 zN − tan zN
4. Fibre with point contact (e), (h) and (k)

The fibre is again in point contact with the surfaces, and θ̄ has a single zero
within the starting section. We vary the solution with the parameter α, use the
continuity conditions
− tan zN
α=
zN − tan zN
to match with the end section, and
tan zmid
2
α=
zmid − 2 tan zmid
2
to match with the middle sections.
5. Fibre containing flat sections (d), (g) and (j)

If σ again represents the total flat length, and there are M middle sections then
√ (3 + 2M )π
λ=
1−σ
2.2.6.3 Results
Figure 2.26 plots t̄ against λ for the different types of solution discussed above. No-
tice the similarity with figure 2.17 which is the same plot for the original geometry.
Firstly, see the similarity between dimple solutions in the first geometry and those
where the first section has a zero in θ between the contact points.
Once again, we need to resort to energy arguments to decide which solution will
be observed in practice. Figure 2.27 shows a plot of energy against t̄. Note that, in
contrast to example 1a, solutions with a flat section are never energetically favoured.
57
(j)
(i)
(g)
Figure 2.26: t̄ vs. λ for a fibre entering midway
(f)
800
600
(d)
58
400
(c)
(b) 200
(a)
2 4 6 8 10
(e) (h) (k)

3500
Energy
Figure 2.27: Energy vs. t̄ for a fibre entering midway
3000
2500
(j)
(g) 2000
(d) 1500
59
(c) 1000
(b)
500
(a)
2 4 6 8 10
t
(e) (f) (h) (i) (k)
Figure 2.28: Geometry for a fibre in a wedge
2.2.7 Example 2: Fibre confined in a narrow wedge

We will now consider the more relevant geometry of a fibre entering a wedge, as shown
in figure 2.28. This is a similar configuration to that encountered in the crimping ma-
chine, where fibre enters into the nip region between the two rollers. We will begin by
discussing the types of behaviour possible between the contact points. From these,
we will form solutions in much the same way as before, once again making use of any
symmetry to simplify the problem. Finally we will discuss the results for this model
and compare them to those for the rectangular box geometries.
We use the same scalings as before and take the walls of the wedge to be at
1 1
y= ± x;
2 2
thus the wedge is of small angle and we can continue to use the linear approximation.
60
Figure 2.29: The unconfined fibre in a wedge
Figure 2.30: The fibre confined by the bottom surface of a wedge
2.2.8 Section Types

We will now consider the types of behaviour that can be exhibited by the fibre between
contact points.
1. Unconfined Fibre (figure 2.29)

See figure 2.29. The boundary conditions and equations are the same as for
the unconfined fibre for example 1b (see page 52).
2. Fibre confined by the base (figure 2.30)

The boundary conditions are the same as for example 1b, page 53 except that
the upper bound on λ needs to be changed. To do so, we must first realise that
√
for contact just to occur, θ̄ = 1/2. If this occurs at the point λri = ζ1/2 , then
we wish to solve
ζ1/2 √
= p̄i tan λ · (1 − cos ζ1/2 ) + sin ζ1/2 + ζ1/2
2
√
to find the maximum λ. When this is reached, the fibre has touched the the
top surface, and we shall need a new type of solution.
3. Starting section (figure 2.31)

The starting wedge has the following boundary conditions:
θ̄(0) = 0
1
θ̄(di ) =
2
1
ȳ(0) =
2
1 di
ȳ(di ) = +
2 2
61
Figure 2.31: Starting section for a fibre in a wedge
Figure 2.32: The mid section for a wedge
1 ri 1 ri
− ≤ ȳ(ri ) ≤ + 0 ≤ r i ≤ di .
2 2 2 2
The boundary condition for ȳ(di ) fixes the value of p̄i to be
1 tan zi /2 − zi
p̄i = .
2 zi − 2 tan zi /2
The condition that the fibre remains within the wedge will fix the maximum
value of zi . As with the fibre confined only by the bottom plate, we need to
find the point of contact with the wedge wall, which occurs at position ci where
θ̄ = 1/2. The maximum value of zi satisfies the condition
1 ci
ȳ(ci ) = + .
2 2
4. Middle section (see figure 2.32)

The mid section has the following boundary conditions
1
θ̄(0) = −
2
1
θ̄(di ) =
2
ȳ(0) = −y0
di
ȳ(di ) = 2y0 +
2
and the analysis proceeds much as before.
62
Figure 2.33: The ending section for a wedge
5. End Section (see figure 2.33)

The end section has the following boundary conditions applied
1
θ̄(0) =
2
θ̄0 (di ) = 0
ȳ(0) = y0

di
ȳ(di ) = − 2y0 +
2
and the analysis is similar to the earlier end section.
2.2.9 Results
The analysis of these section types, and the continuity conditions, proceed much as
before. Once again, the symmetry present means that λ can be determined at the
end, as
√ X
λ= zi .
i
In addition, we can eliminate λ from both the joining conditions and boundary con-
ditions on the section types. Figure 2.34 shows the results of this analysis with a
plot of applied force against excess fibre. Note that there is less of a drop in λ as t̄ is
increased at the transition points (see figure 2.26).
63
t
1.75
1.5
1.25
1
0.75
0.5
0.25
2
12
4
8
10
6
λ
Figure 2.34: λ against t̄ for a fibre in a wedge
64
2.3 3D Confined Fibre
So far, our models for the fibre as it enters the crimper have concentrated on the
relationship between the forces exerted on the fibre and the extra length we have
added. They have not given any insight into another important physical property,
the three-dimensional nature of crimp. That is observed experimentally. If you take
a selection of tow as it comes out of the industrial crimper and examine a number
of fibres, you will see sections where the crimp is vertical, others where it is at some
slanted angle and still more when it forms a helix (see figure 2.35). The different types
of crimp may even be seen on the same fibre. For many applications, it is important
to know the amount of three dimensional crimp.15 Therefore some understanding as
to what causes the different configurations would be useful. To this end, we shall
attempt to extend the ideas we used to discuss two dimensional crimp and apply
them when the fibre has an extra direction in which to deform.
Our model problem will be a fibre confined between two horizontal plates. These
it can touch at multiple points (or over a region). At each end of the fibre we will
apply no moments, just equal and opposite thrusts along the undeformed direction
(see figure 2.36). The idea in this is to be as close to the configuration of the two-
dimensional example 1a (page 35) as is possible. We will begin by discussing the
equations of an elastica in three dimensions. We will then linearise these for the
case of small deflections. Under these circumstances the deformations in each of the
principle directions of the fibre are independent. This will allow us to use a very
similar method to solve the problem as we employed for the two-dimensional crimp.
Once again, we will produce plots of how the force varies with the amount of fibre
entering and discuss which configurations are observed in practice.
2.3.1 The equations of an elastica in three dimensions

16
Let the position of the fibre in the channel be denoted by r(s) where s is the distance
along the elastica. The orientation of the cross-section of the fibre is given by the
orthonormal vector pair d1 (s) and d2 (s), where d1 and d2 point along the directions
of the principle axes of the cross-section. We obtain the following relationship between
15
One use of crimped fibre is to make the filters for cigarettes. The more three dimensional the
crimp, the more tortuous the path through the filter and the more chemicals removed.
16
From Love[10]
65
Vertical
Slanted
Helical
Figure 2.35: Observed 3d crimp formations
66
z
xy
Figure 2.36: A three dimensional elastica confined in a narrow channel
67
the position of the fibre and the orientation
r0 (s) = d3 (s) = −d1 × d2 . (2.53)
As the triad {dk } is orthonormal, there exists a vector u, called the curvature vector,
such that
d0k = u × dk
and we choose to write this vector in the form
u = u k dk .
Let m(s) be the internal moment within the fibre at point s, and n(s) the internal
force. By balancing forces and moments, and in the absence of body forces and applied
moment loads we find
m0 + r 0 × n = 0 (2.54)
n0 = 0
The last equation shows that n is determined by the external forces. We now choose
to write that
m = m k dk
and assume the following linear constitutive relations
u1 = σ 1 m 1 ,
u2 = σ 2 m 2 ,
u3 = σ 3 m 3
where the σi are constant17 . For an arbitrary cross section, these constants may all
be different. Rearranging and combining with (2.54) we find
m01 + (σ2 − σ3 )m2 m3 − n · d2 = 0, (2.55)
m02 + (σ3 − σ1 )m3 m1 + n · d1 = 0, (2.56)
m03 + (σ1 − σ2 )m1 m2 = 0. (2.57)
A similar set of substitutions gives that
d01 = −σ2 m2 d3 + σ3 m3 d2 (2.58)
d02 = −σ3 m3 d1 + σ1 m1 d3 (2.59)
d03 = −σ1 m1 d2 + σ2 m2 d1 . (2.60)
We therefore have 6 vector quantities (r, d1 , d2 , u, m and n) and six vector equations
describing their evolution.
17
In Love’s Notation, σ1−1 = EIA , σ2−1 = EIB and σ3−1 = C
68
2.3.2 Relationship with the co-ordinates in fixed space
Let the fixed coordinates be denoted by {x, y, z} where the z-axis lies along the
undeformed fibre. The position of the fibre in the channel is hence written as
 
x
r=  y .
z
Let the undeformed fibre be aligned such that

     
sin φ cos φ 0
d1 =  cos φ  d2 =  − sin φ  d3 =  0 .
0 0 1
2.3.3 Boundary conditions

We will assume the boundary conditions that are equivalent to those used to describe
example 1a, a fibre confined to a box (see page 35) whose walls are located at x =
0, Zb , where is small. Firstly, we will assume that the ends of the fibre are both
free to rotate, such that they exert no moment on the fibre. Hence
m1 (s) = m2 (s) = m3 (s) = 0 when s = 0, L.
Secondly, we will require that at the contact points, m and d are continuous. Finally,
at a boundary point, the fibre lies along the boundary, so
x0 (s) = 0.
2.3.4 Non-dimensionalisation
We choose similar non-dimensionalisations as for the 2d problem (see page 30), choos-
ing Zb (the length of the box) as a typical length scale, σ1 as a typical beam constant
and obtaining other typical dimensions from suitable combinations of these such that
L = Zb (1 + 2t)
1
n = n̂
σ1 Zb2
1
mk = m̂k
σ1 Z b
σ2 = σ1 σˆ2
σ3 = σ1 σˆ3 .
69
We will also assume that the fibre has only a small deflection from its original direction
so
x = Zb x̂
y = Zb ŷ
z = Zb ẑ.
Our equations (2.55)-(2.60) become
m̂01 + (σˆ2 − σˆ3 )m̂2 m̂3 − n̂ · d2 = 0,

m̂02 + (σˆ3 − 1)m̂3 m̂1 + n̂ · d1 = 0,
m̂03 + (1 − σˆ2 )m̂1 m̂2 = 0,
and
dd1
= −σˆ2 m̂2 d3 + σˆ3 m̂3 d2
dŝ
dd2
= −σˆ3 m̂3 d1 + m̂1 d3
dŝ
dd3
= −m̂1 d2 + σˆ2 m̂2 d1 . (2.61)
dŝ
2.3.5 Linearisation
We will now seek to solve these equations with an externally applied compressive force
(of non-dimensional magnitude λ) and with a small component in the x direction.
Writing this in non-dimensional terms (having droped theˆs) as
 
−aλ
n= 0 
−λ
we now expand each property as a series around the undeformed state of the rod.
Hence
m1 = m11 + O(2 )
m2 = m21 + O(2 )
m3 = m31 + O(2 )
 
sin φ
d1 =  cos φ  + d11 + O(2 )
0
70
 
cos φ
d2 =  − sin φ  + d21 + O(2 )
0
 
0
d3 =  0  + d31 + 2 d32 + O(3 )
1
Substituting into the equations for mk and dk we find that, to order
m011 + λγ21 + aλ cos φ = 0 (2.62)

m021 − λγ11 − aλ sin φ = 0 (2.63)
m031 = 0
and
d011 = −σ2 m21 d30 + σ3 m31 d20

d021 = −σ3 m31 d10 + m11 d30
d031 = −m11 d20 + σ2 m21 d10
where  
αk1
dk1 =  βk1 
γk1
As m31 is constant along the section and continuous at contact points we can imme-
diately see that m31 = 0 everywhere. This in turn implies that
0 0 0 0
α11 = β11 = α21 = β21 =0 (2.64)
and hence the equations simplify further to :

0
γ11 = −σ2 m21 (2.65)
0
γ21 = +m11 (2.66)
d031 = −m11 d20 + σ2 m21 d10 (2.67)
Realising from (2.53) that
d031 = − (d011 × d20 + d10 × d021 )
and hence integrating and making the substitutions d11 = γ11 d30 , d21 = γ21 d30 18 we
find
d31 = −γ11 d30 × d20 − γ21 d10 × d30
18
This comes from integrating (2.64)
71
As {dk } are orthogonal, this in turn simplifies to
d31 = −γ11 d10 − γ21 d20 (2.68)
Writing that
r0 = r00 + r01
and remembering the relationship to the fibre orientation (2.53) we find
dx0 dy0
ds
= 0, dz
= 0, ds
0
= 1.
ds
 0  
  
x1 sin φ cos φ
 y10  = −γ11  cos φ  − γ21  − sin φ  . (2.69)
z10 0 0
2.3.6 Boundary Condition at s = 1 + 2t

As with the two dimensional fibre, we require that at s = 1 + 2t, z = 1 (i.e. the fibre
fills the box). We can rewrite this as
 
Z 1+2t 0
d3 ·  0  ds = 1.
0 1
Now let us use the series expansion for d3 , and write
 
Z 1+2t 0

d30 − γ21 d20 − γ11 d10 + 2 d32 ·  0  ds = 1.
0 1
Collecting terms, we see that this can only be satisfied if
t = 2 t̄
and  
Z 1 0
d32 ·  0  ds = −2t̄. (2.70)
0 1
Consider the second order expansion of (2.61). This can be written as
d032 = −m11 d21 − m12 d20 + σ2 m21 d11 + σ2 m22 d10 .
Taking the dot product of each side with the unit vector in the z direction and using
the results for d21 and d11 this simplifies to
 
0
0
d32 ·  0  = −γ21 γ21
0 0
− γ11 γ11 .
1
72
Integrating twice and substituting into (2.70) we arrive at
Z 1 2 2
γ11 γ21
2t̄ = + ds.
0 2 2
2.3.7 Summary of Linearized Equations

Hence using (2.65) to eliminate m11 and m21 from (2.62) and (2.63) we reach the
linearized equations we will be solving for γ11 , γ21 , x01 and y10 :
00
γ21 + λγ21 + aλ cos φ = 0, (2.71)
00
γ11 + σ2 λγ11 + aσ2 λ sin φ = 0, (2.72)
and
x01 = −γ11 sin φ − γ21 cos φ, (2.73)

y10 = −γ11 cos φ + γ21 sin φ. (2.74)
At the free ends, we apply no moments, but will apply sufficient twist to fix the angle
of inclination. At the contact point, the fibre must be tangent to the surface, and
there will be no moment applied. Hence at s = 0, 1
0 0
γ11 = γ21 =0 (2.75)
and at the contact points
γ11 sin φ + γ21 cos φ = 0, (2.76)

0 +
[γk1 ]− = 0, (2.77)
[γk1 ]+
− = 0. (2.78)
In addition, we have the global boundary condition

Z
1 1 2 2
t̄ = γ + γ21 ds. (2.79)
4 0 11
We will now solve these equations for simple geometries.
73
2.3.8 3D Confined Fibre : Results
We will now consider solutions of equations (2.71)-(2.74) with boundary conditions
at the ends of no applied moments (2.75). If contact points exist, they will satisfy
the conditions on tangency and continuity (2.76)-(2.78). Without loss of generality,
we can choose that
σ2 ≤ 1.
Let us begin by considering the initial deformation of the fibre before it makes contact
with any surface. This has no normal force acting on it, so a = 0. In addition, we
imagine that we are prescribing t̄ (the amount of excess fibre fed into the box), but
not the compressive force λ. First consider the equation for γ21 . This has the general
solution (with a = 0) of
√ √
γ21 = A sin λs + B cos λs.
In order to satisfy the boundary conditions (2.75) non-trivially we require that

√
A = 0 and B sin λ = 0.
Similarily for γ11 (2.72)

p p
γ11 = α sin λ̄s + β cos λ̄s
with p
α = 0 and β sin λ̄ = 0
where
λ̄ = σ2 λ.
These conditions have three possible outcomes, depending on the value of σ2 :
σ2 = 1 The elastica has no prefered direction in which to buckle.
Any direction will be the consequence of inhomogeneities in
the enviroment or the elastica itself. The amplitude is
determined by the value of t̄.
2 2
σ2 = m /n for some pair of integers m,n (m < n).
Buckles can occur in both directions, and λ = n2 π 2 . As
for the Euler-strut, the amplitudes of the buckles are again
determined by the value of t̄. Once again, the values are
not unique (as t̄ ∝ B 2 + β 2 ).
Otherwise The elastica will buckle in only one direction. As σ2 < 1,
the most energetically favourable buckle will be that which
occurs for the lowest λ, and hence the fibre will buckle
in the “easiest” direction (hence β = 0). Again, the
amplitude of the buckle must be determined by the value of t̄.
74
2.3.8.1 Normal contact φ = 0
We will now consider what happens when contact has occured. Our first case con-
sidered will be that of normal contact between the fibre and the plate, that is φ = 0.
We shall also assume that σ2 < 1, so that we imagine that the fibre has buckled in
the x direction until it has touched the top surface. It is now necessary to determine
the value of a, but as we know that the fibre is touching the top surface we can can
use the equations for the position of the fibre in the channel (2.73) to fix its value19 .
We will be solving the following pair of equations
00
γ21 + λγ21 + aλ = 0 (2.80)
00
γ11 + σ2 λγ11 = 0
0 0
γ21 (0) = γ21 (1) = 0 (2.81)
0 0
γ11 (0) = γ11 (1) = 0
at the ends, and at the contact point the free boundary conditions
γ21 = 0 (2.82)
0 +
[γ21 ]− = [γ21 ]+
− = 0 (2.83)
0 + +
[γ11 ]− = [γ11 ]− = 0
Now let us make use of the work we have already done for the two dimensional elastica.
Look at equations (2.80), (2.81), (2.82) and (2.83). These describe the same system
as for example 1a on page 35. We will therefore make full use of these solutions
in solving the problem. The remaining equations describe an unconfined euler strut,
whose solutions are once again well known. Realising that if t̄conf ined and t̄F ree are the
values of t̄ for the two dimensional confined elastica and the two dimensional Euler
strut respectively, then
R1 2
γ11 R1 2
γ21
t̄3d = 0 2
ds + 0 2
ds, (2.84)
= t̄F ree + t̄conf ined
It is simple to modify the plots of λ vs. t̄ to include this effect, as shown in figure 2.37
√
(for which we have taken σ2 = 2/5) Two curves are plotted. The first (black, dashed
19
This is the same technique we used to calculate the value of p¯i for the starting section for
example 1a (see page 37)
75
line), has deformation only in the y direction, and hence corresponds to the first mode
of buckle for the second direction. The second curve (red, dashed) is the same mode
superimposed on the solution of the two dimensional problem.
2.3.8.2 Energy in three dimensions
We have created a new type of solution. Now we need to find out if it is energetically
favoured. In section 2.2.5.3 we noted that the energy stored in the two dimensional
elastica was given by Z L
θ02
Energy = EI ds.
0 2
In a similar manner, and making the connection that EI = σ1−1 we have for the three
dimensional elastica
3
X Z L
m2k
Energy = σk ds.
k=1 0 2
Rewriting this in terms of the non-dimensional, linearised quantities we have
Z 1 0 2 Z 0 2
γ21 1 1 γ11
Energy = ds + ds. (2.85)
0 2 σ2 0 2
Again, this has the components from the confined elastica and free elastica added
together, and it is therefore simple to modify our plots of log(Energy) vs. t̄ to show
this (see figure 2.38). Note how, for σ2 = 0.25 the 3d configuration is energetically
favoured. This suggests that for even moderately non-circular cross-sections the fibre
adopts a crimp with components in the third direction.
We can also plot the position of the fibre as shown in figure 2.39. Notice how the
fibre is slanted rather than helical.20 . We will now look more closely at this type of
configuration for a more general orientation of the fibre (0 < φ < π/2).
2.3.8.3 General Fibre Orientation - Single contact Point
We will now solve the model, (2.71)-(2.78), for the case of a single contact point
(λ > π 2 ). In the same way as we solved the earlier, two-dimensional, problems, we
will divide the fibre into sections before and after the contact point and solve in these
regions separately. We will then apply the matching conditions to produce a solution.
20
This would not be the case if we had observed the second buckling mode for γ11 , however this
occurs for higher λ
76
30
25
t
20
15
10
3d Buckle
5
0
1750
1500
1250
1000
750
500
250
0
λ
√
Figure 2.37: λ vs. t̄ for a 3d elastica, σ2 = 2/5
77
20
17.5
t
15
12.5
10
7.5
3d Buckle
5
2.5
12
14
2
8
10
6
Log(Energy)
√
Figure 2.38: Energy vs. t̄ for a 3d elastica, σ2 = 2/5
78
x
Figure 2.39: Typical path for a 3d crimped fibre, σ2 = 0.25

.
79
1. First section
We have the boundary conditions
0 0
γ11 = γ21 = 0,
at the start of the fibre (s = 0), and
γ11 sin φ + γ21 cos φ = 0
at s = 1/2. Note that we have three boundary conditions for two second order
ODE’s, with one unknown parameter a (which we will denote a1 to associate it
with the first section). We can use equation (2.69) and the boundary conditions

1
x1 (0) = 0 and x1 =1 (2.86)
2
to fix the value of a1 , but we are still short of one boundary condition. We
will overcome this by introducing a new parameter that is constant between the
sections and use the joining conditions to obtain its value. To this end let us
write that
1 Γ
γ21 =
2 cos φ
for some yet-undetermined constant Γ. By (2.76)

1 Γ
γ11 =−
2 sin φ
Solving the equations and applying the boundary conditions gives
√
γ21 = B1 cos λs + a1 cos φ
p
γ11 = β1 cos λ̄s + a1 sin φ
where

1 Γ
B1 = − √ a1 cos φ −
cos λ/2 cos φ

1 Γ
β1 = − √ a1 sin φ +
cos λ̄/2 sin φ
p 2 √ 2
λ̄ = σ2 λ
Integrating (2.69) between s = 0 and s = 1/2 and applying (2.86) gives

√ √
a1 β1 sin φ λ̄ B1 cos φ λ
1+ =− √ sin − √ sin . (2.87)
2 λ̄ 2 λ 2
80
2. End section
For the end section, we apply the boundary conditions
0 0
γ11 (1) = γ21 (1) = 0,
at the end, and by continuity of γk1

Γ
γ11 = −
sin φ
Γ
γ21 =
cos φ
at s = 1/2. Applying these, we find
√ √ √
γ21 = B2 tan λ sin λs + cos λs + a2 cos φ,
p p p
γ11 = β2 tan λ̄ sin λ̄s + cos λ̄s + a2 sin φ,
where

1 a2 cos2 φ − Γ
B2 = − √ √ √
cos φ tan λ · sin λ/2 + cos λ/2
!
1 a2 sin2 φ + Γ
β2 = − √ √ √ .
sin φ tan λ̄ · sin λ̄/2 + cos λ̄/2
The value of a2 is found in the same way as for the start section by requiring
that
1
x(1) − x = −1.
2
This gives
√ " √ #
a2 β2 sin φ λ̄ p λ̄
−1 = − √ cos tan λ̄ − tan
2 λ̄ 2 2
√ " √ #
B2 cos φ λ √ λ
− √ cos tan λ − tan . (2.88)
λ 2 2
In (2.87) and (2.88) we have two equations from which we can deduce a1 (Γ) and
a2 (Γ). If we now apply
+
0 1
γ21 =0
2 −
we find that √ !
√ λ
B2 tan λ · cot − 1 = −B1 . (2.89)
2
81
Combining (2.87), (2.88) and (2.89) we obtain
√ √ √ √
2Γ λ sec φ tan λ/2 λ̄ − 2 tan λ̄/2
√ √ √ √ √ (2.90)
−2 λ̄ cos2 φ tan λ/2 + λ( λ̄ − 2 sin2 φ · tan λ̄/2)
This has the solution

Γ=0
unless √
p λ̄
λ̄ = 2 tan
2
or √
λ
tan = 0.
2
√
The first exception will only occur for λ > 2π, and hence we can ignore this case.
√
λ = 2π is more promising, as we know that this is an important parameter in the
two-dimensional case. Unfortunately, if we find Γ using the continuity condition
+
0 1
γ11 =0
2 −
we find √ √ √ √
2Γ λ̄ csc φ tan λ̄/2 λ − 2 tan λ/2
√ √ √ √ √ =0
−2 λ̄ cos2 φ tan λ/2 + λ( λ̄ − 2 sin2 φ · tan λ̄/2)
√
which requires that Γ = 0 for λ = 2π. From this, we conclude that Γ must always
be zero. The consequence of this is that the fibre never enters a helical configuration.
Figure 2.40 show typical configurations of the fibre for increasing force (with constants
φ = π/4, σ2 = 0.25).
82
x
y
Figure 2.40: 3d configurations for a confined fibre with increasing λ. φ = π/4,

σ2 = 0.25
83
2.4 Conclusions
In this chapter we have discussed models for the deformation of the fibre as it first
enters the stuffer box. We chose to model the fibre as an elastica contained in a box
of fixed length, and allowed the fibre to make multiple contacts with the sides. To
gain understanding of the problem, we concentrated on special case when the aspect
ratio of the box was small, and hence when we could use the linearised equations for
the elastica.
2.4.1 Two dimensional fibre

When we treated the fibre as a two-dimensional elastica, we were left with equations
that were simple to solve between the contact points. We then proceeded to discuss
a methodology for joining these solutions together to produce a full solution. This
was done for a simple rectangular box, a rectangular box fed midway and a wedge.
Different types of solution were observed in each case. We then applied energy argu-
ments to determine which configuration was energetically favoured, although we also
discussed the possible non-uniqueness of solutions created by this methodology.
We must also consider the results of this work on primary crimp formation as
part of an understanding of the machine as a whole. Firstly, we saw how the force
exerted by the fibre on the walls that confine it at the ends changes as more fibre
is forced in. Secondly, we saw that, for a given number of contact points, the force
reaches a maximum as more fibre enters, after which point the fibre will begin to form
a new contact point. For the rectangular box we can see that the maximum force is
quadratic in the number of contact points. In particular, when we apply no moment
to either end we had
EI
Fx = (M + 1)2 4π 2 (2.91)
Xb2
where M is the number of contact points (not counting the ends). In addition there
were (M + 1)/2 crimps,21 hence the crimp length was, on average
2Xb
. (2.92)
M +1
Note that for these boundary conditions, M had to be an odd number.
21
where one crimp goes from bottom to top to bottom again
84
For a rectangular box fed midway, the solutions containing sections with flat fibre
were seen to be energetically unfavoured. The maximum force was found to be
EI
Fx = (2M + 1)2 π 2
Xb2
and the average length of the crimp

2Xb
.
M +1
For the wedge, this is much less clear. In particular, the force is not exactly
quadratic in the number of contact points, nor is it as clear where the maximum force
occurs. However, for large numbers of sections, each contributes approximately the
same amount to the force, and has the same length, hence
EI 2
Fx ∝ M
Xb2
and the crimp length

2Xb
∼
M
Note that all these results are only valid for the small amplitude crimp. Although the
maximum force is likely to be similar, if there are large deflections the crimp length
may be very different.
2.4.2 Three dimensional crimp

Experimentally, we have seen that the fibre can be deformed into a three dimensional
crimped shape, especially a helix. Starting from the equations for a three dimen-
sional elastica, we showed that, for small deflections, the deformations in each of the
principal directions were independent except where they were coupled at the contact
points. We considered the geometry of a fibre confined between two plates. When
the principle axis of the fibre was parallel to the two plates, the fibre behaved as the
two-dimensional elastica until a critical force was reached when it adopted a three-
dimensional form. When the axis was at a general angle, the fibre was not confined
to a plane for any force once contact had been achieved. In neither case did the fibre
adopt a helical configuration.
85
Chapter 3
Secondary Crimp
3.1 Introduction
Having modelled the primary deformation that occurs as the material first enters
between the rollers, it is now time to turn our attention to the subsequent behaviour
and secondary crimp formation. To this end, we need to consider the behaviour of
the crimped tow as it is compressed.
• Hinged Rods Model

Looking at typical primary crimped material, we see a fairly uniform small scale
structure. We also see that much of the fibre is straight, with large changes in
angle occurring at the corners where the material has undergone an irreversible
plastic deformation. A similar structure can be formed by folding a piece of pa-
per in alternate directions to form a concertina shape. When we compress such
a structure, we see that as the force increases, the angle between adjacent rods
decreases. This observation leads us to propose our first model for the crimped
tow. We will describe the fibre as a series of rigid rods, joined by hinges. These
hinges will exert both reaction forces and torques on the rods and in the absence
of any applied force the tow will have a “concertina” shape (figure 3.1a).
By considering the forces and moments acting on such a system, we will derive
a second-order difference equation for the orientation of each rod. In order to
gain a better understanding of the behaviour of this model, we will look for an
asymptotic solution when the ratio of the couple exerted by external forces to
that exerted by the elastic hinges is small. The form of the solution thus found
is not valid for a sufficiently large numbers of rods, and we will then require a
different expansion.
86
(a) Hinged Rods
(b) Naturally Curved Fibre
Figure 3.1: The two ways of seeing secondary crimp
We show that the dependent variable in the equations derived for the “con-
certina” can be rewritten as a sum of two angles that vary over different length
scales. This enables us to seek solutions where the change in these angles be-
tween one hinge and the next is very small, and leads to a more convenient
asymptotic procedure that is uniformly valid for large numbers of rods. Unlike
the first expansion, this procedure will predict a bifurcation of the “concertina”
from a nearly flat configuration to an arched shape similar to the Euler strut
under certain circumstances. Both of these asymptotic solutions are also com-
pared with the numerical response diagrams calculated from solving the full
problem.
The hinged rods model has a number of parameters which are unknown. In or-
der to find these, experiments have been carried out by Acordis1 by stretching2
a crimped tow and measuring the force and extension3 . The model’s predictions
were compared with these, and a fit taken to obtain the parameters.
• Multiple Scale model

A second model for the crimped fibre is to think of it as an elastica that has a
curved shape when unstressed as in figure 3.1b. Starting from the equations of
such a fibre, and, using the small scale of the crimp in comparison to its length,
we will again carry out an analysis to derive an expression for the slow mod-
ulations that occur along the fibre; now however this analysis can be achieved
by the method of multiple scales. This predicts a buckling when a critical con-
dition is satisfied, which is similar to that for the classical Euler strut. We also
1
Formerly Courtaulds.
2
Pull rather than push.
3
Here we use extension as experimentally observed, i.e. the change in the macroscopic length of
the sample.
87
compare this result with that obtained from the above mentioned continuum
limit for the hinged rods model, and show the similarities in the structure of
the solutions.
88
3.2 Hinged Rods : Model
3.2.1 Derivation
Our model will make the following assumptions
• The tow consists of N rods, numbered n = 1 · · · N (see figure 3.2)
• The rods are rigid and inextensible.
• The only external forces acting on the rods are ±F applied at the end rods.
Thus each rod is acted on by forces ±F at the ends due to the internal reaction
in the hinges.
• All the rods are of the same length Λ.
• The hinges joining the rods exert a torque which resists deviation from the
equilibrium angle φ and is given by
K(Θn − φ)
where Θn is the acute angle between rod n and n + 1 and K is a spring constant
(units N m).
Consider the equilibrium of a single rod denoted by n(6= 1). It has two torques acting
on it due to hinges n − 1 and n. The externally applied force, transmitted to the rod
by the hinges, gives rise to a couple F × Λ, where Λ is a vector along the rod. Now
we see a potential complication with the notation so far. We have measured the angle
Θn relative to the previous rod, but it is more convenient to write θn to be the angle
between the nth rod and the direction of the external force (as shown in figure 3.2).
Careful examination of the geometry of the rods tells us that
Θn = π + (θn − θn+1 )(−1)n
where the (−1)n term differentiates between the odd and even hinges. In addition,
balancing the moments acting on the nth rod (n 6= 1) requires that, for equilibrium,
(−1)n (K(Θn−1 − φ) + K(Θn − φ)) = F Λ sin θn ,
where once again we must differentiate between odd and even hinges. Hence we arrive
at an equation necessary for the moments on the nth rod to be balanced as
K (θn+1 − 2θn + θn−1 − 2(−1)n (π − φ)) + F Λ sin θn = 0 for n = 2, · · · , N − 1.
89
F
N
n+1
−θ n+1
Tn
Hinge n
Θn
Hinge n-1
θn
n
T n-1
n-1
Λ
Hinge 1
1
θ
Rod 1
Figure 3.2: Hinged Rod configuration for an even number of rods (N )
90
Note that this does not apply to the end rods, where only one hinge exerts a torque.
For these, we have the conditions
K (θ2 − θ1 + π − φ) + F Λ sin θ1 = 0,

K θN −1 − θN − (−1)N (π − φ) + F Λ sin θN = 0.
3.2.2 Non-dimensionalization
We can non-dimensionalize by writing
FΛ
K=
κ
and dividing. We also write
χ = π − φ.
This gives the following system of equations with which we shall deal from now on
θn+1 − 2θn + θn−1 + κ sin θn = 2(−1)n χ (3.1)

for n = 2, · · · , N − 1
θ2 − θ1 + κ sin θ1 = −χ (3.2)
θN −1 − θN + κ sin θN = (−1)N χ. (3.3)
91
3.2.3 Asymptotics
Equation (3.1) is a nonlinear second order difference equation, and, as such, is difficult
to solve. In absence of the (−1)n term, we might be able to take a continuum limit as
κ → 0 to obtain an ODE for θ, but we do not expect it to be possible when the angle
θn is both O(1) and can change rapidly. Instead, we will make use of the assumption
that4 κ 1 and seek a solution to (3.1) with boundary conditions (3.2) and (3.3) in
the form
θn = θ0,n + κθ1,n + κ2 θ2,n + · · · .
Expanding and collecting terms of the same magnitude we see that
• O(1)
θ0,n+1 − 2θ0,n + θ0,n−1 − 2(−1)n χ = 0, (3.4)

θ0,2 − θ0,1 + χ = 0, (3.5)
θ0,N −1 − θ0,N − (−1)N χ = 0. (3.6)
• O(κ)
θ1,n+1 − 2θ1,n + θ1,n−1 + sin θ0,n = 0, (3.7)

θ1,2 − θ1,1 + sin θ0,1 = 0. (3.8)
θ1,N −1 − θ1,N + sin θ0,N = 0. (3.9)
• O(κ2 )
θ2,n+1 − 2θ2,n + θ2,n−1 + θ1,n cos θ0,n = 0, (3.10)

θ2,2 − θ2,1 + θ1,1 cos θ0,1 = 0, (3.11)
θ2,N −1 − θ2,N + θ1,N cos θ0,N = 0. (3.12)
3.2.3.1 O(1)
The general solution to (3.4) is

χ
θ0,n = α0 n + β0 − (−1)n . (3.13)
2
Applying boundary conditions (3.5) and (3.6) requires that
α0 = 0
but imposes no conditions on β0 , for which we will need to look to higher order terms.5
4
Although it is difficult to estimate κ directly, we shall see later that this assumption leads to
pysically relvant solutions
5
A physical interpretation for this is that the O(1) problem corresponds to the case when no
force is applied, and hence there is no direction imposed by the model
92
3.2.3.2 O(κ)
Applying this result to (3.7) we have

χ χ
θ1,n+1 − 2θ1,n + θ1,n−1 + sin β0 cos − (−1)n sin cos β0 = 0.
2 2
Now seeking a solution of the form
θ1,n = A1 (−1)n + B1 n2 + α1 n + β1 (3.14)
and comparing like terms we see that

1 χ
A1 = − sin cos β0 ,
4 2
1 χ
B1 = − cos sin β0 .
2 2
Applying boundary conditions (3.8) and (3.9) gives
1 χ 1 χ
α1 −
cos sin β0 + sin cos β0 = 0,
2 2 2 2
(−1)N χ 2N + 1 χ
−α1 − sin cos β0 + cos sin β0 = 0.
2 2 2 2
If N is even then
χ
N cos
sin β0 = 0,
2
χ χ
2α1 + sin cos β0 − (N + 1) cos sin β0 = 0;
2 2
otherwise
χ χ
sin cos β0 + N cos sin β0 = 0,
2 2
χ
2α1 − (N + 1) cos sin β0 = 0,
2
giving the solutions
• N even
β0 = 0, (3.15)
1 χ
α1 = − sin . (3.16)
2 2
• N odd
1 χ
tan β0 = − tan , (3.17)
N 2
N +1 χ
α1 = cos sin β0 . (3.18)
2 2
93
3.2.3.3 O(κ2 )
Once again, we have been unable to determine the value of the constant β1 . Using
our results so far, and applying them to equation (3.10) gives
θ2,n+1 − 2θ2,n + θ2,n−1 = (3.19)

χ 1 χ 1 χ

n n 2
cos − (−1) + β0 · α1 n + β1 − sin cos β0 (−1) − cos sin β0 n .
2 4 2 2 2
Considering the solution to the associated homogeneous equation and boundary con-
ditions (which is β2 ) we see, by the Fredholm Alternative, (3.19) will only have a
non-trivial solution if
N
X χ
n
0 = β2 cos − (−1) + β0 ·
n=1
2

1 χ n 1 χ 2
α1 n + β1 − sin cos β0 (−1) − cos sin β0 n .
4 2 2 2
Hence we find a value for β1 .
Note that these solutions are not valid unless κN 2 1. In order to redress this
problem, we will seek an alternative asymptotic procedure that will be uniformly valid
as N increases.
3.2.4 A reformulation of the problem

We want to find an asymptotic solution to the equations that is uniformly valid for
all N . We have seen that writing the solution as a series in κ is insufficient when
N 2 > O(κ−1 ). To understand the difficulty, let us consider the simpler difference
equation
yn+1 − 2yn + yn−1 + κ sin yn = 0. (3.20)
Seeking a solution where y = y0,n + κy1,n + · · · will only pick up the trivial solution
y ≡ 0. Instead, we assume that y is sufficiently smooth that
n 1
y0,n ' y0 (x) where x= and κ = 2 κ0
N N
and hence
y0,n+1 − 2y0,n + y0,n−1 d2 y0
∼ + O(κ).
1/N 2 dx2
For κN 2 = O(1) our difference equation can therefore be approximated as
y000 + κ0 sin y0 = 0
94
which will have a non-trivial solution.
Unfortunately, the (−1)n term in (3.1) precludes such a procedure. Instead, let
us take our inspiration from the form of the solutions encountered above ((3.13)
and (3.14)) and write that
θn = Σn − (−1)n ∆n
for some functions Σn and ∆n that are as yet undefined6 , but that we shall insist vary
gradually in n. Now let us insert this into the equation (3.1) to give :
(Σn+1 − 2Σn + Σn−1 ) + (−1)n (∆n+1 + 2∆n + ∆n−1 )

+κ cos ∆n sin Σn − (−1)n κ cos Σn sin ∆n = 2(−1)n χ.
Grouping together terms of same type ((−1)n or 1) we define Σn , ∆n by letting
Σn+1 − 2Σn + Σn−1 + κ cos ∆n sin Σn = 0 (3.21)

and ∆n+1 + 2∆n + ∆n−1 − κ cos Σn sin ∆n = 2χ. (3.22)
We now seek a solution that is valid when N is large and following our discussion
of (3.20) assume that κN 2 ∼ O(1). Writing7
1
=
N
κ = 2 κ0
Σn ∼ Σ(1) 2 (2)
n + Σn + · · ·
∆n ∼ ∆(0) (1)
n + ∆n + · · ·
and inserting into our equations gives
• From equation (3.21)

(1) (1) (1) 2 (2) (2) (2)
Σn+1 − 2Σn + Σn−1 + Σn+1 − 2Σn + Σn−1 +

(3) (3) 3 0
3 Σn+1 − 2Σ(3) n + Σ (0)
n−1 + κ cos ∆n · Σn +
(1)

4 κ0 − sin ∆(0)
n · ∆ (1) (1)
n Σ n + cos ∆ (0) (2)
n Σ n = O(5 ). (3.23)
6
There are physical interpretations for this tactic. We can obtain the same pair of equations by
writing the difference equation for odd and even hinges separately, or by combining the equations
for hinge n and n + 1 together. However, each of these forces choice of Σn and ∆n . Our purpose in
choosing to write θ in this way is that it be composed of smooth functions and that neither Σ n nor
∆n vary appreciably as n is increased to n + 1
7
If we were to take Σn ∼ O(1) then we would get an equation similar to (3.28) and with Neumann
(0)
boundary data, giving the solution Σn = 0.
95
• From (3.22)

(0) (0)
∆n+1 + 2∆(0)
n + ∆ n−1 − 2χ +

(1) (1) (1) 2 (2) (2) (2)
∆n+1 + 2∆n + ∆n−1 + ∆n+1 + 2∆n + ∆n−1 +

(3) (3) (4) (4)
3 ∆n+1 + 2∆(3) n + ∆ n−1 + 4
∆ n+1 + 2∆ (4)
n + ∆ n−1 −
2 κ0 sin ∆(0) − 3 cos ∆(0)

n · ∆n −
(1)
n
1
4 κ0 − sin ∆(0) (1) 2 (0) (2)
n · (∆n ) + cos ∆n · ∆n −
2

1
cos ∆n · (Σn ) = O(5 ).
(0) (1) 2
(3.24)
2
• From the boundary condition at n = 1 (3.2)

(0) (0)
− ∆2 + ∆ 1 + χ +

(1) (1) (1) (1)
Σ 2 − Σ 1 − ∆2 + ∆ 1 +

(2) (2) (2) (2)
2 Σ 2 − Σ 1 − 2 ∆2 + ∆ 1 +

3 (3) (3) 3 (3) (3)
Σ 2 − Σ 1 − ∆2 + ∆ 1 +
(0) (0) (1) (1)
2 κ0 sin ∆1 + 3 κ0 cos ∆1 · (Σ1 + ∆1 ) +

4 0 (0) (2) (2) 1 (0) (1) (1) 2
κ cos ∆1 · (Σ1 + ∆1 ) − sin ∆1 · (Σ1 + ∆1 )
2
5
= O( ). (3.25)
• From the boundary condition at n = N (3.3).

(0) (0)
(−1)N ∆N −1 + ∆N − (−1)N χ +

(1) (1) (1) (1)
ΣN −1 − ΣN + (−1)N ∆N −1 + ∆N +

2 (2) (2) N 2 (2) (2)
ΣN −1 − ΣN + (−1) ∆N −1 + ∆N +

(3) (3) (3) (3)
3 ΣN −1 − ΣN + (−1)N 3 ∆N −1 + ∆N −
(0) (0) (1) (1)
(−1)N 2 κ0 sin ∆N + 3 κ0 cos ∆N · (ΣN − (−1)N ∆N ) +
2
4 0 (−1)N (0) (1) (1)
κ sin ∆N · ΣN − (−1)N ∆N
2

(0) (2) N (2)
+ cos ∆N · ΣN − (−1) ∆N = O(5 ). (3.26)
We now follow the example used for equation (3.20) and write
(m) (m) d2 Σ(m)
Σn+1 − 2Σn(m) + Σn−1 ∼ 2
ds2
96
(m) dΣ(m)
Σn+1 − Σn(m) ∼
ds
(m) (m)
∆n+1 + 2∆n(m) + ∆n−1 ∼ 4∆(m) (s) + O(2 )
(m)
∆n+1 + ∆n(m) ∼ 2∆(m) (s) + O()
where
s = n ≤s≤1
and group accordingly.8 Hence we have
• From (3.24), (3.25) and (3.26) to O(1)
4∆(0) (s) = 2χ
2∆(0) (0) = χ
2∆(0) (1) = χ
which gives
χ
∆(0) (s) = (3.27)
2
Now looking at equation (3.23) to O(3 ) we have
d2 Σ(1) χ
2
+ κ0 cos · Σ(1) (s) = 0 (3.28)
ds 2
Solving (3.28) for Σ(1) (s) gives9

Σ(1) (s) = A(1) sin λ(1) s − ω (1) (3.29)
where r
(1) χ
λ = κ0 cos . (3.30)
2
The boundary conditions come from the O(2 ) from equations (3.24), (3.23),
(3.25) and (3.26)
4∆(2) (s) = κ0 sin ∆(0) (s)

dΣ(1)
− 2∆(2) (0) = −κ0 sin ∆(0) (0)
ds s=0
dΣ(1)
− + 2(−1)N ∆(2) (1) = (−1)N κ0 sin ∆(0) (1)
ds s=1
8
This is possible as we have the freedom to define Σ and ∆. It would not have worked if we had
limited ourselves by writing (for example)
θn+1 + θn θn+1 − θn
Σn = ∆n = (−1)n .
2 2
9
Notice the similarity to the Euler Strut equation. We will return to this idea later in this chapter
97
Hence
1 1 χ
∆(2) (s) = κ0 sin ∆(0) (s) = κ0 sin (3.31)
4 4 2
dΣ(1) 1 χ
= −κ0 sin (3.32)
ds s=0 2 2
(1)
dΣ 1 χ
and − = (−1)N κ0 sin . (3.33)
ds s=1 2 2
Conditions (3.32), (3.33), when applied to equation (3.29) give different results
depending on whether N is even or odd:
– N even
We require that
dΣ(1) dΣ(1)
=
ds s=0 ds s=1
and hence
λ(1) λ(1)
sin · sin − ω (1) = 0,
2 2
so we either have that
λ(1) = 0, ±2π, ±4π etc.
or
λ(1) − 2ω (1) = 0, ±2π, ±4π etc.
– N odd
This time, we need
dΣ(1) dΣ(1)
=−
ds s=0 ds s=1
and hence
λ(1) λ(1)
cos · cos − ω (1) = 0,
2 2
so we either have that
λ(1) = ±π, ±3π etc.
or
λ(1) − 2ω (1) = ±π, ±3π etc.
In addition, we have that
dΣ(1) κ0 χ
= − sin .
ds s=0 2 2
98
Since λ(1) is a physical parameter, this is sufficient to determine both ω (1) and
A(1) (although not uniquely) except when λ(1) is an appropriate multiple of π.
Thus we could now plot a bifurcation diagram for Σ(1) in terms of parameter
λ(1) and can expect that including nonlinear terms would produce pitchfork
bifurcations at λ(1) = nπ.
3.2.5 Results of the asymptotic analysis

Our asymptotic analysis has produced the following results:
• Using a regular peturbation we predicted a unique solution for small κ. This

was valid only when N 2 κ O(1).
• We then showed that our equations could be rewritten as a pair of second-order

difference equations for two smoothly varying functions. This allowed us to
approximate these with an ordinary differential equation when N >> 1 and
N 2 κ = O(1).
• The solution of these equations predicted that the observed behaviour depends
upon whether the number of hinges is odd or even. In particular, there exist
bifurcation points at values of λ(1) for which the linear solution is not uniquely
defined. In dimensional variables these are located at
χ
N 2 κ cos = 4π 2 , 16π 2 etc. (3.34)
2
when N is even, and
χ
N 2 κ cos = π 2 , 9π 2 etc. (3.35)
2
when N is odd.
We will now compare these predictions with the numerical solution of equations (3.1), (3.2)
and (3.3).
99
3.2.6 Numerical solution of the hinged rods model
Let us now consider the numerical solution of the full problem given by equations (3.1), (3.2)
and (3.3). Our purpose in doing thus is to confirm (or otherwise) the predictions made
by the asymptotic consideration discussed in the previous sections. The procedure we
will adopt is very simple. We define a function F (θ1 ; κ, N, χ) in the following manner
1. Assume θ1 for given κ, N and χ, then use (3.2) to calculate θ2 .
2. Now we can repeatedly apply (3.1) to obtain θ3 , · · · θN .
3. Looking at the end condition (3.3), we set F (θ1 ; κ, N ) to be given by
F (θ1 ; κ, N ) = θN −1 − θN + κ sin θN − (−1)N χ. (3.36)
The roots of the equation F = 0 are values of θ1 that will generate valid solutions to
the full problem. When implementing this procedure, we can consider two separate
techniques to locate the zeros of this function. The simplest is to look at the surface
generated by F (θ1 ; κ, N ) for fixed N and varying θ1 and κ, and find the contours
of F = 0. Such a scheme was implemented in Matlab, and typical results shown in
figures 3.3 and 3.4
Alternatively, we can use a standard root finding algorithm for suitable starting
guesses. This gives the position of the roots more accurately, and generates full
solutions (sets of θn for n = 1, · · · N ) for plotting (see figures 3.5 and 3.7)
3.2.6.1 Comparisons and Observations
Having developed a procedure through which numerical solutions may be found, we

are now in the position to make comparisons with the asymptotic solutions found
previously. Figures 3.3 and 3.4 plot the values of θ1 that correspond to solutions for
• the numerical solution (blue lines),
• the asymptotic expansion in powers of κ (red lines - given by (3.14) and (3.13)),
and
• the asymptotic expansion in powers of (black lines - given by (3.27), (3.31)

and (3.29)).
100
θ1
3
0
χ=π−0.9
N=20
−1
−2
−3
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
θ1 3
0
χ=π−0.9
N=21
−1
−2
−3
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 3.3: Comparison of asymptotics and Numerical Solution for N = 20 and

N = 21
101
θ1
3
χ=π−0.9
−1
N=100
−2
−3
1 2 3 4 5 6 7 8 9 10
−3
x 10
θ1 3
χ=π−0.9
−1
N=101
−2
−3
1 2 3 4 5 6 7 8 9 10
−3
x 10
Figure 3.4: Comparison of asymptotics and Numerical Solution for N = 100 and
N = 101
102
π2
N κcrit obtained numerically κ = N 2 cos χ/2
5 0.8412 0.9076
7 0.6482 0.4631
9 0.3351 0.2801
11 0.2092 0.1875
13 0.1447 0.1343
15 0.1065 0.1008
17 0.0819 0.0785
19 0.0651 0.0629
21 0.0531 0.0515
23 0.0441 0.0429
25 0.0372 0.0363
27 0.0316 0.0311
29 0.0274 0.0270
Table 3.1: Comparison of the numerically measured and the asymptotically predicted
values of κ for the first bifurcation point for N rods
(Note that for all figures and table 3.1 we have taken χ to be π − 0.9).
Note the close correlation between the numerical and κ expansion solutions. For very
small κ, we see a good match, much as we expected. However, the match between
the numerical solution and the expansion in powers of is much worse, especially at
small κ. This can be attributed to our assuming that κ ∼ 2 which limits the validity
of this solution. Indeed, figure 3.4 (which has a smaller value of ) shows a much
better agreement.
Figures 3.3, 3.4, 3.5 and 3.7 all show bifurcations for certain values of κ. For N
odd, we predicted that the first such bifurcation should occur when
π2
κ=
N 2 cos χ/2
(from equation (3.35)). Table 3.1 shows a comparison between measured (from plots
such as 3.3) and asymptotic values of the critical value for κ. This shows good
agreement between predicted and actual bifurcation points, especially as N increases.
Our numerical results also underline the important difference between odd and
even numbers of rods. Firstly, we see that, as predicted, there are bifurcations near
the predicted values of κ (when λ(1) is an odd multiple of π for N odd, and an even
multiple of π when N is even). Secondly, the multiples of π for which there are no
predicted bifurcations seem to have a fold bifurcation. This is similar to the symmetry
breaking that is observed for the Euler strut when gravity is present.
103
(e)
(f)
(d)
0.4
κ
φ=0.9
0.3
0.2
0.1
0
-2
2
-1
0
-3
3
Σ
(b)
(c)
(a)
Figure 3.5: Σ0 against κ in the case N = 20, φ = 0.9
104
(b)
(d)
(f)
(a)
(c)
(e)
Figure 3.6: Key to Figure 3.5
105
(d)
(e)
0.4
κ
0.3
φ=0.9
0.2
0.1
(b)
(c)
0
-2
2
-1
0
-3
3
(a)
Σ
Figure 3.7: Σ0 against κ in the case N = 21, φ = 0.9
106
(b)
(d)
(a)
(c)
(e)
Figure 3.8: Key to Figure 3.7
107
3.3 Hinged Rods : Fitting Experimental Curves
The hinged rods model requires the following parameters with which to calculate the
length of primary crimped material that will buckle
• Applied force F
• Rod length Λ
• Equilibrium Angle φ
• Hinge constant K
Of these, only the first two can be readily estimated in a practical situation, the force
because it must be the same as that buckling the material at the nip and the rod
length as that is half the primary crimp length (chapter 2). Only by modelling the
plastic deformation of the fibres can we begin to get a theoretical understanding of
how they depend on the past history.
The only realistic possibility is to use the results from experiments to fix these
values for typical operating regimes, and then assume that they are constant over the
parameter range we believe the crimper to work in. Suitable experiments are easy to
propose. One of the simplest to perform is a force-extension measurement, for which
the apparatus is available.
We apply the hinged rods model to the problem of a crimped tow under tension.
Consider the tow as comprising M individual fibres, with each having N hinged rods.
In terms of the angle made by the rods with the horizontal (θ) we see that the total
length of the tow is given by
n=N
X
Length of tow = Λ cos θn
n=1
We also know that if the rods are in equilibrium, then10

FΛ
θn+1 − 2θn + θn−1 − 2(−1)n (π − φ) − sin θn = 0. (3.37)
K
Now let us look for a uniform solution where θn+1 = −θn = −(−1)n θ. From (3.37)
we have that
n FΛ
(−1) −4θ + 2(π − φ) − sin θ = 0
K
10
Note that the tow is being elongated and hence the sign of the force F has changed
108
or
FΛ
sin θ = −4θ + 2(π − φ).
K
The total tension in the tow is therefore
M K 2(π − φ) − 4θ
, (3.38)
Λ sin θ
where M is the number of fibres in the tow. As the angles are constant, we also have
that the extension is
φ
N Λ cos θ − sin , (3.39)
2
and, when we apply no forces we have that
1
θ = (π − φ).
2
3.3.1 Fitting Experimental Data

In contrast to the situation where the model is used as part of a larger description of
an industrial crimper, when we consider an experimental run where we take a sample
of tow and measure the extension against tension. In this case we only know the
following11
• M From the specification of the fibre line
• a series of Force–Extension pairs.
We wish to find the parameters φ, N Λ and K/Λ. A difficulty arises as our model
gives tension and extension in terms of the parameter θ which is not measured in the
experiment and so the normal curve fitting routines are not easy to apply. Instead
we use information from (3.38) for special values of θ and (3.39).
Consider the situation as the parameter θ decreases. As it tends to 0, the tension

approaches ∞ and the extension reaches its maximum value

φ
N Λ 1 − sin .
2
Dividing through by the original length we get
Maximum length 1
= −1
Original length sin φ/2
11
As the fibre is crimped, we do not know the total length (N Λ), nor the number of crimps N .
The experiments were performed by Courtaulds using standard force-extension measuring devices.
109
Tension (N)
14
12
10
0
0 100 200 300 400 500
Extension (mm)
Figure 3.9: Fitting Data
from which we can deduce φ, and hence, from the original length, N Λ.
At the other extreme, consider the case when θ = (π − φ)/2 − , where 1.

Expanding (3.38) and (3.39) in powers of we get

φ 2 φ
extension = N Λ cos − sin + O(3 ),
2 2 2
2

4M K φ
tension = + tan + O(3 ).
Λ cos φ/2 2 2
Thus we expect the slope of the tension-extension near the origin to be given by
4KM
+ O().
N Λ2 cos2 φ/2
This slope can be measured either graphically, or by fitting the data for low extension
to a linear function. We now have sufficient information to infer K/Λ.
3.3.1.1 Typical Results
Figure 3.9 shows a typical set of experimental force-extension data (black line). The
red lines correspond to the limiting value of extension and a fit to the slope of the
curve for small force. The mauve line is the predicted shape based on the hinged rods
model discussed above. Some points to note :
• The fit is very poor, especially for medium extension values
110
• Finding the maximum value of the extension is very difficult from this curve,
as the tow had not been extended enough. The tow was discarded at the end
of the experiment, preventing any further measurments being made.
Can we improve on this? Physically, we might expect the curve not to hold when the
rods are nearly straight. In this situation, the assumption of rigidity (i.e. that they
do not extend or bend) is far from the actual situation. This is particularly important
for the fibres in tow, which are relatively elastic. The apparent “asymptote” of the
experimental data is not vertical, so we expect that in practical secondary crimp,
extension will become important.
3.3.2 Extensible, Unbending rods

Consider the tow comprising a number of unbending, extensible rods joined by elastic
hinges. We suppose a linear relationship between tangential force and extension of
the rod, such that if τn is the force along the rod and Λn the length, then
τn
Λn = Λ 0 1 +
E
for an elastic constant E. Once again, the hinges are such that the torque (Tn ) is
K(Θn − φ).
Now consider the equilibrium of the rods. Balancing moments gives

F Λn
θn+1 − 2θn + θn−1 − 2(−1)n (π − φ) − sin θn = 0 (3.40)
K
whilst a force balance gives
Λn − Λ 0
F cos θn = E . (3.41)
Λ0
Let us now seek a solution such that
Λn = Λ,
θn = −(−1)n θ.
This transforms (3.40) into

FΛ
4θ − 2(π − φ) + sin θ = 0.
K
Solving for the tension we get
K 2(π − φ) − 4θ
F = (3.42)
Λ sin θ
111
so that the tension is
M K 2(π − φ) − 4θ
.
Λ sin θ
A simple rearrangement of (3.41) gives

F cos θ
Λ = Λ0 1 + (3.43)
E
and the extension is
π φ
N Λ cos θ − Λ0 cos − .
2 2
Eliminating F from (3.42) and (3.43) gives
s
Λ0 Λ20 Λ0 K (2(π − φ) − 4θ) cos θ
Λ= ± +
2 4 E sin θ
and we require the larger value (as we expect the rods to elongate). So, in summary
we have that
s
Λ0 Λ20 Λ0 K (2(π − φ) − 4θ) cos θ
Λ = + + , (3.44)
2 4 E sin θ
M K 2(π − φ) − 4θ
Tension = , (3.45)
Λ sin θ

π φ
Extension = N Λ cos θ − Λ0 cos − . (3.46)
2 2
As E becomes large, Λ → Λ0 and we return to the original equations for the rigid
rods.
3.3.3 Fitting the data

Let us consider the behaviour of (3.44)-(3.46) for the cases when θ is small and when
θ is near to π/2 − φ/2.
3.3.3.1 Large Tension
We write θ 1 and expanding (3.44) we get

r p
KΛ0 2(π − φ) Λ0 √
Λ∼ √ + + O( θ). (3.47)
E θ 2
Using this, and substituting in (3.45) gives the tension as
s
2EK(π − φ) 1 EM √
∼M √ + + O( θ) (3.48)
Λ0 θ 2
112
and (3.46) gives the extension as
r
2Λ0 K(π − φ) 1 N Λ0 φ √
∼N √ + 1 − sin + O( θ). (3.49)
E θ 2 2
Now we are in the position to be able to calculate the slope of the tension-extension
graph for large values of extension which is
ME √
+ O( θ).
N Λ0
This suggests that for large forces, the slope is governed by the extensional properties
of the fibre itself as one would expect on physical grounds.
Now consider the intercept of the asymptote given by (3.48) and (3.49). Simple
manipulation gives that, as θ → 0, the intercept on the extension axis is
N Λ0 φ
− sin
2 2
which gives us information about the equilibrium angle φ.
3.3.3.2 Small Tension (θ = π/2 − φ/2 − )
We write θ = π/2 − φ/2 − where 1 and we again expand (3.44)-(3.46) in a series

in powers of to get
4K φ
Λ ∼ Λ0 + tan + O(2 )
E 2
4KM φ
Tension ∼ sec + O(2 ) (3.50)
Λ0 2

φ 4K sin2 φ/2
Extension ∼ N Λ0 cos + + O(2 ). (3.51)
2 E cos φ/2
Hence the slope as θ → π/2 − φ/2 is given by
4KM

Λ20 N cos2 φ/2 + 4K
EΛ0
sin2 φ/2
which again reverts to that for the inextensible rods when K/(EΛ0 ) is small.
3.3.3.3 Numerical Fitting
From the above analysis, we have found the following quantities

ME
End Slope ∼ (3.52)
N Λ0
113
Tension (N)
12
10
100 200 300 400 500

Extension (mm)
Figure 3.10: A comparison of a theoretical and experimental curve

φ
End Intercept ∼ M E sin − 1 (3.53)
2
ME
4K N Λ0
Start Slope ∼ (3.54)
Λ0 E cos2 φ/2 + Λ4K
0E
sin2 φ/2
which can easily be estimated from the experimental data. We already know that if
we now make a guess for the value of E, and use experimental values for (3.52), (3.53)
K
and (3.54) we can evaluate N Λ0 , φ and Λ0 E
. We can now assess the accuracy of the
model as follows :
1. Calculate pairs of tension–extension points for a range of θ from the extensible

hinged rods model with the values of the parameters described above.
2. Construct an interpolating function F (extension) that passes through these

points.
3. Define the error as

X
(tensioni − F (extensioni ))2
i
where ()i denotes a data-point.
We can now treat this error as a nonlinear function of M E, and we can search for a
minimum in this function. For our trial dataset, we find a minimum for M E ' 1000.
A comparison of the predicted and dataset curve is shown in figure 3.10.
114
3.4 Naturally-curved Fibres
The second model of the crimped fibre that we will discuss treats the fibre as an elas-
tica whose unstressed configuration is curved on a short length-scale (see figure 3.11).
Let us denote θ(s) ∼ 0(1) to be the angle between the tangent to the curve at point
s and the direction of an externally applied force F. In addition, we will describe the
natural shape of the fibre by Ξ(s) such that, when no forces are applied
θ(s) = Ξ(s).
We consider the case when the natural shape varies over a very short length scale .
If we let t be the distance along the fibre at this scale, then
s
t= and we write Ξ(s) = α(t).

For convenience we will also choose α to be periodic in t, such that
α(t) = α(t + 1),

Z 1
sin αdt = 0, (3.55)
Z 01
cos αdt = Υ, (3.56)
0
where Υ is the ratio of the horizontal extent of the curved fibre to the fully extended
length and the integral is taken over one period of the fibre. Equation (3.55) implies
that, in the unstressed configuration, the fibre lies along the x-axis when viewed over
a long length scale.
The relationship between the moment M (s) and curvature is given by

dθ dΞ
M (s) = EI − ,
ds ds
where E and I are the Young’s modulus and moment of area as in the usual Euler-
strut notation. Balancing moments and forces, and writing
F
λ2 =
EI
as the non-dimensional force, we arrive at the equilibrium equation
d2 θ d2 Ξ
− 2 + λ2 sin θ = 0. (3.57)
ds2 ds
115
ε
F F
εΥ
θ
x
Figure 3.11: A smoothly curved fibre
3.4.1 Multiple scales analysis

Because there are two length scales s and t in this problem we can use the multiple
scales method to analyse this problem (3.57). We write that
d ∂ 1∂
= +
ds ∂s ∂t
and seek a solution of the form
θ(s) = θ0 (s, t) + θ1 (s, t) + 2 θ2 (s, t)
where we assume that θn (s, t) does not grow in t.

Expanding (3.57) as a series in and equating like powers of we have, (if λ ∼ O(1))
to O(−2 )
θ0tt − αtt = 0; (3.58)
to O(−1 )
θ1tt + 2θ0st = 0; (3.59)
to O(1)
θ2tt + 2θ1st + θ0ss + λ2 sin θ0 = 0. (3.60)
116
After imposing that θ does not grow indefinitely as t → ∞, (3.58) has the solution
θ0 = α(t) + B0 (s).
Substituting into (3.59) we find
θ1tt = 0 ⇒ θ1 = A1 (s)t + B1 (s).
However, our desire that θ be periodic in t means that A1 (s) = 0. Substituting now
into (3.60) we obtain
−θ2tt = B0ss + λ2 sin B0 cos α(t) + λ2 cos B0 sin α(t) (3.61)
By the Fredholm Alternative, if there is to be a non-trivial solution of (3.61) bounded

as t → ∞, then we require
Z 1

B0ss + λ2 sin B0 cos α(t) + λ2 cos B0 sin α(t) dt = 0.
0
Applying (3.55) and (3.56) we deduce that
B0ss + Υλ2 sin B0 = 0.
If we now think in terms of a variable measuring the horizontal extent of the fibre in
the natural state, rather than the distance along the fibre, we set s0 = Υs to give
λ2
B0s0 s0 + sin B0 = 0. (3.62)
Υ
This is the Euler Strut equation for a beam of stiffness EIΥ, i.e. a less stiff beam
than the original unbuckled material.
3.4.2 Hinged Rods revisited

Let us now cast our minds back to the hinged rods model we spent so much time
considering in section 3.2. Through an appropriate asymptotic expansion, we found
from (3.28) that
d2 Σ(1) χ
2
+ κ0 cos · sin Σ(1) = 0.
ds 2
(1)
where Σ is the inclination of the “concertina” to its mean direction. In this equation,
n
s= N
and hence is equivalent to the distance measured along the rods. A simple
change of variables to
χ
s0 = s cos
2
117
s
χ
2
s’
Figure 3.12: s and s0 for the hinged rods model
leads to
d2 Σ(1) κ0
+ sin Σ(1) = 0 (3.63)
ds02 cos χ/2
where s0 is the distance measured macroscopically (see figure 3.12).
Now consider the case when the unstressed configutation of the hinged rods model
is straight (i.e. φ = π). For this special case, equation (3.1) reduces to
θn+1 − 2θn + θn−1 + κ sin θn = 0
Writing s = n, and expanding as for equation (3.20) we get
d2 θ
+ κ0 sin θ = 0 (3.64)
ds2
where
κ
. κ0 =
2
Thus the hinged rods model and the naturally curved fibre models have both led to
the same model which can be interpreted as an “equivalent” Euler Strut model. In
each case, we have simplified the model by asymptotically separating the short and
long scale variations. The resulting equations show the same behaviour as a less stiff
system without the small scale variations. To summarise
Euler Strut Hinged Rods
d2 θ d2 θ
ds2
+ λ2 sin θ = 0 ds2
+ κ0 sin θ = 0
d2 B 0 λ2 d2 Σ(1) κ0
ds02
+ Υ
sin B0 = 0 ds02
+ cos χ/2
sin Σ(1) = 0
118
We now have some understanding of the behaviour of the crimped fibre under com-
pression and extension. This predicts that under certain circumstances a bifurcation
can occur from a nearly flat configuration to an arched formation. This could be
a model for secondary crimp formation. In the next chapter, we shall discuss the
subsequent behaviour of the material as it moves through the hinged plates of the
stuffer box.
119
Chapter 4
Modelling the Block Motion
4.1 Introduction
In this chapter we will discuss models for the material inside the stuffer box between
the plates. As distinct from primary and secondary crimp, the orientation of the tow
in this area undergoes no major changes but it is subject to compressive and small
‘elastic’ strains. The material in this region is therefore referred to as “the block”.
Although there are no major structural changes occurring in this region, the prop-
erties of the block change, as new material is introduced at one end after secondary
crimp, and removed from the other end by a conveyor belt. The resulting flow of
material through the block is resisted by frictional forces between the fibre and the
plates and side walls.
The block is compressed between the top and bottom plates of the stuffer box.
The upper plate is hinged at the end that the fibre enters (figure 4.2, and has a fixed
moment applied to it by the addition of a weight on the top plate. This moment is
one of the main control parameters for the stuffer box crimper. It will partly govern
the magnitude of the frictional forces that oppose the motion of the block, and hence
the force acting on the fibre in the nip that causes primary and secondary crimp.
In this chapter we will show how simple models for the block can be coupled with
models for the primary and secondary crimp to determine a complete model for the
crimper. By doing so, we shall see how the choice of model for the block material
alters the stability of the crimper, and that the work undertaken in chapters 2 and 3
need to be rethought in order to give physically realistic results for the modelling of
the stuffer box.
120
4.2 Concept
Let us begin by considering the motion of the block in isolation. We must include a
number of factors when modelling the stuffer box:
• Properties of the material

The small scale structure of the material in the block is very complicated (see
figure 4.1). Modelling all the interactions that take place between fibres would
be excessively difficult. Instead, we shall treat the block as a homogenised mate-
rial whose properties depend on the position along the stuffer box. In addition,
although it is possible that plastic deformation and enviromental conditions1
within the stuffer box could lead to fibre properties that change with time, we
shall ignore these effects and concentrate on modelling the block as a quasi-
static elastic material.
We shall consider two different models for the material in the block. Firstly, we
will use a greatly simplified picture of the material as a series of vertical springs.
In contrast to this, the second model will imagine that the disordered jumble
of fibre is equivalent to an isotropic elastic material. In each case, we shall
have a number of parameters that must be determined. There are two main
methods of doing this. The first is to measure them experimentally, and assume
that their value does not depend on the small scale structure. An alternative
technique is to look at a model for the micro-structure, and determine how
this responds to certain applied forces. This can then be compared with the re-
sponses made by the homogenised material, and hence bulk properties deduced.
Finally, our model for the block will require knowing the amount of material
present at each point of the stuffer box. This will depend on the conditions under
which the material entered between the plates, and the choice of secondary
crimp model.
1
As material passes through the stuffer box, it loses some of the solvents, resulting in a change
in physical properties.
121
Figure 4.1: Video still of a section of the block.
• Interaction between the block and stuffer box

The fibre interacts with walls of the stuffer box through two mechanisms : re-
action forces and frictional forces. The reaction forces exert a moment on the
hinged plate which, in equilibrium, must balance the externally applied mo-
ment. This condition will determine the angle the top plate makes with the
horizontal. It is important to note that the block material can lose contact with
a section of the upper plate of the stuffer box. When this occurs we have a free
boundary problem for the point P in figure 4.2.
The frictional forces in the stuffer box determine the compressive force applied to
the fibres in the primary and secondary crimp regions. Unfortunately, frictional
forces between elastic solids and rigid surfaces introduce many complications
to otherwise simple contact problems. We consider two options in how we
implement friction. The simplest is to assume that all points at the surface of
the block are in limiting friction such that
q = µp
where p is the stress normal to the plate and q is the stress tangential to the
boundary. A more physically realistic option is to divide the boundary into two
regions. In the first, slip has occurred and
u>0 and q = µp
where u is the displacement of the block parallel to the boundary relative to its
orriginal position. In the second, no slip has occurred, and hence
u=0 and q ≤ µp.
122
This adds another free boundary to the problem. Some solutions can be found
for such problems (for example, see Spence [14] for a discussion of a Hertz con-
tact problem with finite friction), but the problem is usually difficult. Thank-
fully, in our case we are interested in situations where the force is sufficient to
move all the block, and hence the assumption that all contact points are in
limiting friction can be justified.
• Motion of material through the stuffer box

Consider the material labelled A at time t and at position x in figure 4.2. After
a small time δt this has advanced through the box to A0 due to material entering
the box at position x = 0. Let Q(t) be the rate of arrival of material at x = 0.
If the material at A at time t entered the block at time t0 then, in the absence
of compression within the block, it is now at
Z t
x(t, t0 ) = Q(t0 )dt0
t
Z 0t Z t0
0 0
= Q(t )dt − Q(t0 )dt0 .
0 0
If we make the reasonable assumption that Q(t) ≥ 0 then it makes sense to

write2 Z t
χ= Q(t0 )dt0
0
and hence
x = χ − χ0 , (4.1)
where χ0 is the value of the Lagrangian variable χ when the material entered
the stuffer box. If we let f (χ0 ) be a property of the material when it enters the
box then the property F (x, t) of the material at position x at time t is given by
F (x, t) = f (χ − x). (4.2)
This assumes that the rate of motion of material is proportional to the rate of
arrival of material, and constant throughout the length of the box. Changes
is the height of the material as it moves through the box are taken care of by
vertical elasic expansion and compression. Furthermore, there is no horizontal
compression. In the first model for the material entering the stufferbox (where
we shall consider it composed of springs) this latter assumption is always valid.
2
Note that χ is the amount of material which has entered the stuffer box in time t and so if
material is fed in at a constant rate χ is proportional to t.
123
P
A x, t A’ x+ δx, t+ δt
Figure 4.2: Diagram of the stuffer box, showing movement of material from A to A0
For the isotropic model, the horizontal compression is of the same magnitude
as the vertical displacement, and hence the relationship between (say) F (x, t)
and f (χ − x) will only hold for small height changes.
124
Figure 4.3: Formation of the block through secondary crimp
4.2.1 Models for the amount of material entering the block

We expect the amount of material entering the block to depend on the events oc-
curring inside the nip of the stuffer box crimper. In chapter 2 we discussed how the
peak force exerted on the block increases as more fibre is added. Our treatment of
the secondary crimp in turn predicts that a buckle in the crimped material will occur
when a critical condition is achieved. Our aim is to use these models to determine
the amount of material entering the block at the hinged end and the force required
to move the block. To this end we make the assumption, based on experimental
evidence, that the length of crimped material that buckles to form secondary crimp
is twice the equilibrium height of material entering the block (see figure 4.3). In
addition we will equate the force associated with this buckle to the force required to
move the block.
Under these assumptions, and recalling the discussions in chapter 3, we can write
the equilibrium height of the block as either (from 3.35)
r
π ΛK cos χ/2
H0 =
2 F
or r
π EI
H0 = (4.3)
2 ΥF
125
depending on whether we use the hinged rods or crimped elastica models for secondary
crimp formation. Unfortunately we are still left with the undetermined parameters
(Λ, K and χ or Υ). However, we can either deduce these from the primary crimp
model or assume that they are independent of the operating conditions of the crimper.
126
4.2.2 Models for the micro-structure of the block
We will now consider two simplified models for the micro-structure of the block. Our
aim is to understand the homogenised elastic properties. To this end, we shall consider
two illustrative fibre arrangements (see figure 4.4). The first, a thought experiment,
is a loop of fibre being compressed due to contact with other loops (I). The second
is a folded elastica (II). Each will be modelled using the equations of the two dimen-
sional elastica that we are familiar with from chapter 2. We shall then consider the
response (i.e. deflection of the top surface) of these when acted on by a compressive
force. This can then be compared with the results of applying the same force to an
identical sized section of the block, and hence deducing the constants of the elastic
model.
The problem of a hoop-shaped elastica under point contact has been extensively
described in the literature [7]. We will briefly describe the analysis performed for this
configuration before showing how it can be extended to
• A hoop touching others above and below
• A hoop touching others above, below and to each side
• A folded elastica (II) touching others above and below.
4.2.2.1 Loops of Elastica
Remember that the elastica obeys the equation
d2 θ Fx Fy
2
+ sin θ = cos θ
ds EI EI
dx
= cos θ
ds
dy
= sin θ.
ds
We consider the half loop of material from A to B (total circumference 2L) with
boundary conditions
θ(0) = π
θ(L) = 0
Fx = 0
Fy = F,
127
B
(I)
A
s
y, F y
θ
x, F x
(II)
Figure 4.4: Two possible fibre arrangements
which assumes that the vertical force is F and that there is no horizontal external
force. Non-dimensionalizing length such that
s = Ls0
rearranging and dropping the primes we find that

d2 θ
− λ2 cos θ = 0 (4.4)
ds2
L 2 Fy
λ2 =
EI
θ(0) = π (4.5)
θ(1) = 0. (4.6)
Integrating and rearranging leads to

dθ √
√ = ±ds 2
λ sin θ + C
where C is the constant of integration. By making the substitutions
π
θ = − + 2α
2
2
C +1 =
sin2 β
128
this can be rewritten as
sin βdα
p = ±ds
2λ 1 − sin2 β sin2 α
3π
α(0) =
4
π
α(1) = .
4
Integrating over the length and taking − sign since α is decreasing gives
π
sin β 3π
EllipticF , β − EllipticF ,β =1 (4.7)
2λ 4 4
where EllipticF is the elliptic integral of the first kind, defined by

Z α
dα0
EllipticF (α, β) = p .
0 1 − sin2 β sin2 α
Since √
0 cos 2β
θ (0) = λ ,
sin β
the solution of (4.7) for β(λ) has turned our boundary value equation into an initial
value problem that is much more amenable to numerical solution. From this, it is a
trivial matter to find computationally the shape (x(s) and y(s)).
We are primarily interested in this loop of elastica as a model for the micro-
structure of the block. Take a small section of block of the same dimensions as the
loop, and apply a compressive force to it (see figure 4.5). Now adjust the elastic
properties of the block material until the response (movement of the top surface) is
the same as that of the loop of elastica. In order to do this we need to know the
response of the elastica to the same force. Figure 4.6 is a plot of the strain
change in height
original Height
against stress
applied force
projected area of loop normal to applied force
for a loop of elastica of unit width3 in non-dimensional units. Notice that the graph is
nearly linear, which will justify the assumptions we shall later make about the elastic
properties of the material.
3
Unit width in this case means that the elastica is of unit width, not the loop
129
W=1
W=1
Figure 4.5: Comparison of loop microstructure with the material from the block
∆H 0.25
H
0.2
0.15
0.1
0.05
10 20 30 40 50 60
F
A
Figure 4.6: Strain vs. stress for a loop of elastica
130
4.2.2.2 Loop touching others above and below
This is identical to the above analysis, except that the fibre can be in continuous
contact over a region. By symmetry, the contact region between the adjacent loops
is flat, and there are no body forces acting, so we will be able to treat the contact in
the same manner as the flat sections in chapter 2. If we this time denote z to be the
length of the flat section, then the curved length is trivially given by (1 − z). The
curved section has the boundary conditions
θ(0) = π
θ0 (0) = 0
θ(1 − z) = 0
θ0 (1 − z) = 0.
Hence, identical analysis to that above, gives
cos 2β = 0
π
⇒β =
4
λ(1 − z) 3π π π π
= EllipticF , − EllipticF , .
4 4 4 4 4
Once again, this will give us our initial condition to solve the problem numerically,
and from which we can determine the elastic properties. Figure 4.7 shows the strain
vs. stress graph for this elastica. It has a more pronounced curve, but has a similar
slope.
4.2.2.3 Loop touching others above, below and to the sides
We now have forces being applied in both directions. Consider a quarter circle as
shown in figure 4.8. The non-dimensional equations of the elastica can be written as
d2 θ
+ λ2x sin θ = λ2y cos θ
ds2
π
θ (0) =
2
1
θ = 0.
2
This can be rewritten

d2 θ
+ λ2 sin (θ − ϕ) = 0
ds2
131
0.45
∆H
H 0.425
0.4
0.375
0.35
0.325
80 90 100 110
0.275
F
A
Figure 4.7: Strain vs. stress for a loop of elastica touching similar loops above and
below
with
λ2 = λ2x + λ2y
λ2x = λ2 cos ϕ
λ2y = λ2 sin ϕ.
Making the substitution ψ = θ − ϕ we obtain

d2 ψ
+ λ2 sin ψ = 0 (4.8)
ds2
subject to
π
ψ (0) = −ϕ (4.9)
2
1
ψ = −ϕ. (4.10)
2
Multiplying (4.8) by ψ 0 and integrating gives

dψ √
√ = ±ds 2
λ C + cos ψ
and the substitution (C + 1) sin2 β = 2 transforms this into
√
2 sin β dψ
p = ±2ds
λ 1 − sin2 β sin2 ψ/2
132
Fy
L2
Fx
Figure 4.8: A loop touching others at four points
Integrating and applying the boundary conditions (4.9) and (4.10) results in
π ϕ ϕ 2λ
EllipticF − , β − EllipticF − , β = .
4 2 2 sin β
The boundary conditions (e.g. prescribing the width and applied force on the loop)
will determine the constants β and ϕ, and the deflection of the top surface, ∆H, can
then be found. Note that for this configuration, we have a new experiment to carry
out on our section of block material. As well as prescribing the vertical force, we
can also imagine that the width of the block cannot change. Figure 4.9 shows the
response of the loop under these conditions. Note two things in comparison with 4.6
and 4.7: firstly, the curve is very straight, and second that the slope is much less.
4.2.2.4 A folded Elastica touching others above and below
Perhaps a more realistic configuration for the block micro-structure is as a fold of

elastica (see figure 4.4II). To describe this, we divide the folded elastica into two
sections of non-dimensional length z and 1 − z (see figure 4.10).
• Semicircular Section
d2 θsc
2
= λ2sc cos θsc
ds
θsc (0) = π
θsc (z) = 0;
133
∆H 0.004
H
0.003
0.002
0.001
1 2 3 4 5 6
F
A
Figure 4.9: Strain vs. stress for a loop of elastica touching similar loops on four sides
• Kinked Section
d2 θk
= λ2k cos θk
ds2
θk (0) = 0
θk (1 − z) = 0,
where λ2sc and λ2k are the non-dimensional forces applied in the vertical direction in
the semicircular and kinked sections respectively. Note that we are assuming that the
elasticas only make point contact. We require that the moment be continuous, and
so
θ0 (z) = θ0 (0) (4.11)
and from symmetry we know that the height of the kink is half that of the semicircular
section. Solving as before gives that
π
zλsc 3π
√ = EllipticF , βsc − EllipticF , βsc
4 2 sin βsc 4 4

(1 − z)λk π φk
= EllipticF , βk − EllipticF , βk ,
4 sin βk 2 2
where βk and φk are related by

π φk
sin = sin βk sin .
4 2
134
z (1-z)
Figure 4.10: Structure of the folded Elastica
135
∆H 0.25
H
0.2
0.15
0.1
0.05
10 20 30 40 50 60
F
A
Figure 4.11: Strain vs. Stress for a folded elastica touching similar loops above and
below
The continuity of curvature (4.11) requires that

r q
2λ 2 2 π 2
1 − sin βsc sin = λk 4 sin βk − 2
sin βsc 4
and we have enough equations to determine βsc , z, βk and λsc , but note that we have
an undetermined parameter φk . This is fixed by applying the condition that says that
the height of the semicircular section is twice that of the kinked, and hence
Z z Z 1−z
sin θsc ds = 2 sin θk ds.
0 0
Once again, we can compare the response of this elastica with that of a section of
material from the block as shown in figure 4.11.
4.2.3 Stress, strain and nondimensionalisation

Figures 4.6, 4.7, 4.9 and 4.11 are all plots of strain against non-dimensional stress for
elasticas of unit width. They all can be seen to be approximately straight (to one
degree or another), and all but the confined loop have a similar slope. The elastic
properties of the block we shall discuss shortly will be dimensional. Now is a good
time to recall the relationship between these. Firstly, we rewrote all forces in terms
of the natural force scale for the elastica
EI
L2
136
where L is the primary crimp length (i.e. half the length of one loop or fold). Secondly,
all the distances (with the exception of the width of the elastica) were nondimension-
alised with the same length. The stress must therefore be nondimensionalised with
the quantity
EI
.
L3
We must take care to remember that EI is the stiffness for an elastica of unit width,
not the stiffness of the elastica fed into the stufferbox.
4.2.4 The primary crimp length and the force applied to the
block
Recall the discussion at the end of chapter 2. There we compared the three, two
dimensional models for primary crimp, and showed that for each
EIW 2
Fx ∼ π 2 M
Xb2
where EI is the stiffness of the tow per unit width, W the width of the stufferbox, Xb
the length of the region over which the primary crimp occurred and M the number
of contact points. In addition, we found that the average crimp length was
2Xb
L∼ .
M
Combining these we have that
r
EIW
L ∼ 2π . (4.12)
Fx
This will prove important when we consider how the material properties depend on
the applied force.
137
Figure 4.12: An idealized view of the crimped material in the stuffer box
4.3 First model of the Block : Vertical Springs

Let us take an idealised view of the block. We imagine it as being composed of a
series of vertical springs (see figure 4.12) which have an equilibrium height (h(x))
and a spring constant k(x), both depending on the position along the box (x). If the
height of the block in the stuffer box is H(x) then a section of the block of width δx
exerts a normal force
Normal Force = P (x)W δx (4.13)
where W is the width of the stufferbox and P (x) is the pressure exerted, given by

h(x) − H(x)
P (x) = k(x) . (4.14)
h(x)
Consider the forces acting on an elemental section of the block (See figure 4.13), and
assume limiting friction. Let ψ be the angle made by the plate with the horizontal.
Balancing the pressures acting on the trapezoidal element and requiring that, for
stability, the total force exerted on the top plate by the block must balance the
external moment applied we find that
P
P1 =
cos ψ + µ sin ψ
and Z
L
W h(x) − H(x)
M= xk(x) dx, (4.15)
cos ψ + µ sin ψ 0 h(x)
where L is the length of the plates. Similarily, if we assume that no horizontal force
is exerted on the block at the exit from the box, the total force at the entrance (F )
138
µ P1 δx
P1 δx
P δx
µ P δx
Figure 4.13: Forces acting on an elemental section of the block
can be shown to be given by

Z L
µ cos ψ − sin ψ h(x) − H(x)
F =W µ+ k(x) dx. (4.16)
cos ψ + µ sin ψ 0 h(x)
4.3.1 Contact
So far we have only said that H(x) is the height of the block in the stufferbox. The
precise form of this depends on what boundary conditions we apply. We must have
that
H(x) ≤ H0 + x sin ψ
where H0 is the height of the top plate at its left most end. Secondly, if the block is
not stuck to the top plate, we must require that
H(x) ≤ h(x)
This is the same as saying that the block can only be compressed by the top plate,
not elongated.
We will consider two cases here
• Full contact
H(x) = H0 + x sin ψ
• Loss of contact
H(x) = min (H0 + x sin ψ, h(x))
139
4.3.2 Time varying Parameters
If we let the properties h(x) and k(x) change with time in the manner discussed in
section 4.2 then the equations of the system become from (4.15) and (4.16)
W
M =
cos ψ(χ) + µ sin ψ(χ)
Z L
h(χ − x) − H(χ, x)
x · k(χ − x) dx, (4.17)
0 h(χ − x)

µ cos ψ(χ) − sin ψ(χ)
F (χ) = W µ + .
cos ψ(χ) + µ sin ψ(χ)
Z L
h(χ − x) − H(χ, x)
k(χ − x) dx, (4.18)
0 h(χ − x)
where χ is the amount of material which has entered the box in time t.
4.3.3 Non-dimensionalization
The following non-dimensionalisations are appropriate for the system
x = Lx̄,
H0 = L,
ψ(χ) = ψ̄(χ̄),
k = K0 k̄,
M = W L2 K0 M̄ ,
χ = Lχ̄,
L
h(χ) = ,
g(χ̄)
H(χ, x) = LH̄(χ̄, x̄),
and introduce the small parameter , the aspect ratio of the box. If the force at
the entrance to the stuffer-box is denoted F , then by considering equation (4.18), a
suitable non-dimensionalisation is
F = µW LK0 F̄ .
140
After dropping the ¯ ’s, the moments and force balances((4.17) and (4.18)), give
the following system to describe the state of the stuffer-box at any instance in time
1
M =
cos ψ + µ sin ψ
Z 1
xk(χ − x) (1 − H(χ, x)g(χ − x)) dx (4.19)
0

µ cos ψ − sin ψ
F = 1+
µ cos ψ + µ2 sin ψ
Z 1
k(χ − x) (1 − H(χ, x)g(χ − x)) dx. (4.20)
0
4.3.4 Linearization
We now linearize equations (4.19) and (4.20) by making the assumption that the
height of the block varies slowly, in such a way that loss of contact occurs only for
ψ > 0 and at a point xC (χ). We therefore have that

1 + x sin ψ 0 < x < xC (χ)
H(χ, x) = 1 .
g(χ−x)
xC (χ) ≤ x ≤ 1
Expanding equations (4.19) and (4.20) in powers of gives, to zeroth order

Z xC
M = xk(χ − x) (1 − (1 + xψ)g(χ − x)) dx, (4.21)
0
Z xC
F = 2 k(χ − x) (1 − (1 + xψ)g(χ − x)) dx. (4.22)
0
4.3.5 Closing the model

We now have a model for the block that predicts the force required to move it in
terms of the properties of the material that entered in the past. In order to close the
model, we shall need to know how the properties of the material now entering the
block depend on this force (the box marked “?” in figure 4.14). In a later section, we
shall show how we can use our model of the primary and secondary crimp regions to
give this information, but for ease of exposition we shall concentrate on the general
case where we have that
F (χ) = F̂ (g(χ))
k(χ) = k̂(g(χ)).
141
Height
Eqm.
Elastic
Prop.
Model
Block
?
F
Figure 4.14: Closing the springs model
142
4.3.6 Uniform equilibrium states and their stability
Observation of the stuffer box crimper sugguests that under most operating condi-
tions, the angle made by the plate is constant in time, and the crimp regular. However,
when the box is nearing the end of its useful life, the system seems to become unsta-
ble, and exhibits large temporal oscillations. To help understand this, we shall now
seek a solution of the form
g(χ) = g0 + εg1 (χ)

xC = x0 + εx1
ψ(χ) = ψ0 + εψ1 (χ)
F (χ) = F0 + εF1 (χ)
= F̂0 (g0 ) + εF1 (χ)
k(χ) = k0 + εk1 (χ)
= k̂0 (g0 ) + εk1 (χ)
where g0 , F0 , x0 , k0 and ψ0 relate to the uniform equilibrium state. Realizing that
dF̂
F1 (χ) = g1 (χ) = αg1 (χ)
dg
and
dk̂
k1 (χ) = g1 (χ) = βg1 (χ),
dg
we can expand the equations (4.21) and (4.22) to obtain both stationary solutions
and information about their stability.
To zeroth order we have that

!
3 1 M g0
ψ0 = − − , (4.23)
x0 g 0 2 x20 k̂0 (g0 ) 2

ψ0 g0 x20
F̂0 (g0 ) = 2k̂0 (g0 ) x0 − − g 0 x0 (4.24)
2
where
either x0 ≤ 1 and (1 + x0 ψ0 )g0 = 1

or x0 = 1 and (1 + ψ0 )g0 ≤ 1
143
These will give us any uniform stationary states. To find their stability, we must
continue to first order, where we arrive at
Z x0
x30
k 0 ψ1 g 0 = (β(1 − g0 ) − k0 ) xg1 (χ − x)dx
3 0
Z x0
−ψ0 (k0 + βg0 ) x2 g1 (χ − x)dx
0
Z x0
α
and g1 (χ) = (β(1 − g0 ) − k0 ) g1 (χ − x)dx
2 0
Z x0
x2
−ψ0 (βg0 + k0 ) xg1 (χ − x)dx − k0 ψ1 g0 0
0 2
Eliminating ψ1 , we obtain
Z x0 Z x0
α
g1 (χ) = A1 g1 (χ − x)dx − A2 xg1 (χ − x)dx
2 0 0
Z x0
3
− A1 xg1 (χ − x)dx
2x0 0
Z x0
2
−A2 x g1 (χ − x)dx (4.25)
0
where
A1 = β(1 − g0 ) − k0
A2 = ψ0 (βg0 + k0 ).
We will now seek a trial solution to this of the form
g1 (χ) = Ceλχ .
Inserting into (4.25) gives the following equation for λ

α 3 2 3A1 3A2
λ − A1 λ + + A2 λ − =
2 2x0 x0

A1 − x 0 A2 2 3A1 3A2 −λx0
λ + − 2A2 λ − e (4.26)
2 2x0 x0
Note that λ will, in general, be complex, and the sign of the real part will determine
if the steady state is stable or not. Specifically, the system will be unstable for
<(λ) > 0.
144
4.3.6.1 Finding the roots
The infinity of roots of the trancendental equation (4.26) are difficult to find, although
we can asymptotically find the location of those well away from the origin, these all
lie well into the negative real half plane4 and hence correspond to stable solutions.
This means that any roots with positive real part will be found for O(1) values of λ,
and hence will need to be found numerically.
Our method to do this will involve locating roots near the origin for a given
value of the parameters, and then tracking where they move for small changes in the
parameters. To make this simpler, we can start by dividing (4.26) by A1 , and hence
writing

3 3 2 3B2
B1 λ + λ − − B2 λ − =
2x0 x0

1 + B 2 x0 2 3 3B2 −λx0
− λ + + 2B2 λ + e
2 2x0 x0
where
α A2
B1 = − B2 = − .
2A1 A1
The roots were found using Mathematica’s inbuilt FindRoot command, using the
root for the previous parameter values as the seed. Figures 4.15 and 4.16 show the
regions of stability (black) and instability (white) for a range of control parameters
(B1 , B2 and x0 ).
4
Consider equation (4.26) when we write that
X + IY
λ=

for some small constant . Inserting into the equation, and dividing through by λ2 we have
α X Y A1 − x0 A2
(X + IY ) + O(1) = exp −( + I )( + O())
2 2
Taking the modulus of this and squaring, we have that
α 2 A1 − x0 A2
(X 2 + Y 2 ) + O(1) = e−2X/ ( + O())2
2 2
Rearranging, this becomes
2 2
α α
Y 2 = 2 e−2X/ − X2
A1 − x0 A2 A1 − x0 A2
which only has real roots for large negative X.
145
2
x0 = 1.0
1.5
- Ψ0 ( β g 0 + k 0 )
B2 = 0.5
b
β (1- g 0 )- k 0
-0.5
-1
-4 -2 0 2 4
αa
B1 =
2( β (1- g 0 )- k 0 )
x0 = 0.9
1.5
- Ψ0 ( β g 0 + k 0 )
B2 = 0.5
b
β (1- g 0 )- k 0
-0.5
-1
-4 -2 0 2 4
a
α
B1 =
2( β (1- g 0 )- k 0 )
Figure 4.15: Regions of stability (black) and instability (white) for the springs model,
with x0 = 1.0 and 0.9
146
2
x0 = 0.8
1.5
- Ψ0 ( β g 0 + k 0 )
B 2= 0.5
b
β (1- g 0 )- k 0
-0.5
-1
-4 -2 0 2 4
αa
B 1=
2( β (1- g 0 )- k 0 )
x0 = 0.7
1.5
- Ψ0 ( β g 0 + k 0 )
B 2= 0.5
b
β (1- g 0 )- k 0
-0.5
-1
-4 -2 0 2 4
a
α
B 1=
2( β (1- g 0 )- k 0 )
Figure 4.16: Regions of stability (black) and instability (white) for the springs model,
with x0 = 0.8 and 0.7
147
Note first that, contrary to what one might expect, the presence of the free bound-
ary (x0 < 1) makes the system more stable. One can attribute this to small variations
having a lesser effect on the angle of the top plate due to decreased leaverage (acting
closer to the pivot point as x0 decreases). We can also see that instabilities occur for
values of B2 > 1. Remembering the definition,
ψ0 (βg0 + k0 )
B2 = −
β(1 − g0 ) − k0

β
= ψ0 1 − ,
β(1 − g0 ) − k0
we observe that one situation where instability may occur corresponds to a positive
equilibrium angle ψ0 and
• Either (for positive β)

k0
g0 > 1 −
β
• or
k0
g0 < −
β
for negative β. This can be interpreted as saying that unstable solutions can occur
when the angle is large and positive, and where the equilibrium height of the block is
less than the height of the stufferbox. These two situations are mutualy exclusive. In-
stability could also occur when β(1−g0 ) is very close to k0 from below. Until we obtain
values of the constants, we will be unable to determine whether this situation is likely.
The second condition for instability is that of small |B1 |, when

α

2(β(1 − g0 ) − k0 )
is small, which occurs when α is small and β(1 − g0 ) − k0 is large. Small α implies
that small changes in the force gives rise to large changes in the equilibrium height
(which one would expect physically to be more unstable), whilst a large β implies
that small changes in force give rise to large changes in the block properties. Again,
we might intuitively expect this to be an unstable situation.
From this model, we would not, in general, expect the block to be unstable. Any
instability observed would need to be attributed to factors we have left out. A test of
the model’s accuracy would be to see if it predicts the equilibrium angle of the stuffer
box plate correctly. This we shall now discuss.
148
Primary Crimp Microstructure
Λ
Model Model
Elastic
Prop.
F
Secondary Crimp Eqm.
Model Height
Figure 4.17: Closing the springs model
4.3.7 Predicting the equilibrium state

We shall now consider a specific example of the feedback between the current state of
the block (represented by the force required to move it) and the properties of the fibre.
To do so, we shall combine the models of the microstructure as looped or folded elas-
tica, with the models of primary and secondary crimp discussed in chapters 2 and 3
(See figure 4.17). Firstly, let us consider the response of a small section of the spring
material to a compressive force. It was an assumption of equations (4.13) and (4.14)
that there was a linear relation between stress and strain (as was borne out in sec-
tion 4.2.2). If we nondimensionalise stress and strain as discussed in section 4.2.3, we
can write the linear relationship as
Λ3 k
stress = strain,
EI
where Λ is the primary crimp length. Reading the slopes from figures 4.6 and 4.11
gives that
EI
k= C, (4.27)
L3
149
where C = Cloop = 285 for a loop of elastica, and C = Cf old = 230 for a fold. We can
also relate this to the applied force through equation (4.12) to give
1
k=C √ F 3/2 . (4.28)
3
π 8 EIW 3/2
When we nondimensionalised the spring constant in section 4.3.3, we did so using an

unspecified scale K0 . We can now choose the value of this such that (4.28) becomes
k̄ = F̄ 3/2 (4.29)
which requires that

64EIπ 6
K0 = ,
C 2 µ3 L 3
(remember that in this case EI is the stiffness of the tow per unit width, and L is
the length of the plates).
In addition, we can recall the earlier discussion on the height of the material
entering the stufferbox (4.3) to write
r
π EIW
h= ,
2 ΥF
and hence, when nondimensionalised,
√
16π 2 Υ p
g= F̄ . (4.30)
Cµ
Combining this with the equation for k̄ (4.29) we have

3
Cµ
k̄ = √ ḡ 3 . (4.31)
16π 2 Υ
4.3.7.1 Steady State Solutions
We shall now solve equations (4.23) and (4.24) using the relationships (4.30) and (4.31)
to close the model, to find ψ0 , g0 , F0 and k0 . We should first note that there are two
possible cases. Either
x0 < 1 and g0 (1 + x0 ψ0 ) = 1
or
x0 = 1 and g0 (1 + x0 ψ0 ) < 1.
150
We shall start by considering the first of these. We start by solving for
1 − g0
ψ0 =
g 0 x0
and hence obtain from (4.23)
x20
3M̄ = (1 − g0 )k¯0 (4.32)
2
and from (4.24)
F¯0 = k¯0 x0 (1 − g0 ) (4.33)
Now we write
3/2
k¯0 = F¯0
from (4.29) and

k¯0 = α3 g03
from (4.31) where

Cµ
α= √ .
16π 2 Υ
Eliminating x0 from (4.32) and (4.33) we have that
6M
g0 = .
α + 6M
In order that x0 < 1 we require that
(α + 6M )2
≤1
6M α2
or
α2 + 6M α(2 − α) + 36M 2 ≤ 1.
This can be rewritten as

36(M − λ1 )(M − λ2 ) ≤ 0
where
1 h p i
λ1,2 = 3α(α − 2) ± 3α (α − 2)2 − 4 .
36
λ1,2 will be real if α > 4 and then a solution will exist if M is between λ1 and λ2 .
Taking some approximate values for the constants5 :
5
A note on measurements.
Quantities such as Length, Height and Width are all easily measured. This is not the case for the
coefficient of friction, applied moment and stiffness of paper. Hence, these quantities are more an
indication of order of magnitude rather than exact values. The quantity Υ (the shortening of the
tow due to primary crimp) can only ever be an estimate.
151
Length of Stuffer box L = 0.20 m
Height of Stuffer box H0 = 0.015 m
Width of Stuffer box W = 0.03 m
Coefficient of Friction µ = 0.2
Applied Moment M = 0.8 N m
Stiffness of paper p.u.w. EI = 0.44 N m
Shortening due to crimp Υ = 0.2 (a guess)
We have that
α ∼ 10 ⇒ λ1,2 ∼ 0.222, 13.1
and
M̄ = 0.13
which does not satisfy the condition. The second possibility for the solution of (4.23)
and (4.24) is that
x0 = 1 and g0 (1 + x0 ψ0 ) < 1.
Using the previous relationship for k̄0 and F̄0 leads to
α3 g04 − α3 g03 + 2α2 g02 − 6M̄ = 0 (4.34)
This has at least one positive real root and one negative real root, and may have 3
positive real roots depending on the relative sizes of M̄ and α. When α ∼ 10 (as for
the above values of the parameters), there is a single positive root, which gives
g0 ∼ 0.77.
Using this value, we can calculate ψ0 to be ∼ 0.34. Unfortunately, this means that
g0 (1 + ψ0 ) ∼ 1.03
which violates our inequality. For the values of the parameters given above, we have
been unable to find a leading order solution. It is worth remembering the assumptions
that we have made in reaching this point. In doing so, we hope to consider alternative
models that may explain the behaviour of the stuffer box:
• Modelling the block as springs

It is always possible that the block does not behave in any way like the springs
model would predict. Later in this chapter, we will consider two further options
: the block as an isotropic elastic material, and the block as an anisotropic
elastic material. The choice of model will then be discussed. For now, we shall
continue to discuss the spring model.
152
• Relationship between force and equilibrium height of the block
When we discussed secondary crimp formation, we assumed that the crimp
formed without any constraints. In the stuffer box crimper, this is not the case,
as the secondary crimp occurs in the nip region, and it may be possible that
under these conditions the model we chose does not hold. In addition, we looked
at the case when we had a fixed number of rods and applied a force until the
collection buckled. In reality, we are increasing the applied force and also adding
new rods. Whether this will behave in the same way is open to question. Indeed,
qualitative arguments made by observing the paper crimper suggest that as the
force required to move the block increases (for example, when the paper gets
caught between the roller and the top plate), the equilibrium height of the block
material also increases. This is the opposite to the result predicted by both the
hinged rods and small scale elastica models, where an increase in force decreased
the secondary crimp length.6
• Relationship between force and primary crimp length

Our discussion on how the force exerted on the secondary crimp region (and
by implication the block) depended on the excess of fibre within the nip region
made the assumption that the aspect ration of the channel was small, and hence
we could linearise the equations of the elastica. Indeed, when we compared our
results with the solutions to the full equations given by Domokos, Holmes and
Royce [8] we saw very similar graphs. Observations from the video suggest that
the crimp length depends on the geometry of the nip and this has not yet been
taken into account in our models.
• Models of the microstructure

There may also be problems with the models for the microstructure. In both
the loop and the folded elastica we assumed that the lengthscale of the shape
was that of the crimp length. If this is not the case, for example if there were
large regions of contact such that the majoritory of the load was supported by
small regions of elastica (that do not scale with the primary crimp length), then
the analysis would need to be redone.
6
Both the models for secondary crimp predict that
1
F ∝ g2 ∝
h̄20
(where h̄0 is the equilibrium height of the material). Hence as we increase F the equilibrium height
must decrease.
153
'()( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )'

'()( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )'
*(+(*(+( *(+(*(+( *(+(*(+( *(+(*(+(!" *(+(*(+(#$#*('(+ * )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )' H
*(+(*(+(*(+(*(+(*(+(*(+( *(+(*(+(! *(+(*(+(%& *('(+ * )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )'
'()( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )( ' )'
lP lS
L nip
Figure 4.18: Geometry for the new model for secondary crimp
We shall now discuss in more detail the relationship between the primary and
secondary crimp lengths, and the force required to move the block to take into account
the geometry of the nip.
4.3.7.2 A new model for secondary crimp
Suppose that the primary and secondary crimp regions lie in a well defined region of
the stuffer box within the nip (see figure 4.18). Let Lnip denote the length of this
region, and H0 the distance between the plates. A trivial application of geometry
allows us to write that r
H02
Lnip = H0 R −
4
where R is the radius of the roller. In addition, let the primary crimp occur in a
region of length lP and the secondary in a region lS such that
Lnip = lP + lS .
As primary crimp occurs, the amount of crimped material increases. However, we

suppose that no secondary crimp can occur until the block has moved and space to
accept it in the block region has been created. If we imagine that the fibre in the
primary crimp region is an Euler strut of length lP and stiffness EIW then the force
required to buckle it is given by
π 2 EIW
Fbuckle = . (4.35)
lP2
Primary crimp will continue to occur until this force is greater than that required
to move the block. At this point, the block moves up and the fibre in the secondary
154
H0
L nip
Figure 4.19: The nip as a wedge
crimp region folds to become the new block material.
Now think of the material in the secondary crimp region. When the block is
about to move, it is being compressed by a force Fbuckle , and is taking up a length
lS = Lnip − lP . In a moment, this material is going to be entering the block, and so
let us use the same “springs” model for it now. Hence if Λ denotes the primary crimp
length, then
Fbuckle l0 − l S
=k .
ΛW l0
where l0 is the equilibrium length of the secondary crimp region. We can also relate
k to the primary crimp length using the relationship derived before (4.27), and hence
EIW l0 − lS
Fbuckle = C . (4.36)
Λ2 l0
l0 can be related to the equilibrium height of the material by the simple assumption
that
l0 = 2h0
Finally, we need to consider how the primary crimp length Λ depends on the
applied force. Let us imagine that the crimp is forming in a wedge of length Lnip and
width H0 (see figure 4.19). We imagine that primary crimp forms at the end of the
length lP and hence has a crimp length of
H0 l P
Λ= (4.37)
Lnip
Now we can put all this together. Eliminating lP from (4.35) and solving for lS
we have r
EIW
lS = Lnip − π .
Fbuckle
155
Combining this with (4.36) and (4.37) we get
r !
1 π 2 H02 EIW
h0 = +1 Lnip − π
2 CL2nip Fbuckle
and r
EIW H0
Λ=π
Fbuckle Lnip
4.3.7.3 Nondimensionalisation
We can now apply the necessary nondimensionalisations. In the same manner as

before, we see that
k̄ = F̄ 3/2
where k is now nondimensionalised by
π 6 EIγ 6
K0 =
µ3 L 3 C 2
and
H0
γ=
Lnip
In addition, we can nondimensionalise the equation for the equilibrium height of the
block, and rearrange to get
C 4 µ2 g2
F̄ =
π 4 γ 4 2 (C(g − 2γ) − 2gπ 2 γ 2 )2
Notice that, as desired, F decreases as g increases (for sufficiently large values of g).
This is all the information we shall need to find the equilbrium position of the plate.
Unfortunately, solving equations (4.23) and (4.24) with these relationships between
F̄ , k and g requires finding the roots of a sixth order polynomial. Thankfully, modern
computer algebra systems can easily find approximate roots to such problems, and it
proves easy to get Mathematica to find all six for suitable values of the constants (as
used before, but also including the radius of the rollers (R = 0.10m). Of the six, only
one is realisitc, and this gives a value of
ψ = 0.017
where is the aspect ratio of the block. This gives us a solution to the problem
(with x0 = 1). We now have a second model for how the primary / secondary crimp
properties depend on the force acting on the block. We shall now consider how the
model for the material of the block affects the results.
156
Axy =+f, A xx =-P
ψ
H0
Ay = v = u y= 0
Figure 4.20: The block composed of an isotropic elastic solid
4.4 The Block as an Isotropic Elastic Material

We shall now discuss our second model for the block, where we imagine that it con-
sists of an isotropic elastic material. Unlike the springs model, in an elastic material
a vertical stress can give rise to a horizontal strain. In this section, we will derive
the equations for the block under these assumptions, and in the same geometry as
the springs model. We shall then look for steady state solutions and consider the
stability. We can then proceed to look for specific solutions when we couple the block
model to models for primary and secondary crimp.
4.4.1 Constitutive relations

We shall use the standard, quasi-static equations for an isotropic linear elastic solid,
written in terms of an Airy stress function A(x, y) such that
∂ 2A
= σxx = λ(ux + vy ) + 2µux , (4.38)
∂y 2
∂ 2A
− = σxy = µ(uy + vx ), (4.39)
∂x∂y
∂ 2A
= σyy = λ(ux + vy ) + 2µvy , (4.40)
∂x2
Here, u is the displacement in the x direction and v that in the y (see figure 4.20).
On the bottom plate, we apply the boundary conditions
v = 0
and τxy = 0.
157
(where τ is the stress). These can be rewritten in terms of the stress function as
uy = 0 (4.41)
v = 0 (4.42)
Ay = 0 (4.43)
(which corresponds to no vertical displacement, no friction), whilst at the upper

surface, we prescribe the normal and tangential stresses to be
Axy = f (4.44)
Axx = −P (4.45)
(with f the horizontal shear stress and P the pressure exerted by the plate). Note
that these only hold exactly if the top plate is horizontal, otherwise they are an
approximation valid for the nearly horizontal case.
4.4.1.1 Nondimensionalisation
We will now nondimensionalise the equations and boundary conditions (4.38)-(4.43)

using the following scalings
x = Lx̄
u = Lū
y = Lȳ
v = Lv̄
λ = Λλ̄
A = ΛL2 Ā
P = ΛP̄
f = Λf¯
µ = Λµ̄
where
H0
=
L
the aspect ratio of the box. Inserting into our equations, and dropping the bars gives
Ayy = 2 [λ(ux + vy ) + 2µux ] (4.46)

−Axy = µ uy + 2 vx (4.47)
Axx = λ(ux + vy ) + 2µvy (4.48)
158
with
uy = 0 (4.49)
v = 0 (4.50)
Ay = 0 (4.51)
on y = 0 and
Axy = f (4.52)
Axx = −P (4.53)
on y = 1 + x sin ψ. We can now proceed to linearise the problem. Writing
A ∼ A(0) + 2 A(1) + · · · and ψ = ψ̄
we get from (4.46) and (4.51)

A(0) = A(0) (x)
and hence from (4.53)
A(0)
xx = −P
From (4.47) this implies that
µu(0) (0)
y = −Axy = 0 ⇒ u(0) = u(0) (x, t)
and then by using (4.48)

−y
v (0) = P + λu(0)
x (4.54)
λ + 2µ
Now look at (4.46) to O(2 )
(0)
4µ(λ + µ)ux − λP
A(1) (0) (0)
yy = (λ + 2µ)ux + λvy =
λ + 2µ
A(1)
y = 0 on y=0
f
A(1)
xy = on y = 1 + ψx

Note the −1 term on the top boundary condition. This arises due to an O(1) change
in the shear force occuring over an O() distance. In order to balance this, we will
require that the shear stress is O() (f = f¯). Now the total force per unit width on
a section of the block will be
Z 1+xψ
T (x) = A(1)
yy dy.
0
159
Hence
(0)
4µ(λ + µ)ux − λP
T (x) = (1 + ψx) (4.55)
λ + 2µ
and
(0)
dT 4µ(λ + µ)ux − λP
= f¯ + ψ . (4.56)
dx λ + 2µ
(This last equation comes from the O(2 ) terms in (4.47)). Now we are in the position
to close the model. Firstly, we shall prescribe the displacement on the top surface
(V , measured downwards) to be
V = (1 + β(x) − (1 + ψx))
where β(x) is the “excess” height of the block in equilibrium at position x, and will
be determined by the model for secondary crimp formation. Secondly, we will assume
that the block is in limiting friction everywhere (as must be the case prior to the
block moving). From this, we can say that
f¯ = ξP
where ξ is a coefficient of friction. Combining this with equations (4.54), (4.55)

(0)
and (4.56) allow us to eliminate ux to give

dT λ + 2µ 4µ(λ + µ)ψ xψ − β
= ξ−ψ P− (4.57)
dx λ λ 1 + xψ
λ + 2µ 4µ(λ + µ)
T = − (1 + xψ)P − (xψ − β) (4.58)
λ λ
Comparing (4.57) with the differential of (4.58) gives, assuming constant µ and λ7

λ + 2µ dP 4µ(λ + µ)ψ 1 + xψ dβ
ξP + (1 + xψ) = (xψ − β) + ψ−
λ dx λ(1 + ψx) ψ dx
Writing
ξλ 4µ(λ + µ)
A1 = and A2 =
λ + 2µ λ + 2µ
this can be transformed using an integrating factor to

d A1/ψ
(A1/ψ−2) d 1 + xψ
(1 + xψ) P (x) = −A2ψ(1 + xψ) (xψ − β) (4.59)
dx dx ψ
and hence solved to find P in terms of the excess height of the block and ψ. In order
to do so, we need to fix the constant this introduces. This is obtained by remembering
7
We shall look only at the case of constant coefficients, as to do otherwise would make the
resulting equations increasingly difficult to solve
160
that the block loses contact at x = x0 , and at this point P = 0. The position of x0
must be determined by the requirement that we apply no force on the block at the
exit of the system, and hence can say that T (x0 ) = 0, and hence from (4.58)

β
x0 = Max ,1 (4.60)
ψ
We also know that the pressure supports the applied moment, and so
Z x0
P xdx = M (4.61)
0
where M is the nondimensionalised moment. This is now sufficient information to

determine both P (x) and ψ for a given β(x). In turn, if we equate
−T = Force required to form primary crimp
then we can close the model and describe the evolution of the stuffer box.
4.4.1.2 Steady State Solution
We shall begin by looking for a steady state solution where β, µ and λ are constant.
Under these circumstances (4.59) simplifies to
d
(1 + xψ)A1/ψ P (x) = A2ψ(1 + xψ)(A1/ψ−2) (1 + 2xψ − β)
dx
This can be integrated to give P (x) and then inserted into (4.61) to give an equation
relating ψ, A1, A2 and M . Unfortunately, this cannot be inverted, so we shall need
to use root finding techniques to find ψ.
4.4.1.3 Relationship between micro-structure and elastic properties
We shall now consider how the elastic properties of the block depend on the micro-
structure. Consider two simple experiments carried out on a cube of the block mate-
rial. In the first, we simply apply a uniform pressure to the top surface, letting the
sides expand as required. In the second, we apply a similar uniform pressure, but
prevent the block expanding by adding walls. The equations for a linear, isotropic
elastic medium are easy to solve in each case. For the block with free sides,
Stress 4µ(λ + µ)
=
Strain λ + 2µ
whilst when the sides are fixed
Stress
= λ + 2µ
Strain
161
Comparing these with the graphs of stress vs. strain for the loop of elastica under
the same conditions (figures 4.6 and 4.9) gives, in dimensional variables
4µ(λ + µ) EI
= 3 CF ree
λ + 2µ Lp
and
EI
λ + 2µ = CConf ined
L3p
where (reading the slopes off the graph)
CF ree ∼ 285 and CConf ined ∼ 1250
and Lp is the primary crimp length. We can then solve for λ and µ to get
EI q 2
λ = CConf ined − CConf ined CF ree
L3p
EI
∼ 1100 3
Lp

1 1250EI
µ = −λ
2 L3p
EI
∼ 75 3
Lp
We are more interested in the quantities A1 and A2, which become in dimensional
variables
ξλ
A1 = λ+2µ
∼ 0.9ξ
4µ(λ+µ) EI
A2 = λ+2µ
∼ 3 CF ree
Lp
This will influence our choice of scalings for λ, M , P and f , as we wish A2 to be

an O(1) quantity, and therefore must choose a scale for Λ appropriately. The ex-
act scaling will depend on the relationship we choose between T and the primary
crimp length. We shall once again compare two choices. The first uses the results
of the chapters on primary (2) and secondary (3) crimp. The second will be the
new model we introduced to improved model allowing for variation in the nip region
(section 4.3.7.2).
162
• Basic models for Primary and Secondary crimps
Recall that, for the primary crimp model as discussed in chapter 28
r
EIW
Lp = 4π
TW
giving
1
A2 = √ CF ree T 3/2
64π 3
EI
in dimensional terms. We now realise that
¯
A2 = ΛA2 and T = ΛLT̄
and hence
¯ = 1
A2 √ CF ree 3/2 L3/2 Λ1/2 T̄ 3/2
64π 3 EI
We now choose Λ such that
¯ = T̄ 3/2
A2
and so
4096π 6 EI
Λ= 3 3 2
L CF ree
From the work on secondary crimp formation (chapter 3, (3.62) ) we had that
r
π EIW
L(1 + β) =
2 ΥW T
and so r
π EI
(1 + β) =
2 Υ3 L3 ΛT̄
Hence
CF ree 1
β= √ −1
128π 2 ΥT̄
We can now use these relationships, coupled with the equations for ψ (4.59-4.61)
to solve the problem and seek equilibrium values of ψ and β. Unfortunately, we
find that there is no solution satisfying the constants of this problem.
• New model for Primary and Secondary crimp

We shall now attempt to find solutions by using the second model for primary
and secondary crimp we derived for the springs model. The analysis needed is
8
This equation comes from combining equations (2.91) and (2.92), and noting that in the notation
of this chapter (where we discuss a problem in terms of a sheet of unit width), Fx ⇒ T W , and
EI ⇒ EIW .
163
the same as that performed in section 4.3.7.2, but with k replaced by A2, in
keeping with the model for the material in the block. This gives us
r !
2 2
1 π H0 EI
L(1 + β) = 2
+1 Lnip − π (4.62)
2 CF ree Lnip T
CF ree L3nip √
A2 = EIT 3/2 (4.63)
3 π 3 L3
This allows us to suitably nondimensionalise our equations to
¯ = T̄ 3/2
A2
which requires
π 6 EIγ 6
Λ=
CF2 ree L3 3
and
1 π2γ 2 1 CF ree 1
β= +1 − 3 2 √
2 CF ree γ γ π T
where γ is again defined by
L
Lnip =
γ
For suitable values of the constants (see page 152), this gives an equilibrium
angle of
ψ ∼ −0.42
Although this is negative, it might be a possible value. Unfortunately, this

corresponds to an equilibrium height of
h ∼ 0.67H0
which is < 1 and hence outside the range of validity for the model (which only
makes physical sense if the pressure on the plate is positive everywhere. Seeing
the effect of changing parameters, we qualitatively observe that γ (the aspect
ratio of the nip) plays an important role whereas M does not. This suggests
that the geometry of the stuffer box has a very important role to play.
4.4.2 Stability of the stuffer box

The set of equations that govern the stuffer box ((4.58), (4.59), (4.60) and (4.61))
are very difficult to work with. Indeed, when we attempt to do an expansion for a
small peturbation, the number of terms rapidly increases, and it becomes increasingly
164
F
Figure 4.21: An alternative design of stuffer box
difficult to make progress, even when taking advantage of computer algebra packages.
This is primarily due to the terms of the form
(1 + xψ)A1/ψ
that appear. In an attempt to overcome this problem, let us begin by considering the
case when the top plate is nearly horizontal, i.e.
ψ = O(2 ).
This has two effects. Firstly, we will find it difficult to balance the applied moment,
and so we must either consider
• The case when the system is designed to run with the plate nearly horizontal.
This prescribes what M should be. Note that for a nearly flat plate, contact is
made throughout the length (i.e. from x0 = 1.)
• The alternative design of stuffer box, where the top plate is horizontal and a
hinged flap at x = 1 provides extra resistance to the motion and control. (see
figure 4.21))9
We shall now consider the stability of these two special cases of the stuffer box.
4.4.2.1 Stability of the stuffer box with a nearly horizontal upper plate
When the upper plate is nearly horizontal, we can write that
P = P0 (x) + εP1 (x)

ψ = 0 + εψ1
β = β0 + εβ(x)
T = T0 (x) + εT1 (x)
9
The effect of adding the hinged flap can be modelled by writing
T (1) = Tcontrol
(instead of T (1) = 0) where Tcontrol represents the effect of the flap applied at x = 1.
165
where ε is the size of the disturbance, and insert into our equations for the stuffer
box ((4.58)-(4.61)). Expanding to O(1) gives
λ + 2µ 4µ(λ + µ)
T0 = − P0 + β0
λ λ
λξ dP0
0 = P0 +
λ + 2µ dx
Z 1
M = xP0 (x)dx
0
with the boundary condition

T0 (1) = 0
Solving these gives

P0 = A2β0 eA1(1−x)
where A1 and A2 are as defined on page 160. This can then be used to calculate the
required value of M . If the functional relationship between the excess material in the
stuffer box (β) and the force needed to move the block −T (0) can be written as
T (0) = f (β)
then we can close the model and write

A2ξ A1
f (β0 ) = e − 1 β0
A1
which can then be solved to find β0 . The stability will require us to consider how the
material moves through the stuffer box, and so applying the ideas in section 4.2 we
have
β1 (x) = βˆ1 (χ − x)
and so, expanding to O(ε) we have
ξ A2ξ
T1 (x, χ) = − (xψ1 P0 (x) + P1 (x, χ)) − (xψ1 (χ) − β1 (χ − x))
A1 A1
dP1 dP0 dβ1

A1P1 (x, χ) + = A2ψ1 (1 − β0 ) − xψ1 − A2
dx dx dx
Z 1
0 = xP1 (x, χ)dx (4.64)
0
with the boundary condition

T0 (1, χ) = 0
166
We shall now seek a solution where
β1 = ez(χ−x)
for an unknown z. Solving for P1 we find

A2
P1 (x, χ) = 2ez(χ−1) + 2(ez(χ−x) − ez(χ−1) ) z
2
−ψ1 (2 + 2β0 + A1β0 − A1β0 x2 eA1(1−x) )
The value of ψ can be deduced from this by balancing the moments (4.64). With this
knowledge, we can now apply our closing condition, by writing that

df
T1 (χ, 0) = β1 (χ)
dβ β0
We now proceed to seek values of z which satisfy this condition. Once again, this
analysis gives a transcendental equation for z, and so in the same manner as before we
shall use a numerical technique to track the roots as control parameters are changed.
Looking at the form of the equation that forms, we see that by writing

A1 df
α=
A2ξ dβ β0
then there are three main parameters we can vary:

α which measures the sensitivity of the coupling between the block
and the secondary crimp region,
β0 the equilibrium excess height of the block,
and A1 which measures the friction between the block and the
stuffer box.
Figures 4.22 and 4.23 show the contours of stability for eight values of A1 and a range
of β0 and α. Note that for large A1 the system becomes unstable for large β0 . We
shall discuss the importance of this later.
4.4.2.2 Stability of the Alternative Stuffer box design
The equations for this case are much simpler, as we have the condition that ψ = 0.
Equation (4.59) becomes
dP dβ
+ A1P = A2 (4.65)
dx dx
167
2 2
1.5 1.5
1 1
β0 β0
0.5 0.5
0 0
-0.5 -0.5
-1 -1
-4 -2 0 2 4 -4 -2 0 2 4
α α
A1 = 1.0 A1 = 1.2
2 2
1.5 1.5
1 1
β0 β0
0.5 0.5
0 0
-0.5 -0.5
-1 -1
-4 -2 0 2 4 -4 -2 0 2 4
α α
A1 = 1.4 A1 = 2.8
Figure 4.22: Stability contours for the stuffer box crimper with isotropic elastic block
and nearly flat top plate
168
2 2
1.5 1.5
1 1
β0 β0
0.5 0.5
0 0
-0.5 -0.5
-1 -1
-4 -2 0 2 4 -4 -2 0 2 4
α α
A1 = 3.15 A1 = 3.5
2 2
1.5 1.5
1 1
β0 β0
0.5 0.5
0 0
-0.5 -0.5
-1 -1
-4 -2 0 2 4 -4 -2 0 2 4
α α
A1 = 4.2 A1 = 5.6
Figure 4.23: Stability contours for the stuffer box crimper with isotropic elastic block
and nearly flat top plate
169
with boundary conditions at x = 0 and x = 1 of
A2ξ ξ
Tcontrol = β(1) − P (1) (4.66)
A1 A1
A2ξ ξ
F (β(0)) = β(0) − P (0) (4.67)
A1 A1
The first two of which can be readily solved to give
Z 1
−A1x 0 A1Tcontrol A1(1−x)
P (x) = A2β(x) + A2A1e β(x0 )eA1x dx0 − e
x ξ
and hence Z 1
A1
F (β(0)) = e Tcontrol − ξA2 β(x)eA1x dx (4.68)
0
Now we shall seek a solution where material moves through the stuffer box and hence
β(x) = β(χ − x)
and look for a solution of the form

dF
β = β0 + εβ1 (χ − x) + · · · and F = F (β0 ) + ε β1 + · · ·
dβ β0
Inserting into (4.68) we find
ξA2
F (β0 ) = eA1 Tcontrol − β0 eA1 − 1
Z 1 A1
dF
β1 (χ) = −A2ξ β1 (χ − x)eA1x dx.
dβ β0 0
Seeking a solution where

β1 (χ − x) = ez(χ−x)
we find that z must satisfy

1
γ=− eA1−z − 1 (4.69)
A1 − z
where
1 dF
γ= .
A2ξ dβ β0
We shall return to discuss the relative stability of this equation shortly, after we have
first discussed the behaviour of the original springs model when contained within the
“alternative” design of stufferbox.
170
4.4.3 Springs model revisited
Recall the springs model for the block. If we have the springs material filling the
alternative stuffer box, with a control force Fcontrol applied at the right hand end then
the (dimensional) force required to move it would be given by
Z 1
β(χ − x)
Fcontrol + dx
0 1 + β(χ − x)
Expanding as

dF
F = F0 (β0 ) + ε β1 (χ) and β = β0 + εβ1 (χ)
dβ β0
and seeking a solution of the form
β1 = ez(χ−x)
requires that
F0 (β0 ) = Fcontrol + β0
and
dF β0 1 −z
= 2
e −1 .
dβ β0 (1 + β0 ) z
Let us compare this with the equation obtained for the stability with an isotropic
elastic material (4.69). Direct comparisons of the stability of the two models is difficult
as the β0 will be different in each case, however if we realise that by choosing our
scalings such that
A2 = 1 and k=1
we can show that the equations determining stability can be written as

1 1 + β0 1 −z
γ= eA1−z − 1 and γ= e −1 (4.70)
z − A1 2β0 z
Stability corresponds to the real parts of the roots of z being negative. Making the
transformation
z = Z + A1
we can reduce both problems (4.70) to finding the roots of the equation
0
Z 0 eZ = a (4.71)
0
where Z 0 = Z − γ 0−1 and a−1 = γ 0 e1/γ for appropriate γ 0 (as defined previously). If
0
Zmin,a is the root of (4.71) with the largest real part then the blocks are stable if
171
• Springs model:
0 2β0 1
<(Zmin,a )<−
1 + β0 γ
or
• Isotropic Elastic model:
0 1
<(Zmin,a ) < − − A1
γ
For the same a, the latter of these is the more restrictive (assuming that A1 > 0, see
footnote), and hence we have predicted that the springs model will be more stable. In
addition, we can observe that small a favours stability, and hence (if β0 < 1) we again
see that the springs model is more stable. Note that this will not necessarily be the
case for large heights (β0 > 1) and materials where A1 < 0.10 Finally, we should note
that stability is improved for both models by having a < 0, i.e. increasing force giving
decreasing height of material. In contrast, observe from figures 4.15 and 4.16 that
in the stability analysis of the springs model with hinged top, the stability diagrams
are very nearly symmetric around the line B1 = 0. As B1 is proportional to α, this
result is in contrast to the predictions of the alternative stuffer box design. We could
attribute this to the presence of the hinged lid stabilising the system (indeed, the lid
will lift if the amount of material becomes too large, reducing the force). When we
look at the stability diagrams for the small angle isotropic elastic model (figures 4.22
and 4.23), we see a slight asymmetry along the line α = 0, as one may expect from the
10
What sort of materials have A1 < 0? Remember that
λ
A1 = ξ
λ + 2µ
and it would be unphysical to have ξ < 0 (i.e. a negative coefficient of friction). Therefor A1 < 0
needs
λ
<0
λ + 2µ
Remembering the definition of the poisson’s ratio (σ) of
1 3K − 2µ
σ=
2 3K + µ
where
2
K =λ− µ>0
3
we see that
3K + µ
A1 = σξ
3K + 4µ
and hence A1 < 0 requires that the poisson’s ratio of the block be negative.
172
Primary/Secondary Model for Stuffer Stuffer Box Design Section and Result
Models Box Material
4.3.6.1 Stability
Hinged
4.3.7.1 No solns
Springs
Model I Alternative 4.4.3 Stability
(ch. 2,3)
Hinged 4.4.2.1 Stability
Isotropic Elastic 4.4.1.3 No solns
Alternative 4.4.2.2 Stability
Hinged 4.3.7.3 Solution
Springs
Alternative NA
Model II
(4.2.7.2)
Hinged 4.4.1.3 No solns
Isotropic Elastic
Alternative NA
Figure 4.24: The different models considered in this chapter
action of a stabalising hinged lid. Note that all these observations must be qualitative
only, as there are no direct correlations between the parameters plotted in each figure.
4.5 Summary
In this chapter we have looked at
• 2 models for the material the block is made from,
• 2 models for the coupling between primary/secondary crimp regions and the
properties of the block, and
• 2 models for the design of stuffer box.
The results are summarised below in figure 4.24
173
Chapter 5
Conclusions
In this thesis, we have shown the rich seam of interesting mathematics that underlies
the crimping process. As we progressed along our journey through the stuffer box
crimper, we needed to look at the process on different scales. At each stage, we
have needed to derive new models to describe the process, and new mathematics to
seek solutions. When we looked at the crimper as a whole, we were forced to think
about the intermediate processes, and the boundaries between the models on the
different scales. Indeed, when we sought solutions for the systems as a whole, we
found that these transition processes governed the whole system. We presented a
number of different models at each stage, and followed them through to show how
the predictions of each case compared with experimental results.
5.1 Primary Crimp

We began our stroll where the fibre enters the crimper, in the nip region between the
rollers. In chapter 2 we model this as an increasing length of elastica being forced into
a narrow channel of fixed dimensions. To make progress in what was a potentially
complicated problem, we began by considering a quasi-static elastica confined to two
dimensions. In addition, in order to see more of the details of the solutions, we took
advantage of the narrowness of the channel to linearise the equations of the system.
Even though the equations were linear, the problem was still nonlinear, as the
confining channel could give rise to numerous contact points. Thankfully, in the re-
gions between the contact points the equations for the elastica were simple to solve.
This enabled us to divide the elastica into a series of sections, and from symmetries
of the confining box, classify these sections based on the type of contact at each end.
The boundary conditions at the ends of the box, and knowledge of the physics of
174
the contact, allowed us to write down both global conditions on these sections, and
continuity conditions which must be satisfied at the joins. This gave us sufficient
information to solve the problem.
Previous attempts at solving the problem in two dimensions had required assump-
tions to be made about the overall shape of the elastica. The technique we had used
overcame this, and by combining all possible section types, we could find all possible
fibre configurations for a given number of contact points. This gave rise to rich bifur-
cation diagrams. To determine what would be observed experimentally, we needed
to use energy arguments, supposing that the observed solution minimises the energy
stored in the system. This enabled us to describe the evolution of shape, contact and
confining force as more fibre was added in.
This progression was found to depend on the boundary conditions imposed at the
ends, especially the amount of freedom at the entrance to the stuffer box. If fibre was
added at the base, then extra contact points formed through the growth of a “dim-
ple”1 . This dimple grew out of a flat section of fibre. If the fibre entered half way up
the wall, then these flat sections never formed, and extra contact points involved a
change in the direction of curvature of the fibre.
The shape of the channel was also found to influence the behaviour of the fibre,
as expected. For a fibre entering a rectangular channel, the formation of another con-
tact point involved an increase in force followed by a sudden decrease as either the
dimple began to form, or the direction of the fibre changed. This was observed not to
be the case for a fibre entering a wedge, where the force was observed to only increase.
5.1.1 Crimp in three dimensions

The work on an elastica confined in two dimensions is certainly applicable to the pa-
per crimper, however when the stuffer box is part of the fibre manufacturing process,
three dimensional effects are an observed feature. Indeed, for some industrial uses
(such as use in filters), the increased tortuosity of the filaments is a welcome feature.
In the second section of chapter 2 we discussed the equations of an elastica in three
1
A section of fibre that goes between two contact points on the same wall
175
dimensions, and linearised for the case of small deflections.
When the deformations were small, we saw that the deflections in each of the two
principal directions uncoupled. Under these circumstances, we considered an elastica
confined to a channel, with the principle directions held at a fixed alignment. We then
applied a thrust to the ends of the fibre, and allowed it to buckle. As expected, the
fibre buckled first in one of its principal directions. However once the fibre touched a
confining surface, we observed more interesting behaviour. If the principal direction
was normal to the surface, then the fibre behaved like the two dimensional case, up
to a critical value of the applied thrust. At this point, it became possible for the fibre
to “escape” from the plane it had deformed in, by buckling along the second principal
direction. Energy arguments showed that this configuration was favoured.
When the principal direction of the elastica was not normal to the confining sur-
face, more complicated behaviour was observed. The boundary conditions that the
fibre be tangential to the surfaces coupled together the deformations in each of the
principal directions, meaning that it was no longer possible to satisfy the boundary
conditions with a planar solution. Instead, the fibre immediately took on a three
dimensional shape. Unfortunately, with only one contact point, the fibre never ex-
hibited the helical structure observed in the tow.
5.2 Secondary Crimp

Continuing our saunter through the stuffer box crimper, we arrived at the secondary
crimp region, where the primary crimped fibre buckled again, before entering the
region between the plates. In chapter 3 we considered two models that captured the
highly oscillatory nature of the tow. The first, the hinged rods model supposed that
the crimped fibre was comprised of a series of rigid rods, joined by elastic hinges where
all deformation occurred. In contrast, the second model thought of it as an elastica
whose natural configuration was oscillatory or crimped, and where deformation could
occur throughout.
176
5.2.1 The hinged rods model
Once again we considered the static case and we derived a difference equation de-
scribing the relationship between the angles θn of the different rods. This had two
features that made solution difficult. Firstly, it was nonlinear, and secondly, it had
a large driving term that oscillated rapidly. This meant that the usual techniques of
linearisation could not work.
To overcome these problems, we started by looking for solutions when the applied
force, κ, was asymptotically small. This proved successful, except when the total
number of rods N was such that
N · κ = O(1),
at which point our expansion broke down. We noted that the series in κ had two
types of terms, those proportional to (−1)n (where n is the number of the rod) and
those which were not. We decided to make use of this fact when we looked for a
solution that was uniformly valid for all n.
We started by seeking a solution of the form
θn = Σn − (−1)n ∆n
where Σ and ∆ would vary smoothly. This enabled us to rewrite the difference
equation as a pair of coupled difference equations, which we then proceeded to solve
by approximating them by smooth functions of n (where = N −1 ). Under these
assumptions, we were able to change the difference equations into an ordinary dif-
ferential equation. This analysis predicted that there were critical values of force for
which there was a bifurcation from a flat solution to an arched one, and that these
values satisfied different equations depending on whether N was odd or even. When
we proceeded to solve the equations numerically, this difference between odd and even
was observed.
The final work on the hinged rods model attempted to fit experimental force
extension curves. This proved difficult with the model as presented, and in order
to improve the fit, we allowed the individual rods to extend slightly. This gave
much better agreement to the experimental curves, and allowed us to determine the
parameters for the hinged rods model. Another application for this was to model
177
the tow under tension, which is of importance when handling the tow (see “Open
Problems” below).
5.2.2 Curved fibre model

The second model had the crimped fibre described as an elastica with a highly oscil-
latory natural shape. We considered the behaviour of this under an applied thrust
along its length. The equation was solved approximately by carrying out a multiple
scales analysis, the result of which was that the elastica behaved as an Euler strut of
lower stiffness than the original fibre.
When we compared this model to the differential equation formulation of the

hinged rods problem we noticed a similarity. In each case, the effect of the oscillatory
nature of the tow led to it behaving as a less stiff version of the straight elastica (of
the same macroscopic length). The ratio of the stiffness of the oscillatory fibre to
that of the straight was the same as the ratio of the macroscopic length of the fibre
to its arc length2 .
5.3 The Block

Our journey through the stuffer box crimper now brought us to the largest scale of
the machine, the plates. In this region, the shape of the fibre was observed to vary
little (although some compression did occur) and hence we referred to the material
as the block. Our modelling of the block had two main avenues. Firstly, we needed a
description of the material contained within. Secondly, we had to relate the properties
of the material currently entering the block to that currently found within. These
together would allow us to describe the evolution of the stuffer box crimper as more
fibre was fed in.
We considered two models for the material contained in the block - a springs
model where the compression of the tow caused only a vertical force, and an isotropic
elastic model, where the compression resulted in a horizontal force. Certain aspects
of the modelling was the same for each:
2
The length measured including all the oscillations
178
• Motion of material through the stuffer box
The material was assumed to move uniformly through the stuffer box so that
the current state of the block depended on the past history.
• Friction between plates and block

Frictional forces acted between the plates and the block, such that the block
was in limiting friction all along its length. This assumption was justified as we
were interested only in the peak force exerted.
• Boundary conditions
We applied boundary conditions that required no force exerted on the block at
the exit from the stuffer box, and that the torque exerted by the compression
of the block must balance the externally applied moment.
• Contact
The upper plate was assumed to not be attached to the block, and thus unable
to stretch the material, only compress it. This meant that there was a free
boundary in the problem where contact was lost.
As well as a model for the material contained within the block, we needed to relate
the properties of the material entering to the block’s current state, in particular to
the current force needed to move the block. In both the springs and isotropic elastic
models, we had two attributes to consider, the elastic constants of the material and
its unstressed height.
To determine values for the elastic constants, we began by proposing a grossly

simplified micro-structure for the material, with it being composed of either loops of
elastica, or folded elasticas. Having written down the necessary equations and nondi-
mensionalised, we then considered the deformation each shape would exhibit under
simple applied forces. These could then be compared with the response of the block
material to the same stresses, and hence elastic constants deduced. Our principal
result come from the dimensional analysis, that predicted that the elastic properties
would be proportional to Λ−3 , where Λ was the primary crimp length. Our original
intention was to use the results of chapter 2 to give a value for Λ in terms of the force
required to move the block. Later it became necessary to use a different primary
crimp length (see next section).
179
The natural place to start when modelling how the height of the material entering
the block was influenced by the force need to move the block was with the discussion
of secondary crimp from chapter 3. For both of the models, this predicted that for
an applied force F , the length that would buckle was proportional to F −1/2 . Unfor-
tunately, we could not find any steady state solutions. We proposed that this could
be because the model did not include any effects from the geometry of the nip region.
To overcome this limitation, we introduced a new model for primary and secondary
crimp.
Closing the models of the block (both as springs and isotropic elastic material)
using models derived by considering primary and secondary crimp separately proved
inadequate. Instead, we began to consider the two processes in the nip region as
linked. Our starting point was to imagine that the nip consisted of two parts, a sec-
tion of fibre that would buckle to form primary crimp, and a section of crimped tow
that would eventually fold to form secondary crimp. The same forces acted on each.
The block was assumed to advance when the force required to buckle the fibre into
primary crimp was greater than that required to move the block. At this point, we as-
sumed that the block would move, and the crimped fibre fold to form secondary crimp.
This model had the advantage over the previous in that it predicted that as the
force increased, the size of the secondary crimp would increase, as was qualitatively
observed in the experimental paper crimper. When we used this model to close the
springs and elastic material models, we found that we could find solutions (which
were physically realistic for the springs model, but not for the other).
5.3.1 Stability of the stuffer box

As well as predicting the equilibrium states of the stuffer box, we were also interested
in the stability of those states. We investigated this by considering a small perturba-
tion of the state, and seeing if it grew or shrank as fibre went through. We found that
the springs model was almost always stable, except under conditions which were un-
likely to occur in reality. The stability of the isotropic elastic block model was more
difficult to determine, and so we concentrated on a more limited class of solution,
those where the equilibrium angle of the plate was zero. This could occur for one of
two reasons. Firstly, because of the operating conditions, and secondly due to the
180
design of stuffer box.
If the plate was free to move (i.e. the plate was horizontal due to the running con-
ditions), then the stuffer box could become unstable if the coefficient of friction was
large enough, and the equilibrium height of the material large. This was in contrast
to the springs model which was almost always stable.
A second design of stuffer box, which exerted a fixed force at the exit, was easier to
analyse for stability. We compared the springs and isotropic elastic material models,
and noted that the latter was always more unstable, except when the Poisson’s Ratio
for the material was negative. We attributed this instability to the added coupling
between the stresses over the springs model. As an interesting aside, we also noted
that the springs model could become unstable in this case, unlike the prediction from
the model when the hinged lid was present. We postulated that the hinged lid adds
to the stability.
This thesis cannot claim to have solved the problem of the stuffer box crimper,
although it has extended the areas of research to examine both the individual sections
of the process and the system as a whole. The discussion in chapter 5 shows that the
coupling between the systems requires much more work before any definite predictions
can be made, but shows that there are some promising avenues to explore. Of the
models for the individual processes, those for the secondary crimp region are least
satisfactory, for it is unclear whether they would apply in the case when more fibre
is being added, nor when the tow was constrained by the top surfaces.
5.4 Open Problems

There are a number of problems that can be considered “open”, either because they
require a considerable extension to the work, or they have not been considered at all.
Some have direct application to the crimping problem, others have arisen as part of
the work.
• Tow under tension

During the use of crimped fibres, it is necessary to transport the tow. This is
often results in crimped fibres hanging between rollers. One example when this
occurs is when fibre is removed from a box by being pulled over a roller above
(see figure 5.1). This problem is similar to that of the periodic folding of thin
181
Tensioner
Tow
-, -, -, -, -, -, -, -, -, -, -, -, -,
-, -, -, -, -, -, -, -, -, -, -, -, -,
-, -, -, -, -, -, -, -, -, -, -, -, -,
,- ,- ,- ,- ,- ,- ,- ,- ,- ,- ,- ,- ,-
-, -, -, -, -, -, -, -, -, -, -, -, -,
Figure 5.1: Tow being removed from a box
sheets [11], but has the added complication that the fibre might stretch under
its own weight.
• Buckling of the block

While carrying out experimental runs with the paper crimper, it was observed
that, under heavy loading, the block could buckle. This could prove an interest-
ing free boundary problem to study. The block is in compression only where it
touches the plate (and must support the applied load). When the buckle occurs,
some of the block moves whilst other regions don’t (a second free boundary).
The contact will be over a region, and hence it may not be appropriate to use
the elastica equations.
• Primary crimp in a nip

All the work done on primary crimp assumed thin aspect ratios so the defor-
mations were small. This is not the case in reality, as the crimp occurs in the
182
nip, and often results in the fibre going vertical.
• Plastic deformation of fibres in the stuffer box

The fibres in the stuffer box have all been modelled as elasticas. This is not the
case in reality, as there must be some permanent deformation in order for the
crimp to take. How to include this into our model could prove an interesting
topic for discussion.
• Multiple fibre effects

All the work has concentrated on modelling the tow as a sheet, and therefore has
not considered the interaction between the fibres. This could be an important
step in understanding effects such as the three dimensional crimp observed in
practice.
• Dynamic effects
We have assumed throughout that inertia could be ignored. In reality, the fibre
enters rapidly, and the boundary of the nip is moving. Both of these effects
may be important.
183
References
[1] Reports on the 22nd European Study Group with Industry, 1989.
[2] 26th European Study Group with Industry, 1993.
[3] G. G. Adams and R. C. Benson. Postbuckling of an elastic plate in a rigid

channel. International Journal of Mechanical Science, 28(3):153–162, 1986.
[4] Warrick Cooke. Buckling in the crimping process, 1997. (Transfer thesis).
[5] C. M. Elliot and J. R. Ockendon. Weak and Variational Methods for Moving
Boundary Problems. Number 59 in Pitman Advanced Publishing Program. Pit-
man, 1982.
[6] Avner Friedman. Mathematics in Industrial Problems Part 4, volume 38 of IMA

Volumes in Mathematics and its Application, chapter 1. Springer-Verlag, 1988-
1997.
[7] Robert Frisch-Fay. Flexible Bars. Butterworths, 1962.
[8] P. Holmes G. Domokos and B. Royce. Constrained euler buckling. Journal of

Nonlinear Science, 7:281–314, 1997.
[9] D. H. Hodges and P. R. Bless. Analysis of beam contact problems via optimal
control theory. AIAA Journal, 33(3):551–556, 1995.
[10] A. E. H. Love. A Treatise on the Mathematical Theory of Elasticity. Dover

Publications, 4 edition, 1944.
[11] L. Mahadevan and J. B. Keller. Periodic folding of thin sheets. SIAM Journal
of Applied Mathematics, 55(6):1609–1624, 1995.
[12] L. Mahadevan and Joseph B. Keller. Coiling of flexible ropes. Proc. R. Soc.
Lond. (A), 452:1679–1694, 1996.
184
[13] T.C. Soong and Injae Choi. An elastica that involves continuous and multiple
discrete contacts with a boundary. Int. J. Mech. Sci., 28:1–10, 1986.
[14] D. A. Spence. The hertz contact problem with finite friction. Journal of Elas-
ticity, 5(3-4):297–319, 1975.
[15] D. P. Vailette and G. G. Adams. An elastic beam contained in a frictionless

channel. Journal of Applied Mechanics, 50(3):693–694, 1983.
[16] D. R. Westbrook. Contact problems for elastic beams. Computers and Structures,
15(4):473–479, 1982.
[17] D. R. Westbrook. The obstacle problem for beams and plates. Journal of Com-
putational and Applied Mathematics, 30(3):295–311, 1990.
185

Mathematics of Crimping

Uploaded by

Copyright:

Available Formats

Mathematics of Crimping

Uploaded by

Document Information

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

Mathematics of Crimping

Uploaded by

Copyright:

Available Formats

Mathematics of

A thesis submitted for the degree of

Thanks to my supervisors, John and Hilary Ockendon, for the many

The aim of this thesis is to investigate the mathematics and modelling of

We begin by modelling the process by which the fibre is deformed as it is

After the fibre has undergone a preliminary deformation, a second process

4 Modelling the Block Motion 120

1.1 Crimped fibres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2.1 Model problem for the industrial drier [1] . . . . . . . . . . . . . . . . 16

3.1 The two ways of seeing secondary crimp . . . . . . . . . . . . . . . . 87

4.1 Video still of a section of the block. . . . . . . . . . . . . . . . . . . . 122

5.1 Tow being removed from a box . . . . . . . . . . . . . . . . . . . . . 182

1.1 1997 : The Year of the Man-made fibre

1.2 The fibre manufacturing process

Figure 1.2: The flat plated stuffer box

1.2.1 The stuffer box crimper

Figure 1.3: Schematic of the Hinged Plate Crimper

1.2.2 Terminology and length scales of the stuffer box crimper

1.3 Structure of this thesis

The remainder of chapter 2 extends these techniques to the problem of a three-

Secondary crimp involves the deformation of highly “concertinaed” fibres. In

1.4 Statement of originality

1.5 Experimental Observations and Results

1.5.1 Observations of crimp formation

• The material between the plates

1.5.2 Quantitative results

1.5.3 General observations on the fibre crimper

Figure 1.9: Angle Against Applied Moment

Figure 1.10: Crimp Length against Applied Moment

y (4) + λy 00 + γ = 0, d < x < t.

where λ is the nondimensional compressive force and γ the nondimensional gravity

and at the feeder (x = t)

Total length − Distance moved along table = Excess Fibre

which can be rewritten as Z t 2

Figure 2.1: Model problem for the industrial drier [1]

Figure 2.3: Infinite elastica confined to a channel [15]

u(±1) = δ B(±1)u00 (±1) = M,

then u is the unique solution of

3. Line contact in the region −ρ ≤ x ≤ ρ.

4. Point contact at x = ±ρ.

Figure 2.6: Contact problem for an elastica [16]

Figure 2.7: Contact problem for a freely supported elastica [9]

1. Pinned-pinned (no contact)

2. Clamped-pinned without contact

5. Contact-contact (same wall)5

Figure 2.8: Section types for constrained Euler buckling [8]

Figure 2.9: A general geometry for a fibre confined in a 2d box.

2.2 2D Confined Fibre

• the fibre touches the surface at a point.

We shall call these points boundary points denote them as si (i = 1, · · · N + 1). We

must be normal to the surface. In particular, if the wall is at an angle ψ to the

2.2.1 Boundary Conditions

These lead us to the following variables

It is perfectly possible to begin solving these equations as written. The results

Equation (2.6) becomes

2.2.3.1 Linearisation of the boundary and free boundary conditions

and hence one condition is Z 1

2.2.3.2 Summary of Linearised equations

At the boundary points, we assume no friction, and hence

In addition, we assume that x, ȳ, θ̄ and θ̄0 are all continuous.

which can be rewritten as Z t 2