Diss. ETH No.

19928

A Procedure for the Simulation of the

Cross Roll Straightening Process

A dissertation submitted to the

ETH Zurich
for the degree of
Doctor of Sciences

presented by
Alexandre Mutrux
Dipl.-Ing. ETH
born February 17, 1983
citizen of Geneva, GE

accepted on the recommendation of

Prof. Dr. Pavel Hora, examiner
Prof. Dr. Konrad Wegener, co-examiner

Zurich, 2011
Acknowledgements
The present thesis is the result of my doctoral studies at the Insti-
tute of Virtual Manufacturing (IVP) of the ETH Zurich from 2007 to
2011. I am deeply grateful to my advisor Prof. P. Hora for giving me
the opportunity to conduct my research at the IVP and for his con-
tinuous support, advice and guidance during my graduate studies.
My sincere thanks go to Prof. K. Wegener for accepting to review
my thesis and to be the co-examiner.

I am indebted to the members of the IVP for providing me such a

good working environment during four years. I am especially grate-
ful to Dr. B. Berisha who, through his commitment to research, was
always an important source of drive and inspiration. My thanks also
go to Dr. L. Tong and my fellow doctoral students for the stimulat-
ing and fruitful discussions.

The financial support of the Swiss Innovation Promotion Agency

(CTI) is gratefully acknowledged.

Many thanks to my family and friends for their support through-

out this period. Finally, this thesis would not exist without the
people who inspired me to pursue doctoral studies: Dr. C. Gonseth,
Dr. T. Hay-Edie, Dr. S. Mutrux, Dr. M. Urabe and Dr. A. Yoshi-
take are only a few of them.

A. Mutrux
Zurich, autumn 2011
Contents

Summary vi

Résumé vii

Nomenclature viii

1 Introduction 1
1.1 Bright Steel Bars . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Uses and requirements . . . . . . . . . . . . . . 1
1.1.2 Production process . . . . . . . . . . . . . . . . 2
1.2 Straightening . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Overview of the processes . . . . . . . . . . . . 3
1.2.2 Principle . . . . . . . . . . . . . . . . . . . . . 7
1.3 Cross Roll Straightening . . . . . . . . . . . . . . . . . 9
1.3.1 Description of cross roll straightening . . . . . 9
1.3.2 Kinetics of cross roll straightening . . . . . . . 11
1.4 Objectives of the Study . . . . . . . . . . . . . . . . . 13

2 State of the Art 17

2.1 Analytical Approach . . . . . . . . . . . . . . . . . . . 17
2.1.1 Theory . . . . . . . . . . . . . . . . . . . . . . 17
2.1.2 Example . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Mixed Analytical-Numerical Approach . . . . . . . . . 21

iii
iv CONTENTS

2.3 Finite Element Approaches . . . . . . . . . . . . . . . 23

2.3.1 Kuboki and co-workers . . . . . . . . . . . . . . 23
2.3.2 Asakawa and co-workers . . . . . . . . . . . . . 24
2.4 Modelling of Related Processes . . . . . . . . . . . . . 25
2.4.1 Roller straightening of wires . . . . . . . . . . . 25
2.4.2 Roller straightening of rails . . . . . . . . . . . 26

3 Experiments and Observations 31

3.1 Yield Stress . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1.1 Methodology . . . . . . . . . . . . . . . . . . . 31
3.1.2 Phenomena observed . . . . . . . . . . . . . . . 32
3.2 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.1 Methodology . . . . . . . . . . . . . . . . . . . 36
3.2.2 Phenomena observed . . . . . . . . . . . . . . . 36
3.3 Residual Stresses . . . . . . . . . . . . . . . . . . . . . 38

4 Material Behaviour and Modelling 43

4.1 Tension-Compression Tests . . . . . . . . . . . . . . . 44
4.2 Microstructural Phenomena . . . . . . . . . . . . . . . 46
4.2.1 Strain hardening . . . . . . . . . . . . . . . . . 47
4.2.2 Bauschinger effect . . . . . . . . . . . . . . . . 47
4.2.3 Cyclic softening . . . . . . . . . . . . . . . . . . 48
4.2.4 Tension-compression asymmetry . . . . . . . . 54
4.3 Macroscopic Constitutive Relations . . . . . . . . . . . 55
4.3.1 General requirements . . . . . . . . . . . . . . 55
4.3.2 Candidate models . . . . . . . . . . . . . . . . 56
4.3.3 Slightly modified Chaboche model . . . . . . . 60
4.3.4 FE implementation . . . . . . . . . . . . . . . . 62
4.3.5 Parameter fitting . . . . . . . . . . . . . . . . . 64

5 Process Modelling 71
5.1 Full Model Approach . . . . . . . . . . . . . . . . . . . 72
5.1.1 Background . . . . . . . . . . . . . . . . . . . . 72
5.1.2 Problem specific procedure . . . . . . . . . . . 75
5.2 Slice Model Approach . . . . . . . . . . . . . . . . . . 77
 5.2.1 Background . . . . . . . . . . . . . . . . . . . . 77
5.2.2 Problem specific procedure . . . . . . . . . . . 81
5.2.3 Tentative incorporation of lateral stamping . . 83
5.3 Total Strain Field Approach . . . . . . . . . . . . . . . 84
5.3.1 Background . . . . . . . . . . . . . . . . . . . . 84
5.3.2 Problem specific procedure . . . . . . . . . . . 90

6 Predictive Capabilities 103

6.1 Inhomogeneity of the Deformation . . . . . . . . . . . 104
6.2 Yield Stress . . . . . . . . . . . . . . . . . . . . . . . . 107
6.2.1 Mixed material modelling approach . . . . . . 107
6.2.2 Predictions . . . . . . . . . . . . . . . . . . . . 108
6.3 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.3.1 Predictions on the assumption of a constant
in-plane curvature . . . . . . . . . . . . . . . . 110
6.3.2 Discussion concerning the assumption of con-
stant in-plane curvature . . . . . . . . . . . . . 110
6.4 Residual Stresses . . . . . . . . . . . . . . . . . . . . . 115
6.4.1 Predictions for the reference configuration . . . 115
6.4.2 Predictions for different configurations . . . . . 115

7 Conclusions and Outlook 121

7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . 121
7.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . 122
7.2.1 Material modelling . . . . . . . . . . . . . . . . 122
7.2.2 Process modelling . . . . . . . . . . . . . . . . 123
7.2.3 Predictive capabilities . . . . . . . . . . . . . . 124

Bibliography 127

Curriculum Vitae 136

List of Publications 137

v
Summary
The present study aims at developing a simulation model for the
cross roll straightening process.

The existing modelling approaches can be grouped in two main

categories: the analytical and the numerical methods. On the one
hand, the analytical methods rely on simplistic assumptions and can-
not take into account the influences of certain process parameters.
On the other hand, numerical models are computationally too expen-
sive to allow the use of reasonably fine meshes, making them unable
to make predictions with a sufficiently high level of accuracy.

Straightening experiments show that, for the product considered,

cross roll straightening leads to: (1) an increase in straightness, (2)
a decrease in the yield stress value and (3) a decrease in the tensile
residual stresses compared to bars before straightening.

The correct description of the above-mentioned phenomena re-

quires to adequately model the cyclic behaviour of the material. To
this end, the tension-compression behaviour of the material is de-
scribed using a mixed isotropic-kinematic hardening model.

In order to overcome the limitations of the existing modelling ap-

proaches mentioned above, a new modelling approach is developed.

This model allows to quantitatively predict the effect of the prin-

cipal process parameters on the yield stress value of material after
straightening. The evolution of the curvature as well as the residual
stress distribution can be predicted qualitatively. The model requires
a reasonable amount of CPU ressources to make predictions with an
acceptable level of accuracy.

vi
Résumé
L’objet de la présente étude est le développement d’un modèle
numérique du procédé de dressage à cylindres obliques.

Les modèles existants se divisent en deux catégories principales :

les méthodes analytiques et les modèles numériques. D’une part, les
méthodes analytiques s’appuient sur des hypothèses simplistes et ne
sont pas en mesure de prendre en compte l’influence de certains pa-
ramètres. D’autre part, les modèles numériques requièrent d’impor-
tants moyens de calcul, ce qui restreint l’usage de maillages fins et
entravant, par conséquent, leur capacité à atteindre des niveaux de
précision élevés.

Les expériences de dressage effectuées montrent que, pour les

barres considérées, le procédé de dressage entraı̂ne : (1) une aug-
mentation de la rectitude, (2) une diminution de la limite élastique
et (3) une diminution des contraintes résiduelles de traction.

La description des phénomènes sus mentionnés requiert une mo-

délisation adéquate du comportement cyclique du matériau. Pour ce
faire, la réponse du matériau à un chargement cyclique est décrite à
l’aide d’une loi d’écrouissage mixte isotrope et cinématique.

Afin de surmonter les limitations des modèles existants pour la

simulation du procédé, une approche mixte analytique-numérique est
développée.

Le modèle développé est en mesure de prédire quantitativement

l’influence des principaux paramètres du procédé sur la limite élastique
après dressage. L’évolution de la rectitude ainsi que de la distribu-
tion des contraintes résiduelles peut être prédite de façon qualitative.
Le modèle est en mesure d’effectuer des prédictions d’un niveau de
précision acceptable avec des ressources de calcul limitées.

vii
Symbols and Abbreviations
Roman symbols

a, a(i) Backstress, backstress components

b Constitutive model parameter
C, Ci Constitutive model parameters
Ce Isotropic elastic stiffness tensor
d Minimum distance in Y between the rolls
E Young’s modulus
Etan Tangent modulus
F Yield surface
G Modulus of rigidity
h Thickness
I, I Identity tensors (2nd and 4th order)
M Torque
My Maximum elastic torque
n Hardening exponent
n Normal to the yield surface
p Accumulated plastic strain
ptrig Trigger value of p
Q Constitutive model parameter
rrod Radius of the rod
rroll Radius of the driving roll
s Deviatoric part of σ
t Time
u Material
√ displacement vector
Y 3/2 times the current radius of F
X, Y, Z Coordinate system defined on page 10
 Diameter

Greek symbols

α, αur , αlr Angles between rolls and bar

viii
 α Center of yield surface in σ space
α0 Constitutive model parameter
β Angle between ξ and ξ˜
γ, γi Constitutive model parameters
δ Stamping applied to the bar
ε Hencky total strain tensor
Υ Half amplitude of deviation from straight line
θ, r, ζ Coordinate system defined on page 94
κ Curvature
κy Maximum elastic curvature
κ Pitch of the helical path of a point
dλ Plastic mutliplier
ν Poisson’s ratio
µ Friction coefficient
ξ, η, ζ Coordinate system defined on page 81
˜ η̃
ξ, Coordinate system defined on page 98
σ Cauchy stress tensor
σy Yield stress
σ0 Constitutive model parameter
σ0.2 Yield stress at 0.2% plastic strain
Ω̇ Angular velocity of the rolls

Operators, subscripts and superscripts

(·) Normalized value
bs as
(·) , (·) Value before resp. after straightening
e
(·) Elastic
p
(·) Plastic
(·)mean Mean value
(·)std Standard deviation
H (·) Heaviside’s step function
J2 (·) 2nd invariant
∆ (·) Incremental value
h·i Macauley’s brackets

ix
Abbreviations
AISI American Iron and Steel Institute
CPU Central Processing Unit
DIN Deutsches Institut für Normung
FE Finite Elements
ISO International Organization for Standardization
SAE Society of Automotive Engineers
SD Strength Differential
TCA Tension Compression Asymmetry
TEM Transmission Electron Microscopy

x
Chapter 1

Introduction

1.1 High Strength Bright Steel Bars

1.1.1 Uses and requirements
Bright steel bars are semi-finished products that are used principally
to manufacture turned parts. The use of high-strength steels pro-
vides an obvious advantage for components subjected to high loads.
The higher oil pressures that can be withstood by hydraulic com-
ponents made from such steels is a typical example. The increasing
use of high-strength steels is one of the consequences of the quest for
lighter components, for example in the automotive industry.

High levels of straightness and low batch fluctuations are among

the other requirements that are set on high quality bright steel bars.
Straightness is crucial in that any deviation from it induces detri-
mental lathe vibrations during the turning process, leading to re-
duced achievable levels of precision. The increasing rotation speeds
used in turning operations lead to a constant increase in straigthness
requirements. Low batch fluctuations are a prerequisite for robust
manufacturing processes.

1
2 CHAPTER 1. INTRODUCTION

Figure 1.1: Production process of the bright steel bars investigated.

1.1.2 Production process

The products investigated in the framework of the present study
are bars of SAE 1144 25 mm in diameter. SAE 1144 is a resulfur-
ized medium carbon steel and belongs to the category of free cutting
steels, i.e. steels that can be well machined as they form small chips.

High end steels, such as the high strength steel sheets used in the
automotive industry, usually stem from primary steelmaking pro-
cesses, i.e. from the processing of pig iron. The production process
of the SAE 1144 on which the present study focuses (a high end
steel) belongs to the category of secondary steel making, i.e. scrap
material is used as raw material. The production of high steel qual-
ities from scrap steel is a technologically demanding process. Figure
1.1 summarizes the main production steps.

After the drawing operation, the bars exhibit the following char-
acteristics:
• a high yield stress value;
• large tensile residual stresses near their surface;
• a relatively large curvature.
1.2. STRAIGHTENING 3

Figure 1.2: Straightening / flattening processes according to DIN

24500-7.

Whereas a high yield stress is desirable, the tensile residual stresses as

well as the curvature still have to be corrected in order to meet the
requirements for high speed turning operations. These corrections
are made through the straightening process.

1.2 Straightening

1.2.1 Overview of the processes

The DIN 24500-7 standard groups the straightening processes in four

main categories, as illustrated in figure 1.2.
4 CHAPTER 1. INTRODUCTION

Roll straighteners
Processes belonging to this category are based on the principle of
straightening through alternate bending. Profile straighteners and
sheet levellers operate in a similar manner. The product passes
through a set of rotating rolls, as pictured in figure 1.3. The number
of rolls may vary but the principle remains the same. A series of
curvatures κi with |κi | < |κi−1 |, all lying in a single plane, is ap-
plied to the product over its journey through the rolls. The amount
of bending applied is defined by the positions δj , which can be varied.

For sheet leveling, a process in which 13 rolls are used is described

by Moses [1934]. The levelling process studied by Menz [2002] re-
lies on 7 rolls. Higher numbers of rolls allow to reach higher levels
of flatness. For wires1 and bars having a round cross section, this
type of straightening is generally used for small diameters. The wire
straightening process studied by Asakawa et al. [2010] (further details
in section 2.4.1) is an example of such a process. For the straighten-
ing of wire, bending is usually applied in two perpendicular planes,
as in the process studied by Balic and Nastran [2002]. Another ex-
ample of profile staightening is the railway rails straightening process
investigated by Schleinzer and Fischer [2001], which is described in
section 2.4.2.

In a cross roll straightener, at least one of the rolls is concave.

The main difference between this process and the ones mentioned
above is that the rotation axis of the rolls is set at a skew angle to
the forward motion of the product. While bending the product, the
rolls induce to it a rotation around its main axis. Obviously, this type
of processes can only be applied to products having a round cross
section, i.e. round bars or tubes. In addition to bending, the rolls
may also be positionned in such a way that the product is laterally
1 According to DIN EN 10079:2007, the term wire refers to drawn or rolled

products that are to be coiled. Generally, their diameters are smaller than 8 mm.
A straightened wire cut to a given length becomes a bar.
1.2. STRAIGHTENING 5

Figure 1.3: Principle of the five rollers straightener studied, for exam-
ple, by Asakawa et al. [2010]. The rolls impose a series of curvatures
κi on the product. The process parameters are the positions δj of the
lower rolls. Sheet levellers are based on the same principle although
more rolls are used (see e.g. Menz [2002]).

stamped (or ’crushed’). Figure 1.4 is a schematic representation of

the tube straightening process studied by Kuboki et al. [2010], which
is further explained in section 2.3.

The kinetics of cross roll straightening processes are explained in

section 1.2.2. Further in the text, if not otherwise specified, the term
cross roll straightening refers to the two-rolls straightening process,
on which the present study focuses.

Wing straighteners
Wing straighteners operate in a similar manner as the cross roll
straighteners. They allow the straightening of round products and
are mostly used to straighten bars of small diameters. The main
difference compared to the cross roll straighteners is that the rolls
are mounted in a frame that rotates around the bar. No rotation is
imposed on the product, just a forward motion. As the product does
not rotate, this kind of process is well suited to uncoil and straighten
wire.
6 CHAPTER 1. INTRODUCTION

Figure 1.4: An example of cross roll straightening process: a three

stands tube straightener.

Stretching machines

In the above mentioned processes, the products are straightened by

alternate bending. As it will be shown in section 1.2.2, this approach
presents the drawback of inducing a residual stress distribution in the
straightened product even in the absence of residual stresses before
the straightening operation. By stretching the product in question,
its whole cross section undergoes plastification. If the operation is
mastered, it is hence possible to produce almost residual stress free
products.

The main disadvantage of straightening through stretching lies in

the difficulty of continuously straightening, for example, coiled prod-
ucts.

The tension levelling process studied, among others, by Yoshida

and Urabe [1999] is used to uncoil and flatten metal sheets. The
setup is similar to the one pictured in figure 1.3 but, in addition to
alternate bending, an axial tension is imposed on the sheet. Using
this method, the difficulty of inducing plastification in the core of
the product, as explained in section 1.2.2, is overcome.
1.2. STRAIGHTENING 7

1.2.2 Principle of straightening through repeti-

tive bending
Pure bending
An initially straight beam, whose cross section is constant over its
length and has a plane of symmetry in axial direction, is considered.
A pair of symmetric torques are applied to both ends of this beam
in its plane of symmetry: the beam exhibits a constant curvature κ.
The neutral surface is defined as the surface at the interface between
the fibers in tension and those in compression. Its neutral axis (or
bending line) is defined as the intersection of the neutral surface and
the plane of curvature.

According to the Euler-Bernoulli assumption, plane cross sections

of the beam remain (1) plane after bending and (2) normal to the
neutral axis of the beam, (3) while the shape and dimensions of the
cross sections remain unchanged after bending. According to this
assumption, the axial engineering strain ǫ in a fiber situated at a
distance y from the neutral surface is related to the curvature of the
bar by
ǫ = yκ (1.1)
A straight round beam with a yield stress σy can be bent without
plastification up to a curvature
σy
κy = (1.2)
Errod
where rrod is the radius of the rod. Conversely, such a beam with an
initial curvature κbs < κy can be held straight without plastification,
as illustrated by the two sketches on top of figure 1.5. κy = 3.58·10−4
1/mm for SAE 1144 bars (25 mm).

Straightening
In order to illustrate the principle of straightening through alternate
bending, a simple example is considered. A beam with a constant
8 CHAPTER 1. INTRODUCTION

Figure 1.5: Principle of straightening through reverse bending with

a decreasing amplitude.

round cross section has a constant initial curvature κbs < κy in a sin-
gle plane. The beam is assumed to be free of initial residual stresses
and its mechanical behaviour is assumed to be isotropic and elastic-
perfectly plastic. Because κbs < κy , holding the beam straight does
not lead to plastification and, upon unloading, it would spring back
to κbs .

The example is illustrated in figure 1.5. A series of curvatures κi

such as |κi | < |κi−1 | is applied to the bar. Each one of the curvatures
imposed lead to plastification in the reverse direction with regard to
its predecessor. After unloading, the beam springs back to a curva-
ture κas and an axial residual stress distribution σres remains in the
beam. The goal is to determine κi in such a way that κas < κbs .
1.3. CROSS ROLL STRAIGHTENING 9

Of course, knowing κbs , the beam could be straightened in a sin-

gle bending operation by applying a curvature κ1 such as κas = 0. In
this case, κ1 would have to be determined for each κbs considered. By
applying mutliple bendings to curvatures κi and by correctly choos-
ing κi , the method allows to reach an approximately unique final
curvature κas that is relatively independent of κbs .

The main drawback of this approach is that, due to the non-

uniform deformation that occurs in the bar, a residual stress distri-
bution remains in the straightened product. The superposition of an
axial tensile load, as for the tension levelling of sheets mentioned on
page 6, allows to circumvent this problem.

1.3 Cross Roll Straightening

1.3.1 Description of cross roll straightening
As mentioned in section 1.1.2, cross roll straightening aims at im-
proving some of the defaults that remain in the bar after the drawing
operation. The process is also referred to as Medart process after its
inventor Medart [1896]. The principal improvements achieved in the
bar through cross roll straightening are:

• the reduction of its undesirable curvature;

• the reduction of the tensile residual stresses on and close to its

surface;

• the homogenization of its curvature;

• the polishing effect (hence the name bright steel ).

On the downside, for materials that exhibit cyclic softening, the

straightening operation leads to a reduction of the yield stress of
the bars. Experimental results and further explanations are given
10 CHAPTER 1. INTRODUCTION

Figure 1.6: Sketch of the cross roll straightening process.

in chapter 3. The specific investigation of the polishing effect is not

within the scope of the present study.

Figure 1.6 shows a sketch of the cross roll straightening process.

During straightening, the bar passes between two rotating rolls, a
concave and a convex one. The bar is bent and stamped in the Y Z
(vertical) plane. The axes of the rolls are parallel to the XZ plane.
The axis of the concave (upper) roll lies at an angle αur to the Y Z
plane; the convex (lower) roll lies at an angle αlr to the Y Z plane.
The rolls rotate and induce both a forward feed to the bar and a
rotation around its main axis. If the distance d between the rolls is
set to a value smaller than the diameter of the bar, stamping δ is
applied in addition to repetitive bending:

δ = h2rrod − di (1.3)

The amount of stamping applied increases with increasing bar

diameter: bars of small diameter are straightened without stamping.
Stamping is particularly useful to achieve high surface qualities and
compressive residual stresses at the surface of the bar. For the pro-
cess under study, neither ovalization nor change in diameter could
be observed on straightened bars.
1.3. CROSS ROLL STRAIGHTENING 11

Figure 1.7: Tangential and forward speed components of a point P

on the surface of the bar.

1.3.2 Kinetics of cross roll straightening

As mentioned above, the rolls induce to the bar both a rotation
around its main axis and a forward motion. Neglecting stamping,
the bar undergoes bending and a material point endures hence alter-
nate tension-compression, as shown in figure 1.8. The amplitude of
the deformation first increases, reaches a peak in the vicinity of the
center of the setup, and decreases again.

The velocity of the rotating rolls splits into tangential and axial
components in the bar. Neglecting slip, the velocities of a point on
the surface of the bar (see Figure 1.7) are:

θ̇ = Ω̇rroll cos α tangential velocity

(1.4)
ζ̇ = Ω̇rroll sin α axial velocity

where θ̇ and ζ̇ define, respectively, its tangential speed and its forward
velocity (a formal definition of the coordinate system is given in
section 5.3.2). The angular velocity Ω̇ of the rolls is transmitted to
the bar through the friction force. In cross roll straightening with
lateral stamping, the highest pressure exerted by the rolls on the bar
is in the central region. Therefore, although the rolls are curved,
rroll corresponds to the radius of the rolls at the point of maximum
stamping, i.e. in the middle. For the same reason, the torsion that is
applied to the bar due to the different velocities at the surface of the
12 CHAPTER 1. INTRODUCTION

Figure 1.8: Neglecting stamping, the principal deformation induced

to a material point of the bar is uniaxial tension-compression.

Figure 1.9: The angle α influences the pitch κ and the amount of
bending applied to the bar, i.e. primarily the axial strains. The
amount of stamping δ influences primarily the magnitude of the ra-
dial strains in the bar.
1.4. OBJECTIVES OF THE STUDY 13

rolls is assumed to be negligible. The value κ is the axial distance

required for a point of the bar to complete a rotation: the pitch of
its helical path. This value is computed from relations (1.4) as

κ = 2πrrod tan α (1.5)

The influence of the angle α on the deformation is twofold, as shown

on the left hand side of figure 1.9:
• a small α leads to a decreases of the pitch κ: the forward
velocity of the bar is reduced and the number of rotations per
given length increases;
• a small α increases the bending imposed on the bar: the am-
plitude of the axial deformation increases.
Without stamping, the deformation is uniaxial according to the as-
sumptions made in section 1.2.2. By decreasing the distance between
the rolls, i.e. by increasing the stamping, the radial deformation im-
posed on the bar increases. The influence of δ is illustrated qualita-
tively on the right hand side of figure 1.9.

The mechanism of the development of the residual stress distri-

bution in a beam during straightening through alternate bending in
a single plane is explained in section 1.2.2. Tokunaga [1961] con-
ducted a similar development for the case of a round bar subjected
to multiple bendings and took into account the rotation of the bar
around its principal axis. Due to the simultaneaous rotation and
forward motion of the bar, the resulting residual stress distribution
in a section of the bar exhibits the helical periodicity of the path of
the contact line.

1.4 Objectives of the Study

The state-of-the-art in the virtual modelling of the cross roll straight-
ening process is presented in chapter 2. The existing approaches to
14 CHAPTER 1. INTRODUCTION

the modelling of the process can be grouped in two main categories:

the analytical ones and the ones based on FE modelling. The former
allow fast computations but, as they rely on important simplifica-
tions, they are not able to take into account such influences as the
stamping δ. The FE based approaches allow to lift some of the sim-
plifying hypotheses but, due to their large computation costs, they
impose the use of rough meshes that do not allow to reach sufficiently
high levels of precision for demanding engineering uses.

Cross roll straightening is not yet supported by a virtual model.

Straightening operations are set up based on empirical knowledge,
and trial and error. Apart from being time consuming, this approach
does not guarantee operations under optimal process parameters.
The development of a virtual support to cross roll straightening op-
erations shall allow, among others, to:

• develop a simulation-based control system;

• optimize the geometries of the rolls;

• reduce the time required for the setup.

The present study focuses on the first step towards these long term
goals: the development of a simulation model for the cross roll
straightening process. More specifically, this model is required to
allow the prediction of:

• the yield stress of straightened bars;

• the remaining curvature after straightening;

• the resulting residual stress distribution.

The principal boundary condition for this model is a reasonable use

of CPU resources. To reach the goals mentioned above, the main
milestones are: (1) the thorough understanding of the deformations
taking place, (2) the correct description of the cyclic behaviour of the
1.4. OBJECTIVES OF THE STUDY 15

material and (3) the development of an adequate simulation model.

In this study, a unique set of rolls is considered and the process

variables are the angle α = αur = αlr and the stamping δ. The
material considered is a special type of SAE 1144 (25 mm) bars.

Summary of Chapter 1
The present study aims at developing a simulation model for the cross
roll straightening process. The straightening of SAE 1144 (25 mm)
bars is taken as an example.

Cross roll straightening is a sub-category of the roll straightening

processes. Straightening occurs through the alternate bending of the
product. The rotating rolls induce both a rotation of the product
around its principal axis and its forward motion.

In the process under study, lateral stamping is applied to the

product in addition to bending. This stamping leads to a three
dimensional cyclic deformation of the material.
16 CHAPTER 1. INTRODUCTION
Chapter 2

State of the Art

The cross roll straightening technology is presented in chapter 1.

This chapter presents a review of the state of the art of the mod-
elling techniques.

A summary of the analytical modelling methods is presented in

section 2.1. The mixed analytical-numerical approach proposed by
Furugen and Hayashi [1985], which is particularly relevant for the
procedure proposed in chapter 5, is explained in section 2.2. In sec-
tion 2.3, FE modelling approaches are commented. Selected strate-
gies to model related processes, such as the roller leveller straighten-
ing of wire, are presented in section 2.4.

2.1 Analytical Approach

2.1.1 Theory
Assuming that straightening occurs through pure alternate bending,
Das Talukder and Johnson [1981], Das Talukder and Singh [1991], Yu
and Johnson [1981], Yu and Zhang [1996], Wu et al. [2000] proposed
similar methods to predict the remaining curvature of a bar after

17
18 CHAPTER 2. STATE OF THE ART

Figure 2.1: (a) Bar subjected to a bending torque around the axis
X, (b) bilinear elastic-plastic hardening curve and (c) axial stress
distribution in a section of the bar.

cross roll straightening. These methods are all based on the same
ideas and the equations presented in this section are a synthesis of
them.

A bilinear isotropic hardening model is considered


Eǫ for |y| ≤ yc
σ= σ
(2.1)
σy + Etan ǫ − Ey for |y| > yc

where the axial strain ǫ is computed according to relation (1.1) and

yc is defined as
σy
yc = (2.2)
Eκ
yc correspond to the distance in the Y Z plane from the neutral axis
under which no plastification occurs. The bending torque on a sec-
tion of a round bar is given by
Z rrod q
1 2
M= σzz (y) y rrrod − y 2 dy (2.3)
4 0

The bending torque required to bend the beam to a curvature (1.2)

is referred to as the maximum elastic bending torque and is written
2.1. ANALYTICAL APPROACH 19

My . The normalized curvature and torque are defined as

κ M
κ= M= (2.4)
κy My

Computing the integral of equation (2.3) for the purely elastic case
(κ ≤ κy ) yields the well known relation
π 4
M= κErrod (2.5)
4
The limit elastic torque My can be computed from relations (1.2)
and (2.6). Using the normalized values introduced above, relation
(2.6) becomes
M (κ) = κ for κ ≤ κy (2.6)
and relation (2.3) yields
"  3  
4 1 2 Etan
M (κ) = sgn (κ) 1− 2 1−
3π κ E
  r !
2 Etan 1 1 (2.7)
+ 1− 1 − 2 + |κ| arcsin
π E κ |κ|

Etan
+|κ| for κ > κy
E

for a mixed elastic-plastic deformation.

A series of curvatures κtot

i with i ∈ [1, n] are imposed by the tools
to the bar. The corresponding curvatures of the bar after springback
are κi , with κ0 = κbs and κn = κas . The value ∆κ is defined as

κtot
i − κi−1
∆κ = (2.8)
κy

The negative sign before κi−1 takes into account the half rotation of
the bar between the steps i − 1 and i. Depending on the value ∆κ,
20 CHAPTER 2. STATE OF THE ART

Figure 2.2: Relation between curvature and torque (relations (2.6)

and (2.7)).

two cases are distinguished (see Figure 2.2):


κi =
κi−1

if |∆κ| ≤ κy (2.9)
κi−1 + κy ∆κ − sgn M (∆κ) else

2.1.2 Example
The analytical procedure presented above is applied to the case of
the SAE 1144 (25 mm) bars. A batch of such bars is investigated
and the experimental results are presented in chapter 3. The mean
value of the initial curvatures measured on the batch under study
(κbs
mean = 3.12 · 10
−6
1/mm) is taken as initial curvature. The bend-
ing imposed by the rolls when set at α = 18◦ , taking into account
the pitch κ = 25.5 mm, is plotted in Figure 2.3 (left). The analyti-
cal procedure is applied assuming the material has an elastic-plastic
behaviour defined as follows: E = 210 GPa, σy = 935 MPa and
Etan = 3000 MPa. The computations are made twice: once with κbs
as initial curvature and once with −κbs .

The resulting straightness as a function of the number of bend-

ings is plotted in Figure 2.3 (right). It appears that only one bending
leads to the plastification of the bar. When the initial curvature is
2.2. MIXED ANALYTICAL-NUMERICAL APPROACH 21

Figure 2.3: Curvature imposed by the rolls set at α = 18◦ (left)

and absolute values of the computed bar curvature according to the
analytical procedure (right).

taken as positive, the absolute value of the final curvature is higher.

When taken as negative, the absolute value of the final curvature is
lower. The mean curvature of straightened bars (see Figure 3.5) is
κas
mean = 6.34 · 10
−7
1/mm.

This example highlights the fact that the final curvature mea-
sured cannot be reasonably predicted using the analytical method.

2.2 Mixed Analytical-Numerical Approach

Furugen and Hayashi [1985] used a mixed analytical-numerical ap-
proach to simulate the straightening of tubes in a four stands cross
roll straightener, which is similar to the process pictured in Figure
1.4. The total strain field method presented in chapter 5 is based on
similar assumptions as the ones made here.

A central aspect of the procedure is that the computation of

the strains is conducted separately from the analysis of the stresses.
22 CHAPTER 2. STATE OF THE ART

For the determination of the strains1 , the following assumptions are

made:
• the bending and the crushing2 of the tube can be analyzed
separately;
• the tube is not rotated and fed;
• the tube is straight and round before and after straightening.
On the basis of these assumptions, the longitudinal (εζ ) and tan-
gential (εθ ) strains are computed as a function of the longitudinal,
tangential and radial coordinates in the tube. Technically, the εζ
strains are obtained by means of a numerical simulation while the εθ
strains are computed analytically knowing that a crush δ is applied
by each pair of rolls. The natural ovalization that occurs when a
tube is bent is introduced into the calculation through an empirical
function.

For the stress analysis, the position of any given point on a cross
section of the tube as it passes through the rolls is computed knowing
the rotation and the forward feed of the tube. The strain increments
dεζ and dεθ are computed on the basis of those paths and the dis-
tribution of strains in the tube. Assuming a plane stress state (no
stress component in the radial direction, σr = 0), the stress incre-
ments dσζ and dσθ are computed using the mixed isotropic-kinematic
set of constitutive equations proposed by Kishi and Tanabe [1973].

The predictive capabilities of the model were tested by compar-

ing the theoretical and experimental loads on the rolls. The values
are in good agreement over a large force domain. Using the Sachs
method3 , axial and tangential residual stresses in a straightened AISI
1 Furugen and Hayashi [1985] does not explicitly define which definitions of

strain and stress are considered.

2 crushing corresponds, in the tube straightening terminology, to stamping.
3 The Sachs method is used to assess the residual stresses in bars and tubes

by measuring the strains that appear subsequently to the removal of material

from the center of the cylinder (see Olson and Bert [1966] for further details).
2.3. FINITE ELEMENT APPROACHES 23

1020 tube (out 60 mm and in 50 mm) were measured. The maxi-
mum deviation of the predicted values from the experimental ones is
approximately 40%. Considering the limitations of the Sachs method
regarding the assumption of linear material behaviour, as Parker
[2004] pointed out, the results obtained by Furugen and Hayashi
[1985] can be considered very good.

2.3 Finite Element Approaches

2.3.1 Kuboki and co-workers
Kuboki et al. [2010] investigated different FE schemes to model the
staightening of AISI 1045 tubes (out 34 mm and in 30.8 mm). The
three stands tube straightener under study is pictured in Figure 1.4.

The tube was meshed with solid elements: 2 in the thickness

and 43 on the circumference while the material behaviour was as-
sumed isotropic. Different models were built, using both implicit
and explicit formulations and different simplifications were tested.
Predictions from the different models regarding geometrical quanti-
ties were compared. The computation times varied between 2 and
25 days on a dual core workstation (in 2010) for, respectively, a
simplified explicit model and a more precise implicit model. The
long computation times can be attributed to the long deformation
paths imposed to the bar and, simultaneously, to the small time steps
which are necessary to take into account the complex contact that
takes place between the tube and the rolls due to the incremental
forming process.

This study highlights the large CPU cost associated with the
approach that consists in modelling the whole process. Also, the
precision of the model is still to be proved, as no comparison between
experiments and predictions was provided.
24 CHAPTER 2. STATE OF THE ART

Figure 2.4: In the modelling approach presented by Onoda et al.

[2002], the deformation is induced in the bar through the control of
the node displacements ∆u.

2.3.2 Asakawa and co-workers

Asakawa and co-workers [Onoda et al., 2002, Yanagihasi et al., 2005]
modelled the cross roll straightening of S45C bars (10 mm) using a
static explicit FE scheme. In the model, it is assumed that straight-
ening occurs purely through atlernate bending; lateral stamping is
not taken into account.

Figure 2.4 shows the modelling approach used. The incremental

position of each node of the bar is defined by a vector ∆ui , where i
is the node number. ∆ui is computed from the profile of the tools
and the progress of the bar. The exact procedure is not explained.
Presumably, it is obtained either through a FE simulation of the
bending of the bar or through an analytical approach, as described
by Fangmeier [1966]. The procedure was applied to a 80 mm seg-
ment of the bar, meshed with three elements in the radial direction,
10 around the circumference and 20 lengthwise.

Comparisons between axial strains measured during straighten-

ing and strains predicted with the procedure yielded a very good
agreement. The initial curvature is κbs = 7.8 · 10−5 1/mm. Af-
ter straightening, the predicted curvatures lie in the range 2 · 10−7 <
2.4. MODELLING OF RELATED PROCESSES 25

κas < 2·10−6 1/mm depending on the number of bending cycles. The
measured ones lied about an order of magnitude higher. To achieve
those values, the rolls induce a maximum curvature κtot = 2 · 10−3
1/mm to the bar.

2.4 Modelling of Related Processes

The studies presented above deal either directly with the cross roll
straightening of bars or with closely related processes, like the straight-
ening of tubes. In this section, reports of simulations of similar pro-
cesses are commented. The studies presented here contain insights
that lead to a better understanding of the modelling possibilities for
the cross roll straightening process.

2.4.1 Roller straightening of wires

The process studied by Asakawa et al. [2010] is pictured in figure
1.3. The deformation that takes place is uniaxial. Normally, the
straightening of wires is conducted with two sets of rolls normal to
each other. In the present case, only one set of rolls is considered.

Static implicit simulations of the whole process were conducted

with the commercial FE code MSC. Marc. The wire (6 mm)
was discretized with solid elements (three over the radius). The
Bauschinger effect observed in the medium carbon steel investigated
was described using a non-linear mixed isotropic-kinematic set of
constitutive equations (the Chaboche model, which is explained in
more detail in chapter 4). During the simulation, the forward mo-
tion of the central node on the front face of the wire was prescribed.
Wires curved in two and three dimensions were studied. The result-
ing straightness of the bar as well as its tilting4 during the process
were evaluated.

4 The tilting angle is a measure of the torsion encountered by the wire.

26 CHAPTER 2. STATE OF THE ART

The comparison between straightness and experiments are in very

good agreement. In both cases, a limit value of the end straightness
of bars initially bent in three dimensions is observed. The fact that
the deformation that takes place is purely uniaxial is a possible ex-
planation of the quality of the results in spite of the rough meshing
adopted to discretize the wire. In fact, the curvatures obtained after
straightening are of the order 0.5 · 10−4 < κas < 2.5 · 10−4 1/mm
(the curvature of the bars before straightening is not mentioned in
the article).

The limit value of straightness that can be reached with bars hav-
ing an initial three dimensional curvature is explained by the tilting
of the bar that occurs during the process. Having recognized this, a
device in which the straightening plane can be rotated (similar to the
wing straightener mentioned in chapter 1) was proposed, enabling a
better straightening. This last point is interesting in that it shows
the limitation of the attempt to straighten bars with an out-of-plane
curvature by repetitive bending in a single plane. This detrimen-
tal phenomenon is avoided in cross roll straightening through the
continuous rotation of the bending plane of the bar.

2.4.2 Roller straightening of rails

The development of residual stresses in railway rails due to the
straightening process has been the subject of numerous recent stud-
ies [Finstermann et al., 1998, Schleinzer and Fischer, 2000, 2001,
Betegon Biempica et al., 2009]. The popularity of this research topic
is likely due to the high stakes related to the train derailment risk
that can be caused by excessive residual stresses in rails. Among
the studies mentioned above, the approach of Schleinzer and Fischer
[2000, 2001] is of particular interest.

Railway rails are straightened in roller straighteners, as shown in

figure 1.3. The rail is repeatedly bent in its vertical plane (the one
with the largest moment of inertia) as it passes through the rolls. As
2.4. MODELLING OF RELATED PROCESSES 27

for the wire, the deformation induced is hence less complex than the
one that develops during cross roll straightening with lateral stamp-
ing.

Schleinzer and Fischer [2001] conducted their analysis using a

submodelling approach. The computation was made in two steps.
First, a global FE simulation was conducted with the commercial
FE code Abaqus/Explicit. The mesh, pictured on the right hand
side of Figure 2.5, was relatively rough and the material behaviour
was modelled using a simple linear kinematic hardening model. The
position of the nodes on two sections, which define the submodel
in the subsequent step, within the bar were recorded. Then, in the
second step, only one submodel, i.e. a segment of the rail, was con-
sidered. For this detailed analysis, the mesh was much finer and the
cyclic behaviour of the material was modelled using a similar model
as [Asakawa et al., 2010]. The analysis was run a second time with
the submodel alone. The deformation was imposed to the submodel
by prescribing the displacements of the nodes that lied on its faces.
To do so, the displacement values of the nodes recorded during the
global simulation were interpolated.

The purpose of the study of Schleinzer and Fischer [2001] was to

predict the axial residual stresses that appear in the rail. Schleinzer
and Fischer [2001] do not provide a quantitative comparison between
measurements and their predicted results. A computation time of
about 10 days on a single core workstation (in 2001) is reported.

The results presented in the different studies mentioned above,

along with the experimental results obtained by [Webster et al., 1993]
using neutron diffraction are plotted in Figure 2.6. This provides a
measure of the precision that can be reached in the prediction of
residual stresses in straightened products.
28 CHAPTER 2. STATE OF THE ART

Figure 2.5: Sketch of the submodelling approach used by Schleinzer

and Fischer [2001] to model the rail straightening process.

Figure 2.6: Axial residual stresses in a straightened railway rail ac-

cording to different authors and corresponding experimental results.
2.4. MODELLING OF RELATED PROCESSES 29

Summary of Chapter 2
For the simple case of straightening under pure alternate bending,
predictions of the resulting curvature of bars can be made using an-
alytical approaches. In the case of the bars investigated, the results
obtained are not accurate enough.

Existing FE approaches are CPU cost intensive and relatively

coarse meshes are used (see Kuboki et al. [2010], Onoda et al. [2002],
Yanagihasi et al. [2005]). To the knowledge of the author, the only
study in which the influence of lateral stamping is taken into account
is the one of Kuboki et al. [2010].

The mixed analytical-numerical approach used by Furugen and

Hayashi [1985] allowed to make good predictions of the residual stress
state in straightened tubes. The submodelling approach used by
Schleinzer and Fischer [2001] allowed to predict accurately the axial
stress distribution in straightened rails.

It appears from the review presented in this chapter that the

existing modelling approaches do not allow to model in a satisfactory
manner the cross roll straightening process with lateral stamping. As
it will be shown in chapter 3, the phenomena observed in straightened
bars are relatively subtle and their prediction requires hence a high
level of accuracy.
30 CHAPTER 2. STATE OF THE ART
Chapter 3

Experiments and
Observations

This chapter presents experimental results showing the principal phe-

nomena induced by cross roll straightening. Experiments related to
the investigation of the material behaviour are presented and dis-
cussed in chapter 4. In section 3.1, experimental results showing the
influence of the process parameters on the yield stress after straight-
ening are presented. The straightening effect of the process is pre-
sented in section 3.2 by comparing measurements of the curvature
of bars before and after straightening. In section 3.3, the result of
straightening on the residual stress distribution is dealt with on the
basis of selected values from the literature.

3.1 Yield Stress

3.1.1 Methodology
The principal mechanical property defining the quality, and hence
the price, of bright steel bars is σ0.2 , i.e. the yield stress at 0.2%

31
32 CHAPTER 3. EXPERIMENTS AND OBSERVATIONS

plastic strain (as defined in ISO 6892-1). In the industrial practice,

the type of specimen used to perform the required tensile tests de-
pends on the diameter of the product. For bars 25 mm in diameter,
the tensile tests are conducted on the whole section of the bar and
are referred to as type I tests further in text. The stress-strain curves
obtained using type I tests represent therefore an average behaviour
of the material over the cross section of the bar. For bars in larger
diameter, σ0.2 is determined on the basis of tensile tests which are
conducted on specimens that are cut from the bars. In order to as-
sess the homogeneity of the bars, tensile tests using this second type
of specimens are also conducted. These tests, named type II tests
further in text, are conducted with DIN 50125–B 6x30 specimens
cut at half the radius of the bars1 . The sections of the two types
of specimens are shown in Figure 3.1. Both types of tensile tests
are conducted on bars before and after straightening with different
values of the process parameters.

3.1.2 Phenomena observed

A stress-strain curve obtained from a type II tensile test is shown
in Figure 3.2. The σ0.2 values obtained from the different tests are
plotted in Figure 3.3.

Considering the σ0.2 values obtained from type I tests on bars

bs as
before and after straightening (respectively σ0.2 and σ0.2 further in
text), it appears that:
as bs
• σ0.2 is lower than σ0.2 ;
as
• σ0.2 increases with increasing values of α;
as
• σ0.2 decreases with increasing values of δ.
1 The presence of a thread on DIN 50125 specimens prevents from cutting the

tensile test specimens closer to the surface of the bars.

3.1. YIELD STRESS 33

Figure 3.1: Type I tensile tests are conducted on the whole sections of
the bars, i.e. no specimen is cut. Type II tensile tests are conducted
on 6 mm specimens cut at half the radius of the bars.

Figure 3.2: Engineering stress-strain curve obtained from a Type II

tensile test on a bar before straightening.
34 CHAPTER 3. EXPERIMENTS AND OBSERVATIONS

Figure 3.3: Measured values of σ0.2 using the two types of tensile tests
before and after straightening with different process parameters.

The reduction of σ0.2 observed with type I tensile tests does not
appear clearly with type II tensile tests. The only experiment for
as bs
which σ0.2 < σ0.2 is the one for which the largest amount of stamping
bs
is applied. The value of σ0.2 is 940 MPa using type I tests and 929
MPa using type II tests. This deviation of 11 MPa can be explained
either by the inaccuracy of the measurement or by inhomogeneities
as
due to the drawing operation. The discrepancy between the σ0.2
values obtained using the two types of tests is too important to be
attributed solely to measurement errors. The results obtained using
type II tests do not show a clear dependency on the parameters α and
as
δ. Considering the small difference between the σ0.2 values obtained
using the two different tests before straightening, it seems that the
inhomogeneous distribution of the properties results primarily from
the straightening operation.
3.2. CURVATURE 35

Figure 3.4: Experimental setup to measure the curvature of bars.

L = 1 m in the device used.

3.2 Curvature
The curvature of the bars before straightening can be principally
explained by

• imperfections in the drawing operation;

• an unsuitable handling between the drawing and the straight-

ening operations.

Examples of imperfections in the drawing operation are: (1) the non

perfectly axis-symmetric friction between the bar and the drawing
die and, (2) geometrical imperfections of the die itself. These im-
perfections lead to an inhomogeneous flow of material through the
die, which, itself, results in a curvature of the bar. This first type
of curvature is generally constant over the length of the product and
lies in a single plane. Important impacts during the handling may
induce plastic deformation in the product. This second type of de-
fects is generally located close to the ends of the product and leads
to angular shapes of the bar. This study focuses on the first type of
curvature before straightening: constant and in a single plane. The
characteristics of the curvature of the bars after straightening are
discussed in chapter 6.
36 CHAPTER 3. EXPERIMENTS AND OBSERVATIONS

3.2.1 Methodology
The curvature of the bars is experimentally measured using the setup
pictured in figure 3.4. A dial gauge is fixed in the middle of the de-
vice, the bar is rotated and 2Υ, the maximum amplitude of the dial
gauge, is recorded. This setup allows, in a simple manner, to com-
pensate for the effect of gravity on the bar. If not otherwise stated,
the curvatures are measured in the middle of the bars. The main
drawback of this method is that it can only assess constant curva-
tures that lie in a single plane of bending. Asakawa et al. [2010]
investigated the roller leveller straightening of wires and measured
the curvatures using a similar procedure, the dial gauge being re-
placed by a laser gauge. The curvature κ is computed as

8Υ 8Υ
κ= ≈ 2 (3.1)
4Υ2 +L 2 L
A batch of drawn bars with similar curvatures after the drawing
operation is selected. In order to assess the robustness of the process,
40 bars from the batch are taken and straightened in a reference
configuration: α = 18◦ and δ = 0.50 mm. The influence of variations
of the process parameters is investigated by measuring the curvature
of bars straightened with different process parameters. In this second
case, the curvatures of 3 bars are measured for each configuration.

3.2.2 Phenomena observed

The curvatures κ before and after straightening (κbs and κas , respec-
tively, further in text) of the 40 bars straightened with α = 18◦ and
δ = 0.50 mm are plotted in Figure 3.5. With a correlation coefficient
of 0.04, κbs and κas are not linearly correlated. From Figure 3.5, it
appears that a higher order correlation is not worth investigating.
The straightening effect is obvious: the mean value of the curvatures
of the bars before straightening is κbs mean = 3.12 · 10
−6
1/mm and
after straightening κas
mean = 6.34 · 10 −7
1/mm.
3.2. CURVATURE 37

Figure 3.5: Curvatures measured on N = 40 bars before and after

straightening with α = 18◦ and δ = 0.50 mm.

Figure 3.6: Curvatures measured after straightening with different

values for the process parameters. The mean values are plotted as
crosses. The distribution shown schematically on the right corre-
sponds to values of κas plotted in Figure 3.5.
38 CHAPTER 3. EXPERIMENTS AND OBSERVATIONS

In the framework of the investigation concerning the influence

of the process parameters on κas , the angle α and the stamping δ
are varied independently. The κas measured on the bars straight-
ened with the different process parameters are plotted in Figure 3.6.
It appears that all the mean values of the curvatures measured lie
within the standard deviation of the results obtained with the 40 bars
staightened in the reference configuration. Also, no logical relation
can be established between the variation of the process parameters
and the corresponding curvatures after straightening. Different in-
terpretations can be given to these observations: (1) the variations of
the process parameters within the range studied do not lead to sig-
nificant changes in the curvatures after straightening and/or (2) the
setup employed is not accurate enough to assess the effect and/or (3)
the assumption that the bars have a constant curvature in a single
plane after straightening does not hold.

3.3 Residual Stresses

Tensile residual stresses at the surface of parts promote crack initia-
tion and can considerably reduce the life of engineering components.
The drawing operation induces this kind of undesirable tensile resid-
ual stresses at the surface of the bars. Straightening is a way to
reduce, or even to annihilate, them.

The proper measurement of residual stresses requires expertise

that is beyond the scope of the present study. In order to measure
the residual stresses in parts, two types of techniques are commonly
used: the destructive ones and the ones based on diffraction. Withers
and Bhadeshia [2001] provide a review of residual stress measurement
methods. The destructive methods basically rely on the removal of
material and the measurement of the deformation of the material in
the vicinity. Residual stresses induce changes in interplanar spacing
that can be detected by measuring the Bragg scattering angle of a
beam whose wavelength is known. X-ray beams only allow to make
3.3. RESIDUAL STRESSES 39

measurements at the surface of the material (down to a few microm-

eters) whereas measurements down to a few centimeters in steel can
be done with neutron beams.

Davis and Mills [1990] measured residual stresses in SAE 1045

bars 38 mm in diameter before and after straightening. The method
employed for the measurement of these stresses is not specified but,
as values are given over the radius of the bar, it is likely that either
a classical destructive procedure or a neutron scattering method was
used. The effect of straightening on the reduction of the axial residual
stresses close to the surface appears clearly from the values plotted
in Figure 3.7. It also appears that the tangential residual stresses
at the surface are tensile before straightening and compressive after-
wards. Fangmeier [1991] presents axial residual stresses before and
after straightening measured on DIN SPb23 bars 20 mm in diameter
that agree qualitatively with the values presented in Figure 3.7.

X-ray goniometer measurements of axial residual stresses at the

surface of a SAE 1144 rod 20 mm in diameter are presented in Fig-
ure 3.8. Although the absolute values obtained may be subject to
measurement errors, it appears that the values vary cyclically over
the length of the bar. Similar values measured before straightening
are approximately constant. The inhomogeneity of the deformation
induced by straightening is already shown in section 3.1.2. The fact
that the residual stresses are not axis-symmetric is a second indica-
tion of this inhomogeneity. Hence, the measurements presented by
Davis and Mills [1990], which are considered axis-symmetric, do not
provide a full picture of the residual stress distribution after straight-
ening.
40 CHAPTER 3. EXPERIMENTS AND OBSERVATIONS

Figure 3.7: Measurements of residual stresses in a SAE 1045 (38

mm) presented by Davis and Mills [1990].

Figure 3.8: Axial residual stresses measured at the surface of a SAE

1144 rod (20.3 mm) straightened with stamping.
3.3. RESIDUAL STRESSES 41

Summary of Chapter 3
The yield stresses obtained from tests conducted on straightened
bars are lower than the ones before straightening. This effect does
not appear when specimens are cut from the bars. The difference
in the results obtained is a first indication that an inhomogeneous
deformation takes place during straightening.

Measurements of the curvatures of bars before and after straight-

ening show that a straightening effect does take place. The cur-
vatures measured after straightening are not correlated to the cor-
responding values before straightening. Also, no influence of the
straightening parameters is detected on the curvatures after straight-
ening.

The measurements presented by Davis and Mills [1990] show the

reduction of the residual stresses due to straightening. Measurements
of the residual stresses on the surface of straightened bars exhibit a
periodic pattern. This pattern is a second indication of an inhomo-
geneous deformation during straightening.
42 CHAPTER 3. EXPERIMENTS AND OBSERVATIONS
Chapter 4

Material Behaviour and

Modelling

A typical stress-strain curve obtained from a type II specimen cut

from a bar before straightening is plotted in figure 3.2. Without
further information about the behaviour of the material, it could be
assumed, as Wu et al. [2000] did, that the material can be modelled
using a bilinear elastic-plastic isotropic hardening model. A material
model which accounts only for isotropic hardening is obviously not
a judicious choice for the prediction of the softening effects shown
in Figure 3.3. As explained in chapter 2, the cross roll straightening
process induces a cyclic deformation in the bar.

The cyclic behaviour of the material is investigated in section

4.1. Microstructural explanations of the phenomena identified are
presented in section 4.2. In section 4.3, the macroscopic constitutive
equations chosen to describe the material are presented and the fit-
ting procedure used to determine the model parameters is explained.

43
44 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING

Figure 4.1: Geometry of the specimens used for the tension-

compression tests.

4.1 Tension-Compression Tests

The material under study is a SAE 1144 medium carbon steel. Tension-
compression tests are conducted on specimens as pictured in Figure
4.1. A clip-on length extensometer with 15 mm between the arms,
centered in the middle of the central region of the specimen, is used
to record the displacements. The tests are carried at a strain rate of
approximately 5 · 10−3 1/s. The force is recorded during the tests.

The cyclic stress-strain curve obtained from a test conducted with

a total strain amplitude ∆ε/2 = ±1.5 · 10−2 is shown in Figure 4.2.
The peak stress values (σmax for tension branches1 and |σmin | for
compression branches) are plotted in Figure 4.3.

Figure 4.4 summarizes the principal effects observed in the tension-

compression tests. Appart from strain hardening, the material ex-
hibits a strong Bauschinger effect (Figure 4.4 (b)), i.e. its yield stress
after a load reversal is lower than the one in its pristine state. Fig-
ures 4.4 (c) and (d) represent the stress extrema reached at each
branch. Considering only the maximum stresses from the tensile
1 A cycle consists of two branches: a tension branch (upgoing) and a com-

pression one (downgoing).

4.1. TENSION-COMPRESSION TESTS 45

1000
800
600
400
200
σ [MPa]

0
−200
−400
−600
−800
−1000
−0.02 −0.015 −0.01 −0.005 0 0.005 0.01 0.015 0.02
ε [−]

Figure 4.2: Experimental cyclic stress-strain curve obtained from a

tension-compression test on SAE 1144.
1000

950
|σ peak| [MPa]

900

850

800

750
0 5 10 15 20 25
j (branch number)

Figure 4.3: Peak stress values from the stress-strain curve plotted in
Figure 4.2.
46 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING

Figure 4.4: Principal phenomena observed in the tension-

compression tests: (a) strain hardening, (b) Bauschinger effect, (c)
cyclic softening, and (d) apparent tension-compression asymmetry.

branches, it appears that their values decrease with the increase in

the number of cycles (Figure 4.4 (c)). This phenomenon is referred
to as cyclic softening. Figure 4.4 (d) shows that the peak stresses
exhibit an asymmetry between their values in tension versus those
in compression.

4.2 Microstructural Phenomena

The principal phenomena that are observed in the tension-compression
experiments conducted are shown in Figure 4.4. In this section, dif-
ferent microstructural theories explaining those phenomena are pre-
sented.
4.2. MICROSTRUCTURAL PHENOMENA 47

4.2.1 Strain hardening

The positive slope of the stress-strain curve in Figure 4.4 (a) is char-
acteristic of the strain (or work) hardening phenomenon. When
medium carbon steels are plastically deformed at room temperature,
the energy brought to the system induces both a movement of the
dislocations and the creation of new dislocations. Each newly cre-
ated dislocation acts as an impediment to the movement of existing
dislocations, leading to an increase of the stress required to induce
further plastic deformation.

4.2.2 Bauschinger effect

The Bauschinger effect is a phenomenon that can be observed in
metals at load reversal. In a tension-compression text, the material
hardens in tension but, when subjected to subsequent compression, it
yields at a lower stress. This effect was first described by Bauschinger
[1881], hence its name. According to Mollica et al. [2001] the micro-
scopic reasons of the effect are far from clear. The most common
explanation advanced to justify this phenomenon is:

Explanation 1 The Bauschinger effect is a manifestation of the

anisotropic behaviour induced by permanent deformations.

To ensure compatibility between grains that deform differently, elas-

tic deformations appear at the grain boundaries. Some of these
elastic deformations remain locked when the material is unloaded,
leading to microscopic residual stresses. Depending on the direc-
tion of subsequent loading, these residual stresses may increase or
decrease the load required to induce plastic deformation. Lemaitre
and Chaboche [1990] explain that the Bauschinger effect is one of the
consequences of the presence of these microscopic residual stresses.
This does not mean that it represents the unique explanation of the
Bauschinger effect. The same point of view is also expressed by Bar-
lat [2007]. Paterson [1955] observed a strong Bauschinger effect upon
48 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING

load reversal on single crystal copper specimens. Single crystals hav-

ing per definition no grain boundary, Explanation 1 cannot be the
sole cause of this phenomenon.

An intra-granular explanation of the Bauschinger effect is also

found in the literature:
Explanation 2 Cell structures build up during pre-straining and di-
solve during load reversal, leading to a lower subsequent yield stress.
This theory has been advanced by Hasegawa et al. [1975] to ex-
plain the Bauschinger effect observed in polycrystalline aluminium.
Christodoulou et al. [1985] performed similar experiments and trans-
mission electron microscopy (TEM further in the text) observations
on copper and came to the same conclusion. This theory, which also
applies to the effect observed in single crystals, was commonly ac-
cepted until Rauch [1997] published experimental results from shear
tests conducted on a low carbon steel. Specimens were prestrained
in shear at different temperatures ranging from 125 K to room tem-
perature. The specimens prestrained at room temperature exhibited
a cellular dislocation structure whereas the dislocations in the ones
prestrained at 125 K were homogeneously distributed. Under reverse
shear, all the specimens exhibited the same mechanical behaviour.
This leads to the conclusion that the dislocation structure is not
decisive for the behaviour after stress reversal.

4.2.3 Cyclic softening

When cyclically deformed, a metal may harden, soften, or remain
stable depending on its microstructure and previous deformation his-
tory. These behaviours are schematically shown in Figure 4.5. Cyclic
softening under strain control corresponds to the trend observed in
Figure 4.2. Whether softening occurs under load or displacement
control, the microstructural phenomena involved are basically the
same.
4.2. MICROSTRUCTURAL PHENOMENA 49

Figure 4.5: Typical stress responses under strain or stress controlled

cyclic deformation: cyclic softening, cyclic hardening and stability.

Morrow [1965] showed, based on strain controlled cyclic exper-

iments on copper, that fully annealed materials tend to cyclically
harden while cold worked materials exhibit cyclic softening. Klesnil
and Lukas [1967] showed that the same effect is observed in low
carbon steels. Landgraf [1970] observed that materials with a hard-
ening exponent n > 0.1 tend to cyclically harden while those with
n < 0.1 cyclically soften. With n = 0.06, the SAE 1144 investi-
gated in the present study does not contradict this trend. Sankaran
et al. [2003] conducted strain controlled tension-compression tests on
a multiphase microalloyed medium carbon steel and observed that,
depending on the strain amplitude under which the experiments are
performed, the material may cyclically harden or soften. This last
behaviour is called mixed hardening and depends on the material and
its previous deformation history. The reader will note here that the
softening effect studied is independent of an eventual development
of micro-cracks or pores in the material. Also, the cyclic behaviours
studied here are restricted to those for which the mean strain, re-
spectively the mean stress, is zero. Cycling with a mean strain other
than zero may lead to the phenomenon of relaxation while cycling
under a non-symmetric stress leads to ratcheting or shakedown, as
described e.g. by Lemaitre and Chaboche [1990]. Ratcheting man-
ifests itself by the progressive increase in strain at each cycle and
50 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING

appears after the material has ceased to harden or to soften isotrop-

ically. As for the Bauschinger effect, the microstructural reasons for
cyclic softening are not completely clear—different explanations of
the phenomenon are reviewed in the following paragraphs.

Explanation 1 The dislocations are rearranged in a cellular struc-

ture which results in an increase in the mean free path for disloca-
tions.

According to the early works of Plumbridge and Ryder [1969] and

Klesnil and Lukas [1965], a cellular arrangement of dislocations is a
common phenomenon in metals that endured high strain amplitude
fatigue. The cells are characterized by a high density of dislocations
within the walls while the interior of the cells is clean. This effect can
be seen in the TEM pictures shown in Figure 4.6. Chai and Laird
[1987] explain that, once the dislocations begin to cluster in walls,
the dislocations left in the interior of the cells can move relatively
freely until they reach a wall and get captured there. They observed
the development of a cellular structure after only 3 cycles at ∆εp /2 =
10−4 on an AISI 1018 steel whilst the material was still in a transient
hardening domain. The cleaning of the interior of the cells results,
according to Wiese [2010], in an increase of the mean free path2 of the
remaining dislocations, thus enhancing their mobility. The softening
would then occur through the mutual annihilation of dislocations
(Explanation 2), which is subsequent to the formation of a cellular
structure.

Explanation 2 The annihilation rate of dislocations exceeds their

generation rate.

Feltner and Laird [1967] studied the cyclic behaviour of copper and
a copper alloy under plastic strain control with amplitudes 5 · 10−3 <
∆εp < 5 · 10−2. Using TEM, they observed the development of a cel-
lular structure of dislocations. They attributed the softening effect
2 According to Sankaran et al. [2006], the mean free path of a dislocation

refers to the average distance it can go through before encountering an obstacle.

4.2. MICROSTRUCTURAL PHENOMENA 51

Figure 4.6: Evolution of the dislocation structure in a 45# medium

carbon steel cyclically loaded under stress control with ∆σ/2 = 405
MPa. Number N of cycles: (a) N = 0; (b) N = 100; (c) N = 500;
(d) N = 1000; (e) N = 2000; (f) N = 3000 (from [Duyi and Zhenlin,
2001]).

observed to the mutual annihilation that takes place between disloca-

tions of opposite signs through their to and fro motion. This reduc-
tion in the dislocation density is the contrary of the strain hardening
effect explained in section 4.2.1. Leber et al. [2007] conducted fatigue
experiments at a total strain amplitude ∆ε/2 = 3 · 10−3 on different
stainless steels. The cyclic stress strain curve obtained exhibited an
initial transient hardening up to 10 ∼ 100 cycles, depending on the
material, followed by softening. They attributed the initial harden-
ing to the generation of the dislocation network (strain hardening)
52 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING

and the following softening to Explanation 2. This behaviour could

be compared to the SAE 1144 investigated in the present study in
the sense that SAE 1144 bars have a deformation history before the
cyclic deformation, i.e. a dislocation network is already present in
the material due to the drawing operation. In most of the studies
about cyclic softening, this effect is attributed to a combination of
different phenomena. For example, in the following studies, both
Explanations 2 and 1 can be found:

• Bhanu Sankara Rao et al. [1993] in the framework of their study

on type 304 stainless steel;

• Ganesh Sundara Raman and Padmanabhan [1996] while study-

ing AISI 304 LN;

• Chai and Laird [1987] while investigating AISI 1018.

In the study conducted by Chai and Laird [1987], softening begins

after a transient hardening that lasts approximately until the 10th
cycle.

Explanation 3 Proliferation of dislocation sources (Lüders’ band

formation).

Klesnil and Lukas [1967] conducted stress controlled tension-compression

tests on a low alloyed carbon steel with a stress amplitude a bit lower
than the yield stress of the material. The softening was attibuted
to the spreading of locally plastic zones, i.e. the same mechanism
that causes the appearance of the Lüders strains. The ferritic steel
studied by Roven and Nes [1991] exhibited softening when cyclically
deformed with a plastic strain amplitude smaller than approximately
∆εp /2 = 1.2 · 10−2 and hardened when cycled at larger strain ampli-
tudes. The disparities in the dislocation structure between the grains
observed in TEM examinations of samples that were cycled with a
low amplitude was attributed to this phenomenon.
4.2. MICROSTRUCTURAL PHENOMENA 53

Explanation 4 The initially pinned dislocations are released and,

through their movement in the material, a reduction in the resistance
to dislocation motion is induced.
Pinned dislocations act as an inhibitor of the movement of disloca-
tions. Their unpinning eases the displacements of all dislocations.
Feltner and Beardmore [1970] used the idea of microyield stress,
which is lower than the yield stress, to describe the value above
which, under cyclic loading, dislocations get unpinned and fatigue
starts to occur. Later, Duyi and Zhenlin [2001] used the same idea
to explain the sharp initial drop in σ0.2 of a medium carbon steel
cycled under stress control at ∆σ/2 equal to the initial value of σ0.2 .
Explanation 1 is used to justify the susbequent drop in softening,
that occurs in the second part of the fatigue life. Micrographs at
different stages of the fatigue life show a cell structure that appears
after about 1000 cycles, which is much more than the few cycles en-
countered by the material in cross roll straightening.

Explanation 5 In precipitation hardened alloys, the to and fro mo-

tion of the dislocations may lead to a shearing of the precipitate par-
ticles (dislocation cutting) so that the subsequent passage of disloca-
tions is facilitated.

In precipitation hardened alloys, movements of dislocations are im-

peded by the hardening precipitates. According to Barralis and
Maeder [1997], the repeated cutting of the precipitates by dislo-
cations eventually leads to their shearing. Once the particles are
sheared, the dislocations can move more freely and, hence, the mate-
rial softens. If the precipitation hardening is mastered, cyclic soften-
ing should not appear. As SAE 1144 is not precipitation hardened,
this explanation does not apply to the present phenomenon.

It appears from the above review that the effect of cyclic soft-
ening is not as well understood as, for example, strain hardening.
This is probably be due to the complexity of the observation of the
54 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING

phenomenon itself. Most likely, several different mechanisms lead

to cyclic softening. Also, in the cross roll straightening process, the
material endures neither uniaxial cyclic loading nor loading at a con-
stant amplitude.

4.2.4 Tension-compression asymmetry

Tension-Compression Asymmetry refers to the difference in yield
strength when a specimen is deformed under tension or compres-
sion. As pointed out by Barlat [2007], this effect is different from the
Bauschinger effect, for which the early plastification in compression
is subsequent to a deformation in tension. According to Kuwabara
et al. [2009], different mechanisms can lead to this behaviour. Met-
als that plastically deform through the formation of twins exhibit
this phenomenon due to the fact that twinning is pressure depen-
dent. The term Strength Differential (SD further in the text) is also
used to describe the behaviour of metals that yield earlier in ten-
sion than in compression (the opposite effect encountered with SAE
1144) without necessarily building twins. For example, Lowden and
Hutchinson [1975] studied a titanium alloy that exhibits this prop-
erty but which does not exhibit twinning. In this last case, the SD
effect was attributed to the pressure sensitivity of the texture hard-
ening phenomenon.

As Barlat [2007] points out, SD is referred to as the effect that

occurs when the tension and compression tests are conducted inde-
pendently from an annealed state. The material on which the present
study focuses has a strong previous deformation history, which may
induce an apparent SD. Due to the drawing operation that precedes
cross roll straightening, it seems that the asymmetry between the
yield point in tension and the one in compression is more likely to
result from the Bauschinger effect than from a real SD effect. From
the experimental data at hand, one can nevertheless not exclude that
the material also exhibits a real SD effect, although this seems un-
4.3. MACROSCOPIC CONSTITUTIVE RELATIONS 55

likely (or insignificant) having regard to its defomation mechanisms.

4.3 Macroscopic Constitutive Relations

As this study is being conducted, microstructure based material
models are still too CPU cost intensive to be used for engineering pur-
poses. Phenomenological material models, based on the macroscopic
behaviour of materials, are still state-of-the-art. The requirements
imposed on the model capabilities to describe the aforementioned
phenomena are discussed, candidate models are presented, and the
chosen Chaboche model, slightly modified, is described as well as its
implementation and the procedure used to fit its parameters.

4.3.1 General requirements

In order to describe the Bauschinger effect discussed above, the con-
stitutive equations chosen will have to take into account kinematic
hardening, i.e. a term has to allow the translation of the yield sur-
face. In the general case, the yield surface is described as

F = fY (σ − α) − Y (4.1)

where α is the backstress, i.e. a tensorial quantity that represents

the center of the yield surface. Assuming that the yield surface is
represented by the von Mises criterion, relation (4.1) becomes

F = J2 (σ − α) − Y (4.2)
where J2 is the second invariant of s − a, which consists of the
deviatoric parts of σ and α respectively. With an astute choice of
the evolution equation of a, the Bauschinger effect can be described.
With Y being a constant, relation (4.2) describes pure kinematic
hardening, i.e. the softening effect observed in the experiments can-
not be taken into account. To do so, the size of the yield surface
56 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING

must also be a function of the deformation. A common evolution

variable used to describe the evolution of the isotropic component
of hardening is the accumulated plastic strain p3 . Writing relation
(4.2) as
r
3
F = (s − a) : (s − a) − Y (p) (4.3)
2
provides a means to take into account both the kinematic and the
isotropic components of a deformation, hence the name of mixed
isotropic-kinematic given to this type of hardening models.

If not otherwise stated, it is assumed in this study that the nor-

mality hypothesis of plasticity holds, i.e.

∂F
dεp = dλ (4.4)
∂σ
dλ is the plastic multiplier. dλ is obtained through the consistency
condition, i.e. the requirement for the load point to remain on the
yield surface during plastic deformation.

4.3.2 Candidate models

Prager [1956] proposed what may be the simplest evolution law for
the backstress. The backstress is assumed to be a linear function of
the plastic strain increment, hence its name linear kinematic hard-
ening model.
da = Cdεp (4.5)
Armstrong and Frederick [1966] introduced a recall term to the
backstress evolution law, making it nonlinear

2
da = Cdεp − γadp (4.6)
3
3 The accumulated plastic strain is associated principally with the density of

dislocations in the current state (see e.g. Lemaitre and Chaboche [1990])
4.3. MACROSCOPIC CONSTITUTIVE RELATIONS 57

the accumulated plastic strain rate is computed as

r
p 2 p
dp = |dε | = dε : dεp (4.7)
3
A simple (in retrospect) but major improvement to the Armstrong-
Frederick model has been proposed by Chaboche [1986]4. The non-
linear backstress evolution law is decomposed in a sum of backstresses
M
X
da = da(i) (4.8)
i=1

in which each backstress a(i) evolves according to relation (4.6) with

Ci and γi . The principal advantage of this decomposition is that it
allows to model separately the different parts of a hysteresis curve.
For example, with M = 3, the first backstress component a(1) can
follow the steep initial slope with C1 large, while a large γ1 quickly
stabilizes it. a(2) , with medium values C2 and γ2 , depicts the tran-
sient domain. a(1) may be assumed to be linear (with γ3 = 0) and
models the stress response to higher levels of strain.

Following the work of Chaboche, numerous authors proposed en-

hancements to this kind of models (Chaboche [2008] reviews and
discusses the most popular ones). Particularly, predictions of the
ratcheting phenomenon, mostly under biaxial loading, are sensitive
to parameters that have minor influences when considering only a
few uniaxial cycles, for example a small value γ3 .

The major drawback of the Prager evolution law (see relation

(4.5)) is its linearity, which only allows to model bilinear stress-strain
hystereses. The introduction by Armstrong and Frederick [1966] of a
recall term paved the way for the development of the type of models
4 According to R.P. Skelton in [Frederick, 2007], J.L. Chaboche developed his

model without prior knowledge of the work of Armstrong and Frederick [1966],
which was an internal report.
58 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING

presented above. The principal alternative to circumvent this lin-

earity is the so-called multilayer models. Mroz [1967] replaced the
constant C in relation (4.5) by the concept of field of work-hardening
moduli. The single yield surface F = 0 is replaced by a series of yield
surfaces f0 , f1 , . . . , fm . f0 is the initial yield surface, within which
the deformation is purely elastic, and f1 , . . . , fm are separate (larger)
regions5. In f1 , . . . , fm the hardening moduli Ci are constant. The
surfaces cannot intersect but can act and move together. During a
plastic loading between fl and fl+1 , the active surface is fl = 0 and
the surfaces within it are tangent at the loading point (see Figure
4.7) This is expressed as

f0 = f1 = · · · = fl = 0 (4.9)
∂f0 ∂f1 ∂fl
dλ0 = dλ1 = · · · = dλl (4.10)
∂σ ∂σ ∂σ
The normals of the yield surfaces are coincident when two surfaces
are in contact. In the formulation of Mroz, the plastic modulus is
not directly coupled with the kinematic hardening rule through the
consistency condition, hence the term uncoupled used by Bari and
Hassan [2000] to qualify this type of models. Considering von Mises
plasticity, each surface is written as

fi = J2 (σ − αi ) − Yi (4.11)

Yi , which can also be a function of p, represents the current radius

of the ith surface.

In addition to the two main approaches described above, some

models rely on both the concepts of non-linear backstress evolution
equations and nested surfaces (not necessarily yield surfaces). For
example, Ohno and Wang [1993] proposed a model in which the
5 As a metaphor, the yield surfaces represent intricate layers as in a Russian

nesting doll.
4.3. MACROSCOPIC CONSTITUTIVE RELATIONS 59

Figure 4.7: Schematic representation of the model of Mroz [1967].

The stress-strain curve is approximated by linear segments leading
to different hypersurfaces in the stress space.

backstresses saturate fully upon reaching critical values. Relation

(4.6) is modified as
    
2 a(i)
da(i) = γi ri dεp − H f˜i εp : a(i) (4.12)
3 ri

where ri = Ci /γi , H (·) is the Heaviside’s step function, h·i are the
Macauley’s brackets and f˜i is a hypersphere of radius ri in the space
of a(i) defined as
3
f˜i = a(i) : a(i) − ri2 = 0 (4.13)
2
60 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING

The backstress a(i) saturates fully when reaching f˜i = 0, i.e. the
backstress does not develop beyond f˜i = 0. Other versions of this
model have been developed allowing, for example, the translation of
the surfaces f˜i = 0.

4.3.3 Slightly modified Chaboche model

The main focus of the studies presented in section 4.3.2 is the descrip-
tion of phenomena like ratcheting, which take place at a relatively
high number of cycles. In most of the studies, the number of cycles
considered vary between a few tens and a few hundreds. As shown
in chapter 2, the cross roll straightening process induces only a few
plastic cycles in the material. The goal is therefore not to reasonably
predict the material response over a large number of cycles but to
be as precise as possible on a few cycles. To this aim, the super-
position of Armstrong-Frederick rules according to Chaboche [1986]
represents a pragmatic option.

Assuming small strains, the total strain tensor ε is additively

decomposed in an elastic strain εe and a plastic strain εp

ε = εe + εp (4.14)

The elastic part of the strain obeys Hooke’s law

σ = Ce : ε e (4.15)

where Ce is the isotropic elastic stiffness tensor. The plastic strain

components are governed by the associated flow rule (equation (4.4)).
The yield function is expressed as
3
F = (s − a) : (s − a) − Y 2 (4.16)
2
and Y , the isotropic hardening (or softening) part, is

Y = Q (1 − exp (−bp)) + σ0 (4.17)

4.3. MACROSCOPIC CONSTITUTIVE RELATIONS 61

a (i.e. its evolution equation) is decomposed as

M
X
a= a(i) + a0 (4.18)
i=1

where a0 allows to take into account the initial asymmetry between

tension and compression. a(i) is defined through the (Armstrong-
Frederick) differential equation
2
da(i) = Ci dεp − γi a(i) dp (4.19)
3
The term slightly modified used above refers to the component a0
in relation (4.18), making it a little bit different from the Chaboche
decomposition (relation (4.8)). The tensor a0 defines the initial po-
sition of the center of the yield surface, which is non-evanescent, i.e.
the term remains even after a load history. The purpose of this term
is to take into account the apparent tension-compression asymmetry
discussed earlier. As the apparent SD results from a previous defor-
mation, another option would be to initialize the different a(i) . A
tension-compression test provides only information about the value
of a, not about the respective influences of the different terms.

Other possibilities exist in order to describe the strength differen-

tial effect. For example, Cazacu and Barlat [2004] proposed a yield
criterion to model the plastic behaviour of titanium, which exhibits
the same effect. Although this type of models is very promising for
sheet forming applications, it would lead to unreasonably complex
systems of equations if the kinematic hardening effects were also to
be taken into account for three dimensional deformations. The addi-
tion of the initial backstress a0 is therefore a frugal means to account
for this asymmetry.
62 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING

4.3.4 FE implementation
Kobayashi and Ohno [2002] proposed an efficient algorithm to im-
plement constitutive models of cyclic plasticity based on the return
mapping algorithm. The main relations, adapted to the set of con-
stitutive equations presented in section 4.3.3, are presented here.

Using the backward Euler method, relations (4.14) to (4.19) are

discretized for an interval n to n + 1 as

εn+1 = εen+1 + εpn+1 (4.20)

εpn+1 = εpn + ∆εpn+1 (4.21)


σn+1 = C : εn+1 − εpn+1
e
(4.22)
r
3
∆εpn+1 = ∆pn+1 nn+1 (4.23)
2
r
3 sn+1 − an+1
nn+1 = (4.24)
2 Yn+1
3 2
Fn+1 = (sn+1 − an+1 ) : (sn+1 − an+1 ) − Yn+1 (4.25)
2
M
X (i)
an+1 = an+1 + a0 (4.26)
i=1
(i) 2 p (i)
an+1 = a(i)
n + Ci ∆εn+1 − γi an+1 ∆pn+1 (4.27)
3

with Ce = 2GI + λII (I is the fourth-order identity tensor and

G the modulus of rigidity). To solve this system of equation, a trial
stress is first defined
trial
σn+1 = Ce : (εn+1 − εpn ) (4.28)

The trial yield condition is checked using the deviatoric trial stress
strial
n+1
trial 3 trial

2
Fn+1 = sn+1 − an : strial
n+1 − an − Yn (4.29)
2
4.3. MACROSCOPIC CONSTITUTIVE RELATIONS 63

trial trial
If Fn+1 < 0, the increment is elastic and σn+1 = σn+1 . If this
is not the case, the accumulated plastic strain increment ∆pn+1 is
sought so that
trial
σn+1 = σn+1 − Ce : ∆εpn+1 (4.30)
satisfies the condition Fn+1 = 0, hence the name return mapping
given to this kind of algorithms. Noting that Ce : ∆εpn+1 = 2G∆εpn+1 ,
relation (4.30) becomes
M
X
p (i)
sn+1 − an+1 = strial
n+1 − 2Gεn+1 − a0 − an+1 (4.31)
i=1

(i)
By defining a θn+1 as

(i) 1
θn+1 = (4.32)
1+ γ (i) ∆p n+1

(i)
an+1 can be written, from relation (4.26), as
 
(i) (i) 3 p
an+1 = θn+1 a(i) n + Ci ∆ε n+1 (4.33)
2
An auxiliary variable A is defined as
M
X (i)
A = strial
n+1 − a0 − θn+1 a(i)
n (4.34)
i=1

sn+1 − an+1 can then be written as

Yn+1 A
sn+1 − an+1 =  PM (i)  (4.35)
Yn+1 + 3G + i=1 θn+1 Ci ∆pn+1

which, substituted in the yield function Fn+1 = 0, yields

q
3
2 A : A − Yn+1
∆pn+1 = PM (i) (4.36)
3G + i=1 θn+1 Ci
64 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING

Relation (4.36) is solved iteratively until

∆pn+1 (k − 1)
1− ≤ tol (4.37)
∆pn+1 (k)
where tol is a prescribed tolerance and k denotes the iteration num-
ber. In the present study, it appears that tol = 10−4 yields sat-
isfactory results. Once the tolerance is reached, εpn+1 is computed
p
according to relation (4.21), σn+1 according to (4.30), and the back-
stress components according to (4.33). The model is implemented in
the commercial FE code LS-Dyna c .

4.3.5 Parameter fitting

A discretionary grouping of the parameters is made: the kinematic
parameters and the isotropic ones. The kinematic parameters are Ci
and γi . The isotropic parameters are Q, b, σ0 and α0 . E, the elas-
tic modulus, is fitted separately. Following the convention adopted
according to which the main axis of the bar in the undeformed con-
figuration corresponds to the axis Z in the global coordinate system,
a0 , the deviatoric part of the initial backstress tensor, is written
 1 
− 3 α0 0 0
a0 =  0 − 31 α0 0  (4.38)
2
0 0 3 α 0

when tension-compression tests are conducted on specimens cut in

the axial direction of the bar, as pictured in Figure 4.1. The idea
behind the splitting of the parameters into two groups is to use the
kinematic parameters to fit the model to the shape of the hysteresis
curve whereas the isotropic parameters are used to fit the softening
over the cycles. A brute force approach, in which the parameters are
fitted all at once using, for example, a genetic algorithm, does not
allow to make this clear distinction between the roles of the different
parameters. Poor results are obtained using such approaches.
4.3. MACROSCOPIC CONSTITUTIVE RELATIONS 65

Bari and Hassan [2000] present a procedure to fit the values of

the kinematic parameters of the Chaboche model on a stabilized
stress-strain cycle, i.e. after the material has ceased to harden/soften
isotropically. The fitting is conducted on a stabilized hysteresis curve
of reasonable strain limit6 . Neglecting isotropic hardening, the closed
form of the Chaboche model with M = 3 can be written
3
X
α(i) + σ0 = σ (4.39)
i=1

with
Ci
α(i) = (1 − 2 exp (−γi (εp + ∆εp /2))) (4.40)
γi
for i = 1 and 2 (∆εp is the plastic strain amplitude in a single
branch). C3 is then determined so that

C1 C2 C3 p
+ + σ0 = σ − (ε + ∆εp ) (4.41)
γ1 γ2 2

holds at or close to ∆εp (the end of each loading branch). The pa-
rameter γ3 is then determined separately from a uniaxial ratcheting
experiment. As the present study only accounts for the first few
deformation cycles, γ3 = 0 is taken. It is conventional to order the
parameters as C1 > · · · > CM and γ1 > · · · > γM , large values of Ci
being associated with large values of γi .

For the purpose of the present study, the isotropic components

play a decisive role.

For the uniaxial case, the stress σ as a function of plastic strain

is
M
X
α(i) + α0 + Q (1 − exp (−bp)) + σ0 = σ (4.42)
i=1

6 Bari and Hassan [2000] use a hysteresis curve in which ∆εp /2 = 8.5 · 10−3
66 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING

where the backstress can be written as the closed form of relation

(4.19) as
p
α(i) = υC p
γi (1 − exp (−υγi (ε − ε0 )))
i

(i) (4.43)
+ α0 exp (−υγi (εp − εp0 ))

(i)
α0 and εp0 being the branch initial values of, respectively, backstress
and plastic strain, υ = +1 for tension branches and υ = −1 for
(i)
compression branches. For a stabilized hysteresis, α0 = −υCi /γi
and relation (4.43) becomes

Ci
α(i) = υ (1 − 2 exp (−υγi (εp − εp0 ))) (4.44)
γi

A branch of the experimental stress-strain curve is chosen and a

root mean square error function is defined with the predicted val-
ues. By keeping the isotropic parameters constant and varying just
the kinematic ones, the error function is minimized using a Newton-
Raphson procedure.

The isotropic parameters of the model are used to describe the ob-
served cyclic softening effect. At this stage, the kinematic parameters
obtained are assumed to be constant and the isotropic parameters
are varied to produce the best fit. To this end, a 1D implementation
of the model is made and its response to the experimental strains
over time is computed. The model is fitted on the largest absolute
stress values of each branch (|σjmin | for compression branches and
σjmax for tension branches) by minimizing an error function similar
to the one described above. The asymmetry between the values in
tension vs. the values in compression is taken into account by the
parameter α0 . The values obtained for the isotropic parameters are
then used as an input for a second fitting of the kinematic ones.
The iteration is repeated until convergence. In the present case, no
significant change in the parameters is observed after three iterations.
4.3. MACROSCOPIC CONSTITUTIVE RELATIONS 67

Table 4.1: Fitted parameter values for the Chaboche model (elastic
modulus E = 210 GPa, Poisson’s ratio ν = 0.3).

parameters for
kin. hard. iso. hard.
C1 = 44400 Q = -214
C2 = 18100 b = 5.5
C3 = 1430 σ0 = 757
γ1 = 665 α0 = 11
γ2 = 87
γ3 = 0

The parameters obtained for the Chaboche model are presented

in Table 4.1 and the corresponding curves are plotted in Figures 4.8
and 4.9.

For certain computations in this study, a linear isotropic model is

also used. Its elastic behaviour is described by the same parameters
as for the Chaboche model and its plastic behaviour is defined by
a yield stress σy = 935 [MPa] and a tangent modulus Etan = 3000
[MPa].

As it can be seen from Figure 4.8, the Chaboche model is not

able to describe simultaneously the sharp initial transition from the
elastic to the plastic deformation as well as the subsequent curved
transitions with a suitable level of accuracy. This is a common weak-
ness of kinematic hardening material models. Jiao and Kyriakides
[2009] attempted to model the cyclic behaviour of a (stainless) SAF
2507 steel using a modified [Dafalias and Popov, 1975] type two-
surface non-linear kinematic hardening model and encountered sim-
ilar difficulties. They circumvented the problem by using two sets of
parameters, one for the first deformation and another for the second
68 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING

1000
Experiment
800 Model
600
400
200
σ [MPa]

0
−200
−400
−600
−800
−1000
−0.02 −0.015 −0.01 −0.005 0 0.005 0.01 0.015 0.02
ε [−]

Figure 4.8: Fitted stress-strain curve on the experimental values

presented in Figure 4.2.
1000
Experiment
Model
950
|σ peak| [MPa]

900

850

800

750
0 5 10 15 20 25
j (branch number)

Figure 4.9: Fitted model response on the peak stress values presented
in Figure 4.3.
4.3. MACROSCOPIC CONSTITUTIVE RELATIONS 69

one. The problem under consideration was a one-dimensional defor-

mation and the sets of material model parameters were switched at
the first load reversal. The sets of parameters were chosen in such
a way that, at the point of transition, the predicted stresses were
continuous over time. This approach suits well to one-dimensional
problems, where the difference between the first and the subsequent
deformations is obvious. For three-dimensional problems, the def-
inition of unloading is not clear, making the approach of Jiao and
Kyriakides [2009] inadequate. An alternative approach, specific to
the present problem, is presented in chapter 6.

Summary of Chapter 4
From the tension-compression tests conducted, it appears that the
material exhibits: (1) strain hardening in monotonic tension, (2) a
strong Bauschinger effect, (3) cyclic softening and (4) an apparent
tension-compression asymmetry. From the microstructural explana-
tions presented, it appears that the effects observed are normal for
a cold-worked SAE 1144, except for the tension-compression asym-
metry. This asymmetry is likely to be due to a Bauschinger effect
resulting from the previous drawing operation.

In order to model the different phenomena observed, a slightly

modified Chaboche model is chosen. The Bauschinger effect is de-
scribed through the kinematic terms of the model, cyclic soften-
ing through an isotropic term and the apparent tension-compression
asymmetry through the incorporation of a constant tensor that de-
fines the initial position of the yield surface.

The model is implemented in the commercial FE code LS-Dyna c .

The parameters’ fitting procedure is presented and it appears that,
apart from the first transition from elastic to plastic, the model can
be fitted correctly to the experiments.
70 CHAPTER 4. MATERIAL BEHAVIOUR AND MODELLING
Chapter 5

Process Modelling

The existing modelling strategies presented in chapter 2 are not able

to predict with a satisfying level of accuracy the experimental results
presented in chapter 3. In order to reach the objectives presented in
section 1.4, three modelling approaches, two of them being developed
on purpose, are investigated in the context of this study. The first
approach, described in section 5.1, is a classical three dimensional FE
model of the process. The failure of this first approach to provide
satisfying results with a reasonable CPU cost leads to the develop-
ment of two new models. The slice model, described in section 5.2 is
a modified submodelling approach. Although an improvement over
the first approach, its main drawback is its failure to incorporate
lateral stamping. The third method, named total strain field model
and presented in section 5.3, takes advantage of some particularities
of the process studied. As shown in chapter 6, this last approach
is able to predict the most important phenomena that occur during
cross roll straightening.

71
72 CHAPTER 5. PROCESS MODELLING

5.1 Full Model Approach

Intermediate results obtained with this model are presented in [Mutrux
et al., 2008]. This approach relies on classical FE modelling tech-
niques.

5.1.1 Background
Dynamic equilibrium
During the straightening process, strong formulation of the dynamic
equilibrium holds for each material point of the bar
divσ + fb − ρü − cρu̇ = 0 (5.1)
where fb are the body forces, ρ is the mass density and u is the mate-
rial displacement vector. Relation (5.1) is subjected to the boundary
conditions
u = ũ on Γu
(5.2)
σn = t on Γt
where Γ is the boundary of the analysis domain, ũ are prescribed
displacements on the region Γu and t are prescribed tractions on the
region Γt . A weak formulation of equation (5.1) can be obtained
using, for example, Hamilton’s principle. The analysis domain is
discretized in a number of elements delimited by nodes. Applied to
those elements, the matrix formulation of the weak form is
M Ü + |{z}
|{z} |{z} −Fext = 0
CU̇ + KU (5.3)
Finert Fdamp Fint

where U is the matrix of nodal displacements, M is the mass matrix,

C is the damping matrix and K is the stiffness matrix. Considering
that a simulation of cross roll straightening is such that at least one
section of the bar passes completely through the rolls and that the
material is considered to be elastic-plastic, the problem is obviously
non-linear. This non-linearity implies a time discretization of the
problem in a series of time steps ∆t (not necessarily constant).
5.1. FULL MODEL APPROACH 73

Contact formulation
A central problem for the simulation of cross roll straightening lies
in the description of the contact between the bar and the rotating
rolls.

The simple case of a node n of the rod penetrating a segment s

of a roll is considered. In the standard penalty contact formulation,
an interface force fn is applied to n so that

fn = lkns (5.4)

where ns is the normal to the segment considered, l is the penetra-

tion depth of the node and k is a stiffness factor. Reaction forces
(i)
fs are applied to the nodes of the segment. For complex contact
interactions, this algorithm requires small ∆t to yield satisfactory
results. The only boundary conditions applied on the bar during the
simulation of cross roll straightening are through its contact with the
different parts: Γu = ∅ in relation (5.2).

The contact between the bar and the rotating rolls during cross
roll straightening belongs to the category of rolling contacts. The
correct description of this type of contacts is the subject of ongoing
research (see e.g. [Wriggers and Laursen, 2007]). For challenging
problems, such as the modelling of rolling tyres, the trend is now
towards Arbitrary Lagrangian Eulerian approaches which are beyond
the scope of the present study.

Solution schemes
Equation (5.3) is to be solved for U at time t + ∆t (t+∆t U further
in the text). To do so, two main types of approaches exist.

In explicit methods, t+∆t U is evaluated directly. Using the Taylor

series expansions of t+∆t U and t−∆t U, the central difference approx-
74 CHAPTER 5. PROCESS MODELLING

imations of t U̇ and t Ü are written

t+∆t
t U − t−∆t U
U̇ = (5.5)
2∆t
t+∆t
t U − 2t U + t−∆t U
Ü = 2 (5.6)
(∆t)


A relation t+∆t U t U, t−∆t U is then obtained by inserting (5.5)
and (5.6) in (5.3). The conditional stability of the method imposes
∆t < ∆tcrit .

The implicit time integration methods represent the second cat-

egory of solution schemes. The principal difference with the explicit
methods lies in the iterative search of t+∆t U. The displacement
vector t+∆t U is expressed as t+∆t U = t U + ∆U. Relation (5.3) is
rewritten as

Finert + Fdamp + Fint − Fext = r (∆U) (5.7)

and the term ∆U is iteratively sought until the residual r (∆U) van-
ishes. The unconditional stability of the implicit methods for the so-
lution of equation (5.7) imply that the choice of ∆t is dictated only
by accuracy considerations. Implicit integration methods are pre-
ferred for static computations (inertia forces (Finert ) and damping
forces (Fdamp ) are neglected) as their iterative nature tends to make
them more accurate than the explicit methods (see e.g. [van den
Boogaard, 2002]).

A thorough discussion on the pros and cons of the different types

of integration schemes is given by van den Boogaard [2002]. In the
present case, the use of the simple contact formulation presented
above imposes the choice of a small ∆t. Small ∆t being one of the
main attributes of the explicit methods, this type of solution scheme
is preferred for the full model simulation of cross roll straightening.
5.1. FULL MODEL APPROACH 75

In this study, implicit time integration methods, both dynamic

and static, are used for different computations which are explained
further in the text.

5.1.2 Problem specific procedure

The FE model used is shown in Figure 5.1. The rolls are meshed
with shell elements and are assumed to be rigid1 . The bar is divided
into two parts: a section is meshed with fully integrated solid ele-
ments and the rest of it is discretized with beam elements. The part
of the bar that is meshed with solid elements is the one that gets
in contact with the rolls. The length of the segment meshed with
solid elements is chosen in such a way that approximately 150 mm
of the bar pass completely through the rolls during the simulation.
The purpose of meshing the rest of the bar with beam elements is to
avoid the appearance of dynamic effects at the rear extremity of the
bar: the bar is stabilized by the contact between the beams and the
box that can be seen on the right side of Figure 5.1.

The simulation is divided into three steps: (1) gravity, (2) closing,
and (3) rolling. During the gravity step, the rolls are separated and
the bar positions itself on the convex roll and in the box. During the
closing step, the rolls are brought together, bending the bar. The
rolling step is the core of the simulation: the rolls rotate, driving
the bar forward and bending it. In order to reduce the computa-
tion time, a mixed integration strategy is adopted: the gravity step
is computed using a dynamic implicit integration scheme and, for
the subsequent steps, the integration type is switched to explicit. A
linear elastic material model describes the behaviour of the segment
of the bar that is meshed with beams. The behaviour of the solid
elements is described using the linear isotropic elastic-plastic model
presented at the end of section 4.3.5.

1 Strictly speaking, a rigid flat element is not really a shell in the sense that

no integration takes place.

76 CHAPTER 5. PROCESS MODELLING

Figure 5.1: The FE model used for the full model approach. A
section of the mesh used to discretize the bar is shown on the top
right.

The main drawback of this FE approach lies in its computational

cost. The mesh pictured in Figure 5.1 is far too rough to depict the
as
subtle differences in, for example, the value of σ0.2 . Even with this
rough mesh, the simulation requires a wall clock time of about 30
hours to run on 8 Intel Xeon 2.5 GHz CPUs. This large compu-
tation time is due mainly to the large number of elements required
to discretize the bar, the lenghts of the trajectories of the elements
during the process and the complexity of the contact conditions (ap-
proximatively 35% of the CPU time). Reducing the CPU cost and
reaching a level of accuracy suitable to model the experiments mo-
tivate the development of the approaches presented in sections 5.2
and 5.3. The full model proves nevertheless useful to understand the
kinematics of the process and to verify some of the assumptions on
which the other methods are based.
5.2. SLICE MODEL APPROACH 77

5.2 Slice Model Approach

Results obtained with this model are presented in [Mutrux et al.,
2009]. The main assumption on which this approach is based is
that the bending line of the bar during cross roll straighening can
be obtained from a much simpler simulation than the full model. A
submodelling technique, similar to the one used by Schleinzer and
Fischer [2001] for the simulation of the staightening of railway rails,
is then applied to a thin layer, a slice, of the bar. This method ap-
parently shares similarities with the one presented by Kinkori [1998].

5.2.1 Background
This approach is primarily developed to model the straightening pro-
cess without stamping. A tentative method to incorporate lateral
stamping is discussed in section 5.2.3. The procedure presented here
neglects the influence of lateral stamping.

The main assumptions on which the slice model relies are listed
and commented below.
Assumption 5.1 The Euler-Bernoulli assumption holds (see sec-
tion 1.2.2).
Assumption 5.1 is a basic assumption of beam bending theory. Yu
and Zhang [1996] provide a thorough explanation of the different
theories of bending.
Assumption 5.2 The pitch can be expressed by the analytical rela-
tion κ = 2πrrod tan α.
The analytical approaches presented in section 2.1 are all based
on the expression of the pitch according to relation (1.5) (κ =
2πrrod tan α). This relation should hold if no slip occurs between
the rolls and the bar. The slice model is also based on this expres-
sion of the pitch.
78 CHAPTER 5. PROCESS MODELLING

The validity of assumption 5.2 can be tested by observing straight-

ened bars. Imperfections on the surface of the rolls leave thin helical
lines on the surface of the straightened rods. The helixes left on bars
straightened under different configurations are measured. The he-
lixes on the rods straightened with α = 18◦ and δ = {0.35, 0.50, 0.65}
mm have a pitch of approximately 25.8 mm (no influence from the
stamping is observed) and equation (1.5) yields κ = 25.5 mm. For
α = 17◦ and δ = 0.50 mm, the measured pitch is 23.5 mm and the
theoretical one is κ = 24.0 mm. The lines on the rod staightened
with α = 19◦ and δ = 0.50 mm can hardly be seen: the pitch is
approximately 26 mm, whereas the theoretical one is κ = 27.04 mm.
These comparisons prove that, for the bars investigated, expression
(1.5) provides a decent approximation of the pitch.

A full model simulation (with α = 17◦ and δ = 0) where the

Coulomb friction coefficient between the rod and the rolls is set to
0.1 yields a pitch of 22.5 mm (equation (1.5) yields κ = 24.0 mm).
This value of the pitch is the same when the friction coefficient is
set to 0.2. The values obtained with equation (1.5) are closer to the
measured ones. Apparently, the pitch obtained from the full model
depends on such factors as the contact formulation between the bar
and the rolls.

Assumption 5.3 The curvature of the bar in a closing simulation

is similar to the one in the steady state of straightening.
Forming processes are generally transient processes, i.e. internal vari-
ables depend on time. The cross roll straightening process, when
considered as a whole, belongs to this category. Nevertheless, the
straightening of the central part of a bar, where the influence of
the extremities of the bar is negligible, can be considered as being a
steady state. Simulations of sheet rolling are generally carried out on
the assumption of steady state (see e.g. [Yuan et al., 2009]). In order
to verify assumption 5.3, the bending line from a full model simula-
tion is compared to the one obtained from a closing simulation, i.e.
5.2. SLICE MODEL APPROACH 79

Figure 5.2: Bending lines from a closing simulation and from a full
model simulation, both conducted with α = 17◦ and δ = 0. The
corresponding curvatures κtot of the curves are also plotted.The Z
axis is defined in Figure 1.6.

where the bar is bent between two static rolls, without rotation or
forward feed. The closing simulation is conducted using a dynamic
implicit integration scheme. The results are plotted in Figure 5.2.

As it can be seen from Figure 5.2, the curvatures of the bar ob-
tained from the two types of simulations are relatively similar. A
small discrepancy between the two curves appears in the middle of
the bar. The principal explanation for this discrepancy lies in the
difficulty of properly defining the stamping δ in the full model sim-
ulation. The effective δ, the amount by which the bar is stamped,
depends on the contact formulation chosen and is significantly dif-
ferent from the value of δ that is set on the cross roll straightening
machine. Also, the value of δ set on the machine does not corre-
spond effectively to δ = 2rrod − d: for example, the value set does
80 CHAPTER 5. PROCESS MODELLING

not account for the elastic deformation of the machine or the ther-
mal expansion of the rolls. The method used to circumvent this
problem for the total strain field model is explained at the beginning
of chapter 6.

Assumption 5.4 The curvature of the bar during straightening is

marginally influenced by the behaviour of the material.

This assumption specifies assumption 5.3 in regard to the material

behaviour. In a closing simulation, the material is assumed to behave
homogeneously over the length of the bar. As shown in chapter 1, the
mechanical properties change due to the straightening operation, i.e.
the material properties are not constantly distributed over the length
of the bar during straightening. An indication that this assumption
holds for the present case is the fact that the curvature of the bar
obtained from the full model (shown in Figure 5.2) is approximately
symmetric about the origin of the abscissa. A generalization of this
assumption is dealt with in section 5.3.1.

Assumption 5.5 The curvature of the bar during straightening lies

exclusively in the YZ plane.

During straightening, the bar is guided by teflon plates as it passes

through the rolls. Both in the closing simulation and in the full
model simulation, rigid shell elements guide the bar between the
rolls, forcing its main axis to remain in the Y Z plane. In the case
where no or little stamping is applied, the reaction forces on the
lateral plates are negligible—the bar remains naturally in the YZ
plane. When stamping is applied in the full model simulation, the
bar tends to bend in the XY plane and reaction forces appear on
the guides. Assumption 5.5 is not at the core of the slice model and
lifting it would not lead to a major increase in the complexity of the
method.
5.2. SLICE MODEL APPROACH 81

5.2.2 Problem specific procedure

A closing simulation is first conducted, during which the bar is
bent between two static rolls. The computation is made using a
dynamic implicit integration scheme with the commercial FE code
LS-Dyna c . The coordinates of the nodes lying on the neutral axis
of the bar are extracted and a spline is fitted through them. This
bending line is a parameterised curve b (t) with 0 ≤ t ≤ 1. Assuming
5.5 holds, the bending line lies in the ZY plane.

A moving coordinate system {ξ (t) , η (t) , ζ (t)} is associated with

b (t)
 
1
eξ (t) =  0 
0
 
0
1  b′z (t) 
eη (t) = (5.8)
kb′ (t) k −b′y (t)
 
0
1  b′y (t) 
eζ (t) =
kb′ (t) k
b′z (t)

where b′ (t) is the directional derivative of b (t).

If assumption 5.2 holds, the pitch κ is computed according to

equation (1.5). The local curvature of the bar in the deformed con-
figuration is computed as

∂ 2 b (t)
κtot (t) = (5.9)
∂t2
The angle ω (t), defining the rotation of the bar around its main axis,
is given by
Lbar
ω (t) = 2π t (5.10)
κ
82 CHAPTER 5. PROCESS MODELLING

Figure 5.3: Slice model: (a) computation of the normals to the faces
of a slice h thick and (b) FE mesh of solid elements used to discretize
the layer.

where Lbar is the length of the bar in the undeformed configuration.

Considering a thin layer of the bar h thick (see Figure 5.3 (a)), the
normal n (t) to one of its faces in the ξηζ frame is written
 
hκ sin (ω)
1  hκ cos (ω) 
n (t) = p (5.11)
κ2tot h2 + 4 2
A slice of the bar is meshed using solid elements, as pictured in
Figure 5.3 (b). The nodes have translational degrees of freedom in ξ
and η. The ∆ζ displacements of the nodes are imposed as boundary
conditions to the model. The displacement of the nodes are given by
−υ (n1 (t) ξ (t) + n2 (t) η (t))
∆ζ (t) = (5.12)
n3 (t)
where υ = 1 for the nodes lying on the upper (ζ > 0) side of the slice
and υ = −1 for the nodes lying on the lower side.

Technically, there are different ways to apply the displacements

∆ζ (t) to the nodes of the slice in the FE calculation. The first op-
tion is to define boundary prescribed motions in ζ for each node.
5.2. SLICE MODEL APPROACH 83

The method presents the disadvantage of requiring a relatively large

number of displacement curves over time (one per node). Another
option is to define two layers of rigid shell plates, one on each side
of the slice, in sliding contact with the nodes of the mesh and to
define the orientation of the plates over time. This second approach
implies to check carefully the contact formulation employed but it
considerably reduces the amount of input data required.

In order to make predictions about the evolution of the curvature

of the bar due to the straightening operation, the initial curvature
κbs is reduced to an axial stress distribution. After the deformation,
the resulting axial stress distribution is checked once again, the max-
imum torque is sought, and the corresponding output curvature is
computed. This procedure is explained in a more detailed manner
on page 98.

5.2.3 Tentative incorporation of lateral stamping

as
It is shown in chapter 1 that the value of σ0.2 is influenced by the
amount of stamping applied during straightening. In order to take
into account the effect of stamping in the simulation, two rolls are
added to the model presented above.

The two rolls, meshed with rigid shells, are revolving around the
slice pictured in Figure 5.3 and around their own axes. The mini-
mum distance d (t) between the rolls is defined as the thickness of
the bar in the Y Z plane from the closing simulation.

This method presents two major drawbacks:

• due to the complex boundary conditions imposed to the nodes,
the contact between the rotating rolls and the slice is unstable
and, therefore, only very low levels of stamping can be applied;
• as mentioned above, the effective level of stamping is difficult to
84 CHAPTER 5. PROCESS MODELLING

define (the distance d between the rolls during straightening is

unknown). With the slice model, a reference point to calibrate
the stamping in the simulation δsim to the stamping on the
machine δmach could not be found.

The model presented in section 5.3 allows to bypass these prob-

lems and the focus is hence laid on the next approach. Interested
readers may find results obtained with the slice model in [Mutrux
et al., 2009].

5.3 Total Strain Field Approach

This approach, presented in [Mutrux et al., 2010] and [Mutrux et al.,
2011], represents an evolution of the slice model and shares similari-
ties with the approach proposed by Furugen and Hayashi [1985]. It
is based on the idea that the closing simulation provides not only the
curvature of the bar, from which the axial strains are re-computed
in the slice model, but the whole total strain distribution in the bar
during the deformation.

5.3.1 Background
The total strain field model relies on the generalizations of assump-
tions 5.3 and 5.4 presented in section 5.2, which can be summarized
as follows:

Assumption 5.6 The total strain distribution in the bar in the steady
state of straightening can be approximated by the one that develops
during a closing simulation.

Figure 5.4 shows the total strain components in the XY Z frame

obtained from a full model simulation and from a closing simulation
for elements in contact with the concave roll. The shear components
are small in comparison to the ones plotted. It appears that the val-
ues obtained from the full model and from the closing simulation are
5.3. TOTAL STRAIN FIELD APPROACH 85

Figure 5.4: Total strains components in elements in contact with the

upper roll from the full model and from a closing simulation (α = 17◦
and δ = 0).

similar. The values obtained from the full model simulation exhibit
noise on the right part of the plot. This phenomenon is more likely
to be due to numerical instabilities in the full model than to an effect
that appears effectively during the process.

Neglecting dynamic effects, assumption 5.6 relies principally on

two sub-assumptions. The first one, assumption 5.7, refers to the
effect of the rolling contact in the process. The second one, assump-
tion 5.8, is about the influence of the material behaviour on the total
strain distribution.

Assumption 5.7 The effect of the rolling contacts between the bar
and the rolls on the deformation that takes place within the bar can
be neglected.

Assumption 5.7 refers to a typical problem of rolling contact as, for

example, a rolling tyre. Figure 5.5 (a) shows the schematic deflection
86 CHAPTER 5. PROCESS MODELLING

Figure 5.5: (a) Deflected cylinder under a static vertical load and
(b) while rolling under a vertical load (from [Popov, 2010]).

of a deformable cylinder under a static load N . Coulomb friction,

characterized by the friction coefficient µ, is assumed to occur be-
tween the cylinder and the ground. Independently of µ, the deforma-
tions in the regions A1 and A2 are symmetrical about the vertical
axis. When a torque M is applied in addition to the load N , the
elements in region B2 are in compression and those in B1 in tension,
as pictured in Figure 5.5 (b). The asymmetry of the deformation
is obvious. If the cylinder is allowed to rotate, the elements enter-
ing the contact region are in compression and remain in that state
during the time of their travel through the contact region. When
they approach the rear end of the contact region, the normal force
decreases and the elements slip in the B1 region.

The problem of rolling contact plays a key role when processes like
rolling tyres are studied. The magnitudes of the deformations taking
place in the case of cross roll straightening are much smaller than in
the case of a rolling tyre. Based on the seminal work of Hertz [1882]
on the contact pressure at the contact interface between two elastic
bodies, Carter [1926] derived the first relations describing the influ-
ence of a rolling motion on the pressure distribution. Kalker [1990]
provides a clear summary of the approach. Be Pmax the maximum
Hertzian pressure between two elastic bodies and be −a ≤ x ≤ a
5.3. TOTAL STRAIN FIELD APPROACH 87

Figure 5.6: Normal pressure according to Hertz [1882] and tangential

traction according to Carter [1926].

the contact zone (x refers to the tangential coordinate), the normal

pressure distribution is
(
0 for |x| > a
Pnorm = Pmax 2

1
2 2
(5.13)
a a − x for |x| ≤ a

The contact region is divided in an adhesion area that begins at

the leading edge of the contact and an area where slip occurs, near
the trailing edge. Be 2a′ the length of the adhesion area and x′
a tangential coordinate centered in the middle of it, the tangential
traction is

 0 for |x| > a

1
Pmax 2 2 2
Ptan = µ a a − x for x ≤ a − 2a′

1
1
 µ Pmax a2 − x2 2 − a′2 − x′2 2 for x > a − 2a′
a
(5.14)
Figure 5.6 shows the normal pressure and the tangential traction,
both normalized, computed according to relations (5.13) and (5.14).
88 CHAPTER 5. PROCESS MODELLING

From relation (5.14), it appears that, according to Carter’s model,

the tangential traction does not exceed µ times the normal pressure.
For the cross-roll straightening process, it can be assumed that the
friction coefficient µ does not exceed 0.10 ∼ 0.15 and that the tan-
gential traction remains small.

Assumption 5.8 The total strain distribution in the bar is marginally

influenced by the chosen set of constitutive equations.

In other words, it is assumed that the changes of the material

properties that occur during straightening do not influence the shape
of the bar in the deformed configuration.

In order to verify this assumption, a closing simulation is made

with a bar segmented in different parts, each of them having dif-
ferent material properties, as shown in Figure 5.7. The reference
material model used for this simulation is the linear elastic-plastic
isotropic model described at the end of section 4.3.5 (σy = 935 MPa
and Etan = 3000 MPa). This reference material model represents
the uniaxial tensile behaviour of the material before straightening.
Assuming the rod moves towards larger Z coordinates, the part of
the bar on the left side of the model is described with the present ma-
terial model. Subsequent segments in the Z direction are described
with the same material model, in which σy is stepwise decreased by
20 MPa, i.e. the segment on the right of the model has a yield stress
σy = 815 MPa.

The results of this simulation are compared to the ones obtained

with a similar model in which the behaviour of the whole bar is
described with the reference material model parameters (see Figure
5.8). The xx, yy and zz components of the total strain tensors of ele-
ments lying on a line in the Y Z plane, 10 mm above the neutral fiber
of the bar (25 mm), are compared in Figure 5.8. The resultant force
on the upper roll in the simulation made with the reference material
5.3. TOTAL STRAIN FIELD APPROACH 89

Figure 5.7: Closing simulation (α = 18◦ and δ = 0.50 mm) on a

segmented bar. The material model parameters of each segment are
different.

Figure 5.8: Comparison of strains obtained with a single material

model (ref) vs. those obtained with the heterogeneous rod shown in
Figure 5.7. The total strains are considered in elements lying 10 mm
above the neutral fiber of the bar (25 mm).
90 CHAPTER 5. PROCESS MODELLING

model is 153 kN, whereas it is 149 kN in the simulation conducted

with the segmented rod—the influence of variations of the material
model on the resultant force is limited. This last point is important
for the calibration of the stamping in the simulation, as explained at
the beginning of chapter 6. The effect of the addition of stamping is
clear upon inspection of Figures 5.4 (δ = 0) and 5.8 (δ = 0.50 mm).
Stamping leads to a massive reduction of the yy total strain compo-
nents and to a reversal of the sign of the xx total strain component
in the vicinity of the center of the bar.

The aspect of the influence of the material behaviour dealt with

above refers principally to the symmetry of the total strain distribu-
tion when the constitutive behaviour of the bar is not constant along
the main axis of the bar. This first point is not directly related to
the influence of the material behaviour when it varies as a whole over
the length of the bar. Murata et al. [2008] studied the influence of
variations of the hardening exponent on the results of tube bending
simulations and came to the conclusion that the influence can be
neglected. It is hence assumed that variations of the material model
parameters do not influence the total strain distribution as a whole.
In other words, the total strain distribution is defined primarily from
the geometries of rolls and the parameters α and δ.

5.3.2 Problem specific procedure

A flowchart of the procedure is presented in Figure 5.9. The first
step is a closing simulation of the bar. The closing simulation pro-
vides the total strain distribution in the bar and the bending line
during the straightening operation. Material points are chosen on a
cross section of the bar and their paths while passing through the
rolls are computed. The solid elements lying on each path are iden-
tified and the sequence of their total strains build up the total strain
histories of the material points considered. Total strain increments
are obtained by substracting consecutive total strain tensors on each
path. The corresponding stress histories are then computed using
5.3. TOTAL STRAIN FIELD APPROACH 91

Figure 5.9: Flowchart of the total strain field method.

92 CHAPTER 5. PROCESS MODELLING

Figure 5.10: εxx , εyy , and εzz total strain components in the center
plane of the bar after the closing simulation (process parameters:
α = 18◦ and δ = 0.50 mm). Only one fourth of a segment of the bar
centered at the origin is plotted.

adequate constitutive equations and initial stresses. The computa-

tion of the stresses being uncoupled from the computation of the
strains, the stresses and strains after the computations may slightly
deviate from a state of equilibrium. Those results are mapped to a
slice of the bar and, under proper boundary conditions, the equilib-
rium is computed. Depending on the type of the desired prediction,
different final computation steps are carried out. The different steps
of the procedure are commented below.

Total strain distribution

According to assumption 5.6, the total strain distribution in the bar
during straightening can be approximated by the one obtained from a
closing simulation, which is much less CPU cost intensive. A closing
simulation, similar to the one from which the curvature is obtained
in the slice model, is conducted. Only the final state of the simula-
tion is relevant for the present procedure. The two main quantities
5.3. TOTAL STRAIN FIELD APPROACH 93

extracted from the simulation are:

• the bending line b (t) of the bar, as for the slice model

• the total strain distribution in the bar

The parameterized bending line b (t) is defined in section 5.2.2. Tech-

nically, the total strain distribution is the collection of the total strain
tensors associated with the solid elements and the coordinates of their
nodes in the deformed configuration.

A typical total strain distribution in the XY Z frame from a clos-

ing simulation is shown in Figure 5.10. There is also a small xy shear
strain component near the surface of the rod. The other shear strain
components are negligible.

Paths of material points

The path p (t) of a material point P as it passes through the rolls
during straightening is pictured in Figure 5.11. The same operation
is repeated for a sufficient number of points on a cross section of the
bar, typically one per element of the mesh also pictured in Figure
5.11.

The moving coordinate system {ξ (t) , η (t) , ζ (t)} is defined in sec-

tion 5.2.2. Let p̃ (ω) be the position of P relative to {ξ (t) , η (t) , ζ (t)}:
 
r cos (ω (t) + ω0 )
p̃ (ω (t)) =  r sin (ω (t) + ω0 )  (5.15)
0

with ω ∈ [0, ωtot ]. The initial position of P in {ξ (0) , η (0) , ζ (0)} is

defined by the parameters r and ω0 . A rotation about the axis X is
defined by R (t) in such a way that

R (t) ez = eζ (t) (5.16)

94 CHAPTER 5. PROCESS MODELLING

Figure 5.11: Schematic path of a point P as it passes through the

rolls and mesh of a section of the bar. One point per element of the
mesh is considered.

The path p (t) of the point P in the XY Z frame is

p (t) = b (t) + R (t) p̃ (ω (t)) (5.17)

ω (t) being defined as for the slice model (see equation (5.10) ω (t) =
2πtLbar /κ). A cylindrical coordinate system {θ (t) , r (t) , ζ (t)} is
attached to the position of P at time t:
 
0
1  b′y (t) 
eζ (t) =
kb′ (t) k
b′z (t)
p (t) − b (t)
er (t) = (5.18)
kp (t) − b (t) k
eθ (t) = er × eζ

Total strain histories

The solids lying on each path are identified. The sequence of the to-
tal strain tensors associated to these elements builds the total strain
5.3. TOTAL STRAIN FIELD APPROACH 95

Figure 5.12: Elements lying on the path of a point in the domain

−50 ≤ Z ≤ 50. The simulation is the same as the one presented in
Figure 5.10.

history of each material point considered. Figure 5.12 shows the el-
ements lying along a path.

In the current form of the procedure, the elements are assumed to

have a constant strain, i.e. a segment of path within an element has
a constant total strain in the XY Z frame. Smoother values could
be obtained by using fully integrated volume elements, considering
the strains at each integration point and interpolating the strain
values along a segment taking into account the shape functions. The
predictions presented in chapter 6 are made on the assumption of
constant strain elements.

Initial stress distribution

An initial stress corresponding to the residual stresses after the draw-
ing operation can be attributed to each material point. Such distri-
butions are axis-symmetric about the axis ζ and are used for predic-
as
tions of the residual stress distribution and of the σ0.2 .

In order to predict the evolution of the curvature of the bar, axis

asymmetric axial initial stresses are considered. An initial curvature
κbs of an elastic beam in the ηζ plane can be reduced to a stress
96 CHAPTER 5. PROCESS MODELLING

distribution over its cross section using the relation

in
σζζ (η) = Eκbs η (5.19)

in the {ξ, η, ζ} coordinate system. For an elastic-plastic material

exempt from initial stresses, relation (5.19) is valid for curvatures
smaller than κy , which is the maximum elastic curvature of the bar,
defined in section 2.1. An axis-symmetric initial stress distribution
can be added as long as the initial total stress distribution obtained
does not lead to an initial plastification, as discussed by Withers and
Bhadeshia [2001].

The initial axial stress (σζζ ) distribution corresponding to a cur-

vature κbs = 1 · 10−4 1/mm is shown in Figure 5.13.

Computation of stresses
From the strain histories defined for each path by the strain tensors
ε(i) at the states i ∈ [1, n], the strain increment for each state on a
given path is defined as

∆ε = ε(i) − ε(i−1) (5.20)

assuming ε(0) is the zero tensor. These strain increment tensors are
then transformed in the local coordinate systems {θ (t) , r (t) , ζ (t)}.
The stress increment ∆σ is then computed as

∆σ = Cep : ∆ε (5.21)

where Cep , the fourth rank material elastoplastic tangent operator,

depends on the chosen constitutive equations. The stress tensor at
state i is then
σ (i) = σ (i−1) + ∆σ (5.22)
where σ (0) corresponds to the initial stress tensor of the material
point considered. The constitutive equations and the integration
procedure are described in chapter 4.
5.3. TOTAL STRAIN FIELD APPROACH 97

Equilibrium in slice

After the computation of the stresses, the equilibrium condition may

not be completely satisfied within a slice. To correct this deviation,
a one-step static implicit simulation is conducted. The results from
the computation of stresses (stresses, accumulated plastic strains and
various internal variables) are mapped to the mesh depicted in Fig-
ure 5.3 (b). All the nodes can move freely in the ξ and η directions.
To avoid rigid body motion, the two nodes in the middle of the slice
are completely constrained and a third one is only free in radial di-
rection. The ζ motion of the lower nodes (in ζ direction) is blocked.
A small displacement ∆ζ is imposed on the upper nodes. ∆ζ is iter-
atively varied until the total resultant force in ζ vanishes.

Prediction of residual stresses

The residual stress distribution in a cross section of the bar is a di-

rect result from the computation of the equilibrium in a slice.

In section 3.3, measurements of residual stresses on the surface

of the bar over a certain length in axial direction are presented.
These values exhibit a cyclic pattern of period κ. In order to make
comparable predictions over a pitch length of the bar, a normalized
pitch value κ is introduced:

κ
κ= (5.23)
Lbar

Instead of defining the start positions of the material points by r and

θ0 in the {ξ (0) , η (0) , ζ (0)}coordinate


system,

they are defined in
different coordinate systems ξ t̃ , η t̃ , ζ t̃ with 0 ≤ t̃ ≤ κ. The
whole procedure is then repeated for different values of t̃, considering
t̃ ≤ t ≤ 1 for the subsequent steps.
98 CHAPTER 5. PROCESS MODELLING

Prediction of residual curvature

The output stress distribution obtained using the initial stress dis-
tribution pictured in Figure 5.13 is shown in Figure 5.14.

An axial stress distribution over a cross section generates a torque

about an axis ξ˜ according to the relation
Z
Mξ̃ = σζζ η̃dA (5.24)
A
n o
˜ η̃ is obtained by a rotation of {ξ, η}
where the coordinate system ξ,
by an angle β about the axis ζ (β is highlighted in Figure 5.15).
Mmax , the maximum torque acting on a section, is defined as
Mmax = max (M (β)) (5.25)
This torque Mmax is related to a curvature by equation (2.6) (M =
EIκ), which corresponds to the curvature of the bar after straight-
ening.

Using this method, the residual curvature κas of a bar having an

initial curvature κbs can be computed. κbs is first reduced to an axial
in in
stress distribution σζζ using relation (5.19). Considering σζζ as ini-
out
tial stresses, the stresses σζζ after straightening are computed. The
out
maximum torque corresponding to σζζ is computed using relations
as
(5.24) and (5.25). The curvature κ = κ (Mmax ) after straightening
is obtained from relation (2.6).

Prediction of yield stress

The deformation induced in the bar during straightening is strongly
inhomogeneous. Hence, the tensile behaviour of the material cannot
be obtained directly from step equilibrium in slice. In order to pre-
as
dict σ0.2 , a virtual tensile test is conducted.
5.3. TOTAL STRAIN FIELD APPROACH 99

Figure 5.13: A typical linear input stress distribution.

Figure 5.14: Axial stress distribution after straightening considering

the initial stress distribution plotted in Figure 5.13.
100 CHAPTER 5. PROCESS MODELLING

Figure 5.15: Same axial residual stress distribution as in Figure 5.14.

The angle β is highlighted (β = −20◦ in this case). The core of the
bar is the region in the which the linear stress distribution that was
present before straightening remains.

The internal variables (residual stresses, accumulated plastic strain,

values of the different backstresses’ components etc. ) obtained for
a cross section of the bar from the equilibrium in slice step (on page
97) are mapped to a 3D segment of the bar. A virtual tensile test
simulation is made and the stress-strain curve is obtained in a sim-
ilar manner as for an experimental tensile test: the apparent axial
strain is obtained by recording the distance between two nodes on
the surface of the bar and the stress is computed by dividing the
axial force on a cross section by the original area of the bar (σ0.2 is
defined on the basis of engineering values of stress and strain). The
mapping of the results is made under consideration of the helical
periodicity inherent to the deformation. The same mesh as for the
bending simulation is used. As mentioned above, the elements lying
on each path of the material points are identified for the determina-
tion of the deformation histories. The end values obtained for the
different points are then attributed to their respective elements.
5.3. TOTAL STRAIN FIELD APPROACH 101

Summary of Chapter 5
A conventional FE model of the cross roll straightening process is
developed. As with the similar procedures presented in chapter 2,
its computational cost is high and it is lacking the precision required
to model the effects presented in chapter 3. The second approach,
i.e. the slice model, does not allow to correctly take into account the
effect of lateral stamping.

The third approach developed, i.e. the total strain field model,
relies on the main hypothesis that the total strain distribution in the
bar can be approximated by a closing simulation. This hypothesis
holds if the effect of rolling contact can be neglected and if the strain
distribution in the bar is only marginally influenced by the mate-
rial behaviour. The procedure consists of a sequence of main steps
followed by specific steps for the prediction of the residual stress dis-
tribution, the residual curvature and the yield stress after straight-
ening.
102 CHAPTER 5. PROCESS MODELLING
Chapter 6

Predictive Capabilities

This chapter presents comparisons between the experimental results

presented in chapter 1 and predictions made using the total strain
field method introduced in section 5.3. In section 6.1, an explana-
as
tion is given to the fact that no significant difference in σ0.2 could
be observed with type II tensile tests. Section 6.2 focuses on the
as
prediction of σ0.2 according to type I tensile tests. The chosen strat-
egy aiming at circumventing the inability of the Chaboche model
to represent correctly the behaviour of SAE 1144 is explained and
discussed. The ability of the model to predict the effect of straight-
ening on the curvature of the bars is discussed in section 6.3, as well
as the assumption of a constant in-plane curvature of the bar after
straightening. Section 6.4 presents a comparison between computed
residual stresses and values from the literature.

Remarks on the computations

If not otherwise stated, the predictions are made according to the
following:
• The procedure used is the total strain field method explained

103
104 CHAPTER 6. PREDICTIVE CAPABILITIES

in section 5.3;
• The parameters used for the Chaboche material model are the
ones listed in Table 4.1. The parameters used for the linear
isotropic model can also be found in section 4.3.5;
as
• The predictions of σ0.2 and of the residual stress distributions
are made taking into account the initial stress distribution pre-
sented by Davis and Mills [1990] (described in section 3.3).
The values are assumed to be representative of the stress dis-
tributions before and after straightening and are scaled to the
radius of the bars studied. This assumption is motivated by
the following facts: (1) Fangmeier [1991] presented qualita-
tively similar results for DIN 9SPb23 (20 mm), (2) both SAE
1144 and SAE 1045 are medium carbon steels, (3) Mughrabi
and Christ [1997] reported that SAE 1045 also exhibits cyclic
softening;
• Except for the initial stress distribution, the bars are assumed
to be homogeneous before straightening1;
• The closing simulations are conducted with a stamping δsim
so that the resultant forces on the rolls correspond to the ones
measured during the straightening experiments. For example,
the mean value of the resultant forces measured on the upper
roll during the straightening of bars in the configuration con-
figuration α = 18◦ and δmach = 0.50 is 155 kN. In order to
achieve the same resultant force in the closing simulation, a
stamping δsim = 0.154 mm is required.

6.1 Inhomogeneity of the Deformation

as
Measurements of σ0.2 using two types of tensile tests are shown in
Figure 3.3. Type I tensile tests, for which the bar is taken as a whole,
1 Although partly true, this assumption is commonly made in forming simu-

lations.
6.1. INHOMOGENEITY OF THE DEFORMATION 105

exhibit a clear dependency between the process parameters set up for

as
straightening and the resulting σ0.2 . Type II tensile tests, for which
specimens are cut from the bars at mid radius, do not lead to sig-
as bs
nificantly differentiated results and the σ0.2 values are similar to σ0.2 .

The distribution of accumulated plastic strain over a cross section

of the bar is obtained directly from the step computation of stresses
presented on page 96. Figure 6.1 presents these distributions for the
different process configurations considered. It appears that only the
outer layers of the bar endure strong deformations, whereas the de-
formations taking place in the core of the bar are, for most of them,
purely elastic. The transition between the regions of high and low
plastic deformation is relatively abrupt. The positions of the type
II tensile tests’ specimens are also highlighted. The specimens are
taken from regions that endure, according to the model, only low
levels of plastic deformation. Thus, the reason why no difference
as
in σ0.2 can be detected becomes obvious. The only type II exper-
as
iment (see Figure 3.3) that exhibits a lower σ0.2 is the one made
on a bar straightened with the configuration α = 18◦ and δ = 0.65
mm. It appears from Figure 6.1 that, for this configuration, the zone
from which the specimen is taken lies closer to the regions of large
plastic deformations. The experiment was apparently conducted on
a specimen that exhibited a little plastic deformation only on one
as
side, which explains the lower σ0.2 value observed. It is also worth
noting that when σ0.2 is defined on the basis of stress-strain curves
obtained from the type I tensile tests, the values represent an av-
erage behaviour of the material over its cross section, which differs
considerably from the local properties.
106 CHAPTER 6. PREDICTIVE CAPABILITIES

Figure 6.1: Accumulated plastic strain distributions over cross sec-

tions after straightening with different process configurations. The
position of type II tensile tests considered in chapter 1 is marked as
a grey circle. α is in degrees and δ in mm.

Figure 6.2: Mapping of the plastic distribution on a segment of the

bar (view from the side) straightened with α = 18◦ and δ = 0.50
mm.
6.2. YIELD STRESS 107

6.2 Yield Stress

6.2.1 Mixed material modelling approach
It is shown in chapter 4 that, for SAE 1144, the Chaboche model is
not able to depict satisfactorily both the first tension branch, which
exhibits a sharp transition from elastic to plastic deformation, and
the subsequent branches, which are more curved. The solution pro-
posed by Jiao and Kyriakides [2009] to circumvent this problem in
the case of one-dimensional deformations is described in section 4.3.5.
The drawback of the Chaboche model regarding the description of
the first transition from the elastic to the plastic state is particularly
as
detrimental to the prediction of σ0.2 . As shown in section 6.1, the
core of the bar undergoes almost no plastic deformation. Hence, the
yield stress of the material of the core of the bar is very similar to
the one of the material before straightening, which corresponds to
the domain that is poorly represented by the Chaboche model.

For the purpose of the present study, two material models are
considered: the Chaboche model (described in section 4.3.3), which
is well suited to the description of the cyclic phenomena, and a sim-
ple linear isotropic model, which correctly depicts the initial tensile
loading branch. The computations in the total strain field approach
are made with the Chaboche model. The discrimination between
the two models is made during the mapping of the results obtained
for a slice to the 3D mesh that is used for the virtual tensile test.
The value of accumulated plastic strain p reached at the end of the
straightening operation is used as a criterion. The integration points
for which p ≤ ptrig are assumed to behave according to the linear
isotropic model, whereas those for which p > ptrig are modelled with
the Chaboche model initialized with the internal variables resulting
from the equilibrium in slice computation step. Figure 6.3 shows the
domains that are modelled using the linear and the Chaboche model.

The major problem with the simultaneous use of two material

108 CHAPTER 6. PREDICTIVE CAPABILITIES

Figure 6.3: Discrimination between the regions described with the

two material models at ptrig = 1.3% (bar straightened with α = 18◦
and δ = 0.50 mm).

models to describe the behaviour of the bar lies in the discontinuity

of the stress distribution that appears during the virtual tensile test,
as shown in Figure 6.4. The stress continuity could be artificially
improved by replacing the linear model by one based on a power
law and by judiciously choosing the material parameters for each
element depending, for example, on the position or on the value
of accumulated plastic strain reached. The approach used in this
study represents a pragmatic way to deal with the current material,
that cannot be well represented by the Chaboche model. As shown
in section 6.1, the stresses from the tensile test simulation represent
integral values over the cross section of the bar. Hence, a parallel with
homogenization techniques could be made. From a purely pragmatic
point of view, the main justification for the use of this artifice is the
clear improvement of the quality of the predictions.

6.2.2 Predictions
The accumulated plastic strains computed for the configuration α =
18◦ and δ = 0.50 mm are plotted in Figures 6.1 and 6.2. It is
shown in section 6.1 that the deformation induced in the bars during
6.2. YIELD STRESS 109

Figure 6.4: Longitudinal section of a bar (straightened with α = 18◦

and δ = 0.50 mm) during the virtual tensile test. The coexistence of
two material models induces an inhomogeneous axial stress distribu-
tion.

straightening is inhomogeneous and that only the outer layers of the

bars endure significant plastic deformation. The virtual tensile tests
are hence evaluated in a similar way as the experimental tests. In
Figure 6.5, both the results obtained with the isotropic-kinematic
hardening model (Chaboche model) and with the one mixed with
the linear isotropic model (with ptrig = 1.3%) are plotted. The value
as
ptrig = 1.3% is chosen so that the predicted σ0.2 value for the config-
◦
uration α = 18 and δ = 0.50 mm is the best. It appears from Figure
6.5 that the consequence of introducing the linear material model for
as
some elements in the bar is twofold: the predicted σ0.2 are shifted
towards higher values and the differences between the predictions are
emphasized.

as
Figure 6.6 shows the predicted values of σ0.2 for the configuration
◦
α = 18 and δ = 0.50 mm as a function of the value attributed to
ptrig . It appears that, for 1 · 10−2 < ptrig < 1.5 · 10−2 , the choice
as
of ptrig plays a minor role in the prediction of σ0.2 . Figure 6.1 pro-
vides a simple explanation for this: the distributions of accumulated
plastic strain show a sharp transition from low to high values. Small
changes in the value of ptrig do not change considerably the number
of elements being described by one model or by the other.
110 CHAPTER 6. PREDICTIVE CAPABILITIES

6.3 Curvature
6.3.1 Predictions on the assumption of a constant
in-plane curvature
Measurements of the curvature of bars before and after straightening
are presented in section 3.2. Figure 6.7 shows the curvature predic-
tions according to this model and the experimental results for 40
bars straightened in the reference configuration. It appears that this
model is able to predict a reduction in curvature due to the cross roll
straightening operation when large initial curvatures are considered.
For small curvatures, the predictions lie approximately an order of
magnitude higher than the measurements. It is shown in section
6.3.2 that the predicted values are in fact local curvatures and that
a comparison with measured (global) curvatures is of limited interest.

6.3.2 Discussion concerning the assumption of con-

stant in-plane curvature
The curvature measurements presented in section 3.2 are based on
the assumption that the bar has a constant curvature in a single
plane both before and after straightening. The analytical methods
presented in section 2.1 are also based on this assumption. In order
to verify it, the curvature κas and its corresponding angle β to the
axis η (defined in section 5.3.2) are computed for different sections
along the axis of the bar over one pitch. The computations are made
for the reference configuration and two initial curvatures: κbs = 0
(a perfectly straight bar) and κbs = 1 · 10−4 1/mm. The results are
shown in Figures 6.8 and 6.9.

It is understood from Figure 6.8 (κbs = 0) that:

6.3. CURVATURE 111

as
Figure 6.5: Predicted and measured values of σ0.2 with different pro-
cess parameters (stamping δ and angle α) and the different material
modelling approaches (ptrig = 1.3% for the mixed approach).

as
Figure 6.6: Predicted σ0.2 for the configuration α = 18◦ and δ = 0.50
mm as a function of the chosen value for ptrig .
112 CHAPTER 6. PREDICTIVE CAPABILITIES

Figure 6.7: The curvatures before (κbs ) and after (κas ) straightening
(with α = 18◦ and δ = 0.50 mm) of 40 bars are measured. The
values predicted by the model are also plotted. The model can not,
apparently, predict curvatures below κas = 6.5 · 10−6 [1/mm].

• the curvature of the bar after straightening is approximately

constant over the length of one pitch;
• the angle β between the axis of principal curvature and the axis
ζ changes almost linearly over one pitch, making a complete
rotation.
The shape of the bar after straightening is hence a helix. This ob-
servation leads to the major conclusion according to which the as-
sumption of constant curvature in a single plane after staightening
does not hold. The curvature measurements, which are made on this
assumption, represent an average global curvature, different from
the local curvature of the bar. The predictions shown in Figure 6.7
are local curvatures and a comparison with measurements conducted
with the setup pictured in Figure 3.4 is not relevant.
6.3. CURVATURE 113

Figure 6.8: The curvature κas and its corresponding angle β to the
axis ζ over an axial distance equal to one pitch κ. The results are
obtained for a bar straightened with α = 18◦ and δ = 0.50 mm
(κ = 25.52 mm) with no initial curvature.

Figure 6.9: Curvature over over pitch length obtained for the same
configuration as plotted in Figure 6.8 except for κbs = 10−4 1/mm.
The strip in which κas varies when the same computations are made
with κbs = 0 (see Figure 6.8) is highlighted.
114 CHAPTER 6. PREDICTIVE CAPABILITIES

Knowing that the period of the helix is κ and using relation (3.1)
(κ ≈ 8Υ/L2 ), the corresponding amplitude Υ that would have to be
measured in order to determine the local curvature of a bar can be
computed. For example, a curvature κ = 6.5 · 10−6 [1/mm] over a
length L = κ/2 would imply a deviation Υ = 0.14 [µm]. The surface
of the straightened bars having a roughness Ra ≈ 0.80 [µm], the
measurement of such deviations may prove challenging.

The analytical method presented in section 2.1 yields κas =

6.5 · 10−6 [1/mm] when κbs = 0 is taken as input. It is interest-
ing to observe that this value is relatively close to the one obtained
with the total strain field method. Of course, the analytical method
provides only little understanding of the role of β. Also, when larger
curvatures are considered, the limitation of the analytical method
regarding sensitivity to the sign of κbs remains a major impediment
to its use.

From Figure 6.9 (κbs = 10−4 1/mm) it appears that:

• the curvature of the bar after straightening is not constant over

the length of one pitch;
• the angle β between the axis of principal curvature and the axis
ζ also varies over one pitch length but in a narrower domain
than it does for κbs = 0.

Figure 5.15 provides an explanation for the behaviour of the value

β. The curvature of the bar is mostly determined by the remaining
linear stress distribution in the core of the bar. This stress distribu-
tion leads to a persistent torque about the axis ξ (the axis of initial
constant curvature). The orientation of the stress distribution in the
outer layers of the bar depends on the axial position of the section
considered. This is shown in Figure 6.9, where no initial stress dis-
tribution is considered (κbs = 0), and κas depends exclusively on
the outer stress distribution. The outer stress distribution leads to
the part of the torque that changes over the length of the bar. The
6.4. RESIDUAL STRESSES 115

strip in which β varies in Figure 6.9 is the result of the equilibrium

between the influences of the stress distribution in the core of the
bar vs. the one in its peripheral layers.

6.4 Residual Stresses

6.4.1 Predictions for the reference configuration
Due to the inhomogeneous deformation during straightening (high-
lighted in section 6.1), the residual stress distribution is not sym-
metrical about the axial direction on a section. The residual stresses
in a cross section of the bar are nevertheless in equilibrium (within
the tolerance of the FEM). For this reason, the computed residual
stresses are plotted as a domain (the envelope of the curves from
the simulation) in Figure 6.10 and not as a line. It appears from
Figure 6.10 that the axial residual stresses computed are in qualita-
tive agreement with the measurements provided by Davis and Mills
[1990] when the corresponding values measured before straightening
are taken as initial residual stresses. The agreements between the en-
velopes obtained for the radial and tangential residual stresses after
straightening are comparatively worse, although they remain quali-
tatively acceptable. It is shown in section 3.3 that the axial residual
stresses on the surface of a straightened bar vary over its length. The
predictions shown in Figure 6.11 exhibit the same behaviour over one
pitch.

6.4.2 Predictions for different configurations

Figure 6.12 shows axial stress distributions computed for different
values of the process parameters. It appears that the shape of the
distributions remains qualitatively similar. The process parameters
seem to have an influence on the size of the envelopes. The areas of
those envelopes, normalized to the value obtained for the reference
116 CHAPTER 6. PREDICTIVE CAPABILITIES

Figure 6.10: Residual stresses before and after straightening mea-

sured by Davis and Mills [1990] and predictions according to the
present procedure (with α = 18◦ and δ = 0.50 mm) : (top) radial
component, (center) tangential component, (bottom) axial compo-
nent. The bar is assumed to be perfectly straight before straightening
(κbs = 0).
6.4. RESIDUAL STRESSES 117

configuration, are noted A and provide a simple means of comparing

how inhomogeneous the stress distributions are on a section of the
bar. The measurements of Davis and Mills [1990] are based on the
assumption that the distributions are axis-symmetric and are hence
plotted as lines. It appears from the values of A in Figure 6.12 that:
• an increase in the value of α leads to an increase in the value
of A

• an increase in the value of δ leads to an increase in the value

of A
The role of α can be explained by its direct influence on the pitch:
the smaller α, the smaller the pitch. A smaller pitch means a larger
number of rotations over a given length of the bar, hence a more
homogeneous deformation on a cross section.
118 CHAPTER 6. PREDICTIVE CAPABILITIES

Figure 6.11: Axial residual stresses (after straightening with α = 18◦

and δ = 0.50 mm) computed for two elements lying close to the
surface of the bar over an axial distance of one pitch.

Figure 6.12: Axial stress distributions computed for different pa-

rameter configurations (α and δ). The areas of the envelopes of the
distributions of residual stresses are noted A (normalized to the area
of the one for the reference configuration).
6.4. RESIDUAL STRESSES 119

Summary of Chapter 6
The inhomogeneous deformation that takes place during cross roll
straightening is modelled: only the peripheral layers of the bar en-
dure plastic deformation. This inhomogeneity explains the difference
between the outcomes of tensile tests that were conducted with dif-
ferent specimen geometries.

The model is able to predict the yield stress of bars straight-

ened with different process parameters provided a parameter ptrig
(a value of accumulated plastic strain) is calibrated on one experi-
mental result. The calibration parameter is necessary to compensate
for the inability of the Chaboche model to depict both the initial
transition from elastic to plastic and the subsequent deformations.
The elements for which p < ptrig during cross roll straightening are
described with a linear isotropic model during the virtual tensile test.

The model is able to predict local curvatures of straightened bars.

Straightened bars exhibit a 3D curvature and the assumption of con-
stant in plane curvature does not hold. The curvature measurements
are based on the assumption of a curvature in a single plane. Thus,
a comparison between experimental results and predictions is not
relevant.

Residual stress measurements from the literature before and af-

ter straightening are taken as references. Considering the measured
values as initial stresses for the model, the predicted residual stress
distributions are in good qualitative agreement with the experimen-
tal values after straightening. The cyclic pattern of the axial residual
stresses on the surface of straightened bars can also be qualitatively
predicted.
120 CHAPTER 6. PREDICTIVE CAPABILITIES
Chapter 7

Conclusions and
Outlook

7.1 Conclusions
The objective of the present study is defined in section 1.4: the de-
velopment of a model able to predict the curvature, the yield stress
and the residual stress distribution of bars after straightening.

In chapter 2, existing modelling approaches are presented. For

cross roll straightening, the FE approaches presented by Onoda et al.
[2002] and Yanagihasi et al. [2005] are able to predict the resulting
curvature for the case of straightening without stamping, i.e. when
the deformation is one-dimensional. No report of the prediction of
the curvature after straightening when stamping is applied could be
found. The prediction of the yield stress of staightened bars is not
documented either. Predictions of the residual stress distribution in
railway rails are presented, among others, by Schleinzer and Fischer
[2000, 2001]. The straightening of rails is based principally on a one-
dimensional deformation. The approach presented by Furugen and

121
122 CHAPTER 7. CONCLUSIONS AND OUTLOOK

Hayashi [1985] proved to be able to successfully predict the residual

stress distribution in straightened tubes.

The total strain field approach presented in chapter 5 is able, at

least in the case of the particular process under study, to reach the
goals defined in chapter 1. Using the set of mixed isotropic-kinematic
constitutive relations presented in chapter 4, the model is able to
make predictions concerning the yield stress after straightening and
the residual stress distribution, as shown in chapter 6. Predictions of
the local curvature of bars after straightening can also be made. The
inadequacy of the current definition of the straightness is discussed
in chapter 6.

The model developed in the present study allows to make pre-

dictions concerning the principal phenomena induced by cross roll
straightening. A meta-modelling technique was applied to the sim-
as
ulation results, allowing for fast predictions of σ0.2 , and a prototype
software for the process control was developed.

7.2 Outlook
The total strain field model developed in the context of this study
is a step towards the optimization of the geometries of the rolls. In
order to be viable for optimization tasks, the simulation of cross roll
straightening should be further improved. This section presents the
principal steps which should be the subject of further investigations.

7.2.1 Material modelling

The review of investigations on microstructural phenomena presented
in section 4.2 highlights the fact that the mechanisms of cyclic soft-
ening are still not fully understood. A better understanding of these
phenomena would be an asset for the development of materials or
processes which would allow to reach the goals of straightening while
7.2. OUTLOOK 123

limiting the downside of softening.

As shown in section 4.3.5, the Chaboche model could not be fitted

to the experimental stress strain hysteresis curve in a completely sat-
isfying way. The simultaneous description of the sharp initial elastic
to plastic transition and of the subsequent, more curved transitions
is difficult. A material model able to depict these phenomena would
allow to describe the material without requiring the artifice of using
simultaneously two material models, as explained in section 6.2.1.

The fitting of the parameters of the material model explained

in section 4.3.5 is based on experimental data from uniaxial tension-
compression tests whereas the deformation during cross roll straight-
ening is three-dimensional. Devising a way to fit the material model
parameters on data from non-coaxial cyclic tests would allow to em-
bed the 3D response of the material in the model. To this end, a
possible solution would be to make a few tension-compression cy-
cles using a specimen like the one shown in Figure 4.1, to remove a
small cuboid from the deformed region and submit it to lateral com-
pression. A more advanced procedure would be to fit the material
parameters by minimizing a multi-objective function based on a set
of such tests. For example, the fitting could be made simultaneously
on hystereses of different strain amplitudes or with different numbers
of tension-compression cycles before the lateral compression.

7.2.2 Process modelling

The important CPU cost and its implications on the achievable pre-
cision of the conventional FE approaches are described in sections
2.3 and 5.1. These limitations motivate the development of the total
strain field model. With the steady rise in accessible computation
capabilities, it is likely that the actual limitations of the full FE
approaches will be lowered. For example, as the full model was de-
veloped with LS-Dyna c in the autumn of 2008 a typical simulation
was computed on 4 CPUs and took about 105 wall clock hours to
124 CHAPTER 7. CONCLUSIONS AND OUTLOOK

run. The same deck could be run on 8 CPUs during spring 2011 and
took 35 hours to complete.

The total strain field model is much faster than the full model.
The closing simulation takes approximately 2.5 hours to run on a 4
CPUs workstation, the identification of the solids along the paths
takes about 3 hours, the computation of the stresses about 10 min-
utes and the other operations, such as the generation of the paths,
only a few minutes. For the prediction of σ0.2 , the simulation takes
about 1 hour.

As the computations in the total strain field model are conducted

separately for the different material points considered, it is, in its
essence, better suited to parallel computing than the full model.
Even with increasing CPU capabilities, the total strain field model
should remain faster. It is therefore unclear whether future simula-
tions of cross roll straightening will rely on full FE approaches or on
enhanced total strain field approaches.

Among the possible enhancements of the total strain field model,

an iterative approach to come closer to the state of equilibrium during
the deformation could be introduced. This could be achieved by
iteratively varying the distribution of the material properties of the
bar during the closing simulation. Also, it should be possible to
devise a way of taking into account the effect of the rolling contact
on the strain distribution during straightening.

7.2.3 Predictive capabilities

The predictive capabilities of the total strain field model are com-
pared to a relatively limited amount of experimental data in chapter
6. Further validation on different materials, diameters and process
configurations is required.

In order to assess more precisely the capabilities of the model,

7.2. OUTLOOK 125

micro tensile or compressive specimens could be machined from pe-

ripheral regions of the straightened bars. Comparisons of mechanical
properties from such experimental data and predictions would allow
to eliminate the mitigation due to the plastically undeformed central
region of the bar.
126 CHAPTER 7. CONCLUSIONS AND OUTLOOK
Bibliography

P.J. Armstrong and C.O. Frederick. A mathematical representa-

tion of the multiaxial bauschinger effect. Technical Report No.
RD/B/N 731, Central Electricity Generating Board, 1966. re-
published in Frederick [2007].

M. Asakawa, M. Urabe, K. Nishimura, R. Hamada, S. Aizawa, and

M. Amari. Theoretical and experimental analysis of roller leveller
straightening for coiled bar. Steel Research International, 81(9):
242–245, 2010.

J. Balic and M. Nastran. An on-line predictive system for steel wire

straightening using genetic programming. Engineering Applica-
tions of Artificial Intelligence, 15:559–565, 2002.

S. Bari and T. Hassan. Anatomy of coupled constitutive models

for ratcheting simulation. International Journal of Plasticity, 16:
381–409, 2000.

F. Barlat. Constitutive modeling for metals. In D. Banabic, editor,

Advanced Methods in Material Forming, pages 1–18, Berlin, 2007.
Springer.

B. Barralis and G. Maeder. Précis de métallurgie – élaboration,

structures-propriétés, normalisation. Nathan, 1997. ISBN 2-09-
177491-X. in French.

127
128 BIBLIOGRAPHY

J. Bauschinger. Über die Veränderung der Elastizitätsgrenze und

des Elastizitätsmoduls verschiedener Metalle. Zivilingenieur, 27:
289–348, 1881.

C. Betegon Biempica, J.J. del Coz Diaz, P.J. Garcia Nieto, and
I. Peñuelas Sanchez. Nonlinear analysis of residual stresses in a rail
manufacturing process by fem. Applied Mathematical Modelling,
33:34–53, 2009.

K. Bhanu Sankara Rao, M. Valsan, R. Sandhya, S.L. Mannan, and

P. Rodriguez. An assessment of cold work effects on strain-
controlled low-cycle fatigue behavior of type 304 stainless steel.
Metallurgical Transactions A, 24A:913–924, 1993.

F.W. Carter. On the action of a locomotive driving wheel. In Pro-

ceedings of the Royal Society of London. Series A, Containing Pa-
pers of a Mathematical and Physical Character, volume 112, pages
151–157. The Royal Society, 1926.

O. Cazacu and F. Barlat. A criterion for description of anisotropy

and yield differential effects in pressure-insensitive metals. Inter-
national Journal of Plasticity, 20:2027–2045, 2004.

J.L. Chaboche. Time-independent constitutive theories for cyclic

plasticity. International Journal of Plasticity, 2:149–188, 1986.

J.L. Chaboche. A review of some plasticity and viscoplasticity consti-

tutive theories. International Journal of Plasticity, 24:1642–1693,
2008.

H.-F. Chai and C. Laird. Mechanisms of cyclic softening and cyclic

creep in low carbon steel. Materials Science and Engineering, 93:
159–174, 1987.

N. Christodoulou, O.T. Woo, and S.R. MacEven. Effect of stress re-

versals on the work hardening behaviour of polycrystalline copper.
Acta Metallica, 34(8):1553–1562, 1985.
BIBLIOGRAPHY 129

Y.F. Dafalias and E.P. Popov. A model of nonlinearly hardening

materials for complex loading. Acta Mechanica, 21:173–192, 1975.
N.K Das Talukder and W. Johnson. On the arrangement of rolls
in cross-roll straighteners. International Journal of Mechanical
Sciences, 23:213–220, 1981.

N.K Das Talukder and A.N. Singh. Mechanics of bar straightening.

Transactions of the ASME, 113:224–232, 1991.

J.R. Davis and K.M. Mills, editors. Metals Handbook, vol. 1: Prop-
erties and Selection: Irons, Steels, and High-Performance Alloys,
chapter Cold-Finished Steel Bars, page 258. ASM International,
Materials Park (OH), 10 edition, 1990.
Y. Duyi and W. Zhenlin. Change characteristics of static mechanical
property parameters and dislocation structures of 45# medium
carbon structural steel during fatigue failure process. Materials
Science and Engineering A, 297:54–61, 2001.
R. Fangmeier. Untersuchungen über das Richten von Rundstäben
in Zwei-Walzen-Richtmaschninen. PhD thesis, Technischen
Hochschule Claustahl, 1966. in German.
R. Fangmeier. Ziehen und Richten von Stäben aus Ringen. Stahl u.
Eisen, 111(3):121–126, 1991. in German.
C.E Feltner and P. Beardmore. Strengthening mechanisms in fatigue.
In Achievement of high fatigue resistance in metals and alloys,
ASTM STP 467, pages 77–112, 1970.
C.E. Feltner and C. Laird. Cyclic stress-strain response of f.c.c met-
als and alloys – II: dislocation structures and mechanisms. Acta
Metallurgica, 15:1633–1653, 1967.
G. Finstermann, F.D. Fischer, G. Shan, and G. Schleinzer. Residual
stress in rails due to roll straightening. Steel Research, 69(7):272–
278, 1998.
130 BIBLIOGRAPHY

P.J. Frederick, C.O. Armstrong. A mathematical representation of

the multiaxial bauschinger effect. Materials at High Temperatures,
24(1):1–26, 2007.

M. Furugen and C. Hayashi. Theory of tube deformation on cross roll

straightening. In Proceedings of the 3rd International Conference
on Steel Rolling, pages 717–724, Tokyo, Japan, 1985.

S. Ganesh Sundara Raman and K.A. Padmanabhan. Effect of prior

cold work on the room-temperature low-cycle fatigue behaviour of
aisi 304ln stainless steel. International Journal of Fatigue, 18(2):
71–79, 1996.

T. Hasegawa, T. Yakou, and S. Karashima. Deformation behaviour

and dislocation structures upon stress reversal in polycrystalline
aluminium. Materials Science and Engineering, 20:267–276, 1975.

H. Hertz. Über die Berührung fester elastischer Körper. Journal für

die reine und angewandte Mathematik, pages 156–171, 1882.

R. Jiao and S. Kyriakides. Ratcheting, wrinkling and collapse of

tubes under axial cycling. International Journal of Solids and
Structures, 46:2856–2870, 2009.

J.J. Kalker. Three-dimensional elastic bodies in rolling contact.

Kluwer Academic Publishers, 1990.

Y. Kinkori. Study on (precise) straightening of bars by two-roll pro-

cess. Master’s thesis, Waseda University Tokyo, 1998. in Japanese.

T. Kishi and T. Tanabe. The Bauschinger effect and its role in

mechanical anisotropy. Journal of the Mechanics and Physics of
Solids, 21:303–315, 1973.

M. Klesnil and P. Lukas. Dislocation arrangements in the surface

layer of α-iron grains during cyclic loading. Journal of the Iron
and Steel Institute, 203(10):1043–1048, 1965.
BIBLIOGRAPHY 131

M. Klesnil and P. Lukas. Fatigue softening and hardening of annealed

low-carbon steel. Journal of the Iron and Steel Institute, 205(7):
746–749, 1967.
M. Kobayashi and N. Ohno. Implementation of cyclic plasticity
models based on a general form of kinematic hardening. Inter-
national Journal for Numerical Methods in Engineering, 53:2217–
2238, 2002.
K. Kuboki, H. Huang, M. Murata, Y. Yamaguchi, and K. Kuroda.
Fem analysis of tube straightener adopting implicit scheme. Steel
Research International, 81(9):584–587, 2010.
T. Kuwabara, R. Saito, T. Hirano, and N. Oohashi. Difference in
tensile and compressive flow stresses in austenitic stainless steel
alloys and its effect on springback behavior. International Journal
of Material Forming, 2(1):499–502, 2009.
R.W. Landgraf. The resistance of metals to cyclic deformation. In
Achievement of high fatigue resistance in metals and alloys, ASM
STP, volume 467, pages 3–36, 1970.
H.J. Leber, M. Niffenegger, and B. Tirbonod. Microstructural as-
pects of low cycle fatigued austenitic stainless tube and pipe steels.
Materials Characterization, 58:1006–1015, 2007.
J. Lemaitre and J.-L. Chaboche. Mechanics of solid materials. Cam-
bridge University Press, 1990.
M.A.W. Lowden and W.B Hutchinson. Texture strengthening and
strength differential in titanium-6al-4v. Metallurgical Transactions
A, 6a:441–448, 1975.
P. Medart. Machine for straightening and polishing shafting. US
Patent 558591, United States Patent Office, 1896.
R. Menz. Entwicklng eines analytischen Simulationsmodells als
Grundlage einer geregelten Richtmaschine. PhD thesis, Univer-
sity of Hanover, 2002. in German.
132 BIBLIOGRAPHY

F. Mollica, K.R. Rajagopal, and A.R. Srinivasa. The inelastic be-

havior of metals subject to loading reversal. International Journal
of Plasticity, 17:1119–1146, 2001.

J. Morrow. Cyclic plastic strain energy and fatigue of metals. In

Internal friction, damping, and cyclic plasticity, ASTM STP 467,
pages 45–87, 1965.

R.W. Moses. Roller leveler. US Patent 1959492, United States Patent

Office, 1934.

Z. Mroz. On the description of anisotropic workhardening. Journal

of the Mechanics and Physics of Solids, 15:163–175, 1967.

H. Mughrabi and H.J. Christ. Cyclic deformation and fatigue of

selected ferritic and austenitic steels: specific aspects. ISIJ Inter-
national, 37(12):1154–1169, 1997.

M. Murata, T. Kuboki, K. Takahashi, M. Goodarzi, and Y. Jin. Ef-

fect of hardening exponent on tube bending. Journal of Materials
Processing Technology, 201:189–192, 2008.

A. Mutrux, B. Berisha, B. Hochholdinger, and P. Hora. Numerical

modelling of cross roll straightening. In Proceedings of the 7th LS-
Dyna Anwederforum, pages C–I–33–40, Bamberg, Germany, 2008.

A. Mutrux, B. Berisha, M. Weber, and P. Hora. FE simulation of

cross roll straightening: a submodel approach. In E. Õnate and
D.R.J. Owen, editors, Proceedings of the 10th International Con-
ference on Computational Plasticity (COMPLAS X), page CD,
Barcelona, Spain, 2009.

A. Mutrux, B. Berisha, and P.Hora. FE simulation of cross roll

straightening: a strain tensor field approach. In F. Barlat, Y.H.
Moon, and M. Lee, editors, Proceedings of the 10th International
Conference on Numerical Methods in Industrial Forming Processes
(NUMIFORM 2010), pages 941–948, Pohang, Korea, 2010.
BIBLIOGRAPHY 133

A. Mutrux, B. Berisha, and P. Hora. Prediction of cyclic softening

in a medium carbon steel during cross roll straightening. Journal
of Materials Processing Technology, 211:1448–1456, 2011.

N. Ohno and J.-D. Wang. Kinematic hardening rules with critical

state of dynamic recovery. International Journal of Plasticity, 9:
375–403, 1993.

W.A. Olson and C.W. Bert. Analysis of residual stresses in bars and
tubes of cylindrically orthotropic materials. Experimental Mechan-
ics, 6(9):451–457, 1966.

Y. Onoda, T. Yanagihasi, T. Hama, and M. Asakawa. A fem simu-

lation of two-roll straightening for bars and wires. In Proceedings
of the 7th International Conference on Technology of Plasticity
(ICTP), pages 625–630, Yokohama, Japan, 2002.

A. Parker. A critical examination of Sachs’ material-removal method

for the determination of residual stress. Transactions of the ASME,
126:234–236, 2004.

M.S. Paterson. Plastic deformation of copper crystals under alternat-

ing tension and compression. Acta Metallurgica, 3:491–500, 1955.

W.J. Plumbridge and D.A. Ryder. The metallography of fatigue.

Metallurgical Reviews, 14:119–142, 1969.

B. Popov. Contact Mechanics and Friction: Physical Principles and

Applications. Springer, Berlin, 2010.

W. Prager. A new method of analyzing stresses and strains in work

hardening plastic solids. Journal of Applied Mechanics, 23:493–
496, 1956.

E.F. Rauch. The stresses and work hardening rates of mild steel with
different dislocation patterns. Materials Science and Engineering
A, 234-236:653–656, 1997.
134 BIBLIOGRAPHY

H.J. Roven and E. Nes. Cyclic deformation of ferritic steel–I. stress-

strain response and structure evolution. Acta Metallurgica et Ma-
terialia, 39(8):1719–1733, 1991.
S. Sankaran, V. Subramanya Sarma, and H.A. Padmanabhan. Low
cycle fatigue behavior of a multiphase microalloyed medium car-
bon steel: comparison between ferrite-pearlite and quenched and
tempered microstructures. Materials Science and Engineering A,
345:328–335, 2003.
S. Sankaran, Gouthama, S. Sangal, and K.A. Padmanabhan. Trans-
mission electron microscopy studies of thermomechanically control
processed multiphase medium-carbon microalloyed steel: forged,
rolled, and low-cycle fatigued microstructures. Metallurgical and
Materials Transactions A, 37A:3259–3273, 2006.
G. Schleinzer and F.D. Fischer. Residual stresses in new rails. Ma-
terials Science and Engineering A, 288:280–283, 2000.
G. Schleinzer and F.D. Fischer. Residual stress formation during
the roller straightening of railway rails. International Journal of
Mechanical Sciences, 43:2281–2295, 2001.
H. Tokunaga. On the roller straightener: report 2, straightening of
round bars, pipe, tubings. Bulletin of JSME, 4:605–611, 1961.
A.H. van den Boogaard. Thermally enhanced forming of aluminium
sheet. PhD thesis, University of Twente, 2002.
P.J. Webster, L.C. Mills, X. Wang, and W.P. Kang. Residual
stress measurements in rails by neutron diffraction, pages 307–314.
Kluwer Academic Publishers, Delft, Netherlands, 1993.
S. Wiese. Verformung und Schädigung von Werkstoffen der Aufbau-
und Verbindungstechnik. Springer, Berlin, 2010. in German.
P.J. Withers and H.K.D.H. Bhadeshia. Residual stress. part 1 mea-
surement techniques. Materials Science and Technology, 17:355–
365, 2001.
P. Wriggers and T. Laursen, editors. Computational Contact Me-
chanics, chapter Modern approaches on rolling contact, pages 83–
127. Springer, Berlin, 2007.
B.J. Wu, L.C. Chan, T.C. Lee, and L.W. Ao. A study on the preci-
sion modeling of the bars produced in two cross-roll straightening.
Journal of Materials Processing Technology, 99:202–206, 2000.
T. Yanagihasi, T. Hama, Y. Onoda, and M. Asakawa. Effect of plas-
tic ratio and repeat bending on straightness in two-roll straight-
ening. Journal of the Japan Society for Technology Plasticity, 46:
972–976, 2005. in Japanese.
F. Yoshida and M. Urabe. Computer-aided process design for the
tension-levelling of metallic strips. Journal of Materials Processing
Technology, 89-90:218–223, 1999.
T.X. Yu and W. Johnson. Estimating the curvature of bars after
cross roll straightening. In Proceedings of the 22nd International
Machine Tool Design and Research Conference (MTDR), pages
517–521, Manchester, UK, 1981.
T.X. Yu and L.C. Zhang. Plastic bending: theory and applications.
World Scientific Publishing, 1996.
S.Y. Yuan, L.W. Zhang, S.L. Liao, G.D. Jiang, and M. Qi. Sim-
ulation of deformation and temperature in multi-pass continuous
rolling by three-dimensional fem. Journal of Materials Processing
Technology, 209:2760–2766, 2009.

135
Curriculum Vitae
Alexandre MUTRUX
born February 17, 1983, in Geneva, Switzerland
Swiss and Greek

Education

2007–2011 Doctoral studies

Institute of Virtual Manufacturing, ETH Zurich (CH)

2002–2007 Diploma in Management and Manufacturing

EPFL, Lausanne (CH) and ETH Zurich (CH)

2002 Matura
Collège Calvin, Geneva (CH)

Work Experience

2007 Diploma thesis

6 months BMW, Munich (DE)

2006 Internship
3 months JFE Steel, Chiba (JP)

2006 Internship
5 months ThyssenKrupp Steel, Duisburg (DE)

136
List of Publications

A. Mutrux, B. Berisha and P. Hora. Prediction of cyclic softening in

a medium carbon steel during cross roll straightening. Journal of
Materials Processing Technology, 211(8):1448–1456, 2011.

A. Mutrux, B. Berisha and P. Hora. On the prediction of the cur-

vature of cross roll straightened bars. Proceedings of the 11th In-
ternational Conference on Computational Plasticity, Barcelona,
Spain, 2011.

A. Mutrux, B. Berisha and P. Hora. Numerical methods for the eval-

uation and the optimization of the cross roll straightening process.
Proceedings of the 4th Forming Technology Forum, pages 157–161,
Zurich, Switzerland, 2011.

A. Mutrux, B. Berisha and P. Hora. FE simulation of cross roll

straightening: a strain tensor field approach. Proceedings of the
10th International Conference on Numerical Methods in Industrial
Forming Processes, pages 941–948, Pohang, Korea, 2010.

A. Mutrux, B. Berisha, M.Weber and P. Hora. FE simulation of cross

roll straightening: a submodel approach. Proceedings of the 10th
International Conference on Computational Plasticity, Barcelona,
Spain, 2009.

A. Mutrux, B. Berisha, B. Hochholdinger and P. Hora. Numerical

modelling of cross roll straightening. Proceedings of the 7th LS-
Dyna Anwenderforum, Bamberg, Germany, 2008.

A. Mutrux, B. Hochholdinger and P. Hora. A procedure for the eval-

uation and validation of the hydraulic biaxial experiment. Pro-
ceedings of the 7th International Conference and Workshop on Nu-
merical Simulation of 3D Sheet Metal Forming Processes, pages
67–71, Interlaken, Switzerland, 2008.

137