A Displacement Based FE Formulation

for Steady State Problems

This research was carried out under project number M E97034 in the frame-
work of the Strategic Research programme of the Netherlands Institute of
Metals Research in the Netherlands (www.nimr.nl)

Samenstelling van de promotiecommissie:

voorzitter en secretaris:
Prof. dr. ir. H.J. Grootenboer Universiteit Twente
promotor:
Prof. dr. ir. J. Huétink Universiteit Twente
leden:
Prof. dr. ir. R. Akkerman Universiteit Twente
Dr. ir. H.J.M. Geijselaers Universiteit Twente
Prof.dr.ir. F.J.A.M. van Houten Universiteit Twente
Prof. ir. L.Katgerman Technische Universiteit Delft
Prof.dr.ir. H.J.J. Slot DSM en Universiteit Twente

A Displacement Based FE Formulation for Steady State Problems

Yuhong Yu
Met samenvatting in het Nederlands
ISBN 90-77172-13-0
Keywords: elastic-plasticity, steady state flow processes, material evolution
equation, finite element

This thesis was prepared with LATEX by the author and printed by Ponsen
& Looijen, Wageningen, from an electronic document.

c 2004 by Y. Yu, Eindhoven, The Netherlands

Copyright °

All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted in any form or by any means, electronic, mechanical,
photocopying, recording or otherwise, without prior written permission of the copyright
holder.
A DISPLACEMENT BASED FE FORMULATION FOR STEADY
STATE PROBLEMS

PROEFSCHRIFT

ter verkrijging van

de graad van doctor aan de Universiteit Twente,
op gezag van de rector magnificus,
prof. dr. W.H.M. Zijm,
volgens besluit van het College voor Promoties
in het openbaar te verdedigen
op woensdag 12 januari 2005 om 13.15 uur

door

Yuhong Yu

geboren op 9 augustus 1970

te Wuhan, China
Dit proefschrift is goedgekeurd door de promotor:

Prof. dr. ir. J. Huétink

Contents

Summary ix

Samenvatting xi

1 Introduction 1
1.1 Industrial Revolution and Development . . . . . . . . . . . . 1
1.2 Project Definition . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Overview of the Thesis . . . . . . . . . . . . . . . . . . . . . . 3
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Continuum Mechanics 7
2.1 Basic Concepts of Deformation . . . . . . . . . . . . . . . . . 7
2.2 Strain and Stress . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.1 Strain Measure . . . . . . . . . . . . . . . . . . . . . . 9
2.2.2 Stress Measure . . . . . . . . . . . . . . . . . . . . . . 10
2.2.3 Objectivity or Frame Indifference . . . . . . . . . . . . 11
2.3 Motion and Equilibrium . . . . . . . . . . . . . . . . . . . . . 13
2.3.1 Cauchy’s Equation of Motion . . . . . . . . . . . . . . 13
2.3.2 Equilibrium Equations . . . . . . . . . . . . . . . . . . 14
2.4 Constitutive Equations . . . . . . . . . . . . . . . . . . . . . . 15
2.4.1 Hypoelastic-plasticity . . . . . . . . . . . . . . . . . . 16
2.4.2 Hyperelastic-plasticity . . . . . . . . . . . . . . . . . . 17
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3 Literature Review – Steady State 23

3.1 Definition of a Steady State Process . . . . . . . . . . . . . . 23
3.2 Steady State Formulations . . . . . . . . . . . . . . . . . . . . 24
3.2.1 Lagrangian Formulation . . . . . . . . . . . . . . . . . 25
3.2.2 Eulerian Formulation . . . . . . . . . . . . . . . . . . 26
3.2.3 ALE Formulation . . . . . . . . . . . . . . . . . . . . . 26
3.3 Referential Formulation (Balagangadhar) . . . . . . . . . . . 29
3.3.1 Construction of a Reference Configuration . . . . . . . 29
vi Contents

3.3.2 Steady State at the Reference Configuration . . . . . . 30

3.3.3 Contact Problem . . . . . . . . . . . . . . . . . . . . . 33
3.3.4 Solution Procedure . . . . . . . . . . . . . . . . . . . . 34
3.3.5 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 Objectives of This Project . . . . . . . . . . . . . . . . . . . . 35
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4 Finite Element Method 39

4.1 Principle of Virtual Work . . . . . . . . . . . . . . . . . . . . 39
4.2 Linearization of the Weak Form . . . . . . . . . . . . . . . . . 42
4.3 Finite Element Discretization . . . . . . . . . . . . . . . . . . 44
4.4 Iteration Procedure . . . . . . . . . . . . . . . . . . . . . . . . 47
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

5 Displacement Based Steady State Formulation 51

5.1 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . 52
5.1.1 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . 52
5.1.2 Constitutive Relations . . . . . . . . . . . . . . . . . . 52
5.1.3 Convection Equations . . . . . . . . . . . . . . . . . . 53
5.2 Material Evolution Equations . . . . . . . . . . . . . . . . . . 53
5.2.1 Developed Material Evolution Equations . . . . . . . . 53
5.2.2 Stabilized FEM for Convection . . . . . . . . . . . . . 54
5.2.3 Tests with Three Weighting Methods . . . . . . . . . . 55
5.2.4 Material Time Derivatives of E . . . . . . . . . . . . . 60
5.3 Solution Algorithm . . . . . . . . . . . . . . . . . . . . . . . . 61
5.4 Contact Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.4.1 A Contact Method . . . . . . . . . . . . . . . . . . . . 65
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

6 Test Results 69
6.1 Linear Material Model . . . . . . . . . . . . . . . . . . . . . . 69
6.1.1 Pure Shear Flow . . . . . . . . . . . . . . . . . . . . . 70
6.1.2 Simple Extrusion Case . . . . . . . . . . . . . . . . . . 73
6.2 Nonlinear Material Model . . . . . . . . . . . . . . . . . . . . 78
6.2.1 One-Element Test with HYPEP Model . . . . . . . . 79
6.2.2 Simple Extrusion Test with HYPEP Model . . . . . . 80
6.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
 Contents vii

7 Conclusions and Recommendations 85

7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7.2 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . 87
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

A Proof of Equation ( 3.45) 91

B Voigt Notation 93

C dN, ∇u dN, ∇u ∇u dN 95

D dt, ∇u dt for Impact Problem 97

List of Symbols 99

Acknowledgements 103
Summary

The efficiency and accuracy of many traditional industrial forming processes

may still be improved by application of numerical simulations. Many form-
ing processes such as rolling, extrusion, continuous casting etc., behave as a
steady state process. In recent years special methods have been developed
for numerical simulation of steady processes and new contributions are still
being published.
In this thesis a new displacement based formulation is developed for
elasto-plastic deformations in steady state problems. In this formulation the
displacements are the primary variables, which is in contrast to the more
common formulations in terms of the velocities as the primary variables. In
a steady state process, a transient calculation is not required and only space
discretizations are needed, without time discretizations. The evolution of
the material variables is expressed as an integration along the streamlines.
The resulting differential equation describes steady convection with source
terms.
The thesis is outlined as follows. The second chapter involves the the-
ory of continuum mechanics. Hypo-elasticity and hyper-elasticity are intro-
duced. In Chapter 3 a review of recent literature on the simulation of the
steady state forming processes is presented. The method published by Bala-
gangadhar appears to be most interesting. The development of our method
is based on this work.
In chapter 5 our new method for simulation of steady elasto-plastic processes
is presented. Several bottlenecks in Balagangadhar’s work have been iden-
tified and solutions to them are presented. The formulation is very much
simplified. The convection equation that describes the integration along the
streamlines is transformed to the undeformed configuration. To calculate the
material evolution by the convection equation, the Least Squares method ap-
pears to be the most suitable method. In this way the numerical oscillations,
which are common in other formulations, are avoided almost totally. The
calculation of the material derivatives of the state variables is improved by
using an approximation by means of a continuous field. Also an algorithm
is developed for modelling of the contact problem between tools and work
x Summary

piece, which is consistent with the steady state solution method.

In Chapter 6 the developed method is applied successfully to a pure shear
test and a simple two-dimensional extrusion problem.
Even though our numerical experiments are limited to a simple geometry
and a rigid tool contact, the applicability of our method is demonstrated.
It provides the possibility to improve the steady state flow simulation and
points out a new direction in the field of the modelling of the steady state
processes.
Samenvatting

Bij veel traditionele omvormprocessen kunnen numerieke simulaties bijdra-

gen aan de verbetering van het procesverloop. Nauwkeurigheid en doel-
matigheid kunnen op deze wijze worden verhoogd en de kosten verlaagd.
Bij een aantal processen, zoals walsen, extrusie en continu gieten, is sprake
van een stationair proces. In de afgelopen jaren is de ontwikkeling van al-
goritmen voor simulatie van stationaire processen onderwerp van onderzoek
geweest binnen een selecte kring van onderzoekers. Ook nu nog worden
regelmatig bijdragen op dit gebied gepubliceerd.
In dit proefschrift wordt een nieuwe formulering voor stationaire elasto-
plastische omvormprocessen uitgewerkt. In deze formulering zijn de ver-
plaatsingen de primaire variabelen. Dit is in tegenstelling tot de meer ge-
bruikelijke formulering in termen van snelheden. Omdat er sprake is van
een stationair proces, is een transiënte berekening niet nodig, de tijdsinte-
gratie is vervangen door integratie langs de stroomlijnen. De evolutie van
de toestandsvariabelen langs de stroomlijnen wordt uitgedrukt in een differ-
entiaalvergelijking, die stationaire convectie beschrijft.
In het proefschrift, dat voor U ligt, komen de volgende onderwerpen aan
bod. In het tweede hoofdstuk worden de basis vergelijkingen uit de con-
tinuum mechanica behandeld. Vergelijkingen voor zowel een hypo-elastisch
als een hyper-elastisch materiaalmodel worden afgeleid. Hoofdstuk 3 be-
helst een literatuur onderzoek naar simulatie van stationaire elasto-plastische
omvormprocessen. De methode van Balagangadhar komt hierbij als meest
belovend te voorschijn. De ontwikkeling van onze methode is gebaseerd op
zijn werk.
Onze nieuwe methode voor simulatie van de stationaire elasto-plastic
processen wordt gepresenteerd in hoofdstuk 5. Een aantal knelpunten in
de methode van Balagangadhar zijn geı̈dentificeerd en oplossingen hiervoor
worden aangedragen. De formulering is sterk vereenvoudigd. De convectiev-
ergelijking, die de stroomlijn integratie beschrijft, wordt getransformeerd
naar de onvervormde configuratie. Voor het oplossen van de convectiev-
ergelijking blijkt de kleinste kwadraten methode het meest geschikt. De
numerieke instabiliteiten, die bij veel oplossingsmethoden voor convectiev-
xii Samenvatting

ergelijkingen optreden zijn vrijwel afwezig. De berekening van de materiële

afgeleiden van de toestandsvariabelen is verbeterd door gebruik te maken
van een benadering door middel van continue velden. Ook is voor modeller-
ing van het contact tussen gereedschap en werkstuk een contact-algoritme
ontwikkeld, dat past bij een stationaire oplosmethode.
In hoofdstuk 6 is de ontwikkelde methode met succes toegepast op een
modelprobleem met zuivere afschuiving en op een eenvoudig twee-dimensionaal
extrusieprobleem.
Hoewel onze numerieke experimenten tot een eenvoudige geometrie en
een stijf gereedschapscontact beperkt zijn, wordt de bruikbaarheid van onze
methode aangetoond. Het verstrekt de mogelijkheid om de simulatie van de
stationaire vormgevings processen te verbeteren en wijst een nieuwe richting
op het gebied van de modellering van dit soort processen.
Chapter 1

Introduction

Why do research? History provides the answer

1.1 Industrial Revolution and Development

Beyond all doubt technology increasingly plays a strategic role in the devel-
opment of the economy all over the world. Industry is the key generator of
resources for further economic development. Therefore, without the applica-
tion of inventions and new technologies in industrial life, social and economic
evolution would and will terminate.
Two well-known industrial revolutions started with the utilization of new

Figure 1.1: A model of the Watt engine for pumping

2 Introduction

technologies ([2],[3]). The First Industrial Revolution took place in Great

Britain around 1750. This First Industrial Revolution resulted from the
invention of three types of the machine: the steam engine (see Fig. 1.1),
machines for spinning thread and weaving cloth and furnaces (e.g. to make
ironstone into finished metal). These machines made mass production pos-
sible and put the cottage industry into historic memory.
After the harsh depression of the 1870s, the Second Industrial Revolution
results from innovations in the production of materials (metals etc.), chem-
icals and foodstuffs and the centre of the revolution moved from the UK to
the US and Germany. Three stages were achieved during this revolution.
First, the transportation and communication networks (telegraph, railroad
and cable) were built up; second, the development of electricity offered a
new source of power for industry field; third, science began to be applied
to industrial processes to create the products as demanded. In brief, the
industrial revolutions shaped the modern society into what it is today.
Since the second industrial revolution, with the extensive development of
science and technology, the computer-driven information revolution is exert-
ing a strong influence on all corners of the world. In industry, computers are
playing an important role in industrial design, quality control, business oper-
ation and process modelling. The modelling of metal forming processes aims
to reach a better understanding and optimization of the processes by use of
numerical techniques. A process model requires the combination of different
fields of knowledge, such as mathematics, mechanics, material science and
also computational skills.

1.2 Project Definition

As is well known, many manufacturing processes such as rolling, extrusion,

continuous casting, laser welding etc., behave as steady state processes.
Finite element analysis of these steady state processes has mainly been
done with Lagrangian, Eulerian and Arbitrary Lagrangian-Eulerian(ALE)
descriptions. However, these models have some disadvantages for steady
state problems.
In the Lagrangian formulation the evolutionary nature of the plasticity
can be followed by an incremental analysis. The simulation terminates when
the steady state is reached. But it is time-consuming and the interaction
between the time and space discretizations could lead to numerical oscilla-
tions. In a Eulerian formulation the material flows through a fixed domain in
space. The problems are calculated on the whole concerned domain at once,
instead of using incremental steps as in a Lagrangian formulation. In order
to obtain the material state variables (e.g. the stress or the cumulative plas-
tic strain), integration of the evolution equations must be carried out along
 1.3 Overview of the Thesis 3

the streamlines. These streamlines are not known a priori in the deformed
configuration. Therefore, iterative techniques are required. When a segment
of the boundary is a free surface, iterative steps are needed to adjust these
parts of the boundary in order for it to coincide with the streamlines. A com-
bined Lagrangian and Eulerian formulation called the Arbitary Lagrangian-
Eulerian (ALE) method has been developed to solve the problems which
occur in the two traditional methods described above. As in the Eulerian
formulation, in order to handle history dependent material variables, convec-
tion must be taken into account. As in the Lagrangian formulation, ALE is
still calculated transiently. In conclusion, all three methods cause problems
to some extent during the simulation.
The present research project was proposed as result of a thorough lit-
erature review for steady state process simulation. The challenge was to
develop a displacement based steady state formulation. Indeed, the idea of
the project was suggested by the work of Balagangadhar ([1]). The objective
of this project will be discussed in more detail in Chapter 3.

1.3 Overview of the Thesis

The thesis includes another six further chapters:

Chapter 2 concerns continuum mechanics. It provides the theoretical back-

ground for our work. In this chapter the material models are described
that were used for the simulation. For our case, the available material
models were chosen for different situations during processing. For the
small deformation case, the linear elastic-plastic model can be applied,
but for the large deformation case, the hyperelastic-plastic model was
chosen. The question of when to use hyperelasticity instead of hypoe-
lasticity the answer is investigated in this chapter.

Chapter 3 contains a literature review of the previous work on models for

the simulation of steady state manufacturing processes. Balagangad-
har’s work is the central theme of this chapter.

Chapter 4 discusses the finite element method. In this chapter the lin-
earized theory and finite element discretizations are included.

Chapter 5 includes the displacement based formulation posed on the steady

state nature. The stabilized FEM and the contact analysis for this case
are included in this chapter.

Chapter 6 includes the results for the different situations. The results are
also discussed here.
4 Introduction

Chapter 7 concludes the thesis with the conclusions and recommendations

for future improvements.
At the end of every chapter a bibliography is listed.
Bibliography

[1] Balagangadhar, D., Tortorelli, D.A.: Design of steady state processes

via a displacement based reference frame formulation. In Simulation of
Materials Processing: Theory, Methods and Applications, Huetink and
Baaijens (Eds), NUMIFORM’98, Enschede, Balkema, 1998, 77-83
[2] Chandler, Jr.Alfred D.: The Visible Hand, The Managerial Revolution
in American Business, 1977
[3] Mathias, P.: The First Industrial Nation: An Economic History of
Britain, 2nd ed., 1983
[4] Kitano, K., Morejon, A.: Essay on the industrial revolution: A trip to
past, http://members.aol.com/mhirotsu/essay.htm
Chapter 2

Continuum Mechanics

Basis for finite element analysis

This chapter is a summary of continuum mechanics and it provides the

theoretical support for the finite element (FE) method, which was used for
the modelling work. In the first part of this chapter, some concepts of de-
formation are presented. Further on, the definitions of stress and strain are
described. The conservation equations, also called the balance equations,
are introduced and hence the equilibrium equations that are described in
the different configurations.

2.1 Basic Concepts of Deformation

Figure 2.1: Initial and current configurations of a body

8 Continuum Mechanics

At time t0 we consider a body (see Fig. 2.1) within a domain Ω0 bounded by

Γ0 , which we call the initial configuration. We call the initial configuration
the undeformed configuration. In order to describe the motion and deforma-
tion of the body, a referential configuration is needed to define the motion.
The initial configuration is mostly used as the referential configuration. After
a certain time, the body is moved or deformed to the domain Ω with bound-
ary Γ. This current configuration is called the deformed conf iguration.
The position field of a material point in the referential configuration is ex-
pressed by X, which is called the material or Lagrangian coordinate. Mean-
while, the position vector of a point in the current or deformed configuration
is given by x, which is the spatial or Eulerian coordinate. The relation be-
tween these two positions is described by the motion of the body:
x = φ (X, t) (2.1)
where φ is the mapping function from the referential configuration to the
current configuration at time t. The position x at time t = 0 coincides with
the material coordinates, when the referential configuration coincides with
the initial configuration. There are two descriptions for the deformation and
motion of a continuum. One is called a material or Lagrangian description,
in which the independent variables are the material coordinates. The other
is called a spatial or Eulerian description in which the independent variables
are the spatial coordinates. When the same field is expressed in terms of
different descriptions, such as the Eulerian or Lagrangian descriptions, the
different function symbols are used as follows:
φ(X, t), t)
(X, t) = f(φ (2.2)
The displacement u of a material point can be obtained from the difference
between its original and current positions:
u(X, t) = φ (X, t) − φ (X, 0) = x − X (2.3)
An important notion is the deformation gradient, which is defined as:
φ
∂φ ∂x ∂u
F= = =I+ (2.4)
∂X ∂X ∂X
The determinant J of F is the relative volume. I is the unit tensor.

2.2 Strain and Stress

The concepts of strain and stress are introduced to define the deformation
and motion of the continuum body. In this section the strain and stress mea-
sures are described respectively. Later the frame invariance and objectivity
will be defined.
 2.2 Strain and Stress 9

2.2.1 Strain Measure

The Green (Green-Lagrange) strain E is defined to measure the change in the

squared length of an infinitesimal segment with reference to the material ele-
mental vector dX, while the body deforms from the referential (undeformed)
configuration to the current configuration:

ds2 − dS2 = dx · dx − dX · dX = 2dX · E · dX (2.5)

Using Eq. (2.4), we have:

dx = F · dX (2.6)
Thus:
dx · dx = (FdX)T · (FdX) = dX · (FT · F) · dX (2.7)
Substitute Eq. (2.7) into Eq. (2.5):

dX · (FT · F − I) · dX = 2dX · E · dX (2.8)

Therefore:
1
E = (FT · F − I) (2.9)
2
where the right Cauchy-Green deformation tensor is defined as:

C = FT · F (2.10)

Then the Green Lagrange strain tensor is defined as:

1
E = (C − I) (2.11)
2
This strain tensor is expressed in terms of the displacement gradient by:
1
E = ((∇0 u)T + ∇0 u + ∇0 u · (∇0 u)T ) (2.12)
2
with:
∂u T
∇0 u = ( ) (2.13)
∂X
Alternatively, the same length change can be derived with reference to the
spatial elemental vector dx as:

ds2 − dS2 = 2dx · e · dx (2.14)

where e is called the Eulerian or Almansi strain tensor. This strain tensor
can be expressed in terms of the displacement gradient by:
1
e = ((∇u)T + ∇u − ∇u · (∇u)T ) (2.15)
2
10 Continuum Mechanics

with:
∂u T
∇u = (
) (2.16)
∂x
For the small deformation case, when the initial and final positions of a
material point are practically the same, the quadratic term can be ignored.
These two strain definitions become identical for small deformation.

2.2.2 Stress Measure

In classical continuum mechanics, four stress descriptions are used:

• The Cauchy stress tensor σ;
• The nominal stress P, PT is called the first Piola-Kirchhoff stress tensor;
• The Kirchhoff stress tensor τ ;
• The second Piola-Kirchhoff stress tensor S.

Figure 2.2: Force vectors for stress definitions

The Cauchy stress is defined in the current configuration using the equilib-
rium of the deformable body. The result is that the traction vector t at a
point on the surface dΓ with outward normal n can be expressed in terms
of the Cauchy stress tensor as:
dp = tdΓ = n · σdΓ (2.17)
where dp is the force acting on a deformed element with area dΓ and t = n·σ.
The nominal tensor P can be derived as follows, see Fig. 2.2. We have
([1],[3],[4]):
ndΓ = Jn0 · F−1 dΓ0 (2.18)
 2.2 Strain and Stress 11

We equal dp written in terms of the Cauchy stress and the nominal stress:
dp = Jn0 · F−1 · σdΓ0 = n0 · PT dΓ0 (2.19)
The nominal stress tensor P has the form:
P = Jσ · F−T (2.20)
where P is a non-symmetric tensor and expresses the force in the current
configuration in terms of the area in the initial configuration.
The Kirchhoff stress tensor τ is defined as:
τ = Jσ = P · FT (2.21)
The second Piola-Kirchhoff stress S is obtained by pulling back the spatial
force dp to a material force vector dp 0 :
dp 0 = F−1 · dp = n0 · SdΓ0 (2.22)
Using Eq. (2.19) we have:
S = JF−1 · σF−T (2.23)
Therefore S is symmetric.

2.2.3 Objectivity or Frame Indifference

Objectivity or frame indifference is an important concept in solid mechanics.

The material constitutive relation should be independent of any rigid body
motions. Many quantities describing the material behaviour should be ob-
served as being the same by two different observers in relative rotation and
translation. This is called the principle of material objectivity or material
frame indifference.
Let x̃ result from x by rotation (Q) and translation (l ) of the reference
frame:
x̃ = Q · x + l and Q−1 = QT (2.24)
It follows that:
dx̃ = Q · dx = Q · F · dX (2.25)
Although this vector dx̃ differs from dx, their magnitudes are equal. In this
case the dx is objective under rigid motion.
For Eulerian, Lagrangian and Eulerian-Lagrangian tensors, in general
there are different definitions to judge whether these tensors are objective
or not ([1]). A Eulerian vector a and a Eulerian second order tensor A are
said to be objective if:
ã = Q · a, Ã = Q · A · QT (2.26)
12 Continuum Mechanics

A Lagrangian vector a and a Lagrangian second order tensor A are said to

be invariant if:
ã0 = a0 , Ã0 = A0 (2.27)

A Eulerian-Lagrangian second order tensor A is said to be objective if:

à = Q · A (2.28)

For example, the deformation tensor F is called an objective Eulerian-

Lagrangian second order tensor since:

F̃ = Q · F (2.29)

The time derivative of the second Piola-Kirchhoff stress tensor S is invariant.

But the time derivative of the Cauchy stress tensor σ is clearly non-objective:
T
σ̃˙ = Q̇ · σ · QT + Q · σ̇ · QT + Q · σ · Q̇ (2.30)

In the rate-type constitutive equations the objective rates should be con-

structed by the so-called pull-back push-forward procedure. The Truesdell
stress rate σ̃ 5 is thus defined in terms of the transformation of the time
derivative of the second Piola-Kirchhoff stress as ([3],[4]):

d
σ̃ 5 = J−1 · F[ (J · F−1 · σ · F−T )] · FT (2.31)
dt
This can be worked out as:

σ̃ 5 = σ̇ − L · σ − σ · LT + tr(L)σ (2.32)

The Green-Naghdi stress rate σ̃ 4 is defined when the pull back and push
forward operation is performed using only the rotation tensor R:

d
σ̃ 4 = R · [ (RT · σ · R)] · RT = σ̇ − σ · Ṙ · RT − Ṙ · RT σ (2.33)
dt

If the antisymmetric tensor Ṙ · RT is approximated by the spin tensor W,

the resulting objective stress rate is called the Jaumann stress rate:

σ̃ ◦ = σ̇ + σ · W − W · σ (2.34)

The both above expressions remain objective even when these approxima-
tions do not apply ([3],[4]).
 2.3 Motion and Equilibrium 13

Figure 2.3: Body force and surface force

2.3 Motion and Equilibrium

2.3.1 Cauchy’s Equation of Motion

The equations of motion are the key equations in finite element analysis.
These equations emanate from the momentum conservation principle. Con-
sider a body with the mass density ρ and with a volume Ω and a boundary
Γ as shown in Fig. 2.3, under action of body forces b per unit volume and
surface forces t per unit area. The total force r is then:
Z Z
r= tdΓ + bdΩ (2.35)
Γ Γ

With the help of the definition of Cauchy stress and Gauss’s divergence
theorem, the first part of the right-hand side can be written as:
Z Z Z
tdΓ = n · σdΓ = ∇ · σdΩ (2.36)
Γ Γ Ω

Newton’s second law of motion states that the material derivative of the
linear momentum equals the total force, i.e.:
R
D( Ω ρv dΩ)
=r (2.37)
Dt
According to Reynolds’ transport theorem and the mass conservation theo-
rem: R Z
D( Ω ρv dΩ) d(v )
= ρ dΩ (2.38)
Dt Ω dt
14 Continuum Mechanics

Combine Eq. ( 2.35), Eq. ( 2.36) and Eq. ( 2.37), for an arbitrary volume
Ω, at each point in this volume, the momentum balance is satisfied if:
dv
∇·σ+b=ρ (2.39)
dt
This is called Cauchy’s first law of motion.

2.3.2 Equilibrium Equations

If the body is in equilibrium, the acceleration (the right-hand side of equation

Eq. (2.39) will reduce to zero:

∇·σ+b=0 (2.40)

This equilibrium equations apply to the current deformed configuration.

Above we pointed out the strains and stresses in different reference states.
When the strains and stresses are defined in a reference state rather than the
current state, the equations of motion in the reference state are needed. The
equilibrium equations in the reference configuration are also derived from
the external force in Ω. The body force in the initial volume is obtained
from the body force b in the current configuration with b0 = Jb as :

bdΩ = bJdΩ0 = b0 dΩ0 (2.41)

Using this form, the total force on the body is given in terms of the integral
over the reference volume in the Lagrangian description and it will be equal
to zero: Z Z
b0 (X, t)dΩ0 + t0 (X, t)dΓ0 = 0 (2.42)
Ω0 Γ0

Using n0 · PdΓ = t0 dΓ0 and Gauss’s theorem, the second part of right-hand
side of the equation above has the form:
Z Z Z
t0 dΓ0 = n0 · PdΓ0 = ∇0 · PdΩ0 (2.43)
Ω0 Γ0 Ω0

So the equilibrium equation for the Lagrangian description is obtained in

the first PK stress tensor form:

∇0 · P + b0 = 0 (2.44)

Substituting the definition of the second PK stress tensor into the equation
above, the equilibrium equation for the second PK stress tensor becomes:

∇0 · [F · S] + b0 = 0 (2.45)
 2.4 Constitutive Equations 15

The motion and equilibrium equations together with their boundary condi-
tions are called the strong form.
The finite element method was developed by introducing the weak form
of these equations and boundary conditions. The procedure to derive the
weak form of these equations is found in Chapter 4.

2.4 Constitutive Equations

The constitutive equations represent the mathematical description of mate-
rial macroscopic behaviour resulting from the stress and strain of the ma-
terial. In this section the emphasis is placed on elastic-plastic constitutive
equations. The common forms of the Hypoelastic-plastic and Hyperelastic-
plastic constitutive equations are discussed.
The plastic strain is defined as below. A material shows elastic and plastic
deformation under loading. When the load is removed, the remaining strain
is called the plastic strain. The stress state therefore becomes zero. The mul-
tiplicative decomposition of the deformation gradient F is the mathematical
description for the statement above. An intermediate local configuration is
introduced as a configuration, obtained from the current configuration by
releasing stresses to the zero state or from the original configuration by plas-
tic deformation. Therefore, the deformation of dX is deformed into dxp by
plastic flow:
dxp = Fp · dX (2.46)
where Fp is the plastic deformation gradient tensor. By elastic deformation,
dxp is then deformed into dx in the current configuration as follows:
dx = Fe · dxp (2.47)
Substituting these two equations into Eq. (2.4), the multiplicative decompo-
sition of the deformation gradient tensor is formed as:
F = Fe · Fp (2.48)
The velocity gradient L can be worked out as:
p
L = Ḟ · F−1 = Ḟe · (Fe )−1 + Fe · Ḟ · (Fp )−1 · (Fe )−1 (2.49)
L is split into elastic and plastic parts, i.e.,
L = Le + Lp (2.50)
In this expression, Le is the first term on the right-hand side and Lp is the
second term on the right-hand side
e
Le = Ḟ · (Fe )−1 (2.51)
16 Continuum Mechanics

and:
p
Lp = Fe · Ḟ · (Fp )−1 · (Fe )−1 (2.52)
where Le and Lp can also be split into a symmetric part (De and Dp ) and
an asymmetric part (We and Wp ):

L = De + We + Dp + Wp (2.53)

Therefore:
D = De + Dp (2.54)
e p
W=W +W (2.55)
where D is the deformation rate tensor and W is the spin tensor.

2.4.1 Hypoelastic-plasticity

In hypoelastic-plastic material model a suitable objective rate of stress is a

function of the elastic part of the deformation rate tensor De . In pure elastic
deformation, the stress rate is related to the total deformation-rate tensor:

σ 5 = Cel · De = Cel : (D − Dp ) (2.56)

Energy is not conserved in a close deformation for a hypoelastic material.

But when the elastic deformation is very small, this energy error is insignif-
icant([4]).
The general plastic flow theory includes a plastic flow rule, the yield func-
tion and a consistency condition. The evolution of plastic strains is governed
by the plastic flow rule:
Dp = γ̇m(σ, q) (2.57)
Where γ̇ is a scalar plastic flow rate and m is the plastic flow direction and
is defined as:
∂φ
m= (2.58)
∂σ
where φ is called the plastic flow potential. q is a set of internal variables.
The yield function is:
f (σ, q) = 0 (2.59)
Evolution equations for the internal variables for most forms of plasticity are
specified as:
q̇ = γ̇h(σ, q) (2.60)
For plastic behaviour not only the yield condition must hold, f = 0, but the
plastic consistency condition must also be satisfied. The plastic consistency
condition of Prager:
ḟ = 0 (2.61)
 2.4 Constitutive Equations 17

restricts the stress path to remain on the yield surface.

The plastic rate parameter γ̇ can be calculated from the consistency con-
dition. f is the function of σ and q, and by the chain rule the material time
derivative of f is given as:
∂f ∂f
ḟ = : σ̇ + : q̇ (2.62)
∂σ ∂q

In [4] the first term in the above equation can be rewritten as:

∂f ∂f
: σ̇ = : σ5 (2.63)
∂σ ∂σ
Substitute Eq. (2.56) into Eq. (2.62) becomes:

∂f ∂f
ḟ = : Cel : (D − Dp ) + : q̇ = 0 (2.64)
∂σ ∂q

Using Eq. (2.57) and Eq. (2.60), Eq. (2.64) can be solved for γ̇:
∂f
∂σ : Cel : D
γ̇ = ∂f ∂f
(2.65)
− ∂q ·h+ ∂σ : Cel : m

Substituting the above equation into Eq. (2.56) with Eq. (2.57), a relation
between the objective rate of Cauchy stress and the total rate-of-deformation
tensor is obtained:

σ 5 = Cel : (D − γ̇m) = Ct : D (2.66)

The fourth-order tensor Ct is called the continuum elastic-plastic tangent

modulus and is expressed as:
∂f
(Cel : m) ⊗ ( ∂σ : Cel )
Ct = Cel − ∂f ∂f
(2.67)
− ∂q ·h+ ∂σ : Cel : m

2.4.2 Hyperelastic-plasticity

To avoid the drawback of the hypoelastic-plastic model, hyperelastic-plastic

constitutive models were developed. First, hyperelastic materials are elastic
materials for which the work is independent of load path, and are charac-
terized by the stress, which is obtained from a stored energy function ψ
([3],[5]and [6]):
∂ψ(C, Cp )
S=2 (2.68)
∂C
18 Continuum Mechanics

where ψ is the stored energy potential. There is no need for computing the
stress using the stress rate equations (as in hypoelastic models). In terms of
the variables E, Ep , S can also be described as:

∂ ψ̂(E, Ep )
S= (2.69)
∂E
In hyperelastic-plastic materials a multiplicative decompostion of the defor-
mation gradient is applied:
F = Fe · Fp (2.70)
Cp is defined as:
Cp = FpT · Fp (2.71)
p
E is therefore written as:
1
Ep = (Cp − I) (2.72)
2
The yield condition in the material description is:
φ(E, Ep , q) ≤ 0 (2.73)
The evolution of the internal plastic variable vector q can also be written
as:
q̇ = γ̇H(E, Ep , q) (2.74)
Where H(E, Ep , q) is a prescribed function relating to the hardening or soft-
ening law. The plastic parameter γ̇ is determined by the so-called consistency
condition:
γ̇ φ̇ = 0 (2.75)
The plastic flow rule is obtained by the principle of maximum plastic dissi-
pation (see the proof in [5]):

∂ 2 ψ̂ p ∂φ(E, Ep , q)
: Ė = − γ̇ (2.76)
∂E∂Ep ∂E
We assume:
∂ 2 ψ̂
M= (2.77)
∂E∂Ep
Eq. (2.76) has the form:
p ∂φ(E, Ep , q)
M : Ė = −γ̇ (2.78)
∂E
Sp is defined from time differentiation of the expression for S:
∂S ∂S p p
Ṡ = : Ė + p : Ė := A : Ė − Ṡ (2.79)
∂E ∂E
 2.4 Constitutive Equations 19

Therefore, the plastic flow rule can be written as:

p ∂φ
Ṡ = γ̇ (2.80)
∂E
Eq. (2.73) can be worked out using the chain rule:
∂φ ∂φ ∂φ
φ̇(E, Ep , q) = : Ė + p
p : Ė + : q̇ = 0 (2.81)
∂E ∂E ∂q
From this expression and Eqs. (2.79)and (2.80), γ̇ is obtained:
∂φ
∂E : Ė
γ̇ = ∂φ ∂φ ∂φ
(2.82)
∂Ep : M−1 : ∂E − ∂q :H

By substituting Eq. (2.82) and Eq.( 2.80) into Eq. (2.79) the following ex-
pression is obtained:
∂φ ∂φ
∂E ⊗ ∂E
Ṡ = (A − ∂φ −1 ∂φ ∂φ
) : Ė (2.83)
∂Ep :M : ∂E − ∂q :H

Also, from Eq. (2.69):

∂φ ∂φ
p ⊗
Ė = −M−1 ( ∂φ
∂E
−1
∂E
∂φ ∂φ
)Ė (2.84)
∂Ep :M : ∂E − ∂q :H

Here, we have obtained a similar form as that of the hypoelastic-plastic

material model.
Bibliography

[1] Malveren, L.E.: Introduction to the mechanics of a continuous medium,

Prentice-Hall, New York, 1969
[2] Zienkiewicz, O.C.: The finite element method (Third edition), Mcgraw-
Hill Book Company (UK) Limited, 1979
[3] Bonet, J., Wood, R.D.: Nonlinear continuum mechanics for finite ele-
ment analysis, Cambridge University Press, 1997
[4] Belytschko, T., Liu, W.K., Moran, B.: Nonlinear finite elements for
continua and structures. John Wiley and Sons Ltd., England, 2001
[5] Simo, J.C.: A framework for finite strain elastoplasticity based on max-
imum plastic dissipation and the mulitplicative decomposition: part I.
Continuum Formulation, Computer Methods in Applied Mechanics and
Engineering, Vol 66, 199-219, 1988
[6] Simo, J.C., Hughes, T.J.R.: Computational inelasticity, Springer- Ver-
lag, New York, Inc., 241-261 301-307, 1988
Chapter 3

Literature Review – Steady State

Start from the previous works

In this chapter a detailed review is made of literature on the modelling

of steady state processes.The definition of a steady state is provided ed by
means of mathematics and physics. First the common methods are discussed
using the sources of previous works; Further on a referential steady state
method has been presented by Balagangadhar([20−21]). This background
and basic idea led to our research plan. The detailed project objective and
problem definition have been written as the conclusions of this chapter.

3.1 Definition of a Steady State Process

In many forming processes such as extrusion, drawing, rolling, welding etc.,
the metal flows continuously. During these processes an invariant pattern
of deformation develops (see Fig. 3.1). Hence the spatial distribution of the
major field variables does not change with time during a steady state situa-
tion. The steady state characteristics should be expressed in a mathematical
description so that these processes can be analyzed with numerical methods.
For example, if an arbitrary history-dependent field variable, f , is expressed
by either the referential or the spatial description:
φ(X, t), t) = f(x, t)
f = f̄ (X, t) = f(φ (3.1)
The rate of change of f of a particle, the material time derivative (with the
material coordinates X held constant), can be written in both a material
(Lagrangian) and spatial (Eulerian) description:
Df ∂f
ḟ = = |X Lagrangian or material (3.2)
Dt ∂t
∂f
= |x + v · ∇x f Eulerian or spatial (3.3)
∂t
24 Literature Review – Steady State

Die

Workpiece

Die Flowing out

Flowing in

Figure 3.1: A steady state extrusion process

in which v is the velocity of a particle with material coordinates X:

∂x
v = ẋ = |X (3.4)
∂t
If we know only the expression of f in the spatial description and not under
the material (referential) description, the rate of change of any material
property may also be calculated from the spatial configuration, as long as v
can be calculated.
The first term of the spatial description gives the local rate of change
(local or spatial time derivative) of f. It is the rate of change which is
observed at the fixed position x; the second term gives the convective rate
of change of f of a particle. In a steady flow, the local time derivative is zero
everywhere, but the convective term may be nonzero. The reason is that the
material property of a particle changes as it moves from one place to another
place during flowing. In the steady flow, the streamline pattern is constant
with time and each particle follows one of the unchanging streamlines.
Fig. 3.2 shows the characteristics of the steady state of a rolling process.
When the steady state is reached at time t, the values of f which are measured
at position x are the same from time t to a certain time t0 :
∂f
|x = 0 (3.5)
∂t

3.2 Steady State Formulations

The finite element method has been applied for simulating steady state metal
forming processes during many years using different methods. There are
three main formulations: Lagrangian, Eulerian, and the Arbitrary Lagrangian-
Eulerian (ALE) method.
 3.2 Steady State Formulations 25

Loading region
Flow direction
x= (X,t)

Vt
t

Vt (Eulerian) : A fixed region in space

0
X
Vt
t
x= (X,t)

Steady state: f(x,t)=f(x,t ) f (X,t) = f (X,t)

Figure 3.2: Physical characteristics of a steady rolling process

3.2.1 Lagrangian Formulation

In Lagrangian meshes, the nodes move with the material points. This means
that the mesh motion coincides with the material motion. In solid mechan-
ics this method is used. Since the mesh follows the material points, the
history-dependent materials can be treated naturally ([23]), and free sur-
face movement is automatically calculated. For example, the equations of
conservation of mass in algebraic form are given simply by:
ρJ = ρ0 (3.6)
In which ρ represents the density of a material point and J is the Jacobian.
There are two Lagrangian methods: the Updated Lagrangian(UL) and Total
Lagrangian (TL) formulation. In UL, the strong form ([23]) is expressed in
spatial coordinates, but in TL it is expressed in material, i.e. Lagrangian
coordinates.
Use of the FE Lagrangian formulation started from the 1970s ([1]). The
analysis is transient to calculate the evolution of the material state variables
and terminates when the steady state is reached([2]−[5]). A large number of
increments is needed to complete the whole process and the mesh includes a
large upstream area so that the steady state situation can be reached([6],[7]).
These disadvantages lead to increasing CPU time and computer storage. In
addition, for history-dependent material, a time discretization is required
in order to integrate an equation like Eq. (3.2). The interaction between
the time and space discretizations can lead to numerical oscillations during
26 Literature Review – Steady State

calculations ([1]). Due to the connection between the material points and the
nodal points, the meshes follow the deformation of the material completely
during the calculation, and become extremely distorted in the case of the
large deformation.

3.2.2 Eulerian Formulation

In Eulerian meshes, the nodes and elements are fixed in space, so that the
material flows through a fixed mesh ([8]∼[15]). This method is used often
to simulate fluid mechanics problems, because the interest is normally con-
centrated on the spatial domain where the gradients in the flow patterns are
large ([13]). The state variables are a function of the spatial coordinates and
time, while the undeformed configuration does not exist. For example, the
equation of conservation of mass has to be written as a partial differential
equation rather than in algebraic form as in the Lagrangian formulation Eq.
(3.6):
∂ρ ∂(ρv )
+ =0 (3.7)
∂t ∂x
Similarly, the material derivatives of any state variable have the form as in
Eq. (3.3).
In most cases the mesh is constructed according to the deformed shape.
The boundary conditions are imposed on fixed mesh nodes. Hence, the
problems are calculated on the whole domain at once, instead of using in-
cremental steps as in a Lagrangian formulation. In order to obtain the
material state variables (e.g. the stress or the cumulative plastic strain), in-
tegration of the evolution equations must be carried out along the unknown
streamlines. Therefore, iterative techniques are required. One of the prob-
lems in this method is the free surface treatment, since the mesh has to be
adjusted to make the surface a streamline ([11], [14] and [15]). In solid prob-
lems, Eulerian formulations have mostly been used for solving steady-state
processes with rigid plastic material models.

3.2.3 ALE Formulation

During the last two decades a combined Lagrangian and Eulerian formu-
lation called the Arbitrary Lagrangian-Eulerian (ALE) method has been
developed to combine the advantages and solve the problems which occur in
the above two traditional methods ([23]).
In an ALE method, the motion of the material is followed in the same
way as Eq. (3.1). Another referential domain Vm is used to define the mo-
tion of the mesh. Assuming that χ are the ALE coordinates relating to the
motion of the mesh, the mapping among three configurations is drawn in
 3.2 Steady State Formulations 27

Figure 3.3: Mapping among Lagrangian, Eulerian and ALE

Fig. 3.3. The motion of the mesh is expressed by:

φ(χ, t)
x = φ̄ (3.8)

In ALE the material motion can be obtained from the mesh motion using
mapping, since:

φ(χ, t) = φ̄
x = φ (X, t) = φ̄ φ(ΦΦ(X, t), t) (3.9)

with:
φ(x, t)
χ = Φ (X, t) = φ̃ (3.10)
The mesh displacement ū is defined by:

ū = x − χ (3.11)

The material derivative can be represented as a function of ALE coordinates

χ and time t in a similar process as that used in a Eulerian coordinate:

˘˙f =˘˙f (χ, t) = ∂ f̆ (χ, t) |χ + ∂ f̆ ∂χ |X (3.12)

∂t ∂χ ∂t
Also, a chain rule expression can be developed for the material velocity:
φ
∂φ φ
∂φ̄ φ(χ, t) ∂χ
∂φ̄ φ(χ, t) ∂χ
∂φ̄
vl = |X = |χ + = ve + |X (3.13)
∂t ∂t ∂χ ∂t ∂χ ∂t
where vl and ve represent the material velocity and the mesh velocity, re-
spectively. The referential velocity w is defined as:
∂x
v l = ve + w (3.14)
∂χ
28 Literature Review – Steady State

Therefore we have:
∂Φ(X, t)
w= |X (3.15)
∂t
Substituting above two equations into Eq. ( 3.12):

Df ∂ f̆ (χ, t) ∂ f̆
= |χ + w (3.16)
Dt ∂t ∂χ
∂ f̆ (χ, t) ∂ f̆ ∂ f̆ (χ, t) ∂ f̆
= |χ + (vl − ve ) = |χ + c (3.17)
∂t ∂x ∂t ∂x
with:
c = vl − ve (3.18)
In which the difference between the material and mesh velocities, c, is called
the convective velocity.
In a Eulerian formulation (χ = x):

ve = 0 c = vl w=c (3.19)

In an updated Lagrangian formulation(χ = X):

ve = vl c=w=0 (3.20)

The equations of mass conservation Eq. (3.6) can be described in the ALE
configuration via a transformation and will have the form:

ρ̇ + ρ∇x · vl = 0 (3.21)

Substituting Eq. (3.15) into the equations above:

∂ ρ̆ ∂ ρ̆x
|χ + w + ρ∇x · vl = 0 (3.22)
∂t ∂χ

In this formulation the reference frame is associated neither with the mate-
rial displacement as in the Lagrangian formulation nor fixed in space as in
the Eulerian formulation. In this method the transient analysis is still re-
quired. At the same time, the free surface problem can be solved too. Since
in ALE meshes the element coordinates are free from the material configu-
ration, mesh distortion in the Lagrangian configuration can be prevented at
a certain stage. However, handling the convective term could lead to spatial
oscillations due to time discretizations and therefore the suitable discretiza-
tion method must be taken into account. ALE has been successfully used for
the simulation of steady state forming processes by a transient calculation
until the processes are stationary([16]∼[19]).
 3.3 Referential Formulation (Balagangadhar) 29

3.3 Referential Formulation (Balagangadhar)

In 1998, D. Balagangadhar([20] and [21]) proposed his reference frame formu-

lation for the simulation of steady state forming processes. His method con-
centrates on finding a referential coordinate by use of a mapping technique
to eliminate the time effect in the convective equation (evolution equation).
In this section we analyse his work.

3.3.1 Construction of a Reference Configuration

Figure 3.4: Relation between undeformed and deformed configurations

In Fig. 3.4, Vt represents the deformation region. At time t, Vt is the

steady state deformed volume of a region V0 in the undeformed configuration.
Therefore:
Vt = φ (V0 , t) (3.23)
Balagangadhar considers Vt as a fixed region in space, therefore V0 changes
with time:
V0 = φ −1 (Vt , t) (3.24)
A fixed reference configuration Vr is introduced (see Fig. 3.5). At any time
t, V0 is defined to be related to Vr by the mapping g:

V0 = g(Vr , t) (3.25)

Hence, position vectors X in V0 can be described by r in Vr as:

X = g(r, t) (3.26)

Any state variables f can be expressed as f̄ (X, t) on V0 and f̃ (r, t) on Vr

with equal values on both these two configurations at corresponding points:

f̃ (r, t) = f̄ (X, t) = f̄ (g(r, t), t) (3.27)

30 Literature Review – Steady State

Figure 3.5: Relation among undeformed, deformed and referential configurations

The gradients and partial derivatives for the fields defined over both unde-
formed and reference configurations are defined as:

∂ f̄
∇X f̄ (X, t) = (X, t) (3.28)
∂X
∂ f̃
∇r f̃ (r, t) = (r, t) (3.29)
∂r
V0 and Vr are related to each other by:

X = r + vt (3.30)

v is defined as the uniform inflow velocity and in fact V0 is defined as the

control volume of Vr when Vr moves to V0 at time t with velocity v.

3.3.2 Steady State at the Reference Configuration

According to the definition in Eq. (3.5), the steady state is reached when
the local rate changes of parameters f become zero at the fixed position x
in Eulerian coordinates (see Fig. 3.6), i.e.:

f(x, t) = f(x, t0 ) (3.31)

since the fixed position x can be described at any time (e.g. t and t0 in the
undeformed configuration:

x = φ(X, t) = φ(X0 , t0 ) (3.32)

with:
φ(X, t), t) = f̄ (X, t)
f(x, t) = f(φ (3.33)
 3.3 Referential Formulation (Balagangadhar) 31

Figure 3.6: Physical explanation of steady state under undeformed and deformed con-
figuration

f(x, t0 ) = f(φ
φ(X0 , t0 ), t0 ) = f̄ (X0 , t0 ) (3.34)
therefore, for steady state at any time:
f̄ (X, t) = f̄ (X0 , t0 ) (3.35)
Meanwhile the transformation can be made between the undeformed config-
uration V0 and the referential configuration Vr by help of Eq. ( 3.26) and
X0 = g(r, t0 ):
f̄ (X, t) = f̄ (g(r, t), t) = f̃ (r, t) (3.36)
and:
f̄ (X0 , t0 ) = f̄ (g(r, t0 ), t0 ) = f̃ (r, t0 ) (3.37)
From Eq. ( 3.36), the steady state situation can be expressed for the refer-
ential configuration for any time (see Fig. 3.7):
f̃ (r, t) = f̃ (r, t0 ) (3.38)
i.e.:
∂ f̃
(r, t) = 0 (3.39)
∂t
32 Literature Review – Steady State

Figure 3.7: Physical explanation of steady state under reference configuration

When the process reaches steady state, the values of the state variables of
points with the same fixed position r will be equal for any time.
Because of Eq. ( 3.26) and Eq. ( 3.36), for steady state there we have:
∂ f̃ (r, t) ∂ f̄ (X, t) ∂ f̄ (X, t) ∂g(r, t)
|r = |X + |r = 0 (3.40)
∂t ∂t ∂X ∂t
or:
¯˙f = ∂ f̄ (X, t) |X = ∂ f̃ (r, t) |r + ∂ f̃ (r, t) ∂r |X (3.41)
∂t ∂t ∂r ∂t
Using the chain rule:
∂ f̄ (X, t) ∂ f̃ (r, t) ∂r ∂ f̃ (r, t)
= = Fr (3.42)
∂X ∂r ∂X ∂r
where Fr is the transformation tensor between the undeformed configuration
and the reference configuration. Eq. ( 3.40) can therefore be rewritten as:
∂ f̄ (X, t) ∂ f̃ (r, t) ∂r ∂g(r, t)
|X = − |r (3.43)
∂t ∂r ∂X ∂t
 3.3 Referential Formulation (Balagangadhar) 33

Eq. ( 3.41) gives the form in steady state because of Eq. ( 3.39):

¯˙f = ∂ f̄ (X, t) |X = ∂ f̃ (r, t) ∂r |X (3.44)

∂t ∂r ∂t
This gives the same result as in Eq. ( 3.43), due to:

∂r ∂r ∂g(r, t)
|X + |r = 0 (3.45)
∂t ∂X ∂t
The proof can be found in Appendix A.
The expression Eq. ( 3.43) shows that in steady state the material deriv-
ative only depends on the gradient of f under the reference configuration r,
Fr and ∂g(r,t)
∂t (i.e. ġ|r = v). Since Fr is defined on the basis of the trans-
formation between the undeformed field and the reference configuration, the
material derivative can be obtained via the gradient of the reference fields.
In conclusion, the reference fields do not vary in time in the reference config-
uration. This result can be used to obtain the material derivatives without
applying time differentiation.

3.3.3 Contact Problem

Balagangadhar made several assumptions:

1. The tool is rigid.

2. The contact surface is smooth.

3. The contact surface ∂Vr c in the reference configuration is treated as a

priori.

4. The Coulomb friction model is applied.

5. The normal traction over the boundary of the contact area is zero (an
equality contact constraint, unlike in Lagrangian analysis an inequality
constraint).

The last item aims to obtain the contact area in the undeformed configu-
ration because the location of the contact area is unknown. An additional
response field α has to be introduced in the relation between the initial and
referential configurations. The mapping X = g(r, t, α) is defined to divide
several flexible subregions in order to satisfy the 5th assumption.
It is found that Fr is no longer a unit tensor as in the non-contact sit-
uation. But it still only depends on the element lengths of the undeformed
and reference configurations.
34 Literature Review – Steady State

3.3.4 Solution Procedure

The momentum balance equations represented by the second PK stress ten-

sor are:
∇0 · (FS) + b = 0 (3.46)
The momentum balance equations can be transformed from these equations:
∇r · (JFSFr ) + Jb = 0 (3.47)
Furthermore, the evolution equations in the reference configuration, using
the consistency conditions, can be described as:
Ċp = Θ(Ċ, C, Q) (3.48)
in which Q represents other variables.
For steady state, due to Eq. ( 3.42), the material derivatives are:
p
p ∂ C̃ (r, t) ∂g
Ċ = − Fr (r, t) (3.49)
∂r ∂t
Using the integration technique, Cp can obtained directly without time in-
tegration. The streamline Upwind Petrov-Galerkin method is applied for re-
ducing the numerical instabilities when solving the material evolution equa-
tions.
Balagangadhar used the coupled methods to solve the above three equa-
tions using FEM. The stress can be calculated with the hyperelastic model
with linear isotropic hardening, since this model provides a direct relation
between stress and strain.

3.3.5 Remarks

Some remarks are made concerning the work of Balagangadhar:

1. A transient analysis is not required, as the reference fields do not vary
in time in the steady state situation. The displacements and plastic
strains are the primary variables.
2. Elastic spring-back can be obtained directly because the plastic parame-
ter can be calculated from its rate form.
3. The material evolution equations in the reference configuration are simi-
lar to those in the Eulerian configuration. Integration of these equations
does not require time integration, since these equations are dependent
only of inflow velocity and the undeformed contact region. Therefore
the integration of the material evolution equations was carried out along
the known streamlines.
 3.4 Objectives of This Project 35

4. The contact problem cannot be treated as in the conventional La-

grangian, Eulerian and ALE formulations.
5. A mathematical coordinate transformation is required to obtain the
reference configuration, which makes the procedure complicated and
more difficult to understand.

3.4 Objectives of This Project

Balagangadhar’s work opens a new field for modelling steady state flow prob-
lems. The objectives of our project were:
1. To find a simple definition of the evolution equations based on the
undeformed configuration. Balagangadhar’s referential configuration
was too complicated for use in practical situations.
2. To use a different solution methodology and numerical techniques.
3. To obtain more insight into steady state process modelling.
Bibliography

[1] Lee, E.H., Mallett, R.L., and Yang, W.H.: Stress and deformation
analysis of the metal extrusion processes. Computer Methods in Ap-
plied Mechanics Engineering. Vol. 10, 1977, 339-353
[2] Appleby, E., Lu, Y., Rao, R.S., Devenpeck, M.L., Wright, P.K., and
Richmond, O.: Strip drawing: A theoretical-experimental comparison.
International Journal of Mechanical Sciences. Vol. 26, No.5, 1984, 351-
362
[3] Liu, C., Hartley, P., Sturgess, C., Rowe, W.: Elastic-plastic finite-
element modelling of cold rolling of strip. International Journal of Me-
chanical Sciences. Vol. 27, No.7/8, 1985, 531-541
[4] Carroll III, J.T., Strenkowski, J.S.: Finite element models of orthogonal
cutting with application to single point diamond turning. International
Journal of Mechanical Sciences. Vol.30, No.12, 1988, 899-920
[5] Suh, Y.S., Agah-Tehrani, A., Chung, K.: Stress analysis of axisym-
metric extrusion in the presence of strain-induced anisotropy models as
combined isotropic-kinematic hardening. Computer Methods in Applied
Mechanics and Engineering. Vol. 93, 1991, 127-150
[6] Rowe, G.W., Sturgess, C.E.N., Hartley, P., Pillinger, I.: Finite-element
plasticity and metalforming analysis. Cambridge University Press, 1991
[7] Gouveia, B.P.P.A., Rodrigues, J.M.C., and Martins, P.A.F.: Steady-
state finite element analysis of cold forward extrusion. Journal Material
Processing Technology. Vol. 73, 1998, 281-288
[8] Dawson, P., R., Thompson, E., G.: Finite element analysis of steady-
state elasto-visco-plastic flow by the initial stress-rate method, Inter-
national Journal for Numerical Methods in Engineering. Vol.12, 1978,
47-57
[9] Abo-elkhier, M., Oravas, G.A., and Dokainish, M.A.: A consistent
Eulerian formulation for large deformation analysis with reference to
metal-extrusion process. International Journal of Non-Linear Mechan-
ics. Vol 23, No.1 1988, 37-52
[10] Yu,S.W. , Thompson, E.: A direct Eulerian finite element method for
steady state elastic plastic flow. NUMIFORM 89, Thompson et al. (eds),
38 Bibliography

Balkema, Rotterdam, 1989,95-103

[11] Thompson, E.G., Yu, S.W.: A flow formulation for rate equilibrium
equations. International Journal for Numerical Methods in Engineering.
Vol. 30, 1990, 1619-1632
[12] Maniatty, A., M., Dawson, P., R., Weber, G., G.: An Eulerian elastic-
viscoplastic formulation for steady-state forming processes, Interna-
tional Journal of Mechanical Sciences. Vol. 33, No.5, 1991, pp.361-377
[13] Legat, V., Marchal, J. M.: Prediction of three-dimensional general
shape extrudates by an implicit iterative scheme. International Jour-
nal for Numerical Methods in Fluids. Vol. 14, No.1, 1992, 609-625
[14] Lefebvre, L., Keunings, R.: Finite element modelling of the flow of
chemically reactive polymeric liquids. IInternational Journal for Nu-
merical Methods in Fluids. Vol. 20, 1995, 319-334
[15] Ramanan, N., Engelman, M.S.: An algorithm for simulation of steady
free surface flows. International Journal for Numerical Methods in Flu-
ids. Vol. 22, 1996, 103-120
[16] Malinowski, Z., Lenard, J., G.: Experimental substantiation of an
elastoplastic finite element scheme for flat rolling, omputer Methods
in Applied Mechanics Engineering. 1993, 1-17
[17] Hu, Y.K. , Liu., W.K.: An ALE hydrodynamic lubrication finite ele-
ment method with application to strip rolling. International Journal for
Numerical Methods in Engineering. Vol. 36, 1993, 855-880
[18] Huétink, J.: On the simulation of thermo-mechanical forming processes
- A mixed Eulerian-Lagrangian finite element method, Ph.D thesis, Uni-
versity of Twente, Enschede, The Netherlands, 1986
[19] Wisselink, H.: Analysis of guillotining and slitting - finite element simu-
lations, PhD thesis, University of Twente, Enschede, The Netherlands,
2000, 33-48
[20] Balagangadhar, D., Tortorelli,D.A. Design of large-deformation steady
elastoplastic manufacturing processes. Part I: a displacement-based ref-
erence frame formulation. International Journal for Numerical Methods
in Engineering. Vol. 49, 2000, 899-932
[21] Balagangadhar, D., Dorai, G.A., Tortorelli, D.A.: A displacement-based
reference frame formulation for steady-state thermo-elasto-plastic mate-
rial processes. International Journal of Solids Structures. Vol. 36, 1999,
2397-2416
[22] Malvern, L., E.,: Introduction to the mechanics of a continuous medium,
Prentice-Hall, New York, 1969
[23] Belytschko, T., Liu, W.K., Moran, B.: Nonlinear finite elements for
continua and structures. John Wiley and Sons Ltd., England, 2001
Chapter 4

Finite Element Method

Stepping stone to structure modelling

Some simulations of the mechanical processes have already been carried out
based on the finite element method. This method aims to find the numerical
solutions for the motion equation together with the boundary conditions.
The essence of this method is to obtain the weak form of the momentum
equations (strong form) with the help of the principle of virtual work. In
this chapter, the relationship between the weak form and the strong form is
discussed. Secondly, the finite element method procedure is introduced by
the linearization, FE discretization and iterative solution procedure, step by
step.

4.1 Principle of Virtual Work

Assume a body is in a certain equilibrium state with the boundary condi-
tions:

u = ū on Γu (4.1)
t = t̄ on Γt (4.2)

and give every point of the body an infinitesimal virtual displacement δui
from the equilibrium configuration. These virtual displacement (or velocity)
functions are called the test functions. The space of these functions are
defined by:

δui ∈ Π Π = {δui |δui ∈ C 0 , δui = 0 on Γu } (4.3)

A function is C n continuous, if the nth derivative of this function is a con-

tinuous function.
40 Finite Element Method

The virtual work of the static spatial equilibrium equations Eq. ( 2.40)
of one point in the body is represented as follows:

δw = δu · (∇ · σ + b) = 0 (4.4)

We integrate over the volume of the body to obtain the virtual work of the
body in the equilibrium state:
Z
δW = δu · (∇ · σ + b)dΩ = 0 (4.5)
Ω

Applying the divergence theorem:

∇ · (δu · σ) = σ : (∇δu) + δu · (∇ · σ) (4.6)

Eq. ( 4.5) has the form:

Z
δW = [∇ · (δu · σ) − σ : (∇δu + δu · b)]dΩ = 0 (4.7)
Ω

Together with Gauss theorem the equation above can be rewritten as:
Z Z Z
δW = δuσ · ndΓ − ∇δu : σdΩ + δu · bdΩ = 0 (4.8)
Γt Ω Ω

in which:
∇δu = δε + δω (4.9)
with the virtual strain and rotation tensor, respectively:
1
δε = (∇u + (∇u)T ) (4.10)
2
and:
1
δω = (∇u − (∇u)T ) (4.11)
2
Because δω is skew-symmetric and σ is symmetric:

∇δu : σ = δε : σ + δω : σ = δε : σ (4.12)

With t = σ · n, finally, the spatial virtual work equation is expressed as:

Z Z Z
δW = − δε : σdΩ + δu · tdΓ + δu · bdΩ = 0 (4.13)
Ω Γt Ω

The proof that the weak form implies the strong form can be found in the
literature ([1],[8]).
The virtual work equations in the reference states can be obtained in
 4.1 Principle of Virtual Work 41

terms of the different material stress tensors, which are introduced in Chap-
ter 2.
Following a similar derivation, the weak form of the equilibrium equation
based on the first PK stress tensor can be written as:
Z Z Z
δW = − ∇0 δu : PdΩ0 + δu · t0 dΓ0 + δu · b0 dΩ0 = 0 (4.14)
Ω0 Γt Ω0

because u = x − X and therefore δu = δx. Hence ([8]):

∇0 δu : P = δFT : P (4.15)

Eq. ( 4.14) is rewritten as:

Z Z Z
δFT : PdΩ0 − δu · t0 dΓ0 − δu · b0 dΩ0 = 0 (4.16)
Ω0 Γt Ω0

The weak form in terms of the second PK stress tensor can be obtained from
the equation above. The second and third terms are similar, and only the
first term need be worked out. Using Eq. ( 2.23) this term becomes:
Z
δFT : (F · S)dΩ0 (4.17)
Ω0

because:
tr(δFT · F · S) = tr(FT · δF · ST ) (4.18)
With ST = S we have:
1
tr(δFT · F · S) = tr(FT · δF · S) = (δFT · F + FT · δF) : S (4.19)
2
Then:
1
δFT : (F · S) = (δFT · F + FT · δF) : S (4.20)
2
Taking the virtual rate of the right Cauchy-Green tensor C = FT F:

δC = δFT · F + FT · δF (4.21)

Therefore, Z Z
1
∇0 δu : (F · S)dΩ0 = δC : SdΩ0 (4.22)
Ω0 Ω0 2
The weak form of the equilibrium equations in terms of the second PK stress
tensors has the form:
Z Z Z
δW = − δE : SdΩ0 + δu · t0 dΓ0 + δu · b0 dΩ0 = 0 (4.23)
Ω0 Γt Ω0
42 Finite Element Method

4.2 Linearization of the Weak Form

In general, the virtual work expression of the equilibrium equations is non-

linear with respect to the material and the geometry. In order to solve
these nonlinear equations in the finite element method, a Newton-Raphson
iterative solution is used. Hence, the linearization of the weak form of the
equilibrium equations is required. In this part the linearization procedure is
described with respect to the material description of the equilibrium equa-
tions Eq. ( 4.23).
First here the directional derivative is introduced. The directional
Q deriv-
ative represents the gradient of the potential energy function in the di-
rection of an incremental displacement u:
Y Y Y
D (x)[u] ≈ (x + u) − (x) (4.24)
Q
where D (x)[u] gives a linear approximation to the increment in the po-
tential energy due to the increment in position u.
Let us consider the solution of a set in the nonlinear weak form,

Figure 4.1: Linearized equilibrium

ϕ, δu) = 0
δW (ϕ (4.25)

where ϕ is the solution of the equilibrium state. Consider a trial solution

ϕ t , the above equations can be linearized if the equilibrium state is satisfied.
Then:

ϕt + u, δu) ≈ δW (ϕ
δW (ϕ ϕt , δu) + DδW (ϕ
ϕt , δu)[u] = 0 (4.26)
 4.2 Linearization of the Weak Form 43

The virtual displacement δu does not alter during the incremental change u
(see Fig. 4.1).
The linearization of the virtual work can be split into two parts: the
internal and external virtual work components according to Eq. ( 4.23):
ϕt , δu)[u] = DδWin (ϕ
DδW (ϕ ϕt , δu)[u] − DδWex (ϕ
ϕt , δu)[u] (4.27)
where: Z
δWin = δE : SdΩ0 (4.28)
Ω0
and: Z Z
δWex = δu · bdΩ + δu · tdΓ (4.29)
Ω0 Γ0
First the directional derivative of the internal virtual work is obtained:
Z
DδWin [u] = D( (δE : S)[u]dΩ0 )
Ω0
Z Z (4.30)
= δE : DS[u]dΩ0 + S : DδE[u]dΩ0
Ω0 Ω0

Because of the constitution relation between the stresses and the strains:
∂S
DS[u] = : DE[u] = Ce : DE[u] (4.31)
∂E
where Ce is called the tangent modulus tensor.
The term DδE[u] can be worked out from:
1
δE = (δFT · F + FT · δF) (4.32)
2
with:
δF = (∇0 δu)T (4.33)
and:
DF[u] = (∇0 u)T (4.34)
Because ∇0 δu is constant as it is independent both of configuration and the
symmetry of S, the result can be expressed as:
1
S : DδE[u] = S : [∇0 δu · (∇0 u)T + ∇0 u · (∇0 δu)T ]
2 (4.35)
= S : [(∇0 u)T · ∇0 δu]
Substituting Eq. ( 4.35) into Eq. ( 4.30), the directional derivative of the
internal virtual work has the form:
Z Z
DδWin = δE : Ce : DE[u]dΩ0 + S : [(∇0 u)T ∇0 δu]dΩ0 (4.36)
Ω0 Ω0
44 Finite Element Method

The body force and the surface force contribute to the external forces. The
directional derivatives of both external virtual work terms are considered
separately. One of the most common body forces is self-weight or gravity
loading with b = ρ0 g in the initial state. The virtual work of the body force
is therefore given as:
Z
ϕ, δu) =
δWex (ϕ ρ0 g · δudΩ (4.37)
Ω0

Because the terms in the above equation are independent of the current state
and by definition Dδu[u] = 0, then:
ϕ, δu) = 0
DδWex (ϕ (4.38)

The friction contact is the most common cause of surface traction, and its
linearization is complex. In this section it will not be discussed further and
the surface traction is assumed to be zero. The issue about the contact force
will be discussed in Chapter 5, where the contact analysis is discussed for
this case.
It is found that the equilibrium equations and their corresponding lin-
earizations have the integration forms. As a consequence, the finite ele-
ment method has been developed to deal with these problems. Besides
linearization of the weak form as one of the main steps in the finite element
method, there are three important procedures in this method: finite element
discretization, approximation to integral formulations (weighted residuals),
and the iteration solution procedure (Newton-Raphson method). In the next
section, finite element discretization is discussed together with the Galerkin
weighted method. Furthermore, after having obtained the discretized equi-
librium equations, the solution of these equations was carried out via itera-
tive methods, for example, Newton-Raphson method.

4.3 Finite Element Discretization

In the finite element method the domain Ω is divided into subdomains Ωe ,
which are called elements, so that:

Ω = ∪Ωe (4.39)

These elements are defined by means of nodes. The independent variables

are calculated in these nodes and they reprensent the degrees of freedom
(dof) system. The node coordinates in the current configuration can be
denoted by xiI , where I is the nodal number and i is the degree of free-
dom, such that xiI = [xI , yI ]. The nodal coordinates in the initial state
are denoted by XiI . The discretization is established using the elements to
 4.3 Finite Element Discretization 45

interpolate the quantities or geometry in terms of the nodal values in the

initial configuration:
Xi = NI (ξ)XiI or X = NI (ξ)XI (4.40)
In the following discussion the second representation is chosen for simplicity.
Similarly, the motion can be fully described by means of the current position
xI of the nodal particles:
x = NI xI (4.41)
where N are the interpolation(shape) functions and ξ represents the natural
coordinates. For the same procedure, the displacement u and the gradient
of it also can be interpolated as:
u = NI uI (4.42)
∇0 u = uI ⊗ ∇0 NI (4.43)
For the interpolations of different qualities, the shape functions can be chosen
the same or different, depending on the choices of the element types. In most
situations the quantities are interpolated by the same shape function. In the
Galerkin weighted method the weighting function is chosen to be the same
as the shape function, i.e.:
δu = NI δuI (4.44)
The linearized integral equation over the internal virtual work Eq. ( 4.36) can
be discretized as follows. We recall Eq. ( 4.36), and start the discretization
in one element:
Z Z
e
DδWin = δE : Ce : DE[u]dΩe0 + S : [(∇0 u)T · ∇0 δu]dΩe0 (4.45)
Ωe0 Ωe0

It is known that the Green strain tensor E is a function of u. We have:

δE = δuI ⊗ B0I (∇0 N, u) (4.46)
DE = uJ ⊗ B0J (∇0 N, u) (4.47)
where δE represents the values in the integration points of one element,
and B0I is called B matrix relating the strains with the displacements .
After some tensor calculations and a transformation between tensor form
and matrix form, the right-hand part of the first term of Eq. ( 4.45) is
expressed in the matrix form ([7]) as :
[δuI ]T Ke mat [uI ] (4.48)
with: Z
Kemat
IJ = BT0I : Ce : B0J (∇0 N )dΩe0 (4.49)
Ωe0
46 Finite Element Method

This stiffness matrix is called the material tangent stiffness matrix of one
element because this term is associated with the rate of stress and then
depends on the material response.
Consider now the second term of Eq. ( 4.45) in one element. We have:
∇0 δu = δuI ⊗ ∇0 NI (4.50)
∇0 u = uJ ⊗ ∇0 NJ (4.51)
Here we take the same shape function for interpolating u and δu in this case.
Then:
Z Z
T e
S : [(∇0 u) · ∇0 δu]dΩ0 = S : [(δuI · uJ )∇0 N ⊗ ∇0 N ]dΩe0 (4.52)
Ωe0 Ωe0

Eq. ( 4.52) is rewritten as:

Z
(δuI · uJ ) [∇0 NI · S · ∇0 NJ ]dΩe0 (4.53)
Ωe0

Because the integral part is observed as the scalar and δu · u = δu · I · u with

the unit tensor I, the second term can also be worked out in matrix form
δu · Ke geo · u with ([7]):
Z
e
K geoIJ = [∇0 NI · S∇0 NJ ]IdΩe0 (4.54)
Ωe0

This term includes the current state of the stress and also illustrates that the
deformation affects the geometry. For this reason the matrix above is called
geometric stiffness matrix. Concluding for the discrete directional derivative
of the weak form in one element, which account for the internal and external
virtual works,
DδW e = −δuTI · (Ke matIJ + Ke geoIJ ) · uJ + DδWex e (= 0) (4.55)
KeIJ = Ke matIJ + Ke geoIJ (4.56)
The discrete directional derivative of the weak form in the total domain can
be obtained as:
X X
DδW = DδW e = δuI · (Ke matIJ + Ke geoIJ ) · uJ = δuI · KIJ · uJ
(4.57)
Furthermore, the virtual work of the equilibrium equations can also be dis-
cretized in a way similar to that above:
Z Z Z
e e e
δW = δE : SdΩ0 − δu · b0 dΩ0 − δu · tdΓe0
e e e
Ω0 Ω0 Γ0
Z Z Z (4.58)
e e e
= δuI ⊗ B0 : SdΩ0 − N δuI · b0 dΩ0 − N δuI · tdΓ0
Ωe0 Ωe0 Γe0
 4.4 Iteration Procedure 47

Rearranging this equation in matrix form:

Z Z Z
e T T e e
δW = δuI · ( B0 · SdΩ0 − N f0 dΩ0 − N tdΓe0 ) (4.59)
e e e
Ω 0 Ω 0 Γ 0

In terms of the nodal internal force TeI and the external force FeI in one
element, the equation above can be rewritten as:
δW e = δuI · (TeI − FeI ) (4.60)
where: Z
TeI = B0 T · SdΩ0 (4.61)
Ω0
Z Z
FeI = N f0 dΩ0 + N tdΓ0 (4.62)
Ω0 Γ0
Finally the virtual work in the finite element mesh is:
X X
δW = δW e = δuI · (TI − FI ) = δuI · RI (4.63)
When the body is in the equilibrium state, the nodal residual forces become
zero.

4.4 Iteration Procedure

Recalling Eq. ( 4.26), there is a Newton-Raphson equation:
δW + DδW = 0 (4.64)
with the discretized form as:
δuT · (K · u + R) = 0 (4.65)
Therefore:
K · u = −R = F − T (4.66)
F is the external force. The iterative procedure is as follows for each step:
1 Initialize F = F0 ,T = 0,i = 0, u = 0 and set tolerance errors (Tor).
2 IF (kRk/kFk) > Tor,i = i + 1, ELSE GO TO 7.
3 Calculate K, and solve Ki u = −R.
4 Update x = x + u.
5 Find other parameters (stresses, strains, deformation tensor).
6 Calculate T and R, GO TO 2
7 END.
Bibliography

[1] Malveren, L., E.: Introduction to the mechanics of a continuous

medium, Prentice-Hall, New York, 1969
[2] Zienkiewicz, O.,C.: The finite element method (Third edition), Mcgraw-
Hill Book Company (UK) Limited, 1979
[3] Spiering, R.,M.,E.,J., Grootenboer, H.J.: Eindige elementen methode in
de werktuigbouwkunde, Dictaat Unversity of Twente, The Netherlands,
1993 (in Dutch)
[4] Jongschaap, R.,J.,J.: Toegepaste tensorrekening, Dictaat Unversity of
Twente, The Netherlands (in Dutch)
[5] Huétink, J.: Plasticiteit en kruip, Dictaat Unversity of Twente, The
Netherlands, 1994 (in Dutch)
[6] Mooi, H.: Finite element simulations of aluminium extrusion, PhD the-
sis Unversity of Twente, The Netherlands, 1996
[7] Bonet, J., Wood, R.D.: Nonlinear continuum mechanics for finite ele-
ment analysis, Cambridge University Press, 1997
[8] Belytschko, T., Liu, W.K., Moran, B.: Nonlinear finite elements for
continua and structures. John Wiley and Sons Ltd., England, 2001
Chapter 5

Displacement Based Steady State

Formulation

Go for details

Our project is concentrated on the simulation of steady state flow processes.

In [1] coupled FE calculations with the material evolution equations in a ref-
erential configuration are used. However, this requires a complicated coor-
dinate system transformation. In this chapter, a displacement based steady
state formulation is constructed based on the ideas from [1].
Two basic expressions are involved in this steady state formulation: the
material evolution equations and the equilibrium equations. These can be
solved as a coupled or an uncoupled set using FEM. Only the uncoupled
approach is examined here. The equilibrium equations are expressed in the
undeformed configuration. The material evolution equations in our case ([3])
are different from the usual steady state solution methods (See Chapter 3).
These equations are not expressed by the velocity v of a particle with ma-
terial coordinates X in Eulerian description, but by the initial velocity v 0 .
In Balagangadhar’s method, the material evolution equations are expressed
by the velocity in the referential configuration v r .
In the first section the governing equations for the steady state formu-
lation are discussed. The stabilized FEM must be taken into account for
reducing numerical oscillation problems, when solving the differential equa-
tions. In the second part this issue is considered; furthermore, a calculation
procedure is proposed in the third section. Finally, since a forming process
involves the interaction of two or more bodies, the contact problem is an
important issue. In order to apply this steady state formulation to the real
industrial processes, the contact problem cannot be ignored. In our case,
since the solution procedure is different from other classical methods for
steady state flow problems, the contact analysis becomes unique for the de-
52 Displacement Based Steady State Formulation

veloped steady state formulation. This discussion is discussed in Section

5.4.

5.1 Governing Equations

In the steady flow calculations the governing equations include the equilib-
rium equations, the constitution relations, boundary conditions and material
evolution (convection) equations.

5.1.1 Equilibrium

The equilibrium equations are Eq.(2.40):

∇ · (FS) + b0 = 0 (5.1)

With boundary conditions:

u = u0 on Γu 0 (5.2)

S · n0 = t0 on Γt 0 (5.3)
where n0 is the outward normal with respect to the boundary, t0 represents
the surface traction on surface Γt 0 and u0 is the prescribed displacement on
surface Γu 0 .

5.1.2 Constitutive Relations

The constitutive relations depend on the material models that are chosen. In
our case, hyperelastic-plastic constitutive models are chosen, and the stresses
can be described as the function of Lagarangian strains:

S = f (Ee ) (5.4)

where Ee represents the elastic strain.

The yield function is written as:

φ(S, q) = 0 (5.5)

where q is a set of internal variables which is governed by the hardening or

softening law of metal deformation.
The flow rule is specified as (see Eq. (2.84)):
p
Ė = Y : Ė (5.6)

where Y is a four order tensor, which can be obtained from the consistency
condition and the yield function.
 5.2 Material Evolution Equations 53

5.1.3 Convection Equations

In order to obtain the plastic strain further in the steady state problem, the
convection equations must to be considered. The convection equations in
Eulerian description:
p
v · ∇Ep = Ė (5.7)
A detailed description of the convection equations in our steady state for-
mulation can be found in the following section.

5.2 Material Evolution Equations

5.2.1 Developed Material Evolution Equations

The material evolution equations are derived from the expression of the
material derivative of an arbitrary state variable in Eulerian way. The ex-
pression is repeated for steady state:

f˙ = v · ∇f (5.8)

We then transform this convection form from the deformed configuration to

the undeformed configuration for the steady state:

f˙ = v · ∇f = (v 0 · FT ) · (F−T · ∇0 f) = v 0 · ∇0 f (5.9)

where ∇0 is the gradient in the undeformed configuration. In this form

the coordinate system transformation is not required as in the method of
Balagangadhar (see Chapter 3). The analysis is steady and the material
state variables can be obtained via integration along the known path-line in
the undeformed configuration. The characteristics of this form are described
in Figure 5.1. The solid line block and grey block represent the initial and

t=0 0 t=t’
v0 A0 v
initial u A
deformed

flow in X flow out

x

Figure 5.1: New description of steady flow processes

deformed situations, respectively. The blocks are connected using two dotted
lines. These two lines indicate the flow pattern between time t0 (initial time)
54 Displacement Based Steady State Formulation

and time t (certain process time). The characters A0 and A stand for any
material particle in the initial mesh and deformed mesh, respectively. Each
material particle A in the deformed situation corresponds to a particle A0
in the initial situation during the steady flow process.

5.2.2 Stabilized FEM for Convection

The convection-diffusion equation system can be written as:

∂f
+ v · ∇f + ∇ · K∇f = g (5.10)
∂t
where g is the source vector. The second term in the above equation is the
convection term and the third term is the diffusion term.
An important parameter indicating the character of the flow is the ele-
ment P éclet number which is defined as:
vL
P é = (5.11)
κ
where L is the length scale and κ is the diffusion coefficient. If P é À 1
convection dominates the solution of the convection-diffusion problem, the
results show an abrupt change within very few elements and the numerical
oscillations occur using the Galerkin method ([4],[5]). The oscillations be-
come worse for a larger P éclet number.
The material evolution expression as in Eq.(5.9) is the convection equa-
tion with a convection term only. The P éclet number is equal to infinity.
Clearly, if the Galerkin formulation is applied for solving these equations,
numerical oscillations cannot be avoided. For this reason stabilized methods
have been developed for convection dominated equations.
Two stabilized methods are investigated here, the Streamline Upwind
Petrov Galerkin functions (SUPG) and the Least Squares method (LS).
SUPG, from a recent perspective, is one of the methods showing reason-
able accuracy and stability properties. The Least Squares FE method has
also been tested for the stability of the solutions in the convection equation.
It is based on the minimization of the residuals in a least squares sense, and
the theory and applications can be found in [4], [6] and [7]. When these
kinds of methods are applied to finite element calculations, we call these
methods ”stabilized finite element methods”.
The schematic descriptions of Galerkin, SU P G and LS weighting func-
tions are illustrated in Fig. 5.2.
For the Galerkin formulation, the convection equation Eq. (5.9) has the
weak form as: Z Z
δf · (v 0 · ∇0 f)dV = δf · f˙ dV (5.12)
 5.2 Material Evolution Equations 55

Figure 5.2: Three different weighting functions

For SU P G and the LS method, the weak forms are expressed, respectively:
Z Z
(δf + γv 0 · ∇0 δf) · (v 0 · ∇0 f)dV = (δf + γv 0 · ∇0 δf) · f˙ dV (5.13)

and: Z Z
(v 0 · ∇0 δf) · (v 0 · ∇0 f)dV = (v 0 · ∇0 δf) · f˙ dV (5.14)

where the parameter γ is the intrinsic time scale.

5.2.3 Tests with Three Weighting Methods

First, in order to compare the numerical results with the analytical solutions
when solving Eq. (5.9), a one-dimensional test with 4-node elements is de-
fined with nine integration points in one element. The material derivatives
ḟ are known at the integration points. A prescribed nodal value of f is given
at the inflow boundary.
In Fig. 5.3 the material derivatives ḟ are given as constants within one

0.3
Values of time derivative of f

0.25
Element nodes
0.2
Values by integration points
0.15

0.1

0.05

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 5.3: Case 1: Input and element mesh in 1D case

element by the integration points. The analytical solutions are compared

to the numerical solutions obtained from Galerkin, SU P G and LS method.
The results are shown in Fig. 5.4. From this figure it is found that the
results from these three weighting methods agree with those from analytical
56 Displacement Based Steady State Formulation

0.035

0.03

0.025
Results of f by nodal points

0.02

0.015 Using Galerkin weighting

Using LS weighting
Using SUPG weighting
Analytical solutions
0.01

0.005

−0.005
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
X

Figure 5.4: Case 1: Comparison between the analytical solutions and the results from
three weighting functions

solutions No difference, however, cannot be seen between these three meth-

ods. The input values in this case within one element are constant, so that
the nodal values in FEM are exact.
A further case was also checked with a non-constant input of the material
time derivatives within one element. The mesh and input of the material
time derivatives are shown in Fig. 5.5 and the results are shown in Fig. 5.6.
It is apparent that the Galerkin method gives most spurious oscillation. Even
though the SU P G method gives better solutions than the Galerkin method
, the LS method gives a more stable result than the SU P G method.
The further applications of Galerkin, SU P G and LS methods were
Values of time derivative of f

0.3

0.25 Element nodes

Values by integration points
0.2

0.15

0.1

0.05

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
X

Flow direction

Figure 5.5: Case 2: Input and element mesh in 1D case

 5.2 Material Evolution Equations 57

0.035

0.03

0.025
f by nodal points

0.02

0.015

Using Galerkin
0.01 Using SUPG
Using LS
Analytical solutions
0.005

−0.005

−0.01
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 5.6: Comparison between the results from three weighting functions in 1D case

Tool

Flow direction
Tool

L-- Prescribed displacements area

Figure 5.7: Set-up of the extrusion-like problem

implemented in a so-called extrusion-like problem with a purely elastic de-

formation. The set-up is shown in Fig. 5.7. Due to symmetry only half of
the geometry needs to be modeled. The elastic deformed mesh can be found
in Fig. 5.8.
The calculation procedure is described as follows:
Step 1 Initialize all the data. Set S0 = 0, F = I, u = 0 etc.
Step 2 Process prescribed loads or displacements.
Step 3 Solve [K] · ∆u = F − R using FEM with pure elastic deformation
58 Displacement Based Steady State Formulation

with stiffness matrix K. In case of nonlinearity an iterative method

should be used. The iteration stops when the unbalance criterion is
satisfied.
Step 4 With the help of the results for u calculate the strain E = 12 ((∇0 u)T +
∇0 u + ∇0 u · (∇0 u)T ) directly.
Step 5 Calculate the flux of total strains Ė = v0 · ∇0 E.
Step 6 Calculate the strains using v0 ∇0 · E∗ = Ė from the flux of total
strains .

0.12

0.1

0.08

0.06
Y

0.04

0.02

0
−0.4 −0.35 −0.3 −0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15 0.2 0.25
X

Figure 5.8: Deformed mesh with pure elastic deformation

0.25

Direct soultion from equilibrium

Using Galerkin method in convection equations
0.2
Using SUPG method in convection equations
Using LS method in convection equations
EXX of surface nodes

0.15

0.1

0.05

−0.05
−0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3

Figure 5.9: Results of EXX with different weighting ways

 5.2 Material Evolution Equations 59

0.2

0.15 Direct results from Equilibrium calculation

From using Galerkin method in convection equations
From using SUPG method in convection equations
0.1 From using LS method in convection equations

0.05
EYY of the surface nodes

−0.05

−0.1

−0.15

−0.2

−0.25

−0.3

−0.35

−0.4
−0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3
X

Figure 5.10: Results of EY Y with different weighting ways

0.5

0.4

0.3
EXY of surface nodes

0.2

0.1

−0.1

Direct soultion from equilibrium calculation

−0.2 Using Galerkin method in convection equations
Using SUPG method in convection equations
Using LSmethod in convection equations

−0.3

−0.4

−0.5
−0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3
X

Figure 5.11: Results of EXY with different choices of weighting

It is obvious that E∗ must be equal to E. The results obtained from using

the different weighting functions in these convection equations, are shown in
Fig. 5.9, Fig. 5.10 and Fig. 5.11. The same conclusions are drawn as for the
previous case. Also here the superior stability of the Least Squares method
is obvious. In fact, The Galerkin finite element method leads to oscillatory
results. A similar behaviour is observed in the central difference method
in the finite difference method. The SU P G uses upwinding strategies to
represent the uni-directional character of the propagation phenomenon dur-
ing the flow so that the numerical oscillations are reduced compared to the
Galerkin method. However, the matrix of the resulting algebraic system is
60 Displacement Based Steady State Formulation

nonsymmetric in SU P G. In [3] it is shown that the LS method has better

control of the streamline derivative than the SU P G and the matrix of the
resulting algebraic system is symmetric and positive definite.
In our further research, in order to avoid the influence of the numerical
oscillations when solving the convection equations, the LS method was cho-
sen.
However even when using the LS method, the strains from the convection
calculation E∗ deviate from those obtained from direct calculation from the
equilibrium equations E. This is due to how the material time derivatives of
E are calculated. The following section is therefore dedicated to this issue.

5.2.4 Material Time Derivatives of E

The material time derivatives of strains can be expressed as a convection

equation (Ė = v 0 · ∇0 E). There are two ways to calculate ∇0 E:
• Method 1: Directly from displacements, which means that the second
order derivatives of displacements are needed. The interpolation func-
tions for the displacements are quadratic, and C0 continuous across
the element boundaries. Then, the strains are C−1 continuous. This
means that the gradient of the strains does not exist at the element
boundaries. This method was used in the previous section.
• Method 2: First obtain the nodal averaged strains En from the element
strains and then use Ė = v0 · ∇0 En . The field of En is now continuous
across the element boundary.
To check these two proposed methods we apply them again to the trivial
problem, in which the material only deforms elastically (see Fig. 5.7 and
Fig. 5.8). In this case the stresses are independent of the deformation history
and hence can also be obtained in one step. They are both direct and indirect
ways to calculate strains in pure elastic deformation:
• Direct: From the relation between the displacements and the strains.
• Indirect: From integration of the flux of the strains, which can be
obtained by both Method 1 and Method 2 above.
Comparisons of strains obtained from both Direct and Indirect methods are
shown in figs. 5.12, 5.13 and 5.14 for the strain components EXX , EY Y and
EXY . In the Indirect method, both Method 1 and Method 2 are used, respec-
tively. It is clearly seen that Method 2 gives better results than Method 1
for calculating EXX and EXY . Method 1 is better for EY Y , but there is
only local deviation in Method 2. For this reason, Method 2 was chosen for
further applications.
 5.3 Solution Algorithm 61

0.3

0.25

Direct solution from equilibrium equations

Indirect method 1 for convection equations
0.2
of surface nodes

Indirect method 2 for convection equations

0.15
XX

0.1
E

0.05

−0.05
−0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3
X

Figure 5.12: EXX of surface nodes along along the direction of flow

0.1

0
EYY of surface nodes

−0.1

Direct soultion from equilibrium equations

Indirect method 1 for convection equations
Indirect method 2 for convection equations

−0.2

−0.3
−0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3
X

Figure 5.13: EY Y of surface nodes along the direction of flow

5.3 Solution Algorithm

The calculation procedure for an elastic plastic model based on the developed
evolution equations is as follows:
Step 1 Initialize all the data. Set S0 = 0, F = I, u = 0 etc.

Step 2 Process prescribed loads or displacements.

Step 3 Solve [K]·∆u = F −R using FEM with pure elastic deformation with
a stiffness matrix K. In the case of nonlinearity an iterative method
62 Displacement Based Steady State Formulation

0.2

0.1

0
EXY of surface nodes

−0.1

Direct solution from equilibrium equations

Indirect method 1 from convection equations
−0.2 Indirect method 2 from convection equations

−0.3

−0.4

−0.5
−0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3
X

Figure 5.14: EXY of surface nodes along the direction of flow

should be used. The iteration stops when the unbalance criterion is

satisfied.
Step 4 Calculate the strains E = 12 ((∇0 u)T + ∇0 u + ∇0 u · (∇0 u)T ).

Step 5 Calculate the flux of total strains Ė = v 0 · ∇0 E.

p
Step 6 Obtain the time derivative of the plastic strains Ė = Y · Ė.
Step 7 Obtain the plastic strains with the developed evolution equations
p
v 0 · ∇0 Ep = Ė
Step 8 Calculate the total stresses from Eq. (5.4). Because tensor Y de-
pends on the total stresses, an iterative method should be used until a
convergence criterion is satisfied. The iterative steps begin from Step
6.
Step 9 Solve:
[K] · ∆u = F − R
where R represents the element reaction forces. Iterative steps should
be taken as from Step 3.
It can be observed that this computation procedure does not require the
transient solution of the process. First the elastic calculation should be
carried out so that the displacements are obtained to calculate the fluxes of
the total strains and the fluxes of the plastic strains. Once these results are
known, the nodal plastic strains can be calculated from the integration of
the fluxes of the plastic strains. Afterwards, the discontinuous stresses are
 5.4 Contact Analysis 63

calculated. Plastic deformation takes place in the loading region. Elastic

spring-back is expected in the unloading part.

5.4 Contact Analysis

When the points on a boundary of one body contact with points on the
boundary of the same or another bodies, the contact problems will be con-
sidered. In many industrial processes, the contact problems cannot be ig-
nored due to the contact forces generated during processing. In order to find
an appropriate manner for describing the contact for our case, the analysis
based on [8] and [9] is studied for general problems, such as impact problems
or flow problems with transient calculations.
The sum of the internal virtual work and the virtual work of applied
forces and traction for the bodies should balance the virtual work of the
contact forces acting on contact bodies Γi :
Z Z Z
i i i i i i i i
ϕ , δu ) =
G (ϕ S : δE dΩ0 − b : δu dΩ0 − t : δui dΓi s0
Ωi0 Ωi0 Γis 0
Z (5.15)
= tc : δui dΓic 0
Γic 0

This equation must hold for each body (i) and for all time t. The quantity
Gi (ϕ
ϕi , δui ) is the sum of the internal virtual work and the virtual work of
the applied forces and traction for body i. Briefly([8],[9]):

Gi (ϕ
ϕi , δui ) + Gic (ϕ
ϕi , δui ) = 0 (5.16)

with: Z
Gic (ϕ
ϕi , δui ) =− tic · δui dΓi (5.17)
Γic

The contact force induced on body 2 is equal and opposite to that induced
on body 1 at the contact area, i.e.

t2c = −t1c (5.18)

Therefore: Z
Gic (ϕ
ϕi , δui ) =− t1c · (δu1 − δu2 ) dΓ1c (5.19)
Γ1c

Resolving the contact forces and the virtual displacements into normal and
tangential components gives the following statement of the contact virtual
work ([1]): Z
ϕ, δu) =
G(ϕ [t1cn · δdN + t1ct · δdt] dΓ1 (5.20)
Γ 1
64 Displacement Based Steady State Formulation

where dN is the distance from body 1 to body 2 (see Fig. 5.15) and dt is
the tangential shift. If using a penalty method, the contact traction can be
described as:
tcn = P dN (5.21)
provided:
dN ≤ 0 (5.22)
P is the penalty number. In the Coulomb friction model, calculation of
the tangential force follows the following conditions with both slip and stick
states:

φ = tct − µtcn < 0 stick (5.23)

tct = Pt dt stick (5.24)
tct = µtcn slip (5.25)

Pt is the penalty number for tangential traction and dt is the the slip distance.
As described earlier, the contact algorithm is split into two parts: normal
and tangential contact traction. The virtual work in the normal direction
can be described as:
Z
Gcn = [δu] · P dN · δdN dΓ
Γ
Z (5.26)
= P dN · ∇u dN dΓ
Γ

From this equation, the normal reaction forces can be worked out:
Z
Rcn = P dN · ∇u dN dΓ (5.27)
Γ

The stiffness Kcn will be:

Z
Kcn = (P (∇u dN ) + P dN (∇u ∇u dN )) dΓ (5.28)
Γ

The closest distance between the two bodies should be sought during the
simulation as described in Figure 5.15; i.e. the distance between a surface
point a (nodal point or integration point) on surface of body 1 and the
surface elements of body 2 should be calculated one by one to find which
element is closest to point a. The details of how to determine dN , ∇u dN
and ∇u ∇u dN can be found in Appendix C.
For slip, the virtual contact work in the tangential direction is written
as: Z
Rct = (P µdN ∇u dt) dΓ (5.29)
Γ
 5.4 Contact Analysis 65

Figure 5.15: Schematic depicting gap definition between two contacting bodies

Figure 5.16: Description of slip in our steady state problem

Therefore, the stiffness Kct is expressed as:

Z Z
Kct = ∇u dt P µ∇u dN dΓ + P µdN ∇u ∇u dtT dΓ
T
(5.30)
Γ

In impact problems or transient calculations, the slip theory concerning the

relative movement between two calculation steps is used (see Appendix D).
However, if the steady state calculations are carried out, the slip for steady
state calculations should be defined as the relative movement between the
bodies (See Fig 5.16). One conclusion is that the slip theory in the contact
method used in most impact problems or transient calculations could not be
used directly for our steady state case. However, the theory to obtain the
normal tensor can still be applied for our formulation.

5.4.1 A Contact Method

In our case, assuming that the tool is considered to be rigid, the weak form
is: Z
G(u, δu) = t1c · δu1 dΓ1 (5.31)
Γ(1)
with: · ¸
t
t1c = cx (5.32)
tcy
The contact forces can be split into the normal and tangential directions to
the surface: · ¸
t
tc = ct
01
(5.33)
tcn
66 Displacement Based Steady State Formulation

The values should be transformed from the local to the referenced configu-
ration.
The reaction force is given by:
Z
Rc = Q · t01
c dΓ
1
(5.34)
Γ

where Q is the transformation tensor. As well known:

tcx = tct tt · ex + tcn n · ex (5.35)

tcy = tt tct · ey + tcn n · ey (5.36)

therefore:
tc = (tt ⊗ ex + n ⊗ ey ) · t0 c (5.37)
with:
Q = tt ⊗ ex + n ⊗ ey (5.38)
tt and n are the unit tensors with relation to the angle between the normal
tensor and the reference configuration and can be obtained by the calculation
of the shortest distance between two bodies.
Hence, the stiffness will be:
Z
Kc = (Q · ∇u t0 c + ∇u Q · t0 c ) dΓ1 (5.39)
Γ

with:
∇u Q = ∇u tt ⊗ ex + ∇u n ⊗ ey (5.40)
Due to:
(S ⊗ v) · u = (u · v)S (5.41)
therefore:

∇u Q·t0 c = (∇u tt ⊗ex +∇u n⊗ey )·t0c = (ex ·t0 c )∇u tt +(ey ·t0 c )∇u n (5.42)

This contact algorithm was chosen for further applications, as shown in the
following chapter.
Bibliography

[1] Balagangadhar, D., Tortorelli, D.A.: Design of large-deformation steady

elastoplastic manufacturing processes. Part I: A displacement-based ref-
erence frame formulation. International Journal for Numerical Methods
in Engineering. Vol. 49, 2000, 899-932
[2] Balagangadhar, D., Dorai, G.A., Tortorelli, D.A.: A displacement-based
reference frame formulation for steady-state thermo-elasto-plastic mate-
rial processes. International Journal of Solids Structures. Vol. 36, 1999,
2397-2416
[4] Yu, Y., Geijselaers, H.J.M., Huétink, J.: A new displacement based FE
formulation for steady state problems, Simulation of Materials Process-
ing: Theory, Methods and Applications, Mori (Eds), Proceedings of
NUMIFORM’2001, Toyohashi, Japan, 2001, 121-126
[4] Jiang, B.: The least-squares finite element method, Theory and applica-
tions in computational fluid dynamics and electromagnetics, Springer-
verlag Berlin Heidelberg New York, 1998, 1-30, 201-213
[5] Mooi, H.: Finite element simulations of aluminium extrusion, phD the-
sis University of Twente, Enschede, The Netherlands, 1996
[6] Jiang, B.: On the least-squares method, Computer Methods in Applied
Mechanics and Engineering, Vol. 152, 1998, 239-257
[7] Moussaoui, F.: A unified approach for inviscid compressible and nearly
incompressible flow by least-squares finite element method, Applied Nu-
merical Mathematics, Vol. 44, 2003, 183-199
[8] Laursen, T.A., Simo, J.C.: A continuum-based finite element formula-
tion for the implicit solution of multibody, large deformation frictional
contact problem, International Journal for Numerical Methods in En-
gineering. Vol. 36, 1985, 3451-3485
[9] Simo, J., Lauren, T.A.: An augmented Lagrangian treatment of contact
problems involving Friction, Computers Structures. Vol. 42, No.1, 1992,
97-116
[10] Wriggers, P., Vu Van, T., Stein, E.: Finite element formulation of
large deformation impact-contact problems with friction, Computers
and structures. Vol.37, No.3, 1990, 319-331
68 Bibliography

[11] Kloosterman, G., van Damme, R.M.J., van den Boogaard, A.H.: A
geometrical-based contact algorithm using a barrier method, Interna-
tional Journal for Numerical Methods in Engineering. Vol.51, 2001,
865-882
[12] Belytschko, T., Liu, W.K., Moran, B.: Nonlinear finite elements for
continua and structures. John Wiley and Sons Ltd., England, 2001
Chapter 6

Test Results

A step further

In Chapter 5 a developed steady state FE formulation is introduced the-

oretically and some related FEM problems are discussed. In this chapter
we attempt to obtain the test results with two different material models: a
linear material model and an nonlinear material model. The steady state
formulation, which is outlined in preceding chapter, is implemented in sev-
eral cases with two different material models. All the problems are plane
strain problems. The simulations were all run using Matlab. We used a
quadrilateral element with 9 nodes to interpolate the displacement u and 4
nodes to interpolate the strain E and the plastic strain Ep .

6.1 Linear Material Model

First, the finite element simulations were carried out with a linear material
model which can be used for small strain deformations. In this material
model the stress relates to the elastic strains linearly as:

S = Cel : Ee = Cel : (E − Ep ) (6.1)

This is called a Saint Venant-Kirchhoff material or a Kirchhoff material

for brevity ([1]). Cel is the constant fourth order elastic moduli tensor.
Many materials (such as metals) can be modelled as being isotropic for
small strains. An isotropic tensor Cel was chosen for our calculation:

Cel = λI ⊗ I + 2µI (6.2)

where the two independent material constants λ and µ are called the Lamé
constants. I is the fourth order tensor with the components Iijkl = 21 (δik δjl +
δil δjk ). The Lamé constants can be expressed in terms other constants, the
70 Test Results

Young’s modulus E and the Possion’s ratio ν:

E νE
µ= , λ= (6.3)
2(1 + ν) (1 + ν)(1 − 2ν)
The associative plastic theory in these calculations is treated as being similar
to that in the hypoelastic-plastic model for linear isotropic hardening:
r
2
φ(S, q) = kdev(S)k − (Sy0 + Et · q) ≤ 0 (6.4)
3
where dev(·) is the deviator of the tensor argument. Sy0 is the initial yield
stress, Et is a constant with a positive value which represents the evolution
of the isotropic hardening and q is the equivalent strain in this case. The
p
evolution of Ė is described as:
p
Ė = λ̇dev(S) ⊗ dev(S) : Ė (6.5)

For pure elastic deformation and an unloading situation,

λ̇ = 0 (6.6)

For plastic deformation:

1
λ̇ = Et
(6.7)
kdev(S)k2 (1 + 3µ )

Two steady state flow processes were modelled with this linear material
model in the displacement-based formulation. One is a pure shear flow and
second is a simple extrusion case.

6.1.1 Pure Shear Flow

The set-up of the pure shear test is shown in Fig. 6.1. The vertical arrows
illustrate the prescribed displacement direction. The material flows in and
then out through the loading region, and the load gives the material only
pure shear deformation. Therefore the material by or after loading endure
the pure shear deformation. The material will only retain the plastic de-
formation after flowing out of the loading region. The elastic spring-back
should be observed at the flow-out part for an elastic-plastic material. The
relation between shear stresses and shear strains is described as:

τxy = µγxy (6.8)

In this test the contact region is known in the undeformed configuration,

and a contact analysis is therefore not required.
 6.1 Linear Material Model 71

0.02

0.015

0.01

0.005
Load
0
0 0.02 0.04 0.06 0.08 0.1

Flow direction

Figure 6.1: Set-up of pure shear flow

E ν Et Sy0 inflow velocity v 0

70Gpa 0.35 1Gpa 2Gpa [1 0]

Table 6.1: The material parameters of the shear flow test

0.03

0.02

0.01
Y

−0.01

Flow direction
−0.02
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
X

Figure 6.2: Pure elastic shear deformation

0.03

0.025

0.02

0.015

0.01

0.005
Y

−0.005
Flow direction
−0.01

−0.015

−0.02
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
X

Figure 6.3: Elastic plastic deformation mesh

The parameters were chosen as given in Table 6.1. In order to see the
spring-back more clearly, we chose the yield stress Sy0 higher than that in
practice.
In Fig. 6.2 the deformed mesh is shown when only a pure elastic defor-
72 Test Results

0.03

0.025

0.02

0.015

0.01

0.005
Y

−0.005
Flow direction
−0.01

−0.015

−0.02
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
X

0 −0.02 −0.04 −0.06 −0.08 −0.1 −0.12 −0.14 −0.16 −0.18 −0.2

Figure 6.4: Total shear strains distribution

0.03

0.025

0.02

0.015

0.01

0.005
Y

−0.005
Flow direction
−0.01

−0.015

−0.02
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
X

0 −0.02 −0.04 −0.06 −0.08 −0.1 −0.12 −0.14 −0.16

Figure 6.5: Plastic shear strains distribution

mation takes place in the simulation. It is found that the material recovers
to the undeformed situation after unloading. In Fig. 6.3 the mesh with
elastic plastic deformation is shown. The elements keep the plastic deforma-
tion after flowing out of the loading region. In order to observe the elastic
spring-back in the simulation, the total shear strain distribution is shown in
Fig. 6.4. It was found that the maximum shear strain is at the nodes which
have the maximum prescribed displacements by the loading regions. In the
flowing-out area the nodes have less total shear strains than those in the
loading region, and that means that the material springs back elastically af-
ter unloading. The plastic shear strains reach a certain constant value after
flowing out in Fig. 6.5.
The plastic strains and total strains along the streamline on the surface
are shown in Fig. 6.6. In the loading area the plastic strains are smaller than
the total strains due to elasticity. After the material flows out of the loading
 6.1 Linear Material Model 73

0.1

0.05
γXY at integration points of the surface

Total shear strains

Plastic shear strains
0

−0.05

−0.1

−0.15

−0.2

−0.25
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
X

Figure 6.6: Total and plastic strains along streamline on surface

0.03

0.025

0.02

0.015

0.01

0.005
Y

−0.005
Flow direction
−0.01
E=200Gpa
−0.015 E=70Gpa

−0.02
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
X

Figure 6.7: Elastic plastic deformed meshes with different elastic modulus

region, both strains overlap due to elastic spring-back after unloading. The
elastic deformation becomes larger with decreasing elastic modulus, which
can be seen in Fig. 6.7. With larger elastic deformation, the elements on the
flow-out side spring back more.

6.1.2 Simple Extrusion Case

In this section a simple extrusion case is studied. Here, the contact region is
unknown and a contact analysis must be carried out. Because the problem is
axis-symmetric, only half of set-up is shown in Fig. 6.8. The material flows
in, enters the tool and then flows out. The in-flow boundary is suppressed.
The material properties were chosen as in Table 6.2.
74 Test Results

Figure 6.8: Undeformed mesh

The deformed mesh with the pure elastic deformation is shown in Fig. 6.9.

E ν Et Sy0 [v X , vY ](m/s) Friction coefficient (µ)

70Gpa 0.35 1Gpa 2Gpa [1 0] 0./0.2/0.4

Table 6.2: The material parameters of extrusion case in linear model

Fricti
o n forc
0.05 e

0.04

0.03
Y

0.02

0.01

Flow direction
−0.01
−0.2 −0.1 0 0.1 0.2 0.3 0.4
X

Figure 6.9: Pure elastic deformation mesh

In the extrusion process the friction between the tools and the specimen
 6.1 Linear Material Model 75

One piece of surface

0.05

0.04

0.03
Y

0.02

0.01

Flow direction
−0.01
−0.2 −0.1 0 0.1 0.2 0.3 0.4
X

(a) Elastic-plastic deformed mesh with different friction coefficients

0.0485

0.0484

0.0483

0.0482
µ=0.4
0.0481
Friction force µ=0.2
µ=0.
0.048
Y

0.0479

0.0478

0.0477

0.0476

0.0475 Flow direction

−0.062 −0.061 −0.06 −0.059 −0.058 −0.057 −0.056 −0.055 −0.054 −0.053
X

(b) Nodes on surface part with µ=0., µ=0.2, µ=0.4

Figure 6.10: Comparison of deformed mesh with three different friction coefficients
76 Test Results

does not change direction. The contact algorithm described in the previous
chapter was applied for this case. The results with different friction coeffi-
cients are shown in Fig. 6.10. With a larger friction the mesh flows more
slowly.
The deformed mesh with the velocity plot is shown in Fig. 6.11 when the

0.05

0.04

0.03
Y

0.02

0.01

Flow direction
−0.01
−0.2 −0.1 0 0.1 0.2 0.3 0.4
X

Figure 6.11: Deformed mesh with velocity vector plot

values of the initial velocities at the in-flow boundary are equal to 1.0m/s.
The velocities of the integration points were obtained from:

v = F · v0 (6.9)

0.05

0.04

0.03
Y

0.02

0.01

0
Flow direction

−0.01
−0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5
X

0.05 0.1 0.15 0.2 0.25

Figure 6.12: Equivalent total strain distribution

 6.1 Linear Material Model 77

0.05

0.04

0.03
Y

0.02

0.01

0
Flow direction
X
−0.01
−0.2 −0.1 0 0.1 0.2 0.3 0.4

0.2 0.15 0.1 0.05 0

Figure 6.13: Equivalent plastic strain distribution

E=35Gpa
E=70Gpa
E=200Gpa
0.05

0.04

0.03
Y

0.02

0.01

0
Flow direction

−0.01
−0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5
X

Figure 6.14: Deformed meshes with three different elastic moduli

The deformed mesh with equivalent strains distributions is shown in Fig. 6.12.
The values at the largest loading area is larger than those in the flow-out
region due to the elastic spring-back. Further, the deformed mesh with
equivalent plastic strains is shown in Fig. 6.13. It is seen that the equivalent
plastic strains reach the constant values from deformed area.
We know that the material properties play an important role in the de-
formation. For elasto-plastic materials the elastic deformation will be more
when the specimen has a smaller elastic modulus. In our case the specimen
78 Test Results

will spring back more after flowing-out with a smaller elastic modulus. In
Fig. 6.14 three deformed meshes have different elastic spring-back when the
deformation in these three cases is based on the different elastic moduli.

6.2 Nonlinear Material Model

In this section the hyperelastic plastic material model is applied. The intro-
duction of this model was described in Chapter 2. In our case a stored-energy
function was chosen of the form([2]):
1 1 1
ψ̂ = κ[ (Je2 − 1) − lnJe ] + µ(tr[b̄e ] − 3) (6.10)
2 2 2
where:
Je := det[Fe ] (6.11)
e
b̄ := Je−2/3 Fe · FeT (6.12)
κ and µ are interpreted as the bulk modulus and the shear modulus, re-
spectively. The relations between these four material parameters κ, elastic
modulus E and Possion’s ratio ν is as follows:
E
κ= (6.13)
3(1 − 2ν)
From this stored energy function the Second Piola-Kirchhoff stress has the
form:
1 2
S = κC−1 (Je2 − 1) + µJe− 3 DEV (Cp−1 ) (6.14)
2
in which DEV represents DEV := (·) − 31 [C : (·)]C−1 .
The Von Mises yield condition is used in the material description as ([2]):
r
1
φ(S, C) = (C · S) : (S · C) − (S · C)2 − R (6.15)
3
where R is the radius of the Von Mises sphere. The Kuhn-Tucker optimality
condition then yields the associative flow rule:
p ∂φ
−M : Ė = γ̇ (6.16)
∂E
where γ̇ is the plastic consistency parameter, and M is defined as:
∂S
M(E, Ep ) = (6.17)
∂Ep
In order to test this model, a one-element test was carried out. Further, this
model was applied to the same case as in section 6.1.1.
 6.2 Nonlinear Material Model 79

Y_pre Y_pre Y_pre

1
3 7 4
0.8
Y_pre =-1.2
0.6

0.4

0.2

0
8 9 6
-0.2

-0.4 Y_pre :Prescribed displacement

-0.6
Y
-0.8

-1 1 X 5 2
-1 -0.5 0 0.5 1

Figure 6.15: Set-up of one-element test

1
3 4
7
0.8

0.6
Deformed element with 60% elastic reduction
Element with 100% spring−back after unloading
0.4

0.2

0 Y 8 9 6

−0.2

−0.4

−0.6

−0.8

1 5 2
−1
X

−1 −0.5 0 0.5 1 1.5

Figure 6.16: Pure elastic deformation in one element test

6.2.1 One-Element Test with HYPEP Model

A one-element test was carried out to verify the hyperelastic plastic (HY P EP )
material model that we introduced above. A 9-Node element was chosen in
this test and the set-up can be seen in Fig. 6.15. The prescribed displace-
ments are given by Node 3, Node 7 and Node 4. Node 1 was suppressed and
Node 5 and Node 2 can only move along the X-direction.
There are two ways to test this model. First with only a pure elastic de-
formation. The element should be spring back 100 percent after unloading.
Since the elastic response is derived from a hyperelastic potential, the work
done in a closed elastic deformation path vanishes exactly. In Fig. 6.16 one
mesh is the deformed mesh with an elastic deformation and the other mesh
is the mesh with pure elastic spring-back. It was found that the mesh had
80 Test Results

1
3 7 4

0.8
Undeformed element
0.6 Compressed element with κ=4
Spring−back after unloading with κ=4
Compressed element with κ=8
0.4
Spring−back after unloading with κ=8

0.2

0 8
Y

−0.2

−0.4

−0.6

−0.8

1 5 2
−1

−1 −0.5 0 0.5 X 1 1.5 2 2.5 3

Figure 6.17: Two elastic plastic deformations in one-element test

sprung back to the undeformed state. Secondly an elasto-plastic deformation

was carried out in the procedure above. After a certain amount of reduction
the element was unloaded with two different elasto-plastic deformations. In
these two deformations the material parameter κ was changed. It can found
in Fig. 6.17. From this one-element test, this hyperelastic plastic material
model can be applied for further research.

6.2.2 Simple Extrusion Test with HYPEP Model

This extrusion test has the same upsetting and calculation as the example
in Section 6.1.2, but the material model is changed to a hyperelastic plastic
material model (HY P EP ). Another difference is that a larger reduction has
been applied in this case. The material proprieties was chosen as follows in
Table 6.3. The friction coefficient was chosen to be 0.1.
The deformed mesh with the velocity plot is shown in Fig. 6.18. The

κ µ Et Sy0 [v X , vY ](m/s)
160Gpa 80Gpa 2Gpa 1Gpa [1 0]

Table 6.3: The material parameters of extrusion case in nonlinear model

deformed mesh with equivalent strains distribution is shown in Fig. 6.19.

The values in the largest loading region were larger than those in the flow-
out area due to the elastic spring-back. The deformed mesh with equivalent
plastic strains is shown in Fig. 6.20. It is seen that the equivalent plastic
strains reach constant values from the deformed area.
The material properties play an important role in the deformation.
In Fig. 6.21 two deformed meshes have a different deformation when these
 6.2 Nonlinear Material Model 81

0.15
Y

0.1

0.05

Flow direction

X
−0.05
−1.2 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4

Figure 6.18: Deformed mesh with velocity vector plot

0.15

0.1

0.05
Y

Flow direction X

−0.05
−1.2 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4

0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26

Figure 6.19: Equivalent total strain distribution

0.15

0.1

0.05
Y

Flow direction
X

−0.05
−1.2 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4

0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18

Figure 6.20: Equivalent plastic strain distribution

82 Test Results

0.15

G=200Gpa
G=100Gpa

0.1
Y

0.05

Flow direction

−0.05
−1.2 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6

Figure 6.21: Equivalent total strains distribution

deformations were carried out with different bulk moduli.

6.3 Conclusions
In this chapter the developed steady state formulation was applied for differ-
ent material models, which can be used for small or large deformation cases.
As shown in these samples the elastic spring-back can be obtained directly
after flowing out.
Bibliography

[1] Belytschko, T., Liu, W.K., Moran, B.: Nonlinear finite elements for
continua and structures. John Wiley and Sons Ltd., England, 2001
[2] Simo, J.C.: A framework for finite strain elastoplasticity based on max-
imum plastic dissipation and the mutliplicative decomposition: part I
Continuum Formulation, Computer Methods in Applied Mechanics and
Engineering. Vol. 66, 199-219, 1988
[3] Simo, J.C.: A framework for finite strain elastoplasticity based on max-
imum plastic dissipation and the mulitplicative decomposition: part II
Computational Aspects,Computer Methods in Applied Mechanics and
Engineering. Vol. 68, 1-31, 1988
[4] Simo, J.C., Hughes, T.J.R.: Computational inelasticity, spring-Verlag,
New York, Inc., 241-261 301-307, 1988
[5] Bonet, J., Wood, R.D.: Nonlinear continuum mechanics for finite ele-
ment analysis, Cambridge University Press, 1997
[6] Balagangadhar, D., Tortorelli,D.A.: Design of large-deformation steady
elastoplastic manufacturing processes. Part I: A displacement-based ref-
erence frame formulation. International Journal for Numerical Methods
in Engineering. Vol. 49, 2000, 899-932
[10] Balagangadhar, D., Dorai, G.A., Tortorelli, D.A.: A displacement-based
reference frame formulation for steady-state thermo-elasto-plastic mate-
rial processes. International Journal of Solids Structures. Vol. 36, 1999,
2397-2416
[8] Belytschko, T., Liu, W.K., Moran, B.: Nonlinear finite elements for
continua and structures. John Wiley and Sons Ltd., England, 2001
[10] Yu, Y., Geijselaers, H.J.M., Huétink, J.: A new displacement based FE
formulation for steady state problems, Simulation of Materials Process-
ing: Theory, Methods and Applications,Mori (Eds), Proceedings of NU-
MIFORM 2001, Toyohashi, Japan, 2001, 121-126
[10] Yu, Y.: A new displacement based FE formulation for steady state
problems, Intern report of N.I.M.R (P.00.4.021), 2000
Chapter 7

Conclusions and Recommendations

An end and also a beginning

7.1 Conclusions

The aim of the research was to develop a new displacement based formulation
for steady state flow processes and to test this formulation using the finite
element method. In our study the material evolution equation of any state
variable for steady state is derived as:

f˙ = v 0 · ∇0 f (7.1)

This is a key equation for the development of our new displacement based
formulation for the steady state problems. Even though in Balagangdhar’s
work (See Chapter 3) the material evolution is also related to the known
velocity field, the material evolution equation must be transformed to that
reference configuration. This kind of performance makes the procedure com-
plicated and not easy to understand. However, in our displacement based
formulation, two basic equations (the equilibrium equation and the material
evolution equation) can be expressed in the undeformed configuration and
hence be solved more simply than in the work of Balagangdhar. In our case
the equations to be solved are uncoupled.
The developed formulation differs from the three common descriptions:
Lagrangian, Eulerian and ALE descriptions, which have been used widely
to model steady state flow processes in recent years. Compared to these
descriptions our method has the following advantages and differences:

1. A transient analysis is not required as the reference fields do not vary

with time in the steady state situation. The displacements and plastic
strains are the primary variables.
86 Conclusions and Recommendations

2. Elastic spring-back can be obtained directly because the plastic parame-

ter can be calculated from its rate form.
3. The material evolution equations in the reference configuration have a
similar expressions as those in the Eulerian configuration. Integration
of these equations does not require time integration, since these equa-
tions are dependent only on the inflow velocity and the undeformed
contact region. Therefore the integration of the material evolution
equations was carried out along the known streamlines. The free sur-
face corrections are unnecessary.
4. Contact problems cannot be treated as in the conventional way in the
Lagrangian, Eulerian and ALE formulations (Chpater 5).
In fact, this new method combines features of the Lagarangian and Eulerian
formualtions.
During the solution of this formulation using FEM, the following should
be taken into account:
1 One problem is how to obtain the time derivatives of the state variables.
When using the second gradient of the displacement ([1]) to directly ob-
tain the time derivatives of the state variables, a disagreement with the
exact solutions in the elastic deformation was found in our study (See
Section 5.2.3), especially for the shear stresses. Therefore, an improve-
ment was made, in which the time derivatives of the state variables
were calculated form the average nodal strains. The results were much
better than using the second gradient of the displacement, even though
there were still local deviations.
2. Mostly the streamline upwind Petrov-Galerkin (SUPG) method has been
applied to solve the convection equations in the literature. However, in
our work the Least Squares (LS) method was applied to solve the ma-
terial state evoultion equations because this method gives more stable
solutions than the SUPG method, as shown in Section 5.2.2.
3. A contact algorithm was developed in which the contact force is treated
as the external force, in our case in global equilibrium equation. This
contact method is different from the contact methods in transient cal-
culations or impact problems.
Overall, it should be mentioned that this research field is still new and
more detailed work must still to be carried out because the current research
is restricted to simple geometry, elastoplasticity, and a rigid tool contact.
However, this formulation provides the researcher and the engineer with a
new strategy in simulation of steady state flow processes.
 7.2 Recommendations 87

7.2 Recommendations
As mentioned above, our current research presented some limits and prob-
lems as follows:
1. In our work, a simple tool geometry was applied. In order to apply this
steady state formulation in real industrial process modelling, a more
complicated tool should be included in further research.
2. The elastic plastic material model was used in our work, however, as
we know visco-elastoplastic material models are used widely in metal
forming process simulation. It is important that this kind of model is
implemented in the developed steady state formulation.
3. How to apply a 3D analysis to this formulation is also of great interest.
4. The contact problem is always a big issue which should not be ignored.
In our work the tool is assumed rigid and smooth, and hence the contact
algorithm must be improved for the non-rigid problem.
5. Other steady manufacturing processes should be tested using this for-
mulation.
Bibliography

[1] Balagangadhar, D.: A reference frame formulation for the analysis and
design of steady manufacturing processes, PhD thesis, UIUC, USA, 1999
[2] Yu, Y., Geijselaers, H.J.M., Huétink, J. : A new displacement based FE
formulation for steady state problems, Simulation of Materials Process-
ing: Theory, Methods and Applications, Mori (Eds), Proceedings of
NUMIFORM 2001, Toyohashi, Japan, 2001, 121-126
Appendix A

Proof of Equation ( 3.45)

X and r represent two Cartesian coordinate systems, with related to:

X = R(t)r + XT (t) (A.1)

in which r is the rotation tensor and XT is the transition vector. In contrast,

r = R−1 (t)(X − XT (t)) (A.2)

Therefore:
∂r ∂r ∂X
|X + |r
∂t ∂X ∂t
−1 −1
= R−1 (Ṙr + ẊT ) + (Ṙ X − Ṙ XT − Ṙ−1 ẊT )
−1
= R−1 Ṙr + Ṙ X − Ṙ−1 XT
−1 (A.3)
= R−1 Ṙr + Ṙ (Rr + XT ) − Ṙ−1 XT
−1
= R−1 Ṙr + Ṙ Rr
−1 ·
= (R R) r
=0
Appendix B

Voigt Notation

In finite element implementations, symmetric second-order tensors (e.g. for

the stresses and the strains) are written as column matrices in order to
simplify the tensor calculations. Any other conversion procedure of higher-
order tensors to matrices is called the Voigt notation.
For example, the material tangent moduli tensor CSijkl is a fourth order
tensor and it complicates the programming procedure. In index notation the
linear elastic law is written as:

Sij = CSijkl Ekl (B.1)

The Voigt matrix form of the above is:

Sa = CSab Eb (B.2)
94 Voigt Notation

where a ← ij and b → kl as in Table A2.1 in plane strain case.

CSijkl CSab
ij kl a b
11 11 1 1
11 22 1 2
11 33 1 3
11 12 1 4
22 11 2 1
22 22 2 2
22 33 2 3
22 12 2 4
33 11 3 1
33 22 3 2
33 33 3 3
33 12 3 4
12 11 4 1
12 22 4 2
12 33 4 3
12 12 4 4
Table A2.1 voigt rule for plane strain
Appendix C

dN, ∇udN, ∇u∇udN

See the figure below,

n T
dN = · b1 a (C.1)
knk
n = M · b1 b2 (C.2)
M is a rotating matrix to obtain the normal vector. b1 a represents a vector
from point b1 to point a:
b1 a = a − b1 (C.3)

Therefore:
n T n
∇u dN = · ∇u ab1 + ab1 T · ∇u (C.4)
knk knk
since a is a linear function of u:
n T n
∇∇u dN T ≈ ∇u · ∇u ab1 + ∇u ab1 T · ∇u (C.5)
knk knk

Body 1
b1 b2

a dN
Body 2
a

Figure C.1: Digram illustrating the gap definition between two contacting bodies
Appendix D

dt, ∇udt for Impact Problem

The contact slave node a of body 1 moves from calculation step i to cal-
culation step i + 1. At every step the projection of a on body 2 can be
calculated. The slip of this slave node will be zero within one step and only
occurs between two calculation steps. Therefore the slip distance can be
treated as the distance between the projections of a on body 2 during two
steps. Body 2 is divided into elements in the finite element method. There-
fore, the gradient of dt and the second order difference of dt can be obtained
through several tensor calculations as follows(see Fig. D.1): Step i:
1
ᾱ = b̄k ā · b̄k b̄k+1 ᾱ ∈ [0, 1] (D.1)
kb̄k b̄k+1 k
The overbar denotes the quantities related to the configuration of step i.
Step i + 1:
Body 2 changes its position related to reference configuration (due to
stretch or shrink), and this kind of change should be considered while the
slip is calculated.
The projection of a in step i on Body 2 will also change due to stretching

Figure D.1: Schematic depicting slip vector between two bodies

98 dt, ∇u dt for Impact Problem

or shrinking of body 2 in step i+1, but the relative position of the projection
in bk bk+1 will be the same as in step i, therefore:

kbk ā · bk bk+1 k
= ᾱ (D.2)
bk bk+1

The coordinate of the projection of a on body 2 in step i (xia−b ) related to

the configuration in step i + 1 can be calculated as:

xia−b = (1 − ᾱ)xbk + ᾱxbk+1 (D.3)

Further on, in step i + 1, a moves to the new situation due to deformation

from step i to step i + 1, the coordinate of projection a in step i + 1 can be
expressed as:
ni+1
xi+1
a−b = xa − dN
i+1
· i+1 (D.4)
kn k
The slip vector can be obtained:
n
dt = xi+1 i
a−b − xa−b = xa − dN · − (1 − ᾱ)xb1 − ᾱxb2 (D.5)
knk
The size of dt is: p
dt = kdtk = dtT · dt (D.6)
The gradient of dt is:
dtT · ∇u dt
∇u dt = (D.7)
dt
with:
n
∇u dt = ∇u xa − ∇u (dN · ) − (1 − ᾱ)∇u xb1 − ᾱ∇u xb2 (D.8)
knk
n n n
∇u (dN · ) = dN · ∇u + ⊗ ∇u dN (D.9)
knk knk knk
Furthermore, the second derivative of the slip distance should be known for
the stiffness matrix:
dt T dt
∇(∇u dt) = · ∇(∇u dt) + (∇u dt)T · (∇u ) (D.10)
dt dt
The first term can be treated as zero due to the linearity of the slip vector.
Therefore:
∇u dt 1
∇(∇u dt) = (∇u dt)T · ( + dt ⊗ ∇u ) (D.11)
dt dt
List of Symbols

Scalars

E Modulus of elasticity
G Shear modulus
t Time
f An arbitrary history-dependent field variable
φ Yield function
J Jocobian
K Bulk modulus
γ̇ Plastic parameter
Ω Volume
Ω0 Initial volume
Γ Surface
Γ0 Initial surface
V Volume
ρ Density
dN Normal distance between two bodies
W Power
λ Lam constant
µ Lam constant, friction cofficient
ψ̂ Stored-energy function
ν Poisson’s ratio
κ Bulk modulus

Tensors
Vectors

b Body force
q Internal variables
100 list of symbols

t Traction
tc Contact force
u Displacement
n Normal vector
v Velocity
x A material point in the current configuration
X A material point in the reference configuration

2nd order tensors

e Eulerian or Almansi strain tensor

C Cauchy-Green strain tensor
D Rate of deformation
E Green (Green-Lagrange) strain tensor
F Deformation tensor
I Unit tensor
L Velocity tensor
P First Piola-Kirchhoff stress tensor
Q Rotation tensor
S Second Piola-Kirchhoff stress tensor
W Spin tensor
τ Kirchhoff stress tensor
ε Linear strain tensor
σ Cauchy stress tensor

4th order tensor

Cel Linear elasticity tensor

Y Yield tensor

Operators

φ Mapping operator of points from

the material domain to the spatial domain
φ̄ Mapping function of points from
the reference domain to the spatial domain
 list of symbols 101

Φ Mapping function of points from

the material domain to the reference domain
∇ Gradient with respect to current coordinates
∇0 Gradient with respect to reference coordinates
∇u Gradient with respect to displacements u
β̇ Material time derivative of β
β −1 Inverse of β
βT Transpose of β
· Tensor contraction
: Double tensor contraction
kβk Two-norm of β

Sub- and superscripts

i, j Index
c Contact issue
e Elastic
i Body i
n Normal
t Tangential
0 Initial
x Concerning the deformed configuration
X Concerning the undeformed configuration

Others

V0 Undeformed domain
Vm Referential domain
V Deformed domain
tr(A) Trace of matrix A
K Stiffness matrix
Acknowledgements

This project is funded by the Netherlands Institute for Metals Research

(NIMR), and takes place within the group of Technical Mechanics of the
University of Twente. Their assistance is greatly acknowledged and appre-
ciated.
Many persons have contributed to this thesis:
First I would like to express my sincere appreciation to Han Huétink for
offering me the opportunity to do this research. He helped me to better un-
derstand the theory of Continuum Mechanics and provided many valuable
suggestions and comments to this research.
This thesis could not have been written without Bert Geijselaers who
not only served as my supervisor but also encouraged and challenged me
throughout this project. He provided me with his insight, recommendations
to this research and guidance on writing this thesis. I thank him.
I am grateful to all my colleagues for making my stay at the university
of Twente a memorable and valuable experience. Getjan Kloosterman gave
significant help on the contact algorithm in this thesis. In addition to that
we exchanged lots of views about the world and life in general. The com-
puter assistance and support of Herman Corbach and Nico van Vliet are
gratefully acknowledged. I am indebted for the administrative support from
Annnemarie Teunissen, Debbie Vrieze, Jacqueline Emmerich and Tanja Ger-
ritsen. Especially to Tanja who helped greatly on my documentation work
after I moved to Eindhoven two years ago. Also my thanks to Harm Wis-
selink, who dedicated time reading part of my thesis and offered comments
for improvements, to Ton van den Boogaard, who provided me with the lay-
out template of LATEX and to Katrina Emmett, who checked the English
language of this manuscript.
On the home front, I would like to give thanks to my friends in Delft,
Enschede and Eindhoven for having made my life more enjoyable. The most
pleasant times were playing cards in the weekends, even during my preg-
nancy.
Last but not least, I would like to dedicate this thesis to my family for
their never-ending support. My two sisters who have always supported me
104 acknowledgements

from my childhood. Without the encouragements and love from my parents,

this thesis would not have been completed. Papa, I know you are happy to
see this come true. Now my words go to my husband Liujin Tang. Thanks
for your love, understanding and the happy time you made for me in my life.
You always encouraged and supported me to finish this thesis. My lovely
little son, Charlie, I admire what you have done and achieved in the last
two years (better than your mom). Your pure smile had always refreshed
me while writing this thesis.

Yuhong Yu
Eindhoven, November 2004