This page intentionally left blank
Introduction to
Computational Plasticity
FIONN DUNNE AND NIK PETRINIC
Department of Engineering Science
Oxford University, UK
1
3
Great Clarendon Street, Oxford ox2 6dp
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford NewYork
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
NewDelhi Shanghai Taipei Toronto
With ofﬁces in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
© Oxford University Press, 2005
The moral rights of the authors have been asserted
Database right Oxford University Press (maker)
First published 2005
Reprinted 2006 (with corrections)
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,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose the same condition on any acquirer
British Library Cataloguing in Publication Data
(Data available)
Library of Congress Cataloging in Publication Data
(Data available)
Typeset by Newgen Imaging Systems (P) Ltd., Chennai, India
Printed in Great Britain
on acidfree paper by
Biddles Ltd, King’s Lynn
ISBN 0198568266 (Hbk) 9780198568261
3 5 7 9 10 8 6 4 2
To Hannah and Roberta, with love
This page intentionally left blank
Preface
The intention of this book is to bridge the gap between undergraduate texts in
engineering plasticity and the many excellent books in computational plasticity aimed
at more senior graduate students, researchers, and practising engineers working in
solid mechanics. The book is in two parts. The ﬁrst introduces microplasticity and
covers continuum plasticity, the kinematics of large deformations and continuum
mechanics, the ﬁnite element method, implicit and explicit integration of plasticity
constitutive equations, and the implementation of the constitutive equations, and the
associated material Jacobian, into ﬁnite element software. In particular, the implemen
tation into the commercial code ABAQUSis addressed (and to help, we provide a range
of ABAQUS material model UMATs), together, importantly, with the tests necessary
to verify the implementation. Our intention, wherever possible, is to develop a good
physical feel for the plasticity models and equations described by considering, at
every stage, the simpliﬁcation of the equations to uniaxial conditions. In addition, we
hope to provide a reasonably physical understanding of some of the large deformation
quantities (such as the continuum spin) and concepts (such as objectivity) which are
often unfamiliar to many undergraduate engineering students who demand more than
just a mathematical description.
The second part of the book introduces a range of plasticity models including those
for superplasticity, porous plasticity, creep, cyclic plasticity, and thermomechanical
fatigue (TMF). We also describe a number of practical applications of the plasticity
models introduced to demonstrate the reasonable maturity of continuum plasticity in
engineering practice.
We hope, above all, that this book will help all those—undergraduates, graduates,
researchers, and practising engineers—who need to move on from knowledge of
undergraduate plasticity to modern practice in computational plasticity. Our aims
have been to encourage development of understanding, and ease of passage to the
more advanced texts on computational plasticity.
September 2004 F. P. E. D. and N. P.
This page intentionally left blank
Contents
Acknowledgements xii
Notation xiii
Part I. Microplasticity and continuum plasticity
1. Microplasticity 3
1.1 Introduction 3
1.2 Crystal slip 5
1.3 Critical resolved shear stress 7
1.4 Dislocations 8
Further reading 10
2. Continuum plasticity 11
2.1 Introduction 11
2.2 Some preliminaries 11
2.3 Yield criterion 17
2.4 Isotropic hardening 23
2.5 Kinematic hardening 27
2.6 Combined isotropic and kinematic hardening 36
2.7 Viscoplasticity and creep 38
Further reading 45
3. Kinematics of large deformations and continuum mechanics 47
3.1 Introduction 47
3.2 The deformation gradient 48
3.3 Measures of strain 49
3.4 Interpretation of strain measures 52
3.5 Polar decomposition 57
x Contents
3.6 Velocity gradient, rate of deformation, and continuum spin 60
3.7 Elastic–plastic coupling 66
3.8 Objective stress rates 69
3.9 Summary 81
Further reading 82
4. The ﬁnite element method for static and dynamic plasticity 83
4.1 Introduction 83
4.2 Hamilton’s principle 84
4.3 Introduction to the ﬁnite element method 96
4.4 Finite element equilibrium equations 100
4.5 Integration of momentum balance and equilibrium equations 136
Further reading 142
5. Implicit and explicit integration of von Mises plasticity 143
5.1 Introduction 143
5.2 Implicit and explicit integration of constitutive equations 143
5.3 Material Jacobian 150
5.4 Kinematic hardening 154
5.5 Implicit integration in viscoplasticity 161
5.6 Incrementally objective integration for large deformations 167
Further reading 168
6. Implementation of plasticity models into ﬁnite element code 169
6.1 Introduction 169
6.2 Elasticity implementation 170
6.3 Veriﬁcation of implementations 171
6.4 Isotropic hardening plasticity implementation 172
6.5 Large deformation implementations 176
6.6 Elastoviscoplasticity implementation 180
Part II. Plasticity models
7. Superplasticity 185
7.1 Introduction 185
7.2 Some properties of superplastic alloys 185
Contents xi
7.3 Constitutive equations for superplasticity 189
7.4 Multiaxial constitutive equations and applications 192
References 197
8. Porous plasticity 199
8.1 Introduction 199
8.2 Finite element implementation of the porous material
constitutive equations 201
8.3 Application to consolidation of Ti–MMCs 205
References 207
9. Creep in an aeroengine combustor material 209
9.1 Introduction 209
9.2 Physically based constitutive equations 209
9.3 Multiaxial implementation into ABAQUS 212
References 217
Appendix 9.1 218
10. Cyclic plasticity, creep, and TMF 219
10.1 Introduction 219
10.2 Constitutive equations for cyclic plasticity 219
10.3 Constitutive equations for C263 undergoing TMF 222
References 227
Appendix A: Elements of tensor algebra 229
Differentiation 231
The chain rule 232
Rotation 233
Appendix B: Fortran coding available via the OUP website 235
Index 239
Acknowledgements
The authors would like to express their sincere gratitude to Esteban Busso for reading
a draft and providing many helpful comments and suggestions, to Paul Buckley for
the provision of the ﬁgures in Chapter 1, and to Jinguo Lin for permission to use
Figures 7.9–7.14.
The authors acknowledge, with gratitude, permission granted to reproduce the
following ﬁgures:
Figures 7.9–7.14: Elsevier Ltd, Oxford, UK.
Figures 8.1–8.6: Institute of Materials Communications Ltd, London, UK.
Figures 9.1–9.4, 10.1–10.3: Elsevier Ltd, Oxford, UK.
Notation
• Regular italic typeface (v, σ, . . .): scalars, scalar functions.
• Bold italic typeface (P, v, A, σ, . . .): points, vectors, tensors, vector and tensor
functions.
• Helvetica bold italic typeface (C, c, I, . . .): fourth order tensors.
Operations
f (·) function of (·)
det[·] determinant of [·]
Tr[·] trace of [·]
ln(·) logarithm of (·)
[·] increment of [·]
∂
∂x
[·] partial derivative of [·] with respect to x
∇(·) = grad[·] gradient of [·]
div[·] = tr[∇(·)] divergence of [·]
x · y scalar product of vectors
x ⊗y dyadic product of vectors
σ : ε double contraction of tensors
u =
√
u · u norm of vector
A =
√
A : A norm of tensor
Some commonly used notation
C fourthorder tensor of material constants
D rate of deformation tensor
E Lagrangian strain tensor
E Young’s modulus
ε strain tensor
xiv Notation
F deformation gradient
f force vector ﬁeld
ρ density
I secondorder identity tensor
J Jacobian
K stiffness matrix
M mass matrix
ν Poisson’s ratio
P material particle
P material point ∈ R
n
R rotation tensor
R real set
σ Cauchy stress tensor
t time
t surface traction vector
u displacement vector ﬁeld
˙ u velocity vector ﬁeld
¨ u acceleration vector ﬁeld
W work
Part I. Microplasticity and
continuum plasticity
This page intentionally left blank
1. Microplasticity
1.1 Introduction
This chapter brieﬂy introduces the origins of yield and plastic ﬂow, and in particular,
attempts to explain the usual assumptions in simple continuum plasticity of isotropy,
incompressibility, and independence of hydrostatic stress. While short, we introduce
grains, crystal slip, slip systems, resolved shear stress, and dislocations; the minimum
knowledge of microplasticity for users of continuum plasticity.
The origin of plasticity in crystalline materials is crystal slip. Metals are usually
polycrystalline; that is, made up of many crystals in which atoms are stacked in
a regular array. A typical micrograph for a polycrystalline nickelbase superalloy is
shown in Fig. 1.1 in which the ‘crystal’ or grain boundaries can be seen. The grain size
is about 100 µm. The grain boundaries demarcate regions of different crystallographic
orientation.
If we represent the crystallographic structure of a tiny region of a single grain by
planes of atoms, as shown in Fig. 1.2(a), we can then visualize plastic deformation
taking place as shown in Fig. 1.2(a) and (b); this is crystallographic slip. Unlike
elastic deformation, involving only the stretching of interatomic bonds, slip requires
the breaking and reforming of interatomic bonds and the motion of one plane of
atoms relative to another. After shearing the crystal from conﬁguration 1.2(a) to
conﬁguration 1.2(b), the structure is unchanged except at the extremities of the crystal.
Anumber of very important phenomena in macroscopic plasticity become apparent
from just two Figs 1.1 and 1.2:
(1) plastic slip does not lead to volume change; this is the incompressibility condition
of plasticity;
(2) plastic slip is a shearing process; hydrostatic stress, at the macrolevel, can often
be assumed not to inﬂuence slip;
(3) in a polycrystal, plastic yielding is often an isotropic process.
As we will see later, the incompressibility condition is very important in macro
scale plasticity and manifests itself at the heart of constitutive equations for plasticity.
4 Microplasticity
100 mm
Fig. 1.1 Micrograph of polycrystal nickelbase alloy C263.
τ τ
τ τ
(a) (b)
Fig. 1.2 Schematic representation of the crystallographic structure within a single grain undergoing
slip.
However, not all plastic deformation processes are incompressible. A porous metal,
for example, under compressive load may undergo plastic deformation during which
the pores reduce in size. Consequently, there is a change of volume and a dependence
on hydrostatic stress. However, the volume change does not originate from the plastic
slip process itself, but from the pore closure.
The fact that plastic slip is a shearing process gives more information about the
nature of yielding; in principle, it tells us that plastic deformation is independent of
hydrostatic stress (pressure). For nonporous metals, this is one of the cornerstones
of yield criteria. The von Mises criterion, for example, is one in which the initiation
of macroscale yield is quite independent of hydrostatic stress. If we take a sample of
the theoretical material shown schematically in Fig. 1.2(a) and submerge it to an ever
deeper depth in an imaginary sea of water, the hydrostatic stress becomes ever larger
but causes no more than the atoms in the theoretical material to come closer together.
It will never in itself be able to generate the shearing necessary for crystallographic slip.
Figure 1.1 shows a micrograph of a polycrystal. If we assume that there is no pre
ferred crystallographic orientation, but that the orientation changes randomly fromone
Crystal slip 5
(a) (b)
Fig. 1.3 (a) Photograph of a single zinc crystal and (b) a schematic diagram representing single slip
in a single crystal.
grain to the next, and if our sample of material contains a sufﬁciently large num
ber of grains, we can get a reasonable physical feel that macroscale yielding of
the material will be isotropic. This is a further cornerstone of the von Mises yield
criterion.
1.2 Crystal slip
The evidence for crystal slip being the origin of plasticity comes from mechanical
tests carried out on single crystals of metals.
The single crystal of zinc shown in Fig. 1.3(a) is a few millimetres in width and has
been loaded beyond yield in tension. The planes that can be seen are those on which
slip has occurred resulting from many hundreds of dislocations running through the
crystal and emerging at the edge. Each dislocation contributes just one Burger’s vector
of relative displacement, but with many such dislocations, the displacements become
large. Figure 1.3(b) shows schematically what is happening in Fig. 1.3(a). The ends of
the test sample have not been constrained in the lateral directions. It can be seen that
single slip in this case leads to the horizontal displacement of one end relative to
the other. Had this test been carried out in a conventional uniaxial testing machine,
the lateral motion would have been prevented. In order to retain compatibility, then,
with the imposed axial displacement, the slip planes would have to rotate towards the
loading direction. The uniaxial loading therefore leads not only to crystallographic
slip, but also to rotation of the crystallographic lattice.
1.2.1 Slip systems: slip directions and slip normals
Observations on single crystals show that slip tends to occur preferentially on certain
crystal planes and in certain speciﬁc crystal directions. The combination of a slip
6 Microplasticity
(111)
[11
–
0]
Fig. 1.4 The particular slip system (111)[1
¯
10] in an fcc lattice.
(101)
(112) (123)
(111
–
)
Fig. 1.5 Slip systems in bcc materials.
plane and a slip direction is called a slip system. These tend to be the most densely
packed planes and the directions in which the atoms are packed closest together.
This is explained in terms of dislocations. In face centred cubic (fcc) materials,
the most densely packed planes are the diagonal planes of the unit cell. In fact the
crystal is ‘closepacked’ in these planes; the observed slip systems are shown in
Fig. 1.4.
The full family of slip systems in such crystals may be written as 1
¯
10{111}.
There are 12 such systems in an fcc crystal (four planes each with three directions).
In body centre cubic (bcc) crystals, there are several planes that are of similar
density of packing, and hence there are several families of planes on which slip occurs.
However, there is no ambiguity about the slip direction, since the atoms are closest
along the [11
¯
1] direction and those equivalent to it. The slip systems observed in bcc
crystals are shown in Fig. 1.5.
Thus, there are three families of slip systems operative:
11
¯
1{101}; 11
¯
1{112}; 11
¯
1{123}.
Table 1.1 shows the slip systems found in single crystals of some of the important fcc
and bcc pure metals together with the resolved shear stress required to cause slip—the
critical resolved shear stress (CRSS).
Critical resolved shear stress 7
Table 1.1 Slip systems and CRSS for some pure metals.
Metal Structure Slip systems CRSS (MPa)
Cu fcc 1
¯
10{111} 0.64
Al fcc 1
¯
10{111} 0.40
Ni fcc 1
¯
10{111} 5.8
αFe bcc 11
¯
1{101}, 11
¯
1{112}, 11
¯
1{123} 32
Mo bcc 11
¯
1{101} 50
Ta bcc 11
¯
1{101}, 11
¯
1{112}, 11
¯
1{123} 50
Slip direction
Slip plane normal
t
n
s
s
l
f
Fig. 1.6 A single crystal containing slip plane with normal n, slip direction s, and loaded in
direction t.
1.3 Critical resolved shear stress
Suppose a single crystal in the shape of a rod is tested in tension. The axis of the rod
is parallel to unit vector t. The crystal has an active slip plane, normal in direction
of unit vector n. It has an active slip direction parallel to unit vector s, as shown in
Fig. 1.6.
When the applied tensile stress is σ, the shear stress acting on the slip plane and in
the slip direction is τ which may be found as follows: if the crosssectional area of
the rod is A, the force in the slip direction is Aσ cos λ and it acts on an area A/cos φ
of the slip plane. Hence the resolved shear stress is
τ = σ cos φ cos λ = σ(t · n)(t · s). (1.1)
Slip will take place on the slip system, that is, the crystal will yield, when τ reaches the
CRSS. This is known as Schmid’s law. The data shown in Fig. 1.7 were obtained from
tensile tests on cadmium single crystals which are hexagonal close packed (hcp). The
measured yield stress depends upon the angle between the tensile loading direction
and the basal plane. The minimum stress to cause yield occurs when the tensile axis is
45
◦
to this plane, that is, when the shear stress is maximized. Schmid’s law provides
a good explanation for the observed behaviour.
8 Microplasticity
7
6
5
4
3
2
1
0
0 30 60 90
Data
Schmid’s law
Angle between tensile axis and basal plane
Y
i
e
l
d
s
t
r
e
s
s
(
M
P
a
)
Fig. 1.7 The dependence of yield stress on the angle between the tensile axis and basal plane in an hcp
cadmium single crystal.
b
Fig. 1.8 An edge dislocation.
1.4 Dislocations
The theoretical shear strength of a crystal, calculated assuming that the shear is
homogeneous (the entire crystal shears simultaneously on one plane), is given by
τ
th
=
G
2π
.
The expected shear strength is therefore very large; many orders of magnitude greater
thanthe CRSSvalues are showninTable 1.1. The assumptionof homogeneous shear is,
of course, wrong and plastic deformation in crystals normally occurs by the movement
of the line defects known as dislocations, that are usually present in large numbers.
The glide of a dislocation involves only very local rearrangements of atoms close to it,
and requires a stress much lower than τ
th
, thereby explaining the low observed values
of CRSS.
Each dislocation is associated with a unit of slip displacement given by the Burgers
vector b. Since the dislocation is a line defect, there are two extreme cases. Figure 1.8
shows a schematic representation of an edge dislocation.
In this case the Burgers vector b is perpendicular to the line of the dislocation,
andthe dislocationcorresponds tothe edge of a missinghalfplane of atoms. Thus band
Dislocations 9
b
Fig. 1.9 A screw dislocation.
the dislocation line deﬁne a plane: a speciﬁc slip plane. If l is a unit vector parallel to
the dislocation line, then
s =
b
b
and n =
b ×l
b
. (1.2)
Figure 1.9 shows a schematic representation of a screw dislocation. This is the
other extreme case where the Burger’s vector is parallel to the dislocation line. The
dislocation itself corresponds to a line of scissorslike shearing of the crystal. It follows
from the fact that b is parallel to l, that, although the slip direction s is deﬁned, the slip
plane n is not. Hence, a screwdislocation can cause slip on any slip plane containing l.
Dislocations are therefore vital in understanding yield and plastic ﬂow.
We have said nothing about the important role of dislocations in strain hardening and
softening. Nor have we discussed the effect of temperature on diffusivity which inﬂu
ences or controls many plastic deformation mechanisms including thermally activated
dislocation climb leading to recovery, vacancy core and boundary diffusion, and grain
boundary sliding. All of these subjects can be found in existing materials text books,
some of which are listed below. Our aim has been to include sufﬁcient material on
microplasticity to ensure that the physical bases of at least some of the assumptions
made in macrolevel continuum plasticity are understood.
There have been many developments over the last 25 years in physically based
microplasticity modelling, including the development of timeindependent and rate
dependent crystal plasticity. Here, for a given crystallographic lattice (e.g. fcc),
using ﬁnite element techniques, the resolved shear stresses on all slip systems can
be determined to ﬁnd the active slip systems. Within either a timeindependent or
ratedependent formulation, the slip on each active system can be determined from
which the overall total deformation can be found. Such models are being used suc
cessfully for the plastic and creep deformation of single crystal materials which ﬁnd
application in aeroengines. The modelling of single crystal components has become
10 Microplasticity
possible with the development of highperformance computing. More recently, using
ﬁnite element techniques, polycrystal plasticity models have been developed. Here,
the understanding of microplasticity discussed above is also employed and again, as
in single crystal plasticity, the active slip systems can be identiﬁed and the corres
ponding slips determined to give the overall deformation within a given grain. This
is now done for all the grains in the polycrystal, which all have their own measured
or speciﬁed crystallographic orientations, subject to the requirements of equilibrium
and compatibility which are imposed within the ﬁnite element method. In order to do
this, it is necessary to generate many ﬁnite elements within each grain. It is clear that
for large numbers of grains, such polycrystal plasticity modelling becomes unten
able. The consequence is that while desirable, it is not (for the foreseeable future)
going to be feasible to carry out polycrystal plasticity modelling at the engineering
component level. Currently, andfor manyyears tocome, the plastic deformationoccur
ring in both the processing to produce engineering components, and occurring under
inservice conditions at localized regions of a component, will continue to be mod
elled using continuumlevel plasticity. This is particularly so in engineering industry
where pressures of time and cost demand rapid analyses. We will now, therefore, leave
microplasticity and address, in the remainder of the book, continuumlevel plasticity.
Further reading
Dieter, G.E. (1988). Mechanical Metallurgy. McGrawHill Book Co., London.
Meyers, M.A., Armstrong, R.W., and Kirchner, H.O.K. (1999). Mechanics and
Materials. Fundamentals and Linkages. John Wiley & Sons Inc., New York.
2. Continuum plasticity
2.1 Introduction
This chapter introduces the fundamentals of timeindependent and ratedependent
continuum, or phenomenological plasticity: multiaxial yield, normality hypothesis,
consistency condition, isotropic and kinematic hardening, and simple constitutive
equations for viscoplasticity and creep. We assume throughout the chapter that we
are dealing with small strain problems in the absence of large rigid body rotations.
The kinematics of large deformations are left until the following chapter.
2.2 Some preliminaries
2.2.1 Strain decomposition
Figure 2.1 shows the idealized stress–strain behaviour which might be obtained froma
purely uniaxial tensile test. Plasticity commences at a uniaxial stress of σ
y
, after which
the material strain hardens. It is called hardening because the stress is increasing
relative to perfect plastic behaviour, also shown in the ﬁgure. If, at a strain of ε,
the loading were to be reversed, the material would cease to deform plastically (at
least in the absence of timedependent effects) and would show a linearly decreasing
stress with strain such that the gradient of this part of the stress–strain curve would
«
«
p
«
e
E
s
s
s
y
E
Linear strain hardening
Perfect plasticity
Fig. 2.1 The classical decomposition of strain into elastic and plastic parts.
12 Continuum plasticity
again be the Young’s modulus, E, shown in Fig. 2.1. Once a stress of zero is achieved
(provided the material remains elastic on full reversal of the load), the strain remaining
in the test specimen is the plastic strain, ε
p
. The recovered strain, ε
e
, is the elastic
strain and it can be seen that the total strain, ε, is the sum of the two
ε = ε
e
+ ε
p
. (2.1)
This is called the classical additive decomposition of strain. It is also apparent from
Fig. 2.1 that the stress achieved at a strain of ε is given by
σ = Eε
e
= E(ε − ε
p
). (2.2)
In many practical situations, particularly in materials processing operations such as
forging or superplastic forming, for example, the strains achieved can be very large
indeed, and of order 1–4. Compare the magnitude of this strain with that of typical
elastic strains of order 0.001 (the 0.1% proof strain) which are generated in metals,
even in forming processes (you can estimate this from the measured forces to give
a stress, and the elastic strains are of order σ/E). In such circumstances, it is entirely
reasonable to make the assumption that ε
e
≈ 0 so that ε
p
≈ ε.
2.2.2 Incompressibility condition
We sawin Chapter 1 that plastic deformation satisﬁes the incompressibility condition;
that is, the deformation takes place without volume change. The consequence of this
is that the sum of the plastic strain rate components is zero:
˙ ε
p
X
+ ˙ ε
p
Y
+ ˙ ε
p
Z
= 0. (2.3)
This is easily veriﬁed by considering a cube of material, with dimensions shown in
Fig. 2.2, which undergoes purely plastic, uniform deformation (or simply argue that
the strains are very large so that the elastic components are negligibly small).
Constancy of volume requires
xyz = x
0
y
0
z
0
.
Differentiating both sides with respect to time, and dividing by xyz gives
˙ x
x
+
˙ y
y
+
˙ z
z
= 0. (2.4)
The strains are deﬁned by
ε
X
= ln
_
x
x
0
_
and similarly for ε
Y
and ε
Z
, and the strain rate in the Ydirection is therefore
˙ ε
Y
=
1
y
˙ y.
Equation (2.4) is therefore the incompressibility condition given in Equation (2.3).
Some preliminaries 13
z
0
z
x
0
x
y
y
0
Z
Y
X
Fig. 2.2 An element of material undergoing plastic, incompressible, elongation in the Ydirection.
2.2.3 Effective stress and plastic strain rate
Identifying the yield condition for uniaxial, monotonically increasing load is
straightforward:
If σ < σ
y
then the material is elastic,
If σ ≥ σ
y
then the material has yielded.
It is not quite so straightforward for a multiaxial stress state; that is, one in which
more than one direct stress exists. Awhole range of multiaxial yield criteria exist. The
most commonly used in engineering practice, particularly for computational analysis,
is that of von Mises (to which we will return later), which relies on the knowledge of an
effective stress, sometimes called (von Mises) equivalent stress. In terms of principal
stresses, the effective stress is deﬁned as
σ
e
=
1
√
2
[(σ
1
− σ
2
)
2
+ (σ
2
− σ
3
)
2
+ (σ
3
− σ
1
)
2
]
1/2
, (2.5)
or in terms of direct and shear stresses,
σ
e
=
_
3
2
(σ
2
11
+ σ
2
22
+ σ
2
33
+ 2σ
2
12
+ 2σ
2
23
+ 2σ
2
31
)
_
1/2
, (2.6)
where we use the numerical subscripts 1, 2, and 3 in an equivalent way to X, Y, and Z
to indicate direction. σ
e
is, of course, a scalar quantity and its origin lies in the postulate
that yielding occurs when a critical elastic shear energy is achieved. Mohr’s circle tells
us that in a given plane, the maximumshear stress is τ = (σ
1
−σ
2
)/2 so that with γ =
τ/G, the elastic shear energy per unit volume is τγ/2 = τ
2
/2G = (σ
1
− σ
2
)
2
/8G.
The origin of the terms in Equation (2.5) then becomes apparent. An effective plastic
strain rate, ˙ p, is deﬁned, similarly, as
˙ p =
√
2
3
[(˙ ε
p
1
− ˙ ε
p
2
)
2
+ (˙ ε
p
2
− ˙ ε
p
3
)
2
+ (˙ ε
p
3
− ˙ ε
p
1
)
2
]
1/2
. (2.7)
14 Continuum plasticity
Writing the stress and plastic strain rate tensors (dropping the superscript p on the
plastic strain rate components) as
σ =
⎛
⎝
σ
11
σ
12
σ
13
σ
21
σ
22
σ
23
σ
31
σ
32
σ
33
⎞
⎠
and ˙ ε =
⎛
⎝
˙ ε
11
˙ ε
12
˙ ε
13
˙ ε
21
˙ ε
22
˙ ε
23
˙ ε
31
˙ ε
32
˙ ε
33
⎞
⎠
(2.8)
then the effective stress and plastic strain rate may be written as
σ
e
=
_
3
2
σ
: σ
_
1/2
˙ p =
_
2
3
˙ ε
p
: ˙ ε
p
_
1/2
≈
_
2
3
˙ ε : ˙ ε
_
1/2
(2.9)
in which σ
is the deviatoric stress given by
σ
= σ −
1
3
Tr(σ)I, (2.10)
for example
σ
11
= σ
11
−
1
3
(σ
11
+ σ
22
+ σ
33
)
=
2
3
σ
11
−
1
3
(σ
22
+ σ
33
)
and
σ
12
= σ
12
− 0 ≡ σ
12
.
The deviatoric stress can be seen to be the difference between the stress and the mean
stress, σ
m
=
1
3
(σ
11
+ σ
22
+ σ
33
); the latter is often called the hydrostatic stress. The
symbol ‘:’ in Equations (2.9) is called the double contracted product, or double dot
product, of two secondorder tensors, which will be deﬁned a little later in the chapter.
The equations in (2.9) for the effective stress and plastic strain rates are a useful and
compact way of writing the equations given in (2.5) and (2.7), respectively.
The coefﬁcients in Equations (2.5) and (2.7) (and equivalently in Equations (2.9))
are chosen to ensure that under purely uniaxial loading, the effective stress σ
e
is
identical to the uniaxial stress and the effective plastic strain rate, ˙ p, is identical to the
uniaxial plastic strain rate.
Let us look at this in detail for the case in which a test specimen is loaded uniaxially
upto large plastic strain (so that ε
e
ε
p
, and ε ≈ ε
p
) under an applied stress σ
11
,
shown schematically in Fig. 2.3. We will now use Equation (2.9) to determine the
effective stress and plastic strain rate.
Some preliminaries 15
2
1
3
s
s
Fig. 2.3 Uniaxial loading of a schematic test piece.
In these circumstances, σ
11
= σ, σ
22
= σ
33
= 0, and all the shear components are
zero. The incompressibility condition leads to the requirement that
˙ ε
11
+ ˙ ε
22
+ ˙ ε
33
= 0. (2.11)
Because of symmetry, ˙ ε
22
= ˙ ε
33
so that incompressibility gives ˙ ε
22
= ˙ ε
33
=
−
1
2
˙ ε
11
= −
1
2
˙ ε. The deviatoric stress components can be determined using
Equation (2.10) as
σ
11
= σ
11
−
1
3
(σ
11
+ σ
22
+ σ
33
)
=
2
3
σ
11
σ
22
= σ
22
−
1
3
(σ
11
+ σ
22
+ σ
33
)
= −
1
3
σ
11
.
Similarly,
σ
33
= −
1
3
σ
11
and
σ
12
= σ
23
= σ
33
= 0
so that the deviatoric stress tensor becomes, for uniaxial loading,
σ
=
⎛
⎝
2
3
σ
11
0 0
0 −
1
3
σ
11
0
0 0 −
1
3
σ
11
⎞
⎠
. (2.12)
We can now use Equation (2.9) to determine the effective stress for this trivial case,
but we ﬁrst need to ﬁnd the contracted tensor product, σ
: σ
. This is deﬁned, for two
16 Continuum plasticity
secondorder tensors, A and B, by
A : B =
n
i=1
n
j=1
A
ij
B
ij
. (2.13)
That is, multiply component by component and sumthe terms to give a scalar quantity.
σ
: σ
is therefore given by
σ
: σ
=
4
9
σ
2
11
+
1
9
σ
2
11
+
1
9
σ
2
11
=
6
9
σ
2
11
≡
2
3
σ
2
and so
σ
e
=
_
3
2
σ
: σ
_
1/2
=
_
3
2
2
3
σ
2
_
1/2
≡ σ.
For a uniaxial stress state, therefore, σ
e
≡ σ. Let us follow the same procedure for
the effective plastic strain rate.
With the incompressibility condition and symmetry conditions discussed above the
plastic strain rate tensor becomes
˙ ε =
⎛
⎝
˙ ε
11
0 0
0 −
1
2
˙ ε
11
0
0 0 −
1
2
˙ ε
11
⎞
⎠
=
⎛
⎝
˙ ε 0 0
0 −
1
2
˙ ε 0
0 0 −
1
2
˙ ε
⎞
⎠
so that
˙ p =
_
2
3
˙ ε : ˙ ε
_
1/2
=
_
2
3
_
˙ ε
2
+
1
4
˙ ε
2
+
1
4
˙ ε
2
__
1/2
=
_
2
3
3
2
˙ ε
2
_
1/2
= ˙ ε.
For uniaxial loading, therefore, ˙ p = ˙ ε. Note that the plastic strain rates appearing
in Equation (2.9) are not given as deviatoric quantities, that is, ε
. This is because
the plastic strain rate components are, in themselves, deviatoric due to the incom
pressibility condition. For example, consider the deviatoric component ˙ ε
11
of plastic
strain rate.
˙ ε
11
= ˙ ε
11
−
1
3
(˙ ε
11
+ ˙ ε
22
+ ˙ ε
33
).
Because of the incompressibility condition, therefore,
˙ ε
11
≡ ˙ ε
11
.
Yield criterion 17
2.3 Yield criterion
Only the von Mises yield criterion is considered here. There are many others including
that of Tresca and the Gurson model for porous materials. In Chapter 1, we saw that
there were several general requirements for yield in isotropic, nonporous materials.
Let f be a yield function such that f = 0 is our yield criterion. Then:
1. Yield is independent of the hydrostatic stress.
Since f is independent of σ
m
, it must be expressible in terms of the deviatoric
stresses σ
i
(i = 1, . . . , 3) alone.
2. Yield in polycrystalline metals can be taken to be isotropic (provided we are
concerned with yield in a volume of material containing many grains) and must
therefore be independent of the labelling of the axes.
Thus f must be a symmetric function of σ
i
(i = 1, . . . , 3).
3. Yield stresses measured in compression have the same magnitude as yield stresses
measured in tension.
Thus f must be an even function of σ
i
(i = 1, . . . , 3).
The von Mises yield function is deﬁned by
f = σ
e
− σ
y
=
_
3
2
σ
: σ
_
1/2
− σ
y
(2.14)
and, with reference to Equation (2.5), it can be seen that it satisﬁes the three
requirements given above. The yield criterion is given by
f < 0: Elastic deformation
f = 0: Plastic deformation.
(2.15)
The second stress invariant, J
2
, is deﬁned as J
2
= (
1
2
σ : σ)
1/2
and it is for this
reason that plastic ﬂow based on the von Mises yield criterion is often referred to
as J
2
plasticity. Geometrically, Equation (2.14) corresponds to a cylinder in three
dimensional principal stress space with axis lying along the line σ
1
= σ
2
= σ
3
. It
is apparent from this that hydrostatic stress has no effect on yield according to the
von Mises criterion. Even inﬁnite, but equal, principal stresses σ
1
, σ
2
, and σ
3
will
never cause yield, since σ
e
remains zero (see Equation (2.5)) and f < 0.
Let us consider the yield function in twodimensional principal stress space by
putting σ
3
= 0 and so imposing conditions of plane stress. Geometrically, this
corresponds to ﬁnding the intersection between the von Mises cylinder and the
plane σ
3
= 0.
18 Continuum plasticity
d«
p
2
d«
p
1
d«
p
s
y
s
y
s
y
s
y
s
2
s
1
Tangent to
yield surface
Load point
Elastic region
Yield surface
( f =0)
Fig. 2.4 The von Mises yield surface for conditions of plane stress, showing the increment in plastic
strain, dε
p
, in a direction normal to the tangent to the surface.
From Equations (2.14) and (2.5),
f = σ
e
− σ
y
=
_
3
2
σ : σ
_
1/2
− σ
y
=
1
√
2
[(σ
1
− σ
2
)
2
+ (σ
2
− σ
3
)
2
+ (σ
3
− σ
1
)
2
]
1/2
− σ
y
,
which for plane stress, at yield, becomes
f =
1
√
2
[(σ
1
− σ
2
)
2
+ σ
2
2
+ σ
2
1
]
1/2
− σ
y
= 0 (2.16)
so
σ
2
1
− σ
1
σ
2
+ σ
2
2
= σ
2
y
,
which is the equation of an ellipse. The yield criterion is shown, for plane stress,
in Fig. 2.4. Naturally, when σ
1
= 0, σ
2
= σ
y
at yield and similarly for the other points
where the ellipse intersects the lines σ
1
= 0 and σ
2
= 0.
2.3.1 The normality hypothesis of plasticity
We have now looked at the conditions necessary to initiate yielding. The question
then is what happens after that if loading continues? After yield comes plastic ﬂow
and the normality hypothesis of plasticity enables us to determine the ‘direction’ of
ﬂow. For what is termed associated ﬂow, the hypothesis states that the increment in
the plastic strain tensor is in a direction (i.e. relative to the principal stress directions)
which is normal to the tangent to the yield surface at the load point. This is shown
schematically in Fig. 2.4, but may be written in terms of the yield function, f , as
dε
p
= dλ
∂f
∂σ
or ˙ ε
p
=
˙
λ
∂f
∂σ
. (2.17)
Yield criterion 19
In these expressions, the direction of the plastic strain increment (or equivalently,
plastic strain rate) is given by ∂f /∂σ while the magnitude is determined, for the
plastic strain rate, by
˙
λ. This is called the plastic multiplier which we shall return
to later.
In order to look at this in more detail, and to make understanding easier, let us write
the plastic strain rate and stress tensors in vector form (using what is known as Voigt
notation, which we shall address in greater depth in Chapter 4). We will continue to
consider principal components only, for the time being. Then for plane stress,
σ =
_
σ
1
σ
2
_
.
Note that the vector representation of the stress, σ, is not italicized. This is the con
vention to be adopted throughout the book for stress and strain: an italicized bold σ
and ε indicates a tensor, a nonitalicized bold symbol a vector, or Voigt notation. The
yield function f , written in terms of principal stresses is, from Equation (2.16)
f =
1
√
2
[(σ
1
− σ
2
)
2
+ σ
2
2
+ σ
2
1
]
1/2
− σ
y
= 0,
and we can then determine the direction of plastic ﬂow from Equation (2.17) as
∂f
∂σ
= grad(f ) =
⎛
⎜
⎜
⎝
∂f
∂σ
1
∂f
∂σ
2
⎞
⎟
⎟
⎠
=
_
1
2
(σ
2
1
− σ
1
σ
2
+ σ
2
2
)
−1/2
(2σ
1
− σ
2
)
1
2
(σ
2
1
− σ
1
σ
2
+ σ
2
2
)
−1/2
(2σ
2
− σ
1
)
_
,
which has direction
_
2σ
1
− σ
2
−σ
1
+ 2σ
2
_
. This is normal to the yield surface at all points
(e.g. choose σ
1
= σ
2
= α, say, and the direction of the normal is clearly along
the line σ
1
= σ
2
). Let us now derive the plastic strain increment using the normality
hypothesis with the von Mises yield criterion given in Equation (2.14), but using the
other expression for effective stress for threedimensions given in Equation (2.5). We
will look at just one component, namely dε
p
1
, to see the general pattern emerge. From
Equation (2.17), the ﬁrst component of direction of plastic ﬂow is given by
∂f
∂σ
1
=
1
2
1
√
2
[(σ
1
− σ
2
)
2
+ (σ
2
− σ
3
)
2
+ (σ
3
− σ
1
)
2
]
−1/2
[2(σ
1
− σ
2
) − 2(σ
3
− σ
1
)]
=
(3/2)[σ
1
− (1/3)(σ
1
+ σ
2
+ σ
3
)]
σ
e
so
∂f
∂σ
1
=
3
2
σ
1
σ
e
20 Continuum plasticity
remembering the deﬁnition of deviatoric stress given in Equation (2.10). Similar
results are obtained for the other components so that Equation (2.17) may be rewritten,
for the von Mises yield criterion, as
dε
p
= dλ
∂f
∂σ
=
3
2
dλ
σ
σ
e
(2.18)
in which the stress and strain are now written as tensor quantities. Let us now look
at the meaning of the plastic multiplier (at least, that is, for a von Mises material).
We can use the expression for effective plastic strain rate in Equation (2.9) to write
a similar expression for the increment in effective plastic strain as
dp =
_
2
3
dε
p
: dε
p
_
1/2
. (2.19)
If we substitute Equation (2.18) into (2.19) we obtain
dp =
_
2
3
3
2
dλ
σ
σ
e
:
3
2
dλ
σ
σ
e
_
1/2
= dλ
((3/2)σ
: σ
)
1/2
σ
e
so that with Equation (2.9) we obtain
dp = dλ (2.20)
and similarly,
˙ p =
˙
λ. (2.21)
So, for a von Mises material, the plastic multiplier, dλ, turns out simply to be
the increment in effective plastic strain. We can then rewrite the ﬂow rule in
Equation (2.18) as
dε
p
=
3
2
dp
σ
σ
e
. (2.22)
In order for Equation (2.22) to be useful to us, we need to be able to calculate
the increment in effective plastic strain, dp, or equivalently, the plastic multiplier,
so that with prescribed loading, we can then calculate the increment in plastic strain
components. We will address this next.
2.3.2 Consistency condition
Let us consider the case of uniaxial, tensile loading, for which the stress path taken
relative to the yield surface is shown in Fig. 2.5.
Starting from zero stress, σ
2
then increases while the material deforms elastically
until the load point (i.e. the point in stress space corresponding to the current loading)
meets the yield surface at σ
2
= σ
y
. At this point, the material behaves plastically,
Yield criterion 21
s
y
s
y
s
y
s
2
s
1
s
2
«
2
s
y
s
y
Yield surface
( f =0)
Load point
Fig. 2.5 The von Mises yield surface for plane stress and the corresponding stress–strain curve obtained
for uniaxial straining in the 2direction.
with no hardening, as shown in the resulting stress–strain curve in Fig. 2.5. With
further plastic deformation, the load point remains on the yield surface and the stress
remains constant and equal to σ
y
. The requirement for the load point to remain on
the yield surface during plastic deformation (at least, that is, for timeindependent
plasticity—we will look at viscoplasticity later) is called the consistency condition,
and it is this that enables us to determine the plastic multiplier, or equivalently for a
von Mises material, the increment in effective plastic strain.
The yield function given in Equation (2.14) includes a dependence on the stress
components and the yield stress, σ
y
. We will see later, when considering hardening,
that the yield stress can increase (and sometimes decrease), and does so often as
a function of effective plastic strain, p. It is therefore convenient to write the yield
function as
f (σ, p) = σ
e
− σ
y
= σ
e
(σ) − σ
y
(p) = 0. (2.23)
The consistency condition is written, for an incremental change in stress and effective
plastic strain
f (σ + dσ, p + dp) = 0. (2.24)
We can expand this as
f (σ + dσ, p + dp) = f (σ, p) +
∂f
∂σ
: dσ +
∂f
∂p
dp. (2.25)
Note that all terms in the equation are scalar. Let us work in principal stress space only.
The advantage of this is simplicity; the disadvantage is that we ignore the complicat
ing features of dealing with tensor versus engineering strain components. However,
this will be dealt with in Chapters 4–6. We will therefore store stress and strain
principal components as vectors, in Voigt notation, as discussed above. The product
22 Continuum plasticity
∂f /∂σ : dσ then becomes the scalar (dot) product, (∂f/∂σ)· dσ of two vectors (this
is only true in principal stress space, as we shall see later) which gives a scalar.
Combining Equations (2.23)–(2.25) gives
∂f
∂σ
· dσ +
∂f
∂p
dp = 0. (2.26)
We can use Hooke’s law in incremental form to relate the stress and elastic strains,
written as column vectors, by
dσ = C dε
e
= C(dε − dε
p
) (2.27)
in which C is the elastic stiffness matrix. We will use the most general form for the
plastic strain increment, given in Equation (2.17), for now, but simplify for the case
of the von Mises yield criterion later.
Substituting (2.17) into (2.27) gives
dσ = C
_
dε − dλ
∂f
∂σ
_
(2.28)
and (2.28) into (2.26)
∂f
∂σ
· C
_
dε − dλ
∂f
∂σ
_
+
∂f
∂p
dp = 0. (2.29)
We can obtain the most general form for dp, that is, without assuming a von Mises
material for the time being, by using Equations (2.17) and (2.19) to give
dp =
_
2
3
dε
p
: dε
p
_
1/2
=
_
2
3
dλ
∂f
∂σ
: dλ
∂f
∂σ
_
1/2
=
_
2
3
dλ
∂f
∂σ
· dλ
∂f
∂σ
_
1/2
(2.30)
for principal stress space, in which the contracted tensor product simpliﬁes to the scalar
product. Substituting (2.30) into (2.29) and rearranging gives the plastic multiplier dλ
dλ =
(∂f /∂σ) · C dε
(∂f /∂σ) · C(∂f /∂σ) − (∂f /∂p)((2/3)(∂f /∂σ) · (∂f /∂σ))
1/2
. (2.31)
The stress increment can then be determined by substituting (2.31) into (2.28) to give
dσ = C
_
dε −
∂f
∂σ
(∂f /∂σ) · C dε
(∂f/∂σ) · C(∂f/∂σ) − (∂f/∂p)((2/3)(∂f/∂σ) · (∂f/∂σ))
1/2
_
=
_
C −C
∂f
∂σ
(∂f/∂σ) · C
(∂f/∂σ) · C(∂f/∂σ) − (∂f/∂p)((2/3)(∂f/∂σ) · (∂f/∂σ))
1/2
_
dε
(2.32)
Isotropic hardening 23
or
dσ = C
ep
dε. (2.33)
C
ep
is called the tangential stiffness matrix. In the absence of plastic deformation,
dλ = 0, and in this case, C
ep
≡ C, the elastic stiffness matrix. In the case of plastic
deformation, with knowledge of the total strain increment, the stress increment can be
obtained from Equation (2.32). Before looking at the plastic multiplier in more detail,
to gain better physical insight, we will ﬁrst introduce isotropic hardening.
2.4 Isotropic hardening
Many metals, when deformed plastically, harden; that is, the stress required to cause
further plastic deformation increases, often as a function of accumulated plastic
strain, p, which can be written as
p =
_
dp =
_
˙ p dt , (2.34)
where ˙ p and dp are given in (2.9) and (2.19), respectively. A uniaxial stress–strain
curve with nonlinear hardening is shown in Fig. 2.6 together with schematic repres
entations of the initial and subsequent yield surfaces. In this instance, the subsequent
yield surface is shown expanded compared with the original. When the expansion is
uniform in all directions in stress space, the hardening is referred to as isotropic. Let
us just consider what happens in going from elastic behaviour to plastic, hardening
behaviour in Fig. 2.6.
Loading is in the 2direction, so the load point moves in the σ
2
direction from zero
until it meets the initial yield surface at σ
2
= σ
y
. Yield occurs at this point. In order
Saturation
Hardening
Initial yield
surface
Subsequent, expanded yield
surfaces after plastic deformation
Yield
s
2
s
1
s
2
s
y
s
y
s
y
r
s
y
«
2
s
y
s
y
Fig. 2.6 Isotropic hardening, in which the yield surface expands with plastic deformation, and the
corresponding uniaxial stress–strain curve.
24 Continuum plasticity
for hardening to take place, and for the load point to stay on the yield surface (the con
sistency condition requires this), the yield surface must expand as σ
2
increases, shown
in Fig. 2.6. The amount of expansion is often taken to be a function of accumulated
plastic strain, p, and for this case, the yield function becomes that given in (2.23),
f (σ, p) = σ
e
− σ
y
(p) = 0, (2.35)
where σ
y
(p) might be of the form
σ
y
(p) = σ
y0
+ r(p) (2.36)
in which σ
y0
is the initial yield stress and r(p) is called the isotropic hardening
function. There are many forms used for r(p) but a common one is
˙ r(p) = b(Q − r) ˙ p or dr(p) = b(Q − r) dp (2.37)
in which b and Q are material constants, which gives an exponential shape to the
uniaxial stress–strain curve which saturates with increasing plastic strain, since
integrating (2.37) with initial condition r(0) = 0, gives
r(p) = Q(1 − e
−bp
). (2.38)
Q is the saturated value of r so that the peak stress achieved with this kind of harden
ing, fromEquation (2.36), is therefore (σ
y0
+Q). The constant b determines the rate at
which saturation is achieved. Figure 2.6 shows an example of the uniaxial stress–strain
behaviour predicted using this kind of isotropic hardening function.
Let us now consider a slightly simpler form of isotropic hardening and determine
the plastic multiplier in Equation (2.31) and hence the stress increments in (2.32).
2.4.1 Linear isotropic hardening
We write the linear isotropic hardening function as
dr(p) = hdp (2.39)
in which h is a constant. With this hardening, we expect the uniaxial stress–strain
curve to look like that shown in Fig. 2.7. For uniaxial conditions, dp = dε
p
, and
referring to Figs 2.6 and 2.7, the stress increase due to isotropic hardening is just dr,
so using Equation (2.39) we may write
dε
p
=
dσ
h
and, of course, the increment in elastic strain is just
dε
e
=
dσ
E
Isotropic hardening 25
Gradient =E
E+h
E
(
1
)
r
s
y
s
y
«
«
p
s
–
Fig. 2.7 Stress–strain curve for linear strain hardening with dr = hdε
p
.
so that the total strain is
dε =
dσ
E
+
dσ
h
= dσ
_
E + h
Eh
_
,
giving
dσ = E
_
1 −
E
E + h
_
dε.
The plastic multiplier has been derived, in Equation (2.31), in terms of the total strain
increment. This is the most appropriate form when considering the development of
computational techniques, including the ﬁnite element method in which, often, the
total strain increments are provided, step by step, and it is then necessary to calculate
the correspondingstress increments. However, inorder toobtainbetter physical insight
into the process, it is more appropriate (and simpler) to write the plastic multiplier in
terms of a known stress increment. In this section, we will obtain the plastic multiplier
in terms of the stress increment so that we can then examine the equations for a simple
uniaxial problem using the von Mises yield criterion in Section 2.4.2.
Now, returning to the plastic multiplier and combining Equations (2.26) and (2.30),
we obtain
∂f
∂σ
· dσ +
∂f
∂p
dp =
∂f
∂σ
· dσ +
∂f
∂p
dλ
_
2
3
∂f
∂σ
·
∂f
∂σ
_
= 0. (2.40)
Rearranging (2.40) gives the plastic multiplier, but now in terms of the stress
increment, dσ, as
dλ =
−(∂f/∂σ) · dσ
(∂f/∂p)((2/3)(∂f/∂σ) · (∂f/∂σ))
. (2.41)
2.4.2 Uniaxial loading with linear isotropic hardening
Let us consider purely uniaxial loading, in the 1direction, for a material that yields
according to the von Mises criterion. In other words, we will apply a stress σ
1
in the
26 Continuum plasticity
1direction, and all other stresses are zero. For simplicity, we will continue to work in
principal stress space and represent the stresses in Voigt notation; that is, the stress is
written in vector form as
σ =
⎡
⎣
σ
1
σ
2
σ
3
⎤
⎦
.
We will now determine the plastic multiplier using Equation (2.41). We will consider
the denominator ﬁrst. With the hardening function given in (2.39), the yield function
becomes
f (σ, p) = σ
e
(σ) − σ
y
(p) = σ
e
(σ) − σ
y0
− r(p) = 0 (2.42)
from which we can obtain
∂f
∂p
= −
∂r
∂p
= −h. (2.43)
For the von Mises yield criterion, ∂f/∂σ can be obtained using Equation (2.18),
with the deviatoric stress components taken from (2.12), as
∂f
∂σ
=
3
2
σ
σ
e
=
3
2
1
σ
1
⎡
⎢
⎣
2
3
σ
1
−
1
3
σ
1
−
1
3
σ
1
⎤
⎥
⎦
=
⎡
⎢
⎣
1
−
1
2
−
1
2
⎤
⎥
⎦
so that
∂f
∂p
_
2
3
∂f
∂σ
·
∂f
∂σ
_
= −h
⎛
⎜
⎝
2
3
⎡
⎢
⎣
1
−
1
2
−
1
2
⎤
⎥
⎦
·
⎡
⎢
⎣
1
−
1
2
−
1
2
⎤
⎥
⎦
⎞
⎟
⎠
1/2
= −h. (2.44)
The numerator can be determined as follows:
−
∂f
∂σ
· dσ = −
⎡
⎢
⎣
1
−
1
2
−
1
2
⎤
⎥
⎦
·
⎡
⎣
dσ
1
dσ
2
dσ
3
⎤
⎦
= −
⎡
⎢
⎣
1
−
1
2
−
1
2
⎤
⎥
⎦
·
⎡
⎣
dσ
1
0
0
⎤
⎦
= dσ
1
(2.45)
and substituting (2.44) and (2.45) into (2.41) gives the plastic multiplier as
dλ =
dσ
1
h
. (2.46)
For a von Mises material, under uniaxial loading, therefore,
dλ = dp = dε
p
1
=
dσ
1
h
. (2.47)
This is telling us that the increment in stress for unit increase in plastic strain is h.
This arises because we chose linear hardening with gradient h. We know, of course,
Kinematic hardening 27
that because we are considering uniaxial loading and the incompressibility condition
applies, the other plastic strain increments must be
dε
p
2
= dε
p
3
= −
1
2
dε
p
1
= −
1
2
dσ
1
h
. (2.48)
We can, of course, obtain the other plastic strain components formally from
Equation (2.17) as
dε
p
= dλ
∂f
∂σ
=
dσ
1
h
⎡
⎢
⎣
1
−
1
2
−
1
2
⎤
⎥
⎦
(2.49)
in agreement with (2.48).
The total strain increment in the loading direction is given by
dε
1
= dε
e
1
+ dε
p
1
=
dσ
1
E
+
dσ
1
h
. (2.50)
Rearranging (2.50), and omitting the subscript 1 to indicate the uniaxial loading
direction, gives
dσ = E
_
1 −
E
E + h
_
dε, (2.51)
which is what we obtained earlier for linear strain hardening shown in Fig. 2.7.
Equation (2.51) shows the linear hardening obtained, under uniaxial loading, for the
isotropic hardening model chosen in Equation (2.39). If we choose perfect plasticity,
that is, with no hardening (h = 0), Equation (2.51) gives a stress increment of zero.
The plastic multiplier can, using (2.51) and (2.46), be rewritten as
dλ =
E
E + h
dε
1
. (2.52)
2.5 Kinematic hardening
In the case of monotonically increasing loading, it is often reasonable to assume that
any hardening that occurs is isotropic. For the case of reversed loading, however,
this is often not appropriate. Consider a material which hardens isotropically, shown
schematically in Fig. 2.8. At a strain of ε
i
, corresponding to load point (1) shown
in the ﬁgure, the load is reversed so that the material behaves elastically (the stress
is now lower than the yield stress) and linear stress–strain behaviour results up until
load point (2). At this point, the load point is again on the expanded yield surface, and
any further increase in load results in plastic deformation. Figure 2.8(b) shows that
isotropic hardening leads to a very large elastic region, on reversed loading, which is
often not what would be seen in experiments. In fact, a much smaller elastic region
28 Continuum plasticity
Load point (1)
s
2
s
1
«
2
«
i
s
2
s
y
s
y
s
y
s
y
s
y
Initial yield
surface
Subsequent, expanded
yield surface
Load point (2)
(a) (b)
E
E
Fig. 2.8 Reversed loading with isotropic hardening showing (a) the yield surface and (b) the resulting
stress–strain curve.
E
E
Load point (1)
s
2
«
2
s
2
s
1
s
y
s
y
s
y
s
y
s
y
Initial yield
surface
Subsequent, translated
yield surface
Load point (2)
(a)
(b)
 x
Fig. 2.9 Kinematic hardening showing (a) the translation, x of the yield surface with plastic strain,
and (b) the resulting stress–strain curve with shifted yield stress in compression—the Bauschinger effect.
is expected and this results from what is often called the Bauschinger effect, and
kinematic hardening. In kinematic hardening, the yield surface translates in stress
space, rather than expanding. This is shown in Fig. 2.9.
In Fig. 2.9(a), the stress increases until the yield stress, σ
y
, is achieved. With
continued loading, the material deforms plastically and the yield surface translates.
When load point (1) is achieved, the load is reversed so that the material deforms
elastically until point (2) is achieved when the load point is again in contact with the
yield surface. The elastic region is much smaller than that for isotropic hardening
shown in Fig. 2.8(b). In fact, for the kinematic hardening in Fig. 2.9, the elastic region
Kinematic hardening 29
is of size 2σ
y
whereas for the isotropic hardening, it is 2(σ
y
+r). In the case of plastic
ﬂowwith kinematic hardening, note that the consistency condition still holds; the load
point must always lie on the yield surface during plastic ﬂow. In addition, normality
still holds; the increment in plastic strain has direction normal to the tangent to the
yield surface at the load point.
The yield function describing the yield surface must now also depend on the
location of the surface in stress space. Consider the initial yield surface shown in
Fig. 2.9. Under applied loading and plastic deformation, the surface translates to the
new location shown such that the initial centre point has been translated by x. We
now need to determine the stresses relative to the new yield surface centre to check
for yield. In the absence of kinematic hardening, the yield function written in terms
of tensor stresses is
f = σ
e
− σ
y
=
_
3
2
σ
: σ
_
1/2
− σ
y
.
With kinematic hardening, however, it is
f =
_
3
2
(σ
−x
) : (σ
−x
)
_
1/2
− σ
y
(2.53)
in which x is the kinematic hardening variable and is often called a back stress.
Because it is a variable deﬁned in stress space, it has the same components as stress,
and we can write it as a tensor, or, using Voigt notation, as a vector. In order to under
stand Equation (2.53), let us consider purely uniaxial loading with linear kinematic
hardening.
2.5.1 Kinematic hardening under uniaxial loading
We will take the increment in kinematic hardening to be proportional to the increment
in plastic strain, hence
dx =
2
3
c dε
p
or equivalently, ˙ x =
2
3
c˙ ε
p
(2.54)
in which c is a material constant, and the coefﬁcient of
2
3
will be discussed below.
This is called Prager linear hardening. Equation (2.54) has the similarity with isotropic
hardening, that the hardening variable depends linearly on the plastic strain. The major
difference is that isotropic hardening is described by a scalar variable, r, whereas
for kinematic hardening, the hardening variable (or back stress) is a tensor, just as
for stress. Because of the incompressibility condition, we know that the plastic strain
increment is a deviatoric quantity, that is
dε
p
= dε
p
−
1
3
Tr(dε
p
) ≡ dε
p
. (2.55)
30 Continuum plasticity
From Equation (2.54), dx is therefore also a deviatoric quantity so that for uniaxial
loading in the 1direction, we may write
x = x
=
⎛
⎝
x
11
0 0
0 x
22
0
0 0 x
33
⎞
⎠
=
⎛
⎜
⎝
x
11
0 0
0 −
1
2
x
11
0
0 0 −
1
2
x
11
⎞
⎟
⎠
. (2.56)
The magnitude or norm of x is deﬁned by
x = x = (x : x)
1/2
(2.57)
=
⎡
⎢
⎣
⎛
⎜
⎝
x
11
0 0
0 −
1
2
x
11
0
0 0 −
1
2
x
11
⎞
⎟
⎠
:
⎛
⎜
⎝
x
11
0 0
0 −
1
2
x
11
0
0 0 −
1
2
x
11
⎞
⎟
⎠
⎤
⎥
⎦
1/2
=
__
x
2
11
+
1
4
x
2
11
+
1
4
x
2
11
__
1/2
=
¸
¸
¸
¸
3
2
x
11
¸
¸
¸
¸
(2.58)
so the uniaxial component of x, that is x
11
, is
2
3
times the magnitude of x. Bear
in mind that under uniaxial loading, the effective stress is identical to the uniaxial
applied stress, and the effective plastic strain increment is identical to the uniaxial
plastic strain increment. Note also that like the plastic strain, uniaxial loading leads
to the development of not only the uniaxial component of x, but also the other direct
components as well. This is because x depends directly on the plastic strain, given
in its evolution equation in (2.54). Consider now the components of dx, for uniaxial
loading, in Equation (2.54).
dx =
⎛
⎜
⎝
dx
11
0 0
0 −
1
2
dx
11
0
0 0 −
1
2
dx
11
⎞
⎟
⎠
=
2
3
c
⎛
⎜
⎝
dε
p
11
0 0
0 −
1
2
dε
p
11
0
0 0 −
1
2
dε
p
11
⎞
⎟
⎠
so that
dx
11
=
2
3
c dε
p
11
≡
2
3
c dp. (2.59)
It is often simpler, and in particular, for uniaxial loading, to write the equations in
terms of the magnitude of x rather than the loading direction component. Combining
Equations (2.59) and (2.58), therefore, gives
dx = c dε
p
11
≡ c dp, (2.60)
which is why the coefﬁcient of
2
3
appears in Equation (2.55).
Kinematic hardening 31
We may nowdetermine the terms necessary for the yield function in Equation (2.53)
for uniaxial loading as
σ
−x
=
⎛
⎜
⎝
2
3
σ
11
− x
11
0 0
0 −
1
2
_
2
3
σ
11
− x
11
_
0
0 0 −
1
2
_
2
3
σ
11
− x
11
_
⎞
⎟
⎠
(2.61)
which can be written in terms of the magnitude, x, of the back stress as
σ
−x
=
2
3
⎛
⎜
⎝
σ
11
− x 0 0
0 −
1
2
(σ
11
− x) 0
0 0 −
1
2
(σ
11
− x)
⎞
⎟
⎠
so that
(σ
−x
) : (σ
−x
) =
4
9
⎛
⎜
⎝
(σ
11
− x) 0 0
0 −
1
2
(σ
11
− x) 0
0 0 −
1
2
(σ
11
− x)
⎞
⎟
⎠
:
⎛
⎜
⎝
(σ
11
− x) 0 0
0 −
1
2
(σ
11
− x) 0
0 0 −
1
2
(σ
11
− x)
⎞
⎟
⎠
=
_
4
9
(σ
11
− x)
2
+
1
9
(σ
11
− x)
2
+
1
9
(σ
11
− x)
2
_
=
2
3
(σ
11
− x)
2
so that
f =
_
3
2
(σ
−x
) : (σ
−x
)
_
1/2
− σ
y
= σ
11
− x − σ
y
= 0. (2.62)
Because we are considering here only uniaxial loading, σ
11
is just the uniaxial
stress, σ, so Equation (2.62) can be written
f = σ − x − σ
y
= 0. (2.63)
The physical interpretation of Equation (2.63) can be seen in Fig. 2.9. Plastic deforma
tion leads to the translation of the yield surface in stress space. Under uniaxial
conditions, therefore, further yield occurs if σ − x is equal to the yield stress, σ
y
.
There are many forms of kinematic hardening available. What often distinguishes
them is how the direction of translation of the yield surface is chosen, and the rate
of its evolution as a function of the plastic strain. We shall now look at what is often
called Armstrong–Frederick, or Chaboche nonlinear kinematic hardening.
32 Continuum plasticity
2.5.2 Nonlinear kinematic hardening
In its multiaxial form, the increment, dx, in the back stress is given by
dx =
2
3
c dε
p
− γ x dp (2.64)
or equivalently,
˙ x =
2
3
c˙ ε
p
− γ x˙ p (2.65)
in which γ is a further material constant. In its uniaxial form, for monotonically
increasing plastic strain, Equation (2.64) may be written in terms of the magnitude,
x, as
dx = c dε
p
− γ x dε
p
,
which can be integrated, taking x to be 0 at ε
p
= 0, to give
x =
c
γ
(1 − e
−γ ε
p
). (2.66)
The resulting form of the stress–strain curve, for this nonlinear kinematic hardening,
is shown together with the translated yield surface, in Fig. 2.10. As the plastic strain
increases, so the back stress, x, in Equation (2.66) saturates to the value c/γ giving
a maximum saturated stress of σ
y
+ c/γ . The constant γ is the time constant and
determines the rate of saturation of stress. c/γ determines the magnitude. We will
now examine the ﬂow rule for kinematic hardening.
Saturated stress
c/g
x
E
Initial yield
surface
Subsequent, translated
yield surface
s
2
s
2
«
2
s
1
s
y
s
y
s
y
s
y
s
y
s
y
Fig. 2.10 Nonlinear kinematic hardening and the resulting nonlinear hardening stress–strain curve
which saturates at stress c/γ .
Kinematic hardening 33
2.5.3 Flow rule with kinematic hardening
We will use the normality hypothesis in Equation (2.17) together with the yield
function in (2.53) to determine the ﬂow rule for plastic deformation with kinematic
hardening. First, let us write the yield function, f , as
f =
_
3
2
(σ
−x
) : (σ
−x
)
_
1/2
− σ
y
= J(σ
−x
) − σ
y
. (2.67)
The normality hypothesis then gives
dε
p
= dλ
∂f
∂σ
= dλ
3
2
σ
−x
J(σ
−x
)
. (2.68)
We now need to use the consistency condition to determine the plastic multiplier.
First, let us use Equation (2.68) just to show that
dp =
_
2
3
dε
p
: dε
p
_
1/2
= dλ
[(3/2)(σ
−x
) : (σ
−x
)]
1/2
J(σ
−x
)
= dλ (2.69)
for the case of kinematic hardening with a von Mises yield criterion. The yield function
depends upon the stress, σ, and the back stress, x, which are both tensor quantities.
However, for simplicity, we shall now work in principal stress space again, and use
Voigt notation. The consistency condition becomes
∂f
∂σ
· dσ +
∂f
∂x
· dx = 0, (2.70)
which, when combined with Hooke’s law in (2.27), the normality hypothesis in
(2.17), and Equation (2.69) gives
∂f
∂σ
· C
_
dε − dλ
∂f
∂σ
_
+
∂f
∂x
·
_
2
3
c dε
p
− γ x dλ
_
=
∂f
∂σ
· C
_
dε − dλ
∂f
∂σ
_
+
∂f
∂x
·
_
2
3
c dλ
∂f
∂σ
− γ x dλ
_
= 0
so that
dλ =
(∂f/∂σ) · C dε
(∂f/∂σ) · C(∂f/∂σ) + γ (∂f/∂x) · x − (2/3)c(∂f /∂x) · (∂f /∂σ)
(2.71)
and the plastic strain increment is
dε
p
=
(∂f/∂σ) · C dε
(∂f/∂σ) · C(∂f/∂σ) + γ (∂f/∂x) · x − (2/3)c(∂f/∂x) · (∂f/∂σ)
∂f
∂σ
(2.72)
34 Continuum plasticity
and so the stress increment can be obtained from Equation (2.27). We will simplify
this to the case of uniaxial loading to interpret the terms in Section 2.5.4, but ﬁrst,
we will start by obtaining the plastic multiplier in terms of stress rather than strain
increments.
Starting from the consistency condition in (2.70) gives
∂f
∂σ
· dσ +
∂f
∂x
· dx =
∂f
∂σ
· dσ +
∂f
∂x
·
_
2
3
c dε
p
− γ x dp
_
= 0. (2.73)
From the yield function in (2.67), ∂f/∂x = −(∂f /∂σ) and dp = dλ and with
(2.68), (2.73) becomes
∂f
∂σ
· dσ −
∂f
∂σ
·
_
2
3
c dλ
∂f
∂σ
− γ x dλ
_
= 0
so that
dλ =
−(∂f/∂σ) · dσ
γ (∂f/∂σ) · x − (2/3)c(∂f/∂σ) · (∂f/∂σ)
. (2.74)
We will examine this for uniaxial loading in a von Mises material in the following
section.
2.5.4 Simple uniaxial loading
For uniaxial loading, we can determine the terms in the denominator of Equation (2.74)
as follows:
γ
∂f
∂σ
· x −
2
3
c
∂f
∂σ
·
∂f
∂σ
= γ
⎡
⎢
⎣
1
−
1
2
−
1
2
⎤
⎥
⎦
·
⎡
⎣
x
1
x
2
x
3
⎤
⎦
−
2
3
c
⎡
⎢
⎣
1
−
1
2
−
1
2
⎤
⎥
⎦
·
⎡
⎢
⎣
1
−
1
2
−
1
2
⎤
⎥
⎦
. (2.75)
We know that x is a deviatoric quantity so that for uniaxial loading in the 1direction,
x
2
= x
3
= −
1
2
x
1
so that Equation (2.75) becomes just
3
2
γ x
1
− c. Now, we can write x
1
=
2
3
x so that
this becomes γ x − c. The numerator in (2.74) is, as before, just −dσ
1
so that the
plastic multiplier for uniaxial loading is just
dλ =
dσ
1
c − γ x
. (2.76)
In a similar manner to that for the isotropic hardening example, we can obtain the
uniaxial stress increment as
dσ = E
_
1 −
E
E + c − γ x
_
dε. (2.77)
Kinematic hardening 35
Let us look at this in a little more detail. If the material behaviour is purely elastic,
then the plastic multiplier is zero and the righthand term in the bracket is zero so that
(2.77) simply reduces to
dσ = E dε.
If we have elastic, perfectly plastic behaviour, that is, no strain hardening, then
c = γ = 0 and then (2.77) tells us that during plastic deformation, the increment
in stress is zero. If the material undergoes linear kinematic hardening, that is, γ = 0,
the stress–strain relation during plasticity becomes
dσ = E
_
1 −
E
E + c
_
dε (2.78)
which is linear during plastic deformation since c is constant. This equation for linear
kinematic hardening and that for linear isotropic hardening in Equation (2.51) can
be seen to be near identical. The two types of hardening are in fact the same for
monotonically increasing loading (provided c = h) and only differ under uniaxial
conditions on a load reversal when the Bauschinger effect becomes important.
If, ﬁnally, the material undergoes nonlinear kinematic hardening, the back stress,
x, increases according to its evolution equation, (2.64), so that the stress increment
progressively decreases until saturation is achieved at which point, no further stress
increase occurs. This can be seen by substituting the equation for x, given in (2.66),
into (2.77) to give
dσ = E
_
1 −
E
E + c − γ x
_
dε
= E
_
1 −
E
E + c − γ [(c/γ )(1 − e
−γ ε
p
)]
_
dε = E
_
1 −
E
E + ce
−γ ε
p
_
dε.
As the plastic strain increases, so the exponential term diminishes until ultimately,
the material becomes perfectly plastic with dσ = 0.
Before ﬁnishing this section, Equation (2.77) demonstrates quite nicely the
incremental nature of plasticity. Let us consider ﬁrst the case of elasticity only.
Equation (2.77) gives
dσ = E dε,
where dε = dε
e
. Integrating this equation is clearly possible such that at any particular
value of elastic strain, the stress can be calculated. This is simply not possible for the
case of plasticity, for which all we can do (normally) is consider increments in stress
resulting from an increment in strain. This follows from Equation (2.77) since x itself
depends upon strain, so that the increment in stress varies from step to step. The
increment in stress, and indeed plastic strain, often depend then, in plasticity, on the
history of prior deformation. Plasticity must, in general, therefore, be considered to
be an incremental process.
36 Continuum plasticity
D
C
A
Expanded yield surface
(isotropic hardening)
after many cycles
Plasticity
recommences
Initial yield
surface
Translated yield surface
(kinematic hardening)
Expanded hysteresis
loop resulting from
isotropic hardening
s
2
s
2
s
1
«
2
s
y
s
y
s
y
s
y
s
y
B
Fig. 2.11 Combined kinematic and isotropic hardening.
S
t
r
a
i
n
Time
Fig. 2.12 Strain imposed resulting in cyclic plasticity shown in Fig. 2.11.
2.6 Combined isotropic and kinematic hardening
We will consider, ﬁnally, materials which harden both kinematically and isotropically.
This is particularly appropriate for applications to cyclic plasticity where within an
individual cycle, kinematic hardening is the dominant hardening process such that
the Bauschinger effect can be represented, but over (normally) quite large numbers
of cycles, the material also hardens isotropically such that the peak tension and com
pression stresses in a given cycle increase from one cycle to the next until saturation is
achieved. Such a process is represented schematically in Fig. 2.11. Starting from the
point of zero stress and strain, the material is subjected to the strain shown in Fig. 2.12.
The stress increases until yield is achieved at point A, and the material kinematically
hardens leading to the translation of the yield surface, as shown in Fig. 2.11. Once the
peak strain is achieved, the strain reversal occurs so that the material becomes elastic
Combined isotropic and kinematic hardening 37
at point B. Elastic deformation continues until the load point reaches the yield surface
again at point Cwhere plasticity recommences until the next strain reversal at point D.
The yield surface is translated again because of the kinematic hardening. The stress–
strain loop BCDBproduced in this way is called a hysteresis loop. If, in addition to the
kinematic hardening, the material also isotropically hardens, then superimposed upon
the translation of the yield surface is a progressive expansion, shown by the broken
line hysteresis loop in Fig. 2.11. This process, by which the peak stress and strain in
a hysteresis loop increase, due to isotropic hardening, is often called cyclic hardening,
as it often occurs from cycle to cycle over many cycles. Kinematic hardening, on the
other hand, occurs within each cycle.
We will consider the case of nonlinear kinematic and isotropic hardening, given by
Equations (2.64) and (2.37), respectively. In order to determine the plastic multiplier,
we will use the consistency condition as before. The yield function, for combined
kinematic and isotropic hardening, depends on the stress, back stress, and accumulated
plastic strain as follows
f = J(σ
−x
) − r(p) − σ
y
(2.79)
so that the consistency condition becomes
∂f
∂σ
· dσ +
∂f
∂x
· dx +
∂f
∂p
dp = 0. (2.80)
Substituting for dσ and dx from Equations (2.27) and (2.64) and writing ∂f /∂p =
−(∂r/∂p) from (2.79), together with (2.37) gives
∂f
∂σ
· C(dε − dε
p
) +
∂f
∂x
·
_
2
3
c dε
p
− γ x dp
_
− b(Q − r(p)) dp = 0. (2.81)
Following the procedure for isotropic and kinematic hardening, (2.81) can be
rearranged to give the plastic multiplier. Assuming von Mises behaviour, dp = dλ,
and substituting for dε
p
from (2.17), gives the plastic multiplier as
dλ =
(∂f/∂σ)· C dε
(∂f/∂σ)· C(∂f/∂σ)− γ (∂f/∂σ)· x +(2/3)c(∂f/∂σ)· (∂f/∂σ)+b(Q − r(p))
.
(2.82)
In a similar manner as before, we can determine the plastic multiplier in terms of the
increment of stress, rather than strain, as follows. Equation (2.80) is written as
∂f
∂σ
· dσ +
∂f
∂x
·
_
2
3
c dε
p
− γ x dp
_
− b(Q − r(p)) dp = 0
38 Continuum plasticity
so that the plastic multiplier becomes
dλ =
(∂f/∂σ) · dσ
(2/3)c(∂f/∂σ) · (∂f/∂σ) − γ (∂f/∂σ) · x + b(Q − r(p))
. (2.83)
As in previous examples, we can reduce this to the form for uniaxial loading of a
von Mises material and obtain for the uniaxial stress increment
dσ = E
_
1 −
E
E + c − γ x + b(Q − r(p))
_
dε. (2.84)
2.7 Viscoplasticity and creep
So far, we have considered timeindependent plasticity only; that is, the stress–strain
behaviour has been assumed to be independent of the rate of loading, whether strain
or stress controlled. The plasticity of materials which do exhibit rate effects is
called viscoplasticity. We shall see that the same formalism for viscoplasticity
is also appropriate for creep—the time dependent, irreversible deformation of a
material under load. The usual convention for terminology is that viscoplasticity
describes ratedependent plasticity in which crystallographic slip is the dominant
deformation process, though likely enhanced by thermallyactivated processes such
as diffusionactivated dislocation climb. Creep is normally used to describe low strain
rate, timedependent irreversible deformation which is either diffusion controlled, or
inﬂuenced by diffusion, though there may well be crystallographic slip still occurring.
In viscoplasticity, the elastic–plastic strain decomposition still holds, and yield is
determined as for timeindependent plasticity with a yield function. In addition, the
plastic ﬂow rule is obtained as before using the normality hypothesis of plasticity, and
once yielding has occurred, the material may harden isotropically or kinematically. An
important difference occurs, however, in that the consistency condition is no longer
formally applied so that the load point may now lie outside of the yield surface. As
a result, some viscoplasticity models are referred to as overstress models. We shall
start by addressing uniaxial loading in the 1direction, as before, for conditions of
plane stress.
Figure 2.13 shows schematically the material’s stress–strain response and the
corresponding yield surface which we assume to expand due to linear isotropic
hardening. At load point (1) shown on the yield surface in Fig. 2.13(a) and at the
corresponding point on the uniaxial stress–strain curve in Fig. 2.13(b), for the case of
timeindependent plasticity, the stress achieved is the yield stress, σ
y
, together with
the contribution from the linear isotropic hardening, r(p) so that
σ = σ
y
+ r(p).
Viscoplasticity and creep 39
r (p)
E
Load point (1)
Initial yield
surface
Subsequent, expanded
yield surface
(a)
(b)
«
2
s
2
s
v
s
y
s
1
s
2
s
y
s
y
s
y
s
y
s
y
Fig. 2.13 (a) The von Mises yield surface in plane stress for viscoplasticity with linear isotropic
hardening, with viscous (overstress) stress σ
v
, and (b), the corresponding stress strain curve.
The resulting stress–strain curve is that shown by the broken line in Fig. 2.13(b).
However, for the case of viscoplasticity, the stress is augmented by the viscous stress,
σ
v
, also shown schematically in Fig. 2.13(b) by the solid line. There are many types
of equations used to represent the viscous stress, but they often contain a dependence
on the effective plastic strain rate (which for uniaxial conditions is identical to the
uniaxial plastic strain rate), ˙ p. This, of course, is how the strain rate dependence of
the plasticity is introduced. Let us consider the commonly used power law function
σ
v
= K˙ p
m
(2.85)
in which K and m are material constants. The constant m is called the material’s
strain rate sensitivity. From Fig. 2.13(b), the stress–strain curve which includes the
ratedependence of stress is therefore
σ = σ
y
+ r(p) + σ
v
= σ
y
+ r(p) + K˙ p
m
. (2.86)
In viscoplasticity, the uniaxial stress depends on the yield stress, the hardening of the
yield stress, and the plastic strain rate. The stress response is, therefore, strain rate
dependent. For this reason, viscoplasticity is sometimes referred to as timedependent
plasticity or ratedependent plasticity. Let us consider the uniaxial behaviour described
by Equation (2.86) in a little more detail, but to simplify, let us assume perfect plasticity
so that there is no isotropic hardening and r(p) = 0. In this case, the equation becomes
σ = σ
y
+ K˙ p
m
. (2.87)
40 Continuum plasticity
Plastic strain, p
S
t
r
a
i
n
Time
S
t
r
e
s
s
(a) (b)
s
3
s
2
s
1
s
y
«
3
•
«
2
•
«
1
•
Fig. 2.14 (a) Applied strain and (b) the resulting ratedependent stress response.
If we apply strain controlled loading to the uniaxial sample, as shown in Fig. 2.14(a),
at a range of different strain rates as shown, the stress response (for given σ
y
, K, and
m) as a function of the plastic strain is shown in Fig. 2.14(b).
The strain rate dependence of stress is clear and for the three strain rates applied,
the corresponding stresses are given by Equation (2.87). Once yield is achieved, for
uniaxial perfect plasticity, since dσ = 0, dp = dε, or equivalently, ˙ p = ˙ ε so that the
stresses are
σ
1
= σ
y
+ K˙ ε
m
1
,
σ
2
= σ
y
+ K˙ ε
m
2
,
σ
3
= σ
y
+ K˙ ε
m
3
.
(2.88)
Many materials exhibit strain ratedependent plasticity, but unlike the stress–strain
curves in Fig. 2.14, would also show isotropic or kinematic hardening. Let us return
to the case of isotropic hardening given in Equation (2.86). We can rearrange the
equation to give
˙ p =
_
σ − r − σ
y
K
_
1/m
. (2.89)
If in addition to isotropic hardening, kinematic hardening also occurs, Equation (2.89)
becomes
˙ p =
_
σ − x − r − σ
y
K
_
1/m
. (2.90)
Equations (2.89) and (2.90) are constitutive equations relating uniaxial plastic strain
rate to uniaxial stress, and which depend upon internal variables; the isotropic harden
ing, r, and the kinematic hardening, x. For a von Mises material, the uniaxial stress
is identical to the effective stress and similarly for the effective plastic strain rate.
Equation (2.90) can therefore be written
˙ p =
_
J(σ
−x
) − r − σ
y
K
_
1/m
. (2.91)
Viscoplasticity and creep 41
In viscoplasticity, we need a constitutive equation, such as (2.91), which relates the
effective plastic strain rate to the stress and internal hardening variables. This replaces
the consistency condition in timeindependent plasticity, which forces the load point
to stay on the yield surface during plastic deformation. In viscoplasticity, the load point
may lie outside the yield surface because of the viscous or overstress. In addition to the
viscoplastic constitutive equation, as before, we make use of the normality hypothesis
and the elastic constitutive equation; namely Hooke’s law. For viscoplasticity, we
write the normality hypothesis, from Equation (2.17), as
˙ ε
p
=
˙
λ
∂f
∂σ
. (2.92)
The yield function is, from Section 2.6,
f = J(σ
−x
) − r(p) − σ
y
(2.93)
so that
∂f
∂σ
=
3
2
σ
−x
J(σ
−x
)
and
˙ ε
p
=
3
2
˙
λ
σ
−x
J(σ
−x
)
. (2.94)
We showed in Section 2.5.3 that for a von Mises material, dp = dλ, or equivalently
˙ p =
˙
λ so that (2.94) becomes
˙ ε
p
=
3
2
˙ p
σ
−x
J(σ
−x
)
. (2.95)
We may now combine the constitutive equation given in (2.91) with (2.95) to give the
plastic ﬂow rule for viscoplasticity with isotropic and kinematic hardening as
˙ ε
p
=
3
2
_
J(σ
−x
) − r − σ
y
K
_
1/m
σ
−x
J(σ
−x
)
. (2.96)
To complete the model, we need the evolution equations for the isotropic and kinematic
hardening variables r and x, and the rate form of Hooke’s law.
FromSection 2.4, Equation (2.37) and Section 2.5.2, Equation (2.65), the hardening
rates are given by
˙ r(p) = b(Q − r) ˙ p, (2.97)
˙ x =
2
3
c˙ ε
p
− γ x ˙ p (2.98)
and we will nowwrite Hooke’s lawin terms of tensor rather than vector strain terms as
˙ σ = 2G˙ ε
e
+ λ Tr(˙ ε
e
)I, (2.99)
42 Continuum plasticity
where
˙ ε
e
= ˙ ε − ˙ ε
p
(2.100)
and in which G is the shear modulus, λ the Lame constant given by
λ =
Eν
(1 − 2ν)(1 + ν)
and I is the identity tensor given by
I =
⎛
⎝
1 0 0
0 1 0
0 0 1
⎞
⎠
.
Equations (2.96)–(2.100) form the complete elastic–viscoplastic model. At a given
time, t , with knowledge of the current total strain rate, ˙ ε and the hardening variables
r and x, together with the stress, σ, these equations enable the stresses at the end of
a given step forward in time to be determined. We will address this in a later chapter,
but for now, let us examine the equations for a simple problem of uniform, uniaxial,
and axisymmetric upsetting.
2.7.1 Uniform, uniaxial, and axisymmetric compression
Upsetting is the name given to the opendie forging of cylindrical billets of material.
We will consider the uniaxial compression of a cylinder between two platens such
that the frictional effects on the contacting surfaces are negligible so that the cylinder
remains under uniform, uniaxial compression. The process is shown schematically in
Fig. 2.15.
z
r
Platens
Undeformed
material
Deformed
material
u
Fig. 2.15 Frictionless, uniaxial compression representing opendie forging.
Viscoplasticity and creep 43
We will consider constant strain ratecontrolled loading and write the plastic strain
rates as
˙ ε
p
=
⎛
⎜
⎜
⎝
˙ ε
p
rr
˙ ε
p
rz
0
˙ ε
p
rz
˙ ε
p
zz
0
0 0 ˙ ε
p
θθ
⎞
⎟
⎟
⎠
. (2.101)
Because we are assuming friction to be negligible, the shear terms are zero. The
boundary conditions give us that
σ
rr
= σ
θθ
= σ
rz
= 0
so that the stress and deviatoric stress tensors for the problem become
σ =
⎛
⎝
0 0 0
0 σ
zz
0
0 0 0
⎞
⎠
, σ
=
⎛
⎜
⎝
−
1
3
σ
zz
0 0
0
2
3
σ
zz
0
0 0 −
1
3
σ
zz
⎞
⎟
⎠
. (2.102)
We will now use Equation (2.96) to determine the plastic strain rate components for
the problem.
Under this loading,
J(σ
−x
) = σ
zz
− x (2.103)
and from Section 2.5.1,
σ
−x
=
2
3
⎛
⎝
−
1
2
(σ
zz
− x) 0 0
0 σ
zz
− x 0
0 0 −
1
2
(σ
zz
− x)
⎞
⎠
(2.104)
so that (2.96) becomes
⎛
⎜
⎜
⎝
˙ ε
p
rr
˙ ε
p
rz
0
˙ ε
p
rz
˙ ε
p
zz
0
0 0 ˙ ε
p
θθ
⎞
⎟
⎟
⎠
=
3
2
_
σ
zz
− x − r − σ
y
K
_
1/m
×
1
σ
zz
− x
2
3
⎛
⎜
⎝
−
1
2
(σ
zz
− x) 0 0
0 σ
zz
− x 0
0 0 −
1
2
(σ
zz
− x)
⎞
⎟
⎠
which reduces to
⎛
⎜
⎜
⎝
˙ ε
p
rr
˙ ε
p
rz
0
˙ ε
p
rz
˙ ε
p
zz
0
0 0 ˙ ε
p
θθ
⎞
⎟
⎟
⎠
=
_
σ
zz
− x − r − σ
y
K
_
1/m
⎛
⎜
⎝
−
1
2
0 0
0 1 0
0 0 −
1
2
⎞
⎟
⎠
. (2.105)
44 Continuum plasticity
We may now look at individual components of the plastic strain rate. We see that
˙ ε
p
zz
=
_
σ
zz
− x − r − σ
y
K
_
1/m
(2.106)
which is, of course, just what we would get for purely uniaxial loading, and
Equation (2.105) also shows that
˙ ε
p
rr
= ˙ ε
p
θθ
= −
1
2
˙ ε
p
zz
= −
1
2
_
σ
zz
− x − r − σ
y
K
_
1/m
demonstrating that the incompressibility condition of plasticity is satisﬁed.
2.7.2 Powerlaw creep
Let us look at one further case in which we assume there to be neither isotropic nor
kinematic hardening. Equation (2.96) then becomes, for the plastic strain rate,
˙ ε
p
=
3
2
_
J(σ
) − σ
y
K
_
1/m
σ
J(σ
)
(2.107)
and J(σ
) = σ
e
so the equation reduces further to
˙ ε
p
=
3
2
_
σ
e
− σ
y
K
_
1/m
σ
σ
e
. (2.108)
Often in creep problems, where the timedependent deformation is not dependent
upon yield, creep deformation is assumed to occur with the application of a nonzero
stress. For creep problems, therefore, Equation (2.108) is often written
˙ ε
c
=
3
2
_
σ
e
K
_
1/m
σ
σ
e
=
3
2
A(σ
e
)
n
σ
σ
e
=
3
2
Aσ
n−1
e
σ
(2.109)
which is the multiaxial form of Norton’s creep law, where A and n are material
constants. For uniaxial loading, it reduces to
˙ ε
c
=
3
2
Aσ
n−1
e
2
3
σ = Aσ
n
.
For completeness, we can write out Equation (2.109) in full component form as
⎛
⎜
⎝
˙ ε
c
xx
˙ ε
c
xy
˙ ε
c
xz
˙ ε
c
yx
˙ ε
c
yy
˙ ε
c
yz
˙ ε
c
zx
˙ ε
c
zy
˙ ε
c
zz
⎞
⎟
⎠
=
3
2
Aσ
n−1
e
⎛
⎜
⎝
σ
xx
σ
xy
σ
xz
σ
yx
σ
yy
σ
yz
σ
zx
σ
zy
σ
zz
⎞
⎟
⎠
(2.110)
in which the stress and strain rate tensors are symmetric.
Further reading 45
2.7.3 Potential and yield function equivalence
In the last example, we saw how the plastic strain rate, based on the normality hypo
thesis together with an appropriate constitutive equation for plastic strain rate, was
determined for viscoplasticity, but then reduced to Norton’s law for creep deforma
tion. Because creep processes may occur independently of plastic yielding, it is not
appropriate to use a yield surface, in the conventional sense discussed earlier. Instead,
a potential function is deﬁned from which creep strain rates are determined.
We will deﬁne a potential, φ, such that
φ =
˙ ε
0
σ
0
n + 1
_
σ
e
σ
0
_
n+1
(2.111)
in which ˙ ε
0
and ˙ σ
0
are material constants with units of strain rate and stress, respect
ively. The potential function, φ, therefore has units of Joules per second per unit
volume. It therefore represents an energy per second per unit volume.
The plastic (creep) strain rate is determined from
˙ ε
c
=
∂φ
∂σ
(2.112)
giving
˙ ε
c
=
3
2
˙ ε
0
σ
0
σ
n+1
0
σ
n−1
e
σ
=
3
2
Aσ
n−1
e
σ
, (2.113)
which is identical to Equation (2.109). The potential function, φ, in creep plays a
similar role to the yield function in plasticity. In fact, often the yield function is
considered to be a potential function. The surfaces in stress space represented by φ
are often called equipotential surfaces; the energy per second per unit volume is the
same at each point on the potential surface, just as the value of σ
e
takes the same value
on every point of a yield surface in plasticity.
Further reading
Dieter, G.E. (1988). Mechanical Metallurgy. McGrawHill Book Co., London.
Hill, R. (1998). The Mathematical Theory of Plasticity. OUP (Oxford Classics Series).
(ﬁrst published in 1950).
Khan, A.S. and Huang, S. (1995). ContinuumTheory of Plasticity. John Wiley &Sons
Inc, New York.
Lemaitre, J. and Chaboche, J.L. (1990). Mechanics of Solid Materials. CUP,
Cambridge, UK.
Lubarda, V.A. (2002). Elastoplasticity Theory. CRC Press, Florida, USA.
This page intentionally left blank
3. Kinematics of large deformations
and continuum mechanics
3.1 Introduction
The strains, be they elastic or plastic, which most engineering components undergo
in service are usually small, that is, <0.001 (or 0.1%). At yield, for example, a nickel
alloy may have undergone a strain of about σ
y
/E ∼ 0.002 and it is hoped that this
occurs rarely in nickelbased alloy aeroengine components in service! During man
ufacture, however, the strains may be much bigger; the forging of an aeroengine
compressor disc, for example, requires strains in excess of 2.0 (i.e. >200%). This
is three orders of magnitude larger than the strain needed to cause yield. In manu
facturing processes, another very important feature is material rotation. Deformation
processing leading to large plastic strains often also leads to large rigid body rotations.
The bending of a circular plate, to large deformation, is an example which shows the
rigid body rotation. Figure 3.1 shows the result of applying a large downward dis
placement at the centre of an initially horizontal, simply supported circular disc. Only
one half of the disc section is shown. While the displacements and rigid body rotations
can be very large, the strains remain quite small.
×
×
Fig. 3.1 Simulatedlarge elastic–plastic deformationof aninitiallyhorizontal, simplysupportedcircular
plate. Only one half of the plate section is shown.
48 Large deformations and continuum mechanics
Towards the outer edge of the disc, where it is supported, the ﬁnite element simu
lation shows that the strains generated are quite small, but that because of the disc
bending, the rigid body rotations are very large. Generally, deformation comprises of
stretch, rigid body rotation, and translation. The stretch provides the shape change.
The rigid body rotation neither contributes to shape change nor to internal stress.
Because a translation does not lead to a change in stress state, we will not address
it in detail here. In this chapter, we will give an introduction to measures of large
deformation, rigid body rotation, elastic–plastic coupling in large deformation, stress
rates, and what is called material objectivity, or frame indifference. Later chapters
dealing with the implementation of plasticity models into ﬁnite element code will rely
on the material covered here and in Chapter 2.
3.2 The deformation gradient
In order to look at deformation, let us consider a small lump of imaginary material
which is yet to be loaded so that it is in the undeformed (or initial ) conﬁguration
(or state). This is shown schematically as state A in Fig. 3.2.
We will now apply a load to the material in state A so that it deforms to that shown
in state B, the deformed or current conﬁguration. We will assume that the material
undergoes combined stretch (i.e. relative elongations with respect to the three ortho
gonal axes), rigid body rotation, and translation. We will measure all quantities relative
to the global XYZ axes, often known as the material coordinate system. Consider an
inﬁnitesimal line, PQ, or vector, dX, embedded in the material in the undeformed
conﬁguration. The position of point P is given by vector X, relative to the material
reference frame. The line PQ undergoes deformation from state A to the deformed
Z
X
Deformed (current)
configuration, state B
Undeformed (original)
configuration, state A
P9
Q9
dx
x
u
X
Q
dX
P
Y
O
Fig. 3.2 An element of material in the reference or undeformed conﬁguration undergoing deformation
to the deformed or current conﬁguration.
Measures of strain 49
conﬁguration in state B. In doing so, point P has been translated by u to point P
.
Relative to the material reference frame, point P
is given by vector x where
x = X +u. (3.1)
The inﬁnitesimal vector dX is transformed to its deformed state, dx, by the
deformation gradient, F, where
dx = F dX. (3.2)
We can write this in component form as
⎛
⎝
dx
dy
dz
⎞
⎠
=
⎛
⎝
F
xx
F
xy
F
xz
F
yx
F
yy
F
yz
F
zx
F
zy
F
zz
⎞
⎠
⎛
⎝
dX
dY
dZ
⎞
⎠
=
⎛
⎜
⎜
⎜
⎜
⎜
⎝
∂x
∂X
∂x
∂Y
∂x
∂Z
∂y
∂X
∂y
∂Y
∂y
∂Z
∂z
∂X
∂z
∂Y
∂z
∂Z
⎞
⎟
⎟
⎟
⎟
⎟
⎠
⎛
⎝
dX
dY
dZ
⎞
⎠
(3.3)
or
F =
∂x
∂X
. (3.4)
The deformation gradient, F, provides a complete description of deformation (exclud
ing translations) which includes stretch as well as rigid body rotation. Rigid body rota
tion does not contribute to shape or size change, or internal stress. In solving problems,
it is necessary to separate out the stretch from the rigid body rotation contained within
F. In the following sections, we will see examples of stretch, rigid body rotation, and
their combination, and how they are described by the deformation gradient.
3.3 Measures of strain
Let us consider the length, ds, of the line dx in the deformed conﬁguration. We may
write
ds
2
= dx · dx = (F dX) · (F dX) = dX
T
F
T
F dX
= dX
T
_
∂x
∂X
_
T
∂x
∂X
dX = dX
T
C dX
so that
C = F
T
F. (3.5)
C is called the (left) Cauchy–Green tensor. Consider also the length, dS, of the
element, dX, in the undeformed state:
dS
2
= dX
T
dX. (3.6)
50 Large deformations and continuum mechanics
Now
dx = F dX
so
dX = F
−1
dx.
Substituting into Equation (3.6) gives
dS
2
= (F
−1
dx)
T
F
−1
dx = dx
T
(F
−1
)
T
F
−1
dx = dx
T
B
−1
dx,
where
B
−1
= (F
−1
)
T
F
−1
(3.7)
and B is called the (right) Cauchy–Green tensor. Both B and C are in fact measures
of stretch as we will now see.
A measure of the stretch is given by the difference in lengths of the lines PQ and
P
Q
in Fig. 3.2 in the undeformed and deformed conﬁgurations, respectively. We can
write
ds
2
− dS
2
= dx · dx − dx · B
−1
dx = dx · (I −B
−1
) dx (3.8)
in which I is the identity tensor. It can be seen, therefore, that B is related to the
change in length of the line; in other words, a measure of stretch. It is independent
of rigid body rotation because the orientations of the lines in the undeformed and
deformed conﬁgurations are irrelevant. If ds and dS are the same length, then there is
no stretch and
ds
2
− dS
2
= 0.
From Equation (3.8), this means that
B = I (3.9)
and there is no stretch, so that in this instance, the deformation gradient contains only
rigid body rotation. The Cauchy–Green tensor, B, could itself be used as a measure of
strain, since it is independent of rigid body rotation, but depends upon the stretch. This
is an important criterion for any strain measure in large deformation analysis in which
rigid body rotation occurs. A strain which depends upon rigid body rotation would
not be appropriate since it would give a different measure of the strain depending
upon orientation. However, the Cauchy–Green tensor given in (3.9) contains non
zero components even though the stretch is zero. An alternative and more appropriate
strain measure called the Almansi strain was introduced:
e =
1
2
(I −B
−1
) (3.10)
so that for zero stretch,
e = 0.
Measures of strain 51
This strain measure behaves more like familiar strains, such as engineering strain,
since for zero stretch, it gives us strain components of zero.
A further measure of strain is the logarithmic, or true strain deﬁned as
ε = −
1
2
ln B
−1
, (3.11)
which we shall consider in more detail later. In determining the change of length of
the line OP, we could have chosen the original conﬁguration to work in as follows:
ds
2
− dS
2
= dx
T
dx − dX
T
dX = (F dX)
T
F dX − dX
T
dX
= dXF
T
F dX − dX
T
dX = dX
T
(F
T
F −I) dX
= dX
T
(C −I) dX = dX
T
(2E) dX,
where
E =
1
2
(C −I) =
1
2
(F
T
F −I). (3.12)
E is called the large strain or the Green–Lagrange strain tensor. We can make this
look a little more familiar, perhaps, by combining Equations (3.4) for F and (3.1)
for x
F =
∂x
∂X
=
∂(u +X)
∂X
=
∂u
∂X
+I
and then substituting into (3.12) to give
E =
1
2
(F
T
F −I) =
1
2
_
_
∂u
∂X
+I
_
T
_
∂u
∂X
+I
_
−I
_
=
1
2
_
∂u
∂X
+
_
∂u
∂X
_
T
+
_
∂u
∂X
_
T
∂u
∂X
_
. (3.13)
If we ignore the secondorder term, this reduces to
E =
1
2
_
∂u
∂X
+
_
∂u
∂X
_
T
_
. (3.14)
We will examine this and other strains for simple uniaxial loading in Section 3.4.
Before doing so, let us examine the symmetry of the strain quantities presented above.
The symmetric part of a tensor, A, is given by
sym(A) =
1
2
(A+A
T
) (3.15)
and the antisymmetric or skew symmetric (or sometimes simply skew) part of A by
asym(A) =
1
2
(A−A
T
). (3.16)
52 Large deformations and continuum mechanics
To be pedantic, let us write out the symmetric and antisymmetric parts of the 2 × 2
matrix A given by
A =
_
a
11
a
12
a
21
a
22
_
in which a
12
= a
21
.
sym(A) =
⎛
⎝
a
11
a
12
+ a
21
2
a
12
+ a
21
2
a
22
⎞
⎠
and
asym(A) =
⎛
⎝
0
a
12
− a
21
2
−
a
12
− a
21
2
0
⎞
⎠
.
Clearly, sym(A) is symmetric and the leading diagonal of an antisymmetric tensor
always contains zeros. Let us now consider the quantity F
T
F which appears above in
a number of strain quantities:
sym(F
T
F) =
1
2
[F
T
F + (F
T
F)
T
] = F
T
F
and
asym(F
T
F) =
1
2
[F
T
F − (F
T
F)
T
] = 0.
We see, therefore, that F
T
F is a symmetric tensor so that, in fact, the quantities
B
−1
, C, ε, and E are all themselves symmetric. In general, F will not necessarily be
symmetric. If it is, the deformation it represents is made only up of stretch.
3.4 Interpretation of strain measures
Let us determine the deformation gradient for the simple case of a uniaxial rod which
is subjected, ﬁrst to purely rigid body rotation and then to uniaxial stretch and then
consider some of the measures of deformation and strain introduced in Section 3.3.
3.4.1 Rigid body rotation only
Figure 3.3 shows a rod lying along the Yaxis which undergoes a clockwise rotation
about the Zaxis of angle θ. There is no stretch imposed, so the deformation gradient
is simply the rotation matrix, R, given by
F = R =
⎛
⎝
cos θ −sin θ 0
sin θ cos θ 0
0 0 1
⎞
⎠
. (3.17)
Interpretation of strain measures 53
Y, y
X, x
Y
u
x
y
X
Fig. 3.3 A rod undergoing rigid body rotation through angle θ.
The uppercase letters in Fig. 3.3, XY, refer to the material reference frame directions.
Also shown in the ﬁgure is a coordinate system which rotates with the deforming
material (in this case, the rotating rod). We will generally use lowercase letters, xy,
to indicate the reference frame which rotates with the material. This is called the
corotational reference frame.
We can now determine the Cauchy–Green tensor, B
−1
, and the strain quantities.
From Equation (3.7),
B
−1
= (F
−1
)
T
F
−1
=
⎛
⎝
cos θ −sin θ 0
sin θ cos θ 0
0 0 1
⎞
⎠
⎛
⎝
cos θ sin θ 0
−sin θ cos θ 0
0 0 1
⎞
⎠
=
⎛
⎝
1 0 0
0 1 0
0 0 1
⎞
⎠
.
The tensor B
−1
is found to be the identity tensor for rigid body rotation because there
is no stretch. Measures of deformation that are appropriate for large deformations with
rigid body rotation must have this property: that they depend upon the stretch but are
independent of the rigid body rotation. The Almansi strain, from (3.10), is
e =
1
2
(I −B
−1
) = 0
and the true strain, (3.11), is given by
ε =
1
2
ln B
−1
= 0.
Similarly, the Green strain is
E =
1
2
(F
T
F −I) = 0.
We see that the three strain measures give zero for the case of rigid body rotation.
Before moving away from rotation, it is important to note that a rotation tensor,
such as R, is always orthogonal, that is,
RR
T
= I (3.18)
54 Large deformations and continuum mechanics
Z, z
Undeformed rod
Deformed rod
after stretch
r
0
r
l
l
0
Y, y
X, x
Fig. 3.4 A rod undergoing pure stretch in the absence of rigid body rotation.
so that
R
T
= R
−1
.
We will make much use of this property in the subsequent sections.
3.4.2 Uniaxial stretch
Let us now consider uniaxial stretch of the circular rod in the Ydirection. This is
shown schematically in Fig. 3.4. Note that because there is no rotation in this case, the
corotational reference frame is directionally coincident with the material reference
frame.
For the uniaxial stretch shown, the stretch ratios, λ, are
λ
x
=
r
r
0
, λ
y
=
l
l
0
, λ
z
=
r
r
0
. (3.19)
Let us consider the case of large strain such that the elastic strains can be ignored.
For large plastic deformation, the incompressibility condition written in terms of
stretches is
λ
x
λ
y
λ
z
= 1 (3.20)
so that
λ
x
= λ
z
≡
1
_
λ
y
and therefore
λ
x
= λ
z
=
_
l
l
0
_
−1/2
. (3.21)
Considering the stretch along the Yaxis, any point, Y, lying on the undeformed rod
becomes the point y = λ
y
Y on the deformed rod. The deformation can therefore be
represented by
x = λ
x
X, y = λ
y
Y, z = λ
z
Z
Interpretation of strain measures 55
so that
∂x
∂X
= λ
x
,
∂y
∂Y
= λ
y
,
∂z
∂Z
= λ
z
and
∂x
∂Y
=
∂x
∂Z
= 0 etc.
The deformation gradient can then be determined using Equation (3.3),
F =
∂x
∂X
=
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎝
∂x
∂X
∂x
∂Y
∂x
∂Z
∂y
∂X
∂y
∂Y
∂y
∂Z
∂z
∂X
∂z
∂Y
∂z
∂Z
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎠
=
⎛
⎝
λ
x
0 0
0 λ
y
0
0 0 λ
z
⎞
⎠
and with (3.19) and (3.21)
F =
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎝
_
l
l
0
_
−1/2
0 0
0
_
l
l
0
_
0
0 0
_
l
l
0
_
−1/2
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎠
.
We ﬁrst notice that F is symmetric so that
F
T
= F
and therefore represents stretch only. Let us examine the various deformation and
strain tensors.
The inverse of F is
F
−1
=
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎝
_
l
l
0
_
1/2
0 0
0
_
l
l
0
_
−1
0
0 0
_
l
l
0
_
1/2
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎠
so the Cauchy–Green tensor is given by
B
−1
= (F
−1
)
T
F
−1
= F
−1
F
−1
=
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎝
l
l
0
0 0
0
_
l
l
0
_
−2
0
0 0
l
l
0
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎠
.
56 Large deformations and continuum mechanics
The true plastic strain is, therefore,
ε = −
1
2
ln B
−1
= −
1
2
ln
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎝
l
l
0
0 0
0
_
l
l
0
_
−2
0
0 0
l
l
0
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎠
=
⎛
⎜
⎜
⎜
⎜
⎜
⎝
−
1
2
ln
l
l
0
0 0
0 ln
l
l
0
0
0 0 −
1
2
ln
l
l
0
⎞
⎟
⎟
⎟
⎟
⎟
⎠
.
(3.22)
We see that the true strain components are ε
xx
= ε
zz
= −
1
2
ln(l/l
0
) = −
1
2
ε
yy
and
ε
yy
= ln(l/l
0
), as we would expect for uniaxial plasticity conditions. Before leaving
the true strain, note that we needed to take the logarithm of the Cauchy–Green tensor
which we did by operating on the leading diagonal alone. We were able to do this
in this case because the leading diagonal terms happen, in this simple case, to be the
principal parts of the tensor (i.e. the eigenvalues).
In general, in order to carry out an operation, p, on tensor A having a linearly
independent set of eigenvectors, we need to diagonalize the tensor and operate on
the principal values. In practice, this means transforming the tensor into its prin
cipal coordinates, by ﬁnding the eigenvalues and eigenvectors of A, operating on it
and transforming it back. If the modal matrix (i.e. the matrix containing the eigen
vectors of A) is written as M, and the diagonal matrix (i.e. the matrix containing the
eigenvalues of A along the leading diagonal) as
ˆ
A, then A can be written
A = M
ˆ
AM
−1
so that we operate on A to give p(A) as follows
p(A) = Mp(
ˆ
A)M
−1
, (3.23)
where the operation p is carried out only on the leading diagonal terms. For example,
ln(A) = M ln(
ˆ
A)M
−1
.
In Equation (3.22), the modal matrix of B
−1
in this case is just the identity matrix and
its diagonal matrix
ˆ
B
−1
is the same as B
−1
so that
ln(B
−1
) = M ln(
ˆ
B
−1
)M
−1
= I ln(B
−1
)I
−1
= ln(B
−1
).
Let us ﬁnally determine the Green strain ﬁrst from (3.12) and second directly from
the displacements using (3.14).
Polar decomposition 57
Using (3.12), we need
F
T
F =
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎝
_
r
r
0
_
2
0 0
0
_
l
l
0
_
2
0
0 0
_
r
r
0
_
2
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎠
≈
⎛
⎜
⎜
⎜
⎜
⎜
⎝
1 +
2r
r
0
0 0
0 1 +
2l
l
0
0
0 0 1 +
2r
r
0
⎞
⎟
⎟
⎟
⎟
⎟
⎠
if we write r = r
0
+ r and assume that the strains are small, so that
E =
1
2
⎡
⎢
⎢
⎢
⎢
⎢
⎣
⎛
⎜
⎜
⎜
⎜
⎜
⎝
1 +
2r
r
0
0 0
0 1 +
2l
l
0
0
0 0 1 +
2r
r
0
⎞
⎟
⎟
⎟
⎟
⎟
⎠
−
⎛
⎝
1 0 0
0 1 0
0 0 1
⎞
⎠
⎤
⎥
⎥
⎥
⎥
⎥
⎦
=
⎛
⎜
⎜
⎜
⎜
⎜
⎝
r
r
0
0 0
0
l
l
0
0
0 0
r
r
0
⎞
⎟
⎟
⎟
⎟
⎟
⎠
.
We see that the components of the Green strain are therefore approximately the
engineering strain components. If we now start from (3.14) in which we have also
neglected secondorder terms, we need the displacements, u, which are given by
u
x
= (r − r
0
)
X
r
0
, u
y
= (l − l
0
)
Y
l
0
, u
z
= (r − r
0
)
Z
r
0
so that
∂u
∂X
=
⎛
⎜
⎜
⎜
⎜
⎜
⎝
r
r
0
0 0
0
l
l
0
0
0 0
r
r
0
⎞
⎟
⎟
⎟
⎟
⎟
⎠
and E =
1
2
_
∂u
∂X
+
_
∂u
∂X
_
T
_
=
⎛
⎜
⎜
⎜
⎜
⎜
⎝
r
r
0
0 0
0
l
l
0
0
0 0
r
r
0
⎞
⎟
⎟
⎟
⎟
⎟
⎠
as before.
We have looked at a number of examples in which stretch and rigid body rotation
have taken place exclusively. We will nowlook at howto separate themin cases where
they occur simultaneously, using the polar decomposition theorem.
3.5 Polar decomposition
Recall that the deformation at any material point can be considered to comprise three
parts:
(1) rigid body translation (which we do not need to consider here);
(2) rigid body rotation;
(3) stretch.
58 Large deformations and continuum mechanics
The polar decomposition theorem states that any nonsingular, secondorder tensor
can be decomposed uniquely into the product of an orthogonal tensor (a rotation), and
a symmetric tensor (stretch).
The deformation gradient is a nonsingular, secondorder tensor and can therefore
be written as
F = RU = VR, (3.24)
where R is an orthogonal (R
T
R = I) rotation tensor, and U and V are symmetric
(U = U
T
) stretch tensors.
Using the polar decomposition theorem, we can now examine in more detail the
right and left Cauchy–Green deformation tensors. From Equation (3.7),
B
−1
= (F
−1
)
T
F
−1
= [(VR)
−1
]
T
(VR)
−1
= (R
−1
V
−1
)
T
R
−1
V
−1
= (V
−1
)
T
(R
−1
)
T
R
−1
V
−1
= (V
−1
)
T
V
−1
since R is orthogonal
B
−1
= V
−1
V
−1
= (V
−1
)
2
,
where the squaring operation is carried out on the diagonalized form of V
−1
. This
conﬁrms, therefore, that B is a measure of deformation that depends on stretch only,
and is independent of the rigid body rotation, R.
Similarly, for C,
C = F
T
F = (RU)
T
RU = U
T
R
T
RU
= U
T
U since R is orthogonal
= U
2
since U is symmetric.
The true strain rate can now be written in terms of V since
ε = −
1
2
ln B
−1
= −
1
2
ln(V
−1
)
2
= ln V, (3.25)
which is also independent of the rigid body rotation and dependent on the stretch
alone. Let us have a look at a problem in which both stretch and rigid body rotation
occur together: the problem of simple shear.
3.5.1 Simple shear
Figure 3.5 shows schematically the simple shear by δ, in two dimensions, of a unit
square.
We can represent the deformation, which transforms a point (X, Y) in the
undeformed conﬁguration to (x, y) in the deformed conﬁguration, by
x = X + δY, y = Y
Polar decomposition 59
1
d
X, x
Y, y
1
Deformed
square
Undeformed
square
Fig. 3.5 A unit square undergoing simple shear.
so that
∂x
∂X
= 1,
∂x
∂Y
= δ,
∂y
∂X
= 0,
∂y
∂Y
= 1.
The deformation gradient is, therefore,
F =
_
1 δ
0 1
_
.
We can then determine the Cauchy–Green tensor, C, from
C = F
T
F =
_
1 0
δ 1
__
1 δ
0 1
_
=
_
1 δ
δ 1 + δ
2
_
. (3.26)
Note the symmetry of this deformation tensor. We can use the polar decomposition
theoremto separate out the stretch and rigid body rotation contained in the deformation
gradient. Let us set
U =
_
U
xx
U
xy
U
yx
U
yy
_
R =
_
cos ϕ sin ϕ
−sin ϕ cos ϕ
_
(3.27)
so that
F = RU gives
_
1 δ
0 1
_
=
_
cos ϕ sin ϕ
−sin ϕ cos ϕ
__
U
xx
U
xy
U
yx
U
yy
_
. (3.28)
Solving for the stretches in terms of δ and ϕ gives
U =
_
cos ϕ sin ϕ
sin ϕ cos ϕ + δ sin ϕ
_
and substituting back into (3.28) gives
sin ϕ =
δ
√
2
2
+ δ
2
and cos ϕ =
2
√
2
2
+ δ
2
so that
R =
1
√
2
2
+ δ
2
_
2 δ
−δ 2
_
(3.29)
60 Large deformations and continuum mechanics
and
U =
1
√
2
2
+ δ
2
_
2 δ
δ 2 + δ
2
_
. (3.30)
ComparingEquations (3.30) and(3.26), andwitha little algebra conﬁrms that C = U
2
.
U is the symmetric stretch and Equation (3.29) shows that the rigid body rotation is
nonzero. It is useful to examine why this is. It may not be obvious that the deformation
corresponding to simple shear shown in Fig. 3.5 leads to a rigid body rotation. Let us
ﬁrst determine the Green–Lagrange strain for this deformation. Using Equation (3.12),
E =
1
2
(F
T
F −I) =
1
2
(C −I) =
1
2
__
1 δ
δ 1 + δ
2
_
−
_
1 0
0 1
__
=
1
2
_
0 δ
δ δ
2
_
.
(3.31)
From Equation (3.29), or from the components of strain in Equation (3.31), we see
that the principal stretch directions rotate as the simple shear proceeds. That is, the
‘material axes’ rotate relative to the direction of the applied deformation.
The polar decomposition theorem is fundamental to large deformation kinematics
and we shall return to it in subsequent sections.
3.6 Velocity gradient, rate of deformation, and continuum spin
We have so far considered stretch, rigid body rotation, measures of strain, and the polar
decomposition theorem which enables us to separate out stretch and rotation. All the
quantities considered have been independent of time. However, many plasticity for
mulations (viscoplasticity is an obvious case) are developed in terms of rate quantities
and it is necessary, therefore, to consider how the quantities already discussed can be
put in rate form. Often, in fact, even rateindependent plasticity models are written
in rate form for implementation into ﬁnite element code. This is easy to understand
given that plasticity is an incremental process; rather than deal with increments in
plastic strain, it is often more convenient to work with the equivalent quantity, plastic
strain rate.
Consider a spatially varying velocity ﬁeld, that is, one for which the material point
velocities vary spatially. The increment in velocity, dv, occurring over an incremental
change in position, dx, in the deformed conﬁguration, may be written as
dv =
∂v
∂x
dx.
The velocity gradient describes the spatial rate of change of the velocity and is given by
L =
∂v
∂x
. (3.32)
Velocity gradient, rate of deformation, and continuum spin 61
Consider the time rate of change of the deformation gradient:
˙
F =
∂
∂t
_
∂x
∂X
_
=
∂v
∂X
=
∂v
∂x
∂x
∂X
= LF
or
L =
˙
FF
−1
. (3.33)
The velocity gradient, therefore, maps the deformation gradient onto the rate of change
of the deformation gradient.
The velocity gradient can be decomposed into symmetric (stretch related) and
antisymmetric (rotation related) parts:
L = sym(L) + asym(L),
where
sym(L) =
1
2
(L +L
T
) (3.34)
and
asym(L) =
1
2
(L −L
T
). (3.35)
The symmetric part is called the rate of deformation, D, and the antisymmetric part
the continuum spin, W, so that
L = D +W, (3.36)
where the rate of deformation
D =
1
2
(L +L
T
) (3.37)
and the continuum spin is given by
W =
1
2
(L −L
T
). (3.38)
We shall nowexamine both quantities, the rate of deformation and the continuumspin,
for some simple examples, including uniaxial stretch and purely rigid body rotation
in order to gain a physical feel for these quantities.
3.6.1 Rigid body rotation and continuum spin
We have previously looked at the rigid body rotation of a uniaxial rod. Figure 3.3
shows this schematically. We will now look not only at the rigid body rotation, but
also the rate at which it occurs. Consider the rotation of the rod shown in Fig. 3.3,
62 Large deformations and continuum mechanics
at time t making an angle θ with the Yaxis, with no stretch, rotating at constant rate
˙
θ. The deformation gradient at any instant is given by
F =
⎛
⎝
cos θ −sin θ 0
sin θ cos θ 0
0 0 1
⎞
⎠
(3.39)
so that the rate of change of deformation gradient is
˙
F =
˙
θ
⎛
⎝
−sin θ −cos θ 0
cos θ −sin θ 0
0 0 0
⎞
⎠
. (3.40)
In order to determine the velocity gradient, we need the inverse of F which we can
obtain from (3.39) as
F
−1
(= F
T
in this case) =
⎛
⎝
cos θ sin θ 0
−sin θ cos θ 0
0 0 1
⎞
⎠
.
We may then determine the velocity gradient using Equation (3.33) as
L =
˙
FF
−1
=
˙
θ
⎛
⎝
−sin θ −cos θ 0
cos θ −sin θ 0
0 0 0
⎞
⎠
⎛
⎝
cos θ sin θ 0
−sin θ cos θ 0
0 0 1
⎞
⎠
=
˙
θ
⎛
⎝
0 −1 0
1 0 0
0 0 0
⎞
⎠
.
The transpose of L is
L
T
=
˙
θ
⎛
⎝
0 1 0
−1 0 0
0 0 0
⎞
⎠
so that the deformation gradient is given by
D =
1
2
(L +L
T
) =
˙
θ
⎛
⎝
0 0 0
0 0 0
0 0 0
⎞
⎠
= 0
and the continuum spin is
W =
1
2
(L −L
T
) =
˙
θ
⎛
⎝
0 −1 0
1 0 0
0 0 0
⎞
⎠
. (3.41)
That is, there is no stretch rate contributing to the velocity gradient so that the rate
of deformation is zero, but rigid body rotation occurs so that the continuum spin is
nonzero. Let us now examine the signiﬁcance of the continuum spin for the case of
rigid body rotation of the uniaxial rod without stretch.
Velocity gradient, rate of deformation, and continuum spin 63
Consider the rate of rotation,
˙
R, given by
˙
R =
˙
θ
⎛
⎝
−sin θ −cos θ 0
cos θ −sin θ 0
0 0 0
⎞
⎠
. (3.42)
Next, consider the product of W and R
WR =
˙
θ
⎛
⎝
0 −1 0
1 0 0
0 0 0
⎞
⎠
⎛
⎝
cos θ −sin θ 0
sin θ cos θ 0
0 0 1
⎞
⎠
=
˙
θ
⎛
⎝
−sin θ −cos θ 0
cos θ −sin θ 0
0 0 0
⎞
⎠
.
That is, for this particular case of rigid body rotation only, we see that
˙
R = WR (3.43)
so that W is the tensor that maps R onto
˙
R. Remembering that R is orthogonal so
that R
−1
= R
T
, W can be written for this simple case as
W =
˙
RR
T
. (3.44)
The spin is not itself, therefore, a rate of rotation, but it is closely related to it. Let us
use the polar decomposition theorem to examine the continuum spin in a little more
detail and more generally, and introduce the angular velocity tensor.
3.6.2 Angular velocity tensor
From Equation (3.38), the continuum spin is given by
W =
1
2
(L −L
T
)
so substituting for the velocity gradient from (3.33) gives
W =
1
2
(
˙
FF
−1
− (
˙
FF
−1
)
T
) =
1
2
(
˙
FF
−1
− (F
−1
)
T
˙
F
T
).
If we now substitute for F using the polar decomposition theorem in (3.24), after
a little algebra we obtain
W =
1
2
[
˙
RR
T
−R
˙
R
T
+R(
˙
UU
−1
− (
˙
UU
−1
)
T
)R
T
]. (3.45)
We will simplify this further by considering the product
RR
T
= I,
which we can differentiate with respect to time to give
˙
RR
T
+R
˙
R
T
= 0
64 Large deformations and continuum mechanics
so that
˙
RR
T
= −R
˙
R
T
= −(
˙
RR
T
)
T
. (3.46)
We see, therefore, that
˙
RR
T
is antisymmetric, since any tensor Z for which Z = −Z
T
is antisymmetric because
sym(Z) =
1
2
(Z +Z
T
) =
1
2
(Z −Z) = 0.
Substituting (3.46) into (3.45) gives
W =
˙
RR
T
+
1
2
R(
˙
UU
−1
− (
˙
UU
−1
)
T
)R
T
(3.47)
or
W = Ω+
1
2
R(
˙
UU
−1
− (
˙
UU
−1
)
T
)R
T
= +Rasym(
˙
UU
−1
)R
T
(3.48)
in which Ω =
˙
RR
T
is called the angular velocity tensor which depends only on the
rigid body rotation and its rate of change and is independent of the stretch. If we con
sider a deformation comprising of rigid body rotation only, as we did in Section 3.6.1,
or if the stretch is negligibly small, then Equation (3.48) simply reduces to
W = Ω =
˙
RR
T
(3.49)
as we saw in Equation (3.44). In general, the angular velocity tensor and continuum
spin are not the same; Equation (3.48) shows that they differ depending on the
stretch, U. Both W and Ω are important when considering objective stress rates,
as we shall see in Section 3.8. We will look at one further simple example in which we
examine in particular the rate of deformation and the continuum spin; that of uniaxial
stretch with no rigid body rotation.
3.6.3 Uniaxial stretch
We considered the uniaxial elongation of a rod lying along the Ydirection earlier
(see Fig. 3.4) for which we obtained the deformation gradient, assuming purely
plastic deformation and the incompressibility condition, in terms of the current, l,
and original, l
0
, rod lengths as
F =
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎝
_
l
l
0
_
−1/2
0 0
0
l
l
0
0
0 0
_
l
l
0
_
−1/2
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎠
.
Velocity gradient, rate of deformation, and continuum spin 65
We will now determine the rate of deformation and continuum spin for the case of
uniaxial stretch. Differentiating the deformation gradient gives
˙
F =
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎝
−
1
2
_
l
l
0
_
−3/2
1
l
0
˙
l 0 0
0
˙
l
l
0
0
0 0 −
1
2
_
l
l
0
_
−3/2
˙
l
l
0
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎠
and
F
−1
=
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎝
_
l
l
0
_
1/2
0 0
0
_
l
l
0
_
−1
0
0 0
_
l
l
0
_
1/2
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎠
so that the velocity gradient is
L =
˙
FF
−1
=
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎝
−
1
2
_
l
l
0
_
−1
˙
l
l
0
0 0
0
˙
l
l
0
_
l
l
0
_
−1
0
0 0 −
1
2
_
l
l
0
_
−1
˙
l
l
0
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎠
=
˙
l
l
⎛
⎜
⎜
⎜
⎜
⎝
−
1
2
0 0
0 1 0
0 0 −
1
2
⎞
⎟
⎟
⎟
⎟
⎠
.
This is a symmetric quantity and therefore equal to the rate of deformation. In addition,
its antisymmetric part is zero so that the continuum spin is zero. Formally,
D =
1
2
(L +L
T
) =
˙
l
l
⎛
⎝
−
1
2
0 0
0 1 0
0 0 −
1
2
⎞
⎠
,
W =
1
2
(L −L
T
) = 0.
(3.50)
There is, therefore, no rigid body rotation occurring; just stretch. Let us look at the
rate of deformation for this case of uniaxial stretch.
For uniaxial stretchinthe Ydirection, the true plastic straincomponents are givenby
ε
yy
= ln
l
l
0
, ε
xx
= ε
zz
= −
1
2
ε
yy
66 Large deformations and continuum mechanics
so that the strain rates are
˙ ε
yy
=
˙
l
l
, ˙ ε
xx
= ˙ ε
zz
= −
1
2
˙
l
l
and ˙ ε
xy
= ˙ ε
yz
= ˙ ε
zx
= 0.
Examination of the components of the deformation gradient in (3.50) shows, therefore,
that D, for this case, can be rewritten
D =
⎛
⎝
˙ ε
xx
0 0
0 ˙ ε
yy
0
0 0 ˙ ε
zz
⎞
⎠
. (3.51)
For uniaxial stretch, therefore, in the absence of a rigid body rotation, it can be seen
that the rate of deformation is identical to the true strain rate. This is not generally
the case; it happens to be so for this example because there is no rigid body rotation.
However, the example gives us a reasonable feel for what kind of measure the rate of
deformation is.
We nowreturn to the consideration of elastic–plastic material behaviour under large
deformation conditions. We will consider cases in which the elastic strains, while not
negligible, can always be assumed to be small compared with the plastic strains.
3.7 Elastic–plastic coupling
Consider now an imaginary lump of material in the undeformed conﬁguration, shown
in Fig. 3.6. The material contains a line vector, dX. As before, after deformation
Z
X
Current
configuration
dx
x
X
Y
dX
O
x
dp
Intermediate, stressfree,
configuration
Initial
configuration
F
F
p
F
e
Fig. 3.6 Schematic diagram showing an element of a material in the initial and current conﬁgurations
and in the intermediate, stressfree, conﬁguration.
Elastic–plastic coupling 67
to the deformed or current conﬁguration, the line vector is transformed to dx. The
transformation mapping of dX to dx is, of course, the deformation gradient, F. We
now introduce what is called the intermediate conﬁguration. In transforming from
the initial to deformed conﬁguration, the line vector dX has undergone elastic and
plastic deformation. The intermediate conﬁguration is that in which line vector dx has
been unloaded to a stressfree state; this is an imaginary state corresponding to one in
which dX in the undeformed conﬁguration has undergone purely plastic deformation
to become dp in the intermediate conﬁguration. The transformation mapping of dX
to dp is the plastic deformation gradient so that
dp = F
p
dX
and the plastic deformation gradient is deﬁned as
F
p
=
∂p
∂X
. (3.52)
In the current conﬁguration dp is deformed into dx by the elastic deformation so that
dx = F
e
dp
and the elastic deformation gradient is deﬁned as
F
e
=
∂x
∂p
. (3.53)
We may then write
dx = F
e
dp = F
e
F
p
dX
so that
F = F
e
F
p
. (3.54)
This is the classical multiplicative decomposition of the deformation gradient into
elastic and plastic parts, due to Erastus Lee.
Note that for general inhomogeneous plastic deformation, unloading a body will not
generally lead to uniform zero stress; rather a residual stress ﬁeld exists. Considering
a ﬁnite number of material points within a continuum, at each point an unstressed
conﬁguration can be obtained, but F
p
and F
e
are, strictly, no longer pointwise
continuous. Within the context of a ﬁnite element analysis, however, in which a
discretization is necessary and the discontinuity of stress and strain resulting from the
discretization is the norm, this is not a problem. In addition, note that the interme
diate conﬁguration described by p is, in general, not uniquely determined since an
arbitrary rigid body rotation can be superimposed on it, still leaving it unstressed. In
Equation (3.54), both the elastic and plastic deformation gradients may contain both
stretch and rigid body rotation.
68 Large deformations and continuum mechanics
However, in order to overcome the uniqueness problem, often, by convention, all
the rigid body rotation is lumped into the plastic deformation gradient, F
p
, such that
the elastic deformation gradient, F
e
, includes stretch only (no rigid body rotation).
As a result, F
e
is written
F
e
= V
e
(symmetric stretch)
and so
F
p
= V
p
R
in which R is the equivalent total rigid body rotation. With this convention, let us now
look at the velocity gradient and address the decomposition of the elastic and plastic
rates of deformation.
3.7.1 Velocity gradient and elastic and plastic rates of deformation
We will now determine the velocity gradient in terms of the elastic and plastic
deformation gradient decomposition given in (3.54). The velocity gradient is given by
L =
˙
FF
−1
=
∂
∂t
(F
e
F
p
)(F
e
F
p
)
−1
= (F
e
˙
F
p
+
˙
F
e
F
p
)F
p−1
F
e−1
=
˙
F
e
F
e−1
+F
e
˙
F
p
F
p−1
F
e−1
=
˙
V
e
V
e−1
+V
e
˙
F
p
F
p−1
V
e−1
.
Now,
L
e
=
˙
V
e
V
e−1
= D
e
+W
e
L
p
=
˙
F
p
F
p−1
= D
p
+W
p
so
L = L
e
+V
e
L
p
V
e−1
= D
e
+W
e
+V
e
D
p
V
e−1
+V
e
W
p
V
e−1
. (3.55)
Now,
D = sym(L) W = asym(L).
Therefore, using Equation (3.55),
D = D
e
+ sym(V
e
D
p
V
e−1
) + sym(V
e
W
p
V
e−1
) (3.56)
and
W = W
e
+ asym(V
e
D
p
V
e−1
) + asym(V
e
W
p
V
e−1
). (3.57)
In general, therefore, we see from (3.56) the result that the elastic and plastic rates of
deformation are not additively decomposed, that is,
D = D
e
+D
p
.
Objective stress rates 69
This is unlike the additive decomposition of elastic and plastic strain rates for the case
of small deformation theory, given in Equation (2.1). However, if the elastic strains
are small, then
V
e
= V
e−1
≈ I.
Also, sym(D
P
) = D
P
, sym(W
P
) = 0, since rate of deformation and spin are sym
metric and antisymmetric, respectively. Hence, for small elastic stretches (V
e
≈ I),
from Equations (3.56) and (3.57),
D = D
e
+D
P
(3.58)
and
W = W
e
+W
P
. (3.59)
This is an important and wellknown result and we will shortly see that Equation (3.58)
is often used in the implementation of plasticity models into ﬁnite element code. The
plastic rate of deformation, D
p
, like the plastic strain rate in small strain theory, is
speciﬁed by a constitutive equation. Often, in ﬁnite element implementations, the total
rate of deformation, D, is known such that if D
p
is speciﬁed by a constitutive equation,
then D
e
can be determined using (3.58) so that the stress rate may be determined using
Hooke’s law. Once we know the stress rate, we can integrate over time to determine
stress. This brings us to the ﬁnal step in our brief examination of continuummechanics.
We need to address stress rate and how to determine it in a material undergoing rigid
body rotation with respect to a ﬁxed coordinate system. We will do this in Section 3.8,
and then summarize the most important steps required in going from knowledge of
deformation to determination of stress, before introducing ﬁnite element methods in
Chapter 4.
3.8 Objective stress rates
The primary focus of this section is stress rate, but before looking at this, we need to
look at what is called material objectivity. We will start not by considering stress rate,
but by something more familiar; transformation of stress.
3.8.1 Principle of material objectivity or frame indifference
Consider the transformation of the stress tensor, σ, which undergoes a rotation
through θ relative to the (XYZ) coordinate system. The stress transformation equations,
or Mohr’s circle, tell us that the transformed stresses, σ
, with respect to the
70 Large deformations and continuum mechanics
s∆A
u
n
Reference plane
Body
∆A
∆F
τ∆A
Fig. 3.7 Schematic diagram showing a body cut by a plane with normal n generating an area of
intersection of A, subjected to resultant force F.
(XYZ) coordinate system become
σ
XX
= σ
XX
cos
2
θ + 2σ
XY
sin θ cos θ + σ
YY
sin
2
θ
σ
YY
= σ
XX
sin
2
θ − 2σ
XY
sin θ cos θ + σ
YY
cos
2
θ
σ
XY
= (σ
YY
− σ
XX
) sin θ cos θ + σ
XY
(cos
2
θ − sin
2
θ).
(3.60)
The last of the three equations enables us to determine the direction of the principal
axes, relative to the applied stress direction, and of course, the principal stresses.
It does this because along the principal directions, σ
XY
= 0 so that
tan 2θ =
σ
XY
σ
XX
− σ
YY
. (3.61)
We will look at an alternative way of dealing with stress (and indeed, other tensor
quantities such as strain) transformation. In order to do this, we need to introduce the
stress vector or surface traction.
3.8.1.1 The stress vector or traction. Figure 3.7 shows a lump of material, making
up a body which has been cut by a plane with normal direction n, with an inﬁnitesimal
area of intersection, A.
Under uniform stress state, the resultant force acting on the area A is F. The
stress vector acting on the area A is deﬁned by
t =
_
F
A
_
A→0
. (3.62)
By deﬁnition, t is a vector quantity with normal, σ, and shear, τ, components, as
shown in Fig. 3.7. We will now look at how the stress vector, t, on a particular plane
with normal n is related to the stress tensor, σ. Consider the three orthogonal planes
shown in Fig. 3.8(a) and the resulting plane ABC, reproduced in Fig. 3.8(b).
Objective stress rates 71
n
t
A
s
xy
s
xz
s
yz s
yx
s
yy
s
xx
s
zx
s
s
zz
s
zy
t
y
t
x
t
z
C
B
A
y
z
B
C
x
(a) (b)
τ
Fig. 3.8 Schematic diagram showing (a) the stress vectors acting on three orthogonal planes with
components in the x, y, and zdirections shown and (b) the resultant stress vector acting on plane ABC
which has normal n, with components n = [n
x
n
y
n
z
]
T
.
On each of the orthogonal planes in Fig. 3.8(a), a stress vector acts. For example,
on plane ‘x’, that is, the plane orthogonal to the xdirection, the stress vector is t
x
.
The stress vectors acting on the three orthogonal planes have components given by
t
x
=
⎡
⎣
σ
xx
σ
xy
σ
xz
⎤
⎦
, t
y
=
⎡
⎣
σ
yx
σ
yy
σ
yz
⎤
⎦
, t
x
=
⎡
⎣
σ
zx
σ
zy
σ
zz
⎤
⎦
. (3.63)
The convention is that on plane ‘x’, the stress components are labelled σ
xx
in the
xdirection, σ
xy
in the ydirection, and so on. If the area of plane ABC is A, then the
areas of the three orthogonal planes are given by
A
x
=
⎡
⎣
1
0
0
⎤
⎦
· An = An
x
, A
y
=
⎡
⎣
0
1
0
⎤
⎦
· An = An
y
, A
z
=
⎡
⎣
0
0
1
⎤
⎦
· An = An
z
.
(3.64a)
By consideration of equilibrium, the resultant force on plane ABC must be balanced
by the forces on the three orthogonal planes so that
tA = t
x
A
x
+t
y
A
y
+t
z
A
z
=
⎡
⎣
σ
xx
σ
xy
σ
xz
⎤
⎦
An
x
+
⎡
⎣
σ
yx
σ
yy
σ
yz
⎤
⎦
An
y
+
⎡
⎣
σ
zx
σ
zy
σ
zz
⎤
⎦
An
z
. (3.64b)
Therefore
t =
⎡
⎣
σ
xx
σ
xy
σ
xz
⎤
⎦
n
x
+
⎡
⎣
σ
yx
σ
yy
σ
yz
⎤
⎦
n
y
+
⎡
⎣
σ
zx
σ
zy
σ
zz
⎤
⎦
n
z
=
⎛
⎝
σ
xx
σ
xy
σ
xz
σ
yx
σ
yy
σ
yz
σ
zx
σ
zy
σ
zz
⎞
⎠
⎡
⎣
n
x
n
y
n
z
⎤
⎦
(3.65)
72 Large deformations and continuum mechanics
remembering that for reasons of moment, or rotational equilibrium, the stress tensor
is symmetric so that σ
xy
= σ
yx
etc. Equation (3.65) can be written in a simple way as
t = σn. (3.66)
3.8.1.2 Transformation of stress. Let us now return to the problem of transforma
tion of stress. Consider the stress vector, t, acting on a surface with normal n. At the
point of interest, the stress is fully described by the tensor σ. Under some rotation, R,
the stress vector t is transformed to t
acting on a plane with normal n
, and similarly,
the stress σ is transformed to σ
such that
t = σn and t
= σ
n
.
Under rotation, R, the vectors t and n transform according to
t
= Rt and n
= Rn
which give
t
= Rσn and n = R
T
n
.
Combining these gives
t
= RσR
T
n
but
t
= σ
n
so that ﬁnally
σ
= RσR
T
. (3.67)
We see, therefore, that unlike a vector, the stress tensor, σ, transforms according to
Equation (3.67). To see what this means in detail, let us consider the twodimensional
transformation giving Equations (3.60). The rotation matrix, R, corresponding to the
rotation of θ is
R =
_
cos θ −sin θ
sin θ cos θ
_
so that
σ
=
_
cos θ −sin θ
sin θ cos θ
__
σ
xx
σ
xy
σ
xy
σ
yy
_ _
cos θ sin θ
−sin θ cos θ
_
.
Multiplying these equations will give Equation (3.60). The stress transformation
equations, or Mohr’s circle representation, are beautifully and succinctly summarized,
therefore, in Equation (3.67). This equation also tells us how other tensor quantities
(such as strain, for example) transform. More formally, a tensor, A, is said to be frame
indifferent or objective if it rotates according to the following:
A
= QAQ
T
. (3.68)
Comparing Equation (3.67) with (3.68) shows that the Cauchy stress tensor is
objective. Let us consider a simple example to understand what is meant by objective.
Objective stress rates 73
3.8.2 Stressed rod under rotation: corotational stress
Consider a rod initially lying parallel to the Yaxis, of crosssectional area A, under
constant, axial, force P, shown schematically in Fig. 3.9(a). In this conﬁguration,
with respect to the XY coordinate system, the stresses in the rod are:
σ
XX
= 0, σ
YY
=
P
A
, σ
XY
= 0.
Acorotational (xy) coordinate systemhas been introduced which rotates with the rod.
The stresses in the rod relative to the corotational (xy) coordinate system initially are,
similarly,
σ
xx
= 0, σ
yy
=
P
A
, σ
xy
= 0.
The rodundergoes a rotationof θ as showninFig. 3.9(b), relative tothe (XY) coordinate
system. With respect to the corotational system, the rod is subjected to an unchanging
stress in the ydirection of P/A; all other stress components remain zero. The stresses
(a) Y Y
y
x
P/A
P/A
X
y
x
P/A
P/A
X
(b)
(c) Y
y
x
P/A
P/A
X
s
XX
= 0
s
YY
=P/A
s
XY
= 0
s
XX
= P/A
s
YY
= 0
s
XY
= 0
s
XX
≠ 0
s
YY
≠ 0
s
XY
≠ 0
u
Fig. 3.9 A rod of crosssectional area A undergoing rigid body rotation while subjected to axial
force P (a) in the initial conﬁguration, (b) having rotated through angle θ, and (c) after rotating through
90
◦
showing the changing stresses with respect to the material (XY) reference frame.
74 Large deformations and continuum mechanics
in the rod, however, when measured with respect to the material (XY) reference frame
can be seen to change. For example, when the rod has rotated through 90
◦
as shown
in Fig. 3.9(c), it lies parallel to the Xaxis so that now, the stresses with respect to the
material (XY) reference frame are
σ
XX
=
P
A
, σ
YY
= 0, σ
XY
= 0.
They are, therefore, completely different to what they were in the initial state shown
in Fig. 3.9(a). However, the stresses relative to the corotational system are as before;
that is,
σ
xx
= 0, σ
yy
=
P
A
, σ
xy
= 0.
The stresses relative to the corotational coordinate system, which we shall designate
by σ
, are related to those relative to the material (XY) coordinate system, which we
shall designate by σ, by
σ
= RσR
T
(3.69)
in which R is just the rotation matrix. σ
is called an objective or corotational stress
because with respect to the corotational reference frame (and indeed, the mater
ial undergoing the rotation), its stress state has not changed; it has simply rotated.
Note that the corotational stress, given in Equation (3.69) follows the requirement
of objectivity given in Equations (3.67) and (3.68). The objective stress, therefore,
results from the constitutive response of the material; it is independent of orientation
and derives from the material response rather than the rigid body rotation.
Before addressing stress rate and its objectivity, we will ﬁrst look at the objectiv
ity, or otherwise, of a number of other important quantities in large deformation
kinematics.
3.8.3 Objectivity of deformation gradient, velocity gradient, and
rate of deformation
Consider the deformation gradient F.
dx = F dX. (3.70)
After a transformation, Q, the quantities in Equation (3.70) become dx
, F
, and dX
so that
dx
= F
dX
. (3.71)
Now,
dx
= Qdx = QF dX
and dX remains unchanged under deformation, by deﬁnition, so that dX
= dX.
Objective stress rates 75
Therefore, from Equation (3.71),
F
= QF. (3.72)
F is, therefore, in fact objective and behaves like a vector because it is what is called
a two point tensor, that is, only one of its two indices is in the spatial coordinate, x.
We will now look at the objectivity of the velocity gradient, the rate of deformation,
and the continuum spin.
Differentiating Equation (3.72) gives
˙
F
=
˙
QF +Q
˙
F
so that
L
=
˙
F
F
−1
= (
˙
QF +Q
˙
F)F
−1
Q
−1
=
˙
QQ
T
+Q
˙
FF
−1
Q
T
and
L
=
˙
QQ
T
+QLQ
T
. (3.73)
Comparing with the requirement for objectivity in Equation (3.68), therefore, shows
that the velocity gradient is not objective.
The transformed velocity gradient, L
, from Equation (3.73) can be written
L
=
1
2
Q(L +L
T
)Q
T
. ,, .
Symmetric
+
1
2
Q(L −L
T
)Q
T
. ,, .
Antisymmetric
+
˙
QQ
−1
. (3.74)
From Equation (3.46) we see that
˙
QQ
T
is antisymmetric. Therefore, from (3.74),
D
= sym(L
) =
1
2
Q(L +L
T
)Q
T
= QDQ
T
and
W
= asym(L
) = QWQ
T
+
˙
QQ
−1
. (3.75)
Therefore, the deformation rate, D, is objective, but the continuumspin, W, is not. It is
useful to ask whether this is important or not. In Sections 3.8.1 and 3.8.2, we examined
the objectivity of the Cauchy stress and found that it is indeed objective. The Cauchy
stress is, therefore, a quantity for which the properties are independent of the reference
frame. This is veryimportant inthe development of constitutive equations. Anequation
which relates elastic strain to stress, for example, must be independent of the reference
frame in which the relationship is used; in other words, the constitutive equation must
provide information about the material response which is independent of rigid body
rotation. The same holds for constitutive equations relating rate of plastic deformation
(which we have just shown to be objective) to Cauchy stress (also objective), or the
constitutive equation relating stress rate to elastic rate of deformation. In fact, plasticity
76 Large deformations and continuum mechanics
problems, especially within the context of ﬁnite element implementations, are often
formulated in rate form. It is therefore necessary for us to address stress rate and, in
particular, to examine whether stress rate is objective or otherwise. It turns out that
there are many measures of objective stress rate, and we will focus on one in particular;
the Jaumann stress rate.
3.8.4 Jaumann stress rate
We will examine, ﬁrst, a rather contrived problemin order to get a physical feel for the
meaning of the Jaumann stress rate. We will then look more generally at the objectivity
of this stress rate, and ﬁnish by looking at a simple example.
Consider again a rod under axial stress σ, shown in Fig. 3.10(a). The stress tensor,
with respect to the material axes is σ
t
. During a time increment of t , the rod undergoes
a rigid body rotation such that at time t +t , the rod lies as shown in Fig. 3.10(b) in
which the corotational reference axes (xy) now coincide with the material reference
axes (XY).
(a) Y
y
x
X
(b)
(c) Y
X
Y
y
x
X
Time =t
Time =t +∆t
s+∆s
s+∆s
Time =t +∆t
y
x
s
s
s
s
∆u
0 0 0
0 0
0 0 0
s
R
t +∆t
= s
s+∆s
0 0 0
0 0
0 0 0
s
t +∆t
=
Fig. 3.10 Arod under axial stress, σ, undergoing rigid body rotation (a) at a time t , (b) at a time t +t
having gone through incremental rotation R (corresponding to angle θ about the Zdirection) and
(c) at the same time t + t but having also undergone stress increment σ.
Objective stress rates 77
The stress tensor σ
with respect to the corotational reference frame is
σ
t
=
⎛
⎝
0 0 0
0 σ 0
0 0 0
⎞
⎠
,
which is obtained from
σ
t
= Rσ
t
R
T
, (3.76)
where R is the incremental rigid body rotation. Following the rigid body rotation,
the rod is subjected to an additional axial stress, σ, shown in Fig. 3.10(c), so that
the ﬁnal stress tensor, with respect to the corotational frame (and the material frame
since they are coincident at time t + t ) is
σ
t +t
=
⎛
⎝
0 0 0
0 σ + σ 0
0 0 0
⎞
⎠
.
The Cauchy stress increment with respect to the corotational frame may be written
approximately, using Equation (2.99), as
σ
= [2GD
e
+ λTr(D
e
)I]t, (3.77)
where we have used the rate of elastic deformation in place of the elastic strain rate.
The stress increment results, therefore, purelyfromthe material’s constitutive response
and is corotational. We may then write the corotational stress tensor at time t + t
as the sum of (3.76) and (3.77) to give
σ
t +t
≡ σ
t +t
= Rσ
t
R
T
+ [2GD
e
+ λTr(D
e
)I]t. (3.78)
In order to investigate this further, let us consider the case in which the incremental
rigid body rotation, R, is small. We may then approximate the rotation matrix as
R = exp[ˆ r] ≈ I + ˆ r
in which ˆ r is the associated antisymmetric tensor which is approximately given
by Wt (a full explanation for this can be found, e.g., in Belytschko et al., 2000).
Substituting into (3.78) then gives
σ
t +t
= (I +Wt )σ
t
(I +Wt )
T
+ [2GD
e
+ λTr(D
e
)I]t
= σ
t
+σ
t
W
T
t +Wσ
t
t +Wσ
t
W
T
t
2
+ [2GD
e
+ λTr(D
e
)I]t
so that
σ
t +t
−σ
t
t
= σ
t
W
T
+Wσ
t
+σ
t
Wσ
t
t + [2GD
e
+ λTr(D
e
)I].
78 Large deformations and continuum mechanics
Now, both σ
t
and σ
t +t
are given with respect to the material reference frame so that
taking the limit, and letting t → 0, gives us the material stress rate, ˙ σ as
˙ σ = σ
t
W
T
+Wσ
t
+ 2GD
e
+ λTr(D
e
)I
and as W is antisymmetric (we saw earlier that this means W
T
= −W), we obtain
˙ σ = Wσ
t
−σ
t
W + 2GD
e
+ λTr(D
e
)I. (3.79)
We can rewrite this as
˙ σ =
∇
σ +Wσ
t
−σ
t
W, (3.80)
where
∇
σ = 2GD
e
+ λTr(D
e
)I. (3.81)
∇
σ in Equations (3.80) and (3.81) is called the Jaumann stress rate and it is the stress
rate that results purely fromthe constitutive response of the material and not fromrigid
body rotation. It is, therefore, as we shall see in Section 3.8.4.1, an objective stress
rate. We may therefore use it in constitutive equations such as (3.81) in which both
the Jaumann stress rate and the elastic rate of deformation are objective quantities.
The material stress rate, ˙ σ however, does depend upon rigid body rotation. ˙ σ is the
Cauchy stress rate with respect to the material reference frame; that is, in Figs 3.9
and 3.10, ˙ σ gives the stress rate with respect to the material (XY) axes. In ﬁnite element
simulations, we are ordinarily interested in the stresses with respect to the material
axes. Equation (3.80) is therefore important in that it enables us to determine the
required stresses from knowledge of the material’s constitutive response given by
the Jaumann stress rate in (3.81). It is useful to remember that the stresses, σ
t
, in
Equation (3.80), are also given with respect to the material reference frame.
3.8.4.1 Objectivity of Jaumann stress rate. Let us now show that the Jaumann
stress rate is objective. We saw earlier that a quantity A is said to be objective if it
transforms according to (3.68), that is
A
= QAQ
T
and that the Cauchy stress transforms in this way (σ
= QσQ
T
). Differentiating
the stress with respect to time gives
˙ σ
=
˙
QσQ
T
+Q( ˙ σQ
T
+σ
˙
Q
T
) =
˙
QσQ
T
+Q˙ σQ
T
+Qσ
˙
Q
T
. (3.82)
We see from Equation (3.82) that the material rate of Cauchy stress is not objective
(even though the stress itself is) since it does not transformaccording to (3.68). We saw
in Section 3.8.3, Equation (3.75) that
W
= QWQ
T
+
˙
QQ
−1
.
Objective stress rates 79
Rearranging and remembering that Q is orthogonal gives
˙
Q = W
Q−QW (3.83)
so that
˙
Q
T
= −Q
T
W
+WQ
T
(3.84)
since both W and W
are antisymmetric. Substituting (3.83) and (3.84) into (3.82)
gives
˙ σ
= W
QσQ
T
−QWσQ
T
+Q˙ σQ
T
+QσWQ
T
−QσQ
T
W
so that
˙ σ
= Q(σW −Wσ + ˙ σ)Q
T
+W
QσQ
T
−QσQ
T
W
. (3.85)
Substituting for QσQ
T
= σ
into (3.85) gives
σ
W
−W
σ
+ ˙ σ
= Q(σW −Wσ + ˙ σ)Q
T
. (3.86)
Equation (3.86) therefore shows that the Jaumann stress rate,
∇
σ, satisﬁes the
requirement for objectivity in Equation (3.68), where
∇
σ = ˙ σ +σW −Wσ. (3.87)
There are objective stress rates other than that of Jaumann. Details of these may be
found in many of the more advanced text books on continuum mechanics, and in
particular, that of Belytschko et al. (2000).
3.8.4.2 Example of Jaumann stress rate for a rotating rod. We will conclude
Section 3.8.4 with a simple example of a rotating rod under constant, uniaxial stress.
We refer back to Fig. 3.9(b) which shows the rotating bar at an instant at which it
makes an angle θ with the vertical. It is subject to a constant, uniaxial stress of P/A
at all times. This, of course, gives a constant corotational stress (with respect to
corotational xyaxes) of
σ
=
⎛
⎝
0 0
0
P
A
⎞
⎠
. (3.88)
The stresses with respect to the material (XY) axes, however, change with the rotation,
as shown in Fig. 3.9(a)–(c). This means that there is a rate associated with each
stress component which also changes with the rotation. We will now determine the
stresses in the rod with respect to the material axes as it rotates. We will do this in two
ways; ﬁrst, we will set up the problem in rate form and integrate to obtain the stresses
and second, use the standard method for transformation of stress using Equation (3.67)
(or equivalently, Mohr’s circle).
80 Large deformations and continuum mechanics
We may obtain the objective, or corotational stress rate,
∇
σ, simply by differentiating
(3.88) with respect to time. As the corotational stress is always constant, its derivative
is zero, so
∇
σ =
∂
∂t
⎛
⎝
0 0
0
P
A
⎞
⎠
= 0.
The stress rate with respect to the original conﬁguration, ˙ σ, is given by rearran
ging (3.87)
˙ σ =
∇
σ −σW +Wσ = Wσ −σW. (3.89)
If there is no rotation, then W = 0 and (3.89) tells us that the stress rates with respect
to the material axes are also zero. However, the rod is rotating, with constant angular
speed
˙
θ. We showed for this case, Equation (3.41), that this results in a continuum
spin given by
W =
˙
θ
_
0 −1
1 0
_
. (3.90)
The stresses with respect to the material axes may be written
σ =
_
σ
XX
σ
XY
σ
XY
σ
YY
_
. (3.91)
Substituting (3.90) and (3.91) into (3.89) gives
˙ σ =
˙
θ
_
−2σ
XY
σ
XX
− σ
YY
σ
XX
− σ
YY
2σ
XY
_
.
That is
dσ
XX
dt
= −2
˙
θσ
XY
≡ −
dσ
YY
dt
(3.92)
and
dσ
XY
dt
=
˙
θ(σ
XX
− σ
YY
). (3.93)
Equation(3.92) gives σ
XX
= −σ
YY
+k where k is just a constant. The initial conditions
are σ
XX
(0) = 0, σ
YY
(0) = P/A, σ
XY
(0) = 0 so that k = P/A. Differentiating
(3.93) and substituting into (3.92) and using k = P/A gives
d
2
σ
XX
dt
2
= +4
˙
θ
2
σ
XX
= 2
˙
θ
2
P
A
.
This has general solution
σ
XX
= Asin 2θ + B cos 2θ +
P
2A
,
Summary 81
where θ =
˙
θt , so that with the initial conditions, the full solution is
σ
XX
=
P
A
sin
2
θ, σ
YY
=
P
A
cos
2
θ, σ
XY
=
P
A
sin θ cos θ. (3.94)
We see, therefore, that the stresses with respect to the material reference frame change
correctly with angle θ. For example, when θ = 0, σ
XX
= 0, σ
YY
= P/A, σ
XY
= 0,
and when θ = π/2, σ
XX
= P/A, σ
YY
= 0, σ
XY
= 0.
Finally, we will determine the same stresses using the stress transformation
equation in (3.67). That is,
σ
=RσR
T
↑ ↑
(x, y) (X, Y)
reference reference
so that the stresses in the material (XY) reference are given by
σ =
_
cos θ sin θ
−sin θ cos θ
_
⎛
⎝
0 0
0
P
A
⎞
⎠
_
cos θ −sin θ
sin θ cos θ
_
=
P
A
_
sin
2
θ sin θ cos θ
sin θ cos θ cos
2
θ
_
,
which just gives the expressions in Equation (3.94).
3.9 Summary
Before leaving the kinematics of large deformations, we will summarize some of
the important steps required in going from knowledge of deformation through to the
determination of stresses for an elastic–plastic material undergoing large deforma
tions. This is often required in the implementation of plasticity models into ﬁnite
element code so it is something we shall return to later. We assume that any deforma
tion taking place is such that the stretches due to elasticity are small compared with
those for plasticity so that the additive decomposition of rate of deformation given in
Equation (3.58) holds. We also assume full knowledge of the deformation gradient,
F and its rate,
˙
F. The steps required in determining stresses are then as follows.
1. Determine the velocity gradient
L =
˙
FF
−1
.
2. Determine the rates of deformation and continuum spin
D = sym(L) =
1
2
(L +L
T
),
W = asym(L) =
1
2
(L −L
T
).
82 Large deformations and continuum mechanics
3. The rate of plastic deformation is speciﬁed by a constitutive equation. For example,
from Chapter 2, for combined isotropic and kinematic hardening with power
law dependence of effective plastic strain rate on stress, we have in the current
conﬁguration
D
p
=
3
2
_
J(σ
−x
) − r − σ
y
K
_
1/m
σ
−x
J(σ
−x
)
.
4. Determine the rate of elastic deformation
D
e
= D −D
p
.
5. Determine the Jaumann stress rate using the tensor form of Hooke’s law,
Equation (3.81), from the rate of elastic deformation
∇
σ = 2GD
e
+ λTr(D
e
)I.
6. Determine the material rate of stress using (3.80)
˙ σ =
∇
σ +Wσ −σW.
7. Use a numerical technique to obtain the stresses with respect to the material
reference frame by integrating ˙ σ.
We will address all of these steps in some detail in later chapters.
Further reading
Belytschko, T., Liu, W.L., and Moran, B. (2000). Nonlinear Finite Elements for
Continua and Structures. John Wiley & Sons Inc, New York.
Khan, A.S. and Huang, S. (1995). ContinuumTheory of Plasticity. John Wiley &Sons
Inc, New York.
Lubarda, V.A. (2002). Elastoplasticity Theory. CRC Press, Florida, USA.
Simo, J.C. and Hughes, T.J.R. (1997). Computational Inelasticity. SpringerVerlag,
Berlin.
4. The ﬁnite element method for static
and dynamic plasticity
4.1 Introduction
There are fewpractical problems in plasticity which can be solved analytically. This is
usually because of irregular geometry and/or complicated boundary and loading condi
tions. Computational mechanics, and in particular, the ﬁnite element method, enables
the approximate solution of these types of problems. The important requirements of
equilibrium, andcompatibility, together witha material’s constitutive response, enable
solutions to be obtained subject to the satisfaction of initial and boundary conditions.
Satisfying the requirements of equilibrium, compatibility, constitutive equations, and
boundary conditions is essential for the solution of any solid mechanics problem.
Within the ﬁnite element method, the body under consideration is discretized into
a ﬁnite number of elements and nodes, with the latter each having a speciﬁed number
of degrees of freedom. The ﬁnite element model representing the body therefore con
tains a ﬁnite number of degrees of freedom and the implication is that the requirement
for equilibrium cannot be satisﬁed exactly at every point in the continuum (in what is
called the strong sense). Instead, within the ﬁnite element technique, a weak formu
lation of equilibrium is used in which global equilibrium for the body as a whole is
imposed even though this does not necessarily ensure pointwise equilibrium. Further
details of strong and weak formulations are available in texts that are more specialized.
A weak formulation can be obtained by consideration of the principle of virtual work,
which appears in many text books, but here, we shall use Hamilton’s principle. The
advantage, as we shall see, is that the equilibriumequations of motion can be obtained
in a fully uniﬁed manner, for both quasistatic and dynamic problems, and that the
boundary conditions for the problem are also obtained. We shall use Hamilton’s prin
ciple to obtain the equations of motion (and boundary conditions) for a number of
well known discrete and continuous systems. In this way, we aim to develop a good
physical feel for what the principle is doing before addressing more complicated ﬁnite
element applications.
84 Finite element method
We introduce Hamilton’s principle in Section 4.2. A reasonable knowledge of
vector calculus is useful, but the important results are not impenetrable without it!
Unfamiliar readers may like to skip on to the introduction to the ﬁnite element method
in Section 4.3. Here, in keeping with the aims of this book, we address ﬁrst a simple
onedimensional rod element subjected to elastic deformation alone and then very
brieﬂy look at element assemblage and some other ﬁnite element types. We then
return to elastic–plastic problems and address plasticity in a onedimensional rod
element.
4.2 Hamilton’s principle
Hamilton’s principle, which results from conservation of energy, is one of the most
general principles of mechanics. It provides a means for ﬁnding the equilibrium
equations (equations of motion) of a dynamical system by determining the stationary
value of a scalar integral. The principle states that the variation of the kinetic and
potential energy plus the variation of the work done by nonconservative (external)
forces acting during any time interval t
1
to t
2
must be zero. We will shortly see what
is meant by variation, and of course, nonconservative implies forces that cannot
be described by the change in a potential energy function (as can strain energy, for
example), which are not already included in the potential energy term.
Let the total kinetic energy of the system be T , the potential energy of the system
be U, and the work done by nonconservative forces be W. The Lagrangian, L, is
deﬁned as
L = T − U + W (4.1)
and the Jintegral is deﬁned as
J =
_
t
2
t
1
Ldt =
_
t
2
t
1
(T − U + W) dt . (4.2)
Hamilton’s principle states that the ﬁrst variation (denoted by δ) of J is zero thus
δJ =
_
t
2
t
1
δLdt =
_
t
2
t
1
δ(T − U + W
nc
) dt = 0. (4.3)
In other words, the motion of a system between speciﬁed, realizable initial, and ﬁnal
conditions at times t
1
to t
2
is such that the average value of Lrelative to any dynamical
path compatible with the physical constraints has a stationary value. An illustration
of such a motion is given in Fig. 4.1 for a single degree of freedom system (SDOF).
Generally, the kinetic energy, potential energy, and work of nonconservative forces
depend on some function (e.g. displacement, temperature, etc.) of time y(t ) which
Hamilton’s principle 85
t
2
t
1
t
x
x(t )
dx
Two dynamical paths compatible
with constraints (the difference
between trajectories is dx)
x(t )
Fig. 4.1 Two dynamical paths subjected to having the same state at times t
1
and t
2
.
can be expressed as follows
J =
_
t
2
t
1
Ldt =
_
t
2
t
1
L(y, y
, t ) dt (4.4)
in which y
is just ˙ y = dy/dt .
Hamilton’s principle tells us that the dynamical path, y(t ), is that which leads to a
stationary value of J. Let us now look at how Equation (4.4), called a functional, may
be minimized by use of the calculus of variations.
Let y = y(t ) be the actual minimizing curve (as distinct fromany admissible curve)
and let
Y(t ) = y(t ) + εη(t ) (4.5)
be the family of comparison curves, where η(t ) is an arbitrary function subjected to
the constraint that
η(t
1
) = η(t
2
) = 0 (4.6)
and ε is an arbitrary parameter. The corresponding J integral is
¯
J(ε) =
_
t
2
t
1
L(Y, Y
, t ) dt (4.7)
and its variation with respect to ε is
∂
∂ε
¯
J(ε) =
_
t
2
t
1
_
∂L
∂Y
∂Y
∂ε
+
∂L
∂Y
∂Y
∂ε
_
dt . (4.8)
For ε = 0, and using (4.5), this variation yields by integrating by parts, as
∂
∂ε
¯
J(0) =
_
t
2
t
1
_
∂L
∂y
η +
∂L
∂y
η
_
dt =
_
t
2
t
1
∂L
∂y
η dt +
_
∂L
∂y
η
_
t
2
t
1
−
_
t
2
t
1
η
d
dt
∂L
∂y
dt
=
_
t
2
t
1
_
∂L
∂y
−
d
dt
∂L
∂y
_
η dt +
_
∂L
∂y
η
_
t
2
t
1
,
86 Finite element method
which for η(t
1
) = η(t
2
) = 0 yields
∂
∂ε
¯
J(0) =
_
t
2
t
1
_
∂L
∂y
−
d
dt
∂L
∂y
_
η dt . (4.9)
Therefore, the necessary condition for minimization of J is
∂
∂ε
¯
J(0) =
∂
∂ε
¯
J(ε) = 0. (4.10)
Hence,
_
t
2
t
1
_
∂L
∂y
−
d
dt
∂L
∂y
_
η dt = 0.
Since η is arbitrary, it follows that
∂L
∂y
−
d
dt
∂L
∂y
= 0, (4.11)
which is known as Euler–Lagrange equation and which must be satisﬁed for J to have
a minimum.
If the function y(t ) is replaced by y + δy where δy is the variation of y such that
δy(t
1
) = δy(t
2
) = 0, (4.12)
then it follows using Taylor’s theorem that
J(y + δy) = J(y) +
_
t
2
t
1
_
∂L
∂y
δy +
∂L
∂y
(δy)
_
dt
+
1
2!
_
t
2
t
1
_
∂
2
L
∂y
2
δy
2
+ 2
∂
2
L
∂y∂y
δy(δy)
+
∂
2
L
∂y
2
(δy
)
2
_
dt
+
1
3!
_
t
2
t
1
_
δy
∂
∂y
+ δy
∂
∂y
_
3
L(y, y
, t ) dt + · · · (4.13)
since
F(x + r, y + s, z + t ) ≈ F(x, y, z) + r
∂F
∂x
+ s
∂F
∂y
+ t
∂F
∂z
+
1
2!
_
r
∂
∂x
+ s
∂
∂y
+ t
∂
∂z
_
2
F(x, y, z) + · · · .
The quantity
δJ =
_
t
2
t
1
_
∂L
∂y
δy +
∂L
∂y
(δy)
_
dt (4.14)
Hamilton’s principle 87
is called the ﬁrst variation while
δ
2
J =
1
2!
_
t
2
t
1
_
∂
2
L
∂y
2
δy
2
+ 2
∂
2
L
∂y∂y
δy(δy)
+
∂
2
L
∂y
2
(δy
)
2
_
dt
is the second variation of J.
If the ﬁrst variation is equal to zero for a particular function y(t ) of the admissible
class, the functional J(y) is saidtohave a stationary value for that particular functiony.
Before returning to equilibrium equations, let us look at one simple example to help
understand the process of ﬁnding a stationary value.
4.2.1 Stationary value: minimizing the distance between two points
Figure 4.2 shows two points (x
1
, y
1
) and (x
2
, y
2
) which are joined by an inﬁnite
number of possible paths, y(x). Only one is shown in the ﬁgure! Here note that y(x)
is a function only of position, x, and not time. Our aim is to use the above approach
to determine that path which minimizes the distance between the two points.
An element of length dl lies on the path shown, and its length can be written as
dl
2
= dx
2
+ dy
2
= dx
2
_
1 +
_
dy
dx
_
2
_
such that
dl = (1 + y
2
)
1/2
dx
and ﬁnally, we obtain the functional, or J integral as
J = l =
_
x
2
x
1
(1 + y
2
)
1/2
dx. (4.15)
Equation (4.15) is the functional which we wish to minimize to obtain the shortest
path. We will use two approaches to do this. The ﬁrst is by obtaining directly the
dl
dy
dx
x
2
(x
2
, y
2
)
y (x)
(x
1
, y
1
)
x
1
x
y
Fig. 4.2 A particular path, y(x), between two points, (x
1
, y
1
) and (x
2
, y
2
), and an element, dl, lying
on the path.
88 Finite element method
stationary value of (4.15) and the second is to do the same by using the Euler–Lagrange
Equation (4.11). The ﬁrst variation of (4.15) is given by
δJ = δ
_
x
2
x
1
(1 + y
2
)
1/2
dx =
_
x
2
x
1
1
2
(1 + y
2
)
−1/2
δ(y
2
) dx
=
_
x
2
x
1
(1 + y
2
)
−1/2
y
δy
dx.
Now,
δy
= δ
_
dy
dx
_
=
d
dx
(δy)
so
δJ =
_
x
2
x
1
(1 + y
2
)
−1/2
y
d
dx
(δy) dx.
Integrating by parts gives
δJ = [y
(1 + y
2
)
−1/2
δy]
x
2
x
1
−
_
x
2
x
1
d
dx
((1 + y
2
)
−1/2
y
) δy dx
and since δy(x
1
) = δy(x
2
) = 0, and for a stationary value, we obtain
δJ = −
_
x
2
x
1
d
dx
((1 + y
2
)
−1/2
y
) δy dx = 0.
Since δy is arbitrary, we obtain the differential equation
d
dx
((1 + y
2
)
−1/2
y
) = 0, (4.16)
which (unsurprisingly) has the solution y = Ax +B in which A and B are constants
of integration. We see, therefore, that the process of ﬁnding the ﬁrst variation
(or equivalently, the stationary value) of J, results in ﬁnding the function y(x) which
minimizes the functional, J. To ﬁnish off, let us do the same thing but now using the
Euler–Lagrange Equation (4.11).
The functional, J, is a function of y, y
, and x and from Equation (4.15) may be
written as
J =
_
x
2
x
1
L(y, y
, x) dx,
where L(y, y
, x) = (1 + y
2
)
1/2
so the Euler–Lagrange equation becomes
∂L
∂y
−
d
dx
∂L
∂y
= 0. (4.17)
Now,
∂L
∂y
= 0 and
∂L
∂y
=
1
2
(1 + y
2
)
−1/2
2y
and substituting into (4.17) just gives us the differential Equation (4.16) which has the
solution as before.
Hamilton’s principle 89
4.2.2 Equilibrium equations
Let us nowreturntoHamilton’s principle tosee howit canbe usedtoobtainequilibrium
equations. For a solid elastic body, in the absence of body forces, the components of
the Lagrangian can be expressed as follows:
T =
1
2
_
ρ ˙ u · ˙ udV,
U =
1
2
_
σ : ε dV, (4.18)
W =
_
∂
t · udA
in which u and ˙ u are the displacement and velocity vectors, ρ the density, t the stress
vector, or traction, σ and ε the stress and strain tensors respectively, and ∂and are
domains of area, A and volume, V, respectively. Substituting for the stress vector, t,
from Equation (3.66), and making use of the divergence theorem gives
W =
_
∂
σn · udA =
_
div [σu] dV. (4.19)
The J integral is, therefore
J =
_
t
2
t
1
(T − U + W) dt
=
_
t
2
t
1
_
1
2
_
ρ ˙ u · ˙ udV −
1
2
_
σ : ε dV +
_
div [σu] dV
_
dt.
Taking the ﬁrst variation of J gives
δJ =
_
t
2
t
1
δ(T − U + W) dt
=
_
t
2
t
1
_
1
2
_
2ρ ˙ u · δ ˙ udV −
1
2
_
(σ : δε + δσ : ε) dV +
_
div [σδu] dV
_
dt
=
_
t
2
t
1
__
ρ ˙ u ·
∂
∂t
δudV −
_
σ : δε dV +
_
div [σδu] dV
_
dt = 0
and noting that div [σδu] = div [σ] · δu +σ : ∇δu and ∇δu = δε and integrating by
parts gives
δJ =
_
t
2
t
1
_
−
_
ρ ¨ u · δudV +
_
div σδudV
_
dt
=
_
t
2
t
1
_
−
_
ρ ¨ udV +
_
div σ dV
_
· δudt = 0
90 Finite element method
so that
_
(−ρ ¨ u + div σ) · δudV = 0 (4.20)
and since δu is arbitrary,
−ρ ¨ u + div σ = 0.
These are the wellknown equilibrium equations of stress analysis. If we consider
quasistatic conditions, ρ ¨ u = 0, and expand div σ, we obtain the more familiar
expressions
div σ =
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
∂σ
xx
∂x
+
∂σ
xy
∂y
+
∂σ
xz
∂z
∂σ
yx
∂x
+
∂σ
yy
∂y
+
∂σ
yz
∂y
∂σ
zx
∂x
+
∂σ
zy
∂y
+
∂σ
zz
∂z
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
= 0.
Hamilton’s principle has therefore given us the very general form of the equilibrium
equations of stress analysis. In addition, Equation (4.20) can be seen to be a statement
of the principle of virtual work. The term
_
(−ρ ¨ u + div σ) dV is just the residual
force, r, so that the expression in (4.20) is simply δW = r · δu = 0. That is, in
the course of an arbitrary, inﬁnitesimal virtual displacement, δu, from a position of
equilibrium, the work done is zero. Hamilton’s principle has therefore provided us
with the condition for equilibrium and, in doing so, a statement of the principle of
virtual work.
Before moving on to its application to ﬁnite elements, let us consider a few more
examples, one for a discrete system, and two further continuous systems, in which
we shall employ Hamilton’s principle to obtain the momentum balance equations for
a number of simple problems.
4.2.3 Further examples of Hamilton’s principle
4.2.3.1 Discrete spring–mass problem. Consider the SDOF spring–mass system
shown in Fig. 4.3. We will use Hamilton’s principle to derive the equation of motion
for the mass, m.
m
k
x(t )
Fig. 4.3 Simple SDOF mass and spring system.
Hamilton’s principle 91
We may write expressions for the kinetic and potential energies of the
system as
U =
1
2
kx
2
, T =
1
2
m˙ x
2
.
The J integral may then be written, in the absence of external forces, as
J =
_
t
2
t
1
(T − U) dt =
_
t
2
t
1
_
1
2
m˙ x
2
−
1
2
kx
2
_
dt .
Taking the ﬁrst variation gives
δJ =
1
2
_
t
2
t
1
(2m˙ x δ ˙ x − 2kx δx) dt =
_
t
2
t
1
_
m˙ x
d
dt
(δx) − kx δx
_
dt
and integrating by parts, we get
δJ = [m˙ x δx]
t
2
t
1
−
_
t
2
t
1
(m¨ x δx + kx δx) dt
and since δx(t
1
) = δx(t
2
) = 0, and δx is arbitrary,
δJ = −
_
t
2
t
1
(m¨ x + kx) δx dt = 0 so m¨ x + kx = 0.
Hamilton’s principle has given us the governing equation of motion, or equilibrium
equation, for the well known mass–spring problem.
4.2.3.2 Transverse vibration of a continuous beam. We shall next consider
a continuous system and use Hamilton’s principle to obtain the equations of motion
for it. Figure 4.4(a) shows an elastic beam undergoing bending with deﬂection, y.
The bending moment variation with angle, φ, is shown in Fig. 4.4(b) assuming elastic
behaviour.
y
y
x
M
2
1
dU= Mdf
df f
(a)
(b)
f
f+
−f
−x
Fig. 4.4 (a) An elastic beam in bending undergoing deﬂection y and rotation φ and (b) the linear
variation of bending moment M with rotation φ leading to stored elastic strain energy U.
92 Finite element method
The elastic strain energy stored in an element of length dx of the beam is
dU =
1
2
M dφ ≈
1
2
M
∂φ
∂x
dx
and
φ =
∂y
∂x
so that
dU =
1
2
M
∂
2
y
∂x
2
dx.
For an elastic beam, the bending moment is related to the deﬂection, y, by
M = EI
∂
2
y
∂x
2
,
where E is Young’s modulus and I is the second moment of area. Thus,
dU =
1
2
EI
_
∂
2
y
∂x
2
_
2
dx
and ﬁnally, the potential energy for a beam of length L is
U =
1
2
_
L
0
EI
_
∂
2
y
∂x
2
_
2
dx. (4.21)
The kinetic energy, in an element of length dx, of the beam, is
dT =
1
2
ρA
_
∂y
∂t
_
2
dx
in which A is the beam’s crosssectional area. Hence the total system kinetic
energy is
T =
1
2
_
L
0
ρA
_
∂y
∂t
_
2
dx. (4.22)
The work done by external forces, F(x, t ) per unit length, is given by
W =
_
L
0
F(x, t )y dx. (4.23)
Consider the simply supported beam undergoing transverse vibration shown
in Fig. 4.5.
We apply Hamilton’s principle in the usual way. The J integral is
J =
_
t
2
t
1
Ldt
=
1
2
_
t
2
t
1
_
_
L
0
ρA
_
∂y
∂t
_
2
dx −
_
L
0
EI
_
∂
2
y
∂x
2
_
2
dx +
_
L
0
F(x, t )y dx
_
dt .
Hamilton’s principle 93
F(x, t)
dx
Fig. 4.5 A simply supported beam with distributed load F(x, t ) in transverse vibration.
We take the ﬁrst variation of J to give the stationary value and equate to zero
δJ =
_
t
2
t
1
δLdt = 0.
Let us take the variation termby term. The contribution fromthe work done by external
forces is obtained by substituting from (4.23) to give
_
t
2
t
1
δW dt =
_
t
2
t
1
_
L
0
F(x, t ) δy dx dt .
The kinetic energy term, substituting for (4.22), is
_
t
2
t
1
δT dt =
1
2
_
t
2
t
1
ρAδ
_
L
0
_
∂y
∂t
_
2
dx dt
=
1
2
_
t
2
t
1
ρA
_
L
0
2
_
∂y
∂t
_
δ
_
∂y
∂t
_
dx dt
=
_
t
2
t
1
ρA
_
L
0
_
∂y
∂t
_
δ
_
∂y
∂t
_
dx dt
= ρA
_
t
2
t
1
_
L
0
∂y
∂t
∂
∂t
δy dx dt
= ρA
_
L
0
_
_
∂y
∂t
δy
_
t
2
t
1
−
_
t
2
t
1
δy
∂
2
y
∂t
2
dt
_
dx
but
δy(t
1
) = δy(t
2
) = 0
thus
_
t
2
t
1
δT dt = −ρA
_
L
0
_
t
2
t
1
δy
∂
2
y
∂t
2
dt dx.
94 Finite element method
The elastic strain energy term is, using (4.21),
_
t
2
t
1
δU dt =
1
2
_
t
2
t
1
δ
_
L
0
EI
_
∂
2
y
∂x
2
_
2
dx dt =
_
t
2
t
1
_
L
0
EI
2
2
∂
2
y
∂x
2
δ
_
∂
2
y
∂x
2
_
dx dt
= EI
_
t
2
t
1
_
L
0
∂
2
y
∂x
2
∂
2
∂x
2
δy dx dt
= EI
_
t
2
t
1
_
_
∂
2
y
∂x
2
∂
∂x
δy
_
L
0
−
_
L
0
∂
3
y
∂x
3
∂
∂x
δy dx
_
dt
= EI
_
t
2
t
1
_
_
∂
2
y
∂x
2
∂
∂x
δy
_
L
0
−
_
∂
3
y
∂x
3
δy
_
L
0
+
_
L
0
∂
4
y
∂x
4
δy dx
_
dt.
Finally, summing the terms,
δJ =
_
t
2
t
1
_
L
0
−
_
EI
∂
4
y
∂x
4
+ ρA
∂
2
y
∂t
2
_
δy dx dt
− EI
_
t
2
t
1
_
_
∂
2
y
∂x
2
∂
∂x
δy
_
L
0
+
_
∂
3
y
∂x
3
δy
_
L
0
_
dt +
_
t
2
t
1
_
L
0
F(x, t ) δy dx dt
=
_
t
2
t
1
_
L
0
−
_
EI
∂
4
y
∂x
4
+ ρA
∂
2
y
∂t
2
− F
_
δy dx dt
+ EI
_
t
2
t
1
_
_
∂
2
y
∂x
2
∂
∂x
δy
_
L
0
−
_
∂
3
y
∂x
3
δy
_
L
0
_
dt = 0. (4.24)
For δJ tobe zerothe integrands between0andLmust be zeroandsince δy is arbitraryit
follows that the individual terms in (4.24) must also vanish so that (assuming uniform,
constant transverse loading, F),
EI
∂
4
y
∂x
4
+ ρA
∂
2
y
∂t
2
= F (4.25)
and in addition, the boundary conditions are
∂
2
y
∂x
2
∂
∂x
δy = 0 for x = 0 and x = L,
∂
3
y
∂x
3
δy = 0 for x = 0 and x = L,
Hamilton’s principle 95
which means
either
∂
2
y
∂x
2
= 0 or
∂
∂x
δy = 0 for x = 0 and x = L
and (4.26)
either
∂
3
y
∂x
3
= 0 or δy = 0 for x = 0 and x = L.
Equation (4.25) is the equilibrium equation for the vibrating beam problem. However,
note that Hamilton’s principle has also given us the boundary conditions for the
problem in Equations (4.26). We shall ﬁnish by looking at one further example.
4.2.3.3 Transverse vibrationof a beamwithaxial loading. We shall nowconsider
the previous problem but with the addition of an externally applied axial force, P, as
shown in Fig. 4.6.
The axial force does work as the beam shortens due to transverse displacement
(here, we are ignoring axial displacement which results from the axial strain due to
the force, P). An element of the beam, ds, shortens by
ds − dx =
_
dx
2
+ dy
2
− dx =
_
1 +
_
dy
dx
_
2
dx − dx
=
⎧
⎨
⎩
_
1 +
_
dy
dx
_
2
− 1
⎫
⎬
⎭
dx ≈
1
2
_
∂y
∂x
_
2
dx.
The work done by the axial force per differential length
dW
P
=
1
2
P
_
∂y
∂x
_
2
dx.
F(x, t )
dx
P P
ds
dy
dx
Fig. 4.6 Simply supported beam subjected to axial force P while undergoing transverse vibration.
96 Finite element method
The ﬁrst variation of the work done by external forces then becomes
_
t
2
t
1
δWdt =
_
t
2
t
1
_
L
0
_
F(x, t ) δy +
1
2
δP
_
∂y
∂x
_
2
_
dx dt
=
_
t
2
t
1
_
L
0
_
F(x, t ) δy + P
_
∂y
∂x
_
∂
∂x
δy
_
dx dt
=
_
t
2
t
1
_
L
0
_
_
F(x, t ) δy − P
∂
2
y
∂x
2
δy
_
dx +
_
P
∂y
∂x
δy
_
L
0
_
dt.
Thus, for uniform transverse loading F(x, t ) = F, the equilibrium equation becomes
EI
∂
4
y
∂x
4
+ P
∂
2
y
∂x
2
+ ρA
∂
2
y
∂t
2
= F
and the boundary conditions are
∂
2
y
∂x
2
∂
∂x
δy = 0 for x = 0 and x = L,
EI
∂
3
y
∂x
3
+ P
∂
2
y
∂x
2
= 0 for x = 0 and x = L,
which means
either
∂
2
y
∂x
2
= 0 or
∂
∂x
δy = 0 for x = 0 and x = L
and
either EI
∂
3
y
∂x
3
+ P
∂
2
y
∂x
2
= 0 or δy = 0 for x = 0 and x = L.
4.3 Introduction to the ﬁnite element method
In this section, we will make use of Hamilton’s principle to obtain ﬁnite element
equilibrium equations which we shall then apply to some simple, uniaxial problems,
before considering some further ﬁnite elements, and their application in statics and
dynamics. We shall start with an introduction to the ﬁnite element method.
Figure 4.7 shows a representation of an undeformed body when the value of time
is zero, which has been discretized into a ﬁnite number of tetrahedral ﬁnite elements
which approximate the initial geometry of the body.
A particular, single ﬁnite element is shown which has nodes at points P
1
, P
2
, P
3
,
and P
4
in the undeformed conﬁguration. On the application of a load (which
may be mechanical or thermal, for example), the body deforms (and undergoes
Introduction to the ﬁnite element method 97
X, x
Z, z
Y, y
Time =0
Time =t
P
3
P
2
P
4
P
1
p
3
p
2
p
4
p
1
F
F
Fig. 4.7 Schematic diagramshowing the ﬁnite element discretization of a body with threedimensional
tetrahedral elements.
transformation ) to that shown in the current conﬁguration at time t . The nodes
of the single element, after deformation, are now located at p
1
, p
2
, p
3
, and p
4
.
For simplicity, let us suppose that the quantity we are interested in determining is
temperature, chosen because it is a scalar variable. The basis of the ﬁnite element
method is to assign nodes to the elements and to assume that we can determine shape
functions to enable interpolation to give the value of the temperature at any point within
the element in terms of the nodal values of temperature. The tetrahedral element shown
in the ﬁgure has four nodes. If the temperatures at the four nodes are θ
1
, θ
2
, θ
3
, and θ
4
respectively, then the temperature anywhere within the element is given by
θ = N
1
θ
1
+ N
2
θ
2
+ N
3
θ
3
+ N
4
θ
4
=
n
i=1
N
i
θ
i
, (4.27)
where N
i
are called shape (or interpolation) functions. Let us consider in isolation
the element shown in Fig. 4.7 in the deformed conﬁguration. The element is shown in
Fig. 4.8(a) with respect to the current conﬁguration, and in (b) with respect to a local
element reference frame (ξ
1
, ξ
2
, ξ
3
).
A transformation is clearly needed to map the element from the local reference
frame to the current conﬁguration, and similarly, from the local reference frame to
the original conﬁguration, and we shall address this later. The shape functions for this
element, in terms of the local variables, are
N
1
(ξ
1
, ξ
2
, ξ
3
) = 1 − ξ
1
− ξ
2
− ξ
3
,
N
2
(ξ
1
, ξ
2
, ξ
3
) = ξ
1
,
N
3
(ξ
1
, ξ
2
, ξ
3
) = ξ
2
,
N
4
(ξ
1
, ξ
2
, ξ
3
) = ξ
3
.
98 Finite element method
x
3
x
2
x
1
4
1
3
2
4
1
3
2
j
1
j
2
j
3
(a) (b)
Fig. 4.8 Fournoded tetrahedral element shown with respect to (a) the current conﬁguration and (b) the
local element reference frame.
An important feature of the shape functions in the ﬁnite element method is that they
generally take a value of unity at their own node and are zero at all others; at node 1,
for example, ξ
1
= ξ
2
= ξ
3
= 0, so N
1
= 1. They generally sum to unity: N
1
+ N
2
+
N
3
+N
4
= 1; there are however, some special elements for which this is not the case.
With knowledge of position (ξ
1
, ξ
2
, ξ
3
) within an element, the shape functions can be
used together with Equation (4.27) to determine the value of the temperature at any
point within the element, given the nodal temperatures. The shape functions may be
used in a similar way for any variable of interest, but often, the ﬁnite element equilib
rium equations are set up with displacement as the basic quantity for which solutions
are obtained. Such an approach is often referred to, therefore, as the displacement
based ﬁnite element method. We will look at a further, very simple ﬁnite element in
order to examine the displacementbased approach.
Figure 4.9 shows a uniformbar under axial force P which has been discretized with
a number of uniaxial truss elements. Each element has two nodes and is of length L.
Each node has just one degree of freedom; namely axial displacement, u. The bar lies
along the xdirection in the current conﬁguration.
The shape functions are
N
1
(ξ) = 1 − ξ, N
2
(ξ) = ξ, (4.28)
where ξ = x/L and 0 ≤ ξ ≤ 1, and the element displacements are given by
u(ξ) = N
1
u
1
+ N
2
u
2
.
This is often written in vector form as
u(ξ) = Nu
I
= [N
1
N
2
]
_
u
1
u
2
_
. (4.29)
Introduction to the ﬁnite element method 99
u
2
u
1
x
L
x
P
P
N
1
1
1
N
2
1
1
N
1
(j) =1–j N
2
(j) =j
j j
Fig. 4.9 A uniform bar discretized using uniaxial truss elements with shape functions shown.
Because we know the displacement everywhere within the element, we can determine
the small strain, which, for this simple, uniaxial displacement is just
ε =
∂u
∂x
=
_
∂N
1
∂x
∂N
2
∂x
_
_
u
1
u
2
_
=
_
∂N
1
∂ξ
∂ξ
∂x
∂N
2
∂ξ
∂ξ
∂x
_
_
u
1
u
2
_
=
1
L
[−1 1]
_
u
1
u
2
_
. (4.30)
The derivatives of the type ∂ξ/∂x relate the current conﬁguration to the local element
reference frame and, in effect, provide the mapping of the element from the current
conﬁguration to the local element reference frame. In this case, because ξ = x/L, the
mapping is trivial and the derivatives ∂ξ/∂x are easily obtained. We will see how to
do this for more general cases a little later.
The matrix of spatial derivatives of the shape functions, given in Equation (4.30),
is often referred to as the B matrix where for this particular element,
B =
1
L
[−1 1]. (4.31)
It can be seen from (4.30) that the strain is constant everywhere within the element;
it is an example (as is the fournoded tetrahedran above) of a constant strain
element. In order to progress with the analysis of the loaded bar, we need to obtain the
equations of motion or equilibrium. We shall do this in a general way using Hamilton’s
principle, and then return to the loaded bar and to uniaxial truss elements.
100 Finite element method
4.4 Finite element equilibrium equations
4.4.1 Some preliminaries: tensor and Voigt notation, tensorial versus
engineering strain
Tensorial notation is very elegant when it comes to theoretical derivations. However,
for the purposes of developing numerical algorithms for implementation into computer
programs, it is often more practical to work with arrays thus representing secondorder
tensor (strain, stress) as onedimensional arrays and constitutive tensors (the elasticity
tensor) as twodimensional arrays.
Symmetry of stress and strain tensors is used to obtain the following (memory
saving) notation (Voigt notation):
σ =
⎡
⎢
⎣
σ
xx
σ
xy
σ
xz
σ
xy
σ
yy
σ
yz
σ
xz
σ
yz
σ
zz
⎤
⎥
⎦
→ σ =
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
σ
xx
σ
yy
σ
zz
σ
xy
σ
yz
σ
xz
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
, (4.32)
ε =
⎡
⎢
⎣
ε
xx
ε
xy
ε
xz
ε
xy
ε
yy
ε
yz
ε
xz
ε
yz
ε
zz
⎤
⎥
⎦
→ ε =
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
ε
xx
ε
yy
ε
zz
γ
xy
γ
yz
γ
zx
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
ε
xx
ε
yy
ε
zz
2ε
xy
2ε
yz
2ε
zx
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
. (4.33)
Note that in vector, or Voigt notation, the shear strain components are stored as
engineering shears, that is, twice the tensor shears. This sometimes causes confu
sion, so let us clarify by recalling Hooke’s law. Writing stress and elastic strain as
column vectors, Hooke’s law in three dimensions becomes
σ =
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
σ
xx
σ
yy
σ
zz
σ
xy
σ
yz
σ
xz
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
= Cε =
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
λ + 2µ λ λ 0 0 0
λ λ + 2µ λ 0 0 0
λ λ λ + 2µ 0 0 0
0 0 0 µ 0 0
0 0 0 0 µ 0
0 0 0 0 0 µ
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
ε
xx
ε
yy
ε
zz
γ
xy
γ
yz
γ
xz
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
, (4.34)
Finite element equilibrium equations 101
where λ = Eν/(1 + ν)(1 − 2ν) and µ = G = E/2(1 + ν) are the Lame constants.
Let us take one of the shear terms, for example,
σ
xy
= µγ
xy
≡
E
2(1 + ν)
γ
xy
. (4.35)
We will now obtain the same shear stress, this time using the stress–strain tensor
relation, given in Equation (2.99), that is,
σ =
⎛
⎜
⎜
⎝
σ
xx
σ
xy
σ
xz
σ
xy
σ
yy
σ
yz
σ
xz
σ
yz
σ
zz
⎞
⎟
⎟
⎠
= 2Gε + λTr(ε)I
= 2G
⎛
⎜
⎜
⎝
ε
xx
ε
xy
ε
xz
ε
xy
ε
yy
ε
yz
ε
xz
ε
yz
ε
zz
⎞
⎟
⎟
⎠
+ λ(ε
xx
+ ε
yy
+ ε
zz
)
⎛
⎜
⎜
⎝
1 0 0
0 1 0
0 0 1
⎞
⎟
⎟
⎠
.
The shear stress is, therefore,
σ
xy
= 2Gε
xy
≡
E
(1 + ν)
ε
xy
. (4.36)
Comparison of (4.35) and (4.36) shows that we must have
γ
xy
= 2ε
xy
and similarly for the other shear strains. A more elegant explanation of this is as
follows. Let us consider the case of pure elastic shear, as described by the stress and
strain tensors
σ =
⎛
⎜
⎜
⎝
0 σ
xy
σ
xz
σ
xy
0 σ
yz
σ
xz
σ
yz
0
⎞
⎟
⎟
⎠
, ε =
⎛
⎜
⎜
⎝
0 ε
xy
ε
xz
ε
xy
0 ε
yz
ε
xz
ε
yz
0
⎞
⎟
⎟
⎠
.
The elastic strain energy per unit volume is given by the tensor product
1
2
σ : ε =
1
2
⎛
⎜
⎜
⎝
0 σ
xy
σ
xz
σ
xy
0 σ
yz
σ
xz
σ
yz
0
⎞
⎟
⎟
⎠
:
⎛
⎜
⎜
⎝
0 ε
xy
ε
xz
ε
xy
0 ε
yz
ε
xz
ε
yz
0
⎞
⎟
⎟
⎠
= (σ
xy
ε
xy
+ σ
yz
ε
yz
+ σ
xz
ε
xz
)
and substituting for the shear stresses using (4.36) and similar expressions gives
1
2
σ : ε =
E
1 + ν
(ε
2
xy
+ ε
2
yz
+ ε
2
xz
). (4.37)
102 Finite element method
We can also calculate the strain energy using engineering shear strain as
1
2
(τ
xy
γ
xy
+ τ
yz
γ
yz
+ τ
xz
γ
xz
)
=
1
2
G(γ
2
xy
+ γ
2
yz
+ γ
2
xz
) =
E
4(1 + ν)
(γ
2
xy
+ γ
2
yz
+ γ
2
xz
). (4.38)
Comparison of (4.37) and (4.38), given that the three shear strains are independent,
gives
γ
xy
= 2ε
xy
as before, and similarly for the other shear strains. The strain tensor is, of course,
an objective quantity; that is, one that satisﬁes the criteria for objectivity discussed
in Chapter 3, and this means a quantity whose properties remain unchanged under
rotation. We must therefore use the tensorial shear strain components (as opposed to
the engineering shear) when rotating a strain. It might be of interest to note that it is
for this reason that Mohr’s circle for strain is drawn in terms of γ
xy
/2(= ε
xy
) rather
than γ
xy
.
We write the dot product of two ﬁrstorder tensors, or vectors, as
u
T
· v = u
i
v
i
=
n
i=1
u
i
v
i
,
v
T
· u = v
i
u
i
=
n
i=1
v
i
u
i
so that
v
T
· u = u
T
· v.
If, therefore, we calculate the product of the stress vector, for example, then
from (4.32),
σ
T
· σ = (σ
xx
σ
yy
σ
zz
σ
xy
σ
yz
σ
xz
) · (σ
xx
σ
yy
σ
zz
σ
xy
σ
yz
σ
xz
)
T
= σ
2
xx
+ σ
2
yy
+ σ
2
zz
+ σ
2
xy
+ σ
2
yz
+ σ
2
xz
.
It is important to note that this does not give the same result as the equivalent product
of the stress tensor, which is
σ : σ = σ
2
xx
+ σ
2
yy
+ σ
2
zz
+ 2(σ
2
xy
+ σ
2
yz
+ σ
2
xz
).
It is therefore necessary to take care when using the vector (Voigt) notation for stress
and strain in carrying out calculations. So, for example, the norm of σ, written σ is
σ = (σ : σ)
1/2
= [σ
2
xx
+ σ
2
yy
+ σ
2
zz
+ 2(σ
2
xy
+ σ
2
yz
+ σ
2
xz
)]
1/2
Finite element equilibrium equations 103
but if we work from the stress written as a column vector, then
σ =
_
σ
T
Sσ,
which includes the following scaling matrix
S =
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
1 0 0 0 0 0
0 1 0 0 0 0
0 0 1 0 0 0
0 0 0 2 0 0
0 0 0 0 2 0
0 0 0 0 0 2
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
.
The following are further examples of possible dangers in using Voigt notation:
1. Second invariant of deviatoric stress in tensorial notation
J
2
=
1
2
σ
: σ
and its correct expression in Voigt notation
J
2
=
1
2
σ
T
Sσ
.
2. Gradient of J
2
invariant in tensorial notation
∂J
2
∂σ
= σ
and its correct expression in Voigt notation
∂J
2
∂σ
= Sσ
.
3. Strain norm in tensorial notation
ε =
√
ε : ε
and its correct expression in Voigt notation
ε =
_
ε
T
S
−1
ε,
where
S
−1
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
1 0 0 0 0 0
0 1 0 0 0 0
0 0 1 0 0 0
0 0 0
1
2
0 0
0 0 0 0
1
2
0
0 0 0 0 0
1
2
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
.
104 Finite element method
4.4.2 Finite element equations using Hamilton’s principle
For the purposes of obtaining the ﬁnite element equilibriumequations, we shall conﬁne
ourselves initially to small elastic deformation problems so that we will not need to
worry about the kinematics of large deformations. We will address large deformations
later.
The components of the Lagrangian for a general body subject to tractions t are
given in Equations (4.18) using tensorial notation. We shall write them, here, using
Voigt notation for a particular (the ‘m’th) ﬁnite element as follows
T
m
=
1
2
_
ρ ˙ u
T
˙ udV,
U
m
=
1
2
_
ε
T
σ dV, (4.39)
W
m
=
_
∂
u
T
t dA,
in which all nonscalar quantities are represented as column vectors and in particular,
the stress and strain vectors are given by Equations (4.32) and (4.33), respectively
and the engineering shear strains are used. Note that ˙ u
T
˙ u ≡ ˙ u · ˙ u, and similarly for
other vector quantities. For completeness, the displacement and traction vectors are
given by
u =
⎡
⎢
⎢
⎣
u
x
u
y
u
z
⎤
⎥
⎥
⎦
and t =
⎡
⎢
⎢
⎣
t
x
t
y
t
z
⎤
⎥
⎥
⎦
.
Hooke’s law, from Equation (4.34), is written as
σ = Cε
e
. (4.40)
The displacement within the particular ﬁnite element may be written in terms of the
nodal displacements, u
I
, as we did in (4.29) so that
u = Nu
I
(4.41)
and the strains can be written, in general, in terms of the nodal displacements and the
B matrix, as we did for the truss element in Equation (4.30),
ε = Bu
I
. (4.42)
For the uniaxial truss element, ε becomes just the unaxial, scalar, strain ε, but for other
element types (in two and three dimensions), the other strain components are needed.
We shall see an example of this later.
Finite element equilibrium equations 105
In applying Hamilton’s principle, we shall start from the Lagrangian, L
m
, for the
single element
L
m
= T
m
− U
m
+ W
m
= T
m
=
1
2
_
ρ ˙ u
T
˙ udV −
1
2
_
ε
T
σ dV +
_
∂
u
T
t dA
and substituting Equations (4.40)–(4.42) gives
L
m
=
1
2
˙ u
T
I
_
ρN
T
N dV ˙ u
I
−
1
2
u
T
I
_
B
T
CB dVu
I
+u
T
I
_
∂
N
T
t dA.
If we deﬁne element mass and stiffness matrices as
m =
_
ρN
T
N dV (4.43)
and
k =
_
B
T
CB dV (4.44)
and the vector of nodal forces as
f =
_
∂
N
T
t dA, (4.45)
then the Lagrangian becomes
L
m
=
1
2
˙ u
T
I
m˙ u
I
−
1
2
u
T
I
ku
I
+u
T
I
f.
The J integral is, therefore
J =
_
t
2
t
1
L
m
dt =
_
t
2
t
1
_
1
2
˙ u
T
I
m˙ u
I
−
1
2
u
T
I
ku
I
+u
T
I
f
_
dt
and applying Hamilton’s principle gives
δJ =
_
t
2
t
1
_
1
2
δ( ˙ u
T
I
m˙ u
I
) −
1
2
δ(u
T
I
ku
I
) + δ(u
T
I
f)
_
dt
=
_
t
2
t
1
δu
T
I
(−m¨ u
I
−ku
I
+f) dt = 0.
Since δu
I
is an arbitrary displacement, this gives for the equilibrium equation, for
a single ﬁnite element
m¨ u
I
+ku
I
= f. (4.46)
Equation (4.46) is the ﬁnite element equilibrium equation, or equation of motion,
written in terms of nodal displacements, u
I
. In the absence of inertia forces, it reduces
to the standard, quasistatic equation
ku
I
= f. (4.47)
106 Finite element method
Some authors differentiate between momentum balance equations and equilibrium
equations; those that are derived which include inertia (dynamic) effects—for
example, that given in (4.46)—are often referred to as momentum balance equations,
and those which do not include inertia effects (for quasistatic problems), are simply
called equilibrium equations. In most cases in what follows, we shall simply refer to
both types as equilibrium equations. In Section 4.4.2.1, we shall determine the mass
and stiffness matrices for a truss element and examine a singleelement problem.
4.4.2.1 Single truss element problem. Consider the uniform bar with cross
sectional area A, made of material with density ρ, and Young’s modulus E, shown in
Fig. 4.10(a).
This is discretized with a single, uniaxial truss element shown in Fig. 4.10(b). The
shape functions are those given in (4.28),
N
1
(ξ) = 1 − ξ, N
2
(ξ) = ξ
so that
N = [1 − ξ ξ].
From above, (4.31), the B matrix is
B =
1
L
[−1 1].
The mass matrix can then be determined from (4.43) as
m =
_
ρN
T
N dV =
_
L
x=0
ρ
_
1 − ξ
ξ
_
[1 − ξ ξ]Adx.
This integral is currently given with respect to the deformed conﬁguration, x, so it
needs to be transformed to the local element reference frame, ξ. While trivial in this
case, since ξ = x/L, in general we would need to obtain the Jacobian, mapping dV
(or equivalently, dx) inthe current conﬁgurationtodV
(or, dξ) inthe element reference
L
r, E, A
P
(a)
x
2
u
2
x
1
u
1
2 1
L
r, E, A
(b)
Fig. 4.10 (a) Auniformbar under axial force, P, and (b) its discretization using a single truss element.
Finite element equilibrium equations 107
frame, as we shall see later. For now, however,
m =
_
L
x=0
ρ
_
1 − ξ
ξ
_
[1 − ξ ξ]Adx =
_
1
ξ=0
ρ
_
1 − ξ
ξ
_
[1 − ξ ξ]ALdξ
= ρAL
_
1
0
_
(1 − ξ)
2
ξ(1 − ξ)
ξ(1 − ξ) ξ
2
_
dξ
so
m = ρAL
_
1
3
1
6
1
6
1
3
_
. (4.48)
The stiffness matrix is, using (4.44)
k =
_
B
T
CB dV =
_
L
x=0
1
L
_
−1
1
_
E
1
L
[−1 1]Adx
=
_
1
ξ=0
E
L
2
_
1 −1
−1 1
_
ALdξ,
where in this simple uniaxial case, the elasticity matrix C is simply Young’s modulus,
E, so
k =
EA
L
_
1 −1
−1 1
_
(4.49)
and the nodal force vector, (4.45), for this problem is simply
f =
_
F
1
P
_
(4.50)
in which P is just the prescribed force at node 2 and F
1
the currently unknown reaction
at node 1, and no integration over ξ is necessary. Note that because t is a traction (stress
vector), t dA is the force acting on the area dA at nodes 1 and 2. The integration of
dAat the two ends therefore just gives Aand is independent of ξ, so that tAat node 2
is P and that at node 1 is F
1
, both of which act in a direction parallel to t; that is, the
xdirection. The equilibrium Equation (4.46) therefore gives
ρAL
_
1
3
1
6
1
6
1
3
__
¨ u
1
¨ u
2
_
+
EA
L
_
1 −1
−1 1
__
u
1
u
2
_
=
_
F
1
P
_
. (4.51)
Let us assume inertia forces are negligible and that this is, therefore, a quasistatic
problem so that we may eliminate the inertia term from (4.51) to give
EA
L
_
1 −1
−1 1
__
u
1
u
2
_
=
_
F
1
P
_
.
108 Finite element method
With the boundary condition that u
1
= 0, these equations can be solved to give the
expected result that
u
2
=
PL
EA
, F
1
= −P
and using u = Nu
I
, that
u
2
= [1 − ξ ξ]
_
0
u
2
_
= ξ
PL
EA
.
where 0 ≤ ξ ≤ 1. Despite using just a single element, these results are exact. This is
only because, in this particular problem, we have been using a constant strain element
to discretize what is, in any case, a constant strain problem.
In the following section, we shall give a more general description of the ﬁnite
element method which includes the kinematics of large deformations and recognizes
that the displacement of the nodes can become large; the problem then becomes
geometrically nonlinear. In addition, we need to give a fuller description of the
mappingprocess betweenthe current (andoriginal) conﬁgurationandthe local element
reference frame. We will then look at several further examples.
4.4.3 General ﬁnite element approach
We will now return to the body in the initial and current conﬁgurations shown in
Fig. 4.7. We will formulate the ﬁnite element equations with respect to the current,
or deformed, conﬁguration. When carried out incrementally, this is usually called
an updated Lagrangian formulation. An alternative is to set up the equations with
respect to the original conﬁguration, in which case, it is called a total Lagrangian
formulation. The initial position of the material particle P within the element shown
can be speciﬁed as follows
X(ξ, t ) ≈
NNODE
I=1
N
I
(ξ)X
I
(t ), (4.52)
where NNODE denotes the number of ﬁnite element nodes, N
I
(ξ) represents, as
before, the element shape functions and X
I
indicates the initial positions of the ﬁnite
element nodal points P
I
, which are given in terms of the local element reference,
(ξ
1
, ξ
2
, ξ
3
) shown in Fig. 4.8. Assuming that the material particle P remains attached
to the same ﬁnite elements during the motion, the current positions of the material
particles at the time t are speciﬁed by
x(ξ, t ) ≈
NNODE
I=1
N
I
(ξ)x
I
(t ) (4.53)
Finite element equilibrium equations 109
where x
I
(t ) denotes the current positions of the ﬁnite element nodal points p
I
. In
the example in Section 4.4.2.1, the displacements were assumed to be small and
there were no rigid body rotations. X(ξ, t ) and x(ξ, t ) were therefore identical and in
addition, because the element considered was onedimensional, ξ in Equations (4.52)
and (4.53) was simply the scalar ξ. For twodimensional elements, for example, two
local reference frame independent variables are needed, (ξ
1
, ξ
2
), to specify position.
For the previous truss element example, x(ξ, t ) becomes
x(ξ, t ) ≈
2
I=1
N
I
(ξ)x
I
(t ) = (1 − ξ)x
1
(t ) + ξx
2
(t ). (4.54)
Equation (4.54), and for the general case (4.53), therefore, maps any point within an
element, speciﬁed by ξ with respect to the local element reference system, onto the
corresponding position in the current conﬁguration. The displacement and the velocity
ﬁelds within each ﬁnite element are approximated as follows
u(ξ, t ) ≈
NNODE
I=1
N
I
(ξ)u
I
(t ), (4.55)
v(ξ, t ) ≈
NNODE
I=1
N
I
(ξ)v
I
(t ). (4.56)
In Equations (4.52)–(4.55), the same shape functions are used for interpolating posi
tion and displacement. Elements for which this holds true are called isoparametric.
The deformation gradient tensor is obtained by differentiating Equation (4.53) with
respect to the initial conﬁguration as follows
F(ξ, t ) ≈
NNODE
I=1
x
I
(t ) ⊗ ∇
X
N
I
(ξ, t ), (4.57)
where
∇
X
N
I
(ξ, t ) =
∂N
I
(ξ)
∂X(t )
=
_
∂N
I
∂X
∂N
I
∂Y
∂N
I
∂Z
_
(4.58)
are the derivatives of the shape functions N
I
. The gradient term, ∇, and the dyadic
product, ⊗, are explained in Appendix A. The derivative can be rewritten as follows
∇
X
N
I
(ξ, t ) =
∂N
I
(ξ)
∂X(t )
=
∂N
I
(ξ)
∂ξ
∂ξ
∂X(t )
=
∂N
I
(ξ)
∂ξ
_
∂X(t )
∂ξ
_
−1
. (4.59)
It is normally not possible to determine ∂ξ/∂X in Equation (4.59) directly since
Equation (4.52) speciﬁes Xin terms of ξ. It is therefore necessary to determine ∂X/∂ξ
instead and obtain the inverse. The derivative ∂X/∂ξ is called the Jacobian and relates
110 Finite element method
inﬁnitesimal quantities in the material conﬁguration to those in the local element
coordinate system. We shall see this in several examples later. Using Equation (4.52),
∂X
∂ξ
=
∂N
I
(ξ)
∂ξ
X
I
(t ), (4.60)
where the members of ∂X/∂ξ are given by
∂X
i
∂ξ
j
=
NNODE
I=1
X
iI
∂N
I
∂ξ
j
. (4.61)
The small strain tensor can be approximated
ε(ξ, t ) =
1
2
[∇u + (∇u)
T
] ≈
1
2
NNODE
I=1
[u
I
(t ) ⊗ ∇
X
N
I
(ξ) + ∇
X
N
I
(ξ) ⊗u
I
(t )].
(4.62)
The velocity gradient is obtained as the spatial derivative of the velocity in
Equation (4.56) as
L(ξ, t ) ≈
NNODE
I=1
v
I
(t ) ⊗ ∇
x
N
I
(ξ, t )
and the rate of deformation tensor is obtained by using D = (1/2)(L +L
T
) as follows
D(ξ, t ) ≈
1
2
NNODE
I=1
(v
I
(t ) ⊗ ∇
x
N
I
(ξ, t ) + ∇
x
N
I
(ξ, t ) ⊗v
I
(t )), (4.63)
where
∇
x
N
I
(ξ, t ) =
∂N
I
(ξ)
∂x(t )
(4.64)
are the spatial derivatives with respect to the current conﬁguration, and are obtained
from the shape functions as
∇
x
N
I
(ξ, t ) =
∂N
I
(ξ)
∂x(t )
=
∂N
I
(ξ)
∂ξ
∂ξ
∂x(t )
=
∂N
I
(ξ)
∂ξ
_
∂x(t )
∂ξ
_
−1
, (4.65)
where members of the Jacobian, ∂x/∂ξ, are given by
∂x
i
∂ξ
j
=
NNODE
I=1
x
iI
∂N
I
∂ξ
j
. (4.66)
We return nowto the equilibriumEquation (4.20), obtained fromHamilton’s principle,
but write the increment in virtual work per unit volume and per unit time as
δW =
_
(div[σ] − ρ ¨ u) · δv dV = 0, (4.67)
Finite element equilibrium equations 111
where δv is an arbitrary virtual velocity in the current conﬁguration. By noting that
div[σ · δv] = div[σ] · δv +σ : ∇δv,
n · σδv = δv · t
(4.68)
that ∇δv = δL and that because σ is symmetric, it can be shown (with a little
algebra) that σ : δL = σ : δD, substituting into (4.67) with the application of the
divergence theorem gives
δW =
_
ρ ¨ u · δv dV +
_
σ : δDdV −
_
∂
t · δv dA = 0. (4.69)
The second term in the righthand side of (4.69) is the internal work per unit volume
per second. A stress quantity is called work conjugate to the strain if their double
contracted tensor product yields work. The corotational Cauchy, or true, stress is
work conjugate to the rate of deformation.
The spatial virtual work equation, (4.69), which describes the dynamic equilibrium
of the element can be rewritten in terms of the ﬁnite element discretization as follows
δW ≈
_
_
ρ
_
NNODE
I=1
N
I
¨ u
I
___
NNODE
I=1
N
I
δv
I
_
dV
+
_
σ :
1
2
NNODE
I=1
(δv
I
⊗ ∇
x
N
I
+ ∇
x
N
I
⊗ δv
I
) dV
−
_
∂
t ·
_
NNODE
I=1
N
I
δv
I
_
dA. (4.70)
As σ is a symmetric tensor, this reduces to
δW ≈
_
_
ρ
_
NNODE
I=1
N
I
¨ u
I
___
NNODE
I=1
N
I
δv
I
_
dV
+
_
σ :
NNODE
I=1
δv
I
⊗ ∇
x
N
I
dV −
_
∂
t ·
_
NNODE
I=1
N
I
δv
I
_
dA
and as the nodal virtual velocities are independent of the integration, this may be
rewritten to give
δW ≈
NNODE
I=1
δv
I
·
_
N
T
I
ρ ¨ u
I
[N
1
, N
2
, . . . , N
NNODE
] dV
+
NNODE
I=1
δv
I
·
_
σ∇
x
N
I
dV −
NNODE
I=1
δv
I
·
_
N
I
t dA. (4.71)
112 Finite element method
The virtual work equation can ﬁnally be written in vector form as
dW ≈
NNODE
I=1
dv
I
· (f
inert
I
+f
int
I
−f
ext
I
) (4.72)
by introducing expressions for the equivalent element nodal force vectors as follows
f
inert
I
=
_
N
T
I
ρ ¨ u
I
[N
1
, N
2
, . . . , N
NNODE
] dV, (4.73)
f
int
I
=
_
σ∇
x
N
I
dV, (4.74)
f
ext
I
=
_
∂
N
I
t dA, (4.75)
where f
inert
I
, f
int
I
, and f
ext
I
represent the inertial, the internal, and the external element
nodal forces, respectively.
Since the discretized rate of virtual work, Equation (4.72) must be satisﬁed for all
cases of the arbitrary virtual velocities, the element equation can be rewritten
f
inert
I
+f
int
I
−f
ext
I
= 0. (4.76)
The nodal forces can be further assembled into global vectors (column matrices) by
summing over all elements to give
F
inert
=
⎡
⎢
⎢
⎢
⎢
⎢
⎣
f
inert
1
f
inert
2
.
.
.
f
inert
n
⎤
⎥
⎥
⎥
⎥
⎥
⎦
, F
int
=
⎡
⎢
⎢
⎢
⎢
⎢
⎣
f
int
1
f
int
2
.
.
.
f
int
n
⎤
⎥
⎥
⎥
⎥
⎥
⎦
, F
ext
=
⎡
⎢
⎢
⎢
⎢
⎣
f
ext
1
f
ext
2
.
.
.
f
ext
n
⎤
⎥
⎥
⎥
⎥
⎦
, (4.77)
where n is the total number of points used in the discretization. Finally, the ﬁnite
element discretization can be expressed by the set of nonlinear equilibrium equations
as follows:
F
inert
(u) + F
int
(u) − F
ext
(u) = 0 (4.78)
for the set (column matrix) of nodal displacements
u =
⎡
⎢
⎢
⎢
⎢
⎣
u
1
(t )
u
2
(t )
.
.
.
u
n
(t )
⎤
⎥
⎥
⎥
⎥
⎦
(4.79)
at time t .
Finite element equilibrium equations 113
We can rewrite Equation (4.76) in a more familiar way by considering the
internal work term and writing it in terms of Voigt notation. We will assume small
strain elasticity for simplicity. Returning to Equation (4.69), the internal work term
originates from
δW
I
=
_
σ : δDdV
and may be written using Voigt notation as
δW
I
=
_
σ : δDdV =
_
δD
T
σ dV, (4.80)
where
D
T
= (D
xx
D
yy
D
zz
2D
xy
2D
yz
2D
zx
)
in which the offdiagonal terms (shear terms) are doubled to satisfy work conjugacy
as speciﬁed in (4.80). The rate of deformation is given in terms of nodal velocities in
(4.63) by
D =
1
2
NNODE
I=1
(v
I
⊗ ∇
x
N
I
+ ∇
x
N
I
⊗v
I
).
After some algebra, this can be written in Voigt notation as
D =
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
D
xx
D
yy
D
zz
2D
xy
2D
yz
2D
zx
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
=
NNODE
I=1
B
I
· v
I
=
NNODE
I=1
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
∂N
I
∂x
0 0
0
∂N
I
∂y
0
0 0
∂N
I
∂z
∂N
I
∂y
∂N
I
∂x
0
0
∂N
I
∂z
∂N
I
∂y
∂N
I
∂z
0
∂N
I
∂x
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎡
⎢
⎢
⎣
v
xI
v
yI
v
zI
⎤
⎥
⎥
⎦
(4.81)
so that from (4.80) the internal energy term becomes
δW
I
=
_
δD
T
σ dV =
_
_
NNODE
I=1
B
I
dv
I
_
T
σ dV. (4.82)
114 Finite element method
If we assume purely elastic small deformation, then from (4.34), we may write
δW
I
=
_
_
NNODE
I=1
B
I
dv
I
_
T
Cε dV
=
_
_
NNODE
I=1
B
I
δv
I
_
T
C
_
NNODE
J=1
B
J
u
J
_
dV
so that
δW
I
= δv
T
I
⎛
⎝
NNODE
I,J=1
_
B
T
I
CB
J
dV
⎞
⎠
u
J
(4.83)
since we may also write the small strains in terms of the nodal displacements as
ε =
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
ε
xx
ε
yy
ε
zz
2ε
xy
2ε
yz
2ε
zx
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
=
NNODE
I=1
B
I
· u
I
=
NNODE
I=1
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
∂N
I
∂x
0 0
0
∂N
I
∂y
0
0 0
∂N
I
∂z
∂N
I
∂y
∂N
I
∂x
0
0
∂N
I
∂z
∂N
I
∂y
∂N
I
∂z
0
∂N
I
∂x
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎡
⎢
⎢
⎣
u
xI
u
yI
u
zI
⎤
⎥
⎥
⎦
.
Equation (4.83) can be written as
δW
I
= δv
T
I
ku
I
,
where
k =
NNODE
I,J=1
_
B
T
I
CB
J
dV. (4.84)
When combined with (4.71) and considering an arbitrary virtual velocity, δv, gives
m¨ u +ku = f (4.85)
and for quasistatic problems,
ku = f. (4.86)
This equation, in general, is nonlinear if the problem considered is geometrically
nonlinear. This is because the stiffness matrix, k, depends upon the B matrix which
Finite element equilibrium equations 115
contains the derivatives of the shape functions with respect to the spatial coordinates.
In a geometrically nonlinear problem, the spatial coordinates change so that the
stiffness matrix also changes. For geometrically nonlinear problems, Equation (4.86)
therefore has to be solved incrementally, and at each increment, the stiffness matrix
must be updated.
In the following sections, we look at a number of small deformation (geometrically
linear) examples including a twoelement axial vibration problem in which element
assembly is addressed, a singleelement and twoelement bending problem in which
beam elements are introduced, and a single square element static elastic problem in
which linear fournoded (quad) elements are introduced.
4.4.3.1 Axial vibration of a beam using two truss elements. The uniform bar
shown in Fig. 4.11(a) undergoes axial vibration. It is discretized using two truss
elements shown in Fig. 4.11(b).
The truss element formulation, relative to the local element reference frame, is
shown in Fig. 4.12.
∂t
2
A dx
∂
2
u
∂P
∂x
dx
P+ dx
P
L
, E, A
(a)
x
1
x
2
x
3
u
3
u
1
u
2
1
2
2 1
3
L/2 L/2
, E, A , E, A
(b)
Fig. 4.11 A uniform bar undergoing axial vibration (a) shown schematically and (b) its discretization
using two truss elements.
1 2
1
j
0
l =1
Fig. 4.12 Linear truss element formulation shown relative to the local element reference frame, ξ.
116 Finite element method
The shape functions are
N
1
= 1 − ξ, N
2
= ξ
and in this problem, the two elements are identical so that their mass and stiffness
matrices, respectively, are the same. Also, because this is a problem in which the
displacements are inﬁnitesimal, the current and original conﬁgurations are taken to
be the same. We shall determine the mass and stiffness matrices for a single element,
and then consider their assemblage.
The mass matrix given by (4.43) with respect to the spatial (which in this case
coincides with the material) coordinate system is
m =
_
ρN
T
N dV =
_
L/2
X=0
ρN
T
NAdX
but of course, the shape functions are given with respect to the local element reference
frame. Therefore, we write
m =
_
L/2
X=0
ρN
T
NAdX =
_
1
ξ=0
ρN
T
NAdet(J) dξ
= ρA
_
1
ξ=0
_
1 − ξ
ξ
_
[1 − ξ ξ] det(J) dξ (4.87)
in which J is the Jacobian, mapping the spatial coordinate system onto the local
element reference frame. In this onedimensional problem, the determinant of the
Jacobian is just
det(J) =
¸
¸
¸
¸
∂X
∂ξ
¸
¸
¸
¸
=
¸
¸
¸
¸
∂
∂ξ
[N
1
N
2
]
_
X
1
X
2
_¸
¸
¸
¸
=
¸
¸
¸
¸
_
∂N
1
∂ξ
∂N
2
∂ξ
_ _
X
1
X
2
_¸
¸
¸
¸
,
which is of the general form given in Equation (4.60). Using the shape functions, we
obtain
det(J) =
¸
¸
¸
¸
[−1 1]
_
X
1
X
2
_¸
¸
¸
¸
= [−X
1
+ X
2
] =
¸
¸
¸
¸
_
0 +
L
2
_¸
¸
¸
¸
=
¸
¸
¸
¸
_
L
2
_¸
¸
¸
¸
=
L
2
so that the element mass matrix becomes
m =
ρAL
2
_
1
ξ=0
_
1 − 2ξ + ξ
2
ξ − ξ
2
ξ − ξ
2
ξ
2
_
dξ =
ρAL
2
_
1
3
1
6
1
6
1
3
_
. (4.88)
Finite element equilibrium equations 117
Let us write the mass matrix for elements 1 and 2 as
m
1
=
_
m
1
11
m
1
12
m
1
21
m
1
22
_
=
ρAL
2
_
1
3
1
6
1
6
1
3
_
, m
2
=
_
m
2
11
m
2
12
m
2
21
m
2
22
_
=
ρAL
2
_
1
3
1
6
1
6
1
3
_
,
then the global mass matrix is given by
M =
⎡
⎢
⎣
m
1
11
m
1
12
0
m
1
21
m
1
22
+ m
2
11
m
2
12
0 m
2
21
m
2
22
⎤
⎥
⎦
= ρA
L
2
⎡
⎢
⎣
1
3
1
6
0
1
6
2
3
1
6
0
1
6
1
3
⎤
⎥
⎦
= ρAL
⎡
⎢
⎣
1
6
1
12
0
1
12
1
3
1
12
0
1
12
1
6
⎤
⎥
⎦
.
(4.89)
The element stiffness matrix is given by (4.44) in terms of the spatial coordinates by
k =
_
B
T
CB dV =
_
L
X=0
B
T
CBAdX.
Introducing again the Jacobian in (4.87), the integral can be written with respect to
the local element reference frame by
k =
_
1
ξ=0
B
T
CBAdet(J) dξ =
_
1
ξ=0
B
T
CBA
L
2
dξ.
Because the truss elements are onedimensional, the elasticity matrix, C, becomes
just the scalar Young’s modulus, E, so
k =
EAL
2
_
1
ξ=0
B
T
B dξ.
The B matrix contains the derivatives of the shape functions with respect to the spatial
coordinates. We are assuming small displacements, however, and that the problem
remains geometrically linear so that the spatial and material coordinates remain the
same. Hence,
B =
_
∂N
1
∂x
∂N
2
∂x
_
≡
_
∂N
1
∂X
∂N
2
∂X
_
=
_
∂N
1
∂ξ
∂ξ
∂X
∂N
2
∂ξ
∂ξ
∂X
_
.
In general, ∂ξ/∂Xcannot be determined directly, since we ordinarily knowXin terms
of ξ. In its most general form, the relationship is given by (4.52), and for the truss
element in this problem, this is
X = [N
1
N
2
]
_
X
1
X
2
_
.
We have already determined the Jacobian derivative, in (4.87), and note that this is
the inverse of ∂ξ/∂X. That is, therefore,
∂ξ
∂X
=
_
∂X
∂ξ
_
−1
= J
−1
=
_
L
2
_
−1
=
2
L
.
118 Finite element method
In general, therefore, the derivatives needed for the B matrix are obtained from the
inverse of the Jacobian. The B matrix is, therefore,
B =
_
∂N
1
∂ξ
∂ξ
∂X
∂N
2
∂ξ
∂ξ
∂X
_
=
_
−1 ×
2
L
1 ×
2
L
_
=
2
L
[−1 1]
so that the element stiffness matrix becomes
k =
EAL
2
_
1
ξ=0
_
2
L
_
2
_
−1
1
_
[−1 1] dξ =
2EA
L
_
1 −1
−1 1
_
. (4.90)
As for the mass matrices, we write the stiffness matrices for the two elements as
k
1
=
_
k
1
11
k
1
12
k
1
21
k
1
22
_
=
2EA
L
_
1 −1
−1 1
_
, k
2
=
_
k
2
11
k
2
12
k
2
21
k
2
22
_
=
2EA
L
_
1 −1
−1 1
_
so that the global stiffness matrix become
K =
⎡
⎢
⎢
⎣
k
1
11
k
1
12
0
k
1
21
k
1
22
+ k
2
11
k
2
12
0 k
2
21
k
2
22
⎤
⎥
⎥
⎦
=
2EA
L
⎡
⎣
1 −1 0
−1 2 −1
0 −1 1
⎤
⎦
=
EA
L
⎡
⎣
2 −2 0
−2 4 −2
0 −2 2
⎤
⎦
.
(4.91)
The ﬁnite element equilibrium equation, given in (4.46) and (4.85) then becomes
M¨ u + Ku = 0
so
ρAL
⎡
⎢
⎢
⎣
1
6
1
12
0
1
12
1
3
1
12
0
1
12
1
6
⎤
⎥
⎥
⎦
⎡
⎣
¨ u
1
¨ u
2
¨ u
3
⎤
⎦
+
EA
L
⎡
⎣
2 −2 0
−2 4 −2
0 −2 2
⎤
⎦
⎡
⎣
u
1
u
2
u
3
⎤
⎦
=
⎡
⎣
0
0
0
⎤
⎦
.
With the boundary condition u
1
= 0, this reduces to
ρAL
_
1
3
1
12
1
12
1
6
_
_
¨ u
2
¨ u
3
_
+
EA
L
_
4 −2
−2 2
_ _
u
2
u
3
_
=
_
0
0
_
.
4.4.3.2 Cantilever beam in bending using a single element. The problem geo
metry and ﬁnite element discretization are shown in Figs 4.13 and 4.14.
The Hermitian shape functions in the element local coordinate system (note that
dx = Ldξ) are obtained as follows:
N = a
0
+ a
1
ξ + a
2
ξ
2
+ a
3
ξ
3
.
Finite element equilibrium equations 119
L
r, E, A, I
Fig. 4.13 Cantilever beam of uniform section with second moment of area I.
x
v
2
1
1
2
L
r, E, A, I
v
1
u
1
u
1
Fig. 4.14 Single beamelement discretization of a cantilever. The element has four degrees of freedom;
two translational and two rotational.
First degree of freedom (y
1
)
ξ = 0, N
1
= 1,
dN
1
dξ
= 0
ξ = 1, N
1
= 0,
dN
1
dξ
= 0
⎫
⎪
⎪
⎬
⎪
⎪
⎭
⇒ N
1
= 1−3ξ
2
+2ξ
3
.
0 1 j
1
Second degree of freedom (θ
1
)
ξ = 0, N
2
= 0,
dN
2
dξ
= 1
ξ = 1, N
2
= 0,
dN
2
dξ
= 0
⎫
⎪
⎪
⎬
⎪
⎪
⎭
⇒ N
2
= ξ−2ξ
2
+ξ
3
.
0 1 j
1
Third degree of freedom (y
2
)
ξ = 0, N
3
= 0,
dN
3
dξ
= 0
ξ = 1, N
3
= 1,
dN
3
dξ
= 0
⎫
⎪
⎪
⎬
⎪
⎪
⎭
⇒ N
3
= 3ξ
2
−2ξ
3
.
0 1 j
1
120 Finite element method
Fourth degree of freedom (θ
2
)
ξ = 0, N
4
= 0,
dN
4
dξ
= 0
ξ = 1, N
4
= 0,
dN
4
dξ
= 1
⎫
⎪
⎪
⎬
⎪
⎪
⎭
⇒ N
4
= −ξ
2
+ξ
3
.
0 1 j
1
We shall now apply Hamilton’s principle in order to obtain the ﬁnite element equi
librium equations. For a beam in bending, we may write the kinetic, T , and potential
(strain), U, energies as
T =
1
2
_
L
0
ρA
_
∂v
∂t
_
2
dx, U =
1
2
_
L
0
EI
_
∂
2
v
∂x
2
_
2
dx.
The ﬁnite element discretization is given by
v(x, t ) = N
1
(ξ)v
1
(t ) + N
2
(ξ)θ
1
(t ) + N
3
(ξ)v
2
(t ) + N
4
(ξ)θ
2
(t )
= N
1
(ξ)v
1
(t ) + N
2
(ξ)
∂v
1
∂ξ
+ N
3
(ξ)v
2
(t ) + N
4
(ξ)
∂v
2
∂ξ
= N
1
(ξ)v
1
(t ) + N
2
(ξ)L
∂v
1
∂x
+ N
3
(ξ)v
2
(t ) + N
4
(ξ)L
∂v
2
∂x
,
which we shall write as
v = Nu = [N
1
(ξ) N
2
(ξ) N
3
(ξ) N
4
(ξ)]
_
v
1
∂v
1
∂x
v
2
∂v
2
∂x
_
T
.
The derivatives in the strain energy term are given by
∂
2
v
∂x
2
=
1
L
2
∂
2
v
∂ξ
2
=
1
L
2
_
∂
2
N
1
∂ξ
2
v
1
+
∂
2
N
2
∂ξ
2
L
∂v
1
∂x
+
∂
2
N
3
∂ξ
2
v
2
+
∂
2
N
4
∂ξ
2
L
∂v
2
∂x
_
=
1
L
2
_
∂
2
N
1
∂ξ
2
∂
2
N
2
∂ξ
2
L
∂
2
N
3
∂ξ
2
∂
2
N
4
∂ξ
2
L
_
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
v
1
∂v
1
∂x
v
2
∂v
2
∂x
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
= Bu.
The element kinetic and strain energies are given by
T =
1
2
_
L
0
ρA(N˙ u)
T
(N˙ u) dx =
1
2
˙ u
T
__
L
0
ρAN
T
N dx
_
˙ u
Finite element equilibrium equations 121
and
U =
1
2
_
L
0
EI
_
∂
2
N
∂ξ
2
u
_
T
_
∂
2
N
∂ξ
2
u
_
dx
=
1
2
u
T
_
_
L
0
EI
_
∂
2
N
∂ξ
2
_
T
_
∂
2
N
∂ξ
2
_
dx
_
u
=
1
2
u
T
__
L
0
EIB
T
B dx
_
u.
The element Lagrangian, L, is
L = T − U =
1
2
˙ u
T
m˙ u −
1
2
u
T
ku, (4.92)
where
m =
_
L
0
ρAN
T
N dx (4.93)
and
k =
_
L
0
EIB
T
B dx. (4.94)
Integrating the Lagrangian with respect to time, and taking the ﬁrst variation gives
m¨ u +ku = 0.
The element mass matrix is
m =
⎡
⎢
⎢
⎢
⎢
⎢
⎣
m
e
11
m
e
12
m
e
13
m
e
14
m
e
21
m
e
22
m
e
23
m
e
24
m
e
31
m
e
32
m
e
33
m
e
34
m
e
41
m
e
42
m
e
43
m
e
44
⎤
⎥
⎥
⎥
⎥
⎥
⎦
=
_
L
0
ρAN
T
N dx
= ρA
_
1
0
⎡
⎢
⎢
⎢
⎢
⎢
⎣
1 − 3ξ
2
+ 2ξ
3
(ξ − 2ξ
2
+ ξ
3
)L
3ξ
2
− 2ξ
3
(−ξ
2
+ ξ
3
)L
⎤
⎥
⎥
⎥
⎥
⎥
⎦
×
_
1 − 3ξ
2
+ 2ξ
3
(ξ − 2ξ
2
+ ξ
3
)L 3ξ
2
− 2ξ
3
(−ξ
2
+ ξ
3
)L
_
Ldξ
=
ρAL
420
⎡
⎢
⎢
⎣
156 22L 54 −13L
22L 4L
2
13L −3L
2
54 13L 156 −22L
−13L −3L
2
−22L 4L
2
⎤
⎥
⎥
⎦
.
122 Finite element method
The element stiffness matrix is given by
k =
⎡
⎢
⎢
⎢
⎢
⎢
⎣
k
e
11
k
e
12
k
e
13
k
e
14
k
e
21
k
e
22
k
e
23
k
e
24
k
e
31
k
e
32
k
e
33
k
e
34
k
e
41
k
e
42
k
e
43
k
e
44
⎤
⎥
⎥
⎥
⎥
⎥
⎦
=
_
L
0
EIB
T
B dx
=
_
1
0
1
L
2
⎡
⎢
⎢
⎢
⎢
⎢
⎣
−6 + 12ξ
(−4 + 6ξ)L
6 − 12ξ
(−2 + 6ξ)L
⎤
⎥
⎥
⎥
⎥
⎥
⎦
EI
1
L
2
× [−6 + 12ξ (−4 + 6ξ)L 6 − 12ξ (−2 + 6ξ)L]Ldξ
=
EI
L
3
⎡
⎢
⎢
⎣
12 6L −12 6L
6L 4L
2
−6L 2L
2
−12 −6L 12 −6L
6L 2L
2
−6L 4L
2
⎤
⎥
⎥
⎦
.
The ﬁnite element equilibrium equation becomes
ρAL
420
⎡
⎢
⎢
⎣
156 22L 54 −13L
22L 4L
2
13L −3L
2
54 13L 156 −22L
−13L −3L
2
−22L 4L
2
⎤
⎥
⎥
⎦
⎡
⎢
⎢
⎢
⎢
⎣
¨ y
1
¨
θ
1
¨ y
2
¨
θ
2
⎤
⎥
⎥
⎥
⎥
⎦
+
EI
L
3
⎡
⎢
⎢
⎣
12 6L −12 6L
6L 4L
2
−6L 2L
2
−12 −6L 12 −6L
6L 2L
2
−6L 4L
2
⎤
⎥
⎥
⎦
⎡
⎢
⎢
⎣
y
1
θ
1
y
2
θ
2
⎤
⎥
⎥
⎦
=
⎡
⎢
⎢
⎣
0
0
0
0
⎤
⎥
⎥
⎦
.
With the boundary conditions v
1
= 0 and θ
1
= 0 at x = 0, the problem reduces to
one with two degrees of freedom
ρAL
420
_
156 −22L
−22L 4L
2
_ _
¨ x
2
¨
θ
3
_
+
EI
L
3
_
12 −6L
−6L 4L
2
_ _
x
2
θ
3
_
=
_
0
0
_
.
4.4.3.3 Free transverse vibration of a propped cantilever using two beam
elements. A uniform, propped cantilever beam is shown in Fig. 4.15 and its ﬁnite
element discretization using two beam elements in Fig. 4.16.
Finite element equilibrium equations 123
L
r, E, A, I
Fig. 4.15 A uniform propped cantilever beam.
1
1
3
L/2
r, E, A, I
y
1
y
2
y
3
u
3
u
2
u
1
r, E, A, I
2
2
L/2
Fig. 4.16 Finite element discretization of a propped cantilever beam. The two beam elements are
identical.
The element mass matrix (valid for both elements) is
m
e
=
ρA(L/2)
420
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
156 22
L
2
54 −13
L
2
22
L
2
4
_
L
2
_
2
13
L
2
−3
_
L
2
_
2
54 13
L
2
156 −22
L
2
−13
L
2
−3
_
L
2
_
2
−22
L
2
4
_
L
2
_
2
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
and the stiffness matrix (valid for both elements) is
k
e
=
EI
(L/2)
3
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
12 6
L
2
−12 6
L
2
6L 4
_
L
2
_
2
−6
L
2
2
_
L
2
_
2
−12 −6
L
2
12 −6
L
2
6
L
2
2
_
L
2
_
2
−6
L
2
4
_
L
2
_
2
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
.
124 Finite element method
With the boundary conditions, we may write for the ﬁrst element
u =
⎡
⎢
⎢
⎣
y
1
θ
1
y
2
θ
2
⎤
⎥
⎥
⎦
=
⎡
⎢
⎢
⎣
0
0
y
2
θ
2
⎤
⎥
⎥
⎦
→ k
1
=
EI
(L/2)
3
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
X X X X
X X X X
X X 12 −6
L
2
X X −6
L
2
4
_
L
2
_
2
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
and for element 2 that
u =
⎡
⎢
⎢
⎣
y
2
θ
2
y
3
θ
3
⎤
⎥
⎥
⎦
=
⎡
⎢
⎢
⎣
y
2
θ
2
0
θ
3
⎤
⎥
⎥
⎦
→ k
2
=
EI
(L/2)
3
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
12 6
L
2
X 6
L
2
6
L
2
4
_
L
2
_
2
X 2
_
L
2
_
2
X X X X
6
L
2
2
_
L
2
_
2
X 4
_
L
2
_
2
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
.
We proceed in a similar way for the mass matrix and obtain the ﬁnite element
equilibrium equation
M¨ u + Ku = 0
to be
ρAL
420
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎣
156 0 −
13
4
L
0 L
2
−
3
8
L
2
−
13
4
L −
3
8
L
2
1
2
L
2
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎡
⎢
⎢
⎣
¨ y
2
¨
θ
2
¨
θ
3
⎤
⎥
⎥
⎦
+
EI
L
3
⎡
⎢
⎢
⎣
172 0 24L
0 16L
2
4L
2
24L 4L
2
8L
2
⎤
⎥
⎥
⎦
⎡
⎢
⎢
⎣
y
2
θ
2
θ
3
⎤
⎥
⎥
⎦
=
⎡
⎢
⎢
⎣
0
0
0
⎤
⎥
⎥
⎦
.
4.4.3.4 Static analysis of a single square element. We will examine one further
example before moving on; a single, twodimensional, fournoded, isoparametric
element subjected to a static force. We take the material behaviour to be elastic, and
assume conditions of plane strain. The problem is shown schematically in Fig. 4.17,
together with the elastic constants.
Finite element equilibrium equations 125
Geometry
a=1 mm
Material data
E=210 N/ mm
2
n =0.25
Loading
F=10 N
Discretization data (nodal coordinates)
n
1
(0, 1), n
2
(0, 0), n
3
(1, 0), n
4
(1, 1)
45°
1
2 3
4
F
a
a
Fig. 4.17 Schematic diagram of a single, fournoded element subjected to force F directed at 45
◦
to
the vertical.
The shape functions for this element are as follows:
N
1
(ξ, η) =
1
4
(1 − ξ)(1 − η),
N
2
(ξ, η) =
1
4
(1 + ξ)(1 − η),
N
3
(ξ, η) =
1
4
(1 + ξ)(1 + η),
N
4
(ξ, η) =
1
4
(1 − ξ)(1 + η).
Element local coordinate system
1 –1
1
–1
1
2
3 4
j
h
We will ﬁnd that we need to carry out integrations (in order to obtain the stiffness
matrix) of the shape functions. Previously, we have been able to do the integrations
analytically. Here, we will need to use a numerical technique. Often, in order to
simplify the process, the integration is carried out with respect to a particular point
in the element; this point is known as an integration point. The integration will be
performed numerically using a single integration point at P(ξ, η) = P(0, 0). Integrals
over the element domain of the type
I =
_
f (ξ, η) dV
are expressed in the element local coordinate system by use of the Jacobian, J, as
I =
_
1
−1
_
1
−1
f (ξ, η) det[J] dξ dη, (4.95)
where the Jacobian for a twodimensional problem is given by
J =
⎡
⎢
⎢
⎣
∂X
∂ξ
∂X
∂η
∂Y
∂ξ
∂Y
∂η
⎤
⎥
⎥
⎦
. (4.96)
126 Finite element method
The integral in (4.95) will be approximated using Gauss quadrature (details may be
found in any of the more specialized books on ﬁnite elements) by
I ≈
_
_
1
−1
_
1
−1
det[J] dξ dη
_
f (ξ, η).
The shape function derivatives at the integration point P(ξ, η) = P(0, 0) are
obtained from
∂N
∂X
=
∂N
∂ξ
∂ξ
∂X
=
∂N
∂ξ
_
∂X
∂ξ
_
−1
(4.97)
and the shape function derivatives with respect to element local coordinates are
given by
∂N(0, 0)
∂ξ
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
∂N
1
(0, 0)
∂ξ
∂N
1
(0, 0)
∂η
∂N
2
(0, 0)
∂ξ
∂N
2
(0, 0)
∂η
∂N
3
(0, 0)
∂ξ
∂N
3
(0, 0)
∂η
∂N
4
(0, 0)
∂ξ
∂N
4
(0, 0)
∂η
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
=
1
4
⎡
⎢
⎢
⎢
⎢
⎣
(−1)(1 − η)
(0,0)
(1 − ξ)(−1)
(0,0)
(1)(1 − η)
(0,0)
(1 + ξ)(−1)
(0,0)
(1)(1 + η)
(0,0)
(1 + ξ)(1)
(0,0)
(−1)(1 + η)
(0,0)
(1 − ξ)(1)
(0,0)
⎤
⎥
⎥
⎥
⎥
⎦
=
1
4
⎡
⎢
⎢
⎢
⎢
⎣
−1 −1
1 −1
1 1
−1 1
⎤
⎥
⎥
⎥
⎥
⎦
.
In order to determine the Jacobian matrix ∂X/∂ξ, let us ﬁrst write down the
relationship between element and nodal positions; that is,
X = N
1
X
1
+ N
2
X
2
+ N
3
X
3
+ N
4
X
4
,
Y = N
1
Y
1
+ N
2
Y
2
+ N
3
Y
3
+ N
4
Y
4
.
We see, therefore, that
∂X
∂ξ
=
∂N
1
∂ξ
X
1
+
∂N
2
∂ξ
X
2
+
∂N
3
∂ξ
X
3
+
∂N
4
∂ξ
X
4
= X
I
∂N
I
∂ξ
Finite element equilibrium equations 127
and similarly for other terms. The Jacobian matrix, therefore, from (4.96) is
J =
⎡
⎢
⎢
⎣
∂X
∂ξ
∂X
∂η
∂Y
∂ξ
∂Y
∂η
⎤
⎥
⎥
⎦
=
⎡
⎢
⎢
⎣
X
I
∂N
I
∂ξ
X
I
∂N
I
∂η
Y
I
∂N
I
∂ξ
Y
I
∂N
I
∂η
⎤
⎥
⎥
⎦
=
_
X
1
X
2
X
3
X
4
Y
1
Y
2
Y
3
Y
4
_
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
∂N
1
∂ξ
∂N
1
∂η
∂N
2
∂ξ
∂N
2
∂η
∂N
3
∂ξ
∂N
3
∂η
∂N
4
∂ξ
∂N
4
∂η
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
.
This can be written in the general form given in Equation (4.61) as
J = X
∂N
∂ξ
.
With the shape functions given above, evaluated at the integration point, and the nodal
coordinates, the Jacobian becomes
∂X
∂ξ
= X
∂N
∂ξ
=
_
X
1
X
2
X
3
X
4
Y
1
Y
2
Y
3
Y
4
_
1
4
⎡
⎢
⎢
⎣
−1 −1
1 −1
1 1
−1 1
⎤
⎥
⎥
⎦
=
1
4
_
−X
1
+ X
2
+ X
3
− X
4
−X
1
− X
2
+ X
3
+ X
4
−Y
1
+ Y
2
+ Y
3
− Y
4
−Y
1
− Y
2
+ Y
3
+ Y
4
_
=
1
4
_
−0 + 0 + 1 − 1 −0 − 0 + 1 + 1
−1 + 0 + 0 − 1 −1 − 0 + 0 + 1
_
=
1
4
_
0 2
−2 0
_
=
_
0 0.5
−0.5 0
_
.
The determinant of the Jacobian is
det[J] = det
_
∂X
∂ξ
_
= det
_
0 0.5
−0.5 0
_
= 0.25
and its inverse is
_
∂X
∂ξ
_
−1
=
_
0 −2
2 0
_
.
128 Finite element method
Finally, the shape function derivatives with respect to the spatial (here, equivalent to
the material) coordinates are
∂N
∂X
=
∂N
∂ξ
_
∂X
∂ξ
_
−1
=
1
4
⎡
⎢
⎢
⎣
−1 −1
1 −1
1 1
−1 1
⎤
⎥
⎥
⎦
_
0 −2
2 0
_
=
1
4
⎡
⎢
⎢
⎣
−1 −1
1 −1
1 1
−1 1
⎤
⎥
⎥
⎦
_
0 −2
2 0
_
=
⎡
⎢
⎢
⎣
−1 −1
1 −1
1 1
−1 1
⎤
⎥
⎥
⎦
_
0 −0.5
0.5 0
_
=
⎡
⎢
⎢
⎣
−0.5 0.5
−0.5 −0.5
0.5 −0.5
0.5 0.5
⎤
⎥
⎥
⎦
.
The B matrix can now be obtained (for plane strain) from (4.81)
B
I
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
∂N
I
(0, 0)
∂x
0
0
∂N
I
(0, 0)
∂y
∂N
I
(0, 0)
∂y
∂N
I
(0, 0)
∂x
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
,
B =
⎡
⎣
−0.5 0







−0.5 0







0.5 0







0.5 0
0 0.5 0 −0.5 0 −0.5 0 0.5
0.5 −0.5 −0.5 −0.5 −0.5 0.5 0.5 0.5
⎤
⎦
.
The elasticity matrix for plane strain is
C =
⎡
⎣
λ + 2µ λ 0
λ λ + 2µ 0
0 0 µ
⎤
⎦
,
λ =
νE
(1 + ν)(1 − 2ν)
= 84 N/mm
2
,
µ = G =
E
2(1 + ν)
= 84 N/mm
2
,
C =
⎡
⎣
252 84 0
84 252 0
0 0 84
⎤
⎦
N/mm
2
.
Finite element equilibrium equations 129
The stiffness matrix can then be determined by using the approximate integration
formula given above as
k =
_
V
B
T
CB dV =
_
1
0
_
1
0
B
T
CB dx dy =
_
1
−1
_
1
−1
B
T
CB det[J] dξ dη
≈ [B(0, 0)]
T
CB(0, 0) det[J(0, 0)].2.2
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
−0.5 0 0.5
0 0.5 −0.5
−0.5 0 −0.5
0 −0.5 −0.5
0.5 0 −0.5
0 −0.5 0.5
0.5 0 0.5
0 0.5 0.5
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎡
⎣
252 84 0
84 252 0
0 0 84
⎤
⎦
×
⎡
⎣
−0.5 0







−0.5 0







0.5 0







0.5 0
0 0.5 0 −0.5 0 −0.5 0 0.5
0.5 −0.5 −0.5 −0.5 −0.5 0.5 0.5 0.5
⎤
⎦
1
4
4
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
84 −42 42 0 −84 42 −42 0
−42 84 0 −42 42 −84 0 42
42 0 84 42 −42 0 −84 −42
0 −42 42 84 0 42 −42 −84
−84 42 −42 0 84 −42 42 0
42 −84 0 42 −42 84 0 −42
−42 0 −84 −42 42 0 84 42
0 42 −42 −84 0 −42 42 84
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
1
4
4
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
84 −42 42 0 −84 42 −42 0
−42 84 0 −42 42 −84 0 42
42 0 84 42 −42 0 −84 −42
0 −42 42 84 0 42 −42 −84
−84 42 −42 0 84 −42 42 0
42 −84 0 42 −42 84 0 −42
−42 0 −84 −42 42 0 84 42
0 42 −42 −84 0 −42 42 84
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
.
The equilibrium equation for this quasistatic problem is
ku = f, (4.98)
130 Finite element method
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
84 −42 42 0 −84 42 −42 0
−42 84 0 −42 42 −84 0 42
42 0 84 42 −42 0 −84 −42
0 −42 42 84 0 42 −42 −84
−84 42 −42 0 84 −42 42 0
42 −84 0 42 −42 84 0 −42
−42 0 −84 −42 42 0 84 42
0 42 −42 −84 0 −42 42 84
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
0
0
0
0
u
x3
u
y3
0
u
y4
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
R
x1
R
y1
R
x2
R
y2
F cos
π
4
−F cos
π
4
R
x4
0
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
so that with the boundary conditions it simpliﬁes to
⎡
⎢
⎢
⎣
84 −42 0
−42 84 −42
0 −42 84
⎤
⎥
⎥
⎦
⎡
⎢
⎢
⎣
u
x3
u
y3
u
y4
⎤
⎥
⎥
⎦
=
⎡
⎢
⎢
⎣
10
√
2/2
−10
√
2/2
0
⎤
⎥
⎥
⎦
.
The solution is obtained from
u = k
−1
f,
⎡
⎢
⎢
⎣
u
x3
u
y3
u
y4
⎤
⎥
⎥
⎦
=
⎡
⎢
⎢
⎣
0.017857 0.011905 0.005952
0.011905 0.023810 0.011905
0.005952 0.011905 0.017857
⎤
⎥
⎥
⎦
⎡
⎢
⎢
⎣
10
√
2/2
−10
√
2/2
0
⎤
⎥
⎥
⎦
,
⎡
⎢
⎢
⎣
u
x3
u
y3
u
y4
⎤
⎥
⎥
⎦
=
⎡
⎢
⎢
⎣
0.04209
−0.08418
−0.04209
⎤
⎥
⎥
⎦
mm.
The reactions can be obtained from
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
84 −42 42 0 −84 42 −42 0
−42 84 0 −42 42 −84 0 42
42 0 84 42 −42 0 −84 −42
0 −42 42 84 0 42 −42 −84
−84 42 −42 0 84 −42 42 0
42 −84 0 42 −42 84 0 −42
−42 0 −84 −42 42 0 84 42
0 42 −42 −84 0 −42 42 84
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
0
0
0
0
u
x3
u
y3
0
u
y4
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
R
x1
R
y1
R
x2
R
y2
F cos
π
4
−F cos
π
4
R
x4
0
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
,
Finite element equilibrium equations 131
R
x1
= −84u
x3
+ 42u
y3
+ 0u
y4
= −10
√
2/2N,
R
y1
= 42u
x3
− 84u
y3
+ 42u
y4
= 10
√
2/2N,
R
x2
= −42u
x3
+ 0u
y3
− 42u
y4
= 0,
R
y2
= 0u
x3
+ 42u
y3
− 84u
y4
= 0,
R
x4
= 42u
x3
+ 0u
y3
+ 42u
y4
= 0.
We can carry out an equilibrium check as follows:
F
x
= R
x1
+ R
x2
+ R
x4
+
10
√
2
2
= −
10
√
2
2
+ 0 + 0 +
10
√
2
2
= 0,
F
y
= R
y1
+ R
y2
−
F
√
2
2
=
10
√
2
2
+ 0 −
10
√
2
2
= 0.
The element strains at the integration point may be determined as
ε(ξ, η) = ε(0, 0) ≈
NNODE
I=1
B
I
(0, 0) · u
I
=
⎡
⎣
−0.5 0







−0.5 0







0.5 0







0.5 0
0 0.5 0 −0.5 0 −0.5 0 0.5
0.5 −0.5 −0.5 −0.5 −0.5 0.5 0.5 0.5
⎤
⎦
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
0
0
0
0
u
x3
u
y3
0
u
y4
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
=
⎡
⎣
−0.5 0







−0.5 0







0.5 0







0.5 0
0 0.5 0 −0.5 0 −0.5 0 0.5
0.5 −0.5 −0.5 −0.5 −0.5 0.5 0.5 0.5
⎤
⎦
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
0
0
0
0
0.04209
−0.08418
0
−0.04209
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
=
⎡
⎣
0.021045
0.021045
−0.08418
⎤
⎦
132 Finite element method
and the stresses are given by
σ(ξ, η) = Cε(ξ, η) =
⎡
⎣
252 84 0
84 252 0
0 0 84
⎤
⎦
⎡
⎣
0.021045
0.021045
−0.08418
⎤
⎦
=
⎡
⎣
7.071068
7.071068
−7.071068
⎤
⎦
=
⎡
⎣
10
√
2/2
10
√
2/2
−10
√
2/2
⎤
⎦
N/mm
2
.
A further check is that the internal forces at nodal points must be equal to nodal
external forces (note that the reactions are external forces too):
f
int
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
f
int
x1
f
int
y1
f
int
x2
f
int
y2
f
int
x3
f
int
y3
f
int
x4
f
int
y4
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
=
_
V
B
T
σ dV =
_
1
0
_
1
0
B
T
σ dx dy
=
_
1
−1
_
1
−1
B
T
σ det[J] dξ dη = B
T
σ
1
4
_
1
−1
_
1
−1
dξ dη
= B
T
σ
1
4
4 = B
T
σ =
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
−0.5 0 0.5
0 0.5 −0.5
−0.5 0 −0.5
0 −0.5 −0.5
0.5 0 −0.5
0 −0.5 0.5
0.5 0 0.5
0 0.5 0.5
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎡
⎢
⎢
⎣
10
√
2/2
10
√
2/2
−10
√
2/2
⎤
⎥
⎥
⎦
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
−10
√
2/2
10
√
2/2
0
0
10
√
2/2
−10
√
2/2
0
0
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
N.
We need to summarize a few important points before leaving this section. In all the
examples considered, we have used only one or two ﬁnite elements. This is intended
Finite element equilibrium equations 133
to aid in the development of understanding of the ﬁnite element method rather than to
suggest that practical problems can be solved in this way. Most engineering problems
require very large numbers of ﬁnite elements in order to obtain an accurate repres
entation and solution. Naturally, the calculations are then carried out on a computer.
We should also note that we have gone on to determine solutions for only quasistatic
problems. This required the solution of the Equation (4.98). However, we have set
up the ﬁnite element equilibrium (also often known as momentum balance) equations
for a number of dynamic problems which we have not attempted to solve. Solution of
these problems requires in general the numerical integration of the momentumbalance
Equation (4.85), using, for example, the explicit central difference time integration
scheme, which is introduced brieﬂy later on, or an implicit scheme. So far, in this
chapter, we have been concerned with both geometrically linear and nonlinear prob
lems in which the material behaviour has always been assumed to be linear (elasticity).
We shall continue by introducing incremental ﬁnite element techniques for nonlinear
material behaviour; that is, in particular, for plasticity.
4.4.4 Finite element formulation for plasticity
The equilibrium equations derived in (4.71)–(4.75) from virtual work are applicable
in general to linear and nonlinear material behaviour. In order to address a particular
example, we return to the equilibriumEquation (4.69) but for simplicity, assume quasi
static conditions and so ignore inertia terms, and we consider small deformations.
Using Voigt notation the equation of virtual work becomes
δW =
_
δD
T
σ dV −
_
∂
t · δv dA = 0.
With the ﬁnite element discretization as before, the equation becomes
δW =
_
_
NNODE
I=1
B
I
dv
I
_
T
σ dV −
_
∂
t ·
_
NNODE
I=1
N
I
δv
I
_
dA.
For small strain elastic–plastic material behaviour, we
σ = Cε
e
= C(ε −ε
p
)
so
δW =
_
_
NNODE
I=1
B
I
δv
I
_
T
C(ε −ε
p
) dV −
_
∂
t ·
_
NNODE
I=1
N
I
δv
I
_
dA
134 Finite element method
and with the ﬁnite element discretization for the strain presented in Section 4.4.3, this
becomes
δW
I
= δv
T
I
⎛
⎝
NNODE
I,J=1
_
B
T
I
CB
J
dV
⎞
⎠
u
J
− δv
T
I
NNODE
I=1
_
B
T
I
Cε
p
dV − δv
T
I
NNODE
I=1
_
∂
N
I
t dA.
As before, for an arbitrary virtual velocity, we may then write the element equilibrium
equation as
ku −f
p
= f, (4.99)
where
f
p
=
NNODE
I=1
_
B
T
I
Cε
p
dV.
In contrast to linear elastic problems in which there exists a unique relationship
between stress and elastic strain, no such uniqueness holds for plasticity problems. An
incremental approach to solving (4.99) is therefore almost always necessary, and to
emphasize this, the displacement, plastic strain, and external force terms are written
as increments so that
ku − f
p
= f (4.100)
and
f
p
=
NNODE
I=1
_
B
T
I
Cε
p
dV. (4.101)
Equation (4.100) may alternatively be written as
ku = f
, (4.102)
where
f
= f
p
+ f.
4.4.4.1 Single truss element undergoing elastic–plastic deformation. We now
return to the single truss element which we examined for the case of elastic material
behaviour in Section 4.4.2.1. The problem is shown in Fig. 4.10(a) and the single
element discretization in 4.10(b). For this problem, the B matrix, and the stiffness
matrix, k, remain unchanged as
B =
1
L
[−1 1], k =
EA
L
_
1 −1
−1 1
_
Finite element equilibrium equations 135
and we write the incremental external force vector as
f =
_
F
1
P
_
.
The plasticity ‘force’ is determined from (4.101) for this element as
f
p
=
_
B
T
Cε
p
dV =
_
1
ξ=0
1
L
_
−1
1
_
Eε
p
A det(J) dξ = EAε
p
_
−1
1
_
so that from (4.100) the equilibrium equation becomes
EA
L
_
1 −1
−1 1
_ _
u
1
u
2
_
− EAε
p
_
−1
1
_
=
_
F
1
P
_
.
With the boundary condition that u
1
= 0, this reduces to
EA
L
u
2
− EAε
p
= P.
Rearranging gives
u
2
=
PL
EA
+ Lε
p
. (4.103)
In the absence of plasticity, this just gives us an incremental formof the expression we
obtained before for elasticity in Section 4.4.2.1. For the case of plasticity, however,
the increment indisplacement nowbecomes that due tothe elastic deformationtogether
with that resulting from the plastic strain, which unsurprisingly in this small strain
formulation has magnitude Lε
p
. To demonstrate further the need for an incremental
approach, let us now assign elastic linear strain hardening plasticity properties to the
rod material and determine the rod extension on application of a given force, P.
In Section 2.4.2, we saw that for uniaxial linear isotropic hardening the relation
between the increment in uniaxial stress and plastic strain was given by
ε
p
=
σ
h
in which h is the strain hardening constant. Writing σ = P/A gives
ε
p
=
P
Ah
and substituting into (4.102) gives
u
2
=
PL
EA
+
LP
Ah
=
PL
A
_
1
E
+
1
h
_
. (4.104)
Therefore, the incremental displacement for elastic and elastic–plastic conditions
becomes
Elastic : σ < σ
y
, u
2
=
PL
EA
.
Elastic–plastic : σ ≥ σ
y
, u
2
=
PL
A
_
1
E
+
1
h
_
.
136 Finite element method
E=1000 MPa
A=1.0 mm
2
s
y
=10 MPa
h =100 MPa
L=10 mm
11 N
Fig. 4.18 A uniform bar under axial loading with the properties shown.
Let us analyse incrementally the problem shown in Fig. 4.18.
We will apply the load incrementally. The ﬁrst load increment, however, will be
chosen to cause ﬁrst yield. That is, P
1
= σ
y
A = 10 N. The displacement increment
is given by
u
(1)
=
PL
EA
=
10 × 10
1000 × 1
= 0.1 mm
so
u
(1)
= u
(0)
+ u
(1)
= 0 + 0.1 = 0.1 mm.
Subsequent displacement increments must take account of the plasticity so that
u
(2)
=
PL
EA
+
LP
Ah
=
1 × 10
1000 × 1
+
10 × 1
1 × 100
= 0.11 mm
and
u
(2)
= u
(1)
+ u
(2)
= 0.1 + 0.11 = 0.21 mm.
We see from Equation (4.104) that since the rate of hardening is constant (i.e. h is
ﬁxed), the displacement—load relationship becomes independent of increment size,
once the material has yielded. However, were we to introduce nonlinear hardening
where, for example, h depends on plastic strain, Equation (4.104) shows that the
displacement—load relationship then becomes dependent on the increment size.
4.5 Integration of momentum balance and equilibrium equations
4.5.1 Explicit integration using the central difference method
The ﬁnite element discretization of the momentum balance equation for a damped
system can be expressed in matrix form as follows:
M¨ u + C˙ u + F
int
= F
ext
,
where M is the mass matrix, F
int
the internal force vector, and F
ext
the external force
vector. The integration of the equations can be carried out by means of the explicit
central difference time integration scheme. The scheme is derived from Taylor series
Integration of momentum balance and equilibrium equations 137
expansions of the displacements, u, as follows.
u(t + t ) = u(t ) + ˙ u(t )t + ¨ u(t )
t
2
2
+ · · ·
u(t − t ) = u(t ) − ˙ u(t )t + ¨ u(t )
t
2
2
− · · · .
The central difference approximations are obtained taking the difference of the above
expressions to give the velocity
˙ u
N
=
u
N+1
− u
N−1
2t
and by summing the expressions for the accelerations
¨ u
N
=
u
N+1
− 2u
N
+ u
N−1
(t )
2
,
where
u
N
= u(t
N
).
Assuming that the acceleration is constant between t
N
and t
N+1
the central difference
approximations can be rearranged to give the following secondorder integration
scheme
u
N+1
= u
N
+ ˙ u
N
t +
1
2
¨ u
N
(t )
2
,
˙ u
N+1
= ˙ u
N
+
1
2
(¨ u
N
+ ¨ u
N+1
)t.
These are more commonly rewritten by deﬁning the intermediate velocities based
on the assumption that the acceleration is constant between t
0
and t
0+1/2
as well as
between t
N−1/2
and t
N+1/2
so that
˙ u
1/2
= ˙ u
0
+
1
2
¨ u
0
t
to give the leap frog explicit method
˙ u
N+1/2
= ˙ u
N−1/2
+ ¨ u
N
t,
u
N+1
= u
N
+ ˙ u
N+1/2
t.
The central difference method can be applied with a varying time increment (which
is particularly important if the response of the continuum is nonlinear), as illustrated
in Fig. 4.19.
Let t
N+1
be the time increment between t
N
and t
N+1
with u
N
= u(t
N
) as
illustrated in Fig. 4.19. The midstep velocities are deﬁned by
˙ u
N−1/2
= ˙ u(t
N−1/2
), ˙ u
N+1/2
= ˙ u(t
N+1/2
),
138 Finite element method
ü
N
t
t
N+1/2
t
N
∆t
N
∆t
N+1
t
N–1/2
t
N–1 t
N+1
u
N+1
u
N
N–1/2 N+1/2
Fig. 4.19 Central difference integration scheme.
where
t
N−1/2
=
1
2
(t
N−1
+ t
N
), t
N+1/2
=
1
2
(t
N
+ t
N+1
).
The central difference formula for velocity is
˙ u
N+1/2
=
u
N+1
− u
N
t
N+1
,
while the central difference formula for acceleration is
¨ u
N
=
˙ u
N+1/2
− ˙ u
N−1/2
t
N+1/2
,
where
t
N+1/2
=
1
2
(t
N
+ t
N+1
).
The discretization of the momentumbalance equation for an undamped systemcan be
obtained by substituting the acceleration term with its ﬁnite difference approximation
as follows:
M
˙ u
N+1/2
− ˙ u
N−1/2
t
N+1/2
+ F
int
N
= F
ext
N
,
which yields
˙ u
N+1/2
= M
−1
(F
ext
− F
int
)t
N+1/2
+ ˙ u
N−1/2
and
u
N+1
= u
N
+ ˙ u
N+1/2
t
N+1
.
Subsequently, the internal and external force vectors can be calculated as
F
int
N+1
= F
int
(u
N+1
), F
ext
N+1
= F
ext
(u
N+1
),
which completes the Nth time step. Furthermore, if the mass matrix is diagonalized
the system of differential equations uncouples and can be solved independently for
each degree of freedom (see, e.g. Newland).
Integration of momentum balance and equilibrium equations 139
4.5.1.1 Stability of the explicit time stepping scheme. The solution of a SDOF
equilibrium equation
¨ u + ω
2
u = 0
can be obtained using the central difference time stepping scheme as follows
¨ u
N
=
u
N+1
− 2u
N
+ u
N−1
(t )
2
= −ω
2
u
N
giving the difference equation
u
N+1
− (2 − ω
2
(t )
2
)u
N
+ u
N−1
= 0.
Trying solutions u
N
= A
n
where A is the ampliﬁcation factor gives
A
2
− (2 − ω
2
(t )
2
)A + 1 = 0.
The roots of this polynomial are
A =
_
1 −
ω
2
(t )
2
2
_
±
_
_
1 −
ω
2
(t )
2
2
_
2
− 1.
For stability
A ≤ 1,
which gives the stability condition as
t ≤
2
ω
.
For a general multidegree of freedom system, the stability condition becomes
t ≤
2
ω
max
,
where
ω
max
= max
i
{ω
i
}
is the element maximum eigenvalue. A conservative estimate of the stable time
increment is given by the following minimum taken over all the elements
t ≤ min
L
c
c
d
,
where L
c
is the characteristic element length and c
d
is the current effective dilatational
wave speed of the material.
140 Finite element method
4.5.2 Introduction to implicit integration
Implicit integration schemes are often preferred to their explicit counterparts since they
involve the determination of a residual force at each step and iteration within the step to
minimize the residual force to within a speciﬁed tolerance. We brieﬂy introduce three
techniques; the tangential stiffness method, the initial tangential stiffness method, and
the Newton–Raphson method. We conﬁne ourselves here to quasistatic problems so
assume inertia effects are negligible.
4.5.2.1 Tangential stiffness method. The discretized static equilibrium equation
given in (4.102), for example, will not generally be satisﬁed unless a convergence
occurs which can be expressed in terms of residual forces as follows
k(u)u −f = Ψ = 0.
The iteration starts from an initially guessed solution u
0
and the corresponding
tangential stiffness matrix k(u
0
). The residual forces are calculated from
Ψ
0
= k(u
0
)u
0
−f.
The correction u is calculated as follows:
u
n
= [K(u
n
)]
−1
Ψ
n
(4.105)
and an improved solution is then obtained as follows:
u
n+1
= u
n
+ u
n
.
The process continues until the residual forces Ψ
n
are smaller than a speciﬁed
tolerance.
4.5.2.2 Initial tangential stiffness method. The initial tangential stiffness method
differs fromthe previous method only in that the correction to the displacement, given
in Equation (4.105), is nowcalculated always based upon the initial tangential stiffness
matrix so that
u
n
= [k(u
0
)]
−1
Ψ
n
and as before, the improved solution is obtained as
u
n+1
= u
n
+ u
n
and the process continues until the residual forces Ψ
n
are smaller than a speciﬁed
tolerance.
Integration of momentum balance and equilibrium equations 141
4.5.2.3 Newton–Raphson method. As in the previous methods, the residual force
is determined from
= k(u)u −f = 0. (4.106)
Using a Taylor expansion, may be approximated by
(u) +
∂(u)
∂u
u + O(u
2
) = 0. (4.107)
The matrix J = ∂/∂u is called the Jacobian (which has nothing to do with the
Jacobian introduced above for the mapping of spatial to local element conﬁgurations),
or effective tangent stiffness, and from (4.106) it can be seen that this comprises a
term corresponding to the internal forces, termed the tangent stiffness matrix, and a
further term corresponding to the ‘external’ forces, called the load stiffness matrix.
Equation (4.107) provides a linearization of (4.106) and may be written as
+ Ju = 0
so that
J(u
n
)u
n
= −(u
n
)
and the displacement is updated by
u
n+1
= u
n
+ u
n
.
The iteration continues until the tolerance limit on residual force is achieved.
The two schemes introduced above, namely explicit and implicit integration, may
be used for the integration of the momentum balance, or equilibrium, ﬁnite element
equations. In implementing plasticity models into ﬁnite element formulations, it is
often necessary, in addition, to integrate a set of constitutive equations (e.g. to give the
plastic strain increment). For this additional integration, it is also possible to employ
either implicit or explicit integration methods. In the overall solution process, there
fore, there are several possible combinations of implicit and explicit integration. Finite
element methods which employ implicit schemes for the integration of the momentum
balance, or equilibrium equations, regardless of whether the constitutive equations
are integrated using implicit or explicit integration, are referred to as implicit ﬁnite
element methods. Similarly, ﬁnite element methods which employ explicit schemes
for the integration of the momentum balance, or equilibrium equations, are called
explicit ﬁnite element methods. Both implicit and explicit formulations are available
in ABAQUS. It can be seen that the implicit scheme offers the more robust overall
approach, because of the iteration necessary in order to achieve convergence. Explicit
schemes, however, can often produce more rapid solutions, and are more appropriate
for dynamic analyses. Care has to be taken, however, in choosing time step size and
ensuring the calculated solution does not drift away from the true solution. We shall
examine this further in the context of integration of constitutive equations in Chapter 5.
142 Finite element method
Further reading
Bathe, K.J. (1996). Finite Element Procedures. Prentice Hall, New Jersey, revised
edition.
Belytschko, T., Liu, K.W., and Moran, B. (2000). Nonlinear Finite Element Analysis
for Continua and Structures. John Wiley & Sons, New York.
Bonet, J. andWood, R.D. (1997). Nonlinear ContinuumMechanics for Finite Element
Analysis. Cambridge University Press.
Hinton, E. and Owen, D.R.J. (1979). An Introduction to Finite Element Computations.
Pineridge Press, Swansea.
Newland, D.E. (1989). Mechanical Vibration Analysis and Computation. Longman
Scientiﬁc and Technical, John Wiley and Sons, New York.
Owen, D.R.J. and Hinton, E. (1980). Finite Elements in Plasticity: Theory and
Practice. Pineridge Press, Swansea.
Timoshenko, S. and Goodier, J.N. (1983). Theory of Elasticity. McGrawHill,
New York.
Zienkiewicz, O.C. and Taylor, R.L. (1989). The Finite Element Method. McGrawHill,
London, 4th edition.
5. Implicit and explicit integration of
von Mises plasticity
5.1 Introduction
In this chapter, we shall return to the constitutive equations for plasticity introduced
in Chapter 2 and see how they may be integrated and how the Jacobian or tangent
stiffness may be obtained from them. We shall explore both implicit and explicit
schemes for the integration of the constitutive equations. To start, we shall consider
small strain, timeindependent, linear isotropic hardening plasticity before looking at
kinematic hardening and viscoplasticity. First, we return to the determination of the
plastic multiplier.
5.2 Implicit and explicit integration of constitutive equations
We saw from Chapter 2 that the yield function for isotropic hardening can be written
f = σ
e
− r − σ
y
=
_
3
2
σ
: σ
_
1/2
− r − σ
y
= 0 (5.1)
and that the plastic multiplier, in principal stress space is given by
dλ =
(∂f/∂σ) · C dε
(∂f/∂σ) · C(∂f/∂σ) − (∂f/∂p)[(2/3)(∂f/∂σ) · (∂f/∂σ)]
1/2
, (5.2)
where all terms are written in Voigt notation. Before proceeding, let us redetermine
the plastic multiplier more generally, without the constraint of working in principal
stress space. The consistency condition is written in terms of the stress tensor
df (σ, p) =
∂f
∂σ
: dσ +
∂f
∂p
dp = 0.
If we write the tensor normal, n, for a von Mises material
n =
∂f
∂σ
=
3
2
σ
σ
e
,
144 Implicit and explicit integration
the consistency condition becomes
df (σ, p) = n : dσ +
∂f
∂p
dp = 0,
where without loss of generality, we assume out of plane shears are zero, so
n =
⎡
⎣
n
11
n
12
0
n
12
n
22
0
0 0 n
33
⎤
⎦
.
Let us also write, in Voigt notation, the vector normal, n,
n =
⎡
⎢
⎢
⎣
n
11
n
22
n
33
2n
12
⎤
⎥
⎥
⎦
.
We can then see that
n : dσ =
⎡
⎣
n
11
n
12
0
n
12
n
22
0
0 0 n
33
⎤
⎦
:
⎡
⎣
dσ
11
dσ
12
0
dσ
12
dσ
22
0
0 0 dσ
33
⎤
⎦
=
⎡
⎢
⎢
⎣
n
11
n
22
n
33
2n
12
⎤
⎥
⎥
⎦
·
⎡
⎢
⎢
⎣
dσ
11
dσ
22
dσ
33
dσ
12
⎤
⎥
⎥
⎦
≡ n · dσ
so that, provided the normal vector, n, in Voigt notation contains twice the tensorial
normal shear terms, then
n : dσ ≡ n · dσ
and the consistency condition becomes
df (σ, p) = n · dσ +
∂f
∂p
dp = 0.
We may now determine the plastic multiplier with Hooke’s law written in Voigt
notation, as before
dσ = C(dε − dε
p
).
The increment in tensorial plastic strain is obtained from the normality hypothesis,
dε
p
= dλ
∂f
∂σ
= dλn,
which we may write in Voigt notation as
dε
p
= dλn
remembering that the shear strain terms in the Voigt strain vector are engineering
shears; that is, twice the tensorial shears. Hooke’s law then becomes
dσ = C(dε − dλn)
Implicit and explicit integration of constitutive equations 145
and combining with the consistency condition, in which, for a von Mises material,
dp = dλ, we obtain
n · C(dε − dλn) +
∂f
∂p
dλ = 0
so that the plastic multiplier is given by
dλ =
n · C dε
n · Cn − (∂f /∂p)
.
For linear isotropic hardening, r = hp so that ∂f /∂p = −∂r/∂p = −h and
dλ =
n · C dε
n · Cn + h
. (5.3)
Finally, the stress increment is given by
dσ = C dε
e
= C(dε − dε
p
) = C(dε − dλn). (5.4)
With knowledge of the total strain increment, dε, together with the stress, σ,
Equation (5.3) allows the plastic multiplier to be determined so that the stress incre
ment can be obtained from (5.4). The updated stress, σ + dσ may then be obtained.
This is called an explicit integration process. If we denote all quantities at time t with
subscript t , and those at the next increment forward in time, t +t , in a similar way,
then we may write
dλ
t
=
n
t
· C dε
t
n
t
· Cn
t
+ h
,
dσ
t
= C(dε
t
− dλn
t
), (5.5)
dr = hdp = hdλ
t
.
The integration to obtain all quantities at the end of the time step, t , may then be
written as
σ
t +t
= σ
t
+ dσ
t
,
ε
p
t +t
= ε
p
t
+ dε
p
t
, (5.6)
r
t +t
= r
t
+ dr
t
.
This is called a ﬁrstorder forward Euler explicit integration scheme. Its advantage is
its simplicity and it is straightforward to implement as we shall see later. However,
there are a number of disadvantages which need to be considered. First, because it is
an explicit scheme, it is conditionally stable; that is, it may become unstable. Second,
the accuracy of the integration depends, of course, on the time step size, t, chosen.
Great care is therefore required in ensuring that the time step does not become too
large such that erroneous results are obtained. Third, and perhaps most importantly,
146 Implicit and explicit integration
the plastic multiplier given in Equation (5.5) was obtained to ensure that at time, t ,
the yield condition in (5.1) is satisﬁed. However, the forward integration process does
not ensure that the yield condition is also satisﬁed at time t + t , and as a result,
it is possible for the solution, over many time steps, to drift away from the yield
surface. This is overcome by means of implicit integration of the equations which
has the additional advantage of being unconditionally stable. The accuracy remains
dependent, however, on the time step size. We introduce an implicit scheme, known
as the radial return method for von Mises plasticity, in Section 5.2.1.
5.2.1 Implicit integration: the radial return method
Figure 5.1(a) shows a von Mises yield surface with a schematic representation of
the explicit integration method described above. A step forward in time takes the
updated stresses outside of the yield surface. Figure 5.1(b) shows a representation of
an implicit scheme. A trial stress increment is chosen which again takes the updated
stresses, σ
tr
t +t
, outside of the yield surface. The stress is then updated with a plastic
correction to bring it back onto the yield surface at time t + t . In deviatoric stress
space, the plane stress von Mises ellipse becomes a circle, and the plastic correction
termis always directed towards the centre of the yield surface (because of the normality
condition). The technique has therefore come to be known as the radial return method.
In what follows, we shall take all quantities to be those at the end of a time step,
t + t , unless speciﬁcally stated. So, the stress at t + t is just σ and that at the
beginning of the time step, at time t , is σ
t
.
We may write Hooke’s lawin multiaxial formin terms of stress and strain tensors as
σ = 2Gε
e
+ λ Tr(ε
e
)I.
(a) (b)
s
t +∆t
s
t +∆t
s
y
s
y
s
y
s
y
s
2
s
y
s
2
s
1
s
1
s
t
s
y
s
y
s
y
s
t
Fig. 5.1 Schematic representations of (a) explicit integration and (b) implicit integration, using the
radial return method, of von Mises plasticity equations.
Implicit and explicit integration of constitutive equations 147
The elastic strain at the end of the time step may be written as
ε
e
= ε
e
t
+ ε
e
= ε
e
t
+ ε − ε
p
so that
σ = 2G(ε
e
t
+ ε − ε
p
) + λ Tr(ε
e
t
+ ε − ε
p
)I
and so
σ = 2G(ε
e
t
+ ε) + λ Tr(ε
e
t
+ ε)
. ,, .
I − 2Gε
p
. ,, .
(5.7)
Elastic predictor Plastic corrector
since
Tr(ε
p
) = 0.
The elastic predictor, or trial stress, is denoted by
σ
tr
= 2G(ε
e
t
+ ε) + λ Tr(ε
e
t
+ ε)I (5.8)
so that from (5.7),
σ = σ
tr
− 2Gε
p
= σ
tr
− 2Gpn, (5.9)
which we may write as
σ = σ
tr
− 2Gp
3
2
σ
σ
e
. (5.10)
The stress may be expressed in terms of its deviatoric and mean as
σ = σ
+
1
3
(σ : I)I (5.11)
so that with (5.10) we obtain
σ
+
1
3
(σ : I)I = σ
tr
− 3Gp
σ
σ
e
(5.12)
and rearranging gives
_
1 + 3G
p
σ
e
_
σ
= σ
tr
−
1
3
(σ : I)I. (5.13)
With some algebra, using Equation (5.8), we show that σ
tr
−
1
3
(σ : I)I is just the
deviatoric of the trial stress, σ
tr
as follows, where K is the elastic bulk modulus.
σ
tr
−
1
3
(σ : I)I = 2G(ε
e
t
+ ε) + λI(ε
e
t
+ ε) : I − Kε
e
: II
= 2G(ε
e
t
+ ε) + λI(ε
e
t
+ ε) : I − K(ε
e
t
+ ε − ε
p
) : II
= 2G(ε
e
t
+ ε) + λI(ε
e
t
+ ε) : I − KI(ε
e
t
+ ε) : I
= 2G(ε
e
t
+ ε) + (λ − K)I(ε
e
t
+ ε) : I ≡ σ
tr
.
148 Implicit and explicit integration
We therefore obtain
_
1 + 3G
p
σ
e
_
σ
= σ
tr
. (5.14)
If we take the contracted tensor product of each side of this with itself, we obtain
_
1 + 3G
p
σ
e
_
2
σ
: σ
= σ
tr
: σ
tr
or
_
1 + 3G
p
σ
e
_
σ
e
=
_
3
2
σ
tr
: σ
tr
_
1/2
≡ σ
tr
e
. (5.15)
This gives, ﬁnally,
σ
e
+ 3Gp = σ
tr
e
. (5.16)
The multiaxial yield condition is
f = σ
e
− r − σ
y
= σ
tr
e
− 3Gp − r − σ
y
= 0. (5.17)
This is generally a nonlinear equation in p which may be solved using Newton’s
method. We write
f +
∂f
∂p
dp + · · · = 0. (5.18)
For linear hardening, r = hp so that
∂r
∂p
=
∂r
∂p
= h. (5.19)
Substituting (5.17) into (5.18), using (5.19) therefore gives
σ
tr
e
− 3Gp − r − σ
y
+ (−3G − h) dp = 0.
Rearranging gives
dp =
σ
tr
e
− 3Gp − r − σ
y
3G + h
.
We may write the integration in iterative form then, as
r
(k)
= r
t
+ hp
(k)
,
dp =
σ
tr
e
− 3Gp
(k)
− r
(k)
− σ
y
3G + h
, (5.20)
p
(k+1)
= p
(k)
+ dp.
The effective stress may then be determined from (5.16); the deviatoric stress tensor
from (5.14), so that the plastic strain tensor increment is
ε
p
=
3
2
p
σ
σ
e
≡
3
2
p
σ
tr
σ
tr
e
Implicit and explicit integration of constitutive equations 149
and the elastic increment
ε
e
= ε − ε
p
and the stress increment is given by
σ = 2Gε
e
+ λIε
e
: I.
We have now introduced both explicit and implicit integration of the plasticity con
stitutive equations. It is important to note that in the implicit scheme, all quantities are
written at the end of the time increment. This ensures that the yield condition, given in
Equation (5.17), is satisﬁed at the end of the time increment, therefore avoiding ‘drift’
from the yield surface which can occur in the explicit scheme. The implicit scheme,
because it enables signiﬁcantly larger time increments to be used, generally leads to
much more rapid solutions.
In an implementation within implicit ﬁnite element code, we saw in Section 4.5.2.3
that the implicit integration of the momentum balance or equilibrium equations
requires the determination of the Jacobian that comprises both the tangent stiffness
matrix and the load stiffness matrix. The tangent stiffness matrix depends very much
on the material behaviour, and hence on the constitutive equations. In the implement
ation of a plasticity model into implicit ﬁnite element code, it is therefore necessary to
provide the material tangent stiffness matrix in addition to the integration of the plas
ticity constitutive equations. In explicit ﬁnite element code, however, which does not
depend upon knowledge of the Jacobian, the tangent stiffness matrix is not required.
The determination of the material Jacobian for implicit ﬁnite element code is very
much bound up with the integration of the constitutive equations used. Typically, in
implementing a plasticity model into commercial codes such as ABAQUS explicit or
LSDyna (which is an explicit code), for example, we need to provide a subroutine
which contains the integration of the plasticity constitutive equations (whether impli
cit or explicit). For implementations into implicit code, such as ABAQUS standard,
we need to provide a subroutine which contains both the integration of the plasti
city constitutive equations (whether implicit or explicit) together with the material
Jacobian or tangent stiffness matrix. It is useful to note from Section 4.5.2.3 that the
Jacobian is required in the iterative procedure in minimizing the force residual. If
convergence occurs after a given number of iterations, the Jacobian does not inﬂuence
the accuracy of the solution, but the rate at which convergence is achieved. It is for this
reason that often approximate Jacobians are used (e.g. in the initial tangent stiffness
method in Section 4.5.2.2). A further reason is that depending on the complexity of
the plasticity model, the material Jacobian may not be derivable in analytical terms
so that a numerical, approximate implementation has to be developed. Perturbation
methods can allow the accurate numerical determination of the Jacobian. In the fol
lowing section, we shall address ﬁrst, the material Jacobian for an elastic material, and
150 Implicit and explicit integration
then that for the timeindependent linear strain isotropic hardening plasticity model
for which the implicit integration scheme was presented in this section.
5.3 Material Jacobian
5.3.1 Isotropic elasticity
Hooke’s law may be written incrementally as
σ = 2Gε
e
+ λIε
e
: I, (5.21)
which may be written more succinctly as
σ = (2GI + λII) : ε
e
in which I is the fourthorder identity tensor with the properties I : I = I : I = I and
I : ε = ε : I = ε. We will deﬁne the material Jacobian here in the way it is
required in the ABAQUS ﬁnite element code as ∂σ/∂ε where
dσ =
∂σ
∂ε
dε
in which the shear strains are taken to be engineering shears. For example,
∂σ
11
∂γ
12
=
∂σ
11
∂ε
12
∂ε
12
∂γ
12
=
1
2
∂σ
11
∂ε
12
,
where ∂σ
11
/∂ε
12
is obtained from Equation (5.21). The material Jacobian
becomes, therefore, for conditions of plane strain or axial symmetry (γ
13
=
γ
23
= 0),
∂σ
∂ε
=
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎝
∂σ
11
∂ε
11
∂σ
11
∂ε
22
∂σ
11
∂ε
33
∂σ
11
∂γ
12
∂σ
22
∂ε
11
∂σ
22
∂ε
22
∂σ
22
∂ε
33
∂σ
22
∂γ
12
∂σ
33
∂ε
11
∂σ
33
∂ε
22
∂σ
33
∂ε
33
∂σ
33
∂γ
12
∂σ
12
∂ε
11
∂σ
12
∂ε
22
∂σ
12
∂ε
33
∂σ
12
∂γ
12
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎠
=
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎝
2G + λ λ λ 0
λ 2G + λ λ 0
λ λ 2G + λ 0
0 0 0
1
2
2G
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎠
.
Note that in general, ∂σ/∂ε and ∂σ/∂ε are not the same thing; the former Jacobian
quantity is required in ABAQUS, for example, for quadratic convergence. Unfortu
nately, the material Jacobian for plasticity is not quite so easy to obtain. It is derived in
Section 5.3.2 for the plasticity model considered above; that is, for timeindependent
isotropic linear strain hardening plasticity.
Material Jacobian 151
5.3.2 Material Jacobian for timeindependent isotropic linear strain
hardening plasticity
We start from Equation (5.14) by applying the differential operator, δ (in a similar
way as used to ﬁnd the ﬁrst variation of an integral in Chapter 4) so that we obtain
_
1 + 3G
p
σ
e
_
δσ
+
3G
σ
e
δpσ
−
3Gp
σ
2
e
δσ
e
σ
= δσ
tr
. (5.22)
Also, from (5.16), we may write
δσ
e
+ 3Gδp = δσ
tr
e
. (5.23)
The yield condition is written as
δf = δσ
e
− δr = 0
so that for linear hardening,
δσ
e
= δr = hδp,
where δp is not the plastic strain in the increment, but an inﬁnitesimal quantity, so
that we may write
δσ
e
= δr = hδp.
Combining with (5.23) gives
hδp + 3Gδp = δσ
tr
e
so
δp =
δσ
tr
e
h + 3G
. (5.24)
Combining with (5.23) gives
δσ
e
= δσ
tr
e
_
1 −
3G
h + 3G
_
. (5.25)
We now use (5.25), (5.24), and (5.19) in Equation (5.22) to eliminate δσ
e
, δp, and
p respectively to give (after some algebra)
σ
tr
e
σ
e
δσ
+
δσ
tr
e
σ
e
σ
e
_
σ
e
−
σ
tr
e
1 + (3G/h)
_
σ
= δσ
tr
. (5.26)
Consider the term,
δσ
tr
e
= δ
_
3
2
σ
tr
: σ
tr
_
1/2
=
1
2
_
3
2
σ
tr
: σ
tr
_
−1/2
_
3
2
δσ
tr
: σ
tr
+
3
2
σ
tr
: δσ
tr
_
=
3
2
1
σ
tr
e
σ
tr
: δσ
tr
.
152 Implicit and explicit integration
Substituting this together with the expression for σ
in Equation (5.14) into (5.26)
gives
δσ
=
3
2
_
1
1 + (3G/h)
−
σ
e
σ
tr
e
_
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: δσ
tr
+
σ
e
σ
tr
e
δσ
tr
. (5.27)
If we write
Q =
3
2
_
1
1 + (3G/h)
−
σ
e
σ
tr
e
_
and R =
σ
e
σ
tr
e
then (5.27) becomes
δσ
=
_
Q
σ
tr
σ
tr
e
σ
tr
σ
tr
e
+ RI
_
: δσ
tr
. (5.28)
Remember that in deriving the Jacobian, we are trying to relate δσ to δε. We may
write the deviatoric trial stress in terms of the deviatoric trial strain using Hooke’s law
(because the trial stress is obtained assuming elasticity) as
δσ
tr
= 2Gδε
tr
= 2G
_
δε
tr
−
1
3
II : δε
tr
_
and
δε
tr
≡ δε
so
δσ
tr
= 2G
_
δε −
1
3
II : δε
_
.
Substituting into Equation (5.28) gives
δσ
=
_
Q
σ
tr
σ
tr
e
σ
tr
σ
tr
e
+ RI
_
:
_
2G
_
δε −
1
3
II : δε
__
= 2GQ
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: δε −
1
3
Q
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: (II : δε) + 2GRδε −
2
3
GRI : (II : δε).
The second term of the righthand side is zero since σ
tr
is deviatoric so
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: (II : δε) =
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: I Tr(δε) = 0
and therefore
δσ
= 2GQ
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: δε + 2GRδε −
2
3
GRII : δε (5.29)
since I : I = I. Finally, the stress is given in terms of its deviatoric by
δσ = δσ
+
1
3
II : δσ = δσ
+ KII : δε
e
= δσ
+ KII : (δε − δε
p
)
= δσ
+ KII : δε.
Material Jacobian 153
Substituting into (5.29) gives
δσ = 2GQ
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: δε + 2GRδε +
_
K −
2
3
GR
_
II : δε. (5.30)
We may write this in the shortened form as
δσ =
_
2GQ
σ
tr
σ
tr
e
σ
tr
σ
tr
e
+ 2GRI +
_
K −
2
3
GR
_
II
_
: δε. (5.31)
Equation (5.30) provides the Jacobian or the material tangent stiffness matrix. In this
case, because it has been derived fromthe implicit backward Euler integration scheme,
which is used to integrate the plasticity constitutive equations, it is called the consistent
tangent stiffness. Let us determine some of the terms using (5.30).
δσ
11
= 2GQ
σ
tr
11
σ
tr
e
1
σ
tr
e
× (σ
tr
11
δε
11
+ σ
tr
22
δε
22
+ σ
tr
33
δε
33
+ 2σ
tr
12
δε
12
+ 2σ
tr
13
δε
13
+ 2σ
tr
23
δε
23
)
+ 2GRδε
11
+
_
K −
2
3
GR
_
(δε
11
+ δε
22
+ δε
33
).
The ﬁrst term of the Jacobian is, therefore,
D
11
=
∂δσ
11
∂δε
11
= 2GQ
σ
tr
11
σ
tr
e
σ
tr
11
σ
tr
e
+ 2GR +
_
K −
2
3
GR
_
and the next is
D
12
=
∂δσ
11
∂δε
22
= 2GQ
σ
tr
11
σ
tr
e
σ
tr
22
σ
tr
e
+
_
K −
2
3
GR
_
and so on. A shear term is given by
D
44
=
∂δσ
12
∂δγ
12
=
1
2
∂δσ
12
∂δε
12
=
1
2
_
2GQ
σ
tr
12
σ
tr
e
2σ
tr
12
σ
tr
e
+ 2GR
_
= 2GQ
σ
tr
12
σ
tr
e
σ
tr
12
σ
tr
e
+ GR
and D
14
is
D
14
=
∂δσ
11
∂δγ
12
=
1
2
∂δσ
11
∂δε
12
= 2GQ
σ
tr
11
σ
tr
e
σ
tr
12
σ
tr
e
.
For conditions of axial symmetry (so that the out of plane shears, σ
13
, ε
13
and σ
23
, ε
23
do not exist) therefore, the Jacobian is the symmetrical matrix
∂δσ
∂δε
=
⎛
⎜
⎜
⎝
D
11
D
12
D
13
D
14
D
22
D
23
D
24
D
33
D
34
D
44
⎞
⎟
⎟
⎠
.
154 Implicit and explicit integration
5.4 Kinematic hardening
In this section, we shall use implicit (backward Euler) integration for the case of linear
kinematic hardening, and obtain the consistent tangent stiffness. We will then go on to
address combined linear kinematic hardening with nonlinear isotropic hardening, and
to introduce combined nonlinear kinematic and isotropic hardening. We will ﬁnish
by introducing semiimplicit integration schemes for plasticity.
5.4.1 Linear kinematic hardening
With the backward Euler scheme, as given in Section 5.2.1.2 for isotropic hardening,
Hooke’s law may be written in its predictor–corrector form as
σ = σ
tr
− 2Gpn (5.32)
in which σ
tr
is the elastic trial stress. We employ linear kinematic hardening so that
the back stress increment is given by
x =
2
3
cε
p
and as a result, if we write the back stress at time, t , as x
t
, then at the end of the time
step, t + t , it becomes
x = x
t
+
2
3
cε
p
(5.33)
and with the normality hypothesis for von Mises plasticity,
x = x
t
+
2
3
cpn, (5.34)
where
n =
3
2
σ
−x
σ
e
(5.35)
for kinematic hardening, in which x is, of course, itself deviatoric, and p is the
increment in effective plastic strain. Combining (5.35) and (5.34) gives
σ
= x
t
+
2
3
pcn +
2
3
σ
e
n. (5.36)
The deviatoric stress is
σ
= σ −
1
3
II : σ
so that with (5.32), we may write (5.36) as
σ
tr
− 2Gpn −
1
3
II : σ = x
t
+
2
3
pcn +
2
3
σ
e
n.
Kinematic hardening 155
We saw earlier that we may write
σ
tr
−
1
3
II : σ ≡ σ
tr
so that
σ
tr
− 2Gpn = x
t
+
2
3
pcn +
2
3
σ
e
n. (5.37)
Taking the contracted product of both sides of (5.37) with n gives
(σ
tr
−x
t
) : n = n : n
_
2Gp +
2
3
pc +
2
3
σ
e
_
(5.38)
and
n : n =
3
2
σ
−x
σ
e
:
3
2
σ
−x
σ
e
=
3
2σ
2
e
_
3
2
(σ
−x) : (σ
−x)
_
.
But the effective stress, σ
e
, is deﬁned by
σ
e
=
_
3
2
(σ
−x) : (σ
−x)
_
1/2
so that
n : n =
3
2
.
Equation (5.38) therefore reduces to
(σ
tr
−x
t
) : n = 3Gp + pc + σ
e
. (5.39)
From (5.32) we may write
σ
= σ
tr
− 2Gpn
so that from (5.35), we may write n as
n =
3
2
_
σ
tr
− 2Gpn −x
σ
e
_
=
3
2
_
σ
tr
− 2Gpn −x
t
− (2/3)cpn
σ
e
_
.
Rearranging this equation for n gives
n =
3
2
_
σ
tr
−x
t
σ
e
+ 3Gp + cp
_
(5.40)
and substituting into (5.39) gives
_
3
2
(σ
tr
−x
t
) : (σ
tr
−x
t
)
_
1/2
≡ σ
tr
e
= 3Gp + pc + σ
e
. (5.41)
Note that the effective trial stress is determined with respect to the back stress taken at
time, t , rather than at the end of the time increment. Rearranging this equation gives
p =
σ
tr
e
− σ
e
3G + c
. (5.42)
156 Implicit and explicit integration
The yield function for the case of kinematic hardening is
f = σ
e
− σ
y
= 0 (5.43)
so that
σ
e
= σ
y
.
In Equation (5.42), therefore, for the case of linear kinematic hardening, none of the
terms depends upon the effective plastic strain. Equation (5.42) is therefore an exact,
closedform expression for p and in this instance, no iteration is required for its
determination. Plastic and elastic strains and the stress may then be updated using
(5.42) to give
n =
3
2
_
σ
tr
−x
t
σ
tr
e
_
.
ε
p
= pn,
ε
p
= ε
p
t
+ ε
p
,
ε
e
= ε
e
t
+ ε
e
= ε
e
t
+ ε − ε
p
,
σ = 2Gε
e
+ λII : ε
e
.
5.4.1.1 Consistent tangent stiffness for linear kinematic Hardening. In implicit
ﬁnite element code, in addition to the integration of the plasticity constitutive
equations, it is also necessary to provide the material Jacobian—the tangent stiff
ness matrix. Here, we determine the consistent (i.e. with the implicit integration given
above) tangent stiffness for linear kinematic hardening.
Using (5.40) and (5.41), n may be written as
n =
3
2
_
σ
tr
−x
t
σ
tr
e
_
.
We have therefore
σ
= σ
tr
− 2Gpn = σ
tr
− 2Gp
3
2
_
σ
tr
−x
t
σ
tr
e
_
and substituting for p from (5.42) gives
σ
= σ
tr
− 2G
_
σ
tr
e
− σ
e
3G + c
_
3
2
_
σ
tr
−x
t
σ
tr
e
_
and after a little algebra, we obtain
σ
= σ
tr
_
c/3G + σ
e
/σ
tr
e
1 + c/3G
_
+
_
1 − σ
e
/σ
tr
e
1 + c/3G
_
x
t
.
Kinematic hardening 157
Applying the differential operator, and rearranging, we obtain
(1 + c/3G)
σ
tr
e
σ
e
δσ
= δσ
tr
(1 + (c/3G)(σ
tr
e
/σ
e
))
+σ
tr
(δσ
e
/σ
e
− δσ
tr
e
/σ
tr
e
) − (δσ
e
/σ
e
− δσ
tr
e
/σ
tr
e
)x
t
.
From the yield function in (5.43), δf = δσ
e
= 0 so this reduces to
(1 + c/3G)
σ
tr
e
σ
e
δσ
= δσ
tr
(1 + (c/3G)(σ
tr
e
/σ
e
)) − (σ
tr
−x
t
)(δσ
tr
e
/σ
tr
e
).
(5.44)
Now,
σ
tr
e
=
_
3
2
(σ
tr
−x
t
) : (σ
tr
−x
t
)
_
1/2
so
δσ
tr
e
=
1
2
_
3
2
(σ
tr
−x
t
) : (σ
tr
−x
t
)
_
−1/2
_
3
2
(σ
tr
−x
t
) : δσ
tr
+
3
2
δσ
tr
: (σ
tr
−x
t
)
_
=
1
σ
tr
e
3
2
(σ
tr
−x
t
) : δσ
tr
.
Substituting into (5.44), and writing
Q =
σ
e
/σ
tr
e
+ c/3G
1 + c/3G
R = −
3
2
σ
e
σ
tr
e
1
1 + c/3G
gives
δσ
= Qδσ
tr
+ R
(σ
tr
−x
t
)
σ
tr
e
(σ
tr
−x
t
)
σ
tr
e
: δσ
tr
.
As before,
δσ
tr
= 2G
_
δε −
1
3
II : δε
_
so that
δσ
= 2GQδε + 2GR
(σ
tr
−x
t
)
σ
tr
e
(σ
tr
−x
t
)
σ
tr
e
: δε −
2
3
GQII : δε
and ﬁnally,
δσ = 2GQδε + 2GR
(σ
tr
−x
t
)
σ
tr
e
(σ
tr
−x
t
)
σ
tr
e
: δε −
_
K −
2
3
GQ
_
II : δε
(5.45)
which gives the consistent tangent stiffness.
158 Implicit and explicit integration
5.4.2 Combined isotropic and linear kinematic hardening
The isotropic hardening evolution equation is
dr = hdp
in which h can be a function of p so that r is not necessarily linear, and for the linear
kinematic hardening,
dx =
2
3
c dε
p
.
Because of the assumption of linearity, as before we may write
x = x
t
+
2
3
cpn
and from Section 5.4, we have
σ
e
= σ
tr
e
− (3G + c)p (5.46)
in which σ
tr
e
and σ
e
are given by
σ
tr
e
=
_
3
2
(σ
tr
−x
t
) : (σ
tr
−x
t
)
_
1/2
, σ
e
=
_
3
2
(σ
−x) : (σ
−x)
_
1/2
.
The yield function is
f = σ
e
− r − σ
y
= 0
which becomes, after substituting (5.46)
f = σ
tr
e
− (3G + c)p − r − σ
y
= 0. (5.47)
In this case, r depends on h which may, for nonlinear hardening, depend on p and
the equation can be nonlinear. We use Newton’s method to solve it by writing
f +
∂f
∂p
dp = 0
and substituting (5.47) gives
dp =
σ
tr
e
− (3G + c)p − r − σ
y
3G + c + h
. (5.48)
Let us simplify it for the cases of linear kinematic hardening and isotropic hardening
only to compare with the previous equations. If there is no isotropic hardening, then
r = h = 0 and (5.48) becomes
dp =
σ
tr
e
− (3G + c)p − σ
y
3G + c
(5.49)
Kinematic hardening 159
and for linear kinematic hardening, σ
e
= σ
tr
e
− (3G + c)p and σ
e
= σ
y
so (5.49)
gives
dp = 0.
That is, there is no iteration required and p is given by
p =
σ
tr
e
− σ
y
3G + c
as before. If there is no kinematic hardening, x = 0 and c = 0 so (5.49) becomes
dp =
σ
tr
e
− 3Gp − r − σ
y
3G + h
,
where
σ
tr
e
=
_
3
2
σ
tr
: σ
tr
_
1/2
which is what we obtained before in Section 5.2.1.
5.4.3 Introduction to implicit integration of combined nonlinear kinematic
and isotropic hardening
In the previous sections in which we assumed linear kinematic hardening, we wrote
the kinematic hardening variable in terms of its value at the start of the increment,
x
t
, and the incremental value. As a result, subsequent equations were set up in terms
of x
t
. Here, we will now introduce the more general case of nonlinear kinematic
hardening, and will ﬁnd the equations depending on quantities at the end of the time
increment.
As before, the deviatoric stress can be written in predictor–corrector form in terms
of the trial stress and plastic return as
σ
= σ
tr
− 2Gpn (5.50)
and the normal, n, is given by
n =
3
2
σ
−x
σ
e
. (5.51)
Combining these equations and after a little algebra, we obtain
(σ
tr
−x) : n =
_
2
3
σ
e
+ 2Gp
_
n : n
and we have shown before that n : n =
3
2
, so
(σ
tr
−x) : n = σ
e
+ 3Gp. (5.52)
160 Implicit and explicit integration
We can also obtain from (5.50) and (5.51)
n =
3
2
σ
tr
− 2Gpn −x
σ
e
so that rearranging,
n =
3
2
σ
tr
−x
σ
e
+ 3Gp
.
Combining with (5.52) gives
σ
e
= σ
tr
e
− 3Gp, (5.53)
where
σ
tr
e
=
_
3
2
(σ
tr
−x) : (σ
tr
−x)
_
1/2
. (5.54)
The yield function is
f = σ
e
− r − σ
y
= σ
tr
e
− 3Gp − r − σ
y
= 0 (5.55)
with σ
tr
e
given by (5.54). Note now that because x depends on p (nonlinearly for
the case of nonlinear kinematic hardening) as may r, Equation (5.55) is a nonlinear
equation in p for which we will need to use Newton’s iterative solution method, for
which we obtain
dp =
σ
tr
(k)
e
− 3Gp
(k)
− r
(k)
− σ
y
3G + (∂r
(k)
/∂p) − (∂σ
tr
(k)
e
/∂p)
(5.56)
and
p
(k+1)
= p
(k)
+ dp.
Both σ
tr
e
and r are the two derivatives with respect to p need to be updated at every
iteration for a fully implicit integration. Often, with complex plasticity hardening
laws, this may be challenging. A simpler approach is to integrate the effective plastic
strainimplicitlybut toupdate the normal nandthe internal variables—the isotropic and
kinematic hardening—explicitly. Such an approach is called semiimplicit integration.
5.4.4 Semiimplicit integration of combined nonlinear kinematic and
isotropic hardening
In the semiimplicit scheme, Newton’s method is used to determine the effective
plastic strain increment using Equation (5.56) and the yield function (5.55) written at
the end of the time increment ensures that drift from the yield surface does not occur.
However, the updates of all other quantities are carried out explicitly as follows:
ε
p
(k+1)
= p
(k+1)
n
t
,
Implicit integration in viscoplasticity 161
where
n
t
=
3
2
σ
t
−x
t
σ
e
t
,
ε
p
(k+1)
= ε
p
t
+ ε
p
(k+1)
,
ε
e
(k+1)
= ε −ε
p
(k+1)
,
σ
(k+1)
= 2Gε
e
(k+1)
+ λII : ε
e
(k+1)
,
r
(k+1)
= r
t
+ r,
x
(k+1)
= x
t
+ x,
where, for example, the nonlinear kinematic hardening increment, x, is given by
x =
2
3
cε
p
(k+1)
+ γ x
t
p
(k+1)
and the isotropic hardening increment by
r = b(Q − r)p
(k+1)
if the nonlinear evolution equations given in Chapter 2 are adopted, and
σ
tr
(k+1)
e
=
_
3
2
(σ
tr
−x
(k+1)
) : (σ
tr
−x
(k+1)
)
_
1/2
.
The semiimplicit scheme is not unconditionally stable and we need to be concerned,
therefore, about both stability and accuracy when using it.
5.5 Implicit integration in viscoplasticity
We saw in Chapter 2 that the plastic strain rate can be written, for a viscoplastic
von Mises material, as
˙ ε
p
= ˙ p
∂f
∂σ
in which, for viscoplasticity, ˙ p is nowspeciﬁed by a constitutive equation (as opposed
to being determined by the consistency condition) which, for both isotropic and
kinematic hardening, can be written as
˙ p = φ(σ, x, r).
We may write this incrementally as
p = φ(σ, x, r)t = φt. (5.57)
For both kinematic and isotropic hardening, the yield function is
f = J(σ
−x
) − r − σ
y
.
162 Implicit and explicit integration
5.5.1 Uniaxial viscoplasticity equations
We will present implicit backward Euler integration for the uniaxial form of the
equations ﬁrst, for simplicity, and we start from the viscoplastic constitutive equation.
Equation (5.57) may be written in uniaxial form, if we consider isotropic hardening
only for now, as
p = φ(σ
e
, r)t (5.58)
and we may write this in a form suitable for Newton’s iterative solution as
ψ = p − φ(σ
e
, r)t = 0. (5.59)
In preparation for differentiation, we may write the stress in terms of p by
remembering (5.16)
σ
e
+ 3Gp = σ
tr
e
(5.60)
so that (5.59) becomes
ψ(p, r) = p − φ(p, r)t = 0. (5.61)
Using Newton’s method, we have
ψ +
∂ψ
∂p
dp +
∂ψ
∂r
dr = 0 (5.62)
and differentiating (5.61) and substituting into (5.62) gives
p − φt +
_
1 −
∂φ
∂p
t
_
dp −
∂φ
∂r
t dr = 0. (5.63)
We will write, from here on,
φ
p
=
∂φ
∂p
and φ
r
=
∂φ
∂r
and similarly for other derivatives. If we assume for now linear isotropic harden
ing, then
dr = hdp = hdp.
Substituting into (5.63) and rearranging gives
dp =
φ(p, r) − p/t
1/t − φ
p
− hφ
r
. (5.64)
As an example, let us consider a particular viscoplasticity constitutive equation,
given by
˙ p = φ(σ
e
, r) = α sinh β(σ
e
− r − σ
y
)
in which σ
y
, α, and β are material constants. Using (5.59) this becomes
˙ p = φ(p, r) = α sinh β(σ
tr
− 3Gp − r − σ
y
)
Implicit integration in viscoplasticity 163
and the required derivatives are
φ
p
= −3Gαβ cosh β(σ
tr
− 3Gp − r − σ
y
)
and
φ
r
= −αβ cosh β(σ
tr
− 3Gp − r − σ
y
)
so that (5.64) gives
dp
=
α sinh β(σ
tr
− 3Gp − r − σ
y
) − p/t
1/t +3Gαβ cosh β(σ
tr
−3Gp−r −σ
y
)+hαβ cosh β(σ
tr
−3Gp−r −σ
y
)
with
r = r
t
+ hp.
As before in rateindependent plasticity, an iterative procedure is used to determine
p from
p
(k+1)
= p
(k)
+ dp.
5.5.2 Multiaxial viscoplasticity
For the case of isotropic hardening, Equation (5.64) still holds for the multiaxial case,
and the plastic strain tensor increment can then be determined from
ε
p
= pn =
3
2
p
σ
σ
e
≡
3
2
p
σ
tr
σ
tr
e
so that the elastic strain and hence stress increments can be determined in the usual way.
We shall next consider combined multiaxial linear isotropic and kinematic harden
ing such that the constitutive equation is written as
˙ p = φ(p, x, r)
in which x is the tensorial kinematic hardening variable. The problem may be written,
as before,
ψ(p, x, r) = p − φ(p, x, r)t = 0
and
ψ +
∂ψ
∂p
dp +
∂ψ
∂x
: dx +
∂ψ
∂r
dr = 0. (5.65)
We may write the derivatives as
∂ψ
∂p
= 1 − φ
p
t,
∂ψ
∂x
= −φ
x
t,
∂ψ
∂r
= −φ
r
t
164 Implicit and explicit integration
so that (5.65) becomes
p − φt + (1 − φ
p
t ) dp −φ
x
: dxt − φ
r
drt = 0. (5.66)
Let us write the linear isotropic and kinematic hardening equations as
dr = hdp = hdp,
dx =
2
3
c dε
p
=
2
3
c dpn
so that substituting into (5.66) and rearranging gives
dp =
φt − p
1 − φ
p
t − hφ
r
t − (2/3)cφ
x
: nt
. (5.67)
If we consider again a viscoplastic constitutive equation with kinematic and isotropic
hardening of the form
˙ p = φ(σ, x, r) = α sinh β(J(σ
−x
) − r − σ
y
) = α sinh β(σ
e
− r − σ
y
),
then
φ
x
= −
∂φ
∂σ
e
∂σ
e
∂x
= −φ
σ
e
3
2
σ
−x
J(σ
−x
)
= −φ
σ
e
n
and the term φ
x
: n may be simpliﬁed to
φ
x
: n = −φ
σ
e
n : n = −
3
2
φ
σ
e
.
Substituting into (5.67) gives
dp =
φ − p/t
1/t − φ
p
− hφ
r
+ cφ
σ
e
(5.68)
which is solved iteratively. The plastic strain increment is then determined from
ε
p
=
3
2
p
σ
−x
J(σ
−x
)
≡
3
2
p
σ
tr
−x
σ
tr
e
.
5.5.3 Consistent tangent stiffness for viscoplasticity with isotropic hardening
The predictor–corrector form of the stress is
σ
= σ
tr
− 2Gpn or σ = σ
tr
− 2Gpn, (5.69)
where
n =
3
2
σ
tr
σ
tr
e
=
3
2
σ
σ
e
(5.70)
Implicit integration in viscoplasticity 165
and
σ
e
= σ
tr
e
− 3Gp.
Rearranging (5.70) gives
σ
=
σ
e
σ
tr
e
σ
tr
and applying the differential operator
δσ
=
σ
e
σ
tr
e
δσ
tr
+
_
δσ
e
σ
tr
e
−
σ
e
σ
tr
e
δσ
tr
e
σ
tr
e
_
σ
tr
. (5.71)
Now,
δσ
tr
e
=
1
σ
tr
e
3
2
σ
tr
: δσ
tr
(5.72)
and
δσ
e
=
1
σ
e
3
2
σ
: δσ
. (5.73)
Substituting (5.72) and (5.73) into (5.71) and eliminating σ
using (5.70) gives
δσ
=
3
2
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: δσ
+
σ
e
σ
tr
e
δσ
tr
−
σ
e
σ
tr
e
3
2
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: δσ
tr
.
Because of the contracted product of δσ
in the ﬁrst term on the righthand side, we
will substitute for this using the differential form of (5.50)
δσ
= δσ
tr
− 2Gδpn
to give
δσ
=
3
2
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: (δσ
tr
− 2Gδpn) +
σ
e
σ
tr
e
δσ
tr
−
σ
e
σ
tr
e
3
2
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: δσ
tr
= −
3
2
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: 2Gδpn +
σ
e
σ
tr
e
δσ
tr
+
_
1 −
σ
e
σ
tr
e
_
3
2
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: δσ
tr
.
Now,
3
2
σ
tr
σ
tr
e
: n = n : n =
3
2
so
δσ
= −
σ
tr
σ
tr
e
: 3Gδpn +
σ
e
σ
tr
e
δσ
tr
+
_
1 −
σ
e
σ
tr
e
_
3
2
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: δσ
tr
. (5.74)
The viscoplastic constitutive equation is written as, from (5.57),
p = φ(σ, r) t (5.75)
166 Implicit and explicit integration
so that
δp = (φ
σ
: δσ + φ
r
δr) t. (5.76)
Let us assume isotropic hardening of the form
δr = q(r)δp.
Because δp is again an inﬁnitesimal increment in p (as opposed to the plastic strain
increment), this can be written as
δr = q(r)δp. (5.77)
Substituting this together with the second equation of (5.69) for δσ into (5.76) gives
δp = (φ
σ
: (δσ
tr
− 2Gδpn) + φ
r
q(r)δp)t
and rearranging
δp =
φ
σ
: δσ
tr
1/t + 2Gφ
σ
: n + φ
r
q(r)
. (5.78)
Substituting into (5.74) gives
δσ
=−
σ
tr
σ
tr
e
: 3G
φ
σ
: δσ
tr
1/t +2Gφ
σ
: n+φ
r
q(r)
n+
σ
e
σ
tr
e
δσ
tr
+
_
1−
σ
e
σ
tr
e
_
3
2
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: δσ
tr
.
Let us write
α = −
3G
1/t + 2Gφ
σ
: n + φ
r
q(r)
, β =
σ
e
σ
tr
e
, γ =
3
2
_
1 −
σ
e
σ
tr
e
_
,
then
δσ
= α
σ
tr
σ
tr
e
φ
σ
: δσ
tr
+ βδσ
tr
+ γ
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: δσ
tr
. (5.79)
Now,
δσ
tr
= 2Gδε
= 2Gδε −
2
3
G(δε : I)I.
Because φ is a function of the effective stress, its derivatives with respect to the stress
components are deviatoric, and hence φ
σ
: δσ
tr
= φ
σ
: (δσ
tr
+ (1/3)δσ
tr
: II) =
φ
σ
: δσ
tr
since φ
σ
: I = 0. In addition,
δσ = δσ
+ KII : δε
so that (5.79) becomes
δσ =
_
2Gα
σ
tr
σ
tr
e
φ
σ
+
_
K −
2
3
Gβ
_
II + 2Gγ
σ
tr
σ
tr
e
σ
tr
σ
tr
e
_
: δε + 2Gβδε (5.80)
which gives the consistent tangent stiffness.
Incrementally objective integration for large deformations 167
5.6 Incrementally objective integration for large deformations
So far, we have addressed the update of stress over a given time increment on the basis
that there is no incremental rotation. In large deformation analyses, however, in which
deformations and rigid body rotations can become large, we need to take account of
the effect of incremental rotation on the determination of the updated stress. Figure 5.1
shows the representation of a body in its original conﬁguration and then in deformed
conﬁgurations at times t and t + t .
In the deformed conﬁgurations, the local material (or corotational) coordinate
systems are shown indicating that an incremental rotation has occurred between the
conﬁgurations at times t and t + t . We need to ensure that the stress update is
carried out with respect to the same reference frame. For example, in using a user
deﬁned subroutine for plasticity in ABAQUS, the user is supplied with the strains at
the beginning and end of the time increment, together with the stress at the beginning
of the time increment, but all these quantities and other tensor quantities have usually
already been incrementally rotated to account for the incremental rigid body rotation.
We shall returntothis inChapter 6. For now, however, we shall lookat anincrementally
objective stress update.
First, referring to Fig. 5.2, let us assume we know the stresses, σ
t
, at time t with
respect to the material reference frame, that is, the undeformed conﬁguration at that
time. Also, we have determined the spin, W, given by the antisymmetric part of the
velocity gradient.
The stress rate, ˙ σ, with respect to the material reference frame, or undeformed
conﬁguration, at time t , taking full account of the incremental rigid body rotation, is
F
t
F
t +∆t
X
Deformed
configuration at t +∆t
Undeformed (original)
configuration
Y
x
x
xЈ
y
y
yЈ
Deformed
configuration at t
Fig. 5.2 A body in its original (undeformed) conﬁguration and in deformed conﬁgurations at times t
and t + t .
168 Implicit and explicit integration
given by
˙ σ =
∇
σ +W
t
σ
t
−σ
t
W
t
(5.81)
so that the updated stress (written explicitly) with respect to the undeformed
conﬁguration at time t is simply
σ
t +t
= σ + ˙ σt.
We can write Equation (5.81) in incremental form, if we wish, as
σ =
∇
σ +(W
t
σ
t
−σ
t
W
t
)t (5.82)
which we may also write as
σ =
∇
σ +Rσ
t
R
T
(5.83)
in which
∇
σ is the increment in objective, or corotational, stress, and R is a rotation
to be determined.
We shall further consider these approaches in Chapter 6, concerned with the
implementation of plasticity models into ﬁnite element code.
Further reading
Bathe, K.J. (1996). Finite Element Procedures. Prentice Hall, New Jersey, revised
edition.
Belytschko, T., Liu, K.W., and Moran, B. (2000). Nonlinear Finite Element Analysis
for Continua and Structures. John Wiley & Sons, New York.
Simo, J.C. and Hughes, T.J.R. (1998). Computational Inelasticity. SpringerVerlag,
New York.
Chaboche, J.L. (1986). ‘Timeindependent constitutive theories for cyclic plasticity’.
International Journal of Plasticity, 2(2), 149–188.
Hughes, T.J.R. (1984). ‘Numerical implementation of constitutive models: rate
independent deviatoric plasticity’. In Nemat Nasser S., Asaro R.J., and Hegemier
G.A. (eds), Theoretical Foundation for LargeScale Computations of Nonlinear
Material Behaviour, Martinus Nijhoff Publication, The Netherlands, 29–57.
Krieg, R.D. and Key, S.W. (1976). ‘Implementation of a time dependent plas
ticity theory into structural computer programs’, In Constitutive Equations in
Viscoplasticity: Computational and Engineering Aspects (eds) Stricklin J.A. and
Saczalski K.J.), ASME.
6. Implementation of plasticity
models into ﬁnite element code
6.1 Introduction
Several commercial ﬁnite element software packages (e.g. ABAQUS, LSDyna, and
MARC) provide the facility for users to specify their own material models. In
ABAQUS, the user is required to provide a Fortran subroutine called a ‘UMAT’. This
chapter is concerned with the implementation of plasticity models into ﬁnite element
code, which is carried out, by example, in using ABAQUS. In particular, we start by
developing an ABAQUS UMAT for elasticity, and discuss the tests necessary to verify
the model implementation into ABAQUS. We go on to consider isotropic hardening
plasticity with explicit and implicit integration, with continuum and consistent tan
gent stiffnesses, large deformation formulations using rotated variables provided by
ABAQUS, and from ﬁrst principles using the deformation gradient. We use the prob
lem of simple shear with elasticity to verify the large deformation implementations.
We then present an implicit implementation for elastoviscoplasticity (and creep).
All the Fortran coding, together with the necessary ABAQUS input ﬁles, are available
through the OUP website.
Finite element code is often modular in structure, whether it be commercial code or
written inhouse. An important module is that relating to material behaviour; in other
words, the constitutive stress response of the material given prescribed conditions of
deformation. In ABAQUS, but in a similar way for all codes, a range of information
is passed into the material module relating to both the beginning and end of a time
increment. In particular, stress, strain, and deformation gradient are provided at the
beginning of the time increment. Strain and the deformation gradient are also provided
at the end of the increment. Within the module, it is then necessary to execute three
tasks. First, the stresses at the end of the time increment must be determined and,
second, for the case of an implicit analysis (i.e. the ﬁnite element momentum balance
or equilibriumequations are solved implicitly) using ABAQUS standard, for example,
the material Jacobian, or tangent stiffness, must also be provided. Third, any state
variables (such as the isotropic hardening variable or effective plastic strain) must
170 Implementation of plasticity models
be updated to the end of the time increment. In fact, in coding an ABAQUS UMAT,
a large range of information is provided, some of which will be referred to later.
However, the job of the UMAT is clear: to update the stresses and state variables to
the end of the time increment and to provide the Jacobian. We start by addressing
elasticity.
6.2 Elasticity implementation
We discuss elasticity and its implementation into a UMAT for two main reasons.
First, it provides a good introduction to writing and testing a UMAT subroutine
for those who are new to the process. Second, it provides a basis from which elasto
plasticity models may be developed. For the latter reason, while is no way essential
for elasticity, we use an incremental approach in implementing the linear elastic
equations.
In Chapter 5, Hooke’s law was given in an incremental form in Equation (5.21)
together with the material Jacobian, for the case in which there are no out of plane
shears (e.g. plane strainandaxisymmetric problems). Withknowledge of the increment
in strains (provided to the UMAT), together with the speciﬁcation of elastic constants
(we shall take E = 210 GPa and ν = 0.3 throughout), the stress increment is obtained
either from Equation (5.21), or its equivalent written in Voigt notation,
σ = Cε
e
=
⎛
⎜
⎜
⎝
2G + λ λ λ 0
λ 2G + λ λ 0
λ λ 2G + λ 0
0 0 0 G
⎞
⎟
⎟
⎠
⎛
⎜
⎜
⎝
ε
11
ε
22
ε
33
γ
12
⎞
⎟
⎟
⎠
. (6.1)
Because ABAQUS provides most tensor quantities in vector (Voigt) form, it may be
more convenient to use Equation (6.1) than the tensor form given in (5.21). However,
this is not always the case, and later we shall use the tensor form in preference,
particularly where it is necessary to work fromthe deformation gradient. Note that the
shear strain quantities provided by ABAQUS are always engineering shears, that is,
twice the tensorial shear strains. It is also important to check the ordering of shear
quantities, which can vary depending on element type used. Equation (6.1) is suitable
for plane strain and axisymmetric problems, but not for problems of plane stress
or three dimensions, for which different stiffness matrices are required. For linear
elasticity, the material Jacobian is just the elastic stiffness matrix so that speciﬁcation
of the Jacobian in the UMAT is easy. The coding required for the UMAT for linear
elasticity for plane strain, axisymmetric, and threedimensional problems, is available
through the OUP website. A complete list of all the UMAT coding, together with
ABAQUS input ﬁles, is given in Appendix B.
Veriﬁcation of implementations 171
6.3 Veriﬁcation of implementations
The veriﬁcation of the model implementation is vital. For complex material models,
this requires the development of an independent solver (which will often be numerical)
for uniaxial and pure shear problems so that direct comparison with the results obtained
using the UMAT can be made. In addition, it is necessary to test the UMAT using
a single, and where appropriate, multiple elements, for conditions of strain control
and load control under uniaxial and pure shear conditions. If possible, perhaps by
‘switching off’ parts of the model implemented into the UMAT, make comparisons
with a multiaxial problem with nonuniform strain and stress distributions for which
the solution is known (either an independent solution or one obtained using an internal
ABAQUS model). In this chapter, we shall not carry out all of the veriﬁcation tests
described for all the model implementations, but it is certainly advisable to do so for
newmodel implementations into UMATs. Because of their importance, we go through
some of the possible veriﬁcation tests step by step.
1. Single and multiple element uniaxial tests. Figure 6.1 shows an axisymmetric
single and fourelement unit square which is subjected to uniaxial displacement or
force control in the zdirection producing uniform, uniaxial stress, σ
zz
, and strain,
ε
zz
, with ε
rz
= σ
rr
= σ
θθ
= σ
rz
= 0, which can be compared with independent
closedform solutions. The fourelement problem is important since it introduces
a ‘free’ node, which does not exist for the single fournoded axisymmetric element.
The force controlled test is important for checking errors in the Jacobian.
2. Single element simple shear test. The uniaxial tests in (1) do not involve the shear
terms at all, so it is important to include a problemwhich tests these terms, particularly
because of the potential for errors with the use of engineering shear strains rather than
their tensorial counterparts. Figure 6.2 shows a plane strain singleelement unit square
under simple shear loading. For small deformations, ε
xx
= ε
yy
= σ
xx
= σ
yy
= 0,
r
u
z
=0.0
Displacement
or force
0.0
0.0
1.0
1.0
z
u
r
=0.0
Displacement
or force
u
z
=0.0
0.0
0.0
1.0
1.0
z
r
u
r
=0.0
Fig. 6.1 Schematic diagram showing an axisymmetric single and fourelement unit square under
uniaxial displacement or force controlled loading.
172 Implementation of plasticity models
g
u
x
=u
y
=0.0
x
0.0
0.0
1.0
1.0
y
Fig. 6.2 Schematic diagram showing a plane strain singleelement unit square under simple shear
loading.
and uniform shear strain, γ
xy
, and stress, σ
xy
, are produced which can be compared
with independent closedform solutions.
3. Nonuniform strain and stress ﬁeld. Often, a comparison test which generates
nonuniform strain and stress ﬁelds will not be possible because a comparison may
well not exist, and a closedform solution is now no longer possible. However, it is
sometimes possible to simplify the implemented plasticity model (e.g. by turning
off the porosity in a Gursontype porous plasticity model) such that a comparison
with another solution (e.g. produced using one of the many builtin models contained
within ABAQUS) is then possible. While this will not test all the features of the
model implemented into the UMAT, it nonetheless may test a good number of them,
and may therefore be worthwhile. Later in the chapter, both an implicit and explicit
implementation of isotropic hardening plasticity are tested in this way by comparing
the results obtained with those produced using the builtin ABAQUS model.
6.4 Isotropic hardening plasticity implementation
In Chapter 5, we presented both explicit and implicit integration of the equations
for linear strain hardening isotropic plasticity, together with the consistent tangent
stiffness for the implicit scheme. We will implement both integration schemes into
ABAQUS UMATs and discuss the advantages and disadvantages of each. We start
with the explicit scheme which was described at the beginning of Section 5.2, and
introduce the continuum Jacobian.
6.4.1 Explicit integration for isotropic hardening plasticity with
continuum Jacobian
We may summarize the implementation as follows. All quantities are assumed to
be given at time, t , that is, at the start of the time increment, unless indicated
Isotropic hardening plasticity implementation 173
otherwise:
(i) Determine the yield function
f = σ
e
− r − σ
y
=
_
3
2
σ
: σ
_
1/2
− r − σ
y
. (6.2)
(ii) Determine if actively yielding
Is f > 0?
(iii) Determine the plastic multiplier
f > 0, dλ =
n · C dε
n · Cn + h
,
f < 0, dλ = 0.
(6.3)
(iv) Determine stress and isotropic hardening increments
dσ = C dε
e
= C(dε − dλn),
dr = hdp = hdλ.
(6.4)
(v) Update all quantities to the end of the time increment, using explicit integration
σ
t +t
= σ + dσ,
ε
p
t +t
= ε
p
+ dε
p
, (6.5)
r
t +t
= r + dr.
(vi) Determine Jacobian.
(vii) End.
We now address the determination of the Jacobian. In Chapter 5, we derived
the consistent tangent stiffness for the implicit integration scheme. For the purposes
of the explicit integration considered here, we introduce what is sometimes referred
to as the continuum Jacobian. That is, it is not derived explicitly on the basis of the
integration scheme, but directly from the constitutive equations. We start from the
stress–strain relationship written in Voigt notation
dσ = C dε
e
= C(dε − dλn), (6.6)
which we may write out fully for conditions of axial symmetry or plane strain as
⎛
⎜
⎜
⎝
dσ
11
dσ
22
dσ
33
dσ
12
⎞
⎟
⎟
⎠
=
⎛
⎜
⎜
⎝
dσ
1
dσ
2
dσ
3
dσ
4
⎞
⎟
⎟
⎠
=
⎛
⎜
⎜
⎝
C
11
C
12
C
13
C
14
C
22
C
23
C
24
C
33
C
34
sym C
44
⎞
⎟
⎟
⎠
⎛
⎜
⎜
⎝
⎛
⎜
⎜
⎝
dε
1
dε
2
dε
3
dε
4
⎞
⎟
⎟
⎠
− dλ
⎛
⎜
⎜
⎝
n
1
n
2
n
3
n
4
⎞
⎟
⎟
⎠
⎞
⎟
⎟
⎠
.
(6.7)
174 Implementation of plasticity models
We may write the Jacobian (here symmetric) as
J =
∂ dσ
∂ dε
=
⎛
⎜
⎜
⎝
J
11
J
12
J
13
J
14
J
22
J
23
J
24
J
33
J
34
sym J
44
⎞
⎟
⎟
⎠
(6.8)
so that, for example, the ﬁrst term is
J
11
=
∂ dσ
1
∂ dε
1
= C
11
−
∂ dλ
∂ dε
1
(C
11
n
1
+ C
12
n
2
+ C
13
n
3
+ C
14
n
4
).
With some algebra we may show, therefore, that
J = C −Cn ⊗
∂ dλ
∂ dε
, (6.9)
where ⊗ is the dyadic product of two vectors, details of which may be found in
Appendix A. In a similar way, by considering the numerator of the plastic multiplier
given in Equation (6.3), we may show that
∂ dλ
∂ dε
=
Cn
n · Cn + h
so that the continuum Jacobian is
J = C −
Cn ⊗Cn
n · Cn + h
. (6.10)
The explicit integration and provision of Jacobian is then complete. An ABAQUS
UMAT containing this formulation, together with various input ﬁles for uniaxial dis
placement and load control, together with a fourpoint beam bending problem, are
available via the OUP website (full details are given in Appendix B). In addition, the
very same problems are analysed using the builtin ABAQUS linear strain hardening
plasticity model (chosen to represent a material with E = 210,000 GPa, ν = 0.3,
σ
y
= 240 MPa, and h = 1206 MPa). In all the analyses using the explicit UMAT,
the maximum time increment allowed in the analysis is carefully chosen to ensure
stability and accuracy. Despite this, at the elastic–plastic transition, the stress at ﬁrst
yield is overestimated because the stress at the start of the increment was determined
on the basis of elastic behaviour in the previous increment. With an explicit scheme,
this is never corrected so that the stresses remain slightly overestimated throughout the
analysis. The error in stress can be reduced by decreasing the time increment size, but
at the cost of the computer CPU time and error accumulation. In any case, many more
time increments are required using the explicit UMAT than are required using the
builtin ABAQUS implicit plasticity integration. This is a further serious shortcoming
of the explicit integration method, which needs to be weighed against the advantage
Isotropic hardening plasticity implementation 175
of simplicity, particularly for complex constitutive equations. However, despite con
siderably longer computer CPU times, the results obtained for the fourpoint bend
simulation from the explicit UMAT and the builtin ABAQUS plasticity model are
found to be nearidentical. We consider an implicit implementation in Section 6.4.2.
6.4.2 Implicit integration for isotropic hardening plasticity with
consistent Jacobian
We may summarize the implementation as follows. As opposed to the explicit case,
all quantities are now assumed to be given at the end of the time increment, that is,
at time t + t , unless otherwise indicated.
(i) Determine the elastic trial stress
σ
tr
= σ
t
+ 2Gε + λIε : I. (6.11)
(ii) Determine the trial yield function
f = σ
tr
e
− r − σ
y
=
_
3
2
σ
tr
: σ
tr
_
1/2
− r − σ
y
. (6.12)
(iii) Determine if actively yielding
Is f > 0?
(iv) If yes, use Newton iteration to determine the effective plastic strain increment
r
(k)
= r
t
+ hp,
dp =
σ
tr
e
− 3Gp
(k)
− r
(k)
− σ
y
3G + h
, (6.13)
p
(k+1)
= p
(k)
+ dp.
Otherwise,
p = 0.
(v) Determine plastic and elastic strain and stress increments
ε
p
=
3
2
p
σ
tr
σ
tr
e
,
ε
e
= ε − ε
p
, (6.14)
σ = 2Gε
e
+ λIε
e
: I.
(vi) Update all quantities to the end of the time increment
σ = σ
t
+ σ,
p = p
t
+ p.
(6.15)
176 Implementation of plasticity models
(vii) Determine consistent Jacobian
δσ = 2GQ
σ
tr
σ
tr
e
σ
tr
σ
tr
e
: δε + 2GRδε +
_
K −
2
3
GR
_
II : δε. (6.16)
(viii) End.
An ABAQUS UMAT containing this formulation, together with various input ﬁles
for uniaxial displacement and load control, together with a fourpoint beam bending
problem, are available through the OUP website (full details are given in Appendix B).
Inaddition, as before, the verysame problems are analysedusingthe builtinABAQUS
linear strain hardening plasticity model. Using implicit integration with the consistent
tangent stiffness eliminates the problemwhich occurred at the elastic–plastic transition
when using explicit integration. In addition, the use of implicit integration enables
much larger time increments to be used, therefore signiﬁcantly reducing CPU times.
The disadvantage of the implicit formulation is simply the difﬁculty in obtaining the
consistent tangent stiffness for complex constitutive equations. Often, analytical forms
are simply not obtainable, in which case a numerical procedure may be possible. The
results obtained for the fourpoint bend simulation from the implicit UMAT and the
builtin ABAQUS plasticity model are found to be identical.
6.5 Large deformation implementations
We have not yet differentiated between small and large deformation implementations
in this chapter. In fact, the UMATs discussed above for both explicit and implicit
schemes are suitable for both small and large deformation problems using ABAQUS.
This is because the necessary rigid body rotations for the strains and stresses have
already been carried out by ABAQUS before they are provided to the UMAT routine.
That is, referring back to Section 5.6, the large deformation stress update necessary is
σ =
∇
σ + (Wσ
t
−σ
t
W)t =
∇
σ + Rσ
t
R
T
(6.17)
in which
∇
σ is the corotational stress increment. The stress, σ
t
, at the start of the
time increment (and the strain) has already been rotated by ABAQUS (i.e. the stress
provided at the start of the time increment is effectively Rσ
t
R
T
) so that all we need
to do within the UMAT is to carry out the stress update. Sometimes, depending on
the plasticity model employed, it is necessary to use internal variables which are also
tensor quantities. An example is the back stress in a kinematic hardening model. The
components of the back stress will need to be updated within the UMAT ( just like
the scalar isotropic hardening variable in the previous sections) and stored in what
are called state variable arrays in ABAQUS. Because state variables are not modiﬁed
Large deformation implementations 177
by ABAQUS, it is necessary for the user to carry out the rigid body rotations on
tensorial state variables. In fact, ABAQUS provides a utility subroutine to simplify
this process. If, for example, the tensorial variable recovered from the state variable
array at the start of the increment is, say, x
t
, then the user must carry out the rigid body
rotation Rx
t
R
T
before updating x to the end of the time increment within the UMAT.
The rotation is carried out by a single call of the utility subroutine rotsig detailed in
the ABAQUS manuals.
Sometimes, constitutive equations for elasticity and plasticity are formulated in
terms of the deformation gradient, F. Examples include hyperelasticity—large strain
nonlinear elasticity and crystal plasticity. It may be, however, that the user simply
wishes to work from the deformation gradient rather than use the strain and stress
quantities provided by ABAQUSto the UMATsubroutine. In Section 6.5.1, we present
a large deformation implementation based on an explicit scheme.
6.5.1 Implementation using the deformation gradient
We may summarize the implementation as follows. In this explicit approach, all
quantities are assumed to be given at time, t , that is, at the start of the time increment,
unless indicated otherwise.
(i) Determine the velocity gradient
L =
˙
FF
−1
. (6.18)
(ii) Determine the rate of deformation and spin
D =
1
2
(L +L
T
), W =
1
2
(L −L
T
). (6.19)
(iii) Determine the yield function
f = σ
e
− r − σ
y
=
_
3
2
σ
: σ
_
1/2
− r − σ
y
. (6.20)
(iv) Determine if actively yielding
Is f > 0? (6.21)
(v) Determine rate of plastic deformation
f > 0, D
p
= as speciﬁed by constitutive equation,
f < 0, D
p
= 0.
(6.22)
178 Implementation of plasticity models
(vi) Determine rate of elastic deformation and Jaumann stress rate and isotropic
hardening rate
D
e
= D −D
p
,
∇
σ = 2GD
e
+ λID
e
: I, (6.23)
˙ r = h˙ p.
(vii) Determine stresses with respect to material reference
˙ σ =
∇
σ +Wσ −σW. (6.24)
(viii) Update all quantities to the end of the time increment, using explicit integration
σ
t +t
= σ + ˙ σt,
r
t +t
= r + ˙ rt.
(6.25)
(ix) Determine Jacobian.
(x) End.
We will next address the veriﬁcation of a large deformation implementation, which
is valid for both the implicit and explicit implementations in Sections 6.4.1 and 6.4.2.
Note that uniaxial displacement or force controlled loading will not test whether
the rigid body rotations are being calculated correctly, since for these cases, the
continuum spin, W, is zero. A good test, however, is provided in the form of
simple shear if we allow the strains to become quite large. To simplify matters,
we will ‘switch off’ the plasticity and allow elasticity only, since the test is being
carried out to check the rigid body rotation calculation rather than the constitutive
response.
We obtained the deformation gradient for simple shear in Section 3.5.1 as
F =
_
1 δ
0 1
_
.
Considering a constant rate of shearing,
˙
δ, the velocity gradient is
L =
_
0
˙
δ
0 0
_
so that the rate of deformation and spin are
D =
1
2
_
0
˙
δ
˙
δ 0
_
, W =
1
2
_
0
˙
δ
−
˙
δ 0
_
.
Large deformation implementations 179
Note that the spin is nonzero so that rigid body rotation, together with stretch, is
occurring.
The Jaumann stress rate is given by
∇
σ = 2GD + λID : I = G
_
0
˙
δ
˙
δ 0
_
(6.26)
and the stress rate with respect to the material (undeformed) reference is
˙ σ =
_
˙ σ
xx
˙ σ
xy
˙ σ
xy
˙ σ
yy
_
, (6.27)
which is given in terms of the Jaumann stress rate by
˙ σ =
∇
σ +Wσ −σW
so that substituting for the spin and (6.26) and (6.27) gives
_
˙ σ
xx
˙ σ
xy
˙ σ
xy
˙ σ
yy
_
= G
˙
δ
_
0 1
1 0
_
+
1
2
˙
δ
_
σ
xy
σ
yy
−σ
xx
−σ
xy
_
−
1
2
δ
_
−σ
xy
σ
xx
−σ
yy
σ
xy
_
so that
˙ σ
xy
= G
˙
δ +
1
2
˙
δ(σ
yy
− σ
xx
), (6.28)
˙ σ
xx
=
˙
δσ
xy
, (6.29)
˙ σ
yy
= −
˙
δσ
xy
. (6.30)
Equations (6.29) and (6.30) give, with the initial condition that all stresses are zero,
σ
yy
= −σ
xx
.
Differentiating (6.28) and substituting for (6.29) and (6.30) gives
¨ σ
xy
+
˙
δ
2
σ
xy
= 0,
which has solution
σ
xy
= Asin
˙
δt + B cos
˙
δt.
The initial conditions require that B = 0. Solving for σ
xx
and σ
yy
using
Equations (6.29) and (6.30) and imposing the initial conditions gives the solution
σ
xy
= Gsin
˙
δt, σ
xx
= −σ
yy
= G(1 − cos
˙
δt ). (6.31)
180 Implementation of plasticity models
–200
–150
–100
–50
0
50
100
150
0.00 0.20 0.40 0.60 0.80 1.00
s
yy
s
xy
x
y
Time (s)
S
t
r
e
s
s
(
G
P
a
)
s
xx
Fig. 6.3 Unit square under simple shear at a rate of 5.0 s
−1
and the corresponding stresses.
For small strain, δ, at constant strain rate, these reduce to
σ
xy
= Gδ and σ
xx
= −σ
yy
= 0. (6.32)
The harmonic variation in (6.31) results from the large deformation, and in particu
lar, the rigid body rotation taking place. This nonphysical result arises because we
are using a small strain, linear elasticity model under conditions of large deformation.
Despite its nonphysicality, it provides a good test of the calculation of the rigid body
rotations in the large deformation UMAT. In order to do this, a single plane strain
element, as shown in Fig. 6.2, has been subjected to simple shear to large strain.
This has been carried out using the UMAT with elasticity described in Section 6.2,
which uses ABAQUSprovided stresses and strains, a further UMAT based on the
deformation gradient, described in this section, and using the builtin elasticity model
in ABAQUS. The results obtained are identical and are shown in Fig. 6.3. The various
UMATs, together with the ABAQUS input ﬁles, are detailed in Appendix B and are
available via the OUP website.
6.6 Elastoviscoplasticity implementation
Viscoplasticity, meaning ratedependent plasticity in which the plastic multiplier is
determined through the use of a viscoplastic constitutive equation as opposed to the
use of the consistency condition, was introduced in Chapter 4. The radial return,
implicit backward Euler integration for viscoplasticity was discussed in Chapter 5.
Here, we present an implicit implementation for linear isotropic strain hardening
elastoviscoplasticity. Such an implementation can readily be simpliﬁed for the
implicit analysis of creep. We employ a sinhtype viscoplastic constitutive equation
and for simplicity, use the initial tangent stiffness (i.e. the elastic stiffness) for the
material Jacobian.
The viscoplastic constitutive equation is taken to be
˙ p = φ(σ
e
, r) = α sinh β(σ
e
− r − σ
y
)
Elastoviscoplasticity implementation 181
and the multiaxial plastic strain increments are given by
ε
p
= pn =
3
2
p
σ
σ
e
.
We determine the increment in effective plastic strain as described in Chapter 5 as
follows. Note that all quantities are now assumed to be given at the end of the time
increment, that is, at time t + t , unless otherwise indicated.
(i) Determine the elastic trial stress
σ
tr
= σ
t
+ 2Gε + λIε : I. (6.33)
(ii) Determine the trial yield function
f = σ
tr
e
− r − σ
y
=
_
3
2
σ
tr
: σ
tr
_
1/2
− r − σ
y
. (6.34)
(iii) Determine if actively yielding
Is f > 0?
(iv) If yes, use Newton iteration to determine the effective plastic strain increment
φ(σ
e
, r) = α sinh β(σ
tr
e
− 3Gp − r − σ
y
),
φ
p
= −3Gαβ cosh β(σ
tr
e
− 3Gp − r − σ
y
),
φ
r
= −αβ cosh β(σ
tr
e
− 3Gp − r − σ
y
),
r = r
t
+ hp,
dp =
φ(p, r) − (p/t )
(1/t ) − φ
p
− hφ
r
,
p
(k+1)
= p
(k)
+ dp.
(6.35)
(v) Determine plastic and elastic strain and stress increments
ε
p
=
3
2
p
σ
tr
σ
tr
e
,
ε
e
= ε − ε
p
, (6.36)
σ = 2Gε
e
+ λIε
e
: I.
182 Implementation of plasticity models
(vi) Update all quantities to the end of the time increment
σ = σ
t
+ σ,
p = p
t
+ p.
(6.37)
(vii) Determine Jacobian.
(viii) End.
An ABAQUS UMAT containing this formulation, together with various input ﬁles
for uniaxial displacement and load control, together with a fourpoint beam bending
problem, are available via the OUP website (full details are given in Appendix B).
Inaddition, uniaxial, closedformimplicit andexplicit solutions are providedinFortran
programs for veriﬁcation of the implementation.
Part II. Plasticity models
This page intentionally left blank
7. Superplasticity
7.1 Introduction
Superplasticity is the ability of some materials to undergo very large, irreversible,
tensile elongations without necking and failing. Generally, a very ﬁne grain structure
is required (a typical grain size will be of the order of 1 µm) and a deformation temper
ature of about 0.5 T
m
is necessary to enable the appropriate superplastic deformation
mechanisms to operate. Two common superplastically formed classes of material are
aluminium and titanium alloys, which are used extensively in the aerospace and aero
engine industry. Superplastic forming exploits the ability of the material to undergo
very large tensile elongations in metal sheet stretch forming and blow moulding
processes, and has many advantages for the manufacture of complex shapes in sheet
metal using simple low pressure pneumatic forming equipment.
In this chapter, we shall introduce superplasticity and its characteristics, constitutive
equations for superplastic deformation which are coupled with the accompanying
microstructural evolution, the multiaxial form of the equations, and an industrial
application.
7.2 Some properties of superplastic alloys
Superplasticity is very much a viscoplastic process in the sense that the stress
response is highly strainrate dependent. The uniaxial response to variable constant
true straincontrolled loading is simpliﬁed and schematically shown in Fig. 7.1.
There is often a very strong strainrate sensitivity, and in addition, strain hardening is
seentooccur. This results froma number of possible causes includingthe effect of grain
growth which occurs in superplasticity, and from dislocation hardening processes. If
we choose a particular strain in Fig. 7.1(b) and pick off the corresponding stress for
each of the stress–strain curves at all the strain rates, we may then plot log(stress)
versus log(strain rate), which often produces a curve of the form shown in Fig. 7.2.
Region(i) corresponds toverylowstrainrates inwhichdiffusionprocesses dominate
when the temperature is higher than ∼0.5 T
m
. Region (iii) corresponds to very high
186 Superplasticity
s
3
s
2
s
1
s
y
Strain
S
t
r
a
i
n
Time
S
t
r
e
s
s
(a) (b) «
3
•
«
3
•
«
2
•
«
1
•
«
2
•
«
1
•
Fig. 7.1 Schematic showing simpliﬁed superplastic stress response which is dependent on strain rate.
(ii)
l
n
(
s
t
r
e
s
s
)
ln(strain rate)
Superplastic
region
(i) (iii)
Fig. 7.2 log(stress) versus log(strain) showing the region usually considered to be superplastic.
strain rate at which diffusion is largely inhibited so that deformation occurs through
dislocation motion, some of which is perhaps thermally activated, but diminishing
with increasing strain rate. Region (ii) is that in which superplasticity takes place;
that is, large tensile strains (∼1–2) are achievable which are many times larger than
those obtainable in regions (i) and (iii).
Because of the approximate linearity in region (ii), the ln(stress)–ln(strain rate)
relationship can be written as
ln(σ) = mln(˙ ε) + k (7.1)
in which m and k are constants. Rearranging (7.1) gives
σ = K˙ ε
m
(7.2)
in which K is a further constant. In Equation (7.2), mis called the strainrate sensitivity
and is, of course, the gradient of the ln(stress) versus ln(strain rate) curve. In fact, the
higher the value of m, the better the superplastic deformation, and the larger the
tensile elongations achievable in the absence of necking and failure. Generally, for
what would be described as superplastic deformation, m > 0.35. If, from Fig. 7.2,
we now plot the strainrate sensitivity versus ln(strain rate), we will obtain the graph
shown schematically in Fig. 7.3.
Some properties of superplastic alloys 187
0.35
Superplastic
region
m
ln(strain rate)
Fig. 7.3 Schematic representation of strainrate sensitivity versus log(strain rate).
L+dL
L
P
P
Uniform region with
Area A
s
«
Stress
Strain rate
•
Necked region with
Area A+dA
Stress s+ds
Strain rate «+ds
•
Fig. 7.4 Schematic diagram showing a uniform, circular test piece with crosssectional area A under
load P before and after the introduction of a neck.
We shall now examine further the signiﬁcance of the strainrate sensitivity, m,
in superplasticity, by considering a uniform, uniaxial circular test piece containing
a single, idealized neck shown schematically in Fig. 7.4. We employ the incompress
ibility condition, apply constant load, P, to the test piece during necking and assume
the stress–strain rate relationship given in (7.2).
The incompressibility condition gives
AL = (A + dA)(L + dL) ≈ AL + LdA + AdL
188 Superplasticity
so that
dA
A
= −
dL
L
= −dε = −˙ ε dt
and
dA
dt
= −˙ εA (7.3)
or, in the neck,
d(A + dA)
dt
= −(˙ ε + d˙ ε)(A + dA),
which, when combined with (7.3) gives
d
dt
(dA) ≈ −(Ad˙ ε + ˙ ε dA) = −dA˙ ε
_
A
dA
d˙ ε
˙ ε
+ 1
_
. (7.4)
The constancy of load, P, gives
σA = (σ + dσ)(A + dA)
so that
dσ
σ
= −
dA
A
. (7.5)
The constitutive equation σ = K˙ ε
m
gives
σ + dσ = K(˙ ε + d˙ ε)
m
= K˙ ε
m
_
1 +
d˙ ε
˙ ε
_
m
= K˙ ε
m
_
1 + m
d˙ ε
˙ ε
+ · · ·
_
so that
dσ
σ
≈ m
d˙ ε
˙ ε
. (7.6)
Combining (7.6) and (7.5) gives
A
dA
d˙ ε
˙ ε
= −
1
m
and substituting into (7.4) gives
d
dt
(dA) ≈ −dA˙ ε
_
1 −
1
m
_
. (7.7)
This equation tells us that the rate of development of the neck depends upon the
quantity 1 − (1/m). As the strainrate sensitivity increases, and approaches unity,
the rate of necking decreases to zero. We therefore see the signiﬁcance of the strain
rate sensitivity in superplasticity; the higher the value of m, the more necking can
be inhibited therefore allowing greater elongations prior to the onset of necking and
failure. A much more complete introduction to superplasticity can be found in Pilling
and Ridley (1989).
Constitutive equations for superplasticity 189
7.3 Constitutive equations for superplasticity
The constitutive equation in (7.2) is simple, but is often inadequate for simulat
ing superplasticity processes because the ln(stress)–ln(strain rate) relationship is not
often linear, particularly over strainrate regimes which occur in practical processing.
In addition, it says nothing about the inﬂuence of changing microstructure during
superplastic deformation, which is known to be important (Ghosh and Hamilton,
1979; Ghosh and Raj, 1981; Hamilton, 1984; Zhou and Dunne, 1996). We present
here constitutive equations for superplasticity for the particular commercially import
ant titanium alloy, Ti–6Al–4V. This alloy ﬁnds application in a range of aeroengine
components, but for component manufacture using superplastic forming, the engine
fan blades are perhaps the most important.
Uniaxial stress–strain curves for the Ti–6Al–4V alloy undergoing superplastic
deformation at 927
◦
C are shown in Fig. 7.5.
A strong strainrate effect is seen together with signiﬁcant strain hardening. The
stresses required to cause the deformation are, however, really quite small but the
tensile strains achieved during the superplastic deformation are large. The hardening
is due to several processes; perhaps the most important is the increasing average grain
size. The average grain size at the start of all the tests shown in Fig. 7.5 is 6.4 µm. At
the end of the deformation, it has increased to about 10 µm, and depends on the rate
of deformation. The evolution of the average grain size for each of the stress–strain
curves shown in Fig. 7.5 is shown in Fig. 7.6.
0.0 0.2 0.4 0.6 0.8 1.0
Strain
0
5
10
15
20
25
S
t
r
e
s
s
(
M
P
a
)
1.0×10
–3
s
–1
2.0×10
–4
s
–1
5.0×10
–5
s
–1
Fig. 7.5 Superplastic stress–strain curves for Ti–6Al–4V at 927
◦
C at the strain rates shown. The
initial grain size is 6.4 µm.
190 Superplasticity
0.006
0.008
0.01
0.012
0 100 200 300 400
G
r
a
i
n
s
i
z
e
(
m
m
)
Time (min)
1.0 ×10
–3
s
–1
2.0×10
–4
s
–1
5.0×10
–5
s
–1
0.0 s
–1
Fig. 7.6 Grain size versus time curves for Ti–6Al–4V at 927
◦
C at the strain rates shown. The initial
grain size is 6.4 µm.
An important feature is that for a strain rate of 0.0 s
−1
, that is, a static test
in which no deformation takes place, grain growth still takes place; this is called
static grain growth. As the strain rate is increased, so the rate of growth of grain
size increases. That part of the grain growth resulting from the straining is often
referred to as deformationenhanced grain growth. Normal grain growth (Shewmon,
1969) gives rise to a kinetic equation for static grain growth of the form
˙
l =
α
1
l
γ
(7.8)
in which l is the average grain size, and α
1
and γ are material constants. The
deformationenhanced grain growth is accounted for with an additional term so that
the complete kinetic equation can be written as
˙
l =
α
1
l
γ
+ β
1
˙ p (7.9)
in which β
1
is a further material constant and ˙ p is, as before, the effective plastic strain
rate. The symbols in Fig. 7.5 are, in fact, experimental data and the lines result from
ﬁtting Equation (7.9) to the data. The material constants for this particular temperature
of 927
◦
Care given in Table 7.1 and assume grain size to be speciﬁed in millimetres. We
now return to the microstructure–deformation coupling (Zhou and Dunne, 1996; Kim
and Dunne, 1999) and employ the following elasticviscoplastic constitutive equations
˙ ε
p
=
α
l
µ
sinh β(σ − r − σ
y
), (7.10)
˙ r = (c
1
− γ
1
r) ˙ p, (7.11)
˙ σ = E(˙ ε − ˙ ε
p
), (7.12)
Constitutive equations for superplasticity 191
Table 7.1 Material constants for Ti–6Al–4V at 927
◦
C.
α β µ σ
y
(MPa) E (MPa)
0.437 × 10
−5
0.0919 1.06 0.5 1000
c
1
γ
1
α
1
β
1
γ
8.397 0.666 0.128 × 10
−16
0.9625 × 10
−13
5.0
1.0
0.6
0.2
10
–6
10
–5
10
–4
10
–3
10
–2
10
–1
Strain rate (s
–1
)
1
10
100
S
t
r
e
s
s
(
M
P
a
)
Fig. 7.7 Experimental (symbols) and predicted (lines) stress versus strain rate curves for Ti–6Al–4V
at 927
◦
C with an initial grain size of 6.4 µm for the strain levels shown.
which are coupled with the grain growth equation in (7.9). Equation (7.10) is the
viscoplastic constitutive equation in which, as before, α and β are material constants,
l is the current grain size, and µ the deformation–microstructure coupling constant.
The hardening seen in Fig. 7.4 does not all result from the grain growth, although
a substantial part of it does. A further isotropic hardening term, r, has therefore been
introduced with evolution equation given in (7.11) in which c
1
and γ
1
are mater
ial constants. Equation (7.12) is Hooke’s law. Equations (7.9)–(7.12) constitute the
uniaxial material model for superplasticity. With the material constants for the grain
growth kinetic equation having been already determined, the unknown constants in
Equations (7.10) and (7.11) are obtained by ﬁtting the equations (Zhou and Dunne,
1996) to the experimental data in Fig. 7.5. The resulting computed stress–strain curves
are given by the solid lines in the ﬁgure.
Comparisons of predicted and experimental log(stress) versus log(strain rate)
behaviour for the Tialloy with an initial grain size of 6.4 µm obtained at the strain
levels shown are given in Fig. 7.7, and the corresponding predicted variation of
strainrate sensitivity with strain rate (at a strain of 1.0) is shown in Fig. 7.8.
192 Superplasticity
0.0
0.2
0.4
0.6
0.8
1.0
1E–06 1E–05 1E–04 1E–03 1E–02 1E–01
Initial grain size
6.4 mm
9 mm
11.5 mm
Strain rate (s
–1
)
S
t
r
a
i
n

r
a
t
e
s
e
n
s
i
t
i
v
i
t
y
Fig. 7.8 Predicted variation of strainrate sensitivity with strain rate for the initial grain sizes shown
for Ti–6Al–4V at 927
◦
C.
7.4 Multiaxial constitutive equations and applications
We may write the multiaxial equations within the framework of viscoplasticity by
assuming the normality hypothesis and von Mises material behaviour. The multiaxial
viscoplastic strain rate, given these assumptions, is
D
p
=
3
2
˙ p
σ
σ
e
, (7.13)
where
˙ p =
α
l
µ
sinh β(σ
e
− r − σ
y
) (7.14)
and Equations (7.9) and (7.11) remain unchanged. If considering large deformations
(which is normally the case for superplasticity), the Jaumann stress rate is, as before,
∇
σ = ˙ σ +Wσ −σW. (7.15)
These equations, together with (7.9) and (7.11), have been implemented into ABAQUS
by means of the CREEP routine, which is described in Chapter 8. This facilitates
an easier implicit implementation, but a UMAT implementation would be carried
out as described for viscoplasticity in Chapters 5 and 6, although if a plane stress
implementation is required, additional problems would need to be addressed. The
application we consider is the superplastic blowforming of a rectangularsection box
made from 1.25 mm thick Ti–6Al–4V sheet and processed at 900
◦
C (Lin and Dunne,
2001, with thanks to Dr. Lin). The model is shown in Fig. 7.9 in which just one quarter
Multiaxial constitutive equations and applications 193
Die surface
normal
Rigid die
surface
Material mesh
Fig. 7.9 Finite element model for the rectangularsection box die surface and the material blank.
is included, for reasons of symmetry, and both the initially ﬂat sheet and the die surface
are included.
The superplastic forming process consists of clamping the ﬂat Tialloy sheet (mod
elled using shell elements) against the die, the surface of which forms a cavity in
the shape required. Gas pressure is applied to the top face of the sheet, forcing it to
acquire the die shape. In these analyses, the maximum strain rate over the deforming
sheet is controlled to be close to the optimum deforming rate of the material; that
is, the strain rate required to give the highest strainrate sensitivity obtained from the
equivalent of Fig. 7.7, but for 900
◦
C. This is achieved by varying the applied gas pres
sure. The process is considered completed at a forming time, t
f
, when all nodes on the
deforming sheet are in contact with the die. Figure 7.10 shows the deformation of the
superplastic metal sheet at three stages during the forming process at times t /t
f
= 0.1,
t /t
f
= 0.6, and t /t
f
= 1.0. The contours show the magnitude of the effective plastic
strain rate. The maximum target strain rate speciﬁed for the analysis is 1.0×10
−5
s
−1
which is achieved by application of uniformly distributed gas pressure. Figure 7.10
shows that the distribution of effective plastic strain rate is highly nonuniform. This is
typical of many practical superplastic forming processes; while ideally it is necessary
to deform superplastically all points in the material at the same optimum strain rate
(to maximize strainrate sensitivity and hence elongation), this is rarely achievable in
practice. It is possible, however, to ensure that a target maximum strain rate is not
exceeded, and in this analysis, the gas pressure is varied to ensure a nominal maximum
strain rate of 1.0 × 10
−5
s
−1
. The variation of the maximum strain rate during the
194 Superplasticity
(a)
(b)
(c)
+8.00E–06
+7.00E–06
+6.00E–06
+5.00E–06
+4.00E–06
+3.00E–06
+2.00E–06
+0.00E–00
Strain rate
Strain rate
+8.00E–06
+7.00E–06
+6.00E–06
+5.00E–06
+4.00E–06
+3.00E–06
+2.00E–06
+0.00E–00
+8.00E–06
+7.00E–06
+6.00E–06
+5.00E–06
+4.00E–06
+3.00E–06
+2.00E–06
+0.00E–00
Strain rate
Fig. 7.10 The simulated superplastically deforming sheet showing the effective plastic strain rate at
fractional processing times of t /t
f
= 0.1, t /t
f
= 0.6, and t /t
f
= 1.0. (See also Plate 1.)
forming process is shown in Fig. 7.11 for target strain rates of both 1.0 × 10
−5
s
−1
and 1.0 ×10
−4
s
−1
, and the corresponding variations in gas pressure to achieve these
are given in Fig. 7.12. Higher gas pressure is needed for the higher target strain rate
because of the higher ﬂow stresses required. The gas pressure increases during the
Multiaxial constitutive equations and applications 195
1E–05
1E–04
0.0 0.2 0.4 0.6 0.8 1.0
t /t
f
M
a
x
i
m
u
m
s
t
r
a
i
n
r
a
t
e
(
s
–
1
)
1×10
– 4
s
–1
1×10
– 5
s
–1
1E–03
1E–06
Fig. 7.11 Variation of the maximum strain rate over the deforming material blank during the
superplastic forming process for the two target strain rates ˙ ε = 1 × 10
−5
and ˙ ε = 1 × 10
−4
s
−1
.
t /t
f
1×10
– 4
s
–1
1×10
– 5
s
–1
0
100
200
300
400
0.0 0.2 0.5 0.8 1.0
G
a
s
p
r
e
s
s
u
r
e
(
K
P
a
)
Fig. 7.12 Variation of gas pressure to ensure a maximum strain rate in the deforming sheet of
1.0 × 10
−5
and 1.0 × 10
−4
s
−1
.
superplastic forming because of increasing geometrical and frictional constraint and,
in addition, because the material hardens during deformation which is due to the grain
growth taking place. High gas pressure is required to ﬁll the corner part of the die,
which is the last stage of the forming process.
In Fig. 7.10, it is apparent that the last part of the sheet to be formed is the corner,
which is also the area of maximum thinning and where tearing is most likely to occur.
This is made clear by looking at the throughthickness strain ﬁelds which are shown
for the two target strain rates of 1.0×10
−5
and 1.0×10
−4
s
−1
in Fig. 7.13(a) and (b),
respectively.
196 Superplasticity
(a)
(b)
Thickness strain
– 1.20E+00
– 1.07E+00
– 9.15E–01
– 7.60E–01
– 6.05E–01
– 4.50E–01
– 2.95E–01
– 1.40E–01
Thickness strain
–1.20E+00
–1.07E+00
–9.15E–01
–7.60E–01
–6.05E–01
–4.50E–01
–2.95E–01
–1.40E–01
Fig. 7.13 Throughthickness strain ﬁelds at the end of the superplastic forming carried out with target
maximum strain rates of (a) 1.0 × 10
−5
and (b) 1.0 × 10
−4
s
−1
. (See also Plate 2.)
The lower strain rate is seen to lead to a higher spatial variation in through
thickness strain; that is, thinning, which is often not desirable in practical processing.
In fact, the ranges of outofplane strain for the low and high strain rates are 1.06
and 0.84, respectively. The reason for this becomes clear on looking at the grain size
distributions, shown in Fig. 7.14.
At the lower target strain rate, the average grain size increase is larger than that at
the higher strain rate because of static grain growth, which is inhibited in the latter.
In addition, at the lower strain rate, the range of ﬁnal grain size is less than that at
the higher rate. For the case of the higher strain rate, therefore, the larger grain size
in the deforming material in the corner (relative to that at the boundaries) leads to
a higher stress in the corner region to generate the same strains seen for the lower
strain rate. This inhibits straining in the corner region producing larger strains in the
boundary regions. The net result is a more uniform strain distribution for the high
strain rate, and a correspondingly more uniform thinning. This is generally preferable
References 197
Grain sizes
Grain sizes
+1.14E–02
+1.15E–02
+1.16E–02
+1.16E–02
+1.17E–02
+1.18E–02
+1.18E–02
+1.19E–02
(a)
(b)
+7.93E– 03
+8.06E– 03
+8.20E– 03
+8.33E– 03
+8.60E– 03
+8.74E– 03
+8.88E– 03
+9.01E– 03
Fig. 7.14 Average grain size ﬁelds at the end of the superplastic forming carried out with target
maximum strain rates of (a) 1.0 × 10
−5
and (b) 1.0 × 10
−4
s
−1
. (See also Plate 3.)
in practice, and results directly from the interactions of the superplastic deformation
and microstructural evolution. These industrially important effects could not have
been obtained without microstructurally based constitutive equations. An additional
important effect is the distribution of grain size in commercial Ti–6Al–4V, which has
not been addressed here. However, modelling techniques to incorporate it, its effects
on strainrate sensitivity and hence on necking and failure have been addressed by
Kim and Dunne (1999), where further information may be found.
References
Ghosh, A.K. and Hamilton, C.H. (1979). ‘Mechanical behaviour and hardening
characteristics of a superplastic Ti–6Al–4V alloy’. Metallurgical Transactions, 10,
699–706.
Ghosh, A.K. and Raj, R. (1981). ‘Grain size distribution effect in superplasticity’.
Acta Metallurgica, 29, 607–616.
198 Superplasticity
Hamilton, C.H. (1984). ‘Superplasticity in titanium alloys’. In Agrawal, S.P. (ed.),
Superplastic Forming, Proceedings of the Symposium, California, March 13–22,
American Society for Metals.
Kim, T.W. and Dunne, F.P.E. (1999). ‘Modelling heterogeneous microstructures in
superplasticity’. Proceedings of the Royal Society, 455, 701–718.
Lin, J. and Dunne, F.P.E. (2001). ‘Modelling grain growth evolution and necking
in superplastic blowforming’. International Journal of Mechanical Science, 43,
595–609.
Pilling, J. and Ridley, N. (1989). Superplasticity in Crystalline Solids. The Institute
of Metals, London.
Shewmon, P.G. (1969). Transformations in Metals. McGrawHill, New York.
Zhou, M. and Dunne, F.P.E. (1996). ‘Mechanismsbased constitutive equations for the
superplastic behaviour of a titanium alloy’. Journal of Strain Analysis, 31, 65–73.
8. Porous plasticity
8.1 Introduction
We have so far considered plasticity and viscoplasticity processes in which it has
been assumed that the deforming material has been incompressible. There are some
materials, however, for which this is not the case. An important example is in the
processing of metal powders in which, generally at elevated temperature, the powder
is consolidated by plastic deformation and the elimination of the porosity. During
the process, signiﬁcant volume changes occur because of the removal of the porosity.
A further example of compressible or porous plasticity is that in which signiﬁcant
voiding develops in a material undergoing plastic deformation resulting froma damage
process (e.g. creep cavitation). Again, while possibly small, volume changes occur
during the deformation and the incompressibility condition no longer holds. A feature
of the constitutive equations in porous plasticity is that there becomes a dependence
on meanstress as we shall see shortly. In this chapter, we introduce the porous plasticity
model of Duva and Crow and outline its implementation into ABAQUS using both
a UMATmaterial subroutine andthe simpler ABAQUSCREEProutine. For simplicity,
within the UMAT, we use explicit integration of the constitutive equations and employ
the initial stiffness as the material Jacobian.
Anumber of constitutive relations for the consolidation of metal powders have been
developed. They have been used to predict the dependence of densiﬁcation rate on
consolidation pressure and temperature as well as volume fraction of voids. One of
the ﬁrst constitutive models for consolidation of metal powder was developed by
WilkinsonandAshby(1975). Theyanalysedthe creepcollapse of a thickwalledspher
ical shell subjected to externally applied hydrostatic loading. In general, consolidation
occurs under the action of a range of stress states that are not purely hydrostatic, for
which the Wilkinson and Ashby model is therefore inappropriate. Subsequent research
has therefore broadened the Wilkinson and Ashby model to more general loading
conditions. The models have taken into account the effects of deviatoric and hydro
static components of stress state by introducing potentials which make possible the
200 Porous plasticity
development of relationships between macroscopic strain rate and stress state (Cocks,
1989; Ponte Castaneda, 1991; Duva andCrow, 1992; Sofronis andMcMeeking, 1992).
The strain rate potential (φ) for a porous material can be written as a function of both
deviatoric and hydrostatic stresses
φ =
˙ ε
0
σ
0
n + 1
_
S
σ
0
_
n+1
, (8.1)
where
S
2
= aσ
2
e
+ bσ
2
m
, (8.2)
σ
2
e
=
3
2
σ
: σ
, (8.3)
σ
m
=
1
3
II : σ. (8.4)
S is an effective effective stress (Duva and Crow, 1994) given in terms of the effective,
σ
e
, and hydrostatic, σ
m
, stresses, n is the creep exponent, and the coefﬁcients a and
b are functions of current relative density, D, which is equivalent to the solid volume
fraction of the porous material, a is associated with the deviatoric component and
b with the hydrostatic component. The coefﬁcients a and b are chosen such that, at the
fully dense stage, that is, D = 1, the coefﬁcient b becomes 0 and a becomes 1. The
effective effective stress, S, is then reduced to σ
e
. The various densiﬁcation models
for monolithic materials in the literature use different approaches to obtain the strain
rate potential. This leads to different expressions for the coefﬁcients a and b (Duva
and Crow, 1994).
Duva and Crow (1994) proposed a strain rate potential for computing densiﬁcation
rates. Based on the strain rate potentials of Cocks (1989) and Ponte Castaneda (1991),
the coefﬁcients a and b were chosen to ensure that the potential gives densiﬁcation
rates identical to those of Wilkinson and Ashby (1975) in the hydrostatic load limit,
and agree with both Cocks (1989) and Ponte Castaneda (1991) in the limit when
the hydrostatic stress vanishes. The densiﬁcation rates predicted by Duva and Crow’s
strainrate potential are consistent withtheir cell model calculations whichwere derived
based on Hill’s minimum principle for velocity. Duva and Crow predict densiﬁcation
rates which compare favourably with the prediction from the Wilkinson and Ashby
model in the hydrostatic limit. The Duva and Crow potential also satisﬁes the lower
bounds derived by both Cocks and Ponte Castaneda in the limit that the hydrostatic
stress vanishes. The densiﬁcation rate predicted fromthe Duva and Crowmodel, in the
presence of a large deviatoric stress component, is closer to the experimental results
than the predictions of Sofronis and McMeeking (1992).
Implementation of porous material constitutive equations 201
8.2 Finite element implementation of the porous material
constitutive equations
Strain rates are determined by differentiating the strain rate potential, φ, to give
˙ ε =
∂φ
∂σ
= AS
n−1
_
3
2
aσ
+
1
3
bIσ
m
_
(8.5)
in which
a =
1 + (2/3)(1 − D)
D
2n/(n+1)
and b =
_
n(1 − D)
(1 − (1 − D)
1/n
)
n
_
2/(n+1)
_
3
2n
_
2
and where A is a material parameter given by
A =
˙ ε
0
σ
n
0
. (8.6)
The dilatation rate can be obtained from
˙ ε
kk
= ˙ ε
xx
+ ˙ ε
yy
+ ˙ ε
zz
(8.7)
and hence, the densiﬁcation rate is given as
˙
D = −D˙ ε
kk
, (8.8)
where D is the relative density.
8.2.1 Implementation into ABAQUS UMAT
The constitutive equations for porous metals developed by Duva and Crow (1992) are
implemented into the ﬁnite element software ABAQUS within a large deformation
formulation using a UMAT subroutine. A simple, explicit, forward Euler integration
is adopted. ABAQUS supplies to the UMAT subroutine the deformation gradient at
the beginning and the end of each time step, F
t
and F
t +δt
. The user is required to
supply the Cauchy stress at the end of the time step. The algorithms, ﬁrst, need to
deﬁne the rate of deformation gradient,
˙
F, which can be calculated, for small time
steps, as
˙
F =
1
δt
(F
t +δt
−F
t
). (8.9)
Then, the velocity gradient, L, is
L = FF
−1
. (8.10)
The total rate of deformation, D, and the spin tensor, W, are given by
D =
1
2
(L +L
T
), (8.11)
W =
1
2
(L −L
T
). (8.12)
202 Porous plasticity
The strain components are calculated independently from
ε = −
1
2
ln(FF
T
)
−1
. (8.13)
The dilatation rate, ˙ ε
kk
, and the densiﬁcation rate,
˙
D, can be calculated from
Equations (8.7) and (8.8), respectively. The relative density at the end of each time
step is determined using the ﬁrstorder Euler integration scheme
D
t +δt
= D
t
+
˙
Dδt. (8.14)
The corotational stress rate
∇
σ is given by
∇
σ =
E
(1 + ν)
D
e
+
Eν
(1 + ν)(1 − 2ν)
II : D
e
, (8.15)
where E is Young’s modulus, ν is Poisson’s ratio, and D
e
is the rate of elastic
deformation given by
D
e
= D −D
p
. (8.16)
D
p
is the rate of plastic deformation as given in Equation (8.5). The stress increment,
with respect to the material reference frame, is calculated as
∇
σ =
∇
σ +Rσ
t
R
t
. (8.17)
This, therefore, provides for an objective update of the stress with respect to a ﬁxed
coordinate system during the time step. The stress increment,
∇
σ, for each time step
can be determined by utilizing the ﬁrstorder Euler integration scheme
σ =
∇
σ t. (8.18)
The updated stress is returned to ABAQUS through the UMAT subroutine. Because
here, we adopt an explicit ﬁrstorder forward Euler integration, great care is neces
sary in choosing an appropriate time step, and in ensuring meaningful results are
obtained.
8.2.2 Implementation into ABAQUS CREEP subroutine
The implementation of the Duva and Crow constitutive equations for consolidation
is also carried out using the ABAQUS CREEP facility. This subroutine is suitable for
constitutive equations in which the increments of inelastic strain are functions of the
hydrostatic stress and the equivalent deviatoric stress described by Mises’ or Hill’s
deﬁnitions, while the UMAT subroutine allows more general forms of constitutive
laws to be implemented, and easier handling of internal state variables.
Implementation of porous material constitutive equations 203
ABAQUS computes the incremental creep strain components as
ε
cr
= ¯ ε
cr
n +
1
3
¯ ε
sw
I, (8.19)
where n is the direction normal to the yield surface, given by
n =
∂σ
e
∂σ
. (8.20)
¯ ε
cr
and ¯ ε
sw
are the equivalent creep strains conjugate to the deviatoric stress and the
mean stress, respectively. Therefore, ¯ ε
cr
corresponds to the conventional deviatoric
creep strain and ¯ ε
sw
to the volumetric strain occurring because of void closure.
In order to implement the porous material constitutive equations using the CREEP
subroutine, the plastic strain rate equation deﬁned in Equation (8.5) has to be
decomposed into two parts as follows
˙ ε
p
= AS
n−1
3
2
aσ
+
1
3
AS
n−1
bσ
m
I. (8.21)
The ﬁrst part is associated with the deviatoric stress and the second with the mean
stress. The user is required to provide ¯ ε
cr
, ∂¯ ε
cr
/∂σ
e
, ∂¯ ε
cr
/∂p, ¯ ε
sw
, ∂¯ ε
sw
/∂σ
e
,
and ∂¯ ε
sw
/∂p, in which σ
e
is the equivalent stress and p is the equivalent pressure
stress given as
p = −
1
3
(σ
xx
+ σ
yy
+ σ
zz
). (8.22)
The variables are obtained as follows:
¯ ε
cr
= AS
n−1
aσ
e
t, (8.23)
∂¯ ε
cr
∂σ
e
= ¯ ε
cr
_
1
σ
e
+
(n − 1)aσ
e
S
2
_
, (8.24)
∂¯ ε
cr
∂p
= −¯ ε
cr
_
(n − 1)bσ
m
S
2
_
, (8.25)
¯ ε
sw
= Abσ
m
S
n−1
t, (8.26)
∂¯ ε
sw
∂σ
e
= ¯ ε
sw
_
(n − 1)aσ
e
S
2
_
, (8.27)
∂¯ ε
sw
∂p
= −¯ ε
sw
_
1
σ
m
+
(n − 1)bσ
m
S
2
_
. (8.28)
In order to calculate σ
e
and σ
m
, the stress components are required. The CREEP
subroutine does not provide these quantities, and therefore the USDFLD subroutine
has to be utilized to access material point data and assign stress components to state
variables, which are then passed into the CREEP subroutine. The densiﬁcation rate
204 Porous plasticity
Table 8.1 Material parameters.
T < 750
◦
C T ≥ 750
◦
C
A
0
8.49 95.67
n 2.18 1.53
0.78
0.83
0.88
0.93
0.98
0 10 20 30 40 50
Time (min)
R
e
l
a
t
i
v
e
d
e
n
s
i
t
y
Predicted using UMAT subroutine
Predicted using CREEP subroutine
Fig. 8.1 Graph showing comparisons of relative density–time curves obtained from the porous
plasticity material model utilizing both an ABAQUS UMAT subroutine and a CREEP subroutine.
can therefore be obtained from Equation (8.8). The relative density for each time
increment can be determined by utilizing the ﬁrstorder Euler integration as given in
Equation (8.14).
The porous plasticity material model implemented using the CREEPsubroutine was
veriﬁed against that using the UMAT subroutine by simulating a simple compression
process using a single plane strain element subjected to inplane compressive load of
20 MPa at 925
◦
C. The material constants, A
0
and n, used are given in Table 8.1, and
A is calculated from
A = A
0
exp
_
−Q
RT
_
(8.29)
in which T is the temperature, R the gas constant, and Q activation energy.
Figure 8.1 shows comparisons of the relative density evolution over time calculated
by the porous material model using the CREEP and the UMAT subroutines. The
results can be seen to be identical.
The Duva and Crow porous plasticity model has been used (empirically) to approx
imate the behaviour in consolidation of continuous (SiC) ﬁbre, (Ti) metal matrix
Application to consolidation of Ti–MMCs 205
Reinforced
area
300mm
Fig. 8.2 Schematic diagram of an aeroengine bling showing region of reinforcement with Ti–MMC
and a demonstrator disc (from King, 1998).
composite materials (Ti–MMCs) being developed for potential application to aero
engine components together with shafts, discs, brakes, and aircraft landing gear.
Application to an aeroengine ‘bling’—a bladed ring—is shown in Fig. 8.2.
The Ti–MMC material in the unconsolidated and fully consolidated state is shown
in Fig. 8.3(a) and (b), respectively.
8.3 Application to consolidation of Ti–MMCs
The simple consolidation model, which was implemented using the CREEP sub
routine as described above, was set up to simulate (empirically) the consolidation
of Ti–MMCs under a range of processing conditions. The consolidation process is
206 Porous plasticity
(a) (b)
Fig. 8.3 Micrographs showing Ti–6Al–4Vmatrix, SiCcontinuous ﬁbre composite material (a) uncon
solidated and (b) consolidated at 925
◦
C for 30 min at 15 MPa.
Die
Ti–MMC
Fig. 8.4 Schematic diagram showing arrays of Ti–MMCs undergoing consolidation.
0.75
0.8
0.85
0.9
0.95
1
0 40 80 120 160 200
Time (min)
R
e
l
a
t
i
v
e
d
e
n
s
i
t
y
Model prediction
Experimental data
Fig. 8.5 Predicted and measured relative density evolution with time for Ti–6Al–4V/SiC composite
consolidated at a constant temperature of 925
◦
C with a pressure of 20 MPa.
References 207
0.75
0.8
0.85
0.9
0.95
1
0 40 80 120 160 200
Time (min)
R
e
l
a
t
i
v
e
d
e
n
s
i
t
y
Model prediction
Experimental data
Fig. 8.6 Predicted and measured relative density evolution with time for Ti–6Al–4V/SiC composite
consolidated at a constant temperature of 700
◦
C with a pressure of 20 MPa.
shown schematically in Fig. 8.4. Generally, there are many more layers of ﬁbres and
the pressure is applied by means of a mechanical punch. The mechanical die ensures
that the Ti–MMCs undergo plane strain compression.
Some of the results of the simulations, which were carried out at a constant tem
perature of 925
◦
C under a pressure of 20 MPa, and at a temperature of 700
◦
C under
a constant pressure of 20 MPa, are shown in Figs 8.5 and 8.6, respectively, showing
the important effect of temperature.
References
Cocks, A.C.F. (1989). ‘Inelastic deformation of porous materials’. Journal of the
Mechanics and Physics of Solids, 37(6), 693.
Duva, J.M. and Crow, P.D. (1992). ‘The densiﬁcation of powders by powerlaw creep
during hot isostatic pressing’. Acta Metallurgica, 40(1), 31.
Duva, J.M. and Crow, P.D. (1994). ‘Analysis of consolidation of reinforced materials
by powerlaw creep’. Mechanics of Materials, 17, 25.
King, J. (1998). ‘Composites take off without a parachute’. Materials World,
5(6), 324.
Ponte Castaneda, P. (1991). ‘The effective mechanical properties of nonlinear isotropic
composites’. Journal of the Mechanics and Physics of Solids, 39(1), 45.
Sofronis, P. and McMeeking, R.M. (1992). ‘Creep of powerlaw material containing
spherical voids’. Journal of Applied Mechanics, Transactions of ASME, 59(2), 88.
Wilkinson, D.S. and Ashby, M.F. (1975). ‘Pressure sintering by power law creep’.
Acta Metallurgica, 23, 1277.
This page intentionally left blank
9. Creep in an aeroengine
combustor material
9.1 Introduction
Aeroengine components operating under inservice conditions are often subjected to
a range of complex cyclic mechanical and thermal loading, leading to combined
creep and cyclic plasticity. The polycrystalline nickelbase superalloy (C263) is
a commercial alloy used for stationary components in aeroengines such as com
bustion chambers, casings, liners, exhaust ducting, and bearing housings. It is a
ﬁneprecipitate strengthened alloy at 800
◦
C, with a precipitate solvus temperature of
925
◦
C (Betteridge and Heslop, 1974). Combustion chamber applications require
the material to undergo temperature ﬂuctuations between 20
◦
C and 950
◦
C, and the
temperature range is therefore such that the precipitate solvus can be exceeded dur
ing inservice operation. Signiﬁcant microstructural change is therefore likely to occur
during ordinary operation, leading to quite profound changes in the creep mechanisms
in the material, controlling both deformation and component life.
In this chapter, we shall introduce a physically based creep model which explicitly
accounts for microstructural change, and its inﬂuence on creep deformation and
failure, in polycrystalline nickelbase alloy C263 for temperatures both above
and below the γ
solvus. The implementation into ABAQUS is carried out using
a forward Euler integration scheme and we again use the initial stiffness as the material
Jacobian for simplicity.
9.2 Physically based constitutive equations
Creep in nickel alloy C263 occurs through diffusionactivated dislocation climb and
precipitate ‘cutting’, depending on the precipitate spacing, λ
p
(or equivalently, size,
r
s
, for a given precipitate volume fraction, φ
p
, which in turn depends on temperat
ure, T ). Above a critical particle spacing, λ
pc
, dislocation climb dominates, whereas
belowthe critical spacing, precipitate ‘cutting’ dominates. The material alsoundergoes
coarsening at temperature such that the particle spacing increases with time.
210 Creep in aeroengine components
All of these microstructural processes are embodied within creep constitutive
equations. The creep strain rate, ˙ p
c
, also depends upon the density, ρ
n
, of mobile
dislocations which in turn depends on the accumulated creep strain. The con
stitutive equation set embodying all the mechanisms discussed above is given below.
The activation volume, V, depends on the active obstruction mechanism. In the
case of dislocation pinning by precipitates, V is dependent on the pinning distance,
which in turn depends on the precipitate spacing. For precipitate cutting, the activa
tion volume is V
c
. In the creep strain rate equation, the uniaxial strain rate, ˙ ε
c
, in
response to a uniaxial stress, σ, is determined from the corresponding shear values
using the usual relationships (Dieter, 1988): ˙ ε = ˙ γ /
¯
M and σ =
¯
Mτ, in which
¯
M
is the Taylor factor. F is the Helmholtz free energy, b the Burger’s vector, ν the
frequency of dislocation jumping energy barriers, k
b
the Boltzman constant, and ρ
i
the initial density of mobile dislocations.
˙ p
c
=
ρ
n
ρ
i
([4π/3φ
p
]
1/3
− 2)
[4π/3φ
p
]
1/3
¯
M
b
2
ν exp
_
−F
k
b
T
_
sinh
_
σ
e
V
k
b
T
¯
M(1 − ω)
_
, (9.1)
where
V = V
c
, if φ
p
> 0 and λ
p
< λ
pc
(cutting),
V = λ
p
b
2
, if φ
p
> 0, λ
p
> λ
pc
and λ
p
< λ
d
(climbing),
V = λ
d
b
2
, if φ
p
= 0 (dislocation network)
and
λ
p
= r
s
_
_
4π
3φ
p
_
1/3
− 2
_
,
˙ ρ
n
= ψ˙ ε
c
.
In Equation (9.1), ω is a scalar damage variable which has evolution equation
˙ ω =
_
σ
1
σ
e
_
χ
H(σ
1
) ˙ p
c
(9.2)
in which σ
1
is the maximum principal stress, H the Heaviside function, and and χ
are material constants (χ being the sensitivity of stress state parameter).
The equations contain just two unknown ﬁtting constants, namely ψ and associ
ated with the multiplication of mobile dislocations and cavitation, occurring largely
within the tertiary creep regime. All the other material constants are physical properties
or measurable physical quantities.
Creep tests have been carried out by RollsRoyce over the temperature range
700–750
◦
C, and by Zhang and Knowles (2001) on C263 over the temperature range
800–950
◦
C. Prior to creep testing, the material had been subjected to the standard
Physically based constitutive equations 211
0.0
0.0
0.5
1.0
S
t
r
a
i
n
(
%
)
1.5
380 MPa
320MPa
200 MPa
250MPa
160MPa
270MPa
[×10
6
] [×10
6
]
2.0
1.0 2.0 3.0
Time (s)
4.0 5.0 6.0
Experimental results Computed results
0 2 4 6 8 10
0
2
4
6
8
10
S
t
r
a
i
n
(
%
)
350MPa
450MPa
250MPa
200MPa
40MPa
50MPa
Time (s)
0 1 2 3
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
S
t
r
a
i
n
(
%
)
[×10
6
] [×10
6
] Time (s)
0.0 0.1 0.2 0.3 0.4 0.5 0.6
0
5
10
15
20
25
30
S
t
r
a
i
n
(
%
)
Time (s)
(a)
(c) (d)
(b)
Fig. 9.1 Comparison of experimental and computed isothermal creep results at (a) 700
◦
C, (b) 750
◦
C,
(c) 800
◦
C, and (d) 950
◦
C.
heat treatment. This involves solutioning at 1150
◦
C for 2 h, quenching and ageing
at 800
◦
C for 8 h followed by air cooling. The results of the tests are shown in Fig. 9.1
by the broken lines.
The constants and constant groups appearing in the equations have been determined
by standard optimization techniques. The volume fraction of γ
precipitate is known
at each test temperature, Burger’s vector for this material is taken as 2.5 × 10
−10
m
(Frost and Ashby, 1982), and the initial density of dislocations can be estimated to
212 Creep in aeroengine components
Table 9.1 Physical constants determined from creep data by optimization.
F (J/atom) λ
pc
(nm) V
c
(m
3
) ν (s
−1
)
¯
M
7.4 × 10
−19
64.9 4.05 × 10
−27
15.1 × 10
21
3.57
be 10
10
m
−2
(Hull, 1975). The remaining temperatureindependent physical prop
erties can then be determined using the results of the optimization and the model
equations. The results of this process are shown in Table 9.1, which includes the
Helmholtz free energy, F, the critical particle spacing, λ
pc
, above which disloca
tion bowing dominates over particle cutting, the activation volume for cutting, V
c
,
the frequency of dislocations jumping energy barriers, ν, and the Taylor factor. The
temperature dependence of the two empirical constants for the evolution of mul
tiplication of mobile dislocation density and cavitation is shown in Appendix 9.1.
The computed creep curves resulting from the determination of the creep parameters
are also shown in Fig. 9.1.
9.3 Multiaxial implementation into ABAQUS
We assume multiaxial creep deformation in nickel alloy C263 to obey von Mises
behaviour; that is, the relationship between uniaxial creep strain rate and stress is
identical to that between the effective creep strain rate and effective stress. Under
multiaxial conditions, the creep strain rate ‘direction’ is taken to be normal to the
dissipative surface, in an analogous way to that in plasticity, giving the multiaxial
creep strain rate as
˙ ε
c
=
3
2
˙ p
c
σ
σ
e
in which σ
is the deviatoric stress tensor, and σ
e
the effective stress, given by
σ
e
=
_
3
2
σ
: σ
_
1/2
and the effective creep strain rate, ˙ p
c
, is given in Equation (9.1). The equations have
been implemented into an ABAQUS UMAT using a simple forward Euler explicit
integration scheme with the initial stiffness method; that is, the material Jacobian is
speciﬁed simply as the elasticity matrix.
9.3.1 Finite element modelling of biaxial creep tests
A range of tests have been carried out on nickel alloy C263 in order to investigate its
stress state sensitivity of creep failure. The tests have been carried out on thinwalled
Multiaxial implementation into ABAQUS 213
Table 9.2 Summary of tests and loading conditions, and corresponding
stress state.
Shear stress (MPa) Axial stress (MPa) σ
1
/σ
e
Tension 0 320 1.0
Tension–torsion 160 160 0.81
Torsion 184 0 0.58
Compression–torsion 160 −160 0.31
tubular specimens which have been subjected to uniaxial tension, tension–torsion,
torsion, and compression–torsion. The material was subjected to the standard heat
treatment such that its uniaxial creep behaviour, through to failure, is correctly mod
elled using the above constitutive equations with the material properties given in
Manonukul et al. (2002). The test specimen gauge section has length 16 mm, with
internal and external radii of 3.12 and 4.12 mm, respectively. The tests were carried
out at 800
◦
C such that the von Mises effective stress for each test was the same,
namely 320 MPa. The loading conditions applied are summarized in Table 9.2.
The loads were increased linearly from zero to the maximum (to give the desired
stresses) over 14 s, and were then held at the maximum value. The experimental test
results are shown in Fig. 9.2(a). The quality of the creep data obtained for the torsion
test is, unfortunately, not as good as desired because of problems with temperature
control. However, the results do demonstrate the strong dependence of creep life on
stress state. The stress state, given by σ
1
/σ
e
, for each test, is shown in Table 9.2.
The test specimens were modelled by developing threedimensional meshes of the
gauge sections (i.e. uniformsection tubes). Because of the need to carry out compres
sion tests, the wall thickness was constrained by buckling problems. As a result, the
specimen wall thickness was such that an assumption of uniformshear strain and stress
through the wall thickness would not have been appropriate. A twodimensional ﬁnite
element model was therefore necessitated. However, because of the ease of speciﬁca
tion of torsional loading conditions using threedimensional elements (as opposed to
axisymmetric elements with ‘twist’ in ABAQUS), 240 eightnoded, threedimensional
solid elements were used so that the specimen thickness contained two elements.
Geometrical nonlinearity was accounted for using the standard ABAQUS large strain
formulation. Under torsional loading, because of the radial variation of stress, the
loads applied were chosen to ensure that the average effective stress through the thick
ness was that desired, that is, 320 MPa. The creep behaviour of the material was
described by the constitutive equations given above. The only unknown parameter in
the equations is the stress state sensitivity, χ. The four tests described above were
simulated using the ﬁnite element model. Parametric studies were carried out in order
214 Creep in aeroengine components
0
0.05
0.1
0.15
0.2
0.25
0.3 (a)
(b)
0 100000 200000 300000 400000 500000 600000 700000 800000
Time (s)
E
q
u
i
v
a
l
e
n
t
c
r
e
e
p
s
t
r
a
i
n
Tension
Tension–torsion Torsion
Compression–torsion
0
0.05
0.1
0.15
0.2
0.25
0.3
0 100000 200000 300000 400000 500000 600000 700000 800000
Time (s)
E
q
u
i
v
a
l
e
n
t
c
r
e
e
p
s
t
r
a
i
n
Tension
Tension–torsion
Torsion
Compression–torsion
Compression
Fig. 9.2 (a) Experimental and (b) computed equivalent creep strain versus time for isothermal creep
tests carried out at 800
◦
C with effective stress 320 MPa for the loading conditions shown.
to determine χ. This was done by choosing that value of χ which provided computed
results closest to those seen in the experiments. The computed results obtained using
a value for χ of 0.38 are shown in Fig. 9.2(b). This value lies within the range for
nickel alloys discussed by Dyson and Loveday (1981). Considerably different creep
lives are seen to occur depending on the stress state, and this is captured well by the
model. The computed creep curve for uniaxial compression shown in Fig. 9.2(b) can
be seen to have a slowly increasing gradient, even though there is no creep cavitation
Multiaxial implementation into ABAQUS 215
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
0 500 1000 1500 2000 2500
Time (h)
D
i
s
p
l
a
c
e
m
e
n
t
o
f
t
h
e
t
o
p
b
o
u
n
d
a
r
y
(
m
m
)
194MPa
161MPa
104MPa
Indicates predicted
failure on the basis
of a damage level
of 0.3 through the
specimen section
Fig. 9.3 Predicted displacement of the top boundary versus time (solid lines) for the double notch
specimen subjected to the stresses shown. Broken straight lines indicate experimentally determined
times to failure.
occurring. This results from both the precipitate coarsening and the multiplication of
mobile dislocations, both of which are included in the model.
9.3.2 Prediction of notched bar creep behaviour
Three experimental tests have been carried out on double notched bar test speci
mens. The specimens have gauge length 25.4 mm and diameter 5.62 mm. The
circular notches have diameter 1.12 mm and they are separated by 7 mm along the
length of the specimen. The effect of the notches is to introduce multiaxial stress
states local to the notches, which inﬂuence creep damage evolution, as discussed
above. Uniaxial loading was applied to generate nominal section stresses of 104, 161,
and 194 MPa, respectively. The test specimens were modelled using axisymmetric
elements. Because of symmetry, just one quarter of the specimen section was mod
elled explicitly. The specimen lifetimes have been predicted and the lives compared
with those obtained in the experiments.
The predicted displacements of the top boundary of the test specimen are shown for
the three applied, nominal stresses of 104, 161, and 194 MPa in Fig. 9.3. The creep
damage evolution for the case of the 161 MPa nominal stress is shown in Fig. 9.4(a).
At this stress level, creep damage is predicted to initiate at the notch root after creeping
for about 13 h. By the time creep has continued for 166 h, the damage has evolved
right across the test specimen, in which the damage level is 0.26 or higher. This is
216 Creep in aeroengine components
(c)
Number of
creep voids
Cavitation
damage parameter
0
+0.00E+00
+2.00E–02
+4.36E–02
+6.73E–02
+9.09E–02
+1.15E–01
+1.38E–01
+1.62E–01
+1.85E–01
+2.09E–01
+2.33E–01
+2.56E–01
+2.80E–01
+3.00E–01
1–3
4–6
7–9
10–12
13
(a)
(b)
1
2
2
4
3
6
2
3
5
9
4
7
8
7
4
1
9
15
12
2
2
22
17
5
6
7
5
2
8
30
2
3
2
0
0
1
0
0
4
0
2
Fig. 9.4 Creep damage ﬁelds (a) predicted by the model, (b) observed in the microstructure, and
(c) from a surface void count using the micrograph. (See also Plate 4.)
References 217
Table 9.3 Predicted and experimentally determined top boundary
displacement at failure (deﬁned as experimental specimen separation).
Stress (MPa) Predicted displacement Experimental displacement
at failure (mm) at failure (mm)
104 1.9 1.97
161 0.1 0.11
shown in Fig. 9.4(a), together with a micrograph of the corresponding region of the
test specimen in Fig. 9.4(b). Microcracking can be seen to initiate at the notch root at
an angle of about 22
◦
from the horizontal. A surface void count has been carried out
and the results are shown in Fig. 9.4(c).
The damage ﬁelds in Fig. 9.4(a) can be seen to predict reasonably well both the site
of major cracking and the distribution of experimentally observed creep cavitation.
Dyson and McLean (1990) have argued that creep rupture occurs typically when the
creep cavity damage achieves a value of about 0.3. This results from the coalescence
of creep cavitation and the propagation of macroscopic creep cracking. The experi
mentally determined failure times are shown in Fig. 9.3 by the dashed lines. The
measured and calculated displacements of the top boundary at failure, which show
good agreement, are shown in Table 9.3. The predicted displacement at failure was
determined as the predicted displacement when a damage level of about 0.3 had been
achieved through the specimen section.
However, the analyses were not stopped once a creep damage level of 0.3 had
been achieved across the specimen section, and hence Fig. 9.3 shows the calculated
top boundary displacement continuing to increase as the damage level exceeds 0.3.
Afurther reason for continuing the analysis is the difﬁculty of deﬁning creep failure in
the calculations. In the experiments, this is easier; it is simply the measured time until
the specimen breaks. An alternative way to deﬁne calculated specimen rupture time
is simply to examine the top boundary displacement rate, and to assume that rupture
occurs once the rate becomes large. On this basis, the comparison of predicted and
experimental time to rupture results shown in Fig. 9.3 is reasonable.
References
Betteridge, W. and Heslop, J. (1974). The Nimonic Alloys. Edward Arnold,
London, UK.
Dyson, B.F. and Loveday, M.S. (1981). In Ponter, A.R.S. and Hayhurst, D.R. (eds),
Proceedings of the Creep in Structures IUTAM Symposium, Springer, Berlin,
406–421.
218 Creep in aeroengine components
Dyson, B.F. and McLean, M. (1990). ‘Creep deformation of engineering alloys—
development from physical modelling’. Iron and Steel Institute of Japan Interna
tional (ISIJ), 30, 802–811.
Dieter, G.E. (1988). Mechanical Metallurgy. McGrawHill Book Co., London, UK.
Frost, H.J. and Ashby, M.F. (1982). DeformationMechanism Maps: The Plasticity
and Creep of Metals and Ceramics. Pergamon Press, Oxford, UK.
Hull, D. (1975). Introduction to Dislocations, 2nd edition. Pergamon Press,
Oxford, UK.
Manonukul, A., Dunne, F.P.E., and Knowles, D. (2002). ‘Physicallybased model
for creep in nickelbase superalloy C263 both above and below the γ
solvus’.
Acta Materialia, 50(11), 2917–2931.
Zhang, Y.H. and Knowles, D. (2001). ‘Deformation behaviour and development
of microstructure during creep of a nickelbase superalloy’. In Parker, J.D.
(ed.), Proceedings of the 9th International Conference on Creep and Fracture of
Engineering Materials and Stuctures, April 2001, Gomer Press, UK.
Appendix 9.1
The dependence of precipitate volume fraction, φ, on temperature, T , in C263
φ = −4.537 × 10
−9
T
3
+ 1.261 × 10
−5
T
2
− 1.187 × 10
−2
T + 3.915.
The dependence of constants and on temperature.
700
◦
C 750
◦
C 800
◦
C 950
◦
C
496.1 435.2 10.0 1.1×10
−19
ψ 1.58×10
−2
3.01×10
−2
18.62 19.84
10. Cyclic plasticity, creep, and TMF
10.1 Introduction
In Chapter 9, we examined creep in the aeroengine combustor material nickelbase
superalloy C263. In this chapter, we address combined anisothermal cyclic plasticity
and creep. We use timeindependent cyclic plasticity with both isotropic and kin
ematic hardening which we combine with the physically based creep model described
in Chapter 9. The plasticity model is implemented into ABAQUS Implicit by means of
a UMAT subroutine which employs a simple ﬁrstorder forward (explicit) integration
scheme and the initial stiffness method. We then address isothermal cyclic plasti
city and creep, followed by an examination of thermomechanical fatigue in nickel
alloy C263.
10.2 Constitutive equations for cyclic plasticity
C263 is largely rateindependent and shows very limited creep response at temper
atures below ∼600
◦
C. During zeromean, straincontrolled reversed plasticity below
this temperature, the material exhibits both kinematic hardening within individual
cycles, and isotropic strain softening/hardening over many cycles, depending on the
temperature. The resulting stress–strain hysteresis loops show considerable change
as plastic strain accumulates. Above ∼600
◦
C, under similar loading, the material
starts to exhibit strainrate dependence. Viscoplastic constitutive equations for a poly
crystalline nickelbase alloy have been presented by Yaguchi et al. (2002a,b), who
considered uniaxial creep, isothermal, and anisothermal cyclic plasticity between
450
◦
C and 950
◦
C.
Because of the wide temperature range considered here (20–950
◦
C), a uniﬁed
viscoplasticitycreep model is not adopted because of the numerical problems that
can arise when using such a model for timeindependent plasticity. In any case, it
would, in addition, prevent the retention of the physical basis of the creep model
to be used here, presented by Manonukul et al. (2002). A nonuniﬁed approach has
therefore been adopted in which conventional timeindependent reversed plasticity
220 Cyclic plasticity, creep, and TMF
theory for lowtemperature deformation has been combined with the physically based
creep model for elevated temperature deformation. The further justiﬁcation for the
separation of creep and timeindependent plasticity terms is that at 20
◦
C, for example,
(thermally activated) creep simply does not occur. Deformation takes place by plastic
slip which is not aided by thermally activated dislocation climb. Conversely, at 950
◦
C,
deformation is dominated by diffusioncontrolled processes, and plastic slip, aided by
thermal activation or otherwise, is negligible. At about 800
◦
C, both sets of mech
anisms operate, and a uniﬁed viscoplasticity theory would be appropriate. However,
for the main application considered here—thermomechanical fatigue with consider
able temperature variations, 20–950
◦
C, our approach can be justiﬁed. It is therefore
assumed that the elastic, dε
e
, timeindependent plastic, dε
p
, creep, dε
c
, and thermal,
dε
θ
, strain increments may be additively decomposed such that the total strain, dε, is
given by
dε = dε
e
+ dε
p
+ dε
c
+ dε
θ
. (10.1)
There have been many developments in the modelling of kinematic hardening that
take account, for example, of nonproportional loadings (Voyiadjis and Abu AlRub,
2003). However, for simplicity here, the Chaboche combined isotropic and kinematic
hardening model (Lemaitre and Chaboche, 1990) is employed for timeindependent
plasticity.
The increment in nonlinear kinematic hardening dx and the increment in isotropic
hardening dr are given, respectively, by
dx =
2
3
c dε
p
− γ x dp (10.2)
and
dr = b(Q − r) dp, (10.3)
where
dp =
_
2
3
dε
p
: dε
p
_
1/2
. (10.4)
Here, x is the kinematic hardening stress tensor, r the isotropic hardening stress,
and b and Q are temperature dependent material parameters associated with iso
tropic hardening. Temperature and accumulated plastic strain dependent material
parameters associated with kinematic hardening are c and γ , respectively. A large
number of fully reversed cyclic plasticity tests have been carried out in order to
determine the temperaturedependent material parameters arising in the kinematic
hardening equations. dp is the effective timeindependent plastic strain increment.
The increment in plastic strain is determined in the conventional way assuming a von
Constitutive equations for cyclic plasticity 221
Mises yield function, f , given by
f =
_
3
2
(σ
−x
) : (σ
−x
)
_
1/2
− r − σ
y
= 0
so that
dε
p
= dλ
∂f
∂σ
=
3
2
dλ
σ
−x
σ
e
(10.5)
in which dλ is the plastic multiplier which, for combined isotropic and kinematic
hardening, is given, using Voigt notation, by
dλ =
(∂f /∂σ) · C dε
(∂f /∂σ) · {(2/3)c(∂f /∂σ) − γ x} + b(Q − R) + (∂f /∂σ) · C(∂f /∂σ)
=
_
(
3
2
)(σ
/σ
e
) · C dε
(
3
2
)(σ
/σ
e
)·{(
2
3
)c(
3
2
)(σ
/σ
e
) − γ x} + b(Q − R) + (
3
2
)(σ
/σ
e
)· C(
3
2
)(σ
/σ
e
)
.
_
(10.6)
The effective stress is, of course, given by
σ
e
=
_
3
2
(σ
−x
) : (σ
−x
)
_
1/2
.
The increment in thermal strain is calculated in the usual way as
dε
θ
= α dθI (10.7)
in which α is the coefﬁcient of thermal expansivity, dθ the increment in temperature,
and I the identity matrix.
The multiaxial constitutive equations presented above for creep and plasticity have
been implemented into a UMAT user subroutine using the simple ﬁrstorder explicit
Euler forward integration scheme as follows
dσ
t
= C
_
dε
t
−
3
2
dλ
σ
σ
e
− dε
c
− dε
θ
_
in which the increment in creep strain is obtained from Equation (9.33) and those in
Table 9.1, and
σ
t +t
= σ
t
+ dσ
t
,
x
t +t
= x
t
+ dx
t
,
ε
p
t +t
= ε
p
t
+ dε
p
t
,
r
t +t
= r
t
+ dr
t
.
222 Cyclic plasticity, creep, and TMF
10.3 Constitutive equations for C263 undergoing TMF
The thermomechanical loading cycles to which the nickelbase superalloy C263 is
subjected inservice are such that the γ
solvus is exceeded so that the material is,
in effect, solution treated. This has a profound effect on the behaviour of the material;
under conditions of both creep and reversed plasticity. The ratecontrolling mechan
isms for creep deformation change from precipitate cutting and dislocation pinning
at precipitates, and climb, to that of the pinning of dislocations through the estab
lishment of a dislocation network. The physically based creep model employed here
is able to capture these changes and to represent the micro and macroscale creep
deformation. For a given temperature, the precipitate volume fraction is known, and
because the precipitate coarsening kinetics has been quantiﬁed, both size and hence
spacing of the precipitate is known. These quantities appear explicitly in the phys
ically based creep model. The effect of solution treating on cyclic plastic response
is also very signiﬁcant and has been addressed. Cyclic plasticity tests between 20
◦
C
and 950
◦
C have been carried out. With the combined creep–cyclic plasticity model
described above, the temperature dependence of the isotropic and kinematic hardening
constants have been determined (Manonukul et al., 2001). The effect of precip
itate volume fraction on cyclic plasticity behaviour is introduced into the model,
therefore, through the temperature dependence of the material parameters for cyclic
plasticity.
At temperatures of about 800
◦
C and above, C263 creeps quite signiﬁcantly. This
appears in the cyclic plasticity response as a reduction in the overall stress level,
occurring through relaxation. LCF tests have been carried out on C263 at 800
◦
C and
950
◦
C under fully reversed (R = −1), straincontrolled loading through to failure.
The experimentally determined and calculated stress–strain hysteresis loops for the
ﬁrst and second cycles of the standard heat treated material are shown in Fig. 10.1. At
800
◦
C, two strain rates have been used such that ramping times (the time over which
the strain is applied linearly with time) of 1 and 10 s have been imposed resulting
in the hysteresis loops shown in Fig. 10.1(a) and (b), respectively. Figure 10.1(c)
shows the results for 950
◦
C. In these cycles, a 1s strain hold has been imposed at the
tensile and compressive peak strains. At 800
◦
C, this does not result in a distinctive
stress relaxation, but this can be seen at 950
◦
C in the computed result. However,
the experiments show a much more progressive stress relaxation at the peak strains
suggesting that the rate of relaxation is not well represented by the creep model. This
occurs for a number of reasons. First, the creep model at 950
◦
C was developed from
experimental data obtained for a much lower stress level (a maximum of 50 MPa)
whereas here, stresses of over 200 MPa are developed in the hysteresis loop. In
addition, our cyclic plasticity model does not contain an explicit static recovery term,
Constitutive equations for C263 undergoing TMF 223
– 600
– 400
– 200
0
200
400
600
–0.6 –0.4 –0.2 0 0.2 0.4 0.6
– 600
– 400
– 200
0
200
400
600
–0.6 –0.4 –0.2 0 0.2 0.4 0.6
–300
–100
100
300
–0.6 –0.4 –0.2 0 0.2 0.4 0.6
(a)
(b)
(c)
Experimental Computed
Fig. 10.1 Experimental and computed LCF stress–strain curves for the ﬁrst and second cycles at
(a) 800
◦
C, 1 s ramping time, strain range 0.9%, (b) 800
◦
C, 10 s ramping time, strain range 1%, and
(c) 950
◦
C, strain range 0.75%.
and further, the creep and plastic strain increments have, of course, been decoupled in
the present work. Note that the microstructure of the material being loaded at 950
◦
C
is considerably different to that at 800
◦
C because of the dissolution of the precipitate
at the higher temperature. However, for both temperatures considered here, the peak
stress levels are reasonably well predicted.
224 Cyclic plasticity, creep, and TMF
–600
–400
–200
0
200
400
600
0 200 400 600 800 1000 1200 1400
(a)
(c)
(a)
(b)
(b)
(c)
Experimental Computed
Fig. 10.2 Experimental and computed LCF peak stresses versus cycles at (a) 800
◦
C, 1 s ramping time,
strain range 0.9%, (b) 800
◦
C, 10 s ramping time, strain range 1%, and (c) 950
◦
C, strain range 0.75%.
Figure 10.2 shows the comparisons of computed and experimental peak stresses
versus cycles for each of the cases (a), (b), and (c) in Fig. 10.1, through to failure.
Good predictions of life can be seen to be achieved for (b) and (c) for which intergranu
lar creep cavitation dominates. Intergranular creep damage typically observed under
these conditions is shown in Fig. 10.2. The life is overpredicted for (a) since creep
damage accumulation has ceased to be the dominant failure process. A micrograph
for these conditions shows that transgranular fracture is now occurring.
Microstructural examination of specimens tested at 800
◦
C and 950
◦
C has been
carried out after specimen failure. LCF tests carried out at 800
◦
C, with 1s ramp
times, led to sample fracture surfaces showing transgranular cracking, as shown in
Fig. 10.2, indicating that lifetime under these conditions is dominated by fatigue
Constitutive equations for C263 undergoing TMF 225
processes. However, microstructural examination of specimens also tested at 800
◦
C
but witha considerablylonger ramptime of 10s showedthe development of intergranu
lar damage, which dominated in some regions of the material. The lifetime of the
material under these conditions may well be inﬂuenced quite strongly by creep dam
age processes. Examination of microstructures of specimens tested at 950
◦
C showed
that intergranular cracking dominated at these temperatures, even for a ramp time
of 1 s. The dominant failure process for these conditions is intergranular creep frac
ture. The results of the tests carried out at 800
◦
C shown in Fig. 10.2(a) and (b) show
the development of macrocracking towards the end of the test which inﬂuences the
observed stress–strain response which becomes different in tension and compression
because of crack closure effects.
The isotropic creep damage model used in the present work is not capable of pre
dicting crack closure effects. The result at 950
◦
C, however, which is dominated by
creep softening and cavitation, does not show this effect.
10.3.1 TMF in C263 both above and below the γ
solvus
TMF tests have been carried out on C263 under conditions of both in and out of phase
loading. Strain controlled tests have been carried out between strains of zero and a
peak of 0.6%, in which the temperature is also varied between 300
◦
C and 950
◦
C.
The strain and temperature loading imposed, together with the results obtained, are
shown in Fig. 10.3. First, considering the inphase loading shown in Fig. 10.3(a), the
experimental results obtained for the initial loading, below the γ
solvus temperature,
give stress levels that lie a little below that expected from both the solution treated and
the standard treated material. After a time of 45 s, the temperature has exceeded 925
◦
C,
and in fact, the experimental results give stress levels that would be anticipated for
such a material. It should be noted that prior to the imposition of the strain, the test
specimen has undergone a prior temperature loading cycle. This is necessary to enable
temperature compensation to be carried out on the TMF test rig. The consequence is
that the material is, therefore, in effect in the solution treated state. The modelling
of the behaviour has therefore been carried out with this assumption. For the cyclic
plasticity modelling, two sets of material parameters were determined; one for the
solution treated material (with the appropriate temperature dependence), and a further
one for the standard treated material (i.e. for the material containing the precipitate
volume fraction, which is dependent on temperature).
The model for the solution treated plasticity and creep behaviour give the computed
behaviour also shown in Fig. 10.3(a). As the temperature increases, a number of
processes are taking place. The volume fraction of precipitate is decreasing, and at
925
◦
C, becomes zero. The creep rate becomes determined by dislocation network
226 Cyclic plasticity, creep, and TMF
–300
–200
–100
0
100
200
300
–0.1 0.1 0.3 0.5 0.7
Strain (%)
S
t
r
e
s
s
(
M
P
a
)
0
500
1000
0 50 100
Time (s)
T
e
m
p
e
r
a
t
u
r
e
(
8
C
)
0
0.2
0.4
0.6
0 50 100
Time (s)
S
t
r
a
i
n
(
%
)
– 300
– 200
– 100
0
100
200
300
–0.7 –0.5 – 0.3 –0.1 0.1
Strain (%)
S
t
r
e
s
s
(
M
P
a
)
0
500
1000
0 50 100
Time (s)
T
e
m
p
e
r
a
t
u
r
e
(
8
C
)
– 0.6
– 0.4
– 0.2
0
0 50 100
Time (s)
S
t
r
a
i
n
(
%
)
–800
–600
–400
–200
0
200
400
600
800
–0.2 0 0.2 0.4 0.6
Strain (%)
S
t
r
e
s
s
(
M
P
a
)
0
500
1000
0 50 100
Time (s)
T
e
m
p
e
r
a
t
u
r
e
(
8
C
)
– 0.1
0.4
0 50 100
Time (s)
S
t
r
a
i
n
(
%
)
(a)
(b)
(c)
Experimental Predicted
Fig. 10.3 Experimental and predicted stress–strain curves for (a) inphase TMF at temperature range
300–950
◦
C and (b) outofphase TMF at temperature range 300–950
◦
C, and (c) inphase TMF at
temperature range 300–800
◦
C .
pinning (Manonukul et al., 2002) and increases as the temperature increases. The
combined effects of these processes result in the predicted decreasing stress level. At
a time of 45s, boththe strainandtemperature loadingare reversed, anda comparatively
small elastic region (though creep deformation is still occurring) can be seen, which
References 227
is followed by a conventional plasticity region in which creep deformation becomes
negligibly small as the temperature decreases towards 300
◦
C. Note that the material
continues to harden during this phase, with no stress relaxation occurring at the higher
compressive stresses because of the elimination of creep. The stress levels measured
and predicted by the model are those that would be anticipated fromthe solution treated
material. The standard heat treated material would show considerably higher stresses.
With the reversal of the loading again, a rather larger elastic region is observed because
of the increasedyieldstress at lower temperature, andbecause of the existinghardening
that has already taken place. The measured and predicted stresses developed with
increasingplastic deformationshowgoodagreement and, unlike the ﬁrst quartercycle,
are those that would be expected for the solution treated material.
The loading conditions and results obtained for outofphase TMF are shown in
Fig. 10.3(b). In this case, the high tensile stresses occur during the low temperature
parts of the loading cycle, and conventional plastic behaviour is observed. At the higher
temperatures, the material is in compression, but again, stress relaxation can be seen to
occur, and the features of the experimentally obtained results can be seen to be reason
ably reproduced. For inphase TMF with the same strain range, but with the temperat
ure varyingbetween300
◦
Cand800
◦
C, the experimental andpredictedresults obtained
are shown in Fig. 10.3(c). Good comparisons are achieved for these conditions.
References
Lemaitre, J. and Chaboche, J.L. (1990). Mechanics of Solid Materials. Cambridge
University Press, Cambridge.
Manonukul, A., Dunne, F.P.E., and Knowles, D. (2001). ‘Thermomechanical fatigue
in nickelbase superalloy C263’. Internal Report, Department of Engineering
Science, Oxford University, England.
Manonukul, A., Dunne, F.P.E., and Knowles, D. (2002). ‘Physicallybased model
for creep in nickelbase superalloy C263 both above and below the γ
solvus’. Acta
Materialia, 50, 2917–2931.
Voyiadjis, G.Z. and Abu AlRub R.K. (2003). ‘Thermodynamic based model for the
evolution equation of the backstress in cyclic plasticity’. International Journal of
Plasticity, 19(12), 2121–2147.
Yaguchi, M., Yamamoto, M., and Ogata, T. (2002a). ‘Aviscoplastic constitutive model
for nickelbase superalloy Part I: kinematic hardening rule of anisotropic dynamic
recovery’. International Journal of Plasticity, 18(8), 1083–1109.
Yaguchi, M., Yamamoto, M., and Ogata, T. (2002b). ‘A viscoplastic constitutive
model for nickelbase superalloy Part II: modelling under anisothermal conditions’.
International Journal of Plasticity, 18(8), 1111–1131.
This page intentionally left blank
Appendix A: Elements of
tensor algebra
The dot (scalar) product of two ﬁrstorder tensors (vectors) gives a scalar
_
NB. Einstein summation convention: u
i
v
i
=
n
i=1
u
i
v
i
_
u · v = u
i
v
i
=
n
i=1
u
i
v
i
, (A1)
v · u = v
i
u
i
=
n
i=1
v
i
u
i
, (A2)
v · u = u · v. (A3)
Double contraction (double dot product) of two secondorder tensors gives a scalar
σ : ε = σ
ij
ε
ij
=
n
i=1
n
j=1
σ
ij
ε
ij
, (A4)
ε : σ = ε
ij
σ
ij
=
n
i=1
n
j=1
ε
ij
σ
ij
, (A5)
ε : σ = σ : ε. (A6)
The dot product of a ﬁrstorder tensor with a secondorder tensor gives a ﬁrstorder
tensor
σ · n ⇒ (σ
ij
n
j
)
i
=
n
j=1
σ
ij
n
j
, (A7)
n · σ ⇒ (n
j
σ
ji
)
i
=
n
j=1
n
j
σ
ji
, (A8)
230 Appendix A: Elements of tensor algebra
hence the expression
n · σ = σ · n (A9)
is valid only if σ is symmetric (i.e. if σ
ij
= σ
ji
for any i, j).
The dyadic (direct) product of two ﬁrstorder tensors gives a secondorder tensor
u ⊗v ⇒ (u ⊗v)
ij
= u
i
v
j
. (A10)
For example, if
u =
⎡
⎣
u
1
u
2
u
3
⎤
⎦
and v =
⎡
⎣
v
1
v
2
v
3
⎤
⎦
,
then
u ⊗v =
⎛
⎝
u
1
v
1
u
1
v
2
u
1
v
3
u
2
v
1
u
2
v
2
u
2
v
3
u
3
v
1
u
3
v
2
u
3
v
3
⎞
⎠
.
The product of two secondorder tensors gives a secondorder tensor
a · b ⇒ (a · b)
ij
= a
ik
b
kj
=
n
k=1
a
ik
b
kj
. (A11)
Double contraction (double dot product) of a thirdorder tensor with a secondorder
tensor gives a ﬁrstorder tensor
ξ : σ =
3
i,j,k=1
ξ
ijk
σ
jk
e
i
. (A12)
Double contraction (double dot product) of a fourthorder tensor with a secondorder
tensor gives a secondorder tensor
c : ε ⇒ (c : ε)
ij
= c
ijkl
ε
kl
=
n
k=1
n
l=1
c
ijkl
ε
kl
(A13)
hence the expression
c : ε = ε : c (A14)
is valid only if c exhibits major symmetry (i.e. if c
ijkl
= c
klij
for any i,j,k,l).
The dyadic (direct) product of two secondorder tensors gives a fourthorder tensor
f ⊗g ⇒ (f ⊗g)
ijkl
= f
ij
g
kl
. (A15)
Kronecker delta—a special secondorder tensor
δ ⇒ δ
ij
_
δ
ij
= 1, if i = j,
δ
ij
= 0, if i = j.
(A16)
Appendix A: Elements of tensor algebra 231
The unit fourthorder tensor (exhibits major but not minor symmetry)
I ⇒ I
ijkl
= δ
ik
δ
jl
(A17)
has the following important property
I : ε = ε : I, (A18)
which is valid for any secondorder tensor ε.
A symmetrized unit fourthorder tensor (exhibits both major and minor symmetry)
I
s
⇒ I
s
ijkl
=
1
2
(δ
ik
δ
jl
+ δ
il
δ
jk
) (A19)
ensures the following identity
I
s
: ε = ε : I
s
, (A20)
which is valid only if the secondorder tensor ε is symmetric.
Differentiation
Differentiation of a tensor valued function (e.g. u, σ) with respect to its tensorial
argument (e.g. x, ε).
Example: Differentiation of the ﬁrstorder tensor u with respect to the ﬁrstorder
tensor x gives the secondorder tensor
∂u
∂x
⇒
_
∂u
∂x
_
ij
=
∂u
i
∂x
j
. (A21)
Special case:
∂u
∂u
⇒
_
∂u
∂u
_
ij
= δ =
∂u
i
∂x
j
= δ
ij
. (A22)
Example: Differentiation of the secondorder tensor σ with respect to the second
order tensor ε gives the fourthorder tensor
∂σ
∂ε
⇒
_
∂σ
∂ε
_
ijkl
=
∂σ
ij
∂ε
kl
. (A23)
Special case:
∂σ
∂σ
⇒
_
∂σ
∂σ
_
ijkl
= I =
∂σ
ij
∂σ
kl
= I
ijkl
. (A24)
232 Appendix A: Elements of tensor algebra
The chain rule
Example: Secondorder tensor f depends on a secondorder tensor u and a scalar v
∂f
∂x
=
∂f
∂u
:
∂u
∂x
+
∂f
∂v
⊗
∂v
∂x
⇒
_
∂f
∂x
_
ijkl
=
∂f
ij
∂x
kl
=
∂f
ij
∂u
mn
∂u
mn
∂x
kl
+
∂f
ij
∂v
∂v
∂x
kl
.
(A25)
The gradient of a scalar ﬁeld gives a ﬁrstorder tensor
∇f ⇒
∂f
∂x
i
. (A26)
The gradient of a ﬁrstorder tensor ﬁeld gives a secondorder tensor
∇v ⇒
∂v
j
∂x
i
. (A27)
The gradient of a secondorder tensor ﬁeld gives a ﬁrstorder tensor with its
components being secondorder tensors
∇σ ⇒ (∇σ)
j
=
∂σ
ij
∂x
i
=
∂σ
ji
∂x
i
. (A28)
Hence,
∇σ =
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎣
∂σ
xx
∂x
∂σ
xy
∂x
∂σ
xz
∂x
∂σ
xx
∂y
∂σ
xy
∂y
∂σ
xz
∂y
∂σ
xx
∂z
∂σ
xy
∂z
∂σ
xz
∂z
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
∂σ
yx
∂x
∂σ
yy
∂x
∂σ
yz
∂x
∂σ
yx
∂y
∂σ
yy
∂y
∂σ
yz
∂y
∂σ
yx
∂z
∂σ
yy
∂z
∂σ
yz
∂z
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎣
∂σ
zx
∂x
∂σ
zy
∂x
∂σ
zz
∂x
∂σ
zx
∂y
∂σ
zy
∂y
∂σ
zz
∂y
∂σ
zx
∂z
∂σ
zy
∂z
∂σ
zz
∂z
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
. (A29)
Appendix A: Elements of tensor algebra 233
The divergence of a secondorder tensor gives a ﬁrstorder tensor
div σ = Tr[(∇σ)
i
]. (A30)
Hence,
div σ =
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
Tr
_
∂σ
ij
∂x
i
_
Tr
_
∂σ
ij
∂x
i
_
Tr
_
∂σ
ij
∂x
i
_
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
Tr
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
∂σ
xx
∂x
∂σ
xy
∂x
∂σ
xz
∂x
∂σ
xx
∂y
∂σ
xy
∂y
∂σ
xz
∂y
∂σ
xx
∂z
∂σ
xy
∂z
∂σ
xz
∂z
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
Tr
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
∂σ
yx
∂x
∂σ
yy
∂x
∂σ
yz
∂x
∂σ
yx
∂y
∂σ
yy
∂y
∂σ
yz
∂y
∂σ
yx
∂z
∂σ
yy
∂z
∂σ
yz
∂y
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
Tr
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
∂σ
zx
∂x
∂σ
zy
∂x
∂σ
zz
∂x
∂σ
zx
∂y
∂σ
zy
∂y
∂σ
zz
∂y
∂σ
zx
∂z
∂σ
zy
∂z
∂σ
zz
∂z
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
∂σ
xx
∂x
+
∂σ
xy
∂y
+
∂σ
xz
∂z
∂σ
yx
∂x
+
∂σ
yy
∂y
+
∂σ
yz
∂y
∂σ
zx
∂x
+
∂σ
zy
∂y
+
∂σ
zz
∂z
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
.
(A31)
Rotation
Finite rotation is mathematically described by the orthogonal secondorder tensor
R
−1
= R
T
, (A32)
which transforms a ﬁrstorder tensor as follows
x
= R · x (A33)
and a secondorder tensor as follows
σ
= R · σ · R
T
. (A34)
Successive ﬁnite rotations do not commute, that is,
R
1
· R
2
= R
2
· R
1
. (A35)
234 Appendix A: Elements of tensor algebra
r
r
r
r
= r
0
x
y
z
Fig. A.1 Finite rotation vector.
If the axis r =
_
r
1
r
2
r
3
_
and the magnitude r = r of a ﬁnite rotation are known the
orthogonal secondorder tensor which describes such rotation can be expressed in the
form of an exponential function as follows:
R = exp[ˆ r] = I + ˆ r +
1
2!
ˆ r
2
+ · · ·, (A36)
where ˆ r is the associated skewsymmetric tensor
ˆ r =
⎡
⎣
0 −r
3
r
2
r
3
0 −r
1
−r
2
r
1
0
⎤
⎦
, (A37)
which satisﬁes
ˆ r · r = 0,
ˆ r · x = r ×x
(A38)
as illustrated in Fig. A.1.
Small rotations can be approximated by
R = exp[ˆ r] = I + ˆ r +
1
2!
ˆ r
2
+ · · · ≈ I + ˆ r =
⎡
⎣
1 0 0
0 1 0
0 0 1
⎤
⎦
+
⎡
⎣
0 −r
3
r
2
r
3
0 −r
1
−r
2
r
1
0
⎤
⎦
,
(A39)
which do commute
(I + ˆ r
1
) · (I + ˆ r
2
) = (I + ˆ r
2
) · (I + ˆ r
1
) = I + ˆ r
1
+ ˆ r
2
+ ˆ r
1
· ˆ r
2
≈ I + ˆ r
1
+ ˆ r
2
.
(A40)
Appendix B: Fortran coding available
via the OUP website
∗
Directory/ﬁle Description
elasticity
elastic.f UMAT for plane strain and axial symmetry for
elastic behaviour using ABAQUS stress and strain
quantities
elas_axidisp.inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial displacement controlled
loading, requiring UMAT elastic.f
elas_axiforce.inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial force controlled loading,
requiring UMAT elastic.f
plasticity_exp
code_exp.f UMAT for plane strain and axial symmetry
for elastic, linear strain hardening plastic behaviour
using explicit integration with continuum Jacobian,
using ABAQUS stress and strain quantities.
Suitable for large deformations
plas_exp_axidisp_aba.inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial displacement controlled
loading, using ABAQUS *PLASTIC
plas_exp_axiforce_aba_inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial force controlled loading,
using ABAQUS *PLASTIC
∗
www.oup.co.uk/isbn/0–19–856826–6
236 Appendix B: Fortran coding
plas_exp_axidisp.inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial displacement controlled
loading, requiring UMAT code_exp.f
plas_exp_axiforce.inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial force controlled loading,
requiring UMAT code_exp.f
plas_exp_beam_aba.inp Four point bend loading using ABAQUS*PLASTIC,
requiring mesh ﬁle beam_mesh.inp
plas_exp_beam.inp Four point bend loading requiring UMAT
code_exp.f, requiring mesh ﬁle beam_mesh.inp
plasticity_imp
code_imp.f UMAT for plane strain and axial symmetry
for elastic, linear strain hardening plastic
behaviour using implicit integration with
consistent Jacobian, using ABAQUS
stress and strain quantities. Suitable for large
deformations
plas_imp_axidisp_aba.inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial displacement controlled
loading, using ABAQUS *PLASTIC
plas_imp_axiforce_aba_inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial force controlled loading,
using ABAQUS *PLASTIC
plas_imp_axidisp.inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial displacement controlled
loading, requiring UMAT code_imp.f
plas_imp_axiforce.inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial force controlled loading,
requiring UMAT code_imp.f
plas_imp_beam_aba.inp Four point bend loading using ABAQUS*PLASTIC,
requiring mesh ﬁle beam_mesh.inp
plas_imp_beam.inp Four point bend loading requiring UMAT
code_imp.f, requiring mesh ﬁle beam_mesh.inp
Appendix B: Fortran coding 237
spin
spin_elastic.f UMAT for threedimensional, plane strain,
and axial symmetry for elastic behaviour
using ABAQUS stress and strain
quantities
spin_elas_def.f UMAT for threedimensional, plane strain, and
axial symmetry for elastic behaviour using the
deformation gradient
spin_axidisp.inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial displacement controlled
loading, requiring UMAT spin_elas_def.f
spin_axiforce.inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial force controlled loading,
requiring UMAT spin_elas_def.f
spin_shear.inp ABAQUS input ﬁle for a single plane strain
element under simple shear, requiring UMAT
spin_elas_def.f
spin_shear_aba.inp ABAQUS input ﬁle for a single plane
strain element under simple shear, using
ABAQUS*ELASTIC
visco
uni_visco_imp.f Closed form Fortran implicit solution for uniaxial
elastoviscoplasticity
uni_visco_exp.f Closed form Fortran explicit solution for uniaxial
elastoviscoplasticity
visco_imp.f UMAT for plane strain and axial symmetry
for elastic, linear strain hardening
viscoplastic behaviour using implicit integration
using the initial tangent stiffness, using ABAQUS
stress and strain quantities. Suitable for large
deformations
visco_imp_axidisp.inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial displacement controlled
loading, requiring UMAT visco_imp.f
238 Appendix B: Fortran coding
visco_imp_axiforce.inp ABAQUS input ﬁle for single axisymmetric
element under uniaxial force controlled loading,
requiring UMAT visco_imp.f
visco_imp_beam.inp Four point bend loading requiring UMAT
visco_beam.f, requiring mesh ﬁle
beam_mesh.inp
Index
ABAQUS 141, 149, 150, 167, 169, 235
ABAQUS input ﬁles 169
accumulated plastic strain 23
Almansi strain 50
angular velocity tensor 63
anisothermal cyclic plasticity 220
antisymmetric 51
Armstrong–Frederick 21
associated ﬂow 18
axial vibration 106, 115
B matrix 99
back stress 29
backward Euler integration 146, 161
Bauschinger effect 28
bcc 6
biaxial creep tests 213
bowing 212
Burger’s vector 8
calculus of variations 85
cantilever beam 118
Cauchy stress 72
Cauchy–Green tensor 49, 50
cavitation 210
central difference method 136
Chaboche 31
coarsening 209
combustion chambers 209
conservation of energy 84
consistency condition 20
consistent Jacobian 153
consistent tangent stiffness 153
consolidation 199
constant strain element 99
constitutive equations 40
continuum tangent stiffness 172
continuum damage 210
continuum Jacobian 172
continuum plasticity 10
continuum spin 61
contracted tensor product 15
convergence 140
corotational 73
creep 209
CREEP subroutine 202
critical resolved shear stress 7
crystal plasticity 5, 9
crystallographic orientation 3
crystallographic slip 5
cyclic hardening 37
cyclic plasticity 219
deformationenhanced grain growth 190
deformation gradient 48
deformed conﬁguration 48
densiﬁcation rate 201
deviatoric stress 14
diagonalization 56
differentiation of a tensor 231
dilatation 201
direction of plastic ﬂow 19
dislocation bowing 212
dislocation density 210
displacementbased ﬁnite element method 98
divergence of a secondorder tensor 233
double contracted product 15
Duva and Crow 200
dyadic product 230
dynamical path 84
effective plastic strain rate 14
effective stress 13
elastic predictor 147
elastic stiffness matrix 22, 100
elastic–plastic deformation 11
engineering shears 100
equations of motion 90
equilibrium 83
equivalent stress 13
Euler–Lagrange equation 86
explicit ﬁnite element methods 1, 141
explicit integration 136, 143
240 Index
fcc 6
ﬁnite element formulation for plasticity 133
ﬁnite element method 83
ﬁnite rotations 57
ﬁrst variation 87
forward integration 145
Gauss quadrature 126
geometric nonlinearity 108
gradient of a ﬁrstorder tensor 232
gradient of a scalar 232
gradient of a secondorder tensor 232
grain boundaries 3
grain growth 190
grain size 190
Hamilton’s principle 84
hardening 23
Helmholtz free energy 210
Hooke’s law 22, 170
hydrostatic stress 14
hysteresis 222
identity tensor 42
implicit 140, 143
implicit ﬁnite element methods 143
implicit implementation for elastoviscoplasticity 161
implicit integration 146
implicit scheme 146
incompressibility 3, 12
incremental nature of plasticity 35
incremental rotation 176
initial tangential stiffness method 140
integration point 126
intergranular cracking 224
intergranular creep fracture 217
intermediate conﬁguration 66
internal variables 40
isoparametric 109
isotropic hardening 150
J
2
plasticity 17
Jacobian 150
Jaumann stress rate 76
Jintegral 84
Kinematic hardening 27
kinematics 47
kinetic energy 84
Lagrangian 84
Lame constant 42
large deformation(s) 47
leap frog explicit method 137
mass and spring system 90
mass matrix 105
material objectivity 69
material reference frame 48
material stress rate 78
mean stress 14
microstructural evolution 190
Mohr’s circle 69
momentum balance equations 106
multiaxial creep strain rate 212
multiaxial stress state 20, 212
multiplicative decomposition 67
necking 187
Newton iteration 148
Newton’s method 148
Newton–Raphson method 141
nickel alloy C263 209
nickelbase superalloy 209
nodal force vector 112
nodal forces 107
nonconservative 84
nonlinear kinematic hardening 32
norm 30
normal grain growth 190
normality hypothesis 18
notched bar test 215
objective stress 72
objectivity 69
onedimensional rod element 99
original conﬁguration 48
orthogonality 58
out of phase loading 225
overstress 38
particle cutting 209
perfect plasticity 11
plane stress 18
plastic correction 146
plastic deformation gradient 66
plastic multiplier 19
plastic strain rate tensor 14
polar decomposition theorem 57
polycrystal 3
polycrystalline 3
porosity 199
porous plasticity 199
potential energy 84
potential function 45
powerlaw creep 44
Prager 29
precipitate coarsening 209
precipitate cutting 209
precipitate spacing 210
predictor–corrector 146
principal coordinates 56
Index 241
principal stresses 13
principle of virtual work 90
quasistatic problems 105
radial return method 146
ratedependent plasticity 38
rate of deformation 60
reaction 130
reversed plasticity 28, 219
rigid body rotation 57
rotation 233
rotation matrix 58
Schmid’s law 7
secondorder tensors 229
second stress invariant 17
semiimplicit integration 160
shape functions 97
shearing 3
simple shear 58
single and multiple element uniaxial tests 171
single element simple shear test 178
skew 51
slip system 5
solvus temperature 209
spin 61
stability of the explicit time stepping 139
static grain growth 190
stationary value 87
stiffness matrix 105
strain decomposition 11
strain measure 49
strainrate sensitivity 39, 186
strain tensor 100
stress space 17
stress tensor 14
stress transformation 72
stress vector 70
stretch 48
stretch ratios 54
superplastic forming 192
Superplasticity 185
tangent stiffness 140, 150
tangential stiffness matrix 150
tension–torsion 213
tensorial notation 229
tensors 229
thermomechanical fatigue 219
thinning 195
timedependent plasticity 38
timeindependent plasticity 11
Ti–MMCs 205
titanium alloy, Ti–6Al–4V 189
traction 70
transformation of stress 72
transverse vibration 118, 122
Tresca 17
trial stress 147
true strain 53
truss element 99
UMAT 169
undeformed conﬁguration 48
uniaxial loading 14, 42
update of stress 143
upsetting 42
velocity gradient 60
veriﬁcation of the model implementation 171
virtual work 90
viscoplasticity 38
viscous stress 39
Voigt notation 19, 100
volume changes 199
von Mises 17
weak formulation 83
work conjugacy 111
yield criteria 17
yield function 17
yielding 3
This page intentionally left blank
+8.00E–06
+7.00E–06
+6.00E–06
+5.00E–06
+4.00E–06
+3.00E–06
+2.00E–06
+0.00E–00
Strain rate
Strain rate
+8.00E–06
+7.00E–06
+6.00E–06
+5.00E–06
+4.00E–06
+3.00E–06
+2.00E–06
+0.00E–00
(c)
+8.00E–06
+7.00E–06
+6.00E–06
+5.00E–06
+4.00E–06
+3.00E–06
+2.00E–06
+0.00E–00
Strain rate
(a) (b)
Plate 1 The simulated superplastically deforming sheet showing the effective plastic strain rate at
fractional processing times of t /t
f
= 0.1, t /t
f
= 0.6, and t /t
f
= 1.0.
(a)
Thickness strain
–1.20E+00
–1.07E+00
–9.15E–01
–7.60E–01
–6.05E–01
–4.50E–01
–2.95E–01
–1.40E–01
(b)
Thickness strain
–1.20E+00
–1.07E+00
–9.15E–01
–7.60E–01
–6.05E–01
–4.50E–01
–2.95E–01
–1.40E–01
Plate 2 Throughthickness strain ﬁelds at the end of the superplastic forming carried out with target
maximum strain rates of (a) 1.0 × 10
−5
and (b) 1.0 × 10
−4
s
−1
.
Grain sizes
+1.14E– 02
+1.15E– 02
+1.16E– 02
+1.16E– 02
+1.17E– 02
+1.18E– 02
+1.18E– 02
+1.19E– 02
(a)
Grain sizes
+7.93E–03
+8.06E–03
+8.20E–03
+8.33E–03
+8.60E–03
+8.74E–03
+8.88E–03
+9.01E–03
(b)
Plate 3 Average grain size ﬁelds at the end of the superplastic forming carried out with target maximum
strain rates of (a) 1.0 × 10
−5
and (b) 1.0 × 10
−4
s
−1
.
(c)
Number of
creep voids
Cavitation
damage parameter
0
+0.00E+00
+2.00E–02
+4.36E–02
+6.73E–02
+9.09E–02
+1.15E–01
+1.38E–01
+1.62E–01
+1.85E–01
+2.09E–01
+2.33E–01
+2.56E–01
+2.80E–01
+3.00E–01
1–3
4–6
7–9
10–12
13
(a)
(b)
1
2
2
4
3
6
2
3
5
9
4
7
8
7
4
1
9
15
12
2
2
22
17
5
6
7
5
2
8
30
2
3
2
0
0
1
0
0
4
0
2
Plate 4 Creep damage ﬁelds (a) predicted by the model, (b) observed in the microstructure, and
(c) from a surface void count using the micrograph.
Introduction to Computational Plasticity
This page intentionally left blank
Introduction to Computational Plasticity
FIONN DUNNE AND NIK PETRINIC
Department of Engineering Science Oxford University, UK
1
Great Clarendon Street, Oxford ox2 6dp Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide in Oxford New York Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid Melbourne Mexico City Nairobi New Delhi Shanghai Taipei Toronto With ofﬁces in Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries Published in the United States by Oxford University Press Inc., New York © Oxford University Press, 2005 The moral rights of the authors have been asserted Database right Oxford University Press (maker) First published 2005 Reprinted 2006 (with corrections) 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, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this book in any other binding or cover and you must impose the same condition on any acquirer British Library Cataloguing in Publication Data (Data available) Library of Congress Cataloging in Publication Data (Data available) Typeset by Newgen Imaging Systems (P) Ltd., Chennai, India Printed in Great Britain on acidfree paper by Biddles Ltd, King’s Lynn ISBN 0198568266 (Hbk) 9780198568261
3
3 5 7 9 10 8 6 4 2
To Hannah and Roberta. with love .
This page intentionally left blank .
Our aims have been to encourage development of understanding. P. The ﬁrst introduces microplasticity and covers continuum plasticity. together. creep. D. and the implementation of the constitutive equations. The book is in two parts.Preface The intention of this book is to bridge the gap between undergraduate texts in engineering plasticity and the many excellent books in computational plasticity aimed at more senior graduate students. We also describe a number of practical applications of the plasticity models introduced to demonstrate the reasonable maturity of continuum plasticity in engineering practice. and N. September 2004 F. is to develop a good physical feel for the plasticity models and equations described by considering. implicit and explicit integration of plasticity constitutive equations. We hope. the kinematics of large deformations and continuum mechanics. with the tests necessary to verify the implementation. researchers. the ﬁnite element method. we hope to provide a reasonably physical understanding of some of the large deformation quantities (such as the continuum spin) and concepts (such as objectivity) which are often unfamiliar to many undergraduate engineering students who demand more than just a mathematical description. at every stage. graduates. and the associated material Jacobian. that this book will help all those—undergraduates. the implementation into the commercial code ABAQUS is addressed (and to help. The second part of the book introduces a range of plasticity models including those for superplasticity. and thermomechanical fatigue (TMF). above all. we provide a range of ABAQUS material model UMATs). wherever possible. E. and ease of passage to the more advanced texts on computational plasticity. . Our intention. importantly. and practising engineers working in solid mechanics. researchers. P. porous plasticity. cyclic plasticity. and practising engineers—who need to move on from knowledge of undergraduate plasticity to modern practice in computational plasticity. In particular. the simpliﬁcation of the equations to uniaxial conditions. into ﬁnite element software. In addition.
This page intentionally left blank .
7 3.1 2.1 1. Microplasticity and continuum plasticity 1.3 3. 1.3 1. 2.6 2.2 2.Contents Acknowledgements xii Notation xiii Part I.2 3.4 3.4 Microplasticity Introduction Crystal slip Critical resolved shear stress Dislocations Further reading Continuum plasticity Introduction Some preliminaries Yield criterion Isotropic hardening Kinematic hardening Combined isotropic and kinematic hardening Viscoplasticity and creep Further reading Kinematics of large deformations and continuum mechanics Introduction The deformation gradient Measures of strain Interpretation of strain measures Polar decomposition 3 3 5 7 8 10 11 11 11 17 23 27 36 38 45 47 47 48 49 52 57 2.2 1. 3.4 2.1 3.5 2.5 .3 2.
5.2 4. 6. rate of deformation.9 Velocity gradient.1 7.x Contents 3.6 3. 7.1 4.3 4.8 3.6 6. and continuum spin Elastic–plastic coupling Objective stress rates Summary Further reading The ﬁnite element method for static and dynamic plasticity Introduction Hamilton’s principle Introduction to the ﬁnite element method Finite element equilibrium equations Integration of momentum balance and equilibrium equations Further reading Implicit and explicit integration of von Mises plasticity Introduction Implicit and explicit integration of constitutive equations Material Jacobian Kinematic hardening Implicit integration in viscoplasticity Incrementally objective integration for large deformations Further reading Implementation of plasticity models into ﬁnite element code Introduction Elasticity implementation Veriﬁcation of implementations Isotropic hardening plasticity implementation Large deformation implementations Elastoviscoplasticity implementation 60 66 69 81 82 83 83 84 96 100 136 142 143 143 143 150 154 161 167 168 169 169 170 171 172 176 180 4. Plasticity models 7.1 6.4 6.3 5.2 Superplasticity Introduction Some properties of superplastic alloys 185 185 185 .2 6.7 3.6 Part II. 4.5 5.4 4.4 5.5 5.5 6.3 6.1 5.2 5.
creep.2 Constitutive equations for cyclic plasticity 10.1 189 192 197 199 199 201 205 207 209 209 209 212 217 218 219 219 219 222 227 229 231 232 233 235 8. 9.3 10.Contents xi 7. and TMF 10.1 8.1 Introduction 10. Cyclic plasticity.3 7.1 9.3 9.4 Constitutive equations for superplasticity Multiaxial constitutive equations and applications References Porous plasticity Introduction Finite element implementation of the porous material constitutive equations Application to consolidation of Ti–MMCs References Creep in an aeroengine combustor material Introduction Physically based constitutive equations Multiaxial implementation into ABAQUS References Appendix 9. 8.2 9.2 8.3 Constitutive equations for C263 undergoing TMF References Appendix A: Elements of tensor algebra Differentiation The chain rule Rotation Appendix B: Fortran coding available via the OUP website Index 239 .
to Paul Buckley for the provision of the ﬁgures in Chapter 1. Oxford.14. UK. UK.1–10.1–8. Figures 8. and to Jinguo Lin for permission to use Figures 7. London. Figures 9. 10.3: Elsevier Ltd.9–7. with gratitude.9–7. The authors acknowledge. .4.6: Institute of Materials Communications Ltd.1–9.14: Elsevier Ltd. UK.Acknowledgements The authors would like to express their sincere gratitude to Esteban Busso for reading a draft and providing many helpful comments and suggestions. Oxford. permission granted to reproduce the following ﬁgures: Figures 7.
vector and tensor functions. v. Operations f (·) det[·] Tr[·] ln(·) [·] ∂ [·] ∂x ∇(·) = grad[·] div[·] = tr[∇(·)] x·y x⊗y σ :ε √ u = √u · u A = A : A function of (·) determinant of [·] trace of [·] logarithm of (·) increment of [·] partial derivative of [·] with respect to x gradient of [·] divergence of [·] scalar product of vectors dyadic product of vectors double contraction of tensors norm of vector norm of tensor Some commonly used notation C D E E ε fourthorder tensor of material constants rate of deformation tensor Lagrangian strain tensor Young’s modulus strain tensor . A. σ . . . . . • Bold italic typeface (P . . . . vectors. .): fourth order tensors. I. .Notation • Regular italic typeface (v. σ.): points. tensors. c.): scalars. scalar functions. • Helvetica bold italic typeface (C.
xiv Notation F f ρ I J K M ν P P R R σ t t u ˙ u ¨ u W deformation gradient force vector ﬁeld density secondorder identity tensor Jacobian stiffness matrix mass matrix Poisson’s ratio material particle material point ∈ Rn rotation tensor real set Cauchy stress tensor time surface traction vector displacement vector ﬁeld velocity vector ﬁeld acceleration vector ﬁeld work .
Part I. Microplasticity and continuum plasticity .
This page intentionally left blank .
The grain size is about 100 µm. crystal slip. this is crystallographic slip. The grain boundaries demarcate regions of different crystallographic orientation.1 and 1.2: (1) plastic slip does not lead to volume change. 1. we can then visualize plastic deformation taking place as shown in Fig.2(a) to conﬁguration 1. (2) plastic slip is a shearing process. we introduce grains. that is.2(a) and (b). 1. slip requires the breaking and reforming of interatomic bonds and the motion of one plane of atoms relative to another. as shown in Fig. The origin of plasticity in crystalline materials is crystal slip. and in particular. hydrostatic stress. . the structure is unchanged except at the extremities of the crystal. this is the incompressibility condition of plasticity.1 in which the ‘crystal’ or grain boundaries can be seen. (3) in a polycrystal. Unlike elastic deformation. 1. While short.2(b). After shearing the crystal from conﬁguration 1. can often be assumed not to inﬂuence slip. made up of many crystals in which atoms are stacked in a regular array. incompressibility. If we represent the crystallographic structure of a tiny region of a single grain by planes of atoms. attempts to explain the usual assumptions in simple continuum plasticity of isotropy. As we will see later. A number of very important phenomena in macroscopic plasticity become apparent from just two Figs 1. the minimum knowledge of microplasticity for users of continuum plasticity. A typical micrograph for a polycrystalline nickelbase superalloy is shown in Fig. plastic yielding is often an isotropic process. involving only the stretching of interatomic bonds.1. Metals are usually polycrystalline. resolved shear stress. and independence of hydrostatic stress. at the macrolevel.1 Introduction This chapter brieﬂy introduces the origins of yield and plastic ﬂow. and dislocations. Microplasticity 1. the incompressibility condition is very important in macroscale plasticity and manifests itself at the heart of constitutive equations for plasticity. slip systems.2(a).
the hydrostatic stress becomes ever larger but causes no more than the atoms in the theoretical material to come closer together. It will never in itself be able to generate the shearing necessary for crystallographic slip. but that the orientation changes randomly from one . Figure 1. If we take a sample of the theoretical material shown schematically in Fig. The fact that plastic slip is a shearing process gives more information about the nature of yielding. Consequently. it tells us that plastic deformation is independent of hydrostatic stress (pressure). A porous metal. in principle. 1. for example. 1. but from the pore closure. For nonporous metals.2(a) and submerge it to an ever deeper depth in an imaginary sea of water.1 shows a micrograph of a polycrystal. for example.4 Microplasticity 100 mm Fig. there is a change of volume and a dependence on hydrostatic stress.2 Schematic representation of the crystallographic structure within a single grain undergoing slip. 1. this is one of the cornerstones of yield criteria. not all plastic deformation processes are incompressible. However. is one in which the initiation of macroscale yield is quite independent of hydrostatic stress. If we assume that there is no preferred crystallographic orientation. under compressive load may undergo plastic deformation during which the pores reduce in size. However. the volume change does not originate from the plastic slip process itself.1 Micrograph of polycrystal nickelbase alloy C263. (a) τ (b) τ τ τ Fig. The von Mises criterion.
1. 1. The planes that can be seen are those on which slip has occurred resulting from many hundreds of dislocations running through the crystal and emerging at the edge. Figure 1.2. with the imposed axial displacement.Crystal slip 5 (a) (b) Fig.3(a) is a few millimetres in width and has been loaded beyond yield in tension. The combination of a slip . 1. the displacements become large. we can get a reasonable physical feel that macroscale yielding of the material will be isotropic. In order to retain compatibility.2 Crystal slip The evidence for crystal slip being the origin of plasticity comes from mechanical tests carried out on single crystals of metals. then.1 Slip systems: slip directions and slip normals Observations on single crystals show that slip tends to occur preferentially on certain crystal planes and in certain speciﬁc crystal directions. the lateral motion would have been prevented.3 (a) Photograph of a single zinc crystal and (b) a schematic diagram representing single slip in a single crystal. the slip planes would have to rotate towards the loading direction. but also to rotation of the crystallographic lattice. but with many such dislocations.3(a). and if our sample of material contains a sufﬁciently large number of grains. This is a further cornerstone of the von Mises yield criterion. The uniaxial loading therefore leads not only to crystallographic slip. Had this test been carried out in a conventional uniaxial testing machine. 1. 1. It can be seen that single slip in this case leads to the horizontal displacement of one end relative to the other.3(b) shows schematically what is happening in Fig. grain to the next. Each dislocation contributes just one Burger’s vector of relative displacement. The single crystal of zinc shown in Fig. The ends of the test sample have not been constrained in the lateral directions.
1 shows the slip systems found in single crystals of some of the important fcc and bcc pure metals together with the resolved shear stress required to cause slip—the critical resolved shear stress (CRSS). However. there are three families of slip systems operative: ¯ 111 {101}. ¯ 111 {123}.4 The particular slip system (111)[110] in an fcc lattice. In face centred cubic (fcc) materials. In body centre cubic (bcc) crystals.5. Table 1. and hence there are several families of planes on which slip occurs. . 1. ¯ The full family of slip systems in such crystals may be written as 110 {111}. 1.6 Microplasticity (111) [110] – ¯ Fig. ¯ 111 {112}. since the atoms are closest ¯ along the [111] direction and those equivalent to it. In fact the crystal is ‘closepacked’ in these planes.4. (101) (112) (123) (111) – Fig. These tend to be the most densely packed planes and the directions in which the atoms are packed closest together. The slip systems observed in bcc crystals are shown in Fig. There are 12 such systems in an fcc crystal (four planes each with three directions). the observed slip systems are shown in Fig. plane and a slip direction is called a slip system. there is no ambiguity about the slip direction. Thus. 1.5 Slip systems in bcc materials. the most densely packed planes are the diagonal planes of the unit cell. 1. there are several planes that are of similar density of packing. This is explained in terms of dislocations.
7 were obtained from tensile tests on cadmium single crystals which are hexagonal close packed (hcp).6. 111 {123} Fig. When the applied tensile stress is σ .3 Critical resolved shear stress Suppose a single crystal in the shape of a rod is tested in tension. that is. (1.64 0. . the shear stress acting on the slip plane and in the slip direction is τ which may be found as follows: if the crosssectional area of the rod is A. 1. 111 ¯ 111 ¯ ¯ 111 {101}. The data shown in Fig. 1. the crystal will yield. normal in direction of unit vector n. Metal Cu Al Ni αFe Mo Ta Structure fcc fcc fcc bcc bcc bcc Slip systems ¯ 110 ¯ 110 ¯ 110 ¯ ¯ 111 {101}.6 A single crystal containing slip plane with normal n. Schmid’s law provides a good explanation for the observed behaviour. 111 t Slip plane normal n s f l s Slip direction 7 CRSS (MPa) 0.8 32 50 50 {111} {111} {111} ¯ {112}. as shown in Fig. This is known as Schmid’s law.Critical resolved shear stress Table 1. The axis of the rod is parallel to unit vector t.40 5. 1. Hence the resolved shear stress is τ = σ cos φ cos λ = σ (t · n)(t · s). It has an active slip direction parallel to unit vector s. when the shear stress is maximized.1) Slip will take place on the slip system. The measured yield stress depends upon the angle between the tensile loading direction and the basal plane. 111 {123} {101} ¯ {112}. and loaded in direction t. the force in the slip direction is Aσ cos λ and it acts on an area A/cos φ of the slip plane. when τ reaches the CRSS. The crystal has an active slip plane. The minimum stress to cause yield occurs when the tensile axis is 45◦ to this plane. 1. slip direction s.1 Slip systems and CRSS for some pure metals. that is.
there are two extreme cases.7 The dependence of yield stress on the angle between the tensile axis and basal plane in an hcp cadmium single crystal. and the dislocation corresponds to the edge of a missing halfplane of atoms.1. b Fig. Thus b and . of course. 1. τth = 2π The expected shear strength is therefore very large. Figure 1. Each dislocation is associated with a unit of slip displacement given by the Burgers vector b. calculated assuming that the shear is homogeneous (the entire crystal shears simultaneously on one plane). wrong and plastic deformation in crystals normally occurs by the movement of the line defects known as dislocations. is given by G . The glide of a dislocation involves only very local rearrangements of atoms close to it. 1. The assumption of homogeneous shear is. Since the dislocation is a line defect. In this case the Burgers vector b is perpendicular to the line of the dislocation.8 shows a schematic representation of an edge dislocation. many orders of magnitude greater than the CRSS values are shown in Table 1. 1.8 Microplasticity 7 6 Yield stress (MPa) 5 4 3 2 1 0 0 30 60 90 Angle between tensile axis and basal plane Data Schmid’s law Fig. that are usually present in large numbers. and requires a stress much lower than τth . thereby explaining the low observed values of CRSS.8 An edge dislocation.4 Dislocations The theoretical shear strength of a crystal.
All of these subjects can be found in existing materials text books. Nor have we discussed the effect of temperature on diffusivity which inﬂuences or controls many plastic deformation mechanisms including thermally activated dislocation climb leading to recovery.2) Figure 1. Our aim has been to include sufﬁcient material on microplasticity to ensure that the physical bases of at least some of the assumptions made in macrolevel continuum plasticity are understood.g. Here. This is the other extreme case where the Burger’s vector is parallel to the dislocation line. that. a screw dislocation can cause slip on any slip plane containing l. There have been many developments over the last 25 years in physically based microplasticity modelling. the slip on each active system can be determined from which the overall total deformation can be found. It follows from the fact that b is parallel to l. If l is a unit vector parallel to the dislocation line. then s= b b and n= b×l . We have said nothing about the important role of dislocations in strain hardening and softening. Dislocations are therefore vital in understanding yield and plastic ﬂow. Such models are being used successfully for the plastic and creep deformation of single crystal materials which ﬁnd application in aeroengines.9 shows a schematic representation of a screw dislocation. and grain boundary sliding. including the development of timeindependent and ratedependent crystal plasticity.9 A screw dislocation. vacancy core and boundary diffusion. Hence. b (1.Dislocations 9 b Fig. using ﬁnite element techniques. 1. The modelling of single crystal components has become . fcc). for a given crystallographic lattice (e. although the slip direction s is deﬁned. the resolved shear stresses on all slip systems can be determined to ﬁnd the active slip systems. some of which are listed below. Within either a timeindependent or ratedependent formulation. The dislocation itself corresponds to a line of scissorslike shearing of the crystal. the dislocation line deﬁne a plane: a speciﬁc slip plane. the slip plane n is not.
and occurring under inservice conditions at localized regions of a component. This is now done for all the grains in the polycrystal. the active slip systems can be identiﬁed and the corresponding slips determined to give the overall deformation within a given grain. McGrawHill Book Co.W. (1999). H.O. In order to do this. using ﬁnite element techniques. it is necessary to generate many ﬁnite elements within each grain.10 Microplasticity possible with the development of highperformance computing. Mechanics and Materials. the understanding of microplasticity discussed above is also employed and again... leave microplasticity and address. London. the plastic deformation occurring in both the processing to produce engineering components. subject to the requirements of equilibrium and compatibility which are imposed within the ﬁnite element method. as in single crystal plasticity. Armstrong. We will now. It is clear that for large numbers of grains. . which all have their own measured or speciﬁed crystallographic orientations.E. New York. such polycrystal plasticity modelling becomes untenable. Here. Meyers. G.A. R. More recently. Currently. it is not (for the foreseeable future) going to be feasible to carry out polycrystal plasticity modelling at the engineering component level.. in the remainder of the book. John Wiley & Sons Inc. The consequence is that while desirable. This is particularly so in engineering industry where pressures of time and cost demand rapid analyses. polycrystal plasticity models have been developed.. continuumlevel plasticity. and for many years to come.K. (1988). Fundamentals and Linkages. therefore. Further reading Dieter. Mechanical Metallurgy. will continue to be modelled using continuumlevel plasticity. and Kirchner. M.
1 Introduction This chapter introduces the fundamentals of timeindependent and ratedependent continuum. also shown in the ﬁgure. at a strain of ε. or phenomenological plasticity: multiaxial yield. It is called hardening because the stress is increasing relative to perfect plastic behaviour. Plasticity commences at a uniaxial stress of σy . consistency condition. and simple constitutive equations for viscoplasticity and creep. We assume throughout the chapter that we are dealing with small strain problems in the absence of large rigid body rotations. isotropic and kinematic hardening. 2.1 shows the idealized stress–strain behaviour which might be obtained from a purely uniaxial tensile test.2.2 Some preliminaries 2. the material would cease to deform plastically (at least in the absence of timedependent effects) and would show a linearly decreasing stress with strain such that the gradient of this part of the stress–strain curve would Linear strain hardening s sy E E s Perfect plasticity « «p «e Fig. . normality hypothesis. If. 2.1 Strain decomposition Figure 2.1 The classical decomposition of strain into elastic and plastic parts.2. the loading were to be reversed. Continuum plasticity 2. The kinematics of large deformations are left until the following chapter. after which the material strain hardens.
the strains achieved can be very large indeed. and the strain rate in the Y direction is therefore 1 ˙ εY = y. (2. and the elastic strains are of order σ/E).3) This is easily veriﬁed by considering a cube of material. and dividing by xyz gives y ˙ z ˙ x ˙ + + = 0. is the elastic strain and it can be seen that the total strain. for example. (2. even in forming processes (you can estimate this from the measured forces to give a stress. 2.1) This is called the classical additive decomposition of strain. εe . . εp . the strain remaining in the test specimen is the plastic strain. 2. ˙ ˙ p p p (2. E. It is also apparent from Fig. shown in Fig. ε.1 that the stress achieved at a strain of ε is given by σ = Eε e = E(ε − εp ). The consequence of this is that the sum of the plastic strain rate components is zero: ˙ εX + εY + εZ = 0.2. The recovered strain.2. the deformation takes place without volume change. and of order 1–4.2) In many practical situations.3).1. 2. Once a stress of zero is achieved (provided the material remains elastic on full reversal of the load). Differentiating both sides with respect to time.001 (the 0. Compare the magnitude of this strain with that of typical elastic strains of order 0.12 Continuum plasticity again be the Young’s modulus.2 Incompressibility condition We saw in Chapter 1 that plastic deformation satisﬁes the incompressibility condition. it is entirely reasonable to make the assumption that ε e ≈ 0 so that εp ≈ ε. with dimensions shown in Fig.4) x y z The strains are deﬁned by x εX = ln x0 and similarly for εY and εZ . uniform deformation (or simply argue that the strains are very large so that the elastic components are negligibly small). (2. 2. ˙ y Equation (2.1% proof strain) which are generated in metals. which undergoes purely plastic.4) is therefore the incompressibility condition given in Equation (2. is the sum of the two ε = εe + εp . In such circumstances. particularly in materials processing operations such as forging or superplastic forming. Constancy of volume requires xyz = x0 y0 z0 . that is.
Y . monotonically increasing load is straightforward: If σ < σy then the material is elastic.Some preliminaries 13 y y0 z0 Y Z X x x0 z Fig. one in which more than one direct stress exists. is that of von Mises (to which we will return later). An effective plastic strain rate. A whole range of multiaxial yield criteria exist. 2. similarly. 2. ε 3 σe = . of course. (2. is deﬁned. The origin of the terms in Equation (2. 2.5) then becomes apparent. Mohr’s circle tells us that in a given plane. and Z to indicate direction. incompressible.2 An element of material undergoing plastic. σe is. the effective stress is deﬁned as 1 σe = √ [(σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ3 − σ1 )2 ]1/2 . (2. that is.7) [(˙ 1 − ε2 )2 + (˙ 2 − ε3 )2 + (˙ 3 − ε1 )2 ]1/2 .3 Effective stress and plastic strain rate Identifying the yield condition for uniaxial. the maximum shear stress is τ = (σ1 − σ2 )/2 so that with γ = τ/G.2. the elastic shear energy per unit volume is τ γ /2 = τ 2 /2G = (σ1 − σ2 )2 /8G. particularly for computational analysis. as ˙ √ 2 p p p p p p p= ˙ ˙ ε ˙ ε ˙ (2. It is not quite so straightforward for a multiaxial stress state. and 3 in an equivalent way to X.6) 2 where we use the numerical subscripts 1. elongation in the Y direction. p. 1/2 3 2 2 2 2 2 2 (σ11 + σ22 + σ33 + 2σ12 + 2σ23 + 2σ31 ) . In terms of principal stresses. The most commonly used in engineering practice. which relies on the knowledge of an effective stress.5) 2 or in terms of direct and shear stresses. a scalar quantity and its origin lies in the postulate that yielding occurs when a critical elastic shear energy is achieved. sometimes called (von Mises) equivalent stress. If σ ≥ σy then the material has yielded.
3 (2. of two secondorder tensors. is identical to the ˙ uniaxial plastic strain rate. Let us look at this in detail for the case in which a test specimen is loaded uniaxially upto large plastic strain (so that ε e ε p . σm = 1 (σ11 + σ22 + σ33 ). The deviatoric stress can be seen to be the difference between the stress and the mean stress. the effective stress σe is identical to the uniaxial stress and the effective plastic strain rate.10) . The coefﬁcients in Equations (2. 2.7) (and equivalently in Equations (2. We will now use Equation (2.5) and (2. The 3 symbol ‘:’ in Equations (2.9) is called the double contracted product. the latter is often called the hydrostatic stress. 1 Tr(σ )I .7).9)) are chosen to ensure that under purely uniaxial loading.5) and (2. which will be deﬁned a little later in the chapter.3. shown schematically in Fig.9) to determine the effective stress and plastic strain rate.8) 1/2 ≈ 2 ˙ ˙ ε:ε 3 1/2 (2. p. and ε ≈ ε p ) under an applied stress σ11 .9) in which σ is the deviatoric stress given by σ =σ− for example 1 σ11 = σ11 − (σ11 + σ22 + σ33 ) 3 1 2 = σ11 − (σ22 + σ33 ) 3 3 and σ12 = σ12 − 0 ≡ σ12 . or double dot product.14 Continuum plasticity Writing the stress and plastic strain rate tensors (dropping the plastic strain rate components) as ⎛ ⎞ ⎛ σ11 σ12 σ13 ε11 ε12 ˙ ˙ ⎝σ21 σ22 σ23 ⎠ and ε = ⎝ε21 ε22 ˙ σ = ˙ ˙ ε31 ε32 ˙ ˙ σ31 σ32 σ33 then the effective stress and plastic strain rate may be written as σe = p= ˙ 3 σ :σ 2 2 p p ˙ ˙ ε :ε 3 1/2 superscript p on the ⎞ ε13 ˙ ε23 ⎠ ˙ ε33 ˙ (2. The equations in (2. respectively.9) for the effective stress and plastic strain rates are a useful and compact way of writing the equations given in (2.
This is deﬁned.Some preliminaries s 15 1 2 s 3 Fig. ⎞ ⎛2 0 0 3 σ11 σ =⎝ 0 0 ⎠.11) Because of symmetry. The incompressibility condition leads to the requirement that ε11 + ε22 + ε33 = 0. for uniaxial loading. σ11 = σ.9) to determine the effective stress for this trivial case.3 Uniaxial loading of a schematic test piece.10) as 1 σ11 = σ11 − (σ11 + σ22 + σ33 ) 3 2 = σ11 3 1 σ22 = σ22 − (σ11 + σ22 + σ33 ) 3 1 = − σ11 . and all the shear components are zero. but we ﬁrst need to ﬁnd the contracted tensor product. − 1 σ11 3 1 0 0 − 3 σ11 (2. 3 Similarly.12) We can now use Equation (2. The deviatoric stress components can be determined using ˙ ˙ Equation (2. ˙ ˙ ˙ (2. ε22 = ε33 so that incompressibility gives ε22 = ε33 = ˙ ˙ ˙ ˙ 1 1 − 2 ε11 = − 2 ε. 2. 1 σ33 = − σ11 3 and σ12 = σ23 = σ33 = 0 so that the deviatoric stress tensor becomes. In these circumstances. σ : σ . for two . σ22 = σ33 = 0.
9) are not given as deviatoric quantities. σe ≡ σ . Note that the plastic strain rates appearing ˙ ˙ in Equation (2. in themselves. With the incompressibility condition and symmetry conditions discussed above the plastic strain rate tensor becomes ⎛ ⎞ ⎛ ⎞ ε11 ˙ ε ˙ 0 0 0 0 1 1 ˙ ε = ⎝ 0 − 2 ε11 ˙ 0 ⎠ = ⎝0 − 2 ε ˙ 0 ⎠ 1 1 ˙ ˙ 0 0 − 2 ε11 0 0 −2ε so that p= ˙ = 2 ˙ ˙ ε:ε 3 23 2 ε ˙ 32 1/2 = 1/2 2 2 1 2 1 2 ε + ε + ε ˙ ˙ ˙ 3 4 4 1/2 = ε. multiply component by component and sum the terms to give a scalar quantity. ε11 ≡ ε11 . p = ε. (2. For a uniaxial stress state. σ : σ is therefore given by 4 2 1 2 1 2 6 2 2 σ : σ = σ11 + σ11 + σ11 = σ11 ≡ σ 2 9 9 9 9 3 and so σe = 3 σ :σ 2 1/2 = 32 2 σ 23 1/2 ≡ σ. therefore. deviatoric due to the incompressibility condition. therefore. A and B. For example. ˙ For uniaxial loading. consider the deviatoric component ε11 of plastic ˙ strain rate. Let us follow the same procedure for the effective plastic strain rate. ε . ˙ ˙ . therefore.13) That is. ˙ ˙ ˙ ˙ ε 3 Because of the incompressibility condition. 1 ε11 = ε11 − (˙ 11 + ε22 + ε33 ).16 Continuum plasticity secondorder tensors. by n n A:B= i=1 j =1 Aij Bij . that is. This is because the plastic strain rate components are.
nonporous materials. . . Since f is independent of σm . J2 . It is apparent from this that hydrostatic stress has no effect on yield according to the von Mises criterion. and σ3 will never cause yield.5). Geometrically. . Let f be a yield function such that f = 0 is our yield criterion. . In Chapter 1. Let us consider the yield function in twodimensional principal stress space by putting σ3 = 0 and so imposing conditions of plane stress. Then: 1. Even inﬁnite. is deﬁned as J2 = ( 2 σ : σ )1/2 and it is for this reason that plastic ﬂow based on the von Mises yield criterion is often referred to as J2 plasticity. it must be expressible in terms of the deviatoric stresses σi (i = 1. .15) 1 The second stress invariant.Yield criterion 17 2. Yield stresses measured in compression have the same magnitude as yield stresses measured in tension. 3). 2. Yield in polycrystalline metals can be taken to be isotropic (provided we are concerned with yield in a volume of material containing many grains) and must therefore be independent of the labelling of the axes.14) and. σ2 . There are many others including that of Tresca and the Gurson model for porous materials.5)) and f < 0. . this corresponds to ﬁnding the intersection between the von Mises cylinder and the plane σ3 = 0. .3 Yield criterion Only the von Mises yield criterion is considered here. it can be seen that it satisﬁes the three requirements given above. with reference to Equation (2. . . The von Mises yield function is deﬁned by f = σe − σy = 3 σ :σ 2 1/2 − σy (2. 3) alone. since σe remains zero (see Equation (2. . The yield criterion is given by f < 0: Elastic deformation f = 0: Plastic deformation. Yield is independent of the hydrostatic stress. Thus f must be an even function of σi (i = 1. . but equal. 3.14) corresponds to a cylinder in threedimensional principal stress space with axis lying along the line σ1 = σ2 = σ3 . principal stresses σ1 . we saw that there were several general requirements for yield in isotropic. . Geometrically. . Thus f must be a symmetric function of σi (i = 1. Equation (2. (2. 3).
18
Continuum plasticity
d«p 2 Load point s2 sy
d«p d«p 1 Tangent to yield surface
Yield surface ( f = 0)
sy
sy
s1
sy
Elastic region
Fig. 2.4 The von Mises yield surface for conditions of plane stress, showing the increment in plastic
strain, dε p , in a direction normal to the tangent to the surface.
From Equations (2.14) and (2.5), f = σe − σy = 3 σ :σ 2
1/2
− σy
1 = √ [(σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ3 − σ1 )2 ]1/2 − σy , 2 which for plane stress, at yield, becomes 1 2 2 f = √ [(σ1 − σ2 )2 + σ2 + σ1 ]1/2 − σy = 0 2 so
2 2 2 σ1 − σ1 σ2 + σ2 = σy ,
(2.16)
which is the equation of an ellipse. The yield criterion is shown, for plane stress, in Fig. 2.4. Naturally, when σ1 = 0, σ2 = σy at yield and similarly for the other points where the ellipse intersects the lines σ1 = 0 and σ2 = 0. 2.3.1 The normality hypothesis of plasticity
We have now looked at the conditions necessary to initiate yielding. The question then is what happens after that if loading continues? After yield comes plastic ﬂow and the normality hypothesis of plasticity enables us to determine the ‘direction’ of ﬂow. For what is termed associated ﬂow, the hypothesis states that the increment in the plastic strain tensor is in a direction (i.e. relative to the principal stress directions) which is normal to the tangent to the yield surface at the load point. This is shown schematically in Fig. 2.4, but may be written in terms of the yield function, f , as dεp = dλ ∂f ∂σ or ˙ ˙ εp = λ ∂f . ∂σ (2.17)
Yield criterion
19
In these expressions, the direction of the plastic strain increment (or equivalently, plastic strain rate) is given by ∂f /∂σ while the magnitude is determined, for the ˙ plastic strain rate, by λ. This is called the plastic multiplier which we shall return to later. In order to look at this in more detail, and to make understanding easier, let us write the plastic strain rate and stress tensors in vector form (using what is known as Voigt notation, which we shall address in greater depth in Chapter 4). We will continue to consider principal components only, for the time being. Then for plane stress, σ= σ1 . σ2
Note that the vector representation of the stress, σ, is not italicized. This is the convention to be adopted throughout the book for stress and strain: an italicized bold σ and ε indicates a tensor, a nonitalicized bold symbol a vector, or Voigt notation. The yield function f , written in terms of principal stresses is, from Equation (2.16) 1 2 2 f = √ [(σ1 − σ2 )2 + σ2 + σ1 ]1/2 − σy = 0, 2 and we can then determine the direction of plastic ﬂow from Equation (2.17) as ⎞ ⎛ ∂f 1 2 (σ 2 − σ1 σ2 + σ2 )−1/2 (2σ1 − σ2 ) ⎜ ∂σ1 ⎟ ∂f ⎟= 2 1 ⎜ , = grad(f ) = ⎝ 1 2 −1/2 (2σ − σ ) 2 ∂f ⎠ ∂σ 2 1 2 (σ1 − σ1 σ2 + σ2 ) ∂σ2 2σ − σ2 which has direction −σ 1 + 2σ . This is normal to the yield surface at all points 1 2 (e.g. choose σ1 = σ2 = α, say, and the direction of the normal is clearly along the line σ1 = σ2 ). Let us now derive the plastic strain increment using the normality hypothesis with the von Mises yield criterion given in Equation (2.14), but using the other expression for effective stress for threedimensions given in Equation (2.5). We p will look at just one component, namely dε1 , to see the general pattern emerge. From Equation (2.17), the ﬁrst component of direction of plastic ﬂow is given by ∂f 1 1 = √ [(σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ3 − σ1 )2 ]−1/2 [2(σ1 − σ2 ) − 2(σ3 − σ1 )] ∂σ1 2 2 (3/2)[σ1 − (1/3)(σ1 + σ2 + σ3 )] = σe so ∂f 3 σ1 = ∂σ1 2 σe
20
Continuum plasticity
remembering the deﬁnition of deviatoric stress given in Equation (2.10). Similar results are obtained for the other components so that Equation (2.17) may be rewritten, for the von Mises yield criterion, as dεp = dλ 3 σ ∂f = dλ ∂σ 2 σe (2.18)
in which the stress and strain are now written as tensor quantities. Let us now look at the meaning of the plastic multiplier (at least, that is, for a von Mises material). We can use the expression for effective plastic strain rate in Equation (2.9) to write a similar expression for the increment in effective plastic strain as dp = 2 p dε : dεp 3
1/2 1/2
.
(2.19)
If we substitute Equation (2.18) into (2.19) we obtain dp = 23 σ 3 σ dλ : dλ 3 2 σe 2 σe = dλ ((3/2)σ : σ )1/2 σe
so that with Equation (2.9) we obtain dp = dλ and similarly, p = λ. ˙ ˙ (2.20) (2.21)
So, for a von Mises material, the plastic multiplier, dλ, turns out simply to be the increment in effective plastic strain. We can then rewrite the ﬂow rule in Equation (2.18) as σ 3 dε p = dp . (2.22) 2 σe In order for Equation (2.22) to be useful to us, we need to be able to calculate the increment in effective plastic strain, dp, or equivalently, the plastic multiplier, so that with prescribed loading, we can then calculate the increment in plastic strain components. We will address this next. 2.3.2 Consistency condition
Let us consider the case of uniaxial, tensile loading, for which the stress path taken relative to the yield surface is shown in Fig. 2.5. Starting from zero stress, σ2 then increases while the material deforms elastically until the load point (i.e. the point in stress space corresponding to the current loading) meets the yield surface at σ2 = σy . At this point, the material behaves plastically,
Yield criterion
s2 s2 sy
21
Load point
sy
sy
sy sy
s1
«2
Yield surface ( f = 0)
Fig. 2.5 The von Mises yield surface for plane stress and the corresponding stress–strain curve obtained for uniaxial straining in the 2direction.
with no hardening, as shown in the resulting stress–strain curve in Fig. 2.5. With further plastic deformation, the load point remains on the yield surface and the stress remains constant and equal to σ y . The requirement for the load point to remain on the yield surface during plastic deformation (at least, that is, for timeindependent plasticity—we will look at viscoplasticity later) is called the consistency condition, and it is this that enables us to determine the plastic multiplier, or equivalently for a von Mises material, the increment in effective plastic strain. The yield function given in Equation (2.14) includes a dependence on the stress components and the yield stress, σy . We will see later, when considering hardening, that the yield stress can increase (and sometimes decrease), and does so often as a function of effective plastic strain, p. It is therefore convenient to write the yield function as f (σ, p) = σe − σy = σe (σ ) − σy (p) = 0. (2.23) The consistency condition is written, for an incremental change in stress and effective plastic strain f (σ + dσ, p + dp) = 0. (2.24) We can expand this as f (σ + dσ, p + dp) = f (σ, p) + ∂f ∂f : dσ + dp. ∂σ ∂p (2.25)
Note that all terms in the equation are scalar. Let us work in principal stress space only. The advantage of this is simplicity; the disadvantage is that we ignore the complicating features of dealing with tensor versus engineering strain components. However, this will be dealt with in Chapters 4–6. We will therefore store stress and strain principal components as vectors, in Voigt notation, as discussed above. The product
22
Continuum plasticity
∂f /∂σ : dσ then becomes the scalar (dot) product, (∂f/∂σ)·dσ of two vectors (this is only true in principal stress space, as we shall see later) which gives a scalar. Combining Equations (2.23)–(2.25) gives ∂f ∂f dp = 0. · dσ + ∂p ∂σ (2.26)
We can use Hooke’s law in incremental form to relate the stress and elastic strains, written as column vectors, by dσ = C dεe = C(dε − dεp ) (2.27)
in which C is the elastic stiffness matrix. We will use the most general form for the plastic strain increment, given in Equation (2.17), for now, but simplify for the case of the von Mises yield criterion later. Substituting (2.17) into (2.27) gives dσ = C dε − dλ and (2.28) into (2.26) ∂f ∂f · C dε − dλ ∂σ ∂σ + ∂f dp = 0. ∂p (2.29) ∂f ∂σ (2.28)
We can obtain the most general form for dp, that is, without assuming a von Mises material for the time being, by using Equations (2.17) and (2.19) to give dp = 2 p dε : dεp 3
1/2
=
2 ∂f ∂f dλ : dλ 3 ∂σ ∂σ
1/2
=
2 ∂f ∂f dλ · dλ 3 ∂σ ∂σ
1/2
(2.30) for principal stress space, in which the contracted tensor product simpliﬁes to the scalar product. Substituting (2.30) into (2.29) and rearranging gives the plastic multiplier dλ dλ = (∂f /∂σ) · C dε (∂f /∂σ) · C(∂f /∂σ) − (∂f /∂p)((2/3)(∂f /∂σ) · (∂f /∂σ))1/2 . (2.31)
The stress increment can then be determined by substituting (2.31) into (2.28) to give dσ = C dε − = C−C ∂f (∂f /∂σ) · C dε ∂σ (∂f/∂σ) · C(∂f/∂σ) − (∂f/∂p)((2/3)(∂f/∂σ) · (∂f/∂σ))1/2
(∂f/∂σ) · C ∂f dε ∂σ (∂f/∂σ) · C(∂f/∂σ) − (∂f/∂p)((2/3)(∂f/∂σ) · (∂f/∂σ))1/2 (2.32)
34) where p and dp are given in (2. Let us just consider what happens in going from elastic behaviour to plastic.6 together with schematic representations of the initial and subsequent yield surfaces. in which the yield surface expands with plastic deformation. p. (2. 2. when deformed plastically. dλ = 0. which can be written as p= dp = p dt. expanded yield surfaces after plastic deformation Initial yield surface Fig. the hardening is referred to as isotropic. we will ﬁrst introduce isotropic hardening. 2.33) C ep is called the tangential stiffness matrix.4 Isotropic hardening Many metals.6 Isotropic hardening. In the absence of plastic deformation. respectively. A uniaxial stress–strain ˙ curve with nonlinear hardening is shown in Fig. so the load point moves in the σ2 direction from zero until it meets the initial yield surface at σ2 = σy . In this instance.32). Before looking at the plastic multiplier in more detail. 2. In the case of plastic deformation. . and in this case. hardening behaviour in Fig. that is. with knowledge of the total strain increment. the stress increment can be obtained from Equation (2. C ep ≡ C.6. ˙ (2. Yield occurs at this point. harden. the elastic stiffness matrix.19). In order s2 s2 r sy Saturation Hardening «2 sy sy sy sy s1 sy Yield Subsequent. and the corresponding uniaxial stress–strain curve.9) and (2.Isotropic hardening 23 or dσ = C ep dε. 2. the subsequent yield surface is shown expanded compared with the original. When the expansion is uniform in all directions in stress space. to gain better physical insight. the stress required to cause further plastic deformation increases. often as a function of accumulated plastic strain. Loading is in the 2direction.
is therefore (σy0 +Q). (2. and referring to Figs 2.23). With this hardening.37) (2.39) we may write dσ h and. 2. f (σ.4.6 and 2.6.31) and hence the stress increments in (2. Figure 2.7.7. and for this case. the yield surface must expand as σ2 increases.38) Q is the saturated value of r so that the peak stress achieved with this kind of hardening. which gives an exponential shape to the uniaxial stress–strain curve which saturates with increasing plastic strain. Let us now consider a slightly simpler form of isotropic hardening and determine the plastic multiplier in Equation (2. the stress increase due to isotropic hardening is just dr. so using Equation (2. dp = dεp . For uniaxial conditions. The amount of expansion is often taken to be a function of accumulated plastic strain. and for the load point to stay on the yield surface (the consistency condition requires this).36) in which σy0 is the initial yield stress and r(p) is called the isotropic hardening function. from Equation (2. gives r(p) = Q(1 − e−bp ). of course. The constant b determines the rate at which saturation is achieved.1 Linear isotropic hardening We write the linear isotropic hardening function as dr(p) = h dp (2. since integrating (2.32).37) with initial condition r(0) = 0.36). p. the increment in elastic strain is just dεp = dεe = dσ E . the yield function becomes that given in (2. 2.39) in which h is a constant.24 Continuum plasticity for hardening to take place. p) = σe − σy (p) = 0. where σy (p) might be of the form σy (p) = σy0 + r(p) (2.6 shows an example of the uniaxial stress–strain behaviour predicted using this kind of isotropic hardening function. we expect the uniaxial stress–strain curve to look like that shown in Fig. There are many forms used for r(p) but a common one is r (p) = b(Q − r)p ˙ ˙ or dr(p) = b(Q − r) dp (2. 2. shown in Fig.35) in which b and Q are material constants.
2.40) · dσ + ∂p ∂σ ∂p 3 ∂σ ∂σ ∂σ Rearranging (2.40) gives the plastic multiplier. (2. so that the total strain is dε = giving dσ = E 1 − dσ dσ + = dσ E h E+h . 2. in terms of the total strain increment.Isotropic hardening 25 Gradient = E s r sy «p sy « ( 1– E E+h ) Fig. the total strain increments are provided.2 Uniaxial loading with linear isotropic hardening Let us consider purely uniaxial loading. in order to obtain better physical insight into the process. Eh dε. However. returning to the plastic multiplier and combining Equations (2.4. for a material that yields according to the von Mises criterion. we will obtain the plastic multiplier in terms of the stress increment so that we can then examine the equations for a simple uniaxial problem using the von Mises yield criterion in Section 2. In this section.30). Now. and it is then necessary to calculate the corresponding stress increments. in the 1direction. (2. dσ .7 Stress–strain curve for linear strain hardening with dr = h dεp . we will apply a stress σ1 in the .26) and (2.31). it is more appropriate (and simpler) to write the plastic multiplier in terms of a known stress increment. often. step by step. in Equation (2. In other words. This is the most appropriate form when considering the development of computational techniques. we obtain ∂f ∂f ∂f 2 ∂f ∂f ∂f dp = · dσ + dλ · = 0. as −(∂f/∂σ) · dσ .4. but now in terms of the stress increment.41) dλ = (∂f/∂p)((2/3)(∂f/∂σ) · (∂f/∂σ)) 2. including the ﬁnite element method in which. E E+h The plastic multiplier has been derived.
This arises because we chose linear hardening with gradient h.18). 3 1 1 −2 −2 ⎛ ⎡ from which we can obtain (2.42) ∂r ∂f =− = −h. under uniaxial loading. ∂f/∂σ can be obtained using Equation (2.44) and (2. the yield function becomes f (σ.44) The numerator can be determined as follows: ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ 1 1 dσ1 dσ1 ∂f ⎢− 1 ⎥ ⎣ ⎦ ⎢− 1 ⎥ ⎣ ⎦ = dσ1 · dσ = − ⎣ 2 ⎦ · dσ2 = − ⎣ 2 ⎦ · 0 − ∂σ 1 1 0 dσ3 −2 −2 and substituting (2. therefore.43) ∂p ∂p For the von Mises yield criterion. and all other stresses are zero. We know. we will continue to work in principal stress space and represent the stresses in Voigt notation. p) = σe (σ ) − σy (p) = σe (σ ) − σy0 − r(p) = 0 (2.45) (2.41) gives the plastic multiplier as dλ = dσ1 . We will consider the denominator ﬁrst. with the deviatoric stress components taken from (2.46) For a von Mises material.26 Continuum plasticity 1direction. as ⎡ 2 ⎤ ⎡ ⎤ σ1 1 ∂f 3σ 3 1 ⎢ 31 ⎥ ⎢ 1 ⎥ = = ⎣− 3 σ1 ⎦ = ⎣− 2 ⎦ ∂σ 2 σe 2 σ1 1 −2 − 1 σ1 3 so that ∂f ∂p 2 ∂f ∂f · 3 ∂σ ∂σ ⎤ ⎡ ⎤⎞1/2 1 1 ⎜ 2 ⎢− 1 ⎥ ⎢− 1 ⎥⎟ = −h ⎝ ⎣ 2 ⎦ · ⎣ 2 ⎦⎠ = −h. .45) into (2. h (2. the stress is written in vector form as ⎡ ⎤ σ1 ⎣σ2 ⎦ . of course.47) This is telling us that the increment in stress for unit increase in plastic strain is h. that is. dλ = dp = dε1 = p (2. σ= σ3 We will now determine the plastic multiplier using Equation (2. (2. For simplicity.12). With the hardening function given in (2. h dσ1 .39).41).
Kinematic hardening 27 that because we are considering uniaxial loading and the incompressibility condition applies.8.5 Kinematic hardening In the case of monotonically increasing loading.48). (2. this is often not appropriate.39). Equation (2. obtain the other plastic strain components formally from Equation (2. 2.50). The plastic multiplier can.52) 2. and omitting the subscript 1 to indicate the uniaxial loading direction. it is often reasonable to assume that any hardening that occurs is isotropic. under uniaxial loading. Equation (2. and any further increase in load results in plastic deformation. 2. using (2. Figure 2.50) E h Rearranging (2.48) 2 2 h We can.51) shows the linear hardening obtained. corresponding to load point (1) shown in the ﬁgure. the load is reversed so that the material behaves elastically (the stress is now lower than the yield stress) and linear stress–strain behaviour results up until load point (2). for the isotropic hardening model chosen in Equation (2. a much smaller elastic region . At this point. gives E dσ = E 1 − dε. shown schematically in Fig. (2. with no hardening (h = 0).51) gives a stress increment of zero.7.17) as ⎡ ⎤ 1 dσ1 ⎢ 1 ⎥ ∂f p dε = dλ (2. the load point is again on the expanded yield surface. At a strain of εi . E+h (2. that is.51) E+h which is what we obtained earlier for linear strain hardening shown in Fig. of course.49) = ⎣− 2 ⎦ ∂σ h 1 −2 in agreement with (2. In fact.51) and (2. (2. the other plastic strain increments must be 1 p 1 dσ1 p p dε2 = dε3 = − dε1 = − . which is often not what would be seen in experiments. be rewritten as e dε1 = dε1 + dε1 = p dλ = E dε1 . on reversed loading. For the case of reversed loading.8(b) shows that isotropic hardening leads to a very large elastic region. If we choose perfect plasticity. however.46). The total strain increment in the loading direction is given by dσ1 dσ1 + . Consider a material which hardens isotropically.
8(b). 2. With continued loading.8 Reversed loading with isotropic hardening showing (a) the yield surface and (b) the resulting stress–strain curve. translated yield surface Fig. for the kinematic hardening in Fig.9 Kinematic hardening showing (a) the translation. In fact. (a) Load point (1) (b) s2 sy x  sy sy s1 s2 sy E E «2 sy Initial yield surface Load point (2) Subsequent. x of the yield surface with plastic strain. the stress increases until the yield stress. is achieved. This is shown in Fig. 2. and kinematic hardening.9(a). is expected and this results from what is often called the Bauschinger effect.9. In Fig. rather than expanding. 2. expanded yield surface Fig. the material deforms plastically and the yield surface translates. The elastic region is much smaller than that for isotropic hardening shown in Fig. 2. σy . 2.9. 2. and (b) the resulting stress–strain curve with shifted yield stress in compression—the Bauschinger effect. the load is reversed so that the material deforms elastically until point (2) is achieved when the load point is again in contact with the yield surface. When load point (1) is achieved. the yield surface translates in stress space. the elastic region . In kinematic hardening.28 Continuum plasticity (a) Load point (1) s2 (b) s2 sy sy E sy sy sy s1 E «i «2 Initial yield surface Load point (2) Subsequent.
that is 1 dε p = dεp − Tr(dεp ) ≡ dεp . as a vector. it is 1/2 3 (σ − x ) : (σ − x ) − σy (2. that the hardening variable depends linearly on the plastic strain.55) 3 . In the absence of kinematic hardening. using Voigt notation. it is 2(σy + r).5. whereas for kinematic hardening. r. it has the same components as stress. σ :σ 2 With kinematic hardening. we know that the plastic strain increment is a deviatoric quantity. the load point must always lie on the yield surface during plastic ﬂow. In order to understand Equation (2. Under applied loading and plastic deformation.Kinematic hardening 29 is of size 2σy whereas for the isotropic hardening. (2. Because of the incompressibility condition. the surface translates to the new location shown such that the initial centre point has been translated by x. The yield function describing the yield surface must now also depend on the location of the surface in stress space.54) 3 3 in which c is a material constant.53) 2 in which x is the kinematic hardening variable and is often called a back stress.1 Kinematic hardening under uniaxial loading We will take the increment in kinematic hardening to be proportional to the increment in plastic strain.54) has the similarity with isotropic hardening. Consider the initial yield surface shown in Fig. In the case of plastic ﬂow with kinematic hardening. just as for stress. and we can write it as a tensor.9. hence 2 2 ˙ ˙ dx = c dεp or equivalently. the increment in plastic strain has direction normal to the tangent to the yield surface at the load point. The major difference is that isotropic hardening is described by a scalar variable. 3 This is called Prager linear hardening. and the coefﬁcient of 2 will be discussed below. or. f = 2. the yield function written in terms of tensor stresses is 1/2 3 f = σe − σy = − σy . In addition. the hardening variable (or back stress) is a tensor. however.53). note that the consistency condition still holds. 2. Equation (2. normality still holds. x = cεp (2. Because it is a variable deﬁned in stress space. let us consider purely uniaxial loading with linear kinematic hardening. We now need to determine the stresses relative to the new yield surface centre to check for yield.
but also the other direct components as well. gives dx11 = dx = c dε11 ≡ c dp.60) appears in Equation (2.55).54). in Equation (2. dx is therefore also a deviatoric quantity so that for uniaxial loading in the 1direction. that is x11 . and in particular.58) so the uniaxial component of x.30 Continuum plasticity From Equation (2.59) and (2. is 2 times the magnitude of x.59) 3 3 It is often simpler. Bear 3 in mind that under uniaxial loading. ⎛ ⎛ p ⎞ ⎞ 0 0 dε11 dx11 0 0 2 ⎜ 1 1 p − 2 dε11 0 ⎟ = c⎜ 0 0 ⎟ − 2 dx11 dx = ⎝ 0 ⎝ ⎠ ⎠ 3 1 1 p 0 0 − 2 dx11 0 0 − 2 dε11 so that 2 2 p c dε11 ≡ c dp. therefore. Consider now the components of dx. .56) ⎠ 1 0 0 x33 0 0 − 2 x11 The magnitude or norm of x is deﬁned by x = x = (x : x)1/2 ⎡⎛ 0 x11 ⎢⎜ 0 − 1 x = ⎣⎝ 2 11 0 = 2 x11 (2. the effective stress is identical to the uniaxial applied stress. (2. and the effective plastic strain increment is identical to the uniaxial plastic strain increment. to write the equations in terms of the magnitude of x rather than the loading direction component. This is because x depends directly on the plastic strain. given in its evolution equation in (2.54).54). x = x = ⎝ 0 x22 0 ⎠ = ⎝ 0 − 2 x11 (2. uniaxial loading leads to the development of not only the uniaxial component of x. Combining Equations (2. for uniaxial loading.57) x11 ⎟ ⎜0 ⎠:⎝ 1 − 2 x11 0 0 0 1/2 ⎞ ⎛ 0 1 − 2 x11 0 0 1 − 2 x11 ⎞⎤1/2 ⎟⎥ ⎠⎦ 0 0 1 2 1 2 + x11 + x11 4 4 = 3 x11 2 (2.58). which is why the coefﬁcient of 2 3 p (2. Note also that like the plastic strain. we may write ⎞ ⎛ ⎞ ⎛ 0 0 x11 x11 0 0 ⎜ 1 0 ⎟. for uniaxial loading.
62) can be written f = f = σ − x − σy = 0. 2. therefore. σ 11 is just the uniaxial stress. further yield occurs if σ − x is equal to the yield stress. or Chaboche nonlinear kinematic hardening.9. σ . (2. Under uniaxial conditions.63) The physical interpretation of Equation (2. What often distinguishes them is how the direction of translation of the yield surface is chosen. σy .62) 2 Because we are considering here only uniaxial loading. of the back stress as ⎞ ⎛ 0 0 σ11 − x 2⎜ ⎟ 1 0 − 2 (σ11 − x) σ −x = ⎝ 0 ⎠ 3 1 0 0 − 2 (σ11 − x) so that ⎞ 0 0 (σ11 − x) 4⎜ ⎟ 1 0 0 − 2 (σ11 − x) (σ − x ) : (σ − x ) = ⎝ ⎠ 9 1 0 0 − 2 (σ11 − x) ⎞ ⎛ 0 0 (σ11 − x) ⎟ ⎜ 1 0 0 − 2 (σ11 − x) :⎝ ⎠ 1 0 0 − 2 (σ11 − x) 4 1 1 (σ11 − x)2 + (σ11 − x)2 + (σ11 − x)2 9 9 9 2 (σ11 − x)2 3 ⎛ = = so that 1/2 3 (σ − x ) : (σ − x ) − σy = σ11 − x − σy = 0. x. There are many forms of kinematic hardening available.53) for uniaxial loading as ⎛2 ⎞ 0 0 3 σ11 − x11 ⎜ ⎟ 1 σ −x =⎝ (2. and the rate of its evolution as a function of the plastic strain.61) 0 − 2 2 σ11 − x11 0 ⎠ 3 0 0 1 −2 2 3 σ11 − x11 which can be written in terms of the magnitude.63) can be seen in Fig. so Equation (2. We shall now look at what is often called Armstrong–Frederick. Plastic deformation leads to the translation of the yield surface in stress space. . (2.Kinematic hardening 31 We may now determine the terms necessary for the yield function in Equation (2.
γ (2. in Equation (2. . dx.2 Nonlinear kinematic hardening In its multiaxial form. taking x to be 0 at εp = 0. as dx = c dεp − γ x dεp . translated yield surface sy s1 sy «2 c/g s2 sy Fig. The constant γ is the time constant and determines the rate of saturation of stress. in Fig. for monotonically increasing plastic strain.65) 2 c dεp − γ x dp 3 (2. c/γ determines the magnitude. ˙ x= 2 p ˙ ˙ cε − γ xp 3 (2. Equation (2. for this nonlinear kinematic hardening.66) The resulting form of the stress–strain curve. is shown together with the translated yield surface. so the back stress. In its uniaxial form. 2. As the plastic strain increases. We will now examine the ﬂow rule for kinematic hardening. which can be integrated. Saturated stress s2 x sy E sy sy Initial yield surface Subsequent.66) saturates to the value c/γ giving a maximum saturated stress of σy + c/γ . x.5.64) may be written in terms of the magnitude. 2.10. in the back stress is given by dx = or equivalently. x.64) in which γ is a further material constant.10 Nonlinear kinematic hardening and the resulting nonlinear hardening stress–strain curve which saturates at stress c/γ .32 Continuum plasticity 2. the increment. to give x= c p (1 − e−γ ε ).
70) ∂σ ∂x which. we shall now work in principal stress space again. σ . and use Voigt notation.68) We now need to use the consistency condition to determine the plastic multiplier.69) for the case of kinematic hardening with a von Mises yield criterion. which are both tensor quantities.3 Flow rule with kinematic hardening We will use the normality hypothesis in Equation (2.69) gives ∂f ∂f · C dε − dλ ∂σ ∂σ = so that dλ = (∂f/∂σ) · C dε (∂f/∂σ) · C(∂f/∂σ) + γ (∂f/∂x) · x − (2/3)c(∂f /∂x) · (∂f /∂σ ) (2. let us use Equation (2. The normality hypothesis then gives dεp = dλ 3 σ −x ∂f . = dλ ∂σ 2 J (σ − x ) (2.27).67) = J (σ − x ) − σy . when combined with Hooke’s law in (2. (2. for simplicity. and the back stress. f .72) . First. let us write the yield function. as f = 3 (σ − x ) : (σ − x ) 2 1/2 − σy (2. x. The consistency condition becomes ∂f ∂f · dσ + · dx = 0. However. the normality hypothesis in (2. and Equation (2.53) to determine the ﬂow rule for plastic deformation with kinematic hardening.17) together with the yield function in (2.71) + ∂f · ∂x + 2 c dεp − γ x dλ 3 ∂f · ∂x 2 ∂f c dλ − γ x dλ = 0 3 ∂σ ∂f ∂f · C dε − dλ ∂σ ∂σ and the plastic strain increment is dεp = ∂f (∂f/∂σ) · C dε (∂f/∂σ) · C(∂f/∂σ) + γ (∂f/∂x) · x − (2/3)c(∂f/∂x) · (∂f/∂σ) ∂σ (2.68) just to show that dp = 2 p dε : dεp 3 1/2 = dλ [(3/2)(σ − x ) : (σ − x )]1/2 = dλ J (σ − x ) (2. First.Kinematic hardening 33 2.17). The yield function depends upon the stress.5.
73) becomes ∂f ∂f 2 ∂f · dσ − · c dλ − γ x dλ = 0 ∂σ ∂σ 3 ∂σ so that −(∂f/∂σ) · dσ dλ = .75) becomes just 2 γ x1 − c.76) c − γx In a similar manner to that for the isotropic hardening example.4. we can determine the terms in the denominator of Equation (2.74) γ (∂f/∂σ) · x − (2/3)c(∂f/∂σ) · (∂f/∂σ) We will examine this for uniaxial loading in a von Mises material in the following section. we can obtain the uniaxial stress increment as E dσ = E 1 − dε.5. ∂σ ∂x ∂σ ∂x 3 From the yield function in (2. Starting from the consistency condition in (2.73) · dσ + · dx = · dσ + · c dεp − γ x dp = 0.74) is.74) as follows: ⎡ ⎤ ⎡ ⎤ ⎤ ⎡ ⎤ 1 1 1 x1 ∂f 2 ∂f ∂f ⎢− 1 ⎥ ⎣ ⎦ 2 ⎢− 1 ⎥ ⎢− 1 ⎥ γ · = γ ⎣ 2 ⎦ · x2 − c ⎣ 2 ⎦ · ⎣ 2 ⎦ . 1 x2 = x3 = − x1 2 3 so that Equation (2.4 Simple uniaxial loading For uniaxial loading.67).77) E + c − γx . The numerator in (2. (2. we can write x1 = 2 x so that 3 this becomes γ x − c. ∂f/∂x = −(∂f /∂σ ) and dp = dλ and with (2.68). ·x− c 3 ∂σ ∂σ 3 ∂σ 1 1 1 x3 −2 −2 −2 ⎡ (2. Now.70) gives ∂f ∂f ∂f ∂f 2 (2. (2.27).75) We know that x is a deviatoric quantity so that for uniaxial loading in the 1direction. just −dσ1 so that the plastic multiplier for uniaxial loading is just dσ1 dλ = . 2. (2.5. as before. We will simplify this to the case of uniaxial loading to interpret the terms in Section 2. but ﬁrst.34 Continuum plasticity and so the stress increment can be obtained from Equation (2. we will start by obtaining the plastic multiplier in terms of stress rather than strain increments. (2.
be considered to be an incremental process. Plasticity must. =E 1− . If the material behaviour is purely elastic. e . the material undergoes nonlinear kinematic hardening. into (2.77) gives dσ = E dε. x. given in (2. and indeed plastic strain. This is simply not possible for the case of plasticity. on the history of prior deformation. then c = γ = 0 and then (2. in general. Before ﬁnishing this section.51) can be seen to be near identical.77) simply reduces to dσ = E dε.77) demonstrates quite nicely the incremental nature of plasticity.77) tells us that during plastic deformation. increases according to its evolution equation. the increment in stress is zero. The increment in stress. (2. no further stress increase occurs.66). so the exponential term diminishes until ultimately. the back stress. that is. Equation (2. Integrating this equation is clearly possible such that at any particular where dε = dε value of elastic strain. therefore. ﬁnally. Let us consider ﬁrst the case of elasticity only.77) to give E dσ = E 1 − dε E + c − γx E E dε = E 1 − dε. γ = 0.78) E+c which is linear during plastic deformation since c is constant. so that the increment in stress varies from step to step. that is. This follows from Equation (2. often depend then.64). This equation for linear kinematic hardening and that for linear isotropic hardening in Equation (2. then the plastic multiplier is zero and the righthand term in the bracket is zero so that (2. If we have elastic. Equation (2. no strain hardening.77) since x itself depends upon strain. so that the stress increment progressively decreases until saturation is achieved at which point. If. the material becomes perfectly plastic with dσ = 0. the stress can be calculated.Kinematic hardening 35 Let us look at this in a little more detail. The two types of hardening are in fact the same for monotonically increasing loading (provided c = h) and only differ under uniaxial conditions on a load reversal when the Bauschinger effect becomes important. p p E + c − γ [(c/γ )(1 − e−γ ε )] E + ce−γ ε As the plastic strain increases. This can be seen by substituting the equation for x. the stress–strain relation during plasticity becomes E dσ = E 1 − dε (2. for which all we can do (normally) is consider increments in stress resulting from an increment in strain. If the material undergoes linear kinematic hardening. perfectly plastic behaviour. in plasticity.
ﬁnally. as shown in Fig. The stress increases until yield is achieved at point A. 2.11.11 Combined kinematic and isotropic hardening. the strain reversal occurs so that the material becomes elastic .11. 2. 2. but over (normally) quite large numbers of cycles. kinematic hardening is the dominant hardening process such that the Bauschinger effect can be represented. This is particularly appropriate for applications to cyclic plasticity where within an individual cycle.12. Starting from the point of zero stress and strain.12 Strain imposed resulting in cyclic plasticity shown in Fig. and the material kinematically hardens leading to the translation of the yield surface.36 Continuum plasticity Expanded yield surface (isotropic hardening) after many cycles s2 s2 B sy sy A sy sy Initial yield surface sy s1 C D «2 Plasticity recommences Expanded hysteresis loop resulting from isotropic hardening Translated yield surface (kinematic hardening) Fig. the material also hardens isotropically such that the peak tension and compression stresses in a given cycle increase from one cycle to the next until saturation is achieved. Strain Time Fig. 2. 2. 2. 2. materials which harden both kinematically and isotropically.11. the material is subjected to the strain shown in Fig. Such a process is represented schematically in Fig.6 Combined isotropic and kinematic hardening We will consider. Once the peak strain is achieved.
in addition to the kinematic hardening. The stress– strain loop BCDB produced in this way is called a hysteresis loop. we will use the consistency condition as before.11.82) In a similar manner as before. In order to determine the plastic multiplier. and accumulated plastic strain as follows f = J (σ − x ) − r(p) − σy so that the consistency condition becomes ∂f ∂f ∂f · dx + dp = 0. occurs within each cycle. shown by the broken line hysteresis loop in Fig. The yield surface is translated again because of the kinematic hardening. as follows. Kinematic hardening. (∂f/∂σ)· C(∂f/∂σ)− γ (∂f/∂σ)· x +(2/3)c(∂f/∂σ)· (∂f/∂σ)+b(Q − r(p)) (2.64) and writing ∂f /∂p = −(∂r/∂p) from (2.80) (2. The yield function.81) can be rearranged to give the plastic multiplier. gives the plastic multiplier as dλ = (∂f/∂σ)· C dε . (2. given by Equations (2. then superimposed upon the translation of the yield surface is a progressive expansion. 2. Elastic deformation continues until the load point reaches the yield surface again at point C where plasticity recommences until the next strain reversal at point D. back stress. we can determine the plastic multiplier in terms of the increment of stress. We will consider the case of nonlinear kinematic and isotropic hardening.64) and (2.37). rather than strain.Combined isotropic and kinematic hardening 37 at point B. due to isotropic hardening. together with (2. on the other hand.79). Equation (2.80) is written as ∂f ∂f · · dσ + ∂x ∂σ 2 c dεp − γ x dp − b(Q − r(p)) dp = 0 3 . This process. by which the peak stress and strain in a hysteresis loop increase. If.17). Assuming von Mises behaviour. as it often occurs from cycle to cycle over many cycles. the material also isotropically hardens. dp = dλ.81) Following the procedure for isotropic and kinematic hardening. respectively. is often called cyclic hardening.27) and (2. 3 (2. · dσ + ∂x ∂p ∂σ (2. depends on the stress.37) gives ∂f ∂f · C(dε − dεp ) + · ∂σ ∂x 2 c dεp − γ x dp − b(Q − r(p)) dp = 0. and substituting for dε p from (2. for combined kinematic and isotropic hardening.79) Substituting for dσ and dx from Equations (2.
(2/3)c(∂f/∂σ) · (∂f/∂σ) − γ (∂f/∂σ) · x + b(Q − r(p)) (2. E + c − γ x + b(Q − r(p)) (2. irreversible deformation of a material under load. We shall see that the same formalism for viscoplasticity is also appropriate for creep—the time dependent. the material may harden isotropically or kinematically. though likely enhanced by thermallyactivated processes such as diffusionactivated dislocation climb. In addition. we have considered timeindependent plasticity only. in that the consistency condition is no longer formally applied so that the load point may now lie outside of the yield surface.84) 2. 2. though there may well be crystallographic slip still occurring.13(b).83) As in previous examples. and once yielding has occurred. σy . the stress–strain behaviour has been assumed to be independent of the rate of loading. Creep is normally used to describe low strain rate. The usual convention for terminology is that viscoplasticity describes ratedependent plasticity in which crystallographic slip is the dominant deformation process.38 Continuum plasticity so that the plastic multiplier becomes dλ = (∂f/∂σ) · dσ . for conditions of plane stress. the plastic ﬂow rule is obtained as before using the normality hypothesis of plasticity. that is. the elastic–plastic strain decomposition still holds. An important difference occurs. for the case of timeindependent plasticity. Figure 2. timedependent irreversible deformation which is either diffusion controlled. some viscoplasticity models are referred to as overstress models. together with the contribution from the linear isotropic hardening. and yield is determined as for timeindependent plasticity with a yield function.13(a) and at the corresponding point on the uniaxial stress–strain curve in Fig. however.7 Viscoplasticity and creep So far. we can reduce this to the form for uniaxial loading of a von Mises material and obtain for the uniaxial stress increment dσ = E 1 − E dε. The plasticity of materials which do exhibit rate effects is called viscoplasticity. At load point (1) shown on the yield surface in Fig. as before. .13 shows schematically the material’s stress–strain response and the corresponding yield surface which we assume to expand due to linear isotropic hardening. In viscoplasticity. or inﬂuenced by diffusion. 2. the stress achieved is the yield stress. As a result. We shall start by addressing uniaxial loading in the 1direction. r(p) so that σ = σy + r(p). whether strain or stress controlled.
the uniaxial stress depends on the yield stress. expanded yield surface Fig. the stress is augmented by the viscous stress. strain rate dependent. the corresponding stress strain curve. ˙ (2. of course. This. ˙ (2.13(b). The stress response is. viscoplasticity is sometimes referred to as timedependent plasticity or ratedependent plasticity. the hardening of the yield stress.85) in which K and m are material constants. In this case.86) In viscoplasticity. The constant m is called the material’s strain rate sensitivity. The resulting stress–strain curve is that shown by the broken line in Fig. therefore. but to simplify. and (b).13(b) by the solid line.13(b). However. let us assume perfect plasticity so that there is no isotropic hardening and r(p) = 0.Viscoplasticity and creep (b) s2 s2 sy E sy sy Initial yield surface sy s1 sy «2 sv r (p) sy 39 (a) Load point (1) Subsequent. but they often contain a dependence on the effective plastic strain rate (which for uniaxial conditions is identical to the uniaxial plastic strain rate). Let us consider the commonly used power law function σv = Kpm ˙ (2. is how the strain rate dependence of ˙ the plasticity is introduced. 2. From Fig. There are many types of equations used to represent the viscous stress. the stress–strain curve which includes the ratedependence of stress is therefore σ = σy + r(p) + σv = σy + r(p) + Kp m .87) . with viscous (overstress) stress σv . For this reason.13 (a) The von Mises yield surface in plane stress for viscoplasticity with linear isotropic hardening. the equation becomes σ = σy + Kp m . p. Let us consider the uniaxial behaviour described by Equation (2. for the case of viscoplasticity. 2. also shown schematically in Fig. and the plastic strain rate. 2. σ v .86) in a little more detail. 2.
2. the corresponding stresses are given by Equation (2. the isotropic hardening. dp = dε.89) and (2. for uniaxial perfect plasticity. For a von Mises material.86). and m) as a function of the plastic strain is shown in Fig. 2. ˙m ˙m σ2 = σy + K ε2 . r. would also show isotropic or kinematic hardening. If we apply strain controlled loading to the uniaxial sample.87).14(a). 2. Let us return to the case of isotropic hardening given in Equation (2.89) becomes σ − x − r − σy 1/m p= ˙ . and which depend upon internal variables. The strain rate dependence of stress is clear and for the three strain rates applied. x. and the kinematic hardening. (2.90) are constitutive equations relating uniaxial plastic strain rate to uniaxial stress. at a range of different strain rates as shown. but unlike the stress–strain curves in Fig. p Fig. Once yield is achieved.14. since dσ = 0.88) Many materials exhibit strain ratedependent plasticity. kinematic hardening also occurs. or equivalently.89) K If in addition to isotropic hardening. 2. the uniaxial stress is identical to the effective stress and similarly for the effective plastic strain rate. Equation (2. Equation (2.14(b). the stress response (for given σy . (2. p = ε so that the ˙ ˙ stresses are σ1 = σy + K ε1 .90) K Equations (2. We can rearrange the equation to give σ − r − σy 1/m p= ˙ .14 (a) Applied strain and (b) the resulting ratedependent stress response. σ3 = σy + K ε3 . ˙m (2.90) can therefore be written p= ˙ J (σ − x ) − r − σy K 1/m . K. (2.40 Continuum plasticity (a) • «3 (b) s3 Stress s2 s1 Strain «2 • «1 • sy Time Plastic strain.91) . as shown in Fig.
94) becomes ˙ ˙ ˙ εp = σ −x 3 p ˙ . or equivalently p = λ so that (2. For viscoplasticity. 2 J (σ − x ) (2.5. such as (2. the load point may lie outside the yield surface because of the viscous or overstress. In addition to the viscoplastic constitutive equation.95) We may now combine the constitutive equation given in (2.91). J (σ − x ) (2. and the rate form of Hooke’s law.6. f = J (σ − x ) − r(p) − σy so that ∂f 3 σ −x = ∂σ 2 J (σ − x ) ˙ εp = (2.4.2. dp = dλ. we make use of the normality hypothesis and the elastic constitutive equation.92) and 3 σ −x ˙ λ .98) 3 and we will now write Hooke’s law in terms of tensor rather than vector strain terms as ˙ σ = 2G˙ e + λ Tr(˙ e )I .Viscoplasticity and creep 41 In viscoplasticity. Equation (2.95) to give the plastic ﬂow rule for viscoplasticity with isotropic and kinematic hardening as 3 ˙ ε = 2 p J (σ − x ) − r − σy K 1/m σ −x . from Section 2.94) 2 J (σ − x ) We showed in Section 2.91) with (2.3 that for a von Mises material. as ˙ ˙ εp = λ The yield function is.96) To complete the model. From Section 2.17). we write the normality hypothesis.99) . ∂σ (2.5. (2. ε ε (2. we need the evolution equations for the isotropic and kinematic hardening variables r and x. Equation (2. as before. ˙ ˙ ˙ x= (2. we need a constitutive equation. This replaces the consistency condition in timeindependent plasticity.37) and Section 2. which relates the effective plastic strain rate to the stress and internal hardening variables. namely Hooke’s law.97) 2 p ˙ cε − γ x p ˙ (2. which forces the load point to stay on the yield surface during plastic deformation. In viscoplasticity. from Equation (2.65).93) ∂f . the hardening rates are given by r (p) = b(Q − r)p.
and axisymmetric compression Upsetting is the name given to the opendie forging of cylindrical billets of material. 1 Eν (1 − 2ν)(1 + ν) (2. but for now. together with the stress. . σ . At a given ˙ time. uniaxial compression. let us examine the equations for a simple problem of uniform.7. with knowledge of the current total strain rate. z Undeformed material Platens Deformed material u r Fig. 2. 2. and axisymmetric upsetting.100) form the complete elastic–viscoplastic model. t.15.1 Uniform. We will address this in a later chapter.42 Continuum plasticity where ˙ ˙ ˙ εe = ε − εp and in which G is the shear modulus.96)–(2. The process is shown schematically in Fig. ε and the hardening variables r and x. λ the Lame constant given by λ= and I is the identity tensor given by 1 I = ⎝0 0 ⎛ 0 1 0 ⎞ 0 0⎠ . uniaxial.100) Equations (2. uniaxial compression representing opendie forging. uniaxial. these equations enable the stresses at the end of a given step forward in time to be determined.15 Frictionless. We will consider the uniaxial compression of a cylinder between two platens such that the frictional effects on the contacting surfaces are negligible so that the cylinder remains under uniform. 2.
J (σ − x ) = σzz − x (2. 0 ⎠. The boundary conditions give us that σrr = σθ θ = σrz = 0 so that the stress and deviatoric stress tensors for the problem become ⎛ 1 ⎞ ⎞ ⎛ 0 0 − 3 σzz 0 0 0 ⎜ ⎟ 2 σ =⎝ 0 σ = ⎝0 σzz 0⎠ . − 1 (σ − x) 0 2 ⎝ 2 zz σ −x = 0 σzz − x 3 0 0 so that (2.105) .96) to determine the plastic strain rate components for the problem.103) and from Section 2.96) becomes ⎞ ⎛ p p ˙ εrr εrz 0 ˙ ⎟ ⎜ p p ⎜εrz εzz 0 ⎟ = 3 ˙ ⎠ 2 ⎝˙ p 0 0 εθθ ˙ ⎛ 0 0 1 − 2 (σzz ⎞ ⎠ − x) (2.101) ˙ ˙ ⎠ ⎝ p 0 0 εθ θ ˙ Because we are assuming friction to be negligible. 1 −2 0 ⎞ (2.104) σzz − x − r − σy K 1/m ⎛ 1 0 − (σzz − x) 2⎜ 2 1 × 0 σzz − x ⎝ σzz − x 3 0 0 ⎛ σzz − x − r − σy K 1/m 0 0 1 − 2 (σzz − x) ⎞ ⎟ ⎠ which reduces to ⎛ p ⎞ p ˙ εrr εrz 0 ˙ ⎜ p ⎟ p ⎜εrz εzz 0 ⎟ = ˙ ⎝˙ ⎠ p 0 0 εθ θ ˙ ⎜ ⎝ 0 0 1 −2 0 1 0 ⎟ 0 ⎠.102) We will now use Equation (2. Under this loading. (2. 3 σzz 0 0 0 0 0 − 1 σzz 3 (2.1.5. the shear terms are zero.Viscoplasticity and creep 43 We will consider constant strain ratecontrolled loading and write the plastic strain rates as ⎞ ⎛ p p ˙ εrr εrz 0 ˙ ⎟ ⎜ p p ˙ ε p = ⎜εrz εzz 0 ⎟ .
2 3 For completeness. For creep problems. just what we would get for purely uniaxial loading.106) which is.109) in full component form as ⎞ ⎛ ⎞ ⎛ c σxx σxy σxz εxx εxy εxz ˙ ˙c ˙c ⎟ 3 ⎟ ⎜ c n−1 ⎜ ˙c ˙c ˙ (2. creep deformation is assumed to occur with the application of a nonzero stress. it reduces to 3 n−1 2 Aσe σ = Aσ n .107) and J (σ ) = σe so the equation reduces further to ˙ εp = 3 2 σe − σy K 1/m σ .110) ⎝εyx εyy εyz ⎠ = Aσe ⎝σyx σyy σyz ⎠ 2 εzx εzy εzz ˙c ˙c ˙c σzx σzy σzz εc = ˙ in which the stress and strain rate tensors are symmetric. of course. Equation (2.44 Continuum plasticity We may now look at individual components of the plastic strain rate. σe (2.108) Often in creep problems.109) which is the multiaxial form of Norton’s creep law.7. Equation (2.108) is often written ˙ εc = 3 σe 2 K 1/m σ σ 3 3 n−1 = A(σe )n = Aσe σ σe 2 σe 2 (2.96) then becomes. where A and n are material constants.2 Powerlaw creep Let us look at one further case in which we assume there to be neither isotropic nor kinematic hardening. and Equation (2. 2. ˙ εp = 3 2 J (σ ) − σy K 1/m σ J (σ ) (2. For uniaxial loading. We see that εzz = ˙ p σzz − x − r − σy K 1/m (2. therefore. where the timedependent deformation is not dependent upon yield. we can write out Equation (2. for the plastic strain rate.105) also shows that 1 p 1 p p εrr = εθ θ = − εzz = − ˙ ˙ ˙ 2 2 σzz − x − r − σy K 1/m demonstrating that the incompressibility condition of plasticity is satisﬁed. .
(2002). . Because creep processes may occur independently of plastic yielding. USA.111) ˙ in which ε0 and σ0 are material constants with units of strain rate and stress. was determined for viscoplasticity. Instead.S. It therefore represents an energy per second per unit volume. (1998). John Wiley & Sons Inc. Mechanics of Solid Materials. V. The Mathematical Theory of Plasticity.112) ˙ 3 ε0 σ0 n−1 3 n−1 σ σ = Aσe σ . London. n+1 e 2 σ0 2 (2. based on the normality hypothesis together with an appropriate constitutive equation for plastic strain rate. UK. and Huang. In fact. A. The surfaces in stress space represented by φ are often called equipotential surfaces. φ. φ. therefore has units of Joules per second per unit volume. and Chaboche. but then reduced to Norton’s law for creep deformation. CUP. New York. Hill.L. in creep plays a similar role to the yield function in plasticity. Further reading Dieter.113) which is identical to Equation (2. (1988). just as the value of σe takes the same value on every point of a yield surface in plasticity.7. McGrawHill Book Co. G. We will deﬁne a potential. Lemaitre. (1995). The plastic (creep) strain rate is determined from ˙ εc = giving ˙ εc = ∂φ ∂σ (2.109).Further reading 45 2. Elastoplasticity Theory. R. often the yield function is considered to be a potential function. CRC Press. J. Florida. Continuum Theory of Plasticity. Khan. Cambridge. Mechanical Metallurgy. OUP (Oxford Classics Series). (1990). The potential function. a potential function is deﬁned from which creep strain rates are determined. such that φ= ε0 σ0 ˙ n+1 σe σ0 n+1 (2. φ. it is not appropriate to use a yield surface. The potential function.A. (ﬁrst published in 1950). Lubarda. we saw how the plastic strain rate.3 Potential and yield function equivalence In the last example. the energy per second per unit volume is the same at each point on the potential surface.E. S.. in the conventional sense discussed earlier. respect˙ ively. J.
This page intentionally left blank .
0 (i. . a nickel alloy may have undergone a strain of about σy /E ∼ 0. simply supported circular disc.1 shows the result of applying a large downward displacement at the centre of an initially horizontal.1 Simulated large elastic–plastic deformation of an initially horizontal. the forging of an aeroengine compressor disc. the strains remain quite small. for example. Deformation processing leading to large plastic strains often also leads to large rigid body rotations. Kinematics of large deformations and continuum mechanics 3. This is three orders of magnitude larger than the strain needed to cause yield. While the displacements and rigid body rotations can be very large.e. be they elastic or plastic. however. another very important feature is material rotation. Figure 3. At yield. the strains may be much bigger. In manufacturing processes.3. Only one half of the disc section is shown. requires strains in excess of 2.001 (or 0. <0. for example. that is.1 Introduction The strains. which most engineering components undergo in service are usually small.002 and it is hoped that this occurs rarely in nickelbased alloy aeroengine components in service! During manufacture. to large deformation.1%). simply supported circular plate. 3. Only one half of the plate section is shown. is an example which shows the rigid body rotation. The bending of a circular plate. × × Fig. >200%).
48
Large deformations and continuum mechanics
Towards the outer edge of the disc, where it is supported, the ﬁnite element simulation shows that the strains generated are quite small, but that because of the disc bending, the rigid body rotations are very large. Generally, deformation comprises of stretch, rigid body rotation, and translation. The stretch provides the shape change. The rigid body rotation neither contributes to shape change nor to internal stress. Because a translation does not lead to a change in stress state, we will not address it in detail here. In this chapter, we will give an introduction to measures of large deformation, rigid body rotation, elastic–plastic coupling in large deformation, stress rates, and what is called material objectivity, or frame indifference. Later chapters dealing with the implementation of plasticity models into ﬁnite element code will rely on the material covered here and in Chapter 2.
3.2 The deformation gradient
In order to look at deformation, let us consider a small lump of imaginary material which is yet to be loaded so that it is in the undeformed (or initial ) conﬁguration (or state). This is shown schematically as state A in Fig. 3.2. We will now apply a load to the material in state A so that it deforms to that shown in state B, the deformed or current conﬁguration. We will assume that the material undergoes combined stretch (i.e. relative elongations with respect to the three orthogonal axes), rigid body rotation, and translation. We will measure all quantities relative to the global XYZ axes, often known as the material coordinate system. Consider an inﬁnitesimal line, PQ, or vector, dX, embedded in the material in the undeformed conﬁguration. The position of point P is given by vector X, relative to the material reference frame. The line PQ undergoes deformation from state A to the deformed
Z Q9 dx O Q dX X P Y Undeformed (original) configuration, state A x
Deformed (current) configuration, state B
P9 u X
Fig. 3.2 An element of material in the reference or undeformed conﬁguration undergoing deformation
to the deformed or current conﬁguration.
Measures of strain
49
conﬁguration in state B. In doing so, point P has been translated by u to point P . Relative to the material reference frame, point P is given by vector x where x = X + u. (3.1)
The inﬁnitesimal vector dX is transformed to its deformed state, dx, by the deformation gradient, F , where dx = F dX. We can write this in component form as ⎛ ⎞ ⎛ Fxx dx ⎝dy ⎠ = ⎝Fyx dz Fzx or F = ∂x . ∂X (3.4) Fxy Fyy Fzy ⎛ ∂x ∂x ∂Y ∂y ∂Y ∂z ∂Y ∂x ⎞ ∂Z ⎟ ⎛dX ⎞ ⎟ ∂y ⎟ ⎝ ⎠ ⎟ dY ∂Z ⎟ dZ ⎠ ∂z ∂Z (3.2)
⎞ ⎛ ⎞ ⎜ ∂X dX Fxz ⎜ ⎠ ⎝ dY ⎠ = ⎜ ∂y Fyz ⎜ ⎜ ∂X dZ Fzz ⎝ ∂z ∂X
(3.3)
The deformation gradient, F , provides a complete description of deformation (excluding translations) which includes stretch as well as rigid body rotation. Rigid body rotation does not contribute to shape or size change, or internal stress. In solving problems, it is necessary to separate out the stretch from the rigid body rotation contained within F . In the following sections, we will see examples of stretch, rigid body rotation, and their combination, and how they are described by the deformation gradient.
3.3 Measures of strain
Let us consider the length, ds, of the line dx in the deformed conﬁguration. We may write ds 2 = dx · dx = (F dX) · (F dX) = dXT F TF dX = dX so that C = F TF . (3.5) C is called the (left) Cauchy–Green tensor. Consider also the length, dS, of the element, dX, in the undeformed state: dS 2 = dXT dX. (3.6)
T T
∂x ∂X
∂x dX = dXT C dX ∂X
50
Large deformations and continuum mechanics
Now dx = F dX so dX = F −1 dx. dS 2 = (F −1 dx)T F −1 dx = dx T (F −1 )TF −1 dx = dx T B −1 dx, where B −1 = (F −1 )TF −1 (3.7)
Substituting into Equation (3.6) gives
and B is called the (right) Cauchy–Green tensor. Both B and C are in fact measures of stretch as we will now see. A measure of the stretch is given by the difference in lengths of the lines PQ and P Q in Fig. 3.2 in the undeformed and deformed conﬁgurations, respectively. We can write (3.8) ds 2 − dS 2 = dx · dx − dx · B −1 dx = dx · (I − B −1 ) dx in which I is the identity tensor. It can be seen, therefore, that B is related to the change in length of the line; in other words, a measure of stretch. It is independent of rigid body rotation because the orientations of the lines in the undeformed and deformed conﬁgurations are irrelevant. If ds and dS are the same length, then there is no stretch and ds 2 − dS 2 = 0. From Equation (3.8), this means that B=I (3.9)
and there is no stretch, so that in this instance, the deformation gradient contains only rigid body rotation. The Cauchy–Green tensor, B, could itself be used as a measure of strain, since it is independent of rigid body rotation, but depends upon the stretch. This is an important criterion for any strain measure in large deformation analysis in which rigid body rotation occurs. A strain which depends upon rigid body rotation would not be appropriate since it would give a different measure of the strain depending upon orientation. However, the Cauchy–Green tensor given in (3.9) contains nonzero components even though the stretch is zero. An alternative and more appropriate strain measure called the Almansi strain was introduced: 1 e = (I − B −1 ) (3.10) 2 so that for zero stretch, e = 0.
Measures of strain
51
This strain measure behaves more like familiar strains, such as engineering strain, since for zero stretch, it gives us strain components of zero. A further measure of strain is the logarithmic, or true strain deﬁned as 1 ε = − ln B −1 , (3.11) 2 which we shall consider in more detail later. In determining the change of length of the line OP, we could have chosen the original conﬁguration to work in as follows: ds 2 − dS 2 = dx T dx − dX T dX = (F dX)T F dX − dXT dX = dXF TF dX − dXT dX = dXT (F TF − I ) dX = dXT (C − I ) dX = dXT (2E) dX, 1 1 (C − I ) = (F TF − I ). (3.12) 2 2 E is called the large strain or the Green–Lagrange strain tensor. We can make this look a little more familiar, perhaps, by combining Equations (3.4) for F and (3.1) for x ∂x ∂(u + X) ∂u F = = = +I ∂X ∂X ∂X and then substituting into (3.12) to give E= E= = 1 T 1 (F F − I ) = 2 2 1 2 ∂u + ∂X ∂u ∂X
T
where
∂u +I ∂X + ∂u ∂X
T
T
∂u +I −I ∂X (3.13)
∂u . ∂X
If we ignore the secondorder term, this reduces to E= 1 2 ∂u + ∂X ∂u ∂X
T
.
(3.14)
We will examine this and other strains for simple uniaxial loading in Section 3.4. Before doing so, let us examine the symmetry of the strain quantities presented above. The symmetric part of a tensor, A, is given by 1 (A + AT ) (3.15) 2 and the antisymmetric or skew symmetric (or sometimes simply skew) part of A by sym(A) = asym(A) = 1 (A − AT ). 2 (3.16)
52
Large deformations and continuum mechanics
To be pedantic, let us write out the symmetric and antisymmetric parts of the 2 × 2 matrix A given by a a12 A = 11 a21 a22 in which a12 = a21 . ⎛ a12 + a21 ⎞ a11 2 ⎠ sym(A) = ⎝ a12 + a21 a22 2 and ⎛ a12 − a21 ⎞ 0 2 ⎠. asym(A) = ⎝ a12 − a21 − 0 2 Clearly, sym(A) is symmetric and the leading diagonal of an antisymmetric tensor always contains zeros. Let us now consider the quantity F TF which appears above in a number of strain quantities: 1 sym(F TF ) = [F TF + (F TF )T ] = F TF 2 and 1 asym(F TF ) = [F TF − (F TF )T ] = 0. 2 We see, therefore, that F TF is a symmetric tensor so that, in fact, the quantities B −1 , C, ε, and E are all themselves symmetric. In general, F will not necessarily be symmetric. If it is, the deformation it represents is made only up of stretch.
3.4 Interpretation of strain measures
Let us determine the deformation gradient for the simple case of a uniaxial rod which is subjected, ﬁrst to purely rigid body rotation and then to uniaxial stretch and then consider some of the measures of deformation and strain introduced in Section 3.3. 3.4.1 Rigid body rotation only
Figure 3.3 shows a rod lying along the Y axis which undergoes a clockwise rotation about the Zaxis of angle θ. There is no stretch imposed, so the deformation gradient is simply the rotation matrix, R, given by ⎞ ⎛ cos θ − sin θ 0 (3.17) F = R = ⎝ sin θ cos θ 0⎠ . 0 0 1
3. that is. B −1 . 3. x X Fig. XY. is 1 e = (I − B −1 ) = 0 2 and the true strain. (3. 2 Similarly.11). The Almansi strain. = 0 0 1 The tensor B −1 is found to be the identity tensor for rigid body rotation because there is no stretch. We can now determine the Cauchy–Green tensor. refer to the material reference frame directions. RR T = I (3. is given by 1 ε = ln B −1 = 0. The uppercase letters in Fig. the rotating rod). 3. Also shown in the ﬁgure is a coordinate system which rotates with the deforming material (in this case.18) . We will generally use lowercase letters. 2 We see that the three strain measures give zero for the case of rigid body rotation.10). and the strain quantities. is always orthogonal. the Green strain is 1 E = (F TF − I ) = 0. y y u Y x X. such as R. from (3.7). Before moving away from rotation. Measures of deformation that are appropriate for large deformations with rigid body rotation must have this property: that they depend upon the stretch but are independent of the rigid body rotation.Interpretation of strain measures 53 Y.3 A rod undergoing rigid body rotation through angle θ. xy. ⎞ ⎞⎛ ⎛ cos θ sin θ 0 cos θ − sin θ 0 B −1 = (F −1 )TF −1 = ⎝ sin θ cos θ 0⎠ ⎝− sin θ cos θ 0⎠ 0 0 1 0 0 1 ⎞ ⎛ 1 0 0 ⎝0 1 0⎠ . it is important to note that a rotation tensor. to indicate the reference frame which rotates with the material. This is called the corotational reference frame. From Equation (3.
y Undeformed rod l0 l Z. x Fig. We will make much use of this property in the subsequent sections. For large plastic deformation.19) Let us consider the case of large strain such that the elastic strains can be ignored.4 A rod undergoing pure stretch in the absence of rigid body rotation.21) l0 Considering the stretch along the Y axis. Note that because there is no rotation in this case. z r r0 X. Y . so that R T = R −1 . the stretch ratios.4. (3.20) so that λx = λz ≡ and therefore 1 λy l −1/2 . r0 (3. For the uniaxial stretch shown. This is shown schematically in Fig. r0 λy = l .2 Uniaxial stretch Let us now consider uniaxial stretch of the circular rod in the Y direction. the incompressibility condition written in terms of stretches is λx λy λz = 1 (3. λ. l0 λz = r . y = λy Y. 3. z = λz Z λx = λz = .4. the corotational reference frame is directionally coincident with the material reference frame. any point. 3. 3. lying on the undeformed rod becomes the point y = λy Y on the deformed rod. The deformation can therefore be represented by x = λx X. are λx = r .54 Large deformations and continuum mechanics Deformed rod after stretch Y.
19) and (3. Let us examine the various deformation and strain tensors. ⎜ 0 0 F =⎜ ⎟ l0 ⎟ ⎜ −1/2 ⎠ ⎝ l 0 0 l0 We ﬁrst notice that F is symmetric so that FT = F and therefore represents stretch only.3). ∂Y ∂Z The deformation gradient can then be determined using Equation (3. ∂X ∂y = λy .21) ⎞ ⎛ l −1/2 0 0 ⎟ ⎜ l ⎟ ⎜ 0 ⎟ ⎜ l ⎟. ⎟ l0 ⎜ ⎟ ⎝ l⎠ 0 0 l0 . ∂Y ∂z = λz ∂Z ∂x ∂x = = 0 etc.Interpretation of strain measures 55 so that and ∂x = λx . The inverse of F is ⎛ ⎞ l 1/2 0 0 ⎟ ⎜ ⎜ l0 ⎟ ⎜ ⎟ l −1 ⎜ ⎟ −1 F =⎜ 0 0 ⎟ ⎜ ⎟ l0 ⎜ 1/2 ⎟ ⎝ ⎠ l 0 0 l0 so the Cauchy–Green tensor is given by ⎛ ⎞ l 0 0⎟ ⎜ l0 ⎜ ⎟ ⎜ ⎟ l −2 −1 −1 T −1 −1 −1 ⎜0 B = (F ) F = F F = ⎜ 0⎟. ⎛ ⎞ ∂x ∂x ∂x ⎜ ∂X ∂Y ∂Z ⎟ ⎛ ⎞ ⎜ ⎟ λx 0 0 ⎜ ∂y ⎟ ∂x ∂y ∂y ⎟ ⎝ ⎠ =⎜ F = ⎜ ∂X ∂Y ∂Z ⎟ = 0 λy 0 ∂X ⎜ ⎟ 0 0 λz ⎝ ∂z ∂z ∂z ⎠ ∂X ∂Y ∂Z and with (3.
In practice. operating on it and transforming it back.56 Large deformations and continuum mechanics The true plastic strain is. If the modal matrix (i. therefore. Before leaving the true strain.12) and second directly from the displacements using (3. note that we needed to take the logarithm of the Cauchy–Green tensor which we did by operating on the leading diagonal alone.e. to be the principal parts of the tensor (i. the eigenvalues). (3.14).22) ⎞ 1 1 We see that the true strain components are εxx = εzz = − 2 ln(l/l0 ) = − 2 εyy and εyy = ln(l/l0 ). in order to carry out an operation.e. we need to diagonalize the tensor and operate on the principal values. by ﬁnding the eigenvalues and eigenvectors of A. In general. . on tensor A having a linearly independent set of eigenvectors. the matrix containing the eigenvectors of A) is written as M. and the diagonal matrix (i.22). in this simple case. Let us ﬁnally determine the Green strain ﬁrst from (3. then A can be written ˆ A = M AM −1 so that we operate on A to give p(A) as follows ˆ p(A) = Mp(A)M −1 . 0⎟=⎜ ⎟ ⎜ ⎟ l0 ⎟ ⎝ 1 l⎠ l⎠ 0 0 − ln 2 l0 l0 (3. ˆ ln(A) = M ln(A)M −1 . We were able to do this in this case because the leading diagonal terms happen. the matrix containing the ˆ eigenvalues of A along the leading diagonal) as A. ⎛ l 0 ⎜ l0 ⎜ 1 1 ⎜ l ε = − ln B −1 = − ln ⎜ 0 ⎜ 2 2 ⎜ l0 ⎝ 0 0 −2 ⎞ ⎛ 1 l 0⎟ − ln 0 0 ⎟ ⎟ ⎜ 2 l0 ⎟ ⎟ ⎜ l ⎟ ⎜ 0 ln 0 ⎟.23) where the operation p is carried out only on the leading diagonal terms.e. p. as we would expect for uniaxial plasticity conditions. the modal matrix of B −1 in this case is just the identity matrix and ˆ −1 its diagonal matrix B is the same as B −1 so that ˆ −1 ln(B −1 ) = M ln(B )M −1 = I ln(B −1 )I −1 = ln(B −1 ). In Equation (3. For example. this means transforming the tensor into its principal coordinates.
⎟ r⎠ r0 0 if we write r = r0 + r and assume that the strains are small. If we now start from (3. We will now look at how to separate them in cases where they occur simultaneously. 3. we need the displacements. (2) rigid body rotation. u. (3) stretch.Polar decomposition 57 Using (3. r0 Y uy = (l − l0 ) . using the polar decomposition theorem. . which are given by ux = (r − r0 ) so that ⎛ r X .12). so that ⎤ ⎛ ⎡⎛ ⎞ 2 r r 1+ 0 0 ⎢⎜ ⎟ ⎛1 0 0⎞⎥ ⎜ r0 r0 ⎥ ⎜ ⎟ 1 ⎢⎜ 2 l ⎥ ⎜ ⎢⎜ ⎟ ⎝ E = ⎢⎜ 0 1+ 0 ⎟ − 0 1 0⎠⎥ = ⎜ 0 ⎥ ⎜ ⎟ 2 ⎢⎜ l0 0 0 1 ⎦ ⎝ ⎣⎝ 2 r⎠ 0 0 0 1+ r0 0 l l0 0 We see that the components of the Green strain are therefore approximately the engineering strain components. l0 uz = (r − r0 ) ⎛ and 1 E= 2 ∂u + ∂X ∂u ∂X T Z r0 ⎞ 0 ⎟ ⎟ ⎟ 0 ⎟ ⎟ r⎠ r0 ⎜ r0 ⎜ ∂u ⎜ =⎜ 0 ∂X ⎜ ⎝ 0 0 l l0 0 ⎞ 0 ⎟ ⎟ ⎟ 0 ⎟ ⎟ r⎠ r0 r ⎜ r0 ⎜ ⎜ =⎜ 0 ⎜ ⎝ 0 0 l l0 0 as before. we need ⎛ r 2 0 ⎜ ⎜ r0 ⎜ l ⎜ F TF = ⎜ 0 ⎜ l0 ⎜ ⎝ 0 0 2 ⎛ 2 r ⎟ ⎟ ⎜1 + r0 ⎟ ⎜ ⎟ ⎜ 0 0 ⎟≈⎜ ⎟ ⎜ ⎟ ⎝ r 2⎠ 0 r0 0 ⎞ ⎞ 0 2 l 1+ l0 0 ⎟ ⎟ ⎟ 0 ⎟ ⎟ 2 r⎠ 1+ r0 ⎞ 0 ⎟ ⎟ ⎟ 0 ⎟. We have looked at a number of examples in which stretch and rigid body rotation have taken place exclusively.14) in which we have also neglected secondorder terms.5 Polar decomposition Recall that the deformation at any material point can be considered to comprise three parts: (1) rigid body translation (which we do not need to consider here).
Let us have a look at a problem in which both stretch and rigid body rotation occur together: the problem of simple shear. by x = X + δY.24) where R is an orthogonal (R TR = I ) rotation tensor.5 shows schematically the simple shear by δ. and U and V are symmetric (U = U T ) stretch tensors. The deformation gradient is a nonsingular. in two dimensions. and is independent of the rigid body rotation. This conﬁrms. y=Y . we can now examine in more detail the right and left Cauchy–Green deformation tensors. where the squaring operation is carried out on the diagonalized form of V −1 . that B is a measure of deformation that depends on stretch only. Similarly. Using the polar decomposition theorem. From Equation (3. y) in the deformed conﬁguration. 2 2 which is also independent of the rigid body rotation and dependent on the stretch alone. Y ) in the undeformed conﬁguration to (x. C = F TF = (RU )TRU = U T R TRU = U TU = U2 since R is orthogonal since U is symmetric. which transforms a point (X.1 Simple shear Figure 3. 3. R.5. and a symmetric tensor (stretch).25) ε = − ln B −1 = − ln(V −1 )2 = ln V . secondorder tensor and can therefore be written as F = RU = VR. secondorder tensor can be decomposed uniquely into the product of an orthogonal tensor (a rotation). (3. for C. therefore. B −1 = (F −1 )TF −1 = [(VR)−1 ]T (VR)−1 = (R −1 V −1 )T R −1 V −1 = (V −1 )T (R −1 )T R −1 V −1 = (V −1 )T V −1 since R is orthogonal B −1 = V −1 V −1 = (V −1 )2 .7). of a unit square. The true strain rate can now be written in terms of V since 1 1 (3. We can represent the deformation.58 Large deformations and continuum mechanics The polar decomposition theorem states that any nonsingular.
F = 1 0 ∂y = 0. ∂Y We can then determine the Cauchy–Green tensor. 3.5 A unit square undergoing simple shear. We can use the polar decomposition theorem to separate out the stretch and rigid body rotation contained in the deformation gradient. 1 δ 1 = 1 δ ∂y = 1.29) and . Uyy (3. ∂X δ .27) Solving for the stretches in terms of δ and ϕ gives U= sin ϕ cos ϕ + δ sin ϕ 2 cos ϕ = √ 2 + δ2 2 2 −δ δ 2 and substituting back into (3. y d 1 Undeformed square 1 X. 1 + δ2 (3.28) gives δ sin ϕ = √ 2 + δ2 2 so that R=√ 1 22 + δ2 (3.Polar decomposition 59 Y. ∂X ∂Y The deformation gradient is. so that ∂x ∂x = 1.26) Note the symmetry of this deformation tensor. x Deformed square Fig. therefore.28) Uxx Uyx Uxy Uyy R= cos ϕ − sin ϕ sin ϕ cos ϕ (3. from C = F TF = 1 δ 0 1 1 0 δ . = δ. C. Let us set U= so that F = RU gives 1 0 δ cos ϕ = 1 − sin ϕ cos ϕ sin ϕ sin ϕ cos ϕ Uxx Uyx Uxy .
29). Consider a spatially varying velocity ﬁeld.6 Velocity gradient. rate of deformation. However. The increment in velocity. E= 1 T 1 1 (F F − I ) = (C − I ) = 2 2 2 1 δ δ 1 2 − 0 1+δ 0 1 = 1 0 2 δ δ . the ‘material axes’ rotate relative to the direction of the applied deformation.26). and with a little algebra conﬁrms that C = U 2 . plastic strain rate. we see that the principal stretch directions rotate as the simple shear proceeds.32) . ∂x The velocity gradient describes the spatial rate of change of the velocity and is given by dv = L= ∂v .12). δ2 (3. or from the components of strain in Equation (3. 3. that is. It is useful to examine why this is. to consider how the quantities already discussed can be put in rate form. may be written as ∂v dx. occurring over an incremental change in position. Using Equation (3. even rateindependent plasticity models are written in rate form for implementation into ﬁnite element code. All the quantities considered have been independent of time. Let us ﬁrst determine the Green–Lagrange strain for this deformation.31) From Equation (3. 3. This is easy to understand given that plasticity is an incremental process. therefore. It may not be obvious that the deformation corresponding to simple shear shown in Fig. dx.5 leads to a rigid body rotation. That is. and continuum spin We have so far considered stretch. The polar decomposition theorem is fundamental to large deformation kinematics and we shall return to it in subsequent sections. measures of strain. ∂x (3. in the deformed conﬁguration. 2 + δ2 (3. Often. and the polar decomposition theorem which enables us to separate out stretch and rotation.29) shows that the rigid body rotation is nonzero.30) and (3.60 Large deformations and continuum mechanics and U=√ 1 22 + δ2 2 δ δ . in fact. it is often more convenient to work with the equivalent quantity. dv. one for which the material point velocities vary spatially.30) Comparing Equations (3. rigid body rotation. many plasticity formulations (viscoplasticity is an obvious case) are developed in terms of rate quantities and it is necessary. rather than deal with increments in plastic strain. U is the symmetric stretch and Equation (3.31).
the rate of deformation and the continuum spin.35) 2 The symmetric part is called the rate of deformation. Consider the rotation of the rod shown in Fig. and the antisymmetric part the continuum spin. and continuum spin 61 Consider the time rate of change of the deformation gradient: ∂ ˙ F = ∂t or ˙ L = FF −1 .3. where sym(L) = and 1 (L − LT ).Velocity gradient. including uniaxial stretch and purely rigid body rotation in order to gain a physical feel for these quantities. W . so that asym(L) = L = D + W.33) The velocity gradient. maps the deformation gradient onto the rate of change of the deformation gradient. D.34) ∂x ∂X = ∂v ∂v ∂x = = LF ∂X ∂x ∂X We shall now examine both quantities. (3.37) (3.38) 1 (L + LT ) 2 (3. . 2 (3.3 shows this schematically. 3. (3. Figure 3. therefore. 3. where the rate of deformation D= and the continuum spin is given by W = 1 (L − LT ). for some simple examples.6. The velocity gradient can be decomposed into symmetric (stretch related) and antisymmetric (rotation related) parts: L = sym(L) + asym(L). but also the rate at which it occurs. rate of deformation.1 Rigid body rotation and continuum spin We have previously looked at the rigid body rotation of a uniaxial rod.36) 1 (L + LT ) 2 (3. We will now look not only at the rigid body rotation.
0 −1 0 0 ⎞ 0 0⎠ . but rigid body rotation occurs so that the continuum spin is nonzero.40) In order to determine the velocity gradient. with no stretch.62 Large deformations and continuum mechanics at time t making an angle θ with the Y axis. rotating at constant rate ˙ θ.33) as ⎞ ⎛ ⎞⎛ ⎛ 0 cos θ sin θ 0 − sin θ − cos θ 0 ˙ ˙ ˙ L = FF −1 = θ ⎝ cos θ − sin θ 0⎠ ⎝− sin θ cos θ 0⎠ = θ ⎝1 0 0 0 1 0 0 0 The transpose of L is 0 ˙ LT = θ ⎝−1 0 so that the deformation gradient is given by ⎛ 1 0 0 ⎞ 0 0⎠ 0 0 0 0 ⎞ 0 0⎠ = 0 0 ⎞ 0 0⎠ . 0 ⎛ 0 1 ˙ D = (L + LT ) = θ ⎝0 2 0 ⎛ 0 1 ˙ W = (L − LT ) = θ ⎝1 2 0 and the continuum spin is −1 0 0 (3. there is no stretch rate contributing to the velocity gradient so that the rate of deformation is zero.39) as ⎞ ⎛ cos θ sin θ 0 F −1 (= F T in this case) = ⎝− sin θ cos θ 0⎠ .41) That is. Let us now examine the signiﬁcance of the continuum spin for the case of rigid body rotation of the uniaxial rod without stretch. we need the inverse of F which we can obtain from (3. . 0 0 1 We may then determine the velocity gradient using Equation (3. The deformation gradient at any instant is given by ⎞ ⎛ cos θ − sin θ 0 (3.39) F = ⎝ sin θ cos θ 0⎠ 0 0 1 so that the rate of change of deformation gradient is ⎛ − sin θ − cos θ ˙ ˙ F = θ ⎝ cos θ − sin θ 0 0 ⎞ 0 0⎠ . 0 (3.
(3. 0 That is. given by ⎛ − sin θ − cos θ ˙ = θ ⎝ cos θ ˙ R − sin θ 0 0 Next. therefore.33) gives W = 1 ˙ 1 ˙ −1 ˙T ˙ (FF − (FF −1 )T ) = (FF −1 − (F −1 )TF ). Let us use the polar decomposition theorem to examine the continuum spin in a little more detail and more generally. consider the product of W and R ⎞⎛ ⎛ cos θ − sin θ 0 −1 0 ˙ WR = θ ⎝1 0 0⎠ ⎝ sin θ cos θ 0 0 0 0 0 ˙ R = WR ⎞ 0 0⎠ . 3.24).Velocity gradient. but it is closely related to it. R.44) The spin is not itself. for this particular case of rigid body rotation only.42) ⎞ ⎛ − sin θ 0 ˙ 0⎠ = θ ⎝ cos θ 0 1 ⎞ 0 0⎠ . W can be written for this simple case as that R ˙ W = RR T . and introduce the angular velocity tensor. a rate of rotation.43) ˙ so that W is the tensor that maps R onto R. 2 We will simplify this further by considering the product W = RR T = I . after a little algebra we obtain W = 1 ˙ T ˙T ˙ ˙ [RR − R R + R(U U −1 − (U U −1 )T )R T ]. Remembering that R is orthogonal so −1 = R T . the continuum spin is given by 1 (L − LT ) 2 so substituting for the velocity gradient from (3.6.38). 2 2 If we now substitute for F using the polar decomposition theorem in (3.45) .2 Angular velocity tensor From Equation (3. we see that (3. 0 − cos θ − sin θ 0 (3. rate of deformation. and continuum spin 63 ˙ Consider the rate of rotation. which we can differentiate with respect to time to give ˙ ˙T RR T + R R = 0 (3.
rod lengths as ⎛ ⎞ l −1/2 0 0 ⎜ l ⎟ ⎜ 0 ⎟ ⎜ ⎟ l ⎜ ⎟. and original.4) for which we obtained the deformation gradient. then Equation (3.48) W = Ω + R(U U −1 − (U U −1 )T )R T = + R asym(U U −1 )R T 2 ˙ in which Ω = RR T is called the angular velocity tensor which depends only on the rigid body rotation and its rate of change and is independent of the stretch.64 Large deformations and continuum mechanics so that ˙ ˙T ˙ RR T = −R R = −(RR T )T .46) ˙ We see.48) simply reduces to ˙ W = Ω = RR T (3. 3. in terms of the current. or if the stretch is negligibly small. since any tensor Z for which Z = −Z T 1 1 (Z + Z T ) = (Z − Z) = 0. In general. (3. that of uniaxial stretch with no rigid body rotation.48) shows that they differ depending on the stretch.1. 3. assuming purely plastic deformation and the incompressibility condition.6.47) as we saw in Equation (3.8. U .46) into (3. therefore. F =⎜ 0 0 ⎟ l0 ⎜ ⎟ −1/2 ⎠ ⎝ l 0 0 l0 . Equation (3. 2 2 Substituting (3. the angular velocity tensor and continuum spin are not the same.6.3 Uniaxial stretch We considered the uniaxial elongation of a rod lying along the Y direction earlier (see Fig.45) gives sym(Z) = 1 ˙ ˙ ˙ W = RR T + R(U U −1 − (U U −1 )T )R T 2 or 1 ˙ ˙ ˙ (3. l.49) (3. We will look at one further simple example in which we examine in particular the rate of deformation and the continuum spin. that RR T is antisymmetric because is antisymmetric. If we consider a deformation comprising of rigid body rotation only. as we shall see in Section 3.44). as we did in Section 3. l0 . Both W and Ω are important when considering objective stress rates.
2 There is. l0 1 εxx = εzz = − εyy 2 . no rigid body rotation occurring. Formally. For uniaxial stretch in the Y direction.Velocity gradient. Let us look at the rate of deformation for this case of uniaxial stretch. rate of deformation. ⎞ ⎛ 1 ˙ −2 0 0 1 l D = (L + LT ) = ⎝ 0 1 0 ⎠ . In addition. its antisymmetric part is zero so that the continuum spin is zero. 2 l 1 (3. ⎟ 1⎠ − 2 This is a symmetric quantity and therefore equal to the rate of deformation. just stretch. and continuum spin 65 We will now determine the rate of deformation and continuum spin for the case of uniaxial stretch. the true plastic strain components are given by εyy = ln l . therefore. Differentiating the deformation gradient gives ⎛ ⎞ 1 l −3/2 1 ˙ l 0 0 ⎜− ⎟ ⎜ 2 l0 ⎟ l0 ⎜ ⎟ ˙ l ⎜ ⎟ ˙ =⎜ F ⎟ 0 0 ⎜ ⎟ l0 ⎜ −3/2 ˙ ⎟ ⎝ l⎠ 1 l 0 0 − 2 l0 l0 and l ⎜ ⎜ l0 ⎜ ⎜ =⎜ 0 ⎜ ⎜ ⎝ 0 ⎛ 1/2 ⎞ 0 l l0 0 −1 F −1 ⎟ ⎟ ⎟ ⎟ 0 ⎟ ⎟ 1/2 ⎟ ⎠ l l0 0 ⎞ so that the velocity gradient is ⎛ ˙ 1 l −1 l ⎜− l0 ⎜ 2 l0 ⎜ ˙ l ⎜ ˙ L = FF −1 = ⎜ 0 ⎜ l0 ⎜ ⎝ 0 0 l l0 0 −1 0 0 − 1 2 l l0 −1 ⎟ ⎟ ⎟ ⎟ ⎟= ⎟ ⎟ ˙⎠ l l0 ⎛ 1 ⎜− 2 ˙ l⎜ ⎜0 l⎜ ⎝ 0 0 1 0 ⎞ 0 ⎟ ⎟ 0 ⎟.50) 0 0 −2 1 W = (L − LT ) = 0.
7 Elastic–plastic coupling Consider now an imaginary lump of material in the undeformed conﬁguration.50) shows. it can be seen that the rate of deformation is identical to the true strain rate. configuration x dx Initial configuration Z Fig. As before. therefore. . dX. can always be assumed to be small compared with the plastic strains. 3. it happens to be so for this example because there is no rigid body rotation. shown in Fig.51) ˙ 0 0 εzz ˙ For uniaxial stretch. ˙ εyy = . while not negligible.6. ˙ l 2l Examination of the components of the deformation gradient in (3. can be rewritten ⎛ ⎞ εxx 0 ˙ 0 D = ⎝ 0 εyy 0 ⎠ . 3. We now return to the consideration of elastic–plastic material behaviour under large deformation conditions.6 Schematic diagram showing an element of a material in the initial and current conﬁgurations and in the intermediate. stressfree. the example gives us a reasonable feel for what kind of measure the rate of deformation is. However.66 Large deformations and continuum mechanics so that the strain rates are ˙ ˙ l 1l ˙ ˙ ˙ εxx = εzz = − ˙ and εxy = εyz = εzx = 0. stressfree. in the absence of a rigid body rotation. that D. conﬁguration. (3. 3. This is not generally the case. after deformation F Current configuration dX X O x Fe Fp X dp Y Intermediate. We will consider cases in which the elastic strains. for this case. The material contains a line vector. therefore.
Considering a ﬁnite number of material points within a continuum. due to Erastus Lee. of course. this is an imaginary state corresponding to one in which dX in the undeformed conﬁguration has undergone purely plastic deformation to become dp in the intermediate conﬁguration. in which a discretization is necessary and the discontinuity of stress and strain resulting from the discretization is the norm. F . the line vector dX has undergone elastic and plastic deformation. both the elastic and plastic deformation gradients may contain both stretch and rigid body rotation. ∂x . rather a residual stress ﬁeld exists. (3. still leaving it unstressed. note that the intermediate conﬁguration described by p is. Note that for general inhomogeneous plastic deformation. no longer pointwise continuous. The transformation mapping of dX to dp is the plastic deformation gradient so that dp = F p dX and the plastic deformation gradient is deﬁned as ∂p . The intermediate conﬁguration is that in which line vector dx has been unloaded to a stressfree state. not uniquely determined since an arbitrary rigid body rotation can be superimposed on it. the deformation gradient.54). The transformation mapping of dX to dx is. ∂p (3.52) ∂X In the current conﬁguration dp is deformed into dx by the elastic deformation so that Fp = dx = F e dp and the elastic deformation gradient is deﬁned as Fe = We may then write dx = F e dp = F eF p dX so that F = F eF p . however. (3. We now introduce what is called the intermediate conﬁguration. in general.53) . the line vector is transformed to dx. Within the context of a ﬁnite element analysis. strictly. In addition. In Equation (3.Elastic–plastic coupling 67 to the deformed or current conﬁguration. this is not a problem. unloading a body will not generally lead to uniform zero stress. but F p and F e are. In transforming from the initial to deformed conﬁguration.54) This is the classical multiplicative decomposition of the deformation gradient into elastic and plastic parts. at each point an unstressed conﬁguration can be obtained.
55) Now. (3. using Equation (3. D = De + Dp. in order to overcome the uniqueness problem. we see from (3. The velocity gradient is given by ˙ L = FF −1 = ∂ ˙e ˙p (F eF p )(F eF p )−1 = (F eF + F F p )F p−1F e−1 ∂t ˙p ˙p ˙e ˙e = F F e−1 + F eF F p−1F e−1 = V V e−1 + V e F F p−1 V e−1 . by convention. Now. ˙e Le = V V e−1 = D e + W e ˙p Lp = F F p−1 = D p + W p so L = Le + V e Lp V e−1 = D e + W e + V e D p V e−1 + V e W p V e−1 . 3.56) W = asym(L). . includes stretch only (no rigid body rotation). that is.68 Large deformations and continuum mechanics However. often. As a result. let us now look at the velocity gradient and address the decomposition of the elastic and plastic rates of deformation. F p . F e is written Fe = Ve and so F p = V pR in which R is the equivalent total rigid body rotation. With this convention.1 Velocity gradient and elastic and plastic rates of deformation (symmetric stretch) We will now determine the velocity gradient in terms of the elastic and plastic deformation gradient decomposition given in (3. therefore.54). (3.57) In general. D = D e + sym(V e D p V e−1 ) + sym(V e W p V e−1 ) and W = W e + asym(V e D p V e−1 ) + asym(V e W p V e−1 ). D = sym(L) Therefore.56) the result that the elastic and plastic rates of deformation are not additively decomposed.7. such that the elastic deformation gradient. all the rigid body rotation is lumped into the plastic deformation gradient.55). (3. F e .
3. This brings us to the ﬁnal step in our brief examination of continuum mechanics.57). for small elastic stretches (V e ≈ I ).58) so that the stress rate may be determined using Hooke’s law. given in Equation (2. respectively. before introducing ﬁnite element methods in Chapter 4. (3. since rate of deformation and spin are symmetric and antisymmetric. σ .58) is often used in the implementation of plasticity models into ﬁnite element code.59) (3. like the plastic strain rate in small strain theory. and then summarize the most important steps required in going from knowledge of deformation to determination of stress. or Mohr’s circle. Often. we can integrate over time to determine stress.1 Principle of material objectivity or frame indifference Consider the transformation of the stress tensor. We will start not by considering stress rate. we need to look at what is called material objectivity. The stress transformation equations. then D e can be determined using (3. with respect to the .8.56) and (3. D p . in ﬁnite element implementations. sym(W P ) = 0. the total rate of deformation. Hence.8.Objective stress rates 69 This is unlike the additive decomposition of elastic and plastic strain rates for the case of small deformation theory. σ . D.8 Objective stress rates The primary focus of this section is stress rate. sym(D P ) = D P . The plastic rate of deformation. tell us that the transformed stresses. then V e = V e−1 ≈ I . Also. is speciﬁed by a constitutive equation. Once we know the stress rate. However. which undergoes a rotation through θ relative to the (XYZ) coordinate system. We will do this in Section 3.1). We need to address stress rate and how to determine it in a material undergoing rigid body rotation with respect to a ﬁxed coordinate system. from Equations (3. is known such that if D p is speciﬁed by a constitutive equation. but by something more familiar. 3. but before looking at this.58) This is an important and wellknown result and we will shortly see that Equation (3. transformation of stress. D = De + DP and W = W e + W P. if the elastic strains are small.
t is a vector quantity with normal. as shown in Fig.70 Large deformations and continuum mechanics Reference plane ∆F n ∆A u s∆ A τ∆ A Body Fig. A. other tensor quantities such as strain) transformation. making up a body which has been cut by a plane with normal direction n. In order to do this. The stress vector acting on the area A is deﬁned by t= F A . τ . It does this because along the principal directions. 3. σXX − σY Y (3. 3. and shear.7 shows a lump of material.1. with an inﬁnitesimal area of intersection. 3. relative to the applied stress direction.7. on a particular plane with normal n is related to the stress tensor. 3.1 The stress vector or traction. σXY = 0 so that tan 2θ = σXY . 2 (3.61) We will look at an alternative way of dealing with stress (and indeed.8(b). subjected to resultant force F . reproduced in Fig. A→0 (3. (XYZ) coordinate system become σXX = σXX cos2 θ + 2σXY sin θ cos θ + σY Y sin2 θ σY Y = σXX sin2 θ − 2σXY sin θ cos θ + σY Y cos2 θ σXY = (σY Y − σXX ) sin θ cos θ + σXY (cos2 θ − sin θ).8. . σ . Figure 3. the resultant force acting on the area A is F . σ . and of course. 3. Consider the three orthogonal planes shown in Fig. the principal stresses. we need to introduce the stress vector or surface traction. components. Under uniform stress state.8(a) and the resulting plane ABC.7 Schematic diagram showing a body cut by a plane with normal n generating an area of intersection of A. We will now look at how the stress vector. t.60) The last of the three equations enables us to determine the direction of the principal axes.62) By deﬁnition.
with components n = [nx ny nz ]T . The stress vectors acting on the three orthogonal planes have components given by ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ σzx σxx σyx t x = ⎣σxy ⎦ . 3. Ax = Ay = Az = 0 1 0 (3. then the areas of the three orthogonal planes are given by ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ 0 0 1 ⎣1⎦ · An = Any . σxy in the ydirection.Objective stress rates (a) B tx sxz sxy sxx n z A syy syz ty syx A C τ y szz szy (b) s t B 71 tz szx C x Fig. t x = ⎣σzy ⎦ . the stress vector is t x . a stress vector acts. (3. and so on.64a) By consideration of equilibrium.8 Schematic diagram showing (a) the stress vectors acting on three orthogonal planes with components in the x.64b) tA = t x Ax + t y Ay + t z Az = σxz σyz σzz Therefore ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎛ σxx σyx σzx σxx ⎣σxy ⎦ nx + ⎣σyy ⎦ ny + ⎣σzy ⎦ nz = ⎝σyx t= σxz σyz σzz σzx ⎞⎡ ⎤ nx σxz ⎠ ⎣ny ⎦ σyz σzz nz σxy σyy σzy (3. that is. and zdirections shown and (b) the resultant stress vector acting on plane ABC which has normal n. If the area of plane ABC is A. ⎣0⎦ · An = Anz . y. ⎣0⎦ · An = Anx . on plane ‘x’. On each of the orthogonal planes in Fig.63) t y = ⎣σyy ⎦ . the resultant force on plane ABC must be balanced by the forces on the three orthogonal planes so that ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ σxx σyx σzx ⎣σxy ⎦ Anx + ⎣σyy ⎦ Any + ⎣σzy ⎦ Anz .65) .8(a). 3. For example. σzz σyz σxz The convention is that on plane ‘x’. (3. the plane orthogonal to the xdirection. the stress components are labelled σxx in the xdirection.
R. Let us now return to the problem of transformation of stress. Equation (3. − sin θ cos θ sin θ cos θ σxy σyy Multiplying these equations will give Equation (3. (3. σ . in Equation (3. the stress tensor. acting on a surface with normal n. (3. corresponding to the rotation of θ is cos θ − sin θ R= sin θ cos θ so that cos θ sin θ cos θ − sin θ σxx σxy σ = . R. Under some rotation. are beautifully and succinctly summarized. and similarly. a tensor. and and t =σ n. Let us consider a simple example to understand what is meant by objective. Consider the stress vector. therefore.67).67) We see.65) can be written in a simple way as t = σ n. the stress is fully described by the tensor σ . therefore. transforms according to Equation (3.60).1. The rotation matrix.68) Comparing Equation (3. let us consider the twodimensional transformation giving Equations (3.67). This equation also tells us how other tensor quantities (such as strain. n = RT n . At the point of interest. the stress σ is transformed to σ such that t = σn t = Rt which give t = Rσ n and Combining these gives t = RσR T n but t =σ n so that ﬁnally σ = RσR T .60). (3.2 Transformation of stress.72 Large deformations and continuum mechanics remembering that for reasons of moment.68) shows that the Cauchy stress tensor is objective. or Mohr’s circle representation.66) 3. for example) transform. or rotational equilibrium. the stress tensor is symmetric so that σxy = σyx etc. More formally.67) with (3.8. A. The stress transformation equations. n = Rn Under rotation. that unlike a vector. To see what this means in detail. R. the stress vector t is transformed to t acting on a plane with normal n . the vectors t and n transform according to . t. is said to be frame indifferent or objective if it rotates according to the following: A = QAQT .
The stresses σXX = 0. 3. under constant. the stresses in the rod are: P . σyy = . A A corotational (xy) coordinate system has been introduced which rotates with the rod. With respect to the corotational system.9(b). σxy = 0. and (c) after rotating through 90◦ showing the changing stresses with respect to the material (XY) reference frame. P σxx = 0.Objective stress rates 73 3. all other stress components remain zero. similarly. .8. shown schematically in Fig. σXY = 0. In this conﬁguration. of crosssectional area A. A The rod undergoes a rotation of θ as shown in Fig. (b) having rotated through angle θ. force P . relative to the (XY ) coordinate system. with respect to the XY coordinate system. The stresses in the rod relative to the corotational (xy) coordinate system initially are.9 A rod of crosssectional area A undergoing rigid body rotation while subjected to axial force P (a) in the initial conﬁguration. σY Y = (a) Y P/A s XX = 0 s Y Y = P/A s XY = 0 (b) Y u P/A x y x s XX ≠ 0 s YY ≠ 0 s XY ≠ 0 y P/A P/A X (c) Y s X X = P/A s YY = 0 s XY = 0 x P/A y P/A X X Fig.2 Stressed rod under rotation: corotational stress Consider a rod initially lying parallel to the Y axis. 3. axial. 3.9(a). the rod is subjected to an unchanging stress in the ydirection of P/A.
74
Large deformations and continuum mechanics
in the rod, however, when measured with respect to the material (XY) reference frame can be seen to change. For example, when the rod has rotated through 90◦ as shown in Fig. 3.9(c), it lies parallel to the Xaxis so that now, the stresses with respect to the material (XY) reference frame are P σXY = 0. , σY Y = 0, A They are, therefore, completely different to what they were in the initial state shown in Fig. 3.9(a). However, the stresses relative to the corotational system are as before; that is, P σxx = 0, σxy = 0. σyy = , A The stresses relative to the corotational coordinate system, which we shall designate by σ , are related to those relative to the material (XY ) coordinate system, which we shall designate by σ , by σ = RσR T (3.69) σXX = in which R is just the rotation matrix. σ is called an objective or corotational stress because with respect to the corotational reference frame (and indeed, the material undergoing the rotation), its stress state has not changed; it has simply rotated. Note that the corotational stress, given in Equation (3.69) follows the requirement of objectivity given in Equations (3.67) and (3.68). The objective stress, therefore, results from the constitutive response of the material; it is independent of orientation and derives from the material response rather than the rigid body rotation. Before addressing stress rate and its objectivity, we will ﬁrst look at the objectivity, or otherwise, of a number of other important quantities in large deformation kinematics. 3.8.3 Objectivity of deformation gradient, velocity gradient, and rate of deformation
Consider the deformation gradient F . dx = F dX. (3.70)
After a transformation, Q, the quantities in Equation (3.70) become dx , F , and dX so that dx = F dX . (3.71) Now, dx = Q dx = QF dX and dX remains unchanged under deformation, by deﬁnition, so that dX = dX.
Objective stress rates
75
Therefore, from Equation (3.71), F = QF . (3.72)
F is, therefore, in fact objective and behaves like a vector because it is what is called a two point tensor, that is, only one of its two indices is in the spatial coordinate, x. We will now look at the objectivity of the velocity gradient, the rate of deformation, and the continuum spin. Differentiating Equation (3.72) gives ˙ ˙ ˙ F = QF + QF so that ˙ L =FF and
−1
˙ ˙ ˙ ˙ = (QF + QF )F −1 Q−1 = QQT + QFF −1 QT ˙ L = QQT + QLQT . (3.73)
Comparing with the requirement for objectivity in Equation (3.68), therefore, shows that the velocity gradient is not objective. The transformed velocity gradient, L , from Equation (3.73) can be written L = 1 1 ˙ Q(L + LT )QT + Q(L − LT )QT +QQ−1 . 2 2
Symmetric Antisymmetric
(3.74)
˙ From Equation (3.46) we see that QQT is antisymmetric. Therefore, from (3.74), D = sym(L ) = and 1 Q(L + LT )QT = QDQT 2 (3.75)
˙ W = asym(L ) = QWQT + QQ−1 .
Therefore, the deformation rate, D, is objective, but the continuum spin, W , is not. It is useful to ask whether this is important or not. In Sections 3.8.1 and 3.8.2, we examined the objectivity of the Cauchy stress and found that it is indeed objective. The Cauchy stress is, therefore, a quantity for which the properties are independent of the reference frame. This is very important in the development of constitutive equations. An equation which relates elastic strain to stress, for example, must be independent of the reference frame in which the relationship is used; in other words, the constitutive equation must provide information about the material response which is independent of rigid body rotation. The same holds for constitutive equations relating rate of plastic deformation (which we have just shown to be objective) to Cauchy stress (also objective), or the constitutive equation relating stress rate to elastic rate of deformation. In fact, plasticity
76
Large deformations and continuum mechanics
problems, especially within the context of ﬁnite element implementations, are often formulated in rate form. It is therefore necessary for us to address stress rate and, in particular, to examine whether stress rate is objective or otherwise. It turns out that there are many measures of objective stress rate, and we will focus on one in particular; the Jaumann stress rate.
3.8.4
Jaumann stress rate
We will examine, ﬁrst, a rather contrived problem in order to get a physical feel for the meaning of the Jaumann stress rate. We will then look more generally at the objectivity of this stress rate, and ﬁnish by looking at a simple example. Consider again a rod under axial stress σ , shown in Fig. 3.10(a). The stress tensor, with respect to the material axes is σt . During a time increment of t, the rod undergoes a rigid body rotation such that at time t + t, the rod lies as shown in Fig. 3.10(b) in which the corotational reference axes (xy) now coincide with the material reference axes (XY ).
(a) Y s
∆u x y s Time = t
(b) Y
s y x s Time = t + ∆t
0 0 0 sR+ ∆t = 0 s 0 t 0 0 0
X (c) Y s + ∆s y x s + ∆s Time = t + ∆t X
X
0 st + ∆t = 0 0
0 s + ∆s 0
0 0 0
Fig. 3.10 A rod under axial stress, σ , undergoing rigid body rotation (a) at a time t, (b) at a time t + t having gone through incremental rotation R (corresponding to angle θ about the Zdirection) and (c) at the same time t + t but having also undergone stress increment σ .
Objective stress rates
77
The stress tensor σ with respect to the corotational reference frame is ⎞ ⎛ 0 0 0 σ t = ⎝0 σ 0⎠ , 0 0 0 which is obtained from σt = Rσt R T , (3.76) where R is the incremental rigid body rotation. Following the rigid body rotation, the rod is subjected to an additional axial stress, σ , shown in Fig. 3.10(c), so that the ﬁnal stress tensor, with respect to the corotational frame (and the material frame since they are coincident at time t + t) is ⎞ ⎛ 0 0 0 σt+ t = ⎝0 σ + σ 0⎠ . 0 0 0 The Cauchy stress increment with respect to the corotational frame may be written approximately, using Equation (2.99), as σ = [2GD e + λTr(D e )I ] t, (3.77)
where we have used the rate of elastic deformation in place of the elastic strain rate. The stress increment results, therefore, purely from the material’s constitutive response and is corotational. We may then write the corotational stress tensor at time t + t as the sum of (3.76) and (3.77) to give σt+
t
≡ σt+
t
=
Rσt R T + [2GD e + λTr(D e )I ] t.
(3.78)
In order to investigate this further, let us consider the case in which the incremental rigid body rotation, R, is small. We may then approximate the rotation matrix as ˆ R = exp[ r ] ≈ I + ˆ r
ˆ in which r is the associated antisymmetric tensor which is approximately given by W t (a full explanation for this can be found, e.g., in Belytschko et al., 2000). Substituting into (3.78) then gives σt+
t
= (I + W t)σt (I + W t)T + [2GD e + λTr(D e )I ] t = σt + σt W T t + W σt t + W σt W T t 2 + [2GD e + λTr(D e )I ] t
so that σt+
t
− σt = σt W T + W σt + σt W σt t + [2GD e + λTr(D e )I ]. t
78
Large deformations and continuum mechanics
Now, both σt and σt+ t are given with respect to the material reference frame so that ˙ taking the limit, and letting t → 0, gives us the material stress rate, σ as ˙ σ = σt W T + W σt + 2GD e + λTr(D e )I and as W is antisymmetric (we saw earlier that this means W T = −W ), we obtain ˙ σ = W σt − σt W + 2GD e + λTr(D e )I . We can rewrite this as where
∇
(3.79) (3.80) (3.81)
˙ σ = σ +W σt − σt W , σ = 2GD e + λTr(D e )I .
∇
∇
σ in Equations (3.80) and (3.81) is called the Jaumann stress rate and it is the stress rate that results purely from the constitutive response of the material and not from rigid body rotation. It is, therefore, as we shall see in Section 3.8.4.1, an objective stress rate. We may therefore use it in constitutive equations such as (3.81) in which both the Jaumann stress rate and the elastic rate of deformation are objective quantities. ˙ ˙ The material stress rate, σ however, does depend upon rigid body rotation. σ is the Cauchy stress rate with respect to the material reference frame; that is, in Figs 3.9 ˙ and 3.10, σ gives the stress rate with respect to the material (XY ) axes. In ﬁnite element simulations, we are ordinarily interested in the stresses with respect to the material axes. Equation (3.80) is therefore important in that it enables us to determine the required stresses from knowledge of the material’s constitutive response given by the Jaumann stress rate in (3.81). It is useful to remember that the stresses, σt , in Equation (3.80), are also given with respect to the material reference frame. 3.8.4.1 Objectivity of Jaumann stress rate. Let us now show that the Jaumann stress rate is objective. We saw earlier that a quantity A is said to be objective if it transforms according to (3.68), that is A = QAQT and that the Cauchy stress transforms in this way (σ = QσQT ). Differentiating the stress with respect to time gives ˙ ˙T ˙ ˙T ˙ ˙ ˙ σ = QσQT + Q(σQT + σQ ) = QσQT + QσQT + QσQ . (3.82)
We see from Equation (3.82) that the material rate of Cauchy stress is not objective (even though the stress itself is) since it does not transform according to (3.68). We saw in Section 3.8.3, Equation (3.75) that ˙ W = QWQT + QQ−1 .
we will set up the problem in rate form and integrate to obtain the stresses and second. 3. We refer back to Fig. as shown in Fig. We will do this in two ways. uniaxial stress.83) and (3. σ .8. change with the rotation.Objective stress rates 79 Rearranging and remembering that Q is orthogonal gives ˙ Q = W Q − QW so that ˙T Q = −QT W + WQT (3. It is subject to a constant. satisﬁes the requirement for objectivity in Equation (3.4. Mohr’s circle).9(a)–(c). We will now determine the stresses in the rod with respect to the material axes as it rotates. and in particular. Substituting (3. 3. use the standard method for transformation of stress using Equation (3.9(b) which shows the rotating bar at an instant at which it makes an angle θ with the vertical. Substituting for QσQ = σ into (3. however. (2000). ﬁrst. uniaxial stress of P/A at all times. ∇ (3. This means that there is a rate associated with each stress component which also changes with the rotation. that of Belytschko et al. . Details of these may be found in many of the more advanced text books on continuum mechanics.2 Example of Jaumann stress rate for a rotating rod. ∇ (3.84) into (3.67) (or equivalently.88) P ⎠.83) (3.85) gives T (3. 0 A The stresses with respect to the material (XY ) axes.87) There are objective stress rates other than that of Jaumann. This.85) ˙ ˙ σ W − W σ + σ = Q(σ W − W σ + σ )QT . where ˙ σ = σ + σW − Wσ.4 with a simple example of a rotating rod under constant.68).84) since both W and W are antisymmetric. 3. of course. gives a constant corotational stress (with respect to corotational xyaxes) of ⎞ ⎛ 0 0 σ =⎝ (3. We will conclude Section 3.86) therefore shows that the Jaumann stress rate.8.82) gives ˙ ˙ σ = W QσQT − QW σQT + QσQT + Qσ WQT − QσQT W so that ˙ ˙ σ = Q(σ W − W σ + σ )QT + W QσQT − QσQT W .86) Equation (3.
90) and (3.89) If there is no rotation.90) 1 0 The stresses with respect to the material axes may be written σ = σXX σXY σXY . or corotational stress rate. simply by differentiating (3. Equation (3. σ . is given by rearranging (3.91) into (3. (3.41).91) Substituting (3. its derivative is zero.80 Large deformations and continuum mechanics ∇ We may obtain the objective. The initial conditions are σXX (0) = 0. ∇ (3.92) dσXX dσY Y ˙ = −2θ σXY ≡ − dt dt dσXY ˙ = θ (σXX − σY Y ). that this results in a continuum spin given by ˙ 0 −1 . σY Y (0) = P/A. Differentiating (3.89) gives ˙ ˙ σ =θ That is −2σXY σXX − σY Y σXX − σY Y . 2A and . As the corotational stress is always constant. σXY (0) = 0 so that k = P/A. 2 A dt This has general solution σXX = A sin 2θ + B cos 2θ + P . σY Y (3. W =θ (3. σ . We showed for this case. so ⎞ ⎛ 0 0 ∂ ⎝ ∇ σ = P ⎠ = 0.88) with respect to time. then W = 0 and (3.93) dt Equation (3. 2σXY (3. with constant angular ˙ speed θ. However.93) and substituting into (3.92) gives σXX = −σY Y +k where k is just a constant.87) ˙ σ = σ − σW + Wσ = Wσ − σW. the rod is rotating.89) tells us that the stress rates with respect to the material axes are also zero. ∂t 0 A ˙ The stress rate with respect to the original conﬁguration.92) and using k = P/A gives d2 σXX P ˙ ˙ = +4θ 2 σXX = 2θ 2 .
Determine the rates of deformation and continuum spin 1 D = sym(L) = (L + LT ). That is. We also assume full knowledge of the deformation gradient. (3. cos2 θ 3. σXY = sin θ cos θ. so that with the initial conditions. Finally. that the stresses with respect to the material reference frame change correctly with angle θ . Determine the velocity gradient ˙ L = FF −1 . 2. we will determine the same stresses using the stress transformation equation in (3. Y ) reference reference so that the stresses in the material (XY ) reference are given by ⎛ ⎞ P cos θ sin θ ⎝0 0 ⎠ cos θ − sin θ sin2 θ σ = = P − sin θ cos θ sin θ cos θ A sin θ cos θ 0 A which just gives the expressions in Equation (3.Summary 81 ˙ where θ = θt. when θ = 0. therefore. σY Y = 0. 1. we will summarize some of the important steps required in going from knowledge of deformation through to the determination of stresses for an elastic–plastic material undergoing large deformations. We assume that any deformation taking place is such that the stretches due to elasticity are small compared with those for plasticity so that the additive decomposition of rate of deformation given in Equation (3. σXY = 0. sin θ cos θ . 2 1 W = asym(L) = (L − LT ).67). and when θ = π/2. σ = RσR T ↑ ↑ (x. σXX = P/A. The steps required in determining stresses are then as follows. For example.94) σY Y = cos2 θ. This is often required in the implementation of plasticity models into ﬁnite element code so it is something we shall return to later. F . 2 . A A A We see. σXY = 0. the full solution is P P P σXX = sin2 θ.58) holds. y) (X.9 Summary Before leaving the kinematics of large deformations.94). ˙ F and its rate. σY Y = P/A. σXX = 0.
For example. CRC Press. New York. T. Continuum Theory of Plasticity. B.. Determine the rate of elastic deformation Further reading Belytschko.J. Computational Inelasticity. Use a numerical technique to obtain the stresses with respect to the material ˙ reference frame by integrating σ . from the rate of elastic deformation σ = 2GD e + λTr(D e )I .. Khan.C.80) ˙ σ = σ +W σ − σ W . John Wiley & Sons Inc. USA. and Moran. T. . Berlin.R. Liu. 7. S. 5. we have in the current conﬁguration Dp = 3 2 J (σ − x ) − r − σy K De = D − Dp. John Wiley & Sons Inc. A. (2002). ∇ ∇ 1/m σ −x . J. Florida. 6. and Hughes. We will address all of these steps in some detail in later chapters. (1995). Equation (3.82 Large deformations and continuum mechanics 3. from Chapter 2. Determine the material rate of stress using (3. Determine the Jaumann stress rate using the tensor form of Hooke’s law. SpringerVerlag. J (σ − x ) 4. The rate of plastic deformation is speciﬁed by a constitutive equation. Lubarda. (2000).81).L.A. Simo. Elastoplasticity Theory. V.S. (1997). New York. and Huang. Nonlinear Finite Elements for Continua and Structures. W. for combined isotropic and kinematic hardening with power law dependence of effective plastic strain rate on stress.
the ﬁnite element method. and that the boundary conditions for the problem are also obtained. This is usually because of irregular geometry and/or complicated boundary and loading conditions. which appears in many text books. within the ﬁnite element technique. we shall use Hamilton’s principle. enables the approximate solution of these types of problems. A weak formulation can be obtained by consideration of the principle of virtual work. we aim to develop a good physical feel for what the principle is doing before addressing more complicated ﬁnite element applications. with the latter each having a speciﬁed number of degrees of freedom. the body under consideration is discretized into a ﬁnite number of elements and nodes. and in particular.4. enable solutions to be obtained subject to the satisfaction of initial and boundary conditions. but here. and boundary conditions is essential for the solution of any solid mechanics problem. The ﬁnite element method for static and dynamic plasticity 4. We shall use Hamilton’s principle to obtain the equations of motion (and boundary conditions) for a number of well known discrete and continuous systems. . Instead. for both quasistatic and dynamic problems. The important requirements of equilibrium. and compatibility. Within the ﬁnite element method. The ﬁnite element model representing the body therefore contains a ﬁnite number of degrees of freedom and the implication is that the requirement for equilibrium cannot be satisﬁed exactly at every point in the continuum (in what is called the strong sense). In this way. compatibility. Computational mechanics. as we shall see. constitutive equations. a weak formulation of equilibrium is used in which global equilibrium for the body as a whole is imposed even though this does not necessarily ensure pointwise equilibrium. is that the equilibrium equations of motion can be obtained in a fully uniﬁed manner. Satisfying the requirements of equilibrium.1 Introduction There are few practical problems in plasticity which can be solved analytically. Further details of strong and weak formulations are available in texts that are more specialized. together with a material’s constitutive response. The advantage.
temperature. A reasonable knowledge of vector calculus is useful. realizable initial. and of course. 4. etc. The principle states that the variation of the kinetic and potential energy plus the variation of the work done by nonconservative (external) forces acting during any time interval t1 to t2 must be zero. Here. the kinetic energy. We will shortly see what is meant by variation. for example). (4. which results from conservation of energy.2. Let the total kinetic energy of the system be T . the motion of a system between speciﬁed.2 Hamilton’s principle Hamilton’s principle. and the work done by nonconservative forces be W . 4. (4.1 for a single degree of freedom system (SDOF). nonconservative implies forces that cannot be described by the change in a potential energy function (as can strain energy. We then return to elastic–plastic problems and address plasticity in a onedimensional rod element. which are not already included in the potential energy term.) of time y(t) which .1) and the Jintegral is deﬁned as J = t2 t1 t2 t1 L dt = (T − U + W ) dt. in keeping with the aims of this book. but the important results are not impenetrable without it! Unfamiliar readers may like to skip on to the introduction to the ﬁnite element method in Section 4. and ﬁnal conditions at times t1 to t2 is such that the average value of L relative to any dynamical path compatible with the physical constraints has a stationary value. L.3) In other words. is one of the most general principles of mechanics. is deﬁned as L=T −U +W (4.84 Finite element method We introduce Hamilton’s principle in Section 4.3. potential energy.g. Generally. the potential energy of the system be U . It provides a means for ﬁnding the equilibrium equations (equations of motion) of a dynamical system by determining the stationary value of a scalar integral. and work of nonconservative forces depend on some function (e.2) Hamilton’s principle states that the ﬁrst variation (denoted by δ) of J is zero thus δJ = t2 t1 δL dt = t2 t1 δ(T − U + Wnc ) dt = 0. displacement. we address ﬁrst a simple onedimensional rod element subjected to elastic deformation alone and then very brieﬂy look at element assemblage and some other ﬁnite element types. An illustration of such a motion is given in Fig. The Lagrangian.
6) and ε is an arbitrary parameter. is that which leads to a stationary value of J . Let us now look at how Equation (4. may be minimized by use of the calculus of variations. Let y = y(t) be the actual minimizing curve (as distinct from any admissible curve) and let Y (t) = y(t) + εη(t) (4. 4.4) in which y is just y = dy/dt. . t) dt (4.7) ∂L ∂Y ∂L ∂Y + ∂Y ∂ε ∂Y ∂ε t2 t1 dt. (4. Y . y . ˙ Hamilton’s principle tells us that the dynamical path. y(t).5). The corresponding J integral is ¯ J (ε) = and its variation with respect to ε is ∂ ¯ J (ε) = ∂ε ∂ ¯ J (0) = ∂ε = t2 t1 t2 t1 t2 t1 t2 t1 L(Y. t) dt (4. can be expressed as follows J = t2 t1 L dt = t2 t1 L(y.8) For ε = 0.4). as ∂L ∂L η+ η ∂y ∂y dt = ∂L ∂L η dt + η ∂y ∂y t2 t1 t2 t1 − t2 t1 η d ∂L dt dt ∂y d ∂L ∂L ∂L − η dt + η ∂y dt ∂y ∂y .5) be the family of comparison curves. and using (4.1 Two dynamical paths subjected to having the same state at times t1 and t2 . this variation yields by integrating by parts.Hamilton’s principle 85 x dx Two dynamical paths compatible with constraints (the difference between trajectories is dx) x(t ) x(t ) t1 t2 t Fig. called a functional. where η(t) is an arbitrary function subjected to the constraint that η(t1 ) = η(t2 ) = 0 (4.
∂y dt ∂y Since η is arbitrary.9) Therefore. − ∂y dt ∂y ∂L d ∂L − = 0. t2 t1 (4. t1 ∂L ∂L δy + (δy) ∂y ∂y dt (4.86 Finite element method which for η(t1 ) = η(t2 ) = 0 yields ∂ ¯ J (0) = ∂ε t2 t1 d ∂L ∂L − η dt.13) ∂ ∂ ∂ 1 r +s +t 2! ∂x ∂y ∂z t2 F (x.11) which is known as Euler–Lagrange equation and which must be satisﬁed for J to have a minimum. y. it follows that (4. If the function y(t) is replaced by y + δy where δy is the variation of y such that δy(t1 ) = δy(t2 ) = 0. y. then it follows using Taylor’s theorem that J (y + δy) = J (y) + + + since F (x + r.12) ∂L ∂L δy + (δy) ∂y ∂y dt 1 2! 1 3! ∂ 2L 2 ∂ 2L ∂ 2L δy + 2 δy(δy) + (δy ) 2 dt 2 2 ∂y∂y ∂y ∂y δy ∂ ∂ + δy ∂y ∂y 3 L(y. t) dt + · · · (4. y . ∂y dt ∂y (4. z) + · · · . the necessary condition for minimization of J is ∂ ¯ ∂ ¯ J (0) = J (ε) = 0. y + s. z + t) ≈ F (x. ∂ε ∂ε Hence.10) ∂L d ∂L η dt = 0.14) . z) + r + The quantity δJ = ∂F ∂F ∂F +s +t ∂x ∂y ∂z 2 t2 t1 t2 t1 t2 t1 (4.
x. 4. or J integral as J =l= x2 x1 dy dx 2 (1 + y 2 )1/2 dx. dl. If the ﬁrst variation is equal to zero for a particular function y(t) of the admissible class. the functional J (y) is said to have a stationary value for that particular function y.Hamilton’s principle 87 is called the ﬁrst variation while δ2J = 1 2! t2 t1 ∂ 2L 2 ∂ 2L ∂ 2L (δy ) 2 dt δy + 2 δy(δy) + ∂y∂y ∂y 2 ∂y 2 is the second variation of J .1 Stationary value: minimizing the distance between two points Figure 4.15) is the functional which we wish to minimize to obtain the shortest path.2 shows two points (x1 . y2 ). y1 ) and (x2 . y1) x1 x2 x Fig. An element of length dl lies on the path shown. . we obtain the functional. The ﬁrst is by obtaining directly the y y (x) dl dy dx (x2. y1 ) and (x2 . We will use two approaches to do this. and not time. 4. let us look at one simple example to help understand the process of ﬁnding a stationary value. lying on the path. (x1 . y(x). and an element. Only one is shown in the ﬁgure! Here note that y(x) is a function only of position.2. Our aim is to use the above approach to determine that path which minimizes the distance between the two points. Before returning to equilibrium equations. y2) (x1.2 A particular path. y2 ) which are joined by an inﬁnite number of possible paths. and its length can be written as dl 2 = dx 2 + dy 2 = dx 2 1 + such that dl = (1 + y 2 )1/2 dx and ﬁnally. y(x). between two points. (4.15) Equation (4.
15) is given by x2 x2 1 (1 + y 2 )1/2 dx = (1 + y 2 )−1/2 δ(y 2 ) dx δJ = δ 2 x1 x1 = Now. The functional.15) may be written as x δJ = [y (1 + y 2 )−1/2 δy]x2 − x1 J = 2 x2 x1 (1 + y 2 )−1/2 y δy dx.16) which has the solution as before. let us do the same thing but now using the Euler–Lagrange Equation (4. y . ∂L ∂L 1 = 0 and = (1 + y 2 )−1/2 2y ∂y ∂y 2 and substituting into (4.17) just gives us the differential Equation (4. we obtain x2 d ((1 + y 2 )−1/2 y ) δy dx = 0. J . that the process of ﬁnding the ﬁrst variation (or equivalently. dx = x2 x1 (1 + y 2 )−1/2 y x2 x1 L(y.11). and x and from Equation (4. the stationary value) of J .16) dx which (unsurprisingly) has the solution y = Ax + B in which A and B are constants of integration.15) and the second is to do the same by using the Euler–Lagrange Equation (4. x) dx.17) ∂y dx ∂y Now. . is a function of y. dy dx d (δy) dx d (δy) dx. (4. y . therefore. δJ = − x1 dx Since δy is arbitrary. We see. (4. we obtain the differential equation d ((1 + y 2 )−1/2 y ) = 0. δy = δ so δJ = Integrating by parts gives d ((1 + y 2 )−1/2 y ) δy dx dx x1 and since δy(x1 ) = δy(x2 ) = 0. where L(y.88 Finite element method stationary value of (4. and for a stationary value. results in ﬁnding the function y(x) which minimizes the functional. x) = (1 + y 2 )1/2 so the Euler–Lagrange equation becomes ∂L d ∂L − = 0.11). The ﬁrst variation of (4. To ﬁnish off. y . J .
For a solid elastic body.66). 2 1 σ : ε dV . (4. or traction. t the stress vector. V . t. and ∂ and are domains of area. σ and ε the stress and strain tensors respectively.Hamilton’s principle 89 4.2 Equilibrium equations Let us now return to Hamilton’s principle to see how it can be used to obtain equilibrium equations.18) U= 2 W = ∂ t · u dA ˙ in which u and u are the displacement and velocity vectors.19) (T − U + W ) dt 1 2 ˙ ˙ ρ u · u dV − 1 2 σ : ε dV + div [σ u] dV dt. Taking the ﬁrst variation of J gives δJ = = = δ(T − U + W ) dt 1 2 ˙ ˙ 2ρ u · δ u dV − ˙ ρu · ∂ δu dV − ∂t 1 2 (σ : δε + δσ : ε) dV + σ : δε dV + div [σ δu] dV div [σ δu] dV dt = 0 dt and noting that div [σ δu] = div [σ ] · δu + σ : ∇δu and ∇δu = δε and integrating by parts gives δJ = = t2 t1 t2 t1 − − ¨ ρ u · δu dV + ¨ ρ u dV + div σ δu dV dt div σ dV · δu dt = 0 . ρ the density. respectively. from Equation (3. (4. A and volume. and making use of the divergence theorem gives W = The J integral is. therefore J = = t2 t1 t2 t1 t2 t1 t2 t1 t2 t1 ∂ σ n · u dA = div [σ u] dV . in the absence of body forces.2. Substituting for the stress vector. the components of the Lagrangian can be expressed as follows: 1 ˙ ˙ T = ρ u · u dV .
Consider the SDOF spring–mass system shown in Fig. we obtain the more familiar expressions ⎡ ⎤ ∂σxy ∂σxx ∂σxz + + ⎢ ∂x ∂y ∂z ⎥ ⎢ ⎥ ⎢ ∂σ ∂σyy ∂σyz ⎥ ⎢ yx ⎥ + + div σ = ⎢ ⎥ = 0. from a position of equilibrium. r.3. Hamilton’s principle has therefore provided us with the condition for equilibrium and. k m x(t ) Fig. These are the wellknown equilibrium equations of stress analysis. in doing so.90 Finite element method so that ¨ (−ρ u + div σ ) · δu dV = 0 and since δu is arbitrary. We will use Hamilton’s principle to derive the equation of motion for the mass. ρ u = 0. the work done is zero.20) is simply δW = r · δu = 0.2. let us consider a few more examples. If we consider ¨ quasistatic conditions. in which we shall employ Hamilton’s principle to obtain the momentum balance equations for a number of simple problems.20) 4. In addition. ⎢ ∂x ∂y ∂y ⎥ ⎢ ⎥ ⎣ ∂σzx ∂σzy ∂σzz ⎦ + + ∂x ∂y ∂z Hamilton’s principle has therefore given us the very general form of the equilibrium equations of stress analysis. in the course of an arbitrary. The term (−ρ u + div σ ) dV is just the residual force. δu. m. Before moving on to its application to ﬁnite elements. That is.3 Further examples of Hamilton’s principle (4. inﬁnitesimal virtual displacement.3. one for a discrete system. and two further continuous systems. a statement of the principle of virtual work. and expand div σ .2.1 Discrete spring–mass problem. .3 Simple SDOF mass and spring system. 4. 4. ¨ −ρ u + div σ = 0. 4. Equation (4.20) can be seen to be a statement ¨ of the principle of virtual work. so that the expression in (4.
2. for the well known mass–spring problem. ¨ Hamilton’s principle has given us the governing equation of motion. 2 2 The J integral may then be written.3. 4.2 Transverse vibration of a continuous beam. . Figure 4. in the absence of external forces. (b) M (a) y y f+ −f −x dU = 1 M df 2 f f x df Fig. or equilibrium equation. as J = t2 t1 (T − U ) dt = t2 t1 1 2 1 2 mx − kx dt. The bending moment variation with angle. We shall next consider a continuous system and use Hamilton’s principle to obtain the equations of motion for it. 4. we get δJ = [mx δx]t2 − ˙ t1 t2 t1 (mx δx + kx δx) dt ¨ and since δx(t1 ) = δx(t2 ) = 0.4(a) shows an elastic beam undergoing bending with deﬂection.4 (a) An elastic beam in bending undergoing deﬂection y and rotation φ and (b) the linear variation of bending moment M with rotation φ leading to stored elastic strain energy U . δJ = − t2 t1 (mx + kx) δx dt = 0 ¨ so mx + kx = 0. is shown in Fig. 4.Hamilton’s principle 91 We may write expressions for the kinetic and potential energies of the system as 1 1 ˙ T = mx 2 . y. U = kx 2 .4(b) assuming elastic behaviour. φ. and δx is arbitrary. ˙ 2 2 t2 t1 Taking the ﬁrst variation gives δJ = 1 2 t2 t1 (2mx δ x − 2kx δx) dt = ˙ ˙ mx ˙ d (δx) − kx δx dt dt and integrating by parts.
is dT = 2 2 (4. F (x. 4. of the beam. . ∂x 2 where E is Young’s modulus and I is the second moment of area.22) ∂t 2 0 The work done by external forces.5.23) Consider the simply supported beam undergoing transverse vibration shown in Fig. the potential energy for a beam of length L is 1 L ∂ 2y U= EI dx.21) 1 ∂y 2 ρA dx 2 ∂t in which A is the beam’s crosssectional area. t)y dx. the bending moment is related to the deﬂection. We apply Hamilton’s principle in the usual way. 2 0 ∂x 2 The kinetic energy. is given by W = 0 L F (x. Thus. by ∂ 2y . Hence the total system kinetic energy is 1 L ∂y 2 T = ρA dx. t) per unit length. (4. 2 ∂x For an elastic beam. M = EI 1 ∂ 2y dU = EI dx 2 ∂x 2 and ﬁnally. in an element of length dx. (4. t)y dx dt. y.92 Finite element method The elastic strain energy stored in an element of length dx of the beam is 1 1 ∂φ dU = M dφ ≈ M dx 2 2 ∂x and ∂y φ= ∂x so that 1 ∂ 2y dU = M 2 dx. The J integral is J = = t2 t1 L dt t2 L 0 1 2 ρA t1 ∂y ∂t 2 dx − 0 L EI ∂ 2y ∂x 2 2 dx + 0 L F (x.
22). We take the ﬁrst variation of J to give the stationary value and equate to zero δJ = t2 t1 δL dt = 0.5 A simply supported beam with distributed load F (x. is t2 t1 δT dt = = = 1 2 1 2 t1 t2 t1 t2 t1 t2 ρAδ 0 L ∂y ∂t ∂y ∂t δ 2 dx dt δ ∂y ∂t ∂y ∂t dx dt ρA 0 L 0 t2 L 0 L 2 ∂y ∂t ρA dx dt = ρA = ρA t1 L 0 ∂y ∂ δy dx dt ∂t ∂t t2 t1 ∂y δy ∂t − t2 t1 δy ∂ 2y dt dx ∂t 2 but δy(t1 ) = δy(t2 ) = 0 thus t2 t1 L 0 t2 t1 δT dt = −ρA δy ∂ 2y dt dx. The kinetic energy term.Hamilton’s principle 93 F (x. Let us take the variation term by term. t) in transverse vibration. 4. ∂t 2 . t) dx Fig. t) δy dx dt.23) to give t2 t1 δW dt = t2 t1 0 L F (x. substituting for (4. The contribution from the work done by external forces is obtained by substituting from (4.
21). F ). EI ∂ 4y ∂ 2y + ρA 2 = F ∂x 4 ∂t (4.24) For δJ to be zero the integrands between 0 and L must be zero and since δy is arbitrary it follows that the individual terms in (4. t) δy dx dt − EI ∂ 4y ∂ 2y + ρA 2 − F δy dx dt ∂x 4 ∂t L + EI ∂ 2y ∂ δy ∂x 2 ∂x − 0 ∂ 3y δy ∂x 3 L dt = 0. ∂x 4 Finally. summing the terms. δJ = t2 t1 0 t2 t1 L 0 t2 t1 L − EI ∂ 4y ∂ 2y + ρA 2 ∂x 4 ∂t L 0 δy dx dt L − EI = t2 t1 ∂ 2y ∂ δy ∂x 2 ∂x ∂ 3y + δy ∂x 3 dt + 0 t2 t1 0 L F (x. constant transverse loading. using (4. the boundary conditions are ∂ 2y ∂ δy = 0 for x = 0 and x = L. ∂x 2 ∂x ∂ 3y δy = 0 for x = 0 and x = L.24) must also vanish so that (assuming uniform. 0 (4.94 Finite element method The elastic strain energy term is.25) and in addition. ∂x 3 . t2 t1 1 δU dt = 2 = EI = EI = EI t2 t1 δ 0 t2 L EI ∂ 2y ∂x 2 2 dx dt = t2 t1 0 L EI ∂ 2 y ∂ 2y 2 2δ 2 ∂x ∂x 2 dx dt L 0 t1 t2 t1 t2 t1 ∂ 2y ∂ 2 δy dx dt ∂x 2 ∂x 2 L ∂ 2y ∂ δy ∂x 2 ∂x ∂ 2y ∂ δy ∂x 2 ∂x − 0 L 0 L ∂ 3y ∂ δy dx dt ∂x 3 ∂x L − 0 ∂ 3y δy ∂x 3 + 0 0 L ∂ 4y δy dx dt.
Hamilton’s principle
95
which means either and either ∂ 3y =0 ∂x 3 or δy = 0 for x = 0 and x = L. ∂ 2y =0 ∂x 2 or ∂ δy = 0 ∂x for x = 0 and x = L (4.26)
Equation (4.25) is the equilibrium equation for the vibrating beam problem. However, note that Hamilton’s principle has also given us the boundary conditions for the problem in Equations (4.26). We shall ﬁnish by looking at one further example. 4.2.3.3 Transverse vibration of a beam with axial loading. We shall now consider the previous problem but with the addition of an externally applied axial force, P , as shown in Fig. 4.6. The axial force does work as the beam shortens due to transverse displacement (here, we are ignoring axial displacement which results from the axial strain due to the force, P ). An element of the beam, ds, shortens by ds − dx = = ⎧ ⎨ ⎩ dx 2 + dy 2 − dx = 1+ dy dx
2
1+
1 − 1 dx ≈ ⎭ 2
⎫ ⎬
dy dx
2
dx − dx ∂y ∂x
2
dx.
The work done by the axial force per differential length dWP = 1 P 2 ∂y ∂x
2
dx.
F (x, t ) P P
dy
ds dx
dx
Fig. 4.6 Simply supported beam subjected to axial force P while undergoing transverse vibration.
96
Finite element method
The ﬁrst variation of the work done by external forces then becomes
t2 t1
δW dt = = =
t2 t1 t2 t1 t2 t1 0 0 0
L
1 F (x, t) δy + δP 2 F (x, t) δy + P F (x, t) δy − P ∂y ∂x
∂y ∂x
2
dx dt
L
∂ δy dx dt ∂x
L
L
∂y ∂ 2y δy δy dx + P 2 ∂x ∂x
dt.
0
Thus, for uniform transverse loading F (x, t) = F , the equilibrium equation becomes ∂ 2y ∂ 2y ∂ 4y + P 2 + ρA 2 = F ∂x 4 ∂x ∂t and the boundary conditions are EI ∂ 2y ∂ δy = 0 for x = 0 and x = L, ∂x 2 ∂x ∂ 3y ∂ 2y EI 3 + P 2 = 0 for x = 0 and x = L, ∂x ∂x which means either and either EI ∂ 2y ∂ 3y +P 2 =0 ∂x 3 ∂x or δy = 0 for x = 0 and x = L. ∂ 2y =0 ∂x 2 or ∂ δy = 0 ∂x for x = 0 and x = L
4.3 Introduction to the ﬁnite element method
In this section, we will make use of Hamilton’s principle to obtain ﬁnite element equilibrium equations which we shall then apply to some simple, uniaxial problems, before considering some further ﬁnite elements, and their application in statics and dynamics. We shall start with an introduction to the ﬁnite element method. Figure 4.7 shows a representation of an undeformed body when the value of time is zero, which has been discretized into a ﬁnite number of tetrahedral ﬁnite elements which approximate the initial geometry of the body. A particular, single ﬁnite element is shown which has nodes at points P1 , P2 , P3 , and P4 in the undeformed conﬁguration. On the application of a load (which may be mechanical or thermal, for example), the body deforms (and undergoes
Introduction to the ﬁnite element method
97
Z, z F
Time = t Y, y p3 Time = 0 X, x P4 P1 P3 P2 F p4 p1 p2
Fig. 4.7 Schematic diagram showing the ﬁnite element discretization of a body with threedimensional
tetrahedral elements.
transformation ) to that shown in the current conﬁguration at time t. The nodes of the single element, after deformation, are now located at p1 , p2 , p3 , and p4 . For simplicity, let us suppose that the quantity we are interested in determining is temperature, chosen because it is a scalar variable. The basis of the ﬁnite element method is to assign nodes to the elements and to assume that we can determine shape functions to enable interpolation to give the value of the temperature at any point within the element in terms of the nodal values of temperature. The tetrahedral element shown in the ﬁgure has four nodes. If the temperatures at the four nodes are θ1 , θ2 , θ3 , and θ4 respectively, then the temperature anywhere within the element is given by
n
θ = N1 θ1 + N2 θ2 + N3 θ3 + N4 θ4 =
i=1
Ni θi ,
(4.27)
where Ni are called shape (or interpolation) functions. Let us consider in isolation the element shown in Fig. 4.7 in the deformed conﬁguration. The element is shown in Fig. 4.8(a) with respect to the current conﬁguration, and in (b) with respect to a local element reference frame (ξ1 , ξ2 , ξ3 ). A transformation is clearly needed to map the element from the local reference frame to the current conﬁguration, and similarly, from the local reference frame to the original conﬁguration, and we shall address this later. The shape functions for this element, in terms of the local variables, are N1 (ξ1 , ξ2 , ξ3 ) = 1 − ξ1 − ξ2 − ξ3 , N2 (ξ1 , ξ2 , ξ3 ) = ξ1 , N3 (ξ1 , ξ2 , ξ3 ) = ξ2 , N4 (ξ1 , ξ2 , ξ3 ) = ξ3 .
98
Finite element method
(a) x3 4 1 1 2 x1 3 x2 j1 2 3 j2 (b)
j3 4
Fig. 4.8 Fournoded tetrahedral element shown with respect to (a) the current conﬁguration and (b) the
local element reference frame.
An important feature of the shape functions in the ﬁnite element method is that they generally take a value of unity at their own node and are zero at all others; at node 1, for example, ξ1 = ξ2 = ξ3 = 0, so N1 = 1. They generally sum to unity: N1 + N2 + N3 + N4 = 1; there are however, some special elements for which this is not the case. With knowledge of position (ξ1 , ξ2 , ξ3 ) within an element, the shape functions can be used together with Equation (4.27) to determine the value of the temperature at any point within the element, given the nodal temperatures. The shape functions may be used in a similar way for any variable of interest, but often, the ﬁnite element equilibrium equations are set up with displacement as the basic quantity for which solutions are obtained. Such an approach is often referred to, therefore, as the displacementbased ﬁnite element method. We will look at a further, very simple ﬁnite element in order to examine the displacementbased approach. Figure 4.9 shows a uniform bar under axial force P which has been discretized with a number of uniaxial truss elements. Each element has two nodes and is of length L. Each node has just one degree of freedom; namely axial displacement, u. The bar lies along the xdirection in the current conﬁguration. The shape functions are N1 (ξ ) = 1 − ξ, N2 (ξ ) = ξ, (4.28)
where ξ = x/L and 0 ≤ ξ ≤ 1, and the element displacements are given by u(ξ ) = N1 u1 + N2 u2 . This is often written in vector form as u(ξ ) = N uI = [N1 N2 ] u1 . u2 (4.29)
Introduction to the ﬁnite element method
99
P x
P
x u1 N1 1 1 j N2(j) = j N2 1 1
L
u2
j
N1(j) = 1– j
Fig. 4.9 A uniform bar discretized using uniaxial truss elements with shape functions shown.
Because we know the displacement everywhere within the element, we can determine the small strain, which, for this simple, uniaxial displacement is just ε= = ∂u ∂N1 = ∂x ∂x 1 [−1 L 1] u1 u2 ∂N2 ∂x . u1 u2 = ∂N1 ∂ξ ∂ξ ∂x ∂N2 ∂ξ ∂ξ ∂x u1 u2 (4.30)
The derivatives of the type ∂ξ /∂x relate the current conﬁguration to the local element reference frame and, in effect, provide the mapping of the element from the current conﬁguration to the local element reference frame. In this case, because ξ = x/L, the mapping is trivial and the derivatives ∂ξ /∂x are easily obtained. We will see how to do this for more general cases a little later. The matrix of spatial derivatives of the shape functions, given in Equation (4.30), is often referred to as the B matrix where for this particular element, B= 1 [−1 L 1]. (4.31)
It can be seen from (4.30) that the strain is constant everywhere within the element; it is an example (as is the fournoded tetrahedran above) of a constant strain element. In order to progress with the analysis of the loaded bar, we need to obtain the equations of motion or equilibrium. We shall do this in a general way using Hamilton’s principle, and then return to the loaded bar and to uniaxial truss elements.
that is. ⎢γxy ⎥ ⎢2εxy ⎥ ⎢ ⎥ ⎢ ⎥ εxz εyz εzz ⎢γyz ⎥ ⎢2ε ⎥ ⎣ ⎦ ⎣ yz ⎦ γzx 2εzx Note that in vector. tensorial versus engineering strain Tensorial notation is very elegant when it comes to theoretical derivations. However.4 Finite element equilibrium equations 4. the shear strain components are stored as engineering shears.4. for the purposes of developing numerical algorithms for implementation into computer programs. Symmetry of stress and strain tensors is used to obtain the following (memory saving) notation (Voigt notation): ⎡ ⎤ σxx ⎢ ⎥ ⎢σyy ⎥ ⎡ ⎤ ⎢ ⎥ σxx σxy σxz ⎢σ ⎥ ⎢ ⎥ ⎢ zz ⎥ σ = ⎣σxy σyy σyz ⎦ → σ = ⎢ ⎥ . 0 0 µ 0 0⎥ ⎢ ⎢σxy ⎥ ⎥ ⎢ ⎢ ⎥ ⎥ ⎢ ⎥⎢ ⎢ ⎥ ⎢ 0 ⎢ σyz ⎥ 0 0 0 µ 0 ⎥ ⎢ γyz ⎥ ⎥ ⎢ ⎥⎢ ⎣ ⎦ ⎦ ⎣ ⎦⎣ γxz 0 0 0 0 0 µ σxz .1 Some preliminaries: tensor and Voigt notation. twice the tensor shears. or Voigt notation. so let us clarify by recalling Hooke’s law.100 Finite element method 4.32) ⎢σxy ⎥ ⎢ ⎥ σxz σyz σzz ⎢σ ⎥ ⎣ yz ⎦ σxz ⎡ ⎤ ⎡ ⎤ εxx εxx ⎢ε ⎥ ⎢ ⎥ ⎢ yy ⎥ ⎢ εyy ⎥ ⎡ ⎤ ⎢ ⎥ ⎢ ⎥ εxx εxy εxz ⎢ εzz ⎥ ⎢ ε ⎥ zz ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ε = ⎣εxy εyy εyz ⎦ → ε = ⎢ ⎥ = ⎢ (4. it is often more practical to work with arrays thus representing secondorder tensor (strain. stress) as onedimensional arrays and constitutive tensors (the elasticity tensor) as twodimensional arrays. (4. Writing stress and elastic strain as column vectors. Hooke’s law in three dimensions becomes ⎤ ⎡ ⎤⎡ ⎡ ⎤ λ + 2µ λ λ 0 0 0 εxx σxx ⎥ ⎢ ⎥⎢ ⎢ ⎥ ⎢ λ λ + 2µ λ 0 0 0 ⎥ ⎢ εyy ⎥ ⎢σyy ⎥ ⎥ ⎢ ⎥⎢ ⎢ ⎥ ⎥ ⎢ ⎥⎢ ⎢ ⎥ ⎢ λ ⎥ ⎢ εzz ⎥ λ λ + 2µ 0 0 0 ⎥ ⎢ ⎢ σzz ⎥ ⎥ ⎢ ⎢ ⎥ ⎥ ⎥⎢ σ = ⎢ ⎥ = Cε = ⎢ (4.33) ⎥. This sometimes causes confusion.34) ⎢ 0 ⎥ ⎢ γxy ⎥ .
therefore.35) stress–strain tensor We will now obtain the same shear stress. (1 + ν) 0 1 0 0 ⎞ ⎟ 0⎟ . as described by the stress and strain tensors ⎛ ⎞ ⎛ ⎞ 0 σxy σxz 0 εxy εxz ⎜ ⎟ ⎜ ⎟ σ = ⎜σxy ε = ⎜εxy 0 εyz ⎟ .37) .99).36) shows that we must have γxy = 2εxy and similarly for the other shear strains.36) and similar expressions gives E 1 2 2 (ε 2 + εyz + εxz ). for example.Finite element equilibrium equations 101 where λ = Eν/(1 + ν)(1 − 2ν) and µ = G = E/2(1 + ν) are the Lame constants. Let us consider the case of pure elastic shear. ⎝ ⎠ ⎝ ⎠ σxz σyz 0 εxz εyz 0 The elastic strain energy per unit volume is given by the tensor product ⎛ ⎞ ⎛ ⎞ 0 σxy σxz 0 εxy εxz ⎟ ⎜ ⎟ 1⎜ 1 σ : ε = ⎜σxy 0 σyz ⎟ : ⎜εxy 0 εyz ⎟ = (σxy εxy + σyz εyz + σxz εxz ) ⎝ ⎠ ⎝ ⎠ 2 2 σxz σyz 0 εxz εyz 0 and substituting for the shear stresses using (4. ⎠ 1 (4. 2(1 + ν) (4.36) Comparison of (4. this time using the relation. σxy = µγxy ≡ E γxy . given in Equation (2. ⎛ ⎞ σxx σxy σxz ⎜ ⎟ σ = ⎜σxy σyy σyz ⎟ = 2Gε + λTr(ε)I ⎝ ⎠ σxz σyz σzz ⎞ ⎛ ⎛ 1 εxx εxy εxz ⎟ ⎜ ⎜ = 2G ⎜εxy εyy εyz ⎟ + λ(εxx + εyy + εzz ) ⎜0 ⎠ ⎝ ⎝ εxz εyz εzz 0 The shear stress is. 0 σyz ⎟ . Let us take one of the shear terms.35) and (4. that is. A more elegant explanation of this is as follows. σxy = 2Gεxy ≡ E εxy . σ :ε= 1 + ν xy 2 (4.
If. for example. σT · σ = (σxx σyy σzz σxy σyz σxz ) · (σxx σyy σzz σxy σyz σxz )T 2 2 2 2 2 2 = σxx + σyy + σzz + σxy + σyz + σxz . vi ui i=1 v T · u = vi ui = so that v T · u = uT · v. written σ  is 2 2 2 2 2 2 σ  = (σ : σ )1/2 = [σxx + σyy + σzz + 2(σxy + σyz + σxz )]1/2 . It might be of interest to note that it is for this reason that Mohr’s circle for strain is drawn in terms of γxy /2(= εxy ) rather than γxy .38) Comparison of (4. for example.37) and (4. or vectors. We must therefore use the tensorial shear strain components (as opposed to the engineering shear) when rotating a strain. and similarly for the other shear strains. which is 2 2 2 2 2 2 σ : σ = σxx + σyy + σzz + 2(σxy + σyz + σxz ).38). It is therefore necessary to take care when using the vector (Voigt) notation for stress and strain in carrying out calculations. one that satisﬁes the criteria for objectivity discussed in Chapter 3.102 Finite element method We can also calculate the strain energy using engineering shear strain as 1 (τxy γxy + τyz γyz + τxz γxz ) 2 E 1 2 2 2 2 2 (γ 2 + γyz + γxz ). that is. we calculate the product of the stress vector. then from (4. given that the three shear strains are independent. therefore. gives γxy = 2εxy as before. It is important to note that this does not give the same result as the equivalent product of the stress tensor. an objective quantity. We write the dot product of two ﬁrstorder tensors. So. The strain tensor is. = G(γxy + γyz + γxz ) = 4(1 + ν) xy 2 (4. the norm of σ . and this means a quantity whose properties remain unchanged under rotation. of course. as n uT · v = ui vi = i=1 n ui vi .32).
∂σ 3. Gradient of J2 invariant in tensorial notation ∂J2 =σ ∂σ and its correct expression in Voigt notation ∂J2 = Sσ . 0⎥ ⎥ ⎥ 0⎥ ⎦ 2 The following are further examples of possible dangers in using Voigt notation: 1. 0 0 1 0 0 0 0 0 0 1 2 0 0 0 0 1 2 S −1 0 0 ⎥ 0⎥ ⎥ ⎥ 0⎥ ⎥.Finite element equilibrium equations 103 but if we work from the stress written as a column vector. Strain norm in tensorial notation √ ε = ε : ε and its correct expression in Voigt notation ε = where ⎡ ⎢ ⎢0 ⎢ ⎢ ⎢0 =⎢ ⎢0 ⎢ ⎢ ⎢0 ⎣ 0 1 0 1 0 0 0 0 εT S −1 ε. Second invariant of deviatoric stress in tensorial notation 1 J2 = σ : σ 2 and its correct expression in Voigt notation 1 J2 = σ T Sσ . 2 2. then σ  = σT Sσ. 0⎥ ⎥ ⎥ 0⎥ ⎦ 1 2 0 ⎤ 0 . 0 0 0 2 0 0 0 0 0 0 2 0 0 ⎤ which includes the following scaling matrix ⎡ 1 0 0 ⎢ ⎢0 1 0 ⎢ ⎢ ⎢0 0 1 S=⎢ ⎢0 0 0 ⎢ ⎢ ⎢0 0 0 ⎣ 0 0 0 ⎥ 0⎥ ⎥ ⎥ 0⎥ ⎥.
⎣ ⎦ ⎣ ⎦ uz tz Hooke’s law.30). The components of the Lagrangian for a general body subject to tractions t are given in Equations (4.32) and (4. the displacement and traction vectors are given by ⎡ ⎤ ⎡ ⎤ ux tx ⎢ ⎥ ⎢ ⎥ u = ⎢uy ⎥ and t = ⎢ty ⎥ .34). We shall see an example of this later. (4. .42) For the uniaxial truss element. and similarly for other vector quantities. strain ε. uI . We shall write them. the stress and strain vectors are given by Equations (4. is written as σ = Cεe .29) so that u = N uI (4.33). ε becomes just the unaxial. in general.39) in which all nonscalar quantities are represented as column vectors and in particular. using Voigt notation for a particular (the ‘m’th) ﬁnite element as follows 1 2 1 Um = 2 Tm = Wm = ∂ ˙ ˙ ρ uT u dV . as we did in (4. as we did for the truss element in Equation (4.41) and the strains can be written. we shall conﬁne ourselves initially to small elastic deformation problems so that we will not need to worry about the kinematics of large deformations. ε = BuI .18) using tensorial notation. (4. here. the other strain components are needed. respectively ˙ ˙ ˙ ˙ and the engineering shear strains are used. but for other element types (in two and three dimensions).40) The displacement within the particular ﬁnite element may be written in terms of the nodal displacements. εT σ dV . in terms of the nodal displacements and the B matrix. Note that uT u ≡ u · u. uT t dA.104 Finite element method 4.2 Finite element equations using Hamilton’s principle For the purposes of obtaining the ﬁnite element equilibrium equations. (4. For completeness.4. We will address large deformations later. scalar. from Equation (4.
(4.46) Equation (4. (4. therefore J = t2 t1 ∂ B T CB dV uI + uT I ∂ N T t dA.Finite element equilibrium equations 105 In applying Hamilton’s principle.46) is the ﬁnite element equilibrium equation.45) 1 T 1 ˙ ˙ uI muI − uT kuI + uT f . it reduces to the standard. for a single ﬁnite element ¨ muI + kuI = f . for the single element Lm = Tm − Um + Wm = Tm = 1 2 ˙ ˙ ρ uT u dV − 1 2 εT σ dV + ∂ uT t dA and substituting Equations (4.40)–(4. quasistatic equation kuI = f . uI .42) gives Lm = 1 T ˙ u 2 I 1 ˙ ρN T N dV uI − uT 2 I m= and k= and the vector of nodal forces as f = then the Lagrangian becomes Lm = The J integral is. In the absence of inertia forces. we shall start from the Lagrangian.44) N T t dA. written in terms of nodal displacements. I Since δuI is an arbitrary displacement. or equation of motion. If we deﬁne element mass and stiffness matrices as ρN T N dV (4.47) . I 2 2 I t2 t1 Lm dt = 1 T 1 ˙ ˙ uI muI − uT kuI + uT f dt I 2 2 I and applying Hamilton’s principle gives δJ = = t2 t1 t2 t1 1 1 ˙I ˙ δ(uT muI ) − δ(uT kuI ) + δ(uT f ) dt I I 2 2 ¨ δuT (−muI − kuI + f ) dt = 0. this gives for the equilibrium equation. Lm .43) B T CB dV (4. (4.
In Section 4.2. This is discretized with a single. mapping dV (or equivalently. Consider the uniform bar with crosssectional area A. P . . uniaxial truss element shown in Fig. 4. since ξ = x/L. those that are derived which include inertia (dynamic) effects—for example.10(b). dx) in the current conﬁguration to dV (or.10 (a) A uniform bar under axial force. This integral is currently given with respect to the deformed conﬁguration. so it needs to be transformed to the local element reference frame. and Young’s modulus E. so that N = [1 − ξ From above. N1 (ξ ) = 1 − ξ. (4. ξ ]. ξ . in general we would need to obtain the Jacobian. x. A 2 x2 u2 Fig. N2 (ξ ) = ξ The mass matrix can then be determined from (4. The shape functions are those given in (4. A P 1 L L (b) x1 u1 r. 4. made of material with density ρ. While trivial in this case. and those which do not include inertia effects (for quasistatic problems). 4. we shall simply refer to both types as equilibrium equations.1 Single truss element problem. E.31). the B matrix is B= 1 [−1 L 1]. shown in Fig.43) as m= ρN T N dV = L x=0 ρ 1−ξ ξ [1 − ξ ξ ]A dx.46)—are often referred to as momentum balance equations.4.2.4.28). E. 4. are simply called equilibrium equations. we shall determine the mass and stiffness matrices for a truss element and examine a singleelement problem.106 Finite element method Some authors differentiate between momentum balance equations and equilibrium equations.10(a). In most cases in what follows. dξ ) in the element reference (a) r. that given in (4.1. and (b) its discretization using a single truss element.
where in this simple uniaxial case. as we shall see later. .44) k= = B T CB dV = 1 ξ =0 L x=0 1 1 −1 E [−1 L 1 L 1]A dx E L2 1 −1 −1 1 AL dξ. t dA is the force acting on the area dA at nodes 1 and 2.51) to give EA 1 L −1 −1 1 u1 u2 = F1 P . that is. The equilibrium Equation (4.49) k= L −1 1 and the nodal force vector.45). (4. a quasistatic problem so that we may eliminate the inertia term from (4. m= L x=0 1 0 ρ 1−ξ ξ [1 − ξ ξ ]A dx = 1 ξ =0 ρ 1−ξ ξ [1 − ξ ξ ]AL dξ = ρAL so (1 − ξ )2 ξ(1 − ξ ) ξ(1 − ξ ) ξ2 m = ρAL dξ 1 3 1 6 1 6 1 3 . for this problem is simply f = F1 P (4. (4.48) The stiffness matrix is. For now. both of which act in a direction parallel to t. The integration of dA at the two ends therefore just gives A and is independent of ξ .46) therefore gives ρAL 1 3 1 6 1 6 1 3 u1 ¨ u2 ¨ + EA 1 −1 L −1 1 u1 u2 = F1 P . so that tA at node 2 is P and that at node 1 is F1 .51) Let us assume inertia forces are negligible and that this is. however. Note that because t is a traction (stress vector). and no integration over ξ is necessary. the elasticity matrix C is simply Young’s modulus.Finite element equilibrium equations 107 frame. using (4. (4. so EA 1 −1 (4. therefore. the xdirection. E.50) in which P is just the prescribed force at node 2 and F1 the currently unknown reaction at node 1.
which are given in terms of the local element reference. we need to give a fuller description of the mapping process between the current (and original) conﬁguration and the local element reference frame. these results are exact. in this particular problem. we have been using a constant strain element to discretize what is. or deformed. that u2 = [1 − ξ ξ] 0 u2 =ξ PL . the element shape functions and X I indicates the initial positions of the ﬁnite element nodal points P I . We will then look at several further examples. t) ≈ I =1 NI (ξ )x I (t) (4. EA where 0 ≤ ξ ≤ 1.7. ξ3 ) shown in Fig. (4.8. in which case. conﬁguration. In addition. these equations can be solved to give the expected result that PL u2 = . The initial position of the material particle P within the element shown can be speciﬁed as follows NNODE X(ξ . 4. (ξ1 .108 Finite element method With the boundary condition that u1 = 0. In the following section. this is usually called an updated Lagrangian formulation. NI (ξ ) represents. we shall give a more general description of the ﬁnite element method which includes the kinematics of large deformations and recognizes that the displacement of the nodes can become large.4. When carried out incrementally. 4.52) where NNODE denotes the number of ﬁnite element nodes. An alternative is to set up the equations with respect to the original conﬁguration. This is only because. F1 = −P EA and using u = N uI . t) ≈ I =1 NI (ξ )XI (t).53) . We will formulate the ﬁnite element equations with respect to the current. it is called a total Lagrangian formulation. a constant strain problem. Despite using just a single element. 4. the problem then becomes geometrically nonlinear. the current positions of the material particles at the time t are speciﬁed by NNODE x(ξ . ξ2 . in any case.3 General ﬁnite element approach We will now return to the body in the initial and current conﬁgurations shown in Fig. Assuming that the material particle P remains attached to the same ﬁnite elements during the motion. as before.
and the dyadic product.55) v(ξ .53). ∇. Elements for which this holds true are called isoparametric. two local reference frame independent variables are needed. are explained in Appendix A.52) speciﬁes X in terms of ξ .55).59) It is normally not possible to determine ∂ξ /∂X in Equation (4.2.1. The gradient term. therefore. t). t) = ∇X NI (ξ . (4. t) were therefore identical and in addition. t) and x(ξ . the same shape functions are used for interpolating position and displacement. The deformation gradient tensor is obtained by differentiating Equation (4. t) becomes 2 x(ξ. The derivative ∂X/∂ξ is called the Jacobian and relates . and for the general case (4. t) ≈ I =1 NNODE NI (ξ )uI (t).54). t) ≈ I =1 x I (t) ⊗ ∇X NI (ξ . In the example in Section 4. (ξ1 .52)–(4. ξ in Equations (4.56) In Equations (4. For twodimensional elements.53) with respect to the initial conﬁguration as follows NNODE F (ξ . ⊗. maps any point within an element. for example. The derivative can be rewritten as follows ∇ X NI (ξ . It is therefore necessary to determine ∂X/∂ξ instead and obtain the inverse. t) ≈ I =1 NI (ξ )xI (t) = (1 − ξ )x1 (t) + ξ x2 (t).59) directly since Equation (4. speciﬁed by ξ with respect to the local element reference system.Finite element equilibrium equations 109 where x I (t) denotes the current positions of the ﬁnite element nodal points pI .54) Equation (4. (4. (4. because the element considered was onedimensional. t) ≈ I =1 NI (ξ )v I (t). The displacement and the velocity ﬁelds within each ﬁnite element are approximated as follows NNODE u(ξ .58) ∂X(t) ∂X ∂Y ∂Z are the derivatives of the shape functions NI . (4. (4. t) = ∂NI (ξ ) ∂NI (ξ ) ∂ξ ∂NI (ξ ) = = ∂X(t) ∂ξ ∂X(t) ∂ξ ∂X(t) ∂ξ −1 where . onto the corresponding position in the current conﬁguration.4. X(ξ .52) and (4.57) ∂NI (ξ ) ∂NI ∂NI ∂NI = (4. x(ξ . For the previous truss element example. ξ2 ). the displacements were assumed to be small and there were no rigid body rotations. to specify position.53) was simply the scalar ξ .
are given by ∂xi = ∂ξj NNODE ∂x(t) ∂ξ −1 . t) + ∇x NI (ξ .65) xiI I =1 ∂NI . t) ≈ where 1 2 NNODE (v I (t) ⊗ ∇x NI (ξ .62) The velocity gradient is obtained as the spatial derivative of the velocity in Equation (4. ∂x/∂ξ . I =1 (4.56) as NNODE L(ξ . ∂ξ ∂ξ where the members of ∂X/∂ξ are given by ∂Xi = ∂ξj NNODE (4. (4. but write the increment in virtual work per unit volume and per unit time as δW = ¨ (div[σ ] − ρ u) · δv dV = 0. We shall see this in several examples later.60) XiI I =1 ∂NI . ∂X ∂NI (ξ ) = XI (t). t) = ∂NI (ξ ) ∂NI (ξ ) ∂ξ ∂NI (ξ ) = = ∇x NI (ξ . obtained from Hamilton’s principle. I =1 (4.61) The small strain tensor can be approximated ε(ξ . and are obtained from the shape functions as ∇x NI (ξ . t) ≈ I =1 v I (t) ⊗ ∇x NI (ξ . (4. ∂ξj (4.63) ∂NI (ξ ) (4.110 Finite element method inﬁnitesimal quantities in the material conﬁguration to those in the local element coordinate system. Using Equation (4.67) . t) = ∂x(t) ∂ξ ∂x(t) ∂ξ where members of the Jacobian.20).64) ∂x(t) are the spatial derivatives with respect to the current conﬁguration. t) and the rate of deformation tensor is obtained by using D = (1/2)(L + LT ) as follows D(ξ . t) ⊗ v I (t)).66) We return now to the equilibrium Equation (4.52). t) = 1 1 [∇u + (∇u)T ] ≈ 2 2 NNODE [uI (t) ⊗ ∇X NI (ξ ) + ∇X NI (ξ ) ⊗ uI (t)]. ∂ξj (4.
this may be rewritten to give NNODE δW ≈ I =1 δv I · NNODE T ¨ NI ρ uI [N1 .70) As σ is a symmetric tensor.69) is the internal work per unit volume per second. (4. A stress quantity is called work conjugate to the strain if their double contracted tensor product yields work. or true. this reduces to NNODE δW ≈ + ρ I =1 ¨ NI uI I =1 NNODE NI δv I dV NNODE σ : I =1 δv I ⊗ ∇x NI dV − ∂ t· I =1 NI δv I dA and as the nodal virtual velocities are independent of the integration. By noting that div[σ · δv] = div[σ ] · δv + σ : ∇δv.69). which describes the dynamic equilibrium of the element can be rewritten in terms of the ﬁnite element discretization as follows NNODE NNODE δW ≈ + − ρ I =1 ¨ NI uI I =1 NI δv I dV σ : t· 1 2 NNODE (δv I ⊗ ∇ x NI + ∇ x NI ⊗ δv I ) dV I =1 NNODE ∂ NI δv I dA. substituting into (4. .71) .Finite element equilibrium equations 111 where δv is an arbitrary virtual velocity in the current conﬁguration. .68) that ∇δv = δL and that because σ is symmetric. (4.69) The second term in the righthand side of (4. I =1 NNODE (4. n · σ δv = δv · t (4. . The corotational Cauchy. NNNODE ] dV NNODE + I =1 δv I · σ ∇x NI dV − I =1 δv I · NI t dA. stress is work conjugate to the rate of deformation. (4. .67) with the application of the divergence theorem gives δW = ¨ ρ u · δv dV + σ : δD dV − ∂ t · δv dA = 0. N2 . The spatial virtual work equation. it can be shown (with a little algebra) that σ : δL = σ : δD.
⎥. the internal.76) The nodal forces can be further assembled into global vectors (column matrices) by summing over all elements to give ⎡ ⎡ ⎤ ⎤ ⎡ ext ⎤ inert int f1 f1 f1 ⎢ inert ⎥ ⎢ int ⎥ ⎢f ext ⎥ ⎢f 2 ⎥ ⎢f 2 ⎥ ⎢ 2 ⎥ ⎢ ⎢ ⎥ ⎥ Finert = ⎢ . N2 . ⎦ un (t) at time t. ⎥. Fext = ⎢ . Since the discretized rate of virtual work. the ﬁnite element discretization can be expressed by the set of nonlinear equilibrium equations as follows: Finert (u) + Fint (u) − Fext (u) = 0 (4.73) (4.72) by introducing expressions for the equivalent element nodal force vectors as follows f Iinert = f Iint = f Iext = ¨ NIT ρ uI [N1 . (4. Finally. ⎦ . .79) . Equation (4. ⎦ ⎣ ⎣ inert fn int fn ext fn where n is the total number of points used in the discretization. ⎥ ⎣ . and f Iext represent the inertial. ⎥ ⎣ . NI t dA. and the external element nodal forces.77) Fint = ⎢ .75) ∂ where f Iinert .78) for the set (column matrix) of nodal displacements ⎡ ⎤ u1 (t) ⎢u (t)⎥ ⎢ 2 ⎥ u=⎢ . NNNODE ] dV .112 Finite element method The virtual work equation can ﬁnally be written in vector form as NNODE dW ≈ I =1 dv I · (f Iinert + f Iint − f Iext ) (4. (4. ⎥. (4. . ⎢ . f Iint . σ ∇ x NI dV . the element equation can be rewritten f Iinert + f Iint − f Iext = 0. ⎥ ⎢ . ⎦ .74) (4. respectively. (4. . . ⎥ ⎢ .72) must be satisﬁed for all cases of the arbitrary virtual velocities. ⎥ ⎢ .
80) in which the offdiagonal terms (shear terms) are doubled to satisfy work conjugacy as speciﬁed in (4.81) ∂z so that from (4. the internal work term originates from δW I = σ : δD dV and may be written using Voigt notation as δW I = where DT = (Dxx Dyy Dzz 2Dxy 2Dyz 2Dzx ) σ : δD dV = δDT σ dV .82) . (4. this can be written in Voigt notation as ⎡ ∂N ⎡ ⎤ I ⎤ 0 ∂NI ∂y 0 ∂NI ∂x ∂NI ∂z 0 ⎥ ⎥ ⎥ 0 ⎥ ⎥ ⎥ ⎥ ∂NI ⎥ ⎡vxI ⎤ ⎥ ∂z ⎥ ⎢ ⎥ ⎥ ⎢vyI ⎥ ⎥⎣ ⎦ 0 ⎥ ⎥ vzI ⎥ ⎥ ∂NI ⎥ ⎥ ∂y ⎥ ⎥ ∂NI ⎦ ∂x 0 ⎢ ∂x ⎢ ⎢ Dxx ⎢ 0 ⎢ ⎢ ⎥ ⎢ ⎢ Dyy ⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥ NNODE NNODE ⎢ 0 ⎢ Dzz ⎥ ⎢ ⎢ ⎥ ⎢ D=⎢ B I · vI = ⎥= ⎢ ∂NI ⎢2Dxy ⎥ I =1 I =1 ⎢ ⎢ ⎥ ⎢ ∂y ⎢ ⎥ ⎢ ⎢2Dyz ⎥ ⎢ ⎣ ⎦ ⎢ ⎢ 0 2Dzx ⎢ ⎢ ⎣ ∂N I (4.76) in a more familiar way by considering the internal work term and writing it in terms of Voigt notation. Returning to Equation (4. (4.80) the internal energy term becomes δW = I NNODE T δD σ dV = T I =1 B I dv I σ dV . The rate of deformation is given in terms of nodal velocities in (4.69).80). I =1 After some algebra.63) by D= 1 2 NNODE (v I ⊗ ∇x NI + ∇x NI ⊗ v I ). We will assume small strain elasticity for simplicity.Finite element equilibrium equations 113 We can rewrite Equation (4.
86) This equation.34).71) and considering an arbitrary virtual velocity. then from (4. δv. (4.J =1 NNODE B T CB J dV .J =1 B T CB J dV ⎠ uJ I (4. ku = f . in general. This is because the stiffness matrix.85) . I where k= I.83) can be written as δW I = δv T kuI . I (4. depends upon the B matrix which (4. k. we may write δW = I =1 NNODE T NNODE I NNODE T B I dv I Cε dV = I =1 B I δv I ⎛ C J =1 B J uJ ⎞ dV so that δW I = δv T ⎝ I NNODE I. gives ¨ mu + ku = f and for quasistatic problems.83) since we may also write the small strains in terms of the nodal displacements as ⎡ ∂N ⎤ I 0 0 ⎢ ∂x ⎥ ⎢ ⎥ ⎤ ⎡ ⎢ ⎥ ∂NI εxx ⎢ 0 0 ⎥ ⎢ ⎥ ⎥ ⎢ ∂y ⎢ ⎥ ⎢ εyy ⎥ ⎢ ⎥ ⎥ ⎢ ∂NI ⎥ ⎡uxI ⎤ ⎢ ⎥ NNODE ⎢ ⎥ 0 NNODE ⎢ 0 ⎢ εzz ⎥ ⎢ ∂z ⎥ ⎢ ⎥ ⎥ ⎢ ⎢ ⎥ ⎢uyI ⎥.84) When combined with (4.114 Finite element method If we assume purely elastic small deformation. B I · uI = ε=⎢ ⎥= ⎢ ∂NI ∂NI ⎥⎣ ⎦ ⎢2εxy ⎥ ⎢ I =1 I =1 ⎢ ⎥ ⎢ 0 ⎥ ⎥ uzI ⎥ ⎢ ∂x ⎢ ∂y ⎥ ⎢2εyz ⎥ ⎢ ⎥ ⎦ ⎣ ⎢ ∂NI ∂NI ⎥ ⎢ 0 ⎥ 2εzx ⎢ ∂z ∂y ⎥ ⎢ ⎥ ⎣ ∂N ∂NI ⎦ I 0 ∂z ∂x Equation (4. is nonlinear if the problem considered is geometrically nonlinear.
the stiffness matrix must be updated.11 A uniform bar undergoing axial vibration (a) shown schematically and (b) its discretization using two truss elements. l=1 1 0 2 1 j Fig. relative to the local element reference frame. A P L dx P + ∂P dx ∂x 2 A ∂ u dx ∂t 2 (b) x1 u1 x2 x3 1 1 . In the following sections.1 Axial vibration of a beam using two truss elements.12. . For geometrically nonlinear problems. A L/2 u3 3 Fig. ξ . The uniform bar shown in Fig. In a geometrically nonlinear problem. we look at a number of small deformation (geometrically linear) examples including a twoelement axial vibration problem in which element assembly is addressed. 4.Finite element equilibrium equations 115 contains the derivatives of the shape functions with respect to the spatial coordinates. The truss element formulation. 4.3. E. and at each increment. E.11(a) undergoes axial vibration. (a) .86) therefore has to be solved incrementally. the spatial coordinates change so that the stiffness matrix also changes. 4.4. Equation (4. 4. 4.12 Linear truss element formulation shown relative to the local element reference frame. a singleelement and twoelement bending problem in which beam elements are introduced. 4. is shown in Fig. A L/2 u2 2 2 . and a single square element static elastic problem in which linear fournoded (quad) elements are introduced. It is discretized using two truss elements shown in Fig. E.11(b).
60). the two elements are identical so that their mass and stiffness matrices. which is of the general form given in Equation (4. respectively. In this onedimensional problem.116 Finite element method The shape functions are N1 = 1 − ξ. are the same. Also. We shall determine the mass and stiffness matrices for a single element.43) with respect to the spatial (which in this case coincides with the material) coordinate system is m= ρN T N dV = L/2 X=0 ρN T NA dX but of course. Using the shape functions. because this is a problem in which the displacements are inﬁnitesimal. the determinant of the Jacobian is just det(J ) = ∂X ∂ = [N1 ∂ξ ∂ξ N2 ] X1 X2 = ∂N1 ∂ξ ∂N2 ∂ξ X1 X2 . Therefore. we write m= L/2 X=0 ρN T NA dX = 1 1 ξ =0 ρN T NA det(J ) dξ (4. N2 = ξ and in this problem. (4. the shape functions are given with respect to the local element reference frame.87) = ρA ξ =0 1−ξ [1 − ξ ξ ξ ] det(J ) dξ in which J is the Jacobian. the current and original conﬁgurations are taken to be the same. The mass matrix given by (4. we obtain det(J ) = [−1 1] X1 X2 = [−X1 + X2 ] = 0+ L 2 = L 2 = L 2 so that the element mass matrix becomes ρAL m= 2 1 ξ =0 1 − 2ξ + ξ 2 ξ − ξ2 ξ − ξ2 ξ2 ρAL dξ = 2 1 3 1 6 1 6 1 3 . and then consider their assemblage. mapping the spatial coordinate system onto the local element reference frame.88) .
becomes just the scalar Young’s modulus.44) in terms of the spatial coordinates by k= B T CB dV = L X=0 B T CBA dX. E. Introducing again the Jacobian in (4. 2 Because the truss elements are onedimensional. and note that this is the inverse of ∂ξ /∂X. 1 6 0 ⎤ (4. The B matrix contains the derivatives of the shape functions with respect to the spatial coordinates. ∂ξ = ∂X ∂X ∂ξ −1 =J −1 = L 2 −1 = 2 . in (4. so k= EAL 2 1 ξ =0 B T B dξ . Hence. the relationship is given by (4. That is. and for the truss element in this problem. and that the problem remains geometrically linear so that the spatial and material coordinates remain the same.89) The element stiffness matrix is given by (4.Finite element equilibrium equations 117 Let us write the mass matrix for elements 1 and 2 as 1 1 m2 m1 m1 ρAL 3 6 11 11 12 = .87). L . however. We are assuming small displacements. since we ordinarily know X in terms of ξ .87). this is X = [N1 N2 ] X1 . ∂ξ /∂X cannot be determined directly. 0 ⎤ = 1⎥ 6⎦ 1 3 1 6 ⎢1 ρAL ⎣ 12 0 1⎥ 12 ⎦ . C. therefore. X2 We have already determined the Jacobian derivative. In its most general form. the integral can be written with respect to the local element reference frame by k= 1 ξ =0 B T CBA det(J ) dξ = 1 ξ =0 B T CBA L dξ . the elasticity matrix. m2 = m1 = 1 1 1 1 2 m21 m22 m2 6 3 21 then the global mass matrix is given by ⎤ ⎡1 ⎡ 1 m1 0 m11 12 3 ⎢ 1 1 + m2 2 ⎥ = ρA L ⎢ 1 M = ⎣m21 m22 ⎣ 11 m12 ⎦ 2 6 2 2 0 m21 m22 0 1 6 2 3 1 6 m2 12 m2 22 ρAL = 2 ⎡ 1 12 1 3 1 12 1 3 1 6 1 6 1 3 . ∂ξ ∂X ∂ξ ∂X ∂x ∂x ∂X ∂X In general.52). ∂N1 ∂ξ ∂N2 ∂ξ ∂N1 ∂N2 ∂N1 ∂N2 B= ≡ = .
3.118 Finite element method In general. −2 0 2 u3 With the boundary condition u1 = 0.85) then becomes ¨ Mu + Ku = 0 so 1 6 ⎢1 ⎢ ρAL ⎣ 12 so that the global stiffness matrix become ⎡ ⎤ 1 1 ⎡ k11 k12 0 1 ⎢ 1 ⎥ 1 + k2 2 ⎥ = 2EA ⎣−1 ⎢k K = ⎣ 21 k22 11 k12 ⎦ L 0 2 2 0 k21 k22 ⎤ ⎡ −1 0 2 ⎦ = EA ⎣−2 2 −1 L −1 1 0 −2 4 −2 ⎤ 0 −2⎦ .91) The ﬁnite element equilibrium equation. The problem geometry and ﬁnite element discretization are shown in Figs 4. 1 1] so that the element stiffness matrix becomes k= EAL 2 1 ξ =0 2 L −1 [−1 1 1] dξ = 2EA 1 L −1 (4. given in (4.90) As for the mass matrices. The Hermitian shape functions in the element local coordinate system (note that dx = L dξ ) are obtained as follows: N = a0 + a1 ξ + a2 ξ 2 + a3 ξ 3 .14. therefore. the derivatives needed for the B matrix are obtained from the inverse of the Jacobian. this reduces to ρAL 1 3 1 12 1 12 1 6 EA 4 u2 ¨ + u3 ¨ L −2 −2 2 u2 0 = . 0 u3 4.4. we write the stiffness matrices for the two elements as k1 = 1 k11 1 k21 1 k12 1 k22 = 2EA 1 −1 . therefore.46) and (4. The B matrix is. B= ∂N1 ∂ξ ∂ξ ∂X ∂N2 ∂ξ 2 = −1 × ∂ξ ∂X L 2 1× 2 2 = [−1 L L −1 . .13 and 4.2 Cantilever beam in bending using a single element. L −1 1 k2 = 2 k11 2 k21 2 k12 2 k22 = 2EA 1 L −1 −1 1 (4. 2 ⎡ 0 1 12 1 3 1 12 ⎡ ⎤ ⎡ ¨ 2 ⎥ u1 EA ⎣ 1 ⎥⎣ ⎦ −2 ¨ 12 ⎦ u2 + L 0 u3 ¨ 1 0 6 ⎤ −2 4 −2 ⎤⎡ ⎤ ⎡ ⎤ u1 0 0 ⎦ ⎣u2 ⎦ = ⎣0⎦ .
⎪ dN2 ⎭ = 0⎪ ξ = 1. ⎪ dN3 ⎭ = 0⎪ dξ 1 0 1 j . 4. ⎫ dN1 ⎪ = 0⎪ ⎬ 1 dξ ⇒ N1 = 1−3ξ 2 +2ξ 3 . The element has four degrees of freedom. ξ = 1. I L Fig. ⎫ dN2 ⎪ = 1⎪ ⎬ dξ ⇒ N2 = ξ −2ξ 2 +ξ 3 . N2 = 0.Finite element equilibrium equations 119 r. dξ 1 0 1 j Third degree of freedom (y2 ) ξ = 0. I u1 1 x 2 L Fig. N3 = 1. First degree of freedom (y1 ) ξ = 0. 0 dξ 1 j Second degree of freedom (θ1 ) ξ = 0. A. E. E. N3 = 0. A. N1 = 0.13 Cantilever beam of uniform section with second moment of area I . ⎫ dN3 ⎪ = 0⎪ ⎬ dξ ⇒ N3 = 3ξ 2 −2ξ 3 . ⎪ dN1 ⎭ = 0⎪ ξ = 1. N2 = 0. v1 u1 v2 1 r.14 Single beam element discretization of a cantilever. 4. two translational and two rotational. N1 = 1.
⎪ dN4 ⎭ = 1⎪ dξ 1 0 1 j We shall now apply Hamilton’s principle in order to obtain the ﬁnite element equilibrium equations. we may write the kinetic. ⎫ dN4 ⎪ = 0⎪ ⎬ dξ ⇒ N4 = −ξ 2 +ξ 3 . U . ξ = 1. The ﬁnite element discretization is given by v(x. N4 = 0. For a beam in bending. and potential (strain). energies as T = 1 2 L 0 ρA ∂v ∂t 2 dx. T . = N1 (ξ )v1 (t) + N2 (ξ )L ∂x ∂x = N1 (ξ )v1 (t) + N2 (ξ ) which we shall write as v = Nu = [N1 (ξ ) N2 (ξ ) N3 (ξ ) N4 (ξ )] v1 ∂v1 ∂x v2 ∂v2 ∂x T . N4 = 0.120 Finite element method Fourth degree of freedom (θ2 ) ξ = 0. U= 1 2 L 0 EI ∂ 2v ∂x 2 2 dx. = 2 L L ⎢ ⎥ ⎢ L ∂ξ 2 ∂ξ 2 ∂ξ 2 ∂ξ 2 ⎢ v2 ⎥ ⎥ ⎢ ⎣ ∂v2 ⎦ ∂x The element kinetic and strain energies are given by T = 1 2 L 0 L 0 ˙ ˙ ρA(N u)T (N u) dx = 1 T ˙ u 2 ˙ ρAN T N dx u . t) = N1 (ξ )v1 (t) + N2 (ξ )θ1 (t) + N3 (ξ )v2 (t) + N4 (ξ )θ2 (t) ∂v2 ∂v1 + N3 (ξ )v2 (t) + N4 (ξ ) ∂ξ ∂ξ ∂v1 ∂v2 + N3 (ξ )v2 (t) + N4 (ξ )L . The derivatives in the strain energy term are given by 1 ∂ 2v 1 ∂ 2 N1 ∂ 2 N2 ∂v1 ∂ 2 N3 ∂ 2 N4 ∂v2 ∂ 2v + = 2 2 = 2 v1 + L v2 + L ∂x 2 L ∂ξ L ∂ξ 2 ∂ξ 2 ∂x ∂ξ 2 ∂ξ 2 ∂x ⎤ ⎡ v1 ⎥ ⎢ ⎢ ∂v1 ⎥ ⎥ ⎢ ⎢ ∂x ⎥ 1 ∂ 2 N1 ∂ 2 N2 ∂ 2 N3 ∂ 2 N4 ⎥ = Bu.
⎢ = 13L 156 −22L⎦ 420 ⎣ 54 −13L −3L2 −22L 4L2 (−ξ 2 + ξ 3 )L L dξ . is L=T −U = where m= 0 1 T 1 ˙ ˙ u mu − uT ku. (4. 2 2 ρAN T N dx EI B T B dx.94) Integrating the Lagrangian with respect to time. L. The element Lagrangian.Finite element equilibrium equations 121 and 1 U= 2 L 0 L 0 L 0 EI ∂ 2N u ∂ξ 2 EI T ∂ 2N u dx ∂ξ 2 T 1 = uT 2 = 1 T u 2 ∂ 2N ∂ξ 2 ∂ 2N ∂ξ 2 dx u EI B T B dx u.92) L (4.93) and k= 0 L (4. and taking the ﬁrst variation gives ¨ mu + ku = 0. The element mass matrix is ⎡ e ⎤ m11 me me me 12 13 14 ⎢ e ⎥ e e e ⎥ ⎢m ⎢ 21 m22 m23 m24 ⎥ m=⎢ ⎥= ⎢me me me me ⎥ 31 32 33 34 ⎦ ⎣ me me me me 41 42 43 44 ⎡ ⎤ 2 + 2ξ 3 1 − 3ξ ⎢ ⎥ 1 ⎢(ξ − 2ξ 2 + ξ 3 )L⎥ ⎢ ⎥ = ρA ⎢ ⎥ 2 − 2ξ 3 ⎥ 0 ⎢ ⎣ 3ξ ⎦ 2 + ξ 3 )L (−ξ L 0 ρAN T N dx × 1 − 3ξ 2 + 2ξ 3 (ξ − 2ξ 2 + ξ 3 )L 3ξ 2 − 2ξ 3 ⎤ ⎡ 156 22L 54 −13L ρAL ⎢ 22L 13L −3L2 ⎥ 4L2 ⎥.
3. −6L 12 −6L⎦ ⎣y2 ⎦ ⎣0⎦ 0 2L2 −6L 4L2 θ2 With the boundary conditions v1 = 0 and θ1 = 0 at x = 0. A uniform.15 and its ﬁnite element discretization using two beam elements in Fig. . = 3 ⎢ L ⎣−12 −6L 12 −6L⎦ 6L 2L2 −6L 4L2 The ﬁnite element equilibrium equation becomes 156 ⎢ 22L ρAL ⎢ 420 ⎣ 54 −13L 12 EI ⎢ 6L + 3 ⎢ L ⎣−12 6L ⎡ ⎡ 22L 4L2 13L −3L2 ⎡ ⎤ ⎤ ¨ 54 −13L ⎢y1 ⎥ ¨ 13L −3L2 ⎥ ⎢ θ1 ⎥ ⎥⎢ ⎥ 156 −22L⎦ ⎢y2 ⎥ ⎣¨ ⎦ −22L 4L2 ¨ θ2 ⎤⎡ ⎤ ⎡ ⎤ 6L −12 6L y1 0 2 2 ⎥ ⎢θ ⎥ ⎢0⎥ −6L 2L ⎥ ⎢ 1 ⎥ ⎢ ⎥ 4L = .3 Free transverse vibration of a propped cantilever using two beam elements. 4. the problem reduces to one with two degrees of freedom ρAL 156 −22L 420 −22L 4L2 EI 12 −6L x2 ¨ + 3 ¨ θ3 L −6L 4L2 x2 0 = . 0 θ3 4. 4. propped cantilever beam is shown in Fig.4.122 Finite element method The element stiffness matrix is given by ⎡ e ⎤ e e e k11 k12 k13 k14 ⎢ e ⎥ e e e ⎥ ⎢k ⎢ 21 k22 k23 k24 ⎥ k=⎢ ⎥= ⎢k e k e k e k e ⎥ 34 ⎦ 32 33 ⎣ 31 e k41 e k42 ⎡ e k43 e k44 L 0 EI B T B dx = 0 1 1 L2 ⎢ ⎥ ⎢(−4 + 6ξ )L⎥ 1 ⎢ ⎥ ⎢ ⎥ EI 2 ⎢ 6 − 12ξ ⎥ L ⎣ ⎦ (−2 + 6ξ )L (−2 + 6ξ )L]L dξ −6 + 12ξ ⎤ × [−6 + 12ξ (−4 + 6ξ )L 6 − 12ξ ⎤ ⎡ 12 6L −12 6L EI ⎢ 6L 4L2 −6L 2L2 ⎥ ⎥.16.
A. E. I L /2 u3 1 2 3 Fig. A.15 A uniform propped cantilever beam. I L Fig.16 Finite element discretization of a propped cantilever beam. I L /2 u2 y3 2 r. E. E. The two beam elements are identical. The element mass matrix (valid for both elements) is ⎡ ⎢ ⎢ ⎢ L ⎢ ⎢ 22 2 ρA(L/2) ⎢ ⎢ me = ⎢ 420 ⎢ ⎢ 54 ⎢ ⎢ ⎣ L −13 2 156 22 4 L 2 2 54 13 L 2 L 2 L 13 2 L 2 156 2 −3 −22 L 2 ⎤ L 2 ⎥ ⎥ L 2⎥ ⎥ −3 ⎥ 2 ⎥ ⎥ L ⎥ ⎥ −22 ⎥ 2 ⎥ ⎥ L 2 ⎦ 4 2 −13 and the stiffness matrix (valid for both elements) is ⎡ 12 ⎢ ⎢ ⎢ ⎢ ⎢ 6L EI ⎢ ⎢ ke = (L/2)3 ⎢ ⎢−12 ⎢ ⎢ ⎢ ⎣ L 6 2 L 6 2 4 L 2 2 L −6 2 2 −12 −6 L 2 12 −6 L 2 L 2 2 ⎥ ⎥ L 2⎥ ⎥ 2 ⎥ 2 ⎥ ⎥. A. 4. y1 u1 y2 1 r.Finite element equilibrium equations 123 r. 4. L ⎥ −6 ⎥ ⎥ 2 ⎥ ⎥ L 2⎦ 4 2 L 6 2 ⎤ .
⎥ ⎥ ⎥ 2⎥ ⎦ L 2 X L 2 L 2 X L 2 2 X 4 We proceed in a similar way for the mass matrix and obtain the ﬁnite element equilibrium equation ¨ Mu + Ku = 0 to be ⎤ 13 ⎡ 156 0 − L ⎡ ⎤ ⎢ 4 ⎥ y2 ¨ 172 ⎥ ⎢ ρAL ⎢ 3 2 ⎥ ⎢ ¨ ⎥ EI ⎢ ⎢ 0 L2 − L ⎥ ⎢θ2 ⎥ + 3 ⎢ 0 420 ⎢ 8 ⎥⎣ ⎦ L ⎣ ⎥ ⎢ ⎣ 13 ¨ 24L 3 2 1 2 ⎦ θ3 L − L − L 2 4 8 ⎡ ⎤⎡ ⎤ ⎡ ⎤ y2 0 ⎥⎢ ⎥ ⎢ ⎥ 4L2 ⎥ ⎢θ2 ⎥ = ⎢0⎥ . fournoded. twodimensional. The problem is shown schematically in Fig. . We will examine one further example before moving on. isoparametric element subjected to a static force. ⎦⎣ ⎦ ⎣ ⎦ 2 8L θ3 0 0 16L2 4L2 24L 4.3. together with the elastic constants. and assume conditions of plane strain.4 Static analysis of a single square element. a single.17.4. we may write for the ﬁrst element ⎡ ⎢ ⎡ ⎤ ⎡ ⎤ ⎢X y1 0 ⎢ ⎢ θ1 ⎥ ⎢ 0 ⎥ EI ⎢ ⎢ u = ⎢ ⎥ = ⎢ ⎥ → k1 = ⎣y2 ⎦ ⎣y2 ⎦ 3 ⎢X (L/2) ⎢ ⎢ ⎢ θ2 θ2 ⎣ X and for element 2 that ⎡ ⎢ 12 ⎢ ⎡ ⎤ ⎡ ⎤ ⎢ y2 y2 ⎢ L ⎢θ2 ⎥ ⎢θ2 ⎥ EI ⎢6 2 ⎢ u = ⎢ ⎥ = ⎢ ⎥ → k2 = ⎢ ⎣y3 ⎦ ⎣ 0 ⎦ (L/2)3 ⎢ ⎢X ⎢ θ3 θ3 ⎢ ⎣ L 6 2 6 4 L 2 2 X X X X X X X 12 −6 L 2 ⎥ ⎥ ⎥ ⎥ L ⎥ −6 ⎥ 2 ⎥ ⎥ 2⎥ L ⎦ 4 2 X X ⎤ X X X 2 6 2 L 2 ⎤ ⎥ ⎥ 2⎥ ⎥ ⎥ ⎥ ⎥. We take the material behaviour to be elastic.124 Finite element method With the boundary conditions. 4.
(4.Finite element equilibrium equations 125 Geometry a = 1 mm 1 a 2 a 3 F 45° Discretization data (nodal coordinates) n1(0. 1 2 4 We will ﬁnd that we need to carry out integrations (in order to obtain the stiffness matrix) of the shape functions. 1) 4 Material data E = 210 N / mm2 n = 0. as I= −1 −1 f (ξ. Often. –1 1 4 j –1 1 N4 (ξ.17 Schematic diagram of a single. n3(1. we will need to use a numerical technique. n2(0. this point is known as an integration point. ∂ξ ∂η (4. n4(1. 0). η) = (1 − ξ )(1 + η). The shape functions for this element are as follows: Element local coordinate system 1 N1 (ξ. η) = (1 + ξ )(1 − η). 1). η) = (1 + ξ )(1 + η). The integration will be performed numerically using a single integration point at P (ξ. we have been able to do the integrations analytically. in order to simplify the process. 0). the integration is carried out with respect to a particular point in the element. Previously. Here. 1 4 1 N3 (ξ. 0). η) = (1 − ξ )(1 − η).25 Loading F = 10 N Fig. η) det[J ] dξ dη. Integrals over the element domain of the type I= 1 1 f (ξ. J .95) where the Jacobian for a twodimensional problem is given by ⎡ ∂X ∂X ⎤ ⎢ ∂ξ ∂η ⎥ ⎥ J =⎢ ⎣ ∂Y ∂Y ⎦ . fournoded element subjected to force F directed at 45◦ to the vertical. 4. η) = P (0. η) dV are expressed in the element local coordinate system by use of the Jacobian.96) . h 4 1 4 3 N2 (ξ.
0) ⎢ ∂N2 (0. 0) are obtained from ∂N ∂N ∂ξ ∂N = = ∂X ∂ξ ∂X ∂ξ ∂X ∂ξ −1 (4. 0) ∂N1 (0. X = N1 X1 + N2 X2 + N3 X3 + N4 X4 . 0) ⎢ ⎥ ⎢ =⎢ ⎢ ∂N (0.0) ⎥ ⎥ (1 + ξ )(1)(0.0) ⎤ ⎥ (1 + ξ )(−1)(0. Y = N1 Y1 + N2 Y2 + N3 Y3 + N4 Y4 . therefore. 0) ⎦ ∂ξ ∂η ⎡ ⎤ −1 −1 ⎢ ⎥ 1 ⎢ 1 −1⎥ ⎥.0) ⎥ ⎢ 3 ⎣ ⎥ ⎢ ∂ξ ∂η ⎥ ⎢ (−1)(1 + η)(0. η). that is. let us ﬁrst write down the relationship between element and nodal positions. that ∂N1 ∂NI ∂N2 ∂N3 ∂N4 ∂X = X1 + X2 + X3 + X4 = XI ∂ξ ∂ξ ∂ξ ∂ξ ∂ξ ∂ξ . 0) ∂N (0.0) ∂ξ ∂η ∂N (0. 0) ⎥ ⎥ ⎢ ⎢ ⎥ 1 ⎢ (1)(1 − η)(0. 0) ∂N2 (0. η) = P (0.97) and the shape function derivatives with respect to element local coordinates are given by ⎤ ∂N1 (0.0) ⎥ ⎦ (1 − ξ )(1)(0. We see. 0) ∂N4 (0.126 Finite element method The integral in (4. 0) ⎥ ⎢ ∂ξ ∂η ⎥ ⎢ ⎡ ⎥ ⎢ (−1)(1 − η)(0. = ⎢ 4⎢ 1 1⎥ ⎣ ⎦ −1 1 ⎡ (1 − ξ )(−1)(0.0) In order to determine the Jacobian matrix ∂X/∂ξ . The shape function derivatives at the integration point P (ξ. 0) ⎥ = 4 ⎢ (1)(1 + η) ∂ξ 3 (0.0) ⎥ ⎢ ⎣ ∂N4 (0.95) will be approximated using Gauss quadrature (details may be found in any of the more specialized books on ﬁnite elements) by I≈ 1 1 −1 −1 det[J ] dξ dη f (ξ.
the Jacobian becomes ⎡ ⎤ −1 −1 ∂X ∂N X1 X2 X3 X4 1 ⎢ 1 −1⎥ ⎢ ⎥ =X = 1⎦ Y1 Y2 Y3 Y4 4 ⎣ 1 ∂ξ ∂ξ −1 1 = = = 1 −X1 + X2 + X3 − X4 4 −Y1 + Y2 + Y3 − Y4 −X1 − X2 + X3 + X4 −Y1 − Y2 + Y3 + Y4 1 −0 + 0 + 1 − 1 −0 − 0 + 1 + 1 4 −1 + 0 + 0 − 1 −1 − 0 + 0 + 1 1 0 4 −2 2 0 = 0 −0.61) as J =X ∂N . and the nodal coordinates. ∂ξ With the shape functions given above.5 . 0 The determinant of the Jacobian is det[J ] = det and its inverse is ∂X 0 0.5 0 ∂ξ ∂X ∂ξ −1 = 0 2 −2 .Finite element equilibrium equations 127 and similarly for other terms. evaluated at the integration point. The Jacobian matrix.25 −0. therefore.5 = det = 0. from (4.5 0. ⎢ = Y1 Y2 Y3 Y4 ⎢ ∂N3 ∂N3 ⎥ ⎥ ⎢ ⎥ ⎢ ∂η ⎥ ⎢ ∂ξ ⎥ ⎢ ⎣ ∂N4 ∂N4 ⎦ ∂η ∂ξ This can be written in the general form given in Equation (4.96) is ⎡ ∂X ∂X ⎤ ⎡ ∂N ∂NI ⎤ I XI XI ⎢ ∂ξ ∂η ⎥ ∂η ⎥ ⎢ ∂ξ ⎥=⎢ ⎥ J =⎢ ⎣ ∂Y ∂Y ⎦ ⎣ ∂NI ∂NI ⎦ YI YI ∂ξ ∂η ∂ξ ∂η ⎤ ⎡ ∂N1 ∂N1 ⎢ ∂ξ ∂η ⎥ ⎥ ⎢ ⎥ ⎢ ⎢ ∂N2 ∂N2 ⎥ ⎥ ⎢ ∂η ⎥ X1 X2 X3 X4 ⎢ ∂ξ ⎥. 0 .
µ=G= 2(1 + ν) ⎤ ⎡ 252 84 0 C = ⎣ 84 252 0 ⎦ N/mm2 .5 0. 0) 0 ⎢ ⎥ ∂x ⎢ ⎥ ⎢ ∂NI (0. 0. 0 0 84 λ= .5 0 0. the shape function derivatives with respect to the spatial (here. 0) ∂NI (0.81) ⎡ ⎤ ∂NI (0.5 ⎢−0. equivalent to the material) coordinates are ⎡ ⎤ −1 −1 ∂N ∂N ∂X −1 1 ⎢ 1 −1⎥ 0 −2 ⎥ = ⎢ = 1⎦ 2 0 ∂X ∂ξ ∂ξ 4⎣ 1 −1 1 ⎡ ⎤ ⎡ ⎤ −1 −1 −1 −1 ⎢ 1 −1⎥ 0 −0.5 0 −0. 0.5 0. 0) ⎦ −0.5⎦ .5⎥ ⎥ =⎢ ⎣ 0.5⎦ .5 −0.5 −0.5 B=⎣ 0 0.5 The B matrix can now be obtained (for plane strain) from (4.5 1 ⎢ 1 −1⎥ 0 −2 ⎥ ⎢ ⎥ = ⎢ ⎣1 ⎦ 2 0 =⎣ 1 1 1 ⎦ 0. 0 0 µ νE = 84 N/mm2 .5 −0.5 0 4 −1 1 −1 1 ⎡ ⎤ −0.5 0.5 0.128 Finite element method Finally.5 −0.5 0.5 The elasticity matrix for plane strain is ⎤ ⎡ λ + 2µ λ 0 C=⎣ λ λ + 2µ 0 ⎦ . ∂y ⎢ ⎥ ⎢ ⎥ ⎣ ∂NI (0.5 0 −0.5 ⎡ ∂y ∂x −0.5 0 0 0. (1 + ν)(1 − 2ν) E = 84 N/mm2 .5 −0.5 0 0 −0.5 ⎤ 0. 0) ⎥ ⎢ ⎥ 0 BI = ⎢ ⎥.
5 −0.5 −0.5 0 −0.5 0. ⎤ 0.5 ⎤ 0 −0.5⎥ ⎣ ⎢ 0 =⎢ ⎥ 84 252 0 ⎦ 0 −0.5 −0.5 0.5 0.5 ⎦ 0 0.2.5⎥ ⎡ ⎥ 252 84 ⎢ 0 ⎥ ⎢ −0.98) . 0) det[J (0.5 −0.5 0 0.5 ⎡ 84 −42 42 0 −84 42 −42 ⎢−42 84 0 −42 42 −84 0 ⎢ ⎢ 42 0 84 42 −42 0 −84 ⎢ ⎢ 42 84 0 42 −42 ⎢ 0 −42 =⎢ 42 −42 0 84 −42 42 ⎢−84 ⎢ ⎢ 42 −84 0 42 −42 84 0 ⎢ ⎣−42 0 −84 −42 42 0 84 0 42 −42 −84 0 −42 42 ⎡ 84 −42 42 0 −84 42 −42 ⎢−42 84 0 −42 42 −84 0 ⎢ ⎢ 42 0 84 42 −42 0 −84 ⎢ ⎢ 0 −42 42 84 0 42 −42 ⎢ =⎢ 42 −42 0 84 −42 42 ⎢−84 ⎢ ⎢ 42 −84 0 42 −42 84 0 ⎢ ⎣−42 0 −84 −42 42 0 84 0 42 −42 −84 0 −42 42 The equilibrium equation for this quasistatic problem is ku = f .5 ⎥ 0 ⎢ 0 84 ⎢ 0 −0.5 ⎥ ⎥ ⎢ ⎣ 0.5 −0.5 0 −0.5⎥ ⎢ 0.5 0 0. 0 ⎥ ⎥ −42 ⎥ ⎥ 42 ⎦ 84  (4.5⎦ 4 4 0. 0)].5 0 −0.5 ⎢ 0 0.5⎥ ⎥ ⎢ ⎢−0.5 0 0.5 0 1 0 0. 0)] CB(0.5 −0.5 ⎡ −0.Finite element equilibrium equations 129 The stiffness matrix can then be determined by using the approximate integration formula given above as k= B T CB dV = 0 T 1 0 1 V B T CB dx dy = 1 1 −1 −1 B T CB det[J ] dξ dη ≈ [B(0.5 0 ×⎣ 0 0.5 ⎤ 0 42 ⎥ ⎥ −42 ⎥ ⎥ ⎥ −84 ⎥ 1 ⎥ 4 0 ⎥4 ⎥ −42 ⎥ ⎥ 42 ⎦ 84 ⎤ 0 42 ⎥ ⎥ −42 ⎥ ⎥ ⎥ −84 ⎥ ⎥.2 ⎤ ⎡ −0.5 0.5 0.
011905 ⎡ ⎤ ⎡ ⎤ ux3 0. ⎦ ⎣ ⎦⎣ ⎦ ⎣ uy4 0 −42 84 0 The solution is obtained from u = k −1 f . ⎥ ⎢ux3 ⎥ ⎢ 4 ⎥ ⎥⎢ ⎥ ⎢ π⎥ ⎥ ⎢uy3 ⎥ ⎢ ⎥ ⎥ ⎢ ⎥ ⎢−F cos ⎥ ⎦⎣ 0 ⎦ ⎢ 4⎥ ⎢ ⎥ ⎣ Rx4 ⎦ uy4 0 Rx1 Ry1 Rx2 Ry2 ⎤ .005952 0.04209 The reactions can be obtained from ⎡ 84 ⎢−42 ⎢ ⎢ 42 ⎢ ⎢ ⎢ 0 ⎢ ⎢−84 ⎢ ⎢ 42 ⎢ ⎣−42 0 −42 84 0 −42 42 −84 0 42 42 0 84 42 −42 0 −84 −42 0 −42 42 84 0 42 −42 −84 −84 42 −42 0 84 −42 42 0 42 −84 0 42 −42 84 0 −42 −42 0 −84 −42 42 0 84 42 0 42 −42 −84 0 −42 42 84 ⎤⎡ ⎤ ⎤ √ 10 2/2 ⎥⎢ √ ⎥ 0.005952 ⎡ ⎢ ⎥ 0 ⎢ ⎥ ⎥⎢ 0 ⎥ ⎢ ⎥ ⎥⎢ ⎥ ⎢ ⎥ ⎥⎢ 0 ⎥ ⎢ ⎥ ⎥⎢ ⎥ ⎢ ⎥ ⎥⎢ ⎥ ⎢ π ⎥ ⎥⎢ 0 ⎥ ⎢ ⎥ ⎥ ⎢ ⎥ = ⎢ F cos ⎥ .011905 ux3 ⎢ ⎥ ⎢ ⎢uy3 ⎥ = ⎢0. ⎦ ⎦⎣ 0. ⎣ ⎦ ⎣ ⎦ uy4 −0.011905⎥ ⎢−10 2/2⎥ .130 Finite element method ⎡ 84 ⎢−42 ⎢ ⎢ 42 ⎢ ⎢ ⎢ 0 ⎢ ⎢−84 ⎢ ⎢ 42 ⎢ ⎣−42 0 −42 84 0 −42 42 −84 0 42 42 0 84 42 −42 0 −84 −42 0 −42 42 84 0 42 −42 −84 −84 42 −42 0 84 −42 42 0 42 −84 0 42 −42 84 0 −42 −42 0 −84 −42 42 0 84 42 0 42 −42 −84 0 −42 42 84 ⎤⎡ ⎤ ⎢ ⎥ 0 ⎢ ⎥ ⎥⎢ 0 ⎥ ⎢ ⎥ ⎥⎢ ⎥ ⎢ ⎥ ⎥⎢ 0 ⎥ ⎢ ⎥ ⎥⎢ ⎥ ⎢ ⎥ ⎥⎢ ⎥ ⎢ π ⎥ 0 ⎥ ⎢ ⎥⎢ ⎥ ⎥ ⎢ ⎥ = ⎢ F cos ⎥ ⎥ ⎢ux3 ⎥ ⎢ 4 ⎥ ⎥⎢ ⎥ ⎢ π⎥ ⎥ ⎢uy3 ⎥ ⎢ ⎥ ⎥ ⎢ ⎥ ⎢−F cos ⎥ ⎢ ⎦⎣ 0 ⎦ 4⎥ ⎢ ⎥ ⎣ Rx4 ⎦ uy4 0 ⎡ Rx1 Ry1 Rx2 Ry2 ⎤ so that with the boundary conditions it simpliﬁes to ⎤ ⎡ ⎤⎡ ⎤ ⎡ √ 10 2/2 84 −42 0 ux3 ⎢ ⎥⎢ ⎥ ⎢ √ ⎥ ⎢−42 84 −42⎥ ⎢uy3 ⎥ = ⎢−10 2/2⎥ .017857 0.08418⎥ mm.017857 0 ⎤⎡ 0.04209 ⎢ ⎥ ⎢ ⎥ ⎢uy3 ⎥ = ⎢−0.023810 ⎣ ⎦ ⎣ uy4 0.011905 0. ⎡ ⎤ ⎡ 0.
Fy = Ry1 + Ry2 − 2 2 2 The element strains at the integration point may be determined as NNODE ε(ξ.5 −0. We can carry out an equilibrium check as follows: √ √ √ 10 2 10 2 10 2 =− +0+0+ = 0. Fx = Rx1 + Rx2 + Rx4 + 2 2 2 √ √ √ F 2 10 2 10 2 = +0− = 0.021045 = ⎣ 0. 0) · uI ⎤ 0 ⎢ 0 ⎥ ⎢ ⎥ ⎤⎢ 0 ⎥ 0.021045 ⎦ −0.5 0.5 =⎣ 0 0.5 0 0 −0.5 −0.5 0.5 0 ⎣ 0 = 0. Ry2 = 0ux3 + 42uy3 − 84uy4 = 0. Rx4 = 42ux3 + 0uy3 + 42uy4 = 0.5⎦ ⎢ ⎥ ⎢u ⎥ 0. √ Ry1 = 42ux3 − 84uy3 + 42uy4 = 10 2/2N.5  ⎡ −0.5 0 −0.5 0 −0.5 .5 0 0 −0.5 0 ⎢ ⎥ ⎢ ⎥ ⎢ 0 ⎥ 0 0.5 0 −0.5 ⎤ 0.04209 ⎡ −0.5   0.5 ⎢ x3 ⎥ ⎢uy3 ⎥ ⎢ ⎥ ⎣ 0 ⎦ uy4 ⎤ ⎡ 0 ⎥ ⎢ 0 ⎥ ⎢ ⎥ ⎢ ⎤⎢ 0 ⎥ 0.5 ⎥ ⎢ 0.5   0 0.5 0.5 0.Finite element equilibrium equations 131 √ Rx1 = −84ux3 + 42uy3 + 0uy4 = −10 2/2N.5 −0. η) = ε(0. Rx2 = −42ux3 + 0uy3 − 42uy4 = 0.08418 ⎡  ⎡ −0.5 −0.5 0 −0.5 ⎢ ⎢−0.08418⎥ ⎥ ⎢ ⎦ ⎣ 0 −0.5 −0.5 0 ⎢ ⎥ 0 ⎥ ⎢ ⎦⎢ 0 0.5 −0. 0) ≈ I =1 B I (0.5 0.04209 ⎥ ⎥ 0.5 0.5 −0.
5 ⎦ ⎢ ⎥ 0 ⎣ ⎦ 0.071068 = ⎣ 7.5 √ ⎤ ⎡ −10 2/2 0. ⎥⎣ ⎦ ⎢ −0.5 ⎢ ⎢ 0 ⎣ 0.5⎥ 10 2/2 ⎥ √ ⎢ ⎥ ⎥ ⎢ ⎥ 0. This is intended .021045 ⎦ −0.5 ⎥ −10 2/2 ⎢−10√2/2⎥ ⎥ ⎢ ⎥ ⎥ ⎢ ⎥ 0.5 T 1 T =B σ 4=B σ=⎢ ⎢ 0.021045 252 84 0 σ(ξ.5 ⎢ √ ⎥ ⎢ 10 2/2 ⎥ ⎥ ⎢ ⎥ −0.071068 ⎦ = ⎣ 10 2/2 ⎦ N/mm2 .5 ⎢ 0 ⎢ ⎢−0.5 4 0 ⎢ ⎢ ⎢ 0 −0.5⎥ ⎢ √ ⎥ ⎥ ⎢ 10 2/2 ⎥ = ⎢ √ ⎥ N.071068 −10 2/2 ⎡ A further check is that the internal forces at nodal points must be equal to nodal external forces (note that the reactions are external forces too): ⎡ int ⎤ fx1 ⎢ ⎥ ⎢f int ⎥ ⎢ y1 ⎥ ⎢ ⎥ ⎢f int ⎥ ⎢ x2 ⎥ ⎢ ⎥ ⎢f int ⎥ 1 1 ⎢ y2 ⎥ ⎥= B T σ dV = B T σ dx dy f int = ⎢ ⎢ int ⎥ 0 0 V ⎢fx3 ⎥ ⎢ ⎥ ⎢ int ⎥ ⎢fy3 ⎥ ⎢ ⎥ ⎢ int ⎥ ⎢fx4 ⎥ ⎣ ⎦ int fy4 = 1 1 −1 −1 B T σ det[J ] dξ dη = B T σ ⎡ −0.5 0 0.5 0 ⎢ ⎢ ⎢ 0 −0.5⎥ ⎡ 10√2/2 ⎤ ⎢ ⎥ ⎢ ⎥ ⎥ ⎢ ⎥ 0 ⎥ ⎢ −0.5 0 ⎢ 0.5 0 ⎤ 1 4 1 1 −1 −1 dξ dη We need to summarize a few important points before leaving this section. we have used only one or two ﬁnite elements.5⎥ ⎢ ⎥ ⎥ ⎢ ⎥ 0 ⎥ −0.08418 0 0 84 ⎤ ⎡ √ ⎡ ⎤ 10√2/2 7.132 Finite element method and the stresses are given by ⎤ ⎤⎡ 0. η) = ⎣ 84 252 0 ⎦ ⎣ 0. η) = Cε(ξ. In all the examples considered. √ −7.
So far.4. for example.85). the explicit central difference time integration scheme.71)–(4. Most engineering problems require very large numbers of ﬁnite elements in order to obtain an accurate representation and solution. For small strain elastic–plastic material behaviour. we return to the equilibrium Equation (4. or an implicit scheme. In order to address a particular example. in this chapter. which is introduced brieﬂy later on.Finite element equilibrium equations 133 to aid in the development of understanding of the ﬁnite element method rather than to suggest that practical problems can be solved in this way. for plasticity. in particular. Solution of these problems requires in general the numerical integration of the momentum balance Equation (4. Naturally. However.69) but for simplicity. This required the solution of the Equation (4. We should also note that we have gone on to determine solutions for only quasistatic problems. Using Voigt notation the equation of virtual work becomes δW = δD T σ dV − t · δv dA = 0. using. and we consider small deformations. the equation becomes NNODE T NNODE δW = I =1 B I dv I σ dV − ∂ t· I =1 NI δv I dA. we σ = Cεe = C(ε − εp ) so NNODE T NNODE δW = I =1 B I δv I C(ε − ε ) dV − p ∂ t· I =1 NI δv I dA . We shall continue by introducing incremental ﬁnite element techniques for nonlinear material behaviour.75) from virtual work are applicable in general to linear and nonlinear material behaviour. that is. we have set up the ﬁnite element equilibrium (also often known as momentum balance) equations for a number of dynamic problems which we have not attempted to solve. ∂ With the ﬁnite element discretization as before. assume quasistatic conditions and so ignore inertia terms. 4. we have been concerned with both geometrically linear and nonlinear problems in which the material behaviour has always been assumed to be linear (elasticity). the calculations are then carried out on a computer.4 Finite element formulation for plasticity The equilibrium equations derived in (4.98).
and external force terms are written as increments so that k u − fp = f (4. (4. k.2.10(b).1 Single truss element undergoing elastic–plastic deformation. and the stiffness matrix. the displacement. plastic strain.J =1 B T CB J dV ⎠ uJ I NNODE NNODE − δv T I I =1 B T Cεp dV I − δv T I I =1 ∂ NI t dA. this becomes ⎛ ⎞ δW = I δv T ⎝ I NNODE I. I (4. As before.99) ku − f p = f . and to emphasize this.4.1. no such uniqueness holds for plasticity problems. we may then write the element equilibrium equation as (4. for an arbitrary virtual velocity.4.134 Finite element method and with the ﬁnite element discretization for the strain presented in Section 4. where NNODE fp = I =1 B T Cεp dV . The problem is shown in Fig.4. We now return to the single truss element which we examined for the case of elastic material behaviour in Section 4. An incremental approach to solving (4. the B matrix. For this problem.4. k= EA 1 L −1 −1 1 f .100) may alternatively be written as k u= where f = fp + f.10(a) and the single element discretization in 4.102) . 4. 4. remain unchanged as B= 1 [−1 L 1]. I In contrast to linear elastic problems in which there exists a unique relationship between stress and elastic strain.3.100) and NNODE fp = I =1 B T C εp dV .101) Equation (4.99) is therefore almost always necessary.
100) the equilibrium equation becomes EA 1 −1 u1 −1 F1 .101) for this element as 1 1 −1 −1 E εp A det(J ) dξ = EA εp B T C εp dV = fp = 1 L 1 ξ =0 so that from (4. let us now assign elastic linear strain hardening plasticity properties to the rod material and determine the rod extension on application of a given force.4. the incremental displacement for elastic and elastic–plastic conditions becomes PL . P . A E h .2. − EA εp = u2 P 1 L −1 1 With the boundary condition that u1 = 0.2. the increment in displacement now becomes that due to the elastic deformation together with that resulting from the plastic strain. this reduces to EA u2 − EA εp = P . u2 = Elastic : σ < σy . this just gives us an incremental form of the expression we obtained before for elasticity in Section 4. To demonstrate further the need for an incremental approach.4. which unsurprisingly in this small strain formulation has magnitude L εp . EA 1 PL 1 + . (4. In Section 2. however.1. we saw that for uniaxial linear isotropic hardening the relation between the increment in uniaxial stress and plastic strain was given by σ εp = h in which h is the strain hardening constant. For the case of plasticity.103) u2 = EA In the absence of plasticity. u2 = Elastic–plastic : σ ≥ σy . (4. P The plasticity ‘force’ is determined from (4.102) gives PL 1 1 PL L P + = + .Finite element equilibrium equations 135 and we write the incremental external force vector as F1 f = . Writing σ = P /A gives P εp = Ah and substituting into (4. L Rearranging gives PL + L εp .104) u2 = EA Ah A E h Therefore.
1 = 0.11 mm EA Ah 1000 × 1 1 × 100 4. however. were we to introduce nonlinear hardening where. However.11 = 0. Let us analyse incrementally the problem shown in Fig. 4.1 mm.18 A uniform bar under axial loading with the properties shown. for example. P1 = σy A = 10 N.1 Explicit integration using the central difference method The ﬁnite element discretization of the momentum balance equation for a damped system can be expressed in matrix form as follows: ¨ ˙ Mu + Cu + Fint = Fext . will be chosen to cause ﬁrst yield.104) shows that the displacement—load relationship then becomes dependent on the increment size. h is ﬁxed). We will apply the load incrementally. Subsequent displacement increments must take account of the plasticity so that u(2) = and u(2) = u(1) + u(2) = 0. The ﬁrst load increment. The integration of the equations can be carried out by means of the explicit central difference time integration scheme.18. 4. We see from Equation (4. and Fext the external force vector. once the material has yielded. The scheme is derived from Taylor series .136 Finite element method 11 N L = 10 mm E = 1000 MPa A = 1. where M is the mass matrix.1 mm u(1) = EA 1000 × 1 so u(1) = u(0) + u(1) = 0 + 0.e. Fint the internal force vector.5. h depends on plastic strain.21 mm. the displacement—load relationship becomes independent of increment size.0 mm2 sy = 10 MPa h = 100 MPa Fig. 1 × 10 10 × 1 PL L P + = + = 0. The displacement increment is given by 10 × 10 PL = = 0.104) that since the rate of hardening is constant (i.1 + 0.5 Integration of momentum balance and equilibrium equations 4. Equation (4. That is.
uN+1 − 2uN + uN −1 . ˙ uN+1 = uN + uN+1/2 t. Let tN+1 be the time increment between tN and tN +1 with uN = u(tN ) as illustrated in Fig. t2 + ··· 2 t2 ˙ ¨ u(t − t) = u(t) − u(t) t + u(t) − ··· . Assuming that the acceleration is constant between tN and tN +1 the central difference approximations can be rearranged to give the following secondorder integration scheme 1 ¨ ˙ uN+1 = uN + uN t + uN ( t)2 . 2 These are more commonly rewritten by deﬁning the intermediate velocities based on the assumption that the acceleration is constant between t0 and t0+1/2 as well as between tN−1/2 and tN+1/2 so that 1 ˙ ¨ ˙ u1/2 = u0 + u0 t 2 to give the leap frog explicit method ˙ ¨ ˙ uN+1/2 = uN−1/2 + uN t. The midstep velocities are deﬁned by ˙ ˙ uN−1/2 = u(tN−1/2 ).19. as follows. ( t)2 . as illustrated in Fig. 4. The central difference method can be applied with a varying time increment (which is particularly important if the response of the continuum is nonlinear). 4. ˙ ˙ uN +1/2 = u(tN +1/2 ). 2 The central difference approximations are obtained taking the difference of the above expressions to give the velocity uN+1 − uN −1 ˙ uN = 2 t and by summing the expressions for the accelerations u(t + ˙ ¨ t) = u(t) + u(t) t + u(t) ¨ uN = where uN = u(tN ).19. u.Integration of momentum balance and equilibrium equations 137 expansions of the displacements. 2 1 ¨ ˙ ¨ ˙ uN+1 = uN + (uN + uN +1 ) t.
the internal and external force vectors can be calculated as int FN +1 = Fint (uN+1 ). tN+1/2 which completes the N th time step. ext FN+1 = Fext (uN +1 ). ˙ ˙ uN+1/2 − uN −1/2 . 4.138 Finite element method ∆tN t N –1 t N –1/2 ∆tN + 1 tN t N + 1/2 üN t N +1 t N –1/2 N +1/2 uN uN +1 Fig. tN +1/2 = (tN + tN +1 ). Furthermore. . 2 2 The central difference formula for velocity is tN −1/2 = ˙ uN+1/2 = uN+1 − uN .19 Central difference integration scheme. 2 The discretization of the momentum balance equation for an undamped system can be obtained by substituting the acceleration term with its ﬁnite difference approximation as follows: ˙ ˙ uN+1/2 − uN−1/2 int ext M + FN = FN . Newland). where 1 1 (tN−1 + tN ).g. e. tN+1/2 tN+1/2 = which yields ˙ ˙ uN +1/2 = M−1 (Fext − Fint ) tN +1/2 + uN −1/2 and ˙ uN+1 = uN + uN+1/2 tN +1 . if the mass matrix is diagonalized the system of differential equations uncouples and can be solved independently for each degree of freedom (see. tN +1 while the central difference formula for acceleration is ¨ uN = where 1 ( tN + tN +1 ). Subsequently.
which gives the stability condition as t≤ 2 .Integration of momentum balance and equilibrium equations 139 4. is the element maximum eigenvalue.1. For a general multidegree of freedom system. A conservative estimate of the stable time increment is given by the following minimum taken over all the elements t ≤ min Lc .1 Stability of the explicit time stepping scheme. the stability condition becomes t≤ where ωmax = max{ωi } i 2 ωmax . Trying solutions uN = An where A is the ampliﬁcation factor gives A2 − (2 − ω2 ( t)2 )A + 1 = 0. The solution of a SDOF equilibrium equation u + ω2 u = 0 ¨ can be obtained using the central difference time stepping scheme as follows uN = ¨ uN+1 − 2uN + uN−1 = −ω2 uN ( t)2 giving the difference equation uN+1 − (2 − ω2 ( t)2 )uN + uN −1 = 0. ω ± ω2 ( t)2 1− 2 2 − 1. .5. The roots of this polynomial are ω2 ( t)2 A= 1− 2 For stability A ≤ 1. cd where Lc is the characteristic element length and cd is the current effective dilatational wave speed of the material.
The iteration starts from an initially guessed solution u0 and the corresponding tangential stiffness matrix k(u0 ).105) The process continues until the residual forces Ψn are smaller than a speciﬁed tolerance. (4.140 Finite element method 4.102). The correction u is calculated as follows: un = [K(un )]−1 Ψn and an improved solution is then obtained as follows: un+1 = un + un . The discretized static equilibrium equation given in (4. the initial tangential stiffness method.2 Initial tangential stiffness method.5. given in Equation (4.5. for example. The residual forces are calculated from Ψ0 = k(u0 )u0 − f . is now calculated always based upon the initial tangential stiffness matrix so that un = [k(u0 )]−1 Ψn and as before.2. . and the Newton–Raphson method.2.5. 4. The initial tangential stiffness method differs from the previous method only in that the correction to the displacement. We conﬁne ourselves here to quasistatic problems so assume inertia effects are negligible. 4. We brieﬂy introduce three techniques. the improved solution is obtained as un+1 = un + un and the process continues until the residual forces Ψn are smaller than a speciﬁed tolerance.1 Tangential stiffness method.105). the tangential stiffness method.2 Introduction to implicit integration Implicit integration schemes are often preferred to their explicit counterparts since they involve the determination of a residual force at each step and iteration within the step to minimize the residual force to within a speciﬁed tolerance. will not generally be satisﬁed unless a convergence occurs which can be expressed in terms of residual forces as follows k(u)u − f = Ψ = 0.
however. The two schemes introduced above. it is also possible to employ either implicit or explicit integration methods. the residual force is determined from = k(u)u − f = 0. regardless of whether the constitutive equations are integrated using implicit or explicit integration. In the overall solution process. In implementing plasticity models into ﬁnite element formulations. therefore. The iteration continues until the tolerance limit on residual force is achieved. in addition.Integration of momentum balance and equilibrium equations 141 4. As in the previous methods. or equilibrium equations.106) and may be written as +J u=0 so that J(un ) un = − (un ) and the displacement is updated by un+1 = un + un . termed the tangent stiffness matrix. to give the plastic strain increment). because of the iteration necessary in order to achieve convergence. Explicit schemes. are called explicit ﬁnite element methods. Finite element methods which employ implicit schemes for the integration of the momentum balance. and from (4. namely explicit and implicit integration. called the load stiffness matrix. may be approximated by ∂ (u) (u) + (4. or equilibrium. there are several possible combinations of implicit and explicit integration. For this additional integration. We shall examine this further in the context of integration of constitutive equations in Chapter 5. It can be seen that the implicit scheme offers the more robust overall approach. . ∂u The matrix J = ∂ /∂u is called the Jacobian (which has nothing to do with the Jacobian introduced above for the mapping of spatial to local element conﬁgurations). ﬁnite element methods which employ explicit schemes for the integration of the momentum balance. ﬁnite element equations. Care has to be taken. however.107) provides a linearization of (4. can often produce more rapid solutions. or effective tangent stiffness. Both implicit and explicit formulations are available in ABAQUS.5. it is often necessary.g. may be used for the integration of the momentum balance.3 Newton–Raphson method. Equation (4. Similarly. and are more appropriate for dynamic analyses. or equilibrium equations. (4.107) u + O( u2 ) = 0. are referred to as implicit ﬁnite element methods.106) it can be seen that this comprises a term corresponding to the internal forces.106) Using a Taylor expansion. to integrate a set of constitutive equations (e.2. and a further term corresponding to the ‘external’ forces. in choosing time step size and ensuring the calculated solution does not drift away from the true solution.
J. Bonet. E. John Wiley & Sons. Hinton. O. E. Owen.R. Prentice Hall. and Wood. K. An Introduction to Finite Element Computations. (1979). R. Theory of Elasticity. Finite Elements in Plasticity: Theory and Practice. New York. New York. Mechanical Vibration Analysis and Computation. (1980). (2000). New Jersey. J.E.W. Zienkiewicz. and Hinton. K.. John Wiley and Sons. Newland.L. Longman Scientiﬁc and Technical. London. Liu. and Moran.. (1989). D. D. (1983). Cambridge University Press. B. McGrawHill. New York. Nonlinear Finite Element Analysis for Continua and Structures. Swansea. S. Belytschko. T. 4th edition. (1989).J. Nonlinear Continuum Mechanics for Finite Element Analysis. McGrawHill. . The Finite Element Method.D. and Owen. (1997). and Goodier. Pineridge Press.R. and Taylor. D.N.J.J. R. Pineridge Press. Swansea. Timoshenko.C. (1996). Finite Element Procedures. revised edition.142 Finite element method Further reading Bathe.
linear isotropic hardening plasticity before looking at kinematic hardening and viscoplasticity. timeindependent. The consistency condition is written in terms of the stress tensor df (σ . To start. we shall return to the constitutive equations for plasticity introduced in Chapter 2 and see how they may be integrated and how the Jacobian or tangent stiffness may be obtained from them. (∂f/∂σ) · C(∂f/∂σ) − (∂f/∂p)[(2/3)(∂f/∂σ) · (∂f/∂σ)]1/2 (5.1) and that the plastic multiplier. Implicit and explicit integration of von Mises plasticity 5. ∂σ 2 σe If we write the tensor normal. p) = ∂f ∂f : dσ + dp = 0. for a von Mises material n= . we return to the determination of the plastic multiplier. let us redetermine the plastic multiplier more generally.2 Implicit and explicit integration of constitutive equations We saw from Chapter 2 that the yield function for isotropic hardening can be written f = σe − r − σy = 3 σ :σ 2 1/2 − r − σy = 0 (5. First. 5.1 Introduction In this chapter. we shall consider small strain. n. ∂σ ∂p ∂f 3σ = . We shall explore both implicit and explicit schemes for the integration of the constitutive equations.5.2) where all terms are written in Voigt notation. Before proceeding. in principal stress space is given by dλ = (∂f/∂σ) · C dε . without the constraint of working in principal stress space.
the vector normal. twice the tensorial shears. ∂p where without loss of generality. n. p) = n · dσ + ∂f dp = 0. provided the normal vector. so ⎡ ⎤ n11 n12 0 n = ⎣n12 n22 0 ⎦ . ⎡ ⎤ n11 ⎢ n ⎥ n = ⎢ 22 ⎥ . then n : dσ ≡ n · dσ and the consistency condition becomes df (σ . The increment in tensorial plastic strain is obtained from the normality hypothesis. as before dσ = C(dε − dεp ). that is.144 Implicit and explicit integration the consistency condition becomes df (σ . in Voigt notation. ∂σ . n. dε p = dλ which we may write in Voigt notation as dεp = dλn remembering that the shear strain terms in the Voigt strain vector are engineering shears. in Voigt notation contains twice the tensorial normal shear terms. 0 0 n33 Let us also write. we assume out of plane shears are zero. ⎣ n33 ⎦ 2n12 We can then see that ⎡ n11 n12 n : dσ = ⎣n12 n22 0 0 ⎤ ⎡ ⎤ ⎡ ⎤ dσ11 n11 0 ⎢ n ⎥ ⎢dσ ⎥ 0 ⎦ = ⎢ 22 ⎥ · ⎢ 22 ⎥ ≡ n · dσ ⎣ n33 ⎦ ⎣dσ33 ⎦ dσ33 2n12 dσ12 ⎤ ⎡ dσ11 0 0 ⎦ : ⎣dσ12 n33 0 dσ12 dσ22 0 so that. p) = n : dσ + ∂f dp = 0. ∂p We may now determine the plastic multiplier with Hooke’s law written in Voigt notation. Hooke’s law then becomes dσ = C(dε − dλn) ∂f = dλn.
σ. it is conditionally stable. dr = h dp = h dλt . The updated stress. .3) (5.6) = rt + drt . r = hp so that ∂f /∂p = −∂r/∂p = −h and n · C dε . and perhaps most importantly.Implicit and explicit integration of constitutive equations 145 and combining with the consistency condition. on the time step size. dε. However. = εt + dεt . dp = dλ. nt · Cnt + h (5. t. and those at the next increment forward in time. may then be dσt = C(dεt − dλnt ). then we may write dλt = nt · C dεt . (5. If we denote all quantities at time t with subscript t. σ + dσ may then be obtained.3) allows the plastic multiplier to be determined so that the stress increment can be obtained from (5. chosen. Second. t + t.4).4) With knowledge of the total strain increment. that is. written as σt+ εt+ rt+ p t t = σt + dσt . because it is an explicit scheme. Equation (5. Great care is therefore required in ensuring that the time step does not become too large such that erroneous results are obtained. The integration to obtain all quantities at the end of the time step. Third. the stress increment is given by dλ = dσ = C dεe = C(dε − dεp ) = C(dε − dλn). it may become unstable. t p p (5. Its advantage is its simplicity and it is straightforward to implement as we shall see later. there are a number of disadvantages which need to be considered. First. This is called a ﬁrstorder forward Euler explicit integration scheme. the accuracy of the integration depends. n · Cn + h Finally. n · Cn − (∂f /∂p) For linear isotropic hardening. in a similar way. This is called an explicit integration process. in which.5) t. for a von Mises material. we obtain ∂f n · C(dε − dλn) + dλ = 0 ∂p so that the plastic multiplier is given by dλ = n · C dε . of course. together with the stress.
146 Implicit and explicit integration the plastic multiplier given in Equation (5. However.1(b) shows a representation of an implicit scheme. in Section 5. In what follows.1 Schematic representations of (a) explicit integration and (b) implicit integration. we shall take all quantities to be those at the end of a time step. the stress at t + t is just σ and that at the beginning of the time step. at time t.1. using the radial return method.5) was obtained to ensure that at time.2. So. (a) (b) s2 sy st sy st + ∆t s2 sy st st + ∆t sy sy s1 sy sy sy s1 Fig. The accuracy remains dependent. Figure 5. In deviatoric stress space. and the plastic correction term is always directed towards the centre of the yield surface (because of the normality condition). on the time step size. is σ t . 5. σ tr t . known as the radial return method for von Mises plasticity.1) is satisﬁed. of von Mises plasticity equations.1 Implicit integration: the radial return method Figure 5. over many time steps. the plane stress von Mises ellipse becomes a circle. A trial stress increment is chosen which again takes the updated stresses. t. to drift away from the yield surface. the yield condition in (5. A step forward in time takes the updated stresses outside of the yield surface. The technique has therefore come to be known as the radial return method. . We may write Hooke’s law in multiaxial form in terms of stress and strain tensors as σ = 2Gεe + λ Tr(εe )I .1(a) shows a von Mises yield surface with a schematic representation of the explicit integration method described above. We introduce an implicit scheme. however. outside of the yield surface. it is possible for the solution. This is overcome by means of implicit integration of the equations which has the additional advantage of being unconditionally stable. unless speciﬁcally stated. the forward integration process does not ensure that the yield condition is also satisﬁed at time t + t.2. and as a result. The stress is then updated with a plastic t+ correction to bring it back onto the yield surface at time t + t. 5. t + t.
9) (5. using Equation (5.13) With some algebra. σe 3 which we may write as (5.7).8) (5. . 2 σe The stress may be expressed in terms of its deviatoric and mean as σ = σ tr − 2G p 1 σ = σ + (σ : I )I 3 so that with (5. ε) + λ Tr(εe + t ε)I (5. or trial stress. The elastic predictor. σ tr as follows. we show that σ tr − 1 (σ : I )I is just the 3 deviatoric of the trial stress. 1 σ tr − (σ : I )I = 2G(εe + ε) + λI (ε e + ε) : I − Kεe : II t t 3 = 2G(εe + ε) + λI (ε e + ε) : I − K(εe + ε − ε p ) : II t t t = 2G(εe + t = 2G(εe + t ε) + λI (ε e + t ε) : I − KI (ε e + t ε) : I ε) + (λ − K)I (ε e + t ε) : I ≡ σ tr .7) ε− ε p ) + λ Tr(εe + t ε− ε p )I εe = εe + t ε− εp Elastic predictor since Tr( ε p ) = 0. where K is the elastic bulk modulus.10) we obtain 1 σ σ + (σ : I )I = σ tr − 3G p 3 σe and rearranging gives 1 + 3G 1 p σ = σ tr − (σ : I )I .11) (5.Implicit and explicit integration of constitutive equations 147 The elastic strain at the end of the time step may be written as εe = εe + t so that σ = 2G(εe + t and so σ = 2G(εe + t ε) + λ Tr(ε e + t ε) I − 2G εp Plastic corrector (5.8). is denoted by σ tr = 2G(εe + t so that from (5.10) σ = σ tr − 2G ε p = σ tr − 2G pn.12) (5. 3σ .
18) f+ ∂ p For linear hardening. tr σe − 3G p − r − σy . (5. ﬁnally.15) (5.19) Rearranging gives d p= r (k) = rt + h p (k) . r = hp so that ∂r ∂r = = h. d p= tr σe − 3G p (k) − r (k) − σy . ∂ p ∂p Substituting (5.17) tr σe + 3G p = σe . 3G + h (5. as (5.16) (5. the deviatoric stress tensor from (5. we obtain 1 + 3G 1 + 3G p σe 2 σ : σ = σ tr : σ tr 3 tr σ : σ tr 2 1/2 tr ≡ σe .14) σe If we take the contracted tensor product of each side of this with itself. We write ∂f d p + · · · = 0. (5. The multiaxial yield condition is tr f = σe − r − σy = σe − 3G p − r − σy = 0.16). using (5. 3G + h We may write the integration in iterative form then.20) p (k+1) = p(k) + d p. p σe = σe (5.19) therefore gives tr σe − 3G p − r − σy + (−3G − h) d p = 0. This is generally a nonlinear equation in p which may be solved using Newton’s method. so that the plastic strain tensor increment is εp = σ σ tr 3 3 p ≡ p tr 2 σe 2 σe . The effective stress may then be determined from (5.17) into (5.18).14). or 1 + 3G This gives.148 Implicit and explicit integration We therefore obtain p σ = σ tr .
however. The implicit scheme. It is important to note that in the implicit scheme.Implicit and explicit integration of constitutive equations 149 and the elastic increment εe = and the stress increment is given by σ = 2G εe + λI ε e : I . In the implementation of a plasticity model into implicit ﬁnite element code. we need to provide a subroutine which contains both the integration of the plasticity constitutive equations (whether implicit or explicit) together with the material Jacobian or tangent stiffness matrix. we saw in Section 4. because it enables signiﬁcantly larger time increments to be used. such as ABAQUS standard. In an implementation within implicit ﬁnite element code. and ε− εp . for example. therefore avoiding ‘drift’ from the yield surface which can occur in the explicit scheme. The tangent stiffness matrix depends very much on the material behaviour. If convergence occurs after a given number of iterations. which does not depend upon knowledge of the Jacobian. and hence on the constitutive equations.5. in implementing a plasticity model into commercial codes such as ABAQUS explicit or LSDyna (which is an explicit code). This ensures that the yield condition. in the initial tangent stiffness method in Section 4. approximate implementation has to be developed. A further reason is that depending on the complexity of the plasticity model. all quantities are written at the end of the time increment.3 that the implicit integration of the momentum balance or equilibrium equations requires the determination of the Jacobian that comprises both the tangent stiffness matrix and the load stiffness matrix. Perturbation methods can allow the accurate numerical determination of the Jacobian. The determination of the material Jacobian for implicit ﬁnite element code is very much bound up with the integration of the constitutive equations used.5.3 that the Jacobian is required in the iterative procedure in minimizing the force residual. the tangent stiffness matrix is not required. given in Equation (5. the Jacobian does not inﬂuence the accuracy of the solution. In the following section. generally leads to much more rapid solutions. is satisﬁed at the end of the time increment. Typically.5.17). we shall address ﬁrst.2.g. it is therefore necessary to provide the material tangent stiffness matrix in addition to the integration of the plasticity constitutive equations. we need to provide a subroutine which contains the integration of the plasticity constitutive equations (whether implicit or explicit).2. but the rate at which convergence is achieved.2). It is useful to note from Section 4. We have now introduced both explicit and implicit integration of the plasticity constitutive equations.2. the material Jacobian may not be derivable in analytical terms so that a numerical. For implementations into implicit code. In explicit ﬁnite element code. the material Jacobian for an elastic material. It is for this reason that often approximate Jacobians are used (e.
∂ σ /∂ ε and ∂σ /∂ε are not the same thing. The material Jacobian becomes.150 Implicit and explicit integration then that for the timeindependent linear strain isotropic hardening plasticity model for which the implicit integration scheme was presented in this section. which may be written more succinctly as σ = (2GI + λII ) : εe (5. ⎛ ⎞ ∂ σ11 ∂ σ11 ∂ σ11 ∂ σ11 ⎜ ∂ ε11 ∂ ε22 ∂ ε33 ∂ γ12 ⎟ ⎛ ⎞ ⎜ ⎟ 2G + λ λ λ 0 ⎜ ⎟ ⎜ ∂ σ22 ∂ σ22 ∂ σ22 ∂ σ22 ⎟ ⎜ ⎟ ⎜ ⎟ 2G + λ λ 0 ⎟ ⎜ ∂ ε11 ∂ ε22 ∂ ε33 ∂ γ12 ⎟ ⎜ λ ⎟ ⎜ ∂ σ ⎜ ⎟ ⎟.1 Isotropic elasticity Hooke’s law may be written incrementally as σ = 2G εe + λI ε e : I . therefore. For example. d σ = ∂ σ11 ∂ ε12 1 ∂ σ11 ∂ σ11 = = . that is. 5. = ⎜∂ σ ∂ σ ∂ σ ∂ σ ⎟ = ⎜ ⎜ λ 33 33 33 33 ⎟ λ 2G + λ 0 ⎟ ⎜ ∂ ε ⎟ ⎜ ⎜ ⎟ ⎠ ⎜ ∂ ε11 ∂ ε22 ∂ ε33 ∂ γ12 ⎟ ⎝ 1 ⎜ ⎟ 0 0 0 2 2G ⎜ ∂ σ12 ∂ σ12 ∂ σ12 ∂ σ12 ⎟ ⎝ ⎠ ∂ ε11 ∂ ε22 ∂ ε33 ∂ γ12 Note that in general. for timeindependent isotropic linear strain hardening plasticity. the material Jacobian for plasticity is not quite so easy to obtain. We will deﬁne the material Jacobian here in the way it is required in the ABAQUS ﬁnite element code as ∂ σ /∂ ε where ∂ σ d ε ∂ ε in which the shear strains are taken to be engineering shears.21) in which I is the fourthorder identity tensor with the properties I : I = I : I = I and I : ε = ε : I = ε. Unfortunately. . for example.2 for the plasticity model considered above.3.3 Material Jacobian 5. It is derived in Section 5. for quadratic convergence. for conditions of plane strain or axial symmetry ( γ13 = γ23 = 0). the former Jacobian quantity is required in ABAQUS.3.21). ∂ γ12 ∂ ε12 ∂ γ12 2 ∂ ε12 where ∂ σ11 /∂ ε12 is obtained from Equation (5.
(5.3. where δp is not the plastic strain in the increment. δσe = δr = hδp.Material Jacobian 151 5.25).14) by applying the differential operator. δ p. δσ + δ pσ − 2 σe σe σe tr δσe + 3Gδ p = δσe .16).23) gives tr δσe .22) to eliminate δσe . h + 3G (5.24) tr δσe = δσe 1 − 3G . but an inﬁnitesimal quantity.19) in Equation (5. so that we may write δσe = δr = hδ p. and p respectively to give (after some algebra) tr σe δσ tr δσ + e σe σe σe σe − tr σe σ = δσ tr .24). and (5. Combining with (5. δ (in a similar way as used to ﬁnd the ﬁrst variation of an integral in Chapter 4) so that we obtain 1 + 3G 3G p 3G p δσe σ = δσ tr .2 Material Jacobian for timeindependent isotropic linear strain hardening plasticity We start from Equation (5. 1 + (3G/ h) (5. (5.25) We now use (5. we may write (5. h + 3G (5. from (5. = tr 2 σe 3 tr σ : σ tr 2 −1/2 3 tr 3 δσ : σ tr + σ tr : δσ tr 2 2 .22) Also.23) The yield condition is written as δf = δσe − δr = 0 so that for linear hardening.23) gives tr hδ p + 3Gδ p = δσe so δ p= Combining with (5. tr δσe = δ 1/2 3 tr 1 = σ : σ tr 2 2 3 1 tr σ : δσ tr .26) Consider the term.
. (5. We may write the deviatoric trial stress in terms of the deviatoric trial strain using Hooke’s law (because the trial stress is obtained assuming elasticity) as 1 δσ tr = 2Gδεtr = 2G δεtr − II : δεtr 3 and δεtr ≡ δε so 1 δσ tr = 2G δε − II : δε .28) gives δσ = Q 1 σ tr σ tr + RI : 2G δε − II : δε tr tr 3 σe σe σ tr σ tr 1 σ tr σ tr 2 : δε − Q tr tr : (II : δε) + 2GRδε − GRI : (II : δε).152 Implicit and explicit integration Substituting this together with the expression for σ in Equation (5. we are trying to relate δσ to δε.28) σe σe Remember that in deriving the Jacobian. tr σ tr 3 σe σe 3 σe e σ tr σ tr σ tr σ tr : (II : δε) = tr tr : I Tr(δε) = 0 tr tr σe σe σe σe and therefore δσ = 2GQ σ tr σ tr 2 : δε + 2GRδε − GRII : δε tr σ tr σe e 3 (5. − tr tr σ tr σe 2 1 + (3G/ h) σe σe e If we write 3 1 σe σe Q= − tr and R = tr 2 1 + (3G/ h) σe σe then (5. the stress is given in terms of its deviatoric by 1 δσ = δσ + II : δσ = δσ + KII : δε e = δσ + KII : (δε − δε p ) 3 = δσ + KII : δε.26) gives σe 3 1 σe σ tr σ tr (5.14) into (5. Finally. 3 Substituting into Equation (5.27) becomes σ tr σ tr δσ = Q tr tr + RI : δσ tr .27) δσ = : δσ tr + tr δσ tr .29) = 2GQ The second term of the righthand side is zero since σ tr is deviatoric so since I : I = I .
3 The ﬁrst term of the Jacobian is. therefore. Let us determine some of the terms using (5.29) gives δσ = 2GQ σ tr σ tr 2 : δε + 2GRδε + K − GR II : δε. it is called the consistent tangent stiffness. (5.Material Jacobian 153 Substituting into (5.30). which is used to integrate the plasticity constitutive equations. because it has been derived from the implicit backward Euler integration scheme.31) Equation (5. In this case. δσ11 = 2GQ tr σ11 1 tr tr σe σe tr tr tr tr tr tr × (σ11 δε11 + σ22 δε22 + σ33 δε33 + 2σ12 δε12 + 2σ13 δε13 + 2σ23 δε23 ) 2 + 2GRδε11 + K − GR (δε11 + δε22 + δε33 ). tr σ tr σe e 3 σ tr σ tr 2 + 2GRI + K − GR II tr σ tr σe e 3 (5. tr tr ∂δγ12 2 ∂δε12 σe σe For conditions of axial symmetry (so that the out of plane shears.30) provides the Jacobian or the material tangent stiffness matrix. ε23 do not exist) therefore. the Jacobian is the symmetrical matrix ⎞ ⎛ D11 D12 D13 D14 ⎜ ∂δσ D22 D23 D24 ⎟ ⎟. =⎜ ⎝ D33 D34 ⎠ ∂δε D44 D14 = and D14 is . ε13 and σ23 . σ13 .30) We may write this in the shortened form as δσ = 2GQ : δε. D11 = and the next is D12 = σ tr σ tr ∂δσ11 2 = 2GQ 11 22 + K − GR tr σ tr ∂δε22 σe e 3 σ tr σ tr ∂δσ11 2 = 2GQ 11 11 + 2GR + K − GR tr σ tr ∂δε11 3 σe e and so on. A shear term is given by D44 = tr σ tr 2σ12 σ tr σ tr ∂δσ12 1 ∂δσ12 1 2GQ 12 tr + 2GR = 2GQ 12 12 + GR = = tr tr tr ∂δγ12 2 ∂δε12 2 σe σe σe σe σ tr σ tr ∂δσ11 1 ∂δσ11 = = 2GQ 11 12 .
of course. 5.2 for isotropic hardening. We will ﬁnish by introducing semiimplicit integration schemes for plasticity. then at the end of the time step. 3 3σ −x 2 σe for kinematic hardening. in which x is.35) and (5.1 Linear kinematic hardening With the backward Euler scheme. 3 3 3 .32). as x t . t. Combining (5. and to introduce combined nonlinear kinematic and isotropic hardening.4 Kinematic hardening In this section.1. and obtain the consistent tangent stiffness. we shall use implicit (backward Euler) integration for the case of linear kinematic hardening. and increment in effective plastic strain.154 Implicit and explicit integration 5.36) 1 σ = σ − II : σ 3 so that with (5. itself deviatoric. t + t.33) 3 and with the normality hypothesis for von Mises plasticity.32) in which σ tr is the elastic trial stress.36) as 1 2 2 σ tr − 2G pn − II : σ = x t + pcn + σe n.34) (5. Hooke’s law may be written in its predictor–corrector form as σ = σ tr − 2G pn (5. We will then go on to address combined linear kinematic hardening with nonlinear isotropic hardening.2. as given in Section 5. we may write (5. We employ linear kinematic hardening so that the back stress increment is given by 2 c εp 3 and as a result.34) gives n= σ = xt + The deviatoric stress is 2 2 pcn + σe n. it becomes 2 x = x t + c εp (5.4.35) p is the (5. if we write the back stress at time. x= 2 x = x t + c pn. 3 3 where (5.
rather than at the end of the time increment.40) and substituting into (5. t. (5.32) we may write σ = σ tr − 2G pn so that from (5.38) 3σ −x 3σ −x 3 3 (σ − x) : (σ − x) .42) .35). 3G + c (5. 3 3 Taking the contracted product of both sides of (5. is deﬁned by σe = 3 (σ − x) : (σ − x) 2 n:n= 3 .Kinematic hardening 155 We saw earlier that we may write 1 σ tr − II : σ ≡ σ tr 3 2 2 pcn + σe n.37) (5.37) with n gives σ tr − 2G pn = x t + (σ tr − x t ) : n = n : n 2G p + and n:n= 2 2 pc + σe 3 3 so that (5. From (5. 2 (5.41) 2 Note that the effective trial stress is determined with respect to the back stress taken at time. Rearranging this equation gives p= tr σe − σe .39) gives 1/2 3 tr tr (σ − x t ) : (σ tr − x t ) ≡ σe = 3G p + pc + σe . : = 2 2 σe 2 σe 2σe 2 But the effective stress.39) 1/2 so that Equation (5. σe .38) therefore reduces to (σ tr − x t ) : n = 3G p + pc + σe . we may write n as n= 3 2 σ tr − 2G pn − x σe = 3 2 σ tr − 2G pn − x t − (2/3)c pn . σe Rearranging this equation for n gives n= 3 2 σ tr − x t σe + 3G p + c p (5.
. no iteration is required for its determination. therefore.42). Plastic and elastic strains and the stress may then be updated using (5. In implicit ﬁnite element code. εp = pn. we obtain σ = σ tr tr c/3G + σe /σe 1 + c/3G tr σe − σe 3G + c 3 2 σ tr − x t tr σe .156 Implicit and explicit integration The yield function for the case of kinematic hardening is f = σe − σy = 0 so that σe = σy . in addition to the integration of the plasticity constitutive equations.41). 3 2 σ tr − x t tr σe 3 2 σ tr − x t tr σe tr 1 − σe /σe 1 + c/3G + xt . with the implicit integration given above) tangent stiffness for linear kinematic hardening. it is also necessary to provide the material Jacobian—the tangent stiffness matrix.42) gives σ = σ tr − 2G and after a little algebra. for the case of linear kinematic hardening.40) and (5.e. closedform expression for p and in this instance. Here.42) is therefore an exact. 5. Using (5.1 Consistent tangent stiffness for linear kinematic Hardening. εp = εt + εe = εe + t εe = εe + t ε− εp . n may be written as n= We have therefore σ = σ tr − 2G pn = σ tr − 2G p and substituting for p from (5.1.42) to give n= 3 2 p (5. In Equation (5. Equation (5.4. none of the terms depends upon the effective plastic strain. we determine the consistent (i.43) σ tr − x t tr σe εp . σ = 2Gεe + λII : εe . .
44). tr tr σe σe tr σe /σe + c/3G 1 + c/3G R=− 1 3 σe tr 1 + c/3G 2 σe . δσ = 2GQδε + 2GR 2 (σ tr − x t ) (σ tr − x t ) : δε − K − GQ II : δε tr tr σe σe 3 (5. σe 2 3 tr 3 (σ −x t ) : δσ tr + δσ tr : (σ tr −x t ) 2 2 Substituting into (5. we obtain (1 + c/3G) tr σe tr δσ = δσ tr (1 + (c/3G)(σe /σe )) σe tr tr tr tr + σ tr (δσe /σe − δσe /σe ) − (δσe /σe − δσe /σe )x t . From the yield function in (5.45) which gives the consistent tangent stiffness. (σ tr − x t ) (σ tr − x t ) 2 : δε − GQII : δε tr tr σe σe 3 (σ tr − x t ) (σ tr − x t ) : δσ tr . tr σe = 3 tr (σ − x t ) : (σ tr − x t ) 2 −1/2 1/2 so tr δσe = 1 3 tr (σ −x t ) : (σ tr −x t ) 2 2 1 3 = tr (σ tr − x t ) : δσ tr .44) Now.Kinematic hardening 157 Applying the differential operator. δf = δσe = 0 so this reduces to (1 + c/3G) tr σe tr tr tr δσ = δσ tr (1 + (c/3G)(σe /σe )) − (σ tr − x t )(δσe /σe ).43). σe (5. 1 δσ tr = 2G δε − II : δε 3 so that δσ = 2GQδε + 2GR and ﬁnally. and writing Q= gives δσ = Qδσ tr + R As before. and rearranging.
46) tr f = σe − (3G + c) p − r − σy = 0. The yield function is f = σe − r − σy = 0 which becomes.47) gives d p= tr σe − (3G + c) p − r − σy . we have tr σe = σe − (3G + c) p tr in which σe and σe are given by tr σe = (5. for nonlinear hardening.4.49) .2 Combined isotropic and linear kinematic hardening The isotropic hardening evolution equation is dr = h dp in which h can be a function of p so that r is not necessarily linear.48) Let us simplify it for the cases of linear kinematic hardening and isotropic hardening only to compare with the previous equations. 2 dx = c dεp . We use Newton’s method to solve it by writing f+ and substituting (5. depend on p and the equation can be nonlinear. after substituting (5. and for the linear kinematic hardening. r depends on h which may.47) In this case. as before we may write 2 x = x t + c pn 3 and from Section 5. σe = 3 (σ − x) : (σ − x) 2 1/2 . If there is no isotropic hardening. then r = h = 0 and (5.46) 3 tr (σ − x t ) : (σ tr − x t ) 2 1/2 .158 Implicit and explicit integration 5.4. (5.48) becomes d p= tr σe − (3G + c) p − σy 3G + c (5. 3 Because of the assumption of linearity. 3G + c + h ∂f d p=0 ∂ p (5.
49) gives d p = 0. n. x t . so (σ tr − x) : n = σe + 3G p.49) becomes d p= where tr σe − 3G p − r − σy . If there is no kinematic hardening. x = 0 and c = 0 so (5. Here. and will ﬁnd the equations depending on quantities at the end of the time increment.50) (5.4.51) 3 and we have shown before that n : n = 2 .2. As before. (5. subsequent equations were set up in terms of x t .3 Introduction to implicit integration of combined nonlinear kinematic and isotropic hardening In the previous sections in which we assumed linear kinematic hardening. That is. σe = σe − (3G + c) p and σe = σy so (5. tr σe = 5. we wrote the kinematic hardening variable in terms of its value at the start of the increment. 3G + h 1/2 3 tr σ : σ tr 2 which is what we obtained before in Section 5. we will now introduce the more general case of nonlinear kinematic hardening.1. and the incremental value. the deviatoric stress can be written in predictor–corrector form in terms of the trial stress and plastic return as σ = σ tr − 2G pn and the normal. we obtain n= (σ tr − x) : n = 2 σe + 2G p n : n 3 (5. As a result. 2 σe Combining these equations and after a little algebra.Kinematic hardening 159 tr and for linear kinematic hardening. is given by 3σ −x . there is no iteration required and p is given by p= tr σe − σy 3G + c as before.52) .
However. .55) is a nonlinear equation in p for which we will need to use Newton’s iterative solution method. Equation (5.51) n= so that rearranging. 5.4 Semiimplicit integration of combined nonlinear kinematic and isotropic hardening In the semiimplicit scheme.56) and the yield function (5.160 Implicit and explicit integration We can also obtain from (5.54). (5. 3 σ tr − 2G pn − x 2 σe 3 σ tr − x . with complex plasticity hardening laws. Newton’s method is used to determine the effective plastic strain increment using Equation (5. this may be challenging.53) 1/2 where tr σe = 3 tr (σ − x) : (σ tr − x) 2 . for which we obtain d p= and tr σe (k) − 3G p (k) − r (k) − σy (k) tr 3G + (∂r (k) /∂ p) − (∂σe /∂ p) (5. Note now that because x depends on p (nonlinearly for the case of nonlinear kinematic hardening) as may r. Such an approach is called semiimplicit integration.55) tr with σe given by (5. n= Combining with (5.56) p(k+1) = p(k) + d p. the updates of all other quantities are carried out explicitly as follows: εp (k+1) = p(k+1) nt .52) gives tr σe = σe − 3G p. Often.54) The yield function is tr f = σe − r − σy = σe − 3G p − r − σy = 0 (5.55) written at the end of the time increment ensures that drift from the yield surface does not occur. A simpler approach is to integrate the effective plastic strain implicitly but to update the normal n and the internal variables—the isotropic and kinematic hardening—explicitly.50) and (5.4. tr Both σe and r are the two derivatives with respect to p need to be updated at every iteration for a fully implicit integration. 2 σe + 3G p (5.
2 (k+1) + γ x t p(k+1) x = c εp 3 and the isotropic hardening increment by r = b(Q − r) p(k+1) if the nonlinear evolution equations given in Chapter 2 are adopted. as ∂f ˙ εp = p ˙ ∂σ in which. the nonlinear kinematic hardening increment. x. x. for example. For both kinematic and isotropic hardening. the yield function is f = J (σ − x ) − r − σy . (5. is given by x (k+1) = x t + where. x. x. 2 σet p = εt + = ε − εp εp (k+1) . and 1/2 3 tr . p is now speciﬁed by a constitutive equation (as opposed ˙ to being determined by the consistency condition) which. tr σe (k+1) = 5.57) . for viscoplasticity. for a viscoplastic von Mises material. therefore. r). r) t = φ t. about both stability and accuracy when using it. (σ − x (k+1) ) : (σ tr − x (k+1) ) 2 The semiimplicit scheme is not unconditionally stable and we need to be concerned. r. (k+1) . (k+1) σ (k+1) = 2Gε r (k+1) = rt + e(k+1) + λII : εe . can be written as p = φ(σ . for both isotropic and kinematic hardening. ˙ We may write this incrementally as p = φ(σ .5 Implicit integration in viscoplasticity We saw in Chapter 2 that the plastic strain rate can be written.Implicit integration in viscoplasticity 161 where nt = εp εe (k+1) (k+1) 3 σ t − xt .
and we start from the viscoplastic constitutive equation. Using (5. as p = φ(σe .64) As an example.62) (5.162 Implicit and explicit integration 5. If we assume for now linear isotropic hardening.63) and rearranging gives d p= φ( p.62) gives p−φ t + 1− We will write.59) this becomes p = φ( p. from here on. Using Newton’s method. r) = α sinh β(σ tr − 3G p − r − σy ) ˙ . r) t (5. r) − p/ t .63) = ∂φ ∂ p and φr = and similarly for other derivatives.58) and we may write this in a form suitable for Newton’s iterative solution as ψ = p − φ(σe .60) so that (5.61) and differentiating (5. then dr = h dp = h d p. (5. r) t = 0. we have ψ+ ∂ψ ∂ψ d p+ dr = 0 ∂ p ∂r ∂φ t ∂ p ∂φ t dr = 0. given by p = φ(σe .16) tr σe + 3G p = σe (5. we may write the stress in terms of p by remembering (5. let us consider a particular viscoplasticity constitutive equation. 1/ t − φ p − hφr (5. φ p d p− (5. for simplicity.59) In preparation for differentiation.59) becomes ψ( p.57) may be written in uniaxial form. r) = p − φ( p. and β are material constants. Substituting into (5. r) = α sinh β(σe − r − σy ) ˙ in which σy . r) t = 0. ∂r ∂φ ∂r (5.61) and substituting into (5. if we consider isotropic hardening only for now. Equation (5.5.1 Uniaxial viscoplasticity equations We will present implicit backward Euler integration for the uniaxial form of the equations ﬁrst. α.
as before. ∂ p ∂x ∂r We may write the derivatives as ψ+ ∂ψ =1−φ ∂ p p (5. an iterative procedure is used to determine p from p(k+1) = p(k) + d p.64) gives d p = with r = rt + h p. r) ˙ in which x is the tensorial kinematic hardening variable. r) = p − φ( p. r) t = 0 and ∂ψ ∂ψ ∂ψ d p+ : dx + dr = 0. and the plastic strain tensor increment can then be determined from εp = pn = σ σ tr 3 3 p ≡ p tr 2 σe 2 σe α sinh β(σ tr − 3G p − r − σy ) − p/ t 1/ t +3Gαβ cosh β(σ tr −3G p−r −σy )+hαβ cosh β(σ tr −3G p−r −σy ) p = −3Gαβ cosh β(σ tr − 3G p − r − σy ) so that the elastic strain and hence stress increments can be determined in the usual way.64) still holds for the multiaxial case. Equation (5. As before in rateindependent plasticity. ψ( p.2 Multiaxial viscoplasticity For the case of isotropic hardening. x. x. We shall next consider combined multiaxial linear isotropic and kinematic hardening such that the constitutive equation is written as p = φ( p.5.65) t. ∂ψ = −φ x t. ∂x ∂ψ = −φr t ∂r . x. 5.Implicit integration in viscoplasticity 163 and the required derivatives are φ and φr = −αβ cosh β(σ tr − 3G p − r − σy ) so that (5. The problem may be written.
66) and rearranging gives dx = d p= 1−φ p φ t− p . x.65) becomes p − φ t + (1 − φ p t) d p − φ x : dx t − φr dr t = 0. tr 2 J (σ − x ) 2 σe 5.70) . 2 2 c dεp = c d pn 3 3 so that substituting into (5. The plastic strain increment is then determined from εp = σ −x 3 σ tr − x 3 p ≡ p .67) gives d p= φ − p/ t 1/ t − φ p − hφr + cφσe (5.3 Consistent tangent stiffness for viscoplasticity with isotropic hardening The predictor–corrector form of the stress is σ = σ tr − 2G pn where n= or σ = σ tr − 2G pn. (5. (5.5. ˙ 3 σ −x ∂φ ∂σe = −φσe = −φσe n ∂σe ∂x 2 J (σ − x ) and the term φ x : n may be simpliﬁed to φx = − 3 φ x : n = −φσe n : n = − φσe . 2 Substituting into (5.164 Implicit and explicit integration so that (5. r) = α sinh β(J (σ − x ) − r − σy ) = α sinh β(σe − r − σy ).66) Let us write the linear isotropic and kinematic hardening equations as dr = h dp = h d p. t − hφr t − (2/3)cφ x : n t (5.67) If we consider again a viscoplastic constitutive equation with kinematic and isotropic hardening of the form p = φ(σ .69) 3 σ tr 3σ = tr 2 σe 2 σe (5.68) then which is solved iteratively.
r) t 3 σ tr σ tr : δσ tr . tr δσe = σe tr σ tr σe tr δσe σe δσe − tr tr tr σe σe σe σe tr δσ + tr σe σ tr .57). (5. tr tr 2 σe σe (5. Rearranging (5.75) .70) gives δσe = δσ = 3 σ tr σ tr σe σe 3 σ tr σ tr : δσ + tr δσ tr − tr : δσ tr . we will substitute for this using the differential form of (5. (5. tr tr 2 σe σe =− Now.73) σe 2 Substituting (5.70) gives σ = and applying the differential operator δσ = Now. tr tr tr tr 2 σe σe σe σe 2 σe σe and Because of the contracted product of δσ in the ﬁrst term on the righthand side.73) into (5.72) 1 3 σ : δσ .71) and eliminating σ using (5.74) 3 σ tr σ tr : δσ tr . from (5. The viscoplastic constitutive equation is written as. (5.50) δσ = δσ tr − 2Gδ pn to give δσ = 3 σ tr σ tr σe σe 3 σ tr σ tr : (δσ tr − 2Gδ pn) + tr δσ tr − tr : δσ tr tr tr tr tr 2 σe σe σe σe 2 σe σe 3 σ tr σ tr σe σe : 2Gδ pn + tr δσ tr + 1 − tr tr σ tr 2 σe e σe σe 3 σ tr 3 :n=n:n= tr 2 σe 2 so δσ = − σ tr σe σe : 3Gδ pn + tr δσ tr + 1 − tr tr σe σe σe p = φ(σ .72) and (5.Implicit integration in viscoplasticity 165 and tr σe = σe − 3G p.71) 1 3 tr σ : δσ tr tr 2 σe (5.
tr σe σe σe (5.74) gives δσ = − σe σe 3 σ tr σ tr σ tr φσ : δσ tr n+ tr δσ tr + 1− tr : δσ tr . and hence φσ : δσ tr = φσ : (δσ tr + (1/3)δσ tr : II ) = φσ : δσ tr since φσ : I = 0.79) which gives the consistent tangent stiffness.76) Substituting this together with the second equation of (5. . δσ = δσ + KII : δε so that (5. 3 Because φ is a function of the effective stress. 1/ t + 2Gφσ : n + φr q(r) (5. Let us assume isotropic hardening of the form δr = q(r)δp. : 3G tr tr tr σe 1/ t +2Gφσ : n+φr q(r) σe σe 2 σe σe 3G . (5.77) (5. this can be written as δr = q(r)δ p.80) σe .69) for δσ into (5.166 Implicit and explicit integration so that δ p = (φσ : δσ + φr δr) t. 1/ t + 2Gφσ : n + φr q(r) δσ = α Now. In addition.79) becomes δσ = 2Gα σ tr σ tr 2 σ tr φσ + K − Gβ II + 2Gγ tr tr tr σe 3 σe σe : δε + 2Gβδε (5.78) Let us write α=− then β= γ = . Because δp is again an inﬁnitesimal increment in p (as opposed to the plastic strain increment). σ tr σ tr σ tr φσ : δσ tr + βδσ tr + γ tr tr : δσ tr . its derivatives with respect to the stress components are deviatoric.76) gives δ p = (φσ : (δσ tr − 2Gδ pn) + φr q(r)δ p) t and rearranging δ p= Substituting into (5. tr σe σe 3 1 − tr 2 σe φσ : δσ tr . 2 δσ tr = 2Gδε = 2Gδε − G(δε : I )I .
but all these quantities and other tensor quantities have usually already been incrementally rotated to account for the incremental rigid body rotation. For now. together with the stress at the beginning of the time increment. however. however. In large deformation analyses. the undeformed conﬁguration at that time. in which deformations and rigid body rotations can become large. we shall look at an incrementally objective stress update.Incrementally objective integration for large deformations 167 5. we have addressed the update of stress over a given time increment on the basis that there is no incremental rotation. is Deformed configuration at t Undeformed (original) configuration y Ft y x Y x Deformed configuration at t + ∆t y x X F t + ∆t Fig. ˙ The stress rate. referring to Fig. at time t. the user is supplied with the strains at the beginning and end of the time increment. 5. with respect to the material reference frame. at time t with respect to the material reference frame. Figure 5. σ .2.2 A body in its original (undeformed) conﬁguration and in deformed conﬁgurations at times t and t + t. We need to ensure that the stress update is carried out with respect to the same reference frame. Also. the local material (or corotational) coordinate systems are shown indicating that an incremental rotation has occurred between the conﬁgurations at times t and t + t.6 Incrementally objective integration for large deformations So far. . we have determined the spin. given by the antisymmetric part of the velocity gradient. let us assume we know the stresses. we need to take account of the effect of incremental rotation on the determination of the updated stress. that is. taking full account of the incremental rigid body rotation. 5. σ t . For example. W . We shall return to this in Chapter 6. First. In the deformed conﬁgurations.1 shows the representation of a body in its original conﬁguration and then in deformed conﬁgurations at times t and t + t. or undeformed conﬁguration. in using a userdeﬁned subroutine for plasticity in ABAQUS.
J. K.R. stress. 149–188. New York.J.J. John Wiley & Sons. In Nemat Nasser S. Belytschko. New Jersey. Further reading Bathe. International Journal of Plasticity. ‘Timeindependent constitutive theories for cyclic plasticity’. Prentice Hall. Finite Element Procedures.A.R. (1976). and R is a rotation to be determined. Chaboche. T. R.J. Krieg. and Moran. (2000).). 29–57. and Hughes. or corotational. S. Theoretical Foundation for LargeScale Computations of Nonlinear Material Behaviour. . The Netherlands. Liu..83) in which σ is the increment in objective. and Hegemier G.C. concerned with the implementation of plasticity models into ﬁnite element code. ‘Numerical implementation of constitutive models: rateindependent deviatoric plasticity’.A. SpringerVerlag..D. revised edition. 2(2). K. (eds). We can write Equation (5.. T. Asaro R. T. (1998).W.L. Martinus Nijhoff Publication. Computational Inelasticity. B.81) in incremental form. Nonlinear Finite Element Analysis for Continua and Structures. In Constitutive Equations in Viscoplasticity: Computational and Engineering Aspects (eds) Stricklin J.82) σ + Rσ t R T (5.J. if we wish. and Saczalski K. Hughes.J.. New York. ‘Implementation of a time dependent plasticity theory into structural computer programs’. as σ = which we may also write as σ = ∇ σ + (W t σ t − σ t W t ) t ∇ (5. Simo. J. (1984). We shall further consider these approaches in Chapter 6.W. (1986). (1996).168 Implicit and explicit integration given by ˙ σ = σ +W t σ t − σ t W t ∇ (5.81) so that the updated stress (written explicitly) with respect to the undeformed conﬁguration at time t is simply σ t+ ∇ t ˙ = σ + σ t. and Key. ASME.
the user is required to provide a Fortran subroutine called a ‘UMAT’. LSDyna. We use the problem of simple shear with elasticity to verify the large deformation implementations. a range of information is passed into the material module relating to both the beginning and end of a time increment. must also be provided. We go on to consider isotropic hardening plasticity with explicit and implicit integration. with continuum and consistent tangent stiffnesses. for example. This chapter is concerned with the implementation of plasticity models into ﬁnite element code. we start by developing an ABAQUS UMAT for elasticity.6. In ABAQUS. An important module is that relating to material behaviour. and from ﬁrst principles using the deformation gradient. Strain and the deformation gradient are also provided at the end of the increment. ABAQUS. Finite element code is often modular in structure. the stresses at the end of the time increment must be determined and. but in a similar way for all codes. for the case of an implicit analysis (i. and discuss the tests necessary to verify the model implementation into ABAQUS. in using ABAQUS. the material Jacobian.e. second. any state variables (such as the isotropic hardening variable or effective plastic strain) must . Third. and deformation gradient are provided at the beginning of the time increment. whether it be commercial code or written inhouse. it is then necessary to execute three tasks. strain. are available through the OUP website. together with the necessary ABAQUS input ﬁles. in other words. by example. In particular. the ﬁnite element momentum balance or equilibrium equations are solved implicitly) using ABAQUS standard. In particular. Implementation of plasticity models into ﬁnite element code 6. the constitutive stress response of the material given prescribed conditions of deformation. First. All the Fortran coding. large deformation formulations using rotated variables provided by ABAQUS. or tangent stiffness. stress.g. Within the module.1 Introduction Several commercial ﬁnite element software packages (e. In ABAQUS. which is carried out. We then present an implicit implementation for elastoviscoplasticity (and creep). and MARC) provide the facility for users to specify their own material models.
In fact. 6.3 throughout). It is also important to check the ordering of shear quantities. For linear elasticity. which can vary depending on element type used.g. However. plane strain and axisymmetric problems). is given in Appendix B. σ = C εe = ⎜ (6. the stress increment is obtained either from Equation (5. particularly where it is necessary to work from the deformation gradient. and threedimensional problems. In Chapter 5.170 Implementation of plasticity models be updated to the end of the time increment. We start by addressing elasticity. it provides a basis from which elastoplasticity models may be developed. we use an incremental approach in implementing the linear elastic equations. it may be more convenient to use Equation (6. is available through the OUP website. in coding an ABAQUS UMAT.2 Elasticity implementation We discuss elasticity and its implementation into a UMAT for two main reasons. or its equivalent written in Voigt notation. this is not always the case. A complete list of all the UMAT coding. Note that the shear strain quantities provided by ABAQUS are always engineering shears.21). for the case in which there are no out of plane shears (e. Hooke’s law was given in an incremental form in Equation (5. Second. the material Jacobian is just the elastic stiffness matrix so that speciﬁcation of the Jacobian in the UMAT is easy. With knowledge of the increment in strains (provided to the UMAT).21). for which different stiffness matrices are required. it provides a good introduction to writing and testing a UMAT subroutine for those who are new to the process. and later we shall use the tensor form in preference. First. For the latter reason. axisymmetric. but not for problems of plane stress or three dimensions. some of which will be referred to later. the job of the UMAT is clear: to update the stresses and state variables to the end of the time increment and to provide the Jacobian. together with the speciﬁcation of elastic constants (we shall take E = 210 GPa and ν = 0. a large range of information is provided. ⎛ ⎞⎛ ⎞ 2G + λ λ λ 0 ε11 ⎜ λ 2G + λ λ 0 ⎟ ⎜ ε22 ⎟ ⎟⎜ ⎟. together with ABAQUS input ﬁles. Equation (6. that is. twice the tensorial shear strains. However. while is no way essential for elasticity.1) ⎝ λ λ 2G + λ 0 ⎠ ⎝ ε33 ⎠ 0 0 0 G γ12 Because ABAQUS provides most tensor quantities in vector (Voigt) form. .1) is suitable for plane strain and axisymmetric problems.21) together with the material Jacobian.1) than the tensor form given in (5. The coding required for the UMAT for linear elasticity for plane strain.
Because of their importance. but it is certainly advisable to do so for new model implementations into UMATs.Veriﬁcation of implementations 171 6.3 Veriﬁcation of implementations The veriﬁcation of the model implementation is vital. 2.0 0. Figure 6.0 1. and strain.0 0.0 ur = 0.0 0.and fourelement unit square under uniaxial displacement or force controlled loading. perhaps by ‘switching off’ parts of the model implemented into the UMAT. Figure 6.0 uz = 0. which can be compared with independent closedform solutions. and where appropriate. Single element simple shear test. particularly because of the potential for errors with the use of engineering shear strains rather than their tensorial counterparts. In this chapter. it is necessary to test the UMAT using a single. For small deformations. σzz . uniaxial stress.and fourelement unit square which is subjected to uniaxial displacement or force control in the zdirection producing uniform.0 Displacement or force z 1. with εrz = σrr = σθ θ = σrz = 0. The fourelement problem is important since it introduces a ‘free’ node.0 Fig. which does not exist for the single fournoded axisymmetric element. εzz . The uniaxial tests in (1) do not involve the shear terms at all. so it is important to include a problem which tests these terms. εxx = εyy = σxx = σyy = 0. this requires the development of an independent solver (which will often be numerical) for uniaxial and pure shear problems so that direct comparison with the results obtained using the UMAT can be made.0 r 0.0 ur = 0.0 1. . we shall not carry out all of the veriﬁcation tests described for all the model implementations. multiple elements.2 shows a plane strain singleelement unit square under simple shear loading. Single and multiple element uniaxial tests. For complex material models. If possible. for conditions of strain control and load control under uniaxial and pure shear conditions. make comparisons with a multiaxial problem with nonuniform strain and stress distributions for which the solution is known (either an independent solution or one obtained using an internal ABAQUS model). Displacement or force z 1.0 r uz = 0. 6. 1.1 shows an axisymmetric single. In addition. we go through some of the possible veriﬁcation tests step by step.1 Schematic diagram showing an axisymmetric single. The force controlled test is important for checking errors in the Jacobian.
it nonetheless may test a good number of them. a comparison test which generates nonuniform strain and stress ﬁelds will not be possible because a comparison may well not exist. 6. σxy .4. 6. γxy . Later in the chapter. However.2. We start with the explicit scheme which was described at the beginning of Section 5.g. at the start of the time increment.1 Explicit integration for isotropic hardening plasticity with continuum Jacobian We may summarize the implementation as follows.2 Schematic diagram showing a plane strain singleelement unit square under simple shear loading.172 Implementation of plasticity models g y 1.0 0. we presented both explicit and implicit integration of the equations for linear strain hardening isotropic plasticity. both an implicit and explicit implementation of isotropic hardening plasticity are tested in this way by comparing the results obtained with those produced using the builtin ABAQUS model. that is. We will implement both integration schemes into ABAQUS UMATs and discuss the advantages and disadvantages of each. and uniform shear strain. it is sometimes possible to simplify the implemented plasticity model (e.0 Fig. and introduce the continuum Jacobian. are produced which can be compared with independent closedform solutions.0 0. unless indicated . by turning off the porosity in a Gursontype porous plasticity model) such that a comparison with another solution (e. and a closedform solution is now no longer possible. together with the consistent tangent stiffness for the implicit scheme. 6. and stress.0 1.0 x ux = uy = 0. All quantities are assumed to be given at time. t. 3.g.4 Isotropic hardening plasticity implementation In Chapter 5. and may therefore be worthwhile. Often. Nonuniform strain and stress ﬁeld. produced using one of the many builtin models contained within ABAQUS) is then possible. While this will not test all the features of the model implemented into the UMAT.
In Chapter 5. dr = h dp = h dλ. it is not derived explicitly on the basis of the integration scheme. = εp + dεp . = r + dr. σt+ εt+ rt+ (vi) Determine Jacobian. For the purposes of the explicit integration considered here.6) which we may write out fully for conditions of axial symmetry or plane strain as ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛⎛ ⎞ ⎛ ⎞⎞ dσ11 dσ1 C11 C12 C13 C14 dε1 n1 ⎜dσ22 ⎟ ⎜dσ2 ⎟ ⎜ ⎟ ⎜⎜dε2 ⎟ ⎜n2 ⎟⎟ C22 C23 C24 ⎟ ⎜⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ − dλ ⎜ ⎟⎟ .2) (iv) Determine stress and isotropic hardening increments dσ = C dεe = C(dε − dλn). We start from the stress–strain relationship written in Voigt notation dσ = C dεe = C(dε − dλn). n · Cn + h dλ = 0. That is. ⎝dσ33 ⎠ = ⎝dσ3 ⎠ = ⎝ ⎝n3 ⎠⎠ C33 C34 ⎠ ⎝⎝dε3 ⎠ dσ12 dσ4 sym C44 dε4 n4 (6.3) f > 0? 3 σ :σ 2 1/2 − r − σy .7) p t t t (6.4) (v) Update all quantities to the end of the time increment. we introduce what is sometimes referred to as the continuum Jacobian. but directly from the constitutive equations. (6. we derived the consistent tangent stiffness for the implicit integration scheme. (vii) End. (6. using explicit integration = σ + dσ.Isotropic hardening plasticity implementation 173 otherwise: (i) Determine the yield function f = σe − r − σy = (ii) Determine if actively yielding Is (iii) Determine the plastic multiplier f > 0. (6. We now address the determination of the Jacobian. dλ = n · C dε . f < 0. (6.5) .
In any case. the stress at ﬁrst yield is overestimated because the stress at the start of the increment was determined on the basis of elastic behaviour in the previous increment. An ABAQUS UMAT containing this formulation. together with various input ﬁles for uniaxial displacement and load control. the maximum time increment allowed in the analysis is carefully chosen to ensure stability and accuracy. n · Cn + h (6. With an explicit scheme. the ﬁrst term is J11 = J13 J23 J33 ⎞ J14 J24 ⎟ ⎟ J34 ⎠ J44 (6. we may show that J = C − Cn ⊗ Cn ∂ dλ = ∂ dε n · Cn + h so that the continuum Jacobian is J =C− Cn ⊗ Cn .9) ∂ dε where ⊗ is the dyadic product of two vectors. In all the analyses using the explicit UMAT.3. at the elastic–plastic transition. many more time increments are required using the explicit UMAT than are required using the builtin ABAQUS implicit plasticity integration.8) ∂ dσ1 ∂ dλ = C11 − (C11 n1 + C12 n2 + C13 n3 + C14 n4 ). details of which may be found in Appendix A.000 GPa. the very same problems are analysed using the builtin ABAQUS linear strain hardening plasticity model (chosen to represent a material with E = 210. by considering the numerator of the plastic multiplier given in Equation (6. In addition. σy = 240 MPa. therefore. This is a further serious shortcoming of the explicit integration method.3). Despite this. for example. ∂ dε1 ∂ dε1 With some algebra we may show. (6. ν = 0. that ∂ dλ . which needs to be weighed against the advantage . are available via the OUP website (full details are given in Appendix B). In a similar way. this is never corrected so that the stresses remain slightly overestimated throughout the analysis. but at the cost of the computer CPU time and error accumulation. The error in stress can be reduced by decreasing the time increment size.10) The explicit integration and provision of Jacobian is then complete.174 Implementation of plasticity models We may write the Jacobian (here symmetric) as ⎛ J11 J12 ⎜ ∂ dσ ⎜ J22 J = =⎝ ∂ dε sym so that. together with a fourpoint beam bending problem. and h = 1206 MPa).
However. 6. 2 σe ε− e εp .14) σ = 2G ε + λI ε : I . particularly for complex constitutive equations. d p= tr σe − 3G p (k) − r (k) − σy .13) p (k+1) = p(k) + d p. (6. at time t + t. (ii) Determine the trial yield function tr f = σe − r − σy = (6.11) 3 tr σ : σ tr 2 1/2 − r − σy . We consider an implicit implementation in Section 6. the results obtained for the fourpoint bend simulation from the explicit UMAT and the builtin ABAQUS plasticity model are found to be nearidentical.4.12) (iii) Determine if actively yielding Is f > 0? (iv) If yes.2. Otherwise. despite considerably longer computer CPU times. e (6. (v) Determine plastic and elastic strain and stress increments εp = εe = σ tr 3 p tr .Isotropic hardening plasticity implementation 175 of simplicity. that is. (i) Determine the elastic trial stress σ tr = σt + 2G ε + λI ε : I . (6. (vi) Update all quantities to the end of the time increment σ = σt + σ. 3G + h (6.15) .4. p = 0. all quantities are now assumed to be given at the end of the time increment. unless otherwise indicated. p = pt + p. use Newton iteration to determine the effective plastic strain increment r (k) = rt + h p. As opposed to the explicit case.2 Implicit integration for isotropic hardening plasticity with consistent Jacobian We may summarize the implementation as follows.
as before. the use of implicit integration enables much larger time increments to be used. in which case a numerical procedure may be possible.5 Large deformation implementations We have not yet differentiated between small and large deformation implementations in this chapter. In addition.176 Implementation of plasticity models (vii) Determine consistent Jacobian δσ = 2GQ (viii) End.6. Often. referring back to Section 5.17) in which σ is the corotational stress increment. An ABAQUS UMAT containing this formulation. The components of the back stress will need to be updated within the UMAT ( just like the scalar isotropic hardening variable in the previous sections) and stored in what are called state variable arrays in ABAQUS. are available through the OUP website (full details are given in Appendix B). An example is the back stress in a kinematic hardening model. analytical forms are simply not obtainable. The disadvantage of the implicit formulation is simply the difﬁculty in obtaining the consistent tangent stiffness for complex constitutive equations. the large deformation stress update necessary is σ = ∇ σ + (W σt − σt W ) t = ∇ σ + Rσt R T ∇ (6. Because state variables are not modiﬁed . σ tr σ tr 2 : δε + 2GRδε + K − GR II : δε.e. σt . That is. depending on the plasticity model employed. Using implicit integration with the consistent tangent stiffness eliminates the problem which occurred at the elastic–plastic transition when using explicit integration. This is because the necessary rigid body rotations for the strains and stresses have already been carried out by ABAQUS before they are provided to the UMAT routine. the stress provided at the start of the time increment is effectively Rσt R T ) so that all we need to do within the UMAT is to carry out the stress update. In addition. the UMATs discussed above for both explicit and implicit schemes are suitable for both small and large deformation problems using ABAQUS. together with various input ﬁles for uniaxial displacement and load control. tr tr σe σe 3 (6. at the start of the time increment (and the strain) has already been rotated by ABAQUS (i. it is necessary to use internal variables which are also tensor quantities.16) 6. The results obtained for the fourpoint bend simulation from the implicit UMAT and the builtin ABAQUS plasticity model are found to be identical. the very same problems are analysed using the builtin ABAQUS linear strain hardening plasticity model. together with a fourpoint beam bending problem. Sometimes. The stress. therefore signiﬁcantly reducing CPU times. In fact.
say. constitutive equations for elasticity and plasticity are formulated in terms of the deformation gradient. ABAQUS provides a utility subroutine to simplify this process.18) (iii) Determine the yield function f = σe − r − σy = (iv) Determine if actively yielding Is f > 0? (6. x t .22) . In Section 6.1.5. 2 W = 1 (L − LT ). (6.5.21) 3 σ :σ 2 1/2 − r − σy . we present a large deformation implementation based on an explicit scheme. however. at the start of the time increment. f < 0. Sometimes. 6. If. 2 (6. D p = 0. (ii) Determine the rate of deformation and spin D= 1 (L + LT ). for example. (6. all quantities are assumed to be given at time. D p = as speciﬁed by constitutive equation.20) (v) Determine rate of plastic deformation f > 0. In this explicit approach. (i) Determine the velocity gradient ˙ L = F F −1 . the tensorial variable recovered from the state variable array at the start of the increment is. The rotation is carried out by a single call of the utility subroutine rotsig detailed in the ABAQUS manuals. t.19) (6. that is. it is necessary for the user to carry out the rigid body rotations on tensorial state variables. Examples include hyperelasticity—large strain nonlinear elasticity and crystal plasticity. that the user simply wishes to work from the deformation gradient rather than use the strain and stress quantities provided by ABAQUS to the UMAT subroutine. It may be. unless indicated otherwise. then the user must carry out the rigid body rotation Rx t R T before updating x to the end of the time increment within the UMAT.Large deformation implementations 177 by ABAQUS. F.1 Implementation using the deformation gradient We may summarize the implementation as follows. In fact.
r = h˙ ˙ p. We will next address the veriﬁcation of a large deformation implementation. 1 t t ˙ = σ + σ t.23) (6.5. we will ‘switch off’ the plasticity and allow elasticity only.1 as F = 1 0 δ .4. = r + r t. ˙ (6. which is valid for both the implicit and explicit implementations in Sections 6. (x) End. 0 . We obtained the deformation gradient for simple shear in Section 3. since for these cases.178 Implementation of plasticity models (vi) Determine rate of elastic deformation and Jaumann stress rate and isotropic hardening rate De = D − Dp. σ = 2GD e + λID e : I . A good test.2. δ. To simplify matters.25) ˙ Considering a constant rate of shearing. however.4. ∇ ∇ (6. W . 0 W = 1 0 ˙ 2 −δ ˙ δ . since the test is being carried out to check the rigid body rotation calculation rather than the constitutive response. using explicit integration σt+ rt+ (ix) Determine Jacobian.24) (viii) Update all quantities to the end of the time increment. the velocity gradient is L= 0 0 ˙ δ 0 so that the rate of deformation and spin are D= 1 0 ˙ 2 δ ˙ δ . is provided in the form of simple shear if we allow the strains to become quite large. is zero. (vii) Determine stresses with respect to material reference ˙ σ = σ +W σ − σ W .1 and 6. Note that uniaxial displacement or force controlled loading will not test whether the rigid body rotations are being calculated correctly. the continuum spin.
Large deformation implementations 179 Note that the spin is nonzero so that rigid body rotation.28) and substituting for (6. Solving for σxx and σyy using Equations (6. ˙ 2 ˙ σxx = δσxy . ˙ σxx = −σyy = G(1 − cos δt).26) and (6. (6.27) gives σxx ˙ σxy ˙ so that 1 ˙ ˙ σxy = Gδ + δ(σyy − σxx ).31) .29) and (6. The Jaumann stress rate is given by ∇ 0 σ = 2GD + λID : I = G ˙ δ ˙ δ 0 (6. with the initial condition that all stresses are zero. ˙ (6.29) (6. Differentiating (6. σyy ˙ (6.29) and (6. The initial conditions require that B = 0. ¨ which has solution ˙ ˙ σxy = A sin δt + B cos δt.26) and the stress rate with respect to the material (undeformed) reference is ˙ σ = σxx ˙ σxy ˙ σxy ˙ .27) which is given in terms of the Jaumann stress rate by ˙ σ = σ +W σ − σ W so that substituting for the spin and (6. together with stretch.30) gives ˙ σxy + δ 2 σxy = 0. ˙ ˙ σyy = −δσxy .30) and imposing the initial conditions gives the solution ˙ σxy = G sin δt.29) and (6. σyy = −σxx .30) give.30) σxy ˙ ˙ 0 = Gδ 1 σyy ˙ 1 1 ˙ σxy + δ 0 2 −σxx 1 −σxy σyy − δ −σxy 2 −σyy σxx σxy ∇ Equations (6.28) (6. is occurring.
00 Time (s) Stress (GPa) syy Fig. The results obtained are identical and are shown in Fig. the rigid body rotation taking place. We employ a sinhtype viscoplastic constitutive equation and for simplicity.32) The harmonic variation in (6. Here. 6.3. The radial return.40 0.60 0. (6. Despite its nonphysicality. Such an implementation can readily be simpliﬁed for the implicit analysis of creep. together with the ABAQUS input ﬁles. 6. This has been carried out using the UMAT with elasticity described in Section 6. are detailed in Appendix B and are available via the OUP website. as shown in Fig. 6. a further UMAT based on the deformation gradient. described in this section. 6.e. the elastic stiffness) for the material Jacobian. implicit backward Euler integration for viscoplasticity was discussed in Chapter 5. has been subjected to simple shear to large strain.180 y Implementation of plasticity models x 150 100 50 0 –50 0. meaning ratedependent plasticity in which the plastic multiplier is determined through the use of a viscoplastic constitutive equation as opposed to the use of the consistency condition.0 s−1 and the corresponding stresses. In order to do this.20 0.2. these reduce to σxy = Gδ and σxx = −σyy = 0. a single plane strain element. The various UMATs. use the initial tangent stiffness (i. The viscoplastic constitutive equation is taken to be p = φ(σe . we present an implicit implementation for linear isotropic strain hardening elastoviscoplasticity. and using the builtin elasticity model in ABAQUS.80 1. δ.31) results from the large deformation. This nonphysical result arises because we are using a small strain. linear elasticity model under conditions of large deformation. at constant strain rate.2. which uses ABAQUSprovided stresses and strains. was introduced in Chapter 4. For small strain. and in particular. r) = α sinh β(σe − r − σy ) ˙ .00 –100 –150 –200 sxx sxy 0.6 Elastoviscoplasticity implementation Viscoplasticity. it provides a good test of the calculation of the rigid body rotations in the large deformation UMAT.3 Unit square under simple shear at a rate of 5.
(6.Elastoviscoplasticity implementation 181 and the multiaxial plastic strain increments are given by εp = pn = σ 3 p . (v) Determine plastic and elastic strain and stress increments εp = εe = σ tr 3 p tr .34) (iii) Determine if actively yielding Is f > 0? (iv) If yes. that is. at time t + t. tr φr = −αβ cosh β(σe − 3G p − r − σy ). r) − ( p/ t) . φ p tr = −3Gαβ cosh β(σe − 3G p − r − σy ). .36) σ = 2G εe + λI ε e : I .33) − r − σy = 3 tr σ : σ tr 2 1/2 − r − σy . use Newton iteration to determine the effective plastic strain increment tr φ(σe . 2 σe We determine the increment in effective plastic strain as described in Chapter 5 as follows. (ii) Determine the trial yield function f = tr σe (6.35) p(k+1) = p(k) + d p. 2 σe ε− εp . r = rt + h p. (i) Determine the elastic trial stress σ tr = σ t + 2G ε + λI ε : I . d p= φ( p. Note that all quantities are now assumed to be given at the end of the time increment. (6. r) = α sinh β(σe − 3G p − r − σy ). (1/ t) − φ p − hφr (6. unless otherwise indicated.
37) . (viii) End.182 Implementation of plasticity models (vi) Update all quantities to the end of the time increment σ = σt + σ. p = pt + p. In addition. An ABAQUS UMAT containing this formulation. together with various input ﬁles for uniaxial displacement and load control. (vii) Determine Jacobian. are available via the OUP website (full details are given in Appendix B). (6. uniaxial. closed form implicit and explicit solutions are provided in Fortran programs for veriﬁcation of the implementation. together with a fourpoint beam bending problem.
Part II. Plasticity models .
This page intentionally left blank .
Region (iii) corresponds to very high . and an industrial application. we shall introduce superplasticity and its characteristics.1 Introduction Superplasticity is the ability of some materials to undergo very large.5 Tm . Region (i) corresponds to very low strain rates in which diffusion processes dominate when the temperature is higher than ∼0.1(b) and pick off the corresponding stress for each of the stress–strain curves at all the strain rates. If we choose a particular strain in Fig. and has many advantages for the manufacture of complex shapes in sheet metal using simple low pressure pneumatic forming equipment. This results from a number of possible causes including the effect of grain growth which occurs in superplasticity. 7. Superplasticity 7. which are used extensively in the aerospace and aeroengine industry. the multiaxial form of the equations.5 Tm is necessary to enable the appropriate superplastic deformation mechanisms to operate. strain hardening is seen to occur. which often produces a curve of the form shown in Fig. we may then plot log(stress) versus log(strain rate). tensile elongations without necking and failing. 7. In this chapter.1. Two common superplastically formed classes of material are aluminium and titanium alloys. a very ﬁne grain structure is required (a typical grain size will be of the order of 1 µm) and a deformation temperature of about 0.7. Superplastic forming exploits the ability of the material to undergo very large tensile elongations in metal sheet stretch forming and blow moulding processes. 7. There is often a very strong strainrate sensitivity. irreversible. and in addition.2 Some properties of superplastic alloys Superplasticity is very much a viscoplastic process in the sense that the stress response is highly strainrate dependent. Generally.2. 7. and from dislocation hardening processes. constitutive equations for superplastic deformation which are coupled with the accompanying microstructural evolution. The uniaxial response to variable constant true straincontrolled loading is simpliﬁed and schematically shown in Fig.
the ln(stress)–ln(strain rate) relationship can be written as ln(σ ) = m ln(˙ ) + k ε in which m and k are constants. In Equation (7. large tensile strains (∼1–2) are achievable which are many times larger than those obtainable in regions (i) and (iii).3. Generally.186 Superplasticity «3 • (a) (b) s3 «2 • «3 • s2 Stress s1 sy «2 • Strain «1 • «1 • Time Strain Fig. that is. the better the superplastic deformation. strain rate at which diffusion is largely inhibited so that deformation occurs through dislocation motion. and the larger the tensile elongations achievable in the absence of necking and failure.2 log(stress) versus log(strain) showing the region usually considered to be superplastic. the higher the value of m. 7. of course. m is called the strainrate sensitivity and is.2.35. Rearranging (7. 7.1) . but diminishing with increasing strain rate. 7.1 Schematic showing simpliﬁed superplastic stress response which is dependent on strain rate. In fact. we will obtain the graph shown schematically in Fig. 7.2) in which K is a further constant. Region (ii) is that in which superplasticity takes place.2). Superplastic region ln(stress) (i) (ii) (iii) ln(strain rate) Fig. Because of the approximate linearity in region (ii). m > 0. the gradient of the ln(stress) versus ln(strain rate) curve. If. for what would be described as superplastic deformation. some of which is perhaps thermally activated. (7. from Fig. we now plot the strainrate sensitivity versus ln(strain rate).1) gives σ = K εm ˙ (7.
idealized neck shown schematically in Fig. L L + dL P P Uniform region with Necked region with Area Stress Strain rate A s • « Area Stress Strain rate A + dA s + ds • « + ds Fig.Some properties of superplastic alloys 187 Superplastic region m 0. 7. 7.4 Schematic diagram showing a uniform. The incompressibility condition gives AL = (A + dA)(L + dL) ≈ AL + L dA + A dL .4. We shall now examine further the signiﬁcance of the strainrate sensitivity. circular test piece with crosssectional area A under load P before and after the introduction of a neck. m.35 ln(strain rate) Fig. uniaxial circular test piece containing a single. by considering a uniform. 7. We employ the incompressibility condition.2). to the test piece during necking and assume the stress–strain rate relationship given in (7. in superplasticity.3 Schematic representation of strainrate sensitivity versus log(strain rate). P . apply constant load.
5) = K εm 1 + m ˙ d˙ ε + ··· ε ˙ (7. A much more complete introduction to superplasticity can be found in Pilling and Ridley (1989). the rate of necking decreases to zero. P .4) gives d 1 (dA) ≈ −dA˙ 1 − ε . ε ε dt which. We therefore see the signiﬁcance of the strainrate sensitivity in superplasticity.4) so that (7.6) d˙ ε dσ ≈m .188 Superplasticity so that dL dA =− = −dε = −˙ dt ε A L dA = −˙ A ε dt (7. (7.3) and d(A + dA) = −(˙ + d˙ )(A + dA). dt m (7.6) and (7. in the neck. σ A The constitutive equation σ = K εm gives ˙ ˙ σ + dσ = K(˙ + d˙ )m = K εm 1 + ε ε so that d˙ ε ε ˙ m or. As the strainrate sensitivity increases. when combined with (7. the more necking can be inhibited therefore allowing greater elongations prior to the onset of necking and failure. . dt dA ε ˙ The constancy of load.5) gives and substituting into (7. σ ε ˙ A d˙ ε 1 =− dA ε ˙ m Combining (7.3) gives d A d˙ ε (dA) ≈ −(A d˙ + ε dA) = −dA˙ ε ˙ ε +1 . gives σA = (σ + dσ )(A + dA) dA dσ =− .7) This equation tells us that the rate of development of the neck depends upon the quantity 1 − (1/m). and approaches unity. the higher the value of m.
0 × 10–3 s–1 20 Stress (MPa) 15 2. it says nothing about the inﬂuence of changing microstructure during superplastic deformation.4 µm. We present here constitutive equations for superplasticity for the particular commercially important titanium alloy. 7.5 Superplastic stress–strain curves for Ti–6Al–4V at 927◦ C at the strain rates shown. which is known to be important (Ghosh and Hamilton.0 × 10–4 s–1 10 5 5. the engine fan blades are perhaps the most important. Ghosh and Raj. Zhou and Dunne. The hardening is due to several processes.5 is shown in Fig.5 is 6. perhaps the most important is the increasing average grain size.5. The evolution of the average grain size for each of the stress–strain curves shown in Fig. At the end of the deformation. The average grain size at the start of all the tests shown in Fig.4 µm. In addition. The initial grain size is 6. 1979.0 × 10–5 s–1 0 0.3 Constitutive equations for superplasticity The constitutive equation in (7. 7. Uniaxial stress–strain curves for the Ti–6Al–4V alloy undergoing superplastic deformation at 927◦ C are shown in Fig. Hamilton.Constitutive equations for superplasticity 189 7.6 Strain 0. 7. A strong strainrate effect is seen together with signiﬁcant strain hardening. but is often inadequate for simulating superplasticity processes because the ln(stress)–ln(strain rate) relationship is not often linear.0 0.4 0. This alloy ﬁnds application in a range of aeroengine components. . 7. 25 1. 1984. 7.2 0. however. but for component manufacture using superplastic forming.0 Fig. 1981.6. it has increased to about 10 µm. Ti–6Al–4V. really quite small but the tensile strains achieved during the superplastic deformation are large. and depends on the rate of deformation. particularly over strainrate regimes which occur in practical processing.8 1.2) is simple. 1996). The stresses required to cause the deformation are.
0 × 10–3 s–1 Grain size (mm) 0.9) to the data.11) r = (c1 − γ1 r)p. experimental data and the lines result from ﬁtting Equation (7.8) l in which l is the average grain size.0 s–1 0. grain growth still takes place. and α1 and γ are material constants. a static test in which no deformation takes place. Normal grain growth (Shewmon. The symbols in Fig.008 0. The initial grain size is 6.5 are. 7. ˙ σ = E(˙ − ε p ). 1999) and employ the following elasticviscoplastic constitutive equations α ε p = µ sinh β(σ − r − σy ). the effective plastic strain ˙ rate. The material constants for this particular temperature of 927◦ C are given in Table 7.9) l in which β1 is a further material constant and p is. As the strain rate is increased.190 Superplasticity 0.10) l ˙ (7. That part of the grain growth resulting from the straining is often referred to as deformationenhanced grain growth. We now return to the microstructure–deformation coupling (Zhou and Dunne. ˙ ε ˙ (7. ˙ (7.012 1.12) . in fact. that is. so the rate of growth of grain size increases. An important feature is that for a strain rate of 0. 1969) gives rise to a kinetic equation for static grain growth of the form ˙ α1 l= γ (7. 7. this is called static grain growth. The deformationenhanced grain growth is accounted for with an additional term so that the complete kinetic equation can be written as ˙ α1 l = γ + β1 p ˙ (7.4 µm.0 s−1 . Kim and Dunne.0 × 10–4 s–1 5.1 and assume grain size to be speciﬁed in millimetres.6 Grain size versus time curves for Ti–6Al–4V at 927◦ C at the strain rates shown. as before.006 0 100 200 Time (min) 300 400 Fig.01 2.0 × 10–5 s–1 0. 1996.
The resulting computed stress–strain curves are given by the solid lines in the ﬁgure. as before.10) and (7. 7. 7.9). A further isotropic hardening term.7. and the corresponding predicted variation of strainrate sensitivity with strain rate (at a strain of 1. .11) are obtained by ﬁtting the equations (Zhou and Dunne. and µ the deformation–microstructure coupling constant.8. l is the current grain size. 7. 7. which are coupled with the grain growth equation in (7.12) constitute the uniaxial material model for superplasticity. α 0.9625 × 10−13 E (MPa) 1000 γ 5.2 1 10–6 10–5 10–4 10–3 Strain rate (s–1) 10–2 10–1 Fig. Comparisons of predicted and experimental log(stress) versus log(strain rate) behaviour for the Tialloy with an initial grain size of 6. Equation (7.0) is shown in Fig. 7. 1996) to the experimental data in Fig.128 × 10−16 σy (MPa) 0.0 0.666 µ 1.06 α1 0.Constitutive equations for superplasticity Table 7.12) is Hooke’s law.6 0.10) is the viscoplastic constitutive equation in which.5 β1 0.4 does not all result from the grain growth.4 µm for the strain levels shown.0 191 100 Stress (MPa) 10 1. although a substantial part of it does.0919 γ1 0. Equation (7. With the material constants for the grain growth kinetic equation having been already determined. α and β are material constants.4 µm obtained at the strain levels shown are given in Fig.397 β 0. r. the unknown constants in Equations (7.1 Material constants for Ti–6Al–4V at 927◦ C.11) in which c1 and γ1 are material constants.5.9)–(7. The hardening seen in Fig. Equations (7.7 Experimental (symbols) and predicted (lines) stress versus strain rate curves for Ti–6Al–4V at 927◦ C with an initial grain size of 6.437 × 10−5 c1 8. has therefore been introduced with evolution equation given in (7.
but a UMAT implementation would be carried out as described for viscoplasticity in Chapters 5 and 6. 7. This facilitates an easier implicit implementation. the Jaumann stress rate is.2 Initial grain size 6.4 Multiaxial constitutive equations and applications We may write the multiaxial equations within the framework of viscoplasticity by assuming the normality hypothesis and von Mises material behaviour. Lin).4 mm 9 mm 11. The application we consider is the superplastic blowforming of a rectangularsection box made from 1. together with (7.0 1E–06 1E–05 1E–04 1E–03 1E–02 1E–01 Strain rate (s–1) Fig. If considering large deformations (which is normally the case for superplasticity).4 0. 7. ∇ (7. ˙ 2 σe (7. 2001.13) α sinh β(σe − r − σy ) (7.8 Predicted variation of strainrate sensitivity with strain rate for the initial grain sizes shown for Ti–6Al–4V at 927◦ C.9) and (7.15) These equations.11).25 mm thick Ti–6Al–4V sheet and processed at 900◦ C (Lin and Dunne.9 in which just one quarter . additional problems would need to be addressed. The model is shown in Fig. 7. with thanks to Dr.9) and (7. ˙ σ = σ + Wσ − σW.8 Strainrate sensitivity 0. which is described in Chapter 8.14) lµ and Equations (7.11) remain unchanged.0 0. The multiaxial viscoplastic strain rate. given these assumptions. is Dp = where p= ˙ 3 σ p .6 0. have been implemented into ABAQUS by means of the CREEP routine. as before.192 Superplasticity 1.5 mm 0. although if a plane stress implementation is required.
and both the initially ﬂat sheet and the die surface are included. the strain rate required to give the highest strainrate sensitivity obtained from the equivalent of Fig. however. 7. The process is considered completed at a forming time. but for 900◦ C. the maximum strain rate over the deforming sheet is controlled to be close to the optimum deforming rate of the material. In these analyses. t/tf = 0. This is typical of many practical superplastic forming processes.0. that is.6.Multiaxial constitutive equations and applications Material mesh 193 Rigid die surface Die surface normal Fig. tf .0 × 10−5 s−1 . while ideally it is necessary to deform superplastically all points in the material at the same optimum strain rate (to maximize strainrate sensitivity and hence elongation).10 shows the deformation of the superplastic metal sheet at three stages during the forming process at times t/tf = 0. The maximum target strain rate speciﬁed for the analysis is 1. and in this analysis.9 Finite element model for the rectangularsection box die surface and the material blank. It is possible. is included. The variation of the maximum strain rate during the . the surface of which forms a cavity in the shape required. forcing it to acquire the die shape.10 shows that the distribution of effective plastic strain rate is highly nonuniform. Gas pressure is applied to the top face of the sheet. this is rarely achievable in practice. The contours show the magnitude of the effective plastic strain rate. 7. when all nodes on the deforming sheet are in contact with the die. Figure 7.1. This is achieved by varying the applied gas pressure. for reasons of symmetry. The superplastic forming process consists of clamping the ﬂat Tialloy sheet (modelled using shell elements) against the die. the gas pressure is varied to ensure a nominal maximum strain rate of 1.7.0 × 10−5 s−1 which is achieved by application of uniformly distributed gas pressure. to ensure that a target maximum strain rate is not exceeded. Figure 7. and t/tf = 1.
00E–00 + 2.00E–06 + 8.00E–06 + 4.00E–00 + 2.11 for target strain rates of both 1. 7.194 Superplasticity (a) Strain rate + 0.1.00E–06 + 5.00E–06 + 5. 7.0 × 10−4 s−1 .00E–06 (c) Strain rate + 0.00E–06 + 5. 7.00E–06 + 3.0.00E–06 + 4.00E–06 + 7. and the corresponding variations in gas pressure to achieve these are given in Fig. Higher gas pressure is needed for the higher target strain rate because of the higher ﬂow stresses required.00E–06 + 6. (See also Plate 1.) forming process is shown in Fig.00E–06 + 7.00E–06 + 8.00E–06 + 3.00E–06 Fig.00E–06 + 6.00E–00 + 2.6. t/tf = 0.00E–06 (b) Strain rate + 0.00E–06 + 8.10 The simulated superplastically deforming sheet showing the effective plastic strain rate at fractional processing times of t/tf = 0.12.0 × 10−5 s−1 and 1.00E–06 + 4. and t/tf = 1. The gas pressure increases during the .00E–06 + 7.00E–06 + 6.00E–06 + 3.
it is apparent that the last part of the sheet to be formed is the corner.5 t /t f 0.2 0.0 × 10−5 and 1.Multiaxial constitutive equations and applications 195 1E–03 Maximum strain rate (s–1) 1 × 10– 4 s–1 1E–04 1 × 10– 5 s–1 1E–05 1E–06 0.0 × 10−5 and 1. High gas pressure is required to ﬁll the corner part of the die. 7. which is also the area of maximum thinning and where tearing is most likely to occur. 7.10. . which is the last stage of the forming process.12 Variation of gas pressure to ensure a maximum strain rate in the deforming sheet of 1.0 t /t f Fig. respectively. ˙ ˙ 400 Gas pressure (KPa) 300 200 1 × 10– 4 s–1 100 1 × 10– 5 s–1 0 0.8 1.6 0. This is made clear by looking at the throughthickness strain ﬁelds which are shown for the two target strain rates of 1.8 1.13(a) and (b). In Fig. 7. in addition.2 0.4 0.0 0. 7. because the material hardens during deformation which is due to the grain growth taking place.0 0.11 Variation of the maximum strain rate over the deforming material blank during the superplastic forming process for the two target strain rates ε = 1 × 10−5 and ε = 1 × 10−4 s−1 .0 Fig.0 × 10−4 s−1 .0 × 10−4 s−1 in Fig. superplastic forming because of increasing geometrical and frictional constraint and.
84.20E+00 – 1. In fact. at the lower strain rate. 7. 7.50E–01 – 2. In addition.) The lower strain rate is seen to lead to a higher spatial variation in throughthickness strain.0 × 10−5 and (b) 1.95E–01 – 1.196 Superplasticity (a) Thickness strain – 1. the larger grain size in the deforming material in the corner (relative to that at the boundaries) leads to a higher stress in the corner region to generate the same strains seen for the lower strain rate. shown in Fig. thinning. At the lower target strain rate. (See also Plate 2.14.60E–01 – 6.40E–01 (b) Thickness strain – 1. The reason for this becomes clear on looking at the grain size distributions. respectively. For the case of the higher strain rate. which is often not desirable in practical processing.40E–01 Fig.05E–01 – 4. The net result is a more uniform strain distribution for the high strain rate.95E–01 – 1. that is. This inhibits straining in the corner region producing larger strains in the boundary regions.15E–01 – 7.13 Throughthickness strain ﬁelds at the end of the superplastic forming carried out with target maximum strain rates of (a) 1.60E–01 – 6.05E–01 – 4.0 × 10−4 s−1 .20E+00 – 1.15E–01 – 7. the average grain size increase is larger than that at the higher strain rate because of static grain growth. the ranges of outofplane strain for the low and high strain rates are 1. This is generally preferable . the range of ﬁnal grain size is less than that at the higher rate.07E+00 – 9.50E–01 – 2. therefore.07E+00 – 9. which is inhibited in the latter. and a correspondingly more uniform thinning.06 and 0.
and results directly from the interactions of the superplastic deformation and microstructural evolution. Acta Metallurgica. However. A.93E – 03 +8.88E – 03 +9. Metallurgical Transactions.17E–02 + 1. An additional important effect is the distribution of grain size in commercial Ti–6Al–4V.0 × 10−5 and (b) 1. its effects on strainrate sensitivity and hence on necking and failure have been addressed by Kim and Dunne (1999). 29.References (a) Grain sizes + 1. 10. R.33E – 03 +8. Ghosh.) in practice.18E–02 + 1. (See also Plate 3.60E – 03 +8.20E – 03 +8. and Raj.06E – 03 +8. 607–616.14E–02 + 1.16E–02 + 1.K. 7. References Ghosh. where further information may be found. . (1981).74E – 03 +8. (1979). which has not been addressed here.0 × 10−4 s−1 .15E–02 + 1.18E–02 + 1.H.19E–02 197 (b) Grain sizes +7. and Hamilton.14 Average grain size ﬁelds at the end of the superplastic forming carried out with target maximum strain rates of (a) 1. 699–706. modelling techniques to incorporate it.01E – 03 Fig. ‘Mechanical behaviour and hardening characteristics of a superplastic Ti–6Al–4V alloy’. These industrially important effects could not have been obtained without microstructurally based constitutive equations. ‘Grain size distribution effect in superplasticity’. A.K. C.16E–02 + 1.
International Journal of Mechanical Science. The Institute of Metals. T. Shewmon. P.H. ‘Superplasticity in titanium alloys’. American Society for Metals. and Dunne. (1999). Kim. Proceedings of the Symposium. Lin. 701–718. London. ‘Modelling heterogeneous microstructures in superplasticity’. ‘Modelling grain growth evolution and necking in superplastic blowforming’. California. (ed.). and Dunne.198 Superplasticity Hamilton. and Dunne. (1984). In Agrawal. N. McGrawHill. Journal of Strain Analysis.E. M. 455.P. and Ridley. Superplastic Forming. J.P. Zhou.E. Superplasticity in Crystalline Solids. 595–609.G.P. March 13–22. F. J. Transformations in Metals. (2001). (1989). (1969). 31. (1996). F. Pilling. New York.P. Proceedings of the Royal Society. S.E. C. . 43. 65–73.W. F. ‘Mechanismsbased constitutive equations for the superplastic behaviour of a titanium alloy’.
A number of constitutive relations for the consolidation of metal powders have been developed. In this chapter. we introduce the porous plasticity model of Duva and Crow and outline its implementation into ABAQUS using both a UMAT material subroutine and the simpler ABAQUS CREEP routine. we use explicit integration of the constitutive equations and employ the initial stiffness as the material Jacobian.g.8. creep cavitation). generally at elevated temperature. They analysed the creep collapse of a thickwalled spherical shell subjected to externally applied hydrostatic loading. They have been used to predict the dependence of densiﬁcation rate on consolidation pressure and temperature as well as volume fraction of voids. volume changes occur during the deformation and the incompressibility condition no longer holds. Subsequent research has therefore broadened the Wilkinson and Ashby model to more general loading conditions. Again. The models have taken into account the effects of deviatoric and hydrostatic components of stress state by introducing potentials which make possible the . within the UMAT. for which the Wilkinson and Ashby model is therefore inappropriate. An important example is in the processing of metal powders in which. while possibly small. Porous plasticity 8. During the process. the powder is consolidated by plastic deformation and the elimination of the porosity. One of the ﬁrst constitutive models for consolidation of metal powder was developed by Wilkinson and Ashby (1975). signiﬁcant volume changes occur because of the removal of the porosity. however. A feature of the constitutive equations in porous plasticity is that there becomes a dependence on mean stress as we shall see shortly. consolidation occurs under the action of a range of stress states that are not purely hydrostatic. for which this is not the case. For simplicity. In general.1 Introduction We have so far considered plasticity and viscoplasticity processes in which it has been assumed that the deforming material has been incompressible. There are some materials. A further example of compressible or porous plasticity is that in which signiﬁcant voiding develops in a material undergoing plastic deformation resulting from a damage process (e.
S. σe . is closer to the experimental results than the predictions of Sofronis and McMeeking (1992). stresses. Based on the strain rate potentials of Cocks (1989) and Ponte Castaneda (1991). and hydrostatic. that is. ε0 σ0 ˙ n+1 S σ0 n+1 .3) (8. 1994). 1992). the coefﬁcient b becomes 0 and a becomes 1.200 Porous plasticity development of relationships between macroscopic strain rate and stress state (Cocks. Sofronis and McMeeking. 3 2 σe = S is an effective effective stress (Duva and Crow. The Duva and Crow potential also satisﬁes the lower bounds derived by both Cocks and Ponte Castaneda in the limit that the hydrostatic stress vanishes. in the presence of a large deviatoric stress component. The densiﬁcation rates predicted by Duva and Crow’s strain rate potential are consistent with their cell model calculations which were derived based on Hill’s minimum principle for velocity. The effective effective stress. a is associated with the deviatoric component and b with the hydrostatic component. Duva and Crow predict densiﬁcation rates which compare favourably with the prediction from the Wilkinson and Ashby model in the hydrostatic limit. Duva and Crow (1994) proposed a strain rate potential for computing densiﬁcation rates. at the fully dense stage. Duva and Crow.1) (8. σm . 2 1 σm = II : σ . . the coefﬁcients a and b were chosen to ensure that the potential gives densiﬁcation rates identical to those of Wilkinson and Ashby (1975) in the hydrostatic load limit. (8.2) (8. is then reduced to σe . 1994) given in terms of the effective. The densiﬁcation rate predicted from the Duva and Crow model.4) 3 σ :σ . 1991. 1989. D = 1. The strain rate potential (φ) for a porous material can be written as a function of both deviatoric and hydrostatic stresses φ= where 2 2 S 2 = aσe + bσm . which is equivalent to the solid volume fraction of the porous material. 1992. n is the creep exponent. The various densiﬁcation models for monolithic materials in the literature use different approaches to obtain the strain rate potential. and agree with both Cocks (1989) and Ponte Castaneda (1991) in the limit when the hydrostatic stress vanishes. Ponte Castaneda. D. The coefﬁcients a and b are chosen such that. This leads to different expressions for the coefﬁcients a and b (Duva and Crow. and the coefﬁcients a and b are functions of current relative density.
is L = FF −1 .2 Finite element implementation of the porous material constitutive equations Strain rates are determined by differentiating the strain rate potential.2.7) (8. which can be calculated. (8. explicit. 8.6) σ0 The dilatation rate can be obtained from ˙ ˙ ˙ εkk = εxx + εyy + εzz ˙ and hence. ˙ where D is the relative density.11) (8. A simple.8) (8. as 1 ˙ F = (F t+δt − F t ). L. to give ∂φ 3 1 ˙ ε= = AS n−1 aσ + bI σm (8. D.12) . The user is required to supply the Cauchy stress at the end of the time step. 2 (8. are given by 1 D = (L + LT ). F . for small time steps.5) ∂σ 2 3 in which 2/(n+1) 1 + (2/3)(1 − D) 3 2 n(1 − D) a= and b = 2n (1 − (1 − D)1/n )n D 2n/(n+1) and where A is a material parameter given by ε0 ˙ A = n. and the spin tensor. The total rate of deformation.9) δt Then. W . (8.10) (8. forward Euler integration is adopted.1 Implementation into ABAQUS UMAT The constitutive equations for porous metals developed by Duva and Crow (1992) are implemented into the ﬁnite element software ABAQUS within a large deformation formulation using a UMAT subroutine. ABAQUS supplies to the UMAT subroutine the deformation gradient at the beginning and the end of each time step. 2 1 W = (L − LT ).Implementation of porous material constitutive equations 201 8. the velocity gradient. F t and F t+δt . ﬁrst. need to ˙ deﬁne the rate of deformation gradient. φ. The algorithms. the densiﬁcation rate is given as ˙ D = −D εkk .
8.15) where E is Young’s modulus. . εkk .16) De = D − Dp. great care is necessary in choosing an appropriate time step.18) The updated stress is returned to ABAQUS through the UMAT subroutine. The relative density at the end of each time step is determined using the ﬁrstorder Euler integration scheme ˙ D t+δt = D t + Dδt. ν is Poisson’s ratio.202 Porous plasticity The strain components are calculated independently from 1 ε = − ln(FF T )−1 . provides for an objective update of the stress with respect to a ﬁxed coordinate system during the time step.2 Implementation into ABAQUS CREEP subroutine The implementation of the Duva and Crow constitutive equations for consolidation is also carried out using the ABAQUS CREEP facility. therefore. and easier handling of internal state variables.7) and (8.17) This. ∇ ∇ (8.8). Because here.5). (8. can be calculated from ˙ Equations (8. The stress increment. (1 + ν) (1 + ν)(1 − 2ν) (8. The stress increment. is calculated as σ = ∇ σ +Rσ t R t . and in ensuring meaningful results are obtained. The corotational stress rate σ is given by σ = ∇ ∇ (8. while the UMAT subroutine allows more general forms of constitutive laws to be implemented. D p is the rate of plastic deformation as given in Equation (8. D. with respect to the material reference frame. we adopt an explicit ﬁrstorder forward Euler integration.14) Eν E De + II : D e . for each time step can be determined by utilizing the ﬁrstorder Euler integration scheme σ =σ ∇ t. σ .2.13) ˙ The dilatation rate. This subroutine is suitable for constitutive equations in which the increments of inelastic strain are functions of the hydrostatic stress and the equivalent deviatoric stress described by Mises’ or Hill’s deﬁnitions. respectively. 2 (8. and the densiﬁcation rate. and D e is the rate of elastic deformation given by (8.
20) ∂σ ¯ ε cr and ε sw are the equivalent creep strains conjugate to the deviatoric stress and the ¯ mean stress. εsw .25) (8. ∂ εcr /∂σe .23) (8. which are then passed into the CREEP subroutine.5) has to be decomposed into two parts as follows 3 1 ˙ ε p = AS n−1 aσ + AS n−1 bσm I . The CREEP subroutine does not provide these quantities. ∂ εcr /∂p. .Implementation of porous material constitutive equations 203 ABAQUS computes the incremental creep strain components as 1 sw ε I. the stress components are required. Therefore.19) ∂σe . the plastic strain rate equation deﬁned in Equation (8. in which σ is the equivalent stress and p is the equivalent pressure and ∂ ε ¯ e stress given as 1 p = − (σxx + σyy + σzz ). (8. ¯ εsw ¯ 1 (n − 1)bσm ∂ εsw ¯ = − εsw ¯ + ∂p σm S2 In order to calculate σe and σm .26) . ∂ εsw /∂σe . respectively. ¯ ¯ sw /∂p.24) (8. given by εcr = εcr n + ¯ n= (8. (8. (8.22) 3 The variables are obtained as follows: ε cr = AS n−1 aσe t. εcr corresponds to the conventional deviatoric ¯ sw to the volumetric strain occurring because of void closure. The user is required to provide ε cr .28) ∂ ε cr ¯ = − εcr ¯ ∂p ∂ εsw ¯ = ∂σe εsw = Abσm S n−1 t. ¯ 3 where n is the direction normal to the yield surface. . The densiﬁcation rate .21) 2 3 The ﬁrst part is associated with the deviatoric stress and the second with the mean ¯ ¯ ¯ stress. creep strain and ε ¯ In order to implement the porous material constitutive equations using the CREEP subroutine.27) (8. (8. and therefore the USDFLD subroutine has to be utilized to access material point data and assign stress components to state variables. ¯ ∂ ε cr ¯ = ∂σe εcr ¯ 1 (n − 1)aσe + σe S2 (n − 1)bσm S2 (n − 1)aσe S2 . (8.
and Q activation energy.1 shows comparisons of the relative density evolution over time calculated by the porous material model using the CREEP and the UMAT subroutines.98 0. can therefore be obtained from Equation (8.67 1.18 T ≥ 750◦ C 95. The Duva and Crow porous plasticity model has been used (empirically) to approximate the behaviour in consolidation of continuous (SiC) ﬁbre. The relative density for each time increment can be determined by utilizing the ﬁrstorder Euler integration as given in Equation (8. The results can be seen to be identical.49 2.8).1. (Ti) metal matrix . The material constants.14).88 0. 8. A0 and n.83 Predicted using UMAT subroutine Predicted using CREEP subroutine 0.204 Porous plasticity Table 8. R the gas constant.78 0 10 20 30 40 50 Time (min) Fig. used are given in Table 8.93 Relative density 0.1 Graph showing comparisons of relative density–time curves obtained from the porous plasticity material model utilizing both an ABAQUS UMAT subroutine and a CREEP subroutine.29) in which T is the temperature. The porous plasticity material model implemented using the CREEP subroutine was veriﬁed against that using the UMAT subroutine by simulating a simple compression process using a single plane strain element subjected to inplane compressive load of 20 MPa at 925◦ C. T < 750◦ C A0 n 8.1 Material parameters. Figure 8. and A is calculated from A = A0 exp −Q RT (8.53 0.
8. Application to an aeroengine ‘bling’—a bladed ring—is shown in Fig.2. 8. respectively. discs. and aircraft landing gear. 8.2 Schematic diagram of an aeroengine bling showing region of reinforcement with Ti–MMC and a demonstrator disc (from King. 8.3 Application to consolidation of Ti–MMCs The simple consolidation model. composite materials (Ti–MMCs) being developed for potential application to aeroengine components together with shafts.Application to consolidation of Ti–MMCs 205 300 mm Reinforced area Fig. 1998). which was implemented using the CREEP subroutine as described above.3(a) and (b). brakes. The consolidation process is . The Ti–MMC material in the unconsolidated and fully consolidated state is shown in Fig. was set up to simulate (empirically) the consolidation of Ti–MMCs under a range of processing conditions.
SiC continuous ﬁbre composite material (a) unconsolidated and (b) consolidated at 925◦ C for 30 min at 15 MPa. .75 0 40 80 Time (min) 120 160 200 Fig. 8.4 Schematic diagram showing arrays of Ti–MMCs undergoing consolidation.5 Predicted and measured relative density evolution with time for Ti–6Al–4V/SiC composite consolidated at a constant temperature of 925◦ C with a pressure of 20 MPa.206 Porous plasticity (a) (b) Fig. Ti–MMC Die Fig.9 0.95 Relative density 0. 8.8 0. 8.3 Micrographs showing Ti–6Al–4V matrix.85 0. 1 Model prediction Experimental data 0.
‘Composites take off without a parachute’. P. ‘Inelastic deformation of porous materials’. respectively. 5(6). ‘The densiﬁcation of powders by powerlaw creep during hot isostatic pressing’. there are many more layers of ﬁbres and the pressure is applied by means of a mechanical punch.4.References 207 1 Model prediction 0. and at a temperature of 700◦ C under a constant pressure of 20 MPa. and McMeeking. ‘Analysis of consolidation of reinforced materials by powerlaw creep’. Acta Metallurgica. Journal of the Mechanics and Physics of Solids.M. (1992). shown schematically in Fig.D.5 and 8. ‘Pressure sintering by power law creep’. are shown in Figs 8. Ponte Castaneda.75 0 40 80 120 Time (min) 160 200 Fig. 37(6). 25. ‘Creep of powerlaw material containing spherical voids’. Sofronis. M. The mechanical die ensures that the Ti–MMCs undergo plane strain compression. Duva.D. P. and Ashby. King. Wilkinson.85 0. A. 8. which were carried out at a constant temperature of 925◦ C under a pressure of 20 MPa. ‘The effective mechanical properties of nonlinear isotropic composites’. 8.C.M. (1998). (1975). Generally. Duva. and Crow. P. Journal of the Mechanics and Physics of Solids. Transactions of ASME. 324.S. J. 1277. showing the important effect of temperature.8 0. (1989). (1994).F. (1991). 17. Some of the results of the simulations.95 Relative density Experimental data 0.9 0. R. 693.F. 23. Acta Metallurgica.6 Predicted and measured relative density evolution with time for Ti–6Al–4V/SiC composite consolidated at a constant temperature of 700◦ C with a pressure of 20 MPa. 40(1).M. P. 39(1). 88.6. References Cocks. Materials World. . D. 59(2). Journal of Applied Mechanics. (1992). and Crow. J. J. 45. Mechanics of Materials. 31.
This page intentionally left blank .
in polycrystalline nickelbase alloy C263 for temperatures both above and below the γ solvus. leading to combined creep and cyclic plasticity.1 Introduction Aeroengine components operating under inservice conditions are often subjected to a range of complex cyclic mechanical and thermal loading. 1974). Signiﬁcant microstructural change is therefore likely to occur during ordinary operation. leading to quite profound changes in the creep mechanisms in the material. Combustion chamber applications require the material to undergo temperature ﬂuctuations between 20◦ C and 950◦ C. which in turn depends on temperature. λpc . precipitate ‘cutting’ dominates. size. λp (or equivalently. The material also undergoes coarsening at temperature such that the particle spacing increases with time. and the temperature range is therefore such that the precipitate solvus can be exceeded during inservice operation. with a precipitate solvus temperature of 925◦ C (Betteridge and Heslop. The implementation into ABAQUS is carried out using a forward Euler integration scheme and we again use the initial stiffness as the material Jacobian for simplicity. and its inﬂuence on creep deformation and failure. exhaust ducting. . dislocation climb dominates. In this chapter. depending on the precipitate spacing. and bearing housings. T ). casings.9. 9. whereas below the critical spacing. Creep in an aeroengine combustor material 9. It is a ﬁneprecipitate strengthened alloy at 800◦ C. liners. Above a critical particle spacing.2 Physically based constitutive equations Creep in nickel alloy C263 occurs through diffusionactivated dislocation climb and precipitate ‘cutting’. The polycrystalline nickelbase superalloy (C263) is a commercial alloy used for stationary components in aeroengines such as combustion chambers. φp . rs . for a given precipitate volume fraction. we shall introduce a physically based creep model which explicitly accounts for microstructural change. controlling both deformation and component life.
pc = ˙ where V = Vc . The creep strain rate. the material had been subjected to the standard .210 Creep in aeroengine components All of these microstructural processes are embodied within creep constitutive equations. if φp > 0. F is the Helmholtz free energy. ν the frequency of dislocation jumping energy barriers. The constitutive equation set embodying all the mechanisms discussed above is given below. σ .2) in which σ1 is the maximum principal stress. the activation volume is Vc . H (σ1 )pc ˙ (9. Prior to creep testing. kb the Boltzman constant. H the Heaviside function. and and χ are material constants (χ being the sensitivity of stress state parameter).1) 4π 3φp −2 . V is dependent on the pinning distance. Creep tests have been carried out by RollsRoyce over the temperature range 700–750◦ C. which in turn depends on the precipitate spacing. ω is a scalar damage variable which has evolution equation ω= ˙ σ1 σe χ 1/3 ρn ρi ([4π /3φp ]1/3 − 2) 2 − F b ν exp ¯ kb T [4π /3φp ]1/3 M sinh σe V . All the other material constants are physical properties or measurable physical quantities.1). of mobile ˙ dislocations which in turn depends on the accumulated creep strain. in ˙ response to a uniaxial stress. the uniaxial strain rate. pc . (climbing). b the Burger’s vector. occurring largely within the tertiary creep regime. and by Zhang and Knowles (2001) on C263 over the temperature range 800–950◦ C. For precipitate cutting. namely ψ and associated with the multiplication of mobile dislocations and cavitation. ¯ kb T M(1 − ω) (cutting). (dislocation network) (9. V . In the case of dislocation pinning by precipitates. λp > λpc and λp < λd V = λd b2 . In the creep strain rate equation. is determined from the corresponding shear values ¯ ¯ using the usual relationships (Dieter. if φp = 0 and λp = rs ˙ ρn = ψ εc . also depends upon the density. ˙ In Equation (9. 1988): ε = γ /M and σ = Mτ . depends on the active obstruction mechanism. if φp > 0 and λp < λpc V = λp b2 . in which M ˙ ˙ ¯ is the Taylor factor. ρn . The equations contain just two unknown ﬁtting constants. The activation volume. and ρi the initial density of mobile dislocations. εc .
The results of the tests are shown in Fig.0 0 1 Time (s) 2 3 [× 106] 0 0.0 0 1.0 2.5 2.5 4.0 0. and (d) 950◦ C.0 6. heat treatment. (b) 750◦ C. quenching and ageing at 800◦ C for 8 h followed by air cooling.5 × 10−10 m (Frost and Ashby. 1982).0 3.5 0. This involves solutioning at 1150◦ C for 2 h.5 2 0.0 Strain (%) 2.1 Comparison of experimental and computed isothermal creep results at (a) 700◦ C.5 (b) 10 450 MPa 350 MPa 8 250 MPa 200 MPa Strain (%) 1.0 [× 106] (d) 0 2 4 6 Time (s) 8 10 [× 106] (c) 4. Burger’s vector for this material is taken as 2.0 250 MPa 200 MPa 160 MPa Strain (%) 30 40 MPa 25 20 15 10 50 MPa 5 0.0 4.3 0.4 Time (s) 0.1 0.2 0.Physically based constitutive equations 211 (a) 2.0 0. Fig.0 380 MPa 1.0 Time (s) 5. 9. The volume fraction of γ precipitate is known at each test temperature.6 [× 106] Experimental results Computed results (c) 800◦ C.0 3.5 0. The constants and constant groups appearing in the equations have been determined by standard optimization techniques.0 1.1 by the broken lines. and the initial density of dislocations can be estimated to .5 3.5 1. 9.0 320 MPa Strain (%) 270 MPa 6 4 0.
3 Multiaxial implementation into ABAQUS We assume multiaxial creep deformation in nickel alloy C263 to obey von Mises behaviour. Under multiaxial conditions. The tests have been carried out on thinwalled . F . which includes the Helmholtz free energy.57 be 1010 m−2 (Hull. F (J/atom) 7. the creep strain rate ‘direction’ is taken to be normal to the dissipative surface. The results of this process are shown in Table 9. is given in Equation (9. 9. 9. that is.1 Finite element modelling of biaxial creep tests A range of tests have been carried out on nickel alloy C263 in order to investigate its stress state sensitivity of creep failure. Vc . 1975).4 × 10−19 λpc (nm) 64. 9. The remaining temperatureindependent physical properties can then be determined using the results of the optimization and the model equations. The equations have ˙ been implemented into an ABAQUS UMAT using a simple forward Euler explicit integration scheme with the initial stiffness method. and σe the effective stress.3. the frequency of dislocations jumping energy barriers. the critical particle spacing. giving the multiaxial creep strain rate as 3 σ ˙ ε c = pc ˙ 2 σe in which σ is the deviatoric stress tensor.1. λpc . above which dislocation bowing dominates over particle cutting. the activation volume for cutting. The temperature dependence of the two empirical constants for the evolution of multiplication of mobile dislocation density and cavitation is shown in Appendix 9.1 × 1021 ¯ M 3.1.212 Creep in aeroengine components Table 9. the relationship between uniaxial creep strain rate and stress is identical to that between the effective creep strain rate and effective stress. and the Taylor factor. ν.9 Vc (m3 ) 4. given by σe = 3 σ :σ 2 1/2 and the effective creep strain rate.1 Physical constants determined from creep data by optimization. in an analogous way to that in plasticity. The computed creep curves resulting from the determination of the creep parameters are also shown in Fig. the material Jacobian is speciﬁed simply as the elasticity matrix.1). that is.05 × 10−27 ν (s−1 ) 15. pc .1.
The loading conditions applied are summarized in Table 9. However. is shown in Table 9. The only unknown parameter in the equations is the stress state sensitivity. Shear stress (MPa) Tension Tension–torsion Torsion Compression–torsion 0 160 184 160 Axial stress (MPa) 320 160 0 −160 σ1 /σe 1. The test specimens were modelled by developing threedimensional meshes of the gauge sections (i. Because of the need to carry out compression tests.12 and 4. The test specimen gauge section has length 16 mm. not as good as desired because of problems with temperature control. the specimen wall thickness was such that an assumption of uniform shear strain and stress through the wall thickness would not have been appropriate. the results do demonstrate the strong dependence of creep life on stress state.2(a). the wall thickness was constrained by buckling problems. A twodimensional ﬁnite element model was therefore necessitated. through to failure. is correctly modelled using the above constitutive equations with the material properties given in Manonukul et al. Geometrical nonlinearity was accounted for using the standard ABAQUS large strain formulation.12 mm. and compression–torsion. The creep behaviour of the material was described by the constitutive equations given above. tension–torsion. The four tests described above were simulated using the ﬁnite element model.81 0.58 0. that is. unfortunately. respectively. However. 240 eightnoded.0 0. The quality of the creep data obtained for the torsion test is. Parametric studies were carried out in order .e. The experimental test results are shown in Fig.2 Summary of tests and loading conditions.31 213 tubular specimens which have been subjected to uniaxial tension. given by σ1 /σe . The loads were increased linearly from zero to the maximum (to give the desired stresses) over 14 s. for each test. The material was subjected to the standard heat treatment such that its uniaxial creep behaviour. because of the radial variation of stress. because of the ease of speciﬁcation of torsional loading conditions using threedimensional elements (as opposed to axisymmetric elements with ‘twist’ in ABAQUS). threedimensional solid elements were used so that the specimen thickness contained two elements. and corresponding stress state. 320 MPa. The stress state. with internal and external radii of 3. 9. and were then held at the maximum value.2.2. Under torsional loading. As a result. (2002). uniformsection tubes).Multiaxial implementation into ABAQUS Table 9. namely 320 MPa. torsion. χ. The tests were carried out at 800◦ C such that the von Mises effective stress for each test was the same. the loads applied were chosen to ensure that the average effective stress through the thickness was that desired.
This value lies within the range for nickel alloys discussed by Dyson and Loveday (1981). and this is captured well by the model. to determine χ. Considerably different creep lives are seen to occur depending on the stress state.1 0. The computed creep curve for uniaxial compression shown in Fig.2 0. The computed results obtained using a value for χ of 0. even though there is no creep cavitation .214 Creep in aeroengine components (a) 0.1 0.15 0. 9.25 Torsion Equivalent creep strain 0.15 Tension–torsion Compression Tension 0.05 0 0 100000 200000 300000 400000 Time (s) Compression–torsion Tension 500000 600000 700000 800000 (b) 0.3 Tension–torsion 0. This was done by choosing that value of χ which provided computed results closest to those seen in the experiments.25 Equivalent creep strain Compression–torsion 0. 9.3 0. 9.2 (a) Experimental and (b) computed equivalent creep strain versus time for isothermal creep tests carried out at 800◦ C with effective stress 320 MPa for the loading conditions shown.05 0 0 100000 200000 300000 400000 Time (s) 500000 600000 700000 800000 Fig.2(b) can be seen to have a slowly increasing gradient.38 are shown in Fig.2 Torsion 0.2(b).
which inﬂuence creep damage evolution. 161. The circular notches have diameter 1. in which the damage level is 0. and 194 MPa in Fig. nominal stresses of 104. This results from both the precipitate coarsening and the multiplication of mobile dislocations. The predicted displacements of the top boundary of the test specimen are shown for the three applied. both of which are included in the model.8 1.4 1.6 1.3. 9. At this stress level.3 Predicted displacement of the top boundary versus time (solid lines) for the double notch specimen subjected to the stresses shown. 9.4(a). Broken straight lines indicate experimentally determined times to failure. The specimens have gauge length 25.2 Prediction of notched bar creep behaviour Three experimental tests have been carried out on double notched bar test specimens. as discussed above. The effect of the notches is to introduce multiaxial stress states local to the notches.4 mm and diameter 5.6 0.62 mm. just one quarter of the specimen section was modelled explicitly. Because of symmetry. The test specimens were modelled using axisymmetric elements. The creep damage evolution for the case of the 161 MPa nominal stress is shown in Fig. creep damage is predicted to initiate at the notch root after creeping for about 13 h.12 mm and they are separated by 7 mm along the length of the specimen. This is . 9. and 194 MPa. The specimen lifetimes have been predicted and the lives compared with those obtained in the experiments. Uniaxial loading was applied to generate nominal section stresses of 104. respectively.26 or higher.Multiaxial implementation into ABAQUS 215 2 Displacement of the top boundary (mm) 1.4 0.3 through the specimen section Fig. the damage has evolved right across the test specimen. occurring.8 0.3.2 1 0.2 0 0 500 1000 Time (h) 1500 2000 2500 161 MPa 104 MPa 194 MPa Indicates predicted failure on the basis of a damage level of 0. 161. 9. By the time creep has continued for 166 h.
73E–02 +9.4 Creep damage ﬁelds (a) predicted by the model.38E–01 +1. and (c) from a surface void count using the micrograph.00E–01 (b) (c) 2 6 3 4 2 1 2 7 8 7 4 9 3 5 2 2 12 15 9 4 1 2 5 7 6 5 22 17 0 0 2 3 2 8 30 1 2 0 4 0 0 Number of creep voids 0 1–3 4–6 7–9 10–12 13 Fig. 9.09E–02 +1.80E–01 +3.00E+00 +2.09E–01 +2.33E–01 +2.00E–02 +4.62E–01 +1.15E–01 +1.216 Creep in aeroengine components (a) Cavitation damage parameter +0.56E–01 +2. (b) observed in the microstructure.) .85E–01 +2. (See also Plate 4.36E–02 +6.
and Loveday. 9.3. 406–421. Springer. The damage ﬁelds in Fig. A further reason for continuing the analysis is the difﬁculty of deﬁning creep failure in the calculations. However. and hence Fig. together with a micrograph of the corresponding region of the test specimen in Fig. it is simply the measured time until the specimen breaks. and Hayhurst.3 Predicted and experimentally determined top boundary displacement at failure (deﬁned as experimental specimen separation).3 by the dashed lines. A surface void count has been carried out and the results are shown in Fig. The Nimonic Alloys. Stress (MPa) 104 161 Predicted displacement at failure (mm) 1. In Ponter. are shown in Table 9. In the experiments. and to assume that rupture occurs once the rate becomes large. 9. W. This results from the coalescence of creep cavitation and the propagation of macroscopic creep cracking. On this basis.97 0. The predicted displacement at failure was determined as the predicted displacement when a damage level of about 0.4(c).3 had been achieved across the specimen section. the analyses were not stopped once a creep damage level of 0. (eds).F. 9.4(a) can be seen to predict reasonably well both the site of major cracking and the distribution of experimentally observed creep cavitation.S. D. 9. The experimentally determined failure times are shown in Fig.9 0. . References Betteridge. Edward Arnold. Proceedings of the Creep in Structures IUTAM Symposium.4(a).1 Experimental displacement at failure (mm) 1. Dyson and McLean (1990) have argued that creep rupture occurs typically when the creep cavity damage achieves a value of about 0. M. J.References Table 9. which show good agreement. and Heslop. A.R. An alternative way to deﬁne calculated specimen rupture time is simply to examine the top boundary displacement rate. UK.11 217 shown in Fig. B. Microcracking can be seen to initiate at the notch root at an angle of about 22◦ from the horizontal.4(b). The measured and calculated displacements of the top boundary at failure. (1981).3 is reasonable.S. 9. 9. this is easier. 9.3. Berlin. Dyson. (1974). London.3.3 had been achieved through the specimen section.R.3 shows the calculated top boundary displacement continuing to increase as the damage level exceeds 0. the comparison of predicted and experimental time to rupture results shown in Fig.
). ‘Deformation behaviour and development of microstructure during creep of a nickelbase superalloy’.. F. Manonukul. A. and McLean.F. Frost. Proceedings of the 9th International Conference on Creep and Fracture of Engineering Materials and Stuctures. (1988).H. ‘Physicallybased model for creep in nickelbase superalloy C263 both above and below the γ solvus’.J.62 . Mechanical Metallurgy. M. (1982). (1990). UK. Pergamon Press. Dieter.0 18. (2002). Dunne. (ed. Oxford. UK. Y.84 800◦ C 10.. J. Acta Materialia. 30. H. 950◦ C 1.187 × 10−2 T + 3. ‘Creep deformation of engineering alloys— development from physical modelling’. T . B. Introduction to Dislocations.E. 50(11). Iron and Steel Institute of Japan International (ISIJ).F.915.P. The dependence of constants 700◦ C ψ 496.1×10−19 19. Appendix 9. D. Pergamon Press. 2917–2931.01×10−2 and on temperature. D. and Knowles. Gomer Press. 2nd edition. (1975). G.. London. Zhang. McGrawHill Book Co. on temperature.261 × 10−5 T 2 − 1. In Parker. and Knowles.537 × 10−9 T 3 + 1. φ.1 1. DeformationMechanism Maps: The Plasticity and Creep of Metals and Ceramics. UK.218 Creep in aeroengine components Dyson. April 2001.1 The dependence of precipitate volume fraction. and Ashby. Oxford.E. in C263 φ = −4. (2001). Hull. UK.D.58×10−2 750◦ C 435.2 3. D. M. 802–811.
We use timeindependent cyclic plasticity with both isotropic and kinematic hardening which we combine with the physically based creep model described in Chapter 9. Cyclic plasticity. we examined creep in the aeroengine combustor material nickelbase superalloy C263. and isotropic strain softening/hardening over many cycles. (2002a. a uniﬁed viscoplasticitycreep model is not adopted because of the numerical problems that can arise when using such a model for timeindependent plasticity. it would. 10. We then address isothermal cyclic plasticity and creep. The plasticity model is implemented into ABAQUS Implicit by means of a UMAT subroutine which employs a simple ﬁrstorder forward (explicit) integration scheme and the initial stiffness method. In this chapter. In any case.10. and anisothermal cyclic plasticity between 450◦ C and 950◦ C. the material exhibits both kinematic hardening within individual cycles.b). Above ∼600◦ C. depending on the temperature. Because of the wide temperature range considered here (20–950◦ C). creep.1 Introduction In Chapter 9. and TMF 10. (2002). the material starts to exhibit strainrate dependence. The resulting stress–strain hysteresis loops show considerable change as plastic strain accumulates. During zeromean. isothermal. A nonuniﬁed approach has therefore been adopted in which conventional timeindependent reversed plasticity . we address combined anisothermal cyclic plasticity and creep. who considered uniaxial creep. under similar loading. straincontrolled reversed plasticity below this temperature.2 Constitutive equations for cyclic plasticity C263 is largely rateindependent and shows very limited creep response at temperatures below ∼600◦ C. followed by an examination of thermomechanical fatigue in nickel alloy C263. in addition. Viscoplastic constitutive equations for a polycrystalline nickelbase alloy have been presented by Yaguchi et al. prevent the retention of the physical basis of the creep model to be used here. presented by Manonukul et al.
1990) is employed for timeindependent plasticity. and TMF theory for lowtemperature deformation has been combined with the physically based creep model for elevated temperature deformation. by dx = and dr = b(Q − r) dp. and a uniﬁed viscoplasticity theory would be appropriate.1) There have been many developments in the modelling of kinematic hardening that take account. The increment in plastic strain is determined in the conventional way assuming a von . dε e . creep. for simplicity here. (10. However. the Chaboche combined isotropic and kinematic hardening model (Lemaitre and Chaboche. creep. However. and plastic slip. both sets of mechanisms operate. Conversely. Temperature and accumulated plastic strain dependent material parameters associated with kinematic hardening are c and γ .2) (10. (10.3) . 2003). our approach can be justiﬁed. for example. dε. dεθ . The increment in nonlinear kinematic hardening dx and the increment in isotropic hardening dr are given.4) Here. It is therefore assumed that the elastic. The further justiﬁcation for the separation of creep and timeindependent plasticity terms is that at 20◦ C. At about 800◦ C. respectively. and thermal. deformation is dominated by diffusioncontrolled processes. A large number of fully reversed cyclic plasticity tests have been carried out in order to determine the temperaturedependent material parameters arising in the kinematic hardening equations. for example. at 950◦ C. r the isotropic hardening stress. of nonproportional loadings (Voyiadjis and Abu AlRub. timeindependent plastic. 20–950◦ C. for the main application considered here—thermomechanical fatigue with considerable temperature variations.220 Cyclic plasticity. respectively. is negligible. aided by thermal activation or otherwise. Deformation takes place by plastic slip which is not aided by thermally activated dislocation climb. (thermally activated) creep simply does not occur. and b and Q are temperature dependent material parameters associated with isotropic hardening. x is the kinematic hardening stress tensor. is given by dε = dεe + dε p + dε c + dε θ . dεp . dp is the effective timeindependent plastic strain increment. strain increments may be additively decomposed such that the total strain. dεc . where dp = 2 p dε : dεp 3 1/2 2 c dεp − γ x dp 3 (10.
by dλ = (∂f /∂σ) · C dε (∂f /∂σ) · {(2/3)c(∂f /∂σ) − γ x} + b(Q − R) + (∂f /∂σ) · C(∂f /∂σ) 3 ( 2 )(σ /σe ) · C dε 3 3 3 3 ( 2 )(σ /σe )·{( 2 )c( 2 )(σ /σe ) − γ x} + b(Q − R) + ( 2 )(σ /σe )· C( 2 )(σ /σe ) 3 = . and σt+ x t+ εt+ rt+ p t t t t = σt + dσt . p p . = rt + drt .Constitutive equations for cyclic plasticity 221 Mises yield function. The increment in thermal strain is calculated in the usual way as (10.5) = dλ 2 σe ∂σ in which dλ is the plastic multiplier which. = εt + dεt . The multiaxial constitutive equations presented above for creep and plasticity have been implemented into a UMAT user subroutine using the simple ﬁrstorder explicit Euler forward integration scheme as follows dσt = C dεt − 3 σ dλ − dεc − dεθ 2 σe in which the increment in creep strain is obtained from Equation (9.7) in which α is the coefﬁcient of thermal expansivity. (10. for combined isotropic and kinematic hardening.6) The effective stress is. dθ the increment in temperature. f . and I the identity matrix. using Voigt notation.1. = x t + dx t .33) and those in Table 9. given by σe = 3 (σ − x ) : (σ − x ) 2 dε θ = α dθI 1/2 . is given. of course. given by 3 f = (σ − x ) : (σ − x ) 2 so that dεp = dλ 1/2 − r − σy = 0 ∂f 3 σ −x (10.
in effect. . under conditions of both creep and reversed plasticity.. creep. through the temperature dependence of the material parameters for cyclic plasticity. the creep model at 950◦ C was developed from experimental data obtained for a much lower stress level (a maximum of 50 MPa) whereas here. our cyclic plasticity model does not contain an explicit static recovery term. the experiments show a much more progressive stress relaxation at the peak strains suggesting that the rate of relaxation is not well represented by the creep model. The physically based creep model employed here is able to capture these changes and to represent the micro. LCF tests have been carried out on C263 at 800◦ C and 950◦ C under fully reversed (R = −1).1(c) shows the results for 950◦ C. respectively. 2001). straincontrolled loading through to failure. but this can be seen at 950◦ C in the computed result.222 Cyclic plasticity. Cyclic plasticity tests between 20◦ C and 950◦ C have been carried out.1. solution treated. two strain rates have been used such that ramping times (the time over which the strain is applied linearly with time) of 1 and 10 s have been imposed resulting in the hysteresis loops shown in Fig. The effect of precipitate volume fraction on cyclic plasticity behaviour is introduced into the model. the precipitate volume fraction is known. and climb. In addition. and because the precipitate coarsening kinetics has been quantiﬁed. However. a 1s strain hold has been imposed at the tensile and compressive peak strains. This appears in the cyclic plasticity response as a reduction in the overall stress level. The effect of solution treating on cyclic plastic response is also very signiﬁcant and has been addressed. First. With the combined creep–cyclic plasticity model described above. both size and hence spacing of the precipitate is known. 10. The ratecontrolling mechanisms for creep deformation change from precipitate cutting and dislocation pinning at precipitates. The experimentally determined and calculated stress–strain hysteresis loops for the ﬁrst and second cycles of the standard heat treated material are shown in Fig. 10. At 800◦ C. occurring through relaxation.3 Constitutive equations for C263 undergoing TMF The thermomechanical loading cycles to which the nickelbase superalloy C263 is subjected inservice are such that the γ solvus is exceeded so that the material is. therefore. and TMF 10. At 800◦ C.and macroscale creep deformation. Figure 10. For a given temperature. this does not result in a distinctive stress relaxation. These quantities appear explicitly in the physically based creep model. the temperature dependence of the isotropic and kinematic hardening constants have been determined (Manonukul et al. C263 creeps quite signiﬁcantly. to that of the pinning of dislocations through the establishment of a dislocation network. This occurs for a number of reasons. At temperatures of about 800◦ C and above.1(a) and (b). stresses of over 200 MPa are developed in the hysteresis loop. In these cycles. This has a profound effect on the behaviour of the material.
6 – 0.2 0.2 0.Constitutive equations for C263 undergoing TMF 223 (a) 600 400 200 0 – 0. and further.6 (b) 600 400 200 0 – 0.6 – 300 Experimental Computed Fig. strain range 0. Note that the microstructure of the material being loaded at 950◦ C is considerably different to that at 800◦ C because of the dissolution of the precipitate at the higher temperature. the creep and plastic strain increments have.4 0.1 Experimental and computed LCF stress–strain curves for the ﬁrst and second cycles at (a) 800◦ C.6 (c) 300 100 – 0.2 – 100 0 0.6 – 0.4 – 0.6 – 0.4 0.4 – 0.2 – 200 – 400 – 600 0 0. . strain range 0.2 0. 10. 10 s ramping time.2 – 200 – 400 – 600 0 0. (b) 800◦ C. strain range 1%.75%. for both temperatures considered here. been decoupled in the present work. and (c) 950◦ C. of course. the peak stress levels are reasonably well predicted.9%. However. 1 s ramping time.4 – 0.4 0.
9%. 1 s ramping time. creep. The life is overpredicted for (a) since creep damage accumulation has ceased to be the dominant failure process. 10. Good predictions of life can be seen to be achieved for (b) and (c) for which intergranular creep cavitation dominates. and TMF 600 400 (b) 200 (c) 0 0 – 200 (b) – 400 (a) 200 400 (c) 600 800 1000 1200 1400 (a) – 600 Experimental Computed Fig.2. through to failure. 10. indicating that lifetime under these conditions is dominated by fatigue .2 shows the comparisons of computed and experimental peak stresses versus cycles for each of the cases (a). Intergranular creep damage typically observed under these conditions is shown in Fig.224 Cyclic plasticity. 10. 10. with 1s ramp times. strain range 0. LCF tests carried out at 800◦ C. and (c) in Fig. A micrograph for these conditions shows that transgranular fracture is now occurring. and (c) 950◦ C. Figure 10.2.2 Experimental and computed LCF peak stresses versus cycles at (a) 800◦ C. led to sample fracture surfaces showing transgranular cracking. as shown in Fig. (b). strain range 1%. strain range 0. Microstructural examination of specimens tested at 800◦ C and 950◦ C has been carried out after specimen failure. 10 s ramping time.75%.1. (b) 800◦ C.
The lifetime of the material under these conditions may well be inﬂuenced quite strongly by creep damage processes. After a time of 45 s. The volume fraction of precipitate is decreasing. considering the inphase loading shown in Fig. below the γ solvus temperature.Constitutive equations for C263 undergoing TMF 225 processes.3(a).1 TMF in C263 both above and below the γ solvus TMF tests have been carried out on C263 under conditions of both in and out of phase loading. which dominated in some regions of the material. 10. As the temperature increases. The creep rate becomes determined by dislocation network . However. becomes zero. in effect in the solution treated state. the test specimen has undergone a prior temperature loading cycle. therefore. The results of the tests carried out at 800◦ C shown in Fig. The dominant failure process for these conditions is intergranular creep fracture. and at 925◦ C. and in fact. First. The model for the solution treated plasticity and creep behaviour give the computed behaviour also shown in Fig. Strain controlled tests have been carried out between strains of zero and a peak of 0. which is dominated by creep softening and cavitation. for the material containing the precipitate volume fraction. 10. The isotropic creep damage model used in the present work is not capable of predicting crack closure effects. 10. are shown in Fig.6%. which is dependent on temperature). give stress levels that lie a little below that expected from both the solution treated and the standard treated material.3. does not show this effect. It should be noted that prior to the imposition of the strain. 10.e. even for a ramp time of 1 s. The consequence is that the material is. in which the temperature is also varied between 300◦ C and 950◦ C. a number of processes are taking place. Examination of microstructures of specimens tested at 950◦ C showed that intergranular cracking dominated at these temperatures. For the cyclic plasticity modelling. and a further one for the standard treated material (i. one for the solution treated material (with the appropriate temperature dependence). two sets of material parameters were determined.2(a) and (b) show the development of macrocracking towards the end of the test which inﬂuences the observed stress–strain response which becomes different in tension and compression because of crack closure effects. together with the results obtained. microstructural examination of specimens also tested at 800◦ C but with a considerably longer ramp time of 10 s showed the development of intergranular damage. The strain and temperature loading imposed. 10. The modelling of the behaviour has therefore been carried out with this assumption. the experimental results obtained for the initial loading.3. however. the experimental results give stress levels that would be anticipated for such a material. This is necessary to enable temperature compensation to be carried out on the TMF test rig. The result at 950◦ C. the temperature has exceeded 925◦ C.3(a).
and (c) inphase TMF at temperature range 300–800◦ C . At a time of 45 s.5 – 0.4 – 0. creep.1 – 100 – 200 – 300 Strain (%) 1000 500 0 0 0.6 Strain (%) 800 0 0.4 – 0. which .2 0 – 200 0 – 400 – 600 – 800 Strain (%) Experimental Predicted 0.1 0.1 – 100 – 200 – 300 Strain (%) 0.7 – 0.2 0. 2002) and increases as the temperature increases.226 Cyclic plasticity.6 1000 500 0 0 300 200 Stress (MPa) 100 – 0. 10.7 50 100 Time (s) Time (s) 50 100 (b) Strain (%) 0 0 – 0.1 Temperature (8C) 1000 500 0 0 50 100 Time (s) Fig.5 0. and a comparatively small elastic region (though creep deformation is still occurring) can be seen.6 0.3 0.3 0 – 0.1 Temperature (8C) 50 100 Time (s) Time (s) 50 100 (c) 600 400 Stress (MPa) 200 – 0.2 0 0 50 100 Time (s) Stress (MPa) 100 Temperature (8C) 0 – 0. both the strain and temperature loading are reversed. The combined effects of these processes result in the predicted decreasing stress level.2 – 0.4 0. pinning (Manonukul et al.4 0. and TMF Strain (%) (a) 300 200 0.3 Experimental and predicted stress–strain curves for (a) inphase TMF at temperature range 300–950◦ C and (b) outofphase TMF at temperature range 300–950◦ C..
2917–2931. ‘Thermodynamic based model for the evolution equation of the backstress in cyclic plasticity’. stress relaxation can be seen to occur. Mechanics of Solid Materials.P. The measured and predicted stresses developed with increasing plastic deformation show good agreement and.3(c). Voyiadjis.. J.References 227 is followed by a conventional plasticity region in which creep deformation becomes negligibly small as the temperature decreases towards 300◦ C. Yaguchi. England. T. M. and Abu AlRub R. unlike the ﬁrst quartercycle. 1083–1109. Cambridge. Yaguchi. International Journal of Plasticity. G. With the reversal of the loading again.E. Cambridge University Press. T. M. Oxford University.. The stress levels measured and predicted by the model are those that would be anticipated from the solution treated material. Note that the material continues to harden during this phase. International Journal of Plasticity. 10. International Journal of Plasticity. M. The loading conditions and results obtained for outofphase TMF are shown in Fig. In this case. 19(12). D. (2001). and Knowles. (2002). the material is in compression. A. but again.. (2002b)..E. ‘A viscoplastic constitutive model for nickelbase superalloy Part II: modelling under anisothermal conditions’. References Lemaitre. a rather larger elastic region is observed because of the increased yield stress at lower temperature..L. D. Yamamoto. ‘Physicallybased model for creep in nickelbase superalloy C263 both above and below the γ solvus’. M. Yamamoto. F.. the high tensile stresses occur during the low temperature parts of the loading cycle. ‘A viscoplastic constitutive model for nickelbase superalloy Part I: kinematic hardening rule of anisotropic dynamic recovery’. 2121–2147. and because of the existing hardening that has already taken place. For inphase TMF with the same strain range. The standard heat treated material would show considerably higher stresses. Manonukul. At the higher temperatures. F. J. and the features of the experimentally obtained results can be seen to be reasonably reproduced. Department of Engineering Science. ‘Thermomechanical fatigue in nickelbase superalloy C263’.Z. and Ogata. Manonukul.3(b). are those that would be expected for the solution treated material. 50. A. and Chaboche. (2003). Good comparisons are achieved for these conditions. with no stress relaxation occurring at the higher compressive stresses because of the elimination of creep. (1990). 18(8). . but with the temperature varying between 300◦ C and 800◦ C. 1111–1131.. Acta Materialia. the experimental and predicted results obtained are shown in Fig. Internal Report. Dunne. Dunne. 10. and Ogata. (2002a).P.K. 18(8).. and Knowles. and conventional plastic behaviour is observed.
This page intentionally left blank .
n (A2) (A3) Double contraction (double dot product) of two secondorder tensors gives a scalar n σ : ε = σij εij = i=1 j =1 n n σij εij . i=1 (A1) v · u = vi ui = v · u = u · v. (A5) (A6) ε : σ = σ : ε. (A8) . (A4) ε : σ = εij σij = i=1 j =1 εij σij . The dot product of a ﬁrstorder tensor with a secondorder tensor gives a ﬁrstorder tensor n σ · n ⇒ (σij nj )i = j =1 n σij nj . Einstein summation convention: ui vi = i=1 n ui vi u · v = ui vi = i=1 n ui vi . (A7) n · σ ⇒ (nj σji )i = j =1 nj σji .Appendix A: Elements of tensor algebra The dot (scalar) product of two ﬁrstorder tensors (vectors) gives a scalar n NB. vi ui .
The dyadic (direct) product of two secondorder tensors gives a fourthorder tensor f ⊗ g ⇒ (f ⊗ g)ijkl = fij gkl . The dyadic (direct) product of two ﬁrstorder tensors gives a secondorder tensor u ⊗ v ⇒ (u ⊗ v)ij = ui vj .j. if cijkl = cklij for any i. if σij = σji for any i.k=1 ξijk σjk ei . if i = j. (A12) Double contraction (double dot product) of a fourthorder tensor with a secondorder tensor gives a secondorder tensor n n c : ε ⇒ (c : ε)ij = cijkl εkl = k=1 l=1 cijkl εkl (A13) hence the expression c:ε=ε:c (A14) is valid only if c exhibits major symmetry (i.e. δij = 0. u3 v3 n (A10) and then u1 v1 u ⊗ v = ⎝u2 v1 u3 v1 The product of two secondorder tensors gives a secondorder tensor a · b ⇒ (a · b)ij = aik bkj = k=1 aik bkj .k. (A11) Double contraction (double dot product) of a thirdorder tensor with a secondorder tensor gives a ﬁrstorder tensor 3 ξ :σ = i.e.230 Appendix A: Elements of tensor algebra hence the expression n·σ =σ ·n (A9) is valid only if σ is symmetric (i. For example. Kronecker delta—a special secondorder tensor δ ⇒ δij δij = 1. if i = j. if ⎡ ⎤ u1 ⎣ u2 ⎦ u= u3 ⎛ ⎡ ⎤ v1 ⎣v2 ⎦ . j ). v= v3 u1 v2 u2 v2 u3 v2 ⎞ u1 v3 u 2 v3 ⎠ . (A16) (A15) .j .l).
g. (A18) (A17) which is valid for any secondorder tensor ε. ∂xj ∂ui = δij . σ ) with respect to its tensorial argument (e. u. which is valid only if the secondorder tensor ε is symmetric. ∂σkl (A24) . ∂εkl (A23) ∂σij = Iijkl . ε).g.Appendix A: Elements of tensor algebra 231 The unit fourthorder tensor (exhibits major but not minor symmetry) I ⇒ Iijkl = δik δjl has the following important property I : ε = ε : I. (A20) Differentiation Differentiation of a tensor valued function (e. x. A symmetrized unit fourthorder tensor (exhibits both major and minor symmetry) s I s ⇒ Iijkl = 1 (δik δjl + δil δjk ) 2 (A19) ensures the following identity I s : ε = ε : I s. Example: Differentiation of the ﬁrstorder tensor u with respect to the ﬁrstorder tensor x gives the secondorder tensor ∂u ⇒ ∂x Special case: ∂u ⇒ ∂u ∂u ∂u =δ= ij ∂u ∂x = ij ∂ui . ∂xj (A21) (A22) Example: Differentiation of the secondorder tensor σ with respect to the secondorder tensor ε gives the fourthorder tensor ∂σ ⇒ ∂ε Special case: ∂σ ⇒ ∂σ ∂σ ∂σ =I= ijkl ∂σ ∂ε = ijkl ∂σij .
⎡⎡ ∂σ ∂σij ∂σji = . ∂xi (A26) The gradient of a ﬁrstorder tensor ﬁeld gives a secondorder tensor ∇v ⇒ (A27) The gradient of a secondorder tensor ﬁeld gives a ﬁrstorder tensor with its components being secondorder tensors ∇σ ⇒ (∇σ )j = Hence. ∂xi ∂vj . ∂xi ∂xi ⎤ ∂σxz ⎤ ∂x ⎥⎥ ⎥⎥ ∂σxz ⎥⎥ ⎥⎥ ∂y ⎥⎥ ⎥⎥ ∂σxz ⎦⎥ ⎥ ⎥ ∂z ⎥ ⎥ ⎤ ∂σyz ⎥ ⎥ ⎥⎥ ∂x ⎥⎥ ⎥ ∂σyz ⎥⎥ ⎥⎥ .232 Appendix A: Elements of tensor algebra The chain rule Example: Secondorder tensor f depends on a secondorder tensor u and a scalar v ∂f ∂u ∂f ∂v ∂f = : + ⊗ ⇒ ∂x ∂u ∂x ∂v ∂x ∂f ∂x = ijkl ∂fij ∂fij ∂umn ∂fij ∂v = + . ∂y ⎥⎥ ⎥⎥ ∂σyz ⎥⎥ ⎦⎥ ∂z ⎥ ⎥ ⎥ ⎤⎥ ∂σzz ⎥ ⎥ ∂x ⎥ ⎥ ⎥⎥ ∂σzz ⎥ ⎥ ⎥⎥ ∂y ⎥ ⎥ ⎥⎥ ∂σzz ⎦ ⎦ ∂z (A28) ⎢⎢ ∂x ⎢⎢ ⎢⎢ ∂σxx ⎢⎢ ⎢⎢ ∂y ⎢⎢ ⎢⎣ ∂σxx ⎢ ⎢ ⎢ ∂z ⎢ ⎢⎡ ∂σ yx ⎢ ⎢⎢ ⎢⎢ ∂x ⎢⎢ ⎢ ∂σyx ∇σ = ⎢⎢ ⎢⎢ ∂y ⎢⎢ ⎢⎢ ∂σyx ⎢⎣ ⎢ ∂z ⎢ ⎢ ⎢⎡ ⎢ ∂σzx ⎢ ⎢ ⎢ ∂x ⎢⎢ ⎢ ⎢ ∂σzx ⎢⎢ ⎢ ⎢ ∂y ⎢⎢ ⎣ ⎣ ∂σ zx ∂z xx ∂σxy ∂x ∂σxy ∂y ∂σxy ∂z ∂σyy ∂x ∂σyy ∂y ∂σyy ∂z ∂σzy ∂x ∂σzy ∂y ∂σzy ∂z (A29) . ∂xkl ∂umn ∂xkl ∂v ∂xkl (A25) The gradient of a scalar ﬁeld gives a ﬁrstorder tensor ∇f ⇒ ∂f .
Appendix A: Elements of tensor algebra 233 The divergence of a secondorder tensor gives a ﬁrstorder tensor div σ = Tr[(∇σ )i ]. Hence. R1 · R2 = R2 · R1. (A35) (A34) (A33) (A32) . ∂σxx ⎢ ⎢ ∂x ⎢ ⎢ ⎢ ⎢ ∂σxx ⎢Tr ⎢ ⎢ ⎢ ∂y ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ∂σxx ⎢ ⎢ ∂z ⎤ ⎢ ⎡ ⎢ ⎢ ∂σyx ⎥ ⎢ ⎢ ⎢ ⎥ ⎢ ⎢ ∂x ⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ∂σyx ⎥ = ⎢Tr ⎢ ⎥ ⎢ ⎢ ∂y ⎥ ⎢ ⎢ ⎥ ⎢ ⎣ ∂σ yx ⎦ ⎢ ⎢ ∂z ⎢ ⎢ ⎡ ⎢ ∂σzx ⎢ ⎢ ⎢ ∂x ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ∂σzx ⎢ Tr ⎢ ⎢ ⎢ ∂y ⎢ ⎢ ⎣ ⎣ ∂σzx ∂z ⎡ ⎡ (A30) ⎡ ⎢Tr ⎢ ⎢ ⎢ div σ = ⎢Tr ⎢ ⎢ ⎢ ⎣ Tr ∂σij ∂xi ∂σij ∂xi ∂σij ∂xi ⎤⎤ ∂σxy ∂σxz ∂x ∂x ⎥⎥ ⎥⎥ ∂σxy ∂σxz ⎥⎥ ⎥⎥ ⎥ ∂y ∂y ⎥⎥ ⎥⎥ ⎥ ∂σxy ∂σxz ⎦⎥ ⎥ ∂z ∂z ⎥ ⎥ ⎤⎥ ⎡ ⎤ ∂σxy ∂σxz ∂σxx ∂σyy ∂σyz ⎥ ⎥ + + ∂y ∂z ⎥ ∂x ∂x ⎥⎥ ⎢ ∂x ⎥⎥ ⎢ ⎥ ⎥⎥ ⎢ ∂σ ∂σyy ∂σyz ⎥⎥ ⎢ yx ∂σyy ∂σyz ⎥ ⎥ + + ⎥⎥ = ⎢ ⎥. Successive ﬁnite rotations do not commute. that is. ∂y ∂y ⎥⎥ ⎢ ∂x ∂y ∂y ⎥ ⎥⎥ ⎢ ⎥ ∂σyy ∂σyz ⎦⎥ ⎣ ∂σzx ∂σzy ∂σzz ⎦ ⎥ + + ∂z ∂y ⎥ ∂x ∂y ∂z ⎥ ⎤⎥ ∂σzy ∂σzz ⎥ ⎥ ∂x ∂x ⎥ ⎥ ⎥⎥ ⎥ ∂σzy ∂σzz ⎥ ⎥ ⎥⎥ ⎥ ∂y ∂y ⎥ ⎥ ⎥⎥ ∂σzy ∂σzz ⎦ ⎦ ∂z ∂z (A31) Rotation Finite rotation is mathematically described by the orthogonal secondorder tensor R −1 = R T . which transforms a ﬁrstorder tensor as follows x =R·x and a secondorder tensor as follows σ = R · σ · RT .
A. A. 2! (A36) ˆ where r is the associated skewsymmetric tensor ⎤ ⎡ 0 −r3 r2 ˆ r = ⎣ r3 0 −r1 ⎦ .1 Finite rotation vector. −r2 r1 0 which satisﬁes ˆ r · r = 0. Small rotations can be approximated by (A37) (A38) 1 1 2 ⎣0 ˆ ˆ ˆ R = exp[ˆ ] = I + r + r + · · · ≈ I + r = r 2! 0 ⎡ 0 1 0 ⎤ ⎡ 0 0 ⎦ + ⎣ r3 0 1 −r2 −r3 0 r1 ⎤ r2 −r1 ⎦ .1. r1 If the axis r = r2 and the magnitude r = r of a ﬁnite rotation are known the r3 orthogonal secondorder tensor which describes such rotation can be expressed in the form of an exponential function as follows: ˆ R = exp[ˆ ] = I + r + r 1 2 ˆ r + · · ·. 0 (A39) which do commute ˆ ˆ ˆ ˆ ˆ ˆ ˆ ˆ ˆ ˆ (I + r 1 ) · (I + r 2 ) = (I + r 2 ) · (I + r 1 ) = I + r 1 + r 2 + r 1 · r 2 ≈ I + r 1 + r 2 . ˆ r ·x =r ×x as illustrated in Fig. (A40) .234 Appendix A: Elements of tensor algebra r z y x r r0 = r r Fig.
f Description UMAT for plane strain and axial symmetry for elastic behaviour using ABAQUS stress and strain quantities ABAQUS input ﬁle for single axisymmetric element under uniaxial displacement controlled loading.oup. using ABAQUS *PLASTIC plas_exp_axidisp_aba.f ABAQUS input ﬁle for single axisymmetric element under uniaxial force controlled loading. linear strain hardening plastic behaviour using explicit integration with continuum Jacobian. using ABAQUS *PLASTIC ABAQUS input ﬁle for single axisymmetric element under uniaxial force controlled loading.uk/isbn/0–19–856826–6 .co.inp plasticity_exp code_exp. requiring UMAT elastic. Suitable for large deformations ABAQUS input ﬁle for single axisymmetric element under uniaxial displacement controlled loading. requiring UMAT elastic.inp plas_exp_axiforce_aba_inp ∗ www.f UMAT for plane strain and axial symmetry for elastic.Appendix B: Fortran coding available via the OUP website∗ Directory/ﬁle elasticity elastic.f elas_axidisp.inp elas_axiforce. using ABAQUS stress and strain quantities.
using ABAQUS stress and strain quantities. requiring UMAT code_imp.f Four point bend loading using ABAQUS*PLASTIC.inp plas_imp_axidisp_aba.inp plas_imp_beam.inp ABAQUS input ﬁle for single axisymmetric element under uniaxial displacement controlled loading.inp plas_exp_beam_aba. Suitable for large deformations ABAQUS input ﬁle for single axisymmetric element under uniaxial displacement controlled loading.inp plas_exp_axiforce. requiring mesh ﬁle beam_mesh.f UMAT for plane strain and axial symmetry for elastic. requiring UMAT code_imp. requiring mesh ﬁle beam_mesh.f Four point bend loading using ABAQUS*PLASTIC.inp . requiring UMAT code_exp.inp plas_imp_axiforce.f. using ABAQUS *PLASTIC ABAQUS input ﬁle for single axisymmetric element under uniaxial force controlled loading.f ABAQUS input ﬁle for single axisymmetric element under uniaxial force controlled loading.f. linear strain hardening plastic behaviour using implicit integration with consistent Jacobian. requiring mesh ﬁle beam_mesh.f ABAQUS input ﬁle for single axisymmetric element under uniaxial force controlled loading.236 Appendix B: Fortran coding plas_exp_axidisp.inp plas_imp_beam_aba.inp plas_imp_axiforce_aba_inp plas_imp_axidisp. requiring UMAT code_exp. requiring mesh ﬁle beam_mesh.inp Four point bend loading requiring UMAT code_imp. using ABAQUS *PLASTIC ABAQUS input ﬁle for single axisymmetric element under uniaxial displacement controlled loading.inp Four point bend loading requiring UMAT code_exp.inp plas_exp_beam.inp plasticity_imp code_imp.
Appendix B: Fortran coding 237 spin spin_elastic.f ABAQUS input ﬁle for a single plane strain element under simple shear. requiring UMAT visco_imp.inp visco uni_visco_imp.inp spin_axiforce.f ABAQUS input ﬁle for a single plane strain element under simple shear.f visco_imp. plane strain.inp spin_shear.f spin_axidisp.f ABAQUS input ﬁle for single axisymmetric element under uniaxial force controlled loading.inp .inp spin_shear_aba.f visco_imp_axidisp. and axial symmetry for elastic behaviour using the deformation gradient ABAQUS input ﬁle for single axisymmetric element under uniaxial displacement controlled loading. using ABAQUS stress and strain quantities. and axial symmetry for elastic behaviour using ABAQUS stress and strain quantities UMAT for threedimensional.f uni_visco_exp. linear strain hardening viscoplastic behaviour using implicit integration using the initial tangent stiffness. using ABAQUS*ELASTIC spin_elas_def. plane strain. requiring UMAT spin_elas_def. requiring UMAT spin_elas_def.f Closed form Fortran implicit solution for uniaxial elastoviscoplasticity Closed form Fortran explicit solution for uniaxial elastoviscoplasticity UMAT for plane strain and axial symmetry for elastic.f UMAT for threedimensional. requiring UMAT spin_elas_def. Suitable for large deformations ABAQUS input ﬁle for single axisymmetric element under uniaxial displacement controlled loading.
f Four point bend loading requiring UMAT visco_beam. requiring mesh ﬁle beam_mesh. requiring UMAT visco_imp.238 Appendix B: Fortran coding visco_imp_axiforce.f.inp .inp ABAQUS input ﬁle for single axisymmetric element under uniaxial force controlled loading.inp visco_imp_beam.
167. 143 . 235 ABAQUS input ﬁles 169 accumulated plastic strain 23 Almansi strain 50 angular velocity tensor 63 anisothermal cyclic plasticity 220 antisymmetric 51 Armstrong–Frederick 21 associated ﬂow 18 axial vibration 106.Index ABAQUS 141. 150. 169. 149. 100 elastic–plastic deformation 11 engineering shears 100 equations of motion 90 equilibrium 83 equivalent stress 13 Euler–Lagrange equation 86 explicit ﬁnite element methods 1. 115 B matrix 99 back stress 29 backward Euler integration 146. 161 Bauschinger effect 28 bcc 6 biaxial creep tests 213 bowing 212 Burger’s vector 8 convergence 140 corotational 73 creep 209 CREEP subroutine 202 critical resolved shear stress 7 crystal plasticity 5. 141 explicit integration 136. 50 cavitation 210 central difference method 136 Chaboche 31 coarsening 209 combustion chambers 209 conservation of energy 84 consistency condition 20 consistent Jacobian 153 consistent tangent stiffness 153 consolidation 199 constant strain element 99 constitutive equations 40 continuum tangent stiffness 172 continuum damage 210 continuum Jacobian 172 continuum plasticity 10 continuum spin 61 contracted tensor product 15 deformationenhanced grain growth 190 deformation gradient 48 deformed conﬁguration 48 densiﬁcation rate 201 deviatoric stress 14 diagonalization 56 differentiation of a tensor 231 dilatation 201 direction of plastic ﬂow 19 dislocation bowing 212 dislocation density 210 displacementbased ﬁnite element method 98 divergence of a secondorder tensor 233 double contracted product 15 Duva and Crow 200 dyadic product 230 dynamical path 84 effective plastic strain rate 14 effective stress 13 elastic predictor 147 elastic stiffness matrix 22. 9 crystallographic orientation 3 crystallographic slip 5 cyclic hardening 37 cyclic plasticity 219 calculus of variations 85 cantilever beam 118 Cauchy stress 72 Cauchy–Green tensor 49.
143 implicit ﬁnite element methods 143 implicit implementation for elastoviscoplasticity 161 implicit integration 146 implicit scheme 146 incompressibility 3. 12 incremental nature of plasticity 35 incremental rotation 176 initial tangential stiffness method 140 integration point 126 intergranular cracking 224 intergranular creep fracture 217 intermediate conﬁguration 66 internal variables 40 isoparametric 109 isotropic hardening 150 J2 plasticity 17 Jacobian 150 Jaumann stress rate 76 Jintegral 84 Kinematic hardening 27 kinematics 47 kinetic energy 84 Lagrangian 84 Lame constant 42 large deformation(s) 47 leap frog explicit method 137 mass and spring system 90 mass matrix 105 material objectivity 69 material reference frame 48 material stress rate 78 mean stress 14 microstructural evolution 190 Mohr’s circle 69 momentum balance equations 106 multiaxial creep strain rate 212 multiaxial stress state 20.240 Index fcc 6 ﬁnite element formulation for plasticity 133 ﬁnite element method 83 ﬁnite rotations 57 ﬁrst variation 87 forward integration 145 Gauss quadrature 126 geometric nonlinearity 108 gradient of a ﬁrstorder tensor 232 gradient of a scalar 232 gradient of a secondorder tensor 232 grain boundaries 3 grain growth 190 grain size 190 Hamilton’s principle 84 hardening 23 Helmholtz free energy 210 Hooke’s law 22. 170 hydrostatic stress 14 hysteresis 222 identity tensor 42 implicit 140. 212 multiplicative decomposition 67 necking 187 Newton iteration 148 Newton’s method 148 Newton–Raphson method 141 nickel alloy C263 209 nickelbase superalloy 209 nodal force vector 112 nodal forces 107 nonconservative 84 nonlinear kinematic hardening 32 norm 30 normal grain growth 190 normality hypothesis 18 notched bar test 215 objective stress 72 objectivity 69 onedimensional rod element 99 original conﬁguration 48 orthogonality 58 out of phase loading 225 overstress 38 particle cutting 209 perfect plasticity 11 plane stress 18 plastic correction 146 plastic deformation gradient 66 plastic multiplier 19 plastic strain rate tensor 14 polar decomposition theorem 57 polycrystal 3 polycrystalline 3 porosity 199 porous plasticity 199 potential energy 84 potential function 45 powerlaw creep 44 Prager 29 precipitate coarsening 209 precipitate cutting 209 precipitate spacing 210 predictor–corrector 146 principal coordinates 56 .
150 tangential stiffness matrix 150 tension–torsion 213 tensorial notation 229 tensors 229 thermomechanical fatigue 219 thinning 195 timedependent plasticity 38 timeindependent plasticity 11 Ti–MMCs 205 titanium alloy. 219 rigid body rotation 57 rotation 233 rotation matrix 58 Schmid’s law 7 secondorder tensors 229 second stress invariant 17 semiimplicit integration 160 shape functions 97 shearing 3 simple shear 58 single and multiple element uniaxial tests 171 single element simple shear test 178 skew 51 slip system 5 solvus temperature 209 spin 61 stability of the explicit time stepping 139 static grain growth 190 stationary value 87 stiffness matrix 105 strain decomposition 11 strain measure 49 strainrate sensitivity 39. 42 update of stress 143 upsetting 42 velocity gradient 60 veriﬁcation of the model implementation 171 virtual work 90 viscoplasticity 38 viscous stress 39 Voigt notation 19. 122 Tresca 17 trial stress 147 true strain 53 truss element 99 UMAT 169 undeformed conﬁguration 48 uniaxial loading 14. 186 strain tensor 100 stress space 17 stress tensor 14 stress transformation 72 stress vector 70 stretch 48 stretch ratios 54 superplastic forming 192 Superplasticity 185 tangent stiffness 140. Ti–6Al–4V 189 traction 70 transformation of stress 72 transverse vibration 118.Index 241 principal stresses 13 principle of virtual work 90 quasistatic problems 105 radial return method 146 ratedependent plasticity 38 rate of deformation 60 reaction 130 reversed plasticity 28. 100 volume changes 199 von Mises 17 weak formulation 83 work conjugacy 111 yield criteria 17 yield function 17 yielding 3 .
This page intentionally left blank .
and t/tf = 1.00E – 06 + 8.6.07E + 00 – 9.40E – 01 (b) Thickness strain – 1.00E – 06 + 4.00E – 06 + 5.00E – 06 Plate 1 The simulated superplastically deforming sheet showing the effective plastic strain rate at fractional processing times of t/tf = 0.15E – 01 – 7.00E – 06 + 8.00E – 00 + 2.0 × 10−5 and (b) 1.00E – 06 + 6.00E – 06 + 8.20E + 00 – 1.(a) Strain rate + 0.1.00E – 06 + 4.00E – 06 + 3.95E – 01 – 1.00E – 06 + 7.00E – 06 + 3.15E – 01 – 7.00E – 06 + 6.60E – 01 – 6.07E + 00 – 9.40E – 01 Plate 2 Throughthickness strain ﬁelds at the end of the superplastic forming carried out with target maximum strain rates of (a) 1.05E – 01 – 4.50E – 01 – 2.05E – 01 – 4.20E + 00 – 1.00E – 06 + 5.0 × 10−4 s−1 .00E – 06 (b) Strain rate + 0.00E – 06 (c) Strain rate + 0. .60E – 01 – 6.00E – 00 + 2.50E – 01 – 2.00E – 06 + 6.00E – 06 + 3. t/tf = 0.00E – 06 + 5. (a) Thickness strain – 1.00E – 00 + 2.95E – 01 – 1.0.00E – 06 + 7.00E – 06 + 4.00E – 06 + 7.
18E – 02 + 1.14E – 02 + 1.33E–01 +2.09E–02 +1.73E–02 +9. .19E – 02 (b) Grain sizes +7.20E – 03 +8.62E–01 +1.74E – 03 +8.0 × 10−4 s−1 . and (c) from a surface void count using the micrograph.00E–02 +4. (b) observed in the microstructure.(a) Grain sizes + 1.38E–01 +1.00E+00 +2. (a) Cavitation damage parameter +0.0 × 10−5 and (b) 1.33E – 03 +8.56E–01 +2.17E – 02 + 1.06E – 03 +8.36E–02 +6.85E–01 +2.60E – 03 +8.09E–01 +2.15E – 02 + 1.01E – 03 Plate 3 Average grain size ﬁelds at the end of the superplastic forming carried out with target maximum strain rates of (a) 1.16E – 02 + 1.80E–01 +3.18E – 02 + 1.93E – 03 +8.88E – 03 +9.16E – 02 + 1.00E–01 (b) (c) 2 6 3 4 2 1 2 7 8 7 4 9 3 5 2 2 12 15 9 4 1 2 5 7 6 5 22 17 0 0 2 3 2 8 30 1 2 0 4 0 0 Number of creep voids 0 1–3 4–6 7–9 10–12 13 Plate 4 Creep damage ﬁelds (a) predicted by the model.15E–01 +1.