Computational Methods For Platicity-Souza
Computational Methods For Platicity-Souza
Computational Methods For Platicity-Souza
METHODS FOR
PLASTICITY
THEORY AND APPLICATIONS
EA de Souza Neto
D Perić
DRJ Owen
Civil and Computational Engineering Centre, Swansea University
D Perić
DRJ Owen
Civil and Computational Engineering Centre, Swansea University
Registered office
John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, United
Kingdom
For details of our global editorial offices, for customer services and for information about how to apply
for permission to reuse the copyright material in this book please see our website at www.wiley.com.
The right of the author to be identified as the author of this work has been asserted in accordance with
the Copyright, Designs and Patents Act 1988.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or
transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or
otherwise, except as permitted by the UK Copyright, Designs and Patents Act 1988, without the prior
permission of the publisher.
Wiley also publishes its books in a variety of electronic formats. Some content that appears in print
may not be available in electronic books.
Designations used by companies to distinguish their products are often claimed as trademarks. All
brand names and product names used in this book are trade names, service marks, trademarks or
registered trademarks of their respective owners. The publisher is not associated with any product or
vendor mentioned in this book. This publication is designed to provide accurate and authoritative
information in regard to the subject matter covered. It is sold on the understanding that the publisher is
not engaged in rendering professional services. If professional advice or other expert assistance is
required, the services of a competent professional should be sought.
ISBN 978-0-470-69452-7
Preface xx
2.5 Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.5.1 The derivative map. Directional derivative . . . . . . . . . . . . . 32
2.5.2 Linearisation of a nonlinear function . . . . . . . . . . . . . . . . 32
2.5.3 The gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.5.4 Derivatives of functions of vector and tensor arguments . . . . . . 33
2.5.5 The chain rule . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.5.6 The product rule . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.5.7 The divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.5.8 Useful relations involving the gradient and the divergence . . . . 38
2.6 Linearisation of nonlinear problems . . . . . . . . . . . . . . . . . . . . . 38
2.6.1 The nonlinear problem and its linearised form . . . . . . . . . . . 38
2.6.2 Linearisation in infinite-dimensional functional spaces . . . . . . 39
11 Viscoplasticity 435
11.1 Viscoplasticity: phenomenological aspects . . . . . . . . . . . . . . . . . 436
11.2 One-dimensional viscoplasticity model . . . . . . . . . . . . . . . . . . . 437
11.2.1 Elastoplastic decomposition of the axial strain . . . . . . . . . . . 437
11.2.2 The elastic law . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
11.2.3 The yield function and the elastic domain . . . . . . . . . . . . . 438
11.2.4 Viscoplastic flow rule . . . . . . . . . . . . . . . . . . . . . . . . 438
11.2.5 Hardening law . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
11.2.6 Summary of the model . . . . . . . . . . . . . . . . . . . . . . . 439
11.2.7 Some simple analytical solutions . . . . . . . . . . . . . . . . . . 439
11.3 A von Mises-based multidimensional model . . . . . . . . . . . . . . . . 445
11.3.1 A von Mises-type viscoplastic model with isotropic strain
hardening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
11.3.2 Alternative plastic strain rate definitions . . . . . . . . . . . . . . 447
11.3.3 Other isotropic and kinematic hardening laws . . . . . . . . . . . 448
11.3.4 Viscoplastic models without a yield surface . . . . . . . . . . . . 448
11.4 General viscoplastic constitutive model . . . . . . . . . . . . . . . . . . . 450
11.4.1 Relation to the general continuum constitutive theory . . . . . . . 450
11.4.2 Potential structure and dissipation inequality . . . . . . . . . . . 451
11.4.3 Rate-independent plasticity as a limit case . . . . . . . . . . . . . 452
11.5 General numerical framework . . . . . . . . . . . . . . . . . . . . . . . . 454
11.5.1 A general implicit integration algorithm . . . . . . . . . . . . . . 454
11.5.2 Alternative Euler-based algorithms . . . . . . . . . . . . . . . . . 457
11.5.3 General consistent tangent operator . . . . . . . . . . . . . . . . 458
11.6 Application: computational implementation of a von Mises-based model . 460
11.6.1 Integration algorithm . . . . . . . . . . . . . . . . . . . . . . . . 460
11.6.2 Iso-error maps . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
11.6.3 Consistent tangent operator . . . . . . . . . . . . . . . . . . . . . 464
11.6.4 Perzyna-type model implementation . . . . . . . . . . . . . . . . 466
11.7 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
11.7.1 Double-notched tensile specimen . . . . . . . . . . . . . . . . . 467
11.7.2 Plane stress: stretching of a perforated plate . . . . . . . . . . . . 469
Appendices 729
References 765
Index 783
PREFACE
T HE purpose of this text is to describe in detail numerical techniques used in small and
large strain finite element analysis of elastic and inelastic solids. Attention is focused
on the derivation and description of various constitutive models – based on phenomenolog-
ical hyperelasticity, elastoplasticity and elasto-viscoplasticity – together with the relevant
numerical procedures and the practical issues arising in their computer implementation
within a quasi-static finite element scheme. Many of the techniques discussed in the text are
incorporated in the FORTRAN program, named HYPLAS, which accompanies this book and
can be found at www.wiley.com/go/desouzaneto. This computer program has been specially
written to illustrate the practical implementation of such techniques. We make no pretence
that the text provides a complete account of the topics considered but rather, we see it as an
attempt to present a reasonable balance of theory and numerical procedures used in the finite
element simulation of the nonlinear mechanical behaviour of solids.
When we embarked on the project of writing this text, our initial idea was to produce
a rather concise book – based primarily on our own research experience – whose bulk
would consist of the description of numerical algorithms required for the finite element
implementation of small and large strain plasticity models. As the manuscript began to take
shape, it soon became clear that a book designed as such would be most appropriate to those
already involved in research on computational plasticity or closely related areas, being of
little use to those willing to learn computational methods in plasticity from a fundamental
level. A substantial amount of background reading from other sources would be required for
readers unfamiliar with topics such as basic elastoplasticity theory, tensor analysis, nonlinear
continuum mechanics – particularly nonlinear kinematics – finite hyperelasticity and general
dissipative constitutive theory of solids. Our initial plan was then gradually abandoned as
we chose to make the text more self-contained by incorporating a considerable amount of
basic theory. Also, while writing the manuscript, we decided to add more advanced (and very
exciting) topics such as damage mechanics, anisotropic plasticity and the treatment of finite
strain single crystal plasticity. Following this route, our task took at least three times as long
to complete and the book grew to about twice the size as originally planned. There remains
plenty of interesting material we would like to have included but cannot due to constraints
of time and space. We are certainly far more satisfied with the text now than with its early
versions, but we do not believe our final product to be optimal in any sense. We merely offer
it to fill a gap in the existing literature, hoping that the reader will benefit from it in some way.
The text is arranged in three main parts. Part One presents some basic material of relevance
to the subject matter of the book. It includes an overview of elementary tensor analysis,
continuum mechanics and thermodynamics, the finite element method in quasi-static non-
linear solid mechanics and a brief description of the computer program HYPLAS. Part Two
xxii PREFACE
deals with small strain problems. It introduces the mathematical theory of infinitesimal
plasticity as well as the relevant numerical procedures for the implementation of plasticity
models within a finite element environment. Both rate-independent (elastoplastic) and rate-
dependent (elasto-viscoplastic) theories are addressed and some advanced models, including
anisotropic plasticity and ductile damage are also covered. Finally, in Part Three we focus
on large strain problems. The theory of finite hyperelasticity is reviewed first together with
details of its finite element implementation. This is followed by an introduction to large strain
plasticity. Hyperelastic-based theories with multiplicative elastoplastic kinematics as well as
hypoelastic-based models are discussed, together with relevant numerical procedures for their
treatment. The discussion on finite plasticity and its finite element implementation culminates
with a description of techniques for single crystal plasticity. Finite element techniques for
large-strain near-incompressibility are also addressed.
We are indebted to many people for their direct or indirect contribution to this text.
This preface would not be complete without the due acknowledgement of this fact and a
record of our sincere gratitude to the following: to J.M.P. Macedo for the numerous valuable
suggestions during the design of the program HYPLAS at the very early stages of this project;
to R. Billardon for the many enlightening discussions on damage modelling; to R.A. Feijóo
and E. Taroco for the fruitful discussions held on many occasions over a long period of time;
to M. Dutko for producing some of the numerical results reported; to Y.T. Feng for helpful
discussions on the arc-length method; to F.M. Andade Pires for his key contribution to the
development of F-bar-Patch elements, for producing the related figures presented and for
thoroughly reviewing early versions of the manuscript; to P.H. Saksono for his involvement
in the production of isoerror maps; to A. Orlando for literally ‘scanning’ through key parts
of the text to find inconsistencies of any kind; to L. Driemeier, W. Dettmer, M. Vaz Jr,
M.C. Lobão, M. Partovi, D.C.D. Speirs, D.D. Somer, E. Saavedra, A.J.C. Molina, S. Giusti
and P.J. Blanco for carefully reviewing various parts of the manuscript, spotting hard-to-find
mistakes and making several important suggestions for improvement. Last, but not least, to
our late colleague and friend Mike Crisfield, for the numerous illuminating and passionate
discussions (often held on the beach or late in the bar) on many topics addressed in the book.
EA de Souza Neto
D Perić
DRJ Owen
Swansea
Part One
Basic concepts
1 INTRODUCTION
O VER the last four decades, the use of computational techniques based on the Finite
Element Method has become a firmly established practice in the numerical solution of
nonlinear solid mechanics problems both in academia and industry.
In their early days, these techniques were largely limited to infinitesimal deformation and
strain problems with the main complexity arising from the nonlinear constitutive characteriza-
tion of the underlying material by means of basic elastoplastic or elasto-viscoplastic theories.
Applications were mostly confined to the modelling of the behaviour of solids in conventional
areas of engineering and analyses were carried out on crude, user-unfriendly software that
typically required highly specialized users. Since those days, this area of solid mechanics
– generally known as computational plasticity – has experienced dramatic developments.
Fuelled by the steady increase in computing power at decreasing costs together with the
continuous industrial demand for accurate models of solids, the evolution of computational
plasticity techniques have made possible the development of refined software packages
with a considerable degree of automation that are today routinely employed by an ever-
increasing number of engineers and scientists. The variety of practical problems of interest
to which such techniques are currently applied with acceptable levels of predictive capability
is very wide. They range from traditional engineering applications, such as stress analysis
in structures, soil and rock mechanics, to the simulation of manufacturing processes such as
metal forming. Also included are much less conventional applications, such as the simulation
of food processing, mining operations and biological tissue behaviour. Many such problems
are characterised by extremely large straining and material behaviour often described by
means of rather complex constitutive equations.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
4 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
1.1.1. READERSHIP
This book is intended for graduate students, research engineers and scientists working in the
field of computational continuum mechanics. The text requires a basic knowledge of solid
mechanics – especially the theory of linear elasticity – as well as the Finite Element Method
and numerical procedures for the approximate solution of ordinary differential equations.
An elementary understanding of vector and tensor calculus is also very helpful. Readers
wishing to follow the computer implementation of the procedures described in the text should,
in addition, be familiar at a fairly basic level with the FORTRAN computer programming
language. It is worth remarking here that the choice of the FORTRAN language is motivated
mainly by the following:
(b) the suitability of procedural languages for codes with relatively low level of complexity,
such as HYPLAS. In the present case, the use of more advanced programming concepts
(e.g. object-oriented programming) could add a further difficulty in the learning of the
essential concepts the HYPLAS code is meant to convey;
(c) its relative clarity in the coding of short algorithmic procedures such as those arising
typically in the implementation of elastic and inelastic material models – the main
subject of this book.
1.2. Layout
In line with the above aims, the book has been divided into three parts as follows.
INTRODUCTION 5
• Part One: Basic concepts. In this part we introduce concepts of fundamental relevance
to the applications presented in Parts Two and Three. The following material is covered:
• Part Two: Small strains. Here, the theory of infinitesimal plasticity is introduced
together with the relevant numerical procedures used for its implementation into a finite
element environment. A relatively wide range of models is presented, including both
rate-independent (elastoplastic) and rate-dependent (elasto-viscoplastic) theories. The
following main topics are considered:
• Part Three: Large strains. This part focuses on finite strain hyperelasticity and
elastoplasticity problems. The models discussed here, as well as their computational
implementation, are obviously more complex than those of Part Two. Their complexity
stems partly from the finite strain kinematics. Thus, to follow Part Three, a sound
knowledge of the kinematics of finite deformations discussed in Chapter 3 (in Part One)
is essential. The following topics are addressed:
The material has been organised into sixteen chapters and four appendices. These will now
be briefly described. The remainder of Chapter 1 discusses the general scheme of notation
adopted in the book.
Chapter 2 contains an introduction to elementary tensor analysis. In particular, the material
is presented mainly in intrinsic (or compact) tensor notation – which is heavily relied upon
thoughout the book.
Chapter 3 provides an introdution to the mechanics and thermodynamics of continuous
media. The material presented here covers the kinematics of deformation, balance laws and
constitutive theory. These topics are essential for an in-depth understanding of the theories
discussed in later chapters.
Chapter 4 shows the application of the Finite Element Method to the solution of problems
in quasi-static nonlinear solid mechanics. A generic dissipative constitutive model, initially
presented in Chapter 3, is used as the underlying material model.
6 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Chapter 5 describes the general structure of the program HYPLAS, where many of the
techniques discussed in the book are implemented. We remark that the program description
is rather concise. Further familiarisation with the program will require the reader to follow
the comments in the FORTRAN source code together with the cross-referencing of the main
procedures with their description in the book. This is probably more relevant to those wishing
to use the HYPLAS program for research and development purposes.
Chapter 6 is devoted to the mathematical theory of infinitesimal plasticity. The main
concepts associated with phenomenological time-independent plasticity are introduced here.
The basic yield criteria of Tresca, von Mises, Mohr–Coulomb and Drucker–Prager are
reviewed, together with the most popular plastic flow rules and hardening laws.
In Chapter 7, we introduce the essential numerical methods required in the finite element
solution of initial boundary value problems with elastoplastic underlying material models.
Applications of the von Mises model with both isotropic and mixed isotropic/kinematic
hardening are described in detail. The most relevant subroutines of the program HYPLAS are
also listed and explained in detail.
Chapter 8 focuses on the detailed description of the implementation of the basic plasticity
models based on the Tresca, Mohr–Coulomb and Drucker–Prager yield criteria. Again, the
relevant subroutines of HYPLAS are listed and explained in some detail.
In Chapter 9 we describe the numerical treatment of plasticity models under plane stress
conditions. Different options are considered and their relative merits and limitations are
discussed. Parts of source code are also included to illustrate some of the most important
programming aspects. The application of the concepts introduced here to other stress-
constrained states is briefly outlined at the end of the chapter.
In Chapter 10 advanced elastoplasticity models are considered. Here we describe the
computational implementation of a modified Cam-Clay model for soils, a capped Drucker–
Prager model for geomaterials and the Hill, Hoffman and Barlat–Lian anisotropic models for
metals. The numerical techniques required for the implementation of such models are mere
specialisations of the procedures already discussed in Chapters 7 and 8. However, due to
the inherent complexity of the models treated in this chapter, their actual implementation is
generally more intricate than those of the basic models.
Chapter 11 begins with an introduction to elasto-viscoplasticity theory within the con-
stitutive framework for dissipative materials described in Chapter 3. The (rate-independent)
plasticity theory is then obtained as a limiting case of viscoplasticity. The numerical methods
for a generic viscoplastic model are described, following closely the procedures applied
earlier in elastoplasticity. Application of the methodology to von Mises criterion-based
viscoplastic models is described in detail.
In Chapter 12 we discuss continuum damage mechanics – the branch of Continuum Solid
Mechanics devoted to the modelling of the progressive material deterioration that precedes
the onset of macroscopic fracturing. Some elastoplastic damage models are reviewed and
their implementation, with the relevant computational issues, is addressed in detail.
Chapter 13 introduces finite strain hyperelasticity. The basic theory is reviewed and some
of the most popular isotropic models are presented. The finite element implementation of the
Odgen model is discussed in detail with relevant excerpts of HYPLAS source code included. In
addition, the modelling of the so-called Mullins dissipative effect by means of a hyperelastic-
damage theory is addressed at the end of the chapter. This concept is closely related to those
already discussed in Chapter 12 for ductile elastoplastic damage.
INTRODUCTION 7
At the same time, we have tried to keep the notation in line with what is generally adopted
in the present subject areas. Unfortunately, these two goals often conflict so that, in many
cases, we choose to adopt the notation that is more widely accepted instead of adhering to
the use of specific fonts for specific mathematical entities. Whenever such exceptions occur,
their meaning should either be clear from the context or will be explicitly mentioned the first
time they appear.
F Deformation gradient
e
F Elastic deformation gradient
p
F Plastic deformation gradient
ext
f Global (finite element) external force vector
ext
f(e) External force vector of element e
int
f Global (finite element) internal force vector
int
f(e) Internal force vector of element e
G Virtual work functional; shear modulus
G Discrete (finite element) full gradient operator
H Hardening modulus
H Generalised hardening modulus
I1 , I2 , I3 Principal invariants of a tensor
I Fourth-order identity tensor: Iijkl = δik δjl
IS Fourth-order symmetric identity tensor: Iijkl = 12 (δik δjl + δil δjk )
Id Deviatoric projection tensor: Id ≡ IS − 13 I ⊗ I
I Second-order identity tensor
IS Array representation of IS
i Array representation of I
J Jacobian of the deformation map: J ≡ det F
J2 , J3 Stress deviator invariants
J Generalised viscoplastic hardening constitutive function
K Bulk modulus
KT Global tangent stiffness matrix
(e)
KT Tangent stiffness matrix of element e
K Set of kinematically admissible displacements
L Velocity gradient
e
L Elastic velocity gradient
p
L Plastic velocity gradient
α
m Unit vector normal to the slip plane α of a single crystal
N Plastic flow vector
N̄ Unit plastic flow vector: N̄ ≡ N/N
O The orthogonal group
+
O The rotation (proper orthogonal) group
INTRODUCTION 11
β Back-stress tensor
γ̇ Plastic multiplier
δij Krönecker delta
ε Strain tensor; also Eulerian logarithmic strain when
under large strains
εe Elastic strain tensor; also elastic Eulerian logarithmic
strain when under large strains
εp Plastic strain tensor
ε, ε , ε
e p
Array representation of ε, εe and εp , respectively
ε, εe , εp Axial total, elastic and plastic strain in one-dimensional
models
ε̄p Effective (or accumulated) plastic strain
η Virtual displacement field; relative stress tensor in kinematic
hardening plasticity models
κ Isotropic hardening thermodynamical force
λ One of the Lamé constants of linear elasticity; axial stretch;
load factor in proportional loading
λe , λp Elastic and plastic axial stretch
λi , λei , λpi Total, elastic and plastic principal stretches
µ One of the Lamé constants of linear elasticity
µp Ogden hyperelastic constants (p = 1, . . . , N ) for a model
with N terms in the Ogden strain-energy function series
ν Poisson ratio
Ξ Dissipation potential
ξ Isoparametric coordinates of a finite element
ρ Mass density
ρ̄ Reference mass density
σ Cauchy stress tensor
σ Axial stress in one-dimensional models
σi Principal Cauchy stress
σy Yield stress (uniaxial yield stress for the conventional
von Mises and Tresca models)
σy 0 Initial yield stress
σ Array representation of σ
τ Kirchhoff stress tensor
τi Principal Kirchhoff stress
INTRODUCTION 13
We remark that subscripts are not employed exclusively in connection with indicial notation.
Different connotations are assigned to subscripts throughout the text and the actual meaning
of a particular subscript should be clear from the context. For example, in the context of
incremental numerical procedures, subscripts may indicate the relevant increment number. In
the expression
∆ε = εn+1 − εn ,
the subscripts n and n + 1 refer to the values of ε, respectively, at the end of increments n
and n + 1.
Superscripts
Superscripts are also used extensively throughout the text. The meaning of a particular
superscript will be stated the first time it appears in the text and should be clear from the
context thereafter.
14 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
T HIS chapter introduces the notation and reviews some fundamentals of vector and
tensor calculus which are extensively employed in this book. Throughout this text,
preference is given to the use of intrinsic (or compact) tensor notation where no indices
are used to represent mathematical entities. However, in many of the definitions introduced
in this chapter, indicial notation is also used. This will allow readers not yet familiar with
compact notation to associate compactly written entities and operations with their indicial
forms, which will be expressed exclusively in terms of Cartesian coordinate systems. We
note that the use of Cartesian, rather than curvilinear, coordinates for indicial representation
is sufficiently general for the applications considered in this book. In the subsequent chapters,
the use of indicial notation will be much less frequent. Readers who are familiar with tensor
analysis and, in particular, the use of compact notation, may comfortably skip this chapter.
We remark that no proofs are given to most relations presented in this chapter. Readers
interested in such proofs and a more in-depth treatment of the subject are referred to other
textbooks such as Gurtin (1981).
2.1. Vectors
Let E be an n-dimensional Euclidean space and let U be the space of n-dimensional vectors
associated with E. Points of E and vectors U satisfy the basic rules of vector algebra, with
which we assume the reader to be familiar.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
18 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
u · v = 0. (2.4)
ei · ej = δij , (2.5)
where
1 if i = j
δij = (2.6)
0 if i
= j
u = u 1 e1 + u 2 e2 + · · · + u n en = u i ei , (2.7)
where
u i = u · ei , i = 1, 2, . . . , n (2.8)
are the Cartesian components of u relative to the basis {ei }. Any vector of U is uniquely
defined by its components relative to a given basis. This allows us to represent any vector u
as a single column matrix, denoted [u], of components
u1
u2
[u] = . . (2.9)
..
un
xi = (x − x0 ) · ei . (2.12)
ELEMENTS OF TENSOR ANALYSIS 19
T:U→U
(S + T )u = S u + T u
(2.14)
(α S)u = α (S u),
where α ∈ R. In addition, the zero tensor, 0, and the identity tensor, I, are, respectively, the
tensors that satisfy
0u = o
(2.15)
Iu=u
∀ u ∈ U.
The product of two tensors S and T is the tensor ST defined by
ST u = S (T u). (2.16)
In general,
ST
= TS. (2.17)
If ST = TS, then S and T are said to commute.
T u · v = u · T T v, ∀ u, v ∈ U. (2.18)
If
T = TT, (2.19)
then T is said to be symmetric. If
T = −T T , (2.20)
then T is said to be skew symmetric (or, simply, skew).
Any tensor T can be decomposed as the sum
Basic properties
The following basic properties involving the transpose, skew and symmetric parts of a tensor
hold:
(i) (S + T )T = ST + T T .
(ii) (S T )T = T T ST .
T
(iii) (T T ) = T.
skew(T ) = 0, sym(T ) = T.
u ⊗ v,
is the tensor that maps each vector w into the vector (v · w)u:
(u ⊗ v) w = (v · w) u. (2.24)
(i) u ⊗ (v + w) = u ⊗ v + u ⊗ w.
(ii) (u ⊗ v)T = v ⊗ u.
(iv) ei ⊗ ei = I.
(v) S (u ⊗ v) = (S u) ⊗ v.
where
Tij = ei · T ej (2.26)
are the Cartesian components of T. Note that in (2.25) no summation is implied over the
index n.
Any second tensor is uniquely defined by its Cartesian components. Thus, by arranging
the components Tij in a matrix, we may have the following matrix representation for T :
T11 T12 ··· T1n
T21 T22 ··· T2n
[T ] = . .. .. . (2.27)
. ..
. . . .
Tn1 Tn2 · · · Tnn
Thus, the array [v] of Cartesian components of v is obtained from the matrix-vector product
v1 T11 T12 ··· T1n u1
v2 T21 T22 ··· T2n u2
[v ] ≡ . = . .. .. . . (2.31)
.. .. ..
. .
. . .
vn Tn1 Tn2 · · · Tnn un
It can be easily proved that the Cartesian components TijT of the transpose T T of a tensor
T are given by
TijT = Tji . (2.32)
22 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Basic properties
The following basic properties involving the internal product of tensors hold for any tensors
R, S, T and vectors s, t, u, v:
(i) I : T = tr T.
(ii) R : (S T ) = (ST R) : T = (R T T ) : S.
(iii) u · S v = S : (u ⊗ v).
(iv) (s ⊗ t) : (u ⊗ v) = (s · u)(t · v).
(v) Tij = T : (ei ⊗ ej ).
(vi) (u ⊗ v)ij = (u ⊗ v) : (ei ⊗ ej ) = ui vj .
Some useful relations involving the cross product are listed in the following:
(i) u × v = −v × u.
(ii) (u × v) · w = (v × w) · u = (w × u) · v.
(iii) u × u = o.
(iv) For any tensor T and a set {u, v, w} of linearly independent vectors,
(T u × T v) · T w
det T = . (2.54)
(u × v) · w
(v) For any skew tensor W , there is a unique vector w, called the axial vector of W ,
such that
W u = w × u. (2.55)
The space of all vectors u satisfying the above equation is called the characteristic space of
T corresponding to ω. The following properties hold:
Spectral theorem
Let S be a symmetric tensor. Then S admits the representation
n
S= si ei ⊗ ei , (2.59)
i=1
Eigenprojections
Alternatively, with p ≤ n defined as the number of distinct eigenvalues of S, we may write
p
S= si Ei , (2.61)
i=1
where the symmetric tensors {Ei } are called the eigenprojections of S. Each eigenprojection
Ei is the orthogonal projection operator on the characteristic space of S associated with si .
The eigenprojections have the property
p
I= Ei , (2.62)
i=1
Two-dimensional space
In the two-dimensional space, det(S − α I ) can be expressed as
det(S − α I ) = α2 − α I1 + I2 , (2.67)
I1 (S) ≡ tr S = Sii
(2.68)
I2 (S) ≡ det S = S11 S22 − S12 S21 .
s2i − si I1 + I2 = 0. (2.69)
The eigenvalues si are the solutions to this quadratic equation. If S is symmetric, then its
principal invariants can be expressed in terms of its eigenvalues as
I1 = s1 + s2
(2.70)
I2 = s1 s2 .
Three-dimensional space
In the three-dimensional space, det(S − α I ) has the representation
where I1 , I2 and I3 are the principal invariants of S, which are now defined by
I1 (S) ≡ tr S = Sii
I2 (S) ≡ 12 [(tr S)2 − tr(S2 )] = 12 (Sii Sjj − Sij Sji ) (2.72)
I3 (S) ≡ det S = 1
6 ijk pqr Sip Sjq Skr .
The characteristic equation (now a cubic equation), whose solution is the set of eigenvalues
of S, reads
−s3i + s2i I1 − si I2 + I3 = 0. (2.73)
If S is symmetric, then we have
I1 = s1 + s2 + s3
I2 = s1 s2 + s2 s3 + s1 s3 (2.74)
I3 = s1 s2 s3 .
28 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Symmetry
We shall call symmetric any fourth-order tensor that satisfies
S : T : U = (T : S) : U , (2.87)
for any second-order tensors S and U . This definition is analogous to that of symmetric
second-order tensors. The Cartesian components of symmetric fourth-order tensors satisfy
the major symmetries
Tijkl = Tklij . (2.88)
It should be noted that other symmetries are possible in fourth-order tensors. If symmetry
occurs in the last two indices, i.e. if
Tijkl = Tijlk , (2.89)
the tensor has the properties
T : S = T : ST , S : T = (S : T)T , (2.90)
for any S. If it is symmetric in the first two indices,
Tijkl = Tjikl , (2.91)
then,
T : S = (T : S)T , S : T = ST : T. (2.92)
‡ Fourth-order tensors are represented in this text by bold face upright sans serif fonts.
30 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
with R a rotation. The components T∗ijkl of a tensor T relative to the basis defined by {e∗i }
are given by
T∗ijkl = Rmi Rnj Rpk Rql Tmnpq , (2.94)
where Tmnpq are the components relative to {ei }.
for all u ∈ U. The definition of contraction operations is completely analogous to that seen
above for fourth-order tensors.
αI,
U = α I + β IT + γ (I ⊗ I ), (2.99)
where α, β and γ are scalars. The tensor I is called the fourth-order identity, given in
component form as
Iijkl = δik δjl . (2.100)
For any second-order tensor T, the fourth-order identity satisfies
I : T = T : I = T, (2.101)
I : T = T : I = T. (2.102)
The tensor IT is the transposition tensor. It maps any second-order tensor onto its trans-
pose, i.e.
IT : S = S : IT = ST , (2.103)
for any S. The components of IT are
(I ⊗ I ) : T = (tr T ) I. (2.106)
Another important isotropic tensor that frequently appears in continuum mechanics is the
tensor defined as
IS = 12 ( I + IT ). (2.107)
This tensor maps any second-order tensor into its symmetric part, i.e.
IS : T = T : IS = sym(T ). (2.108)
2.5. Differentiation
2.5.1. THE DERIVATIVE MAP. DIRECTIONAL DERIVATIVE
Let X and Y be finite-dimensional normed vector spaces and let a function Y be defined as
Y : D ⊂ X → Y, where the domain D of Y is an open subset of X.
The function Y is said to be differentiable at an argument X 0 ∈ D if there exists a linear
transformation DY(X 0 ) : X → Y such that as U ∈ X approaches zero (the zero element
of X ),
Y(X 0 + U ) = Y(X 0 ) + DY(X 0 ) [U ] + o(U ), (2.111)
where
DY(X 0 ) [U ]
denotes a linear transformation on U , represented in abstract notation, and
o(U )
o(U )
lim = 0. (2.112)
U→0 U
If the linear map DY(X 0 ) exists, it is unique and is called the derivative of Y at X 0 . The
derivative satisfies
1 d
DY(X 0 ) [U ] = lim [Y(X 0 + U ) − Y(X 0 )] = Y(X 0 + U ) , (2.113)
→0 d =0
∈R
for each U ∈ X. For a given U , the term DY(X 0 ) [U ] is called the directional derivative of
Y (at X 0 ) in the direction of U .
is the called the linearisation of Y about X 0 , that is, it is the linear approximation to Y
at X 0 . Note by observing (2.111) that the function defined in (2.114) corresponds indeed to
the approximation to Y that ignores all higher-order (nonlinear) terms in U .
We remark that throughout this text we shall use the term derivative also as a synonym for
gradient. We will also use the notation
dY
dX
to denote the gradient and when Y is a function of two or more arguments, we may use the
notation
∂Y
∂X
to emphasise that the derivative is taken with respect to X having the other function arguments
as parameters.
y(x) ≡ x2 . (2.116)
|o(u)| u2
lim = lim = 0. (2.119)
u→0 |u| u→0 |u|
In addition, we have
Dy(x0 ) [u] = 2x0 u. (2.120)
Thus, by comparing the above expression with (2.115), the gradient of y can be promptly
identified as
∇y(x0 ) = 2x0 , (2.121)
and the product ‘∗’ is identified with the standard product between scalars. With the above
and (2.114), we find that the linearisation of y about x0 in the present case reads
y
y(x) = x2
linearisation of
y(x) about x 0
tan--1 2x 0
x0 x
y(x) ≡ x · x = xi xi . (2.123)
The generic product operation indicated in (2.115) becomes the internal product between
vectors in the present case. We may alternatively use the notation
dy
= 2x, (2.127)
dx
or, in component form,
dy dy
= = 2xi . (2.128)
dx i dxi
We have
y(X + U ) = X : X + 2X : U + U : U
= y(X) + Dy(X) [U ] + o(U ). (2.130)
Y (X + U ) = XX + XU + U X + U U
= Y (X) + DY (X) [U ] + o(U ). (2.135)
The derivative in this case is a linear mapping between second-order tensors and can be
identified as the fourth-order tensor ∇Y (X) that satisfies
DY (x) [u] = (a ⊗ I) · u
d (2.141)
(a ⊗ x) = a ⊗ I.
dx
(v) Y (x) ≡ x ⊗ x,
DY (x) [u] = (I ⊗ x + x ⊗ I) · u
d (2.142)
(x ⊗ x) = I ⊗ x + x ⊗ I.
dx
(vi) Y (X) ≡ X,
DY (X) [U ] = U
d (2.143)
X = I.
dX
(vii) Y (X) ≡ X −1 ,
DY (X) [U ] = −X −1 U X −1 = D : U ; −1 −1
Dijkl = −Xik Xlj
d (2.144)
(X −1 ) = D.
dX
Y(X) = W(Z(X))
Z : D ⊂ X → Z and W : C ⊂ Z → Y.
for all U ∈ X.
ELEMENTS OF TENSOR ANALYSIS 37
where the symbol ‘∗’ denotes a generic product between functions W : D ⊂ X → W and
Z : D ⊂ X → Z. If W and Z are vectors, for instance, this product could be the internal or
tensor product between vectors. If W is a fourth-order tensor and Z is a second-order tensor,
‘∗’ could be the double contraction operation, and so on. If W and Z are differentiable at an
argument X, then so is Y and
Now let T be a smooth second-order tensor field on D. The divergence of T is the vector
field, denoted div T, that satisfies
∂Tij
(div T )i = . (2.147)
∂xj
Let B ⊂ E be a closed region with piecewise smooth boundary ∂B and let α, v and T be
smooth scalar, vector and tensor-valued fields on B, respectively. Then,
α n da = ∇α dv,
∂B B
v · n da = div v dv, (2.148)
∂B B
T n da = div T dv,
∂B B
Let α and T be, respectively, smooth scalar and tensor fields on D ⊂ E and let u and v be
smooth vector fields on D. The following useful relations hold:
The linearisation of the problem defined by equation (2.149) about an arbitrary point X 0 ∈ D
at which Y is differentiable consists in finding U ∈ D such that
y
nonlinear
function y(x)
linearisation of
y(x) about x 0
x0 x
solution of the
nonlinear problem solution of the
linearised problem
y(x) = 0. (2.151)
This is obviously a particularisation of the general nonlinear problem (2.149). In the present
case, the generic linearised problem (2.150) takes the form of the following scalar equation
for u at an arbitrary argument x0 :
referred to Marsden and Hughes (1983) where the concept of linearisation is introduced in the context of mappings
between Banach spaces.
40 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
A simple example
Let X now be a space of sufficiently smooth¶ functions x : R n → R. We define a functional
y : X → R as
y(x) ≡ sin(x(p)) dp, (2.153)
Ω
where the directional derivative Dy(x0 ) [u] is now a linear transformation on the function
u ∈ X and can be determined by direct generalisation of (2.113):
d
Dy(x0 ) [u] = y(x0 + u)
d
=0
d
= sin(x0 (p) + u(p)) dp
d
Ω =0
From the above, the linearisation of the functional (2.153) at x0 is then established as
l(u) = sin(x0 (p)) dp + cos(x0 (p)) u(p) dp. (2.156)
Ω Ω
¶ To avoid a precise statement of regularity properties of functions, we frequently use the term sufficiently smooth
in the present text, meaning that functions have a sufficient degree of regularity so that all operations in which they
are involved are properly defined.
3 ELEMENTS OF CONTINUUM
MECHANICS AND
THERMODYNAMICS
T HIS chapter reviews some basic concepts of mechanics and thermodynamics of contin-
uous media. The definitions and notation introduced will be systematically employed
throughout the subsequent chapters of this book. The material presented here is well
established in the continuum mechanics literature and an effort has been made to follow
the notation and nomenclature in use in standard textbooks (Billington and Tate, 1981; Bonet
and Wood, 1997; Ciarlet, 1988; Gurtin, 1981; Lemaitre and Chaboche, 1990; Ogden, 1984;
Spencer, 1980; Truesdell and Noll, 1965).
ϕ:Ω→E
x = ϕ(p) (3.1)
where the particle is positioned in the deformed configuration of B. The region of E occupied
by B in its deformed configuration will be denoted
ϕ(Ω).
x = p + u(p). (3.3)
† For convenience, material particles of B will be identified with their positions in the reference configuration
of B.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
42 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
M:
:
u( p) x=M( p)
p
e2
e1
e3
rigid rotation
rigid translation M:
fixed point
:
: q
x
x
u
p p M:
c(t) M:
t
1
reference
configuration
: xt 1
)
u( p, t1 M:
t2
u( p, t2 ) xt2
so that for each time t, the map ϕ(·, t) is a deformation of B. The deformation map at time t
will be also denoted ϕt . During the motion ϕ, the position x of a material particle p at time
t is given by
x = ϕ(p, t). (3.7)
Similarly,
ϕ(Ω, t)
will denote the region of E occupied by the body B at time t. In terms of the displacement
field the motion is expressed as
for a fixed material point p, describes the trajectory of p during the motion of B.
During a motion ϕ, the velocity of a material particle p is defined by
∂ϕ(p, t)
ẋ(p, t) = . (3.10)
∂t
Since at each time t the map ϕ(·, t) is one-to-one (and hence invertible) by assumption,
material points can be expressed in terms of the place they occupy at a time t as
The map ϕ−1 is called the reference map. Using the reference map, one may define the
function
v(x, t) ≡ ẋ(ϕ−1 (x, t), t). (3.12)
The field v is called the spatial velocity and gives the velocity of the material particle
positioned at x at time t.
44 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
v(x)
x
y)
w x (x Ŧ y v(
)
v( y)
y w
A rigid motion of B is a motion for which, at each time t, the map ϕ(·, t) is a rigid
deformation. A motion ϕ is rigid if and only if at each time t, the spatial velocity v admits
the representation
v(x, t) = v(y, t) + W (t) (x − y) (3.13)
for all x, y ∈ ϕ(Ω, t), with W (t) a skew tensor. The velocity at x is given as the sum of a
uniform velocity v(y, t) and a superimposed rotation about the line that passes through y and
is parallel to the axial vector associated to the skew tensor W . By denoting w(t) the axial
vector of W (t), the velocity field above can be re-written as
which is the standard formula for the velocity field of classical rigid-body dynamics. The
vector w(t) is called the angular velocity of the body. The rigid velocity field is schematically
illustrated in Figure 3.4.
It should be noted that any field associated with a motion of B can be expressed as a
function of time and material particles or spatial position. A material (spatial) field does
ELEMENTS OF CONTINUUM MECHANICS AND THERMODYNAMICS 45
time t
x
T=a+b(x Ŧtv)+c t
time 0
tv =a+b p+c t
p
e2
e1
not necessarily represent a quantity physically associated with the reference (deformed)
configuration of the body.
Example 3.1.1. Consider, for instance, the rectangular body of Figure 3.5 subjected to the
rigid translation:
x = ϕ(p, t) ≡ p + t v,
with constant velocity v. Assume that, during the motion ϕ, the temperature field of the
body in question is linearly distributed along its longitudinal axis and varies uniformly
throughout the body at a constant rate. Taking the initial configuration (at t = 0) as the
reference configuration (and, therefore, labelling material particles of the body with their
position p at time 0), the material description of this temperature field reads
θm (p, t) = a + b p1 + c t,
where a, b and c are constants. In view of the assumed motion ϕ, the spatial description of
the same field is given by
Note that, in spite of having p as one of its arguments, θm (as θs ) expresses a physical quantity
associated with the configuration of time t. The spatial description θs gives the temperature,
at time t, of the material particle whose position at time t is x. In experimental terms, it would
be the temperature read from a thermometer held fixed in space at x. The function θm gives
the temperature, at time t, of the material particle whose position at time 0 is p. It would be
the temperature indicated by a thermometer attached to this material particle.
To avoid notational complexity, the subscripts m and s employed above to denote the
material and spatial descriptions of general fields will not be used throughout this book unless
absolutely necessary. In general, the description employed will be evident either from the
context or from the argument used (p or x).
46 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
θ̇ = c, θ = −b v1 + c.
The material time derivative in this case corresponds to the temperature rate computed from
a thermometer attached to a material particle p whilst θ is the temperature rate observed
in a thermometer held fixed in space at x. Note that the extra term −b v1 added to θ is
a contribution to the rate of change of temperature at x due to the motion of the body
combined with its non-uniform distribution of temperature. This contribution vanishes if the
body moves parallel to e2 (v1 = 0), i.e. the direction of temperature isolines. It would also
vanish if the temperature were uniform throughout the body (b = 0).
Analogously to (2.145) (page 37), we define the spatial and material divergence of a vector
field v, respectively, as
In addition (refer to (2.147)), for a tensor field T, the spatial and material divergence are
given, in Cartesian components, by
∂Tij ∂Tij
(divx T )i = , (divp T )i = . (3.20)
∂xj ∂pj
The compact definition (2.146) is also applicable to the material and spatial divergence of a
tensor.
F = I + ∇p u. (3.22)
ELEMENTS OF CONTINUUM MECHANICS AND THERMODYNAMICS 47
reference
configuration
x+dx
dp
p p+dp Mt x dx=F dp
: M:
t
where xi denote the components of xt . In terms of the reference map (3.11), the deformation
gradient may be equivalently expressed as
Consider the infinitesimal material fibre dp that connects two neighbouring material
particles p and p + dp of a deforming body (Figure 3.6). Under the deformation ϕt , these
particles are mapped, respectively, into x and x + dx. The deformation gradient is the linear
operator that relates infinitesimal material fibres dp with their deformed counterparts dx:
dx = F dp. (3.25)
reference
configuration
dv = det[F] dv0
dv 0
Mt F dc
dc F db x F da
da
db p
M:
t
one has
dv0 = (da × db) · dc. (3.27)
The deformation ϕt maps the infinitesimal vectors, respectively, into F da, F db and F dc,
so that the deformed infinitesimal volume is given by
dv = (F da × F db) · F dc. (3.28)
By making use of identity (2.54), it follows that
dv
det F = , (3.29)
dv0
i.e. the determinant of the deformation gradient represents, locally, the volume after defor-
mation per unit reference volume (or volume change ratio). Throughout this book, we will
adopt the following notation
J ≡ det F . (3.30)
From (3.29) it follows that if det F = 0, then the infinitesimal volume has collapsed
into a material particle. Since the body is not allowed to penetrate itself (this restriction
is embodied in the assumption that the deformation map is one-to-one), this represents a
physically unacceptable situation. Also note that, at the reference configuration, F = I and,
consequently, J = 1. Thus, a configuration with J < 0 cannot be reached from the reference
configuration without having, at some stage, J = 0. Therefore, in any deformed configuration
of a body, J satisfies
J > 0. (3.31)
Isochoric deformations
Isochoric (or volume-preserving) deformations are deformations that do not produce changes
in volume. A locally isochoric deformation is characterised by
J = 1. (3.32)
ELEMENTS OF CONTINUUM MECHANICS AND THERMODYNAMICS 49
Volumetric deformations
Volumetric deformations (i.e. pure contractions/dilations) are deformations consisting purely
of a uniform contraction/dilation in all directions. The deformation gradient of any volumetric
deformation is a spherical tensor:
F = α I, (3.33)
where the scalar α is the corresponding contraction/dilation ratio. With l0 and l denoting,
respectively, the undeformed and deformed lengths of a material fibre, for a locally volumetric
deformation we have:
l
=α (3.34)
l0
in all directions.
F iso ≡ (det F )− 3 F
1
(3.37)
F = R U = V R, (3.40)
where the proper orthogonal tensor R is the local rotation tensor and the symmetric positive
definite tensors U and V are, respectively, the right and left stretch tensors. The right and left
stretch tensors are related by the rotation
V = R U RT . (3.41)
50 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
U dp
R
U deformed
configuration
reference
configuration dx=F dp
F x
dp i2*E
i2 :
p M:
D iiD1*
i1 R
V
R dp
Figure 3.8. Polar decomposition of the deformation gradient. Stretches and rotation.
where C and B – named, respectively, the right and left Cauchy–Green strain tensors – are
defined by
C = U 2 = F TF , B = V 2 = F F T . (3.43)
Example 3.1.2 (A simple plane deformation). To illustrate the meaning of the polar
decomposition of F , a simple example consisting of a body subjected to a homogeneous
deformation, i.e. with F independent of p, is given in what follows. Consider the rectangular
body of Figure 3.8 subjected to homogeneous stretching/compression in the directions of its
longitudinal and transversal axes (respectively, the directions of i1 and i2 in the reference
configuration) with a superimposed rigid rotation of angle α. With pi and xi denoting
coordinates of p and x in the Cartesian system associated with the orthonormal basis {i1 , i2 },
the deformation map is defined as
x1 = p1 λ1 cos α − p2 λ2 sin α
ϕ: (3.44)
x2 = p1 λ1 sin α + p2 λ2 cos α,
where the factors λ1 and λ2 determine how much stretching/compression occurs, respec-
tively, along the longitudinal and transversal axes. In the basis {i1 , i2 }, the matrix represen-
tation of the corresponding deformation gradient is given by
λ1 cos α −λ2 sin α
F= . (3.45)
λ1 sin α λ2 cos α
ELEMENTS OF CONTINUUM MECHANICS AND THERMODYNAMICS 51
and
λ1 cos2 α + λ2 sin2 α (λ1 − λ2 ) sin α cos α
V= . (3.48)
(λ1 − λ2 ) sin α cos α 2
λ1 sin α + λ2 cos α 2
Insight into the meaning of the polar decomposition of the deformation gradient can be gained
by focusing now on the generic infinitesimal fibre represented by dp in Figure 3.8. Under
deformation, dp is mapped into dx = F dp. With use of the polar decomposition of F , this
mapping can be split into two sequential steps. If the right polar decomposition F = R U is
used, the two steps are:
1. dp −→ U dp,
2. U dp −→ R (U dp) = F dp.
In the first operation, dp deforms as if the body were being purely stretched (or compressed)
along the directions of its longitudinal and transversal axes (which at this stage coincide with
i1 and i2 respectively). The second mapping is a pure rotation (of angle α) of the deformed
fibre U dp and corresponds to a rigid rotation of the body. If the left polar decomposition
F = V R is employed instead, the sequence is reversed:
1. dp −→ R dp,
2. R dp −→ V (R dp) = F dp.
In this case, the fibre is first rigidly rotated by an angle α. The second operation corresponds
to the deformation of the fibre under pure stretching/compression of the body along its axial
and transversal directions. However, due to the previous rotation, these directions coincide
now with i∗1 = R i1 and i∗2 = R i2 , respectively. Note that if the basis {i∗1 , i∗2 } is used, the
matrix representation of V reads
λ1 0
V= , (3.49)
0 λ2
so that the transformation (·) → V (·) indeed corresponds to stretchings along the directions
of i∗1 and i∗2 .
The above example has illustrated the significance of the polar decomposition of F . The
discussion has been restricted to a homogeneous deformation only to ease visualisation of the
stretches and rotation involved in the decomposition of the deformation gradient. It should be
52 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
3
3
U= λi li ⊗ li , V= λi ei ⊗ ei , (3.51)
i=1 i=1
where the {λ1 , λ2 , λ3 } are the eigenvalues of U (and V ) named the principal stretches.
The vectors li and ei are unit eigenvectors of U and V respectively. The triads {l1 , l2 , l3 }
and {e1 , e2 , e3 } form orthonormal bases for the space U of vectors in E. They are called,
respectively, the Lagrangian and Eulerian triads and define the Lagrangian and Eulerian
principal directions.
Substitution of (3.41) into (3.51) gives the following relationship between the eigenvectors
of V and U :
l i = R ei , (3.52)
that is, each vector ei differs from the corresponding li by a rotation R.
The spectral decomposition of the right and left stretch tensors implies that in any deforma-
tion, the local stretching from a material particle can always be expressed as a superposition
of stretches along three mutually orthogonal directions. In the example discussed above,
illustrated by Figure 3.8, {λ1 , λ2 } are the principal stretches and the Lagrangian and Eulerian
bases are, respectively, {i1 , i2 } and {i∗1 , i∗2 }.
distance between material particles. Under stretching, we say that the region surrounding p
is strained. To quantify straining, i.e. to evaluate how much U (or V ) departs from I (a rigid
deformation), some kind of strain measure needs to be defined.
Let us consider, again, the generic material fibre represented by the infinitesimal vector
dp that emanates from p (Figure 3.8 serves as an illustration). The deformation maps dp into
dx = F dp. Thus, the square of the deformed length of the material fibre in question reads
where C = F T F = U 2 is the right Cauchy–Green tensor and the strain measure E(2) (the
meaning of the superscript will be made clear below) is the so-called Green–Lagrange strain
tensor defined as
E(2) = 12 (C − I )
= 12 [∇p u + (∇p u)T + (∇p u)T ∇p u]. (3.54)
No straining occurs, that is, the size of any infinitesimal material fibre emanating from
p remains constant (dx = dp, ∀ dp), if and only if E(2) = 0. This condition is
equivalent to C = U = I, implying that F is an orthogonal tensor and the deformation is
rigid (pure translation and/or rotation) in the neighbourhood of p. From the definition of
E(2) , its eigenvectors coincide with the Lagrangian triad so that it can be expressed as
3
E(2) = 1 2
2 (λi − 1) li ⊗ li , (3.55)
i=1
and, since it measures strains along the principal Lagrangian directions, it is called a
Lagrangian strain measure.
It must be emphasised that the Green–Lagrange strain measure is defined by expression
(3.54). It is by no means the unique way of quantifying straining. In fact, the definition of a
strain measure is somewhat arbitrary and a specific choice is usually dictated by mathematical
and physical convenience. An important family of Lagrangian strain tensors, i.e. strain
measures based on the Lagrangian triad, is defined by Seth (1964), Hill (1978) and Ogden
(1984)
1 (U m − I) m
= 0
E(m) = m (3.56)
ln[U ] m=0
where m is a real number and ln[ · ] denotes the tensor logarithm of [ · ]. Equivalently, in
terms of its spectral decomposition, (3.56) may be rephrased as
3
E(m) = f (λi ) li ⊗ li , (3.57)
i=1
where
1 (λm − 1) m
= 0
i
f (λi ) = m (3.58)
ln λi m = 0.
54 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Green-Lagrange (m=2)
2
Biot (m=1)
1
Hencky (m=0)
f (λi )
Almansi (m= --2)
0
-1
-2
0 1 2
λi
Figure 3.9. Strain measures. Principal strain as a function of the principal stretch for various strain
measures.
The Green–Lagrange strain tensor, E(2) , is a particular member of this family (with m = 2).
Other commonly used members of this family are the Biot (m = 1), Hencky (m = 0) and
Almansi (m = −2) strain tensors. Note that for any m, the associated strain tensor vanishes
if and only if the deformation gradient represents, locally, a rigid deformation, i.e.
E(m) = 0 ⇐⇒ U =I ⇐⇒ F = R. (3.59)
To illustrate the relationship between the stretch and strain tensors, the principal strain for
various strain measures is plotted in Figure 3.9 as a function of the corresponding principal
stretch.
Analogously to the strain measures discussed above, it is also possible to define tensors
that measure strain along the principal Eulerian directions or, simply, Eulerian strain tensors.
Based on the left stretch tensor, the Eulerian counterpart of the Lagrangian family of strain
measures above is defined by
1 (V m − I ) m
= 0
(m)
ε = m (3.60)
ln[V ] m = 0,
or, using the Eulerian triad,
3
ε(m) = f (λi ) ei ⊗ ei . (3.61)
i=1
Lagrangian and Eulerian strain tensors are related by
ε(m) = R E(m) RT , (3.62)
that is, they differ by the local rotation R.
ELEMENTS OF CONTINUUM MECHANICS AND THERMODYNAMICS 55
To gain insight into the physical meaning of the tensors D and W , it is convenient to
consider a body undergoing a motion with uniform (independent of x) velocity gradient. For
such a motion the velocity field reads
If the decomposition of L into its symmetric and skew parts is introduced, the velocity field
can be split as
v(x, t) = vR (x, t) + vS (x, t), (3.67)
where the following definitions have been used:
By recalling expression (3.13), the velocity vR , associated with the spin tensor W , can be
immediately identified as a rigid velocity. The only contribution to straining is then provided
by the term vS , associated with the rate of deformation tensor. Note that, due to its symmetry,
D admits the representation
3
D= di ei ⊗ ei , (3.69)
i=1
with di and {ei }, respectively, the eigenvalues and an orthonormal basis of eigenvectors of D.
With the spectral representation above, the velocity field vS can be decomposed as a sum of
three linearly independent velocities of the form:
di (ei ⊗ ei ) (x − y),
ei
x vis ei
xi Ŧ yi
each viS corresponds to a velocity field that purely stretches the body in the direction of ei ,
with the plane perpendicular to ei that passes through y fixed. Thus, the rate of deformation
tensor corresponds indeed to a pure stretching of the body.
If a general motion (in which L is not necessarily uniform) is considered, the above
decomposition of the velocity field into the sum of a rigid velocity and a straining velocity
remains valid in the local sense. In this case, consider a point x and a point x + dx lying
within an infinitesimal neighbourhood of x. The velocity field within this infinitesimal
neighbourhood of x is given by
so that, in any motion, the velocity field can be locally decomposed as a sum of a rigid velocity
D(x, t) dx,
where we have made use of relation (2.140) (page 36) for the derivative of the determinant.
This, together with definition (2.36) (page 22) of the trace of a tensor and the fact that the
skew symmetry of W implies
tr L = tr D, (3.74)
leading to (3.72).
Also note that from the definition (2.145) of the divergence of a vector field we have
tr D = divx v, (3.75)
so that the rate of volume change can be equivalently expressed as
J˙ = J divx v. (3.76)
with o(∇p u) a term of second order in ∇p u. It is clear from this expression that, to within an
error of o(∇p u), only the symmetric part ε of ∇p u is associated with local straining. The skew
part of ∇p u produces no straining and is associated exclusively with local infinitesimal rigid
rotations. For a pure local infinitesimal rigid rotation (dx = dp, ∀ dp) the tensor ε
vanishes or, equivalently, ∇p u is skew.
For a body under an arbitrary homogeneous deformation (∇p u independent of p), the
displacement field can be written as
for all points p and q. For infinitesimal rigid deformations and within an approximation of
second order in the displacement gradient, ∇p u is skew and the field u can be written as
for all points p and q with A ≡ ∇p u a skew tensor. Alternatively, with a denoting the axial
vector of A, u can be expressed as
Any displacement that admits the representation (3.85)–(3.86) is called an infinitesimal rigid
displacement field. Note that infinitesimal rigid displacements have the same representation
as rigid velocity fields (see expressions (3.13) and (3.14)).
ε = ε d + εv , (3.87)
where
ε d ≡ ε − εv (3.88)
is the isochoric component, known as the strain deviator or deviatoric strain, which measures
pure infinitesimal distortions. The tensor
εv ≡ 1
3 εv I (3.89)
ELEMENTS OF CONTINUUM MECHANICS AND THERMODYNAMICS 59
εv ≡ I1 (ε) = tr ε = tr ∇s u = tr ∇u (3.90)
εd = [IS − 1
3 I ⊗ I ] : ε, εv = 1
3 (I ⊗ I ) : ε. (3.92)
tr εd = 0. (3.93)
Id ≡ IS − 1
3 I ⊗ I, (3.94)
dev(S)
dev(S) ≡ Id : S. (3.95)
where
F iso = (det F )− 3 F F = (det F )− 3 C
2 2
Ciso ≡ F iso
T T
(3.97)
and
2
Cv ≡ F vT F v = (det F ) 3 I. (3.98)
Now we proceed to show that, under small strain conditions (small ∇p u), the volumetric
Green–Lagrange strain defined above leads to definition (3.90). From (3.96)2 and (3.98),
we have
2
v = 2 [(det F ) − 1] I.
E(2) 1 3 (3.99)
60 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
From the standard concepts of differentiation discussed in Section 2.5 together with the
definition F = I + ∇p u and the expression given in (iii) of page 36 for the derivative of
the determinant, we find that
det F = det(I + ∇p u)
= det I + (det I) I : ∇p u + o(∇p u)
= 1 + tr ∇p u + o(∇p u) (3.100)
and
2
(det F ) 3 = 1 + 2
3 tr ∇p u + o(∇p u). (3.101)
With the substitution of the above expression into (3.99), we then obtain
E(2)
v = εv + o(∇p u). (3.102)
v ≈ εv .
E(2) (3.103)
= 12 {[1 − 2
3 tr ∇p u + o(∇p u)][I + ∇pT u + ∇p u + o(∇p u)] − I }
= ε − 13 (tr ∇p u)I + o(∇p u)
= εd + o(∇p u). (3.104)
The infinitesimal limits above are valid for all Lagrangian and Eulerian finite strain
measures defined by expressions (3.56) and (3.60).
the so-called polar continuum theories (Cosserrat and Cosserrat, 1909; Toupin, 1962; Truesdell and Noll, 1965).
ELEMENTS OF CONTINUUM MECHANICS AND THERMODYNAMICS 61
1. Boundary forces. Forces applied to the boundary of the body such as those resulting
from contact with another body. The dimension of boundary forces is force per unit
area.
2. Body forces. Forces exerted on the interior of the body. Gravitational and magnetic
forces are typical examples of such forces. The dimension of body forces is force per
unit mass (or volume).
3. Internal interactions between adjacent parts of a body. The dimension of such
interactions is force per unit area.
Internal interaction forces arise from the action of one part of the body upon an adjacent
part and are transmitted across the surface that separate them. Boundary forces represent
interactions between the exterior and the interior of a body and, as internal interactions,
are transmitted across a surface (the boundary of the body in this case). Thus, boundary
forces and interactions between distinct parts of a body are forces of essentially the same
type and will be collectively called surface forces. To describe surface forces mathematically,
the concept of stress as well as the different ways of quantifying it are introduced in this
section.
B
n
t (n)
x
T S
are satisfied, with ρ = ρ(x) denoting the mass density field, i.e. the mass per unit volume in
the deformed configuration of B ’. The right-hand sides of (3.106) and (3.107) contain the
inertia terms, with v̇ = ü denoting the acceleration field of B.
σ = σij ei ⊗ ej , (3.110)
§ The symmetry of σ is a result of the balance of angular momentum.
ELEMENTS OF CONTINUUM MECHANICS AND THERMODYNAMICS 63
B
n
t Vn
x
V22
V2 V1
V21
V12
V23
V32
V11
V13
V31
V33 V3
e2 e2* e1*
e1
e3 e*3
Figure 3.13. Cauchy stress tensor components and principal Cauchy stresses.
with summation on repeated indices implied and the components σij given by
σij = (σ ei ) · ej . (3.111)
From (3.108), it follows that the vector σ ei is the force per unit area exerted across a surface
whose unit normal vector is ei at the point of interest. The component σij of the Cauchy
stress tensor is the magnitude of the projection of σ ei in the direction of ej . The schematic
representation of such projections is illustrated in Figure 3.13 where an infinitesimal cube
with faces normal to the base vectors e1 , e2 and e3 is considered. The components σ11 ,
σ22 and σ33 represent the tractions normal to the faces of the infinitesimal cube whereas the
remaining components, σ12 , σ13 , σ21 , σ23 , σ31 and σ32 are the shear tractions acting parallel
to the faces.
3
σ= σi e∗i ⊗ e∗i , (3.112)
i=1
64 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
that is, there exists an orthonormal basis {e∗1 , e∗2 , e∗3 }, for which all shear components of the
Cauchy stress tensor vanish and only the normal components may be non-zero. The normal
components, σi , are the eigenvalues of σ and are called the principal Cauchy stresses. The
directions defined by the basis {e∗1 , e∗2 , e∗3 } are named the principal stress directions. The
schematic representation of the forces acting on the faces of the infinitesimal cube oriented
according to the principal stress directions is shown in Figure 3.13. The forces are exclusively
normal to the faces of this cube. Note that, analogously to the representation of the stress
tensor in terms of principal stresses, the spectral decomposition has been used in Section 3.1
to represent the stretch tensors U and V in terms of principal stretches (see expression
(3.51)).
σ = s + p I, (3.113)
t
t n
m
reference
F dp2
p x
dp1 dp2 da
da0 F dp1
deformed
S S
ϕ(S )
configuration
Under deformation, the tangent vectors dp1 and dp2 are mapped, respectively, into F dp1
and F dp2 so that the unit normal to the deformed configuration of S reads
n da = F dp1 × F dp2 , (3.119)
where da is the corresponding deformed area element. Pre-multiplication of both sides of the
expression above by F T together with use of the identity
Su × Sv = (det S) S −T (u × v), (3.120)
valid for any invertible tensor S and vectors u and v, leads to
F T n da = J dp1 × dp2 = J m da0 , (3.121)
where J ≡ det F . This is equivalent to
da −T
n=J F m. (3.122)
da0
Finally, with substitution of the expression above into (3.117), t̄ may be written in terms of
the reference unit normal m as
t̄ = J σ F −T m. (3.123)
This last expression motivates the following definition
−T
P ≡J σF , (3.124)
so that the force transmitted across S measured per unit reference area reads
t̄ = P m. (3.125)
The tensor P is called the first Piola–Kirchhoff stress and is often referred to as the Piola–
Kirchhoff stress or nominal stress.¶ The vector t̄ is obtained by applying the first Piola–
Kirchhoff stress to the unit vector m, normal to the reference configuration of S at the point
of interest. Note that in contrast to the Cauchy stress, P is generally unsymmetric.
¶ Some authors (Billington and Tate, 1981; Nemat-Nasser, 1999) define the nominal stress as the transpose of the
reference
a0
e1
Ŧf
f
deformed a
configuration
Example 3.3.1 (The Piola–Kirchhoff stress). Consider a cylindrical bar (Figure 3.15)
with cross-sectional area a0 in its initial configuration (taken as reference). During a
uniaxial experiment this bar is stretched along its longitudinal axis (direction of e1 ) with
a simultaneous reduction of its cross section. Assume that the final deformed configuration
of the bar corresponds to a state of homogeneous deformation with cross-sectional area a.
Furthermore, assume that the bar is subjected to a state of uniaxial stress, with constant σ
given by
σ = σ11 e1 ⊗ e1 .
Let f = f e1 be the total force applied to the deformed configuration of the bar (by the
experimental equipment). Under the assumption of uniform stress distribution in the cross-
section of the bar, force balance requires that the Cauchy stress component σ11 be given by
f
σ11 = .
a
In practice, the force f (and not the stress component) is what can actually be measured in an
experiment. Thus, after f is measured, the Cauchy stress σ11 is determined according to the
expression above. If instead of a, the reference cross-sectional area a0 is used, then the first
Piola–Kirchhoff or nominal stress component is determined
f
P11 = .
a0
It is obvious that, in this case, the corresponding tractions t and t̄, respectively per unit
deformed and reference area, are simply
1 1
t = σ11 e1 = f, t̄ = P11 e1 = f.
a a0
τ ≡ J σ. (3.128)
Due to the symmetry of σ, the Kirchhoff stress is symmetric. Its spectral representation reads
3
τ= τi e∗i ⊗ e∗i , (3.129)
i=1
where the principal Kirchhoff stresses, τi , are related to the principal Cauchy stresses, σi , by
τi = J σi . (3.130)
Later in this book, frequent reference to the principal Kirchhoff stresses will be made in the
formulation of various constitutive models.
ρ̇ + ρ divx u̇ = 0. (3.131)
divx σ + b = ρ ü in ϕ(Ω)
(3.132)
t=σn in ϕ(∂Ω),
Equations (3.132) are also a result of Cauchy’s theorem, alluded to in page 62.
68 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
where n is the outward unit vector normal to the deformed boundary ϕ(∂Ω) of B and t is the
applied boundary traction vector field on ϕ(∂Ω). Equations (3.132) are often referred to as
the strong, local or point-wise form of equilibrium. Equation (3.132)1 is known as Cauchy’s
equation of motion.
The above momentum balance equations are formulated in the spatial (deformed) config-
uration. Equivalently, they may be expressed in the reference (or material) configuration of
B in terms of the first Piola–Kirchhoff stress tensor as
divp P + b̄ = ρ̄ ü in Ω
(3.133)
t̄ = P m in ∂Ω,
where
b̄ = J b (3.134)
is the reference body force, i.e. the body force measured per unit volume in the reference
configuration,
ρ̄ = J ρ, (3.135)
is the reference density (mass per unit volume in the reference configuration), t̄ is the
reference boundary traction (boundary force per unit reference area) and m is the outward
normal to the boundary of B in its reference configuration.
The first principle of thermodynamics postulates the conservation of energy. Before stating
this principle, it is convenient to introduce the product
σ : D,
which represents the stress power per unit volume in the deformed configuration of a body.
The first principle of thermodynamics is mathematically expressed by the equation
ρ ė = σ : D + ρ r − divx q. (3.136)
In words, the rate of internal energy per unit deformed volume must equal the sum of the
stress power and heat production per unit deformed volume minus the spatial divergence of
the heat flux.
Thermodynamic determinism
The basic axiom underlying the constitutive theory discussed here is the principle of
thermodynamically compatible determinism (Truesdell, 1969). It postulates that ‘the history
of the thermokinetic process to which a neighbourhood of a point p of B has been subjected
determines a calorodynamic process for B at p’. In particular, we shall be concerned with
so-called simple materials, for which the local history (history at point p only) of F , θ and
g suffices to determine the history of the thermokinetic process for constitutive purposes. In
this case, regarding the body force b and heat supply r as delivered, respectively, by the linear
momentum balance (3.132)1 and conservation of energy (3.136) and introducing the specific
free energy (3.139), the principle of thermodynamic determinism implies the existence of
constitutive functionals F, G, H and I of the histories of F , θ and g such that, for a point p,
t
σ(t) = F (F t , θ , gt )
t
ψ(t) = G(F t , θ , gt )
t
(3.143)
s(t) = H(F t , θ , gt )
t
q(t) = I (F t , θ , gt )
and the Clausius–Duhem inequality (3.141) holds for every thermokinetic process of B. The
dependence on p is understood on both sides of (3.143) and (·)t on the right-hand sides
denotes the history of (·) at p up to time t.
Material objectivity
Another fundamental axiom of the constitutive theory is the principle of material objectivity
(or frame invariance). It states that ‘the material response is independent of the observer’.
The motion ϕ∗ is related to the motion ϕ by a change in observer if it can be expressed as
where y(t) is a point in space, Q(t) is a rotation and ϕ(p, t) − x0 is the position vector
of ϕ(p, t) relative to an arbitrary origin x0 . This relation corresponds to a rigid relative
movement between the different observers and the deformation gradient corresponding to ϕ∗
is given by
∗
F = Q F. (3.145)
Scalar fields (such as θ, ψ and s) are unaffected by a change in observer but the Cauchy stress
σ(t), heat flux q(t) and the temperature gradient g(t) transform according to the rules
σ −→ σ∗ = Q σ QT
q −→ q∗ = Q q (3.146)
∗
g −→ g = Q g.
ELEMENTS OF CONTINUUM MECHANICS AND THERMODYNAMICS 71
Material symmetry
The symmetry group of a material is the set of density preserving changes of reference
configuration under which the material response functionals F, G, H and I are not affected.
The symmetry group of a solid material is a subset of the proper orthogonal group O + , that
is, a set of rotations. Thus, the symmetry group of a solid material is the set of rotations of
the reference configuration under which the response functionals remain unchanged. This
concept is expressed mathematically as follows. A subgroup S of O + is said to be the
symmetry group of the material defined by the constitutive functionals F, G, H and I if
the relations
t t
F (F t , θ , gt ) = F ([F Q]t , θ , gt )
t t
G (F t , θ , gt ) = G([F Q]t , θ , gt )
t t (3.148)
H (F t , θ , gt ) = H([F Q]t , θ , gt )
t t
I (F t , θ , gt ) = I ([F Q]t , θ , gt )
hold for any time-independent rotation Q ∈ S. A solid is said to be isotropic∗∗ if its symme-
try group is the entire proper orthogonal group. In the development of any constitutive model,
the constitutive functionals must comply with the restrictions imposed by the symmetries of
the material in question.
point p can be completely determined by the knowledge of a finite number of state variables.
The thermodynamic state depends only on the instantaneous value of the state variables and
not on their past history.
Mathematically, state variable models can be seen as particular instances of the general
history functional-based constitutive theory. The relationship between the two approaches
is discussed in detail by Kestin and Bataille (1977) and Bataille and Kestin (1979). In
general terms, state variable models can be obtained from the general history functional-
based description by re-defining the history of the thermokinetic process in terms of a finite
number of parameters (the state variables).
for the stress power, one obtains for the Clausius–Duhem inequality
−T ∂ψ ∂ψ ∂ψ 1
σF −ρ : Ḟ − ρ s + θ̇ − ρ α̇k − q · g ≥ 0. (3.153)
∂F ∂θ ∂αk θ
Equivalently, in terms of power per unit reference volume, we have
∂ψ ∂ψ ∂ψ J
P − ρ̄ : Ḟ − ρ̄ s + θ̇ − ρ̄ α̇k − q · g ≥ 0. (3.154)
∂F ∂θ ∂αk θ
†† The dependence of ψ on the temperature gradient is disregarded as it contradicts the second principle of
∂ψ ∂ψ
P = ρ̄ , s=− , (3.155)
∂F ∂θ
for the first Piola–Kirchhoff stress and entropy. Equation (3.155)1 is equivalent to the
following constitutive relations for the Cauchy and Kirchoff stress tensors:
1 ∂ψ T ∂ψ T
σ= ρ̄ F , τ = ρ̄ F . (3.156)
J ∂F ∂F
Thermodynamical forces
For each internal variable αk of the set α, we define the conjugate thermodynamical force
∂ψ
Ak ≡ ρ̄ . (3.157)
∂αk
With this definition and the identities (3.155), the Clausius–Duhem inequality can be
rewritten as
J
−Ak ∗ α̇k − q · g ≥ 0, (3.158)
θ
where we recall that the symbol ‘∗’ denotes the appropriate product operation between Ak
and α̇k . In what follows, we will adopt for convenience the notation
A ≡ {Ak } (3.159)
for the set of thermodynamical forces, so that (3.158) can be expressed in a more compact
form as
J
−A ∗ α̇ − q · g ≥ 0. (3.160)
θ
α̇ = f (F , θ, g, α)
1 (3.161)
q = h(F , θ, g, α).
θ
Recalling the principle of thermodynamic determinism, the Clausius–Duhem inequality, now
expressed by (3.158), must hold for any process. This requirement places restrictions on the
possible forms of the general constitutive functions f and h in (3.161) (the reader is referred
to Coleman and Gurtin, 1967; Truesdell, 1969, for further details on this issue). It is also
74 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
important to mention that when internal variables of vectorial or tensorial nature are present,
it is frequently convenient to re-formulate (3.161)1 in terms of so-called objective rates rather
than the standard material time derivative of α. Objective rates are insensitive to rigid-body
motions and may be essential in the definition of a frame invariant evolution law for variables
representing physical states associated with material directions. Objective rates are discussed
in Section 14.10 (starting page 615) in the context of the hypoelastic-based formulation of
plasticity models.
where the state variables F , θ and α appear as parameters. The potential Ξ is assumed convex
with respect to each Ak and g, non-negative and zero valued at the origin, {A, g} = {0, 0}.
In addition, the hypothesis of normal dissipativity is introduced, i.e. the flux variables are
assumed to be determined by the laws
∂Ξ 1 ∂Ξ
α̇k = − , q=− . (3.163)
∂Ak θ ∂g
A constitutive model defined by (3.150), (3.155) and (3.163) satisfies a priori the
dissipation inequality. It should be noted, however, that the constitutive description by means
of convex potentials as described above is not a consequence of thermodynamics but, rather,
a convenient tool for formulating constitutive equations without violating thermodynamics.
Examples of constitutive models supported by experimental evidence which do not admit
representation by means of dissipation potentials are discussed by Onat and Leckie (1988).
refined constitutive models, incorporating a larger number of state variables, will be required.
Due account of the possible temperature histories and time span to be considered is also
fundamental. At higher temperatures, the long-term behaviour of the steel bar subjected to
even a very small strain, may no longer be accurately modelled by the linear elasticity theory.
In this case the introduction of time-dependent effects (creep/relaxation) may be essential to
produce an acceptable model. In an extreme situation, if the temperature rises above melting
point, the bar will cease to be a solid. Under such circumstances, a fluid mechanics theory
will be needed to describe the behaviour of the material.
In general, due to the difficulty involved in the identification of the underlying dissipative
mechanisms, the choice of the appropriate set of internal variables is somewhat subtle and
tends to be biased by the preferences and background of the investigator. In simplistic
terms, we may say that constitutive modelling by means of internal variables relies either
on a micromechanical or on a phenomenological approach. The micromechanical approach
involves the determination of mechanisms and related variables at the atomic, molecular
or crystalline levels. In general, these variables are discrete quantities and their continuum
(macroscopic) counterparts can be defined by means of homogenisation techniques. The
phenomenological approach, on the other hand, is based on the study of the response of
the representative volume element, i.e. the element of matter large enough to be regarded as
a homogeneous continuum. The internal variables in this case will be directly associated
with the dissipative behaviour observed at the macroscopic level in terms of continuum
quantities (such as strain, temperature, etc.). Despite the macroscopic nature of theories
derived on the basis of the phenomenological methodology, it should be expected that ‘good’
phenomenological internal variables will be somehow related to the underlying microscopic
dissipation mechanisms.
The phenomenological approach to irreversible thermodynamics has been particularly
successful in the field of solid mechanics. Numerous well-established models of solids, such
as classical isotropic elastoplasticity and viscoplasticity, discussed in Parts Two and Three of
this book, have been developed on a purely phenomenological basis providing evidence of
how powerful such an approach to irreversible thermodynamics can be when the major con-
cern is the description of the essentially macroscopic behaviour. In some instances, however,
the inclusion of microscopic information becomes essential and a purely phenomenological
methodology is unlikely to describe the behaviour of the material with sufficient accuracy.
One such case is illustrated in Chapter 16, where a microscopically-based continuum model
of ductile metallic crystals is described.
where b and t are the body force per unit deformed volume and boundary traction per unit
deformed area and V is the space of virtual displacements of B, i.e. the space of sufficiently
regular arbitrary displacements
η : ϕ(Ω) → U.
We now concentrate on the first term within the square brackets of the above equation. The
divergence theorem (expression (2.148)2, page 37) implies the following identity
divx (σ η) dv = σ η · n da. (3.171)
ϕ(Ω) ϕ(∂Ω)
By taking into account the symmetry of σ, which implies ση · n = σn · η, together with the
above identity, equation (3.170) can be rewritten in the equivalent form
(divx σ + b − ρü) · η dv + (t − σ n) · η da = 0, ∀ η ∈ V. (3.172)
ϕ(Ω) ϕ(∂Ω)
78 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Finally, since this equation holds for all virtual displacement fields η, then it follows from the
fundamental theorem of variational calculus (refer, for instance, to Gurtin 1972; Oden 1979
or Reddy 1998) that each bracketed term of the above equation must vanish pointwise within
their respective domains, i.e. we recover the strong equilibrium equations (3.132).
Conversely, the strong form yields the weak form of equilibrium. This can be shown in a
relatively straightforward manner by applying a weighted residual method to the strong form
together with use of the divergence theorem.
where b̄ and t̄ are, respectively, the body force per unit reference volume and the surface
traction per unit reference area and ρ̄ is the mass density in the reference configuration. The
space of virtual displacements, V, is accordingly defined as the space of sufficiently regular
arbitrary displacement fields
η : Ω → U.
The material version of the virtual work equation is obtained by introducing, in its spatial
counterpart, the identities
1 −1
σ= P F T; ∇x a = ∇p a F , (3.174)
J
where the second expression holds for a generic vector field a, and making use of the standard
relation (Gurtin, 1981)
a(x) dv = J(p) a(ϕ(p)) dv, (3.175)
ϕ(Ω) Ω
b(t), t ∈ [t0 , T ]
(i) The natural boundary condition. The history of the surface traction
t(t), t ∈ [t0 , T ],
with dependence on x implied, is prescribed over the portion of the boundary of B that
occupies the region ∂Ωt in its reference configuration.
(ii) The essential boundary condition. The motion is a prescribed function on the part of
the boundary of B that occupies the region ∂Ωu in the reference configuration
The body B is assumed to be made from a generic material modelled by the internal
variable-based constitutive equations associated with Problem 3.1 (page 76) and the internal
variable field, α, is known at the initial time t0 , i.e.
The fundamental quasi-static initial boundary value problem is stated in its spatial version in
the following.
Problem 3.3 (The spatial quasi-static initial boundary value problem). Find a kinemati-
cally admissible displacement function, u ∈ K, such that, for all t ∈ [t0 , T ], the virtual work
80 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
t
x t
M
: t
: @ :t t
M
@ :t b
x t
u
t
@ :u t
M
@ :u
equation is satisfied
[σ(t) : ∇x η − b(t) · η] dv − t(t) · η da = 0, ∀ η ∈ Vt . (3.179)
ϕ(Ω,t) ϕ(∂Ωt ,t)
where P (t) is the solution of constitutive initial value Problem 3.1 (page 76) with prescribed
deformation gradient
F (t) = ∇p ϕ(p, t) = I + ∇p u(p, t). (3.182)
The problem can be equivalently formulated in the reference configuration of B in terms
of the material version of the principle of virtual work (3.173). For completeness, we state
the material version of the fundamental initial boundary value problem in the following.
Problem 3.4 (The material quasi-static initial boundary value problem). Find a kinemat-
ically admissible displacement function, u ∈ K, such that, for all t ∈ [t0 , T ],
[P (t) : ∇p η − b̄(t) · η] dv − t̄(t) · η da = 0, ∀ η ∈ V, (3.183)
Ω ∂Ωt
where
V = {η : Ω → U | η = 0 on ∂Ωu } (3.184)
and the Piola–Kirchhoff stress, P (t), is the solution of initial value Problem 3.1 with
prescribed deformation gradient (3.182).
ELEMENTS OF CONTINUUM MECHANICS AND THERMODYNAMICS 81
Problem 3.5 (The infinitesimal quasi-static initial boundary value problem). Find a
kinematically admissible displacement, u ∈ K, such that, for t ∈ [t0 , T ],
[σ(t) : ∇η − b(t) · η] dv − t(t) · η da = 0, ∀ η ∈ V, (3.185)
Ω ∂Ωt
where
V = {η : Ω → U | η = 0 on ∂Ωu } (3.186)
and, at each point p, σ(t) is the solution of the constitutive initial value Problem 3.2 (page 76)
with prescribed strain
ε(t) = ∇s u(p, t). (3.187)
4 THE FINITE ELEMENT METHOD
IN QUASI-STATIC NONLINEAR
SOLID MECHANICS
I N this chapter we present a summary of the application of the Finite Element Method
to quasi-static nonlinear solid mechanics. Our aim here is to review the main techniques
involved in the finite element solution of the quasi-static initial boundary value problems
stated in Section 3.7. No attempt is made to provide a comprehensive review of the subject.
For a detailed account on the Finite Element Method, and related techniques we refer to
standard textbooks (Bathe, 1996; Belytschko et al., 2000; Hughes, 1987; Wriggers, 2001;
Zienkiewicz and Taylor, 2000). The material presented here covers the basic procedures that
form the backbone of the computer implementation provided in the finite element program
HYPLAS and should be helpful to readers who wish to learn more about the code in Chapter 5.
Throughout the chapter, reference is frequently made to subroutines of HYPLAS involved in
the computational implementation of the procedures described.
At this point, it is probably worth emphasising that two major numerical approximations
(Figure 4.1) are necessary in the finite element solution of the generic initial boundary value
problems stated in Section 3.7:
1. A time discretisation of the underlying constitutive initial value problem. A numerical
integration scheme is introduced to solve the initial value problem defined by the
constitutive equations of the model that relate stresses to the history of deformations.
The general constitutive initial value problems are those stated in Section 3.5.5.
Upon introduction of the numerical integration scheme, the original time-continuum
constitutive equations are transformed into incremental (or time-discrete) counterparts.
2. A finite element discretisation. This comprises a standard finite element approximation
of the virtual work statement where the domain of the body and the associated
functional sets are replaced with finite-dimensional counterparts generated by finite
element interpolation functions.
With the introduction of the above approximations, the original initial boundary value
problem is reduced to a set of incremental (generally nonlinear) algebraic finite element
equations to be solved at each time station of the considered time interval. In addition to items
1 and 2 above, the present chapter addresses the solution of the associated algebraic system,
with particular emphasis on the quadratically convergent Newton–Raphson algorithm. The
arc-length technique, which becomes crucial in the solution of problems involving structural
instability, is also described.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
84 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
finite element
discretisation of the PVW
incremental
finite element equations
Figure 4.1. Numerical approximations. Reducing the initial boundary value problem to a set of
incremental finite element equations.
so that they can be distinguished from standard vector and second-order tensor quantities that
are represented with italic bold-faced fonts.
the strain tensor only (without dependence on internal variables). Since the strain tensor is in
turn a functional of the displacement field, we have the following constitutive equation
σ = σ(ε(u)) = σ(∇s u). (4.1)
With the above constitutive equation, the infinitesimal problem of Section 3.7.2 reduces to
finding a kinematically admissible displacement field u ∈ K such that
[σ(∇ u) : ∇ η − b · η] dv −
s s
t · η da = 0, ∀ η ∈ V (4.2)
Ω ∂Ωt
holds.
Ni(e) ( x )
N i(e)( x i ) = 1
Ω
(e)
xi
a( x ) ha ( x ) = Σ a i Ni(e)( x )
i
a i = a( x i )
xi
g
Ni ( x i ) = 1
xi
and
npoin
V≡
h h
η(x) = η i
Nig (x) | η = 0 if x ∈ ∂Ωu .
i i
(4.10)
i=1
The finite element approximation to the continuum variational equation (4.2) is then obtained
by replacing the functional sets K and V with the above defined finite-dimensional subsets.
THE FINITE ELEMENT METHOD IN QUASI - STATIC NONLINEAR SOLID MECHANICS 87
Matrix notation
To derive the discretised form of the virtual work equation it is convenient to introduce the
standard matrix notation that follows. Let ndim be the number of spatial dimensions. Firstly,
we define the global interpolation matrix:
g g
Ng (x) = [diag[N1 (x)] diag[N2 (x)] · · · diag[Nngpoin (x)]], (4.11)
where the generic element uji is the i-component of the displacement vector at the global
node j. Any element h u ∈ hK can be represented as
h
u = Ng u, (4.14)
where the components of u corresponding to nodal points on ∂Ωu satisfy the prescribed
kinematic constraints and dependence on x has been omitted for notational convenience.
Analogously, any virtual displacement h η ∈ hV has the representation
h
η = Ng η, (4.15)
where
n
η = [η11 , . . . , ηn1 dim , . . . . . . , η1 poin , . . . , ηnndim
poin T
] (4.16)
is the vector of virtual nodal displacements. In accordance with the definition of hV, the
components of η associated with nodal points belonging to ∂Ωu must vanish.
It is also convenient to introduce the global discrete symmetric gradient operator (or
strain-displacement matrix) which, in two dimensions (plane stress and plane strain prob-
lems), has the format
g g
N1,1 0 N2,1 0 · · · Nngpoin ,1 0
Bg = ,
g g g
0 N 1,2 0 N 2,2 · · · 0 N npoin ,2 (4.17)
g g g g g g
N1,2 N1,1 N2,2 N2,1 · · · Nnpoin ,2 Nnpoin ,1
where
∂(·)i
(·)i,j = . (4.18)
∂xj
88 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Since the above equation is satisfied for all vectors η of the form (4.16), the term within the
curly brackets on the left-hand side must vanish. Also, recall that the stress tensor components
depend on the strain components which, in turn, depend on the displacement field defined by
u. The finite element discrete boundary value problem is then formulated as follows. Find the
global vector of nodal displacements, u, such that
f int (u) − f ext = 0, (4.25)
where f int and f ext are, respectively, the internal and external global force vectors
f int = (Bg )T σ dv
hΩ
(4.26)
ext g T g T
f = (N ) b dv + (N ) t da.
hΩ ∂ hΩt
† The computational array representation of tensors is described in more detail in Appendix D. The reader
The finite element assembly operator, A, implies that each component of the global force
associated with a particular global node is obtained as the sum of the corresponding
contributions from the element force vectors of all elements that share that global node. For
a generic element e, the interpolation matrix N is given by
and matrix B (in plane stress/strain analyses) has the standard format
(e) (e) (e)
N1,1 0 N2,1 0 · · · Nnnode ,1 0
(e) (e) (e)
B= 0 N1,2 0 N2,2 · · · 0 Nnnode ,2 . (4.30)
(e) (e) (e) (e) (e) (e)
N1,2 N1,1 N2,2 N2,1 · · · Nnnode ,2 Nnnode ,1
where ξi (i = 1, . . . , ngaus ) are the positions (coordinates) of the Gauss points in the standard
domain Γ and wi (i = 1, . . . , ngaus ) are the corresponding weights.
Now let g be a generic function defined over the element domain, Ω(e) , and let x : Γ →
Ω map the standard domain Γ onto Ω(e) . By a straightforward transformation of variables
(e)
90 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
the Gaussian quadrature for approximation of the integral of g over Ω(e) is obtained as
ngaus
g(x) dx = g(x(ξ)) j(ξ) dξ ≈ wi gi ji , (4.32)
Ω(e) Γ i=1
where
j(ξ) = det[∂x/∂ξ] (4.33)
is the determinant of the Jacobian of the transformation x and we have defined
ji = j(ξi ) (4.34)
and
gi = g(x(ξi )). (4.35)
Similarly to (4.32), we define the Gaussian quadrature to approximate integrals over the
boundary, ∂Ω(e) , of an element. With ∂Γ denoting the standard integration interval and the
function g defined over ∂Ω(e) we have
ngausb
g(xb ) dxb = g(xb (ξb )) j b (ξb ) dξb ≈ wib gi jib , (4.36)
∂Ω(e) ∂Γ i=1
where wib are the weights at the corresponding boundary integration points ξbi and jib denotes
the Jacobian of the boundary transformation xb : ∂Γ → ∂Ω(e) .
By applying Gaussian quadratures with ngausp and ngausb points for integrals, respec-
(e)
tively, over the element domain and the relevant portion of its boundary (∂Ωt ), the element
int ext
arrays f (e) and f (e) are evaluated as
ngausp
f int
(e) = wi BTi σi ji
i=1
(4.37)
ngausp
ngausb
ext
f (e) = wi NTi bi ji + wib NTi ti jib .
i=1 i=1
Linear triangle
We start by considering the simplest two-dimensional element – the three-noded linear
triangle. This element is named TRI 3 in program HYPLAS. Figures 4.2 and 4.3 have already
illustrated element and global shape functions associated with the linear triangle. Let the
element be defined over the standard domain Γ (Figure 4.4). The element shape functions are
THE FINITE ELEMENT METHOD IN QUASI - STATIC NONLINEAR SOLID MECHANICS 91
x3
η )
x(ξ
(e)
Ω
3 (0,1) x1
y x2
(0,0) (1,0)
x
1 2 ξ
defined as
N1 (ξ) = N1 (ξ, η) = 1 − ξ − η
N2 (ξ) = N2 (ξ, η) = ξ (4.38)
N3 (ξ) = N3 (ξ, η) = η.
3
x(ξ) = Ni (ξ) xi , (4.39)
i=1
3
x(ξ, η) = Ni (ξ, η) xi
i=1
(4.40)
3
i
y(ξ, η) = Ni (ξ, η) y .
i=1
In HYPLAS, the shape functions (as well as their derivatives) for the three-node triangle are
coded in subroutine SFT3.
This element is typically used with a one-point Gauss quadrature. The corresponding
Gauss-point position and weight are
1
ξ1 = 1
3, 3 , w1 = 12 . (4.41)
92 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
x4
x3
)
x(ξ (e)
Ω
η
y x1
(−1,1) (1,1)
4 3 x2
x
ξ
Γ
1 2
(-1,-1) (1,−1)
bi-linear
(e)
shape function, N i
Bi-linear quadrilateral
The bi-linear quadrilateral (named QUAD 4 in HYPLAS) has the following shape functions:
N1 (ξ) = 14 (1 − ξ − η + ξη)
N2 (ξ) = 14 (1 − ξ + η − ξη)
(4.42)
N3 (ξ) = 14 (1 + ξ + η + ξη)
N4 (ξ) = 14 (1 + ξ − η − ξη).
The above shape functions are computed in subroutine SFQ4 of HYPLAS. The element is
illustrated in Figure 4.5. The four-point (2 × 2) Gauss quadrature is usually adopted for this
element. The corresponding sampling point positions and weights can be found in subroutine
GAUS2D.
Eight-noded quadrilateral
of volumetric ‘locking’ observed when low-order elements are used under near incompress-
ibility conditions (volumetric locking is discussed in Chapter 15). Under plane stress, the
nine-point (3 × 3) Gauss quadrature is recommended.
Matrix notation
Let us now define the array of engineering strains.‡ Analogously to the array σ of stress
components, we define for plane strain/stress,
ε = [ε11 , ε22 , 2ε12 ]T . (4.47)
Under axisymmetric conditions, we have
ε = [ε11 , ε22 , 2ε12 , ε33 ]T , (4.48)
and, for three-dimensional analyses,
ε = [ε11 , ε22 , ε33 , 2ε12 , 2ε23 , 2ε13 ]T . (4.49)
The elastic law can then be written equivalently in terms of the arrays σ and ε as
σ = De ε, (4.50)
where the elasticity matrix, De , has the general form
De = 2G IS + A (K − 23 G) i iT , (4.51)
where IS is the array of components of IS (refer to Section D.2, starting on page 761) and i is
the array representation of the second-order identity. For example, in axisymmetric problems
we have T
i= 1 1 0 1 . (4.52)
‡ Refer to Appendix D for further details on the computational array representation of tensors in the finite element
context.
94 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
ε = Bg u, (4.53)
σ = De Bg u. (4.54)
With the substitution of the above relation into (4.26)1, the global internal force vector
reduces to the following linear function of u:
f int (u) = (Bg )T De Bg dv u, (4.55)
hΩ
or, equivalently,
f int (u) = K u, (4.56)
where K is the global stiffness matrix assembled from the element stiffnesses
nelem
K= A K(e) , (4.57)
e=1
with
K(e) = BT De B dv. (4.58)
Ω(e)
Finally, with (4.56), the discrete boundary value problem defined by the equilibrium
equation (4.25) is reduced to the solution of the following linear system of algebraic equations
for the global nodal displacement vector u:
K u = f ext . (4.59)
In program HYPLAS, the solution of the above system is undertaken by the classical frontal
method (Hinton and Owen, 1977; Irons, 1970). The frontal linear equation solver is coded in
subroutine FRONT.
whose outcome, σn+1 , is expected to converge to the exact solution of the actual evolution
problem as the strain increments are reduced. The numerical constitutive law is nonlinear in
general and is path-independent within one increment; that is, within each increment σn+1 is
a function of εn+1 alone (note that the argument αn is constant within [tn , tn+1 ]), analogous
to a nonlinear elastic law. The integration algorithm also defines a similar incremental
constitutive function for the internal variables of the model:
for any η ∈ V, where bn+1 and tn+1 are the body force and surface traction fields prescribed
at time station tn+1 . The set Kn+1 is defined as
Equation (4.64) is generally nonlinear. The source of its nonlinearity is the nonlinearity of
the incremental constitutive function that takes part in the definition of the element internal
force vector above.
The incremental finite element scheme is summarised in Box 4.1 where the particular
case of proportional loading (implemented in HYPLAS) is considered. Proportional loading is
characterised by body force and surface traction fields given, at an arbitrary instant tn+1 , by
bn+1 = λn+1 b̃
(4.66)
tn+1 = λn+1 t̃,
where λn+1 is the prescribed load factor at tn+1 and b̃ and t̃ are prescribed constant (in time)
fields. In this case, the global external force vector reduces to
f ext ¯ext ,
n+1 = λn+1 f (4.67)
ext
where f¯ is computed only once at the beginning of the incremental procedure as the
assembly of element vectors
¯ ext T
f (e) = N b̃ dv + NT t̃ da. (4.68)
(e)
Ω(e) ∂Ωt
ext
(i) Assemble global external force, f¯ and set up the proportional loading curve, λ(t)
(ii) Initialise increment counter, i := 1
(iii) Set load factor λi := λ(ti ) at the end of the current interval [ti−1 , ti ]
(iv) Solve the nonlinear equation
ext
f int (ui ) − λi f¯ = 0
− (N ) t da · η,
g T
∀ η ∈ hV. (4.69)
∂ h Ωt
− (Ng )T t da . (4.70)
∂ h Ωt
(k−1)
At a state defined by the global displacement vector un+1 , the typical iteration (k) of the
Newton–Raphson scheme consists of solving the linear system of equations
for δu(k) , where we have defined the residual (or out-of-balance force) vector
(k−1)
r(k−1) ≡ f int (un+1 ) − f ext
n+1 , (4.72)
With the solution δu(k) of the linear system (4.71) at hand, we apply the Newton correction
to the global displacement
(k) (k−1)
un+1 = un+1 + δu(k) , (4.74)
98 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
The method is schematically illustrated in Figure 4.6. The Newton–Raphson iterations are
repeated until after some iteration (m), the following convergence criterion is satisfied:
|r(m) |
≤ tol , (4.77)
|f ext
n+1 |
where tol is a sufficiently small specified equilibrium convergence tolerance. The corre-
(m)
sponding displacement vector, un+1 , is then accepted as sufficiently close to the solution
of (4.64)
(m)
un+1 := un+1 . (4.78)
(0)
To start up the Newton–Raphson iterations, we need an initial guess, un+1 . The initial guess
is usually taken as the converged (equilibrium) displacement vector at the end of the previous
increment,
(0) (0)
un+1 = un or ∆un+1 = 0. (4.79)
The overall Newton–Raphson algorithm for solution of the nonlinear finite element equations
is summarised in Box 4.2 in pseudo-code format. The procedure is implemented in program
HYPLAS. To help familiarise the reader with the program, the main subroutines of HYPLAS
associated with some of the items listed in Box 4.2 are named within square brackets.
The consistent tangent operator is the derivative of the incremental constitutive function
σ̂. This generally implicit function is typically defined by some numerical algorithm for
integration of the rate constitutive equations of the model. The full (exact) linearisation of
the finite element equations in the context of path-dependent materials, including the exact
THE FINITE ELEMENT METHOD IN QUASI - STATIC NONLINEAR SOLID MECHANICS 99
f int (u)
ext
f n+1
KT
r (1)
KT
r (0)
f next
u(0)
n+1 = u n u(1)
n+1
(2)
un+1 ... un+1 u
δ u(1) δ u(2)
Figure 4.6. The Newton–Raphson algorithm for the incremental finite element equilibrium equation.
linearisation of the incremental constitutive function, has been first addressed by Nagtegaal
(1982). In particular, this author has emphasised the need for linearisation of the incremental
stress updating procedure, rather than to appeal to the rate stress–strain tangential relation
(commonly used at the time), in order to achieve quadratic rates of asymptotic convergence
in the iterative solution of the finite element equilibrium equations. This concept was later
formalised by Simo and Taylor (1985) who coined the term consistent tangent to refer to
the tangent operator consistent with the relevant numerical algorithm for integration of the
path-dependent rate constitutive equations. The derivation of consistent tangent operators
is thoroughly discussed in Part Two of this book, in the context of elastoplasticity and
elastoviscoplasticity. We remark that the full Newton–Raphson scheme, which relies on
consistent tangent operators, is available for all material models implemented in HYPLAS.
Other iterative methods can be used to solve the nonlinear incremental finite element
equations. A straightforward alternative to the above full Newton–Raphson scheme are
the so-called modified Newton methods. These methods consist of replacing the tangential
stiffness KT , which in the standard method is updated in every iteration, by a constant
counterpart. Traditional approaches include (Owen and Hinton (1980) and Zienkiewicz and
Taylor (2000))
(i) the use of the initial tangent stiffness throughout all increments;
(ii) the use of a constant stiffness within each increment. The stiffness is updated at the
beginning of every increment;
100 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Box 4.2. The Newton–Raphson scheme for solution of the incremental nonlinear
finite element equation (infinitesimal strains).
D := ∂ σ̂/∂εn+1
KT δu(k) = −r(k−1)
(vii) Use constitutive integration algorithm to update stresses and other state variables
[MATISU]
(k) (k) (k) (k)
σn+1 := σ̂(αn , εn+1 ); αn+1 := α̂(αn , εn+1 )
(viii) Compute element internal force vectors [INTFOR, IFSTD2]
ngausp
(k)
f int
(e) := i=1 wi ji BTi σn+1
i
(ix) Assemble global internal force vector and update residual [CONVER]
ext
r := f int − λn+1 f¯
(iii) a variant of method (ii) where the stiffness matrix is updated after a certain number of
iterations within each increment.
For such schemes, the stiffness matrix does not need to be computed or factorised whenever it
is reused. Some procedures of this type are available in the HYPLAS program. The convergence
rates of such methods are far slower than the quadratic rates of the classical Newton algorithm
and numerical experience shows that much faster solutions are obtained generally with the
full Newton–Raphson scheme.
THE FINITE ELEMENT METHOD IN QUASI - STATIC NONLINEAR SOLID MECHANICS 101
Other alternatives include the quasi-Newton methods. Such methods, whose roots are in
the theory of optimisation (Haftka et al., 1990), require neither the direct computation of
the tangent stiffness nor its inversion (linear system solution). Instead, at a new iteration,
an updated approximation of the inverse of the stiffness matrix is obtained which depends
on the residual and displacement vectors at the end of the previous two iterations and
the inverse stiffness approximation employed in the previous iteration of the algorithm.
The Broyden–Fletcher–Goldfarb–Shanno algorithm (BFGS) is a particularly popular quasi-
Newton scheme. Its application in the finite element context has been introduced by Matthies
and Strang (1979). At the beginning of an iteration (k), the approximate inverse stiffness of
the BFGS scheme is updated by the formula
T
vwT K−1 K−1
(k−1) wv
T
wT K−1
(k−1) w vw
K−1 K−1
(k−1)
(k) = (k−1) − − + 1+ , (4.83)
vT w vT w vT w vT w
where
v = u(k−1) − u(k−2) , w = r(k−1) − r(k−2) . (4.84)
These methods can be useful, for instance, when the derivation of the tangent operator (4.81)
is difficult – a situation that may arise in dealing with complex material models/integration
algorithms. Again, the rates of convergence are much slower than those produced by the
standard Newton algorithm.
the form
p
i i−1
v (x, t) = v (x, t) + gk (t) wk (x), (4.85)
k=1
where gk are suitably chosen scalar-valued basis functions defined over [t0 , T ] and wk are
vector-valued functions defined over Ω. The functions wk are generated by conventional
finite element interpolation and their corresponding nodal values are unknowns at each
iteration of the method. Analogous approximations are used to describe the sequence of strain
rate and stress rate histories. The overall accuracy of the history approximations depends
crucially upon the choice of the total number, p, of terms used to describe the iterative history
differences. In contrast to the conventional incremental procedure, the single increment of the
LATIN Method comprising the interval [t0 , T ] may contain several cycles of a cyclic load.
This is illustrated, for instance, by Boisse et al. (1990, 1989) in elastoplastic applications.
Further applications are found in Boucard et al. (1997), Cognard and Ladèveze (1993),
Cognard et al. (1999), Ladevèze and Perego (2000) and Dureisseix et al. (2003).
Problem 4.2 (The finite strain incremental boundary value problem). Given the field αn
at time tn and given the body forces and surface traction fields at tn+1 , find a kinematically
admissible configuration ϕn+1 (Ω) ∈ Kn+1 such that the virtual work equation
[σ̂(αn , F n+1 ) : ∇x η − bn+1 · η] dv −
s
tn+1 · η da = 0, (4.88)
ϕn+1 (Ω) ϕn+1 (∂Ωt )
and
F n+1 = ∇p ϕn+1 = I + ∇p un+1 . (4.90)
The B-matrix above is the spatial discrete symmetric gradient operator. It has the same
format as that of its small strain counterpart (such as, for example, that given by (4.30) for the
plane stress and plane strain cases), but its shape function derivatives are spatial derivatives;
that is, derivatives with respect to the spatial coordinates of the finite element mesh (at the
deformed configuration defined by un+1 ).
Starting from (C.32) and following the same arguments as those resulting in (4.70) we
obtain the discrete form of the linearised virtual work equation for large strain problems in
spatial description:¶
(G ) a G dv δu = −
g T g
[(Bg )T σ − (Ng )T b] dv
ϕ(h Ω) ϕ(h Ω)
− g T
(N ) t da , (4.93)
ϕ(∂ h Ω)
where a is the matrix of components of the spatial tangent modulus (C.31) ordered according
to the convention described in Section D.2.1 (starting page 763). The matrix Gg is the global
discrete spatial gradient operator. In plane strain and plane stress analyses, for example, it is
given by
g g
N1,1 0 N2,1 0 · · · Nngpoin ,1 0
0 g
N1,1 0 g
N2,1 ··· 0 Nngpoin ,1
Gg = N g g g
.
(4.94)
1,2 0 N 2,2 0 · · · N npoin ,2 0
g g
0 N1,2 0 N2,2 ··· 0 Nngpoin ,2
Its multiplication by, say, the global vector δu gives the array of components of the spatial
gradient of δu ordered according to the convention of Section D.2.1.
Similarly to the infinitesimal strain case, the generic Newton–Raphson iteration (k) here
requires the solution of the standard linear system for δu:
where the global tangent stiffness matrix KT corresponds to the bracketed term on the left-
hand side of (4.93). This matrix is constructed in practice by assembling the element tangent
stiffness matrices defined as
(e)
KT = GT a G dv, (4.96)
(k)
ϕn+1 (Ω(e) )
where G is the element discrete spatial gradient operator which, in plane stress/strain
analyses, has the format
(e) (e) (e)
N1,1 0 N2,1 0 · · · Nnnode ,1 0
0 N
(e)
0 N
(e)
· · · 0 N
(e)
1,1 2,1 nnode ,1
G= N (e)
. (4.97)
0
(e) (e)
1,2 0 N2,2 0 · · · Nnnode ,2
(e) (e) (e)
0 N1,2 0 N2,2 · · · 0 Nnnode ,2
¶ We remark that this expression is valid when the external loads b and t are configuration-independent, i.e. they
do not depend on the displacement u. Its extension to account for configuration-dependent loads is briefly addressed
in Section 4.3.6.
THE FINITE ELEMENT METHOD IN QUASI - STATIC NONLINEAR SOLID MECHANICS 105
Box 4.3. The Newton–Raphson scheme for the large strain incremental nonlinear
finite element equation. Modifications to Box 4.2.
(vii) Use constitutive integration algorithm to update stresses and other state variables
[MATISU]
(k) (k) (k) (k)
σn+1 := σ̂(αn , Fn+1 ); αn+1 := α̂(αn , Fn+1 )
The overall Newton procedure in the finite strain case is completely analogous to that
shown in Box 4.2 for the infinitesimal theory. The only necessary change is the replacement
of some of the items of Box 4.2 with the finite strain counterparts listed in Box 4.3.
Examples of path-dependent material implementations where τ is defined by an implicit
algorithmic function are given in Part Three of this book. The derivation of explicit
expressions for the above implicit function derivative may become quite intricate but involve
nothing more than standard concepts of linearisation. Finite elasticity (also addressed in
Part Three) is a particular instance where τn+1 is obtained by direct function evaluation.
In such a case, τ̂ is an explicit function of the deformation gradient and its derivative can, as
a general rule, be derived in a relatively straightforward manner.
106 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
defined by
(e)
KM = BT c B dv
(k)
ϕn+1 (Ω(e) )
(4.101)
(e)
KG = GT S G dv,
(k)
ϕn+1 (Ω(e) )
As the infinitesimal tangent operator (4.81), the tensor c has the symmetries
The material stiffness matrix (4.101)1 has the same format as that of the infinitesimal stiffness
matrix (4.80) except that, in the finite strain case, the matrix form of the tensor defined
by (4.102) replaces the matrix form of the infinitesimal consistent tangent operator. In
computational terms, this allows the subroutines for computation of the infinitesimal element
tangent stiffness matrix to be reused to compute the material stiffness of the finite strain case.
If this approach is adopted, the full tangent stiffness under finite strains can be obtained by
simply adding the geometrical term (4.101)2 to the material stiffness. We remark, however,
that in program HYPLAS the form (4.96) is adopted instead; that is, under finite strains,
we firstly compute the matrix form of tensor a and then obtain the element stiffness by
performing a single computation with (4.96).
f ext
n+1 = f
ext
(un+1 ), (4.105)
KT + KL ,
vector:
∂f ext
KL = . (4.106)
∂un+1 u(k−1)
n+1
∆λ ≡ λn+1 − λn (4.107)
be the corresponding incremental load factor. To derive the arc-length method, we allow ∆λ
to become a variable and redefine the residual equation (4.91) as
ext
r(un+1 , ∆λ) ≡ f int (un+1 ) − (λn + ∆λ) f¯ = 0. (4.108)
The arc-length method consists of adding an extra constraint to the above augmented residual
equation so as to limit the ‘length’ of the incremental solution.
108 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
λ λ
snap-through snap-back
A
A
0 u 0 u
For the so-called spherical arc-length method, the constraint equation has the general format
For the more widely used cylindrical arc-length method (implemented in HYPLAS), the
scaling parameter ψ is set to zero and the constraint equation reads simply
∆uT ∆u = l2 . (4.110)
In this case, the constraint equation requires that the Euclidean norm of the converged
incremental displacement be l, i.e. the equilibrium solution at the end of the increment lies
at an intersection between the solution path and a ball of radius l in the space of nodal
displacements (a cylinder in the λ-u space) centred at the equilibrium configuration un of
the beginning of the increment. A graphical representation is shown in Figure 4.8 which
illustrates a system with two degrees of freedom. The possible intersections are denoted A
and B.
(i) Initialise iteration counter, k := 0, and set initial guess for displacement and incremen-
tal load factor
(0) (0) ext
un+1 := un ; λn+1 := λn ; r(0) := f int (un ) − λn f¯
(k)
(ii) Assemble stiffness, KT := KT (un+1 ) [FRONT, ELEIST]
(iii) Set k := k + 1. Solve the linear systems for δu∗ and for the tangential solution, δū
[subroutine FRONT]
KT δu∗ = −r(k−1) ;
ext
KT δū = −f¯
with coefficients defined by (4.117) and choose root δλ(k) according to (4.118)
(v) Apply correction to incremental load factor [ARCLEN]
(k) (k−1)
λn+1 := λn+1 + δλ(k)
arc-length constraint
∆uT ∆u = l
λ
equilibrium
path
λn B
( λ n , un )
0
u2
un A
u1 l
In the iterations (4.111), the arc-length constraint is guaranteed to hold only at the converged
solution whereas, in the non-consistent scheme, the arc-length constraint is enforced at every
iteration. Both iterative schemes lead to identical converged equilibrium solutions {∆λ, u}.
From (4.112)1, we have
δu(k) = δu∗ + δλ δ ū, (4.113)
where δu∗ is the iterative displacement stemming from the Newton–Raphson algorithm for
the standard load controlled scheme:
δu∗ ≡ −K−1
T r
(k−1)
, (4.114)
δ ū ≡ K−1
T f̄
ext
. (4.115)
THE FINITE ELEMENT METHOD IN QUASI - STATIC NONLINEAR SOLID MECHANICS 111
Substitution of (4.113) into the relation ∆u(k) = ∆u(k−1) + δu(k) , followed by the enforce-
ment of the constraint (4.112)2, results in the quadratic equation for the iterative load factor
δλ(k)
2
a δλ(k) + b δλ(k) + c = 0, (4.116)
with coefficients
a = δ ūT δ ū
b = 2(∆u(k−1) + δu∗ )T δ ū (4.117)
∗ T ∗
c = (∆u (k−1)
+ δu ) (∆u (k−1)
+ δu ) − l .
2
(b) Incremental work. Follow the sign of the predictor work increment:
(c) Secant path (Feng et al., 1995, 1996). The sign of δλ(1) is determined as
sign(δλ(1) ) = sign(∆uTn δ ū). (4.123)
112 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Criteria (a) and (c) are implemented in program HYPLAS. Procedure (a) is widely used
in commercial finite element codes and works well in the absence of bifurcations. In the
presence of bifurcations, however, it is known not to be appropriate and fails in most cases.
As pointed out by Crisfield (1991), its ill-conditioned behaviour stems from the fact that
the sign of |KT | changes either when a limit point or when a bifurcation point is passed.
In this case, the predictor cannot distinguish between these two quite different situations,
unless further (usually computationally expensive) analyses are undertaken. In the presence
of a bifurcation, instead of following the current equilibrium path, the solution will oscillate
about the bifurcation point. This property is mathematically proved by Feng et al. (1997).
Procedure (b), on the other hand, is ‘blind’ to bifurcations and can continue to trace an
equilibrium path after passing a bifurcation point. However, this criterion proves ineffective
in the descending branch of the load-deflection curve in ‘snap-back’ problems, where the
predicted positive ‘slope’ will provoke a ‘back tracking’ load increase. One important feature
shared by the criteria (a) and (b) is the fact that they rely exclusively on information relative
to the current equilibrium point (at the beginning of the increment). The decision on the sign
of δλ is made without considering the history of the currently traced equilibrium path. In
situations such as the ones pointed out above, this may result in wrong direction prediction.
In contrast, a key point concerning criterion (c) is the fact that ∆un carries with it information
about the history of the current equilibrium path. The importance of this fact is established in
the discussion that follows, where, in particular, we show by means of geometric arguments
that the secant predictor can easily overcome the problems associated with criteria (a) and (b).
The above inequality is equivalent to (4.123). In essence, a positive product, ∆uTn δ ū,
indicates that the load factor is currently increasing so that the positive sign should be chosen
for the predictor δλ to carry on following the present equilibrium path (Figure 4.10(a)).
Similarly, if ∆uTn δ ū is negative, the load factor is decreasing and the negative sign should be
chosen (Figure 4.10(b)).
THE FINITE ELEMENT METHOD IN QUASI - STATIC NONLINEAR SOLID MECHANICS 113
λ S1 second space of
descending ascending nodal displacements u (λ)
branch branch
-
δu
-
δu
first
ascending S1
branch S2 S2
-
δu
0 ui
Figure 4.9. The tangential solution, δ ū. (Reproduced with permission from On the determination of the
path direction for arc-length methods in the presence of bifurcations and ‘snap-backs’, EA de Souza
Neto and YT Feng, Computer Methods in Applied Mechanics and Engineering, Vol 179, Issue 12
c 1999 Elsevier Science S.A.)
(a) (b)
T - >0
∆u δ u - <0
∆u n δ u
T
n
un -
δu u n-1
∆u n
u n-1 ∆u n S1 S2 u (λ) S1 S2 u (λ)
-
δu un
Figure 4.10. The secant path direction prediction criterion. (a) Load factor is currently increasing:
choose δλ > 0; (b) load factor is currently decreasing: choose δλ < 0. (Reproduced with permission
from On the determination of the path direction for arc-length methods in the presence of bifurcations
and ‘snap-backs’, EA de Souza Neto and YT Feng, Computer Methods in Applied Mechanics and
Engineering, Vol 179, Issue 12 c 1999 Elsevier Science S.A.)
It is important to emphasise that the secant path criterion is insensitive to the presence of
bifurcations and, provided that ∆un is sufficiently small, does not show the basic deficiency
associated with the criterion based on the sign of |KT |. Also, in the descending branch of
‘snap back’ curves, the secant path predictor should indicate that the load is decreasing
(negative δλ), overcoming the problem associated with the predictor incremental work
criterion. Numerical experiments demonstrating the effectiveness of the secant path criterion
in situations where criteria (a) and (b) fail are shown by de Souza Neto and Feng (1999).
Step-size limitation
The need for sufficiently small ∆un has been stressed above as a condition for producing
reliable indications of the ‘forward’ direction. With regard to this aspect, it is important
to recall that a natural limitation on the maximum size of ∆un is already imposed by the
Newton–Raphson algorithm. If the convergence radius of the Newton–Raphson scheme is
smaller than the maximum size of ∆un that produces an accurate direction prediction, then
no further increment size restrictions are introduced by the use of the secant path predictor
criterion. In fact, numerical experience shows that the maximum increment size is usually
dictated by the Newton–Raphson algorithm rather than by the direction predictor criterion.
5 OVERVIEW OF THE PROGRAM
STRUCTURE
I N Chapter 4, we have described a general strategy for the finite element simulation of linear
and nonlinear solid mechanics problems, including generic path-dependent constitutive
models and finite deformations and strains. Most procedures described there are incorporated
in the standard version of the program HYPLAS that accompanies this book. Throughout their
description, whenever appropriate, reference has been made to the subroutines of HYPLAS
where some of the most relevant computational operations associated with such numerical
procedures are carried out.
In the present chapter we provide a more thorough description of the HYPLAS program,
with emphasis on its structure. The depth at which details of the code are described here is
only what the authors believe to be sufficient to help readers find their way through HYPLAS.
It is by no means intended to be a comprehensive guide to the code. The present chapter is
particularly relevant for researchers who wish to understand and/or modify the basic code
by including new procedures, material models, finite elements, etc. Those who are only
interested in the theoretical and numerical aspects of nonlinear solid mechanics may skip
this chapter and move on to Part Two of the book.
5.1. Introduction
HYPLAS is a finite element code for implicit small and large strain analysis of hyperelastic
and elastoplastic solids in plane stress, plane strain and axisymmetric states.
5.1.1. OBJECTIVES
The main purpose of HYPLAS is to illustrate the computational implementation of numerical
procedures described throughout this book. The general framework adopted is that described
in Chapter 4. It comprises:
(i) a general displacement-based incremental finite element procedure;
(ii) iterative schemes (e.g. Newton–Raphson) for the solution of nonlinear incremental
finite element equations;
(iii) an arc-length scheme for problems involving structural instability.
Within this framework, numerous (elastic and elastoplastic) material models have been
incorporated. The theory and numerical methods underlying the implementation of each
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
116 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
material model are thoroughly discussed in Parts Two and Three of this book. Part Two
focuses on the infinitesimal theory and Part Three on the finite strain range. The reader will
find there a detailed description of the associated procedures including, in many cases, the
pseudo-code and the corresponding FORTRAN source code of the relevant routines.
Clarity of coding
When developing a computer code, programmers are frequently faced with the dilemma:
clarity versus efficiency. Finite element codes are no exception. We do not mean that clarity
of coding and computational efficiency are opposing concepts but, as a general rule, excessive
optimisation of computer operations tends to result in cumbersome, difficult-to-follow source
codes.
In view of the educational purposes of HYPLAS, the programming philosophy adopted in
its development has been biased towards code clarity and modularity. This is particularly true
for the numerical procedures whose theory is described in this text.
5.1.3. PORTABILITY
The code of HYPLAS is entirely written in (almost) standard FORTRAN 77 programming
language (ANSI, 1978). The existing exceptions to the FORTRAN 77 standards are known
to be accepted by most currently available compilers.
OVERVIEW OF THE PROGRAM STRUCTURE 117
RSTCHK
INLOAD RSTART
carry on to load
incrementation loop
(ii) Solution-related (variable) data. Data that change during the solution process. Typical
solution-related data are nodal displacements, current nodal coordinates (changes in
large deformation analysis only), stresses and state variables in general at Gauss points,
arrays storing element internal forces, etc.
The global database is defined in the include file GLBDBASE.INC. The reader is referred to
the comment lines of this file for a description of the data stored in the arrays and variables
of the global database. These will be further explained in this chapter only when convenient
for a better understanding of the program. All arrays of the database have fixed dimension
and are grouped in a number of COMMON blocks. File GLBDBASE.INC is included in all
routines that require data stored in any of the COMMON blocks of the global database. The
array-dimensioning parameters defining the dimensions of the global database arrays are set
in the include files MAXDIM.INC, MATERIAL.INC and ELEMENTS.INC (refer to their
comment lines). File MAXDIM.INC sets the parameters that define the maximum admissible
problem size that can be analysed. Files MATERIAL.INC and ELEMENTS.INC set,
respectively, maximum array dimensions required by the currently implemented material
models and element types. These are discussed further in Sections 5.6 and 5.7.
Variables and arrays (or array components) containing data of type (i) are set once and for
all in subroutines INDATA, INLOAD and ININCR, according to the input data file. If in restart
mode, the data otherwise read and set in INDATA and INLOAD are retrieved from the input
restart file by subroutine RSTART. In the data input and initialisation phase of HYPLAS, data
OVERVIEW OF THE PROGRAM STRUCTURE 119
of type (ii) are initialised in subroutine INITIA. If in restart mode, current values at the last
equilibrium (converged) solution are retrieved instead from the input restart file in RSTART.
RLOAD.
where the input data ρ, g and g are, respectively, the material density, the gravity
acceleration and the unit vector defining the direction of the gravity acceleration.
• Distributed load at element boundaries (edges in two dimensions). Assembly of
element vectors
ngausb
Nb t da ≈
T
wi j(ξi ) Nb (ξi )T t, (5.2)
(e)
∂Ωt i=1
(e)
where ∂Ωt is the loaded edge of the boundary of element e, Nb is the boundary shape
function associated with the loaded boundary and ngausb is the number of Gauss points
used for numerical integration over the element boundary. Array t is the array of nodal
pressure vectors of the loaded edge. The above numerical integration is carried out for
each loaded edge of the element.
Some checks are also made in INLOAD to detect incorrect data.
120 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
reset converged
problem variables
SWITCH
increment external
load vector
increment cutting INCREM
split current increment into two
and restart from previous
converged solution assemble and solve
linear equation system
FRONT
failure in
system solution
UPCONF
iteration NO
converged?
YES
maximum number
of iterations exceeded
output converged
results
OUTPUT, RSTART
convergence in a desired number of iterations, nitdes (prescribed in the input data file). With
nitact and lcurr denoting, respectively, the number of iterations for convergence and the arc
length used in the current step, the new length lnext to be used in the next step is set (in
subroutine LENGTH) as
nitdes
lnext = lcurr . (5.3)
nitact
The new length is not allowed to be larger than a maximum value lmax .
122 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
reset converged
problem variables
arc-length equation
update displacements
& coordinates
UPCONF
iteration NO
converged?
YES
maximum number
of iterations exceeded
Start-up length.
Obviously, at the beginning of the first iteration of the very first load increment, the above
length adjustment cannot be applied. Since it is not normally easy to imagine the magnitude
of l that would be reasonable for a particular problem, the start-up length, l0 , is calculated
(in ARCLEN) by specifying (in the input data file) an initial load increment factor, ∆λ0 , and
setting
l0 = ∆λ20 δūT ū (5.4)
OVERVIEW OF THE PROGRAM STRUCTURE 123
where ū is the initial tangential solution (see item (iii) of Box 4.4). The maximum length
lmax is also set at this stage as
lmax = p l0 , (5.5)
where p is the maximum arc-length parameter (defined in the input data file).
Predictor solution.
The present implementation allows the choice of the sign of the predictor solution (refer to
Section 4.4.3) to be made by two criteria: the stiffness determinant sign or the secant path
predictor. The predictor solution is computed in subroutine ARCLEN.
• The load increment is too large. The total prescribed load (with fixed increments
(0)
option) is beyond the limit load of the structure or the initial guess, un+1 (refer to
Box 4.2), falls outside the convergence radius of the iterative solution method. In
this case the residual norm will either diverge or not converge within a reasonable
number of iterations. A possible alternative (not available in HYPLAS) to increase the
convergence radius is the incorporation of line-searches (Crisfield, 1991) within the
equilibrium iterations.
• The linear equation system cannot be solved in FRONT due to a zero pivot in the stiffness
matrix. This may occur for material models whose tangent operator may become
singular under certain states of stress.
124 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
• For arc-length control only. The arc-length constraint equation (4.116), solved in
routine ARCLEN, may have no real roots. This problem is also generally overcome by
reducing the increment size (the arc length, l, in this case) and restarting the current
step with the reduced value of l.
KT δu = −r (5.6)
is solved for δu. If MODE=2, the tangential solution of the arc-length method, δū, is obtained
instead by solving
ext
KT δū = −f¯ . (5.7)
If MODE=3, then both solutions are obtained. The variable KRESL controls whether or not a
new stiffness is computed. If KRESL=1, then a new stiffness is required. When the standard
Newton–Raphson algorithm is used, KRESL is always set to 1. If KRESL=2, then the stiffness
of the previous iteration is to be reused in the current iteration and only the load term will be
reduced (this situation will occur, for instance, when the initial stiffness method is used).
The global arrays of HYPLAS used exclusively by the frontal solver are not included in
the global database (file GLBDBASE.INC). They are grouped in the COMMON block named
FRONTA.
This applies to all arrays of COMMON block STATE (see comments in file GLBDBASE.INC).
Switching between current and last converged data is managed by subroutine SWITCH as
follows (refer to source code). Before a material-specific state updating procedure is called
(this is required during the calculation of the internal force vector in the element class-specific
routines called from ELEIIF), the previous converged values are assigned to the positions
corresponding to the current values. These are passed on to the state-updating procedure
which returns (in the same position) the updated (or current) values. When the iterative
procedure converges, the converged values need updating. The current values (which have
converged at this point) are then assigned to the corresponding array positions that store
converged data. Appropriate switching/resetting is also needed when increment cutting is
activated. Most arrays of COMMON block STATE are switched in material-specific routines
(see Section 5.7). In large deformation analysis, array COORD (in COMMON block MESH) of
nodal coordinates also needs switching between current and last converged values. Other
arrays, such as the total, incremental and iterative nodal displacement vectors also require
switching/resetting. The type of switching operation carried out by SWITCH depends on the
integer argument MODE.
ELEIIF
quantities (strains or deformation
ELEMENT LEVEL
state update
element level to update stress and
MATISU
MATERIAL LEVEL
state update
linear elastic material
SUEL
state update
Tresca model
SUTR
... state update
Ogden model
SUOGD
element internal forces need to be computed within the equilibrium iteration loop, the main
program calls subroutine INTFOR (refer to the flowcharts of Figures 5.2 and 5.3). INTFOR
loops over all elements of the mesh and calls the element interface routine for internal force
calculation ELEIIF (see call tree of 5.4). Before ELEIIF is called, only operations related
to the general incremental procedure, with no distinction between different types of finite
element or material, are carried out.
ngausp
f int
(e) := wi j(ξi ) B(ξi )T σ|ξi . (5.8)
i=1
The Gaussian quadrature weights, wi , the Gauss points positions, ξi and the way in which
the Jacobian, j, and the strain-displacement matrix, B, are computed depend only on
the particular element being used. More precisely, they depend on the element class to
which the adopted element belongs.† The calculation of these quantities is independent of
the underlying material model. The stress array, on the other hand, is the outcome of a
state update procedure which depends exclusively on the adopted material model and the
† The definition of element class is given in Section 5.6.
OVERVIEW OF THE PROGRAM STRUCTURE 127
algorithm used to integrate its constitutive equations (if the material is path-dependent). In the
general path-dependent case, under infinitesimal strains, the state update algorithm defines an
incremental constitutive function
within the typical interval [tn , tn+1 ]. In the above, αn is the set of internal variables at the
converged state at tn and εn+1 is the given strain at the end of the interval. The incremental
constitutive function can be expressed in the equivalent form‡
and the set αn which now contains the converged internal state variables at time station tn
and the strain εn .§ In HYPLAS, the incremental strain is the actual argument passed into the
state update interface in the infinitesimal case. In finite strain analysis, the corresponding
argument is the incremental deformation gradient,
σ = σ̂(αn , F ∆ ). (5.13)
Based on the above comments, the computation of the internal force vector can be split
into two well-defined levels:
1. The element level. Here, all element-related quantities (wi , j, B) are computed together
with the essential kinematic variable ∆ε (F ∆ in large strains). The incremental strain
(or deformation gradient) is passed on to the material level which returns the updated
stress. With the updated stress at hand, the internal force vector is assembled according
to (5.8).
2. The material level. This is the lowest layer of code. It receives the incremental strain
(or deformation gradient) from the element level, retrieves the relevant converged state
variables αn stored in appropriate arrays and updates the stress by using the material
model-specific subroutine that defines the incremental function (5.10). The updated
stress is returned to the element level.
‡ Note that, formally, the function on the right-hand side of (5.9) differs from that on the right-hand side of (5.10)
in that it has different arguments. Throughout the text, however, we shall employ the symbol σ̂ to denote algorithmic
(incremental) constitutive functions for the stress tensor in general, regardless of their list of arguments.
§ As we shall see later, in computations with elastoplastic/viscoplastic models, it will be convenient to have the
HYPLAS implementation
In HYPLAS, the element level starts in subroutine ELEIIF. This routine identifies the element
class and calls the appropriate element class-specific internal force computation routine.
The element class-specific routine (e.g. IFSTD2 for standard two-dimensional isoparametric
elements), in turn, calls the material interface for state update, MATISU. The material
level starts here. MATISU identifies the material model/algorithm in question and calls the
corresponding material-specific state updating procedure (e.g. routine SUTR for the Tresca
elastoplastic model).
As we shall see in the following sections, the modularity concept discussed here is applied
not only to the internal force computation, but also to the evaluation of the element tangent
stiffness and some input/output operations. It makes the program particularly suitable for the
incorporation of new elements and material models and/or modification of the existing ones.
that contain, respectively, integer and real element properties. Typical element properties are
listed below:
• Integer element properties. Number of nodes, number of degrees of freedom, number
of Gauss points, number of boundaries, node ordering on boundaries, etc.
• Real element properties. Gauss quadrature weights and positions for domain and
boundary integration, extrapolation matrices to extrapolate Gauss point values to nodes
(used for output purposes only), etc.
OVERVIEW OF THE PROGRAM STRUCTURE 129
Each group of elements in the structure (as defined in the input data file) is assigned one
element type. The properties of the element type assigned to an element group IGRUP are
stored in the columns
of the element properties arrays. These data are used by the corresponding element class-
specific subroutines for computation of internal force and tangent stiffness matrices. Element
properties are also used in the evaluation of the external load vector carried out in INLOAD.
Element properties are assigned to the arrays of COMMON block ELEMEN during the input phase
of the program (refer to source code of subroutine INDATA). To read the relevant data from
the input data file and assign all necessary properties to IELPRP and RELPRP, each element
type uses its own routine. Such subroutines are named following the convention:
RSxxxx.
For instance, subroutines RST3 and RSQ4 read and set data, respectively, for the isoparametric
three-noded triangle and four-noded bi-linear quadrilateral. These routines are called only
once during the program execution.
The modularisation of elements in the internal force computation has been discussed in
Section 5.5.1. Subroutine ELEIIF in this case is the element interface routine that controls the
evaluation of the element internal force vector. The basic task of the element interface here is
to identify the class of the element whose internal force is required and call the appropriate
class-specific internal force evaluation routine. For standard isoparametric two-dimensional
elements, the class-specific routine is IFSTD2. For F -bar elements (discussed in Chapter 15),
the corresponding class-specific routine is IFFBA2.
The evaluation of the element tangent stiffness matrix is also carried out in a modular
structure completely analogous to that shown in Figure 5.4. The corresponding call tree is
shown in Figure 5.5. At the element level, we have the interface routine ELEIST (called
from FRONT) whose task is to identify element classes and call the class-specific routine for
computation of the element tangent stiffness. The class-specific routines for standard and
F -bar elements are STSTD2 and STFBA2, respectively.
In this section, we provide a summary of the basic path to be followed in order to include a
new finite element into HYPLAS.
130 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
stiffness computation
tangent operator to assemble element
ELEIST
to compute tangent operator. Use
quantities (strains or deformation
ELEMENT LEVEL
STSTD2 STFBA2
MATICT
MATERIAL LEVEL
Assume that we want to include a new element type belonging to an existing element class.
In this case, basic class-related internal force and stiffness computation routines already exist
and do not need to be changed. The implementation of the new element requires coding of
the following new element type-specific procedures:
2. The routine for evaluation of shape functions and shape function derivatives (SFxxxx).
These routines have not been discussed above. They are called by SHPFUN which is
used by element classes available in the program. Subroutine SHPFUN is the interface
for shape function/shape function derivatives computation. It identifies the element type
and calls the corresponding element type-specific shape function/derivative routine
(refer to the source code of subroutine SFT3 or SFQ4). New classes of element that
do not use this structure will not require shape function routines coded in this way.
In addition, a new element-type identification parameter has to be added to the include file
ELEMENTS.INC. For the interested reader a good exercise could be to include, say, the
nine-noded Lagrangian isoparametric quadrilateral.
OVERVIEW OF THE PROGRAM STRUCTURE 131
of COMMON block MATERL (see include file GLBDBASE.INC). The data stored in these arrays
include all parameters required by the continuum constitutive model and, if required, may
also contain other parameters related to the specific numerical algorithm adopted to integrate
the constitutive equations of the model. Examples of integer and real material properties are:
• Real material properties. Usual properties such as Young’s modulus, Poisson’s ratio,
parameters defining hardening curves of elastoplastic materials, etc.
The situation here is analogous to the storage of element properties discussed in Section 5.6.1.
Each group of elements defined by the user in the input data file is assigned one material type
132 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
with one set of material properties. The material properties of a generic group IGRUP are
stored in the columns
RDxxxx.
For example, all data for the von Mises piecewise linear isotropic hardening model are read
and set in subroutine RDVM. For the Ogden hyperelastic model, the corresponding routine is
RDOGD.
The arrays of COMMON block STATE store both current and last converged values of the
corresponding variables. The array positions having the last index equal to 1, such as
STRSG( , ,1)
for the Gauss point stresses, store the current values. Positions with the last index equal to 2,
such as
STRSG( , ,2),
contain the last equilibrium converged values.
In the finite strain case, the above functions are defined in terms of the incremental
deformation gradient, F ∆ , rather than ∆ε. For path-dependent constitutive models, the
format of the incremental functions depends not only on the material model in question but
also on the algorithm used to integrate its constitutive equations.
In the HYPLAS program, each material model implementation has a material-specific state-
updating procedure that defines a pair of functions {σ̂, α̂}. The state-updating procedures
(subroutines) are named according to the rule
SUxxxx.
For the von Mises elastoplastic model, for instance, we have the subroutines SUVM (general
case) and SUVMPS (plane stress algorithm).
SWxxxx.
For the von Mises and Ogden models we have, respectively, the routines SWVM and SWOGD.
CSTxxx
for consistent spatial tangent moduli (defined by (4.98)) needed in finite strain analyses. In
this case, the material-specific routine for the von Mises model is named CTVM. The Ogden
hyperelastic model uses subroutine CSTOGD.
MATICT.
For instance, the von Mises effective stress is important for metals but may be of no relevance
for concrete failure models. Thus, the use of material-specific routines at this stage is
convenient in order to print out only relevant results and keep, at the same time, a clean
and modular structure. These routines are named in HYPLAS according to the convention
ORxxxx.
The output routine for the von Mises model, for instance, is named ORVM.
MATIOR.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
140 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
described in Section 6.4. Plastic flow rules and hardening laws are addressed, respectively, in
Sections 6.5 and 6.6.
1. The existence of an elastic domain, i.e. a range of stresses within which the behaviour
of the material can be considered as purely elastic, without evolution of permanent
(plastic) strains. The elastic domain is delimited by the so-called yield stress. In
Figure 6.1, segments O0 Y0 and O1 Y1 define the elastic domain at two different states.
The associated yield stresses correspond to points Y0 and Y1 .
2. If the material is further loaded at the yield stress, then plastic yielding (or plastic flow),
i.e. evolution of plastic strains, takes place.
3. Accompanying the evolution of the plastic strain, an evolution of the yield stress itself
is also observed (note that the yield stresses corresponding to points Y0 and Y1 are
different). This phenomenon is known as hardening.
THE MATHEMATICAL THEORY OF PLASTICITY 141
σ1 Z1
Z0
σ0
Y1
Y0
O0 O1 ε
εp
It is emphasised that the above properties can be observed not only in metals but also in
a wide variety of materials such as concrete, rocks, soils and many others. Obviously, the
microscopic mechanisms that give rise to these common phenomenological characteristics
can be completely distinct for different types of material. It is also important to note that,
according to the type of material, different experimental procedures may be required for
the verification of such properties. For instance, in materials such as soils, which typically
cannot resist tensile stresses, uniaxial tension tests do not make physical sense. In this case,
experiments such as triaxial shear tests, in which the sides of the specimen are subjected to a
confining hydrostatic pressure prior to the application of longitudinal compression, are more
appropriate.
The object of the mathematical theory of plasticity is to provide continuum constitutive
models capable of describing (qualitatively and quantitavely) with sufficient accuracy the
phenomenological behaviour of materials that possess the characteristics discussed in the
above.
Z1
Y1 Z0
σ0
slope E ep
Y0
slope E
σ
O1
O0
ε
εp
the beginning of unloading and the onset of plastic yielding upon subsequent reloading, are
assumed to coincide. The transition between the elastic region and the elastoplastic regime
is now clearly marked by a non-smooth change of slope (points Y0 and Y1 ). During plastic
yielding, the stress–strain curve always follows the path defined by O0 Y0 Y1 Z1 . This path is
normally referred to as the virgin curve and is obtained by a continuous monotonic loading
from the initial unstressed state O0 .
Under the above assumptions, after being monotonically loaded from the initial unstressed
state to the stress level σ0 , the behaviour of the bar between states O1 and Y1 is considered to
be linear elastic, with constant plastic strain, εp , and yield limit, σ0 . Thus, within the segment
O1 Y1 , the uniaxial stress corresponding to a configuration with total strain ε is given by
σ = E (ε − εp ), (6.1)
where E denotes the Young’s modulus of the material of the bar. Note that the difference
between the total strain and the current plastic strain, ε − εp , is fully reversible; that is,
upon complete unloading of the bar, ε − εp is fully recovered without further evolution of
plastic strains. This motivates the additive decomposition of the axial strain described in the
following section.
tension
σy
elastic
domain
0 ε
compression
− σy
Expressions (6.8), (6.12) and (6.13) define the so-called loading/unloading conditions of
the elasticplastic model; that is, the constraints
Φ ≤ 0, γ̇ ≥ 0, γ̇Φ = 0, (6.16)
Finally, the complete characterisation of the uniaxial model is achieved with the introduction
of the hardening law. As remarked in item 3 of Section 6.1, an evolution of the yield stress
accompanies the evolution of the plastic strain. This phenomenon, known as hardening, can
be incorporated into the uniaxial model simply by assuming that, in the definition (6.5) of Φ,
the yield stress σy is a given function
σy = σy (ε̄p ) (6.17)
of the accumulated axial plastic strain, ε̄p . The accumulated axial plastic strain is defined as
t
ε̄ ≡
p
|ε̇p | dt, (6.18)
0
thus ensuring that both tensile and compressive plastic straining contribute to ε̄p . Clearly, in
a monotonic tensile test we have
ε̄p = εp , (6.19)
The curve defined by the hardening function σy (ε̄p ) is usually referred to as the hardening
curve (Figure 6.4).
From the definition of ε̄p , it follows that its evolution law is given by
σy
σy ( ε p )
hardening slope, H
σA
0
εp
Figure 6.4. One-dimensional model. Hardening curve.
ε = εe + εp
ε̄˙p = γ̇
6. Loading/unloading criterion
Φ ≤ 0, γ̇ ≥ 0, γ̇Φ = 0
• an elastic law;
• a plastic flow rule defining the evolution of the plastic strain; and
The generalisation of these concepts for application in two- and three-dimensional situations
is described in this section.
ε = εe + ε p . (6.34)
The tensors εe and εp are known, respectively, as the elastic strain tensor and the plastic
strain tensor. The corresponding rate form of the additive split reads
ψ(ε, εp , α),
of the total strain, the plastic strain (taken as an internal variable) and a set α of internal
variables associated with the phenomenon of hardening. It is usual to assume that the free
THE MATHEMATICAL THEORY OF PLASTICITY 149
ψ(ε, εp , α) = ψ e (ε − εp ) + ψ p (α)
= ψ e (εe ) + ψ p (α) (6.37)
into a sum of an elastic contribution, ψ e , whose dependence upon strains and internal
variables appears only through the elastic strain, and a contribution due to hardening, ψ p .
Following the above expression for the free energy, the Clausius–Duhem inequality reads
∂ψ e
σ − ρ̄ : ε̇e + σ : ε̇p − A ∗ α̇ ≥ 0, (6.38)
∂εe
where
A ≡ ρ̄ ∂ψ p/∂α (6.39)
is the hardening thermodynamical force and we note that −σ is the thermodynamical force
associated with the plastic strain while the symbol * indicates the appropriate product
between A and α̇. The above inequality implies a general elastic law of the form
∂ψ e
σ = ρ̄ , (6.40)
∂εe
so that the requirement of non-negative dissipation can be reduced to
ρ̄ ψ e (εe ) = 1
2 εe : De : εe
= G εed : εed + 1
2 K (εev )2 (6.43)
where De is the standard isotropic elasticity tensor and G and K are, respectively the
shear and bulk moduli. The tensor εed is the deviatoric component of the elastic strain and
εev ≡ tr[εe ] is the volumetric elastic strain. Thus, the general counterpart of uniaxial elastic
law (6.4) is given by
σ = De : εe
= 2G εed + K εev I. (6.44)
150 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
ε = εe + εp
or
ε̇ = ε̇e + ε̇p , ε(t0 ) = εe (t0 ) + εp (t0 )
2. Free-energy function
ψ = ψ(εe , α)
where α is a set of hardening internal variables
3. Constitutive equation for σ and hardening thermodynamic forces A
∂ψ ∂ψ
σ = ρ̄ , A = ρ̄
∂ εe ∂α
4. Yield function
Φ = Φ(σ, A)
5. Plastic flow rule and hardening law
ε̇p = γ̇ N(σ, A)
α̇ = γ̇ H(σ, A)
6. Loading/unloading criterion
Φ ≤ 0, γ̇ ≥ 0, γ̇Φ = 0
Ψ = Ψ(σ, A) (6.54)
Ψ(0, 0) = 0. (6.57)
152 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
These restrictions ensure that the dissipation inequality (6.41) is satisfied a priori by the
evolution equations (6.49) and (6.50).
Φ̇ γ̇ = 0, (6.59)
Φ̇ = 0 (6.60)
This, together with the definition of A in terms of the free-energy potential (refer to
expression (6.39)) and the evolution law (6.50), allow us to write (6.61) equivalently as
∂Φ ∂Φ ∂ 2 ψp
Φ̇ = : De : (ε̇ − ε̇p ) + ∗ ρ̄ ∗ α̇.
∂σ ∂A ∂α2
2 p
∂Φ ∂Φ ∂ ψ
= : De : (ε̇ − γ̇ N ) + γ̇ ∗ ρ̄ ∗ H. (6.63)
∂σ ∂A ∂α2
Finally, the above expression and the consistency condition (6.60) lead to the following closed
formula for the plastic multiplier
∂Φ/∂σ : De : ε̇
γ̇ = . (6.64)
∂Φ/∂σ : D : N − ∂Φ/∂A ∗ ρ̄∂ 2 ψ p /∂α2 ∗ H
e
THE MATHEMATICAL THEORY OF PLASTICITY 153
σ̇ = De : ε̇. (6.65)
Under plastic flow, the corresponding rate relation can be obtained by introducing expres-
sion (6.64) into (6.62). The rate equation reads
The fourth-order tensor Dep is the multidimensional generalisation of the scalar modulus
E ep associated with the slope of the uniaxial stress–strain curve under plastic flow. In the
computational plasticity literature, Dep is frequently referred to as the continuum elastoplastic
tangent operator.
Remark 6.1 (The symmetry of Dep ). Note that if the plastic flow rule is associative, i.e.
if N ≡ ∂Φ/∂σ, then the continuum elastoplastic tangent operator is symmetric. For models
with non-associative plastic flow, Dep is generally unsymmetric.
V
isosurfaces of
<A
normal to the iso-surfaces of function Ψ in the space of stresses (with fixed A). A schematic
representation of N in this case is shown in Figure 6.5. The generalised modulus, H, can be
interpreted in a completely analogous way.
The requirement of differentiability of the flow potential is, however, too restrictive and
many practical plasticity models are based on the use of a non-differentiable Ψ. Specific
examples are given later in this chapter. For a more comprehensive account of such theories
the reader is referred to Duvaut and Lions (1976), Eve et al. (1990) and Han and Reddy
(1999). In such cases, the function Ψ is called a pseudo-potential or generalised potential
and the formulation of the evolution laws for the internal variables can be dealt with by
introducing the concept of subdifferential sets, which generalises the classical definition of
derivative.†
If the set ∂y is not empty at x̄, the function y is said to be subdifferentiable at x̄. The elements
of ∂y are called subgradients of y. If the function y is differentiable, then the subdifferential
contains a unique subgradient which coincides with the derivative of y,
dy
∂y = . (6.70)
dx
Rockafellar (1970), Part V, for a detailed account of the theory of subdifferentiable functions.
THE MATHEMATICAL THEORY OF PLASTICITY 155
the subdifferential
of y at x
s
y(x) − y(x )
1
s⋅(x −x )
x x
N ∈ ∂σ Ψ, (6.71)
i.e. the flow vector N is now assumed to be a subgradient of Ψ. Analogously, the evolution
law (6.50) for α can be generalised with the replacement of the definition (6.56) by
H ∈ −∂A Ψ. (6.72)
At this point, it should be remarked that differentiability of Ψ with respect to the stress
tensor is violated for some very basic plasticity models, such as the Tresca, Mohr–Coulomb
and Drucker–Prager theories to be seen later. Therefore, the concepts of subgradient and
subdifferential sets introduced above are important in the formulation of evolution laws
for εp .
An alternative definition of the plastic flow rule with non-smooth potentials, which
incorporates a wide class of models, is obtained as follows. Firstly assume that a finite
number, n, of distinct normals (N1 , N2 , . . . , Nn ) is defined at a generic singular point of an
isosurface of Ψ. In this case, any subgradient of Ψ can be written as a linear combination
c1 N1 + c2 N2 + · · · + cn Nn ,
156 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
w<
V
the subdifferential
set of < at V Hp
N1
N2
V
isosurfaces of
<A
with non-negative coefficients c1 , c2 , . . . , cn .‡ Based on this observation, the flow rule (6.49)
can be generalised as
n
ε̇p = γ̇i Ni , (6.73)
i=1
with all n plastic multipliers required to be non-negative
γ̇i ≥ 0, i = 1, . . . , n. (6.74)
The generalisation of the plastic flow law, in this format, was originally proposed by Koiter
(1953).
Multisurface models
The above concepts are particularly useful in defining evolution laws for multisurface
plasticity models. In a generic multisurface model, the elastic domain is bound by a set of
n surfaces in the space of stresses which intersect in a non-smooth fashion. In this case, n
yield functions (Φi , i = 1, . . . , n) are defined so that each bounding surface is given by an
equation
Φi (σ, A) = 0. (6.75)
The elastic domain in this case reads
and the yield surface, i.e. the boundary of E, is the set of all stresses such that Φi (σ, A) = 0
for at least one i and Φj (σ, A) ≤ 0 for all other indices j
= i.
‡ It should be emphasised that this representation is not valid for certain types of singularity where the
Assuming associativity (Ψ ≡ Φ), the situation discussed previously, where the subgradient
of the flow potential is a linear combination of a finite number of normals, is recovered. Thus,
the plastic flow rule can be written in the general form (6.73) with the normals being defined
here as
∂Φi
Ni = . (6.77)
∂σ
In the present case, the standard loading/unloading criterion (6.53) is replaced by the
generalisation
Φi ≤ 0, γ̇i ≥ 0, Φi γ̇i = 0, (6.78)
which must hold for each i = 1, . . . , n. Note that summation on repeated indices is not
implied in the above law.
3
σ= σi ei ⊗ ei , (6.79)
i=1
where σi are the principal stresses and ei the associated unit eigenvectors, and let σmax and
σmin be, respectively, the maximum and minimum principal stresses
σmax = max(σ1 , σ2 , σ3 );
(6.80)
σmin = min(σ1 , σ2 , σ3 ).
According to the Tresca criterion, the onset of plastic yielding is defined by the condition
1
2 (σmax − σmin ) = τy (α), (6.82)
where τy is the shear yield stress, here assumed to be a function of a hardening internal
variable, α, to be defined later. The shear yield stress is the yield limit under a state of pure
shear.
158 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
In view of (6.82), the yield function associated with the Tresca yield criterion can be
represented as
Φ(σ) = 12 (σmax − σmin ) − τy (α), (6.83)
with the onset of yielding characterised by Φ = 0. Alternatively, the Tresca yield function
may be defined as
Φ(σ) = (σmax − σmin ) − σy (α), (6.84)
where σy is the uniaxial yield stress
σy = 2 τy , (6.85)
that is, it is the stress level at which plastic yielding begins under uniaxial stress conditions.
That σy is indeed the uniaxial yield stress for the Tresca theory can be established by noting
that, when plastic yielding begins under uniaxial stress conditions, we have
The substitution of the above into (6.82) gives (6.85). The elastic domain for the Tresca
criterion can be defined as
E = {σ | Φ(σ, σy ) < 0}. (6.87)
Pressure-insensitivity
Due to its definition exclusively in terms of shear stress, the Tresca criterion is pressure
insensitive, that is, the hydrostatic pressure component,
p≡ 1
3 tr[σ] = 1
3 (σ1 + σ2 + σ3 ), (6.88)
of the stress tensor does not affect yielding. Indeed, note that the superposition of an arbitrary
pressure, p∗ , on the stress tensor does not affect the value of the Tresca yield function
We remark that the von Mises criterion described in Section 6.4.2 below is also pressure-
insensitive. This property is particularly relevant in the modelling of metals as, for these
materials, the influence of the hydrostatic stress on yielding is usually negligible in practice.
Isotropy
One very important aspect of the Tresca criterion is its isotropy (a property shared by the
von Mises, Mohr–Coulomb and Drucker–Prager criteria described in the following sections).
Note that, since Φ in (6.83) or (6.84) is defined as a function of the principal stresses,
the Tresca yield function is an isotropic function of the stress tensor (refer to Section A.1,
page 731, for the definition of isotropic scalar functions of a symmetric tensor), i.e. it satisfies
for all rotations Q; that is, rotations of the state of stress do not affect the value of the yield
function.
At this point, it is convenient to introduce the following definition: A plastic yield criterion
is said to be isotropic if it is defined in terms of an isotropic yield function of the stress tensor.
THE MATHEMATICAL THEORY OF PLASTICITY 159
Graphical representation
Since any isotropic scalar function of a symmetric tensor can be described as a function of
the principal values of its argument, it follows that any iso-surface (i.e. any subset of the
function domain with fixed function value) of such functions can be graphically represented
as a surface in the space of principal values of the argument. This allows, in particular, the
yield surface (refer to expression (6.48), page 150) of any isotropic yield criterion to be
represented in a particularly simple and useful format as a three-dimensional surface in the
space of principal stresses.
− √3 p
σ3
Tresca
von Mises
σ2
σ1
Figure 6.8. The Tresca and von Mises yield surfaces in principal stress space.
(a) (b)
V3
V3
tic
sta
dro
hy axis
pla
ne von Mises
SŦ
0
V2
Tresca
V1 V2
V1
Figure 6.9. (a) The π-plane in principal stress space and, (b) the π-plane representation of the Tresca
and von Mises yield surfaces.
160 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
In principal stress space, the Tresca yield surface, i.e. the set of stresses for which Φ = 0,
is graphically represented by the surface of an infinite hexagonal prism with axis coinciding
with the hydrostatic line (also known as the space diagonal), defined by σ1 = σ2 = σ3 . This
is illustrated in Figure 6.8. The elastic domain (for which Φ < 0) corresponds to the interior
of the prism. Due to the assumed insensitivity to pressure, a further simplification in the
representation of the yield surface is possible in this case. The Tresca yield surface may
be represented, without loss of generality, by its projection on the subspace of stresses
with zero hydrostatic pressure component (σ1 + σ2 + σ3 = 0). This subspace is called the
deviatoric plane, also referred to as the π-plane. It is graphically illustrated in Figure 6.9(a).
Figure 6.9(b) shows the π-plane projection of the Tresca yield surface.
Multisurface representation
Equivalently to the above representation, the Tresca yield criterion can be expressed by means
of the following six yield functions
Φ1 (σ, σy ) = σ1 − σ3 − σy
Φ2 (σ, σy ) = σ2 − σ3 − σy
Φ3 (σ, σy ) = σ2 − σ1 − σy
(6.91)
Φ4 (σ, σy ) = σ3 − σ1 − σy
Φ5 (σ, σy ) = σ3 − σ2 − σy
Φ6 (σ, σy ) = σ1 − σ2 − σy ,
Definitions (6.87) and (6.93) are completely equivalent. The yield surface – the boundary of
E – is defined in this case as the set of stresses for which Φi (σ, σy ) = 0 for at least one i with
Φj (σ, σy ) ≤ 0 for j
= i.
Invariant representation
Alternatively to the representations discussed above, it is also possible to describe the yield
locus of the Tresca criterion in terms of stress invariants. In the invariant representation,
proposed by Nayak and Zienkiewicz (1972) (see also Owen and Hinton 1980, and Crisfield
1997), the yield function assumes the format
Φ = 2 J2 cos θ − σy , (6.94)
J2 ≡ −I2 (s) = 1
2 tr[s2 ] = 1
2 s:s= 1
2 s2 . (6.95)
THE MATHEMATICAL THEORY OF PLASTICITY 161
Φ1 = 0
Φ6 = 0
Φ2 = 0
σ1 Φ5 = 0
Φ4 = 0
Φ3 = 0
σ3
σ2
Figure 6.10. The Tresca criterion. Multisurface representation in principal stress space.
27) for i = 1, 2, 3 and taking into account the fact that I1 (s) = 0 (s is a traceless tensor)
equation (2.73) (page
and that tr(S )3 = i s3i for any symmetric tensor S.
162 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
we have, s = σ and
J2 = τ 2 . (6.105)
Thus, a yield function for the von Mises criterion can be defined as
Φ(σ) = J2 (s(σ)) − τy , (6.106)
√
where τy ≡ R is the shear yield stress. Let us now consider a state of uniaxial stress:
σ 0 0
[σ] = 0 0 0. (6.107)
0 0 0
THE MATHEMATICAL THEORY OF PLASTICITY 163
is termed the von Mises effective or equivalent stress. The uniaxial and shear yield stresses
for the von Mises criterion are related by
√
σy = 3 τy . (6.112)
Note that this relation differs from that of the Tresca criterion given by (6.85). Obviously, due
to its definition in terms of an invariant of the stress tensor, the von Mises yield function is an
isotropic function of σ.
The von Mises and Tresca criteria may be set to agree with one another in either uniaxial
stress or pure shear states. If they are set by using the yield functions (6.84) and (6.110) so
√ stress σy , then, under pure shear, the von Mises
that both predict the same uniaxial yield
criterion will predict a yield stress 2/ 3 (≈ 1.155) times that given by the Tresca criterion.
On the other hand, if both criteria are made to coincide under pure shear (expressions (6.83)
and (6.106) with the same τy ), √ then, in uniaxial stress states, the von Mises criterion will
predict yielding at a stress level 3/2 (≈ 0.866) times the level predicted by Tresca’s law.
The yield surface (Φ = 0) associated with the von Mises criterion is represented, in the
space of principal stresses, by the surface of an infinite circular cylinder, the axis of which
coincides with the hydrostatic axis. The von Mises surface is illustrated in Figure 6.8 where
it has been set to match the Tresca surface (shown in the same figure) under uniaxial stress.
The corresponding π-plane representation is shown in Figure 6.9(b). The von Mises circle
intersects the vertices of the Tresca hexagon. The yield surfaces for the von Mises and
Tresca criteria set to coincide in shear is shown in Figure 6.11. In this case, the von Mises
circle is tangent to the sides of the Tresca hexagon. It is remarked that, for many metals,
experimentally determined yield surfaces fall between the von Mises and Tresca surfaces.
A more general model, which includes both the Tresca and the von Mises yield surfaces as
particular cases (and, in addition, allows for anisotropy of the yield surface), is described in
Section 10.3.4 (starting page 427).
V3
r
ea
sh
re
pu
Tresca
T 30o
von Mises
V1 V2
Figure 6.11. Yield surfaces for the Tresca and von Mises criteria coinciding in pure shear.
a strong dependence of the yield limit on the hydrostatic pressure, appropriate description
of plastic yielding requires the introduction of pressure-sensitivity. A classical example of
a pressure-sensitive law is given by the Mohr–Coulomb yield criterion described in the
following.
The Mohr–Coulomb criterion is based on the assumption that the phenomenon of
macroscopic plastic yielding is, essentially, the result of frictional sliding between material
particles. Generalising Coulomb’s friction law, this criterion states that plastic yielding begins
when, on a plane in the body, the shearing stress, τ , and the normal stress, σn , reach the
critical combination
τ = c − σn tan φ, (6.113)
where c is the cohesion and φ is the angle of internal friction or frictional angle. In the above,
the normal stress, σn , was assumed tensile positive.
The yield locus of the Mohr–Coulomb criterion is the set of all stress states such that there
exists a plane in which (6.113) holds. The Mohr–Coulomb yield locus can be easily visualised
in the Mohr plane representation shown in Figure 6.12. It is the set of all stresses whose largest
Mohr circle, i.e. the circle associated with the maximum and minimum principal stresses
(σmax and σmin , respectively), is tangent to the critical line defined by τ = c − σn tanφ.
The elastic domain for the Mohr–Coulomb law is the set of stresses whose all three Mohr
circles are below the critical line. From Figure 6.12, the yield condition (6.113) is found to
be equivalent to the following form in terms of principal stresses
σmax − σmin σmax + σmin σmax − σmin
cos φ = c − + sin φ tan φ, (6.114)
2 2 2
In view of (6.115), a yield function expressed in terms of the principal stresses can be
immediately defined for the Mohr–Coulomb criterion as
W
W c Vtan
n I (critical line)
I
c
I
Vmin Vint Vmax 0 Vn
c cot I
Vmax − Vmin
Due to its definition in terms of principal stresses, this yield function is an isotropic function
of σ. The corresponding yield surface (Φ = 0) is a hexagonal pyramid aligned with the
hydrostatic axis and whose apex is located at
p = c cot φ (6.117)
on the tensile side of the hydrostatic axis. The Mohr–Coulomb surface is illustrated in
Figure 6.13. Its pyramidal shape, as opposed to the prismatic shape of the Tresca surface, is
a consequence of the pressure-sensitivity of the Mohr–Coulomb criterion. It should be noted,
however, that both criteria coincide in the absence of internal friction, i.e. when φ = 0. As no
stress state is allowed on the outside of the yield surface, the apex of the pyramid (point A in
the figure) defines the limit of resistance of the material to tensile pressures. Limited strength
under tensile pressure is a typical characteristic of materials such as concrete, rock and soils,
to which the Mohr–Coulomb criterion is most applicable.
Multisurface representation
Analogously to the multisurface representation of the Tresca criterion, the Mohr–Coulomb
criterion can also be expressed by means of six functions:
Φ1 (σ, c) = σ1 − σ3 + (σ1 + σ3 ) sin φ − 2 c cos φ
Φ2 (σ, c) = σ2 − σ3 + (σ2 + σ3 ) sin φ − 2 c cos φ
Φ3 (σ, c) = σ2 − σ1 + (σ2 + σ1 ) sin φ − 2 c cos φ
(6.118)
Φ4 (σ, c) = σ3 − σ1 + (σ3 + σ1 ) sin φ − 2 c cos φ
Φ5 (σ, c) = σ3 − σ2 + (σ3 + σ2 ) sin φ − 2 c cos φ
Φ6 (σ, c) = σ1 − σ2 + (σ1 + σ2 ) sin φ − 2 c cos φ,
whose roots, Φi (σ, c) = 0 (for fixed c), define six planes in the principal stress space.
Each plane contains one face of the Mohr–Coulomb pyramid represented in Figure 6.13.
166 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
− σ3 3 p
√
Φ=0
φ
ot
cc
√3
A
− σ2
− σ1
The definition of the elastic domain and the yield surface in the multisurface representation
is completely analogous to that of the Tresca criterion.
Invariant representation
Analogously to the invariant representation (6.94) of the Tresca criterion, the Mohr–Coulomb
yield function can be expressed as (Owen and Hinton 1980, and Crisfield 1997):
1
Φ = cos θ − √ sin θ sin φ J2 (s) + p(σ) sin φ − c cos φ, (6.119)
3
where the Lode angle, θ, is defined in (6.97). As for the Tresca model, in spite of its
frequent use in computational plasticity, the invariant representation of the Mohr–Coulomb
surface renders more complex numerical algorithms so that the multisurface representation is
preferred in the computational implementation of the model described in Chapter 8.
is satisfied, where η and c̄ are material parameters. Represented in the principal stress space,
the yield locus of this criterion is a circular cone whose axis is the hydrostatic line. For η = 0,
the von Mises cylinder is recovered. The Drucker–Prager cone is illustrated in Figure 6.14.
THE MATHEMATICAL THEORY OF PLASTICITY 167
− σ3 3 p
√
Φ=0
− σ2
− σ1
where c is the cohesion and the parameters η and ξ are chosen according to the required
approximation to the Mohr–Coulomb criterion. Note that the isotropy of the Mohr–Coulomb
yield function follows from the fact that it is defined in terms of invariants of the stress
tensor (J2 (s) and p). Two of the most common approximations used are obtained by making
the yield surfaces of the Drucker–Prager and Mohr–Coulomb criteria coincident either at
the outer or inner edges of the Mohr–Coulomb surface. Coincidence at the outer edges is
obtained when
6 sin φ 6 cos φ
η= √ , ξ= √ , (6.122)
3 (3 − sin φ) 3 (3 − sin φ)
whereas, coincidence at the inner edges is given by the choice
6 sin φ 6 cos φ
η= √ , ξ= √ . (6.123)
3 (3 + sin φ) 3 (3 + sin φ)
The outer and inner cones are known, respectively, as the compression cone and the
extension cone. The inner cone matches the Mohr–Coulomb criterion in uniaxial tension and
biaxial compression. The outer edge approximation matches the Mohr–Coulomb surface in
uniaxial compression and biaxial tension. The π-plane section of both surfaces is shown in
Figure 6.15. Another popular Drucker–Prager approximation to the Mohr–Coulomb criterion
is obtained by forcing both criteria to predict identical collapse loads under plane strain
conditions. In this case (the reader is referred to Section 4.7 of Chen and Mizuno (1990) for
168 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
V 3
DruckerŦPrager
(outer edges)
MohrŦCoulomb
DruckerŦPrager
(inner edges)
V 1
r
V 2
ea
sh
re
pu
Figure 6.15. The π-plane section of the Mohr–Coulomb surface and the Drucker–Prager approxima-
tions.
σ2
ft’ fbt’
fc’
ft’ σ1
Mohr–Coulomb
fbc’
fc’
Drucker–Prager
biaxial fit
Drucker–Prager
uniaxial fit
Figure 6.16. Plane stress. Drucker–Prager approximation matching the Mohr–Coulomb surface in
uniaxial tension and uniaxial compression.
170 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
In non-associative models, the plastic strain rate is not normal to the yield surface in general.
as the set of all admissible pairs (combinations) of stress and hardening force. The principle
of maximum dissipation postulates that among all admissible pairs (σ∗ , A∗ ) ∈ A, the actual
state (σ, A) maximises the dissipation function (6.42) for a given plastic strain rate, ε̇p , and
rate α̇ of hardening internal variables. The principle of maximum plastic dissipation requires
that, for given (ε̇p , α̇),
In other words, the actual state (σ, A) of stress and hardening force is a solution to the
following constrained optimisation problem:
The Kuhn–Tucker optimality conditions (Luenberger, 1973, Chapter 10) for this optimisation
problem are precisely the associative plastic flow rule (6.129), the associative hardening
rule (6.130) and the loading/unloading conditions
Remark 6.2. The postulate of maximum plastic dissipation, and the corresponding asso-
ciative laws, are not universal. Based on physical considerations, maximum dissipation
has been shown to hold in crystal plasticity and is particularly successful when applied to
the description of metals. Nevertheless, for many materials, particularly soils and granular
materials in general, associative laws frequently do not correspond to experimental evidence.
In such cases, the maximum dissipation postulate is clearly not applicable and the use of
non-associative laws is essential.
Associative Tresca
The associative Tresca flow rule utilises the yield function (6.84) as the flow potential. Since
Φ here is also an isotropic function of σ, the rate of plastic strain has the same principal
directions as σ. The Tresca yield function is differentiable when the three principal stresses
are distinct (σ1
= σ2
= σ3 ) and non-differentiable when two principal stresses coincide (at
the edges of the Tresca hexagonal prism). Hence, the Tresca associative plastic flow rule is
generally expressed as
ε̇p = γ̇N, (6.138)
where N is a subgradient of the Tresca function
N ∈ ∂σ Φ. (6.139)
6
6
∂Φi
ε̇p = γ̇ i N i = γ̇ i , (6.140)
i=1 i=1
∂σ
172 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
σ3 σ
3 p
√
σ2
σ1
with the yield functions Φi defined by (6.91). Each vector N i is normal to the plane defined
by Φi = 0.
The above flow rule can be alternatively expressed as follows. Firstly assume, without loss
of generality, that the principal stresses are ordered as σ1 ≥ σ2 ≥ σ3 , so that the discussion
can be concentrated on the sextant of the π-plane illustrated in Figure 6.18. Three different
possibilities have to be considered in this sextant:
(a) yielding at a stress state on the side (main plane) of the Tresca hexagon (Φ1 = 0, Φ2 < 0
and Φ6 < 0);
(b) yielding from the right corner, R (Φ1 = 0, Φ6 = 0 and Φ2 < 0); and
(c) Yielding from the left corner, L (Φ1 = 0, Φ2 = 0 and Φ6 < 0).
When the stress is on the side of the hexagon, only one multiplier may be non-zero and the
plastic flow rule reads
ε̇p = γ̇ N a , (6.141)
where the flow vector is the normal to the plane Φ1 = 0, given by
∂Φ1 ∂
Na ≡N1 = = (σ1 − σ3 )
∂σ ∂σ
= e1 ⊗ e1 − e3 ⊗ e3 , (6.142)
with ei denoting the eigenvector of σ associated with the principal stress σi . In deriving the
last right-hand side of (6.142), use has been made of the expression (A.27) of page 736 for
the derivative of an eigenvalue of a symmetric tensor.
At the right and left corners of the hexagon, where two planes intersect, two multipliers
may be non-zero. Thus, the plastic flow equation is
ε̇p = γ̇ a N a + γ̇ b N b . (6.143)
THE MATHEMATICAL THEORY OF PLASTICITY 173
sextant with
V1t V2 t V3 V1
Na a
N )1 Hp
Nbb
R
L
)6
Naa Nbb
b
Nb
V
V2 )2 V3
The vector N a is the normal to the plane Φ1 = 0, already defined. In the right corner
(repeated minimum principal stress), the second vector, N b , is normal to the plane Φ6 = 0
and is obtained analogously to (6.142) as
N b ≡ N 6 = e1 ⊗ e1 − e2 ⊗ e2 . (6.144)
In the left corner (repeated maximum principal stress), N b , is normal to the plane Φ2 = 0,
N b ≡ N 2 = e2 ⊗ e2 − e3 ⊗ e3 . (6.145)
It should be noted that, as for the Prandtl–Reuss rule, the plastic flow predicted by the
associative Tresca law is volume-preserving. Indeed, note that, in the above, we have trivially
tr N a = tr N b = 0. (6.146)
sextant with
V1t V2 t V3 V1
(a)
a
Na
Nbb
)1
(b)
R
L )6
b
Nb
V2 )2 V3
Figure 6.19. The Mohr–Coulomb flow rule; (a) faces and edges, and (b) apex.
the normal vectors N a and N b are no longer deviatoric, i.e. they have a non-zero component
along the hydrostatic axis (the vectors shown in Figure 6.19 are deviatoric projections of the
actual normals). For plastic yielding from the face, the flow rule is given by
ε̇p = γ̇ N a , (6.147)
ε̇p = γ̇ a N a + γ̇ b N b . (6.149)
At the right (extension) corner, R, the second vector, N b , is normal to the plane Φ6 = 0 and
is given by
N b = (1 + sin φ) e1 ⊗ e1 − (1 − sin φ) e2 ⊗ e2 , (6.150)
whereas, at the left (compression) corner, L, the tensor N b is normal to the plane Φ2 = 0,
At the apex of the Mohr–Coulomb surface, all six planes intersect and, therefore, six
normals are defined and up to six plastic multipliers may be non-zero. This situation is
schematically illustrated in Figure 6.19(b). The plastic strain rate tensor lies within the
pyramid defined by the six normals:
6
ε̇p = γ̇ i N i . (6.152)
i=1
THE MATHEMATICAL THEORY OF PLASTICITY 175
It is important to note that, due to the pressure sensitivity of the Mohr–Coulomb criterion,
the associative Mohr–Coulomb rule predicts a non-zero volumetric plastic straining. This is in
contrast to the Prandtl–Reuss and associative Tresca laws. The volumetric component of the
plastic strain rate in the associative Mohr–Coulomb law can be obtained by expanding (6.152)
in principal stress space taking into account the definitions of Ni . This gives
1
γ̇
p γ̇ 2
ε̇1 α 0 β β 0 α
γ̇ 3
p
ε̇2 = 0 α α 0 β β
γ̇ 4 , (6.153)
p β β 0 α α 0 5
ε̇3 γ̇
γ̇ 6
where
α ≡ 1 + sin φ, β ≡ −1 + sin φ. (6.154)
The above trivially yields
6
ε̇pv ≡ ε̇p1 + ε̇p2 + ε̇p3 = 2 sin φ γ̇ i . (6.155)
i=1
As all γ̇ i ’s are non-negative, the volumetric plastic strain rate is positive and, therefore,
dilatant. The phenomenon of dilatancy during plastic flow is observed for many materi-
als, particularly geomaterials. However, the dilatancy predicted by the associative Mohr–
Coulomb law is often excessive. To overcome this problem, it is necessary to use a non-
associated flow rule in conjunction with the Mohr–Coulomb criterion. The non-associated
Mohr–Coulomb law adopts, as flow potential, a Mohr–Coulomb yield function with the
frictional angle φ replaced by a different (smaller) angle ψ. The angle ψ is called the dilatancy
angle and the amount of dilation predicted is proportional to its sine. Note that for ψ = 0, the
plastic flow becomes purely deviatoric and the flow rule reduces to the associative Tresca law.
(a) plastic yielding at (smooth portion of) the cone surface; and
At the cone surface, where the Drucker–Prager yield function is differentiable, the flow
vector is obtained by simply differentiating (6.121) which gives (Figure 6.20(a))
1 η
N= s + I, (6.156)
2 J2 (s) 3
176 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
(a) w<
V
Hp
PrandtlŦReuss
associative (b)
DruckerŦPrager
V
V 3
V2
V1
Figure 6.20. The Drucker–Prager flow vector; (a) cone surface, and (b) apex.
ε̇p = γ̇ N. (6.157)
At the apex singularity, the flow vector is an element of the subdifferential of the yield
function (6.121):
N ∈ ∂σ Φ. (6.159)
It lies within the complementary cone to the Drucker–Prager yield surface, i.e. the cone
whose wall is normal to the Drucker–Prager cone illustrated in Figure 6.20(b). From standard
properties of subdifferentials (Rockafellar, 1970; Rockafellar and Wets, 1998) it can be
established that the deviatoric/volumetric split of N in this case is given by
Nd ∈ ∂σ Φd , Nv = η, (6.160)
where Φd ≡ J2 (s). Expressions (6.157), (6.158) and (6.160) result in the following rate of
(dilatant) volumetric plastic strain for the associative Drucker–Prager flow rule:
ε̇pv = γ̇ η. (6.161)
Nv = η̄. (6.166)
If the dilatancy angle of the non-associative potential is chosen as ψ = 0, then the volumetric
component, Nv , vanishes and the flow rule reduces to the Prandtl–Reuss law that predicts
volume-preserving plastic flow (refer to Figure 6.20(a)).
V3 V
uniaxial
cyclic test
S-plane E T
Vy
V1 V2
Vy
von Mises model along with the corresponding π-plane representation of the yield surface.
Perfectly plastic models are particularly suitable for the analysis of the stability of structures
and soils and are widely employed in engineering practice for the determination of limit loads
and safety factors.
Strain hardening
In this case the hardening internal state variable is some suitably chosen scalar measure of
strain. A typical example is the von Mises effective plastic strain, also referred to as the
THE MATHEMATICAL THEORY OF PLASTICITY 179
V3 V
S -plane
uniaxial
cyclic test
initial surface
V1 V2
hardened surface
The above definition generalises the accumulated axial plastic strain (6.18) (page 145) of the
one-dimensional model to the multiaxially strained case. Its rate evolution equation reads
ε̄˙p = 23 ε̇p : ε̇p = 23 ||ε̇p ||, (6.168)
Accordingly, a von Mises isotropic strain-hardening model is obtained by letting the uniaxial
yield stress be a function of the accumulated plastic strain:
σy = σy (ε̄p ). (6.170)
This function defines the strain-hardening curve (or strain-hardening function) that can be
obtained, for instance, from a uniaxial tensile test.
Under a uniaxial stress state with axial stress σ and axial stress rate σ̇ in the direction of the
base vector e1 , the matrix representations of the stress tensor and the stress rate tensor in the
three-dimensional model are given by
1 0 0 1 0 0
[σ] = σ 0 0 0, [σ̇] = σ̇ 0 0 0. (6.171)
0 0 0 0 0 0
The corresponding stress deviator reads
1 0 0
[s] = 23 σ 0 − 12 0 . (6.172)
0 0 −2
1
Φ̇ = N : σ̇ − H ε̄˙p = 0, (6.175)
where N ≡ ∂Φ/∂σ is the Prandtl–Reuss flow vector (6.136) and H = H(ε̄p ) is the hard-
ening modulus defined in (6.27). To conclude the demonstration, we combine (6.175)
with (6.136), (6.171)2 and (6.172) to recover (6.28) and, then, following the same arguments
as in the one-dimensional case we find that, under uniaxial stress conditions, the isotropic
strain hardening von Mises model predicts the tangential axial stress–strain relation
EH
σ̇ = ε̇, (6.176)
E+H
which is identical to equation (6.31) of the one-dimensional model.
Work hardening
In work-hardening models, the variable defining the state of hardening is the dissipated plastic
work,¶ wp , defined by
t
w ≡
p
σ : ε̇p dt. (6.177)
0
¶ The term work hardening is adopted by many authors as a synonym for the phenomenon of hardening in general.
Materials that harden, i.e. materials whose yield stress level depends on the history of strains, are frequently referred
to as work-hardening materials. In this text, however, the term work hardening is reserved for plasticity models in
which the dissipated plastic work is taken as the state variable associated with hardening.
THE MATHEMATICAL THEORY OF PLASTICITY 181
σ
w=
w = wpp +
+ wee P
p
w
wp
e
we
w
ε
Figure 6.23. The plastic work.
In a uniaxial test, for instance (Figure 6.23), the total work w necessary to deform the material
up to point P is given by the total area under the corresponding stress–strain curve. Part of this
work, we , is stored in the form of elastic energy and is fully recovered upon elastic unloading.
The remaining (shaded) area, wp , is the plastic work. It corresponds to the energy dissipated
by the plastic mechanisms and cannot be recovered. From the definition of wp , its evolution
equation is given by
ẇp = σ : ε̇p . (6.178)
An isotropic work-hardening von Mises model is obtained by postulating
σy = σy (wp ). (6.179)
or, equivalently,
dwp
= σy . (6.181)
dε̄p
As σy is strictly positive (σy > 0), the above differential relation implies that the mapping
between wp and ε̄p is one-to-one and, therefore, invertible so that
wp = wp (ε̄p ) (6.182)
and
ε̄p = ε̄p (wp ). (6.183)
182 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
This allows any given strain-hardening function of the type (6.170) to be expressed as an
equivalent work-hardening function,
and any given work-hardening function of the type (6.179) to be expressed as an equivalent
strain-hardening function,
Expressions (6.184) and (6.185) establish the equivalence between the strain and work-
hardening descriptions for the von Mises model with associative flow rule.
α ≡ {ε̄p } (6.187)
and
ψ p = ψ p (ε̄p ). (6.188)
The set of hardening thermodynamic forces in this case specialises as
A ≡ {κ}, (6.189)
∂ψ p
κ ≡ ρ̄ = κ(ε̄p ). (6.190)
∂ ε̄p
THE MATHEMATICAL THEORY OF PLASTICITY 183
If the state of hardening is defined in terms of cohesion (or shear yield stress), c (or τy )
replaces σy in (6.191). Note that the hardening modulus H, initially defined in (6.27),
represents the rate of change of the hardening thermodynamic force with respect to the
hardening internal variable, i.e.
∂σy ∂κ
H(ε̄p ) ≡ p
= p. (6.192)
∂ ε̄ ∂ ε̄
For the strain-hardening von Mises model the evolution law (6.168) and (6.169) for the
internal variable, ε̄p , follows from the hypothesis of associativity, that relies on the choice of
the yield function as the plastic potential. The associative evolution equation for ε̄p in this
case is a specialisation of (6.130); that is, we have
Note that, here, the accumulated plastic strain, ε̄p , is being defined by evolution equa-
tion (6.195). Its actual physical meaning depends on the specific format of the functions
Φi and is generally different from that of (6.167) adopted for the von Mises model.
A simple example of associative isotropic hardening law of the type (6.195) is obtained
for the Tresca model. Here, we refer to the plastic flow equations (6.141) and (6.143),
defined respectively on the side (smooth portion) and corner of the Tresca yield surface.
The corresponding associative evolution equations that define the accumulated plastic strain
ε̄p are
∂Φ1
ε̄˙p = −γ̇ = γ̇ (6.196)
∂κ
and
∂Φ1 ∂Φ6
ε̄˙p = −γ̇ a − γ̇ b = γ̇ a + γ̇ b , (6.197)
∂κ ∂κ
184 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
When flow takes place at the smooth portion of a Mohr–Coulomb pyramid face, this is
reduced to
ε̄˙p = 2 cos φ γ̇. (6.201)
At the corners (refer to the plastic flow equation (6.149)), we have
ε̄˙p = 2 cos φ (γ̇ a + γ̇ b ). (6.202)
Note that if it is insisted to adopt the von Mises accumulated plastic strain rate defini-
tion (6.167) in conjunction, say, with the Tresca model with associative plastic flow, (6.141)
to (6.143), the evolution equation for ε̄p will result in
ε̄˙p = 23 ε̇p : ε̇p = √23 γ̇ (6.203)
for flow from the smooth portions of the Tresca surface, and
ε̄˙p = 23 ε̇p : ε̇p = √23 (γ̇ a )2 + γ̇ a γ̇ b + (γ̇ b )2 , (6.204)
for flow from a corner. In this case, the isotropic hardening law is non-associative in spite of
the associativity of the plastic flow rule.
V uniaxial
V3
cyclic test
E Vy
H
V1 V2
initial surface
Figure 6.24. Kinematic hardening and the Bauschinger effect. Uniaxial test and π-plane representation.
Loading in one direction results in decreased resistance to plastic yielding in the opposite direction.
where β is the axial back-stress component. With the above, we obtain for the relative stress
tensor
1 0 0
[η] = η 0 − 21 0 , (6.214)
0 0 −2 1
where
η = 23 σ − β (6.215)
is the axial relative stress. From (6.212) and (6.214) we obtain
1 0 0
[β̇] = 23 H ε̇p 0 − 12 0 , (6.216)
0 0 − 21
Now, by recalling (6.60) and specialising (6.61) for the present case we have that, under
plastic yielding, the following consistency condition must be satisfied:
∂Φ ∂Φ
Φ̇ = : σ̇ + : β̇ = 0. (6.218)
∂σ ∂β
After some straightforward tensor algebra, taking into account (6.171)2 and the above
expressions for β̇, β, the definition of η, and the identity
%
∂Φ ∂Φ 3 η
=− =− , (6.219)
∂β ∂σ 2 η
equation (6.218) yields
σ̇ = H ε̇p . (6.220)
Then, with the introduction of the elastoplastic split of the axial strain rate, together with the
equation
σ̇ = E ε̇e , (6.221)
188 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
of the linear elastic model under uniaxial stress conditions, into (6.220), we obtain
EH
σ̇ = ε̇, (6.222)
E+H
which coincides with the stress rate equation (6.176) of the von Mises isotropic strain-
hardening model with constant H. To complete the demonstration, let us assume that the
uniaxial loading is monotonic, i.e. we have either ε̇ > 0 or ε̇ < 0 throughout the entire loading
process. In this case, the integration of (6.222) having the initial yield stress (σy 0 for both
models) as the initial condition produces the same stress–strain curve as the isotropic model.
Armstrong–Frederick hardening
A refinement upon the linear kinematic hardening law proposed by Armstrong and Frederick
(1966) is obtained by introducing an extra term in the above expression (refer to Lemaitre
and Chaboche (1990), Chapter 5, or Jirásek and Bažant (2002), Chapter 20, for details) with
the evolution of β given by
2
β̇ = H ε̇p − γ̇ b β
3
2 ∂Φ
= γ̇ H −bβ , (6.223)
3 ∂σ
where b is a material constant. The extra term −γ̇ b β introduces the effect of saturation in the
kinematic hardening rule. In the case of the von Mises criterion, the saturation corresponds to
a maximum limit value for the norm of β, at which the material behaves as perfectly plastic.
Another possible improvement upon Prager’s linear kinematic hardening rule is the introduc-
tion of nonlinearity by replacing the constant kinematic hardening modulus, H, of (6.212)
with a generic function of the accumulated plastic strain, ε̄p ,
2 2 ∂Φ
β̇ = H(ε̄p ) ε̇p = γ̇ H(ε̄p ) . (6.224)
3 3 ∂σ
such that
dβ̄
H(ε̄p ) = , (6.226)
dε̄p
defines the kinematic hardening curve. This curve can be obtained from simple uniaxial tests
in a manner analogous to the determination of the hardening curve for the purely isotropic
hardening model.
THE MATHEMATICAL THEORY OF PLASTICITY 189
ψ p = ψ p (X). (6.227)
The variable X is related to self-equilibrated residual stresses that remain after elastic
unloading. These stresses may increase or decrease resistance to plastic slip according to
the direction considered. The kinematic hardening thermodynamical force – the back-stress
tensor, β – is then defined as the derivative
∂ψ p
β≡ . (6.228)
∂X
For the Armstrong–Frederick kinematic hardening law (6.223), for instance, we have
a
ψ p (X) = X : X, (6.229)
2
where the material constant a has been defined as
a≡ 2
3 H. (6.230)
β = a X. (6.231)
The evolution law for the internal variable X is obtained by postulating a flow potential
b
Ψ≡Φ+ β : β, (6.232)
2a
and assuming normal dissipativity
∂Ψ ∂Φ b
Ẋ ≡ −γ̇ = −γ̇ + β . (6.233)
∂β ∂β a
Obviously (since Ψ
= Φ), this evolution law is non-associative. The equivalence between the
above equation and (6.223) can be established by taking into account (6.231) and the fact that,
since Φ is obtained from a non-kinematic hardening yield function by replacing the argument
σ with σ − β, we have
∂Φ ∂Φ
=− . (6.234)
∂β ∂σ
σ load
reversal
σy
β σy
εp
ε
For example, a relatively simple von Mises-based model with mixed isotropic/kinematic
hardening can be devised by adopting the yield function (6.208) and allowing σy to be a
function of ε̄p . If the nonlinear rule defined by (6.224) and (6.225) is adopted, the hardening
behaviour of the model is determined by the curves
which can be obtained from relatively simple uniaxial tests with load reversal (see schematic
illustration of Figure 6.25). At each point ε̄p , the kinematic hardening stress, β̄, is the
kinematic contribution to overall hardening.
A more refined mixed hardening model can be devised by coupling the Armstrong–
Frederick law (6.223) with the von Mises-type yield function (6.208) where σy , as in (6.235)1,
is a function of the accumulated plastic strain. A model including mixed hardening of
this type is discussed in Section 12.3 (starting on page 478) in the context of damage
mechanics.
7 FINITE ELEMENTS IN
SMALL-STRAIN PLASTICITY
PROBLEMS
I N the previous chapter, the mathematical theory of plasticity has been reviewed. A general
small-strain elastoplastic constitutive model has been established within the formalism of
thermodynamics with internal variables and the most popular theories, namely, the von Mises,
Tresca, Mohr–Coulomb and Drucker–Prager models, have been described in detail.
Obviously, due to the mathematical complexity of such constitutive theories, an exact
solution to boundary value problems of practical engineering interest can only be obtained
under very simplified conditions. The existing analytical solutions are normally restricted
to perfectly plastic models and are used for the determination of limit loads and steady
plastic flow of bodies with simple geometries (Chakrabarty, 1987; Hill, 1950; Lubliner,
1990; Prager, 1959; Skrzypek, 1993). The analysis of the behaviour of elastoplastic structures
and soils under more realistic conditions requires the adoption of an adequate numerical
framework capable of producing approximate solutions within reasonable accuracy. As
pointed out in Chapter 4, the approximate solution to such problems is addressed in this
book within the context of the Finite Element Method. In fact, the Finite Element Method is
by far the most commonly adopted procedure for the solution of elastoplastic problems. Since
the first reported applications of finite elements in plasticity in the mid-1960s, a substantial
development of the related numerical techniques has occurred. Today, the Finite Element
Method is regarded as the most powerful and reliable tool for the analysis of solid mechanics
problems involving elastoplastic materials and is adopted by the vast majority of commercial
software packages for elastoplastic stress analysis.
This chapter describes in detail the numerical/computational procedures necessary for
the implicit finite element solution of small strain plasticity problems within the framework
of Chapter 4. For the sake of generality, the methodologies presented in this chapter are
initially derived taking the general plasticity model introduced in Chapter 6 (summarised
in Box 6.2, page 151) as the underlying constitutive model. Practical application of the
theory and procedures introduced, including a complete description of the algorithms and
corresponding FORTRAN subroutines of the HYPLAS program, is then made to the particular
case of the von Mises model with nonlinear isotropic hardening. The choice of this model
is motivated here by the simplicity of its computational implementation. A set of numerical
examples is also presented. Further application of the theory is made at the end of the chapter
to a mixed isotropic/kinematic hardening version of the von Mises model. This model is also
included in the HYPLAS program. Application to the Tresca, Mohr–Coulomb and Drucker–
Prager models is left for Chapter 8.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
192 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
1. The state update procedure which, in the case of elastoplastic materials, requires the
formulation of a scheme for numerical integration of the rate elastoplastic evolution
equations. Within a (pseudo-) time increment [tn , tn+1 ], the state update procedure
gives the stresses σn+1 and the internal variables αn+1 at the end of the increment as
a function of the internal variables αn at the beginning of the increment and the strains
εn+1 at the end of the increment:
The incremental constitutive functions σ̂ and α̂ are defined by the integration algorithm
adopted and the stress delivered by σ̂ is used to assemble the element internal force
vector
ngausp
f int
e = ji wi BTi σn+1 |i . (7.3)
i=1
∂ σ̂
D≡ , (7.4)
∂εn+1
(e)
ngausp
KT = wi ji BTi Di Bi . (7.5)
i=1
These are the primary procedures that effectively define a particular constitutive
model/algorithm in the program.
This chapter will focus precisely on the two topics listed above, specialised to the case
of elastoplastic materials. Following the nomenclature established in Chapter 5 (refer to
the call trees of Figures 5.4 and 5.5 respectively on pages 126 and 130), the computational
implementation of the state update algorithms and tangent moduli addressed in the present
chapter appears at the lowest layer of the material level in the computation of the element
internal force vector and tangent stiffness matrix.
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 193
Remark 7.1. We refer to the system of differential equations (7.6) as reduced in that it
is obtained from the model of Box 6.2 by incorporating the plastic flow equation into the
additive strain rate decomposition. In this way, the plastic strain does not appear explicitly in
194 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
the system and the only unknowns are the elastic strain, the set of hardening internal variables
and the plastic multiplier. Note that with the solution εe (t) of Problem 7.1 at hand, the history
of the plastic strain tensor is obtained from the trivial relation
so that the history of all variables involved in the definition of the elastoplastic model of
Box 6.2 is determined.
As already mentioned, exact solutions to Problem 7.1, when yield functions and flow rules
such as the ones described in the previous chapter are adopted, may only be obtained for
very simple prescribed strain histories. Even in such cases, the derivation of the analytical
solutions is normally cumbersome. For complex deformation paths, which are more likely to
occur in realistic engineering problems, analytical solutions are not available in general and
the adoption of a numerical technique to find an approximate solution becomes absolutely
essential. A general framework for the numerical solution of the constitutive initial value
problem of elastoplasticity is described below.
Problem 7.2 (The incremental elastoplastic constitutive problem). Given the values εen
and αn , of the elastic strain and internal variables set at the beginning of the pseudo-time
interval [tn , tn+1 ], and given the prescribed incremental strain ∆ε for this interval, solve the
following system of algebraic equations
for the unknowns εen+1 , αn+1 and ∆γ, subjected to the constraints
where
∂ψ ∂ψ
σn+1 = ρ̄ e , An+1 = ρ̄ . (7.12)
∂ε n+1 ∂α n+1
In the above, we have adopted the obvious notation
with (·)n and (·)n+1 denoting the value of (·) respectively at tn and tn+1 . The increment ∆γ
will be called the incremental plastic multiplier. Note that once the solution εen+1 has been
obtained, the plastic strain at tn+1 can be calculated as
so that all variables of the model are known at the end of the interval [tn , tn+1 ].
∆γ = 0. (7.15)
In this case there is no plastic flow or evolution of internal variables within the
considered interval [tn , tn+1 ], i.e. the step is purely elastic. The constraint (7.11)3 is
automatically satisfied, εen+1 and αn+1 are given by
εen+1 = εen + ∆ε
(7.16)
αn+1 = αn
must hold, where σn+1 and An+1 are functions of εen+1 and An+1 defined through
the potential relations (7.12).
2. Strictly positive plastic multiplier,
∆γ > 0. (7.18)
In summary, we have just seen that any solution to Problem 7.2 is either given by (7.16),
in which case it must satisfy constraint (7.17), or it is a solution of the algebraic system of
equations (7.19)–(7.20), subjected to the constraint (7.18); that is, only one of two possible
sets of equations provides a solution to Problem 7.2. It remains now to devise a procedure
whereby the correct solution can be chosen. This is next described in detail.
The corresponding stress and hardening force will be called the elastic trial stress and
elastic trial hardening force, given by
trial trial
trial ∂ψ trial ∂ψ
σn+1 = ρ̄ e , An+1 = ρ̄ . (7.22)
∂ε n+1 ∂α n+1
The above variables are collectively called the elastic trial state. Now note that, to be
the actual solution, the elastic trial state has, in addition, to satisfy (7.17). We then
proceed as follows. If
Φtrial ≡ Φ(σtrial
n+1 , An+1 ) ≤ 0,
trial
(7.23)
that is, if the elastic trial state lies within the elastic domain or on the yield surface, it
is accepted as a solution to Problem 7.2. In this case, we update
(·)n+1 := (·)trial
n+1 (7.24)
and the algorithm is terminated. Otherwise, the elastic trial state is not plastically
admissible and a solution to Problem 7.2 must be obtained from the plastic corrector
step described below.
(b) The Plastic Corrector Step (or Return-Mapping Algorithm).
The only option left now is to solve the system (7.19)–(7.20) of algebraic equations
subject to the constraint (7.18). Using the elastic trial state definition above, we rewrite
the algebraic system equivalently as
trial
εen+1 = εen+1 − ∆γ N(σn+1 , An+1 )
αn+1 = αtrial
n+1 + ∆γ H(σn+1 , An+1 ) (7.25)
Φ(σn+1 , An+1 ) = 0,
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 197
(a) (b)
σn+1
trial
σn+1
trial
plastic
corrector plastic
corrector
elastic
elastic predictor
predictor σn+1 Φ(σ, An+1 ) = 0
σn+1
σn σn
elastic elastic
domain at t n domain
Φ(σ, An ) = 0 Φ(σ) = 0
Figure 7.1. General return mapping schemes. Geometric interpretation: (a) hardening plasticity; and
(b) perfect plasticity.
which are, of course, complemented with the potential relations (7.12). The plastic
corrector stage of the algorithm then consists in finding a solution εen+1 , αn+1 and ∆γ
for (7.25) that satisfies
∆γ > 0. (7.26)
Remark 7.2. The procedure of item (b) above possesses an appealing geometric interpreta-
tion as illustrated in Figure 7.1. Consider the yield surface at the elastic trial state. The elastic
trial stress, σtrial
n+1 , in this case lies outside the plastically admissible domain (i.e. neither in
the elastic domain nor on the yield surface). Upon solution of the algebraic system (7.25),
equation (7.25)3, which is commonly referred to as the plastic consistency equation, ensures
that the stress, σn+1 , at the end of the interval [tn , tn+1 ] lies on the updated yield surface; that
is, the elastic trial stress returns to the yield surface so that plastic consistency is re-established
in the updated state. In the case of perfect plasticity, σtrialn+1 returns to a fixed surface. Due to
this interpretation the procedure of item (b) is referred to as the return mapping algorithm
and (7.25) are called the return mapping equations. The first algorithm of this type appears
to have been the radial return method proposed in the pioneering work of Wilkins (1964).
Euler pseudo-time
discretisation
incremental elastoplastic
constitutive problem (Problem 7.2)
solution
procedure
PLASTIC CORRECTOR
ELASTIC PREDICTOR
(RETURN MAPPING ALGORITHM)
Figure 7.2. From the initial value problem of elastoplasticity to the elastic predictor/return-mapping
integration algorithm. Schematic diagram.
derivation of the overall integration algorithm. The algorithm described above has been
obtained by adopting, in particular, the backward Euler scheme to discretise elastoplastic
constitutive equations and is, therefore, termed the backward, fully implicit or simply implicit
elastic predictor/return mapping scheme. This algorithm is conveniently summarised in
Box 7.1 in pseudo-code format. We remark that different discretisation schemes may be used
instead, each one resulting in a different return mapping algorithm, but all having the same
elastic predictor step. Alternatives to the backward Euler-based algorithm will be discussed
in Section 7.2.7.
(i) Elastic predictor. Given ∆ε and the state variables at tn , evaluate the elastic trial
state
εen+1
trial
= εen + ∆ε
αtrial
n+1 = αn
trial trial
∂ψ ∂ψ
σtrial
n+1 = ρ̄ , Atrial
n+1 = ρ̄
∂ εe n+1 ∂ α n+1
(ii) Check plastic admissibility
trial
n+1 , An+1 ) ≤ 0
IF Φ(σtrial
THEN set (·)n+1 = (·)trial
n+1 and EXIT
(iv) EXIT
the fact that, in contrast to the Newton–Raphson algorithm, they do not require the exact
gradients of the associated residual functions (defined by the left-hand side of the equations
of item (iii) of Box 7.1) and, for complex material models, the derivation and computational
implementation of residual gradients may prove a tedious exercise if performed manually.
However, this argument is substantially weakened by considering that currently available
symbolic manipulation software packages, such as MATHEMATICA R
(Wolfram, 1991),
can handle the closed-form calculation of derivatives quite easily, making the derivation
and computational implementation of residual gradients a relatively straightforward task,
even for complex plasticity models. Added to this is the fact that the relative ease of quasi-
Newton procedures comes at the expense of lower rates of convergence and, generally, less
efficiency in the return-mapping procedure (and poorer performance of the overall finite
element scheme). Also, as will be seen later, the return-mapping residual derivatives are
needed to compute the consistent tangent operator, used to assemble the tangent stiffness
matrix. Thus, if the full Newton–Raphson scheme is adopted to solve the global finite element
equilibrium equations, the return mapping residual derivatives will have to be computed
anyway.
Remark 7.3. One aspect of the return-mapping scheme deserves particular attention. It might
happen that the corresponding nonlinear system of algebraic equations has a solution with
200 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
and then applying the integration algorithm sequentially m times to reach the approximate
state at tn+1 . Each strain sub-increment ∆εi , i = 1, . . . , m, has to be sufficiently small to
ensure convergence of the Newton–Raphson iterations. More recently, Armero and Pérez-
Foguet (2002) and Pérez-Foguet and Armero (2002) have explored the closest point projec-
tion interpretation (Section 7.2.5 below) and the associated variational structure of the return-
mapping equations to devise globally convergent root-finding algorithms (i.e. algorithms that
converge regardless of the initial guess) based on a combination of the Newton–Raphson
scheme and constrained line-search procedures.
σ = De : εe , (7.27)
with constant elasticity tensor, De , equation (7.25)1 can be rewritten equivalently in terms of
stresses as
n+1 − ∆γ D : Nn+1 .
e
σn+1 = σtrial (7.28)
In this case, the updated stress, σn+1 , obtained by the implicit return mapping (Figure 7.3)
is the projection of the trial stress σtrial
n+1 onto the updated yield surface along the direction of
the tensor
De : Nn+1 .
For perfectly plastic materials with associative flow rule, the implicit return mapping can be
interpreted as a closest point projection of the trial stress onto the set
A = {σ | Φ(σ) ≤ 0} (7.29)
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 201
σn+1
trial
σn+1
trial
perfect
plasticity
n+1 ||
D e : Nn+1
n+1 ||
∆γ || D e :
hardening
N
∆γ || D e :
plasticity
D e : Nn+1
σn+1
Φ(σ, An+1 ) = 0
σn+1
σn σn
elastic elastic
domain at t n domain
Φ(σ, An ) = 0
Φ(σ) = 0
Figure 7.3. The fully implicit return mapping. Geometric interpretation for materials with linear elastic
response.
The interpretation of the implicit return mapping as a closest point projection of the trial stress
remains valid for linearly hardening materials provided that a suitable definition of distance
in the space of stress and hardening forces is introduced (see Simo et al. (1988b) for details).
are reviewed in Section 7.2.10) of the overall elastic predictor/return mapping algorithm
will depend on the particular strategy adopted. Two important families of algorithms for
elastoplasticity, which incorporate the backward (or fully implicit) Euler approach as a
particular case, can be derived by adopting generalised versions of the classical trapezoidal
and midpoint rules in the discretisation of Problem 7.1. These procedures have been proposed
by Ortiz and Popov (1985). Another popular scheme is the so-called cutting plane algorithm
proposed by Simo and Ortiz (1985) (see also Ortiz and Simo 1986). These procedures
are described below. It should be emphasised that in all cases discussed here the use of a
different discretisation rule will affect only the return mapping part of the overall integration
algorithm (item (iii) of Box 7.1). The elastic predictor stage, as previously described, remains
unchanged.
σn+1
trial
θ ∆γ || D e : Nn ||
D e : Nn+1
(1− θ) ∆γ || D e : Nn+1 ||
e
D : Nn σn+1
Φ(σ, An+1 ) = 0
σn
elastic
domain
Φ(σ, An ) = 0
Figure 7.4. The generalised trapezoidal return mapping. Geometric interpretation for materials with
linear elastic response.
where
Nn+θ = N(σn+θ , An+θ )
(7.38)
Hn+θ = H(σn+θ , An+θ ),
with the generalised midpoint state defined by the variables
σn+θ = (1 − θ)σn+1 + θ σn
(7.39)
An+θ = (1 − θ)An + θ An+1 .
The parameter θ is, again, a prescribed constant within the interval 0 ≤ θ ≤ 1. In the present
case, the discrete plastic flow rule reads
Note that the generalised midpoint rule coincides with the generalised trapezoidal rule (7.33)
for θ = 1 and θ = 0, which correspond, respectively, to the fully implicit and fully explicit
204 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
σn+1
trial
∆γ || D e : Nn+θ ||
D e : Nn+θ
σn+1
σn
elastic
domain
Φ(σ, An ) = 0
Figure 7.5. The generalised midpoint return mapping. Geometric interpretation for materials with linear
elastic response.
return mappings. The geometric interpretation of the generalised midpoint return algorithm
is given in Figure 7.5 for materials with linear elastic response. The updated stress, in this
case, is given by
n+1 − ∆γ D : Nn+θ ,
e
σn+1 = σtrial (7.41)
and corresponds to the projection of the elastic trial stress, σtrial
n+1 , onto the updated yield
surface at tn+1 along the direction of
De : Nn+θ .
A possible variation of the above generalised midpoint rule is obtained by replacing the
consistency condition (7.37)3 with the alternative
In this case, plastic consistency is enforced upon the generalised midpoint, rather than the
updated, state. This version of the generalised return mapping is discussed in detail by Simo
and Govindjee (1991) who highlight, in particular, the fact that in this case the symmetry
of the associated consistent tangent operators‡ is ensured for fully associative models – a
property not generally preserved by the family of algorithms based on (7.37)3, as noted by
Ortiz and Martin (1989).
Remark 7.4. As for the fully implicit return mapping, both generalised trapezoidal and
midpoint return-mapping algorithms, defined respectively by (7.33) and (7.37), require the
‡ Consistent tangent operators for return-mapping schemes will be discussed in Section 7.4. In particular, a
where ∗ denotes the product of the appropriate type. The above rate constitutive equations
for σ and A are obtained by a straightforward application of the chain rule to (7.8). The
fourth-order tensor De (εe , α) in (7.46)1 is the elastic modulus
∂2ψ
De (εe , α) = ρ̄ , (7.47)
∂εe 2
which, in general, is assumed to be a function of εe and α. The operator E(εe , α) denotes
the tangent modulus associated with the hardening response, defined by
∂2ψ
E(εe , α) = ρ̄ . (7.48)
∂εe ∂α
206 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
∂2ψ
F(εe , α) = ρ̄ (7.49)
∂α∂εe
and
∂2ψ
G(εe , α) = ρ̄
. (7.50)
∂α2
It should be noted that De is constant for materials whose elastic behaviour is linear. Also,
under the assumption (6.37) (page 149) of decoupling between elasticity and hardening,
the tangent operators E and F vanish and so does the second term on the right-hand side
of (7.46)1 and the first term on the right-hand side of (7.46)2. However, for models in which
elasticity is coupled with dissipative phenomena, such as Lemaitre’s damage theory discussed
in Chapter 12, these terms do not vanish in general.
By combining (7.44) with (7.46) and making use of complementary potential relations
that give
εe = εe (σ, A), α = α(σ, A), (7.51)
we redefine the initial value problem (7.44)–(7.45) equivalently in terms of stress and
hardening force as
σ = σtrial
n+1 , A = Atrial
n+1 . (7.53)
In summary, the return-mapping problem now comprises the initial value problem (7.52)–
(7.53) in conjunction with the plastic consistency constraint
Basically, the cutting plane return-mapping algorithm is an iterative procedure for numer-
ical solution of the return-mapping problem whereby the linear approximation to the plastic
consistency equation is solved at each iteration. In a typical cutting-plane iteration, (k), the
plastic consistency equation is linearised§ about the current (known) state, {σn+1 , An+1 },
(k) (k)
Note that for fully associative plasticity (Ψ ≡ Φ), N̄ and H̄ coincide, respectively, with the
flow vector, N, and H.
To solve (7.55) we proceed as follows. We discretise the rate constitutive equations (7.52)
using a forward (or explicit) Euler scheme. This gives the formula
# $
(k+1) (k) e (k) (k) (k) (k)
σn+1 − σn+1 = ∆γ −Dn+1 : N n+1 + En+1 ∗ H n+1
# $ (7.57)
(k+1) (k) (k) (k) (k) (k)
An+1 − An+1 = ∆γ −Fn+1 ∗ N n+1 + Gn+1 ∗ H n+1 .
With the substitution of the above equations into (7.55), the following expression is obtained
for ∆γ in closed form
() # $
(k) (k) e (k) (k) (k) (k)
∆γ = Φn+1 N̄ n+1 : Dn+1 : N n+1 − En+1 ∗ H n+1
# $*
(k) (k) (k) (k) (k)
+ H̄ n+1 ∗ Fn+1 ∗ N n+1 − Gn+1 ∗ H n+1 , (7.58)
(k+1) (k+1)
and the new state {σn+1 , An+1 } is computed by substituting the value obtained for ∆γ
into (7.57).
Starting (with k = 0) from the initial condition of the return-mapping problem
(0) (0)
{σn+1 , An+1 } = {σtrial
n+1 , An+1 },
trial
(7.59)
the repeated application of the above iteration generates a sequence of states
(k) (k)
{σn+1 , An+1 }, k = 0, 1, 2, . . . .
The cutting-plane iterations continue until the value of the yield function is sufficiently close
(k) (k)
to zero, i.e. the iterative process is interrupted at an iteration (k) if the state {σn+1 , An+1 }
satisfies the convergence criterion
(k) (k)
Φ(σn+1 , An+1 ) ≤ tol , (7.60)
where tol is a prescribed convergence tolerance. The cutting-plane return-mapping algorithm
is summarised in Box 7.2 in pseudo-code format.
Remark 7.5. The geometric interpretation of the cutting-plane algorithm is illustrated in
Figure 7.6. The interpretation here remains valid for materials with a nonlinear elastic
(k+1) (k+1)
response. In each iteration, the new stress, {σn+1 , An+1 }, is obtained by projecting the
(k) (k)
current stress, {σn+1 , An+1 }, onto a cutting plane defined by the linearised consistency
condition¶. The projection is made along the direction of the tensor
e (k) (k) (k) (k)
Dn+1 : N n+1 − En+1 ∗ H n+1 .
¶ Note that the linearised consistency condition (7.55) defines a hyperplane in the space of stresses and hardening
forces.
208 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
σ(0) trial
n+1 = σn+1 , A(0) trial
n+1 = An+1
A(k+1) (k) (k) (k)
n+1 = An+1 − ∆γ Fn+1 ∗ N n+1 − Gn+1 ∗ H n+1
(k) (k)
σn+1 = σ(k)
n+1 ; An+1 = A(k)
n+1 ;
In the limit of the iterative process, plastic consistency is restored and the cutting plane is
tangent to the actual yield surface.
Remark 7.6. The iterations of the cutting-plane algorithm converge to the solution Φ = 0 at
quadratic rates. Quadratic rates of convergence are achieved here despite the fact that only
relatively simple function evaluations are performed in each iteration. This is in contrast to the
previously described return-mapping schemes for which the computation of (generally more
complex) residual derivatives is required in order to achieve quadratic rates of convergence.
This makes the cutting-plane algorithm particularly attractive for more complex plasticity
models. However, it is important to emphasise that the incremental stress–strain function
in Section 7.4, to which readers unfamiliar with the concept are referred.
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 209
(0)
σn+1
trial
≡ σn+1
(1)
Φlin (σ , An+1 ) = 0
σn+1
(1)
(2)
σn+1
(2)
Φlin (σ , An+1 ) = 0
.....
cutting
planes
σn+1
Φ(σ, An+1 ) = 0
σn
elastic
domain at tn
Φ(σ, An ) = 0
Within the implicit finite element framework of Chapter 4 (see Section 4.2.5, page 98), the
lack of a consistent tangent modulus represents a serious limitation for it does not allow
the use of the (quadratically convergent) Newton–Raphson algorithm in the solution of the
(global) finite element equilibrium equations. However, in explicit codes, which do not
require the solution of a global system of equilibrium equations, the use of cutting-plane
algorithms could be an attractive option.
where the potential relations (7.8) can be used to express σ and A as explicit functions
of εe and α. Formally, such a system is classed as a system of differential-algebraic
equations (DAE). These generally contain ordinary differential equations (equations (7.63)1,2
in the present case) complemented by a constraint in the form of algebraic equations
(equation (7.63)3).
A detailed account on numerical methods for DAEs is provided in the textbook by
Ascher and Petzold (1998) (see also Gear (1971) for an early reference on the subject).
The identification of plasticity equations with DAEs allows the use of numerical methods
210 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
devised for this type of equation, as well as the corresponding methods of analysis, in the
treatment of plasticity problems. By exploiting these ideas, Papadopoulos and Taylor (1994)
proposed a two-step backward difference formula of second-order accuracy for von Mises
type plasticity. Results from DAE theory were used by these authors to prove the stability
and accuracy order of their method (here, the concepts of accuracy and stability are briefly
reviewed below in Section 7.2.10). The use of such tools in plasticity is also discussed by
Simo (1998).
and stability of elastoplastic integration algorithms. Rather, the purpose of this section is to
provide the reader with some results that can be useful in helping to decide which algorithm
to adopt when considering the computational implementation of a particular plasticity model.
The numerical error, i.e. the maximum difference between exact and numerical
solutions (excluding machine round-off errors) within (t0 , tf ], produced by first-order
accurate algorithms is proportional to the step size, ∆t, as ∆t → 0. If, in addition to
first-order accuracy, the algorithm satisfies
d2
x = ẍ(tn ), (7.68)
d∆t2
n+1
∆t=0
(b) With a more practical (and less precise) definition, the term finite step accuracy will
be adopted here to describe the actual accuracy of particular algorithms under finite
step sizes. In the context of plasticity, finite step accuracy may be measured by means
of numerical experiments in which the state-update algorithm is used to integrate the
elastoplastic equations under a wide range of initial conditions and strain increment
sizes and directions. Finite step accuracy measurements can give important information
on the practical limitations of integration algorithms, especially with regard to the
permissible size of strain increments for which the error remains within reasonable
bounds.
The property of stability of an algorithm for numerical solution of the typical problem (7.64)–
(7.66) relates to how perturbations to the initial condition propagate throughout the stepping
procedure. Generally speaking, an algorithm is said to be stable if the variations in the
numerical solution that result from a perturbation of the initial conditions are bounded within
the domain [t0 , tf ]. The reader is referred, for instance, to Ascher and Petzold (1998) for
various definitions of stability in this context. In the case of perfectly plastic models (Ortiz
and Popov, 1985), an elastic predictor/plastic corrector scheme (under plastic yielding) is
said to be stable if given two arbitrary distinct initial stress states at tn , σn and σ∗n , the
corresponding updated values at tn+1 are bounded by
where d is a measure of distance on the yield surface defined such that for any pair {σa , σb }
of stresses on the yield surface, we have
d(σa , σb ) ≡ inf σ (s) ds, (7.70)
λ λ
with λ denoting arbitrary paths on the yield surface connecting σa and σb . For unstable
algorithms, on the other hand, perturbations to the initial conditions propagate in an
unbounded manner so that small changes in increment size, or even machine round-off
errors, may produce dramatic changes in the numerical results which completely invalidate
the solution.
The property of stability together with first-order accuracy (or consistency of order one)
of the numerical method, are necessary and sufficient conditions for the numerical solution to
converge to the exact solution as the increment size tends to zero. An algorithm that satisfies
the first-order accuracy and stability requirements is said to be convergent. If an algorithm is
not stable or not (at least) first-order accurate, then it does not converge to the exact solution.
implicit Finite Element Method described in Chapter 4 and adopted in the HYPLAS program, it
is desirable that the state-update procedures employed at the Gauss point level be sufficiently
accurate for strain increments as large as possible in order to ensure that the global finite
element solution remains within reasonable bounds of accuracy for large load increments.
From this point of view, the extension of the accuracy order study to the assessment of the
accuracy of elastoplastic integration algorithms under finite steps becomes crucial.
Systematic finite step accuracy analyses of elastoplastic algorithms have been carried
out firstly by Krieg and Krieg (1977) who investigated the behaviour of procedures for
integration of the von Mises perfectly plastic model. The fully implicit algorithm, which for
this particular model is termed the radial return method, was among the procedures assessed
by these authors. Their assessment was based on the use of iso-error maps. It is remarked
that, since the original application by Krieg and Krieg (1977), iso-error maps have proved
very effective and are currently accepted as the most reliable (if not the only) tool for the
assessment of the finite step accuracy of integration algorithms for elastoplasticity (de Souza
Neto et al., 1994a; Dutko et al., 1993; Fuschi et al., 1992; Ortiz and Popov, 1985; Simo and
Taylor, 1986).
In order to generate a typical iso-error map, consider an arbitrary stress state at a point P on
the yield surface as shown in Figure 7.7(a). Here, the perfectly plastic associative von Mises
model is adopted for illustration, with the yield surface of Figure 7.7(a) represented in the
deviatoric plane. From this point a sequence of strain increments is applied corresponding to
specified normalised elastic trial stress increments of the form
∆σT ∆σN
∆σtrial = T+ N, (7.71)
q q
where N and T are, respectively, the unit (in Euclidean norm) normal and tangent vectors to
the yield surface and q is the von Mises equivalent stress at P . As a result of the numerical
integration algorithm, an approximated stress σnum is computed for each increment. Calling
σexact the exact solution of the stress integration problem, the error associated with each
increment is defined as
(σexact − σnum ) : (σexact − σnum )
ERROR = √ . (7.72)
σexact : σexact
By varying the prescribed increment sizes ∆σT and ∆σN , respectively associated with the
tangential and normal directions to the yield surface, an error field is obtained. The contour
plot of this error field is the iso-error map for P . The iso-error map corresponding to the fully
implicit scheme for integration of the perfectly plastic von Mises model under general stress
state is shown in Figure 7.7(b). For this particular model, the starting point P is immaterial.
It is important to note that, as analytical solutions are generally not available, σexact is taken
as the numerical solution obtained by dividing each strain increment into a sufficiently large
number of subincrements∗∗. To give a general idea of reasonable sub-increment sizes, it is
worth mentioning that one thousand sub-increments have been used to produce the ‘exact’
solutions for the map of Figure 7.7(b). It should be emphasised that the particular increment
directions and error measure employed above are suitable for the von Mises perfectly plastic
∗∗ Note that this assumption is valid if the algorithm under study is convergent, i.e. first-order accurate and stable,
∆σT
N
trial
σ trial 5%
10%
∆σ
4
∆σ N ⁄ q
P T ∆σN
15%
2
20%
0
0 2 4 6
∆σ T ⁄ q
(a) (b)
Figure 7.7. Iso-error map: (a) typical increment directions; and (b) a typical iso-error map.
model, but may not be so for different models. In the case of hardening materials, for instance,
it might be useful to include the error associated with the hardening internal variable in the
definition of the error measure.
Having described the concept of iso-error maps, let us now return to the discussion regard-
ing the properties of the integration algorithms of Sections 7.2.3 and 7.2.7 concentrating on
their finite step accuracy characteristics. A finite step accuracy analysis of the generalised
trapezoidal and midpoint rule integration algorithms, based on iso-error maps, has been
carried out by Ortiz and Popov (1985) in the same paper in which they presented the analysis
of accuracy and stability of these algorithms. The iso-error maps discussed by these authors
were restricted to the perfectly plastic von Mises model – a model for which the generalised
trapezoidal and midpoint algorithms coincide. Essentially, these authors have found that, for
small increments, the best accuracy is obtained with θ = 12 . This is in obvious agreement with
the (infinitesimal) accuracy analysis that established second-order accuracy for this particular
choice of θ. With increasing increment size, however, the performance of the second-order
algorithm deteriorates. For reasonably sized increments, likely to occur in real problems, a
choice of θ between 0.7 and 0.8 gives the best accuracy. Also, the fully implicit algorithm
(θ = 1), which will be adopted in the numerical implementation described in the following
sections of this chapter, gives better accuracy than the second-order algorithm. It is worth
remarking that completely analogous results have been found by Fuschi et al. (1992) for the
variant of the midpoint rule based algorithm proposed by Simo and Govindjee (1991).
implicit algorithm is exclusively adopted throughout this book (and in the HYPLAS program)
for integration of elastoplastic models. Its particularisation for the Tresca, Mohr–Coulomb
and Drucker–Prager models is described later, in Chapter 8.
The choice of the fully implicit algorithm here and in the remainder of this book is
essentially motivated by:
(i) its (unconditional) stability which, in conjunction with its first-order accuracy, ensures
that the resulting algorithm is convergent;
(ii) its generally ‘good’ finite step accuracy;
(iii) its suitability for the derivation of associated consistent tangent operators – a property
(not shared by the cutting-plane algorithm) that is absolutely essential for its use in
conjunction with a (quadratically convergent) global Newton–Raphson procedure; and
(iv) its relatively simple computational implementation, as compared to other members
of the families of generalised trapezoidal and midpoint algorithms discussed in
Section 7.2.7.
4. An associative hardening rule, with the evolution equation for the hardening internal
variable given by
ε̄˙p = 23 ε̇p = γ̇. (7.78)
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 217
strial e trial
n+1 = 2G εd n+1 , ptrial e trial
n+1 = K εv n+1 , (7.82)
where s and p denote, respectively, the deviatoric and hydrostatic stresses, G and K are,
respectively, the shear and bulk moduli and the subscripts d and v in the elastic trial strain
denote, respectively, the deviatoric and volumetric components. The trial yield stress is simply
σytrial p
n+1 = σy (ε̄n ) = σy n . (7.83)
Having computed the elastic trial state, the next step in the algorithm is to check whether
σtrial
n+1 lies inside or outside of the trial yield surface:
• If σtrial
n+1 lies inside of the trial yield surface, i.e. if
n+1 , σy n ) ≤ 0,
Φ(σtrial
then the process within the interval [tn , tn+1 ] is purely elastic and the elastic trial state
itself is the solution to the integration problem. In this case,
trial
εen+1 = εen+1
σn+1 = σtrial
n+1
(7.84)
trial
ε̄pn+1 = ε̄pn+1 = ε̄pn
σy n+1 = σytrial
n+1 = σy n
is updated.
• Otherwise, the process is elastoplastic within the interval [tn , tn+1 ] and the return-
mapping procedure described below has to be applied.
Recall that the general implicit return-mapping procedure corresponds to solving the
system (7.25) of nonlinear equations. In the present case, direct specialisation of the general
fully implicit return-mapping equations (7.25) to the von Mises model gives the following set
218 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
of nonlinear equations:
%
3 sn+1
εen+1 = trial
εen+1 − ∆γ
2 sn+1
ε̄pn+1 = ε̄pn + ∆γ (7.85)
3 J2 (sn+1 ) − σy (ε̄pn+1 ) = 0,
which has to be solved for εen+1 , ε̄pn+1 and ∆γ and where
sn+1 = sn+1 (εen+1 ) = 2G dev[εen+1 ]. (7.86)
After the solution of the above system, the plastic strain tensor can be updated according to
the following formula: %
p p 3 sn+1
εn+1 = εn + ∆γ . (7.87)
2 sn+1
i.e. the trial and updated deviatoric stresses are co-linear. This implies that
sn+1 strial
= n+1 ,
sn+1 strial
n+1
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 219
σn+1
trial
updated
yield surface
√
3 p
Nn+1
σ3 σn+1 initial
yield surface
σn
σ2
σ1
Figure 7.8. The implicit elastic predictor/return-mapping scheme for the von Mises model. Geometric
interpretation in principal stress space.
so that the flow vectors at the trial and updated states coincide. Substitution of the above
identity into (7.89)2 leads to the following simpler update formula for the deviatoric stress:
%
3 ∆γ 2G trial
sn+1 = 1 − sn+1
2 strial
n+1
∆γ 3G
= 1 − trial strialn+1 , (7.90)
qn+1
trial
where qn+1 ≡ 3 J2 (strial
n+1 ) is the elastic trial von Mises effective stress. Note that, since
strial
n+1 is a constant tensor in the return mapping, the deviatoric stress, sn+1 , is a (linear)
function of ∆γ only in the above update formula. Expression (7.90) implies that, in the
fully implicit algorithm for the von Mises model, the updated deviatoric stress is obtained
by scaling down the trial deviatoric stress by the factor 1 − ∆γ 3G/qn+1 trial
. The geometric
representations of this update formula in the principal stress space and deviatoric plane are
illustrated, respectively, in Figures 7.8 and 7.9.
Finally, with substitution of (7.90) and (7.85)2 into the plastic consistency condi-
tion (7.85)3, the system (7.85) of equations of the return mapping for the von Mises model is
reduced to the following scalar (generally nonlinear) equation having the incremental plastic
multiplier, ∆γ, as the only unknown:
Φ̃(∆γ) ≡ qn+1
trial
− 3G ∆γ − σy (ε̄pn + ∆γ) = 0. (7.91)
220 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
sn+1
trial
Nn+1
1 ||
sn+trial
π --plane
) ||
n+1 l
ia
G/q tr
sn+1
3
(∆γ
sn
surface at t n
updated surface
at t n+1
Figure 7.9. The implicit elastic predictor/return-mapping scheme for the von Mises model. Geometric
interpretation in the deviatoric plane.
The above equation is then solved by the Newton–Raphson method and, with its solution ∆γ
at hand, the state variables are updated as follows:
∆γ 3G trial
sn+1 = 1 − trial sn+1
qn+1
σn+1 = sn+1 + ptrial
n+1 I
(7.92)
e −1 1 1
εen+1 = [D ] : σn+1 = sn+1 + εev trial
2G 3 n+1
ε̄pn+1 = ε̄pn + ∆γ.
From the update formulae (7.92)1,2 and relations (7.81,7.82) we establish after simple
manipulations that the updated stress tensor, σn+1 , can be expressed in terms of ε̄pn and
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 221
(i) Elastic predictor. Given ∆ε and the state variables at tn , evaluate the elastic trial state
εen+1
trial
:= εen + ∆ε
ε̄pn+1
trial
:= ε̄pn
ptrial e trial
n+1 := K εv n+1 ; strial e trial
n+1 := 2G εd n+1
trial
qn+1 := 32 strial trial
n+1 : sn+1
for ∆γ using the Newton–Raphson method – GOTO Box 7.4 – and update the state
variables
∆γ 3G trial
pn+1 := ptrial
n+1 ; sn+1 := 1 − trial
sn+1
qn+1
σn+1 := sn+1 + pn+1 I
εen+1 = 1
2G
sn+1 + 1
3
εev trial
n+1 I
ε̄pn+1 := ε̄pn + ∆γ
(iv) EXIT
trial
εen+1 by means of the following incremental constitutive function:
2
trial ∆γ 6G
σn+1 = σ̄n+1 (ε̄n , εn+1 ) ≡ D − Ĥ(Φ
p e trial e
) trial
trial
Id : εen+1 , (7.93)
qn+1
= qn+1 (εn+1 ) ≡ 2G 32 Id : εen+1
trial e trial trial
, (7.95)
222 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
∆γ (0) := 0
Φtrial is the value of the yield function at the elastic trial state:
and
trial
∆γ = ∆γ(ε̄pn , εen+1 ) (7.97)
trial
is the implicit function of εen+1and ε̄pn
defined as the solution of the consistency equa-
tion (7.91).
Clearly, (7.93) defines σn+1 as an implicit function of the elastic trial strain and ε̄pn .
trial
Equivalently, since εen+1 = εn+1 − εpn , we may write
σn+1 = σ̂n+1 (ε̄pn , εpn , εn+1 ) ≡ σ̄n+1 (ε̄pn , εn+1 − εpn ). (7.98)
For a given state at tn , the functions (7.93) and (7.98) express the updated stress as implicit
functions, respectively, of the elastic trial stress and the total elastic strain at tn+1 .
Remark 7.7. The use of incremental algorithmic constitutive functions has been first alluded
to in this book in Chapter 4 in the formulation of incremental boundary value problems
involving path-dependent materials. The reader is referred to expression (4.60) (page 95),
and the text surrounding it for details. The function σ̂ defined in (7.98, 7.93) is the
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 223
particularisation of the generic incremental function (4.60) for the von Mises model with
isotropic strain hardening (for which αn = {ε̄pn , εpn }) integrated numerically by the fully
implicit elastic predictor/return-mapping algorithm.
sn+1
trial
Nn+1
sn sn+1
π
- -plane
Figure 7.10. The perfectly plastic von Mises model. Geometric interpretation of the implicit return-
mapping scheme as the closest point projection algorithm.
hardening
curve
σy
.....
( n hard ε p, n hard σ y )
( 3 ε p , 3 σy )
( 2 ε p , 2 σy )
piecewise linear
approximation with
n hard sampling points
( 1 ε p , 1 σy )
εp
Figure 7.11. Piecewise linear hardening.
σy (ε̄p )
has been assumed to be piecewise linear. This means that any (arbitrarily nonlinear)
hardening curve can be adequately approximated by using a sufficiently large number of
sampling pairs
{ε̄p , σy }
with linear interpolation between adjacent pairs. A piecewise linear approximation, with
nhard sampling points, to a generic hardening curve is illustrated in Figure 7.11. In fact, actual
experimental data for hardening curves are normally obtained as a set of points (σy , ε̄p ).
Thus, the present implementation of the von Mises model allows for the experimental data to
be used directly in the definition of the hardening curve.
The FORTRAN source code of SUVM is listed below.
1 SUBROUTINE SUVM
2 1( DGAMA ,IPROPS ,LALGVA ,NTYPE ,RPROPS ,
3 2 RSTAVA ,STRAT ,STRES )
4 IMPLICIT DOUBLE PRECISION (A-H,O-Z)
5 PARAMETER(IPHARD=4 ,MSTRE=4)
6 LOGICAL IFPLAS, LALGVA(2), SUFAIL
7 DIMENSION
8 1 IPROPS(*) ,RPROPS(*) ,RSTAVA(MSTRE+1) ,
9 2 STRAT(MSTRE) ,STRES(MSTRE)
10 DIMENSION
11 1 EET(MSTRE)
12 DATA
13 1 R0 ,RP5 ,R1 ,R2 ,R3 ,TOL /
14 2 0.0D0,0.5D0,1.0D0,2.0D0,3.0D0,1.D-06/
15 DATA MXITER / 50 /
16 C***********************************************************************
17 C STATE UPDATE PROCEDURE FOR THE VON MISES ELASTO-PLASTIC MATERIAL MODEL
18 C WITH NON-LINEAR (PIECEWISE LINEAR) ISOTROPIC HARDENING:
19 C IMPLICIT ELASTIC PREDICTOR/RETURN MAPPING ALGORITHM (BOXES 7.3-4).
20 C PLANE STRAIN AND AXISYMMETRIC IMPLEMENTATIONS.
21 C***********************************************************************
22 C Stop program if neither plane strain nor axisymmetric state
23 IF(NTYPE.NE.2.AND.NTYPE.NE.3)CALL ERRPRT(’EI0013’)
24 C Initialise some algorithmic and internal variables
25 DGAMA=R0
26 IFPLAS=.FALSE.
27 SUFAIL=.FALSE.
28 EPBARN=RSTAVA(MSTRE+1)
29 C Set some material properties
30 YOUNG=RPROPS(2)
31 POISS=RPROPS(3)
32 NHARD=IPROPS(3)
33 C Shear and bulk moduli and other necessary constants
34 GMODU=YOUNG/(R2*(R1+POISS))
35 BULK=YOUNG/(R3*(R1-R2*POISS))
36 R2G=R2*GMODU
37 R3G=R3*GMODU
38 C Elastic predictor: Compute elastic trial state
39 C ----------------------------------------------
40 C Volumetric strain and pressure stress
41 EEV=STRAT(1)+STRAT(2)+STRAT(4)
226 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
42 P=BULK*EEV
43 C Elastic trial deviatoric strain
44 EEVD3=EEV/R3
45 EET(1)=STRAT(1)-EEVD3
46 EET(2)=STRAT(2)-EEVD3
47 EET(4)=STRAT(4)-EEVD3
48 C Convert engineering shear component into physical component
49 EET(3)=STRAT(3)/R2
50 C Compute trial effective stress and uniaxial yield stress
51 VARJ2T=R2G*R2G*(EET(3)*EET(3)+RP5*(EET(1)*EET(1)+
52 1 EET(2)*EET(2)+EET(4)*EET(4)))
53 QTRIAL=SQRT(R3*VARJ2T)
54 SIGMAY=PLFUN(EPBARN,NHARD,RPROPS(IPHARD))
55 C Check for plastic admissibility
56 C -------------------------------
57 PHI=QTRIAL-SIGMAY
58 IF(PHI/SIGMAY.GT.TOL)THEN
59 C Plastic step: Apply return mapping - use Newton-Raphson algorithm
60 C to solve the return mapping equation (Box 7.4)
61 C -------------------------------------------------------------------
62 IFPLAS=.TRUE.
63 EPBAR=EPBARN
64 DO 10 NRITER=1,MXITER
65 C Compute residual derivative
66 DENOM=-R3G-DPLFUN(EPBAR,NHARD,RPROPS(IPHARD))
67 C Compute Newton-Raphson increment and update variable DGAMA
68 DDGAMA=-PHI/DENOM
69 DGAMA=DGAMA+DDGAMA
70 C Compute new residual
71 EPBAR=EPBAR+DDGAMA
72 SIGMAY=PLFUN(EPBAR,NHARD,RPROPS(IPHARD))
73 PHI=QTRIAL-R3G*DGAMA-SIGMAY
74 C Check convergence
75 RESNOR=ABS(PHI/SIGMAY)
76 IF(RESNOR.LE.TOL)THEN
77 C update accumulated plastic strain
78 RSTAVA(MSTRE+1)=EPBAR
79 C update stress components
80 FACTOR=R2G*(R1-R3G*DGAMA/QTRIAL)
81 STRES(1)=FACTOR*EET(1)+P
82 STRES(2)=FACTOR*EET(2)+P
83 STRES(3)=FACTOR*EET(3)
84 STRES(4)=FACTOR*EET(4)+P
85 C compute converged elastic (engineering) strain components
86 FACTOR=FACTOR/R2G
87 RSTAVA(1)=FACTOR*EET(1)+EEVD3
88 RSTAVA(2)=FACTOR*EET(2)+EEVD3
89 RSTAVA(3)=FACTOR*EET(3)*R2
90 RSTAVA(4)=FACTOR*EET(4)+EEVD3
91 GOTO 999
92 ENDIF
93 10 CONTINUE
94 C reset failure flag and print warning message if the algorithm fails
95 SUFAIL=.TRUE.
96 CALL ERRPRT(’WE0004’)
97 ELSE
98 C Elastic step: Update stress using linear elastic law
99 C ----------------------------------------------------
100 STRES(1)=R2G*EET(1)+P
101 STRES(2)=R2G*EET(2)+P
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 227
102 STRES(3)=R2G*EET(3)
103 STRES(4)=R2G*EET(4)+P
104 C elastic engineering strain
105 RSTAVA(1)=STRAT(1)
106 RSTAVA(2)=STRAT(2)
107 RSTAVA(3)=STRAT(3)
108 RSTAVA(4)=STRAT(4)
109 ENDIF
110 999 CONTINUE
111 C Update some algorithmic variables before exit
112 LALGVA(1)=IFPLAS
113 LALGVA(2)=SUFAIL
114 RETURN
115 END
← LALGVA. Array of logical algorithmic flags or variables. For the von Mises model it
contains the plastic yielding flag, IFPLAS, and the state update failure flag SUFAIL. If
the step is elastic, the plastic yielding flag returns as .FALSE., otherwise its return value
is .TRUE.. The state update failure flag returns as .TRUE. only if the Newton scheme of
the return-mapping algorithm fails to converge. In this case, the state variables are not
updated in SUVM and a warning message is sent (by calling subroutine ERRPRT) to the
results file and standard output. When the state update procedure fails (for any material
model), increment cutting is activated in the main program; that is, the calculations for
the current load increment are aborted and restarted with a smaller increment.
→ RPROPS. Array of real material properties. It contains the elastic properties (Young’s
modulus, E, and Poisson’s ratio, ν) and the pairs {i ε̄p , i σy }, i = 1, . . . , nhard , of
sampling points along the hardening curve (see illustration of Figure 7.11):
Array RPROPS is set in subroutine RDVM during the input phase of HYPLAS.
↔ RSTAVA [εe , ε̄p ]. Array of state variables other than the stress components. For the
present material model implementation, this array contains the elastic strain (returned
as the updated value εen+1 ) and the equivalent plastic strain (last converged solution
value ε̄pn on entry, updated value ε̄pn+1 on exit). The last converged elastic strain, εen ,
is used in subroutine MATISU to compute the elastic trial strain before the present
subroutine is called.
228 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
→ STRAT [εen+1
trial
]. Array containing the elastic trial (engineering) strains. It is computed
as shown in item (i) of Box 7.3 in subroutine MATISU. The calculation of array STRAT in
MATISU is common to all elastoplastic material models of the class whose identification
parameter is HYPEPL.
Most remaining local variables of SUVM have been named so as to resemble the corresponding
notation of Section 7.3.2. Thus, the operations carried out in SUVM can be easily identified
with those indicated in Boxes 7.3 and 7.4.
• DPLFUN. This function computes the derivative of a piecewise linear function. Used in
SUVM to compute the slope of the hardening curve, H.
• PLFUN. Piecewise linear function defined by a set of pairs (x, f (x)). Used in SUVM as
the hardening function σy (ε̄p ). Recall that the hardening curve is defined here by the
NHARD pairs of sampling points stored in array RPROPS.
∂σn+1
D≡ , (7.104)
∂εn+1
Elastic materials
In the case of purely elastic materials, the stress is an explicit function of the strain tensor
∂ψ(ε)
σ = σ(ε) = ρ̄ , (7.105)
∂ε
so that the tangent modulus can be explicitly derived as
∂ 2 ψ(ε)
D = De ≡ ρ̄ . (7.106)
∂ε2
For linear elastic materials, the consistent tangent modulus is the standard (constant) elasticity
tensor
D = 2G Id + K I ⊗ I. (7.107)
Algorithmic functions of this type have been first referred to in Section 4.2.1 (page 95),
in the formulation of incremental boundary value problems with general path-dependent
material models. Specific examples of incremental constitutive functions have been obtained
earlier in this chapter for the fully implicit elastic predictor/return-mapping implementation
of the von Mises model with isotropic strain hardening. Expressions (7.93, 7.98) show
the corresponding (implicit) incremental constitutive function for the model with nonlinear
hardening and expression (7.103) shows the particular (explicit) format obtained under linear
hardening.
Within a load increment [tn , tn+1 ], the internal variable set αn given as argument of σ̂
is fixed. Only the guesses for the total strain, εn+1 – associated with the guesses for the
displacement field, un+1 – change during the global Newton–Raphson equilibrium iterations
(refer to Section 4.2, from page 94 for details on the global Newton–Raphson procedure). In
other words, within each global load increment, the stress σn+1 delivered by the integration
algorithm is a function of the total strain tensor only. This function – σ̂(αn , εn+1 ) with
fixed αn – defines a path-independent stress/strain relation within the interval [tn , tn+1 ],
equivalent to a (nonlinear) elastic law. The consistent tangent modulus in this case is precisely
the derivative of this equivalent nonlinear elastic law:
dσn+1 ∂ σ̂
D≡ = , (7.109)
dεn+1 ∂εn+1 αn
i.e. it is the derivative of the algorithmic function σ̂ with respect to εn+1 with αn held
constant.
230 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
algorithmic stress
constitutive function
σn+1 = σ^ (α n , εn+1 )
⇔
state update procedure
input output
Figure 7.12. The algorithmic constitutive function for the stress tensor.
At this point, it is worth remarking that, in the context of the multiplicative finite
strain plasticity framework discussed in Chapter 14, no measure of total nonlinear strain
analogous to εn+1 is used (or needed) in the definition of elastoplastic constitutive models.
trial
An elastic trial strain measure (analogous to the infinitesimal tensor εen+1 ), however, does
appear naturally in the formulation of the corresponding elastic predictor/return-mapping
trial
schemes. Clearly, since εen+1 = εn+1 − εpn , the incremental constitutive function for σn+1
can always be expressed equivalently as††
trial
σn+1 = σ̄(αn , εen+1 ) ≡ σ̂(αn , εen+1
trial
+ εpn ), (7.110)
in terms of the elastic trial strain and the internal variable set at tn , and we also have the trivial
identity
∂ σ̂ ∂ σ̄
D= = e trial (7.111)
∂εn+1 ∂εn+1
for the consistent tangent operator. To make the material presented here formally valid also
for the large-strain case addressed in Chapter 14 (where a total strain tensor is not defined),
we shall adopt in what follows the rightmost term of (7.111) as the definition of consistent
tangent operator.‡‡
strain εn+1 , as the actual input argument to the elastoplastic integration procedures for models of the present type
(see routine SUVM, listed on page 227, for instance).
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 231
YE
σn 1
unloading
εn 1 εn 1
Figure 7.13. The tangent moduli consistent with elastic predictor/return-mapping integration algo-
rithms.
to Φtrial < 0 in (7.93), (7.98), any infinitesimal change of total strain can only be elastic, with
the stress σn+1 evolving along the (smooth) elastic curve (see graphical representation of the
uniaxial stress case in Figure 7.13). In this case the function σ̂ is differentiable. At states with
Φtrial > 0, the function σ̂ is also differentiable if the hardening curve is smooth. Infinitesimal
changes of εn+1 will move the stress along the elastoplastic part of the incremental stress–
strain curve. However, at states with Φtrial = 0 in (7.93, 7.98) – where the Heaviside step
function is non-differentiable – either elastic unloading or plastic straining may occur in
the incremental constitutive law. The incremental constitutive function is obviously non-
differentiable in this case. The tangent modulus D is not uniquely defined and two tangent
stress–strain relations exist: an elastic tangent relation, defined for elastic unloading, and an
elastoplastic tangent relation, defined for plastic loading. Consider the one-dimensional case
illustrated in Figure 7.13. Even though σ̂ is non-differentiable its two one-sided derivatives
– the elastic and the elastoplastic tangents – are well defined. In the multidimensional case,
these are generalised respectively as the elastic tangent modulus, De , and the elastoplastic
consistent tangent modulus, Dep . The elastic tangent is associated with the elastic predictor
procedure whereas the elastoplastic tangent is related to the plastic corrector (return-mapping)
procedure. Clearly, when assembling the tangent stiffness matrix required by the Newton–
Raphson iterative procedure for the global incremental equilibrium problem, the appropriate
choice of tangent operator must be made.
∂2ψ
D = De ≡ ρ̄ , (7.112)
∂εe 2
232 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
i.e. it is the standard elastic modulus. Note that, in this case, the stress σn+1 is the outcome
of the elastic predictor, which employs only the elastic constitutive law. This procedure is
common to all algorithms described in Section 7.2.
For elastoplastic materials whose elastic response is linear, such as the von Mises model
and all other models described in Chapter 6, the elastic consistent tangent is the standard
elasticity operator (7.107).
∂ σ̂
Dep = trial
∂εen+1
is simply the derivative of the implicit function defined by the return-mapping equations and
is derived by following the standard procedure for differentiation of implicit functions.
As an illustration of the above concepts, the elastoplastic tangent consistent with the fully
implicit algorithm for the von Mises model is derived below.
procedure to the isotropically hardening von Mises model, which offers an alternative route
to the derivation presented below, is described in Section 7.4.5.
The incremental algorithmic constitutive function for the implicitly integrated von Mises
model with nonlinear isotropic strain hardening is given by (7.93). Under plastic flow, i.e.
when the return-mapping procedure is used, the update formula for σn+1 reads
∆γ 6G2
σn+1 = D − trial Id : εen+1
e trial
, (7.113)
qn+1
where ∆γ is the solution of the return-mapping equation of the algorithm (Box 7.3),
Φ̃(∆γ) ≡ qn+1
trial
− 3G ∆γ − σy (ε̄pn + ∆γ) = 0. (7.114)
trial
In the above, the elastic trial von Mises effective stress, qn+1 , is the function of the elastic
trial strain defined by (7.95). The elastoplastic consistent tangent modulus for the present
model/algorithm combination is obtained by differentiating (7.113).
A straightforward application of tensor differentiation rules to (7.113) gives
∂σn+1 ∆γ 6G2 6G2 ∂∆γ
e trial
= De − trial Id − trial εed trial
n+1 ⊗ trial
∂εn+1 qn+1 qn+1 ∂εen+1
trial
∆γ 6G2 e trial ∂qn+1
+ trial )2
ε d n+1 ⊗ trial
. (7.115)
(qn+1 ∂εen+1
From (7.95) and relation (2.139) (page 36), for the tensor norm derivative, we obtain
trial
∂qn+1
3
trial
= 2G 2 N̄ n+1 , (7.116)
∂εen+1
where we have conveniently defined the unit flow vector
strial εed trial
n+1 n+1
N̄n+1 ≡ 2
Nn+1 = = (7.117)
3
strial
n+1 ε e trial
d n+1
and we have made use of the trivial identity: εed trial e trial
n+1 : Id = εd n+1 , when applying the chain
rule. Further, the differentiation of the implicit equation (7.114) for ∆γ, taking (7.116) into
account gives
trial
∂∆γ 1 ∂qn+1
e trial
= e trial
∂εn+1 3G + H ∂εn+1
2G
3
= N̄n+1 , (7.118)
3G + H 2
where H is the slope of the hardening curve:
dσy
H ≡ p . (7.119)
dε̄ ε̄pn +∆γ
Finally, by substituting (7.116) and (7.118) into (7.115), we obtain, after a straightforward
manipulation making use of (7.95) and definition (7.117), the following expression for the
234 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
elastoplastic tangent operator consistent with the implicit return-mapping scheme for the
isotropically hardening von Mises model:
∆γ 6G2 ∆γ 1
D = D − trial Id + 6G
ep e 2
trial
− N̄n+1 ⊗ N̄ n+1
qn+1 qn+1 3G + H
∆γ 3G
= 2G 1 − trial Id
qn+1
∆γ 1
+ 6G2 trial − N̄ n+1 ⊗ N̄ n+1 + K I ⊗ I. (7.120)
qn+1 3G + H
It should be noted that the operator Dep in the present case, i.e. for this particular model
and numerical integration algorithm is symmetric. The symmetry of consistent elastoplastic
tangent operators will be further commented upon in Section 7.4.6. In the HYPLAS program,
the above tangent operator is computed in subroutine CTVM. Its implementation is described
in detail in Section 7.4.3.
Remark 7.8. Within the global (equilibrium) Newton–Raphson scheme, the value of ∆γ,
q trial and H, as well as the incremental unit flow vector, N̄ n+1 , that take part in (7.120) are
those obtained for the Gauss point of interest in the return-mapping procedure of the previous
global iteration. For the first iteration of any global load increment, ∆γ is zero.
∂ 2 ψp ∂ 2 ψp ∂κ
ρ̄ = ρ̄ = p = H. (7.121)
∂α 2
∂ ε̄p 2 ∂ ε̄
With the above, together with (6.194) and the associative flow vector definition (6.136) for
the von Mises model, we find that expression (6.67) particularises in the following format:
(De : N ) ⊗ (De : N )
c =D −
Dep e
, (7.122)
N : De : N + H
where we have used the subscript ‘c’ to emphasise that the above operator is the continuum
tangent modulus. With De defined by (7.107), and taking into consideration the fact that for
the von Mises model N is a deviatoric tensor, we have
De : N = 2G N. (7.123)
N : De : N = 3G. (7.124)
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 235
By introducing these results into (7.122), we obtain the following explicit expression for the
continuum tangent operator for the von Mises model with isotropic strain hardening:
6G2
c =D −
Dep N̄ ⊗ N̄,
e
(7.125)
3G + H
where N̄ is the unit flow vector at the current state.
Remark 7.9. The difference between the elastoplastic consistent tangent operator and its
continuum counterpart above lies only in the terms that contain ∆γ in expression (7.120).
Note that we may write
∆γ 6G2
c −
Dep = Dep trial
[ Id − N̄ n+1 ⊗ N̄ n+1 ]. (7.126)
qn+1
If ∆γ is set to zero (as in the first iteration of any load increment), the continuum tangent is
recovered. This fact (Ortiz and Martin, 1989) is a mere consequence of the consistency of the
numerical method (backward Euler-based in the present case) adopted in the discretisation of
the time-continuum elastoplasticity equations. For large steps, when the value of ∆γ is large,
the difference between the continuum and the consistent operator can be substantial. In such
cases, the use of the continuum tangent in the assemblage of the stiffness matrix results in a
dramatic degradation of the convergence rate of the global iterative procedure. Clearly, if the
continuum tangent is used in conjunction with the return-mapping scheme, then the global
iterative procedure is not the Newton–Raphson algorithm. In this case, the global iterations
are a form of approximation to the Newton–Raphson scheme. Early implicit elastoplastic
implementations (Owen and Hinton, 1980) relied exclusively on the use of the continuum
tangent operator. The use of the consistent tangent operator in this context was introduced by
Simo and Taylor (1985).
1 SUBROUTINE CTVM
2 1( DGAMA ,DMATX ,EPFLAG ,IPROPS ,NTYPE ,
3 2 RPROPS ,RSTAVA ,STRES )
4 IMPLICIT DOUBLE PRECISION (A-H,O-Z)
5 PARAMETER(IPHARD=4 ,MSTRE=4)
6 LOGICAL EPFLAG
7 DIMENSION
8 1 DMATX(MSTRE,MSTRE),IPROPS(*) ,RPROPS(*) ,
9 2 RSTAVA(MSTRE+1) ,STRES(MSTRE)
10 DIMENSION
236 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
11 1 DEVPRJ(MSTRE,MSTRE),FOID(MSTRE,MSTRE) ,S(MSTRE) ,
12 2 SOID(MSTRE)
13 DATA
14 1 FOID(1,1),FOID(1,2),FOID(1,3),FOID(1,4)/
15 2 1.0D0 ,0.0D0 ,0.0D0 ,0.0D0 /
16 3 FOID(2,1),FOID(2,2),FOID(2,3),FOID(2,4)/
17 4 0.0D0 ,1.0D0 ,0.0D0 ,0.0D0 /
18 5 FOID(3,1),FOID(3,2),FOID(3,3),FOID(3,4)/
19 6 0.0D0 ,0.0D0 ,0.5D0 ,0.0D0 /
20 7 FOID(4,1),FOID(4,2),FOID(4,3),FOID(4,4)/
21 8 0.0D0 ,0.0D0 ,0.0D0 ,1.0D0 /
22 DATA
23 1 SOID(1) ,SOID(2) ,SOID(3) ,SOID(4) /
24 2 1.0D0 ,1.0D0 ,0.0D0 ,1.0D0 /
25 DATA
26 1 R1 ,R2 ,R3 ,R6 /
27 2 1.0D0,2.0D0,3.0D0,6.0D0/
28 C***********************************************************************
29 C COMPUTATION OF THE CONSISTENT TANGENT MODULUS FOR VON MISES TYPE
30 C ELASTO-PLASTIC MATERIAL WITH PIECE-WISE LINEAR ISOTROPIC HARDENING.
31 C PLANE STRAIN AND AXISYMMETRIC IMPLEMENTATIONS.
32 C***********************************************************************
33 C Stops program if neither plane strain nor axisymmetric state
34 IF(NTYPE.NE.2.AND.NTYPE.NE.3)CALL ERRPRT(’EI0030’)
35 C Current accumulated plastic strain
36 EPBAR=RSTAVA(MSTRE+1)
37 C Set material properties
38 YOUNG=RPROPS(2)
39 POISS=RPROPS(3)
40 NHARD=IPROPS(3)
41 C Shear and bulk moduli
42 GMODU=YOUNG/(R2*(R1+POISS))
43 BULK=YOUNG/(R3*(R1-R2*POISS))
44 R2G=R2*GMODU
45 R1D3=R1/R3
46 C Set deviatoric projection tensor
47 IF(NTYPE.EQ.2)THEN
48 NSTRE=3
49 ELSEIF(NTYPE.EQ.3)THEN
50 NSTRE=4
51 ENDIF
52 DO 20 I=1,NSTRE
53 DO 10 J=1,NSTRE
54 DEVPRJ(I,J)=FOID(I,J)-SOID(I)*SOID(J)*R1D3
55 10 CONTINUE
56 20 CONTINUE
57 IF(EPFLAG)THEN
58 C Compute elastoplastic consistent tangent
59 C ----------------------------------------
60 R3G=R3*GMODU
61 ROO3D2=SQRT(R3/R2)
62 C Hydrostatic pressure
63 P=(STRES(1)+STRES(2)+STRES(4))*R1D3
64 C Deviatoric stress components
65 S(1)=STRES(1)-P
66 S(2)=STRES(2)-P
67 S(3)=STRES(3)
68 S(4)=STRES(4)-P
69 C Recover last elastic trial von Mises effective stress
70 SNORM=SQRT(S(1)*S(1)+S(2)*S(2)+R2*S(3)*S(3)+S(4)*S(4))
71 Q=ROO3D2*SNORM
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 237
72 QTRIAL=Q+R3G*DGAMA
73 C Assemble elastoplastic tangent (upper triangle only)
74 AFACT=R2G*(R1-R3G*DGAMA/QTRIAL)
75 BFACT=R6*GMODU*GMODU*(DGAMA/QTRIAL-
76 1 R1/(R3G+DPLFUN(EPBAR,NHARD,RPROPS(IPHARD))))/
77 2 (SNORM*SNORM)
78 DO 40 I=1,NSTRE
79 DO 30 J=I,NSTRE
80 DMATX(I,J)=AFACT*DEVPRJ(I,J)+BFACT*S(I)*S(J)+
81 1 BULK*SOID(I)*SOID(J)
82 30 CONTINUE
83 40 CONTINUE
84 ELSE
85 C Compute elasticity matrix (upper triangle only)
86 C -----------------------------------------------
87 DO 60 I=1,NSTRE
88 DO 50 J=I,NSTRE
89 DMATX(I,J)=R2G*DEVPRJ(I,J)+BULK*SOID(I)*SOID(J)
90 50 CONTINUE
91 60 CONTINUE
92 ENDIF
93 C Assemble lower triangle
94 C -----------------------
95 DO 80 J=1,NSTRE-1
96 DO 70 I=J+1,NSTRE
97 DMATX(I,J)=DMATX(J,I)
98 70 CONTINUE
99 80 CONTINUE
100 RETURN
101 END
← DMATX [either De or Dep ]. Tangent operator (stored in matrix form) consistent with
the implicit elastic predictor/return-mapping algorithm for the von Mises model. The
implicit algorithm has been described in Section 7.3.2. The associated elastoplastic
consistent tangent, Dep , was derived in Section 7.4.2 above and is given by expres-
sion (7.120).
→ EPFLAG. Elastoplastic tangent logical flag. If .TRUE., DMATX returns as the elastoplastic
consistent tangent operator, Dep . If .FALSE., DMATX returns as the elastic matrix, De .
The value of EPFLAG is set in subroutine MATICT. The procedure for setting EPFLAG in
MATICT is common to all elastoplastic models of the material class HYPEPL.
→ NTYPE. Stress state type flag. NTYPE=2 for plane strain and NTYPE=3 for axisymmetric
problems.
→ RSTAVA [εe , ε̄p ]. Array of updated state variables other than the stress components
(output of SUVM) (see page 227 for complete description).
→ STRES [σn+1 ]. Array containing the updated stress tensor components (output of
SUVM).
• FOID [IS ]. Fourth-order symmetric identity tensor stored in array form according to the
convention shown in (D.16) (page 762).
The names of most local variables and arrays of CTVM follow closely the notation of
Section 7.4.2, where the elastoplastic consistent tangent operator has been derived.
The basic unknowns of the above system of algebraic equations are: the updated elastic strain,
εen+1 ; the updated set of internal variables, αn+1 ; and the incremental plastic multiplier, ∆γ.
The corresponding updated stress tensor delivered by the above return mapping is evaluated
from the standard potential form
∂ψ
σn+1 = ρ̄ e .
∂ε n+1
trial
Clearly, by changing the elastic trial strain, εen+1 , that takes part in the return map
equations, the solution {εn+1 , αn+1 , ∆γ} will change accordingly and so will the updated
e
stress σn+1 . In this way, the algebraic system of equations define an implicit function for
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 239
the stress tensor at tn+1 . As discussed in the preceding sections, the elastoplastic consistent
tangent operator
∂σn+1
Dep ≡ e trial ,
∂εn+1
i.e. the derivative of the implicit function for stress, gives the linear tangent relationship
trial
between εen+1 and σn+1 . Thus, the first step in its derivation is to linearise the return-
mapping equations (having the elastic trial strain – the system input – also as a variable).
Straightforward differentiation of the general return-mapping equations yields the following
linearised form:
∂N ∂N
dεe trial
dε + ∆γ ∂σ : dσ + ∆γ ∂A ∗ dA + d∆γ N
e
∂H ∂H
dα − ∆γ ∗ dσ − ∆γ ∗ dA − d∆γ H = 0 , (7.127)
∂σ ∂A
∂Φ ∂Φ
: dσ + ∗ dA 0
∂σ ∂A
dσ = De : dεe + E ∗ dα
(7.128)
dA = F ∗ dεe + G ∗ dα.
The linear operators E, F and G are defined in (7.48)–(7.50). Inversion of this relation gives
the following general expression
dεe = C : dσ + B ∗ dA
(7.129)
dα = A ∗ dσ + J ∗ dA,
where C, B, A and J are suitably defined linear operators. Note that, under the classical
assumption of decoupling between elasticity and plastic hardening:
the tangent moduli E and F vanish and so do B and A. In this case we end up with the following
relation:
dεe = C : dσ ;
(7.131)
dα = J ∗ dA,
From the above, we obtain the generalised tangent operators consistent with the implicit
return-mapping algorithm. The generalised operators are: The elastoplastic consistent tan-
gent modulus,
dσn+1
D11 = e trial ≡ Dep . (7.134)
dεn+1
The operator
dAn+1
D21 = trial
, (7.135)
dεen+1
which gives the tangent relation between increments of elastic trial strain and thermodynam-
ical force set An+1 ; and
d∆γ
D31 = e trial , (7.136)
dεn+1
which is the tangent operator (a second-order tensor in this case) relating increments of elastic
trial strain and incremental plastic multiplier. Note that the dimension and order of the linear
operator D21 depends on the definition of set A. As far as finite element computations are
concerned, only the elastoplastic consistent tangent operator is of relevance.
tensors, etc.) of appropriate order. The product operations between the elements of the matrix and the elements of
the ‘vector’ containing the differentials dσ, etc. are the corresponding appropriate products. This symbolic matrix
notation will be used frequently in this book to represent linearised systems.
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 241
As discussed in Chapter 6, these sets may contain as many variables as necessary for the
appropriate description of the material behaviour. Depending on the particular model, each
element of these sets may be a scalar, vector or tensor of any order. With such a high degree
of generality, it may not be easy to have a clear picture of what operations are involved in
the above derivation. In order to provide the reader with a better view of the derivation of
the elastoplastic consistent tangent operator, the general procedure described above is here
particularised for the implicit return mapping for the von Mises model with isotropic strain
hardening.
Firstly, we recall the discussion around expressions (6.187–6.194) (page 182) for the
von Mises isotropically hardening model. We then have
∂H ∂H ∂H
≡ = 0; = 0. (7.137)
∂A ∂κ ∂σ
In addition, as elasticity and plastic hardening are decoupled in the von Mises model, the
operators B and A vanish, as pointed out in the discussion following expression (7.129). The
remaining (non-vanishing) operators are
where by taking (6.136) into account, the flow vector derivative is easily obtained as
∂N ∂ 2 Φ 3
= 2
= (Id − N̄ ⊗ N̄), (7.142)
∂σ ∂σ 2q
where N̄ is the unit flow vector (7.117).
The tangential relation (7.141) can be inverted trivially as follows. We start by observing
that the first row of (7.141) is equivalent to
where −1
∂N
P≡ IS + ∆γ De : : De . (7.144)
∂σ
The second row of (7.141), in turn, gives
dσy = H d∆γ. (7.145)
The above expression for Dep is an equivalent representation to (7.120) and could be
used as an alternative in the computational implementation of the model. Also note the
similarity between (7.148) and the continuum tangent operator (7.122). As expected (refer
to Remark 7.9, page 235), the above formula recovers (7.122) if ∆γ = 0.
Alternative formula
Yet another representation, alternative to (7.148), is sometimes used in computational
applications. The alternative formula is obtained simply by replacing the first right-hand side
of (7.146) into the first row of (7.141) and then inverting the resulting differential relation.
The corresponding expression for the elastoplastic consistent tangent reads
−1
∂N 1
Dep = [De ]−1 + ∆γ + N⊗N , (7.149)
∂σ H
or, in view of (7.142), and making use of the unit flow vector definition,
−1
3∆γ 3 1 ∆γ
Dep = [De ]−1 + Id + − N̄ ⊗ N̄ . (7.150)
2q 2 H q
The use of an analogous formula is reported, for instance, by Dutko et al. (1993) in the context
of anisotropic plasticity.
Remark 7.10. The presence of the hardening modulus in the denominator in the above
formula does not allow its use in practical computations if H = 0 (perfect plasticity).
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 243
Incremental potentials
The symmetry of the consistent tangent operator implies that at any state {εn , εpn , αn } there
exists an incremental potential,
trial
ψn = ψ̃n (εn+1 ) = ψ̄n (εen+1 ) (7.151)
analogous to the free-energy function, ψ, such that the updated stress, σn+1 , delivered by the
integration algorithm is given by
∂ ψ̃n ∂ ψ̄n
σn+1 = ρ̄ = ρ̄ e trial . (7.152)
∂εn+1 ∂εn+1
Remark 7.11. For elastoplastic model/algorithm combinations that preserve the consistent
tangent operator symmetry, the incremental equilibrium boundary value problem is endowed
with a potential structure analogous to that of hyperelasticity boundary value problems. This
allows methods conventionally used in hyperelasticity to be adopted in the study of elasto-
plastic incremental boundary value problems. Also, many important properties associated
with the regularity and stability of the solution of (time-continuum) elastoplasticity problems
244 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
depend crucially on the symmetry of the continuum tangent operator. When symmetry is
preserved in the consistent tangent operator, many such properties can be transferred to the
discretised problem. These issues fall outside the scope of this book. Interested readers are
referred to Martin (1975) and Rice (1976) for further details.
1 e e trial trial
ρ̄ψn = (D : εn+1 ) : εen+1
2
6G2 1 e trial 2 σy
− Ĥ(Φtrial
n+1 ) ε − εe trial
. (7.154)
3G + H 2 d n+1 2G 3/2
d n+1
That the incremental constitutive function (7.103) is indeed obtained from the potential
relation (7.152) with the above potential can be established by a straightforward application
of tensor differentiation rules taking expression (2.139) (page 36), into account together with
the identity εed trial e trial
n+1 = Id : εn+1 .
b = 200mm
y
P
30 o
x
P
a = 100 mm
Figure 7.14. Internally pressurised cylinder. Geometry, material properties and finite element mesh.
is carried out assuming plane strain conditions. Due to symmetry, only a 30o segment of
the whole cylinder cross-section is discretised with the appropriate symmetry displacement
constraints imposed on the edge nodes. The pressure, P , prescribed on the inner surface,
is increased gradually until a collapse (limit) load is reached. For the present problem (see
Figure 7.15), plastic yielding starts at the inner surface (with radial coordinate r = a) and
develops gradually, in the form of a cylindrical plastic front (with radius c), toward the outer
face of the cylinder (r = b). Collapse occurs when the plastic front reaches the outer face
(c = b) and the entire cylinder becomes plastified. At the limit load, the cylinder can expand
indefinitely without further increase in the applied pressure. A closed-form solution to this
problem has been derived by Hill (1950). It relates the applied pressure to the radius c of the
plastic front by means of the expression
P c 1 c2
= ln + 1− 2 , (7.155)
Y a 2 b
√
where, for the von Mises model, Y = σy / 3. Plastic yielding begins when c = a, which
corresponds to the yielding pressure
Y a2
P0 = 1− 2 . (7.156)
2 b
Before plastic yielding starts (P < P0 ), the radial displacement, ub , of the outer surface is a
linear function of P , given by
2P b
ub = (1 − ν 2 ), P < P0 . (7.157)
E(b2 /a2 − 1)
246 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
b
P c
a
elastic region
0.20
internal pressure, P (GPa)
0.15
0.10
closed solution (Hill, 1950)
present results
0.05
0.00
0.0 0.2 0.4 0.6
radial displacement at outer face (mm)
Y c2
ub = (1 − ν 2 ), P ≥ P0 , (7.158)
Eb
where c can be evaluated as an implicit function of P through (7.155). A diagram showing
the applied pressure versus the radial displacement at the outer face of the cylinder obtained
in the finite element simulation is plotted in Figure 7.16 together with Hill’s closed-form
solution. The high accuracy of the finite element results is clear. The limit load is reached
when c = b. It then follows from (7.155) that the limit pressure is
2σy
Plim = √ ln(b/a), (7.159)
3
which for the present dimensions and material parameters gives
In the finite element solution, the limit load is assumed to have been reached when equilib-
rium can no longer be obtained (global iterations do not converge) with a reasonably small
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 247
0.26
0
0.22 -0.02
P=0.18 GPa P=0.1 GPa
0.18 -0.06
P=0.18 GPa
0.14 -0.10
P=0.1 GPa
0.10 -0.14
analytical solution (Hill, 1950)
present results
0.06 -0.18
100 120 140 160 180 200 100 120 140 160 180 200
radial coordinate (mm) radial coordinate (mm)
Figure 7.17. Internally pressurised cylinder. Hoop and radial stress distributions at different levels of
applied internal pressure. Finite element results are computed at Gauss integration points.
load increment. The limit pressure obtained in the present simulation is about 0.19209 GPa,
which is virtually identical to Hill’s closed-form solution. The hoop and radial stresses
obtained at the Gauss points are plotted in Figure 7.17 together with Hill’s solution for
P = 0.1 GPa and P = 0.18 GPa. In the plastic region, where the radial coordinate r satisfies
a ≤ r ≤ c, the radial and hoop stresses obtained by Hill are given, respectively, by
1 c c2 1 c c2
σr = Y − − ln + 2 , σθ = Y − ln + 2 . (7.161)
2 r 2b 2 r 2b
In the elastic region, c ≤ r ≤ b, the stresses are
Y c2 b2 Y c2 b2
σr = − 2 − 1 , σθ = + 1 . (7.162)
2b r2 2b2 r2
The finite element results at the Gauss points are also in very close agreement with Hill’s
solution. It is worth mentioning that at the pressure level P = 0.1 GPa the entire cylinder is
still elastic. At the level P = 0.18 GPa the plastic front has already progressed considerably
and c ≈ 159.79. The transition between the elastic and plastic zones is clearly marked by the
drastic change in the slope of the curve for the hoop stress shown in Figure 7.17.
where the radius of the plastic front, c, is associated with the applied pressure, P , through the
expression
c 2σy c3
P = 2σy ln + 1− 3 , (7.164)
a 3 b
and the yielding pressure, P0 , is
2σy a3
P0 = 1− 3 . (7.165)
3 b
When the whole shell is elastic, i.e. when P < P0 , the displacement ub is given by
3P b
ub = (1 − ν), P < P0 . (7.166)
2E(b3 /a3 − 1)
The displacement ub obtained in the finite element solution is plotted against the applied
pressure in Figure 7.18. The analytical curve is also plotted for comparison. The analytical
limit load, corresponding to c = b, is obtained as
Again, in the numerical solution, it is assumed that the limit load has been achieved when
convergence for sufficiently small load increments cannot be obtained. The limit pressure
obtained in the present analysis is approximately 0.33769 GPa, which is nearly identical to
the exact value. The hoop and radial stresses obtained at the Gauss points in the finite element
simulation are plotted in Figure 7.19 together with the corresponding analytical solution.
The agreement between numerical and exact results is excellent. The analytical expressions
for the hoop and radial stresses are
c 1 c3 1 c 1 c3
σr = −2Y ln + 1 − 3 , σθ = 2Y − ln − 1− 3 ,
r 3 2b 2 r 3 2b
(7.169)
in the plastic region (a ≤ r ≤ c). In the elastic region (c ≤ r ≤ b), the stresses are
2σy c3 b3 2σy c3 b3
σr = − − 1 , σθ = +1 . (7.170)
3b3 r3 3b3 2r3
Residual stresses
After complete unloading from a partly plastic state, the shell will be subjected to a field
of (self-equilibrated) residual stresses. The residual stress field produces an increase in the
yield pressure so that pre-loading can be applied to strengthen the shell. For a shell that has
been monotonically loaded to a pressure level P and then completely unloaded without the
occurrence of reverse plastic flow (i.e. the unloading process is purely elastic), the analytical
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 249
0.4
b = 200 mm
0.2
P
analytical solution (Hill, 1950)
0.1 present results
a = 100 mm 0.0
0.00 0.04 0.08 0.12 0.16
radial displacement at outer face (mm)
Figure 7.18. Internally pressurised spherical shell. Geometry and pressure versus displacement curve.
0.2
0.0
P=0.30 GPa P=0.30 GPa
radial stress (GPa)
hoop stress (GPa)
0.1
-0.1
P=0.15 GPa
P=0.15 GPa
-0.2
0.0
-0.1
100 120 140 160 180 200 100 120 140 160 180 200
radial coordinate (mm) radial coordinate (mm)
Figure 7.19. Internally pressurised spherical shell. Hoop and radial stress distributions at different levels
of applied internal pressure. Finite element results are computed at Gauss integration points.
distribution of residual hoop and radial stresses along the sphere radius are given by
3 3
3
2σ c P a a
− 3
y
σrR = − 3
− 3
3 a P0 r b
3 3 , for c ≤ r ≤ b, (7.171)
2σ c P a a 3
+ 3
y
σθR = −
3 a3 P0 2r3 b
and
2σ P a 3
r
σrR = −
y
1 − 3 − 3 ln
3 P0 r a
, for a ≤ r ≤ c. (7.172)
2σy 3 P a3 r
σθ = −
R
− 1 + 3 + 3 ln
3 2 P0 2r a
In the above expressions, c is the radius of the plastic region at the maximum pressure level
attained before unloading. The maximum pressure that can be applied without the occurrence
250 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
0.1
hoop stress
radial stress
-0.1
-0.3
100 120 140 160 180 200
radial coordinate (mm)
Figure 7.20. Internally pressurised spherical shell. Residual hoop and radial stress distributions
resulting from pre-loading to P = 0.28 GPa.
of reverse plastic flow during unloading is P = 2P0 , which for the present dimensions and
material constants is
P = 0.28 GPa. (7.173)
For thinner shells, with b/a < 1.701, the maximum pressure that can be applied in pre-loading
is P = Plim . Here, the finite element simulation of a pre-loading operation (with maximum
load P = 0.28 GPa) has also been carried out. The resulting residual stresses are shown in
Figure 7.20 where the above analytical solution is also plotted for comparison. Again, the
high accuracy of the finite element solution is evident.
300
Material properties – von Mises model
Young’s modulus: E = 10 7
P
100
h = 10 finite element results
Skrzypek’s limit load - 260.8
w (central deflection)
R = 10 0
0.0 0.4 0.8 1.2 1.6
central deflection, w
Figure 7.21. Uniformly loaded circular plate. Geometry and load versus deflection diagram.
0.0
normalised deflection, w/h
P=100
P=200
0.2
0.4 P=250
0.6
0 0.2 0.4 0.6 0.8 1
normalised radial coordinate, r/R
Figure 7.22. Uniformly loaded circular plate. Deflection profiles at different levels of load obtained in
the finite element analysis.
In the finite element analysis, starting from the unloaded state, the pressure P is increased
gradually, in ten increments, until collapse occurs. The deflection at the centre of the plate
obtained in the numerical simulation is plotted against the applied pressure in the diagram
of Figure 7.21. The accuracy of the finite element procedure in capturing the collapse is
evident. The limit load obtained in the finite element analysis (taken as the load above which
equilibrium iterations can no longer converge for a sufficiently small load increment) is
fe
Plim ≈ 259.8. (7.177)
The relative error is about 0.4%. Note that such a high accuracy has been obtained despite the
use of a relatively coarse mesh. Deflection profiles obtained at different stages of the loading
process are shown in Figure 7.22. In the present problem, plastic yielding starts at the top
and bottom surfaces of the plate around its centre and propagates toward the neutral plane
with increasing load until collapse occurs. The propagation of the plastic zones is illustrated
in Figure 7.23. The plastic and elastic regions, represented respectively by the shaded and
white areas, are shown at different stages of the loading process. At P = 100 the plate is still
purely elastic. At P = 200 plastic yield has already taken place at the bottom and top surfaces
252 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
P = 200
Please Wait..
P = 259.5
Please Wait..
Figure 7.23. Uniformly loaded circular plate. Evolution of the plastic region.
B =1m
P
Material properties
L=5m
- Soil: von Mises perfectly plastic
footing E = 107 kPa
ν = 0.48
σy = 848.7 kPa ( τ y = 490 kPa)
soil
around the plate centre and at P = 259.5 only a narrow layer surrounding the neutral plane
remains elastic and collapse is imminent. In order to obtain a more accurate definition of the
plastic zones, a finer mesh of 5 × 10 elements has been used. It is remarked that the limit
load obtained with the finer mesh is virtually identical to the one obtained with the previous
coarser mesh.
u
footing
Please Wait..
mesh refinement
5m
around footing edge
Please Wait..
5m
5
normalised pressure, P/c
0
0 0.001 0.002
normalised settlement, u/B
interface). This corresponds to prescribing the vertical displacement (the settlement), u, of the
nodes under the footing and allowing their horizontal displacement to be unconstrained. A
total displacement of u = 0.002 m is applied in 14 increments. The corresponding (average)
pressure P supported by the footing is computed as the total reaction on the footing divided
by the width B. Solutions to this problem, based on slip-line theory, have been derived by
Prandtl and Hill (Hill, 1950). They give the following limit pressure:
√
where the cohesion or shear strength, c, for the von Mises model is given by c = σy / 3. The
normalised average pressure obtained in the finite element simulation is plotted in Figure 7.26
versus the normalised settlement. The limit pressure obtained is in excellent agreement with
the slip-line solution, with a relative error of approximately 0.9%. It is remarked that the
accuracy can be increased by refining the finite element mesh. The evolution of the plastic
zone during the process of loading is shown in Figure 7.27. The plastic zones correspond
to the shaded area in the plot. At the 12th increment, the collapse load has already been
reached. The incremental nodal displacement field corresponding to increment 12 is shown
in Figure 7.28. The sizes of the nodal displacement vectors plotted in Figure 7.28 have been
greatly exaggerated and are not to the same scale as the underlying finite element mesh. At the
collapse load, the nodes located sufficiently far from the footing are virtually fixed and only
the nodes surrounding the footing move. Note that the mechanism depicted in Figure 7.28 is
in agreement with Hill’s slip-line solution (Hill, 1950) which, as pointed out by Hill, is valid
when the footing is assumed to be smooth (unconstrained horizontal displacements under
the footing). It is important to emphasise here that, except under undrained conditions, the
plastic flow in soils is strongly affected by the hydrostatic stress. Thus, pressure insensitive
plasticity models, such as the von Mises material adopted in the present example, give in
general a rather poor representation of the actual behaviour. The simulation of the collapse of
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 255
footing
Please Wait..
Please Wait..
Figure 7.28. Strip footing. Incremental nodal displacement field at collapse (increment 12).
the strip footing using more appropriate pressure-sensitive plasticity models is carried out in
Chapter 8.
Please Wait..
w=10
u
Figure 7.29. Double-notched tensile specimen. Geometry, material properties and finite element mesh.
(a) (b)
normalised net stress, σ̄/σy
3 3
2 2
0 0
0 4 8 12 16 0 4 8 12 16
Figure 7.30. Double-notched specimen. Load-deflection curves: (a) perfect plasticity; and (b) linear
hardening.
In Figure 7.30(a), the normalised axial net stress, σ̄/σy , obtained in the perfectly plastic finite
element analysis is plotted against the normalised deflection of the top edge, 2uE/σy w. The
limit load obtained with the present mesh is only approximately 0.8% higher than the Prandtl
solution. The load-deflection curve obtained for the linearly hardening model is shown in
Figure 7.30(b). In this case, no limit load exists.
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 257
7.6. Further application: the von Mises model with nonlinear mixed
hardening
Having used the simple set-up of the von Mises model with isotropic hardening, we hope to
have provided the reader with a clear picture of the basic components of the finite element
implementation of plasticity models. In this section we move one step forward and apply
the same concepts to derive an implicit integration algorithm (together with its associated
consistent tangent operator) for a version of the von Mises model that combines general
nonlinear isotropic and kinematic hardening. The mixed hardening rule considered here is
the one described in Section 6.6.5 (page 189).
It is worth remarking that consideration of kinematic hardening may become crucial in
applications involving cyclic loads – situations where the Bauschinger effect may not be
disregarded without significant loss of accuracy. The algorithm/tangent operator derived
here are also incorporated into HYPLAS. As we shall see, the structure of these routines
is very similar to those of the isotropic hardening-only counterparts. The computational
implementation of these procedures is a straightforward extension of routines SUVM (for the
integration algorithm) and CTVM (for the consistent tangent). The corresponding routines for
the mixed hardening model are named, respectively, SUVMMX and CTVMX. Their source code
is not included in the text.
Readers wishing to skip the details of the present derivations are referred directly to
Box 7.5 (page 260) for the integration algorithm and formula (7.213) for the corresponding
elastoplastic consistent tangent modulus.
where β is the backstress tensor which defines the translation of the centre of the
(kinematically hardened) von Mises circle in the deviatoric plane, η is the relative
stress
η ≡ s − β, (7.182)
and σy defines now only the radius of the yield surface and not necessarily the uniaxial
yield stress as in the isotropic hardening-only model.
2. Associative law for the plastic flow:
%
p ∂Φ 3 η
ε̇ = γ̇ = γ̇ . (7.183)
∂σ 2 η
4. Nonlinear kinematic hardening defined by the following evolution law for the back-
stress: %
2 k p p 2 k p η
β̇ = H (ε̄ ) ε̇ = γ̇ H (ε̄ ) , (7.185)
3 3 η
with
dβ̄
H k (ε̄p ) ≡ (7.186)
dε̄p
denoting the slope of the given kinematic hardening curve:
β̄ = β̄(ε̄p ). (7.187)
Recall that the kinematic hardening stress, β̄, is the kinematic contribution to overall
hardening that can be obtained from uniaxial tests with load reversal.
n+1 ≡ sn+1 − βn .
ηtrial trial
(7.191)
which, substituted into (7.190), renders the simpler update formula for the relative stress
∆γ k p
ηn+1 = 1 − trial [3G + H (ε̄n+1 )] ηtrial
n+1 , (7.193)
q̄n+1
trial
where q̄n+1 is the elastic trial relative effective stress, defined by
trial
q̄n+1 ≡ 3
2 ηtrial
n+1 . (7.194)
Then, by substituting (7.193), together with (7.188)3, into (7.188)4, the return mapping
reduces to the following generally nonlinear scalar equation for ∆γ:
Φ̃(∆γ) ≡ q̄n+1
trial
− ∆γ[3G + H k (ε̄pn + ∆γ)] − σy (ε̄pn + ∆γ) = 0. (7.195)
Note that, also consistently with the backward Euler approximation, we have
where β̄k ≡ β̄(ε̄pk ). Since the curve β̄(ε̄p ) (in the form of discrete sampling points) rather than
the slope H k (ε̄p ) is what is normally available from uniaxial experiments, it is convenient to
use the above approximation and work only with the kinematic hardening stress in the return-
mapping equations. With the adoption of this approach, we replace (7.188)2 and (7.193),
respectively, with
%
2 η
βn+1 = βn + (β̄n+1 − β̄n ) n+1 (7.197)
3 ηn+1
and
3G∆γ + β̄n+1 − β̄n
ηn+1 = 1 − trial
ηtrial
n+1 . (7.198)
q̄n+1
Φ̃(∆γ) ≡ q̄n+1
trial
− 3G∆γ − β̄(ε̄pn + ∆γ) + β̄n − σy (ε̄pn + ∆γ) = 0. (7.199)
The mth Newton–Raphson iterative correction to ∆γ in the solution of the above equation
reads
Φ̃(∆γ (m−1) )
∆γ (m) := ∆γ (m−1) − , (7.200)
d
where
d = −3G − H k (ε̄pn + ∆γ (m−1) ) − H i (ε̄pn + ∆γ (m−1) ) (7.201)
Box 7.5. Implicit elastic predictor/return-mapping algorithm for the von Mises
model with mixed nonlinear hardening.
(i) Elastic predictor. Given ∆ε and the state variables at tn , evaluate the elastic trial state
εen+1
trial
:= εen + ∆ε
ε̄pn+1
trial
:= ε̄pn ; βtrial
n+1 := βn
ptrial e trial
n+1 := Kεv n+1 ; strial e trial
n+1 := 2G εd n+1
n+1 := sn+1 − βn ;
ηtrial ηtrial
n+1
trial trial 3
q̄n+1 := 2
e 1 1
σn+1 := sn+1 + pn+1 I; εn+1 = sn+1 + εev trial I
2G 3 n+1
(iv) EXIT
By means of straightforward tensor manipulations, we can easily establish that the incre-
mental constitutive function, analogous to (7.93), for the present stress updating procedure
reads
trial
σn+1 = σ̄n+1 (ε̄pn , βn , εen+1 )
∆γ 6G2 ∆γ 3G
≡ De − Ĥ(Φtrial ) trial
I d
trial
: εen+1 + Ĥ(Φtrial ) trial βn , (7.202)
q̄n+1 q̄n+1
FINITE ELEMENTS IN SMALL - STRAIN PLASTICITY PROBLEMS 261
trial
= q̄n+1 trial
(βn , εen+1 ) ≡ 32 2G Id : εen+1
trial
− βn , (7.203)
and
trial
∆γ = ∆γ(ε̄pn , βn , εen+1 ) (7.205)
is the implicit function of trial
εen+1 and the internal variables {ε̄pn , βn } defined by the
consistency equation (7.199).
with constant σy 0 , H i and H k , then the present return mapping equation has the following
closed form solution:
Φtrial
∆γ = . (7.207)
3G + H k + H i
β̄(ε̄p ),
is assumed to be piecewise linear, defined by the same number of (user supplied) sampling
pairs as the isotropic hardening curve
{ε̄p , β̄}.
The extra properties are read and stored in array RPROPS (real material properties) during the
data input phase of HYPLAS.
262 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
trial
∆γ 3G trial ∂ q̄n+1
+ η
trial )2 n+1
⊗ trial
, (7.209)
(q̄n+1 ∂εen+1
where we have made use of the connection ηtrial n+1 = 2G Id : εn+1 − βn . The derivative of the
e trial
trial relative effective stress is obtained analogously to (7.116). The corresponding expression
in the present case has the same format as (7.116), i.e.
trial
%
∂ q̄n+1 2
trial
= 2G N̄ n+1 , (7.210)
∂εen+1 3
Implementation
The above tangent operator is computed in subroutine CTVMMX (Consistent Tangent operator
for the Von Mises model with nonlinear MiXed hardening) of HYPLAS. As the reader may
verify CTVMMX ¯ is a straightforward variation of routine CTVM (refer to Section 7.4.3, from
page 235) coded for the isotropic hardening-only model. The basic difference is that, in
CTVMMX, the incremental flow vector and the scalar factors multiplying the fourth-order
tensors are redefined according to the above formulae.
8 COMPUTATIONS WITH OTHER
BASIC PLASTICITY MODELS
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
266 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
interest. These summarise the most important results and the tables should be particularly
helpful to readers who wish to skip the details of derivation and concentrate only on the
practical aspects of computational implementation. Also at the beginning of each main
section, a summary of the constitutive equations of the model concerned is presented. This
layout makes the main sections self-contained to a certain extent and has been chosen in order
to avoid readers having to refer back to Chapter 6, where a more detailed description of the
constitutive equations adopted here is given.
Applications of the numerical procedures derived in the main sections are illustrated in
Section 8.4 by means a comprehensive set of benchmarking numerical examples.
The model
All ingredients of the Tresca plasticity model implemented here have been fully described in
Chapter 6. In the following we summarise for convenience only its main equations.
The multisurface representation of the Tresca model is defined by means of the six yield
functions, Φ1 , . . . , Φ6 , given in (6.91) (page 160). The corresponding associative plastic flow
rule is defined by the rate equation
6
ε̇p = γ̇ i N i , (8.1)
i=1
where
∂Φi
Ni ≡ , (8.2)
∂σ
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 267
Φi ≤ 0, γ̇ i ≥ 0, Φi γ̇ i = 0 (8.3)
σy = σy (ε̄p ). (8.4)
N a ≡ N 1 = e1 ⊗ e1 − e3 ⊗ e3 . (8.6)
2. Plastic flow from the right corner, R. The rate of plastic strain in this case is a linear
combination with non-negative coefficients of the normals to the two intersecting
planes at this point. The flow rule reads
ε̇p = γ̇ a N a + γ̇ b N b , (8.7)
where N a is the normal to the main plane, defined by (8.6), and N b is the normal to
the plane on the right of the main plane defined by Φ6 = 0:
N b ≡ N 6 = e1 ⊗ e1 − e2 ⊗ e2 . (8.8)
3. Plastic flow from the left corner, L. The flow rule, in this case, is the same as the one
above, but with N b being the normal to the plane on the left of the main plane (defined
by Φ2 = 0):
N b ≡ N 2 = e2 ⊗ e2 − e3 ⊗ e3 . (8.9)
Under the assumption of associative hardening (for further details, refer to the text surround-
ing equation (6.196), page 183), the accumulated plastic strain for this model is defined by
means of its evolution equations, which are given respectively by
ε̄˙p = γ̇ (8.10)
for plastic flow from the main plane (item 1 above), and
ε̄˙p = γ̇ a + γ̇ b (8.11)
for plastic flow from the right and left corners (items 2 and 3, respectively).
268 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
(·)n+1 := (·)trial
n+1 . (8.12)
particularized for the Tresca model. To derive the simplified equations, consider the implicit
update formula for the elastic strain obtained as a result of the discretisation of (8.1):
6
trial
εen+1 = εen+1 − i
∆γ i Nn+1 . (8.13)
i=1
Due to the isotropic linear elastic law and the fact that the Tresca flow vector is purely
deviatoric, this expression is equivalent, as in the von Mises model, to independent update
formulae for the hydrostatic and the deviatoric stress:
pn+1 = ptrial
n+1
6
(8.14)
n+1 − 2G
i
sn+1 = strial ∆γ i Nn+1 .
i=1
i
Note that by definition of the flow vectors, Nn+1 and sn+1 share the same principal
directions. Thus, expression (8.14)2 implies that these principal directions are also shared
by strial
n+1 so that the update formula for the deviatoric stress can be equivalently expressed in
terms of principal stresses as†
6
sj = strial
j − 2G ∆γ i Nji , (j = 1, 2, 3), (8.15)
i=1
i
where Nji denotes the eigenvalues of Nn+1 .
However, for the von Mises model, the return-mapping algorithm (described in the previous section) turns out to
be more efficient when the standard formulation in terms of the deviatoric stress tensor components (rather than its
eigenvalues) is adopted.
270 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
The return mapping, in the present case, is obtained with the introduction of (8.16)
and (8.17) into the discrete consistency condition Φ1 (σn+1 , σy (ε̄pn+1 )) = 0. This
results in the following scalar (generally nonlinear) equation in ∆γ:
Φ̃(∆γ) ≡ strial
1 − strial
3 − 4G ∆γ − σy (ε̄pn + ∆γ) = 0. (8.18)
2. The updated stress lies on the right corner. Here, two plastic multipliers may be non-
zero and the incremental plastic strain is obtained from the discretisation of (8.7) as
a b
∆εp = ∆γ a Nn+1 + ∆γ b Nn+1 , (8.19)
where N a and N b denote, respectively, the normals to the main plane and to the
plane on its right side given by (8.6) and (8.8). This results in the following explicit
expression for the update of the principal deviatoric stresses:
s1 = strial
1 − 2G (∆γ a + ∆γ b )
s2 = strial
2 + 2G ∆γ b (8.20)
s3 = strial
3
a
+ 2G ∆γ .
The incremental law for ε̄p in this case is the discrete counterpart of (8.11):
This yields the following set of two algebraic equations for ∆γ a and ∆γ b :
3. The updated stress lies on the left corner. The situation now is completely analogous to
item 2 above with the difference that, here, N b is the normal to the plane on the left
of the main plane, given by (8.9). The deviatoric principal stress update formula in the
present case reads
s1 = strial
1 − 2G ∆γ a
s2 = strial
2 − 2G ∆γ b (8.25)
s3 = strial
3
a
+ 2G (∆γ + ∆γ ). b
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 271
The update formula for ε̄p is the same as for the return map to the right corner and the
final equations to be solved in the return-mapping algorithm are
In summary, it has been shown above that the essential return-mapping algorithm for the
Tresca model, formulated in principal deviatoric stresses, may have three different explicit
forms. The particular form to be employed depends on the position of the updated stress
on the yield surface. In any case, similarly to the von Mises return mapping, an equation
(or system of two equations for the corners) is solved firstly for the plastic multiplier ∆γ
(∆γ a and ∆γ b for the corners). With the plastic multiplier(s) at hand, the principal deviatoric
stresses, si , and the accumulated plastic strain, ε̄p , are updated by the formulae corresponding
to the particular case considered (main plane, right or left corner). The updated stress
tensor components, required in the finite element computations, are then obtained simply
by assembling
3
σn+1 := (si + pn+1 ) ei ⊗ ei . (8.28)
i=1
s1 − s3 − σy (ε̄pn+1 ) = 0. (8.29)
Upon application of the main plane return algorithm, the final stress can end up either inside
or outside the main sextant:
s1 ≥ s2 ≥ s3 , (8.30)
represented by the shaded areas in Figure 8.1. If the updated principal stresses satisfy the
above relation, then they are in the main sextant (Figure 8.1(a)), and, clearly, satisfy the
consistency condition, i.e. they lie on the updated yield surface, and the result of the return
mapping to the main plane is valid. If the updated stress falls outside the main sextant
272 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
S n+1
updated yield
a surface at t n+1 S n+1
N
yield surface
σ1 at t n main plane σ1
updated surface
σ2 σ3 σ2 σ3
yield surface
at t n
(a) (b)
Figure 8.1. The implicit algorithm for the Tresca model in principal stresses: (a) valid return to main
plane; and (b) invalid return to main plane – converged stress outside the original sextant. (Reproduced
with permission from A new computational model for Tresca plasticity at finite strains with an optimal
parametrization in the principal space, D Perić and EA de Souza Neto, Computer Methods in Applied
Mechanics and Engineering, Vol 171 c 1999 Elsevier Science S.A.)
trial
S n+1
trial
T: trial
S n+1 > 0 ⇒ right T: S n+1 < 0 ⇒ left
−2
G∆
γ
trial
a
S n+1
N
a
−2G∆γ b N b Na
S n+1 Nb
T
Na
T
S n+1
σ1 σ1
b
N
yield surface σ2 σ3 σ2 σ3
at t n
Figure 8.2. The implicit algorithm for the Tresca model in principal stresses. Selection of the
appropriate return mapping to corner. (Reproduced with permission from A new computational model
for Tresca plasticity at finite strains with an optimal parametrization in the principal space, D Perić
and EA de Souza Neto, Computer Methods in Applied Mechanics and Engineering, Vol 171 c 1999
Elsevier Science S.A.)
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 273
and exit
YES
check validity
of one-vector return: YES
σ 1 ≥ σ2 ≥ σ3 ?
NO
Figure 8.3. Flowchart of the integration algorithm for the Tresca model in principal stresses. Procedure
implemented in subroutine SUTR of program HYPLAS.
274 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
(Figure 8.1(b)), i.e. if relation (8.30) is not satisfied, then, in spite of being on the main
plane, the updated stress lies outside the updated elastic domain and, therefore, violates the
consistency condition. In this case, the result of the return to the main plane is obviously
not valid and the correct algorithm to be applied must be either the return to the left or to
right corner, described, respectively, in items 2 and 3 above. The appropriate algorithm to be
applied (right or left) can be easily determined by, again, considering the geometric properties
of the Tresca yield surface: if strial
n+1 lies on the right side of the line that passes through the
origin of stresses and is orthogonal to the main plane (the pure shear line), represented by the
dotted line in Figure 8.2, then it can be easily visualized that strial
n+1 could only return to the
left corner if ∆γ b were negative, which is unacceptable. In this case, the only possible return
is to the right corner. Using the same argument, one concludes that if the trial deviatoric stress
is on the left side of the dotted line, then the appropriate return is to the left corner. It is very
simple to determine, in actual computations, whether strial n+1 lies on the right or left of the
dotted line. Let T be a tangent vector to the main plane as illustrated in Figure 8.2. Clearly, T
has the same eigenvectors as strial
n+1 and its eigenvalues may be chosen as
T1 = 1, T2 = −2, T3 = 1. (8.31)
3
T: strial
n+1 = Ti strial
i = strial
1 + strial
3 − 2 strial
2 , (8.32)
i=1
Box 8.1. Implicit elastic predictor/return-mapping algorithm for the Tresca model.
(i) Elastic predictor. Given ∆ε and the state variables at tn , evaluate the elastic trial state
εen+1
trial
:= εen + ∆ε; ε̄pn+1
trial
:= ε̄pn
ptrial e trial
n+1 := K εv n+1 ; strial e trial
n+1 := 2G εd n+1
strial
1 ≥ strial
2 ≥ strial
3 and ei (i = 1, 2, 3)
Box 8.2. The Tresca model. One-vector return mapping to main plane.
Φ̃ := strial
1 − strial
3 − σy (ε̄pn )
s3 := strial
3 + 2G ∆γ
ε̄pn+1 := ε̄pn + ∆γ
and RETURN to Box 8.1
∆γ a := 0 ∆γ b := 0
where
strial
1 − strial
2 , for right corner
a
s̄ = strial
1 − strial
3 , s̄ =b
strial
2 − strial
3 , for left corner
∆γ := ∆γ a + ∆γ b
ε̄pn+1 := ε̄pn + ∆γ (update ε̄p )
dσy
H := p (hardening slope)
dε̄ ε̄p
n+1
residual derivative:
dΦ̃a dΦ̃a
d∆γ a d∆γ b −4G − H −2G − H
d := dΦ̃b
=
b
dΦ̃ −2G − H −4G − H
d∆γ a d∆γ b
new guess for ∆γ a and ∆γ b :
a a a
∆γ ∆γ Φ̃
−1
b
:= b
−d
∆γ ∆γ Φ̃b
continued on page 278
278 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Box 8.3 (contd. from page 277). The Tresca model. Two-vector return mappings
to corners (implemented in subroutine SUTR).
s1 := strial
1 − 2G∆γ a
s2 := strial − 2G∆γ b for left corner
2
s3 := strial
3 + 2G(∆γ a + ∆γ b )
1 SUBROUTINE SUTR
2 1( DGAM ,IPROPS ,LALGVA ,NTYPE ,RPROPS ,
3 2 RSTAVA ,STRAT ,STRES )
4 IMPLICIT DOUBLE PRECISION (A-H,O-Z)
5 PARAMETER(IPHARD=4 ,MSTRE=4)
6 C Arguments
7 LOGICAL
8 1 LALGVA(4)
9 DIMENSION
10 1 DGAM(2) ,IPROPS(*) ,RPROPS(*) ,
11 2 RSTAVA(MSTRE+1) ,STRAT(MSTRE) ,STRES(MSTRE)
12 C Local arrays and variables
13 LOGICAL
14 1 DUMMY, IFPLAS, RIGHT, SUFAIL, TWOVEC
15 DIMENSION
16 1 EIGPRJ(MSTRE,2) ,PSTRS(3) ,STREST(3)
17
18 DATA
19 1 R0 ,R1 ,R2 ,R3 ,R4 ,SMALL ,TOL /
20 2 0.0D0,1.0D0,2.0D0,3.0D0,4.0D0,1.D-10,1.D-10/
21 DATA MXITER / 50 /
22 C***********************************************************************
23 C STRESS UPDATE PROCEDURE FOR TRESCA TYPE ELASTO-PLASTIC MATERIAL WITH
24 C PIECE-WISE LINEAR ISOTROPIC HARDENING:
25 C IMPLICIT ELASTIC PREDICTOR/RETURN MAPPING ALGORITHM (Boxes 8.1-3).
26 C PLANE STRAIN AND AXISYMMETRIC IMPLEMENTATIONS.
27 C***********************************************************************
28 C Stops program if neither plane strain nor axisymmetric state
29 IF(NTYPE.NE.2.AND.NTYPE.NE.3)CALL ERRPRT(’EI0029’)
30 C Initialize some algorithmic and internal variables
31 DGAMA=R0
32 DGAMB=R0
33 IFPLAS=.FALSE.
34 SUFAIL=.FALSE.
35 EPBARN=RSTAVA(MSTRE+1)
36 EPBAR=EPBARN
37 C Set some material properties
38 YOUNG=RPROPS(2)
39 POISS=RPROPS(3)
40 NHARD=IPROPS(3)
41 C Set some constants
42 GMODU=YOUNG/(R2*(R1+POISS))
43 BULK=YOUNG/(R3*(R1-R2*POISS))
44 R2G=R2*GMODU
45 R4G=R4*GMODU
46 R1D3=R1/R3
47 C Compute elastic trial state
48 C ---------------------------
49 C Volumetric strain and pressure stress
50 EEV=STRAT(1)+STRAT(2)+STRAT(4)
51 P=BULK*EEV
280 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
178 PSTRS2=PSTRS2+R2G*DGAMB
179 ELSE
180 PSTRS1=PSTRS1-R2G*DGAMA
181 PSTRS3=PSTRS3+R2G*(DGAMA+DGAMB)
182 PSTRS2=PSTRS2-R2G*DGAMB
183 ENDIF
184 GOTO 50
185 ENDIF
186 40 CONTINUE
187 C failure of stress update procedure
188 SUFAIL=.TRUE.
189 CALL ERRPRT(’WE0001’)
190 GOTO 999
191 50 CONTINUE
192 C update stress components
193 C ------------------------
194 PSTRS(II)=PSTRS1
195 PSTRS(JJ)=PSTRS3
196 PSTRS(MM)=PSTRS2
197 STRES(1)=PSTRS(1)*EIGPRJ(1,1)+PSTRS(2)*EIGPRJ(1,2)+P
198 STRES(2)=PSTRS(1)*EIGPRJ(2,1)+PSTRS(2)*EIGPRJ(2,2)+P
199 STRES(3)=PSTRS(1)*EIGPRJ(3,1)+PSTRS(2)*EIGPRJ(3,2)
200 STRES(4)=PSTRS(3)+P
201 C and elastic engineering strain
202 RSTAVA(1)=(STRES(1)-P)/R2G+EEVD3
203 RSTAVA(2)=(STRES(2)-P)/R2G+EEVD3
204 RSTAVA(3)=STRES(3)/GMODU
205 RSTAVA(4)=PSTRS(3)/R2G+EEVD3
206 ELSE
207 C Elastic step: update stress using linear elastic law
208 C ====================================================
209 STRES(1)=STREST(1)+P
210 STRES(2)=STREST(2)+P
211 STRES(3)=STREST(3)
212 STRES(4)=PSTRS(3)+P
213 C elastic engineering strain
214 RSTAVA(1)=STRAT(1)
215 RSTAVA(2)=STRAT(2)
216 RSTAVA(3)=STRAT(3)
217 RSTAVA(4)=STRAT(4)
218 ENDIF
219 999 CONTINUE
220 C Update algorithmic variables before exit
221 C ========================================
222 DGAM(1)=DGAMA
223 DGAM(2)=DGAMB
224 LALGVA(1)=IFPLAS
225 LALGVA(2)=SUFAIL
226 LALGVA(3)=TWOVEC
227 LALGVA(4)=RIGHT
228 RETURN
229 END
The arguments of this subroutine are identical to those of SUVM for the von Mises model
implementation (see list starting on page 227), except for the following:
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 283
← DGAM [∆γ a , ∆γ b ]. This argument now is an array that stores up to two incremental
plastic multipliers. If the increment is elastic, both multipliers are set to 0. Otherwise,
the multiplier(s) are obtained as the solution of one of the three possible return mapping
equation sets (main plane, right or left corner).
← LALGVA. In addition to the plastic yielding flag IFPLAS and the state update failure
flag SUFAIL used also in SUVM, the array LALGVA of logical algorithnmic flags now
stores the two-vector return flag TWOVEC and the right corner return flag RIGHT. The
two-vector return flag is set to .TRUE. if one of the two possible two-vector return
mappings (right or left corner) is used. It is set to .FALSE. otherwise. The right corner
return flag is set to .TRUE. if the possible two-vector return is to the right corner to
.FALSE. if to the left corner. The flags TWOVEC and RIGHT are required by subroutine
CTTR to decide which elastoplastic tangent operator to compute (consistent with main
plane, right or left corner return mapping).
Also, IPROPS(3) and array RPROPS are, for the present material model implementation, set
in subroutine RDTR during the data input phase of HYPLAS.
• STREST [σtrial
n+1 ]. Array of components of the trial stress tensor.
• TOL [tol ]. Convergence tolerance for the Newton–Raphson algorithm used to solve the
return-mapping equations.
• ERRPRT. Called to send warning message to results file and standard output in
case of failure of the return-mapping algorithm. Corresponding message is in file
ERROR.RUN.
∆σT
∆σN N
T
B
N T σ1
A
r
σ3 σ2
Figure 8.4. Iso-error maps for the implicit algorithm for the Tresca model. Increment directions.
symmetry of the yield surface. For the Tresca model, however, the starting point has to be
taken into consideration. Here, two starting points, A and B, lying on the deviatoric plane are
considered. These points are shown in Figure 8.4. Point A is on the singularity of the Tresca
hexagon and point B corresponds to a state of pure shear. The increments ∆σtrial are of the
form
∆σN ∆σT
∆σtrial = N+ T, (8.33)
r r
where N and T are unit (in Euclidean norm) tensors defining the increment directions and
the scaling factor,
r≡ 2
3 σy , (8.34)
is the size of the sides of the Tresca hexagon (same as the radius of the von Mises cylinder
matching the Tresca prism on the edges). Note that, at the singularity (point A), the tensors
N and T are not orthogonal to each other. Two cases are considered in the iso-error maps
constructed in this section:
• perfect plasticity;
In both cases, the error plotted is the (Euclidean) norm of the relative difference between
the numerical and ‘exact’ updated stresses. With isotropic hardening, the hardening curve
adopted is shown in Figure 8.5. A total of 16 sampling points have been used to define the
curve.
The error maps obtained for the perfectly plastic case are shown in Figure 8.6. Note that for
both starting points, there is a large area within which the integration error vanishes. This is
obviously a very desirable feature. However, within a narrow band of increments the relative
error can be as high as 40% for the increment range considered in the present assessment.
The reason for such relatively high errors can be easily explained by graphically performing
the projection of σtrial onto the (fixed) yield surface and comparing the resulting stress with
the (exact) one obtained when the increment ∆σtrial is divided into two substeps of suitably
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 285
0.8
0.4
0.2
0
0 0.04 0.08 0.12 0.16
accumulated plastic strain, ε̄p
Figure 8.5. Iso-error maps for the Tresca model. Isotropic hardening curve.
6 6
0% 4 0% 4
∆σN /r
∆σN /r
2 2
40%
20%
0% 20% 0%
0% 0%
5% 5%
0 0
0 2 4 6 0 2 4 6
∆σT /r ∆σT /r
(a) (b)
Figure 8.6. Iso-error maps. Perfectly plastic Tresca model: (a) point A; and (b) point B.
chosen sizes. In this case (perfect plasticity), the integration error is reduced to zero for any
increment if the substepping procedure is included in the return-mapping algorithm. The
error maps obtained with isotropic hardening are shown in Figure 8.7. Again, the areas in
which the error is reasonably small are quite large, but errors are high within a narrow band
of increments. A comparison of the maps obtained with and without hardening suggests that,
for practical purposes, the accuracy expected from the implicit algorithm for the Tresca model
is not influenced by hardening.
286 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
6 6
4 4
0% 0%
∆σN /r
∆σN /r
2 20% 2
40%
20%
5%
0% 5% 0%
0 0
0 2 4 6 0 2 4 6
∆σT /r ∆σT /r
(a) (b)
Figure 8.7. Iso-error maps. Tresca model with hardening: (a) point A; and (b) point B.
where σn+1 is the outcome of the implicit function defined by the corresponding return-
mapping equations. For the return mapping for the Tresca model presented in Section 8.1.1,
the implicit constitutive function for the stress tensor is defined by the procedures of
items (ii), (iv) and (v) of Box 8.1. It is important to note that, for the Tresca model, there
are three possible sets of equations to be solved:
(i) return to main plane;
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 287
of the principal deviatoric stresses with respect to the principal deviatoric elastic trial strains.
When the return to the main plane is applied, the principal deviatoric stresses are updated
according to expressions (8.16–8.18) (item (iii) of Box 8.2). By using the elastic deviatoric
relation, strial
i = 2G εed trial
i , and differentiating the principal stress update formula (8.16), we
obtain
ds1 = 2G(dεed trial
1 − d∆γ)
With the differentiation of the return-mapping equation (8.18) the linearised form of the
consistency condition is obtained as
where H ≡ dσy /dε̄p is the slope of the hardening curve. From the above expression, we
obtain
2G
d∆γ = (dεed trial
1 − dεed trial
3 ). (8.40)
4G + H
With the substitution of (8.40) into (8.38) it follows that the partial derivatives ∂si /∂εed trial
j ,
arranged in matrix format, are given by
2G(1 − f ) 0 2Gf
∂si
=
0 2G 0
(8.41)
∂εed trial
j
2Gf 0 2G(1 − f )
where the matrix d, also used in the Newton–Raphson algorithm for solution of the return-
mapping equation in Box 8.3, is defined as
daa dab
d=
,
(8.49)
dba dbb
290 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
with
daa = −4G − H,
dab = −2G − H,
(8.50)
dba = −2G − H,
dbb = −4G − H.
which substituted into (8.47) leads, after a straightforward manipulation, to the expressions
for the derivatives of the principal deviatoric stresses consistent with the return mapping to
the right corner. Arranged in matrix format, the resulting derivatives are
2
2 2
2G 1 − 8G 4G 4G
(dab − daa ) (dba − dbb )
det d det d det d
∂si 8G 3
2G d 4G 2
= 2G 1 +
aa
− d . (8.52)
e trial
∂εd j det d det d det d
ba
8G 3
4G 2
2G dbb
− dab 2G 1 +
det d det d det d
With the above deviatoric principal stresses derivatives at hand, the principal stress deriva-
tives, ∂σi /∂εej trial , are obtained by applying formula (8.46).
where, again, the derivatives ∂σi /∂εej trial are obtained by applying (8.46).
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 291
1. Firstly, the principal stresses derivatives consistent with the appropriate return mapping
(main plane, right or left corner) are evaluated according to the expressions derived in
the above.
2. The tangent operator, Dep , (in matrix form) is then assembled in subroutine DGISO2
(Derivative of General ISOtropic tensor functions of a single tensor in 2-D and
axisymmetric conditions). This routine is called by CTTR which passes the principal
stress derivatives (as well as other necessary variables) as arguments.
1 SUBROUTINE CTTR
2 1( DMATX ,EPFLAG ,IPROPS ,LALGVA ,NTYPE ,
3 2 RPROPS ,RSTAVA ,STRAT ,STRES )
4 IMPLICIT DOUBLE PRECISION (A-H,O-Z)
5 PARAMETER(IPHARD=4 ,MDIM=3, MSTRE=4)
6 LOGICAL EPFLAG, LALGVA(4), OUTOFP, RIGHT, REPEAT, TWOVEC
7 DIMENSION
8 1 DMATX(MSTRE,MSTRE) ,IPROPS(*) ,RPROPS(*) ,
9 2 RSTAVA(MSTRE+1) ,STRAT(*) ,STRES(*)
10 DIMENSION
11 1 DPSTRS(MDIM,MDIM) ,DPSTRE(MDIM,MDIM) ,EIGPRJ(MSTRE,2) ,
12 2 FOID(MSTRE,MSTRE) ,PSTRS(MDIM) ,PSTRA(MDIM) ,
13 3 SOID(MSTRE) ,STRAC(MSTRE)
14 DATA
15 1 FOID(1,1),FOID(1,2),FOID(1,3),FOID(1,4)/
16 2 1.0D0 ,0.0D0 ,0.0D0 ,0.0D0 /
17 3 FOID(2,1),FOID(2,2),FOID(2,3),FOID(2,4)/
18 4 0.0D0 ,1.0D0 ,0.0D0 ,0.0D0 /
19 5 FOID(3,1),FOID(3,2),FOID(3,3),FOID(3,4)/
20 6 0.0D0 ,0.0D0 ,0.5D0 ,0.0D0 /
21 7 FOID(4,1),FOID(4,2),FOID(4,3),FOID(4,4)/
22 8 0.0D0 ,0.0D0 ,0.0D0 ,1.0D0 /
23 DATA
24 1 SOID(1) ,SOID(2) ,SOID(3) ,SOID(4) /
25 2 1.0D0 ,1.0D0 ,0.0D0 ,1.0D0 /
26 DATA
27 1 R0 ,RP5 ,R1 ,R2 ,R3 ,R4 /
28 2 0.0D0,0.5D0,1.0D0,2.0D0,3.0D0,4.0D0/
29 C***********************************************************************
30 C COMPUTATION OF CONSISTENT TANGENT MODULUS FOR TRESCA TYPE
31 C ELASTO-PLASTIC MATERIAL WITH PIECE-WISE LINEAR ISOTROPIC HARDENING.
32 C PLANE STRAIN AND AXISYMMETRIC IMPLEMENTATIONS.
33 C***********************************************************************
34 C Stops program if neither plane strain nor axisymmetric state
35 IF(NTYPE.NE.2.AND.NTYPE.NE.3)CALL ERRPRT(’EI0028’)
36 C Current accumulated plastic strain
37 EPBAR=RSTAVA(MSTRE+1)
292 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
159 C -------------------------
160 IF(NTYPE.EQ.2)THEN
161 NSTRE=3
162 ELSEIF(NTYPE.EQ.3)THEN
163 NSTRE=4
164 ENDIF
165 C
166 FACTOR=BULK-R2G*R1D3
167 DO 50 I=1,NSTRE
168 DO 40 J=I,NSTRE
169 DMATX(I,J)=R2G*FOID(I,J)+FACTOR*SOID(I)*SOID(J)
170 40 CONTINUE
171 50 CONTINUE
172 DO 70 J=1,NSTRE-1
173 DO 60 I=J+1,NSTRE
174 DMATX(I,J)=DMATX(J,I)
175 60 CONTINUE
176 70 CONTINUE
177 ENDIF
178 RETURN
179 END
← DMATX [either De or Dep ]. As in the von Mises implementation (refer to page 237).
→ EPFLAG. Elastoplastic tangent logical flag (refer to page 237 for a description).
The elements of the array LALGVA of logical algorithmic flags, set in subroutine SUTR, indicate
which elastoplastic tangent is to be computed here (consistent with main plane, right or left
return algorithm).
• EIGPRJ [ei ⊗ ei , i=1,2]. Matrix containing the components of the in-plane eigenpro-
jection tensors of the elastic trial strain.
• FOID [IS ]. Fourth-order identity tensor stored in array form according to (D.16),
page 762.
• STRAT [εei trial ]. Array containing the principal elastic trial strains.
The names of most local variables defined in CTTR resemble resemble the notation of
Section 8.1.4.
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 295
As far as yield surface singularities are concerned, the algorithm described here is entirely
based on the ideas that underly the numerical integration scheme for the Tresca model of the
previous section. Here, however, an extra return-mapping procedure – the return to the apex
of the Mohr–Coulomb yield surface – is required in the numerical integration scheme.
1. Plastic flow from the smooth (flat) portion of the main plane, where only one multiplier
may be non-zero. The generally non-associative flow vector in this case is reduced to
and
εp = γ̇N a . (8.56)
2. Plastic flow from the right edge. Here, only two plastic multipliers can be non-zero and
the flow rule is explicitly defined as
ε̇p = γ̇ a N a + γ̇ b N b , (8.57)
where N a is the same as in the above, N b is the flow vector of the plane on the right
of the main plane
3. Plastic flow from the left edge. The plastic flow is defined as above but with N b being
the flow vector of the plane on the left of the main plane
4. Plastic flow from the apex of the Mohr–Coulomb pyramid. In this case, up to six
multipliers may be non-zero, as the plastic rate tensor lies within the pyramid formed by
the six vectors of the Mohr–Coulomb model (refer to Figure 6.19.(b) which illustrates
the associative flow case). The plastic strain rate equation in this case is left in its
general format
6
ε̇p = γ̇ i N i . (8.60)
i=1
c = c(ε̄p ). (8.61)
Hardening associativity is assumed. The general equation for the evolution of ε̄p in this case
is given by (6.200) on page 184. For flow from the main plane as discussed above, only one
multiplier may be non-zero and (6.200) is reduced to
For plastic flow from an edge of the Mohr–Coulomb pyramid, the evolution of ε̄p reads
At the apex, the general equation (6.200) applies. In this case, the following general relation
cos φ p
ε̄˙p = ε̇ , (8.64)
sin ψ v
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 297
between the rate of accumulated plastic strain and the volumetric plastic strain rate, will be
crucial in the computational implementation of the model. This relation can be obtained by
combining the counterpart of (6.155) (page 175) under non-associative plastic flow:
6
ε̇pv = 2 sin ψ γ̇ i , (8.65)
i=1
and (6.200).
As for the algorithm for the Tresca model, the essential stress-updating is carried out here in
the principal stress space. Hence, after the computation of the elastic trial state, the spectral
decomposition of the trial stress, σtrial , is performed. With the principal trial stresses arranged
as σ1trial ≥ σ2trial ≥ σ3trial , the plastic consistency check proceeds as follows:
• If
Φtrial ≡ σ1trial − σ3trial + (σ1trial + σ3trial ) sin φ − 2 c(ε̄pn ) cos φ ≤ 0,
then the step is elastic and all state variables are updated as
(·)n+1 := (·)trial
n+1 .
Recall that the general return-mapping update formula for the stress tensor is given by
n+1 − D : ∆ε .
e
σn+1 = σtrial p
(8.66)
With the discretisation of the non-associated Mohr–Coulomb flow rule (8.60), this gives
6
n+1 − D :
e i
σn+1 = σtrial ∆γ i Nn+1 . (8.67)
i=1
Due to the isotropy of the model, the stress update formula can be written equivalently in
terms of principal stresses as
6
σj = σjtrial − ∆γ i (2G [Ndi ]j − K Nvi ), (8.68)
i=1
for j = 1, 2, 3. In the above formula, Nvi ≡ tr[Ni ] is the volumetric component of the flow
vector N i at the updated state and [Ndi ]j denotes the j th eigenvalue of its deviatoric
projection.
298 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
σ1 = σ1trial − ∆γ [2G(1 + 1
3 sin ψ) + 2K sin ψ]
σ3 = σ3trial + ∆γ [2G(1 − 1
3 sin ψ) − 2K sin ψ].
where ∆ε̄p is the (linear) function of ∆γ defined by (8.70) and a is the constant
1
a = 4G(1 + 3 sin φ sin ψ) + 4K sin φ sin ψ. (8.72)
Equation (8.71) is obtained by introducing the update formulae (8.69) into the main
plane equation Φ1 = 0.
2. The updated stress lies on the right edge of the pyramid. The incremental plastic strain,
obtained from the backward Euler discretisation of (8.57), is in this case given by the
two-vector formula
∆εp = ∆γ a N a + ∆γ b N b . (8.73)
The corresponding principal stress update formulae follow from the substitution of the
above expression into (8.68):
σ1 = σ1trial − [2G(1 + 1
3 sin ψ) + 2K sin ψ](∆γ a + ∆γ b )
At the right edge, the updated principal stresses are such that the equations of the main
plane, Φ1 = 0, and the plane on its right, Φ6 = 0, are simultaneously satisfied. This
condition gives the following set of two return-mapping equations to be solved for
∆γ a and ∆γ b :
3. The updated stress lies on the left edge. The return mapping to the left edge is
completely analogous to the return to the right edge derived in the above. The essential
difference is that the tensor N b is now defined by (8.59). The resulting principal stress
updating formulae are
σ1 = σ1trial − [2G(1 + 1
3 sin ψ) + 2K sin ψ]∆γ a + ( 43 G − 2K) sin ψ ∆γ b
σ3 = σ3trial + [2G(1 − 1
3 sin ψ) − 2K sin ψ](∆γ a + ∆γ b ),
(8.78)
and the corresponding increment of accumulated plastic strain is also given by (8.75).
The incremental plastic multipliers are the solution of the following return-mapping
equations:
The above equations represent the intersection of the main plane, Φ1 = 0, with the
plane on its left, Φ2 = 0.
4. The updated stress lies on the apex. The apex of the Mohr–Coulomb pyramid is the
point along the hydrostatic axis for which
p = c cot φ. (8.81)
300 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Now recall the general hydrostatic pressure-updating equation for a return mapping
with an underlying linear description of the elastic behaviour:
n+1 − K ∆εv .
pn+1 = ptrial p
(8.82)
The updated hydrostatic stress in the present case must lie on the apex of the updated
Mohr–Coulomb pyramid. This condition is enforced by introducing the update for-
mula (8.82), together with the hardening curve and the discretised evolution equation
for ε̄p , into the apex equation (8.81). We then have (Figure 8.8)
cos φ
∆ε̄p = α∆εpv ; α≡ . (8.84)
sin ψ
The substitution of this expression into (8.83) yields the final apex return-mapping
equation:
c(ε̄pn + α∆εpv ) cot φ − ptrial p
n+1 + K ∆εv = 0, (8.85)
for the unknown ∆εpv . Once the solution is found, we update
with pn+1 given by (8.82). Note that under zero dilatancy (ψ = 0), the return to apex
does not make sense as the purely deviatoric flow vector (combination of deviatoric
flow vectors) does not produce volumetric plastic flow. In this case, α → ∞.
Remark 8.1. Under the assumption of linear hardening, with c = c0 + H ε̄p , all four sets of
return-mapping equations are linear and, therefore, can be solved in closed form.
Remark 8.2. When φ = ψ = 0 all equations of the one-vector return mapping to the main
plane and the two-vector return mappings to the right and left edge reduce to those of the
Tresca model implementation, derived in Section 8.1.1.
yield surface at t n
σntrial
+1
σn +1= pn +1 I
- K ∆εvp
pntrial
+1
2. Check validity: if the updated principal stresses remain in the same sextant as the trial
stresses, i.e. if
σ1 ≥ σ2 ≥ σ3 , (8.87)
then the return to the main plane is valid and the corresponding updated state is
accepted.
3. Otherwise, try either the return mapping to the right or to the left edge (the procedure
to select which one to try at this stage is described later).
4. Check validity: if the updated principal stresses remain in the same sextant as the trial
stress, then the return mapping carried out in item 3 is valid and the updated state is
accepted.
5. Otherwise, apply the return algorithm to the apex. The result obtained now is necessar-
ily valid and does not require further checks.
The reader should note that any returned stress that fails to pass the validity check of
items 2 and 4 lies necessarily outside the updated elastic domain (which is fixed in the
perfectly plastic case) and is, therefore, not admissible. This can be easily verified by simple
geometrical considerations analogous to those of the Tresca model given in Figure 8.1.
302 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
σ n+1
trial
:T > 0 right σ n+1
trial
:T < 0 left
strial
n+1
a
− 2G ∆γ a Nd strial
n+1
a b
Nd Nd
b
Nd
a
− 2G ∆γ b Nd b
Nd
T sn+1
sn+1
yield surface
at t n σ1
b
Nd b
Nd
σ3
updated σ2
surface
n+1 − 2G(∆γ Nd + ∆γ Nd ),
a b
sn+1 = strial a b
(8.88)
where Nda and Ndb are the deviatoric components of N a and N b , respectively. Clearly,
N b in the above is normal to the right (left) plane if sn+1 lies on the right (left) edge. The
dotted line of Figure 8.9 is the deviatoric projection of the plane parallel to N a that contains
the hydrostatic line. If σtrial trial
n+1 lies on the right side of this plane (sn+1 lies on the right of
the dotted line) the only possible way to return to the left edge consistently with the above
formula is to have negative ∆γ b – an unacceptable solution. In this case, if σtrial n+1 returns to
an edge, it can only be the right edge. Similarly, if strial
n+1 lies on the left of the dotted line, the
only possibly valid edge return algorithm is the return to the left edge.
Based on the above considerations, the procedure for selection of the appropriate edge
return is extremely simple. Let T be a deviatoric tensor orthogonal to the deviatoric
projection Nda , pointing to the right. Here T is chosen with the following eigenvalues:
S ≡ T : σtrial
n+1
is computed first. If S > 0 the possible edge return is to the right. Otherwise, it is the return
mapping to the left edge.
check
consistency: process is elastic – update:
σ1trial−
trial
σ3trial+ (σ1trial+ σ3 ) sin φ
NO trial
_ ( ⋅ )n+1 := ( ⋅ ) n+1
−2 c ( ε np) cos φ > ∈tol ?
and exit
YES
check validity
of one-vector return: YES
σ 1 ≥ σ2 ≥ σ3 ?
NO
check validity
of two-vector return: YES
σ 1 ≥ σ2 ≥ σ3 ?
NO
apply
apply2vector return map
return mapping to
to apex assemble updated stress tensor:
apex obtain:
– obtain: 3
σn+1 := ∑ σ i e i⊗ e i
i=1
∆γ a∆ε
pb
, ∆γ
v and σi (i=1,2,3)
and other state variables, and exit
Figure 8.10. Flowchart of the integration algorithm for the Mohr–Coulomb model in principal stresses.
Procedure implemented in subroutine SUMC of program HYPLAS.
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 305
Box 8.4. Implicit elastic predictor/return mapping algorithm for the Mohr–
Coulomb model.
(i) Elastic predictor. Given ∆ε and the state variables at tn , evaluate the elastic trial state
εen+1
trial
:= εen + ∆ε; ε̄pn+1
trial
:= ε̄pn
σtrial e trial e trial
n+1 := 2G εd n+1 + K εv n+1 I
3
σn+1 := σ i ei ⊗ ei
i=1
Box 8.5. The Mohr–Coulomb model. One-vector return mapping to main plane.
dΦ̃
d := = −4G(1 + 13 sin ψ sin φ) (residual derivative)
d∆γ
− 4K sin ψ sin φ − 4H cos2 φ
∆γ := ∆γ − Φ̃/d (update ∆γ)
(iii) Check for convergence
ε̄pn+1 := ε̄pn + 2 cos φ ∆γ
Φ̃ := σ1trial − σ3trial + (σ1trial + σ3trial ) sin φ
− [4G(1 + 13 sin ψ sin φ) + 4K sin φ sin ψ]∆γ
− 2 c(ε̄pn+1 ) cos φ
IF |Φ̃| ≤ tol THEN update
σ1 := σ1trial − [2G(1 + 1
3
sin ψ) + 2K sin ψ]∆γ
σ2 := σ2trial + ( 43 G − 2K) sin ψ ∆γ
σ3 := σ3trial + [2G(1 − 1
3
sin ψ) − 2K sin ψ]∆γ
and RETURN to Box 8.4
(iv) GOTO (ii)
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 307
∆γ a := 0, ∆γ b := 0; ε̄pn+1 := ε̄pn
where
σ̄ a = σ1trial − σ3trial + (σ1trial + σ3trial ) sin φ
trial
b
σ1 − σ2trial + (σ1trial + σ2trial ) sin φ right edge
σ̄ =
σ2trial − σ3trial + (σ2trial + σ3trial ) sin φ left edge
Box 8.6 (contd. from page 307). The Mohr–Coulomb model. Two-vector return
mappings to edges (implemented in subroutine SUMC)
n+1 − K ∆εv
pn+1 := ptrial p
{i ε̄p , i c}
defined by the accumulated plastic strain and the corresponding cohesion. It is pointed out
that perfect plasticity – the most common assumption accompanying the use of the Mohr–
Coulomb model – is obtained by simply defining two sampling points with identical cohesion.
The FORTRAN source code of subroutine SUMC is listed below.
1 SUBROUTINE SUMC
2 1( DGAM ,IPROPS ,LALGVA ,NTYPE ,RPROPS ,
3 2 RSTAVA ,STRAT ,STRES )
4 IMPLICIT DOUBLE PRECISION (A-H,O-Z)
5 PARAMETER(IPHARD=7 ,MSTRE=4)
6 C Arguments
7 LOGICAL
8 1 LALGVA(5)
9 DIMENSION
10 1 DGAM(2) ,IPROPS(*) ,RPROPS(*) ,
11 2 RSTAVA(MSTRE+1) ,STRAT(MSTRE) ,STRES(MSTRE)
12 C Local variables and arrays
13 LOGICAL
14 1 APEX, DUMMY, EDGE, IFPLAS, RIGHT, SUFAIL
15 DIMENSION
16 1 EIGPRJ(MSTRE,2) ,PSTRS(3) ,STREST(3)
17 DATA
18 1 R0 ,R1 ,R2 ,R3 ,R4 ,SMALL ,TOL /
19 2 0.0D0,1.0D0,2.0D0,3.0D0,4.0D0,1.D-06,1.D-10/
20 DATA MXITER / 50 /
21 C***********************************************************************
22 C STATE UPDATE PROCEDURE FOR MOHR-COULOMB TYPE ELASTO-PLASTIC MATERIAL
23 C WITH ASSOCIATIVE/NON-ASSOCIATIVE FLOW RULE AND PIECE-WISE LINEAR
24 C ISOTROPIC HARDENING:
25 C IMPLICIT ELASTIC PREDICTOR/RETURN MAPPING ALGORITHM (BOXES 8.4-7).
26 C PLANE STRAIN AND AXISYMMETRIC IMPLMENTATIONS.
27 C***********************************************************************
28 C Stops program if neither plane strain nor plane stress state
29 IF(NTYPE.NE.2.AND.NTYPE.NE.3)CALL ERRPRT(’EI0027’)
30 C Initialize some algorithmic and internal variables
31 DGAMA=R0
32 DGAMB=R0
33 IFPLAS=.FALSE.
34 SUFAIL=.FALSE.
35 EDGE=.FALSE.
36 APEX=.FALSE.
37 EPBARN=RSTAVA(MSTRE+1)
38 EPBAR=EPBARN
39 C Set some material properties
40 YOUNG=RPROPS(2)
41 POISS=RPROPS(3)
42 SINPHI=RPROPS(4)
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 311
43 COSPHI=RPROPS(5)
44 SINPSI=RPROPS(6)
45 NHARD=IPROPS(3)
46 C Set some constants
47 GMODU=YOUNG/(R2*(R1+POISS))
48 BULK=YOUNG/(R3*(R1-R2*POISS))
49 R2G=R2*GMODU
50 R4G=R4*GMODU
51 R2BULK=R2*BULK
52 R2CPHI=R2*COSPHI
53 R1D3=R1/R3
54 C Compute elastic trial state
55 C ---------------------------
56 C Elastic trial volumetric strain and pressure stress
57 EETV=STRAT(1)+STRAT(2)+STRAT(4)
58 PT=BULK*EETV
59 C Spectral decomposition of the elastic trial stress
60 EETVD3=EETV*R1D3
61 STREST(1)=R2G*(STRAT(1)-EETVD3)+PT
62 STREST(2)=R2G*(STRAT(2)-EETVD3)+PT
63 STREST(3)=GMODU*STRAT(3)
64 CALL SPDEC2(EIGPRJ,PSTRS,DUMMY,STREST)
65 PSTRS(3)=R2G*(STRAT(4)-EETVD3)+PT
66 C Identify maximum (PSTRS1) and minimum (PSTRS3) principal stresses
67 II=1
68 JJ=1
69 PSTRS1=PSTRS(II)
70 PSTRS3=PSTRS(JJ)
71 DO 10 I=2,3
72 IF(PSTRS(I).GE.PSTRS1)THEN
73 II=I
74 PSTRS1=PSTRS(II)
75 ENDIF
76 IF(PSTRS(I).LT.PSTRS3)THEN
77 JJ=I
78 PSTRS3=PSTRS(JJ)
79 ENDIF
80 10 CONTINUE
81 IF(II.NE.1.AND.JJ.NE.1)MM=1
82 IF(II.NE.2.AND.JJ.NE.2)MM=2
83 IF(II.NE.3.AND.JJ.NE.3)MM=3
84 PSTRS2=PSTRS(MM)
85 C Compute trial yield function and check for plastic consistency
86 C --------------------------------------------------------------
87 COHE=PLFUN(EPBARN,NHARD,RPROPS(IPHARD))
88 SMCT=PSTRS1-PSTRS3+(PSTRS1+PSTRS3)*SINPHI
89 PHIA=SMCT-R2CPHI*COHE
90 RES=PHIA
91 IF(COHE.NE.R0)RES=RES/ABS(COHE)
92 IF(RES.GT.TOL)THEN
93 C Plastic step: Apply return mapping
94 C ==================================
95 IFPLAS=.TRUE.
96 C identify possible edge return: either right or left of main plane
97 SCAPRD=PSTRS1*(R1-SINPSI)+PSTRS2*(-R2)+PSTRS3*(R1+SINPSI)
98 IF(SCAPRD.GE.R0)THEN
99 RIGHT=.TRUE.
100 ELSE
101 RIGHT=.FALSE.
102 ENDIF
103 C Apply one-vector return mapping first (return to MAIN PLANE)
104 C ------------------------------------------------------------
105 SPHSPS=SINPHI*SINPSI
312 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
106 CONSTA=R4G*(R1+R1D3*SPHSPS)+R4*BULK*SPHSPS
107 R4C2PH=R2CPHI*R2CPHI
108 C Start Newton-Raphson iterations for DGAMA
109 DO 20 NRITER=1,MXITER
110 C Compute residual derivative
111 DENOM=-CONSTA-R4C2PH*DPLFUN(EPBAR,NHARD,RPROPS(IPHARD))
112 C Compute Newton-Raphson increment and update variable DGAMA
113 DDGAMA=-PHIA/DENOM
114 DGAMA=DGAMA+DDGAMA
115 C Compute new residual
116 EPBAR=EPBARN+R2CPHI*DGAMA
117 COHE=PLFUN(EPBAR,NHARD,RPROPS(IPHARD))
118 PHIA=SMCT-CONSTA*DGAMA-R2CPHI*COHE
119 C Check convergence
120 RESNOR=ABS(PHIA)
121 IF(SMCT.NE.R0)RESNOR=RESNOR/ABS(SMCT)
122 IF(RESNOR.LE.TOL)THEN
123 C Check validity of 1-vector return (check sextant of converged stress)
124 S1=PSTRS1-(R2G*(R1+R1D3*SINPSI)+R2BULK*SINPSI)*DGAMA
125 S2=PSTRS2+(R4G*R1D3-R2BULK)*SINPSI*DGAMA
126 S3=PSTRS3+(R2G*(R1-R1D3*SINPSI)-R2BULK*SINPSI)*DGAMA
127 DELTA=DMAX1(ABS(S1),ABS(S2),ABS(S3))*SMALL
128 IF(S1+DELTA.GE.S2.AND.S2+DELTA.GE.S3)THEN
129 C converged stress is in the same sextant as trial stress -> 1-vector
130 C return is valid.
131 P=(S1+S2+S3)*R1D3
132 GOTO 70
133 ELSE
134 C converged stress is not in the same sextant -> 1-vector result is
135 C not valid. Go to two-vector return map to edge
136 GOTO 30
137 ENDIF
138 ENDIF
139 20 CONTINUE
140 C failure of stress update procedure
141 SUFAIL=.TRUE.
142 CALL ERRPRT(’WE0003’)
143 GOTO 999
144 30 CONTINUE
145 C Apply two-vector return mapping to appropriate EDGE
146 C ---------------------------------------------------
147 DGAMA=R0
148 EPBAR=EPBARN
149 COHE=PLFUN(EPBARN,NHARD,RPROPS(IPHARD))
150 SMCTA=PSTRS1-PSTRS3+(PSTRS1+PSTRS3)*SINPHI
151 IF(RIGHT)THEN
152 SMCTB=PSTRS1-PSTRS2+(PSTRS1+PSTRS2)*SINPHI
153 ELSE
154 SMCTB=PSTRS2-PSTRS3+(PSTRS2+PSTRS3)*SINPHI
155 ENDIF
156 PHIA=SMCTA-R2CPHI*COHE
157 PHIB=SMCTB-R2CPHI*COHE
158 IF(RIGHT)THEN
159 CONSTB=R2G*(R1+SINPHI+SINPSI-R1D3*SPHSPS)+R4*BULK*SPHSPS
160 ELSE
161 CONSTB=R2G*(R1-SINPHI-SINPSI-R1D3*SPHSPS)+R4*BULK*SPHSPS
162 ENDIF
163 C Start Newton-Raphson iterations for DGAMA and DGAMB
164 DO 40 NRITER=1,MXITER
165 C Compute residual derivative matrix
166 FACTA=R4C2PH*DPLFUN(EPBAR,NHARD,RPROPS(IPHARD))
167 DRVAA=-CONSTA-FACTA
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 313
168 DRVAB=-CONSTB-FACTA
169 DRVBA=-CONSTB-FACTA
170 DRVBB=-CONSTA-FACTA
171 C Compute Newton-Raphson increment and update variables DGAMA and DGAMB
172 R1DDET=R1/(DRVAA*DRVBB-DRVAB*DRVBA)
173 DDGAMA=(-DRVBB*PHIA+DRVAB*PHIB)*R1DDET
174 DDGAMB=(DRVBA*PHIA-DRVAA*PHIB)*R1DDET
175 DGAMA=DGAMA+DDGAMA
176 DGAMB=DGAMB+DDGAMB
177 C Compute new residual
178 EPBAR=EPBARN+R2CPHI*(DGAMA+DGAMB)
179 COHE=PLFUN(EPBAR,NHARD,RPROPS(IPHARD))
180 PHIA=SMCTA-CONSTA*DGAMA-CONSTB*DGAMB-R2CPHI*COHE
181 PHIB=SMCTB-CONSTB*DGAMA-CONSTA*DGAMB-R2CPHI*COHE
182 C Check convergence
183 RESNOR=(ABS(PHIA)+ABS(PHIB))
184 FACTOR=(ABS(SMCTA)+ABS(SMCTB))
185 IF(FACTOR.NE.R0)RESNOR=RESNOR/FACTOR
186 IF(RESNOR.LE.TOL)THEN
187 C Check validity of 2-vector return to edge
188 AUX1=R2G*(R1+R1D3*SINPSI)+R2BULK*SINPSI
189 AUX2=(R4G*R1D3-R2BULK)*SINPSI
190 AUX3=R2G*(R1-R1D3*SINPSI)-R2BULK*SINPSI
191 IF(RIGHT)THEN
192 S1=PSTRS1-AUX1*(DGAMA+DGAMB)
193 S2=PSTRS2+AUX2*DGAMA+AUX3*DGAMB
194 S3=PSTRS3+AUX3*DGAMA+AUX2*DGAMB
195 ELSE
196 S1=PSTRS1-AUX1*DGAMA+AUX2*DGAMB
197 S2=PSTRS2+AUX2*DGAMA-AUX1*DGAMB
198 S3=PSTRS3+AUX3*(DGAMA+DGAMB)
199 ENDIF
200 DELTA=DMAX1(ABS(S1),ABS(S2),ABS(S3))*SMALL
201 IF(S1+DELTA.GE.S2.AND.S2+DELTA.GE.S3)THEN
202 C converged stress is in the same sextant as trial stress -> 2-vector
203 C return to edge is valid.
204 EDGE=.TRUE.
205 P=(S1+S2+S3)*R1D3
206 GOTO 70
207 ELSE
208 C converged stress is not in the same sextant -> 2-vector return to edge
209 C is not valid. Go to two-vector return map to APEX
210 GOTO 50
211 ENDIF
212 ENDIF
213 40 CONTINUE
214 C failure of stress update procedure
215 SUFAIL=.TRUE.
216 CALL ERRPRT(’WE0003’)
217 GOTO 999
218 50 CONTINUE
219 C Apply multi-vector return mapping to APEX
220 C ---------------------------------------
221 C Check conditions for which return to apex does not make sense
222 IF(SINPHI.EQ.R0)CALL ERRPRT(’EE0009’)
223 IF(SINPSI.EQ.R0)CALL ERRPRT(’EE0010’)
224 C Set initial guess for volumetric plastic strain increment DEPV
225 DEPV=R0
226 EPBAR=EPBARN
227 COHE=PLFUN(EPBAR,NHARD,RPROPS(IPHARD))
228 COTPHI=COSPHI/SINPHI
229 RES=COTPHI*COHE-PT
230 C Newton-Raphson iterations for DEPV
314 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
231 DO 60 NRITER=1,MXITER
232 DENOM=COSPHI*COTPHI/SINPSI*DPLFUN(EPBAR,NHARD,RPROPS(IPHARD))+
233 1 BULK
234 DDEPV=-RES/DENOM
235 DEPV=DEPV+DDEPV
236 EPBAR=EPBARN+COSPHI/SINPSI*DEPV
237 COHE=PLFUN(EPBAR,NHARD,RPROPS(IPHARD))
238 P=PT-BULK*DEPV
239 RES=COTPHI*COHE-P
240 C check for convergence
241 RESNOR=ABS(RES)
242 IF(PT.NE.R0)RESNOR=RESNOR/ABS(PT)
243 IF(RESNOR.LE.TOL)THEN
244 APEX=.TRUE.
245 DGAMA=DEPV
246 DGAMB=R0
247 C update principal stresses
248 S1=P
249 S2=P
250 S3=P
251 GOTO 70
252 ENDIF
253 60 CONTINUE
254 SUFAIL=.TRUE.
255 CALL ERRPRT(’WE0003’)
256 GOTO 999
257 70 CONTINUE
258 C update internal variable EPBAR and stress components
259 C -----------------------------------------------------
260 RSTAVA(MSTRE+1)=EPBAR
261 PSTRS(II)=S1
262 PSTRS(JJ)=S3
263 PSTRS(MM)=S2
264 STRES(1)=PSTRS(1)*EIGPRJ(1,1)+PSTRS(2)*EIGPRJ(1,2)
265 STRES(2)=PSTRS(1)*EIGPRJ(2,1)+PSTRS(2)*EIGPRJ(2,2)
266 STRES(3)=PSTRS(1)*EIGPRJ(3,1)+PSTRS(2)*EIGPRJ(3,2)
267 STRES(4)=PSTRS(3)
268 C and elastic engineering strain
269 EEVD3=P/BULK*R1D3
270 RSTAVA(1)=(STRES(1)-P)/R2G+EEVD3
271 RSTAVA(2)=(STRES(2)-P)/R2G+EEVD3
272 RSTAVA(3)=STRES(3)/GMODU
273 RSTAVA(4)=(STRES(4)-P)/R2G+EEVD3
274 ELSE
275 C Elastic step: update stress using linear elastic law
276 C ====================================================
277 STRES(1)=STREST(1)
278 STRES(2)=STREST(2)
279 STRES(3)=STREST(3)
280 STRES(4)=PSTRS(3)
281 C elastic engineering strain
282 RSTAVA(1)=STRAT(1)
283 RSTAVA(2)=STRAT(2)
284 RSTAVA(3)=STRAT(3)
285 RSTAVA(4)=STRAT(4)
286 ENDIF
287 999 CONTINUE
288 C Update algorithmic variables before exit
289 C ========================================
290 DGAM(1)=DGAMA
291 DGAM(2)=DGAMB
292 LALGVA(1)=IFPLAS
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 315
293 LALGVA(2)=SUFAIL
294 LALGVA(3)=EDGE
295 LALGVA(4)=RIGHT
296 LALGVA(5)=APEX
297 RETURN
298 END
RPROPS = [E, ν, sin φ, cos φ, sin ψ, 1 ε̄p , 1 c, 2 ε̄p , 2 c, . . . . . . , nhard ε̄p , nhard c],
where i c denote the cohesion at the different points supplied along the (piecewise linear)
isotropic hardening curve. The array LALGVA of logical algorithmic values also differs from
that of the Tresca model implementation. Here, we have:
← LALGVA. For the Mohr–Coulomb model this array contains the plastic yielding flag,
IFPLAS, the state update failure flag SUFAIL, the edge return flag EDGE, the right edge
return flag RIGHT and the apex return flag APEX. Flags IFPLAS and SUFAIL are set
in the same way as in the other material model implementations already described.
The edge return flag is set to .TRUE. if one of the two possible edge return mappings
(right or left edge) is used. It is set to .FALSE. otherwise. The right edge return flag
is set to .TRUE. if the possible edge return is to the right edge and is set to .FALSE.
otherwise. The flag APEX is set to .TRUE. only if the return mapping to the apex is
applied. The flags EDGE, RIGHT and APEX are required by subroutine CTMC to decide
which elastoplastic tangent operator to compute (consistent with main plane, right, left
edge or apex return mapping).
σ trial
∆σT
∆σN N
N
T T
T B
N A
C
σ1
σ2 σ3
Figure 8.11. Iso-error maps for the Mohr–Coulomb model. Starting points and increment directions.
having the vector T parallel to the corresponding edge of the Mohr–Coulomb pyramid. It is
remarked, however, that the results presented in this section give a very good insight into the
accuracy properties of the present algorithm.
The maps obtained are shown in Figure 8.12. Similarly to the integration algorithm for the
Tresca model (Section 8.1.3), large areas in which the integration error vanishes are observed
in the present case for all starting points. Within some narrow bands, however, the integration
error can be high. Recall that, in the absence of hardening, the present integration algorithm
simply projects the trial stress onto the yield surface along a suitable direction. The reason for
the observed relatively high errors can be verified by performing the projections graphically
in a single step and comparing with the result obtained when the increment is divided into
substeps.
10
0%
20%
5
5%
∆σN /c
0%
5% 0%
-10 -5 0 5 10
∆σT /c
(a)
10 10
0%
0%
5 5
50%
∆σN /c
∆σN /c
0% 20%
5%
10% 5%
0%
0 0
0 5 10 0 5 10
∆σT /c ∆σT /c
(b) (c)
Figure 8.12. Iso-error maps. Associative Mohr–Coulomb model: (a) point A; (b) point B; and,
(c) point C.
defined by the Mohr–Coulomb integration algorithm. Having computed the principal stress
consistent derivatives, the tangent operator
dσn+1
Dep ≡ trial
(8.92)
dεen+1
can then be assembled as described in Section A.3 of Appendix A (in the program, the matrix
form of Dep is assembled in subroutine DGISO2).
318 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Recall that the present model admits four distinct return-mapping sets of equations: return
to main plane, right, left edge and apex.§ For each one there is a different explicit form
for the principal stress derivatives and for Dep . In the finite element program, the actual
form adopted in the assemblage of the tangent stiffness matrix is the one consistent with
the previous application of the return-mapping procedure at the Gauss point in question. The
derivation of the principal stresses derivatives consistent with each of the four possible return
mappings is addressed in what follows.
and the constant a is defined by (8.72). Combination of the two differential forms above,
together with the use of the linear elastic relation
e
σtrial e trial
n+1 = D : εn+1 , (8.96)
renders the final expressions for the derivatives of the principal stresses. The resulting
expressions – not explicitly written in the present text – are implemented in subroutine CTMC
and can be easily identified from the FORTRAN source code listed in Section 8.2.5. In CTMC,
the computed principal stress derivatives are stored in matrix DPSTRS.
sets exist.
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 319
n+1 − K d∆εv
dσi = dptrial p
= K (dεtrial
1 + dεtrial
2 + dεtrial
3 − d∆εpv ) (8.97)
The substitution of this expression into (8.97) yields the principal stress derivatives consistent
with the apex return mapping:
∂σi K
=K 1− , (8.100)
∂εej trial K + cossinφ cot φ
ψ H
for i, j = 1, 2, 3.
Remark 8.3. In the absence of hardening (H = 0) all principal stress derivatives consistent
with the apex return vanish. Clearly, as in this case the apex remains fixed in stress space, we
have trivially Dep = 0.
1 SUBROUTINE CTMC
2 1( DMATX ,EPFLAG ,IPROPS ,LALGVA ,NTYPE ,
3 2 RPROPS ,RSTAVA ,STRAT ,STRES )
4 IMPLICIT DOUBLE PRECISION (A-H,O-Z)
5 PARAMETER(IPHARD=7 ,MDIM=3, MSTRE=4)
6 C Arguments
7 LOGICAL EPFLAG ,LALGVA(5)
8 DIMENSION
9 1 DMATX(MSTRE,MSTRE) ,IPROPS(*) ,RPROPS(*) ,
320 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
132 DPSTRS(MM,JJ)=BULK-R2GD3+(AUX2*((R2BULK*(DRVAB-DRVBB)-
133 1 (DRVBB*R2GD3+DRVAB*R4GD3))*SINPHI+DRVBB*R2G)+
134 2 AUX3*((R2BULK*(DRVBA-DRVAA)+(DRVAA*R4GD3+
135 3 DRVBA*R2GD3))*SINPHI-DRVBA*R2G))*R1DDET
136 DPSTRS(JJ,II)=BULK-R2GD3+((AUX2*(DRVBA-DRVAA)+AUX3*(DRVAB-
137 1 DRVBB))*((R2BULK+R2GD3)*SINPHI+R2G))*R1DDET
138 DPSTRS(JJ,MM)=BULK-R2GD3+(AUX2*(((R2BULK*(DRVBA-DRVAA)-
139 1 (DRVBA*R4GD3+DRVAA*R2GD3))*SINPHI)+DRVAA*R2G)+
140 2 AUX3*(((R2BULK*(DRVAB-DRVBB)+(DRVAB*R2GD3+
141 3 DRVBB*R4GD3))*SINPHI)-DRVAB*R2G))*R1DDET
142 DPSTRS(JJ,JJ)=BULK+R4GD3+(AUX2*(((R2BULK*(DRVBA-DRVAA)+
143 1 (DRVAA*R4GD3+DRVBA*R2GD3))*SINPHI)-DRVBA*R2G)+
144 2 AUX3*(((R2BULK*(DRVAB-DRVBB)-(DRVAB*R4GD3+
145 3 DRVBB*R2GD3))*SINPHI)+DRVBB*R2G))*R1DDET
146 ELSE
147 C ...returned to left edge
148 DPSTRS(II,II)=BULK+R4GD3+(AUX1*(((R2BULK*(DRVBB-DRVAB)+
149 1 (DRVAB*R4GD3+DRVBB*R2GD3))*SINPHI)+DRVBB*R2G)+
150 2 AUX2*(((R2BULK*(DRVBA-DRVAA)+(DRVAA*R4GD3+
151 3 DRVBA*R2GD3))*SINPHI)+DRVBA*R2G))*R1DDET
152 DPSTRS(II,MM)=BULK-R2GD3+(AUX1*(((R2BULK*(DRVBB-DRVAB)-
153 1 (DRVAB*R2GD3+DRVBB*R4GD3))*SINPHI)-DRVAB*R2G)+
154 2 AUX2*(((R2BULK*(DRVBA-DRVAA)-(DRVAA*R2GD3+
155 3 DRVBA*R4GD3))*SINPHI)-DRVAA*R2G))*R1DDET
156 DPSTRS(II,JJ)=BULK-R2GD3+((AUX1*(DRVBB-DRVAB)+AUX2*(DRVBA-
157 1 DRVAA))*(((R2BULK+R2GD3)*SINPHI)-R2G))*R1DDET
158 DPSTRS(MM,II)=BULK-R2GD3+(AUX1*(((R2BULK*(DRVAA-DRVBA)-
159 1 (DRVAA*R4GD3+DRVBA*R2GD3))*SINPHI)-DRVBA*R2G)+
160 2 AUX2*(((R2BULK*(DRVAB-DRVBB)-(DRVAB*R4GD3+
161 3 DRVBB*R2GD3))*SINPHI)-DRVBB*R2G))*R1DDET
162 DPSTRS(MM,MM)=BULK+R4GD3+(AUX1*(((R2BULK*(DRVAA-DRVBA)+
163 1 (DRVAA*R2GD3+DRVBA*R4GD3))*SINPHI)+DRVAA*R2G)+
164 2 AUX2*(((R2BULK*(DRVAB-DRVBB)+(DRVAB*R2GD3+
165 3 DRVBB*R4GD3))*SINPHI)+DRVAB*R2G))*R1DDET
166 DPSTRS(MM,JJ)=BULK-R2GD3+((AUX1*(DRVAA-DRVBA)+AUX2*(DRVAB-
167 1 DRVBB))*(((R2BULK+R2GD3)*SINPHI)-R2G))*R1DDET
168 DPSTRS(JJ,II)=BULK-R2GD3+(AUX3*(((R2BULK*(DRVAB-DRVBB-DRVAA+
169 1 DRVBA)+(DRVAA-DRVAB)*R4GD3+(DRVBA-DRVBB)*
170 2 R2GD3)*SINPHI)+(DRVBA-DRVBB)*R2G))*R1DDET
171 DPSTRS(JJ,MM)=BULK-R2GD3+(AUX3*(((R2BULK*(DRVAB-DRVBB-DRVAA+
172 1 DRVBA)+(DRVAB-DRVAA)*R2GD3+(DRVBB-DRVBA)*
173 2 R4GD3)*SINPHI)+(DRVAB-DRVAA)*R2G))*R1DDET
174 DPSTRS(JJ,JJ)=BULK+R4GD3+(AUX3*(DRVAB-DRVBB-DRVAA+DRVBA)*
175 1 (((R2BULK+R2GD3)*SINPHI)-R2G))*R1DDET
176 ENDIF
177 ELSEIF(APEX)THEN
178 C Tangent consistent with multi-vector return to apex
179 COTPHI=COSPHI/SINPHI
180 DSIDEJ=BULK*(R1-(BULK/(BULK+
181 1 DPLFUN(EPBAR,NHARD,RPROPS(IPHARD))*COTPHI*COSPHI/
182 2 SINPSI)))
183 DPSTRS(II,II)=DSIDEJ
184 DPSTRS(II,MM)=DSIDEJ
185 DPSTRS(II,JJ)=DSIDEJ
186 DPSTRS(MM,II)=DSIDEJ
187 DPSTRS(MM,MM)=DSIDEJ
188 DPSTRS(MM,JJ)=DSIDEJ
189 DPSTRS(JJ,II)=DSIDEJ
190 DPSTRS(JJ,MM)=DSIDEJ
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 323
191 DPSTRS(JJ,JJ)=DSIDEJ
192 ELSE
193 C Tangent consistent with 1-vector return to main active plane
194 SPHSPS=SINPHI*SINPSI
195 CONSTA=R4G*(R1+R1D3*SPHSPS)+R4*BULK*SPHSPS
196 DENOM=-CONSTA-R4C2PH*DPLFUN(EPBAR,NHARD,RPROPS(IPHARD))
197 B1=(R2G*(R1+R1D3*SINPSI)+R2BULK*SINPSI)/DENOM
198 B2=(R4G*R1D3-R2BULK)*SINPSI/DENOM
199 B3=(R2G*(R1-R1D3*SINPSI)-R2BULK*SINPSI)/DENOM
200 DPSTRS(II,II)=R2G*(R2D3+B1*(R1+R1D3*SINPHI))+
201 1 BULK*(R1+R2*B1*SINPHI)
202 DPSTRS(II,MM)=R1D3*(R3*BULK-R2G)*(R1+R2*B1*SINPHI)
203 DPSTRS(II,JJ)=R2G*(-R1D3-B1*(R1-R1D3*SINPHI))+
204 1 BULK*(R1+R2*B1*SINPHI)
205 DPSTRS(MM,II)=R2G*(-R1D3-B2*(R1+R1D3*SINPHI))+
206 1 BULK*(R1-R2*B2*SINPHI)
207 DPSTRS(MM,MM)=R4G*R1D3*(R1+B2*SINPHI)+BULK*(R1-R2*B2*SINPHI)
208 DPSTRS(MM,JJ)=R2G*(-R1D3+B2*(R1-R1D3*SINPHI))+
209 1 BULK*(R1-R2*B2*SINPHI)
210 DPSTRS(JJ,II)=R2G*(-R1D3-B3*(R1+R1D3*SINPHI))+
211 1 BULK*(R1-R2*B3*SINPHI)
212 DPSTRS(JJ,MM)=R1D3*(R3*BULK-R2G)*(R1-R2*B3*SINPHI)
213 DPSTRS(JJ,JJ)=R2G*(R2D3+B3*(R1-R1D3*SINPHI))+
214 1 BULK*(R1-R2*B3*SINPHI)
215 ENDIF
216 C
217 IF(NTYPE.EQ.2)THEN
218 OUTOFP=.FALSE.
219 ELSEIF(NTYPE.EQ.3)THEN
220 OUTOFP=.TRUE.
221 ENDIF
222 CALL DGISO2
223 1( DPSTRS ,DMATX ,EIGPRJ ,PSTRA ,PSTRS ,
224 2 OUTOFP ,REPEAT )
225 ELSE
226 C Compute elasticity matrix
227 C -------------------------
228 IF(NTYPE.EQ.2)THEN
229 NSTRE=3
230 ELSEIF(NTYPE.EQ.3)THEN
231 NSTRE=4
232 ENDIF
233 FACTOR=BULK-R2G*R1D3
234 DO 50 I=1,NSTRE
235 DO 40 J=I,NSTRE
236 DMATX(I,J)=R2G*FOID(I,J)+FACTOR*SOID(I)*SOID(J)
237 40 CONTINUE
238 50 CONTINUE
239 DO 70 J=1,NSTRE-1
240 DO 60 I=J+1,NSTRE
241 DMATX(I,J)=DMATX(J,I)
242 60 CONTINUE
243 70 CONTINUE
244 ENDIF
245 RETURN
246 END
324 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Following the format of the previous sections, a brief summary of the constitutive
equations adopted in the present implementation is presented below.
where
J2 = 12 s : s; s = σ − p(σ) I, (8.102)
p = 1/3 tr[σ] is the hydrostatic pressure and c is the cohesion. The constants η and
ξ are chosen according to the required approximation to the Mohr–Coulomb criterion.
Their formulae for outer edge, inner edge, plane strain, uniaxial tension/compression and
biaxial tension/compression matching are given, respectively, by expressions (6.122), (6.123)
and (6.124), (6.126) and (6.127).
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 325
The generally non-associative flow rule is adopted in the present implementation of the
Drucker–Prager model. The corresponding flow potential is
Ψ(σ, c) = J2 (s(σ)) + η̄ p(σ), (8.103)
where the constant η̄ depends on the dilatancy angle, ψ, and according to the adopted
Drucker–Prager approximation to the Mohr–Coulomb law may be given by (6.163), (6.164)
or (6.165) respectively for the outer, inner edge and plane strain match. The flow rule reads
ε̇p = γ̇ N, (8.104)
where, on the smooth portion of the yield surface, the flow vector is given by
∂Ψ 1 η̄
N= = s + I. (8.105)
∂σ 2 J2 (s) 3
At the cone apex, where the potential is singular, the flow vector is a subgradient of Ψ; that
is, N is a vector contained in the complementary cone (refer to the schematic illustration of
Figure 6.20(b), page 176).
As for the Mohr–Coulomb implementation of the previous section, associative isotropic
strain hardening is adopted here. The hardening law is defined by (8.61), together with the
associative evolution equation for the accumulated plastic strain derived for Drucker–Prager
plasticity in Section 6.6.3:
ε̄˙p = γ̇ ξ. (8.106)
Analogously to the Mohr–Coulomb implementation, the Drucker–Prager counterpart of
relation (8.64):
ξ
ε̄˙p = ε̇pv , (8.107)
η̄
will be needed in the derivation of the return-mapping equation for the apex of the Drucker–
Prager cone. This relation is obtained by combining the non-associative version of (6.161)
with the evolution law (8.106).
of whether σn+1 lies on the smooth portion of the cone or on its apex), the return vector
is always parallel to the plane that contains σtrial
n+1 and the hydrostatic axis (see Figure 8.14,
page 329). Thus, without loss of generality, the return-mapping algorithm can be completely
formulated in such a plane of the stress space. The corresponding algorithms are derived in
what follows. Note that, since the definition of the flow vector Nn+1 in the smooth portion
of the cone differs from that at the apex singularity, two possible explicit forms exist for the
return-mapping algorithm. These are treated separately below.
Simplification of the above expression can be obtained by noting that, due to the definition of
J2 , the following identity holds
sn+1 strial
n+1
= . (8.111)
J2 (sn+1 ) J2 (strial
n+1 )
It should be noted that the deviatoric updated stress, sn+1 , is obtained by simply scaling down
the trial deviatoric stress, strial
n+1 , by a factor that depends on ∆γ. Recall that in the von Mises
implicit return mapping derived in Chapter 7 the updated deviatoric stress is also obtained by
scaling its elastic trial counterpart.
The consistency condition in the present case is given by
Φn+1 ≡ J2 (sn+1 ) + η pn+1 − ξ c(ε̄pn+1 ) = 0, (8.114)
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 327
where the update accumulated plastic strain is obtained from the discrete version of (8.106):
with
∆ε̄p = ξ ∆γ. (8.116)
The substitution of the above expression and (8.113) into the consistency condition results
in the following (generally nonlinear) equation for ∆γ:
n+1 ) − G ∆γ + η (pn+1 − K η̄ ∆γ)
Φ̃(∆γ) ≡ J2 (strial trial
where
ξ ξ
α≡ , β≡ . (8.120)
η η̄
This equation is analogous to the return-mapping equation (8.85) of the Mohr–Coulomb
implementation. Its geometrical representation is shown in Figure 8.13. After the solution
of (8.119) for ∆εpv , we update
Note that for non-dilatant flow (η̄ = 0), the return to apex does not make sense in the present
context and the comments made immediately after expression (8.86) for the Mohr–Coulomb
implementation apply.
Remark 8.4. For perfectly plastic materials, c is constant and, for linearly hardening models,
the hardening function reads c(ε̄p ) = c0 + H ε̄p , where H denotes the hardening modulus.
In such cases, the return-mapping equations (8.117) and (8.119) respectively for the smooth
portion and apex of the Drucker–Prager cone are linear and ∆γ is computed in closed form.
328 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
J2( s )
complementary
cone
σn+1
trial
updated
Drucker–Prager
cone
-K ∆εvp I
trial p
σn+1= pn+1 I pn+1
• Firstly, apply the return algorithm to the smooth part of the Drucker–Prager cone. From
the corresponding deviatoric stress update formula (8.113)1, it follows that
n+1 ) − G ∆γ.
J2 (sn+1 ) = J2 (strial (8.122)
If, after determination of ∆γ from the related consistency condition, the following is
satisfied
acone ≡ J2 (strial
n+1 ) − G ∆γ ≥ 0, (8.123)
then the returned stress indeed lies on the Drucker–Prager cone and the return mapping
is validated.
• Otherwise, the returned stress lies outside the updated elastic domain and is not
admissible. In this case, the return mapping to the apex must be applied. The results
obtained by the apex return are then necessarily valid.
The above selection procedure ensures the consistency of the selected return algorithm
regardless of any prescribed hardening/softening law.
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 329
J2( s )
σ ntrial
+1
1
η
J2( s trial
n +1 ) - G ∆γ > 0
σ n +1 σ ntrial
+1
updated
elastic domain
ξ c(ε n +1 )
p
apex p
J2( s trial
n +1 ) - G ∆γ < 0
σ n +1
check
consistency: process is elastic update:
NO
√
trial )
J2( sn+1 + η pn+1
_p
trial
trial
( ⋅ )n+1 := ( ⋅ ) n+1
− ξ c ( ε n) > ∈tol ?
and exit
YES
check validity
of one-vector return: YES
√
trial ) − G ∆γ ≥ 0
J2( sn+1 ?
NO
apply
apply2vector return map
return mapping to
to apex
apex obtain:
– obtain:
∆γ ∆ε
a p b
, ∆γ
v and σ n+1
Figure 8.15. Flowchart of the implicit elastic predictor/return-mapping scheme for the Drucker–Prager
model. Procedure implemented in subroutine SUDP of program HYPLAS.
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 331
(i) Elastic predictor. Given ∆ε and the state variables at tn , evaluate the elastic trial state
εen+1
trial
:= εen + ∆ε ε̄pn+1
trial
:= ε̄pn
strial e trial
n+1 := 2G εd n+1 ptrial e trial
n+1 := K εv n+1
Box 8.9. The Drucker–Prager model. Return mapping to smooth portion of cone.
r := c(ε̄pn ) β − ptrial
n+1
r := β c(ε̄pn+1 ) − pn+1
{i ε̄p , i c}.
1 SUBROUTINE SUDP
2 1( DGAM ,IPROPS ,LALGVA ,NTYPE ,RPROPS ,
3 2 RSTAVA ,STRAT ,STRES )
4 IMPLICIT DOUBLE PRECISION (A-H,O-Z)
5 PARAMETER(IPHARD=7 ,MSTRE=4)
6 LOGICAL APEX, IFPLAS, LALGVA(3), SUFAIL
7 DIMENSION
8 1 IPROPS(*) ,RPROPS(*) ,RSTAVA(MSTRE+1) ,
9 2 STRAT(MSTRE) ,STRES(MSTRE)
10 DIMENSION
11 1 STRIAL(MSTRE)
12 DATA
13 1 R0 ,RP5 ,R1 ,R2 ,R3 ,TOL /
14 2 0.0D0,0.5D0,1.0D0,2.0D0,3.0D0,1.D-08/
15 DATA MAXRT / 50 /
16 C***********************************************************************
17 C STRESS UPDATE PROCEDURE FOR DRUCKER PRAGER TYPE ELASTO-PLASTIC
18 C MATERIAL WITH ASSOCIATIVE/NON-ASSOCIATIVE FLOW RULE AND PIECE_WISE
19 C LINEAR ISOTROPIC HARDENING:
20 C IMPLICIT ELASTIC PREDICTOR/RETURN MAPPING ALGORITHM (Boxes 8.8-10)
21 C***********************************************************************
22 C Stops program if neither plane strain nor axisymmetric
23 IF(NTYPE.NE.2.AND.NTYPE.NE.3)CALL ERRPRT(’EI0016’)
24 C Initialize some algorithmic and internal variables
25 DGAMA=R0
26 IFPLAS=.FALSE.
27 SUFAIL=.FALSE.
28 EPBARN=RSTAVA(MSTRE+1)
29 EPBAR=EPBARN
30 C Set some material properties
31 YOUNG=RPROPS(2)
32 POISS=RPROPS(3)
33 ETA=RPROPS(4)
34 XI=RPROPS(5)
35 ETABAR=RPROPS(6)
36 NHARD=IPROPS(3)
37 C and some constants
38 GMODU=YOUNG/(R2*(R1+POISS))
39 BULK=YOUNG/(R3*(R1-R2*POISS))
40 R2G=R2*GMODU
41 R1D3=R1/R3
42 C Compute elastic trial state
43 C ---------------------------
44 C Elastic trial volumetric strain and pressure stress
45 EETV=STRAT(1)+STRAT(2)+STRAT(4)
46 PT=BULK*EETV
47 C Elastic trial deviatoric stress
48 EEVD3=EETV*R1D3
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 335
49 STRIAL(1)=R2G*(STRAT(1)-EEVD3)
50 STRIAL(2)=R2G*(STRAT(2)-EEVD3)
51 STRIAL(4)=R2G*(STRAT(4)-EEVD3)
52 C shear component
53 STRIAL(3)=R2G*(STRAT(3)*RP5)
54 C Compute elastic trial stress J2 invariant and cohesion
55 VARJ2T=STRIAL(3)*STRIAL(3)+RP5*(STRIAL(1)*STRIAL(1)+
56 1 STRIAL(2)*STRIAL(2)+STRIAL(4)*STRIAL(4))
57 COHE=PLFUN(EPBARN,NHARD,RPROPS(IPHARD))
58 C Check for plastic consistency
59 C -----------------------------
60 SQRJ2T=SQRT(VARJ2T)
61 PHI=SQRJ2T+ETA*PT-XI*COHE
62 RES=PHI
63 IF(COHE.NE.R0)RES=RES/ABS(COHE)
64 IF(RES.GT.TOL)THEN
65 C Plastic step: Use return mapping
66 C ================================
67 IFPLAS=.TRUE.
68 APEX=.FALSE.
69 C Apply return mapping to smooth portion of cone - Box 8.9
70 C --------------------------------------------------------
71 DO 20 IPTER1=1,MAXRT
72 C Compute residual derivative
73 DENOM=-GMODU-BULK*ETABAR*ETA-
74 1 XI*XI*DPLFUN(EPBAR,NHARD,RPROPS(IPHARD))
75 C Compute Newton-Raphson increment and update variable DGAMA
76 DDGAMA=-PHI/DENOM
77 DGAMA=DGAMA+DDGAMA
78 C Compute new residual
79 EPBAR=EPBARN+XI*DGAMA
80 COHE=PLFUN(EPBAR,NHARD,RPROPS(IPHARD))
81 SQRJ2=SQRJ2T-GMODU*DGAMA
82 P=PT-BULK*ETABAR*DGAMA
83 PHI=SQRJ2+ETA*P-XI*COHE
84 C Check convergence
85 RESNOR=ABS(PHI)
86 IF(COHE.NE.R0)RESNOR=RESNOR/ABS(COHE)
87 IF(RESNOR.LE.TOL)THEN
88 C Check validity of return to smooth portion
89 IF(SQRJ2.GE.R0)THEN
90 C results are valid, update stress components and other variables
91 IF(SQRJ2T.EQ.R0)THEN
92 FACTOR=R0
93 ELSE
94 FACTOR=R1-GMODU*DGAMA/SQRJ2T
95 ENDIF
96 GOTO 50
97 ELSE
98 C smooth wall return not valid - go to apex return procedure
99 GOTO 30
100 ENDIF
101 ENDIF
102 20 CONTINUE
103 C failure of stress update procedure
104 SUFAIL=.TRUE.
105 CALL ERRPRT(’WE0002’)
106 GOTO 999
107 30 CONTINUE
108 C Apply return mapping to APEX - Box 8.10
109 C ---------------------------------------
110 C perform checks and set some variables
336 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
111 APEX=.TRUE.
112 IF(ETA.EQ.R0)CALL ERRPRT(’EE0011’)
113 IF(ETABAR.EQ.R0)CALL ERRPRT(’EE0012’)
114 ALPHA=XI/ETABAR
115 BETA=XI/ETA
116 C Set initial guess for unknown DEPV and start iterations
117 DEPV=R0
118 EPBAR=EPBARN
119 COHE=PLFUN(EPBAR,NHARD,RPROPS(IPHARD))
120 RES=BETA*COHE-PT
121 DO 40 IPTER2=1,MAXRT
122 DENOM=ALPHA*BETA*DPLFUN(EPBAR,NHARD,RPROPS(IPHARD))+BULK
123 C Compute Newton-Raphson increment and update variable DEPV
124 DDEPV=-RES/DENOM
125 DEPV=DEPV+DDEPV
126 C Compute new residual
127 EPBAR=EPBARN+ALPHA*DEPV
128 COHE=PLFUN(EPBAR,NHARD,RPROPS(IPHARD))
129 P=PT-BULK*DEPV
130 RES=BETA*COHE-P
131 C Check convergence
132 RESNOR=ABS(RES)
133 IF(COHE.NE.R0)RESNOR=RESNOR/ABS(COHE)
134 IF(RESNOR.LE.TOL)THEN
135 C update stress components and other variables
136 DGAMA=DEPV/ETABAR
137 FACTOR=R0
138 GOTO 50
139 ENDIF
140 40 CONTINUE
141 C failure of stress update procedure
142 SUFAIL=.TRUE.
143 CALL ERRPRT(’WE0002’)
144 GOTO 999
145 C Store converged stress components and other state variables
146 C -----------------------------------------------------------
147 50 CONTINUE
148 STRES(1)=FACTOR*STRIAL(1)+P
149 STRES(2)=FACTOR*STRIAL(2)+P
150 STRES(3)=FACTOR*STRIAL(3)
151 STRES(4)=FACTOR*STRIAL(4)+P
152 C update EPBAR
153 RSTAVA(MSTRE+1)=EPBAR
154 C compute converged elastic (engineering) strain components
155 FACTOR=FACTOR/R2G
156 EEVD3=P/(BULK*R3)
157 RSTAVA(1)=FACTOR*STRIAL(1)+EEVD3
158 RSTAVA(2)=FACTOR*STRIAL(2)+EEVD3
159 RSTAVA(3)=FACTOR*STRIAL(3)*R2
160 RSTAVA(4)=FACTOR*STRIAL(4)+EEVD3
161 ELSE
162 C Elastic step: update stress using linear elastic law
163 C ====================================================
164 STRES(1)=STRIAL(1)+PT
165 STRES(2)=STRIAL(2)+PT
166 STRES(3)=STRIAL(3)
167 STRES(4)=STRIAL(4)+PT
168 C elastic engineering strain
169 RSTAVA(1)=STRAT(1)
170 RSTAVA(2)=STRAT(2)
171 RSTAVA(3)=STRAT(3)
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 337
172 RSTAVA(4)=STRAT(4)
173 ENDIF
174 999 CONTINUE
175 C Update some algorithmic variables before exit
176 C =============================================
177 LALGVA(1)=IFPLAS
178 LALGVA(2)=SUFAIL
179 LALGVA(3)=APEX
180 DGAM=DGAMA
181 RETURN
182 END
6
2.5
2.5 55.0
7.57.5
4
∆σN /q
10
10.0 2
0.0
0
0
0 2 4 6
∆σT /q
above for the Drucker–Prager model. For this model, there are two possible elastoplastic
tangents: one consistent with the return to the smooth portion of the cone and the other
consistent with the return to the apex. Following the procedure adopted in the implementation
of the Tresca and Mohr–Coulomb models where, respectively, three and four elastoplastic
explicit forms of tangent exist, the actual elastoplastic tangents used to assemble the stiffness
matrix are the ones consistent with the last application of the return-mapping procedure at
each Gauss point.
where the second-order tensor D is the unit tensor parallel to εed trial
n+1 :
εed trial
n+1
D≡ . (8.126)
εed trial
n+1
gives
n+1 − η̄ d∆γ).
dpn+1 = K (dεev trial (8.128)
The expression relating d∆γ and the differentials of trial strain is obtained by linearising the
consistency condition (8.117). The linearisation of (8.117), in conjunction with the use of the
elastic relation, gives the following equation:
√
dΦ̃ = 2 G D : dεed trial e trial
n+1 + Kη dεv n+1
− (G + Kη η̄ + ξ 2 H) d∆γ = 0, (8.129)
the explicit expression for the elastoplastic tangent consistent with the one-vector return is
obtained after some straightforward manipulations as
∆γ ∆γ
Dep = 2G 1 − √ I d + 2G √ − GA D⊗D
n+1
2εed trial n+1
2εed trial
√
− 2 GAK (η D ⊗ I + η̄ I ⊗ D) + K (1 − Kη η̄A) I ⊗ I, (8.132)
n+1 − K ∆εv .
pn+1 = ptrial p
(8.136)
The explicit expression for the d∆εpv in terms of the elastic trial strain differential is obtained
by differentiating the residual equation (8.119). This gives
Finally, with the substitution of the above formula into (8.137) and use of (8.135) we obtain
the expression for the tangent operator consistent with the apex return mapping:
K
Dep = K 1 − I ⊗ I. (8.140)
K + αβH
This formula is in complete analogy with that of the apex return for the Mohr–Coulomb
model, given by (8.100) in terms of principal stress derivatives.
Remark 8.5. The comments made in Remark 8.3 are equally valid for the Drucker–Prager
model. Note that, in the absence of hardening (H = 0) the above tangent operator vanishes.
1 SUBROUTINE CTDP
2 1( DGAM ,DMATX ,EPFLAG ,IPROPS ,LALGVA ,
3 2 NTYPE ,RPROPS ,RSTAVA ,STRAT )
4 IMPLICIT DOUBLE PRECISION (A-H,O-Z)
5 PARAMETER(IPHARD=7 ,MSTRE=4)
6 LOGICAL APEX, EPFLAG, LALGVA(3)
7 DIMENSION
8 1 DMATX(MSTRE,MSTRE),IPROPS(*) ,RPROPS(*) ,
9 2 RSTAVA(MSTRE+1) ,STRAT(MSTRE)
10 DIMENSION
11 1 EETD(MSTRE) ,FOID(MSTRE,MSTRE) ,SOID(MSTRE) ,
12 2 UNIDEV(MSTRE)
13 DATA
14 1 FOID(1,1),FOID(1,2),FOID(1,3),FOID(1,4)/
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 341
15 2 1.0D0 ,0.0D0 ,0.0D0 ,0.0D0 /
16 3 FOID(2,1),FOID(2,2),FOID(2,3),FOID(2,4)/
17 4 0.0D0 ,1.0D0 ,0.0D0 ,0.0D0 /
18 5 FOID(3,1),FOID(3,2),FOID(3,3),FOID(3,4)/
19 6 0.0D0 ,0.0D0 ,0.5D0 ,0.0D0 /
20 7 FOID(4,1),FOID(4,2),FOID(4,3),FOID(4,4)/
21 8 0.0D0 ,0.0D0 ,0.0D0 ,1.0D0 /
22 DATA
23 1 SOID(1) ,SOID(2) ,SOID(3) ,SOID(4) /
24 2 1.0D0 ,1.0D0 ,0.0D0 ,1.0D0 /
25 DATA
26 1 R0 ,R1 ,RP5 ,R2 ,R3 /
27 2 0.0D0,1.0D0,0.5D0,2.0D0,3.0D0/
28 C***********************************************************************
29 C COMPUTATION OF CONSISTENT TANGENT MODULUS FOR DRUCKER-PRAGER TYPE
30 C ELASTO-PLASTIC MATERIAL WITH ASSOCIATIVE/NON-ASSOCIATIVE FLOW RULE AND
31 C PIECE-WISE LINEAR ISOTROPIC HARDENING
32 C***********************************************************************
33 IF(NTYPE.EQ.2)THEN
34 NSTRE=3
35 ELSEIF(NTYPE.EQ.3)THEN
36 NSTRE=4
37 ELSE
38 CALL ERRPRT(’EI0017’)
39 ENDIF
40 C Retrieve accumulated plastic strain, DGAMA and APEX algorithm flag
41 EPBAR=RSTAVA(MSTRE+1)
42 DGAMA=DGAM
43 APEX=LALGVA(3)
44 C Set some material properties
45 YOUNG=RPROPS(2)
46 POISS=RPROPS(3)
47 ETA=RPROPS(4)
48 XI=RPROPS(5)
49 ETABAR=RPROPS(6)
50 NHARD=IPROPS(3)
51 C and some constants
52 GMODU=YOUNG/(R2*(R1+POISS))
53 BULK=YOUNG/(R3*(R1-R2*POISS))
54 R2G=R2*GMODU
55 R1D3=R1/R3
56 ROOT2=SQRT(R2)
57 C
58 IF(EPFLAG)THEN
59 C Compute elastoplastic consistent tangent
60 C ========================================
61 C Hardening slope
62 HSLOPE=DPLFUN(EPBAR,NHARD,RPROPS(IPHARD))
63 IF(APEX)THEN
64 C Elastoplastic tangent consistent with apex return
65 C -------------------------------------------------
66 ALPHA=XI/ETABAR
67 BETA=XI/ETA
68 AFACT=BULK*(R1-BULK/(BULK+ALPHA*BETA*HSLOPE))
69 DO 20 I=1,NSTRE
70 DO 10 J=1,NSTRE
71 DMATX(I,J)=AFACT*SOID(I)*SOID(J)
72 10 CONTINUE
73 20 CONTINUE
74 ELSE
75 C Elastoplastic tangent consistent with smooth cone wall return
342 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
76 C -------------------------------------------------------------
77 C Elastic trial deviatoric (physical) strain
78 EEVD3=(STRAT(1)+STRAT(2)+STRAT(4))*R1D3
79 EETD(1)=STRAT(1)-EEVD3
80 EETD(2)=STRAT(2)-EEVD3
81 EETD(3)=STRAT(3)*RP5
82 EETD(4)=STRAT(4)-EEVD3
83 ETDNOR=SQRT(EETD(1)*EETD(1)+EETD(2)*EETD(2)+
84 1 R2*EETD(3)*EETD(3)+EETD(4)*EETD(4))
85 C Unit deviatoric flow vector
86 IF(ETDNOR.NE.R0)THEN
87 EDNINV=R1/ETDNOR
88 ELSE
89 EDNINV=R0
90 ENDIF
91 DO 30 I=1,NSTRE
92 UNIDEV(I)=EETD(I)*EDNINV
93 30 CONTINUE
94 C Assemble tangent
95 AUX=R1/(GMODU+BULK*ETA*ETABAR+XI*XI*HSLOPE)
96 AFACT=R2G*(R1-DGAMA/(ROOT2*ETDNOR))
97 AFACD3=AFACT*R1D3
98 BFACT=R2G*(DGAMA/(ROOT2*ETDNOR)-GMODU*AUX)
99 CFACT=-ROOT2*GMODU*BULK*AUX
100 DFACT=BULK*(R1-BULK*ETA*ETABAR*AUX)
101 DO 50 I=1,NSTRE
102 DO 40 J=1,NSTRE
103 DMATX(I,J)=AFACT*FOID(I,J)+BFACT*UNIDEV(I)*UNIDEV(J)+
104 1 CFACT*(ETA*UNIDEV(I)*SOID(J)+
105 2 ETABAR*SOID(I)*UNIDEV(J))+
106 3 (DFACT-AFACD3)*SOID(I)*SOID(J)
107 40 CONTINUE
108 50 CONTINUE
109 ENDIF
110 ELSE
111 C Compute elasticity matrix
112 C =========================
113 FACTOR=BULK-R2G*R1D3
114 DO 70 I=1,NSTRE
115 DO 60 J=I,NSTRE
116 DMATX(I,J)=R2G*FOID(I,J)+FACTOR*SOID(I)*SOID(J)
117 60 CONTINUE
118 70 CONTINUE
119 DO 90 J=1,NSTRE-1
120 DO 80 I=J+1,NSTRE
121 DMATX(I,J)=DMATX(J,I)
122 80 CONTINUE
123 90 CONTINUE
124 ENDIF
125 RETURN
126 END
The situation here is completely analogous to that of subroutine CTMC (refer to page 324),
with state update related arguments coming in the present case from subroutine SUDP.
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 343
• UNIDEV [D]. Array of components of the unit tensor parallel to the deviatoric elastic
trial strain, D ≡ εed trial
n+1 /εd n+1 .
e trial
8.4. Examples
This section presents a set of benchmark numerical examples involving all material model
implementations described in the previous sections of this chapter. The examples presented
comprise metal plasticity (with the Tresca model) as well as some typical soil mechanics
applications with pressure-sensitive models (Mohr–Coulomb and Drucker–Prager). It is noted
that all results presented here have been obtained with the standard version of program
HYPLAS that accompanies this book. Whenever available, theoretical solutions are presented
and compared with the results obtained in order to illustrate the accuracy of the numerical
procedures described in the preceding sections. Without exception, the full Newton–Raphson
algorithm has been adopted in the iterative solution of the global finite element equilibrium
equations.
Mu ≈ 0.623 c a2 (8.141)
where a is the thickness of the bar in the neck and the cohesion or shear strength, c, for the
Tresca model is given as c = σy /2. For the present material parameters and geometry, it gives
Mu ≈ 1.869 Nm (8.142)
per millimetre width of the bar. The finite element model adopted in the present analysis is
shown in Figure 8.18. For symmetry reasons, only half of the bar is discretised with symmetry
kinematical constraints imposed on the nodes across the neck. A mesh of 312 eight-noded
elements (with 2 × 2 Gauss quadrature) is adopted with a total of 1001 nodes. The moment
M is applied by means of two opposite nodal forces of equal intensity F prescribed on the
nodes indicated in Figure 8.18. Due to the distance between the neck and the points where the
forces are applied, a condition close to pure bending is obtained near the neck. A total number
of 13 load increments is applied in the finite element analysis. The vertical deflection obtained
for the nodes where the forces are applied is plotted in Figure 8.19 against the normalised
344 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
h = 8 mm
a = 5 mm M
l = 40 mm
Figure 8.17. Bending of a V-notched bar. Problem definition. (Reproduced with permission from
A new computational model for Tresca plasticity at finite strains with an optimal parametrization in
the principal space, D Perić and EA de Souza Neto, Computer Methods in Applied Mechanics and
Engineering, Vol 171 c 1999 Elsevier Science S.A.)
Please Wait.. M = Fh h
F
Figure 8.18. Bending of a V-notched bar. Finite element model. (Reproduced with permission from
A new computational model for Tresca plasticity at finite strains with an optimal parametrization in
the principal space, D Perić and EA de Souza Neto, Computer Methods in Applied Mechanics and
Engineering, Vol 171 c 1999 Elsevier Science S.A.)
bending moment per unit width, M/ca2 . The limit load determined by the present analysis is
fe
Mlim
≈ 0.636. (8.143)
c a2
This result is in very close agreement (about 2% higher) with Green’s upper bound. The slip-
line field proposed by Green (1953) for the present dimensions is schematically illustrated in
Figure 8.20(a). Alongside, in Figure 8.20(b), the contour plot of the incremental accumulated
plastic strain, ∆ε̄p , corresponding to load increment 12 of the finite element computation
is shown for comparison. In this increment the bar is effectively collapsing, i.e. a small
increment of applied load results in extremely large incremental deflections. The contour
plot of ∆ε̄p shows the area where the plastic process is concentrated and clearly illustrates
the collapse mechanism. Note that the collapse mechanism predicted by the finite element
simulation is in agreement with Green’s slip-line field.
0.8
0.4
0
0 0.5 1 1.5 2
edge deflection (mm)
Figure 8.19. Bending of a V-notched bar. Moment-deflection diagram. (Reproduced with permission
from A new computational model for Tresca plasticity at finite strains with an optimal parametrization
in the principal space, D Perić and EA de Souza Neto, Computer Methods in Applied Mechanics and
Engineering, Vol 171 c 1999 Elsevier Science S.A.)
∆ε - contour levels:
p
0.117
0.050
0.010
M M
0.004
0.000
(a) (b)
Figure 8.20. Bending of a V-notched bar: (a) slip-line field of the upper bound solution by Green (1953);
(b) finite element result under collapse (increment 12): incremental accumulated plastic strain, ∆ε̄p .
(Reproduced with permission from A new computational model for Tresca plasticity at finite strains
with an optimal parametrization in the principal space, D Perić and EA de Souza Neto, Computer
Methods in Applied Mechanics and Engineering, Vol 171 c 1999 Elsevier Science S.A.)
illustrated in Figure 8.21. The width of the cantilever is assumed sufficiently large so that a
plane strain analysis is carried out. The cantilever is firmly supported (clamped) on one edge
and the applied load is a uniformly distributed shearing traction, of intensity S, acting on
the opposite edge. The cantilever is assumed to be made of a Tresca elastic-perfectly plastic
material with the same constants as in the previous example. Upper bound limit loads for
tapered cantilevers have been obtained analytically by Green (1954) by means of slip-line
field theory.¶ For the present geometry, the normalised upper bound for the shear traction
¶ Green’s limit loads are exact for taper angles θ ≤ 75o . His solution for θ > 75o , such as in the present example,
are only upper bounds. Nevertheless, due to experimental confirmation of the proposed slip-line fields, Green argues
that it is unlikely that his upper-bound solutions will be above the exact theoretical limit load.
346 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
L = 150 mm
θ = 80 o
Figure 8.21. End-loaded tapered cantilever. Problem definition and finite element mesh.
found by Green is
Su
≈ 0.775 (8.144)
c
where c is the shear strength given, for the Tresca model, by c = σy /2. In the finite element
analysis, the load is applied incrementally until collapse occurs. A diagram showing the
vertical deflection, u, of the mid-node of the free edge versus the applied force, S, obtained
in the finite element analysis is plotted in Figure 8.22. The corresponding collapse load (i.e.
the load above which convergence of equilibrium iterations cannot be attained for sufficiently
small increments) is
fe
Slim ≈ 0.786, (8.145)
which is about 1.4% above Green’s upper bound. The present limit load has been reached in
eight increments. The slip-line field proposed by Green (1954) is schematically illustrated in
Figure 8.23(a). The associated collapse mechanism is of the plastic hinge type. At collapse,
the portion to the right of the slip-line field rotates by sliding over the circular arc BC.
Figure 8.23(b) shows the incremental nodal displacements obtained in the last increment
(increment 8), when the cantilever is effectively collapsing. It can be seen that the finite
element solution has captured Green’s collapse mechanism quite accurately. In this plot, the
sizes of the vectors have been largely exaggerated and do not correspond to the norms of the
actual nodal displacements. The increment of accumulated plastic strain, ∆ε̄p , obtained in the
same load increment is illustrated in Figure 8.23(c). The contour plot shows that, at collapse,
the plastic process predicted by the finite element analysis is confined to an area which is in
very close agreement with Green’s slip-line field.
0.8
0.2
0
0 0.01 0.02 0.03 0.04
normalised edge deflection, u/h
present example, the same strip-footing problem is analysed but the soil is now modelled
as a Tresca, Mohr–Coulomb and Drucker–Prager material. The objective is to illustrate the
performance of the algorithms described earlier in the present chapter for these models. The
adopted common material parameters are
The dimensions, boundary conditions and the finite element mesh adopted are identical to
those of Example 7.5.4 (see Figure 7.24). The soil is assumed to be weightless. Note that
due to symmetry, only half of the problem is discretised. The mesh contains 135 eight-
noded quadrilaterals (four Gauss point quadrature) and a total of 446 nodes. The footing
is assumed to be rigid and perfectly smooth (no friction). Thus, the loading consists of the
prescribed vertical displacement, u, of the nodes under the footing as shown in Figure 7.24.
The horizontal displacement of these nodes is unconstrained.
Tresca
Firstly, the analysis is carried out using the Tresca model. The corresponding uniaxial yield
stress (required by the input data procedure for the Tresca material in HYPLAS) is σy = 2c =
980 kPa. A final settlement u/B = 0.002 is applied in 14 increments. The resulting load-
settlement curve is plotted in Figure 8.24. This curve and the corresponding limit load
fe
Plim
≈ 5.19, (8.146)
c
are practically identical to those obtained with the von Mises model in Example 7.5.4
(plotted in Figure 7.26, page 253), where the same material parameters have been used. The
348 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Green’s slip-line
A field
D
(a)
∆ε - contour levels:
p
(b) (c)
Figure 8.23. End-loaded tapered cantilever. Analytical solution: (a) slip-line field proposed by
Green (1954); finite element results under collapse (increment 8): (b) incremental nodal displacements;
(c) incremental accumulated plastic strain, ∆ε̄p .
numerically determined limit pressure is less than 1% above the slip-line theory solution
Plim
≈ 5.14. (8.147)
c
The failure mechanism captured in the finite element analysis with the Tresca model is in
agreement with that proposed by Hill (1950) for perfectly smooth footings (unconstrained
horizontal displacements under the footing). This mechanism, as obtained in the simulation
with the von Mises model carried out in the previous chapter, is illustrated in Figure 7.28.
Mohr–Coulomb
The analysis of the strip-footing problem with the Mohr–Coulomb material model is carried
out assuming a frictional angle
φ = 20o . (8.148)
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 349
3
TRESCA:
2 finite element results
slip-line theory limit load: 5.14
1
0
0 0.001 0.002
normalised settlement, u/B
• associative (φ = ψ = 20o );
Plim
≈ 14.8. (8.150)
c
Drucker–Prager
Three of the Drucker–Prager approximations to the Mohr–Coulomb criterion discussed in
Section 6.4.4 (and implemented in program HYPLAS) are used here in the analysis of the
footing:
16
MOHR–COULOMB:
8 associative
non-associative
Prandtl solution: 14.8
4
0
0 0.001 0.002 0.003
normalised settlement, u/B
Only the associative law (φ = ψ = 20o ) is considered. For all approximations, a total set-
tlement u/B = 0.004 is imposed gradually in 10 increments. The load-settlement diagrams
obtained are shown in Figure 8.26. The limit load predicted by the outer cone approximation
is well above Prandtl’s solution. The normalised limit pressure in this case is approximately
35.7 (about 140% above Prandtl’s solution). The normalised limit pressure predicted by the
inner cone is around 17.3 (about 17% above Prandtl’s solution). For the plane strain match
on the other hand, the results are in excellent agreement with Prandtl’s solution. In this case,
the predicted limit pressure is
fe
Plim /c ≈ 15.0, (8.151)
which is practically identical to the limit loads obtained with the Mohr–Coulomb model.
40
30 DRUCKER–PRAGER:
normalised pressure, P/c
outer cone
inner cone
plane strain match
20 Prandtl solution: 14.8
10
0
0 0.001 0.002 0.003 0.004 0.005
normalised settlement, u/B
determined (to any desired accuracy) by the finite difference method (the limit loads for
several frictional angles are tabulated in Chen, 1975). For the present frictional angle (φ =
20o ) the limit average pressure on the footing is
The corresponding slip-line net is schematically illustrated in Figure 8.27. The limit load
obtained in the finite element analysis is in close agreement with (8.153). The relative error
is less than 2%. The incremental nodal displacement vectors near the footing obtained in
increment 8 (where the collapse load has effectively been reached) are plotted in Figure 8.29.
The vector sizes have been amplified to ease visualisation. The failure mechanism captured
in the finite element analysis is in agreement with the slip-line field of Cox et al. (1961).
B =1m
Material properties
ν = 0.48
c = 490 kPa
φ = 20 o
soil 20 o (associative flow)
ψ=
25
normalised pressure, P/c
20
15
10 CIRCULAR FOOTING:
finite element results
Cox et al., 1961: 20.1
5
0
0 0.001 0.002
normalised settlement, u/B
β = 45 o
B A
Cohesion: c = 50 kPa
35 m
40 m
Please Wait..
30 m
75 m
reaches a critical limit. Here, h denotes the slope height, c is the cohesion and γ is the specific
weight of the soil
γ = ρ g, (8.155)
with ρ being the mass density and g the gravitational acceleration. For the present geometry
and material constants, the ratio N is
N = 4. (8.156)
The critical value of N , denoted Ns , is called the stability factor. The stability factor tabulated
by Chen (1975) for the angles β = 45o and φ = 20o , adopted in the present example, is
Ns ≈ 16.18.
354 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
0
0 1 2 3 4
settlement at point A (m)
incremental
nodal displacements
Please Wait..
∆ε p - contour levels:
Please Wait..
1.3050
0.9788
0.6525
0.3263
0.0000
Figure 8.33. Slope stability. Increment of accumulated plastic strain and displacement at collapse
(increment 6).
COMPUTATIONS WITH OTHER BASIC PLASTICITY MODELS 355
This value corresponds to a collapse mechanism (Figure 8.30) in which the portion ABC
of the embankment slides along a logarithmic spiral discontinuity line (the dotted line) that
passes through the toe (point C). A safety factor based on limit analysis can be defined as
αlim = Ns /N . For the present dimensions and material properties it gives
ḡ = α g (8.158)
denoting the applied gravity acceleration, the loading process starts from α = 0 (completely
unloaded). The load factor, α, is then increased gradually until collapse occurs. The value of
α at collapse is the safety factor.
The adopted mesh, shown in Figure 8.31, consists of 221 eight-noded quadrilaterals (2 ×
2 integration quadrature) with a total of 718 nodes. The discretised domain is made large
enough to avoid interference of the collapse mechanism with the boundaries of the finite
element mesh. In the finite element analysis, a total of seven load increments are applied. The
settlement at the top corner of the slope (point A of Figure 8.30) is plotted in Figure 8.32
against the load factor α. The limit value of α, i.e. the safety factor, found in the incremental
finite element analysis is
lim ≈ 4.2.
αfe (8.159)
It is about 3.8% above the limit analysis solution. The contour plot of incremental accumu-
lated plastic strain, ∆ε̄p , as well as the incremental nodal displacements obtained at increment
6 are depicted in Figure 8.33. The slope is effectively collapsing during this increment. The
failure mechanism captured by the finite element simulation is in good agreement with the
logarithmic spiral slide line referred to above.
9 PLANE STRESS PLASTICITY
I N the numerical solution of elastoplastic initial value problems addressed in the previous
two chapters, attention has been focused on generic three-dimensional states of stress
and strain – conditions under which the elastoplastic models were formulated in Chapter 6.
In spite of the generality of the algorithmic formulations presented in Chapters 7 and 8,
the actual computational implementation of the elastic predictor/return-mapping algorithms
shown in the corresponding subroutines of program HYPLAS has been restricted to states of
plane strain and axisymmetric deformation. From the theoretical point of view, the classes of
problem with such constraints on the possible strains are those in which certain components
of the strain tensor vanish during the given history of strains in the general elastoplastic initial
value problem – stated in Problem 7.1, page 193. Clearly, having formulated the generic
three-dimensional integration algorithms (as we did in Chapters 7 and 8), particularisation for
states of constrained strain (used in the HYPLAS subroutines presented) is trivially obtained
simply by eliminating the relevant strain components from the formulation. Note that, if the
corresponding stress component is generally non-vanishing (such as the normal out-of-plane
stress in plane strain problems), then it will be determined naturally as an outcome of the
numerical integration algorithm.
The conditions addressed in the present chapter are quite different from those discussed
in Chapters 7 and 8. Here, we are concerned with plane stress problems in which certain
components of stress, rather than strain, are constrained to be zero. In spite of its triviality
within the realm of linear elasticity, the treatment of the plane stress constraint in elastoplastic
problems requires further consideration and justifies a chapter of its own. The main difficulty
lies in the fact that as some components of the stress tensor are now prescribed, the original
statement of the elastoplastic initial value problem – Problem 7.1 (page 193), where all stress
components are unknown – no longer applies and needs to be reformulated. As a result, the
three-dimensional numerical integration algorithms of Chapters 7 and 8, which have been
developed specifically to solve Problem 7.1, cannot be used in general without modifications.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
358 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
σ12 σ11
σ12
e3
σ22
e2
t
e1
where
3K − 2G
α= . (9.5)
3K + 4G
PLANE STRESS PLASTICITY 359
This relationship is derived directly from the three-dimensional elastic law as follows.
Firstly, in the three-dimensional law
K + 43 G K − 23 G K − 23 G 0 0 0
σ
K − 2 G K + 4 G
ε11
0 ε
11
K− 2
3G 0 0
σ22
3 3
22
σ33 K − 23 G K − 23 G K+ 4
3G 0 0 0 ε33
=
, (9.6)
σ12
0 0 0 G 0 02ε12
2ε23
σ
23
0 0 0 0 G 0
σ13 2ε13
0 0 0 0 0 G
we prescribe the in-plane strain components, ε11 , ε22 and ε12 together with the (zero) out-
of-plane stress components, σ33 , σ23 and σ13 . This gives the system
∗ K + 43 G K − 23 G K − 23 G 0 0 0
σ
11
ε11
∗ K − 3 G K + 3 G K − 3 G 0 0 0
2 4 2
σ22
ε22
K − 2 G K − 2 G K + 4 G 0 0 0 ∗
0 ε
=
3 3 3 33
, (9.7)
G 0 0
∗
σ12
0 0 0
2ε12
0 2ε23
∗
0 0 0 0 G 0
∗
0 2ε13
0 0 0 0 0 G
where the star superscript has been used to point out the unknown quantities. Solution of the
above system yields the plane stress elastic law together with the well-known expressions for
the out-of-plane strains
ε13 = 0, ε23 = 0,
(9.8)
3K − 2G ν
ε33 = − (ε11 + ε22 ) = − (ε11 + ε22 ).
3K + 4G 1−ν
In summary, the constitutive equations of plane stress elasticity are obtained by prescribing
the in-plane strains and enforcing the plane stress constraint on the three-dimensional law,
and then obtaining, as a result, the in-plane stresses and out-of-plane strains.
find the functions εe (t), α(t) and γ̇(t), for the elastic strain, internal variable set and
plastic multiplier that satisfy the reduced general elastoplastic three-dimensional constitutive
equations
ε̇(t) = ε̇e (t) − γ̇(t) N(σ(t), A(t))
(9.9)
α̇(t) = γ̇(t) H(σ(t), A(t))
and
γ̇(t) ≥ 0, Φ(σ(t), A(t)) ≤ 0, γ̇(t) Φ(σ(t), A(t)) = 0, (9.10)
with
∂ψ ∂ψ
σ(t) = ρ̄ e
, A(t) = ρ̄ , (9.11)
∂ε ∂α
and, in addition, satisfy the plane stress constraint
(a) direct inclusion of the plane stress constraint into the three-dimensional elastic pre-
dictor and plastic corrector algorithm equations (7.21), (7.22) and (7.25) (refer to
page 196) applied at the Gauss point level. This approach can also be implemented by
means of a nested Newton return-mapping iteration for plane stress enforcement. The
nested iteration procedure is implemented in program HYPLAS for the Drucker–Prager
model;
(b) use of the standard three-dimensional return mapping at the Gauss point level with the
plane stress condition added as a structural constraint at the global structural level;
(c) use of plane stress-projected constitutive equations where, similarly to the procedure of
item (a), the plane stress constraint is enforced at the Gauss point level. This strategy
has been implemented in program HYPLAS for the von Mises isotropically hardening
model.
Remark 9.1. Approaches (a), (b) and (c) are equivalent in that, provided the same
numerical scheme is used in the discretisation of the elastoplastic evolution equations, the
three methodologies produce identical incremental plane stress constitutive equations. The
essential difference between distinct approaches in this case lies in the way that the plane
stress constraint (9.12) imposed upon the three-dimensional model is enforced.
PLANE STRESS PLASTICITY 361
Having the general three-dimensional plasticity model of Chapter 6 and the implicit
elastic predictor/return-mapping scheme of Section 7.2.3 as the underlying model/algorithm,
approaches (a) and (b) are discussed below in Sections 9.2, and 9.3, respectively. In
Section 9.2, an outline of the FORTRAN code of the von Mises model implementation with
nested iterations is shown. The third strategy is described in Section 9.4 where the von Mises
model is used as an example for completeness. The complete description of the computational
implementation, including the FORTRAN code, is also given in Section 9.4.
to the general system of return-mapping equations (7.25) (page 196) and letting the out-of-
plane components of the elastic trial strain tensor become unknowns of the new system. After
performing a plane stress-constrained elastic trial state evaluation,† plastic consistency is
checked in the usual way. If plastic admissibility is violated, i.e. if
Φ(σtrial
n+1 , An ) > 0, (9.14)
then the return-mapping procedure has to be applied. The return mapping now requires the
solution of the augmented algebraic system
trial
εen+1 = εen+1 − ∆γ Nn+1
αn+1 = αn + ∆γ Hn+1
Φ(σn+1 , An+1 ) = 0
(9.15)
σ13 (εen+1 , αn+1 ) = 0
σ23 (εen+1 , αn+1 ) = 0
σ33 (εen+1 , αn+1 ) = 0
for the variables εen+1 , αn+1 , ∆γ and the elastic trial components
As in the standard non-constrained case, the solution of the above system can be undertaken
by the Newton–Raphson algorithm.
† Note that if the elastic behaviour is nonlinear, the plane stress-constrained elastic equations may not be obtained
which ensures that σ13 and σ23 vanish. In this case the only extra equation to be added to the
general three-dimensional return mapping is‡
During a typical equilibrium iteration, the in-plane displacements are prescribed and so is the
in-plane strain array εn+1 . Instead of giving εn+1 (or εen+1
trial
) as the input of an augmented
algebraic system (as described in the above), we proceed as follows. Firstly, we define
some initial guess for the unknown (εe33 )trial
n+1 . One possible guess can be the previously
(equilibrium) converged out-of-plane elastic strain, i.e. we can set
Here the subscript n + 1 has been dropped for notational convenience. Next, we use the
augmented strain array
T
[εe trial εe33trial ]T
as the input of the integration algorithm for the axisymmetric case.§ Note that an out-of-plane
strain component must be prescribed for the axisymmetric algorithm.
‡ The original algorithm proposed by Aravas (1987) was derived under such isotropy conditions, having Gurson’s
void growth model (described in Section 12.5, from page 496) as the underlying material.
§ Axisymmetric implementations of the general integration algorithm have been included in the corresponding
[σTn+1 σ33 ]T .
If σ33 = 0 (or, in computational terms, |σ33 | ≤ tol ) then the guess εe33trial indeed solves the
plane stress problem, and the solution obtained by the axisymmetric algorithm is the one we
are looking for. Otherwise, we apply a Newton–Raphson correction to obtain another guess
σ33
εe33trial := εe33trial − , (9.20)
D22
where D22 is the component of the axisymmetric consistent tangent matrix
e trial
dσ11 dε11
e trial
dσ22 D11 D12
dε22
=
dσ12 2dεe trial , (9.21)
12
dσ33 D21 D22 dεe33trial
that relates the transverse stress and strain components.¶ We repeat this process until we find
the elastic trial thickness strain εe33trial that, together with the in-plane strain kinematically
prescribed by the global equilibrium iteration, results in zero (or sufficiently small) transverse
stress upon application of the axisymmetric algorithm. The resulting overall algorithm is
shown in Box 9.1 in pseudo-code format. An outline of its specialisation to the von Mises
model is presented in the next section. Its complete computational implementation for the
Drucker–Prager model is included in program HYPLAS. The associated subroutine (not listed
in this text) is named SUDPPN (State Update procedure for the Drucker–Prager model in Plane
stress with Nested iterations).
Remark 9.2. The above methodology involves two nested Newton–Raphson iteration loops:
one (outer) global equilibrium loop having a nested (inner) iteration loop to enforce the
plane stress constraint with εe33trial as the unknown. Thus, for each global equilibrium loop, a
number of iterations will be required in each Gauss point to ensure that the final stress is plane
(to within a prescribed numerical tolerance). Since the axisymmetric integration algorithm is
applied in each iteration of the inner loop, the present procedure is computationally expensive.
It is important to remember, however, that the cost of the calculations carried out at the
Gauss point level increases approximately linearly with the problem size, whereas the cost
of the solution of the global linearised problem increases at a much higher nonlinear rate. In
other words, the relative cost of the Gauss point computations decreases considerably as the
problem size increases and the present procedure can be used efficiently in large problems.
Remark 9.3. The solution obtained by the nested iteration is identical to that obtained by
the previously discussed strategy based on the augmented return-mapping system. Its main
advantage over the augmented system procedure lies in its simplicity. Essentially, for any
¶ The implementation of the axisymmetric consistent tangent for the material models of Chapters 7 and 8 is also
Box 9.1. Plane stress constraint at the Gauss point level with nested iterations.
(i) Set initial guess for the elastic trial thickness strain
εe33trial := (εe33 )n
[σT σ33 ]T
(iv) Compute component D22 of the axisymmetric consistent tangent matrix (see expres-
sion (9.21))
(v) Apply Newton–Raphson correction to the thickness trial strain
σ33
εe33trial := εe33trial −
D22
material model, the nested iteration approach allows a straightforward adaptation of existing
general procedures (which include the three-dimensional or axisymmetric implementations)
to cope with the plane stress problem, without the need for modification of the original
integration algorithms.
Remark 9.4. The present strategy is compatible with material models other than elasto-
plastic, such as generic nonlinear elastic materials. Note that for a linear elastic model, for
instance, the nested iteration loop produces the solution that satisfies the standard plane
equilibrium together with the trivial plane stress linear elastic law (9.3, 9.4). This is what
happens before plastic yielding in analyses involving linear elasticplastic materials such as
the von Mises model.
which renders
−1
dεe33trial = D21 dεe trial . (9.23)
D22
Substitution of the above expression into (9.21) results in the following consistent tangent
relation between the in-plane stress and strain components
dσn+1 1
trial
= D11 − D12 D21 . (9.24)
dεen+1 D22
Its incorporation into HYPLAS can be left as an exercise. We remark that the full implemen-
tation of a tangent matrix consistent with the nested iteration approach is already available
in HYPLAS for the Drucker–Prager model. The associated FORTRAN subroutine (not listed
here) is named CTDPPN (Consistent Tangent for the Drucker–Prager model in Plane stress
with Nested iterations).
PLANE STRESS PLASTICITY 367
where B is the in-plane symmetric gradient operator (with only the three rows associated with
ext
the in-plane strain components ε11 , ε22 and ε12 ), fn+1 is the applied external force vector at
tn+1 , which contains only in-plane forces, and σ is the array of in-plane stress components
trial e trial e trial e trial
σ̂11 (εen+1 , ε13 , ε23 , ε33 )
e trial e trial e trial e trial
σn+1 ≡ σ̂12 (εn+1 , ε13 , ε23 , ε33 ), (9.26)
trial e trial e trial e trial
σ̂22 (εen+1 , ε13 , ε23 , ε33 )
where
εen+1
trial
≡ εen + B ∆u = [ εe11trial εe22trial εe12trial ]T , (9.27)
with ∆u denoting the array of incremental nodal displacements.
∆u = un+1 − un . (9.28)
Here and in what follows, σ̂ij (with the superimposed hat) denote the algorithmic constitutive
functionals for the stress components defined by the three-dimensional elastoplastic integra-
tion algorithm. Note that in the plane strain case, the three-dimensional algorithm is used
with zero prescribed out-of-plane strains.
In this case, the dependence of the in-plane stress components on the transverse shear strain
components is removed from (9.26). Under such a condition, the present strategy consists
For notational convenience, the functional dependence of the stress components on the state variables at t is
n
not explicitly indicated.
368 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
simply in adding the zero transverse stress constraint to the finite element equilibrium
equation (9.25). The resulting residual equations are accordingly redefined as
BT σn+1 dΩ − fn+1 ext
=0
Ω (9.29)
trial e trial
σ̂33 (εen+1 , ε33 ) = 0 in each Gauss point,
where the unknowns are the nodal incremental displacements ∆u (which enter the equations
through the dependence of εen+1 trial
upon it) and the thickness strain εe33trial in each Gauss point.
In the above, σ̂33 is the algorithmic constitutive function for the transverse stress component
resulting from the three-dimensional integration algorithm.
Box 9.2. Plane stress constraint at the structural level. The modified global
equilibrium iteration loop.
δu := −[K∗ ]−1 r∗ ; u := u + δu
δε := B δu(e) ; εen+1
trial
:= εen+1
trial
+ δε
IF ||r∗ || < tol AND |σ33 | < tol for all Gauss points of
the structure, THEN EXIT
ELSE GOTO (i)
includes combined isotropic and Prager’s linear kinematic hardening whereas the derivation
presented by Jetteur is restricted to purely isotropic hardening. Another application of the
projected equations concept is reported by Ramm and Matzenmiller (1987) in the context of
large deformation shell analysis. The equations and algorithms derived by these authors are
analogous to those presented here but account for transverse shear stresses and strains, which
become important in the analysis of thick shells.
ε̇ = ε̇e + ε̇p
σ = De : εe
Φ = 3J2 (s) − σy (ε̄p )
% (9.35)
p ∂Φ 3 s
ε̇ = γ̇ = γ̇
∂σ 2 s
ε̄˙p = γ̇
γ̇ ≥ 0, Φ ≤ 0, γ̇ Φ = 0.
Under plane stress, the above equations are complemented with the constraints
As s13 = s23 = 0 under a plane stress state, it also follows from (9.35)4 that
The above relations completely define the out-of-plane strain components as functions of the
in-plane values. Thus, under plane stress, the history of the out-of-plane strain components is
372 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
automatically prescribed as
ε13 (t) = εe13 (t) + εp13 (t) ≡ 0
which holds under plane stress condition. This identity is obtained by introducing the plane
stress constraint into the definition
√
s ≡ s : s; s ≡ σ − 13 tr[σ] I. (9.43)
Matrix notation
A conveniently compact representation of the plane stress-projected von Mises equations can
be obtained by making use of the matrix notation
σ = [σ11 σ22 σ12 ]T , s = [s11 s22 s12 ]T
expressed as
s = Pσ, (9.45)
where P is the matrix defined by
2 −1 0
1
−1
P≡ 2 0
. (9.46)
3
0 0 6
With the above matrix/vector notation at hand, the plane stress-projected von Mises
equations (9.41) can be equivalently written in compact form as
ε̇ = ε̇e + ε̇p
σ = De εe
2 σ Pσ − σy (ε̄ )
3 T p
Φ =
% (9.47)
∂Φ 3 Pσ
ε̇ p
√
= γ̇ = γ̇
∂σ 2 σT Pσ
ε̄˙p = γ̇
γ̇ ≥ 0, Φ ≤ 0, γ̇ Φ = 0,
where De is the plane stress elasticity matrix. For the derivation of the integration algorithm
to be described in the next section, it is more convenient to use the squared form of the yield
function which, in the present case, can be handled more easily. Then, in the set (9.47) of the
evolution equations, we replace (9.47)3, (9.47)4 and (9.47)5, respectively, with
Φ = 12 σT Pσ − 13 σy2 (ε̄p )
∂Φ
ε̇p = γ̇ = γ̇ Pσ (9.48)
∂σ
ε̄˙p = γ̇ 23 σT Pσ.
The complete set of equations of the plane stress-projected von Mises model with
isotropic hardening is conveniently grouped in Box 9.3. The corresponding implicit elastic
predictor/return-mapping algorithm is derived subsequently.
Box 9.3. Plane stress-projected von Mises model with isotropic hardening.
ε̇ = ε̇e + ε̇p
ε ≡ [ε11 ε22 2ε12 ]T , εe ≡ [εe11 εe22 2εe12 ]T , εp ≡ [εp11 εp22 2εp12 ]T
2. Elastic law
σ = De εe
1 ν 0
E ν 1 0
σ ≡ [σ11 σ22 σ12 ]T , De ≡
1 − ν2 0 1−ν
0
2
3. Yield function definition
2 −1 0
Φ = 12 σT Pσ − 13 σy2 (ε̄p ); P≡ 1 −1 2 0
3
0 0 6
6. Loading/unloading criterion
γ̇ ≥ 0, Φ ≤ 0, γ̇ Φ = 0
predictor state
εen+1
trial
= εen + ∆ε
n+1 = D εn+1
σtrial e e trial
(9.49)
trial
ε̄pn+1 = ε̄pn ,
where ∆ε is the given (in-plane) strain increment associated with the interval [tn , tn+1 ]
∆ε = εen+1 − εen . (9.50)
The next step is to check for plastic admissibility of the elastic trial state. We then compute
1 2 p trial
n+1 ) Pσn+1 − 3 σy (ε̄n+1 ).
Φtrial = 12 (σtrial T trial
(9.51)
If the elastic trial state is admissible, i.e. if
Φtrial ≤ 0, (9.52)
PLANE STRESS PLASTICITY 375
(·)n+1 = (·)trial
n+1 . (9.53)
2 σn+1 Pσn+1
1 T
− 13 σy2 (ε̄pn+1 ) = 0,
for εen+1 , ε̄pn+1 and ∆γ where σn+1 is a function of εen+1 defined by the elastic law.
As in the three-dimensional case discussed in Section 7.3.2 (page 217), it is also possible
to reduce the number of unknowns of the plane stress-projected return-mapping equations
for the von Mises model. The five-variable return-mapping system (9.54) can be reduced to
a single scalar nonlinear equation having the incremental plastic multiplier as the unknown.
To this end, we substitute (9.54)2 into (9.54)3 and rearrange (9.54)1 using the inverse elastic
law. The original return-mapping system reduces to
+ , (9.55)
2 σn+1 Pσn+1 −
1 T 1 2 2
3 σy ε̄pn + ∆γ T
3 (σn+1 ) Pσn+1 = 0,
where the unknowns now are the stress array, σn+1 , and the plastic multiplier, ∆γ, and C is
the inverse of the elastic matrix
C ≡ (De )−1 . (9.56)
Finally, by substituting (9.55)1 into the consistency condition (9.55)2, the return mapping for
the plane stress von Mises model is reduced to the following scalar nonlinear equation having
∆γ as the only unknown
+ ,
Φ̃(∆γ) ≡ 12 ξ(∆γ) − 13 σy2 ε̄pn + ∆γ 23 ξ(∆γ) = 0. (9.57)
with
A(∆γ) ≡ [C + ∆γ P]−1 C. (9.59)
Thus, the return mapping for the plane stress von Mises model is carried out as follows.
Firstly, we solve the consistency equation (9.57) using the Newton–Raphson algorithm. With
376 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Remark 9.8. The plane stress-projected return-mapping equation (9.57) is always nonlinear
in ∆γ, regardless of the prescribed hardening function σy . This is in contrast with its three-
dimensional counterpart (see equation (7.91), page 219, and the comments made at the
beginning of Section 7.3.4, page 223) whose only source of nonlinearity is the hardening
curve and which becomes linear for perfectly plastic and linearly hardening models.
The plane stress-projected elastic predictor/return-mapping algorithm for the von Mises
model with isotropic hardening is summarised in Boxes 9.4 and 9.5 in pseudo-code format.
In program HYPLAS, the procedure is implemented in subroutine SUVMPS (State Update
procedure for the Von Mises model in Plane Stress), whose FORTRAN code is shown
in Section 9.4.3. The solution of the plane stress-projected consistency equation (9.57) is
undertaken by the Newton–Raphson algorithm. The Newton–Raphson procedure is shown
in Box 9.5. Its implementation follows the simplification obtained below by exploiting the
properties of matrices P and De .
The reader might have realised already that the Newton–Raphson algorithm for solution
of the consistency equation (9.57) is potentially cumbersome. This is because the scalar
function ξ defined in (9.58) depends on a matrix A(∆γ) which, in turn, is a function whose
definition (9.59) involves the sum, inversion and multiplication of other matrices. Fortunately,
in the present case where the elastic behaviour is isotropic, the explicit expression for (9.57)
can be significantly simplified. The key point to be observed is the fact that matrices P and
De share the same eigenvectors so that they both have diagonal representation on the same
basis (as do C and A). By applying the orthogonal transformation
1 1
√ √ 0
2 2
Q=
− √
1 1
√
0 (9.62)
2 2
0 0 1
PLANE STRESS PLASTICITY 377
(i) Elastic predictor. Given the in-plane incremental strains, ∆ε, and the state variables at
tn , evaluate the elastic trial state
εen+1
trial
:= εen + ∆ε
ε̄pn+1
trial
:= ε̄pn
σtrial
n+1 := D εn+1
e e trial
Φ̃(∆γ) = 0
for ∆γ using the Newton–Raphson method – GOTO Box 9.5 – and update the state
variables
σn+1 := A(∆γ) σtrial
n+1
εen+1 := C σn+1
ε̄pn+1 := ε̄pn + ∆γ 2
3
ξ(∆γ)
and
E
1 − ν 0 0
De∗ ≡ QDe QT = 0 2G 0 . (9.64)
0 0 G
378 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
On the same basis, the matrix A(∆γ) has the diagonal representation
where C∗ = [De ∗ ]−1 . The corresponding representation of the elastic trial stress array reads
1 trial trial
√ (σ
2 11 + σ22 )
∗ 1 trial
σtrial √
n+1 ≡ Q σn+1 = 2 (σ22 − σ11 ).
trial trial
(9.66)
trial
σ12
With the above transformed variables, expression (9.58) can be written in the simpler form
with
3(1 − ν) 1
A∗11 = , A∗22 = , A∗33 = A∗22 . (9.70)
3(1 − ν) + E ∆γ 1 + 2G ∆γ
Φ̃
∆γ := ∆γ − (new guess for ∆γ)
Φ̃
(iii) Check for convergence
the description of nonlinear isotropic hardening, which uses a piecewise linear hardening
curve, is adopted here. Again, the symbolic names of most variables resemble those of
the corresponding name given in the text. This should allow readers easily to correlate the
FORTRAN instructions with the corresponding expressions of Boxes 9.4 and 9.5, where the
plane stress integration algorithm is summarised. The complete list of the FORTRAN source
code of SUVMPS is given below.
380 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
1 SUBROUTINE SUVMPS
2 1( DGAMA ,IPROPS ,LALGVA ,NTYPE ,RPROPS ,
3 2 RSTAVA ,STRAT ,STRES )
4 IMPLICIT DOUBLE PRECISION (A-H,O-Z)
5 PARAMETER( IPHARD=4 ,MSTRE=4 ,NSTRE=3 )
6 LOGICAL IFPLAS, LALGVA(2), SUFAIL
7 DIMENSION
8 1 IPROPS(*) ,RPROPS(*) ,RSTAVA(MSTRE+1) ,
9 2 STRAT(MSTRE) ,STRES(MSTRE)
10 DIMENSION
11 1 EET(MSTRE) ,STREST(NSTRE)
12 DATA
13 1 R0 ,RP5 ,R1 ,R2 ,R3 ,R4 ,R6 ,TOL /
14 2 0.0D0,0.5D0,1.0D0,2.0D0,3.0D0,4.0D0,6.0D0,1.D-08/
15 DATA MXITER / 50 /
16 C***********************************************************************
17 C STATE UPDATE PROCEDURE FOR THE VON MISES ELASTO-PLASTIC MODEL WITH
18 C NON-LINEAR (PIECEWISE LINEAR) ISOTROPIC HARDENING IN PLANE STRESS:
19 C IMPLICIT PLANE STRESS-PROJECTED ELASTIC PREDICTOR/RETURN MAPPING
20 C ALGORITHM (BOXES 9.4-5).
21 C***********************************************************************
22 C Stop program if not plane stress
23 IF(NTYPE.NE.1)CALL ERRPRT(’EI0031’)
24 C Initialise some algorithmic and internal variables
25 DGAMA=R0
26 IFPLAS=.FALSE.
27 SUFAIL=.FALSE.
28 C...set previously (equilibrium) converged accumulated plastic strain
29 EPBARN=RSTAVA(MSTRE+1)
30 C Set some material properties
31 YOUNG=RPROPS(2)
32 POISS=RPROPS(3)
33 NHARD=IPROPS(3)
34 C Shear and bulk moduli and other necessary constants
35 GMODU=YOUNG/(R2*(R1+POISS))
36 BULK=YOUNG/(R3*(R1-R2*POISS))
37 R2G=R2*GMODU
38 R4G=R4*GMODU
39 R1D3=R1/R3
40 R1D6=R1/R6
41 R2D3=R2*R1D3
42 SQR2D3=SQRT(R2D3)
43 R4GD3=R4G*R1D3
44 C Elastic predictor: Compute elastic trial state
45 C ----------------------------------------------
46 C Volumetric strain
47 FACTOR=R2G/(BULK+R4GD3)
48 EEV=(STRAT(1)+STRAT(2))*FACTOR
49 C Elastic trial deviatoric strain
50 EEVD3=EEV/R3
51 EET(1)=STRAT(1)-EEVD3
52 EET(2)=STRAT(2)-EEVD3
53 C Convert engineering shear component into physical component
54 EET(3)=STRAT(3)*RP5
55 C Elastic trial stress components
56 PT=BULK*EEV
57 STREST(1)=R2G*EET(1)+PT
58 STREST(2)=R2G*EET(2)+PT
59 STREST(3)=R2G*EET(3)
60 C Compute yield function value at trial state
61 A1=(STREST(1)+STREST(2))*(STREST(1)+STREST(2))
PLANE STRESS PLASTICITY 381
62 A2=(STREST(2)-STREST(1))*(STREST(2)-STREST(1))
63 A3=STREST(3)*STREST(3)
64 XI=R1D6*A1+RP5*A2+R2*A3
65 SIGMAY=PLFUN(EPBARN,NHARD,RPROPS(IPHARD))
66 C...yield function
67 PHI=RP5*XI-R1D3*SIGMAY*SIGMAY
68 C Check for plastic admissibility
69 C -------------------------------
70 IF(PHI/SIGMAY.GT.TOL)THEN
71 C Plastic step: Apply return mapping - use Newton-Raphson algorithm
72 C to solve the plane stress-projected return mapping
73 C equation for the plastic multiplier (Box 9.5)
74 C -----------------------------------------------------------------
75 IFPLAS=.TRUE.
76 EPBAR=EPBARN
77 SQRTXI=SQRT(XI)
78 B1=R1
79 B2=R1
80 FMODU=YOUNG/(R3*(R1-POISS))
81 DO 10 NRITER=1,MXITER
82 C Compute residual derivative
83 HSLOPE=DPLFUN(EPBAR,NHARD,RPROPS(IPHARD))
84 DXI=-A1*FMODU/(R3*B1*B1*B1)-R2G*(A2+R4*A3)/(B2*B2*B2)
85 HBAR=R2*SIGMAY*HSLOPE*SQR2D3*(SQRTXI+DGAMA*DXI/(R2*SQRTXI))
86 DPHI=RP5*DXI-R1D3*HBAR
87 C Compute Newton-Raphson increment and update equation variable DGAMA
88 DGAMA=DGAMA-PHI/DPHI
89 C Compute new residual (yield function value)
90 B1=R1+FMODU*DGAMA
91 B2=R1+R2G*DGAMA
92 XI=R1D6*A1/(B1*B1)+(RP5*A2+R2*A3)/(B2*B2)
93 SQRTXI=SQRT(XI)
94 EPBAR=EPBARN+DGAMA*SQR2D3*SQRTXI
95 SIGMAY=PLFUN(EPBAR,NHARD,RPROPS(IPHARD))
96 PHI=RP5*XI-R1D3*SIGMAY*SIGMAY
97 C Check for convergence
98 RESNOR=ABS(PHI/SIGMAY)
99 IF(RESNOR.LE.TOL)THEN
100 C update accumulated plastic strain
101 RSTAVA(MSTRE+1)=EPBAR
102 C update stress components: sigma := A sigma^trial
103 ASTAR1=R3*(R1-POISS)/(R3*(R1-POISS)+YOUNG*DGAMA)
104 ASTAR2=R1/(R1+R2G*DGAMA)
105 A11=RP5*(ASTAR1+ASTAR2)
106 A22=A11
107 A12=RP5*(ASTAR1-ASTAR2)
108 A21=A12
109 A33=ASTAR2
110 STRES(1)=A11*STREST(1)+A12*STREST(2)
111 STRES(2)=A21*STREST(1)+A22*STREST(2)
112 STRES(3)=A33*STREST(3)
113 C compute corresponding elastic (engineering) strain components
114 FACTG=R1/R2G
115 P=R1D3*(STRES(1)+STRES(2))
116 EEV=P/BULK
117 EEVD3=R1D3*EEV
118 RSTAVA(1)=FACTG*(R2D3*STRES(1)-R1D3*STRES(2))+EEVD3
119 RSTAVA(2)=FACTG*(R2D3*STRES(2)-R1D3*STRES(1))+EEVD3
120 RSTAVA(3)=FACTG*STRES(3)*R2
121 RSTAVA(4)=-POISS/(R1-POISS)*(RSTAVA(1)+RSTAVA(2))
122 GOTO 999
382 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
123 ENDIF
124 10 CONTINUE
125 C reset failure flag and print warning message if N-R algorithm fails
126 SUFAIL=.TRUE.
127 CALL ERRPRT(’WE0013’)
128 ELSE
129 C Elastic step: Update stress using linear elastic law
130 C ----------------------------------------------------
131 STRES(1)=STREST(1)
132 STRES(2)=STREST(2)
133 STRES(3)=STREST(3)
134 C elastic engineering strain
135 RSTAVA(1)=STRAT(1)
136 RSTAVA(2)=STRAT(2)
137 RSTAVA(3)=STRAT(3)
138 RSTAVA(4)=-POISS/(R1-POISS)*(STRAT(1)+STRAT(2))
139 ENDIF
140 999 CONTINUE
141 C Update some algorithmic variables before exit
142 LALGVA(1)=IFPLAS
143 LALGVA(2)=SUFAIL
144 RETURN
145 END
dσn+1 dσn+1
Dep ≡ = trial , (9.71)
dεn+1 dεn+1
where σn+1 is the outcome of the plane stress-projected return-mapping algorithm. The
explicit expression for Dep is derived analogously to its three-dimensional counterpart
discussed in Section 7.4.2 (page 232). We start by differentiating (9.54)1 which, together
with the elastic law, gives
By using the elastic law and the above definition of matrix E, the scalar ξ defined by (9.58)
can be equivalently written as
n+1 ) E P E εn+1 .
ξ = (εtrial T trial
(9.76)
Substitution of the above formula into (9.75) followed by the substitution of the resulting
expression into (9.72) and straightforward algebraic manipulations gives
The explicit expression for the matrix form of the elastoplastic tangent operator consistent
with the von Mises plane stress-projected return mapping is then given by
The step-by-step procedure for evaluation of the above elastoplastic tangent is summarised
in Box 9.6 in pseudo-code format. Its computation in program HYPLAS is carried out in
subroutine CTVMPS (Consistent Tangent operator for the Von Mises model in Plane Stress).
This subroutine is described in the next section.
Box 9.6. Computation of the elastoplastic tangent operator for the plane stress-
projected von Mises model.
(i) Given σn+1 , ε̄pn+1 and ∆γ (outcome of the return mapping of Box 9.4), compute
ξ := σTn+1 Pσn+1
dσy
H := p (hardening slope)
dε̄ ε̄p
n+1
E := [C + ∆γ P]−1
n := EPσn+1
1
α :=
σTn+1 Pn + 2ξH
3−2H ∆γ
Dep := E − α n ⊗ n
with ∗
∗ 3E ∗ 2G ∗ E22
E11 = , E22 = , E33 = . (9.82)
3(1 − ν) + E ∆γ 1 + 2G ∆γ 2
A similar expression is obtained for the product EP. This representation is exploited in
subroutine CTVMPS, whose FORTRAN source code is listed below.
1 SUBROUTINE CTVMPS
2 1( DGAMA ,DMATX ,EPFLAG ,IPROPS ,NTYPE ,
3 2 RPROPS ,RSTAVA ,STRES )
4 IMPLICIT DOUBLE PRECISION (A-H,O-Z)
5 PARAMETER(IPHARD=4 ,MSTRE=4)
6 LOGICAL EPFLAG
7 C Array arguments
8 DIMENSION
9 1 DMATX(MSTRE,MSTRE),IPROPS(*) ,RPROPS(*) ,
10 2 RSTAVA(MSTRE+1) ,STRES(MSTRE)
11 C Local arrays
12 DIMENSION
13 1 FOID(MSTRE,MSTRE) ,SOID(MSTRE) ,VECN(3)
PLANE STRESS PLASTICITY 385
14 DATA
15 1 FOID(1,1),FOID(1,2),FOID(1,3),FOID(1,4)/
16 2 1.0D0 ,0.0D0 ,0.0D0 ,0.0D0 /
17 3 FOID(2,1),FOID(2,2),FOID(2,3),FOID(2,4)/
18 4 0.0D0 ,1.0D0 ,0.0D0 ,0.0D0 /
19 5 FOID(3,1),FOID(3,2),FOID(3,3),FOID(3,4)/
20 6 0.0D0 ,0.0D0 ,0.5D0 ,0.0D0 /
21 7 FOID(4,1),FOID(4,2),FOID(4,3),FOID(4,4)/
22 8 0.0D0 ,0.0D0 ,0.0D0 ,1.0D0 /
23 DATA
24 1 SOID(1) ,SOID(2) ,SOID(3) ,SOID(4) /
25 2 1.0D0 ,1.0D0 ,0.0D0 ,1.0D0 /
26 DATA
27 1 RP5 ,R1 ,R2 ,R3 ,R4 /
28 2 0.5D0,1.0D0,2.0D0,3.0D0,4.0D0/
29 C***********************************************************************
30 C COMPUTATION OF THE CONSISTENT TANGENT MODULUS FOR VON MISES TYPE
31 C ELASTO-PLASTIC MATERIAL WITH PIECE-WISE LINEAR ISOTROPIC HARDENING.
32 C PLANE STRESS IMPLEMENTATION ONLY.
33 C***********************************************************************
34 C Stops program if not plane stress
35 IF(NTYPE.NE.1)CALL ERRPRT(’EI0032’)
36 C Current accumulated plastic strain
37 EPBAR=RSTAVA(MSTRE+1)
38 C Set material properties
39 YOUNG=RPROPS(2)
40 POISS=RPROPS(3)
41 NHARD=IPROPS(3)
42 C Shear and bulk moduli
43 GMODU=YOUNG/(R2*(R1+POISS))
44 BULK=YOUNG/(R3*(R1-R2*POISS))
45 R2G=R2*GMODU
46 R1D3=R1/R3
47 R2D3=R2*R1D3
48 IF(EPFLAG)THEN
49 C Compute elastoplastic consistent tangent (Box 9.6)
50 C ==================================================
51 C Item (i):
52 C ---------
53 C Compute XI
54 XI=R2D3*(STRES(1)*STRES(1)+STRES(2)*STRES(2)-STRES(1)*STRES(2))+
55 1 R2*STRES(3)*STRES(3)
56 C Hardening slope
57 HSLOPE=DPLFUN(EPBAR,NHARD,RPROPS(IPHARD))
58 C Matrix E components
59 ESTAR1=R3*YOUNG/(R3*(R1-POISS)+YOUNG*DGAMA)
60 ESTAR2=R2G/(R1+R2G*DGAMA)
61 ESTAR3=GMODU/(R1+R2G*DGAMA)
62 E11=RP5*(ESTAR1+ESTAR2)
63 E22=E11
64 E12=RP5*(ESTAR1-ESTAR2)
65 E33=ESTAR3
66 C Components of the matrix product EP
67 EPSTA1=R1D3*ESTAR1
68 EPSTA2=ESTAR2
69 EPSTA3=EPSTA2
70 EP11=RP5*(EPSTA1+EPSTA2)
71 EP22=EP11
72 EP12=RP5*(EPSTA1-EPSTA2)
73 EP21=EP12
74 EP33=EPSTA3
386 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
75 C Vector n
76 VECN(1)=EP11*STRES(1)+EP12*STRES(2)
77 VECN(2)=EP21*STRES(1)+EP22*STRES(2)
78 VECN(3)=EP33*STRES(3)
79 C Scalar alpha
80 DENOM1=STRES(1)*(R2D3*VECN(1)-R1D3*VECN(2))+
81 1 STRES(2)*(R2D3*VECN(2)-R1D3*VECN(1))+
82 2 STRES(3)*R2*VECN(3)
83 DENOM2=R2*XI*HSLOPE/(R3-R2*HSLOPE*DGAMA)
84 ALPHA=R1/(DENOM1+DENOM2)
85 C Item (ii): Assemble elasto-plastic tangent
86 C ------------------------------------------
87 DMATX(1,1)=E11-ALPHA*VECN(1)*VECN(1)
88 DMATX(1,2)=E12-ALPHA*VECN(1)*VECN(2)
89 DMATX(1,3)=-ALPHA*VECN(1)*VECN(3)
90 DMATX(2,1)=DMATX(1,2)
91 DMATX(2,2)=E22-ALPHA*VECN(2)*VECN(2)
92 DMATX(2,3)=-ALPHA*VECN(2)*VECN(3)
93 DMATX(3,1)=DMATX(1,3)
94 DMATX(3,2)=DMATX(2,3)
95 DMATX(3,3)=E33-ALPHA*VECN(3)*VECN(3)
96 ELSE
97 C Compute plane stress elasticity matrix
98 C ======================================
99 NSTRE=3
100 R4GD3=R4*GMODU/R3
101 FACTOR=(BULK-R2G/R3)*(R2G/(BULK+R4GD3))
102 DO 20 I=1,NSTRE
103 DO 10 J=I,NSTRE
104 DMATX(I,J)=R2G*FOID(I,J)+FACTOR*SOID(I)*SOID(J)
105 10 CONTINUE
106 20 CONTINUE
107 C lower triangle
108 DO 40 J=1,NSTRE-1
109 DO 30 I=J+1,NSTRE
110 DMATX(I,J)=DMATX(J,I)
111 30 CONTINUE
112 40 CONTINUE
113 ENDIF
114 RETURN
115 END
Flim = 30 KN.
The numerical solution is obtained here by applying six load increments to the loaded node
shown in Figure 9.2. The resulting vertical deflection of the loaded node is plotted against the
applied force for each increment in Figure 9.3. The load level at the end of the last increment
is F = 30.3594 KN. At this point, the cantilever is effectively collapsing and equilibrium
can no longer be found for reasonably sized load increments. It should be noted that the
numerically obtained collapse load is in excellent agreement with the theoretical limit. The
relative error is approximately 1.2%. Finally, to emphasise the quadratic rates of equilibrium
convergence attained as a result of the consistent linearisation of the algorithmic constitutive
rule, the evolution of the residuals (out-of-balance forces) during the global Newton–Raphson
iterations is shown in Table 9.1 for increments 2, 4 and 6. Note that for increment 6 (the
last load increment) the cantilever is effectively collapsing (see the load-deflection diagram
of Figure 9.3). At that stage, substantial diagonal decay (with consequent round-off errors)
is found in the linearised finite element system of equations requiring a larger number of
iterations for convergence.
Material properties:
E = 210 GPa
ν = 0.3
σy = 0.24 GPa (perfectly plastic)
F
b = 50 mm
h = 100 mm
L = 1000 mm
Figure 9.2. End-loaded cantilever. Problem definition and finite element model.
30
applied load, F (KN)
present results
20 analytical limit load (30 KN)
10
0
0 40 80 120 160
vertical deflection at loaded node (mm)
σy c2 σy c2
σr = − √ and σθ = √ .
3 r2 3 r2
Within the plastic region of the plate – where a ≥ r ≥ c – the analytical stress distributions
are given by
2σy π 2σy π
σr = − √ sin +φ and σθ = √ sin −φ ,
3 6 3 6
c2 √
2
= e 3φ cos φ.
r
390 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
p 10 o
a =10 mm
100 mm
Figure 9.4. Infinite plate with a hole. Problem definition and finite element model.
1.2
normalised pressure, p/σy
0.8
numerical result
analytical limit (1.155)
0.4
0
0 0.01 0.02
radial expansion at edge, u/a
0.2
p=0.15 p=0.20 p=0.25
0.1
stresses, σr and σθ (GPa)
p=0.15
-0.1
p=0.20
-0.2 analytical (hoop)
p=0.25 analytical (radial)
FE results
-0.3
Figure 9.6. Infinite plate with a hole. Stress distribution in the radial direction (r denotes the radial
coordinate). Analytical and finite element results for different applied pressures (indicated pressures
levels in GPa).
increments ∆u = 0.02 mm. The top edge nodes (except that on the symmetry line) are free to
move horizontally. The reaction force obtained over the complete edge is plotted in Figure 9.8
versus the prescribed displacement. In this problem, plastic yielding starts at the intersection
between the bottom symmetry line and the edge of the hole and spreads in an oblique front
until the entire cross-section becomes plastic. The evolution of the plastic front is depicted in
Figure 9.9, which shows contour plots of accumulated plastic strain at different stages of the
loading process.
prescribed
displacement
u
Geometry:
l = 36 mm
w = 20 mm
d = 10 mm
thickness = 1 mm
sym
Please Wait..
Material properties (von Mises) :
l /2
E = 70 GPa
ν = 0.2
linear hardening:
_ _
σy ( ε p ) = 0.243 + 0.2 ε p [GPa]
d/ 2
sym
w/2
Figure 9.7. Stretching of a perforated rectangular plate. Geometry, boundary conditions and finite
element mesh.
3.0
edge reaction (KN)
2.0
1.0
0
0 0.04 0.08 0.12 0.16
edge displacement, u (mm)
Figure 9.8. Stretching of a perforated rectangular plate. Reaction–deflection diagram.
PLANE STRESS PLASTICITY 393
Please Wait..
Please Wait..
0.11
0.10
0.09
0.08
0.07
0.06
0.05
u = 0.02 mm 0.04 u = 0.06 mm
0.03
0.02
0.01
0.00
u = 0.10 mm u = 0.14 mm
Figure 9.9. Stretching of a perforated rectangular plate. Contour plots of accumulated plastic strain, ε̄p .
(b) the biaxial tension and compression limits of the Mohr–Coulomb surface that matches
the given uniaxial data.
The Drucker–Prager parameters η and ξ for cases (a) and (b) are automatically set to the
correct values by selecting the flag for the appropriate Drucker–Prager match of the Mohr–
Coulomb surface in the input data file for HYPLAS. In the numerical simulation, the load
is applied incrementally, with arc-length control, until collapse occurs. A total of 16 and
10 increments are used, respectively, in cases (a) and (b) to reach a numerical limit load.
To emphasise the quadratic (equilibrium) convergence rates resulting from the consistent
linearisation of the nested iteration-based integration algorithm, Table 9.2 shows the evolution
of the normalised residuals (out-of-balance forces) during typical global Newton iterations
at sample load increments. The applied load versus vertical deflection of the midpoint of the
lower edge (point A of Figure 9.10) is shown in Figure 9.11. This example illustrates the huge
discrepancy that may exist between results obtained with different types of Drucker–Prager
approximations to the Mohr–Coulomb criterion. The limit load obtained with the uniaxial
match is about nine times that obtained with the biaxial match. The numerical limit load
in the uniaxial match case is 3045 KN/m whereas with the biaxial match the limit load is
335.5 KN/m. In the present example, failure occurs by crushing of the material near the
support under a combination of shear and compressive stresses. The failure mechanism is
illustrated in Figure 9.12 where the direction of the incremental nodal displacements at failure
are plotted for the biaxial match case. As pointed out in Section 6.4.4, limit loads can be
significantly overestimated when the uniaxial match is used under such conditions, becoming
worse towards the biaxial compression state. This is particularly true in the present case
where the ratio fc /ft is high. The biaxial match on the other hand produces a conservative
prediction. Other predictions can be obtained by using the outer or inner edge Drucker–
Prager approximations to the Mohr–Coulomb surface (refer to Section 6.4.4, from page 166)
corresponding to the given uniaxial data. In these cases (not shown on the load-deflection
graphs) the predicted limit loads are, respectively, 2539 KN/m and 276 KN/m. The most
conservative one is that produced by the inner match approximation.
It should be emphasised that, in practice, shear walls are made from reinforced rather
than plain concrete. Thus, the present example is not meant to be a practical engineering
application. Instead, it serves to illustrate the performance of the (nested iteration) plane
stress Drucker–Prager model. Within the finite element environment of HYPLAS, effects of
reinforcement could be accounted for by adding steel elements (the reinforcement) on top of
an appropriately designed background plain concrete mesh.
PLANE STRESS PLASTICITY 395
Uniformly distributed
load, P
Geometry:
l = 1.18 m
h = 1.20 m
thickness, t = 0.1 m
Material properties: h
E = 2.75 x 104 MPa
ν = 0.2
fc’ = 30.5 MPa
ft ’ = 2.75 MPa
l/8
400
3000
applied load, P (KN/m)
applied load, P (KN/m)
300
2000
200
1000
100
0 0
0 1.0 2.0 3.0 0 0.4 0.8 1.2 1.6
vertical deflection at node A (mm) vertical deflection at node A (mm)
(a) (b)
Figure 9.11. Shear wall. Load versus vertical deflection at point A: (a) uniaxial match; (b) biaxial
match.
396 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
failure
region
σ13
σ13
σ33
σ23 σ23
η1
e1
e2 e3 m
ζ
n
η2
with σ13 , σ23 and σ33 being the only possible non-vanishing components.
Resultant forces
The generalised forces associated with the generalised strain vectors γ and κ are, respectively,
the resultant force, denoted n, and the resultant moment, denoted m. These are obtained by
398 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
integration of the non-vanishing stress components over the beam cross-section. The resultant
force has the local components
n1 = σ13 dA
C
n2 = σ23 dA (9.87)
C
n3 = σ33 dA
C
whereas the resultant moment is defined by
m1 = η2 σ33 dA
C
m2 = − η1 σ33 dA (9.88)
C
where sij denote components of the deviatoric stress tensor. Note that the generally non-
vanishing components s11 and s22 , missing in the above representation of the stress deviator,
are dependent on s33
s11 = s22 = − 21 s33 = − 31 σ33 . (9.90)
PLANE STRESS PLASTICITY 399
s = P σ, (9.91)
σ = De εe
∂Φ Pσ
(9.95)
ε̇p = γ̇ = γ̇ 32 √
∂σ σT Pσ
ε̄˙p = γ̇
where
Φ(σ, ε̄p ) ≥ 0, γ̇ ≥ 0, γ̇ Φ(σ, ε̄p ) > 0. (9.96)
The elastic matrix, De , for the three-dimensional Timoshenko beam is given simply by
G 0 0
De = 0 G 0 . (9.97)
0 0 E
400 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
The above set of evolution equations is analogous to that of the plane stress-projected
von Mises model of Box 9.3. Under stress states that satisfy the constraint (9.86), the rates
ε̇p13 , ε̇p23 and ε̇p33 resulting from the original three-dimensional elastoplastic equations are
identical to the rates resulting from the above constrained equations. The main difference
between the two formulations is the fact that, in the three-dimensional case, the history of
the total strain components, ε11 , ε22 and ε12 , is prescribed in the underlying initial value
problem, whereas, in the constrained case, these components are dependent variables whose
evolution is a consequence of the constraint on the stress state.
Having determined the history of the independent strains by integrating the constrained
elastoplastic evolution equations, the history of the dependent strain components, ε11 , ε12
and ε22 , is obtained as follows. As σ12 = 0 and, consequently, ε̇p12 = 0 for constrained
stress states, the elastoplastic additive strain decomposition together with the elastic relation,
σ12 = 2G εe12 , implies that
ε12 = εe12 = εp12 = 0. (9.98)
−K
εe11 = εe22 = εe . (9.99)
2(G + K) 33
From the three-dimensional von Mises equations, it follows that, for constrained stress states,
we have ε̇p11 = ε̇p22 = − 21 ε̇p33 , so that the components εp11 and εp22 are promptly found to be
given by
εp11 = εp22 = − 12 εp33 . (9.100)
Finally, the total strain components ε11 and ε22 are obtained from the additive elastoplastic
split of the total strain tensor
ε11 = ε22 = εe11 + εp11 . (9.101)
The derivation of the implicit integration algorithm for the beam state-projected von Mises
model follows the same steps as the derivation of the plane stress-projected algorithm
described in Section 9.4.2. The final return-mapping equation for the plastic multiplier can
be cast in the same format as equations (9.57, 9.58), with matrix P redefined by (9.92) and
matrix A(∆γ) now given by the simpler diagonal form
1
1 + 2G ∆γ 0 0
1
A(∆γ) =
0 0 .
(9.102)
1 + 2G ∆γ
1
0 0 2
1 + 3 E ∆γ
PLANE STRESS PLASTICITY 401
With the above matrix, the function ξ(∆γ) defined in (9.58) has the following explicit
expression
trial 2 trial 2 trial 2
2[(σ13 ) + (σ23 ) ] 2(σ33 )
ξ(∆γ) ≡ + 2 . (9.103)
(1 + 2G ∆γ)2 3(1 + 3 E ∆γ)2
The overall elastic predictor/return-mapping algorithm is identical to that shown in
Boxes 9.4 and 9.5 except that A and ξ (as well as De and the stress and strain arrays) have
now the above new definitions and
ξ trial := 2[(σ13
trial 2 trial 2
) + (σ23 trial 2
) ] + 23 (σ33 ) . (9.104)
In the Newton algorithm of Box 9.5, the computation of the derivative ξ is replaced by
−8G[(σ13
trial 2 trial 2
) + (σ23 ) ] trial 2
8E (σ33 )
ξ := − 2 . (9.105)
(1 + 2G ∆γ)3 9(1 + 3 E ∆γ)3
T HIS chapter is devoted to more advanced plasticity models. Recall that, in Chapters 7–
9, all examples of numerical implementation of plasticity models have been limited
to basic theories. Here, we move one step further and apply the same underlying concepts
to more advanced models. Sections 10.1 and 10.2 describe, respectively, the treatment of a
modified Cam-Clay model and a capped Drucker–Prager model – both mainly applicable
to the description of geomaterials. Section 10.3 introduces the modelling of anisotropic
plasticity. In this context, the Hill model (Hill, 1950), a model based on the Hoffman
anisotropic criterion (Hoffman, 1967) – here referred to as the Hoffman model – and the
Barlat–Lian model for sheet metals (Barlat and Lian, 1989) are discussed in detail, together
with the computational treatment of the Hoffman and Barlat–Lian models with isotropic
strain hardening.
The new theoretical concepts introduced in this chapter are, essentially, the modelling of
plastic compaction – a phenomenon of particular relevance in the behaviour of geomaterials
– and the modelling of plastic anisotropy. Note, however, that the computational implemen-
tation of elastoplastic models possessing such new features reduces to mere specialisations
of the general framework discussed in the preceding chapters.
The modified Cam-Clay model was originally proposed by Roscoe and Burland (1968) to
model the plastic behaviour of soils. Models of this type are frequently used in the finite
element modelling of soil mechanics problems. These models are characterised by plastic
compressibility, hardening (softening) associated with compressive (dilatant) plastic flow and
a possibly nonlinear elasticity law to model the hydrostatic pressure/volumetric elastic strain
relation (refer, for instance, to Naylor et al. 1981, and Muir Wood 1990). In many cases,
when dealing with soil consolidation problems, Cam-Clay-type models can be used within
a multiphase environment with coupling between solid behaviour and flow through porous
media. Such problems are outside the scope of this book and we shall describe in this section
a version of the modified Cam-Clay plasticity model featuring a linear elasticity law and
a standard volumetric plastic strain-dependent isotropic hardening rule. In soil mechanics
applications, the use of a linear elastic law can be justified whenever the considered range of
hydrostatic pressures is sufficiently narrow.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
404 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
q
1
M
yield surface
Φ=0
0 p
βa a
pc pt
where the constant M is the ratio between the two radii of the Cam-Clay ellipse, a is the
radius of the ellipse along the pressure axis and pt is the tensile yield hydrostatic stress. The
parameter b takes the values
1 if p ≥ pt − a
b= (10.2)
β if p < pt − a,
where β is a material constant. This parameter modifies the radius of the second half of the
ellipse on the compressive side of the hydrostatic axis. If β = 1, the yield locus becomes
an ellipse with radii a and M a, respectively, along directions p and q. Note that, for any
β > 0, the two horizontal halves of the surface intersect in a smooth fashion at (p, q) = (pt −
a, M a). The dashed line in Figure 10.1 is named the critical state line. The parameter M is
the slope of the critical state line. The portion of the p–q plane on the right of the intersection
between the critical state line and the Cam-Clay ellipse is named the supercritical region and
the portion on its left is the subcritical region.
ADVANCED PLASTICITY MODELS 405
Flow rule
The plastic flow equation can be defined by postulating associativity, which gives
∂Φ
ε̇p = γ̇ = γ̇ N = γ̇ (Nd + Nv I ), (10.3)
∂σ
where we have conveniently split the flow vector, N = ∂Φ/∂σ, into deviatoric and volumet-
ric components, respectively denoted Nd and Nv :
3 2
Nd = s; Nv = (p − pt + a). (10.4)
M2 b2
The deviatoric and volumetric plastic strain rates then read
ε̇pd = γ̇ Nd ; ε̇pv = γ̇ Nv . (10.5)
The plastic flow defined by the above associative rule is compressive† (ε̇pv < 0) if p < pt − a
(subcritical states), dilatant (ε̇pv > 0) if p > pt − a (supercritical states) and isochoric (ε̇pv = 0)
at p = pt − a (critical state).
Hardening law
A simple way of incorporating hardening into the model consists in letting the yield
surface parameter a be a function of a hardening internal variable,‡ α. For many plastically
compressible materials, and for soils in particular, the state of hardening is largely dependent
upon the volumetric plastic strain (or plastic compaction),
εpv ≡ tr[εp ]. (10.6)
Because in actual applications, soils will be predominantly subjected to compressive strains,
the hardening variable, α, will be chosen as the compressive-positive volumetric plastic strain:
α ≡ −εpv (10.7)
and the hardening behaviour will be defined by means of the experimentally determined
hardening function
a = a(α). (10.8)
Equivalently, in terms of the compressive yield pressure (or compaction pressure), pc (also
assumed compressive-positive) hardening can be defined by
pc (α) ≡ (1 + β) a(α) − pt . (10.9)
Under this assumption, the evolution of the volumetric plastic strain will change the size
of the modified Cam-Clay yield surface while maintaining its shape. Hardening in this case is
characterised by a change in compressive yield pressure, having the intersection of the yield
surface with the p-axis at p = pt (the tensile yield pressure) fixed.
† Consistently with the continuum mechanics sign convention adopted throughout this book, compressive
volumetric strains and hydrostatic pressures have a negative sign here. It should be noted that the opposite sign
convention is commonly adopted in soil mechanics texts, where compressive volumetric strains and hydrostatic
pressures are assumed positive.
‡ In a more general setting, other yield surface parameters such as M or β could also be functions of hardening
variables.
406 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Remark 10.1 (Hardening behaviour). The hardening function a(α) (or pc (α)) is usually
expected to be monotonically increasing in its argument, that is, a increases with increasing
plastic compression. In this case, following the comments on the volumetric plastic strain
rate made immediately after expression (10.5), hardening (expansion of the elastic domain)
occurs only in the subcritical region, i.e. under hydrostatic pressures p < pt − a. For pressures
p > pt − a (supercritical states) plastic flow will cause softening (contraction of the elastic
domain) and at p = pt − a (the critical state) the model behaves as perfectly plastic (fixed
yield surface).
Here we shall adopt a linear elasticity law to model the reversible behaviour of the modified
Cam-Clay model. We remark, however, that it is sometimes convenient to adopt nonlinear
elasticity laws with models of this type. The implementation of Cam-Clay-type models with
nonlinear elastic laws is described, for instance, by Simo and Meschke (1993) and Owen
et al. (1998).
2
α n+1 = α n − ∆γ [pn+1 − pt + a(αn+1 )]. (10.12)
b2
In addition, the discrete consistency equation here reads
1 +q ,2
n+1
Φn+1 = [p n+1 − p t + a(αn+1 )]2
+ − [a(αn+1 )]2 = 0, (10.13)
b2 M
where qn+1 is the von Mises effective stress associated with sn+1 .
Particularisation of the general return-mapping equations (7.25) (page 196) for the present
model results in the system of equations comprising (10.11)–(10.13), having αn+1 , pn+1 , the
linearly independent components of sn+1 and the incremental plastic multiplier, ∆γ, as the
unknowns.
ADVANCED PLASTICITY MODELS 407
M2
sn+1 = s(∆γ) ≡ strial , (10.14)
M2 + 6G ∆γ
M2
qn+1 = q(∆γ) ≡ q trial . (10.15)
M 2 + 6G ∆γ
Finally, the substitution of (10.15) and (10.16) in (10.12) and (10.13) leads to the following
reduced system of two scalar equations with unknowns ∆γ and αn+1 :
2
R
1 [p(α) − p + a(α)]2 + q(∆γ) − [a(α)]2
1 2 t 0
b M
≡ = , (10.17)
R 2
α − α n + ∆γ [p(α) − pt + a(α)]
0
2
b2
where the subscript n + 1 of αn+1 has been omitted for notational convenience. Once the
solution to the above return-mapping equations is obtained, the stress-updating procedure is
completed with the computation of sn+1 by means of (10.14), followed by the calculation of
the updated stress tensor through the standard relation
It is worth commenting here that system (10.17) could in principle be further reduced to
a single scalar equation having αn+1 as the only unknown by trivially solving the second
equation of (10.17) for ∆γ:
b2 (αn − αn+1 )
∆γ = ∆γ(αn+1 ) ≡ , (10.19)
2[p(αn+1 ) − pt + a(αn+1 )]
and then replacing this result into the first equation. It should be noted, however, that the
above expression cannot be used in practice at p = pt − a.
Newton–Raphson solution
The solution of (10.17) can be undertaken as usual by the Newton–Raphson algorithm.
For completeness, an explicit expression for the linear system to be solved for the iterative
408 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
corrections δ∆γ (k) and δα(k) at the typical k th Newton iteration is shown below:
2
−12 G q 2p̄ (k−1)
2 (K + H) − 2aH δ∆γ (k) R1
M + 6G∆γ M b 2
= , (10.20)
2p̄ 2∆γ (k) (k−1)
1 + 2 (K + H) δα R 2
b2 b
where H is the hardening modulus (the slope of the hardening curve):
da
H = H(α) ≡ , (10.21)
dα
the scalar p̄ is defined as
p̄ = p̄(α) ≡ p(α) − pt + a(α), (10.22)
and all terms of the derivative matrix on the right-hand side are evaluated at (∆γ, α) =
(k−1)
(∆γ (k−1) , αn+1 ).
∂N ∂Nd ∂Nv
= +I ⊗ (10.24)
∂σ ∂σ ∂σ
with
∂Nd 3 ∂Nv 2
= 2 Id ; = 2 I. (10.25)
∂σ M ∂σ 3b
In addition, from (10.13) we obtain
∂Φ 2
= 2 (p − pt + a)H − 2aH = (Nv − 2a)H. (10.26)
∂α b
Here we should note that tangential relation (10.23) has a similar format to that of (7.141)
which was obtained for the implicit implementation of the isotropically hardening von Mises
ADVANCED PLASTICITY MODELS 409
model. Accordingly, its inversion follows completely analogous steps to those leading
to (7.148) and yields the following formula for the elastoplastic consistent tangent operator:
∂σn+1
Dep ≡
∂εe trial
1
=P− (P : N ) ⊗ (P : Ñ ), (10.27)
Ñ : P : N + (Nv − 2a)Nv H
where
−1
e ∂N
P≡ IS + ∆γ D : : De (10.28)
∂σ
with flow vector derivative given by (10.24), and
∂Nv 2H∆γ
Ñ ≡ −∆γ(Nv − 2a)H + 1+ N
∂σ b2
−2∆γ(Nv − 2a)H 2H∆γ
= 2
I+ 1+ N. (10.29)
3b b2
Since in general Ñ
= N, the elastoplastic tangent operator (10.27) is generally unsym-
metric. Its non-symmetry is a consequence of the non-associativity of the adopted hardening
law. The reader is referred to the discussion at the beginning of Section 7.4.6, from page 243.
Also note that in the absence of hardening, that is, if H = 0, Ñ coincides with N and
symmetry is recovered.
1 q
M Drucker–Prager
cone
Φa = 0
modified Cam-Clay
yield surface
Φb = 0
0 p
βa a
pc pt
and Φb = 0 defines the elliptical cap – the modified Cam-Clay yield surface for subcritical
states discussed in the previous section of this chapter. In (10.30)1,
η = 2/3 M. (10.32)
.
εp cone
cap flow vector
normal
Nb
Na
.
εp
p
Figure 10.3. Capped Drucker–Prager model. Flow vectors.
(page 176), (6.158)1 and (6.166). At the cone apex, the flow vector has deviatoric and
volumetric components (6.160)1 and (6.166), respectively. On the cap, the (associative) flow
rule is that of the modified Cam-Clay model, given by (10.3) and (10.4).
Following the general representation of associative plastic flow rules for multisurface
plasticity models given by (6.73) (page 156), (6.77) and (6.78), the plastic strain rate at the
intersection between the Drucker–Prager cone and the elliptical cap reads
ε̇p = γ̇ a N a + γ̇ b N b , (10.34)
where N a and N b are, respectively, the non-associative Drucker–Prager flow vector at the
smooth portion of the cone and the associative modified Cam-Clay flow vector referred to
above. The flow vector N a has the standard form
1 η̄
Na = s + I. (10.35)
2 J2 (s) 3
At the cap/cone intersection, where p = pt − a, the normal to the cap with general expres-
sion (10.4) is a deviatoric tensor:
3
N b = Ndb = s. (10.36)
M2
A general p-q plane illustration of the plastic flow directions for the resulting model is given
in Figure 10.3.
Hardening
As for the previously discussed modified Cam-Clay model (refer to expressions (10.9)
and (10.8)) isotropic strain hardening is incorporated by letting the compaction pressure pc
(or, equivalently, the yield surface parameter a) be a function of the compressive-positive
412 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
volumetric plastic strain, α, defined in (10.7). The Drucker–Prager cone remains fixed and is
independent of the hardening variable.
As pc increases (decreases), the cap moves in the compression (tensile) direction along the
hydrostatic line, expanding (shrinking) the elastic domain.
and then substituting the resulting expression, together with (10.39) into the cone/cap
intersection consistency equation
Φan+1 ≡ J2 (sn+1 ) + η[pn+1 − pt ] = 0
2 (10.41)
1 q(sn+1 )
Φbn+1 ≡ 2 [pn+1 − pt + a(αn+1 )]2 + − [a(αn+1 )]2 = 0.
β M
The derivation of the final equations for the unknown multipliers ∆γ a and ∆γ b is straight-
forward and will be left as an exercise for the interested reader.
ADVANCED PLASTICITY MODELS 413
START
atrial
Φ >0 Y Y APPLY RETURN TO
atrial
OR Φ >0
trial D-P CONE
Φb >0
N N
b
Φn+1 > 0
APPLY RETURN TO Y N
AND
CAP (invalid return) p n+1 < -an+1
Y RETURN TO
p n+1 > -an+1 CONE/CAP
(invalid return)
INTERSECTION
END
Figure 10.4. Capped Drucker–Prager model. Algorithm for selection of the correct return-mapping
procedure.
The tangent operator consistent with the Drucker–Prager cone and cone apex return
algorithms are those derived in Section 8.3.4 (from page 337) which, for the present model,
must be particularised for the perfectly plastic case. Note that, in this case, at the cone apex,
we have
Dep = 0. (10.42)
For the cap return, the tangent operator is that of (10.27) with b = β in (10.29). The derivation
of an explicit form for the tangent for the cone/cap intersection return follows an analogous
procedure to those of the two-vector return mappings discussed in Chapter 8 and will be left
as an exercise for the reader.
Pressure insensitivity
One important feature of the Hill criterion is the fact that, as for the von Mises and Tresca
criteria, it is pressure insensitive. This allows us to express its yield function in terms of the
components of the stress deviator, s, only. By recalling the trivial identities
σii − σjj = sii − sjj , (10.48)
with no summation on repeated indices, and
σij = sij , for i
= j, (10.49)
the Hill yield function can be equivalently written as
Φ(σ, σ̄) = F1 (s11 − s22 )2 + F2 (s22 − s33 )2 + F3 (s11 − s33 )2
+ F4 s212 + F5 s223 + F6 s213 − σ̄ 2 . (10.50)
416 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
we obtain
3 2
Φ(σ, σ̄) = [s + s222 + s233 + 2(s212 + s223 + s213 )] − 1
2σy2 11
3
= 2 J2 (s) − 1. (10.53)
σy
The corresponding yield surface is that of the von Mises model with uniaxial yield stress σy :
3 J2 (s) = σy . (10.54)
Geometric representation
Unlike isotropic models, whose yield functions can always be expressed in terms of principal
stresses and the corresponding yield surfaces can be visualised in the three-dimensional
principal stress space, yield surfaces for anisotropic models cannot be easily represented
graphically. Anisotropic yield surfaces are truly six-dimensional hypersurfaces in the space
of stress components. Nevertheless, visualisation of projections of such hypersurfaces on
two- or three-dimensional subsets of the six-dimensional stress space is possible and can
provide a very good insight into the properties of the anisotropic model. Two- and three-
dimensional projections are obtained by fixing, respectively, four and three stress components
and then plotting the corresponding yield locus on the subset of the two or three free stress
components. In general, the two- or three-dimensional yield locus will change if the fixed
stress components are changed. Thus, an appropriate visual study of anisotropic yield criteria
requires plotting such projections with different combinations of fixed stress values.
In order to give the reader some visual insight into properties of the Hill criterion, we will
plot here the corresponding yield surfaces under states of plane stress. Before proceeding
to the graphical representation, it is worth remarking that at any given state of shear stress
(defined by the components σ12 , σ23 and σ13 relative to the axes of material orthotropy), the
yield surface defined by the Hill criterion (with material constants within the ‘usable’ range
of the Hill model) in the three-dimensional σ11 -σ22 -σ33 space is a (generally elliptic) cylinder
whose axis is the hydrostatic line (σ11 = σ22 = σ33 ). The size of the cylinder cross-section
(and its intersection with the plane stress space) depends upon the state of shear stress. For
plane stress states (on plane {e1 , e2 }), the Hill yield surface is defined by
2
(F1 + F3 ) σ11 2
+ (F1 + F2 ) σ22 − 2F1 σ11 σ22 + F4 σ12
2
− σ̄ 2 = 0. (10.55)
ADVANCED PLASTICITY MODELS 417
In order to study the effect of the material constants (the yield stresses) on the yield surface,
we conveniently assume the following relations:
0 0 0 0
σ22 = a22 σ11 ; σ33 = a33 σ11 ;
a12 0 a23 0 a13 0 (10.56)
0
σ12 = √ σ11 ; 0
σ23 = √ σ11 ; 0
σ13 = √ σ11 ,
3 3 3
0
where aij are parameters defining all initial yield stresses of the model as a function of σ11 .
0
Note that the von Mises surface, with uniaxial yield stress σ11 is recovered if we set
a22 = a33 = a12 = a23 = a13 = 1. (10.57)
Firstly, let us focus on the effect of the variation of direct yield stresses on the Hill yield
surface. This effect is illustrated in Figures 10.5 and 10.6. In Figure 10.5, the Hill yield
0
surface on the σ11 -σ22 space is plotted for different values of direct yield stress σ22 (i.e.
different values of a22 ), assuming no shear stresses and
a33 = 1. (10.58)
In this case (absence of shear stresses), the parameters a12 , a23 and a13 have no effect on
the Hill yield function. For a22 > 0.5, the Hill surface is an ellipse intersecting the horizontal
axis at ±1 and the vertical axis at ±a22 . The surface with a22 = 1 corresponds to that of the
von Mises model. If a22 increases (see surface plotted for a22 = 1.5), the von Mises ellipse is
stretched in the σ22 direction. If a22 decreases (refer to the surface with a22 = 0.7), the ellipse
is compressed in the vertical direction but substantially stretched along its longer radius. In
the limit, when a22 = 0.5 (one direct yield stress is half of the other two), the original ellipse
degenerates into two parallel straight lines (two hyperplanes in the six-dimensional stress
space). For a22 < 0.5 (see surface with a22 = 0.3), the yield locus produced by Hill’s function
is a set of two hyperbolic (non-convex) surfaces in stress space. Clearly, in such cases, the
elastic domain becomes unbounded in some directions and does not correspond to physical
behaviour. Thus, the Hill criterion is to be used only within certain limits of yield strength
variation among the orthotropy directions. At this point, it is important to emphasise that
the Hill criterion was originally proposed to model anisotropy of formed steel components.
For such materials, the maximum variation of yield strength between the different orthotropy
directions is typically less than about 5 to 10%. Under such conditions, the Hill criterion can
provide reasonable approximations to the actual yield surfaces. In the next representation of
the Hill surface shown in Figure 10.6, we illustrate the effect of the variation of the direct yield
0 0 0
stress σ33 (variation of a33 ) in the absence of shear stresses and with σ22 = σ11 (a22 = 1).
For a33 > 0.5, the Hill surface is always an ellipse intersecting the horizontal and vertical axis
at ±1. As σ330
increases (see surface with a33 = 1.5), the ellipse stretches in the biaxial state
0
direction. When σ33 decreases (see surfaces with a33 = 0.7 and 0.6) the ellipse is compressed
in the biaxial state direction and stretched in a pure shear (σ11 = −σ22 ) direction. When
a33 = 0.5 (one direct yield stress is half of the other two), the ellipse degenerates into two
straight lines and becomes a hyperbola for a33 < 0.5. Finally, in Figure 10.7 we illustrate the
effect of shear stresses on the Hill yield surface. Essentially, any increase in shear stress will
shrink the yield surface isotropically (this effect is also present in the von Mises criterion and
can only be seen if, as opposed to the usual representation in principal stress space, the surface
0
is represented in the space of direct stresses along fixed axes). Obviously, if σ12 = σ̄σ12 (the
shear stress has reached its yield limit) the σ11 -σ22 space surface degenerates to a point
at (0, 0).
418 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
σ22 /σ11y
1.5
a 22 =1.5
1 a 22 = 0.3
a 22 =1.0
a 22 = 0.5
a 22 = 0.7 0.5
-1 1 σ11 /σ11y
-0.5
-1
-1.5
0
Figure 10.5. The Hill orthotropic criterion. Effect of variation of direct yield stress σ22 on the yield
0 0
surface. Surface plot for various values of a22 with no shear stresses and a33 = 1 (σ33 = σ11 ).
σ22 /σ11y
a 33 =1.5
a 33 = 0.5
a 33 =1.0
a 33 =0.6 1
a 33 =0.7
σ11 /σ11y
-1 1
-1
a 33 = 0.4
0
Figure 10.6. The Hill orthotropic criterion. Effect of variation of direct yield stress σ33 (the transversal
yield stress) on the yield surface. Surface plot for various values of a33 with no shear stresses and
0 0
a22 = 1 (σ22 = σ11 ).
ADVANCED PLASTICITY MODELS 419
σ22 /σ11y
1.5 σ12 =0
0.5
σ11 /σ11y
-1 -0.5 0.5 1
-0.5
-1
-1.5
Figure 10.7. The Hill orthotropic criterion. Effect of shear stress on the yield surface. Surface plot for
various values of σ12 with σ23 = σ13 = 0, a22 = 1.5 and a12 = a33 = 1.
Hardening law
σ̄ = σ̄(ε̄p ). (10.59)
The material in this case is assumed to remain orthotropic with constant axes of orthotropy.
In addition, in view of (10.46), all six yield stresses of the Hill criterion will change in strict
proportion (isotropically) as the accumulated plastic strain increases.
Remark 10.2. Since general straining of an initially orthotropic material is usually expected
to change the yield stresses in different directions by different amounts, or even lead to loss
of orthotropy, the model resulting from the above assumptions provides only a first approx-
imation to the phenomenon of hardening. We remark, however, that the phenomenological
modelling of hardening in plastically anisotropic materials is a complex issue which remains
open at present. The approximation provided by simple laws of the above type can be very
useful in the finite element analysis of plastically orthotropic materials and, undoubtedly,
represents a substantial gain in predictive capability (as compared to the use of isotropic
plasticity models) in situations where plastic anisotropy is an important feature of the material
behaviour.
420 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
materials, with no reference to plastic flow modelling. What we refer to as the Hoffman model here, is an extension
of the Hill elastoplastic model with yield criterion and plastic flow rule based on Hoffman’s function (10.66).
ADVANCED PLASTICITY MODELS 421
where σ̄ is the non-dimensional relative yield stress (analogous to that of the Hill criterion)
and C1 , C2 , . . . , C9 are material constants defined as
1 1 1 1
C1 =
2 σ11t σc + σt σc − σt σc ;
11 22 22 33 33
1 −1 1 1
C2 = t σc + t σc + t σc ; (10.67)
2 σ11 11 σ22 22 σ33 33
1 1 1 1
C3 =
2 σ11t σc − σt σc + σt σc ,
11 22 22 33 33
t c
with σii and σii (no summation on repeated indices) denoting the initial (i.e. when σ̄ = 1)
direct yield stresses along the orthotropy direction i, respectively in tension and compres-
sion, and
1 1 1
C4 = 0 2 ; C5 = 0 2 ; C6 = 0 2 , (10.68)
(σ12 ) (σ23 ) (σ13 )
and
c
σ11 − σ11
t c
σ22 − σ22
t c
σ33 − σ33
t
C7 = c σt ; C8 = c σt ; C9 = c σt . (10.69)
σ11 11 σ22 22 σ33 33
0 0 0
The constants σ12 , σ23 and σ13 have the same meaning as in the Hill criterion, i.e. they denote
the initial yield stresses in states of pure shear on the planes of orthotropy.
Remark 10.3 (Hill criterion as a particular case). If for each principal direction of
orthotropy the direct yield stress in tension coincides with the direct yield stress in com-
pression, i.e. if we set
c t 0 c t 0 c t 0
σ11 = σ11 = σ11 ; σ22 = σ22 = σ22 ; σ33 = σ33 = σ33 , (10.70)
This contribution – a linear function of the hydrostatic pressure – is identical to that of the
Drucker–Prager criterion (refer to expression (6.121), page 167). The constraint
C7 + C8 + C9 = 0 (10.72)
is the necessary and sufficient condition for the Hoffman criterion to be pressure-insensitive.
In its general form (within the range of applicability of the criterion), the Hoffman yield
surface is an elliptic cone in the σ11 -σ22 -σ33 space. The cone intersects the σ11 , σ22 and σ33
axes at the corresponding prescribed values of (tensile and compressive) direct yield stresses.
A graphical representation of the Hoffman criterion is shown in Figure 10.8. Again, only the
intersection of the yield surface with the σ11 -σ22 plane, in the absence of shear stresses, is
422 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
σ22 /σ σ11t
1
0.5
-0.5
-1
plotted. For illustration purposes only, the following relations have been chosen:
c t c t t t
σ11 = 1.2 σ11 ; σ22 = 0.9 σ11 ; σ22 = 0.8 σ11 ;
(10.73)
c c t t
σ33 = σ11 ; σ33 = σ11 .
As for the Hill criterion (refer to Figure 10.7), the Hoffman surface shrinks isotropically with
an increase in shear stresses. The effect of (isotropic) hardening can also be incorporated into
the Hoffman criterion by assuming σ̄ to be a function of the accumulated plastic strain.
Again, it is important to observe that (as for the Hill criterion) the Hoffman model can only
be used within certain limits of yield strength variation among the directions of orthotropy.
Excessive variations of direct (tensile or compressive) yield stress can cause the originally
elliptic surface to degenerate into a hyperbola that does not model the actual behaviour of
solids.
Flow rule
Here, an associative flow rule is also postulated. Analogously to (10.64), the associative
component-wise plastic flow equations for the Hoffman model are given by
ε̇p11 = γ̇ [C7 + 2C1 (σ11 − σ22 ) + 2C3 (σ11 − σ33 )],
with
Ei (1 − νjk νkj )
Dii = (10.79)
(1 − νki νik )(1 − νjk νkj ) − (νij + νik νkj )(νji + νjk νki )
for i = 1, 2, 3 (no summation on the repeated index) and (i, j, k) denoting cyclic permuta-
tions of (1, 2, 3),
νji + νjk νki
Dij = Dii (10.80)
1 − νjk νkj
for i
= j and, again, no summation on the repeated index, and
The total number of independent elastic constants is nine (in the full three-dimensional
case). These are: the Young’s moduli, E1 , E2 and E3 , associated with directions 1, 2 and 3
of orthotropy respectively; the shear moduli, G12 , G23 and G31 associated, respectively, with
planes 12, 23 and 31 and satisfying
Gij = Gji ; (10.82)
and three Poisson’s ratios, ν12 , ν23 and ν31 , where νij is defined as the ratio between the
contraction in direction j and extension in direction i under a uniaxial stress state along i.
424 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
where, clearly, the D matrix here is the orthotropic operator (10.78). The trial accumulated
plastic strain (isotropic hardening state variable) is
If the trial state is outside the elastic domain defined by the yield function (10.66), the return
mapping described below is applied.
Return-mapping algorithm
For computer implementation purposes, it is convenient to write the Hoffman yield func-
tion (10.66) in the following equivalent form in terms of the array of stress components:
Φ(σ, σ̄) = 1
2 σT Pσ + qT σ − σ̄ 2 , (10.86)
where
C1 + C3 −C1 0 −C3
−C1 C2 + C1 0 −C2
P=2 (10.87)
0 0 C4 0
−C3 −C2 0 C3 + C2
and
q = [ C7 C8 0 C9 ]T . (10.88)
With the above notation, the flow rule (10.74) can be equivalently expressed in terms of the
engineering plastic strain rate array as
With the above relations at hand, a direct particularisation of the general return-mapping
equations (7.25) (page 196) is obtained as
εen+1 − εe trial + ∆γ (Pσn+1 + q)
0
p 1
ε̄n+1 − ε̄n − ∆γ[ 3 (Pσn+1 + q) Z (Pσn+1 + q)]
p 2 T 2 = 0 . (10.92)
p
2 σn+1 Pσn+1 + q σn+1 − [σ̄(ε̄n+1 )]
1 T T 2 0
that is, the updated stress, σn+1 is a function of ∆γ, exclusively. Further, introduction of the
above function in the second equation of (10.92) gives
1
ε̄pn+1 = ε̄p (∆γ) ≡ ε̄pn + ∆γ{ 23 [Pσ(∆γ) + q]T Z [Pσ(∆γ) + q]} 2 . (10.94)
Finally, with the substitution of the last two expressions in the third equation of (10.92) –
the discretised plastic consistency – the original return-mapping system is reduced to the
following nonlinear scalar equation for ∆γ:
Φ̃(∆γ) ≡ 1
2 [σ(∆γ)]T Pσ(∆γ) + qT σ(∆γ) − [σ̄(ε̄p (∆γ))]2 = 0. (10.95)
In summary, the stress-updating procedure for the Hoffman model comprises the solution
of (10.95), followed by the update of stress and accumulated plastic strain according
to (10.93) and (10.94), respectively.
Newton–Raphson solution
In the Newton–Raphson iterative scheme to solve (10.95), the k th guess for the solution ∆γ
is obtained as
∆γ (k) = ∆γ (k−1) + δ∆γ (k) , (10.96)
where (k−1)
Φ̃(∆γ)
δ∆γ (k) = − . (10.97)
dΦ̃/d∆γ
426 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
dΦ̃ dσ dε̄p
= [Pσ(∆γ) + q]T − 2 σ̄(∆γ) H̄ , (10.98)
d∆γ d∆γ d∆γ
dσ̄
H̄ = , (10.101)
dε̄p
also evaluated at ε̄pn+1 . The flow vector N is given by
N = Pσn+1 + q, (10.102)
The above operator is clearly unsymmetric. Its non-symmetry is a consequence of the fact
that the adopted hardening rule is non-associative (despite the associativity of the plastic
ADVANCED PLASTICITY MODELS 427
flow rule). Refer to the comments made in Section 7.4.6, page 243. Note that the above
formula is analogous to that given by (7.149) for the von Mises isotropically hardening
model. Its use, in this format, is restricted strictly to hardening models (H̄
= 0) owing to
the presence of the hardening modulus in the denominator of the third summand of the
term within square brackets. For use under the assumption of perfect plasticity (H = 0),
an alternative representation can be obtained following completely analogous steps to those
leading to expression (7.148). This will be left as an exercise for the interested reader.
where
f (σ) ≡ a|K1 + K2 |M + a|K1 − K2 |M + (2 − a)|2 K2 |M , (10.106)
with - 2
σ11 + h σ22 σ11 − h σ22 2 ,
K1 = ; K2 = + b2 σ12 (10.107)
2 2
y
where M , a, b and h are material constants and σ11 is the uniaxial yield stress in the principal
orthotropy direction 1. The yield function (10.105) is convex (Barlat and Lian, 1989) if
To give the reader an idea of realistic values for these constants, the following have been
determined by Lege et al. (1989) for an aluminium alloy:
The constant M defines the curvature of the yield surface. This can be seen in Figure 10.9
where the projection of the Barlat–Lian yield surface on the σ11 -σ22 plane is shown (in the
absence of shear stresses) for different values of M and a = h = 1 (note that the value of b is
immaterial in the absence of shear). We remark that the criterion becomes isotropic when a =
b = h = 1. Under such a condition, the standard von Mises locus is recovered when M = 2
and the Tresca yield locus is recovered for M = 1 and M → ∞. The parameter h relates the
y
uniaxial yield strength, σ22 , in the principal orthotropy direction 2 with that of direction 1;
that is,
y σy
σ22 = 11 . (10.110)
h
The effect of the choice of h on the Barlat–Lian yield surface is illustrated in Figure 10.10(a).
The effect of constant a is shown in Figure 10.10(b). Similarly to the previously discussed
428 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
M =2 M =2
y M = 1.5 M =8
σ22 / σ11 y
σ22 / σ11
M = 1.2 M = 40
1 1
M =1
-1 -1
Figure 10.9. The Barlat–Lian criterion. Yield surfaces on σ11 -σ22 plane (in the absence of shear stress)
for various values of M with a = h = 1.
Hill and Hoffman criteria, the presence of shear stresses will shrink the Barlat–Lian yield
surface. This effect is illustrated in Figure 10.11, where we plot yield surfaces obtained with
M = 8, a = b = 1 and h = 2/3 at different levels of shear stress. We remark that (not shown
in Figure 10.11) an increase (decrease) in constant b will increase (decrease) the rate at which
the surface shrinks with increasing shear stress.
For the purposes of computational implementation of the model, it turns out to be more
convenient to describe the yield criterion for the Barlat–Lian model equivalently in terms of
the following alternative definition of the yield function:
y y
Φ(σ, σ11 ) = g(σ) − σ11 , (10.111)
where
g(σ) ≡ [ 12 f (σ)]1/M . (10.112)
Note that the yield function now has dimension of stress. In the original definition (10.105),
the yield function has dimension of stress to the power M , which may produce computation-
ally intractable numbers for large values of material constant M .
Hardening law
Analogously to the Hill and Hoffman criteria, isotropic hardening can be incorporated into
y
the present model by letting σ11 be a prescribed function of a scalar strain-like hardening
internal variable, α:
y y
σ11 = σ11 (α). (10.113)
Here we take the yield function (10.111) as the plastic potential and assume the evolution of
the isotropic hardening variable to be governed by the standard relation
∂Φ
α̇ = −γ̇ y = γ̇. (10.114)
∂σ11
ADVANCED PLASTICITY MODELS 429
y
σ22 / σ11 σ22 / σ11y
h = 2/3 a=0.5
1.5 a=1.0
1.5 a=1.8
h= 1
1
1
h= 2
0.5
0.5
-1 1
y
σ11 / σ11 -1 -0.5 0.5 1 y
σ11 / σ11
-0.5 -0.5
-1
-1
-1.5
-1.5
(a) (b)
Figure 10.10. The Barlat–Lian criterion. Yield surfaces in the absence of shear for M = 8: (a) effect of
parameter h with a = 1; (b) effect of parameter a with h = 2/3.
y
σ22 / σ11
σ12 =0
1.5
y
σ12 =0.3 σ11
1 y
σ12 =0.5 σ11
0.5
-1 1
y
σ11 / σ11
-0.5
-1
-1.5
Figure 10.11. The Barlat–Lian criterion. Effect of shear stress on the yield surface for M = 8, a = 1,
h = 2/3 and b = 1.
430 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Flow rule
The Barlat–Lian yield function (10.111) is represented analogously to the plane stress-
projected von Mises yield function given in (9.41)3, page 372. The in-plane associative plastic
strain rate is given simply by
ε̇p = γ̇ N, (10.115)
where (1−M)/M
∂Φ ∂g 1 f ∂f
N≡ = = . (10.116)
∂σ ∂σ 2M 2 ∂σ
Equivalently, in component form, we have
(1−M)/M
1 f ∂f
ε̇pαβ = γ̇ , α, β = 1, 2. (10.117)
2M 2 ∂σαβ
Note that, in deriving (10.120), we have taken into account the tensorial nature of the
shear stress contribution in (10.105) (refer to the comments surrounding expression (10.63)).
The thickness plastic strain rate is obtained by imposing plastic incompressibility and the
transversal plastic shear strain rates are assumed to vanish. Thus, the complete set of plastic
flow equations comprises (10.117)–(10.120) together with the out-of plane rates
Finally, to complete the definition of the model, we assume the material to be elastically
isotropic. Then, following the procedure applied to the plane stress-projected von Mises
model, all out-of-plane strain components can be recovered a posteriori as functions of the
in-plane elastic and plastic strains according to relations (9.37)–(9.40) (refer to page 371).
ADVANCED PLASTICITY MODELS 431
Newton–Raphson iteration
As usual, the solution of equation (10.126) is undertaken by the Newton–Raphson iterative
scheme. To do this, it is convenient to split the residual vector R̄ as
. e −1 .
R [D ] : (σn+1 − σtrial ) + ∆γ Nn+1
R̄ = ; R≡ . (10.127)
Φ H −1 (κn+1 − κn ) − ∆γ
Accordingly, the typical k th Newton iteration comprises the solution of the linear system
e −1 (k−1)
δσ(k)
R(k−1)
[D ] + ∆γ ∂N/∂σ 0 N
0 1/H −1 δκ(k) =− (10.128)
(k−1)
N −1 0 δ∆γ (k) Φ
for the iterative increments of σn+1 , κn+1 and ∆γ. Following (10.116), the flow vector
derivative can be expressed as
1−M 1−2M
∂N ∂ 2 Φ 1 f M ∂2f 1−M f M
∂f ∂f
= = 2
+ 2
⊗ . (10.129)
∂σ ∂σ 2M 2 ∂σ 4M 2 ∂σ ∂σ
and . .
δσ(k) (k−1)
N
= −A(k−1) R(k−1) + δ∆γ (k) , (10.131)
δκ(k) −1
where we have defined
& '−1
[De ]−1 + ∆γ ∂N/∂σ 0
A≡ . (10.132)
0 H
With the above solution at hand, we update the Newton guess for the unknowns according to
(k) (k−1)
Σn+1 = Σn+1 + δΣ(k) , (10.133)
Line-search procedure
Dutko et al. (1993) observed that the convergence bowl of the above Newton–Raphson
scheme can be dramatically reduced for larger values of material constant M . Such an ill-
conditioned behaviour stems from the high curvature present in the Barlat–Lian model for
large M in the neighbourhood of points in stress space corresponding to the corners of the
Tresca yield surface. Under such conditions, small changes in stress on the yield surface result
in large changes in flow vector direction, characterising a set of stiff evolution equations. To
tackle the problem, Dutko et al. (1993) adopted a line-search procedure which has effectively
stabilised the Newton–Raphson scheme for the range of strain increments expected to be
present in practical finite element computations with values of constant M as high as 40.
Line-search procedures are discussed in detail by Fletcher (1980) (see also Matthies and
Strang 1979, and Crisfield 1991).
The line-search in the present case is activated whenever the rate of convergence of
the Newton–Raphson iterations falls below a prescribed minimum; that is, if for a Newton
iteration k ≥ 2 and a prescribed tolerance , the relation
is satisfied, then the line-search algorithm is carried out. The interested reader is referred to
Dutko et al. (1993) for a complete description of the algorithm.
Note that this differential relation has the same symbolic format as expression (7.141),
obtained for the von Mises isotropically hardening model. Accordingly, the elastoplastic
tangent for the Barlat–Lian model with implicit return mapping is a complete analogy
to (7.148) and can be expressed as
∂σn+1 1
Dep ≡ e trial
=P− (P : N ) ⊗ (P : N ), (10.137)
∂ε N:P:N+H
where −1
e ∂N
P≡ IS + ∆γ D : : De . (10.138)
∂σ
As a result of the associative nature of the plastic flow rule and hardening law (refer to
Section 7.4.6, page 243), the above operator is symmetric.
11 VISCOPLASTICITY
T HE elastoplastic constitutive theories presented so far in Part Two of this book are
classed as rate independent or time independent; that is, the material response is
regarded as independent of the rate of application of loads and/or the timescale of the
problems considered. Time (or, more precisely, pseudo-time) is used merely to describe the
sequence of events that defines the history of the loading process. For such theories, identical
solutions are produced when a given load (or sequence of loads) is applied at different rates.
However, the observed behaviour of real materials is generally time dependent; that is,
the stress response always depends on the rate of loading and/or the timescale considered.
The extent of such dependence may or may not be significant according to the physical
conditions of the problem. In situations where the rates of loading and/or the timescale of the
analysis remain within a range where the time-dependent phenomena can be neglected, rate-
independent elastoplasticity models can provide good descriptions of the material behaviour
(Lemaitre and Chaboche, 1990; Lubliner, 1990; Skrzypek, 1993). If such conditions are
not met, then accurate predictions can only be obtained if rate dependence is adequately
accounted for by the constitutive model. Rate-dependence effects are described by means of
so-called viscoplasticity (or rate-dependent plasticity) models, to which the present chapter
is devoted.
This chapter is organised as follows. Section 11.1 presents a brief introduction to phe-
nomenological aspects of viscoplasticity. It motivates the establishment of a one-dimensional
mathematical model of viscoplasticity in Section 11.2. Here, some simple analytical solutions
are presented to demonstrate the ability of the one-dimensional model in capturing the
fundamental phenomenological features of viscoplastic behaviour. In Section 11.3 the one-
dimensional viscoplastic theory is generalised to the multidimensional case within the context
of von Mises plasticity. A more general multidimensional model is presented in Section 11.4.
The general model can be rigorously described within the constitutive framework of internal
variable theories initially referred to in Chapter 3 (Section 3.5.2, from page 71). Rate-
independent plasticity is shown to be, under some circumstances, a limit case of the general
viscoplasticity model. This establishes a formal link between rate-independent plasticity
and the general constitutive framework of Chapter 3. Section 11.5 proceeds to introduce
a numerical framework to treat the general viscoplasticity model within the finite element
environment of Chapter 4. This includes the numerical integration algorithm for the general
viscoplastic constitutive equations as well as a symbolic form of the associated consistent
tangent modulus. Then, in Section 11.6, the general numerical framework is specialised
to a von Mises-based model presented in Section 11.3. The integration algorithm and
the associated consistent tangent operator are derived step by step. In addition, an error
assessment of the numerical integration procedure is presented by means of iso-error maps.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
436 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
We remark that Section 11.6 is essential for the reader interested in the computational
implementation of viscoplasticity. The chapter ends with finite element examples being
shown in Section 11.7. In the reported examples, the procedures of Section 11.6 are used.
σ ε3
ε2
ε1
ε1 < ε2 < ε3
ε
(a)
ε high moderate σ
stress stress
constant strain
low stress
time time
(b) (c)
Figure 11.1. Viscoplasticity. Phenomenological aspects: uniaxial tensile tests at high temperature.
(a) Strain rate dependence. Uniaxial tests at different strain rates. (b) Creep. Plastic flow at constant
stress. (c) Relaxation. Stress decay at constant strain.
ε = εe + εp . (11.1)
438 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
where the material constants are the viscosity-related parameter µ, whose dimension is time,
and the non-dimensional rate-sensitivity parameter, . Both parameters are strictly positive.
This particular form has been introduced by Perić (1993) similarly to the power law form of
the viscoplastic potential proposed by Perzyna (1963). It is important to emphasise that the
material parameters µ and are temperature dependent. As a general rule, as temperature
increases (decreases) µ and increase (decrease). For many metals, µ, → 0 for sufficiently
low temperatures, when the material behaviour may be assumed rate-independent.
σy = σy (ε̄p ) (11.7)
ε = εe + εp
5. Hardening law
σy = σy (ε̄p )
ε̄˙p = γ̇
instantaneous loading is formally demonstrated in the next example when the strain rate
dependence of the stress response is discussed. However, even without a formal proof, it
makes sense to accept that, as there is no time for plastic strains to develop over an (idealised)
instantaneous loading event, the behaviour must be purely elastic under such a condition.
Assuming zero initial plastic strain, it follows from the elastoplastic split of the total strain
together with the elastic law that the total strain in the bar at t = 0, immediately after the
instantaneous application of load, will be
σ
ε0 = εe0 = , (11.9)
E
where the zero subscript denotes quantities at t = 0. From this moment on, the bar is
kept under a constant stress above the yield limit. Under constant stress, the elastic law
implies that the elastic strain will also remain constant. Thus, the straining of the bar after
the instantaneous loading will be due purely to viscoplastic flow and will be modelled by
constitutive equations (11.5), (11.6). Assuming that σ is positive (tensile), we then have
1/
1 σ
ε̇p = −1 . (11.10)
µ σy
For a perfectly viscoplastic material (constant σy ), the integration of the above equation,
in conjunction with the elastoplastic decomposition of the total strain and the initial condi-
tion (11.9), gives the following solution for the straining of the bar
1/
σ 1 σ
ε(t) = + − 1 t. (11.11)
E µ σy
VISCOPLASTICITY 441
The creep rate in this case is constant and proportional to (σ/σy )1/ − 1. The material
constants (µ and ) could be calibrated, for instance, so as to capture the initial branches
of the creep curves of the material (refer to Figure 11.1(b)). The initial branches describe
the phenomenon of primary creep. A hardening law could be incorporated to include the
follow-up of the curves to their second straight branch that describes secondary creep.
or, equivalently,
σ=Eε if ε < ε∗ , (11.13)
∗
where ε is the strain at which the yield stress is reached
σy
ε∗ = . (11.14)
E
Viscoplastic flow (and rate-dependent behaviour) may only take place when σ ≥ σy or,
in terms of the applied strain, when ε ≥ ε∗ . Then, as our purpose is to illustrate the rate
dependence predicted by the model, our initial value problem will be defined only over
the portion of the loading process where σ ≥ σy (or ε ≥ ε∗ ). To simplify the problem, we
will assume that the material is perfectly viscoplastic (constant σy ) and, in addition, the rate
sensitivity parameter will be set to
= 1,
so that an analytical solution to the initial value problem can be easily found.
corresponding to monotonic stretching with arbitrary (constant) strain rate α ≥ 0. Note that
t = 0 corresponds to the onset of viscoplastic flow (σ = σy ⇔ ε = ε∗ ). The initial condition
for εp (the plastic strain at the onset of viscoplastic flow) is obviously
εp (0) = 0. (11.17)
By placing the above solution together with (11.16) into the elastic law, σ = E(ε − εp ), we
obtain, after a straightforward manipulation, the following solution for the stress as a function
of time −t
σ(t) = σy 1 + α µ 1 − e µ ε∗ . (11.20)
which gives the stress–strain curve for an arbitrary strain rate α. Insight into the problem can
be gained by looking into the limit stress–strain curves, i.e. the curves obtained at infinitely
slow rates (α → 0) and infinitely fast processes (α → ∞). In order to obtain the limit for
infinitely slow rates, we first observe that, in the present monotonic loading process, the term
1 − ε/ε∗
in (11.22) is always negative. In addition, µ and α are positive so that the last term on the
right-hand side of (11.22) is the exponential of a negative number. The limit is then easily
obtained as
lim σ̃(ε) = σy ; (11.23)
α→0
VISCOPLASTICITY 443
αµ = 2.0
αµ oo αµ = 1.0
2
σ / σy
non-dimensional stress,
αµ = 0.5
αµ = 0.1
1
rate-independent limit: αµ 0
lope
tic s
elas
0
0 1 5 10
relative strain, ε / ε∗
Figure 11.2. One-dimensional viscoplasticity model. Analytical solution showing the dependence of
the stress response on the applied strain rate/viscosity parameter.
that is, at infinitely slow rates the perfectly viscoplastic model rigorously recovers the
behaviour of the rate-independent plasticity model with yield stress σy . The rate-independent
model was described in Section 6.2 (page 141). It is also very important to note that the same
limit is obtained for the vanishing viscosity parameter, i.e. when µ → 0. At infinitely fast
rates, the limit is derived by a standard limiting procedure which gives
i.e. the process is purely elastic and the stress–strain curve after the yield limit is the
continuation (with the same slope, E) of the elastic curve. Also note that the identical limit
is found for µ → ∞ (infinitely viscous material). For any other rate (or viscosity parameter),
the corresponding stress–strain curve will lie between these two limits with higher stress
obtained at higher strain rates (or higher viscosity). To illustrate better the behaviour of the
model under the present conditions, the analytical solution (11.22) is shown in the graph of
Figure 11.2, where the non-dimensional stress, σ/σy , is plotted against the relative strain,
ε/ε∗ , for various normalised strain rates µα. The limits µα → 0 (infinitely slow rates or non-
viscous material) and µα → ∞ (infinitely fast rates or infinitely viscous material) are also
included. Clearly, the model is able to capture the experimentally observed rate-dependence
phenomenon illustrated in Figure 11.1(a).
Remark 11.1. In fact, even though it is not formally shown here, the above limits remain
valid for any hardening curve and any rate sensitivity parameter ; that is, at infinitely slow
strain rates, the model recovers the rate-independent behaviour of the plasticity model of
Section 6.2 (this limit is also obtained for µ → 0) and, at infinitely fast rates (or when µ →
∞), the model behaves in a purely elastic manner, regardless of the given hardening curve
and rate-sensitivity parameter. In addition (again not formally shown in this section), the
444 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
εe0 = ε, (11.25)
σ0 = E εe0 = E ε. (11.26)
From this point on, the stress in the bar will be governed by the law
σ = E(ε − εp ) = σ0 − E εp , (11.27)
where εp evolves in time according to the differential equation (11.10) which, in view of the
above expression can be equivalently written as
1/
1 σ0 − E εp
p
ε̇ = −1 . (11.28)
µ σy
To simplify the problem, we will assume, as in the previous example, that the material is
perfectly viscoplastic (constant σy ) and = 1. In this case, the initial value problem is to find
a function εp (t) such that
ε̇p (t) = c1 − c2 εp (t), (11.29)
with initial condition
εp (0) = 0, (11.30)
where the constants c1 and c2 are defined as
1 σ0 E
c1 = −1 , c2 = . (11.31)
µ σy µ σy
σ0
stress
σy
0
0 1 2 3 4 5
non-dimensional time, E t / µσ y
Figure 11.3. One-dimensional viscoplasticity model. Analytical solution to the stress relaxation
problem.
Finally, by placing the above solution into (11.27), and taking into account the definition of
c1 and c2 , we obtain the stress as a function of time
− µE
σy t
σ(t) = σ0 − (σ0 − σy )(1 − e ). (11.33)
Clearly, the above function describes the stress relaxation process of the bar, with the stress
taking the value σ = σ0 > σy at t = 0 and subsequently relaxing asymptotically to σy as
t → ∞. This is illustrated in Figure 11.3 where a graph of the analytical function σ(t) (with
σ0 = 2 σy ) is plotted. The analytical solution with the present one-dimensional model clearly
captures the experimentally observed behaviour referred to in Figure 11.1(c).
ε = εe + εp
ε̄˙p = γ̇
that is, the elastoplastic split of the total strain, the linear elasticity law, the flow rule and
yield function are recast in terms of the corresponding tensor quantities (total, elastic and
plastic strain tensors, stress tensor and flow vector). The yield function is also redefined as a
function of variables of appropriate tensorial order. The usual concept of an elastic domain
bounded by a yield surface in the rate-independent theory will remain valid in the viscoplastic
case. Here, as our multidimensional extension is von Mises-based, the yield function and the
plastic flow rule (including the hardening internal variable) will have the same format as
that of the standard rate-independent von Mises model with (associative) Prandtl–Reuss flow
vector. The resulting model is a viscoplastic version of the rate-independent isotropically
hardening von Mises model summarised in Section 7.3.1 (from page 216). The constitutive
equations of the von Mises-based extension are listed in Box 11.2.
Remark 11.2. Analogously to its rate-independent counterpart, under uniaxial stress condi-
tions, the model of Box 11.2 reduces exactly to the one-dimensional theory of Box 11.1. It is
also important to emphasise that the basic properties of creep, stress relaxation and strain-rate
dependence of the stress response (including the behaviour at limits) as demonstrated for the
VISCOPLASTICITY 447
uniaxial model in Section 11.2.7 are reproduced by the multidimensional theory under any
state of stress.
Bingham model
The Bingham model is the simplest model of viscoplasticity. The multiplier γ̇ in this case is
defined as
1 Φ(σ, σy ) = q(σ) − σy if Φ(σ, σy ) ≥ 0
η η
γ̇(σ, σy ) = (11.37)
0 if Φ(σ, σy ) < 0.
The only material constant in this case is the (temperature-dependent) viscosity parameter η
and the strain rate is modelled as a linear function of the von Mises effective stress. Note
that this law is obtained from Perić’s model given in item 4 of Box 11.2 (and also from the
Perzyna model described below) by setting
η
= 1; µ = . (11.38)
σy
In the uniaxial case, the plastic strain rate for the Bingham model is a linear function of the
axial stress:
1
ε̇p = (|σ| − σy ) sign(σ). (11.39)
η
This may severely limit the ability of the model to fit experimental data as, in many cases,
the observed strain rate may be a markedly nonlinear function of the stress. However, over
a relatively narrow range of stresses, the linear approximation may give good results. Other
models, with more material constants, have, in general, better flexibility to allow a wider
range of experimental data to be fitted.
448 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Perzyna model
This model was introduced by Perzyna (1966, 1971) and is widely used in computational
applications of viscoplasticity. It is defined by
1/
1 q(σ) − 1 if Φ(σ, σy ) ≥ 0
γ̇(σ, σy ) = µ σy (11.40)
0 if Φ(σ, σy ) < 0.
As in Perić’s model, the material constants are the viscosity-related parameter, µ, and the rate
sensitivity, . We remark here that, in spite of its similarity to Perić’s definition, as the rate-
independent limit is approached with vanishing rate-sensitivity → 0 (refer to Remark 11.1
on page 443), the Perzyna model does not reproduce the uniaxial stress–strain curve of the
corresponding rate-independent model with yield stress σy . As shown by Perić (1993), in
this limit, the Perzyna model produces a curve with σ = 2 σy instead. However, for vanishing
viscosity (µ → 0) or vanishing strain rates, the response of both Perzyna and Perić models
coincide with the standard rate-independent model with yield stress σy .
theory. Viscoplasticity models without a yield surface have been used widely, especially in
the analysis of creep and hot metal forming operations. Within the present framework, such
models can be defined simply by postulating the explicit function for γ̇ accordingly.
where N and λ are temperature-dependent material constants. Clearly, plastic flow is assumed
to occur whenever σ
= 0. Its multidimensional generalisation, sometimes referred to as
Odqvist’s law, is obtained by simply replacing the definition of the function for γ̇ in item
4 of Box 11.2 with the following
N
q(σ)
γ̇(σ) = . (11.45)
λ
Here, plastic flow takes place for any stress with non-zero deviator. Note that, by setting
σy = 0 in (11.37) the Bingham model recovers the Norton law with N = 1 and λ = η.
Lemaitre–Chaboche law
A modification of Norton’s law in order to improve its ability to model secondary creep over a
wider range of stresses and strain rates is provided by the Lemaitre–Chaboche law (Lemaitre
and Chaboche, 1990). The function γ̇ in this case reads
N
q(σ)
γ̇(σ) = exp[α q(σ)N +1 ]. (11.46)
λ
In addition to the material parameters N and λ required by Norton’s law, the present model
has a third (also temperature-dependent) parameter α.
where t and T denote, respectively, the time and absolute temperature and fσ , ft and fT are
experimentally defined functions. A comprehensive list of proposed empirical functions is
given by Skrzypek (1993), to which the interested reader is referred. For instance, fσ could
be Norton’s law or the Lemaitre–Chaboche relation above. The temperature function fT is
450 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
ε̇p = G(σ, A)
(11.50)
α̇ = J(σ, A);
that is, the plastic strain rate and the evolution law for the set α of hardening internal variables
are defined by means of the explicit constitutive functions G and J of σ and the set A of
hardening thermodynamic forces. In addition, as we have seen above, an elastic domain may
not exist. Thus, a yield function is not necessarily present in the viscoplastic formulation. The
constitutive equations of the general viscoplasticity model are listed in Box 11.3.
Note that the von Mises-based model of Box 11.2 (which incorporates an elastic domain)
is trivially recovered by defining the functions G and J as well as the free-energy potential
ψ and the internal variable set α accordingly. The same applies to all other models (with or
without an elastic domain) described in Section 11.3.
ε = εe + εp
2. Free-energy function
ψ = ψ(εe , α)
where α is a set of hardening internal variables
3. Constitutive equation for σ and hardening thermodynamic forces A
∂ψ ∂ψ
σ = ρ̄ A = ρ̄
∂ εe ∂α
4. Plastic flow rule and hardening law
ε̇p = G(σ, A)
α̇ = J(σ, A)
that the model of Box 11.3 is a particular case of the general purely mechanical infinitesimal
constitutive law defined by (3.165) on page 76. The general viscoplasticity model is obtained
by simply defining the set α of (3.165) as composed of the plastic strain tensor and the set of
hardening internal variables (as described in Section 6.3.2) and then introducing the explicit
constitutive functions for the rates of plastic strain and hardening variables listed in item 4 of
Box 11.3.
from which, through the hypothesis of normal dissipativity, the evolution of the internal
variables of the problem are derived as
∂Ξ
ε̇p =
∂σ
(11.52)
∂Ξ
α̇ = − .
∂A
At this point, it is important to recall that the plastic dissipation in the present case is given
by (again, refer to Section 6.3.2)
σ : ε̇p − A ∗ α̇,
452 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
where
Υp (σ, A; ε̇p , α̇) ≡ σ : ε̇p − A ∗ α̇ (11.54)
is the dissipation function.
By defining Ξ such that it is convex with respect to both variables, non-negative and zero-
valued at {σ, A} = {0, 0} it is ensured that the dissipation inequality is satisfied a priori by
the model.
In rate-independent plasticity, A is the set of all admissible states (σ, A) of stress and
hardening thermodynamical forces. The set A is convex, i.e. it defines a convex region in the
space of stresses and hardening forces. Following the above considerations, we now introduce
the indicator function, ΨA , of the convex set A as the scalar-valued function defined by
0 if (σ, A) ∈ A
ΨA (σ, A) = (11.56)
∞ if (σ, A) ∈ / A.
At this point we need to make use of the concept of subdifferential.§ In view of the non-
differentiability of the indicator function, the constitutive equations (11.52), which follow
from normal dissipativity, are replaced with the subdifferential relations
ε̇p ∈ ∂σ ΨA
(11.58)
α̇ ∈ −∂A ΨA ,
This last inequality states that, among all states (σ∗ , A∗ ) ∈ A, the actual stress and hardening
force (σ, A) maximise the dissipation function. This is known as the principle of maximum
plastic dissipation, discussed in Section 6.5.2 (page 170) to which the reader is referred for
details. The solution to the maximisation problem associated with the principle of maximum
plastic dissipation is the classical associative laws
∂Φ
ε̇p = γ̇
∂σ
(11.61)
∂Φ
α̇ = −γ̇ ,
∂A
together with the loading/unloading conditions of rate-independent plasticity
In summary, it has been shown above that the classical rate-independent associative plasticity
equations are rigorously recovered from the general viscoplasticity model when the indicator
function of the set A is taken as the dissipation potential.
where
A = {σ | q(σ) − σy ≤ 0}. (11.66)
This completes the demonstration. The schematic illustration of Figure 11.4 shows the
potential Ξ for various choices of the rate-sensitivity parameter . Clearly, as → 0, Ξ tends
to the indicator function of A.
Ξ ∋
0
small
∋
large
∋
0 1
q (σ) σ y
elastic domain
Problem 11.1 (The viscoplastic constitutive initial value problem). Given the initial
values εe (t0 ) and α(t0 ) and given the history of the strain tensor, ε(t), t ∈ [t0 , T ], find the
functions εe (t) and α(t), for the elastic strain tensor and hardening internal variable set that
satisfy the reduced general viscoplastic constitutive equations
Box 11.4. Fully implicit algorithm for numerical integration of general viscoplastic
constitutive equations.
Given the strain and time increment, ∆ε and ∆t, over [tn , tn+1 ] and the state variables at tn ,
compute the updated state by solving the nonlinear system of equations
e
εn+1 − εn − ∆ε + ∆t G(σn+1 , An+1 ) 0
e
=
αn+1 − αn − ∆t J(σn+1 , An+1 ) 0
(i) Elastic predictor. Given ∆ε and the state variables at tn , evaluate the elastic trial state
εen+1
trial
= εen + ∆ε
αtrial
n+1 = αn
trial trial
∂ψ ∂ψ
σtrial
n+1 = ρ̄ , Atrial
n+1 = ρ̄
∂ εe n+1 ∂ α n+1
(ii) Check for viscoplastic flow
trial
n+1 , An+1 ) ≤ 0
IF Φ(σtrial
THEN set (·)n+1 = (·)trial
n+1 and EXIT
and
∂ψ ∂ψ
σn+1 = ρ̄ , An+1 = ρ̄
∂ εe n+1 ∂ α n+1
(iv) EXIT
the considered interval. Nevertheless, the terminology viscoplastic return mapping remains
justifiable in the present case since, upon application of the procedure, the updated stress is
obtained by moving (or returning) the trial stress towards the yield surface.
Similarly to the rate-independent case (refer to Section 7.2.7, page 201), different numerical
integration algorithms can be employed in the stress updating procedure. In what follows we
list the basic equations of the generalised trapezoidal and midpoint algorithms. For further
details on alternative integration algorithms we refer to Cormeau (1975), Zienkiewicz and
Cormeau (1974), Hughes and Taylor (1978), Marques and Owen (1983), Peirce et al. (1984)
and Kojić and Bathe (1987).
458 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
where the prescribed parameter, θ, also lies within the interval [0, 1] and
Again, for θ = 1, the implicit algorithm of Box 11.4 is recovered and θ = 0 defines the explicit
algorithm.
where the symbol * denotes the product of the appropriate type and the subscripts n + 1
have been omitted for notational convenience. With the introduction of the differential
relations (7.129) (page 239), the linearised system is equivalently written as
∂G ∂G
C + ∆t B + ∆t dσ d∆ε
∂σ ∂A
.
= (11.78)
∂J ∂J
A − ∆t J − ∆t dA 0
∂σ ∂A
By inverting the linearised system above, we finally obtain a tangent relation which can be
written symbolically as
dσ D11 D12 d∆ε
= , (11.79)
dA D21 D22 0
where Dij are tensors of appropriate order resulting from the inversion of (11.78). The
consistent tangent operator we are looking for is the fourth-order tensor
dσn+1
D≡ = D11 . (11.80)
d∆ε
For models with a yield surface, the tangent modulus is elastic if the state is within the elastic
domain; that is, as in rate-independent plasticity, when Φ(σn+1 , An+1 ) ≤ 0, we have
∂2ψ
D = De = ρ̄ . (11.81)
∂εe 2
Under viscoplastic flow, i.e. when Φ(σn+1 , An+1 ) > 0, the stress is the result from the
solution of the equation system of item (iii) of Box 11.5. In this case, the tangent operator is a
specialisation of the general tangent modulus (11.80) where the functions G and J taking part
in the symbolic matrix (11.78) are defined by (11.71). The derivatives of G then specialise as
∂G ∂N ∂ γ̇
= γ̇ +N⊗
∂σ ∂σ ∂σ
(11.82)
∂G ∂N ∂ γ̇
= γ̇ +N∗ .
∂A ∂A ∂A
Similarly, the derivatives of J specialise as
∂J ∂H ∂ γ̇
= γ̇ +H∗
∂σ ∂σ ∂σ
(11.83)
∂J ∂H ∂ γ̇
= γ̇ +H∗ .
∂A ∂A ∂A
460 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
εp trial = εpn
(11.84)
ε̄p trial = ε̄pn
σtrial = De : εe trial .
If Φ(σtrial , σy (ε̄p trial )) ≤ 0, then the process is indeed elastic within the interval and
the variables at tn+1 are assigned the values of the trial variables. Otherwise, we apply
the viscoplastic return-mapping algorithm described in the following.
2. Viscoplastic return mapping. At this stage, we solve the system of discretised equations
of item (iii) of Box 11.5 which, for the present model, by taking the linear elastic law
into consideration, are specialised as
∂Φ
σn+1 = σtrial − ∆γ De :
∂σ n+1 (11.85)
p p
ε̄n+1 = ε̄n + ∆γ,
where the incremental multiplier, ∆γ, is given by
1/
∆t q(σn+1 )
∆γ = −1 , (11.86)
µ σy (ε̄pn+1 )
with ∆t denoting the time increment within the considered interval. After solv-
ing (11.85), we can update
p ∂Φ ∂Φ
p
εn+1 = εn + ∆γ e
, εn+1 = ε e trial
− ∆γ . (11.87)
∂σ n+1 ∂σ n+1
VISCOPLASTICITY 461
Single-equation corrector
The viscoplastic corrector can be more efficiently implemented by reducing (11.85) to a sin-
gle scalar equation. The situation here is completely analogous to that of the implementation
of the elastoplastic (rate-independent) von Mises model described in Section 7.3.2 (page 217).
For convenience, the steps leading to the system reduction are repeated here. Firstly, we
observe that the plastic flow vector
%
∂Φ 3 s
= (11.88)
∂σ 2 s
is deviatoric so that the hydrostatic stress is independent of the viscoplastic flow. The stress
update equation (11.85)1 can then be split as
%
3 sn+1
sn+1 = s trial
− ∆γ 2G
2 sn+1 (11.89)
trial
pn+1 = p .
Further, simple inspection of (11.89)1 shows that sn+1 is a scalar multiple of strial so that,
trivially, we have the identity
sn+1 strial
= trial , (11.90)
sn+1 s
which allows us to rewrite (11.89)1 as
%
3 ∆γ 2G trial ∆γ 3G trial
sn+1 = 1 − s = 1 − s (11.91)
2 strial q trial
where q trial is the elastic trial von Mises equivalent stress. Application of the definition of the
von Mises equivalent stress to the above equation gives the update formula
Finally, with the substitution of the above formula together with (11.85)2 into (11.86) we
obtain the following scalar algebraic equation for the multiplier ∆γ
1/
∆t q trial − 3G ∆γ
∆γ − − 1 = 0, (11.93)
µ σy (ε̄pn + ∆γ)
Box 11.6. Integration algorithm for von Mises-type viscoplastic model (over a
generic time interval [tn , tn+1 ] with ∆t = tn+1 − tn ).
(i) Elastic predictor. Given ∆ε, and the state variables at tn , evaluate the elastic trial state
εe trial := εen + ∆ε
ε̄pn+1
trial
:= ε̄pn
ptrial e trial
n+1 := K εv n+1 ; strial := 2G εed trial
n+1
qn+1 := 2 sn+1
trial 3 trial
(iv) EXIT
Remark 11.4 (Rate-independent limit). Note that, as expected, equation (11.94) rigorously
recovers its elastoplastic (rate-independent) counterpart (7.91) (refer to page 219) when µ →
0 (no viscosity), → 0 (no rate-sensitivity) or ∆t → ∞ (infinitely slow straining). Clearly, in
such cases, the algorithm of Box 11.6 reproduces the rate-independent elastoplastic numerical
solution.
the term to the power on the left-hand side can only assume values within the interval [0, 1]
and causes no numerical problems within practical ranges of material constants.
Remark 11.6 (Solution existence and uniqueness). Within a viscoplastic step, we have
Let R(∆γ) be the function defined by the right-hand side of (11.94). The above inequality
clearly implies that R(0) > 0. In addition, taking into account the strict positiveness of the
hardening function σy , we can easily verify that R(q trial /3G) < 0. The continuity of R then
implies that (11.94) has a root within the interval (0, q trial /3G). Let us now consider the
derivative of R,
q trial − 3G∆γ ∆t
R (∆γ) = − 3G + µ − H(ε̄pn + ∆γ),
µ ∆γ + ∆t µ ∆γ + ∆t
where H is the derivative of the isotropic hardening function σy . Upon simple inspection, we
can easily establish that the derivative R is strictly negative for ∆γ ∈ (0, q trial /3G) if the
viscoplastic model is non-softening, i.e. if H is non-negative for any value of accumulated
plastic strain. The strict negativeness of R in conjunction with the existence of a root for R
established in the above implies that the root of R (the solution of the viscoplastic corrector
equation) within the interval (0, q trial /3G) is unique for non-softening materials.
µ ε̇
set respectively to 1 and 1000. For each non-dimensional rate, three values of rate-sensitivity
parameter, , have been used: 100 , 10−1 and 0. Recall that for = 0 the algorithm reproduces
the rate-independent solution. The resulting map in this case is obviously identical to the
rate-independent map of Figure 7.7(b) and is shown here only to emphasise the effect of
rate-dependence on the integration error. The main conclusion drawn from the iso-error
maps is that, in general, increasing (decreasing) rate-sensitivity and/or increasing (decreasing)
strain rates tend to produce decreasing (increasing) integration errors. The largest errors are
expected in the rate-independent limit.
464 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
6 6
5%
4 10% 4
∆σN / q
∆σN / q
10% 15%
2 2
15%
20%
5%
0 0
0 2 4 6 0 2 4 6
∆σT / q ∆σT / q
(a) (b)
6
5%
10% 4
∆σN / q
15%
2
20%
0
0 2 4 6
∆σT / q
(c)
Figure 11.5. Iso-error maps with µ ε̇ = 1; (a) = 100 ; (b) = 10−1 ; (c) = 0 (rate-independent).
0.4%
0.3% 4 4
∆σN / q
∆σN / q
0.2%
5%
2 2
0.1%
7%
7% 9% 10%
5% 9%
1%
0 0
0 2 4 6 0 2 4 6
∆σT / q ∆σT / q
(a) (b)
6
5%
10% 4
∆σN / q
15%
2
20%
0
0 2 4 6
∆σT / q
(c)
Figure 11.6. Iso-error maps with µ ε̇ = 103 ; (a) = 100 ; (b) = 10−1 ; (c) = 0 (rate-independent).
∂∆γ 2G 32
trial
= ∆t − N̄ n+1 , (11.95)
∂εen+1 3G + µ ∆γ+∆t H+ µ qn+1
µ ∆γ+∆t
which is consistent with (11.94). Analogously to the elastoplastic case, this expression is
obtained by taking the differential of the viscoplastic corrector equation (11.94), having ∆γ
trial
and qn+1 as variables, and equating it to zero. With the above differential relation, the final
466 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Remark 11.7 (Rate-independent limit). By simple inspection we find that in the limits →
0 (vanishing rate-sensitivity parameter), µ → 0 (vanishing viscosity) or ∆t → ∞ (infinitely
slow straining), expression (11.96) rigorously recovers the elastoplastic consistent tangent
operator of the isotropically hardening rate-independent von Mises model with implicit return
mapping given by expression (7.120).
The implementation of the von Mises-based model with Perzyna’s viscoplastic law (11.40)
follows exactly the same procedure as described in the above except that, consistently with
the backward Euler time discretization of (11.40), the return-mapping equation (11.94) (or
item (iii) of Box 11.6) is replaced with
µ ∆γ
q trial − 3G ∆γ − 1 + σy (ε̄pn + ∆γ) = 0. (11.97)
∆t
Here, we have assumed isotropic strain hardening. Note that, as µ → ∞ (vanishing viscosity)
or ∆t → ∞ (infinitely slow process) equation (11.97) reduces to that of the elastoplastic rate-
independent von Mises model with yield stress σy . For vanishing rate sensitivity parameter,
→ 0, (11.97) reduces to a von Mises elastoplastic return-mapping equation with yield stress
2σy . This is, as one should expect, in agreement with the theoretical limits of the Perzyna
model discussed in the text immediately following equation (11.40).
trial
The differential relation between the incremental plastic multiplier and εen+1 consistent with
the return-mapping equation (11.97) reads
∂∆γ 2G 32
trial
= ∆γ
µ µ ∆γ −1
N̄n+1 , (11.98)
∂εen+1 3G + 1 + µ∆t H+ σy
∆t ∆t
VISCOPLASTICITY 467
4
16
3
normalised net stress
0 0
0 4 8 12 16 0 4 8 12 16
normalised edge deflection normalised edge deflection
(a) (b)
Figure 11.7. Double-notched specimen. Reaction–deflection diagrams: (a) = 100 ; (b) = 10−2 .
where σy is evaluated at ε̄pn+1 = ε̄pn + ∆γ. This expression is the counterpart of (11.95)
for the present implementation of Perzyna’s viscoplasticity law. The corresponding elasto-
viscoplastic consistent tangent operator is obtained following the usual procedure as
∆γ 3G
Dep = 2G 1 − trial Id
qn+1
2 ∆γ 1
+ 6G trial
− ∆γ
µ µ ∆γ −1
N̄n+1 ⊗ N̄ n+1
qn+1 3G + 1 + µ∆t H + ∆t ∆t σy
+ K I ⊗ I. (11.99)
11.7. Examples
The finite element examples presented in this section illustrate applications of the compu-
tational treatment of viscoplasticity described above. The underlying viscoplastic material
model is the one shown in Box 11.2, which includes isotropic strain hardening.
14
12
10
Deformation rates (1/s):
(a)
14
(b)
3
edge reaction (KN)
(c)
Figure 11.8. Stretching of a perforated plate. Displacement-reaction diagrams. (a) = 100 ; (b) =
10−1 ; (c) = 10−2 .
VISCOPLASTICITY 469
This choice covers very slow to very fast strain rates and is meant to demonstrate the
robustness of the integration algorithm over a wide range of strain rates. The following
material parameters are adopted
The linearly hardening case listed in Figure 7.29 is not considered here. In order to show the
effect of the rate-sensitivity parameter on the behaviour of the model, two values of are
considered
= 100 and 10−2 .
The results of the finite element simulations are presented in Figure 11.7 whose diagrams
show the evolution of the reaction forces on the constrained edge against the corresponding
edge deflection. As in the rate-independent case, the results are plotted in terms of the
normalised net stress and the normalised edge deflection defined in Section 7.5.5. The results
for = 100 and 10−2 are shown, respectively, in Figures 11.7(a) and (b). They illustrate the
expected higher reactions and limit loads for higher rates of stretching. For the lowest non-
dimensional rate of 10−4 , the rate-independent solution is recovered for any rate-sensitivity
parameter. We remark that the rate-independent solution shown in the graphs for comparison
can be obtained with the present model/algorithm simply by setting = 0 or µ = 0.
where v is the stretching velocity imposed on the nodes of the upper edge. The three
graphs of Figure 11.8 show the effect of stretching rates on the response of the plate, with
higher reactions obtained at high rates and the rate-independent solution being approached
as the stretching rate vanishes. The effects of the rate sensitivity parameter are also clearly
illustrated. At higher (lower) values of , a greater (smaller) variation of reaction as a function
of the stretching rate is produced.
12 DAMAGE MECHANICS
I NTERNAL damage can be defined as the presence and evolution of cracks and cavities
at the microscopic level which may, eventually, lead to failure – a complete loss of load-
carrying capability of the material. In many engineering applications, particularly those where
mechanical/structural components are subjected to severe service conditions, the useful life
of components is a crucial item of information which has to be carefully considered during
the design process. In such cases, the ability of the designer to predict mechanical failure
becomes an important factor. In some applications, such as in certain types of industrial
machinery, non-scheduled stops for maintenance owing to unpredicted failure may incur
serious economic consequences. In the design of manufacturing processes, such as metal-
forming operations, prediction of failure is also a crucial issue. In safety-critical applications,
frequently encountered in the aeronautical and nuclear industries, unpredicted failure may
have catastrophic effects with consequences far beyond purely economical issues.
Traditionally, the prediction of useful life/failure of materials is based on mostly empir-
ical experience accumulated over long periods of time. In some cases, failure prediction
is achieved by the systematic (and expensive) testing of real models under laboratory-
reproduced service conditions. However, with the growing knowledge of the mechanisms
of progressive internal damage that cause failure in a wide range of materials, it is becoming
possible to formulate continuum constitutive models capable of accounting for the evolution
of internal deterioration. This relatively new branch of continuum solid mechanics is known
as Continuum Damage Mechanics (CDM). This fact, allied to the fast development of
computational mechanics techniques, has made the use of computational tools to carry out
life/failure prediction a realistic alternative that can be successfully adopted in many design
and damage assessment situations.
The present chapter is devoted to computational continuum damage mechanics. Our inten-
tion here is to provide the reader with an introduction to this new and promising ramification
of computational solid mechanics that has been gaining widespread acceptance over the last
two decades. The material presented in this chapter is summarised as follows. After providing
a brief review of some basic mechanisms that characterise the presence and evolution of
damage in Section 12.1, we give in Section 12.2 a brief historical account of CDM together
with a discussion on the continuum modelling of damage phenomena. Sections 12.3, 12.4
and 12.5, describe, respectively, Lemaitre’s ductile damage model (Lemaitre, 1985b), a
simplified version of Lemaitre’s model where kinematic hardening is not considered and
Gurson’s void growth model (Gurson, 1977). In each of these sections, the computational
implementation of the corresponding constitutive models within an implicit finite element
environment is described in detail. Note that the simplified version of Lemaitre’s model
discussed in Section 12.4 is fully incorporated into program HYPLAS. Further issues, including
crack closure effects and damage anisotropy are addressed in Section 12.6.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
472 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
12.1.1. METALS
In metals, the primary mechanisms that characterise the phenomenon of mechanical degra-
dation may be divided into two distinct classes: brittle and ductile damage. Brittle damaging
occurs mainly in the form of cleavage of crystallographic planes in the presence of negligible
inelastic deformations. This behaviour is observed for many polycrystalline metals, usually at
low temperatures. At high temperatures, brittle damage can also be observed associated with
creep processes. In this case, the decohesion of interatomic bonds is concentrated at grain
boundaries. At low stresses they are accompanied by relatively small strains. Ductile damage,
on the other hand, is normally associated with the presence of large plastic deformations in
the neighbourhood of crystalline defects. The decohesion of interatomic bonds is initiated at
the boundary interface of inclusions, precipitates and particles of alloy elements leading to the
formation of microscopic cracks and cavities. Further evolution of local plastic deformation
may cause the cavities to coalesce, resulting in final rupture. This mechanism is schematically
illustrated in Figure 12.1. For most metallic materials, the damage behaviour is a combination
of brittle and ductile response and the contribution of each mode is, to a significant extent,
dependent on the temperature, loading rate, etc.
Another important mode of material deterioration in metals is fatigue damage. It is
normally observed in mechanical components subjected to a large number of load and/or
DAMAGE MECHANICS 473
straining
temperature cycles. Although fatigue damage occurs at overall stress levels below the
macroscopic plastic yield limit, the nucleation of microcracks is attributed to the accumu-
lation of dislocations observed in connection with cyclic plastic deformation due to stress
concentration near microscopic defects. A large number of complex interactive physical
mechanisms take place from the nucleation of cracks to the complete failure of the material,
and the understanding of fatigue degradation processes in metals remains a challenging issue
in the field of materials science. Some of the most important mechanisms of material damage
are described by Engel and Klingele (1981).
CDM models along with a brief historical review of this new branch of continuum mechanics
are presented below.
In the previous section, some basic microscopic mechanisms associated with internal
damage evolution in solids have been reviewed. It is clear that the underlying phenomena
which characterise damage are essentially different from those characterising deformation.
While damage manifests itself in the form of the irreversible rupture of atomic bonds, defor-
mation can be associated with reversible variations of interatomic spacing (in purely elastic
processes) and movement and accumulation of dislocations (in permanent deformations of
metals). Therefore, it should be expected that in order to describe the internal degradation of
solids within the framework of the continuum mechanics theory, new variables intrinsically
connected with the internal damage process will have to be introduced in addition to the
standard variables (such as the strain tensor, plastic strain, etc.) employed in the description
of deformation. In this context, we shall refer to Continuum Damage Mechanics Model as any
continuum constitutive model which features special internal variables representing, directly
or indirectly, the density and/or distribution of the microscopic defects that characterise
damage.
so that the phenomenon of damaging (increase in D) may produce the marked acceleration of
the plastic strain rate observed during tertiary creep (refer to Figure 11.1(b) and the discussion
in Section 11.1).
† Kachanov in fact used the material continuity or integrity, ω = 1 − D, as the variable associated with the
Since Kachanov–Rabotnov’s original developments, it did not take long before the concept
of internal damage variable was generalised to three-dimensional situations by a number
of authors. Leckie and Hayhurst (1974) exploited the idea of the effective load-bearing
area reduction as a scalar measure of material deterioration to define a model for creep-
rupture under multiaxial stresses. The theories derived later by Chaboche (1978, 1981,
1984) and Murakami and Ohno (1981) deserve special mention. Chaboche proposed a
phenomenological theory for creep-damage based on rigorous thermodynamic foundations,
in which, as a consequence of the hypothesis of strain equivalence, the damage variable
appears as a fourth-order non-symmetric tensor in the most general anisotropic case. In the
theory derived by Murakami and Ohno, the anisotropic damage variable is represented by a
second-rank symmetric tensor. In this case, the definition of the damage variable follows from
the extension of the effective stress concept to three dimensions by means of the hypothesis of
the existence of a mechanically equivalent fictitious undamaged configuration. Murakami’s
fictitious undamaged configuration concept was later extended to describe general anisotropic
states of internal damage in solids with particular reference to the analysis of elastic-brittle
materials (Murakami, 1988). Still within the context of creep-rupture, Saanouni et al. (1989)
used a non-local formulation to predict the nucleation and growth of cracks.
This theory was further elaborated on by Lemaitre (1985a,b) and ageing effects were later
incorporated by Marquis and Lemaitre (1988). More recently, the original isotropic model
has been extended by Lemaitre et al. (2000) to account for the anisotropy of damage as
well as for partial closure of microcracks under compressive stresses. The damage variable
in this case is a second-order tensor whose evolution is linked to the principal directions of
the plastic strain rate. Based on the concept of energy equivalence (as opposed to Lemaitre’s
strain equivalence) another model for elastoplastic damage worth mentioning was proposed
by Cordebois and Sidoroff (1982). The damage variable in this case takes the form of a
second-order tensor under general anisotropy. Also within the theory of elastoplasticity, Simo
and Ju (1987) proposed a framework for the development of (generally anisotropic) strain-
and stress-based damage models. In this case, Lemaitre’s hypothesis of strain equivalence and
its dual hypothesis of stress equivalence are used, respectively, in the formulation of models in
stress and strain spaces. Application of the proposed framework was made in the description
of brittle damage in concrete. The more recent developments by Armero and Oller (2000)
are also worth mentioning. These authors propose a framework whereby the strain tensor is
decomposed additively into elastic, plastic and damage parts. Damage component of ε in this
case is then split as a sum of individual contributions from various damage mechanisms.
A somewhat different approach was followed by Krajčinović and Fonseka (1981) (see also
Fonseka and Krajčinović 1981), in the derivation of a continuum damage theory for brittle
materials. Assuming that damage in this case is characterised mainly by planar penny-shaped
microcracks, a vectorial variable was proposed as the local measure of internal deterioration.
Later, in (Krajčinović, 1983), the model was endowed with a thermodynamical structure
and extended to account for ductile damage. Further developments were introduced by
Krajčinović (1985) with the distinction between active and passive systems of microcracks.
Other vectorial models are described by Kachanov (1977) and Mitchell (1990).
Continuum damage mechanics has also been applied to the description of fatigue pro-
cesses. Janson (1978) developed a continuum theory to model fatigue crack propagation
which showed good agreement with simple uniaxial experiments. A general formulation
incorporating low and high-cycle fatigue as well as creep-fatigue interaction at arbitrary
stress states has been presented by Lemaitre (1987). Further discussion on these models is
provided by Chaboche (1988) and Lemaitre and Chaboche (1990). In order to model the
effects of fatigue, the evolution law for the damage variable is usually formulated in terms of
a differential equation which relates damage growth with the mean stress, maximum stress
and number of cycles.
Physical significance
With regard to the physical significance of damage variables, it is convenient to separate the
CDM theories into two main categories: micromechanical and phenomenological models.
In micromechanical models, the damage internal variable must represent some average
of the microscopic defects that characterise the state of internal deterioration. Despite the
physical appeal of internal variables such as the reduction of load-bearing area, as suggested
by Rabotnov (1963), or distribution of microcracks, as adopted by Krajčinović (1983, 1985)
in his vectorial model, the enormous amount of bookkeeping required in conjunction with the
serious difficulties involved in the experimental identification of damaged states and evolution
laws preclude most micromechanical theories from practical applications at present. This is
especially true if the final objective is the analysis of large-scale problems for engineering
design purposes.
Phenomenological damage variables, on the other hand, can be defined on the basis of
the influence that internal degradation exerts on the macroscopic properties of the material.
In particular, properties such as the elastic moduli (Cordebois and Sidoroff, 1979; Horii and
Nemat-Nasser, 1983), yield stress, density and electric resistance can be strongly affected by
the presence of damage in the form of microscopic cavities. Needless to say, the measurement
of such quantities is, in general, far easier than the determination of the geometry or
distribution of micro-defects. Based on such concepts, the class of models presented by
Lemaitre and Chaboche (1990) rely mostly on the use of the degradation of the elastic moduli
as the macroscopic measure of damage. In its simplest form, i.e. under ideally isotropic
conditions, the damage variable is the scalar defined by expression (12.7). Under anisotropy,
the damage variable is a second-order tensor (Lemaitre et al., 2000). A similar definition
for the isotropic damage variable is employed by Cordebois and Sidoroff (1982). A model
relying on the volume changes due to void growth as a measure of internal degradation is
described by Gelin and Mrichcha (1992).
Current methods of experimental identification of damage, comprising direct as well
as indirect techniques, are described in detail by Lemaitre and Dufailly (1987). Such
techniques range from the direct observation of microscopic pictures to the measurement
of the degradation of the elastic moduli by means of ultrasonic emissions and micro-hardness
tests. The potential as well as the limitations of both micromechanical and phenomenological
approaches to damage mechanics are discussed by Basista et al. (1992). In the present state of
development of CDM it has been verified that, in general, the loss of microscopic information
resulting from a phenomenological approach is compensated for by the gain in analytical,
experimental and computational tractability of the model.
Mathematical representation
In view of the many possibilities regarding the choice of the damage internal variable,
Leckie and Onat (1981) showed that the distribution of voids on the grain boundaries can
be mathematically represented by a sequence of even rank irreducible tensors. Although this
result was obtained in the context of creep-damage theories, Onat (1986) showed that the
same phenomenological representation for the damage variable applies to general micro-
cracked continua, regardless of the underlying deformation processes.
The conclusions drawn by Onat were based on the use of averaging techniques to
transform the distribution of micro-defects into a mathematically well-defined continuum
478 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
where De is the standard isotropic elasticity tensor. For this particular potential, the elasticity
law is given by
∂ψ
σ = ρ̄ e = (1 − D)De : εe . (12.11)
∂ε
Equivalently, the above damaged elastic law can be written as
σeff = De : εe , (12.12)
where σeff is the effective stress tensor that generalises the uniaxial effective stress of (12.2):
1
σeff ≡ σ. (12.13)
1−D
The thermodynamical force conjugate to the damage internal variable is given by
∂ψ
Y ≡ ρ̄ = − 21 εe : De : εe , (12.14)
∂D
or, using the inverse of the elastic stress/strain law,
−1
Y = σ : [De ]−1 : σ
2(1 − D)2
−1
= [(1 + ν) σ : σ − ν (tr σ)2 ]
2E(1 − D)2
2
−q 2 2 p
= (1 + ν) + 3(1 − 2ν)
2E(1 − D) 3
2 q
−q 2 p2
= − , (12.15)
6G (1 − D)2 2K (1 − D)2
where E and ν are, respectively, the Young’s modulus and Poisson ratio associated with G
and K. In the above, p is the hydrostatic stress and q is the von Mises effective stress.
Commonly known as the damage energy release rate, −Y corresponds to the variation of
internal energy density due to damage growth at constant stress. It is the continuum damage
analogue of the J-integral used in fracture mechanics (Rice, 1968). The product −Y Ḋ
represents the power dissipated by the process of internal deterioration (mainly as decohesion
of interatomic bonds).
Remark 12.1. Stress–strain rule (12.11) has an important experimental consequence. With
the elasticity–damage coupling introduced via the hypothesis of strain equivalence (stated in
Section 12.2.2), the effective elastic modulus of the material, which can be measured from
experiments, is given by
Deff = (1 − D)De , (12.16)
where the damage variable assumes values within the interval [0, 1]. In the absence of damage
(D = 0), the effective modulus equals the modulus De of the virgin material. For a completely
damaged state (D = 1), Deff = 0, corresponding to a total loss of stiffness and load bearing
capacity of the material. The identification of a generic damaged state, with D ∈ [0, 1], is
then restricted to the measurement of the degradation of the current effective elastic modulus
with respect to the virgin state (D = 0) as described by Lemaitre (1985a) and Lemaitre and
Chaboche (1990).
480 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
∂ψ p (R, X) ∂ψ I (R)
κ ≡ ρ̄ = ρ̄ = κ(R). (12.18)
∂R ∂R
From (12.17), it follows that the thermodynamic force associated with kinematic hardening
– the back-stress tensor, β – is given by
∂ψ
β ≡ ρ̄ = a X. (12.19)
∂X
where a, b, r and s are material constants. The damage evolution constants r and s can be
identified by integrating the damage evolution law for particular cases of (constant) stress
triaxiality rate as described in Section 7.4 of Lemaitre and Chaboche (1990). The constants
a and b, associated with the resulting Armstrong–Frederick kinematic hardening law (refer
to the text surrounding expression (6.223), page 188, for further details) and can be obtained
from cyclic loading experiments (Lemaitre and Chaboche, 1990). The convexity of the flow
potential Ψ with respect to the thermodynamical forces for positive constants a, b, r and s
ensures that the dissipation inequality is satisfied a priori by the present constitutive model.
The constitutive equations of Lemaitre’s ductile damage model are conveniently grouped in
Box 12.1.
‡ Refer to Subsections 6.6.3 and 6.6.4, from pages 182 and 185, for the definitions of individual contributions
ε = εe + εp
σ = (1 − D)De : εe
ε̇p = γ̇ N
Ṙ = γ̇
β̇ = γ̇ (a N − b β)
s
1 −Y
Ḋ = γ̇
1−D r
Φ ≤ 0, γ̇ ≥ 0, Φ γ̇ = 0
Damage threshold
At low values of accumulated plastic strain, ε̄p , elastic modulus degradation can be hardly
detected in experiments. Thus, we can assume that damage growth starts only at a critical
value, denoted ε̄pD . This critical value will be called the damage threshold and can be included
in the model by redefining the damage evolution law of item (iv) of Box 12.1 as
s
Ĥ(ε̄p − ε̄pD ) −Y
Ḋ = γ̇ , (12.22)
1−D r
where Ĥ here denotes the Heaviside step function defined by (7.94) on page 221.
If such a threshold is adopted, then the evolution law for ε̄p has to be defined for the model
to be complete. From its definition (we recall expression (6.168)) we have
ε̄˙p = 2
3 ε̇p . (12.23)
482 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
(iv) EXIT
By taking the plastic flow rule for the present model into consideration, the above equation
results in the following evolution law for the accumulated plastic strain:
γ̇
ε̄˙p = . (12.24)
1−D
σ2 / q
∆σN / q
1.0
A
∆σT / q
B
1.0 σ1 / q
Figure 12.3. Iso-error maps for Lemaitre’s damage model in plane stress. Departure points.
1. If ε̄pn < ε̄pD , perform the calculations initially without damage evolution; that is, set the
trial updated damage
Dn+1 := Dn = 0,
and solve the equation system of item (ii) of Box 12.2 with the damage equation
(the last equation of the system) ‘switched off’. Then compute the trial updated
accumulated plastic strain
ε̄pn+1 := ε̄pn + ∆γ.
Now check the validity of the trial updated values:
(a) If ε̄pn+1 < ε̄pD , the trial values are accepted as the updated values.
(b) Otherwise, solve the return-mapping equations as in Box 12.2.
2. If ε̄pn ≥ ε̄pD , then the calculations follow the procedure of Box 12.2.
D=0 D = 0.433
6. 6.
5. 5.
4. 4.
∆σT q
∆σT q
3. 3.
2. 2.
1. 1.
0. 0. .
0. 1. 2. 3. 4. 5. 6. 0. 1. 2. 3. 4. 5. 6.
∆σN q ∆σN q
(a)
6. 6.
5. 5.
4. 4.
∆σT q
∆σT q
3. 3.
2. 2.
1. 1.
0. 0.
0. 1. 2. 3. 4. 5. 6. 0. 1. 2. 3. 4. 5. 6.
∆σN q ∆σN q
(b)
Figure 12.4. Lemaitre’s model in plane stress. Iso-error maps at D = 0 and D = 43.3%. (a) Biaxial
state – point A; (b) Uniaxial state – point B. Contour intervals = 4%.
Φ(dev[σproj ], Rn , βn , Dn ) = 0,
with
n+1 − βn ]
dev[σtrial
dev[σproj ] = + βn ,
1 + 32 ∆γ
This procedure stabilises the local Newton–Raphson scheme and assures convergence at any
stage of damage evolution for relatively large increments of elastic trial stress.
The above tangent also applies when the current state is on the yield surface and elastic
unloading is assumed to occur. Under plastic straining, σn+1 is delivered as the solution
of the nonlinear system of the plastic corrector stage (item (ii)). In this case, the nonlinear
system is differentiated leading to the linear form
A1,σ A1,D A1,∆γ A1,β dσn+1 0
trial
A2,σ A2,D A2,∆γ A2,β dDn+1 (1 − Dn+1 )De : dεen+1
= (12.27)
A
3,σ A3,D A3,∆γ A3,β d∆γ 0
A4,σ A4,D A4,∆γ 0 dβn+1 0
where the coefficients A1,σ , A1,D , . . . , are the partial derivatives of the left-hand sides
of item (ii) with respect to the system variables computed at the converged solution of
the nonlinear system of equations of the plastic corrector procedure. Note that the same
coefficients matrix is computed for each trial solution obtained during the Newton–Raphson
iterations of the plastic corrector stage. Inversion of (12.27) gives the tangent relations
trial
between the system variables (σn+1 , Dn+1 , ∆γ and βn+1 ) and εen+1
dσn+1 C11 C12 C13 C14 0
trial
dDn+1 C21 C22 C23 C24 (1 − Dn+1 )De : dεen+1
.
d∆γ = C C34 (12.28)
31 C32 C33 0
dβn+1 C41 C42 C43 C44 0
486 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
(12.30)
ptrial
n+1 := (1 − D n ) K ε e trial
v n+1 .
The elastic trial value of the yield function in the present case is then evaluated as
Φtrial := q̃n+1
trial
− σy (Rn ), (12.31)
DAMAGE MECHANICS 487
ε = εe + εp
σ = (1 − D)D : εe
ε̇p = γ̇ N
Ṙ = γ̇
s
1 −Y
Ḋ = γ̇
1−D r
Φ ≤ 0, γ̇ ≥ 0, Φ γ̇ = 0
where we have defined the effective elastic trial von Mises equivalent stress,
trial
qn+1 3J2 (strial
n+1 )
strial
n+1
trial
q̃n+1 ≡ = = 3
21−D . (12.32)
1 − Dn 1 − Dn n
As usual, if Φtrial ≤ 0 the process is elastic within the step and the elastic trial state coincides
with the updated state at tn+1 . Otherwise, we apply the single equation return-mapping
procedure derived in the following.
The implicit return-mapping equations for a generic plasticity model are given by (7.25),
page 196. In the present case, with the model defined in Box 12.3, the set of return-mapping
488 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Expression (12.34)2, together with the elastic law, gives the following updating relation
for the hydrostatic pressure:
pn+1 = (1 − Dn+1 ) p̃n+1 , (12.35)
where we have defined
p̃n+1 = K εev n+1 = K εev trial
n+1 . (12.36)
Let us now consider the stress deviator updating formula. With the introduction of the
elastic law into (12.34)1, it follows that
%
3 sn+1
sn+1 = (1 − Dn+1 ) 2G εd n+1 = (1 − Dn+1 ) 2G εd n+1 − 2G ∆γ
e e trial
2 sn+1
and we obtain the update equation for the stress deviator:
%
3 sn+1
sn+1 = (1 − Dn+1 ) s̃trial − 2G ∆γ , (12.37)
n+1
2 sn+1
where we have defined
n+1 ≡ 2G εd n+1 .
s̃trial e trial
(12.38)
From (12.37), it is clear that s̃trial
is proportional to sn+1 so that we may equivalently write
n+1
%
3 s̃trial
n+1
sn+1 = (1 − Dn+1 ) s̃n+1 − 2G ∆γ
trial
. (12.39)
2 s̃trial
n+1
After a straightforward manipulation, the above equation yields the following simpler update
formula for sn+1 :
3G ∆γ trial
sn+1 = 1 − Dn+1 − trial s̃n+1 . (12.40)
q̃
From this last expression and the definition of the von Mises equivalent stress, we obtain
qn+1 = (1 − Dn+1 )q̃ trial − 3G ∆γ. (12.41)
DAMAGE MECHANICS 489
Two-equation system.
By introducing (12.41) and (12.33)2 into (12.33)4, we obtain the consistency equation
3G ∆γ
Φ̃(∆γ, Dn+1 ) ≡ q̃ trial − − σy (Rn + ∆γ) = 0. (12.42)
(1 − Dn+1 )
With (12.41) and (12.35) introduced into the definition of the damage energy release
rate, (12.33)3 can be written as
s
∆γ −Y (∆γ, Dn+1 )
Dn+1 − Dn − = 0, (12.43)
1 − Dn+1 r
where
[(1 − Dn+1 )q̃ trial − 3G ∆γ]2 p̃2n+1
−Y (∆γ, Dn+1 ) ≡ + . (12.44)
6G (1 − Dn+1 )2 2K
In summary, the return mapping has been reduced to the set of two scalar equations (12.42)
and (12.43). The unknowns of this system of equations are ∆γ and Dn+1 . After solution,
with ∆γ and Dn+1 at hand, sn+1 and pn+1 are trivially updated respectively by (12.40)
and (12.35). This two-equation return mapping was proposed by Vaz Jr. (1998) in the context
of fracture prediction for metal-cutting applications. A similar two-equation algorithm was
proposed by Steinmann et al. (1994) for a variation of the simplified Lemaitre model, where
the damage energy release rate depends only on the deviatoric part of the strain energy
function. The work of Steinmann et al. (1994) was carried out within the framework of
multiplicative finite strain plasticity. Return-mapping convergence difficulties similar to those
discussed in Section 12.3.2 have been identified by these authors. To tackle the problem, they
introduced a line-search procedure within the Newton–Raphson scheme for solution of the
return-mapping equations.
Further reduction in the above system, leading to a computationally more efficient single-
equation return-mapping algorithm, is possible for the present model. The single-equation
algorithm is derived as follows. Firstly, for convenience, we define the material integrity as
ω ≡ 1 − D. (12.45)
3G ∆γ
ωn+1 ≡ 1 − Dn+1 = w(∆γ) ≡ . (12.46)
q̃ trial − σy (Rn + ∆γ)
In addition, by combining (12.42) and (12.44), the (updated) damage energy release rate may
be expressed as a function of ∆γ only, i.e. we may redefine
Finally, by combining (12.46) and (12.47) with (12.43), our return-mapping procedure is
reduced to the solution of the following scalar equation for ∆γ:
s
∆γ −Y (∆γ)
F (∆γ) ≡ ω(∆γ) − ωn + = 0. (12.48)
ω(∆γ) r
Once a solution ∆γ has been found, we update the hardening and damage variables, the
hydrostatic stress and the stress deviator using the relevant equations listed above. The
overall algorithm has been implemented in subroutine SUDAMA (State Update procedure for
the simplified Lemaitre DAMAge model) of program HYPLAS. The corresponding pseudo-
code is conveniently listed in Box 12.4. As for the other models discussed in this book, the
return-mapping equation is solved by the standard Newton–Raphson iterative scheme.
Remark 12.3. In the Newton–Raphson scheme for the iterative solution of the return
mapping equation (12.48) we adopt the following initial guess for ∆γ:
Box 12.4. Fully implicit elastic predictor/return-mapping algorithm for the simpli-
fied version of Lemaitre’s damage model.
(i) Elastic predictor. Given ∆ε and the state variables at tn , evaluate the elastic trial state
εen+1
trial
:= εen + ∆ε
trial
Rn+1 := Rn
p̃n+1 := K εev trial
n+1 ; s̃trial e trial
n+1 := 2G εd n+1
trial
q̃n+1 := 32 s̃trial
n+1 /(1 − Dn )
for ∆γ using the Newton–Raphson method. Start iterations with the initial
guess (12.49). The functions ω(∆γ) and Y (∆γ) are defined, respectively, by (12.46)
and (12.47). Then update
Rn+1 := Rn + ∆γ
pn+1 := ω(∆γ) p̃n+1 ; qn+1 := ω(∆γ) σy (Rn+1 )
qn+1
sn+1 := trial s̃trial
n+1 ; σn+1 := sn+1 + pn+1 I
q̃n+1
1
εen+1 = sn+1 + 13 εev trial
n+1 I
2G
(iv) EXIT
particular case of the original one given by (12.29) and could be obtained by setting β = 0 and
a = b = 0 in the procedure described in Section 12.3.3. However, due to the relative simplicity
of the one-equation return mapping, it is possible to derive a closed-form expression for
the elastoplastic tangent in a straightforward manner. The evaluation of the elastoplastic
tangent by means of its closed form is far simpler (and computationally more efficient)
than its calculation as a particular case of the original model. The closed-form expression
is implemented in program HYPLAS (subroutine CTDAMA). Its derivation follows the standard
application of consistent linearisation concepts to the one-equation return-mapping scheme
of Box 12.4 and we shall leave it as an exercise for the interested reader. The final closed
492 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
e = K (ωn+1 + a3 p̃n+1 ).
In the above, H denotes the updated slope of the hardening curve at tn+1 :
dσy
H= , (12.53)
dR Rn+1
and a1 , a2 , a3 and a4 are defined as
s
1 ωn+1 1 −Yn+1
a1 = trial −
F q̃n+1 − σy (Rn+1 ) 3G r
s p̃n+1 [q̃n+1 − σy (Rn+1 )] −Yn+1 s−1
trial
a2 = −
3G r K F r (12.54)
a 3 = a2 ω
ωn+1
a 4 = a1 ω − .
trial
q̃n+1 − σy (Rn+1 )
The scalar F is the derivative (at the solution ∆γ) of the return-mapping residual function
defined by (12.48)
s
H −Yn+1
F = ω −
3G r
trial
s H σy (Rn+1 )[q̃n+1 − σy (Rn+1 )] −Yn+1 s−1
+ (12.55)
9G2 r r
and ω denotes the derivative (also at the solution ∆γ) of the function defined by (12.46)
3G + ωn+1 H
ω = . (12.56)
trial
q̃n+1 − σy (Rn+1 )
DAMAGE MECHANICS 493
prescribed
displacement,u
Please Wait..
40 mm
R = 4 mm
D = 18 mm
prescribed
displacement,u
Figure 12.5. Cylindrical notched bar. Geometry, boundary conditions and finite element mesh.
Remark 12.4. In the first equilibrium iteration of any load increment, corresponding to ∆γ =
0 and q̃ trial = σy , the above expression for Dep is used in subroutine CTDAMA by setting a
small perturbation
∆γ := 10−8 .
Remark 12.5. It should also be noted that as c is generally different from d in (12.50), the
elastoplastic tangent operator for the simplified model is also unsymmetric.
E 210 GPa
ν 0.3
Hardening curve σy (R) = 620 + 3300[1 − exp(−0.4 R)] (MPa)
S 1.0
r 3.5 MPa
0.8
800
damage variable, D
0.6
axial stress (MPa)
600
0.4
400
200 0.2
0 0
0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0
axial strain axial strain
Figure 12.6. Cylindrical notched bar. Lemaitre’s model uniaxial behaviour with the present material
constants.
is used in the region surrounding the smallest cross-section. The material parameters adopted
in the analysis are listed in Table 12.1. These parameters have been calibrated by Benallal
et al. (1987) from uniaxial experiments with AISI 1010 low carbon steel in a rolled state.
The corresponding uniaxial stress–strain and damage–strain curves predicted by Lemaitre’s
model with the material constants of Table 12.1 are shown in Figure 12.6.
The analysis was carried out by applying the vertical displacement of the constrained edge
incrementally. The simulation was stopped when the first Gauss point in the structure reached
a damaged state with D > 0.999. This state was attained in 62 increments with a prescribed
edge displacement u = 0.576 mm. The evolution of the damage variable field obtained in
the finite element analysis is illustrated in the contour plots shown in Figure 12.7. It can be
seen that during the early stages of the loading process, maximum damage is detected near
the root of the notch. As the specimen is progressively stretched, the maximum damage area
moves gradually towards the centre of the specimen and localises there. At the final stage
with u = 0.576 mm, damage is highly localised around the centre. It indicates, therefore,
that fracture initiation should be expected in that area. This prediction is in agreement with
experimental observations by Hancock and Mackenzie (1976) and Cescotto and Zhu (1995)
which show that for certain notched specimen configurations, fracturing initiates at the centre
of the specimen and propagates radially towards the notch. The reason for faster damaging at
the centre lies in the fact that damage growth in ductile metals is strongly dependent on the
stress triaxiality ratio
p/q,
DAMAGE MECHANICS 495
0.00366 0.00900
0.00335 0.00824
0.00305 0.00749
0.00274 0.00674
0.00243 0.00599
0.00212 0.00523
0.00181 Please Wait..
0.00448 Please Wait..
0.00151 0.00373
0.00120 0.00298
0.00089 0.00222
0.00058 0.00147
0.00027 0.00072
0.00000 0.00000
u = 0.051 mm u = 0.076 mm
0.105 0.999
0.096 0.916
0.087 0.833
0.079 0.750
0.070 0.666
0.061 0.583
Please Wait..
0.052 0.500 Please Wait..
0.044 0.416
0.035 0.333
0.026 0.250
0.017 0.166
0.009 0.083
0.000 0.000
u = 0.246 mm u = 0.576 mm
1.0
0.8
0.4
0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6
prescribed edge displacement, u (mm)
Figure 12.8. Cylindrical notched bar. Damage variable evolution at the centre of the specimen.
which is highest at the centre of the specimen. A marked decrease in ductility occurs as
the triaxiality ratio increases. This phenomenon is captured by Lemaitre’s ductile damage
model. The evolution of the damage variable at the centre of the specimen is depicted in
Figure 12.8 where the value of D computed at that point is plotted against the prescribed
edge deflection. The corresponding reaction forces on the constrained edge of the specimen
are shown in Figure 12.9. The diamond marks along the reaction–deflection curve correspond
to the converged equilibrium states obtained at the end of every other step.
It is important to emphasise here that if the purely elastoplastic von Mises model (without
damage) is used and a critical value of accumulated plastic strain is taken as the fracture
criterion, crack initiation will be predicted at the notch root. It is, therefore, crucial that a
damage model taking into consideration the effect of stress triaxiality be used in analyses of
this type.
The constitutive model addressed in this section was proposed by Gurson (1977) to describe
the mechanism of internal damaging in the form of void growth in porous metals. The starting
point of Gurson’s theory is the microscopic idealisation of porous metals as aggregates
containing voids of simple geometric shapes embedded in a metallic matrix whose behaviour
is governed by a rigid-plastic von Mises constitutive law. Approximate functional forms for
the corresponding macroscopic yield functions are derived based on the analysis of single
void cells and use of the upper bound plasticity theorem. In contrast to Lemaitre’s damage
model, the evolution of the damage variable of Gurson’s model is not directly associated with
a dissipative mechanism. The damage variable D in this case is the void volume fraction,
i.e. the local fraction of volume occupied by voids and its evolution law follows as a direct
consequence of mass conservation.
DAMAGE MECHANICS 497
0.08
0.04
0.02
0
0 0.1 0.2 0.3 0.4 0.5 0.6
prescribed edge displacement, u (mm)
ρ̄ ψ e (εe ) = 1
2 εe : De : εe , (12.58)
∂ψ
σ = ρ̄ = De : εe . (12.59)
∂εe
As in Lemaitre’s model, the isotropic hardening contribution is left as an arbitrary function
of a single argument, so that the thermodynamic force κ associated with R is given by
∂ψ ∂ψ p
κ = ρ̄ = ρ̄ = κ(R). (12.60)
∂R ∂R
Remark 12.6. Note that, in contrast to Lemaitre’s ductile damage model, the effect of
internal damage on the elastic behaviour of the material is ignored in the present model;
that is, the elasticity tensor is not a function of the damage variable.
498 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
D + vm = 1, (12.66)
ρ = ρm (1 − D). (12.67)
The matrix material is assumed to be plastically incompressible (von Mises type). In addition,
it is assumed that elastic volumetric strains are negligible. Under such hypotheses, mass
conservation requires that
ρ̇m = 0. (12.69)
Note that in the original derivation of Gurson’s model, the matrix material is assumed to be
rigid-plastic and the above identity holds exactly. Substitution of this equation into (12.68)
gives
ρ̇ ρ̇
Ḋ = − = − (1 − D). (12.70)
ρm ρ
Now let us recall that the axiom of mass conservation requires that
ρ̇
ε̇v = − . (12.71)
ρ
Again, by disregarding the elastic volumetric strains, we assume
and, by combining the last three equations, we obtain the following evolution law for D:
Ḋ = (1 − D) ε̇pv . (12.73)
Finally, in view of the constitutive equation (12.64) for the volumetric plastic flow, the
evolution law for the damage variable is obtained as
3p
Ḋ = γ̇ (D − D ) (κ + σy 0 ) sinh
2
. (12.74)
2 (κ + σy 0 )
ε = εe + εp
σ = De : εe
where σy = σy (R)
(iv) Plastic flow and evolution equations for R and D
# $
p 1 3p
ε̇ = γ̇ s + D σy sinh I
3 2 σy
% ! & !
2
3
1 + D2 − 2 D cosh 3p
2 σy
σy + p D sinh 3p
2 σy
Ṙ = γ̇
1−D
3p
Ḋ = γ̇ (D − D2 ) σy sinh
2 σy
Φ ≤ 0, γ̇ ≥ 0, Φ γ̇ = 0
Remark 12.9. Note from (12.74) that if the initial voids ratio is zero (D = 0), no damage
evolution will be predicted by the standard Gurson model, whatever strain history the material
might be subjected to. A non-zero initial voids ratio is required to produce damage growth.
To improve the model, a mechanism of damage nucleation, whereby voids are nucleated
depending on the strain history, should be incorporated. In this context, some nucleation laws
have been proposed by Tvergaard (1982a) and Tvergaard and Needleman (1984) whereby
voids may nucleate in the absence of damage. Further acceleration in voids growth, intended
to produce a more realistic response, can be relatively easily incorporated by introducing
a modification into Gurson’s macroscopic yield criterion as suggested by Tvergaard (1981,
1982a,b).
Remark 12.10. Again from (12.74), it follows that the voids ratio growth rate is positive
(negative) under tensile (compressive) hydrostatic stress. Positive and negative voids ratio
growth rate correspond, respectively, to damaging and healing of the material. Under cyclic
loading (in the plastic range), Gurson’s model will not predict the experimentally observed
continuous increase in damage, as the increase in voids ratio predicted in the tensile part
of the cycle is reversed by the healing process predicted over the compressive part of the
DAMAGE MECHANICS 501
cycle (de Souza Neto et al., 1998). This feature precludes the use of Gurson’s model for
damage prediction in situations where cyclic loads are important. On the other hand, the
healing behaviour can be useful, for instance, in the prediction of voids distribution in powder
compaction processes (Gethin et al., 1998; Ransing et al., 1998). In this case, the initial voids
ratio is relatively high and is gradually reduced by application of hydrostatic pressure under
highly confining boundary conditions. This is in sharp contrast to Lemaitre’s model where
only damaging (no healing) can be predicted.
σtrial e e trial
n+1 := D : εn+1
IF Φtrial ≤ 0 THEN
Set (·)n+1 = (·)trial
n+1 and EXIT
ELSE go to (ii)
(ii) Plastic corrector (solve the system for the unknowns pn+1 , Rn+1 ,
Dn+1 and ∆γ)
2
2G
J2 (εedn+1
trial 1
) − a [σy (Rn+1 )]2
0
1 + 2 G ∆γ 3
pn+1 − κ εev trial + ∆γ κ b σy (Rn+1 ) 0
n+1
=
Dn+1 − Dn − ∆γ b (Dn+1 − Dn+1 ) σy (Rn+1 )
2
0
∆γ 2
0
Rn+1 − Rn − a σy (Rn+1 ) + b pn+1 Dn+1
1 − Dn+1 3
where
2 3 pn+1 3 pn+1
a=1+ Dn+1 − 2Dn+1 cosh , b = sinh
2 σy (Rn+1 ) 2 σy (Rn+1 )
The elastic tangent for the present model is simply the standard linear elasticity tensor. In this
case we have
D = De . (12.78)
Under plastic loading, the elastoplastic operator is obtained by the standard linearisation
procedure. The system of equations of the plastic corrector phase is differentiated at the
DAMAGE MECHANICS 503
where A1,∆γ , A1,p , . . . denote the derivatives of the plastic corrector system components.
Inversion of the above expression then leads to
d∆γ C11 C12 C13 C14 −A1,εed trial : dεedn+1
trial
n+1
trial
dpn+1 C21 C22 C23 C24 −A2,εev trial dεevn+1
= n+1 , (12.80)
dD
n+1 C31 C32 C33 C34 0
dRn+1 C41 C42 C43 C44 0
which provides the tangent relations between the system variables (∆γ, p, D and R) and the
trial
system input εen+1 . Note that, as the stress tensor is one of the system variables in Lemaitre’s
original model, the tangent operator Dep in that case is obtained directly from the inversion of
the system derivative. Here, the consistent tangent operator can be obtained by differentiating
the stress update formula of item (iii) of Box 12.6 which gives
2
2G 2G
dσn+1 = dεe trial − trial
d∆γ εedn+1 + dpn+1 I. (12.81)
1 + 2G ∆γ dn+1 1 + 2G ∆γ
Then, substitution of d∆γ and dpn+1 by the relations given in (12.80) and use of the identities
2
2G
A1,εed trial = trial
εedn+1 ; A2,εev trial = −κ,
n+1 1 + 2 G ∆γ n+1
result, after some straightforward manipulations, in the following expression for the elasto-
plastic consistent tangent operator:
dσn+1
Dep = trial
dεen+1
= g Id + g 2 εedn+1
trial
⊗ [C11 g 2 εedn+1
trial
− C12 κ I ]
− I ⊗ [C21 g 2 εedn+1
trial
− C22 κ I ], (12.82)
where
2G
g≡ .
1 + 2G ∆γ
Remark 12.12. It is important to note that, as in Lemaitre’s model, the resulting elastoplastic
tangent operator Dep is generally unsymmetric.
504 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
σ
virgin material
(no damage)
damaged
E0 material
1
n
sio (1-D) E 0
ten
1
0 ε
n
sio
es
pr
m
co
1
(1-hD) E 0
Figure 12.10. Uniaxial elastic model with damage and partial crack closure effect.
0 ≤ h ≤ 1. (12.86)
This constant describes the effect of partial microcrack/void closure. A value h ≈ 0.2 is
typically observed in many experiments (Lemaitre, 1996). Note that for h = 1, the behaviour
of the original damage model, without crack closure effects, is recovered, whereas the other
extreme value, h = 0, represents full crack closure with E = E0 under compression. Any
other value of h describes a partial crack closure effect. The uniaxial stress–strain diagram of
Figure 12.10 illustrates the behaviour of the model.
The constitutive equation of the above uniaxial model can be more elegantly expressed by
introducing the following tensile/compressive split of the uniaxial stress:
σ = σ+ + σ− , (12.87)
where
σ+ = σ and σ− = −−σ, (12.88)
are, respectively, the tensile and compressive components of σ and is the Macauley bracket,
that is, for any scalar, a,
a if a ≥ 0
a = (12.89)
0 if a < 0.
506 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
With the above notation, the uniaxial stress–strain relation for the damaged elastic material
crack closure effect can be written simply as
1 σ+ σ−
ε= + . (12.90)
E0 1 − D 1 − h D
and
3
σ− = − −σi ei ⊗ ei . (12.95)
i=1
The matrix representation of σ+ and σ− in principal stress basis reads
σ1 0 0
[σ+ ] ≡ 0 σ2 0 (12.96)
0 0 σ3
DAMAGE MECHANICS 507
and
−σ1 0 0
[σ− ] ≡ − 0 −σ2 0 . (12.97)
0 0 −σ3
Note that, under uniaxial stresses, i.e. when only one of the principal stresses is non-zero, the
above tensile/compressive split reduces to that of expression (12.87).
Computational implementation
Now, we turn our attention to the computational implementation of the above model within
an implicit finite element environment. As usual, the basic ingredients of the computer
implementation are: (a) the stress updating procedure which, given a strain tensor, should
compute the stress tensor that satisfies the constitutive law (12.99); and (b) the computation
of the tangent operator associated with (12.99). These will be described in the following.
At the outset, it should be noted that the stress tensor in (12.99) cannot be written explicitly
as a function of the strain tensor. Thus, given ε, the corresponding stress tensor will be
obtained as the solution of the following nonlinear tensorial equation for σ:
1+ν σ+ σ− ν tr σ −tr σ
R(σ) ≡ ε − + − − = 0. (12.100)
2E0 1 − D 1 − hD 2E0 1 − D 1 − hD
This solution can be obtained by means of the Newton–Raphson algorithm as described in
what follows. Before proceeding, however, we note that as σ+ and σ− are defined on the
508 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Array/matrix definitions
For convenience, we define the following arrays of principal stresses and strains
σ∗ ≡ [ σ1 σ2 σ3 ]T , ε∗ ≡ [ ε1 ε2 ε3 ]T , (12.101)
and
H̄(σ1 ) 0 0
1
− ≡
PD 0 H̄(σ2 ) 0 (12.104)
1 − hD
0 0 H̄(σ3 )
where Ĥ is the Heaviside step function given in (7.94), page 221, and H̄ is defined such that,
for any scalar a,
0 if a > 0
H̄(a) ≡ (12.105)
1 if a ≤ 0.
We also define the following:
1
I if tr σ > 0
ID ≡ 1−D (12.106)
+
0 if tr σ ≤ 0
and
0 if tr σ > 0
− ≡
ID 1 (12.107)
I if tr σ ≤ 0,
1 − hD
DAMAGE MECHANICS 509
Box 12.7. Stress-updating algorithm for damaged elastic material model with crack
closure effects.
(i) Given ε, compute initial guess for stress, σ(0) (use damaged elastic constitutive
equation without crack closure effect)
σ(0) := (1 − D) D : ε
ε∗ − [D(σ∗ )]−1 σ∗ = 0
3
σ := σi ei ⊗ ei
i=1
(vi) EXIT
crucial under complex loading conditions and is relevant to many practical applications. An
alternative to describe this phenomenon is provided by the damage evolution law proposed by
Ladevèze and Lemaitre (1984) (described by Lemaitre 1996 in more detail). Their approach
consists in modifying the damage energy release rate (12.15) of the original Lemaitre ductile
damage model by including the tensile/compressive split of the stress tensor. Thus, the
original expression (12.15) is replaced with
−1
Y = [(1 + ν) σ+ : σ+ − ν tr σ2 ]
2E(1 − D)2
h
− [(1 + ν) σ− : σ− − ν −tr σ2 ], (12.118)
2E(1 − hD)2
and the elastoplastic damage evolution equation keeps the same format as in the original
model: s
1 −Y
Ḋ = γ̇ . (12.119)
1−D r
The complete fully coupled elastoplastic model for ductile damage is obtained by considering
the above damage evolution equation together with the damaged isotropic elasticity law with
crack closure effects (discussed in the previous section) and the standard plasticity equations
of the original Lemaitre model.
1. Given the incremental strain, ∆ε, compute the updated stress, σn+1 , and the incre-
mental plastic multiplier, ∆γ, by using the conventional von Mises model (possibly
with kinematic hardening). The computer implementation of the von Mises model was
described in Chapter 7.
2. With the updated stress and incremental multiplier at hand, compute the new damage
state, Dn+1 , ‘a posteriori’ by solving (for Dn+1 ) the backward Euler discrete version
of (12.119) which reads
s
∆γ −Yn+1
Dn+1 = Dn + , (12.120)
1 − Dn+1 r
where seff and peff are, respectively, the effective stress deviator and effective hydrostatic
pressure defined as
p
seff ≡ dev[H s H ], peff ≡ I. (12.124)
1 − dH
DAMAGE MECHANICS 513
H ≡ (I − D)− 2 .
1
(12.125)
σeff = De : εe . (12.126)
D = D I, (12.127)
and
η = 1, (12.128)
which imply
dH = D. (12.129)
Under such states, straightforward algebra shows that (12.126) reduces to the damaged elastic
law (12.11) of the original isotropic model. For a more general non-isotropic damage state D,
the linear elastic constitutive equation (12.126) is orthotropic. The directions of orthotropy
are the principal directions of the damage tensor. In an orthonormal basis aligned with its
principal directions the damage tensor is represented as
D1 0 0
[D] = 0 D2 0 . (12.130)
0 0 D3
The eigenvalues values D1 , D2 and D3 are associated to the degradation of the elasticity
modulus along the corresponding directions of orthotropy. Their experimental identification
is discussed by Lemaitre et al. (2000).
Damage evolution
To complete the definition of the model, an evolution law for the damage tensor is required.
In the present theory, the rate of damage tensor is assumed to follow the directions of plastic
straining. The evolution law for D is defined by
s
−Ỹ
D= ε̃˙ p , (12.131)
r
where 2
−qeff
2
2 peff
Ỹ ≡ (1 + ν) + 3(1 − 2ν) , (12.132)
2E 3 qeff
with qeff defined as
qeff ≡ 3 J2 (seff ). (12.133)
514 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
3
ε̃˙ p ≡ |ε̇pi | epi ⊗ epi , (12.134)
i=1
where ε̇pi are the eigenvalues of the plastic strain rate tensor, ε̇p , and {epi } is an orthonormal
basis of eigenvectors of ε̇p . In the basis {epi }, the absolute plastic strain rate has the matrix
representation p
|ε̇1 | 0 0
[ε̃˙ p ] = 0 |ε̇p2 | 0 . (12.135)
0 0 |ε̇p3 |
σy = σy (R), (12.137)
where R is the strain-hardening variable. Clearly, with this yield function, the softening
experienced by the material as a result of damage is generally anisotropic. The yield
von Mises equivalent stress depends on the orientation of the applied stress with respect
to the principal directions of damage. For an isotropically damaged state, where D = DI,
the above yield function reduces to the isotropic function of Box 12.3.
The flow rule is defined by assuming associativity of the plastic flow. We then have the
equation
ε̇p = γ̇ N, (12.138)
where the associative flow vector,
∂Φ
N≡ , (12.139)
∂σ
is found, after some straightforward algebra, to be given by
%
3 dev[Hseff H ]
N= . (12.140)
2 seff
Ṙ = γ̇. (12.141)
DAMAGE MECHANICS 515
Kinematic hardening
A kinematic hardening-type variable can also be introduced in the model. In this case, we
replace qeff with
q̄eff ≡ 3 J2 (seff − β) (12.142)
in (12.136) and redefine the flow vector as
%
3 dev[H(seff − β)H ]
N≡ , (12.143)
2 seff − β
where β is the backstress tensor. The evolution of the backstress can be defined analogously
to the original Lemaitre model:
β̇ = γ̇ (a N − b β). (12.144)
P ART III of this book is devoted to the formulation and numerical treatment of finite strain
solid mechanics problems. We begin by describing in this chapter some useful finite
hyperelasticity constitutive theories together with their implementation within the implicit
finite element environment of Chapter 4. Two of the models addressed here are implemented
in program HYPLAS: the Hencky (logarithmic strain-based) and the Ogden hyperelastic
models. Recall (refer to Chapter 4) that the most relevant components of the finite element
implementation of a material model in HYPLAS are:
1. the state-update procedure, where the stresses are computed for a given state of
deformation. The stresses computed by such a procedure are used to assemble the
element internal force vector;
2. the computation of the corresponding spatial tangent modulus, which is used in the
assembly of the element tangent stiffness matrix.
To provide the reader with a clear idea of the complete finite element implementation of
finite hyperelastic theories, the FORTRAN source code of these procedures is provided and
explained in detail for the particular case of the Ogden material model.
Hyperelastic constitutive models are suitable to describe the behaviour of a large number
of engineering materials. In the finite strain range, such theories are particularly appropriate
for the analysis of rubber-like solids and in many cases show excellent agreement with
experiments involving strains over 700% (Ogden, 1984).
This chapter is divided as follows: after reviewing some basic concepts of finite hypere-
lasticity in Section 13.1, some popular models are described in Section 13.2. The formulation
of plane stress theories is addressed in Section 13.3, with particular reference to the Ogden
and Hencky models. The explicit formulae for the associated tangent moduli are derived
in Section 13.4. Section 13.5 describes in detail the computational implementation of the
Ogden model, including the listing of the relevant FORTRAN source code. A comprehensive
set of benchmarking finite element examples of hyperelasticity applications is provided in
Section 13.6. Finally, Section 13.7 discusses a relatively new topic: the modelling of internal
damage coupled with hyperelasticity. Hyperelastic damage theories are useful in modelling
the so-called Mullins effect – the loss of stiffness frequently observed in filled polymers at
strain levels below the maximum previously attained strain.
We remark that much of the notation adopted in this chapter as well as throughout the
remainder of Part Three of this book has been set up in Chapter 3. Readers who are unfamiliar
with it are advised to review that chapter before proceeding further.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
520 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
ψ = ψ(F ). (13.1)
Functions of this form (without dependence on internal variables) are particular cases of the
general potential (3.150) (page 72) which include the description of dissipative materials
(with internal variables). In this case (refer to (3.155)), the first Piola–Kirchhoff stress tensor
is given by the constitutive relation
∂ψ(F )
P = P (F ) = ρ̄ , (13.2)
∂F
where ρ̄ is the reference density. The stress here depends solely on the current deformation
gradient and is not affected by the past deformation history.† The free-energy function, ψ,
completely defines a hyperelastic model.
Accordingly, the constitutive equation for the Kirchoff stress tensor,
τ≡P F T
, (13.3)
∂ψ(F ) T
τ(F ) = ρ̄ F (13.4)
∂F
and the Cauchy stress tensor, σ = τ/J, has the following constitutive equation
ρ̄ ∂ψ(F ) T
σ(F ) = F , (13.5)
J ∂F
where
J ≡ det F . (13.6)
In Sections 13.1.1 and 13.1.2 below, we consider some useful constraints imposed by the
general constitutive axioms of Section 3.5.1 (from page 69) upon the constitutive equations
for hyperelastic materials.
first Piola–Kirchhoff stress is a function of the deformation gradient only, but is not necessarily the derivative of a
free-energy function.
FINITE STRAIN HYPERELASTICITY 521
equation to the present case, where the free energy depends exclusively on the current value
of F , requires that the function ψ satisfies
Thus, material objectivity implies that ψ depends on F solely through the right stretch
tensor, U .
Equivalently, the free energy may be expressed as a function of the right Cauchy–Green
strain tensor, C ≡ F T F = U 2 ; that is, the free energy may be expressed by a function ψ̃
defined as √
ψ(F ) = ψ̃(C) ≡ ψ( C ). (13.9)
In terms of the reduced form, ψ̃, the first Piola–Kirchhoff stress tensor given by (13.2) can be
equivalently expressed as
∂ ψ̃ ∂C ∂ ψ̃
P = ρ̄ : = 2ρ̄F . (13.10)
∂C ∂F ∂C
In deriving the rightmost part of the above formula, use has been made of the component form
of the derivative ∂C/∂F , obtained by simply applying the product rule to Cij = Fmi Fmj ,
which gives
∂Cij
= δil Fkj + δjl Fki . (13.11)
∂Fkl
From (13.10) and (13.3), it then follows that the corresponding constitutive equation for the
Kirchhoff stress is
∂ ψ̃
τ = 2ρ̄F F T. (13.12)
∂C
Similarly, for the Cauchy stress, we have
2ρ̄ ∂ ψ̃ T
σ= F F . (13.13)
J ∂C
for all rotations Q. This equation introduces a further restriction on the possible representa-
tions for the free-energy function. Indeed, by choosing Q = RT again, we promptly establish
522 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
that
ψ(F ) = ψ(V ); (13.15)
that is, the free energy of an isotropic hyperelastic material must depend on F only through
the left stretch tensor, V.
The free energy in this case may be equivalently expressed as a function of the left
Cauchy–Green strain tensor, B ≡ F F T = V 2 , as follows:
√
ψ̃(B) = ψ( B). (13.16)
Equation (13.16), together with (13.8), (13.9) and (13.15), imply that
U = RT VR, (13.18)
As the above holds for any deformation (and, hence, any rotation R), it follows from
definition (A.2) (page 731) that ψ is an isotropic function of U (or V ). Following a
completely analogous argument in terms of the right and left Cauchy–Green tensors, we
establish that
ψ̃(C) = ψ̃(R BRT ), (13.20)
for any rotation R, and conclude that ψ̃ is an isotropic function of C (or B).
∂ ψ̃ ∂B ∂ ψ̃
P = ρ̄ : = 2ρ̄ F. (13.21)
∂B ∂F ∂B
Analogously to the derivation of (13.10), in obtaining the rightmost part of (13.21) we have
made use of the component form of ∂B/∂F , which can be derived by differentiating Bij =
Fim Fjm . The corresponding constitutive equation for the Kirchhoff stress is obtained by
combining (13.21) and (13.3) as
∂ ψ̃
τ = τ̃(B) ≡ 2 ρ̄ B. (13.22)
∂B
Further, the isotropy of ψ̃ implies that B and ∂ ψ̃/∂B share the same principal axes and,
thus, commute (refer to property (A.6), page 732). Hence, the Kirchhoff stress constitutive
FINITE STRAIN HYPERELASTICITY 523
∂ ψ̃
τ̃(B) ≡ 2 ρ̄ B . (13.23)
∂B
Accordingly, for the Cauchy stress, we have
2ρ̄ ∂ ψ̃ 2ρ̄ ∂ ψ̃
σ = σ̃(B) ≡ B= B . (13.24)
J ∂B J ∂B
The isotropy of ψ̃ implies that (refer to the general representation formula (A.12), page 733)
the free energy can be equivalently expressed as a function of the principal invariants, I1 (B),
I2 (B) and I3 (B), of the left Cauchy–Green strain tensor (see definition (2.72), page 27, of
the principal invariants of a tensor); that is, there exists a function ψ̄ such that
As (13.22) defines the Kirchhoff stress (what follows is equally true for equation (13.24)
for the Cauchy stress) as an isotropic symmetric tensor-valued function of B – an invertible
symmetric tensor, then it follows from the general representation (A.14) (page 733) that the
stress response of an isotropic hyperelastic solid may be cast in the form
or, equivalently,
σ = β0 I + β1 B + β−1 B−1 , (13.28)
where the scalar coefficients βΓ = βΓ (I1 , I2 , I3 ), (Γ = 0, 1, −1), are the elastic response
functions which, in terms of the strain-energy function (13.26) are expressed as
2 ∂ ψ̄ ∂ ψ̄
β0 = √ I2 ρ̄ + I3 ρ̄
I3 ∂I2 ∂I3
2 ∂ ψ̄
β1 = √ ρ̄ (13.29)
I3 ∂I1
∂ ψ̄
β−1 = −2 I3 ρ̄ .
∂I2
524 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
With the above decomposition at hand, we define the isochoric left Cauchy–Green strain
tensor as
= (det F )− 3 F F T ,
2
Biso ≡ F iso F iso
T
(13.41)
and the principal invariants
where the material constant K is the logarithmic bulk modulus and relates the hydrostatic
pressure to the purely volumetric component of the deformation gradient. The corresponding
neo-Hookean function reads
for instance, is also frequently adopted (see, among others, Crisfield 1997) in the finite
element analysis of nearly incompressible materials.
FINITE STRAIN HYPERELASTICITY 527
where N is the total number of terms in the series and µp and αp (p = 1, . . . , N ) are the
corresponding material constants. Perfect incompressibility is also assumed so that λ1 λ2 λ3 =
1. Thus, equivalently, the strain-energy can be expressed as a function of λ1 and λ2 only
N
µp α α 1
ρ̄ ψ̃(λ1 , λ2 ) = λ1 p + λ2 p + αp αp − 3 . (13.49)
p=1
αp λ1 λ2
where (i, j, k) are cyclic permutations of (1, 2, 3). The strain-energy function for a regu-
larised compressible version of the Ogden model can be defined as
N
µp ∗ αp
ρ̄ ψ̂ ∗ (λ∗1 , λ∗2 , λ∗3 , J) = [(λ1 ) + (λ∗2 )αp + (λ∗3 )αp − 3] + 12 K (ln J)2 . (13.52)
p=1
αp
The corresponding constitutive function for the principal Kirchhoff stresses is obtained from
the potential relation (13.33). With the above regularised strain-energy function, we have
∗ ∗
∂ ψ̂ ∂ ψ̂ ∂λi ∂ ψ̂ ∗ ∂λ∗j ∂ ψ̂ ∗ ∂λ∗k ∂ ψ̂ ∗ ∂J
τi = λi ρ̄ = λi ρ̄ + + + (13.53)
∂λi ∂λ∗i ∂λi ∂λ∗j ∂λi ∂λ∗k ∂λi ∂J ∂λi
528 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
where no summation is implied on repeated indices and (i, j, k) are permutations of (1, 2, 3).
The final expression for the principal Kirchhoff stresses reads
N
µp J −αp /3 [λi p − 13 (λ1 p + λ2 p + λ3 p )] + K ln J.
α α α α
τi = (13.54)
p=1
ε ≡ ln V = 1
2 ln B. (13.55)
D ≡ 2G IS + (K − 23 G)I ⊗ I. (13.57)
Note that the functional format of the Hencky strain-energy function is identical to that of
infinitesimal elasticity. Equivalently, in terms of principal stretches, the Hencky strain-energy
function is given by
The above strain-energy renders the following linear relationship between the Kirchhoff
stress and the Eulerian logarithmic strain:
∂ψ
τ = ρ̄ = D : ε, (13.59)
∂ε
which has the same functional format as the infinitesimal linear elastic stress–strain relation.
That (13.59) follows from definition (13.56) (or (13.58)) can be demonstrated as follows.
Recall equation (13.22), which is valid for any isotropic hyperlastic material. By taking
definition (13.55) into account and applying the chain rule to differentiate the free-energy
function (13.56), we obtain the expression
∂ψ 1 ∂ψ ∂(ln B)
= : , (13.60)
∂B 2 ∂ε ∂B
which, substituted into (13.22), gives
∂ψ ∂(ln B)
τ = ρ̄ : B. (13.61)
∂ε ∂B
FINITE STRAIN HYPERELASTICITY 529
Now, note that by definition of ε, the tensors ε and B share the same principal axes (have the
same eigenvectors/eigenprojection tensors). Also, due to the isotropy of ψ, the tensor ∂ψ/∂ε
has the same principal axes. In addition, it should be noted that the tensor logarithm is a
member of the class of isotropic tensor functions discussed in Section A.5 (from page 740).
These observations in conjunction with the specialisation of the formulae (A.52) (page 744)
for the derivative ∂(ln B)/∂B lead, after some algebra, to the identity
∂ψ ∂(ln B) ∂ψ
: B= , (13.62)
∂ε ∂B ∂ε
which, replaced in (13.61), yields the constitutive equation (13.59) for the Hencky material.
where
εd ≡ ε − 1
3 εv I (13.64)
is the deviatoric logarithmic strain and
εv ≡ tr ε (13.65)
εv ≡ tr ε = 0 ⇐⇒ det F = 1. (13.69)
530 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Let us now look at the properties of the logarithmic strain deviator. Using the definition of
εd , we obtain after a straightforward manipulation the following diagonal representation
2 1 1
ln λ13 − ln λ23 − ln λ33 0 0
[εd ] = .
2 1 1
0 ln λ2 − ln λ3 − ln λ1
3 3 3
0
2 1 1
0 0 ln λ33 − ln λ13 − ln λ23
(13.70)
From the above and the definition (13.50) of the isochoric left stretch tensor, we then find that
εd = ln Viso ; (13.71)
that is, as for the infinitesimal strain tensor, the logarithmic strain deviator is a measure of the
purely isochoric component of the strain.
where µ0 and f are material constants and g(J) is the derivative of the strain-energy function
with respect to J. This derivative is a function of J only. Note that for the Blatz–Ko material
the corresponding elastic response functions βΓ (refer to representation (13.27)) depend
exclusively on J.
To validate the model, these authors have conducted a series of experiments, comprising
uniaxial and biaxial tension tests, on polyurethane rubbers. The explicit expression adopted
for the function g has been chosen by setting
where the constant ν is a finite strain extension to the conventional Poisson’s ratio of the
infinitesimal theory. The model recovers the incompressible limit when ν → 0.5.
page 362) can be applied to the hyperelastic case. As in infinitesimal elastoplasticity, the
derivation of plane stress-projected equations here is feasible when the relevant constitutive
model is sufficiently simple to allow the out-of-plane strain to be eliminated from the
formulation, becoming a dependent variable that can be expressed in closed form as an
explicit function of the in-plane strains. This approach is precisely that adopted in plane stress
linear elasticity reviewed in Section 9.1.1 (from page 358). As we shall see, the same concept
can be trivially applied to derive plane stress constitutive equations for the incompressible
Ogden model and the logarithmic strain-based Hencky model (both implemented in program
HYPLAS). These will be discussed, respectively, in Sections 13.3.1 and 13.3.2. Our discussion
on plane stress ends in Section 13.3.3 where we briefly outline the use of a nested iteration
approach in the computational treatment of finite hyperelasticity.
where Jp is the Kirchhoff pressure. The pressure Jp can be determined by introducing the
plane stress state constraint. Indeed, (13.74) together with (13.75) gives
N
α
τ3 = µp λ3 p + J p = 0, (13.76)
p=1
which implies
N
α
J p=− µp λ3 p , (13.77)
p=1
or, equivalently, since λ1 λ2 λ3 = 1 (incompressibility),
N
J p=− µp (λ1 λ2 )−αp . (13.78)
p=1
Finally, the substitution of (13.78) into (13.75) results in the following plane stress-
projected constitutive relation for the incompressible Ogden model;
N
µp [λβ p − (λ1 λ2 )−αp ] for β = 1, 2.
α
τβ = (13.79)
p=1
Thickness update
At finite strains, the thickness at any point depends on the out-of-plane stretch λ3 . In
spatial finite element formulations, where integrations are performed over the deformed
configuration, the current thickness (thickness in the deformed configuration) is required at
each Gauss point during the computation of the internal force vector and stiffness matrix.
With t0 denoting the thickness of the Gauss point of interest in its reference configuration,
the current thickness can be computed as
t = λ3 t0 . (13.80)
In view of the incompressibility of the model, the above update formula is equivalent to
t0
t= . (13.81)
λ1 λ2
Remark 13.3. One important point to be considered is that for isotropic elastic materials, the
Kirchhoff stress tensor is coaxial with the left Cauchy–Green strain tensor B. Consequently,
since the hypothesis of plane stress implies that the non-zero stress components are in-plane,
the deformation pattern must be such that transverse shear strains vanish. Clearly, that can
only be a good approximation as long as the thickness deformation does not vary abruptly
within the plane. If abrupt thickness variations do occur, then the plane stress assumption no
longer makes sense.
Thickness update
Analogously to the linear elastic case, we have
3K − 2G ν
ε33 = − tr[ε] = − tr[ε], (13.83)
3K + 4G 1−ν
where, here,
tr[ε] = ε11 + ε22 (13.84)
is the trace of the plane tensor ε. In terms of logarithmic stretches, expression (13.83) is
equivalent to
3K − 2G ν
ln λ3 = − (ln λ1 + ln λ2 ) = − (ln λ1 + ln λ2 ). (13.85)
3K + 4G 1−ν
FINITE STRAIN HYPERELASTICITY 533
2. With the solution λ3 at hand, use the three-dimensional constitutive function to update
the in-plane Kirchhoff stresses τ1 and τ2 :
τ1 = τ (λ1 , λ2 , λ3 )
τ2 = τ (λ2 , λ3 , λ1 ).
Clearly, τ3 vanishes within the prescribed tolerance for solution of the nonlinear plane
stress enforcement equation.
3. The current thickness is updated by the formula
t = λ3 t0 .
The above procedure can be applied to any model. The remarks made in Section 9.2.2
concerning the computational implications of this approach, i.e. the introduction of an
iteration loop for plane stress enforcement at the Gauss point level – nested within an outer
equilibrium loop – applies equally to the present case.
∂τ ∂Biso K ∂J
= G Id : + I⊗ (13.91)
∂B ∂B J ∂B
where the fourth-order tensor Id is the deviatoric projection operator in the space of symmetric
tensors, defined by expression (3.94), page 59. Next, a straightforward differentiation of
(13.41) leads to
∂Biso ∂J
= − 32 J − 3 B ⊗ + J − 3 IS .
5 2
(13.92)
∂B ∂B
Substitution of this formula into (13.91), together with the expression
∂J
= 1
2 J B−1 , (13.93)
∂B
gives
∂τ
= G Id : [J − 3 IS − 13 J − 3 Biso ⊗ B−1 ] + 12 K I ⊗ B−1
2 2
∂B
= G J − 3 Id − 13 τd ⊗ B−1 + 12 K I ⊗ B−1 ,
2
(13.94)
where τd is the deviatoric component of the Kirchhoff stress tensor. Expression (13.93) is
obtained by direct application of the formula for the derivative of the determinant
d(det A)
= (det A) A−T , (13.95)
dA
√
valid for any invertible A, to the definition J ≡ det B. Finally, by replacing the component
form of (13.94) into (13.89), we arrive, after some algebra, at the following expression for a:
where
2G K
c= tr[Biso ] Id − 2p IS + I ⊗ I − 23 [s ⊗ I + I ⊗ s]. (13.97)
3J J
In the above expression, s is the deviatoric Cauchy stress and p denotes the Cauchy
hydrostatic pressure, i.e. p ≡ tr[σ]/3.
3
τ(B) = τ̃ (bi , bj , bk ) ei ⊗ ei , (13.99)
i=1
In the present case, it is crucial to observe that in (13.99) τ(B) is represented explicitly in
the same format as the general isotropic function defined by (A.15) in Appendix A (page 734).
Thus, the derivative ∂τ/∂B can be conveniently evaluated simply by setting Y = τ, yi = τi ,
X = B and xi = bi in the general expressions for the derivative of isotropic tensor functions
summarised in Boxes A.3 and A.6, respectively, for the two- and three-dimensional cases. In
the computational implementation of the model (described in Section 13.5) we first evaluate
the components of ∂τ/∂B and then use (13.89) to compute the components of the spatial
elasticity tensor. Note (see Boxes A.3 and A.6) that the eigenvalue derivatives
∂τi
, i, j = 1, 2, 3
∂bj
are required to compute ∂τ/∂B. The explicit form of the eigenvalue derivatives is given in
what follows for the regularised version of the Ogden model as well as for the incompressible
model under plane stress. These expressions are implemented in subroutine CSTOGD of
program HYPLAS.
where no summation is implied on repeated indices and δij is the Krönecker delta.
∂τi µp αp αp
N
= [λ δij + (λ1 λ2 )−αp ] (13.102)
∂bj p=1 2λ2j j
Remark 13.4. As any isotropic model admits a principal stretch-based representation of the
function τ(B), we can, in principle, apply the methodology described above for the Ogden
material to compute the elasticity tensor for any isotropic elastic model. For the regularised
neo-Hookean model, whose expression for a is given by (13.96, 13.97), the principal stretch-
based computation of a can be carried out by setting N = 1, µ1 = G and α1 = 2 in (13.101).
For this particular model, however, it is computationally more efficient to use the more
compact form (13.96, 13.97).
Finally, formula (13.105) together with (13.103) yields the following expression for the
Hencky spatial elasticity tensor
1
aijkl = [D : L : B]ijkl − σil δjk , (13.107)
2J
which has a particularly simple format. We remark that the above formula is equally valid for
the plane stress case discussed in Section 13.3.2.
−1
By differentiation of the identity‡ Bik Bkj = δij the following component expression for the
last derivative on the right-hand side of the above formula is obtained:
−1
∂Bij 1 −1 −1
= − (Bik Blj + Bil−1 Bjk
−1
). (13.109)
∂Bkl 2
Substitution of this expression together with (13.93) into (13.108) then gives
∂τij J −1 f µ0 (1 − f )µ0 −1 −1
= g (J) δij Bkl + (δik δjl + δil δjk ) + (Bik Bjl + Bil−1 Bjk
−1
).
∂Bkl 2 2 2
(13.110)
Finally, by substituting the above expression into (13.89), the following explicit formula for
the Cartesian components of a is obtained:
f µ0
aijkl = g (J) δij δkl + (δik Bjl + δjk Bil )
J
(1 − f )µ0 −1 −1
+ (δil Bjk + δjl Bik ) − σil δjk . (13.111)
J
The computational implementation of this formula is a relatively straightforward program-
ming exercise.
T
(i) Given the deformation gradient, F , compute B := F F
(ii) Perform spectral decomposition of B (routine SPDEC2) and compute principal
stretches √
B= bi Mi , λi := bi
i
N
µp [λβ p − (λ1 λ2 )−αp ]
α
τβ :=
p=1
N
µp J −αp /3 [λi p − 13 (λ1 p + λ2 p + λ3 p )] + K ln J
α α α α
τi :=
p=1
1 SUBROUTINE SUOGD
2 1( B ,IPROPS ,NTYPE ,RPROPS ,RSTAVA ,
3 2 STRES ,THICK )
4 IMPLICIT DOUBLE PRECISION (A-H,O-Z)
5 PARAMETER(IPOGDC=2)
6 LOGICAL DUMMY
7 PARAMETER
8 1( MCOMP=4 ,MSTRE=4 ,NDIM=2 )
9 DIMENSION
10 1 B(MCOMP) ,IPROPS(*) ,RPROPS(*) ,
11 2 RSTAVA(MSTRE) ,STRES(MSTRE)
12 DIMENSION
13 1 EIGPRJ(MCOMP,NDIM) ,EIGB(NDIM) ,PSTRES(3) ,
14 2 PSTRTC(3)
15 DATA R1 ,R3 /
16 1 1.0D0,3.0D0/
17 C***********************************************************************
18 C STRESS UPDATE PROCEDURE FOR OGDEN TYPE HYPERELASTIC MATERIAL MODEL.
19 C PLANE STRESS, PLANE STRAIN AND AXISYMMETRIC IMPLEMENTATIONS.
20 C***********************************************************************
21 C Retrieve Ogden material constants
22 C =================================
23 C Number of terms in Ogden’s strain-energy function
24 NOGTRM=IPROPS(3)
540 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
25 C Bulk modulus
26 BULK=RPROPS(IPOGDC+NOGTRM*2)
27 C Compute principal stretches
28 C ===========================
29 C Perform spectral decomposition of the left Cauchy-Green tensor B
30 CALL SPDEC2
31 1( EIGPRJ ,EIGB ,DUMMY ,B )
32 C Compute in-plane principal stretches
33 PSTRTC(1)=SQRT(EIGB(1))
34 PSTRTC(2)=SQRT(EIGB(2))
35 C...and out-of-plane stretches
36 IF(NTYPE.EQ.1)THEN
37 PSTRTC(3)=R1/(PSTRTC(1)*PSTRTC(2))
38 ELSEIF(NTYPE.EQ.2)THEN
39 PSTRTC(3)=R1
40 ELSEIF(NTYPE.EQ.3)THEN
41 PSTRTC(3)=SQRT(B(4))
42 ENDIF
43 C Compute principal Kirchhoff stresses
44 C ====================================
45 CALL RVZERO(PSTRES,3)
46 IF(NTYPE.EQ.1) THEN
47 C Plane stress: Exact incompressibility assumed
48 C ---------------------------------------------
49 DO 10 I=1,NOGTRM
50 CMU=RPROPS(IPOGDC-1+I*2-1)
51 ALPHA=RPROPS(IPOGDC-1+I*2)
52 PSTRES(1)=PSTRES(1)+CMU*(PSTRTC(1)**ALPHA-
53 1 (PSTRTC(1)*PSTRTC(2))**(-ALPHA))
54 PSTRES(2)=PSTRES(2)+CMU*(PSTRTC(2)**ALPHA-
55 1 (PSTRTC(1)*PSTRTC(2))**(-ALPHA))
56 10 CONTINUE
57 DETF=R1
58 ELSE IF(NTYPE.EQ.2.OR.NTYPE.EQ.3)THEN
59 C Plane strain and axisymmetric: Regularised Ogden constitutive law
60 C -----------------------------------------------------------------
61 C Compute principal deviatoric Kirchhoff stresses
62 R1D3=R1/R3
63 DETF=PSTRTC(1)*PSTRTC(2)
64 IF(NTYPE.EQ.3)DETF=DETF*PSTRTC(3)
65 DO 20 I=1,NOGTRM
66 CMU=RPROPS(IPOGDC-1+I*2-1)
67 ALPHA=RPROPS(IPOGDC-1+I*2)
68 FACTOR=R1D3*(PSTRTC(1)**ALPHA+PSTRTC(2)**ALPHA+
69 1 PSTRTC(3)**ALPHA)
70 FACVOL=DETF**(-ALPHA*R1D3)
71 PSTRES(1)=PSTRES(1)+CMU*FACVOL*(PSTRTC(1)**ALPHA-FACTOR)
72 PSTRES(2)=PSTRES(2)+CMU*FACVOL*(PSTRTC(2)**ALPHA-FACTOR)
73 PSTRES(3)=PSTRES(3)+CMU*FACVOL*(PSTRTC(3)**ALPHA-FACTOR)
74 20 CONTINUE
75 C Add hydrostatic Kirchhoff pressure (incompressibility penalty term)
76 PRESS=BULK*LOG(DETF)
77 DO 30 I=1,3
78 PSTRES(I)=PSTRES(I)+PRESS
79 30 CONTINUE
80 ENDIF
81 C Assemble array of Cauchy stress tensor components
82 C ==================================================
83 CALL RVZERO(STRES,3)
84 R1DDET=R1/DETF
FINITE STRAIN HYPERELASTICITY 541
85 PSTRES(1)=PSTRES(1)*R1DDET
86 PSTRES(2)=PSTRES(2)*R1DDET
87 DO 50 ICOMP=1,3
88 DO 40 IDIR=1,2
89 STRES(ICOMP)=STRES(ICOMP)+PSTRES(IDIR)*EIGPRJ(ICOMP,IDIR)
90 40 CONTINUE
91 50 CONTINUE
92 IF(NTYPE.EQ.2.OR.NTYPE.EQ.3)STRES(4)=PSTRES(3)*R1DDET
93 C Update thickness (plane stress only) and store left Cauchy-Green
94 C tensor components in state variables vector RSTAVA
95 C ======================================================================
96 RSTAVA(1)=B(1)
97 RSTAVA(2)=B(2)
98 RSTAVA(3)=B(3)
99 IF(NTYPE.EQ.1)THEN
100 THICK=THICK*PSTRTC(3)
101 RSTAVA(4)=PSTRTC(3)*PSTRTC(3)
102 ELSEIF(NTYPE.EQ.2)THEN
103 RSTAVA(4)=R1
104 ELSEIF(NTYPE.EQ.3)THEN
105 RSTAVA(4)=B(4)
106 ENDIF
107 C
108 RETURN
109 END
→ IPROPS. Array of integer material properties. IPROPS(3) contains the number of terms
in the Ogden strain-energy function, NOGTRM. This is the only integer material property
required by SUOGD. IPROPS(3) is set in subroutine RDOGD during the input phase of
HYPLAS.
→ NTYPE. Stress state-type flag. Its value is: 1 for plane stress, 2 for plane strain and 3 for
axisymmetric case.
→ RPROPS. Array of real material properties. It contains the Ogden constants {µp , αp }
for p = 1, . . . , NOGTRM and the logarithmic bulk modulus, K. The bulk modulus is
used only in the compressible version of the Ogden model, adopted in the plane strain
and axisymmetric cases only. It is not used in plane stress where the incompressible
version of the model is adopted. Array RPROPS is set in subroutine RDOGD during the
input phase of HYPLAS.
← RSTAVA [B]. Array of state variables other than the stress components. For the present
implementation of the Ogden material, this array stores the current left Cauchy–Green
tensor, B.
← STRES [σ]. Array containing the updated Cauchy stress tensor components.
↔ THICK [t0 , t]. Initial Gauss point thickness, t0 , on entry. Returns as the updated
thickness, t. Used only in the plane stress case.
542 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
• EIGPRJ [Mi , i = 1, 2]. Matrix containing the components of the in-plane eigenprojec-
tion tensors of B.
1 SUBROUTINE CSTOGD
2 1( AMATX ,B ,IPROPS ,NTYPE ,RPROPS ,
3 2 STRES )
4 IMPLICIT DOUBLE PRECISION (A-H,O-Z)
5 LOGICAL OUTOFP ,REPEAT
6 PARAMETER
7 1( MADIM=5 ,MSTRE=4 ,NDIM=2 )
8 PARAMETER(IPOGDC=2)
9 DIMENSION
10 1 AMATX(MADIM,MADIM) ,B(MSTRE) ,IPROPS(*) ,
11 2 RPROPS(*) ,STRES(MSTRE)
12 DIMENSION
13 1 DELTA(3,3) ,DPSTRE(3,3) ,DTAUDB(MSTRE,MSTRE),
14 2 EIGPRJ(MSTRE,NDIM) ,EIGB(NDIM) ,PSTALP(3) ,
15 3 PSTRES(3) ,PSTRTC(3)
16 DATA
17 1 DELTA(1,1) ,DELTA(1,2) ,DELTA(1,3) /
18 2 1.0D0 ,0.0D0 ,0.0D0 /
19 3 DELTA(2,1) ,DELTA(2,2) ,DELTA(2,3) /
20 4 0.0D0 ,1.0D0 ,0.0D0 /
21 5 DELTA(3,1) ,DELTA(3,2) ,DELTA(3,3) /
22 6 0.0D0 ,0.0D0 ,1.0D0 /
23 DATA
24 1 R1 ,R2 ,R3 ,R6 /
25 2 1.0D0,2.0D0,3.0D0,6.0D0/
26 C***********************************************************************
27 C COMPUTATION OF THE CONSISTENT SPATIAL TANGENT MODULUS ’a’ FOR
28 C OGDEN TYPE HYPERELASTIC MATERIAL MODEL.
29 C PLANE STRESS, PLANE STRAIN AND AXISYMMETRIC IMPLEMENTATIONS.
30 C***********************************************************************
FINITE STRAIN HYPERELASTICITY 543
Box 13.2. Computation of the spatial tangent modulus for the Ogden material
model.
T
(i) Given the deformation gradient, F , compute B := F F
(ii) Perform spectral decomposition of B (routine SPDEC2) and compute principal
stretches √
B= bi Mi , λi := bi
i
τi := J σ : Mi
∂τi µp αp αp
N
:= [λ δij + (λ1 λ2 )−αp ]
∂bj p=1
2λ2j j
∂τi µp αp J −αp /3
N
α α α K
:= 2 [f − λi p − λj p + 3λi p δij ] + 2
∂bj p=1
6λ j 2λ j
47 PSTRTC(3)=R1/(PSTRTC(1)*PSTRTC(2))
48 DETF=R1
49 ELSEIF(NTYPE.EQ.2)THEN
50 PSTRTC(3)=R1
51 DETF=PSTRTC(1)*PSTRTC(2)
52 ELSEIF(NTYPE.EQ.3)THEN
53 PSTRTC(3)=SQRT(B(4))
54 DETF=PSTRTC(1)*PSTRTC(2)*PSTRTC(3)
55 ENDIF
56 C Recover principal Kirchhoff stresses (from the given Cauchy stress)
57 PSTRES(1)=(STRES(1)*EIGPRJ(1,1)+STRES(2)*EIGPRJ(2,1)+
58 1 R2*STRES(3)*EIGPRJ(3,1))*DETF
59 PSTRES(2)=(STRES(1)*EIGPRJ(1,2)+STRES(2)*EIGPRJ(2,2)+
60 1 R2*STRES(3)*EIGPRJ(3,2))*DETF
61 IF(NTYPE.EQ.2.OR.NTYPE.EQ.3)PSTRES(3)=STRES(4)*DETF
62 C Compute derivatives of principal Kirchhoff stresses
63 C ===================================================
64 CALL RVZERO(DPSTRE,9)
65 IF(NTYPE.EQ.1) THEN
66 C Plane stress: Perfectly incompressibility assumed
67 C -------------------------------------------------
68 NSTRA=2
69 DO 10 IP=1,NOGTRM
70 ALPHA=RPROPS(IPOGDC+IP*2-1)
71 ALPHMU=ALPHA*RPROPS(IPOGDC+IP*2-2)
72 PSTALP(1)=PSTRTC(1)**ALPHA
73 PSTALP(2)=PSTRTC(2)**ALPHA
74 FACTOR=R1/(PSTALP(1)*PSTALP(2))
75 DO I=1,NSTRA
76 DO J=1,NSTRA
77 DPSTRE(I,J)=DPSTRE(I,J)+ALPHMU/(R2*PSTRTC(J)**2)*
78 1 (PSTALP(J)*DELTA(I,J)+FACTOR)
79 END DO
80 END DO
81 10 CONTINUE
82 ELSE IF(NTYPE.EQ.2.OR.NTYPE.EQ.3)THEN
83 C Plane strain and axisymmetric: Regularised Ogden model
84 C ------------------------------------------------------
85 C compute principal Kirchhoff stresses derivatives
86 R1D3=R1/R3
87 IF(NTYPE.EQ.2)THEN
88 NSTRA=2
89 ELSEIF(NTYPE.EQ.3)THEN
90 NSTRA=3
91 ENDIF
92 DO 40 IP=1,NOGTRM
93 CMU=RPROPS(IPOGDC-1+IP*2-1)
94 ALPHA=RPROPS(IPOGDC-1+IP*2)
95 PSTALP(1)=PSTRTC(1)**ALPHA
96 PSTALP(2)=PSTRTC(2)**ALPHA
97 PSTALP(3)=PSTRTC(3)**ALPHA
98 ALPHMU=ALPHA*CMU
99 FACTOR=R1D3*(PSTALP(1)+PSTALP(2)+PSTALP(3))
100 FACVOL=DETF**(-ALPHA*R1D3)
101 DO 30 I=1,NSTRA
102 DO 20 J=1,NSTRA
103 DPSTRE(I,J)=DPSTRE(I,J)+ALPHMU*FACVOL/(R6*PSTRTC(J)**2)*
104 1 (FACTOR-PSTALP(I)-PSTALP(J)+R3*PSTALP(I)*
105 2 DELTA(I,J))
106 20 CONTINUE
107 30 CONTINUE
FINITE STRAIN HYPERELASTICITY 545
108 40 CONTINUE
109 DO 60 I=1,NSTRA
110 DO 50 J=1,NSTRA
111 DPSTRE(I,J)=DPSTRE(I,J)+BULK/(R2*PSTRTC(J)**2)
112 50 CONTINUE
113 60 CONTINUE
114 ENDIF
115 C Compute the derivative of the Kirchhoff stress with respect to B
116 C (use routine for computation of derivative of general isotropic
117 C tensor functions of one tensor)
118 C ================================================================
119 IF(NTYPE.EQ.3)THEN
120 OUTOFP=.TRUE.
121 NADIM=5
122 ELSE
123 OUTOFP=.FALSE.
124 NADIM=4
125 ENDIF
126 CALL DGISO2
127 1( DPSTRE ,DTAUDB ,EIGPRJ ,EIGB ,PSTRES ,
128 2 OUTOFP ,REPEAT )
129 C Assemble the spatial tangent modulus ’a’
130 C ========================================
131 R2DDET=R2/DETF
132 C upper triangle and diagonal terms
133 AMATX(1,1)=R2DDET*(DTAUDB(1,1)*B(1)+DTAUDB(1,3)*B(3))-STRES(1)
134 AMATX(1,2)=R2DDET*(DTAUDB(1,3)*B(1)+DTAUDB(1,2)*B(3))
135 AMATX(1,3)=R2DDET*(DTAUDB(1,1)*B(3)+DTAUDB(1,3)*B(2))-STRES(3)
136 AMATX(1,4)=R2DDET*(DTAUDB(1,3)*B(3)+DTAUDB(1,2)*B(2))
137 AMATX(2,2)=R2DDET*(DTAUDB(3,3)*B(1)+DTAUDB(3,2)*B(3))
138 AMATX(2,3)=R2DDET*(DTAUDB(3,1)*B(3)+DTAUDB(3,3)*B(2))-STRES(2)
139 AMATX(2,4)=R2DDET*(DTAUDB(3,3)*B(3)+DTAUDB(3,2)*B(2))
140 AMATX(3,3)=R2DDET*(DTAUDB(3,1)*B(3)+DTAUDB(3,3)*B(2))
141 AMATX(3,4)=R2DDET*(DTAUDB(3,3)*B(3)+DTAUDB(3,2)*B(2))-STRES(3)
142 AMATX(4,4)=R2DDET*(DTAUDB(2,3)*B(3)+DTAUDB(2,2)*B(2))-STRES(2)
143 IF(NTYPE.EQ.3) THEN
144 AMATX(1,5)=R2DDET*DTAUDB(1,4)*B(4)
145 AMATX(2,5)=R2DDET*DTAUDB(3,4)*B(4)
146 AMATX(3,5)=R2DDET*DTAUDB(3,4)*B(4)
147 AMATX(4,5)=R2DDET*DTAUDB(2,4)*B(4)
148 AMATX(5,5)=R2DDET*DTAUDB(4,4)*B(4)-STRES(4)
149 ENDIF
150 C lower triangle
151 DO 80 J=1,NADIM
152 DO 70 I=J+1,NADIM
153 AMATX(I,J)=AMATX(J,I)
154 70 CONTINUE
155 80 CONTINUE
156 C
157 RETURN
158 END
→ B [B]. Array containing the components of the current left Cauchy–Green strain tensor.
• DPSTRE [∂τi /∂bj ]. Matrix containing the derivatives of the principal Kirchhoff
stresses.
• EIGPRJ [Mi , i = 1, 2]. Matrix containing the components of the in-plane eigenprojec-
tion tensors of B.
prescribed
radial displacement 30
external edge
thickness internal edge
25
0.0625 in
reaction, R (lb)
20
15
10
R
5
ri = 0.5 in 45o
0
0 1 2 3 4 5 6
re = 5 in radial displacement (in)
(a) (b)
Figure 13.1. Annular plate: (a) geometry and boundary conditions; (b) reaction-displacement diagram.
(Reproduced with permission from Finite elasticity in spatial description: Linearization aspects with
membrane applications, EA de Souza Neto, D Perić and DRJ Owen, International Journal for
Numerical Methods in Engineering, Vol 38 c 1995 John Wiley & Sons, Ltd.)
2R
prescribed displacement U
2 l = 20 in
d = 6 in
thick.= 0.1 in
prescribed displacement U
2R
Figure 13.2. Square perforated sheet. Initial geometry and boundary conditions. (Reproduced with
permission from Finite elasticity in spatial description: Linearization aspects with membrane applica-
tions, EA de Souza Neto, D Perić and DRJ Owen, International Journal for Numerical Methods in
Engineering, Vol 38
c 1995 John Wiley & Sons, Ltd.)
As in the previous example, the corresponding incompressible Ogden material in plane stress
is adopted in HYPLAS. The initial geometry and the boundary conditions are schematically
illustrated in Figure 13.2. We define the non-dimensional factor
U
γ≡ ,
l
and the simulation is carried out up to γ = 1. A mesh containing 528 three-noded triangles
discretises the symmetric quarter of the sheet. It corresponds to a total 549 degrees of freedom
in the structure. Figure 13.3 shows the finite element mesh at the initial configuration (γ = 0)
alongside the final deformed mesh (γ = 1). The final configuration is reached in two equal-
size steps requiring five Newton–Raphson iterations per step for convergence with a tolerance
of 10−12 in the Euclidean norm of the out-of-balance forces. For both increments, the residual
norm of the k th iteration, ||Gk || is plotted in Figure 13.4 against the residual of the subsequent
iteration k + 1. Note that, as a logarithmic scale is used in both axes, the slope 2:1 indicated
corresponds to quadratic convergence. In Figure 13.5, we plot the results obtained for the
reaction force R on the restrained edge versus the load parameter γ. It is in good agreement
with the numerical solutions presented by Parisch (1986) and Oden (1972).
FINITE STRAIN HYPERELASTICITY 549
initial configuration
(γ = 0)
configuration at γ=1
Figure 13.3. Square perforated sheet. Finite element meshes. (Reproduced with permission from Finite
elasticity in spatial description: Linearization aspects with membrane applications, EA de Souza Neto,
D Perić and DRJ Owen, International Journal for Numerical Methods in Engineering, Vol 38 c 1995
John Wiley & Sons, Ltd.)
0
10
10
-5 2
||Gk+1||
increment 2
-10
10
increment 1
-10 -5 0
10 10 10
||Gk||
Figure 13.4. Square perforated sheet. Convergence behaviour. (Reproduced with permission from Finite
elasticity in spatial description: Linearization aspects with membrane applications, EA de Souza Neto,
D Perić and DRJ Owen, International Journal for Numerical Methods in Engineering, Vol 38 c 1995
John Wiley & Sons, Ltd.)
550 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
100
80
edge reaction, R 60
40
20
0
0.0 0.2 0.4 0.6 0.8 1.0
load factor,
Figure 13.5. Square perforated sheet. Edge reactions. (Reproduced with permission from Finite
elasticity in spatial description: Linearization aspects with membrane applications, EA de Souza Neto,
D Perić and DRJ Owen, International Journal for Numerical Methods in Engineering, Vol 38 c 1995
John Wiley & Sons, Ltd.)
These parameters have been determined by Ogden (1972) in order to fit the experiments
carried out by Treloar (1944) on natural vulcanised rubber. The analytical solution for a
general hyperelastic spherical balloon has been presented by Ogden (1972, 1984). Needleman
(1977) has used the Ritz–Galerkin method to investigate aspherical equilibrium configura-
tions of initially imperfect balloons. The elastic stability of this problem is known to be
crucially dependent on the specific strain-energy function adopted (Beatty, 1987; Ogden,
1972). Pressure instability is detected, in particular, for the three-term Ogden function with
§ The derivation of load-stiffness matrices is discussed in Section 4.3.6, page 106.
FINITE STRAIN HYPERELASTICITY 551
Figure 13.6. Spherical balloon. Finite element mesh. (Reproduced with permission from Finite
elasticity in spatial description: Linearization aspects with membrane applications, EA de Souza Neto,
D Perić and DRJ Owen, International Journal for Numerical Methods in Engineering, Vol 38 c 1995
John Wiley & Sons, Ltd.)
the constants chosen above. For this reason, the Arc-Length Method was employed here in
conjunction with the Newton–Raphson algorithm to allow equilibrium to be found beyond the
instability point. The Arc-Length Method has been described in Chapter 4. For convenience,
we define the normalised internal pressure
p r0
p∗ ≡ ,
2 t0
and the expansion ratio of the balloon
r
λ≡ ,
r0
where r and r0 are, respectively, the current and initial radii of the balloon, t0 is the initial
thickness of the rubber membrane and p is the current internal pressure. Ten increments
were employed to reach the final configuration with λ = 747%. In Figure 13.7, the expansion
ratio obtained in the finite element simulation is plotted versus p∗ . For comparison, Ogden’s
membrane analytical solution is also plotted. The numerical result matches the exact pressure-
expansion curve very closely.
normalized pressure, p
2
0
1 2 3 4 5 6 7 8
expansion ratio, λ
Figure 13.7. Spherical balloon. Pressure-expansion diagram. (Reproduced with permission from Finite
elasticity in spatial description: Linearization aspects with membrane applications, EA de Souza Neto,
D Perić and DRJ Owen, International Journal for Numerical Methods in Engineering, Vol 38 c 1995
John Wiley & Sons, Ltd.)
0.20
r2
0.10 A r1
initial configuration
0.05 r1 = 9.5
r2 = 14.5
0.00
0 40 80 120 160
radii r1, r2 (cm)
Figure 13.8. Rugby ball. Pressure-expansion diagram. (Reproduced with permission from Finite
elasticity in spatial description: Linearization aspects with membrane applications, EA de Souza Neto,
D Perić and DRJ Owen, International Journal for Numerical Methods in Engineering, Vol 38 c 1995
John Wiley & Sons, Ltd.)
(a) (b)
Figure 13.9. Rugby ball. (a) Initial configuration, and (b) spherical configuration corresponding to
point A of the pressure–expansion diagram. (Reproduced with permission from Finite elasticity in
spatial description: Linearization aspects with membrane applications, EA de Souza Neto, D Perić and
DRJ Owen, International Journal for Numerical Methods in Engineering, Vol 38 c 1995 John Wiley
& Sons, Ltd.)
554 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
clamped
clamped
symmetric
symmetric
clamped
t 0 = 0.1cm
A symmetric B symmetric
R = 10 cm l/2 = 10 cm
Figure 13.10. Flat membranes. Initial geometry and boundary conditions. (Reproduced with permission
from Finite elasticity in spatial description: Linearization aspects with membrane applications, EA
de Souza Neto, D Perić and DRJ Owen, International Journal for Numerical Methods in Engineering,
Vol 38
c 1995 John Wiley & Sons, Ltd.)
0.12
0.10
internal pressure (kg/cm )
2
0.08
0.06
0.04
circular membrane
square membrane
0.02
0.00
0 10 20 30 40 50 60
mid-deflection (cm)
Figure 13.11. Flat membranes. Pressure-deflection diagram. (Reproduced with permission from Finite
elasticity in spatial description: Linearization aspects with membrane applications, EA de Souza Neto,
D Perić and DRJ Owen, International Journal for Numerical Methods in Engineering, Vol 38 c 1995
John Wiley & Sons, Ltd.)
membrane. The initially flat sheet evolves into a spherical-type shape. Figures 13.12(a),
(b), (c) and (d) correspond, respectively, to the mid-deflections dB = 4.94, dB = 12.01,
dB = 17.49 and dB = 27.34 cm. The deformed state shown in Figure 13.12(b) is very close
to the instability point observed in the pressure–deflection curve.
FINITE STRAIN HYPERELASTICITY 555
(a) (b)
(c) (d)
Figure 13.12. Square membrane. Deformed configurations; (a) dB = 4.94 cm; (b) dB = 12.01 cm;
(c) dB = 17.49 cm, and (d) dB = 27.34 cm. (Reproduced with permission from Finite elasticity in
spatial description: Linearization aspects with membrane applications, EA de Souza Neto, D Perić and
DRJ Owen, International Journal for Numerical Methods in Engineering, Vol 38 c 1995 John Wiley
& Sons, Ltd.)
D= 400 mm
(a)
(b) (c)
Figure 13.13. Rubber cylinder pressed between two plates: (a) geometry and boundary conditions;
(b) initial mesh; and (c) deformed mesh at u = 250 mm. (Reproduced with permission from Design of
simple low-order finite elements for large-strain analysis of nearly incompressible solids, EA de Souza
Neto, D Perić, M Dutko and DRJ Owen, International Journal of Solids and Structures, Vol 33, Issue
20–22 c 1996 Elsevier Science Ltd.)
Note that the high bulk modulus adopted here makes the material virtually incompressible,
so that appropriate treatment of quasi-incompressibility is required for a successful finite
element simulation. Here, the F-bar methodology thoroughly discussed in Chapter 15 is
adopted to overcome the problem. This problem has been considered originally by Sussman
and Bathe (1987) in the context of hybrid u/p finite elements and by Simo and Taylor (1991),
who employed a mixed formulation in conjunction with an augmented Lagrangian procedure
to handle near-incompressibility. The geometry of the problem and the boundary conditions
are illustrated in Figure 13.13(a). For symmetry reasons, only one quarter of the cylinder
cross-section is considered in the simulation. It is discretised with the mesh of 48 four-
noded F-bar elements shown in Figure 13.13(b). We remark that this type of finite element is
available in program HYPLAS. The deformed mesh obtained with a prescribed displacement
u = 250 mm of the plate is depicted in Figure 13.13(c). It is in good qualitative agreement
with the deformed mesh shown by Simo and Taylor (1991). The reaction forces per unit
thickness of the plate, obtained for the Mooney–Rivlin and Ogden models, are plotted in
Figure 13.14 against the plate deflection u. The curves are plotted up to u = 200 mm. For both
materials the results shown here agree with those obtained by Sussman and Bathe (1987).
1.6
Mooney–Rivlin material
1.2 Ogden material
reaction, R (KN)
0.8
0.4
0
0 50 100 150 200
plate displacement, u (mm)
Figure 13.14. Rubber cylinder pressed between two plates. Load-deflection curves. (Reproduced
with permission from Design of simple low-order finite elements for large-strain analysis of nearly
incompressible solids, EA de Souza Neto, D Perić, M Dutko and DRJ Owen, International Journal of
Solids and Structures, Vol 33, Issue 20–22
c 1996 Elsevier Science Ltd.)
assumed clamped to a flat rigid base and both plate and base are idealised as rigid bodies with
frictionless contact condition on the boundaries. The bead is modelled as a regularised neo-
Hookean material with constant C = 2.5 and bulk modulus k = 1000. A mesh of 520 four-
noded F-bar axisymmetric elements is used to discretise the bead. Figure 13.16(a) shows the
mesh in its initial configuration. A total vertical displacement u = 0.25 is applied to the plate
in 25 increments. Deformed meshes obtained at different stages of the compression process
are depicted in Figures 13.16(b), (c) and (d). They correspond, respectively, to u = 0.09,
0.17 and the final configuration with u = 0.25. At the early stage shown in Figure 13.16(b),
the lateral surfaces of the seal make contact only with the top plate. At the later stages of
Figures 13.16(c) and (d), contact also occurs between the bead and the rigid base. The reaction
force obtained in the plate is plotted in the diagram of Figure 13.17 versus the imposed
displacement u.
presc. displ., u
plate
r = 20 seal 0.55
1
2.1
Figure 13.15. Elastomeric bead compression. Initial geometry and boundary conditions. (Reproduced
with permission from Design of simple low-order finite elements for large-strain analysis of nearly
incompressible solids, EA de Souza Neto, D Perić, M Dutko and DRJ Owen, International Journal of
Solids and Structures, Vol 33, Issue 20–22
c 1996 Elsevier Science Ltd.)
(a)
(b)
(c)
..
(d)
Figure 13.16. Elastomeric bead compression. Finite element discretisation on the initial configuration
and deformed meshes obtained with the present element: (a) initial configuration; (b) u = 0.09;
(c) u = 0.17; and (d) final configuration, u = 0.25. (Reproduced with permission from Design of simple
low-order finite elements for large-strain analysis of nearly incompressible solids, EA de Souza Neto,
D Perić, M Dutko and DRJ Owen, International Journal of Solids and Structures, Vol 33, Issue 20–22
c 1996 Elsevier Science Ltd.)
will follow path BC. Then, unloading will follow curve D with a third stretch occurring via
DE and so on. The above description shows that the Mullins effect is a clearly dissipative
phenomenon. Therefore, it cannot be modelled by means of purely hyperelastic theories.
The first attempt to model the Mullins effect through a phenomenological constitutive
theory appears to have been made by Gurtin and Francis (1981) in the one-dimensional
context. The Gurtin–Francis uniaxial model is very simple in concept. It is described in the
following.
FINITE STRAIN HYPERELASTICITY 559
14 000
12 000
10 000
reaction force
8000
6000
4000
2000
0
0 0.05 0.1 0.15 0.2 0.25
plate displacement, u
σ E
A
D
B
0 ε1 ε2 ε
as a measure of damage. Note that this phenomenological measure of damage is in fact closely
related to the microscopic mechanism of internal deterioration in filled polymers discussed
in Section 12.1.2. The number of broken polymer chains between filler particles (effectively
defining the state of internal damage) increases as the material is stretched and will remain
constant at strains below the maximum attained strain.¶
Having defined the damage measure, a constitutive equation expressing the uniaxial stress,
σ, as a function of the current axial strain, ε, and the state of damage (characterised by εm ) is
then postulated as
σ(ε, εm ) = f¯(ζ) σ 0 (εm ), (13.113)
where ζ is the relative strain, defined as
ε
ζ≡ . (13.114)
εm
From the definition of εm , it follows that
ζ ≤1 ∀ ε. (13.115)
¯
The function f(ζ), named the master damage curve, satisfies for any ζ
¯ ≤ 1,
0 ≤ f(ζ) (13.116)
and the factor f¯ defines the loss of stiffness experienced by the polymer at strain levels below
the maximum previously attained strain, εm . The following additional condition is imposed
on f¯(ζ):
f¯(1) = 1, (13.117)
so that when the maximum strain occurs at the current time, i.e. when
εm = ε ⇔ ζ = 1,
The stress–strain curve defined by the function σ 0 is called the virgin curve. It is the stress–
strain curve obtained in a simple uniaxial experiment with monotonically increasing strain.
To characterise completely the material parameters for this model, we need, in addition to
the virgin curve, to determine the master damage curve f¯(ζ). This curve is obtained from
unloading experiments.
¶ Here, we disregard other potentially important phenomena such as thermally activated damage, strain rate
effects, etc.
FINITE STRAIN HYPERELASTICITY 561
1.0 E
0.8 E
0.4 E
0.2 E
0
0 0.2 0.4 0.6 0.8 1
axial logarithmic strain
Example
The above concepts can be illustrated in the following simple example. Let us consider the
uniaxial version of the Hencky hyperelastic model of Section 13.2.3. For this model, the axial
Kirchhoff stress is given by
τ = E ε, (13.119)
where E is the logarithmic Young’s modulus and ε is the axial logarithmic strain. A Gurtin–
Francis-type model with underlying Hencky hyperelastic behaviour can be obtained by taking
the above logarithmic-based stress–strain relation as the virgin curve:
τ 0 (εm ) ≡ E εm , (13.120)
where εm is the maximum logarithmic strain, and postulating a master damage curve defined
by, say,
f¯(ζ) ≡ ζ 2 = (ε/εm )2 . (13.121)
The resulting behaviour is illustrated in the graph of Figure 13.19 where the solid straight line
corresponds to monotonic loading and the parabolas plotted with dashed lines correspond
to the stress–strain relation upon unloading/reloading below the maximum attained strains:
εm = 0.4 and εm = 0.8.
We remark that the master damage curve adopted in the present example has been chosen only for the purpose
f¯ = f¯(ζ, εm ), (13.123)
The above refinement allows different experimental unloading curves, obtained by unloading
from different maximum strain levels, to be accurately reproduced by the model.
∂ψ 0
τi = ρ̄ λi , (13.125)
∂λi
with no summation implied on repeated indices.
The crucial step in the definition of the hyperelastic damage model is the assumption
that the above constitutive equation for stress is valid only upon monotonic loading and
reloading with damage evolution. In addition, similarly to (13.113), we postulate a general
stress constitutive function given by
where
∂ψ 0
τi0 (λ1 , λ2 , λ3 ) ≡ ρ̄ λi , (13.127)
∂λi
again with no summation implied on repeated indices. Analogously to its one-dimensional
counterpart, f¯, the function f : [0, 1] → [0, 1] is expressed in terms of some, as yet not
defined, three-dimensional measure of relative strain ξ. It also satisfies
f (1) = 1. (13.128)
Remark 13.5. Definition (13.129) implies that the damage variable grows equally in tension
and compression so that the model cannot distinguish between these two situations. As filled
rubbers typically suffer substantially more damage in tension than in compression, the use
of the present model in situations where the damageable body is subjected to both tension
and compression is expected to produce unrealistic predictions. One possible alternative to
overcome this deficiency could be to link the damage variable to the maximum principal
stretch instead. This approach was adopted, for instance, by Govindjee and Simo (1992).
ρ̄ ψ 0
ξ≡ . (13.130)
D
This completes the characterisation of the model.
564 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
∂ψ 0
τi = f (ξ) ρ̄ λi (no summation)
∂λi
ρ̄ ψ 0
ξ≡
D
Elastoplasticity analogy
Following the definition (13.129) for the damage internal variable, a straightforward analogy
between classical elastoplasticity (refer to Chapter 6) and the present model for elastic
damage may be established by introducing a damage surface (cf. yield surface) in the space
of principal stretches, defined by the equation
For fixed D, the damage surface delimits the region of the space of principal stretches in
which the behaviour of the material is purely hyperelastic without evolution of damage. The
above definitions also imply the following complementarity law associated with the evolution
of damage or damage loading/unloading (cf. plastic loading/unloading criterion):
Φ ≤ 0, Ḋ ≥ 0, ḊΦ = 0. (13.132)
Thermodynamical aspects
Alternatively to the above arguments, the present constitutive equations for damage in filled
polymers can be obtained by postulating the existence of a free-energy function of the form
ρ̄ψ 0
κ
ρ̄ ψ(λ1 , λ2 , λ3 , D) ≡ ρ̄ ψ(ρ̄ ψ (λ1 , λ2 , λ3 ), D) =
0
f dκ, (13.133)
0 D
∂ψ
−ρ̄ Ḋ ≥ 0. (13.135)
∂D
It has been shown by the authors de Souza Neto et al. (1994b) that a sufficient condition for
the Clausius–Duhem inequality to be satisfied is that f be a non-decreasing function. This,
together with the fact that f is a non-negative function, also guarantees that a ψ 0 convex in
λi produces a potential ψ also convex in λi .
C1 = 135 psi.
The master damage curve adopted is plotted in Figure 13.20 (ρ̄ taken as unity). In a
uniaxial test, when subjected to the cyclic axial strain path shown in Figure 13.21(a), the
material described by the parameters above produces the uniaxial stress–strain curve plotted
in Figure 13.21(b). The axial strain measure referred to in the graph of Figure 13.21(b)
has been defined as λ − 1 where λ is the principal stretch in the axial direction. The stress
plotted in Figure 13.21(b) is the corresponding Cauchy principal stress. The solid line is the
virgin curve followed during monotonic loading. The dotted lines correspond to the material
behaviour with ρ̄ψ 0 < D and illustrate the ability of the model to capture the Mullins effect.
1.0
f( )
0.5
0.0
0.0 0.5 1.0
0
= /D
Figure 13.20. Master damage curve. (Reproduced with permission from A phenomenological three-
dimensional rate-independent continuum damage model for highly filled polymers: Formulation and
computational aspects, EA de Souza Neto, D Perić and DRJ Owen, Journal of the Mechanics and
Physics of Solids, Vol 42, Issue 10
c 1994 Elsevier Science Ltd.)
We remark that the hyperelastic-damage model is not incorporated into the standard version
of program HYPLAS that accompanies this book. However, as we shall see below, its
incorporation is a straightforward programming exercise that we will leave for the interested
reader.
Stress updating
Due to the path-dependent nature of the above elastic damage constitutive equations, a
numerical algorithm is required for the evaluation of stress and damage evolution throughout
the deformation history. The derivation of such an algorithm for the present model is
straightforward. The underlying idea is the following. Considering the typical (pseudo-) time
interval [tn , tn+1 ], if the current value of ρ̄ψ 0 at tn+1 is greater than the previous maximum
value (Dn ), then damage evolution necessarily occurs within the interval. If the current value
of ρ̄ψ 0 is smaller than or equal to the previous maximum, then it will be assumed that no
damage evolution has occurred throughout the interval. The algorithm is listed in Box 13.4
in pseudo-code format.
Remark 13.7. The implementation of the algorithm of Box 13.4 in program HYPLAS with, for
instance, the Ogden material as the underlying hyperelastic law, is a straightforward exercise.
FINITE STRAIN HYPERELASTICITY 567
12
10
strain history
8
strain (%)
0
time
(a)
100
virgin curve
unloading/reloading below
75 maximum attained strain
axial stress (psi)
50
25
0
0 2 4 6 8 10 12
axial strain (x100)
(b)
Figure 13.21. The model problem: (a) strain history, and (b) uniaxial stress–strain curve. (Reproduced
with permission from A phenomenological three-dimensional rate-independent continuum damage
model for highly filled polymers: Formulation and computational aspects, EA de Souza Neto, D Perić
and DRJ Owen, Journal of the Mechanics and Physics of Solids, Vol 42, Issue 10 c 1994 Elsevier
Science Ltd.)
The essential modification required to the existing Ogden stress updating implementation
(refer to subroutine SUOGD) in order to accommodate damage is the incorporation of the
operations carried out in items (ii)–(iv). Also, the storage of one extra state variable – the
damage internal variable, D – is required. This variable is to be handled completely analo-
gously to other internal state variables such as, for instance, the accumulated plastic strain
in conventional elastoplasticity models (refer, for example, to subroutine SUVM described
on page 224 for details). A logical flag similar to IFPLAS of SUVM is also needed. This is
addressed in Remark 13.8 below.
T
(i) Given the deformation gradient F n+1 , compute Bn+1 := F F and perform the
spectral decomposition of Bn+1
3
Bn+1 = λ2i Mi
i=1
ENDIF
(iv) Update principal Kirchhoff stresses
0
ρ̄ψn+1
ξ :=
Dn+1
∂ψ 0
τi := f (ξ) ρ̄ λi
∂λi n+1
3
τn+1 := τi Mi
i=1
In contrast to purely hyperelastic theories, the situation here is similar to that of classical
elastoplasticity (with elastic and elastoplastic tangents) in that two possible tangents exist
at points on the virgin curve: one tangent to the loading path and another tangent to the
unloading path (the situation is clearly illustrated in Figure 13.21(b)). Below the maximum
attained strain level of ψ 0 (the dotted lines of Figure 13.21(b), excluding their intersection
with the virgin curve), the tangent is unique.
Now let us recall (refer to Section 13.4.2) that, when the principal stretch representation
is adopted, the essential model-related derivatives required to assemble the spatial tangent
modulus, a, are the principal Kirchhoff stress derivatives: ∂τi /∂bj , where bj = λ2j are the
eigenvalues of B. The only difference between the present model and a standard hyperelastic
one in the calculation of the tangent modulus is that, following what was said in the previous
paragraph, two distinct situations have to be considered:
FINITE STRAIN HYPERELASTICITY 569
1. unloading or reloading with ρ̄ψ 0 < D is assumed to occur, i.e. the process is purely
elastic. The tangent is unique and the corresponding derivatives are obtained by differ-
entiating the principal stress constitutive equation (ii) of Box 13.3 with constant D:
∂τi 1 ∂τ 0 f (ξ) τi0 τj0
= f (ξ) i + (13.136)
∂bj 2λj ∂λj D λj
with no summation implied on repeated indices and with f denoting the derivative
of f ;
Remark 13.8. The incorporation of the above into an existing hyperelastic computational
implementation is also straightforward. One important practical aspect that needs to be
taken into consideration in the implementation of this hyperelastic damage model is that
the decision on whether to use the expression of item 1 or item 2 above has to be made
in the computation of the tangent modulus. This can be done in the same manner as in the
computation of the tangent modulus in elastoplasticity where a logical flag, set before the
tangent modulus computation subroutine is called, indicates which tangent (either elastic
or elastoplastic) is to be computed. The interested reader is referred to subroutine CTVM
(Consistent Tangent computation for the Von Mises model), described in Section 7.4.3
(page 235), where the logical argument EPFLAG (set in subroutine MATICT before CTVM is
called) defines which tangent will be computed. Damage loading (and use of item 2 above)
and unloading/reloading without damage evolution (with use of item 1 above) should be
assumed under the same conditions that set EPFLAG, respectively, to .TRUE. and .FALSE. in
subroutine MATICT of program HYPLAS). Note that a logical algorithmic variable (analogous
to IFPLAS in subroutine SUVM), indicating whether the step is purely elastic (no damage
evolution) or whether there is damage evolution, must be set in the corresponding stress-
updating procedure. As in elastoplasticity, this flag can be transferred between the stress-
updating routine and the tangent modulus computation routine stored in the logical global
array LALGVA.
Figure 13.22. Damageable rubber balloon. Master damage curve. (Reproduced with permission from
A phenomenological three-dimensional rate-independent continuum damage model for highly filled
polymers: Formulation and computational aspects, EA de Souza Neto, D Perić and DRJ Owen, Journal
of the Mechanics and Physics of Solids, Vol 42, Issue 10
c 1994 Elsevier Science Ltd.)
3
normalized pressure, p
2 inflation
A
1
deflation
0
1 2 3 4 5 6
expansion ratio, λ
Figure 13.23. Damageable rubber balloon. Pressure–expansion diagram. (Reproduced with permission
from A phenomenological three-dimensional rate-independent continuum damage model for highly
filled polymers: Formulation and computational aspects, EA de Souza Neto, D Perić and DRJ Owen,
Journal of the Mechanics and Physics of Solids, Vol 42, Issue 10
c 1994 Elsevier Science Ltd.)
FINITE STRAIN HYPERELASTICITY 571
By means of arc-length control, starting from the initial configuration (λ = 1), the internal
pressure is applied gradually and the membrane is inflated until the configuration defined
by λ = 5.182 (point A of Figure 13.23) is reached. At this stage, the load is reversed
and the balloon is deflated returning to its initial configuration. The pressure–expansion
curve obtained is presented in Figure 13.23. As inflation occurs under monotonically
increasing circumferential stretching, the inflation branch of the pressure–expansion diagram
corresponds to the behaviour governed by the strain-energy function ψ 0 . Indeed, it matches
exactly the hyperelastic solution obtained in the example of Section 13.6.5. The deflation
branch of the curve shows clearly the softening effect of damage at the global level. The
pressure–expansion curve shown here has a good qualitative agreement with the balloon
inflation experiment discussed by Beatty (1987). However, in the experiment studied by this
author, a residual circumferential strain was observed after complete deflation of the balloon
(p = 0). Incorporation of this effect would require the consideration of additional internal
variables leading to a theory that allows for description of inelastic deformations with possible
inclusion of viscous effects.
14 FINITE STRAIN
ELASTOPLASTICITY
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
574 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
The formulation of elastoplastic constitutive models in the geometrically nonlinear range can
be traced back to the late 1950s. Early references to this fascinating subject are provided by
the pioneering works of Hill (1958) and Green and Naghdi (1965). In view of its potential
application in many areas of practical engineering interest, it did not take long before large
strain elastoplastic formulations were brought within the finite element context. The first
papers on the subject were published in the 1970s. Initial developments (Argyris and Kleiber,
1977; Argyris et al., 1978; Hibbitt et al., 1970; McMeeking and Rice, 1975; Nagtegaal and de
Jong, 1981) have relied exclusively on the use of hypoelastic-based constitutive formulations
whereby standard infinitesimal elastoplasticity models are extended to the finite strain range
by recasting the original evolution equations in terms of suitably chosen objective (or frame-
invariant) stress rates. During early stages of development, hypoelastic-based descriptions
have been the subject of intense debate within the finite element community. Many con-
troversial issues have arisen, ranging from the use of different objective stress rates in the
formulation of the constitutive equations (Atluri, 1984; Nemat-Nasser, 1982; Perić, 1992) to
fundamental drawbacks such as the possible lack of objectivity of (algorithmic) incremental
constitutive laws (Hughes, 1984; Hughes and Winget, 1980; Rubinstein and Atluri, 1983)
as well as observed oscillatory stress response under monotonic loading (Nagtegaal and
de Jong, 1982) and dissipative behaviour within the ‘elastic’ range (Kojić and Bathe, 1987;
Simo and Pister, 1984). In this context, hyperelastic-based formulations of finite plasticity
have emerged (Simo, 1985; Simo and Ortiz, 1985). Based on the hyperelastic description
of the reversible behaviour (as in the general constitutive theory alluded to in Chapter 3)
in conjunction with the multiplicative elastoplastic split of the deformation gradient (Lee
and Liu, 1967), such theories naturally by-pass the inherent drawbacks of hypoelastic-
based approaches. In particular, dissipative response becomes impossible within the elastic
range and the requirement of incremental objectivity (frame invariance of the algorithmic
constitutive rule) is trivially satisfied. Moreover, when the Hencky strain energy function
(described in Section 13.2.3, page 528) is adopted to model the elastic behaviour, a small
strain format return mapping-based stress-updating procedure can be elegantly recovered
with the use of a suitable exponential map-based integrator for the plastic flow equation
(Cuitiño and Ortiz, 1992a; Eterovic and Bathe, 1990; Perić and Owen, 1991; Perić et al.,
1992; Simo, 1992; Simo and Miehe, 1992; Weber and Anand, 1990). Throughout the last
decade, the hyperelastic-based multiplicative approach has gained widespread acceptance
(see Simo (1998), for an extensive review of both theory and related numerical methods) and
is currently adopted in many commercial finite element codes for large-strain elastoplastic
analysis. Hyperelastic-based formulations have proven successful not only in the context of
conventional isotropic J2 metal plasticity but also with other underlying theories such as
soil plasticity (Meschke and Liu, 1999), damage mechanics (de Souza Neto and Perić, 1996;
de Souza Neto et al., 1992, 1994a, 1998; Li, 1995; Steinmann et al., 1994) and anisotropic
single crystal plasticity (discussed in detail in Chapter 16) as well as in the modelling of
various structural elements such as membranes (Ibrahimbegović, 1994) and shells (Miehe,
1998b).
FINITE STRAIN ELASTOPLASTICITY 575
l
λ= . (14.1)
l0
Let us now assume that, at the deformed configuration defined by l, the bar has already
been loaded past the elastic limit. Upon complete unloading from this configuration, the
plastified bar will have a (permanently deformed) length lp . The length lp , thus, defines
an unstressed configuration of the bar. This configuration will also be referred to as the
intermediate or plastic configuration. The axial plastic stretch, that is, the stretch associated
with the unstressed configuration is defined, similarly to (14.1), as
lp
λp = . (14.2)
l0
The unloading process, that is, the deformation from the configuration defined by l to the
unstressed configuration defined by lp , is elastic. This motivates the definition of the elastic
axial stretch,
l
λe = . (14.3)
lp
From the above, we establish that any deformed state of the bar is characterised by the
multiplicative split of the axial stretch (14.1) into an elastic stretch, λe , and a plastic stretch,
λp , i.e.
λ = λe λp . (14.4)
This expression is a generalisation of the additive split (6.2) of the infinitesimal strain to the
finite strain case.
576 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
λ = λe λp
ε̄˙p = γ̇
(vi) Loading/unloading criterion
Φ ≤ 0, γ̇ ≥ 0, γ̇Φ = 0
of the accumulated logarithmic axial plastic strain, ε̄p , defined by the evolution equation
The determination of the plastic multiplier γ̇ under plastic flow follows the same steps as in
Section 6.2.7. The final expression reads
E λ̇
γ̇ = sign(τ ) . (14.17)
H +E λ
Equivalently, in terms of the total logarithmic axial strain,
ε = ln λ, (14.18)
578 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
E
γ̇ = |ε̇|, (14.19)
H +E
which has the identical functional format as its infinitesimal counterpart given by equa-
tion (6.30).
The main hypothesis underlying the finite strain elastoplasticity constitutive framework
described here is the multiplicative decomposition of the deformation gradient, F , into
elastic and plastic contributions; that is, it is assumed that the deformation gradient can be
decomposed as the product
F = F eF p, (14.20)
where F e and F p are named, respectively, the elastic and plastic deformation gradients. The
multiplicative split of F , introduced by Lee and Liu (1967) and Lee (1969), embodies the
assumption of the existence of a local unstressed intermediate configuration defined by the
plastic deformation gradient, F p . At each material point, the local intermediate configuration
is obtained from the fully deformed configuration by a purely elastic unloading (associated
with the inverse of F e ) of its neighbourhood. The concept is schematically illustrated in
Figure 14.1. It is a multiaxial generalisation of the one-dimensional multiplicative split (14.4).
It must be emphasised that, unlike in the one-dimensional case, the unstressed intermediate
configuration concept in the multiaxial case is valid only in the local (pointwise) sense.
Compatible unstressed configurations of multiaxially stretched elastoplastic bodies do not
exist in general; that is, it is generally not possible to find a deformation ϕp such that
dϕp
Fp = .
dp
For a number of years since its introduction, the multiplicative kinematic description has
been the subject of intense investigation (Dafalias, 1984, 1985; Dashner, 1986; Lubarda and
Lee, 1981; Mandel, 1973; Nemat-Nasser, 1979, 1982) and much light has been shed on its
constitutive implications. Some aspects and kinematic consequences of the multiplicative
decomposition assumption are addressed in the following.
FINITE STRAIN ELASTOPLASTICITY 579
initial
configuration current
configuration
p
x = ϕ( p)
F = F e Fp
Fp
Fe
local
intermediate
configuration
Remark 14.1. Rather than being just a convenient mathematical extension of the standard
additive decomposition of the strain tensor adopted in the infinitesimal theory, the multiplica-
tive decomposition of F finds a solid and consistent physical justification in the slip theory
of crystals (Asaro, 1983; Peirce et al., 1982, 1983; Rice, 1971). Essentially, the macroscopic
plastic deformation of metallic crystals is the result of microscopic sliding between blocks
of crystals along certain crystallographic planes (for more details on the plastic deformation
of crystals, we refer to Chapter 16, where the computational treatment of an elastoplastic
single crystal model is addressed). This mechanism is schematically illustrated in Figure 14.2.
Under generic finite straining, crystal blocks slide over each other under the simultaneous
action of lattice rotation and distortion. The plastic deformation gradient, F p , is associated
with the pure sliding between crystal blocks. The elastic contribution, F e , on the other hand,
represents the crystal lattice rotation and distortion.
pure
crystallographic
slip
e
F
p F
lattice
lattice rotation
distortion
initial
crystallographic
arrangement
The meaning of the above tensors is analogous to that of the tensors obtained from the
polar decomposition (3.40) (page 49) of the total deformation gradient; that is, U p represents
a pure plastic stretch along three mutually orthogonal axes starting from the reference
configuration, U e represents a pure elastic stretch along three mutually orthogonal axes
starting from the intermediate configuration, Rp is the pure rotation between the reference
and intermediate configurations, and so forth.
With the introduction of (14.20) into the above formula, a straightforward differentiation
using the product rule gives the following additive decomposition of L,
L = Le + F e Lp (F e )−1 , (14.24)
We shall refer to Le and Lp , respectively, as the elastic and plastic velocity gradients.
Analogously to the velocity gradient discussed in Section 3.1.8, the plastic velocity
gradient as defined in the above is a quantity of the intermediate configuration. The velocity
gradient, L, itself is a quantity of the spatial (or deformed) configuration. Note that,
FINITE STRAIN ELASTOPLASTICITY 581
where dpiis the eigenvalue of D associated with the eigenvector ei . Each dpi gives the
p
will be called the spatially rotated (or simply spatial) plastic stretching (or rate of plastic
deformation). It can be represented as
p
3
D̃ = dpi ẽi ⊗ ẽi , (14.29)
i=1
where ẽi ≡ Re ei .
Be = F e (F e )T = (V e )2
where
εed ≡ εe − 1
3 tr[εe ] I,
and the volumetric logarithmic elastic strain is given by
εev ≡ tr[εe ] = ln J e ,
with
J e ≡ det F e .
Due to the properties of the logarithmic strain measure (refer to the discussion starting on
page 529), analogously to the infinitesimal theory, a traceless εe (εev = 0) corresponds to a
finite volume-preserving elastic deformation, i.e. a deformation with det F e = 1.
FINITE STRAIN ELASTOPLASTICITY 583
∂Ψ
D̃p = γ̇ , (14.37)
∂τ
complemented by postulating a zero plastic spin†
p
W = 0. (14.38)
The two equations above completely define the evolution of F p . Indeed, the definition (14.28)
of D̃p together with the constitutive laws (14.37) and (14.38) are equivalent to the following
evolution law for the plastic deformation gradient
p ∂Ψ e
Lp ≡ Ḟ (F p )−1 = γ̇ (Re )T R , (14.39)
∂τ
which is a multidimensional extension of the one-dimensional plastic flow equation (14.14).
Remark 14.2. Plastic isotropy implies that τ and ∂Ψ/∂τ are coaxial (i.e. they share the
same principal axes). Thus, in the above constitutive equation, the (instantaneous) principal
directions of stretching of the intermediate configuration are the directions obtained by
rotating the principal axes of the Kirchhoff stress back to the intermediate configuration.
In the absence of elastic rotations, the principal directions of plastic stretching coincide with
those of the Kirchhoff stress.
Under elastoplastic isotropy (a condition to which the present model is restricted), the
above plastic flow rule is equivalent to the following constitutive equation for the total plastic
contribution, F e Lp (F e )−1 , to the velocity gradient (14.39),
∂Ψ
F e Lp (F e )−1 = γ̇ . (14.40)
∂τ
The equivalence between the two equations can be established by first making use of the left
polar decomposition of F e , which after a straightforward rearrangement gives
∂Ψ e e
Lp = γ̇ (Re )T (V e )−1 V R . (14.41)
∂τ
Due to the assumed elastoplastic isotropy, V e and ∂Ψ/∂τ commute so that the terms V e and
(V e )−1 are cancelled out of the above equation, leading to (14.39).
† The assumption of zero plastic spin is compatible with plastic isotropy – the condition to which the present
model is restricted. Introduction of plastic anisotropy requires the definition of an appropriate constitutive equation
for the plastic spin. In the anisotropic single-crystal model described in Chapter 16, a micromechanically-based
constitutive law which defines a generally non-vanishing plastic spin is adopted. The reader is referred to Dafalias
(1984, 1985) for an in-depth discussion on this topic.
FINITE STRAIN ELASTOPLASTICITY 585
∂Ψ
Le = L − γ̇ . (14.43)
∂τ
The above equation is a finite strain counterpart of the elastic strain evolution law (7.6)1 and
is obtained simply by introducing the plastic flow equation (14.40) into the velocity gradient
expression (14.24)–(14.25).
Another alternative definition of the present model in terms of an elastic rate can be made
by exploiting the so-called Lie derivative concept (the reader is referred to Marsden and
Hughes 1983, for a precise mathematical definition of the Lie derivative.) Let
denote the Lie derivative of Be with respect to the velocity field v. The above derivative
is identified with the Oldroyd objective rate‡ introduced in (14.130)–(14.131). The plastic
flow rule (14.39) (or (14.40)) is equivalent to the following constitutive equation for the Lie
derivative of the elastic left Cauchy–Green tensor
1 ∂Ψ e
£v Be = −γ̇ B . (14.45)
2 ∂τ
The equivalence between the two constitutive equations can be established as follows. We
start by post-multiplying both sides of (14.40) by Be and expanding the resulting expression
with use of (14.24). This gives
∂Ψ e
γ̇ B = F e Lp (F e )−1 Be
∂τ
= [L − Le ]Be
= L Be − F˙ e (F e )T . (14.46)
‡ The concept of objective rates will be discussed only in Section 14.10.1 in connection with the definition of
hypoelastic-based elastoplasticity models. The reader who is unfamiliar with the concept of objective rates of a
tensor may prefer to read that section before proceeding further. We remark, however, that the alternative formulation
presented here in terms of the Lie derivative (or Oldroyd objective rate) of the elastic left Cauchy–Green tensor is
not required for the computational treatment of the model discussed in Section 14.4.
586 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Due to the isotropy of Ψ and the elastic response, the tensors Be and ∂Ψ/∂τ commute (refer
to Section A.1.2, page 732) so that the left-hand side of the above equation is symmetric and
the right-hand side must equal its symmetrised counterpart; that is, we have
∂Ψ e 1
γ̇ B = [L Be + Be LT − F˙ e (F e )T − (F e )T F˙ e ]. (14.47)
∂τ 2
Now, note that the definition of Be gives
Ḃe = [F e (F e )T ]˙ = F˙ e (F e )T + (F e )T F˙ e . (14.48)
Equation (14.45) is recovered by substituting this formula into (14.47) and taking the
definition (14.44) into account.
If the potential formulation of the internal variable evolution is adopted, then the constitutive
function H is given by
∂Ψ
H(τ, A) = − . (14.50)
∂A
In the above, the plastic multiplier, γ̇, is required to satisfy the standard complementarity
relation
Φ ≤ 0, γ̇ ≥ 0, γ̇Φ = 0. (14.51)
F = F eF p
Φ ≤ 0, γ̇ ≥ 0, γ̇Φ = 0
To obtain the first summand of the last expression in the above, we have made use of identity
(ii) of page 22. Note that the tensors εe , Be and ∂ψ/∂εe in (14.53) share the same principal
axes. The situation here is completely analogous to that leading to equation (13.62) (page 529)
for the Hencky hyperelastic model. Accordingly, the following identity holds in the present
case
∂ψ ∂(ln Be ) e ∂ψ
: B = e, (14.54)
∂εe ∂Be ∂ε
and then, (14.53) can be rewritten as
1 ∂ψ 1
ψ̇ = : Ḃe Be −1 + A ∗ α̇. (14.55)
2 ∂εe ρ̄
and
(F p −T )˙F p T = −F p −T (F p T )˙= −Lp T , (14.58)
588 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
where λp(i) are the principal plastic stretches, i.e. the eigenvalues of the plastic stretch tensor
V p ≡ F pF pT .
FINITE STRAIN ELASTOPLASTICITY 589
det F p = 1,
εpv = 0.
In other words, the (local) plastic deformation is volume-preserving if and only if the above
defined volumetric plastic strain vanishes.
Attention is now focused on the condition under which isochoric plastic flow occurs within
the present finite strain plasticity theory. As J p is constant under volume-preserving plastic
flow, it follows that the plastic flow is isochoric if and only if the rate of the above-defined
volumetric plastic strain vanishes,
ε̇pv = 0.
By using the chain rule together with the standard rule for differentiation of the determinant
of an invertible tensor, we obtain the identity
1
ε̇pv ≡ (ln[det F p ]). = (det F p ).
det F p
p
= tr[Ḟ (F p )−1 ].
From the evolution law (14.39) for F p , it follows that the rate of the above defined volumetric
plastic strain is given by
p ∂Ψ
ε̇v = γ̇ tr , (14.65)
∂τ
so that, clearly, as in the infinitesimal theory, flow potentials whose derivative with respect
to stress – the flow vector – is traceless (such as the classical von Mises and Tresca
functions) produce isochoric plastic flow. In summary, within the present framework, the
crucial infinitesimal plasticity property of plastic incompressibility under a traceless flow
vector is naturally extended to the finite strain range.
Analogously to the small strain theory, if Φ is taken as the flow potential, then the principle
of maximum plastic dissipation discussed in Chapter 6 (see page 170) is extended to the
finite strain range. In that case, the loading/unloading criterion (14.51) is the Kuhn–Tucker
optimality condition for the left-hand side of (14.63) to reach a maximum subjected to the
plastic admissibility constraint, Φ ≤ 0.
590 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
and
γ̇(t) ≥ 0, Φ(τ(t), A(t)) ≤ 0, γ̇(t)Φ(τ(t), A(t)) = 0 (14.67)
for each instant t ∈ [t0 , T ], with
∂ψ ∂ψ
τ(t) = ρ̄ , A(t) = ρ̄ (14.68)
∂εe t ∂α t
Remark 14.3. In the above definition of the constitutive initial value problem, we have
used the plastic deformation gradient as the primary kinematic unknown in the system
of differential equations of evolution. This is conceptually at variance with Problem 7.1
(page 193) for the infinitesimal strain case, where the elastic strain has been chosen as the
primary kinematic variable. The choice of the plastic deformation gradient in Problem 14.1
is motivated only by the fact that it allows, in the finite strain context, a more straightforward
derivation of the integration algorithm. We stress, however, that in both cases (small and
large strains) the basic initial value problem (and the corresponding numerical integration
FINITE STRAIN ELASTOPLASTICITY 591
algorithm) may be equivalently defined in terms of evolution of either the elastic or the plastic
kinematic variable. In Problem 14.1, the equivalent formulation having F e as the primary
kinematic unknown is obtained by replacing (14.66)1 with the differential equation (14.42)
for the elastic deformation gradient.
The above expressions have the same functional format as their infinitesimal counterparts.
or, equivalently, in view of the isotropy of the tensor exponential function (refer to (B.8)),
p eT ∂Ψ
F n+1 = Rn+1 exp ∆γ Ren+1 F pn . (14.73)
∂τ n+1
Remark 14.4. The incompressibility of the plastic flow for pressure insensitive flow
potentials is carried over exactly to the incremental rule (14.73). Indeed, for a traceless flow
592 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
vector, ∂Ψ/∂τ, det[exp[∆γ ∂Ψ/∂τ]] = 1 which ensures that the updating formula (14.73)
is volume preserving. This basic property of the exponential map is listed in Section B.1.1
of Appendix B. Had we adopted, for instance, a standard backward Euler difference scheme
to discretise the plastic flow equation (14.66)1, the resulting updating formula for the plastic
deformation gradient would be
−1
e T ∂Ψ
F pn+1 = I − ∆γ Rn+1 Re
F pn . (14.74)
∂τ n+1 n+1
By simple inspection, we can easily establish that the above update formula for F p is not
volume-preserving in general; that is, given F pn such that det F pn = 1 and a traceless ∂Ψ/∂τ,
the resulting F pn+1 is such that, in general, det F pn+1
= 1. This approach would result in
substantial accuracy loss in the numerical integration of elastoplastic constitutive equations
of plastically incompressible models.
i.e. the gradient of the deformation that maps the configuration of time tn onto the
configuration of tn+1 , with ∆u denoting the corresponding incremental displacement field.
Note that (14.75) is a finite strain version of the infinitesimal elastic strain incremental
equation (7.10)1 (page 194).
With the above discretised evolution equations at hand, we now proceed to state the
incremental version of Problem 14.1. In order to remain consistent with the infinitesimal
counterpart (Problem 7.2, where the elastic strain is the primary kinematic unknown), rather
than using (14.73), we shall here adopt the equivalent formula (14.75) in the definition of
the incremental problem. This will result in a reduced system having the elastic deformation
gradient as the primary kinematic unknown.
Problem 14.2 (The incremental finite plasticity problem). Given F en and αn at the
beginning of the interval [tn , tn+1 ] and given the prescribed incremental deformation
gradient, F ∆ , for this interval, solve the following system of algebraic equations
∂Ψ
F n+1 = F ∆ F n Rn+1 exp −∆γ
e e eT
Re
∂τ n+1 n+1 (14.77)
αn+1 = αn + ∆γ Hn+1
FINITE STRAIN ELASTOPLASTICITY 593
for the unknowns F en+1 , αn+1 and ∆γ, subjected to the constraints
with
∂ψ ∂ψ
τn+1 = ρ̄ e , An+1 = ρ̄ (14.79)
∂ε n+1 ∂α n+1
and the kinematic relations
e
εen+1 = ln Vn+1
e 1
Vn+1 = [F en+1 (F en+1 )T ] 2 (14.80)
Ren+1 = e
[Vn+1 ]−1 F en+1 .
After the solution of the above problem, the updated plastic strain can be determined promptly
from the multiplicative elastoplastic split as
Clearly, Problem 14.2 has the same structure as its infinitesimal counterpart (Problem 7.2)
and all arguments leading to the establishment of the elastic predictor/return-mapping scheme
in Section 7.2.3 remain valid in the present case. The resulting large-strain algorithm is
described next.
n+1 , An+1 }, obtained from the potential constitutive relations at the elastic
If the pair {τtrial trial
n+1 , An+1 ) ≤ 0,
trial
Φ(τtrial (14.83)
then the elastic trial state is accepted as the actual state at tn+1 . Otherwise, we solve the
return-mapping equations
∂Ψ
F n+1 = F n+1 Rn+1 exp −∆γ
e e trial eT
Re
∂ τ n+1 n+1
αn+1 = αtrial + ∆γ Hn+1 (14.84)
n+1
Φ(τn+1 , An+1 ) = 0,
for F en+1 , αn+1 and ∆γ, with τn+1 and An+1 obtained from their potential constitutive
relations.
594 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
or, equivalently, by making use of (B.9) in the inversion of the exponential term,
e ∂Ψ e trial
Vn+1 exp ∆γ = F n+1 Ren+1
T
. (14.86)
∂ τ n+1
Then, a further post-multiplication of each side by its transpose, together with the use of
property (B.10), gives
e ∂Ψ e trial 2
Vn+1 exp 2 ∆γ V e = (Vn+1 ) . (14.87)
∂ τ n+1 n+1
Recall that, due to the assumed elastic and plastic isotropy, V e and ∂Ψ/∂τ commute. Then,
by rearranging the terms and taking the square root of both sides of (14.87), we obtain
∂Ψ
e
Vn+1 e trial
= Vn+1 exp −∆γ . (14.88)
∂ τ n+1
Further, by taking the tensor logarithm of both sides of (14.88), we obtain the much simpler
formula in terms of logarithmic Eulerian strain tensors
∂Ψ
εn+1 = εn+1 − ∆γ
e e trial
. (14.89)
∂τ n+1
Remarkably, the above expression has the same format as the elastic strain update formula
of the backward return-mapping algorithms of the infinitesimal theory derived in Chapter 7
(compare (14.89) with expression (7.25)1 of page 196).
Finally, by replacing (14.84)1 with its equivalent (14.89) in terms of logarithmic strains,
the return-mapping equation system of the finite strain incremental problem is reduced to
∂Ψ
trial
εen+1 = εen+1 − ∆γ
∂τ n+1
αn+1 = αn + ∆γ Hn+1 (14.90)
Φ(τn+1 , An+1 ) = 0,
which has the same functional format as the infinitesimal counterpart (7.25).
FINITE STRAIN ELASTOPLASTICITY 595
• firstly, after retrieving the elastic logarithmic strain, εen , of tn , we recover the corre-
sponding elastic left Cauchy–Green tensor according to
• the elastic trial left Cauchy–Green tensor is computed next by means of the formula
trial
Ben+1 ≡ F n+1
e trial e trial T
(F n+1 ) = F ∆ Ben (F ∆ )T ; (14.94)
• with Ben+1
trial
at hand, the elastic trial logarithmic strain tensor is computed as
trial
εe trial = 1
2 ln[Ben+1 ].
In program HYPLAS, this calculation is carried out in subroutine LOGSTR. The actual
computation of 12 ln[Bn+1 etrial
] is performed in the general subroutine ISO2, called by
LOGSTR. Subroutine ISO2 computes general isotropic tensor functions of a single
tensor belonging to the family described in Section A.5 of Appendix A. The above
function, 12 ln[·], is a particular member of this class of isotropic tensor functions;
αtrial
n+1 = αn ;
• having computed the elastic trial state, we proceed now to the standard small strain
integration algorithm to update εe , α and the corresponding forces τ and A. This
procedure comprises the standard admissibility check followed by the application of
the return-mapping procedure – solution of the system (14.90) – when the trial state is
not plastically admissible;
596 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Box 14.3. General integration algorithm for isotropic multiplicative finite strain
elastoplasticity.
F ∆ := I + ∇n [∆u], F n+1 := F ∆ F n
• finally, the updated Cauchy stress tensor, required to assemble the internal finite
element force vector f int
(e) , is evaluated
1
σn+1 = τn+1 .
det F n+1
The overall algorithm for integration of the large-strain elastoplastic constitutive equations is
listed in Boxes 14.3 and 14.4.
Remark 14.5. The operations carried out in items (i), (ii) and (iv) of Box 14.3 are related
exclusively to the kinematics of finite strains. Due to the use of the logarithmic elastic
strain measure in conjunction with the backward exponential approximation (14.73) to
the plastic flow rule, the essential material-related stress-updating procedure, shown in
Box 14.4, preserves the format of the general elastic predictor/return-mapping algorithm for
infinitesimal plasticity described in Chapter 7 and summarised in Box 7.1 (page 199). Further,
if the Hencky model (linear relation between the Kirchoff stress and the elastic logarithmic
strain) is adopted to describe the elastic behaviour, then the resulting algorithms will have the
same functional format as the specific procedures seen in Part Two of this book, for which
the return mapping is effectively carried out in the space of stresses. In computational terms,
it means that all small-strain elastoplastic integration subroutines described in Part Two can
be reused in the finite strain range without modification. For instance, in the finite strain
extension to the von Mises model, subroutine SUVM (Boxes 7.3 and 7.4 and Section 7.3.5)
is used in program HYPLAS to carry out the procedure of Box 14.4. Clearly, the kinematic
operations of Box 14.3 have to be performed before and after these routines are called.
FINITE STRAIN ELASTOPLASTICITY 597
trial trial
trial ∂ψ trial ∂ψ
τn+1 = ρ̄ , An+1 = ρ̄
∂ εe n+1 ∂ α n+1
α − α − ∆γ H ( τ , A ) = 0
n+1 n n+1 n+1
0
Φ(τn+1 , An+1 )
(iv) EXIT
Remark 14.6. It is important to emphasise that the simplicity of the integration algorithm
of Boxes 14.3 and 14.4 came as a result of the assumptions of elastoplastic isotropy and
the particular implicit exponential approximation adopted to discretise the plastic flow rule.
As already mentioned, other schemes such as the standard backward Euler method of
equation (14.74) or the fully explicit (forward) Euler method may be used to discretise the
plastic flow law. However, the adoption of schemes other than the exponential approximation
will generally result in quite complex integration algorithms in which kinematic and material-
related operations cannot be carried out separately.
Firstly, note that in the small strain integration algorithm of Box 14.4, the updated stress
Kirchhoff stress, τn+1 , is obtained as a function of the internal variable set αn , at tn , and the
elastic trial logarithmic strain. This procedure can be regarded as an incremental constitutive
function of the form
trial
τn+1 = τ̃ (αn , εen+1 ). (14.96)
This algorithmic function has the same functional format and properties as the function σ̄
referred to in expression (7.110), page 230, in the small-strain context.
trial e trial
In the general procedure of Box 14.3, εen+1 is computed as a function of Bn+1
p e trial
which, in turn, is a function of F n and F n+1 . With εn+1 at hand, the Kirchhoff stress
is then updated by means of the incremental constitutive function τ̃ (small strain algorithm –
Box 14.4). Thus, the overall operator split algorithm, comprising the operations carried out in
Boxes 14.3 and 14.4, defines an implicit function τ̂ (the incremental constitutive function),
for the updated Kirchhoff stress, that can be generally expressed as
e trial
trial
τ̂ (αn , F n+1 ) = τ̃ (αn , εen+1 (Bn+1 (F pn , F n+1 ))). (14.97)
The term ∂τij /∂Fkq taking part in (14.95) is the derivative of this implicit function. To obtain
this derivative, we apply the chain rule to (14.97) and obtain
trial e trial
∂ τ̂ ∂ τ̃ ∂εen+1 ∂Bn+1
= e trial : e trial
: . (14.98)
∂F n+1 ∂εn+1 ∂Bn+1 ∂F n+1
Substitution of this expression into (14.95) results, after straightforward manipulations, in the
following closed formula for the components of the spatial tangent modulus consistent with
the present operator split algorithm
1
aijkl = [ D : L : B ]ijkl − σil δjk , (14.99)
2J
where D is the small-strain elastic or elastoplastic consistent tangent operator, associated
exclusively with the integration algorithm of Box 14.4
∂τ̃
D= trial
. (14.100)
∂εen+1
e trial
i.e. it is the derivative of the tensor logarithm function at Bn+1 . The fourth-order tensor B
is defined by the Cartesian components
trial trial
Bijkl = δik (Ben+1 )jl + δjk (Ben+1 )il . (14.102)
FINITE STRAIN ELASTOPLASTICITY 599
implementations are described in Sections 8.1 and 8.2, respectively. Even though any
isotropic model may be formulated and implemented in terms of principal stresses and strains,
it has been emphasised in Chapter 8 that this approach may not be optimal in certain cases. For
example, the computer implementation of the infinitesimal isotropically hardening von Mises
model turns out to be more efficient and clear when the standard component representation is
adopted.
Equivalent principal stress representations are also possible in finite isotropic plasticity.
The general computational schemes for stress updating and tangent modulus evaluation for
isotropic finite strain plasticity described in Sections 14.4 and 14.5 can be alternatively
implemented in principal stress space-based format. This approach has been extensively
exploited by Simo (1992). We also refer to Rosati and Valoroso (2004) for a recently proposed
general framework, also based on the principal space description. An outline of the principal
stress space-based treatment of finite plasticity is presented in the following.
In this section, we reformulate the general numerical integration algorithm of Boxes 14.3
and 14.4 in an equivalent principal stress-based form. We begin the derivation of the fully
trial
principal stress-based algorithm by recalling that, for any isotropic model, εen+1 and τn+1
have the same principal axes. This means that the small-strain part of the algorithm (Box 14.4)
trial
can be implemented in terms of the principal values of τn+1 and εen+1 . By observing that
e trial e trial
Bn+1 and εn+1 also share the same principal directions, we may reformulate the algorithm
of Boxes 14.3 and 14.4 in the following equivalent form
trial
1. Given ∆u, compute F ∆ , F n+1 and Ben+1 (as in Box 14.3).
trial
2. Perform spectral decomposition of Ben+1
3
trial
Ben+1 = bei trial ei ⊗ ei , (14.103)
i=1
εei trial := 1
2 ln bei trial . (14.104)
3. With the principal elastic trial strains and the set αn of internal variables at tn at hand,
use a principal stress-based integration algorithm to obtain the eigenvalues τi , of τn+1 ,
and εei , of εen+1 .
1
3
σn+1 := τi ei ⊗ ei . (14.105)
J i=1
FINITE STRAIN ELASTOPLASTICITY 601
and then assembling the corresponding fourth-order tensor (the product (14.106)) as an
isotropic tensor-valued function of one tensor of the type alluded to in Section 14.5.2.
Remark 14.7. Further efficiency can be gained if the eigenprojection tensors Mi ≡ ei ⊗ ei
(no summation over i) obtained in the spectral decomposition carried out in the stress-
updating stage (item 2 of Section 14.6.1) is stored in memory and then retrieved later for use
in the computation of the above fourth-order tensor. This approach (not adopted in HYPLAS)
trial
avoids the re-computation of the eigenprojections of Ben+1 during the evaluation of the
tangent modulus.
to review Section 9.4 where the infinitesimal plane stress-projected von Mises model is
described in detail. Of particular relevance to our discussion are the constitutive model and
the corresponding integration algorithm summarised, respectively, in Boxes 9.3 (page 374)
and 9.4 (page 377).
The extension of the infinitesimal plane stress-projected model (Box 9.3) to the finite
strain range within the present logarithmic strain-based multiplicative framework is given by
the following set of constitutive equations:
1. the standard multiplicative split (item (i) of Box 14.2) where, now, only the in-plane
components of F , F e and F p are required;
τ = De εe ,
where τ and εe are the arrays of in-plane components of the Kirchhoff stress and
logarithmic elastic strain
Φ= 1
2 τT P τ − 13 σy2 (ε̄p )
γ̇ ≥ 0, Φ ≤ 0, γ̇ Φ = 0.
The operator split integration algorithm in the present case has essentially the same format as
the general finite strain procedure listed in Box 14.3. It comprises the following steps:
(a) Perform the calculations of items (i) and (ii) of Box 14.3 taking into consideration only
the in-plane components of the tensor variables F ∆ , F n+1 , etc.
FINITE STRAIN ELASTOPLASTICITY 603
(b) Having computed the elastic trial state, apply the integration algorithm of the small
strain plane stress-projected model (Box 9.4) to update the in-plane components of τ
and εe as well as the hardening variable. From the Hencky hyperelastic law, the updated
thickness logarithmic elastic strain
This, together with the use of the logarithmic elastic strain measure, allows det F n+1
to be computed from the relation
The above computation is carried out in subroutine TUVM of HYPLAS). Once det F n+1
is computed, we then update the Cauchy stress
−1
σn+1 := (det F n+1 ) τn+1 . (14.112)
Thickness update
Within the finite element environment, the current thickness at Gauss points is required in
the computation of the internal force vector and tangent stiffness. With t0 denoting the initial
thickness (in the reference configuration) at a generic Gauss point, the current thickness t is
given by
t = λ3 t0 . (14.113)
Within the structure of program HYPLAS, where the determinant of the current total
deformation gradient (in-plane deformation gradient in the plane stress case) is readily
available during the stress-update phase (see subroutine MATISU), the thickness stretch, λ3 ,
for the von Mises model can be conveniently computed as follows. With F̃ n+1 denoting the
current in-plane deformation gradient, we have the trivial relation
Thus, after the computation of det F n+1 by (14.111), the thickness stretch is promptly
evaluated as
det F n+1
λ3 := . (14.115)
det F̃ n+1
The thickness is then updated according to (14.113). These operations are performed in
subroutine TUVM of HYPLAS.
1. Perform the calculations of items (i) and (ii) of Box 14.3 taking into consideration only
the in-plane components of the tensor variables F ∆ , F n+1 , etc.
2. With the elastic trial state at hand, apply the small-strain nested iteration procedure
(refer to Box 9.1). The procedure will update τ, the in-plane components of εe and the
internal variables of the model.
3. Perform the calculations of item (iv) of Box 14.3 again taking into consideration only
in-plane components. Again (see item (c), page 603), the determinant of F n+1 (which
includes the thickness stretching) needs to be computed before the Cauchy stress can
be updated. For a plastically incompressible model, det F n+1 can be computed as
described in item (c) of the plane stress-projected finite von Mises algorithm. For more
general models, the following procedure may be adopted. Firstly, let us recall that at
a (plane stress) converged state of the plane stress enforcement loop, both the elastic
trial and updated elastic thickness logarithmic strains, εe33trial and εe33 (with subscripts
n + 1 dropped for convenience), are available as a by-product of the algorithm. Now
let us define the plastic logarithmic thickness strain as
εp33 ≡ ln λp3 .
FINITE STRAIN ELASTOPLASTICITY 605
where the subscript n denotes the (equilibrium) converged value at the end of the
previous load step. To compute det F n+1 , we proceed as follows:
(a) update the plastic thickness strain using (14.117);
(b) then compute the total logarithmic thickness strain using (14.116) and obtain the
total thickness stretch
λ3 := exp(ε33 );
(c) finally compute
det F n+1 := (det F̃ n+1 ) λ3 ,
and update the Cauchy stress according to (14.112).
Thickness update
With λ3 already computed, the current Gauss point thickness is promptly updated with
expression (14.113).
D̃p = G(τ, A)
(14.118)
α̇ = J(τ, A),
where G and J are given explicit functions of τ and the set A of thermodynamical
forces. The resulting model is a finite strain extension (under isotropic conditions) of the
infinitesimal viscoplasticity model of Box 11.3 (page 451). It incorporates, for instance, the
606 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
finite strain extension of all von Mises-based infinitesimal viscoplasticity models discussed
in Section 11.3 (from page 445).
For viscoplastic models with a yield surface, the above evolution equations can be written
in the more specialised format (refer to equations (11.71) on page 456)
where N is the flow vector and H represents the generalised hardening moduli. The scalar
function γ̇ is zero within the elastic domain or on the yield surface and positive outside the
elastic domain as, for instance, in the definitions given in Sections 11.3.1 and 11.3.2.
14.9. Examples
In this section we present a set of finite element examples where the hyperelastic-based
multiplicative plasticity framework described in the preceding sections has been used. Unless
otherwise stated, the results presented have been obtained with program HYPLAS. Again,
in all examples, the full Newton–Raphson algorithm is selected to solve the incremental
equilibrium problem.
Please Wait..
Figure 14.3. Large strain bending of a V-notched bar. Deformed configuration with eight-noded
quadrilaterals mesh. (Reproduced with permission from A new computational model for Tresca
plasticity at finite strains with an optimal parametrization in the principal space, D Perić and EA
de Souza Neto, Computer Methods in Applied Mechanics and Engineering, Vol 171 c 1999 Elsevier
Science S.A.)
Finite element solutions are obtained here with two different meshes. The first mesh, of
eight-noded quadrilaterals with four-point reduced integration, is identical to that employed
in the infinitesimal analysis. The second solution is obtained with a mesh of 312 F-bar four-
noded quadrilaterals. The F-bar four-noded quadrilateral element is described in detail in
Chapter 15. This element is especially suited for large strain analysis of nearly incompressible
materials. The adopted discretisation coincides with that shown in Figure 8.18 (page 344).
In this case, however, the mesh contains only 345 nodes, as opposed to 1001 nodes used
in the corresponding discretisation with eight-noded quadrilaterals. Due to the occurrence
of geometric softening of the bar under large strains, the arc-length procedure described
in Chapter 4 (see Section 4.4, from page 107) is used in conjunction with the Newton–
Raphson algorithm in the equilibrium iterations. A deformed configuration obtained with
the eight-noded elements is depicted in Figure 14.3. The corresponding vertical displacement
of the mid-node of the edge where the forces are applied is 10.288 mm. This configuration
was reached in 30 steps. The results obtained with the F-bar element mesh are shown in
Figure 14.4. Figures 14.4(a) and (b) show the deformed configurations of the bar, with
the corresponding contour plot of accumulated plastic strains, respectively with 2.50 mm
and 12.77 mm of displacement of the mid-node. These configurations have been reached,
respectively, at load steps 10 and 30. The corresponding moment–deflection diagram obtained
in the simulation is shown in Figure 14.5. The normalised applied moment plotted in the
diagram is measured in the current configuration of the specimen. It should be noted that
after the limit load is reached, global softening is observed. This phenomenon – not captured
in the corresponding small strain analysis – is due to the reduction of cross-sectional area
near the notch. The behaviour of the Newton–Raphson algorithm for equilibrium iterations
is illustrated in Table 14.1, where the evolution of the relative residual norm is shown during
a typical load increment. The quadratic rates of convergence of the Newton algorithm, which
results from the consistent linearisation of the algorithms involved, are evident.
Please Wait..
(a)
Please Wait..
(b)
Figure 14.4. Large strain bending of a V-notched bar. Deformed configuration with four-noded F-bar
quadrilaterals mesh. Mid-node displacements: (a) 2.503 mm; and (b) 12.836 mm. (Reproduced with
permission from A new computational model for Tresca plasticity at finite strains with an optimal
parametrization in the principal space, D Perić and EA de Souza Neto, Computer Methods in Applied
Mechanics and Engineering, Vol 171 c 1999 Elsevier Science S.A.)
0.8
bending moment, M/ca2
0.6
0.4
0.2
0
0 2 4 6 8 10 12 14
edge deflection (mm)
Figure 14.5. Finite strain bending of a V-notched bar. Moment–deflection diagram.
Simo and Armero (1992) used this problem, with the von Mises model, to study the
performance of enhanced assumed strain elements in large strain localisation problems.
We first carry out an axisymmetric analysis of the problem. Due to obvious symmetry,
only one quarter of the bar is discretised with the appropriate boundary conditions being
imposed on the symmetry lines. The mesh of 200 four-node axisymmetric F-bar elements
(refer to Chapter 15) shown in Figure 14.6(a) is used. The mesh contains 231 nodal points. A
geometric imperfection of 1.8% of the radius is introduced at the centre of the bar to trigger
the necking. A vertical displacement u = 7.0 mm is imposed incrementally at the top of the
bar. The final deformed meshes, in which the development of necking in the central zone
can be clearly seen, are plotted in Figure 14.6(b) and (c) and correspond, respectively, to the
Tresca and von Mises material models. The final configuration was reached in 23 steps for
the Tresca model and 17 steps for the von Mises model. Figure 14.7 shows the reaction–
displacement curves obtained in the present computation. It is noted that up to approximately
u = 3.0 mm – where the limit reaction is attained – the reactions obtained with the Tresca
model are virtually identical to those obtained with the von Mises material. This is an obvious
consequence of the fact that the dominant stresses up to that stage are uniaxial – a condition
under which the behaviour of both material models coincide in the present set-up. Some
discrepancy in reactions is observed after strain localisation is triggered, in the descent branch
of the displacement–reaction curve. This is attributed to the stronger influence of shear under
localisation. It should be noted that the final deformed configurations of Figures 14.6(b) and
(c) differ quite considerably in the necking zone.
Three-dimensional analysis
We now carry out a three-dimensional analysis of the cylindrical bar necking problem. Only
the von Mises model is used. The analysis reported here was not carried out in program
HYPLAS. Two different meshes – one with 120 and another with 960 eight-noded F-bar
hexahedra – are used to discretise the symmetric octant of the specimen, with the appropriate
boundary conditions imposed on the symmetry planes. The initial meshes are shown in
Figure 14.8. The final deformed configurations with imposed edge displacement u = 7.0 mm
obtained for both meshes are shown in Figure 14.9 along with the von Mises effective
stress contour plot obtained for the finer mesh. For ease of visualisation, the deformed
610 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
symmetric octant has been mirrored to produce a full deformed specimen in Figures 14.9(a)
and (b). The upper half of Figure 14.9(c) shows the distribution of von Mises effective
stress within the longitudinal cross-section of the specimen whereas its lower half shows
the distribution on the boundary of the specimen. Both fine and coarse meshes are able to
predict correctly the necking phenomenon. The final shape resulting from the simulation with
960 eight-node hexahedra is virtually identical to that predicted by the above axisymmetric
simulation. The deformed transversal cross-section obtained with the fine mesh simulation is
depicted in Figure 14.10 where, again, the deformed symmetric octant has been mirrored for
ease of visualisation. The reaction–displacement curves obtained for both three-dimensional
simulations are plotted in Figure 14.11 along with results of the axisymmetric analysis. The
results for the finer three-dimensional mesh are almost indistinguishable from those obtained
in the axisymmetric simulation.
FINITE STRAIN ELASTOPLASTICITY 611
80
40 Tresca
von Mises
20
0
0 1 2 3 4 5 6 7 8
edge displacement (mm)
(a)
(b)
Figure 14.8. Necking of a cylindrical bar. Initial meshes for three-dimensional analysis: (a) 120 eight-
noded F-bar hexahedra; (b) 960 eight-noded F-bar hexahedra. (Reproduced with permission from
Design of simple low-order finite elements for large-strain analysis of nearly incompressible solids,
EA de Souza Neto, D Perić, M Dutko and DRJ Owen, International Journal of Solids and Structures,
Vol 33, Issue 20–22
c 1996 Elsevier Science Ltd.)
In this example we simulate the occurrence of shear bands during the finite stretching of an
elastoplastic rectangular bar. This problem was considered by Simo and Armero (1992) to test
the ability of assumed enhanced strain element formulations in capturing strain localisation.
Here, we use a mesh of F-bar four-noded quadrilaterals under plane strain condition. Due
to symmetry, only one quarter of the bar is discretised. A total number of 200 elements is
used. Figure 14.12(a) shows the initial geometry and the finite element mesh. The von Mises
model is adopted. The material constants are the same as in the previous example except that
the hardening modulus here is
H = −0.012924 GPa.
612 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
. 1.035E+00
8.450E-01
6.550E-01
4.650E-01
2.750E-01
8.503E-02
Figure 14.10. Necking of a cylindrical bar. Three-dimensional analysis. Deformed transversal cross-
section obtained with fine mesh at u = 7.0 mm.
To trigger strain localisation, a width reduction of 1.8% is introduced in the centre of the bar.
The displacement u on the constrained (top) edge is prescribed incrementally until the final
deformed configuration with u = 5.0 mm is reached. The final deformed mesh obtained in
the simulation is shown in Figure 14.12(b). At that stage, a very localised shear band can be
observed. Figure 14.13 shows the displacement–reaction diagram obtained. The substantial
decrease in reaction forces occurring near u = 3.0 mm corresponds to the development of
the localised shear band. To emphasise the need for an appropriate element formulation in
FINITE STRAIN ELASTOPLASTICITY 613
80
40
Axisymmetric
20 3D - 120 elements
3D - 960 elements
0
0 1 2 3 4 5 6 7 8
edge displacement (mm)
Figure 14.11. Necking of a cylindrical bar. Force–displacement diagrams for axisymmetric and three-
dimensional analysis with the von Mises model. (Reproduced with permission from Design of simple
low-order finite elements for large-strain analysis of nearly incompressible solids, EA de Souza Neto,
D Perić, M Dutko and DRJ Owen, International Journal of Solids and Structures, Vol 33, Issue 20–22
c 1996 Elsevier Science Ltd.)
capturing strain localisation phenomena,§ we also plot the results obtained with the Q1/E4
enhanced strain element of Simo and Armero (1992) as well as the results obtained with
the standard four-noded quadrilateral. Both F-bar and Q1/E4 elements are able to capture
localisation (producing substantial global softening) whereas the conventional four-noded
element produces an over-stiff solution with no localisation.
6.413
..tiaW esaelP
..tiaW esaelP
26.667
(a) (b)
Figure 14.12. Plane strain localisation: (a) initial geometry (dimensions in mm) and finite element
mesh; (b) final deformed mesh with u = 5.0 mm. (Reproduced with permission from Design of simple
low-order finite elements for large-strain analysis of nearly incompressible solids, EA de Souza Neto,
D Perić, M Dutko and DRJ Owen, International Journal of Solids and Structures, Vol 33, Issue 20–22
c 1996 Elsevier Science Ltd.)
change in geometry (reduction of cross-section) that occurs as the plate is stretched. In the
infinitesimal theory, the change in geometry is neglected as all calculations are based on the
initial geometry. Thus, this phenomenon cannot be captured by the infinitesimal strain theory
that predicts global hardening in all cases.
10
total load 6
4 Present element
Q1/E4
Standard element
2
0
0 1 2 3 4 5 6
top displacement
Figure 14.13. Plane strain localisation. Force–displacement diagrams. (Reproduced with permission
from Design of simple low-order finite elements for large-strain analysis of nearly incompressible
solids, EA de Souza Neto, D Perić, M Dutko and DRJ Owen, International Journal of Solids and
Structures, Vol 33, Issue 20–22
c 1996 Elsevier Science Ltd.)
discretisation of the sheet. The penalty based approach described by Perić and Owen (1992)
is employed in the treatment of the frictional contact between the sheet, punch and die.
Coulomb dry friction with friction coefficient µ = 0.3 is assumed between the sheet, punch
and die. The surfaces of the punch and the die are discretised, respectively, by 2145 and
612 flat triangular elements. Figure 14.16 shows the finite element meshes used. Note that
within membrane elements, a plane stress state is assumed. In the present case, the plane
stress projected approach to the finite strain von Mises model (described in Section 14.7.1)
is adopted. The reaction force obtained on the punch is plotted in Figure 14.17 against
the punch travel. Figure 14.18 shows the corresponding distributions of radial strain and
thickness along the sheet radius obtained in the simulation. The results are plotted for 10,
20, 30 and 40 mm of punch displacement (dp ). The peak in thickness reduction and radial
strain observed at dp = 40 mm corresponds to a strain localisation phenomenon leading to
the rupture of the workpiece. Strain localisation initiates when the peak reaction is reached
at around dp = 34 mm. The ability accurately to predict rupture in industrial processes of
this nature is of crucial importance in the design of tools. It can bring substantial savings in
the overall design process. For this reason, the Finite Element Method is currently widely
employed in industry for the simulation of thin sheet-forming operations.
16
14
12
0 1 2 3 4 5 6 7
(a)
10
6 0.555E-03
0.555E-04
rate independent
0 1 2 3 4 5 6 7
(b)
0.555E+02
0.555E+01
0.555E-00
0.555E-01
edge reaction (KN)
0.555E-02
2 0.555E-03
0.555E-04
rate independent
0 1 2 3 4 5 6 7
(c)
Figure 14.14. Stretching of a perforated plate. Finite strain solution. Displacement–reaction diagrams.
(a) = 100 ; (b) = 10−1 ; (c) = 10−2 .
FINITE STRAIN ELASTOPLASTICITY 617
punch Rp
Geometry:
X
3 R p = 50.8 mm
t0
Rd = 6.35 mm
R 0 = 59.18 mm
sheet X Rd
1
die t 0 = 1.0 mm
punch
clamped
sym
blank
sym
die
(a) (b)
Figure 14.16. Thin sheet metal forming. (a) Finite element discretization of the sheet, die and punch;
(b) finite element mesh and boundary conditions for the sheet.
618 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
80
40
20
0
0 10 20 30 40
Punch travel, u3 (mm)
Figure 14.17. Thin sheet metal forming. Reaction forces on punch.
1.6 1.2
LEGEND:
1.4 Punch travel (mm):
... 40.0 1.0
... 30.0
1.2 ... 20.0
... 10.0
0.8
Thickness (mm)
1.0
Radial strain
0.8 0.6
0.6
0.4
LEGEND:
0.4 Punch travel (mm):
... 40.0
0.2 ... 30.0
0.2 ... 20.0
... 10.0
0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Radial position (X1/R0) Radial position (X1/R0)
(a) (b)
Figure 14.18. Thin sheet metal forming: (a) radial strain distributions plotted over the initial configura-
tion; (b) thickness distributions.
Section 14.1, early formulations of finite strain plasticity have relied exclusively on the
hypoelastic-based approach to provide ad hoc finite strain extensions to existing infinitesimal
models. In spite of their relative simplicity, hypoelastic-based plasticity models do not
usually account for truly reversible behaviour, even in the absence of plastic flow. In other
words, dissipative behaviour may be predicted even within what is meant to be an ‘elastic’
(reversible) domain. This is certainly an undesirable feature from the theoretical point of
view. In addition, the formulation of incremental constitutive equations which preserve the
objectivity (frame-invariance) of the rate (time-continuum) forms is not trivial and may,
in some circumstances, result in rather cumbersome numerical procedures. Despite such
disadvantages, hypoelastic-based models are used extensively by many researchers and are
currently available (in some cases as the only option) in many commercial finite element
FINITE STRAIN ELASTOPLASTICITY 619
packages. We, therefore, find it appropriate to devote this section to this class of constitutive
models of plastic material.
for any change in observer. Note, for instance, that the material time derivative σ̇ of the
Cauchy stress is not an objective stress rate. In this case, the transformation reads
T
σ̇ −→ Q σ̇ QT + Q̇ σ QT + Q σ Q̇ , (14.122)
and satisfies (14.121) only for changes in observer with time-independent rotation (Q̇ = 0).
In order to ensure material objectivity in the formulation of finite strain constitutive laws
directly in terms of stress rates, it is essential that the constitutive equation for the stress
tensor be defined in terms of objective stress rates. Objective stress rates are usually defined
by suitably modifying the material time derivative of the stress tensor to ensure that (14.121)
is satisfied. Their definition is somewhat arbitrary and many different objective rates of
stress have been proposed in the formulation of hypoelastic-based constitutive theories. The
definition of some of the most commonly used objective rates is reviewed in the following. We
remark that proof of objectivity is provided here only for the Jaumann rate of Cauchy stress
described below. The objectivity of the remaining stress rates can be proved in a completely
analogous way. This will be left as an exercise for the interested reader.
∇
To show that σ is indeed objective, let us recall (again, refer to page 70) that under a
change in observer, the deformation gradient transforms as
F −→ Q F .
By introducing the above into the definition of L and then into the definition of W , we obtain
the following transformation for the spin tensor
W −→ Q̇ QT + Q W QT . (14.125)
With the substitution of the above transformation for W , together with (14.120) and (14.122),
into the definition of the Jaumann rate, we find, after some straightforward tensor algebra, that
the Jaumann rate transforms objectively
∇ ∇
σ −→ Q σ̇ QT − Q W σ QT + Q σ W QT = Q σ QT . (14.126)
Generalising the above definition, let T be a generic spatial tensor; that is, a tensor that
transforms according to (14.120) under a change in observer. The Jaumann rate of T is
defined as
∇
T ≡ Ṫ − W T + T W . (14.127)
In particular, the Jaumann rate of Kirchhoff stress
∇
τ ≡ τ̇ − W τ + τ W (14.128)
is often employed in the formulation of hypoelastic-based finite strain plasticity models.
Of particular relevance is the model defined in terms of the Jaumann rate of Kirchhoff stress.
This model has been widely used to extend conventional isotropic infinitesimal plasticity
models to the finite strain range. Its basic rate evolution law for stress is the following
∇
τ = De : (D − Dp ), (14.138)
De ≡ 2G IS + (K − 23 G)I ⊗ I,
In the ‘elastic’ range (where Dp = 0) the stress rate equation reads simply
∇
τ = De : D. (14.139)
Plasticity equations
∂Ψ
Dp = γ̇ , (14.140)
∂τ
together with the general evolution equation for the internal variables
and the standard load/unload criterion (14.137). The extension of a given infinitesimal model
to the finite strain range is obtained by adopting in the above equations Ψ, Φ and H with the
same functional format as those of the corresponding infinitesimal model. Analogously to the
rate form of infinitesimal plasticity, the stress rate equation (14.138) is reduced under plastic
flow to the following form
∇
τ = Dep : D, (14.142)
where the elastoplastic tangent operator Dep is obtained from (14.138), (14.140) and (14.141)
together with the consistency condition
Φ̇ = 0.
In fact, the operator Dep above has exactly the same format as the elastoplastic (continuum)
tangent operator derived in Section 6.3.8 (page 153) for the generic infinitesimal plasticity
model. This is demonstrated in the following.
FINITE STRAIN ELASTOPLASTICITY 623
The derivation of the continuum operator Dep shown here is completely analogous to that
shown in Section 6.3.8. We start by taking the time derivative of the yield function Φ which,
in the present case, gives
∂Φ ∂Φ
Φ̇ = : τ̇ + ∗ α̇
∂τ ∂α
∂Φ ∇ ∂Φ
= : (τ + W τ − τ W) + ∗ α̇
∂τ ∂α
∂Φ ∂Ψ ∂Φ
= : D : D − γ̇
e
+ W τ − τ W + γ̇ ∗ H(τ, α). (14.143)
∂τ ∂τ ∂α
The above expression can be simplified initially by observing that, in view of the symmetry
of τ and ∂Φ/∂τ and skew-symmetry of W , we have
∂Φ ∂Φ
: (W τ − τ W ) = : [(W τ)T − τ W ]
∂τ ∂τ
∂Φ
= −2 :τW
∂τ
∂Φ
= −2 τ : W.
∂τ
In addition, due to the isotropy of the model, τ and ∂Φ/∂τ commute and their product
remains symmetric, leading to the identity
∂Φ
: (W τ − τ W ) = 0, (14.144)
∂τ
The above expression has identical format to that of the infinitesimal deformation theory.
Finally, by equating the above formula to zero (the consistency condition), we obtain the
general expression for the plastic multiplier
∂Φ/∂τ : De : D
γ̇ = .
∂Φ/∂τ : De : ∂Ψ/∂τ − ∂Φ/∂α ∗ H
Substitution of this expression into (14.140) and the resulting formula into (14.138) followed
by comparison with (14.142) leads to
which has the format of the small strain counterpart (6.67), shown on page 153.
624 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Remark 14.8. Note that the small strain format of Dep in (14.145) is a direct consequence of
the identity (14.144). It will hold whenever the difference between the material time derivative
of the stress and the corresponding objective rate comprises only terms with products between
the stress and a skew-symmetric tensor. It holds, for instance, for the Green–Naghdi rate-based
model defined by
τ = De : (D − Dp ),
with τ given in (14.135). For similar models defined in terms of other objective stress rates,
such as the Truesdell or the convected rate, different elastoplastic operators have to be derived.
10
8
e2
as
∆u
vn+ 12 = . (14.147)
∆t
626 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
where ∇n+ 12 denotes the spatial gradient taken at the midpoint configuration defined by the
displacement field
un+ 12 = un + 12 ∆u. (14.149)
With the above at hand, the rate of deformation and spin tensors at tn+ 12 are computed as
Finally, the Kirchhoff stress tensor can be updated according to the formula
or, equivalently,
F ∆ ≡ I + ∇n (∆u) = Q∆ , (14.153)
or, equivalently,
F ∆ = [I + 12 ∇n+ 12 (∆u)] F 12 ∆ = Q∆ , (14.154)
with
F 12 ∆ ≡ I + 12 ∇n (∆u). (14.155)
From the above, we can easily obtain
The principle of material objectivity requires that the following identity be satisfied
However, by replacing (14.156) into (14.152) we find that the present algorithm updates τ as
+ skew[Q∆ F 1−1
∆
− I] τn − τn skew[Q∆ F 1−1
∆
− I]), (14.158)
2 2
which, in general, does not satisfy (14.157). To see this, note that, in particular, if τn = 0
then under an incremental pure rotation (Q∆
= I) we should have τn+1 = 0. The algorithm
FINITE STRAIN ELASTOPLASTICITY 627
instead gives
τn+1 = 2 De : sym[Q∆ F 1−1
∆
− I]
= 0. (14.159)
2
It is worth remarking that, in spite of its lack of objectivity, the use of the update for-
mula (14.152) may be justified if displacement increments are sufficiently small. This is
the case, for instance, with explicit transient dynamic finite element schemes, where global
stability criteria impose stringent restrictions on increment sizes. The use of (14.152) within
implicit schemes, where steps can be large, will, however, produce unacceptable results.
Under plastic flow, the elastoplastic tangent operator Dep replaces the elasticity tensor
in (14.152) and the stress update formula reads
(i) Firstly, the original rate constitutive equation is mapped into a rigid motion-insensitive
(or rotation-neutralised) local configuration.
(ii) Time discretisation is then performed over the resulting equation which involves only
rigid motion-insensitive quantities.
(iii) Finally, the discretised equation is mapped back to the spatial configuration.
The reader is referred to Chapter 8 of Simo and Hughes (1998) for a detailed description
of the general methodology. Here we shall limit our discussion to a midpoint rule-based
algorithm that is closely related to the popular Hughes–Winget algorithm originally proposed
by Hughes and Winget (1980).
628 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
τ̄ ≡ ΛT τ Λ, D̄ ≡ ΛT D Λ, (14.162)
where Λ is the rotation tensor that solves the initial value problem
Λ̇ = W Λ
(14.163)
Λ|t=0 = I.
From the above definitions, it follows after straightforward tensor algebra that the material
time derivative of the rotated Kirchhoff stress is given by
∇
τ̄˙ ≡ ΛT τ Λ. (14.164)
Again, considering the ‘elastic’ regime, the first step (item (i) above) in the formulation of
the present algorithm is to rotate both sides of the original Jaumann rate equation (14.139)
with the rotation ΛT . This results in the following equivalent rate form defined in the local
rotated configuration
τ̄˙ = De : D̄. (14.165)
In rotating the right-hand side of (14.139) we have made use of the fact that De is isotropic.
The above rate form of constitutive equation for the stress is rigid motion-insensitive, in
the sense that it involves only quantities which are not affected by superimposed rigid-body
motions. This is proved in the following.
Under the same superimposed motion, it can be easily established that the stretching tensor
transforms as
D −→ D ∗ = Q D QT .
To see this, we firstly note that the spin tensor follows the transformation rule (14.125). This
expression for the spin together with (14.163) yields
∗
Λ̇ Λ−1 −→ Λ̇ (Λ∗ )−1 = Q Λ̇ Λ−1 QT + Q̇ QT . (14.166)
The expression for the transformed rotation Λ∗ is obtained simply by integrating the differ-
ential equation on the right-hand side of (14.166). This gives the following transformation
rule for Λ
Λ −→ Λ∗ = Q Λ. (14.167)
FINITE STRAIN ELASTOPLASTICITY 629
From the above and definition (14.162)2 of the rotated stretching, we then find
D̄ −→ ΛT QT (Q D QT )Q Λ = ΛT D Λ = D̄,
so that the right-hand side of (14.165) is indeed rigid motion-insensitive.
To complete the proof, let us now concentrate on the time derivative of the rotated stress.
By taking the time derivative of (14.162)1 we obtain
T T
τ̄˙ = Λ̇ τ Λ + ΛT τ̇ Λ + ΛT τ Λ̇ .
This, together with (14.167) and the transformation rule for τ
τ −→ Q τ QT ,
yields, after some straightforward algebra,
T T
τ̄˙ −→ ΛT Q̇ Q τ Λ +Λ̇ τ Λ
T
− ΛT Q̇ Q τ Λ + ΛT τ̇ Λ
T T
+ ΛT τ Q̇ Q Λ − ΛT τ Q̇ Q Λ + ΛT τ Λ̇ = τ̄.
˙
This concludes the proof that all quantities taking part in (14.165) are rigid motion-
insensitive.
The midpoint rotation tensor, Λn+ 12 , is obtained through the numerical integration of the
initial value problem (14.163), which we will discuss later.
Finally, with the updated rotated Kirchhoff stress computed by (14.168), the Kirchhoff
stress tensor at tn+1 is then obtained simply by rotating τ̄n+1 back to the spatial configuration
(item (iii) listed on page 627)
τn+1 = Λn+1 τ̄n+1 ΛTn+1 , (14.170)
where the rotation Λn+1 is obtained by solving problem (14.163) numerically. For imple-
mentation purposes it is convenient to recast (14.170) in the equivalent form
τn+1 = Λ∆ τn ΛT∆ + ∆t De : Λδ Dn+ 12 ΛTδ
= Λ∆ τn ΛT∆ + De : Λδ sym[∇n+ 12 ∆u] ΛTδ , (14.171)
630 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Λ∆ = exp[∆t W n+ 12 ], (14.176)
Finally, by combining (14.176), (14.175) and (14.173) and making use of property (B.11) of
the exponential map (see page 748), we find
∆t
Λδ = Λ∆ = exp W n+ 12 , (14.178)
2
Remark 14.10. In practical computations, the tensor exponential above can be evaluated
according to expressions (B.13) or (B.14), noting that the use of (B.14) should be avoided
near its singularities. An alternative way to compute Λδ and Λ∆ is by using a quaternion
parametrisation as suggested by Crisfield (1997) and Simo and Hughes (1998).
FINITE STRAIN ELASTOPLASTICITY 631
αtrial
n+1 := αn
n+1 ) ≤ 0
trial
IF Φ(τn+1 , αtrial
THEN set (·)n+1 := (·)trial
n+1 and EXIT
Next, the plastic consistency check is carried out. If the elastic trial stress lies within the
trial
trial elastic domain, then τn+1 = τn+1 . Otherwise, a small-strain format return-mapping
algorithm is used to compute τn+1 and the updated values of the internal variables of
the model. The overall incrementally objective algorithm for the Jaumann rate-based finite
plasticity model is summarised in Box 14.5.
Λ̃∆ = [I − 1
2 ∆t W n+ 12 ]−1 [I + 1
2 ∆t W n+ 12 ]. (14.181)
632 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
exp[ 12 ∆t W n+ 12 ] ≈ [I + 1
2 ∆t W n+ 12 ]. (14.182)
In view of property (B.9) of the tensor exponential, we have the alternative approximation of
the same order
Ω ≡ Ṙ RT ,
with R being the rotation tensor resulting from the polar decomposition of the deformation
gradient
F = R U = V R.
Trivially, the exact solution to problem (14.184) is
Λ = R. (14.185)
As a result, the incrementally objective algorithm for this model follows the same steps of
Box 14.5 except that the incremental rotation tensors Λ∆ and Λδ are redefined as
Λ∆ = R∆ = Rn+1 RTn
(14.186)
Λδ = Rδ = Rn+1 RTn+ 1 .
2
The rotations Rn , Rn+1 and Rn+ 12 are obtained, respectively from the polar decomposition
of F n , F n+1 and the deformation gradient at the midpoint configuration defined by the
displacement field un+ 12 = un + 12 ∆u
F n+ 12 = F n + 1
2 ∇0 (∆u)
1
= 2 (F n + F n+1 ). (14.187)
FINITE STRAIN ELASTOPLASTICITY 633
Note that R∆ is the actual local rigid rotation associated with the incremental displacement
∆u, which can be equivalently obtained from the polar decomposition of the incremental
deformation gradient
F ∆ = I + ∇n (∆u). (14.188)
Rδ , in turn, is the rigid rotation associated with the deformation mapping between the
midpoint configuration and the configuration at the end of the increment. In can be computed
directly from the polar decomposition of
Fδ = I + 1
2 ∇n+ 12 (∆u). (14.189)
where
s ≡ τ − 13 tr[τ] I (14.191)
is now the Kirchhoff stress deviator and σy is the usual isotropic hardening stress defining
the size of the yield surface. Again, mixed hardening can be accounted for by having σy as a
function of, for example, the accumulated plastic strain
σy = σy (ε̄p ). (14.192)
A class of von Mises-based finite strain kinematic hardening models can be defined within
the multiplicative plasticity framework by adopting Ψ ≡ Φ in Box 14.2, where Φ denotes the
above yield function, together with the standard Hencky hyperelastic law and an appropriate
(frame-invariant) evolution equation for the back-stress tensor.
To see this, we must first recall (refer to expression (3.145), page 70), that, under the
superimposed rigid motions, the (total) deformation gradient transforms as
F −→ QF . (14.196)
FINITE STRAIN ELASTOPLASTICITY 635
F e F p −→ QF e F p , (14.197)
F e −→ QF e ; F p −→ F p . (14.198)
Expression (14.198)1 implies that the elastic rotation obtained from the polar decomposition
F e = R e U e transforms as
R e −→ QR e . (14.199)
The proof that (14.195) holds follows analogous steps to those leading to (14.126); that is,
we simply introduce (14.199) and the obvious transformation rule for the spatial tensor β,
β −→ QβQT , (14.200)
together with their corresponding material time derivatives, into the right-hand side
of (14.193).
where β̂ is a given constitutive function of τ, β and the accumulated plastic strain, ε̄p , having
the same functional format as in the back-stress evolution equation of the corresponding
infinitesimal model. In this context, the finite strain extension of Prager’s linear kinematic
hardening rule (6.212) is defined as
2 ∂Φ
β = γ̇ H , (14.202)
3 ∂τ
with nonlinear hardening extension
2 ∂Φ
β = γ̇ H(ε̄p ) . (14.203)
3 ∂τ
For the Armstrong–Frederick model (6.223), we have
2 ∂Φ
β = γ̇ H −bβ . (14.204)
3 ∂τ
For completeness, we summarise in Box 14.6 a finite strain von Mises-type multiplicative
plasticity model with mixed hardening, based on the Hencky hyperelastic description of the
reversible behaviour and Green–Naghdi rate of back-stress tensor.
636 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
σy = σy (ε̄p ); ε̄˙p = γ̇
β = γ̇ β̂(τ, β, ε̄p )
Φ ≤ 0, γ̇ ≥ 0, γ̇Φ = 0
Model response
To illustrate the behaviour predicted by models of the present type, we plot in Figures 14.20
and 14.21 the response of the finite strain extension (14.202) of Prager’s linear kine-
matic hardening model, respectively, over a cyclic strain-controlled uniaxial test and under
monotonic shearing deformation (refer to Figure 14.19, page 625, for the definition of the
shearing problem). For comparison, the response predicted by the finite strain linear isotropic
hardening model with identical hardening modulus is also shown. The adopted material
constants are
E = 210; ν = 0.3; σy = 0.45; H = 1.0.
In the uniaxial test, the behaviour predicted by the finite strain Prager linear law is in line with
what is expected from an extension of its infinitesimal counterpart. In the shearing problem,
however, the Green–Naghdi rate-based extension to Prager’s rule produces unrealistic stress
oscillations, which start at strains of around 150%. Such unexpected oscillations of shear
stress in the finite strain formulation of Prager’s kinematic hardening law were first reported
by Nagtegaal and de Jong (1981) for a hypoelastic-based finite plasticity model with
a Jaumann rate extension of the infinitesimal Prager kinematic hardening law. A more
FINITE STRAIN ELASTOPLASTICITY 637
1.5
1.0
0.5
axial stress
Prager’s rule
-0.5
isotropic
-1.0
-1.5
0.8 0.9 1.0 1.1 1.2
axial stretch, λ
Figure 14.20. Cyclic strain-controlled uniaxial test. Finite strain extension of Prager’s linear kinematic
hardening model and linear isotropic hardening model.
recent study by Dettmer and Reese (2004) has also reported shear stress oscillations in a
hyperelastic-based plasticity model with a Jaumann rate extension of the infinitesimal linear
kinematic hardening rule. For the model discussed in this section, such oscillations can be
eliminated, for instance, by adopting the Green–Naghdi rate-based extension (14.204) of the
Armstrong–Frederick law (note that this law reduces to Prager’s linear rule if b = 0). This is
shown in Figure 14.21. The curve for the Armstrong–Frederick model was obtained with the
same material parameters as above and the additional non-dimensional saturation hardening
parameter b set to
b = 3.
With this choice of b in the present case, before the oscillations start, the stress–strain curve
reaches a saturation state where the behaviour becomes perfectly plastic in the absence
of load reversal, producing a more realistic behaviour under finite shearing. Dettmer and
Reese (2004) showed that, in the presence of linear kinematic hardening only (Prager’s rule),
stress oscillations can be removed and monotonically increasing shear stress obtained with
increasing shear deformation if suitable alternative finite strain extensions of the infinitesimal
model are adopted (such alternatives are briefly described below in Section 14.11.5).
In summary, the example presented here serves to emphasise that care should be exercised
when modelling kinematic hardening effects within the framework of finite plasticity, partic-
ularly if high strains are expected to occur. The use of the finite strain Prager rule (14.202),
for instance, appears to be justifiable only in the presence of moderately large plastic strains.
1.5
isotropic
shear stress, τ12
Prager’s rule
0.5 Armstrong-Frederick
-0.5
0 2 4 6 8
shear deformation, γ
Figure 14.21. Behaviour under monotonic shearing (see problem definition in Figure 14.19). Finite
strain extension of Prager’s linear kinematic hardening model, Armstrong–Frederick kinematic harden-
ing model and linear isotropic hardening model.
is analogous to that discussed in Section 14.10.7 in connection with the integration of the
hypoelastic-based plasticity model defined in terms of the Green–Naghdi stress rate. Here,
however, we shall adopt a fully implicit integration scheme. In this case, the incremental
version of (14.201) reads
Box 14.7. Numerical integration algorithm for the general multiplicative finite
strain elastoplasticity model with kinematic hardening.
F ∆ := I + ∇n [∆u], F n+1 := F ∆ F n
η ≡ s − β. (14.209)
Before proceeding, we should note the crucial step in establishing the equivalence
between (14.87) and the small-strain format update formula (14.89) was the introduction of
the assumption of elastoplastic isotropy, under which, the flow vector, the updated elastic and
the elastic trial left stretch tensor all commute. Here, due to the plastic anisotropy introduced
e e trial
by kinematic hardening, Nn+1 , Vn+1 and Vn+1 do not commute in general (note that
they do commute if s and β share the same principal axes). Consequently, equivalence
between (14.207) and the infinitesimal format update formula for the logarithmic elastic
strain does not hold in general. Fortunately, however, such an equivalence can be established
approximately as we demonstrate in the following.
trial
In terms of the Eulerian logarithmic strains, εen+1 and εen+1 , expression (14.207) can be
written as
trial
exp[εen+1 ] exp[2 ∆γ Nn+1 ] exp[εen+1 ] = exp[2 εen+1 ]. (14.210)
From the series representation (B.4) for the tensor exponential, it follows that
where o(εe ) is a term of second order in the elastic logarithmic strain. This together
with (14.210) gives
trial
exp[2 εen+1 ] = exp[2 ∆γ Nn+1 ] + εen+1 exp[2 ∆γ Nn+1 ]
+ exp[2 ∆γ Nn+1 ] εen+1 + o(εen+1 ). (14.212)
εp∆ ≡ 1
2 [exp(2∆γ Nn+1 ) − I ]. (14.213)
This tensor is a Green–Lagrange-type measure (refer to Section 3.1.7, from page 52)
of incremental plastic strain within the interval considered. Indeed, within the present
FINITE STRAIN ELASTOPLASTICITY 641
exponential map-based discretisation of the plastic flow (14.73), the incremental plastic
deformation gradient, F p∆ , defined by the relation
F pn+1 = F p∆ F pn , (14.214)
F p∆ = U ∆p = V∆p = Ren+1
T
exp[∆γ Nn+1 ]Ren+1 . (14.215)
The tensor defined by (14.213) is the Green–Lagrange strain measure associated with the
above incremental plastic stretch, rotated to the spatial configuration (by Ren+1 ).
With definition (14.213) at hand, we can rewrite (14.212) equivalently as
trial
exp[2 εen+1 ] = I + 2 εp∆ + 2[εen+1 + εen+1 εp∆ + εp∆ εen+1 ] + o(εen+1 ). (14.216)
We then take the logarithm of both sides of the above equation, making use, on its right-hand
side, of the series representation of the tensor logarithm
∞
(−1)k
ln(I + X) = X k+1 , (14.217)
k+1
k=0
In addition, by noting that, in view of the definition of εp∆ , the first term on the right-hand
side of the above equals 2∆γNn+1 , and expanding the second term, we obtain
trial
εen+1 = εen+1 + ∆γ Nn+1 + o(εp∆ ) + o(εen+1 ), (14.219)
Finally, under infinitesimal elastic strains (typical of metal plasticity), the second-order term
can be neglected in the above expression leading to the approximate update formula
trial
εen+1 = εen+1 − ∆γ Nn+1 (14.221)
which, remarkably, has the same small-strain format as the fully isotropic formula discussed
in earlier sections of this chapter.
642 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
αn ≡ {α∗n , βn }, (14.223)
where α∗n is the subset of state variables at tn , excluding the back-stress tensor.
The key difference between the present and the isotropic case comes from the replacement
of (14.97) with (14.222) in the derivation of the derivative of the Kirchhoff stress with
respect to the deformation gradient that takes part in the general formula (14.95) for the
spatial tangent modulus. Clearly, the elastic tangent here coincides with that of the isotropic
model (as the same elastic law is adopted in both cases). In the elastoplastic case, as a result
of (14.222), the derivative (14.98) is here given by
∂ τ̂ trial
∂εen+1 e trial
∂Bn+1 ∂βtrial
n+1
e trial
∂F n+1
= Dep : e trial
: + Dβ : e trial
: , (14.224)
∂F n+1 ∂Bn+1 ∂F n+1 ∂F n+1 ∂F n+1
where Dep is the standard infinitesimal elastoplastic consistent tangent operator associated
with the particular model/algorithm adopted and
∂τ̃
Dβ ≡ (14.225)
∂βtrial
n+1
is a fourth-order tensor which represents the tangent relation between the updated Kirchhoff
stress and the trial back-stress, also consistent with the infinitesimal return mapping. The
tangential relation Dβ is obtained in a manner analogous to the conventional infinitesimal
consistent tangent operator Dep , through appropriate differentiation of the corresponding
return-mapping equations. Note that the conventional tangent measures the rate of change
of τn+1 with changes in the input value of εen+1 trial
at a frozen value of βtrial
n+1 and the other
β
internal variables, whilst D measures the rate of change of τn+1 with changes in the input
value of βtrial e trial
n+1 at a frozen value of εn+1 and the other internal variables of the model.
FINITE STRAIN ELASTOPLASTICITY 643
Implementation aspects
Those interested in the computational implementation of finite strain kinematic hardening
models of the present type should note that the computation of aτ follows the same steps
as those already included in program HYPLAS for the isotropic plasticity models. The
computation of tensors U and V and subsequent assemblage of aβ need to be added to the
relevant part of the code. In program HYPLAS, this is probably best done in the material
interface subroutine MATICT, having (as suggested in Remark 14.12, page 639) kinematic
hardening models of this type as a new class of material models. In the computation of
U, a new material model-specific quantity – Dβ – is needed. In program HYPLAS, this
would probably be best implemented in a separate subroutine specifically associated with
the particular material model/algorithm type in question (just as with the other state-update
procedures and tangent moduli computation routines of HYPLAS). In the computation of V, as
given by expression (14.229), the derivative of the inverse tensor square root can be computed
in practice by subroutine DISO2 of HYPLAS. Finally, note that for the particular nonlinear
mixed hardening model alluded to in Remark 14.11, the tangent operator Dep is the one coded
in subroutine CTVMMX. This routine can be used in the extension of the mixed hardening model
already implemented in HYPLAS to the finite strain range.
644 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
M ≡ (F e )T τ (F e )−T , (14.232)
T HE present chapter is devoted to special finite element techniques for the analysis of
large deformations of nearly incompressible solids. It is a well-known fact that the
performance of low-order virtual work-based finite elements becomes extremely poor as
the incompressible limit is approached. Problems of practical engineering interest for which
incompressibility plays a crucial role include the analysis of rubbery solids, typically mod-
elled as incompressible hyperelastic materials (refer to Chapter 13), as well as elastoplastic
simulations under plastic dominant deformations and the assumption of isochoric plastic
flow, such as in metal plasticity models. In such situations, spurious volumetric locking, i.e.
overstiff solutions, are frequently obtained as a consequence of the inability of low-order
interpolation polynomials adequately to represent general volume-preserving displacement
fields. The problem can be remedied simply by adopting elements of sufficiently high
order. However, due to their simplicity, low-order elements are often preferred in large-scale
computations. For this reason, several formulations have been proposed to allow the use of
low-order elements near the incompressible limit.
Within the context of the geometrically linear theory, the class of enhanced assumed strain
(EAS) methods described by Simo and Rifai (1990), which incorporates popular procedures
such as the classical incompatible modes formulation proposed by Taylor et al. (1976) and
B-bar methods (Hughes, 1980), is well established and is employed with success in a number
of existing commercial finite element codes.
At this point, it is worth mentioning that such issues have not been raised by the authors
in the infinitesimal strain context of Part Two of this book, the reason for this being that,
under infinitesimal kinematics, the use of underintegrated standard finite elements (such
as the eight-noded quadrilateral with four Gauss point quadrature) can provide simple and
satisfactory solutions near the incompressible limit. However, in the geometrically nonlinear
regime, particularly in the simulation of more realistic industrially relevant problems, more
complex phenomena such as frictional contact with large sliding distances under extreme
strains are often present. In such cases, lower-order elements are usually preferred due to
their robustness and simplicity.
To tackle the problem, different approaches have been proposed. Among others, the class
of mixed variational methods developed by Simo et al. (1985), the mixed u/p (displacement-
pressure) formulation proposed by Sussman and Bathe (1987), the geometrically nonlinear
extension of the B-bar methodology devised by Hughes et al. (1975) and Moran et al. (1990),
the family of enhanced assumed strain elements of Simo and Armero (1992), the F-bar
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
648 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
technique proposed by the authors and co-workers (de Souza Neto et al., 1996), the co-
rotational incompatible modes approach of Crisfield and Moita (1996), the geometrically
nonlinear selective reduced integration scheme of Doll et al. (2000), the more recent versions
of the enhanced assumed strain method proposed by Reese and Wriggers (2000) and Wall
et al. (2000) are possible alternatives that allow the use of low-order elements near the
incompressibility limit.
In the context of explicit transient dynamic analysis, methods based on so-called hourglass
control techniques are almost invariably employed. Essentially, this class of methods relies on
the use of reduced numerical integration quadrature within the element. For linear elements
this implies single point integration which, however, can result in spurious zero energy
(hourglass) modes of deformation (Belytschko et al., 2000, 1984; Hughes, 1987). In order to
obtain reliable results, various control methods have been proposed to eliminate hourglassing
by providing the restraint that the element lacks under single point integration, but without
stiffening the element’s adequate response to other modes (Belytschko and Bindeman, 1991,
1993; Belytschko et al., 2000, 1984). Two principal ways of resisting hourglassing are with
viscous damping and by introduction of artificial stiffness, both of which are capable of
eliminating the spurious singular modes but have a negligible effect on the stable global
modes. However, it should be stressed that hourglass control does not fully remove the
kinematic modes and, in particular, coarse meshes and meshes loaded with large nodal forces,
resulting either from boundary conditions or from contact, are susceptible to hourglassing
despite the use of control techniques.
One aspect that should be pointed out here is that, in addition to handling near incom-
pressibility, robust formulations should also be able to cope with the extra requirements that
different problems may present. For instance, in applications such as the prediction of failure
in metal-forming processes, the ability to capture strain localisation phenomena becomes
crucial; in problems involving extremely large strains, frequently encountered in the analysis
of rubbery materials and metal-forming problems, it is not unusual that a solution can be
obtained only if adaptive mesh refinement is employed. Thus, since a single formulation
is normally not sufficiently robust to produce an optimal performance under a very wide
range of conditions, the design of low-order finite elements for large-strain analysis of quasi-
incompressible materials remains an open issue.
Rather than providing a general overview of a large number of different approaches, we
have opted in this chapter to provide the reader with a reasonably detailed description (to
a level of detail sufficient to allow the reader to produce the necessary computer codes) of
only three methods. These methods, which we consider to be of sufficient generality, are,
namely, the F-bar methodology (including its more recent F-bar-Patch variant for simplex
elements), the geometrically nonlinear EAS method and the mixed u/p formulation. These
are described, respectively, in Sections 15.1, 15.2 and 15.3. We remark that the F-bar method
is incorporated into the standard version of program HYPLAS that accompanies this book.
such that the incompressibility constraint can be enforced in an approximate average (not
pointwise) sense throughout the element. The method is in fact closely related to the classical
B-bar procedures (Hughes, 1980). However, as we shall see, it cannot be regarded as a
direct geometrically nonlinear extension of the B-bar method. Geometrically nonlinear B-
bar elements (Hughes et al., 1975; Moran et al., 1990) possess a more complex structure,
particularly with respect to the corresponding consistent tangent stiffness matrices. Some
basic features of the F-bar approach are summarised below:
2. The computer implementation of F-bar elements is also simple. In fact, F-bar element
routines can be obtained by introducing straightforward modifications into existing
standard element routines. Interested readers are referred to the source code of program
HYPLAS where both standard and F-bar four-noded quadrilaterals are implemented.
3. As for B-bar methods and assumed enhanced strain approaches, F-bar elements can be
used with any material model, elastic or inelastic, regardless of deviatoric/volumetric
coupling of the constitutive equations. This is not true if selective reduced integration
is used.
4. Numerical experience shows that F-bar elements produce good solutions with reason-
ably sized meshes for a wide range of industrially relevant problems. The methodology
avoids volumetric locking and, in addition, is particularly suitable to capture strain
localisation phenomena.
F = Fiso ( F0 )v
F0
F
F
ξi
F0 ξi
ξ0= 0
ξ0= 0
Figure 15.1. The F-bar four-node quadrilateral and eight-node hexahedron. (Reproduced with permis-
sion from Design of simple low-order finite elements for large-strain analysis of nearly incompressible
solids, EA de Souza Neto, D Perić, M Dutko and DRJ Owen, International Journal of Solids and
Structures, Vol 33, Issue 20–22
c 1996 Elsevier Science Ltd.)
tensor is obtained from the deformation gradient at the end of the time interval by means
of incremental constitutive functions of the general form‡
σn+1 = σ̂(αn , F ). (15.1)
where αn denotes the set of internal variables of the model at tn . For conventional
(or standard) finite elements, the deformation gradient F is computed directly from the
standard interpolation (bi-linear for the quadrilateral and tri-linear for the hexahedron) of
the displacement field at the generic integration point i.
of the stress tensor usually involves only straightforward function evaluations. For dissipative materials, σ̂ is in
general associated with numerical algorithms for integration of the constitutive equations of the underlying material
model. Such algorithms have been thoroughly discussed in the preceding chapters of this book.
FINITE ELEMENTS FOR LARGE - STRAIN INCOMPRESSIBILITY 651
F = F iso F v ,
(15.4)
F 0 = (F 0 )iso (F 0 )v .
The F-bar deformation gradient is then defined as the product of the isochoric component of
F with the volumetric component of F 0 , i.e. we compute
1/3
det F 0
F̄ = F iso (F 0 )v = F. (15.5)
det F
Having defined the modified deformation gradient, the F-bar four- and eight-noded elements
are obtained simply by replacing F with F̄ in (15.1); that is, for the present elements, the
Cauchy stress at each Gauss point is computed as
The corresponding F-bar element internal force vector is then given simply by
int
f(e) = BT σ̂(αn , F̄ ) dv, (15.9)
ϕ(Ω(e) )
652 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
where B denotes the standard discrete spatial symmetric gradient operator (the same as
that used to assemble the internal force vector of the corresponding standard element – see
expression (4.92), page 103).
Remark 15.2. As far as the computation of the internal force vector is concerned, the only
difference between F-bar elements and their conventional displacement-based counterparts
is the replacement of F with F̄ in the computation of the Cauchy stress by the incremental
constitutive function σ̂ (the stress-updating procedure). Thus, in addition to the evaluation of
F and σ at each Gauss point – procedures carried out normally for the standard elements
– the F-bar elements require only the evaluation of det F 0 , at their centroids, and the
computation of F̄ which, from its definition (15.5), is obtained simply by multiplying F
by a scalar factor. The added computational effort is, therefore, minimal (refer to Box 15.1).
We emphasise that, as for the conventional elements, four and eight-point Gauss quadratures
are adopted for the F-bar four-node quadrilateral and eight-node hexahedron respectively.
Remark 15.3. Geometrically nonlinear extensions of the so-called B-bar methods, such as
the class of elements proposed by Hughes et al. (1975), Moran et al. (1990) and particular
cases of the general methodology described by Simo et al. (1985), are also based on the
replacement of the compatible deformation gradient with an assumed modified counterpart
defined similarly to (15.5). However, in contrast to such procedures, modified gradient
operators (or B-bar matrices) do not take part in the present formulation. Note that the
standard B-matrix appears in (15.9). This is due to the fact that, in the F-bar procedure, the
assumed deformation gradient has been introduced in the stress constitutive function rather
than in the corresponding strain energy function. As we shall see in what follows, this crucial
difference allows a relatively straightforward adaptation of existing finite strain displacement-
based element routines to incorporate F-bar elements.
where G is the standard discrete spatial gradient operator, G0 is the gradient operator at the
element centroid, a|F=F̄ denotes the matrix form of the consistent spatial tangent modulus
evaluated at F = F̄ , and q is the matrix form of the fourth-order tensor defined by
q= 1
3 a : (I ⊗ I) − 23 (σ ⊗ I), (15.11)
also computed at F = F̄ .
FINITE ELEMENTS FOR LARGE - STRAIN INCOMPRESSIBILITY 653
Proof. This follows from the linearisation of the element internal force vector (15.9). The
main steps are the following. Firstly, we map (15.9) to the material configuration. Then we
linearise the material form and map the linearised expression back to the spatial configuration.
Let us start by observing the following relation for the first Piola–Kirchhoff stress:
− 23
−T −T det F 0
P ≡ (det F ) σ F = (det F ) σ̂(F̄ ) F = P̂ (F̄ ) (15.12)
det F
where we have defined
−T
P̂ (F̄ ) ≡ (det F̄ ) σ̂(F̄ ) F̄ . (15.13)
For notational convenience we have omitted αn from the arguments of σ̂ (and P̂ ). With the
above definition, we write the following material version of the F-bar internal virtual work
functional (15.8):
− 23
det F 0
Gint
(e) (u, η) = P̂ (F̄ ) : ∇p η dv, (15.14)
Ω(e) det F
where
F ≡ I + ∇p (u + d)
(F 0 ) ≡ I + ∇p (u + d)|ξ=0
(15.16)
13
det(F 0 )
F̄ ≡ F .
det F
After a lengthy but straightforward application of the chain and product rules to (15.15)
together with the use of basic tensor algebra relations, we obtain
int
DG(e) (u, η) [d] = {a(F̄ ):∇x d + [ 13 a(F̄ ):(I ⊗ I )
ϕ(Ω(e) )
− 2
3 σ̂(F̄ ) ⊗ I ] : (∇0 d − ∇x d )} : ∇x η dv, (15.17)
where ∇0 d denotes the spatial gradient of d evaluated at the centroid of the element. Finally,
the replacement of the gradient operators, tangent modulus and other relevant tensors with
the corresponding finite element matrices in (15.17) leads to the tangent stiffness formula
(15.10).
Remark 15.4. (The structure of the F-bar tangent stiffness) The first term on the right-
hand side of (15.10) is identical to the tangent stiffness of conventional displacement
elements evaluated at F =F̄ . The computation of the extra term indicated in (15.10) is
relatively simple. In addition to the spatial tangent modulus, stress and the discrete gradient
654 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
at the Gauss point of interest – already used in the calculation of the stiffness of standard
elements – only the discrete gradient at the element centroid, G0 , and the matrix q are
required. The evaluation of G0 is straightforward and, from its definition, the computation
of q also requires little computational effort. Thus, the c.p.u. time needed to compute the
tangent stiffnesses of the proposed elements is just slightly longer than that consumed by
their conventional counterparts. However, note that the additional stiffness term appearing
in (15.10) is generally unsymmetric and, therefore, requires an unsymmetric solver in the
finite element computations. Note that this fact is immaterial whenever the problem to
be solved is characterised by an unsymmetric tangent modulus in the continuum setting,
i.e. before finite element discretisation. This condition is encountered in a wide variety of
practical engineering applications, such as, for instance, the simulation of metal-forming
operations in which frictional contact almost invariably plays an essential role and the use
of an unsymmetric solver is inevitable. More sophisticated constitutive laws, such as the
coupled elastoplastic damage models discussed in Chapter 12, single crystal plasticity laws
(Chapter 16), and non-associative plasticity models in general, also lead to unsymmetric
tangent moduli.
where aij are the components of the spatial tangent modulus matrix a. The matrix form of
the remaining term σ ⊗ I appearing in (15.11) is simply given by
σ11 0 0 σ11 σ11
σ12 0 0 σ12 σ12
[σ ⊗ I ] = σ12 0 0 σ12 σ12 . (15.19)
σ22 0 0 σ22 σ22
σ33 0 0 σ33 σ33
In the expressions above, the matrix index 5 (or tensorial index 33) represents the circumfer-
ential directions, other indices correspond to the in-plane components.
FINITE ELEMENTS FOR LARGE - STRAIN INCOMPRESSIBILITY 655
We now turn our attention to the plane strain implementation of F-bar elements. To
ensure that the modified deformation gradient corresponds to a plane strain deformation,
expression (15.5) is replaced in plane strain F-bar elements with
0
F̄p
0
F̄ =
,
(15.20)
0 0 1
Remark 15.5. Note from (15.20) that the redefined modified deformation gradient represents
indeed a plane strain state. This would not be the case if the general definition (15.5) were
applied under plane strain conditions. Note that, under the present definition, (15.7)2 remains
valid and pressure distributions within elements will be constant for materials with decoupled
volumetric/deviatoric response. Expression (15.7)1, however, is no longer valid, i.e. the
deviatoric component of the modified deformation gradient does not coincide in general with
that of the deviatoric deformation gradient that results from the standard bilinear interpolation
at the Gauss points.
The algorithmic implementation of the plane strain version of the proposed four-node
quadrilateral element follows that described in Boxes 15.1 and 15.2, except that the exponent
1/3 must be replaced by 1/2 in the computation of the modified deformation gradient. Also,
as a consequence of definition (15.20), expression (15.11) for the tensor q, which arises from
the linearisation of the internal force vector, must be replaced by
q= 1
2 a : (I ⊗ I) − 12 (σ ⊗ I), (15.22)
in the computation of the element tangent stiffness matrix (Box 15.2). The explicit form
of the matrices [a : (I ⊗ I)] and [σ ⊗ I], in this case, is obtained from expressions (15.18)
and (15.19) simply by deleting all components related to the circumferential direction, that
is, we have
a + a
11 0
14 0 a +a 11 14
a21 + a24 0 0 a21 + a24
[a : (I ⊗ I)] = (15.23)
a31 + a34
0 0 a31 + a34
a41 + a44 0 0 a41 + a44
656 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
and
σ11 0 0 σ11
σ σ12
12 0 0
[σ ⊗ I ] =
.
(15.24)
σ12 0 0 σ12
σ22 0 0 σ22
Cook’s membrane.
This example is frequently used in the literature to assess the convergence properties of finite
elements under a mixture of shear and bending strains (Glaser and Armero, 1997; Korelc
and Wriggers, 1996; Simo and Armero, 1992; Simo and Rifai, 1990). The purpose here is
only to illustrate the convergence properties of the four-noded F-bar element under near-
incompressibility. It is a well-known fact that, under near-incompressibility, conventional
low-order elements require excessively fine discretisations to produce solutions sufficiently
close to the converged one. ‘Adequate’ elements should converge with more sensible coarser
meshes. The problem consists of a tapered and swept panel of unit thickness, illustrated in
FINITE ELEMENTS FOR LARGE - STRAIN INCOMPRESSIBILITY 657
Figure 15.2(a), with one edge clamped and a uniformly distributed vertical load of linear
density f = 6.25 applied to the opposite edge. The corresponding resultant applied load in
the initial configuration is F = 100. Throughout the loading programme, the applied load
remains vertical (it does not follow the direction tangent to the faces of the loaded elements),
but the resultant load will change according to the variation of length of the loaded faces so
as to preserve the prescribed constant load density. Plane strain condition is assumed and a
regularised neo-Hookean material with shear modulus µ = 80.1938 and bulk modulus k =
40.0942 × 104 is adopted. Note that near incompressibility is achieved with the high ratio
k/µ, of order 104 . Several meshes are considered, so that the convergence of the solution with
mesh refinement can be assessed. A mesh of 4 × 4 elements is depicted in Figure 15.2(a). For
all meshes considered, the total load is applied in five increments. The results are shown in
the graph of Figure 15.2(b) where the final vertical displacement obtained at the upper right
corner of the panel (point A of Figure 15.2(a)) is plotted against the number of elements per
side. Results obtained with the Q1/E4 element of Simo and Armero (1992) are also shown
for comparison. It can be seen that the convergence of the F-bar quadrilateral in this case is
almost identical to the behaviour of the element Q1/E4 with five-point integration rule.§
§ The five-point integration rule for quadrilateral domains (whose corresponding weights and integration point
positions are found in subroutine GAUS2D of program HYPLAS) has been introduced (Simo et al., 1993) as an
improvement to the earlier version of the Q1/E4 enhanced assumed strain element.
658 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
End do
σy = 0.45.
The elastic behaviour is defined by the Hencky material model with Young’s modulus E =
206.9 and Poisson ratio ν = 0.29. This corresponds to the shear and bulk moduli
µ = 80.1938, K = 164.21.
Vertical displacement is applied to the top nodes of the mesh up to a total displacement u =
0.3. The simulation is carried out here using the four-node element presented in this section
as well as the Q1/E4 element and the standard displacement-based four-node quadrilateral.
FINITE ELEMENTS FOR LARGE - STRAIN INCOMPRESSIBILITY 659
48
7
A
6
Present element
Q1/E4 (4 point rule)
Q1/E4 (5 point rule)
5
44
4
0 10 20 30
number of elements per side
(a) (b)
Figure 15.2. Cook’s membrane: (a) geometry and boundary conditions; (b) convergence of the solution
with mesh refinement. (Reproduced with permission from Design of simple low-order finite elements
for large-strain analysis of nearly incompressible solids, EA de Souza Neto, D Perić, M Dutko and
DRJ Owen, International Journal of Solids and Structures, Vol 33, Issue 20–22 c 1996 Elsevier
Science Ltd.)
u
4
3
total edge reaction
2
30
Present element
2 Q1/E4
Standard element
1
10
0
0 0.1 0.2 0.3
u
displacement, u
(a) (b)
Figure 15.3. Stretching of a double notched elastoplastic specimen: (a) geometry and finite element
discretisation; (b) reaction–displacement diagram. (Reproduced with permission from Design of simple
low-order finite elements for large-strain analysis of nearly incompressible solids, EA de Souza Neto,
D Perić, M Dutko and DRJ Owen, International Journal of Solids and Structures, Vol 33, Issue 20–22
c 1996 Elsevier Science Ltd.)
The total edge reactions per unit thickness obtained for each computation is plotted versus the
prescribed displacement in Figure 15.3(b). It can be seen that both the present element and
Q1/E4 are able to predict the existence of a limit load – a phenomenon that is not captured by
the standard four-node quadrilateral. The force–displacement curve predicted by the present
formulation is very close to that obtained with the enhanced element Q1/E4.
660 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
with constants
The corresponding force per unit thickness, P , obtained during the tests is plotted in
Figure 15.4(b). It can be seen that both the F-bar element and, as verified by Simo and Armero
(1992), the Q1/E4 element with four-point quadrature are able to capture the global softening
that characterises strain localisation. This indicates a possible suitability of these elements for
localisation problems. Interestingly, if the five-point integration rule suggested by Simo et al.
(1993) is employed, the forces obtained for the Q1/E4 element increase dramatically after
following closely the results for the four-Gauss point rule up to approximately u = 3.0. This
phenomenon might be attributed to a bifurcation of the solution and was noticed by Schönauer
et al. (1995). The results obtained with the standard four-node quadrilateral are also plotted
and show that, for this element, no softening occurs. The deformed configurations, at
u = 10.0, of each of the elements discussed are depicted in Figure 15.5. The final geometry of
the proposed element lies between the deformed geometries of the Q1/E4 with four and five
integration points. We point out the substantial difference that results from the application
of distinct quadratures in the Q1/E4 element. Severe locking is exhibited by the standard
quadrilateral as a consequence of the pointwise enforcement of the plastic incompressibility
constraint.
u, P
1.2
force, P
0.8
5
5 0
0 4 8 12 16
displacement, u
(a) (b)
Figure 15.4. Single element test: (a) geometry and load; (b) force-displacement curve. (Reproduced
with permission from Design of simple low-order finite elements for large-strain analysis of nearly
incompressible solids, EA de Souza Neto, D Perić, M Dutko and DRJ Owen, International Journal of
Solids and Structures, Vol 33, Issue 20–22
c 1996 Elsevier Science Ltd.)
(a) (b)
(c) (d)
Figure 15.5. Single element test. Deformed configurations at u = 10.0: (a) present element; (b) Q1/E4
with four-point rule; (c) Q1/E4 with five-point rule; (d) standard four-node quadrilateral. (Reproduced
with permission from Design of simple low-order finite elements for large-strain analysis of nearly
incompressible solids, EA de Souza Neto, D Perić, M Dutko and DRJ Owen, International Journal of
Solids and Structures, Vol 33, Issue 20–22
c 1996 Elsevier Science Ltd.)
clear at present. In contrast, such an issue does not arise in methodologies such as the F-
bar procedure described above and geometrically nonlinear extensions of the classical B-
bar method (Moran et al., 1990). The absence of element internal parameters makes the
incorporation of such methods into adaptive remeshing environments rather straightforward.
Exactly as for standard displacement-based formulations, such methods require only nodal
displacements and physical variables (e.g. plastic, elastic strains, hardening parameters, etc.)
to be transferred between meshes. The suitability of the four-noded F-bar quadrilateral in
particular for adaptive analysis is illustrated here in the simulation of the upsetting of an
elastoplastic cylindrical billet. The billet, with radius r = 9 mm and height h = 30 mm is
compressed between two flat tools (assumed rigid) subjected to sticking contact condition
on the interface. The initial tool/workpiece configuration is schematically illustrated in
662 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
h=30mm
r=9mm
(b)
(a)
(d)
(c)
(e)
Figure 15.6. Adaptive analysis: (a) initial tool/workpiece configuration; (b) initial mesh; (c) deformed
mesh at 40% compression; (d) deformed mesh at 70% compression; (e) deformed mesh at 70%
compression without remeshing. (Reproduced with permission from Design of simple low-order finite
elements for large-strain analysis of nearly incompressible solids, EA de Souza Neto, D Perić, M Dutko
and DRJ Owen, International Journal of Solids and Structures, Vol 33, Issue 20–22 c 1996 Elsevier
Science Ltd.)
Figure 15.6(a). The Hencky elastic model and the von Mises elastoplastic law are used to
model the billet. Linear isotropic hardening is assumed:
σy (ε̄p ) = σy 0 + H ε̄p ,
with hardening modulus H = 0.7 GPa and initial yield stress σy 0 = 0.45 GPa. The Young’s
modulus and Poisson ratio are, respectively, E = 200 GPa and ν = 0.3, corresponding to
The initial mesh employed to discretise the symmetric quarter of the cylinder is plotted in
Figure 15.6(b). The criterion for mesh refinement/de-refinement is based on the incremental
plastic work (refer to Perić et al. (1994) for details). Two other meshes, generated during the
adaptive analysis, are shown in Figures 15.6(c) and 15.6(d). The configurations illustrated
correspond to 40% and 70% compression of the billet respectively. Due to the concentration
of the plastic process near the corner of the billet, substantial refinement is detected in that
region of the mesh of Figure 15.6(c). At the later stage of 70% compression, shown in
Figure 15.6(d), the rate of plastic dissipation and, therefore, the elements, are more uniformly
distributed. Note that, at any instant, all elements have a very good aspect ratio. The result of
a similar computation without adaptivity, i.e. maintaining the initial discretisation throughout
the entire process, is shown in Figure 15.6(e) where the excessive distortion of the mesh is
clear.
FINITE ELEMENTS FOR LARGE - STRAIN INCOMPRESSIBILITY 663
where J(ξ̄j ) ≡ det F (ξ̄j ) is the standard deformation gradient obtained as usual from the
interpolated displacement field of the element, evaluated at the volumetric sampling point
with local coordinate ξ̄j . The modified deformation gradient, F̄ , at each Gauss point with
coordinates ξi is obtained as the composition of the isochoric deformation gradient at ξi with
the approximate volumetric deformation gradient (15.26) at ξi ; that is, we compute
¯ 1/3
J(ξi )
F̄ (ξi ) = F (ξi )
J(ξi )
3nvol 1/3
j=1 J(ξ̄j ) N̄j(ξi )
= F (ξi ). (15.27)
J(ξi )
Clearly, (15.27) reduces to the centroid-only formula (15.5) if nvol = 1 with ξ̄1 being the
coordinate of the centroid. Following (15.21), in plane strain, J in the above is replaced
with the determinant of the in-plane deformation gradient Jp = det F p and the exponent
1/3 is replaced with 1/2. Having computed F̄ according to (15.27), the remainder of the
procedure is identical to that of the centroid-only sampling F-bar elements, except for the
tangent stiffness matrix whose explicit expression is given below.
664 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
where
1
nvol
Ḡ(ξ) ≡ ¯ J(ξ¯j ) N̄j (ξ) G(ξ̄j ). (15.29)
J(ξ) j=1
The element stiffness in this case has the same structure as its counterpart (15.10) for centroid
sampling-only F-bar elements. The only difference lies in the additional stiffness term where
the original gradient operator G0 (evaluated at the centroid) used in (15.10) is replaced here
with the operator Ḡ defined by (15.29).
F-bar elements will have fewer volumetric sampling points than Gauss integration points. The
arguments here are similar to those used in mixed u/p formulations described in Section 15.3
(refer to the comments made at the end of Section 15.3.2).
F̂ ≡ α F + (1 − α) F̄ , (15.32)
F = F 0. (15.33)
The use of an F-bar procedure based on the original definition (15.5) of the modified
deformation gradient would not make sense in this case.
However, the use of simplex finite elements is very desirable, particularly in the solution of
three-dimensional problems with complex geometry. The generation of hexahedral element
meshes for complex three-dimensional geometries remains largely an open issue. Many
difficulties exist and this poses serious limitations on the use of such elements in the solution
of many problems of interest.
An effective alternative proposed in de Souza Neto et al. (2005) is to redefine the
F-bar deformation gradient so as to produce the required relaxation of the pointwise
incompressibility constraint in the context of simplex elements. Then, rather than working
separately within individual elements, we will consider a patch of simplex elements. A typical
patch is illustrated in Figure 15.7 (drawn in solid lines). With P denoting the set of elements
forming a predefined patch, for each element e ∈ P, the modified deformation gradient, F̄e ,
666 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
typical mesh of
predefined patch simplex elements
of simplex elements
Figure 15.7. The F-bar-Patch Method. Patch of simplex elements. (Reproduced with permission
from F-bar-based linear triangles and tetrahedra for finite strain analysis of nearly incompressible
solids. Part I: Formulation and benchmarking, EA de Souza Neto, FM Andrade Pires and DRJ Owen,
International Journal for Numerical Methods in Engineering, Vol 62
c 2005 John Wiley & Sons, Ltd.)
is defined as: 13
vpatch
F̄e = Fe , (15.34)
Vpatch (detFe )
where F e is the deformation gradient obtained from the standard linear displacement
interpolation within element e and vpatch and Vpatch are, respectively, the deformed and
reference (undeformed) volume of the patch P :
vpatch = vi , Vpatch = Vi . (15.35)
i∈P i∈P
In the above, vi and Vi denote, respectively, the deformed and reference (undeformed) volume
of the generic element i of patch P.
The above definition implies that the determinant of the modified deformation gradient is
identical for all elements e of patch P and is given by:
vpatch
J¯e ≡ detF̄e = , (15.36)
Vpatch
that is, detF̄e is the ratio between the deformed and undeformed volume of the predefined
patch P of elements. Thus, the use of (15.34) under the incompressibility constraint requires
that the patch P of elements preserves its volume, even though individual elements of the
patch may suffer volume changes during deformation.
10
Total load [K N]
6
4
Present element
F-bar quadrilateral
2 Standard element
0
0 1 2 3 4 5 6
non-overlapping patches of elements. The more elements in each patch, the greater the
constraint relaxation achieved. Then, in defining such patches, it is crucial to have in
mind that excessive constraint relaxation (too many elements in the patch) may lead to
spurious zero-energy mechanisms. On the other hand, insufficient constraint relaxation (too
few elements in the patch) may result in incompressibility locking. Numerical experience
shows that patches of two elements in two-dimensional/axisymmetric problems and eight
elements in three-dimensional analyses produce sufficient constraint relaxation to avoid
incompressibility locking without showing spurious mechanisms. Figure 15.8(a) shows the
detail of the deformed mesh obtained with patches of two triangles in the solution of the plane
strain localisation problem whose four-noded F-bar solution is plotted in Figure 14.12(b)
(page 614). The detail of the localisation zone for the F-bar quadrilateral is shown for
comparison alongside the F-bar-Patch result, in Figure 15.8(b). The deformation pattern
obtained with the F-bar-Patch element is virtually identical to that obtained with the four-
noded F-bar quadrilateral. The force–displacement curves for both elements is shown in
Figure 15.8(c). The curves for both elements virtually coincide. An F-bar-Patch solution
of the three-dimensional necking problem of Section 14.9.2 (from page 607) is shown in
Figure 15.9(a), where patches of eight linear tetrahedra have been used. The deformed mesh
is in close agreement with that produced with the F-bar hexahedron (shown in Figure 14.9).
The force-displacement diagrams of Figure 15.9(b) show that the results obtained with the F-
bar-Patch method using patches of eight tetrahedra virtually coincide with analogous results
obtained in an axisymmetric analysis using patches of two linear triangles. These results are
in very close agreement with those obtained with the F-bar hexahedral and axisymmetric
quadrilateral elements (shown in Figure 14.11).
668 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
80
60
Total load
Please Wait..
40
Present element 3D
20 Present element-Axisymmetric
3D Standard element
0
0 1 2 3 4 5 6 7 8
Top displacement
(a) (b)
Figure 15.9. Necking of a cylindrical bar. F-bar-Patch solution with patches of eight linear tetrahedra:
(a) deformed configuration at u = 7.0 mm; (b) force–displacement diagrams (forces in KN, displace-
ments in mm). (Reproduced with permission from F-bar-based linear triangles and tetrahedra for finite
strain analysis of nearly incompressible solids. Part I: Formulation and benchmarking, EA de Souza
Neto, FM Andrade Pires and DRJ Owen, International Journal for Numerical Methods in Engineering,
Vol 62 c 2005 John Wiley & Sons, Ltd.)
Both Kee and Kes are generally unsymmetric, regardless of the material model adopted.
FINITE ELEMENTS FOR LARGE - STRAIN INCOMPRESSIBILITY 669
Remark 15.6 (The structure of the tangent stiffness). The stiffness contribution given
in (15.37) gives the tangent relation between the internal force vector components and nodal
displacements of element e. Its format is similar to that of the stiffness matrix (15.10) for
the F-bar elements. Its rows and columns are associated with the degrees of freedom of
element e only. The additional contributions (15.38), on the other hand, are the tangent
relations between the internal force components of element e and nodal displacements of
the other elements s of the patch. Their rows are associated with the degrees of freedom
of element e and their columns are associated with the degrees of freedom of the other
elements s of the patch. The global tangent stiffness required in the solution of the equilibrium
problem is obtained by adding, for each element e of the mesh, the contributions of Kee and
Kes (s ∈ P; s
= e) to the appropriate global stiffness matrix position.
where ψ is the specific free-energy function characterising the hyperelastic material, Πext (u)
is the potential energy functional of the external loading at the configuration defined by the
displacement field u and
C̃ = C̃(H ) = [I + H ]T [I + H ]
H = ∇p u + H̃, (15.40)
where H̃ is the enhanced displacement gradient, and then re-parametrise the Hu–Washizu
functional so as to have displacement, first Piola–Kirchhoff stress and enhanced displacement
gradient, H̃, as the independent variables; that is, the three-field enhanced functional is
670 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
defined as
Π̃(u, H̃ , P ) ≡ Π(u, ∇p u + H̃ , P )
= [ρ̄ψ(C̃) + P : H̃ ] dv + Πext (u). (15.41)
Ω
F = I + ∇p u.
In (15.41) we have
T
C̃ = C̃(H̃ ) = F̃ F̃ = [I + ∇p u + H̃ ]T [I + ∇p u + H̃ ]. (15.43)
Stationary condition
The enhanced three-field functional Π̃ is stationary if and only if the following integral
equation holds
∂ψ ∂ψ
2ρ̄ F̃ : ∇p η − H̃ : δP − P − 2ρ̄ F̃ : δ H̃ dv − Gext (η) = 0 (15.44)
Ω ∂ C̃ ∂ C̃
for all variations η, δ H̃ and δP of displacement, enhanced displacement gradient and first
Piola–Kirchhoff stress, respectively. In the above, Gext is the functional of virtual work of
the external forces:
Gext (η) ≡ b̄ · η dv + t̄ · η da, (15.45)
Ω ∂Ω
where b̄ and t̄ are, respectively, the material description of the body force and boundary
surface traction fields.
Strong form
By means of arguments analogous to those leading to the equivalence between the weak and
strong form of the equilibrium in Section 3.6.1, it can be shown that the above integral form
leads to the strong (pointwise) equations
∂ψ
divp 2ρ̄ F̃ + b̄ = 0
∂ C̃
H̃ = 0 in Ω
(15.46)
∂ψ
P = 2ρ̄ F̃
∂ C̃
P m = t̄ in ∂Ω,
FINITE ELEMENTS FOR LARGE - STRAIN INCOMPRESSIBILITY 671
where m denotes the outward unit normal to the reference boundary ∂Ω. Note that the
above strong form has recovered in (15.46)1,3,4 the momentum balance equation (3.133) (see
page 68) together with the standard hyperelasticity law (13.10) (page 521). Equation (15.46)2
enforces the compatibility of the displacement gradient field, i.e. H = ∇p u, and implies
F̃ = F .
is the determinant of the compatible deformation gradient. Note that in view of the definition
(15.42) of the enhanced deformation gradient, the modified spatial gradient, ∇ ˜x η, does not
coincide in general with the true spatial gradient ∇x η. However, they do coincide, i.e.
˜x η = ∇x η
∇
if the strong (pointwise) compatibility equation (15.46)2 holds. As we shall see below, the
modified spatial gradient gives rise to modified discrete gradient operators in the finite
element formulation.
where hHij (ξ) are the corresponding interpolated components. The array h H̃ is assumed to
be constructed as
h
H̃(e) (ξ) = H(ξ) β(e) , (15.51)
where each H is the interpolation matrix for the enhanced displacement gradient and β(e) is
the vector of element internal degrees of freedom. In two dimensions, the matrix H has the
format 11
H1 (ξ) H211 (ξ) · · · Hn11enh (ξ)
H 12 (ξ) H 12 (ξ) · · · H 12 (ξ)
1 2 nenh
H(ξ) = 21
,
(15.52)
H1 (ξ) H221 (ξ) · · · Hn21enh (ξ)
H122 (ξ) H222 (ξ) · · · Hn22enh (ξ)
where Hijk (ξ) denotes the component jk of the ith prescribed enhanced mode:
11
Hi (ξ) Hi12 (ξ)
Hi (ξ) = , (15.53)
Hi21 (ξ) Hi22 (ξ)
and nenh is the total number of enhanced modes or element internal degrees of freedom
adopted. The vector β(e) is generally given by
The interpolated tensor field hH̃(e) (ξ) is obviously a linear combination of the nenh modes:
nenh
h
H̃(e) (ξ) = βi Hi (ξ). (15.55)
i=1
Variations of the enhanced displacement gradient are interpolated in the same way, i.e.
nenh
where δβ(e) (with components δβi ) is the array of variations of the element internal degrees
of freedom.
for all variations δP . The above condition will be identically satisfied in the discretised
version of (15.44) if the space of interpolated first Piola–Kirchhoff stress fields is L2 -
orthogonal to the space of interpolated enhanced displacement gradients; that is, let
h
P (e) (ξ)
be the generic interpolation function for the nominal stress tensor within an element e. At
the outset, we assume that, as for the interpolated enhanced displacement gradients, the
global interpolated first Piola–Kirchhoff stress field may be discontinuous across element
boundaries. In the discrete setting, the L2 -orthogonality is expressed as
h
H̃ (e) : hP (e) dv = 0
Ω(e)
and, in view of the assumed enhanced displacement gradient interpolation (15.51), requires
that
h
P (e) (ξ) : Hi (ξ) dv = 0 (15.58)
Ω(e)
(e)
for all functions hP (e) and Hi . With the interpolation functions hP (e) satisfying the above
condition, equation (15.57) is automatically satisfied. In addition, by observing that (15.58)
also implies that the product P : δ H̃ in (15.44) vanishes in the discretised version of the
variational principle, we find that the nominal stress field P is completely eliminated from
the original three-field formulation.
¶ Readers not familiar with this nomenclature are referred to standard texts on functional analysis (Oden, 1979)
where f̃ int is a modified internal force vector obtained by assembling the element vectors
1 T
f̃(e) (u(e) , β(e) ) =
int
B̃ τ(F̃ ) dv, (15.61)
(e)
ϕ(Ω ) J
where f ext is the standard external load vector – the same as in the virtual work-based
formulation – and the nenh -dimensional element vectors s(e) are defined by
1 T
s(e) (u(e) , β(e) ) = Q̃ τ(F̃ ) dv. (15.62)
(e)
ϕ(Ω ) J
In the above, the matrix B̃ is a modified discrete symmetric gradient operator and Q̃ is the
discrete operator associated with variations of interpolated enhanced displacement gradients.
The structure of B̃ and Q̃ will be discussed below. The function τ gives the finite element
array of Kirchhoff stress tensor components (arranged in the same order as in the arrays
defined in (4.20–4.22)) obtained from the standard hyperelastic law
∂ψ T
τ(F̃ ) = 2ρ̄ F̃ F̃ . (15.63)
∂ C̃
where hjk
i is the component jk of the tensor
(e) (e) −1
hi (ξ) ≡ Hi (ξ) F̃ . (15.69)
where
(k−1) (k−1) (k−1)
r(e) ≡ f̃(e)
int
(u(e) , β(e) ) − f(e)
ext
(15.71)
and (u(k−1) , β(k−1) ), Kuu and Kuβ are, respectively, the derivatives of the global residual
(e) (e)
r with respect to the global vectors u and β. The element tangent stiffnesses Kβu and Kββ
are, respectively, the derivatives of the left-hand side of (15.60)2 with respect to the element
vectors δu(e) and δβ(e) . All stiffnesses are evaluated at (u(k−1) , β(k−1) ). Their expressions
(k)
are given below. Solution of (15.70)2 for δβ(e) gives
in the remainder of this section to denote the iterative increment of (·) within the Newton–Raphson scheme and not
the variation of (·) as we did previously in this chapter.
FINITE ELEMENTS FOR LARGE - STRAIN INCOMPRESSIBILITY 677
(e) (e) (e) (e)
version of (15.60) where the element matrices Kuu , Kuβ , Kβu and Kββ are found to be
given by
(e)
Kuu = G̃T ã G̃ dv
ϕ(Ω(e) )
(e)
Kuβ = G̃T ã R̃ dv
ϕ(Ω(e) )
(15.75)
(e) T
Kβu = R̃ ã G̃ dv
ϕ(Ω(e) )
(e)
Kββ = R̃T ã R̃ dv,
ϕ(Ω(e) )
where G̃ is the modified full discrete gradient operator which, in the two-dimensional case,
is defined by
(e) (e) (e)
N1;1 0 N2;1 0 · · · Nnnode ;1 0
0 Nnnode ;1
(e) (e) (e)
N1;1 0 N2;1 · · · 0
G̃ = , (15.76)
N (e) 0
(e) (e)
1;2 0 N2;2 0 · · · Nnnode ;2
(e) (e) (e)
0 N1;2 0 N2;2 · · · 0 Nnnode ;2
and R̃ is defined as
h11
1 h11
2 ··· h11
nenh
12
h1 h12 ··· h12
2 nenh
R̃ = . (15.77)
h21 h21 ··· h21
1 2 nenh
h22
1 h22
2 ··· h22
nenh
and r̃(k−1) is the global modified residual vector obtained from the element vectors
Remark 15.7. Recall that for hyperelastic models (or dissipative models whose stress is
derived from an incremental potential), a is symmetric. In this case, the modified tangent
stiffness K̃T is also symmetric. Such symmetry is an obvious consequence of the variational
structure of the EAS method and is in contrast with the F-bar approach described earlier in
this chapter. For materials with unsymmetric spatial tangent modulus (such as non-associative
plasticity models in general) the EAS approach does not possess a variational structure (refer
to the text surrounding equation (15.64)) and the tangent stiffness K̃T is unsymmetric.
As can be observed from Box 15.3 the computational implementation of an EAS element
involves far more operations than those of the conventional virtual work-based element of the
same order. The extra computational effort concentrates around the evaluation of the matrices
Kuβ and Kββ (and inversion of Kββ ), computation of the iterative internal parameters and
the element vectors s(e) . The additional matrices increase in size with the number of internal
degrees of freedom and their computation cost may become large particularly for three-
dimensional elements (see Simo et al. 1993). Also, note that the extra matrices are initially
computed during the assembly of the modified tangent stiffness K̃T in the solution of the
condensed linearised system. The same matrices are reused later in the computation of the
internal forces and residuals. To avoid the costly recomputation of those matrices at this stage,
they can be stored in memory immediately after their initial computation and retrieved later.
This approach, however, causes a considerable increase in memory storage and may become
critical for three-dimensional elements. To circumvent this problem, Simo et al. (1993) have
proposed an alternative static condensation procedure whereby the relevant matrices can be
recomputed at a smaller additional cost.
The stability of geometrically nonlinear EAS elements has been for some time the subject of
research by many authors. In spite of the substantial reduction of incompressibility locking,
the ability to capture strain localisation phenomena and the high coarse mesh accuracy
achieved by many EAS formulations, analytical and numerical studies (Crisfield et al.,
1995; de Souza Neto et al., 1996; Glaser and Armero, 1997; Korelc and Wriggers, 1996;
Wriggers and Reese, 1996) have shown that severe instabilities in the form of spurious
hourglass deformation modes may occur in some EAS elements under large strains. This
is particularly true for early formulations such as the original Q1/E4 (plane strain), the Q1/E5
(axisymmetric) and Q1/E9 (three-dimensional) elements proposed by Simo and Armero
(1992). The stability of such elements is briefly discussed below.∗∗
∗∗ A mathematical analysis of the stability and convergence properties of the EAS method, restricted to the
infinitesimal strain theory, is provided by Reddy and Simo (1995). For the large strain formulation, we refer to a
recent contribution by Auricchio et al. (2005).
FINITE ELEMENTS FOR LARGE - STRAIN INCOMPRESSIBILITY 679
(vi) Update deformation gradient (for each Gauss point of each element)
(k)
nenh
(k)
F̃ := I + ∇p u(k) + βi Hi
i=1
(vii) Compute Kirchhoff stresses (for each Gauss point of each element)
(k)
τ(k) := τ̂(F̃ )
ngausp
(B̃(k) )T wi ji τi
int (k) (k)
(f̃(e) ) :=
i=1
ngausp
wi ji (Q̃i )T τi
(k) (k) (k)
s(e) :=
i=1
(x) Check convergence. If r̃(k) ≤ tol then accept (·)(k) as the solution to (15.70) and
EXIT. Else, set k := k + 1 and GOTO (i)
680 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
and defining the following interpolation functions for the enhanced displacement gradient
field:
j0 −T
Hi (ξ) ≡ J Ei (ξ), (15.82)
j(ξ) 0
where J0 denotes the Jacobian of the isoparametric mapping of the element in the reference
configuration evaluated at the centroid (ξ = 0),
j0 = det J0 , (15.83)
and
j(ξ) = det[J(ξ)], (15.84)
where J(ξ) is the Jacobian at the generic point with local coordinates ξ.
The direct extension of the above element to three-dimensions yields the so-called Q1/E9
eight-noded brick which contains a total number of nine enhanced modes. The Q1/E9 element
is obtained as above by setting
ξ 0 0 0 0 0 0 0 0
E1 = 0 0 0, E 2 = ξ 0 0, E3 = 0 0 0,
0 0 0 0 0 0 ξ 0 0
0 η 0 0 0 0 0 0 0
E4 = 0 0 0, E5 = 0 η 0 , E6 = 0 0 0, (15.85)
0 0 0 0 0 0 0 η 0
0 0 ζ 0 0 0 0 0 0
E7 = 0 0 0, E8 = 0 0 ζ , E9 = 0 0 0.
0 0 0 0 0 0 0 0 ζ
Please Wait..
(a)
Pl
(b)
Figure 15.10. Spurious hourglassing in some EAS elements: (a) Q1/E9 three-dimensional element;
(b) Q1/E5 axisymmetric element. ((a) Reproduced with permission from Remarks on the stability of
enhanced assumed strain elements in finite elasticity and elastoplasticity, EA de Souza Neto, D Perić,
GC Huang and DRJ Owen, Communications in Numerical Methods in Engineering, Vol 11 c 1995
John Wiley & Sons, Ltd. (b) Reproduced with permission from Design of simple low-order finite
elements for large-strain analysis of nearly incompressible solids, EA de Souza Neto, D Perić, M Dutko
and DRJ Owen, International Journal of Solids and Structures, Vol 33, Issue 20–22 c 1996 Elsevier
Science Ltd.)
strained configurations can be a source of major problems. This problem seems to have
been first systematically investigated in the analytical study of Wriggers and Reese (1996)
who found that severe hourglass instabilities can occur under moderate-to-large compressive
strains for hyperelastic material models. Numerical studies carried out by the authors and
co-workers de Souza Neto et al. (1996) found in addition that, for elastoplastic models,
hourglass instabilities can be activated not only in compressive regimes but also (although
less severe) under finite stretching. The unstable behaviour is illustrated in Figure 15.10.
Figure 15.10(a) shows the results of the analysis of the compression of a cylindrical block
made of compressible rubber (modelled as a regularised Ogden material). The initially
cylindrical block is clamped at both ends and the analysis considers only the symmetric octant
of the actual geometry. The Q1/E9 eight-noded brick (with 3 × 3 × 3 Gauss quadrature) was
used and the configuration shown corresponds to 70% compression. Figure 15.10(b) depicts
a deformed configuration obtained with the axisymmetric EAS element Q1/E5 in the analysis
of the elastomeric bead compression problem described in Section 13.6.7 (page 556). The
configuration shown was obtained for a prescribed top plate displacement u = 0.235. In both
examples, severe spurious hourglassing invalidates the numerical results.
682 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
in conjunction with the following alternative definition for the enhanced displacement
gradient interpolation functions:
j0 −T
Hi (ξ) ≡ J Ei (ξ) J−1
0 . (15.87)
j(ξ) 0
Glaser and Armero (1997) proposed the elements named Q1/ET4 and Q1/ES4 obtained by
replacing the primary modes Ei of (15.81) respectively with their transpose (as for the GC4
element above) and their symmetrised counterparts
ξ 0 0 ξ 0 η 0 0
E1 = , E2 = , E3 = , E4 = . (15.88)
0 0 ξ 0 η 0 0 η
and
j0
Hi (ξ) ≡ F 0 J0 Ei (ξ) J−1
0 , (15.90)
j(ξ)
where F 0 denotes the (compatible) deformation gradient at the element centroid. The inclu-
sion of F 0 (absent in the formulation of the CG4 element and the original Q1/E4 and Q1/E9
elements) guarantees the frame invariance of the interpolated enhanced displacement gradient
in (15.89) and (15.90); that is, under an arbitrary superimposed rigid-body deformation with
rotation Q, we have the transformations
h
H̃ −→ Q hH̃
(15.91)
F̃ ≡ F + H̃ −→ Q F + Q H̃ = Q F̃ .
h h
The interpolation form (15.90) was introduced by Simo et al. (1993) to ensure objectivity of
improved versions of the original eight-noded EAS hexahedron.
The comparative study carried out by Glaser and Armero (1997) shows that the elements
Q1/ET4, Q1/ES4 and CG4 overcome the hourglass instability under compressive strains.
However, the problem persists in highly stretched configurations. These authors avoid this
instability by introducing an artificial hourglass control term in the formulation of the
elements Q1/ET4 and Q1/ES4.
FINITE ELEMENTS FOR LARGE - STRAIN INCOMPRESSIBILITY 683
where u and p are, respectively, the displacement and the independent Cauchy hydrostatic
pressure fields and p̄ is the (dependent) Cauchy pressure obtained from the pointwise
deformation gradient through the hyperelastic constitutive law
p̄ = p̄(F ) ≡ 1
3 σ̄(F ): I, (15.93)
where σ̄ is the hyperelastic constitutive function for the Cauchy stress tensor
ρ̄ ∂ψ T
σ̄(F ) ≡ F . (15.94)
det F ∂F
Note that ψ depends on u owing to the dependence of the pointwise value of the deformation
gradient, F , on the field u. The dependence of π on u follows from the dependence of p̄ on
F . Πext is the potential energy functional of the external load. The potential π is assumed to
have the general form
(p̄ − p)2
π(u, p) ≡ , (15.95)
2 P (p̄)
where P is the functional defined by
1 ∂ p̄
P (p̄) ≡ F T
: I. (15.96)
3 det F ∂F
The functional P is assumed to satisfy the condition
P (P (p̄)) = 0, (15.97)
684 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
which is equivalent to
∂P (p̄) T
F : I = 0. (15.98)
∂F
Expression (15.98) represents a constraint on the possible constitutive models that can be used
in conjunction with the u/P formulation. In other words, only constitutive models whose
constitutive function p̄ for the hydrostatic pressure satisfy (15.98) are amenable to treatment
via the present u/p formulation. The need for the above constraint is discussed below.
Stationary condition
Mixed u/p finite element formulations are derived simply by applying standard finite element
discretisations to the integral equations expressing the stationary condition of Π. It can be
easily established that the stationary condition of the functional defined by (15.92) is given by
p̄ − p ∂ p̄ (p̄ − p)2 ∂P (p̄)
P̄ − + : ∇p η dv − Gext (η) = 0
Ω P (p̄) ∂F 2 [P (p̄)]2 ∂F
(15.99)
p̄ − p
δp dv = 0
Ω P (p̄)
for all variations η and δ p̄ of displacement and independent pressure fields. In the above,
1
P̄ = P̄ (F ) ≡ σ̄(F ) F T
(15.100)
det F
is the first Piola–Kirchhoff stress obtained from the constitutive law.
For the discussion that follows, and also to outline the computer implementation based on
the spatial configuration, it is convenient to recast (15.99) in its spatial form
σ(F , p̄) : ∇x η dv − Gext (η) = 0
ϕ(Ω)
(15.101)
p̄ − p
δp dv = 0,
ϕ(Ω) J P (p̄)
that is, the hydrostatic component of the modified stress defined by (15.102) coincides
with the interpolated pressure and is not obtained from the deformation map through the
FINITE ELEMENTS FOR LARGE - STRAIN INCOMPRESSIBILITY 685
constitutive equations of the model. The need for the constraint (15.98) stems from the fact
that if it is not satisfied by the constitutive model in question, then (15.103) does not hold
in general. We finish our discussion on the meaning of the stationary conditions of the u/p
formulation, by considering equation (15.101)2. This equation implies that the independent
pressure field satisfies the constitutive equation of the material only in an averaged (weak or
integral) sense. Its corresponding strong (pointwise) form is simply
p = p̄. (15.104)
Remark 15.8. To see how (15.98) limits the possible constitutive models for which the
above interpretation of the stationary conditions remains valid, let us consider the class of
hyperelastic models whose free-energy function can be split into a sum of a purely isochoric
and a purely volumetric contribution; that is, we consider materials with free-energy function
of the type
ψ(F ) ≡ ψiso (F iso ) + ψv (J). (15.105)
The Hencky model as well as the regularised versions of the neo-Hookean, Mooney–Rivlin
and Ogden models discussed in Chapter 13 fall into this category. For such models, it can be
shown that (15.98) is equivalent to the restriction
∂ 2 ψv
= 0, (15.106)
∂J 2
or, equivalently,
∂ p̄
= 0. (15.107)
∂J
Free-energy functions based on the logarithmic bulk term, such as those defined by (13.43)
(13.44), (13.52) and (13.56), do not satisfy the above constraint in general. However, modified
versions of such models, obtained by replacing the volumetric contribution to ψ with the
alternative function referred to in Remark 13.1 (page 526), do satisfy (15.105) and can be
used in conjunction with u/p formulations.
npres
h
p(ξ) = pi N̂i (ξ), (15.108)
i=1
where pi and N̂i (i = 1, 2, . . . , npres ) are, respectively, the element pressure degrees of
freedom and the prescribed pressure interpolation functions and npres is the prescribed total
number of pressure degrees of freedom of the element. The interpolated pressure within the
element can be equivalently expressed as
h
p(ξ) = pT(e) N̂(ξ), (15.109)
686 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
npres N̂1 N̂2 N̂3 N̂4 N̂5 N̂6 N̂7 N̂8 N̂9
1 1
3 1 ξ η
4 1 ξ η ξη
6 1 ξ η ξη ξ2 η2
2
8 1 ξ η ξη ξ η2 ξ2 η ξη 2
9 1 ξ η ξη ξ2 η2 ξ2 η ξη 2 ξ2 η2
Variations of pressure are interpolated in the same way. The interpolation functions are
assumed discontinuous across element boundaries. Typical pressure interpolation functions
for two-dimensional quadrilateral u/p elements are shown in Table 15.1
The above interpolations result in the following discrete version of (15.101):
r(u, p) ≡ f int (u, p) − f ext = 0
(15.111)
s (u , p ) = 0 (e = 1, 2, . . . , n ),
(e) (e) (e) el
where f ext is the global standard external load vector and f int is the assembly of element
internal force vectors
int
f(e) (u(e) , p(e) ≡ BT σ(F , p(e) ) dv, (15.112)
ϕ(Ω(e) )
where the array σ contains the components of the modified stress defined by (15.102).
The arrays p and p(e) are, respectively, the global and element vectors of pressure degrees
of freedom and B is the standard discrete symmetric gradient operator (the same as in
virtual work-based elements). The element residual vector s(e) – associated with the weak
enforcement of the pressure constitutive relation – is given by
p̄ − p
s(e) (u(e) , p(e) ) ≡ N̂ dv. (15.113)
ϕ(Ω(e) ) J P (p̄)
If the algorithmic constitutive function does not possess a potential structure (refer to the text
surrounding expression (15.64) on page 674) the variational basis of the u/p formulation is
lost.
where hU and hP are, respectively, the spaces of interpolated displacements and interpolated
pressures generated by the finite element discretisation. In practice, we are looking for
elements that have the least possible number of displacement degrees of freedom combined
with the greatest number of pressure degrees of freedom that do not lock. In two dimensions,
the nine-noded quadrilateral with three pressure degrees of freedom is known to satisfy the
LBB condition. This element, named 9/3 is recommended by Sussman and Bathe (1987) for
finite strain analysis. For three-dimensional analysis, these authors suggest the use of the
so-called 27/4 elements – 27-noded element with four pressure degrees of freedom.
where r(e) is the element residual vector associated with the global vector r. The matrix
a is the matrix form of the spatial tangent modulus (the same as that computed for the
corresponding standard displacement-based element) and b is the matrix form of the fourth-
order tensor defined as
−1 p̄ − p
b≡ M⊗M+ (M ⊗ N + N ⊗ M )
P (p̄) P (p̄)
(p̄ − p)2 p̄ − p (p̄ − p)2
− N ⊗ N − d + e, (15.116)
[P (p̄)]3 P (p̄) 2 [P (p̄)]2
where
1 ∂ p̄ 1 ∂P (p̄) T
M≡ F T, N ≡ F (15.117)
J ∂F J ∂F
and d and e are the fourth-order tensors defined by the components
2
1 ∂ p̄ 1 ∂ 2 P (p̄)
d ≡ Fjm Fln , e ≡ Fjm Fln . (15.118)
J ∂F ∂F imkn J ∂F ∂F imkn
With δu(k) we apply the Newton–Raphson correction to the vector u of nodal displacements:
u(k) = u(k−1) + δu(k) , (15.124)
and to the element vectors p(e) of pressure degrees of freedom:
Having computed the new displacement and pressure degrees of freedom, we compute,
for each element, the modified Cauchy stress according to (15.102) and then assemble the
internal force vector defined by (15.112) and the element vector defined by (15.113). The
iterative process is stopped when the norm of the global vector r and the element vectors s(e)
fall below a prescribed convergence tolerance.
Remark 15.9. As for EAS elements (refer to Remark 15.7, page 678) the variational basis
of the u/p formulation ensures that the stiffness matrix K̂T is symmetric for materials that
possess a potential structure (materials with symmetric a). For materials with unsymmetric
tangent modulus, the method loses its variational justification and K̂T is unsymmetric.
variation of (·).
16 ANISOTROPIC FINITE
PLASTICITY: SINGLE CRYSTALS
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
692 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
The plastic deformation of crystals is mainly the result of sliding between crystal blocks along
well-defined crystallographic planes. A simple analogy can be made between such planes and
playing cards which slide relative to each other in a pack of cards. As atomic bonds strengthen
with the decrease in atomic distance, crystallographic sliding (which requires atomic bonds to
be cyclically broken and re-established) tends to occur between preferential planes of greatest
atomic density. These planes are called slip planes. This principle is schematically illustrated
in Figure 16.2(a). The bonds between atoms lying within slip planes are the strongest, whilst
bonds between atoms of neighbouring slip planes are weaker. By the same token, the direction
of sliding follows one of the closest-packed directions within the slip plane. The closest-
packed directions are called slip directions. A pair comprising a slip plane and a slip direction
defines a slip-system. In an fcc crystal, for instance, the closest-packed slip planes are the
four possible planes of the type illustrated in Figure 16.2(b). Each slip plane contains three
slip directions (indicated by the arrows in Figure 16.2(b)). The total number of slip systems
in this case is twelve and crystallographic sliding may happen in any such system. Other
crystallographic arrangements found in metal crystals are the body-centred cubic (bcc) and
the hexagonal close-packed (hcp) structures. These are illustrated in Figure 16.3.
ANISOTROPIC FINITE PLASTICITY: SINGLE CRYSTALS 693
slip plane
stronger
bonds
(a) (b)
Figure 16.2. Slip planes and slip directions: (a) the concept; (b) slip planes and slip directions in an fcc
crystal.
bcc hcp
Figure 16.3. The bcc and hcp cell structures. Schematic illustration.
τ ≡ (σ m) · s = σ : (s ⊗ m). (16.1)
The resolved stress τ is the shear force transferred across the slip plane in the direction of
s. Thus, plastic slip occurs in system {s, m} when τ reaches a critical level. This critical
694 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
value is analogous to the uniaxial yield stress of the von Mises theory of Chapter 6 (refer
to Subsection 6.4.2, page 162) and, in general, will be made dependent on temperature as
well as on the history of sliding. The dependence of the critical resolved Schmid stress on
the history of sliding is the crystal analogue of the phenomenon of hardening discussed in
Section 6.6 (from page 177) in the context of purely phenomenological plasticity.
γ pure
crystallographic
slip
1
e
F = I + γ s m
p F
m lattice
lattice rotation
distortion
initial
crystallographic
arrangement
Figure 16.4. The multiplicative decomposition of the deformation gradient. Schematic illustration.
The plastic deformation gradient represents a continuum measure of the microscopic sliding
along crystallographic planes. The elastic deformation gradient, on the other hand, provides
a continuum measure of crystal lattice distortion, which can be recovered upon complete
unloading of the material, as well as the rigid rotation. Consider a crystal containing a single
slip system {s, m} (see schematic illustration of Figure 16.4). A plastic slip of magnitude γ
along the system {s, m} corresponds to the plastic deformation gradient
F p = I + γ s ⊗ m. (16.3)
It is worth remarking here that whilst rigid rotations contained in F e are generally large,
the actual elastic distortion of the lattice observed in metal crystals is usually small
(infinitesimal).
Single slip
For a crystal yielding along a single slip-system {s, m} the evolution of the plastic
deformation gradient is defined by the equation
Lp = γ̇ s ⊗ m. (16.5)
To show that this equation leads to (16.3), note that it can be rephrased as
p
Ḟ = [γ̇ s ⊗ m] F p . (16.6)
696 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Without loss of generality, γ̇ can be assumed constant within any interval where the process
is monotonic. Then, the bracketed term on the right-hand side of (16.6) becomes constant
and the following exact solution (see Section B.1, page 747) is obtained if F p |γ=0 = I (the
virgin state) is taken as the initial condition:
By using the series representation (B.4) (page 747) of the tensor exponential function and
taking into account that the orthogonality between s and m implies
(s ⊗ m)(s ⊗ m) = 0, (16.8)
Multislip
In the presence of nact active slip systems (systems undergoing slip), the above plastic flow
equation is generalised as
nact
Lp = γ̇ α sα ⊗ mα , (16.9)
α=1
where {s , m } are the unit vectors that define slip system α and the multiplier γ̇ α is the
α α
Plastic slip may occur on a slip system only if the resolved shear stress on that system reaches
a critical level. Here, this principle will be expressed in terms of the Kirchhoff stress tensor, τ.
Accordingly, definition (16.1) is generalised and the resolved shear stress on a system α is
defined as
τ α ≡ Re T τ Re : sα ⊗ mα . (16.10)
Note that, as vectors sα and mα are defined on the intermediate configuration, the Kirchhoff
stress tensor (which is defined on the deformed configuration) has been appropriately rotated
back (with Re ) to the intermediate configuration before being resolved along slip system α.
Yield criterion
To formulate a yield criterion for single crystals within the framework of multisurface
plasticity, it is convenient to split each physical slip system of the crystal into two mirrored
systems
{sα , mα } and {sβ , mβ } ≡ {−sα , mα }. (16.11)
where τyα is the critical resolved shear stress for system α. In the presence of hardening, the
value of τyα depends on the history of the deformation process. It is important to note that the
functions Φα are anisotropic functions of τ. For a given τyα a hypersurface in the space of
Kirchhoff stresses is defined by
for each system α. Plastic slip in system α may occur only if the stress state is on the
above defined surface. The 2 × nsyst surfaces of this type form the envelope of the elastic
domain where the mechanical response of the crystal is fully reversible. The elastic domain
is defined by
E ≡ {τ | Φα (τ α (τ), τyα ) < 0, α = 1, . . . , 2nsyst }. (16.14)
The set of plastically admissible stresses is the closure of E, defined as
The boundary of the elastic domain is the yield surface, where plastic slip may occur. The
yield surface is defined by
2nsyst
Lp = γ̇ α sα ⊗ mα , (16.17)
α=1
or, equivalently,
2n
syst
p
Ḟ = γ̇ s ⊗ m
α α α
F p, (16.18)
α=1
Φα ≤ 0, γ̇ α ≥ 0, Φα γ̇ α = 0, (16.19)
2nsyst
τ̇yα = hαβ γ̇ β , α = 1, . . . , 2nsyst ,
β=1
where hαβ defines a matrix of hardening coefficients, whose components may depend on the
history of deformation.
In what follows, we shall adopt a much simpler law, known as Taylor hardening, in which
the resolved yield stress is the same for all systems. With τy denoting the common critical
value, the yield function for the generic system α is redefined as
Φα (τ α , τy ) = τ α − τy . (16.20)
γ≡ γ̇ α dt. (16.22)
0 α=1
ψ e (F e ) (16.24)
is introduced, and the standard constitutive law for the Kirchhoff stress tensor follows as
∂ψ e eT
τ = τ(F e ) = ρ̄ F . (16.25)
∂F e
This completes the definition of the continuum constitutive model of single crystals. The
model is summarised in Box 16.1.
ANISOTROPIC FINITE PLASTICITY: SINGLE CRYSTALS 699
Box 16.1. General finite strain anisotropic single crystal plasticity model.
F = F eF p
τ α = ReT τ Re : (sα ⊗ mα )
2nsyst
γ̇ = γ̇ α
α=1
Φα ≤ 0, γ̇ α ≥ 0, γ̇ α Φα = 0
for α = 1, . . . , 2nsyst
Consistency check
The next step is to check whether the elastic trial state violates plastic admissibility. If the
trial state lies in the elastic domain or on the yield surface, i.e. if
Φα trial ≡ τn+1
α trial
− τy (γn+1
trial
) ≤ 0, (16.31)
(·)n+1 := (·)trial
n+1 (16.32)
and the state update is complete. Otherwise, the multisurface return mapping described below
is applied.
The standard backward Euler discretisation of the Taylor hardening variable evolution
equation gives
γn+1 = γn + ∆γ α . (16.35)
α∈A
is traceless. As a result of the use of the exponential integrator, the corresponding incremental
plastic deformation gradient,
p
F ∆ ≡ exp ∆γ s ⊗ m ,
α α α
(16.37)
α∈A
is isochoric (the basic properties of the tensor exponential are listed in Section B.1.1,
page 748), i.e.
det[F p∆ ] = 1. (16.38)
Thus, the above update formula for F p preserves exactly the plastic incompressibility of the
continuum model. This property is pointed out by Steinmann and Stein (1996) as a vital
source of accuracy of the numerical integration scheme.
Remark 16.2. Recall that in the isotropic context (see derivation in Section 14.4.2, page 591)
the tensor exponential function does not need to be explicitly evaluated in the return-
mapping procedure. In that case, when the logarithmic strain-based model is adopted, the
tensor exponential function is conveniently cancelled out of the relevant update formulae. In
contrast, in the present anisotropic case, the tensor exponential function cannot be eliminated
from the return-mapping equations and will have to be evaluated (expressions (16.33)
and (16.34)) in the computational procedure.
Remark 16.3. Note that the incremental flow vector is generally unsymmetric. Thus, the
evaluation of the exponential of an unsymmetric tensor is required. The evaluation will be
carried out here by means of the procedure described in Section B.1 (page 747) which makes
use of the series representation of the tensor exponential function.
The exact computation of the derivative of the exponential of a generally unsymmetric tensor
is discussed in detail in Section B.2 of Appendix B and is implemented in HYPLAS subroutine
DEXPMP. The term H is the Taylor hardening modulus, defined as
dτy
H≡ . (16.47)
dγ
In deriving the above Jacobian, we have made use of the fact that
dγ
= 1, (16.48)
d∆γ β
which follows by differentiation of (16.35).
ANISOTROPIC FINITE PLASTICITY: SINGLE CRYSTALS 703
{sα ⊗ mα | α = 1, . . . , nsyst }
is linearly dependent in the space of traceless tensors to which the incremental flow vector
belongs. Note that in two dimensions, for instance, the space of traceless tensors is spanned
by three linearly independent base tensors. In three dimensions eight linearly independent
tensors generate the traceless tensors space. Thus, if the total number of active slip systems
is greater than three in two dimensions (greater than eight in three dimensions), then the set
A is necessarily linearly dependent and multiple solutions ∆γ exist for the return-mapping
equations. In this case, the Jacobian of the return-mapping system of equations becomes
singular.
To overcome the problems associated with non-uniqueness, many authors have resorted
to the viscoplastic (rate-dependent) regularisation of the original elastoplastic (rate-
independent) model (Asaro and Needleman, 1985; Needleman et al., 1985; Peirce, 1983;
Peirce et al., 1982, 1983). Non-uniqueness is not an issue in the viscoplastic context and the
rate-independent limit can be recovered by setting suitably low values for the rate-sensitivity
constant. However, the resulting set of equations to be integrated becomes extremely stiff as
the rate-independent limit is approached and the formulation of robust algorithms under such
conditions is not an easy task.
A completely different approach has been proposed by Borja and Wren (1993). Restricted
to the infinitesimal theory, they have developed a fairly robust algorithm for selection of active
systems. However, in spite of its relative robustness, the active set determination algorithm
of Borja and Wren (1993) is reported to fail under certain circumstances. This algorithm is
described next.
generate a sequence of sets {A (1) , A (2) , . . .} as the Newton iterations for solution of (16.39)
are applied. It is expected that, at the end, the algorithm will converge to a state (and an active
set) where the discrete complementarity condition expressed by (16.41, 16.42) is satisfied.
We may then proceed as follows. During a generic Newton iteration k, just after the new
α
guesses ∆γ(k) have been obtained according to (16.43), we reset the incremental multipliers,
which are currently negative, to zero:
α α
∆γ(k) := max[ 0, ∆γ(k) ]. (16.49)
With the above reset plastic multipliers, we compute the corresponding yield function values,
Φα
(k) , and define the following sets:
(k)
AΦ ≡ {α ∈ {1, 2, . . . , 2nsyst } | Φα
(k) > 0}
(16.50)
(k)
A∆γ ≡ {α ∈ {1, 2, . . . , 2nsyst } | ∆γ(k)
α
> 0}.
The above set is then used as the new trial active set in the following Newton iteration k + 1.
A solution to (16.39) that satisfies the discrete complementarity will have been obtained if,
after an iteration k, we have
(k+1)
AΦ = ∅.
To start up the algorithm, Borja and Wren (1993) suggest the following trial active set
A (0) := An , (16.52)
where An is the converged active set obtained for the Gauss point in question at the end of
the previous global load increment. Miehe (1996a,b), on the other hand, adopts
A (0) := A trial
≡ {α ∈ {1, . . . , 2nsyst } | Φα trial > 0}. (16.53)
s2
m1
s1
m 2
β
θ
x
no impact on the results of numerical simulations. Thus, the definition of the elastic law
in the present context will be regarded as a matter of numerical convenience rather than
a physically-based choice. In the isotropic context of Chapter 14, the logarithmic strain-
based Hencky model was found to be particularly convenient. Here a compressible neo-
Hookean model will be adopted instead. Originally proposed by Miehe (1996a,b), the use
of the compressible neo-Hookean material in the single crystal plasticity context was shown
to lead to a relatively simple format of return-mapping equations.
The compressible neo-Hookean model was described in detail in Chapter 13. In the present
elastoplastic context, the regularised neo-Hookean potential (13.44) (page 526) is redefined
in terms of the elastic left Cauchy–Green tensor as
ρ̄ ψ e (Beiso ) ≡ 1
2 G (tr[Beiso ] − 3) + 1
2 K (ln J e )2 , (16.55)
where G and K denote, respectively the shear and bulk moduli and
J e ≡ det[F e ]. (16.56)
Beisois the isochoric (volume-preserving) component of the elastic left Cauchy–Green strain
tensor, Be ,
−2
Beiso ≡ F eiso (F eiso )T = (J e ) 3 Be , (16.57)
with F eiso denoting the isochoric component of F e ,
− 13
F eiso ≡ (J e ) F e. (16.58)
The constitutive law for the Kirchhoff stress is obtained directly from the above free-energy
potential as
∂ψ
τ = 2 ρ̄ Be = G dev[Beiso ] + K (ln J e ) I. (16.59)
∂Be
where Ceiso ≡ (F eiso )T F eiso = ReT Beiso Re is the isochoric right elastic Cauchy–Green
strain tensor. Then, after a straightforward manipulation, we end up with the simpler final
formula
τ α = G s̄α · m̄α , (16.61)
where s̄α (m̄α ) is the isochoric elastic push-forward of sα (mα ),
s̄α ≡ F eiso sα
(16.62)
m̄α ≡ F eiso mα .
ANISOTROPIC FINITE PLASTICITY: SINGLE CRYSTALS 707
where
e trial α e trial
s̄α trial = F iso s , m̄α trial = F iso mα . (16.64)
Next, we proceed to the standard consistency check of page 700. If the process is elastic, the
stresses are updated according to the compressible neo-Hookean relation (16.59). Otherwise
we apply the return-mapping algorithm whose equations are described in the following.
and F eiso is the function of ∆γ defined through the exponential map-based update formula
F iso (∆γ) ≡ F iso
e e trial
exp − ∆γ s ⊗ m .
α α α
(16.67)
α∈A
The determination of the above active set A will be discussed later. For the time being, we
assume A to be known a priori.
Due to the isochoric/volumetric split of the hyperelastic response of the present model,
the return mapping, which effectively involves only the isochoric component of the elastic
deformation gradient, affects only the deviatoric component of the Kirchhoff stress tensor.
The updated hydrostatic Kirchhoff stress is the one at the elastic trial state. Note that this
is not necessarily true in the general case described in Section 16.5 where the underlying
hyperelastic model may have coupling between isochoric and volumetric responses. Here,
after solution of (16.65), we update
where
e
Jn+1 = Jn+1 ≡ det[F n+1 ]. (16.69)
708 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
where H 3
is the Taylor hardening modulus and E denotes the derivative of the exponential
map at − α∈A ∆γ α sα ⊗ mα .
{1}, {2}, {3}, {4}, {1, 2}, {2, 3}, {3, 4}, {1, 4}.
In practice, however, we do not need to try all the above combinations of active systems.
The combinations we try will depend on the set A trial . The tentative active sets shown in
Table 16.1 are sufficient.
Remark 16.4. For large numbers of slip systems, the number of possible combinations of
active systems increases dramatically and the present active set search procedure becomes
computationally prohibitive.
The overall elastic predictor/return-mapping integration algorithm for the planar double-
slip single crystal model is summarised in Boxes 16.2 and 16.3 in standard pseudo-code
format. The procedure is implemented in HYPLAS subroutine SUPDSC (State Update for the
Planar Double-slip Single Crystal model).
ANISOTROPIC FINITE PLASTICITY: SINGLE CRYSTALS 709
A (1) ≡ A trial
1 2 3 4 1, 2 2, 3 3, 4 1, 4
A (2)
1, 4 1, 2 2, 3 3, 4 1 2 3 1
A (3) 1, 2 2, 3 3, 4 1, 4 2 3 4 4
A (4) – – – – 1, 4 1, 2 2, 3 1, 2
A (5) – – – – 2, 3 3, 4 1, 4 3, 4
(i) Given the deformation gradient F ∆ := I + ∇n [∆u], evaluate the elastic trial state
e trial
F n+1 := F ∆ F ne
trial
γn+1 := γn
ELSE GOTO BOX 16.3 – return map (update F eiso and γn+1 )
(v) Update F en+1 and the Cauchy stress tensor
1
F en+1 := J 3 F eiso
Beiso := F eiso (F eiso )T
τn+1 := G dev[Beiso ] + K (ln Jn+1 ) I
σn+1 := J −1 τn+1
Remark 16.5. The algorithm of Boxes 16.2 and 16.3 is equally applicable to three-
dimensional cases as well as to other two-dimensional models, provided that the slip systems
710 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Box 16.3. Return-mapping algorithm for the planar double-slip single crystal
model.
(i) Set tentative active sets according to Table 16.1 and initialise tentative set counter:
itent := 1
(ii) Initial guess for plastic multiplier(s)
∆γ α := 0, α ∈ A (itent )
(vi) Update F eiso (expression (16.67)) and γn+1 (expression (16.35)) and compute yield
function values, Φ̃α , for all systems α = 1, . . . , 2nsyst (expression (16.65))
(vii) Check convergence
are linearly independent. The only modification required is the redefinition of the possi-
ble combinations of active systems described in Table 16.1. However, as pointed out in
Remark 16.4, the use of the present trial-and-error search procedure with a larger number
of slip systems may become impractical.
η
y
s2 s1
initial configuration initial lattice
deformed configuration
β = 60 o orientation
1 θ = 60o
until the final state with a shear deformation parameter η = 3 is reached. During the process,
the state variables (stress, elastic deformation gradient and hardening variable) are obtained
by means of the exponential map-based integration algorithm. In order to assess the finite step
accuracy of the algorithm under the present circumstances, the problem is solved with the
final shear parameter η = 3 attained in 10, 30 and 300 equally sized increments. The material
constants are the same as those adopted by Miehe (1996a,b) and are listed in Table 16.2.
The values of the Taylor hardening variable, γ, obtained along the process are plotted in
Figure 16.7. The rotation of the crystal lattice is also shown. The crystal lattice rotation is
the (in-plane) rotation angle (here assumed counterclockwise-positive) corresponding to the
elastic rotation tensor, Re . The elastic rotation tensor is obtained from the polar decompo-
sition of the elastic deformation gradient, F e . The lattice rotation tends asymptotically to
−60o – orientation where slip-system 1 is aligned with the x-axis. Note that in all cases (10,
30 and 300 increments) virtually identical results are obtained, which emphasises the high
accuracy of the exponential map-based integration algorithm. This is also observed in the
graphs of Figure 16.8 where the in-plane Kirchhoff stress components are plotted. The high
accuracy of the exponential map-based algorithm is due partly to its natural plastic volume
conservation property. As shown by Miehe (1996b), the stress components τxx and τyy are
particularly sensitive to this property in the present problem. When using an integration
algorithm where the plastic flow rule is discretised by a standard backward Euler scheme
together with a posteriori plastic volume corrections, Miehe (1996b) showed that a large
number of increments are required to produce reasonable accuracy for stress components τxx
and τyy .
712 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
3.5 0
10 increments
3 10 increments -10 30 increments
0.5 -50
0 -60
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
shear parameter, η shear parameter, η
Figure 16.7. The model problem. Results for hardening internal variable and lattice rotation.
0.2 0.05
10 increments 10 increments
30 increments 30 increments
0.15 300 increments 0 300 increments
0.1 -0.05
τxx
τyy
0.05 -0.1
0 -0.15
-0.05 -0.2
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
shear parameter, η shear parameter, η
0.2
0.18 10 increments
30 increments
0.16 300 increments
0.14
shear stress, τxy
0.12
0.1
0.08
0.06
0.04
0.02
0
0 0.5 1 1.5 2 2.5 3
shear parameter, η
Figure 16.8. The model problem. In-plane Kirchhoff stress components (kN/mm2 ).
ANISOTROPIC FINITE PLASTICITY: SINGLE CRYSTALS 713
nact
r≡ |Φ̃α /τy |,
α=1
is shown in Table 16.3 for typical iterations. The results of Table 16.3 were obtained in the
10 increment case.
Remark 16.6. Before proceeding, it is worth remarking that the tangent moduli derived in
this section are valid for any (two- or three-dimensional) single crystal models based on
the neo-Hookean hyperelastic description and the present exponential-map based integration
algorithm. Only the active set search procedure has to be changed to accommodate other
models of this class and this procedure does not affect in any way the tangent moduli.
analogous expression
aeijkl = ceijkl + δik σjl , (16.72)
e
where c has the compact representation
2G K 2
ce = tr[Beiso ] Id − 2p IS + I ⊗ I − [σd ⊗ I + I ⊗ σd ], (16.73)
3J J 3
with σd being the Cauchy stress deviator and p denoting the Cauchy hydrostatic pressure,
p ≡ tr[σ]/3.
where the isochoric elastic trial deformation gradient is a function of the total deformation
gradient F n+1 . In view of the hyperelastic law, the incremental constitutive function for the
Kirchhoff stress, τ̂, consistent with the return-mapping algorithm, is defined as
where
τd (F eiso ) ≡ G dev[F eiso F eiso T ]. (16.76)
ep
Our task now is to derive the elastoplastic spatial tangent modulus, a , defined as
ep 1 ∂τ̂
aijkl = (F n+1 )lm − σil δjk . (16.77)
Jn+1 ∂F n+1 ijkm
We then start by concentrating on the derivation of an explicit formula for the derivative
∂τ̂
,
∂F n+1
which is the only term in (16.77) that depends on the algorithmic constitutive function. By
straightforward differentiation of (16.75) we obtain
∂τ̂ ∂τd ∂Beiso ∂F iso K ∂
= : :P: + I⊗ det[F ],
∂F ∂Beiso ∂F eiso ∂F J ∂F
where the subscript n + 1 has been omitted for notational convenience and the fourth-order
tensor
∂F eiso
P≡ , (16.78)
∂F iso
ANISOTROPIC FINITE PLASTICITY: SINGLE CRYSTALS 715
is the only contribution to the derivative of τ̂ that depends on the particular return-mapping
algorithm adopted. After some straightforward derivations considering the elastic law and the
definitions of Beiso and F iso we obtain the components
∂τ̂ G
= 1 d : b : P : f + K I ⊗ F T, (16.79)
∂F ijkl J 3
where
bijkl ≡ [δik (F eiso )jl + δjk (F eiso )il ]
where
(F p∆ )−1 = exp − ∆γ s ⊗ m
α α α
α∈A
and
−1
d(F p∆ ) = −E : d∆γ α sα ⊗ mα .
α∈A
3 that E denotes the derivative of the exponential map evaluated at the argument
Recall
− α∈A ∆γ α sα ⊗ mα . The differentials of the plastic multiplier, d∆γ α , which take part
in the above expression are obtained by the standard procedure that consists in differentiating
the discrete consistency condition (16.65), having F eiso also as a variable, to obtain
dΦα = G [m̄α ⊗ sα + s̄α ⊗ mα ] : dF eiso − H d∆γ β , ∀ α ∈ A,
β∈A
dΦα = 0, ∀ α ∈ A.
These last two expressions together with (16.80) yield, after some tensor algebra, the
following differential relation
With the substitution of (16.81) into (16.80), we obtain, again after some manipulations, the
following differential relation
dF eiso = P : dF iso , (16.82)
where the tangent operator P is defined by the components
Pijkl ≡ δik Fljp + (F iso
e trial
)im Emjpq Qpqkl , (16.83)
and Qpqkl are the components of the fourth-order tensor
Q≡− sα ⊗ mα ⊗ Sα . (16.84)
α∈A
An approximate formula for the elastoplastic tangent modulus tensor has been employed
e trial e
by Miehe (1996a) which can be obtained by setting E := IS and Fiso := Fiso in the
above exact expression. At the beginning of the first global equilibrium iteration of any
load increment – when all plastic multipliers have been reset to zero – we have E = IS and
e trial e
Fiso = Fiso and the approximate operator coincides with the exact one.
prescribed
vertical displacement, u
weaker
elements
L = 60mm
prescribed
vertical displacement, u
W = 20 mm
4
edge reaction (kN)
2 20⫻60 mesh
10⫻30 mesh
1
0
0 1 2 3 4 5
edge deflection, u (mm)
θ = 75o.
The angle between the slip systems remains β = 60o . The slip systems are then no longer
symmetrically oriented with respect to the vertical axis and the whole strip (rather than the
symmetric quarter adopted in the previous example) has to be discretised. A mesh of 20 ×
60 four-noded F -bar elements is used to discretise the strip (Figure 16.13). The vertical
displacement is prescribed for all nodes at the top and bottom edges. Here, we consider two
situations:
(a) clamped condition – constrained horizontal displacement of the top and bottom edges;
(b) free edges – only the mid-side nodes A and B are fixed in the horizontal direction.
The strip is stretched up to u = 9 mm. A total of 77 and 56 increments are used, respectively,
for the clamped and free edge conditions. The reaction–deflection diagrams obtained are
shown in Figure 16.14. A peak reaction of approximately 3.8 kN is observed for both
cases. The corresponding configurations are u = 3.25 mm for the clamped condition and
u = 4.25 mm for the free edge. These configurations are reached, respectively, in ten and
eight displacement increments. The sharp decrease in reactions observed in the graphs
ANISOTROPIC FINITE PLASTICITY: SINGLE CRYSTALS 721
y initial lattice
orientation
s1
s2
β = 60 o
θ = 75o
corresponds to the formation of shear bands which, in contrast to the previous example,
are unsymmetrically oriented with respect to the axis of stretching. The unsymmetric shear
bands are depicted in Figures 16.15 and 16.16, which show the deformed configurations at
u = 4.0 mm, u = 6.5 mm and u = 9.0 mm. Unlike in the previous example, no material
imperfection (weakened element) is required here to trigger localisation. The non-uniform
state of stress generated due to the unsymmetric lattice orientation is sufficient. In fact, if
the four elements that share the central node are weakened (with the same properties as the
weakened elements of the previous example), the results obtained are virtually identical to
those shown here.
One of the important physical properties of single crystal alloys is their distinctive resistance
to creep failure at high service temperatures when the crystallographic microstructure is
favourably oriented with respect to the direction of loading. This makes such materials
particularly suitable for applications such as, for instance, jet aero-engine turbine blades
which may operate at temperatures as high as 700–1000 ◦ C for long periods of time.
Under such high temperatures the time-dependence of the plastic flow becomes an important
property that cannot be disregarded if numerical simulations are to produce realistic results.
722 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
2
free edges
clamped
1
0
0 2 4 6 8 10
edge deflection, u (mm)
u = 4.0 mm
u = 6.5 mm
u = 9.0 mm
Figure 16.15. Unsymmetric localisation. Deformed configurations (free edges).
ANISOTROPIC FINITE PLASTICITY: SINGLE CRYSTALS 723
u = 4.0 mm
u = 6.5 mm
u = 9.0 mm
Figure 16.16. Unsymmetric localisation. Deformed configurations (clamped).
Slip-rate laws alternative to the above definition can be adopted. The suitability of any law
for a particular single-crystal material should be determined by experimental evidence. In
defining alternative slip-rate equations, we could adopt, for instance, models which include
the use of yield surfaces (in analogy with the elastoplastic model discussed in previous
sections of this chapter). In such a case, one possibility could be, for instance, a multisurface
single-crystal generalisation of equation (11.6):
1/
1 |τ α |
−1 if Φα (τ α , τy ) ≥ 0
α
γ̇ = µ τ y (16.91)
0 if Φα (τ α , τy ) < 0,
where the material constants are analogous to those of the corresponding original models and
the yield functions Φα are defined according to (16.20).
In view of the hyperelastic constitutive law, which defines τ as a function of F e , and the
viscoplastic slip rate equation (any of the above) that defines the slip rate γ̇ α as a function of
τ, we may write
τ α = τ α (F e ) and γ̇ α = γ̇ α (F e ). (16.93)
Then, following a standard backward Euler discretisation of any of the above slip-rate laws,
we have
∆γ α = ∆t γ̇ α(F en+1 ). (16.94)
By substituting this last expression into the update formula (16.34) the integration of the
constitutive equations of the general viscoplastic single crystal model† within the interval
[tn , tn+1 ] is reduced to the solution of the following residual equation for F en+1 :
nsyst
S(F e ) ≡ F e − F e trial
exp −∆t γ̇ α(F e ) sα ⊗ mα = 0, (16.95)
α=1
F/2 F/2
y
s2 s1 lattice
60o orientation
L0 60 o
W0 = L0
The exact Jacobian used in the Newton–Raphson scheme for the solution of residual
equation (16.95) is obtained simply by differentiating S with respect to the equation variable
e
Fn+1 . Again omitting the n + 1 subscript, we have in Cartesian components
n
syst
dS dγ̇ α
Jijkl ≡ e trial
= δik δjl + ∆t Fim Emjpq 0 ⊗ m0 ⊗
sα α
, (16.96)
dF e ijkl α=1
dF e pqkl
where only the derivative dγ̇ α /dF e depends on the particular slip-rate model adopted.
Clearly, the use of the above exact Jacobian results in quadratic rates of asymptotic
convergence in the iterative solution of the nonlinear algebraic residual system.
Here we shall assume, as in the rate-independent case discussed earlier, the hyperelastic law to
be given by the regularised neo-Hookean model (16.59). Again, the derivation of the tangent
modulus associated with the above integration algorithm is relatively lengthy. Only the final
726 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
16
100 increments
12 10 increments
elongation (%)
8
0
0 5 10 15 20
time (h)
expression is shown here. In the present case, the spatial tangent modulus is given by
G K
aijkl = F:A :H+ I⊗I − σil δjk , (16.97)
J 4/3 J ijkl
where
Fijkl ≡ δik (F eiso )jl + δjk (F eiso )il − 13 δij (F eiso )kl (16.98)
and
Hijkl ≡ δik Flj − 13 Fij δkl . (16.99)
The tensor A is the only algorithm-related term taking part in the assemblage of a. It is
defined as
∂F eiso
A≡ . (16.100)
∂F iso
The final expression for A reads
A = J−1 : C, (16.101)
where C is defined by the components
p −1
Cijkl ≡ −δik (F n+1 )lj . (16.102)
The above expression for A is obtained by linearising (16.95), also taking into account the
relationship between F e and F and the respective isochoric components.
defined in Figure 16.17. The dimensions of the element are L0 = W0 = 1 mm. Its thickness
is also 1 mm. The material properties of the rate-dependent model are listed in Table 16.3.
Initially, a total load F = 500 N is applied instantaneously (with ∆t = 0), subjecting the
element to a uniform axial stress of 500 MPa. The element is subsequently allowed to creep
for 20 hours under constant load (stresses will increase due to cross-section reduction).
The creeping phase is carried out in 10 and 100 equal-size time increments ∆t = 2h and
∆t = 0.2h, respectively. For both cases, the elongation of the element, L/L0 where L is
the deformed length, obtained in the simulation is plotted against time in Figure 16.18. The
results show that, in the present algorithmic set-up, large time steps are possible without
significant loss of accuracy. In addition, we remark that, as in the rate-independent problems
reported in Section 16.8, the rates of convergence to equilibrium obtained with the rate-
dependent model are also quadratic.
Appendices
A ISOTROPIC FUNCTIONS OF A
SYMMETRIC TENSOR
T his appendix presents some important definitions, properties and expressions involving
isotropic scalar- and symmetric tensor-valued functions of one symmetric tensor. Such
functions and their properties are exploited in various parts of this book in connection with
the definition of elastic and elastoplastic constitutive models as well as with their computer
implementation. Much of the material presented here is quite standard, so the proof to most
relations is omitted.
We remark that some of the formulae presented here for the tensor-valued functions have
been coded in program HYPLAS and are used in practice for the evaluation of functions as
well as function derivatives. The corresponding subroutines of HYPLAS are the following:
In line with the general layout adopted throughout this book, the corresponding formulae
have been conveniently grouped in Boxes A.1–A.6 in the standard pseudo-code format.
Readers who are interested only in the computational implementation of the above-mentioned
formulae are referred directly to those boxes.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
732 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
A.1.1. REPRESENTATION
The derivative of an isotropic function φ of the above type is the symmetric second-order
tensor
∂φ
.
∂X
This tensor is coaxial with X, i.e. any eigenvector of X is an eigenvector of ∂φ/∂X, so that
these two tensors commute
∂φ ∂φ
X =X . (A.6)
∂X ∂X
With {ei } denoting an orthonormal basis of eigenvectors of X, the spectral representation of
∂φ/∂X reads
∂φ ∂ φ̂
= ei ⊗ ei ; (A.7)
∂X i
∂xi
Y (X) : L ⊂ S → S (A.9)
is said to be isotropic if
Q Y (X) QT = Y (QXQT ) (A.10)
for all rotations Q.
The following important property holds for any isotropic function defined as above. If Y
is isotropic, then Y (X) and X are coaxial and Y (X) and X commute
A.2.1. REPRESENTATION
Let us now focus on the representation of isotropic tensor-valued functions of a tensor
in three-dimensional space. A tensor function is isotropic if and only if it admits the
representation
Y (X) = α0 I + α1 X + α2 X 2 , (A.12)
where
α0 = α0 (I1 (X), I2 (X), I3 (X));
are scalar-valued functions of the principal invariants of X. Note that the functions α0 , α1
and α2 are themselves isotropic scalar functions of a symmetric tensor.
Further, if the domain of Y is a set of invertible symmetric tensors, then Y is isotropic if
and only if it has the representation
where the scalars βΓ (Γ = 0, 1, −1) are functions of the principal invariants (2.72) of X.
Assertions (A.12) and (A.14) are well-known representation theorems for isotropic tensor
functions. Their formal proof can be found, for instance, in Gurtin (1981) and Ogden (1984).
where p is the number of distinct eigenvalues of Y (and X) and Ei denotes the eigenprojec-
tions of X (and Y ) (refer to Section 2.2.8, from page 25). In two-dimensional space (n = 2),
the eigenvalues yi of Y are obtained from the eigenvalues xi of X as
y1 = y(x1 , x2 )
(A.16)
y2 = y(x2 , x1 ).
In three-dimensional space (n = 3), the eigenvalues yi are given by
y1 = y(x1 , x2 , x3 )
y2 = y(x2 , x3 , x1 ) (A.17)
y3 = y(x3 , x1 , x2 ),
and the function y has the symmetry property
Remark A.1. If there are no repeated eigenvalues, closed formulae for D(X) in the
two- and three-dimensional cases can be obtained simply by taking the derivatives of
explicit expressions for xi and Ei . However, in the presence of repeated eigenvalues, the
eigenprojections (see expression (2.65)) as well as the eigenvalues are not differentiable –
in spite of the differentiability of Y . The final explicit formulae for the derivative D(X) at
arguments with repeated eigenvalues are obtained as limits of the above explicit expression.
ISOTROPIC FUNCTIONS OF A SYMMETRIC TENSOR 735
y1 := y(x1 , x2 )
y2 := y(x2 , x1 )
(iii) Assemble Y :
p
Y (X ) := yα Eα
α=1
x2α − I1 xα + I2 = 0, α = 1, 2, (A.22)
where
The solution of the quadratic equation (A.22) provides the exact formula for the eigenvalues
of X:
I1 + I12 − 4I2 I1 − I12 − 4I2
x1 = ; x2 = . (A.24)
2 2
If x1
= x2 , (2.65)1 results in the following closed formula for computation of the
eigenprojections of X in two dimensions:
1
Eα = [X + (xα − I1 )I ]. (A.25)
2 xα − I1
y1 − y2 2 2
∂yα
D(X) = [IS − E1 ⊗ E1 − E2 ⊗ E2 ] + Eα ⊗ Eβ , (A.28)
x1 − x2 α=1
∂xβ
β=1
where IS is the fourth-order tensor defined by (2.110) (page 31). As pointed out in
Remark A.2, if the eigenvalues are repeated, (A.25) is singular and cannot be differentiated.
ISOTROPIC FUNCTIONS OF A SYMMETRIC TENSOR 737
Also, the right-hand side of the above formula for D becomes indeterminate. In this case, D
is the limit
lim D(X), (A.29)
x1 →x2
which is obtained by repeated applications of l’Hôspital’s rule to (A.28). This tedious but
straightforward derivation of the limit is omitted here. The final expression for the derivative
at arguments with x1 = x2 , which exists for sufficiently smooth functions y, is
∂y1 ∂y1 ∂y1
D(X) = − IS + I ⊗ I. (A.30)
∂x1 ∂x2 ∂x2
The algorithm for computation of the derivative of Y (X) based on the above closed formulae
is described in Box A.3. The procedure is implemented in subroutine DGISO2 (Derivative of
a General ISOtropic tensor function of one tensor in 2-D) of program HYPLAS.
Otherwise, assume xi
= xj , where tol is a small tolerance whose value depends on the
machine precision.
738 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
2
∂yα dx3 ∂y3
3
dxj
+ Eα ⊗ + E3 ⊗ . (A.32)
α=1
∂x3 dX j=1
∂xj dX
The first summation on the right-hand side of the above equation is identical to the
corresponding two-dimensional expression and is computed by the closed formula listed in
Box A.3 for x1
= x2 . After straightforward manipulations, the contribution of the remaining
terms is found to be
2
∂yα ∂y3 ∂y3
D3 (X) ≡ Eα ⊗ E3 + E3 ⊗ Eα + E3 ⊗ E3 . (A.33)
α=1
∂x 3 ∂xα ∂x 3
The assemblage of the complete derivative is attained by adding D3 (X) to the corresponding
equation of Box A.3. This operation is included in subroutine DGISO2.
With repeated in-plane eigenvalues, x1 = x2 , the above formula is replaced with
∂y1 ∂y3 ∂y3
D3 (X) = Ip ⊗ E3 + E 3 ⊗ Ip + E3 ⊗ E3 , (A.34)
∂x3 ∂x1 ∂x3
where Ip is the second-order plane orthogonal projection tensor. It is important to note that
in Box A.3, IS and I are tensors on the plane. When applying the formulae of Box A.3 in
plane strain/axisymmetric situations, IS and I must be replaced, respectively, with the plane
projection operators Ip and Ip . The tensor Ip has the following matrix representation:
1 0 0
Ip = 0 1 0 (A.35)
0 0 0
in an orthonormal system where the out-of-plane direction is associated with the third line
and column. Analogously, the Cartesian components of Ip are given by
1
(δik δjl + δil δjk ) i, j, k, l ∈ {1, 2}
[Ip ]ijkl = 2 (A.36)
0 otherwise.
ISOTROPIC FUNCTIONS OF A SYMMETRIC TENSOR 739
I1 = tr[X]
As the X is real by assumption, its eigenvalues xi are real and the solution of (A.37) can be
computed exactly by the following expressions:
θ I1
x1 = −2 Q cos + ;
3 3
θ + 2π I1
x2 = −2 Q cos + ; (A.39)
3 3
θ − 2π I1
x3 = −2 Q cos + ,
3 3
If there are three distinct eigenvalues, the explicit formula for D is obtained simply by taking
the derivatives of the closed expressions for xi and Ei , given in Box A.5. Again, if there
are two or three identical eigenvalues, the expression for Ei becomes indefinite and the
final formula for the derivative is obtained by applying a limiting procedure to the general
formula valid for three distinct eigenvalues. Note that the limiting procedure in the three-
dimensional case is quite lengthy but, as in the two-dimensional case, involves only the
repetitive application of l’Hôspital’s rule. The final expressions are gathered in Box A.6.
In Box A.6, dX 2 /dX denotes the derivative of the square of a tensor with Cartesian
components given by
dX 2 1
= (δik Xlj + δil Xkj + δjl Xik + δkj Xil ) (A.46)
dX ijkl 2
y1 := y(x1 , x2 , x3 )
y2 := y(x2 , x3 , x1 )
y3 := y(x3 , x1 , x2 )
(iii) Assemble Y
p
Y (X ) := yi Ei
i=1
p
Y (X) ≡ y(xi ) Ei . (A.48)
i=1
Closed-form expressions for the derivative of functions of this class, employing the
eigenprojection-based representation, were derived by Carlson and Hoger (1986). Complete
algorithms using compact expressions for computation of functions of the above class (and
their derivatives) at invertible arguments have been proposed by Miehe (1993).
Remark A.3. Functions expressed as such define an important class of isotropic tensor-
valued functions of a tensor and are, clearly, particular cases of the general form (A.15).
The families of strain measures defined by (3.56) (page 53) and (3.60), for instance, are
members of this class. Other important functions, such as the tensor square root and the
√
tensor exponential, can also be expressed in the format (A.48) by setting y(xi ) ≡ xi and
y(xi ) ≡ exp(xi ), respectively.
I1 = tr[X ]
I2 = 12 {(tr X )2 − tr[X 2 ]}
I3 = det[X ]
−2 I13 + 9 I1 I2 − 27 I3
R=
54
R
θ = cos−1 "
Q3
I − 3 I2
2
Q= 1
9
√ θ I1
x1 = −2 Q cos +
3 3
√ θ + 2π I1
x2 = −2 Q cos +
3 3
√ θ − 2π I1
x3 = −2 Q cos +
3 3
Ej = I − Ei
– Else (x1 = x2 = x3 ),
E1 = I
After closed-form evaluation of xi and Ei in Box A.2, the following expression is used to
compute Y (X) in two dimensions
2
y(x ) E
if x1
= x2
i i
Y (X) = (A.49)
i=1
y(x1 ) I if x1 = x2 .
ISOTROPIC FUNCTIONS OF A SYMMETRIC TENSOR 743
where the subscripts (a, b, c) are cyclic permutations of (1,2,3). The computation of the
derivative, D(X), follows the expressions
3
y(xa ) dX 2
− (xb + xc ) IS
(xa − xb )(xa − xc ) dX
a=1
− [(xa − xb ) + (xa − xc )]Ea ⊗ Ea
− (xb − xc )(Eb ⊗ Eb − Ec ⊗ Ec ) + y (xa ) Ea ⊗ Ea
D(X) = (A.52)
if x
= x
= x
1 2 3
dX
2
s1 − s2 IS − s3 X ⊗ X + s4 X ⊗ I + s5 I ⊗ X − s6 I ⊗ I
dX
if xa
= xb = xc
y (x1 ) IS if x1 = x2 = x3
Remark A.4. The closed expressions (A.50) and (A.52) for the function derivative are equiv-
alent to those derived by Carlson and Hoger (1986). Their equivalence can be established
(after some algebra) by considering the standard identity for the directional derivative of the
square of a tensor in a generic direction T,
dX 2
[T ] = XT + TX,
dX
together with the identity
(Ei ⊗ Ei ) T = Ei T Ei ,
demonstrated by Carlson and Hoger, and the general closed formulae (2.65) for Ei .
T his appendix describes the computational procedures for evaluation of the tensor expo-
nential function (or exponential map) and its derivative. It also describes the application
of the exponential map in the numerical solution of a class of initial value problems of
particular relevance in computational mechanics. Note that the exponential of a symmetric
tensor is a particular member of the class of isotropic tensor-valued functions described in
Section A.5 (page 740) and is obtained by setting y(xi ) ≡ exp(xi ) in expression (A.48). In
this case, the function as well as its derivatives can be computed by the procedures already
explained in that section. In the present appendix, however, we are concerned with the more
general tensor exponential function whose domain is the entire space of (symmetric and
unsymmetric) real tensors in ndim dimensions. The computational procedures described here
have been implemented in the following subroutines of the program HYPLAS:
• EXPMAP – EXPonential MAP computation.
• DEXPMP – Evaluation of the Derivative of the EXPonential MaP.
The corresponding pseudo-codes are provided, respectively, in Boxes B.1 and B.2.
The above series is absolutely convergent for any argument X and, as its scalar counterpart,
can be used to evaluate the tensor exponential function to any prescribed degree of accuracy.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
748 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
† Note that for symmetric arguments, the computation of the tensor exponential can be performed more efficiently
by the procedure described in Section A.5. For skew-symmetric arguments, the closed formulae (B.13) or (B.14) can
be used for a more efficient computation of the tensor exponential.
750 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Proof. Follows by direct differentiation of the series representation (B.4) of the tensor
exponential and subsequent use of the general formula (B.19) for the Cartesian components
of the derivative of a positive integer power of a tensor.
THE TENSOR EXPONENTIAL 751
1
n
[D exp(X)]ijkl := [D exp(X)]ijkl + [X m−1 ]ik [X n−m ]lj
n! m=1
where nmax satisfies the accuracy requirement (B.17). The computer implementation of
the above expression follows the pseudo-code shown in Box B.2. The procedure has been
implemented in HYPLAS subroutine DEXPMP.
where
∆t ≡ tn+1 − tn , tn+θ ≡ tn + θ ∆t. (B.25)
The prescribed parameter θ satisfies
0 ≤ θ ≤ 1. (B.26)
1
The choice θ = 0 yields the explicit exponential map integrator. With choices θ = and θ = 1
2
the above formula is called, respectively, the midpoint and the backward (or fully implicit)
exponential map integrators. The algorithm is second-order accurate for θ = 12 and only first-
order accurate otherwise.
C LINEARISATION OF THE
VIRTUAL WORK
I N this appendix we present the derivation of the linearised version of the virtual work
equations. Linearisation of the virtual work gives rise to the tangent moduli which take
part in the assemblage of the tangent stiffness matrix – a crucial component of the implicit
finite element solution procedure described in Chapter 4. Tangent moduli associated with
specific material models and constitutive integration procedures are discussed in Parts Two
and Three of this book.
σ = σ(ε), (C.1)
so that the problem is reduced to finding a kinematically admissible field u that satisfies
G(u, η) = 0, ∀ η ∈ V, (C.2)
The dependence of G on the unknown function u follows from the constitutive dependence
of the stress tensor on the strain tensor which, in turn depends on the field u (not only its
pointwise value); that is, σ is a functional of the field u:
Here we want to linearise equation (C.2) with respect to the unknown u about an arbitrary
argument u∗ (refer to Section 2.6, from page 38, where the concept of linearisation is
discussed). In abstract notation, the linearised problem consists in finding the field δu such
† This assumption encompasses elastic constitutive models as well as incremental constitutive functions employed
in the numerical implementation of path-dependent (dissipative) material models. In the incremental case, σ is a
function of ε alone (analogous to an elastic constitutive function) within each time or load step.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
754 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
that
L(δu, η) ≡ G(u∗ , η) + DG(u∗ , η) [δu] = 0, ∀ η ∈ V, (C.5)
with
ε∗ = ∇s u∗ (C.9)
With the above at hand, we arrive at the following final expression for the linearised virtual
work equation at u∗ :
D : ∇s δu : ∇s η dv = − (σ : ∇s η − b · η) dv + t · η da, ∀ η ∈ V, (C.12)
Ω Ω ∂Ωt
where b̄ and t̄ denote, respectively, the reference body force and surface traction fields. In the
above, P is a functional of the displacement field due to its constitutive dependence on the
deformation gradient F = I + ∇p u. The material description of the virtual work equation
for the unknown displacement field u under finite deformations is given by (C.2) with the
functional G defined by (C.14).
The linearisation of equation (C.2, C.14) at a state defined by the field u∗ has the
representation (C.5), with the directional derivative (C.6) in the present case being given by‡
d
DG(u∗ , η) [δu] = [P (F ()) : ∇η − b̄ · η] dv − t̄ · η da
d =0 Ω ∂Ωt
d
= P (F ()) : ∇η dv, (C.15)
d =0 Ω
where we have defined
∗
F () = I + ∇p (u∗ + δu) = F + ∇p δu, (C.16)
with
∗
F = I + ∇p u∗ (C.17)
∗
denoting the deformation gradient for the displacement field u .
where
∂P
A≡ (C.19)
∂F F ∗
is generally termed the material tangent modulus. For materials with an elastic constitutive
law, this is also referred to as the first elasticity tensor.
‡ We have implicitly assumed here that the prescribed body force and surface traction fields are independent of
the displacement field. If the applied loads are configuration-dependent, i.e. functions of the displacement field, then
extra contributions associated to the linearisation of such terms have to be added to the directional derivative.
756 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
where b and t are, respectively, forces per unit volume and area of the deformed configuration
of the body in question and σ is the Cauchy stress tensor which, for an elastic constitutive
model, is a function of the deformation gradient:
valid for any vector field a and scalar field a. Likewise, the spatial version of the virtual
work directional derivative can be obtained from its material counterpart (C.18) simply by
replacing the material description with the corresponding spatial description of the relevant
fields.
By regrouping the terms of the above integrand, the directional derivative may be equivalently
written as
DG(u∗ , η) [δu] = a : ∇x δu : ∇x η dv, (C.25)
ϕ(Ω)
where a is termed generally the spatial tangent modulus. Strictly under elasticity, it is often
called the spatial elasticity tensor. It is defined by the Cartesian components
1
aijkl = Aimkn Fjm Fln . (C.26)
J
LINEARISATION OF THE VIRTUAL WORK 757
1 ∂τij
aijkl = Flm − σil δjk . (C.31)
J ∂Fkm
T his appendix describes the handling of second and fourth-order tensors in finite element
computer programs. It should be particularly helpful to those wishing to follow, in
program HYPLAS, the implementation of techniques discussed in this book that have been
presented almost exclusively in compact tensorial notation.
In finite element computer programs, the components of a symmetric second-order
tensor are usually stored as a single column array, whereas fourth-order tensor components
are stored in two-dimensional arrays. By arranging the relevant components consistently,
operations such as internal products between tensors and products between fourth and
second-order tensors can be conveniently carried out in the computer program as matrix
products.
The order in which components of a tensor can be stored in array format is not unique.
In the following, we show the convention adopted in many finite element programs and, in
particular, in the program HYPLAS.
Let us start with second-order tensors. Expression (2.27) (page 21) shows the matrix repre-
sentation of a generic tensor in terms of its Cartesian components. Here we shall be concerned
only with symmetric tensors (which are of relevance for finite element computations).
Second-order symmetric tensors will be converted into single column arrays and their actual
single array representation will depend on whether the tensor is a stress-like or strain-like
quantity. Let us start by considering the stress tensor, σ, in plane stress and plane strain
problems. In this case, the in-plane components of the matrix [σ] will be converted into a
single column array σ (the computer array representation will be denoted here by upright
bold-faced symbols) according to the rule
σ11
σ11 σ12
[σ] = −→ σ = σ22 . (D.1)
σ12 σ22
σ12
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
760 COMPUTATIONAL METHODS FOR PLASTICITY: THEORY AND APPLICATIONS
Only the three relevant independent components are stored in σ. In axisymmetric problems
we store, in addition to the above, the component σ33 ; that is
σ11
σ11 σ12 0
σ
[σ] = σ12 σ22 0 −→ σ = 22 , (D.2)
0 0 σ33 σ12
σ33
where the index 3 is associated with the circumferential direction. It should be noted here that
the σ33 stress is generally non-zero also under plane strain conditions. In this case, the stress
may be stored as in the above (i.e. including σ33 ), but the last element of the stress array will
be ignored in product operations such as (D.5).
In three-dimensions (this case is not implemented in HYPLAS), we have the conversion rule
σ11
σ22
σ11 σ12 σ13
σ
[σ] = σ12 σ22 σ23 −→ σ = 33 . (D.3)
σ12
σ13 σ23 σ33
σ
23
σ13
Let us now consider the strain tensor, ε. The rule for storage in this case is
ε11
ε22
ε11 ε12 ε13
ε33
ε ε ε
[ε] = −→ ε = 2ε . (D.4)
12 22 23
12
ε13 ε23 ε33
2ε
23
2ε13
Note that the shear components have been multiplied by a factor of two; that is, ε is the array
of engineering strains. The reason for this is that, in this way, the internal product between a
stress- and a strain-like tensor can be computed as a matrix product
σ : ε = σT ε. (D.5)
If we denote by δε a virtual strain tensor, i.e. the symmetric gradient virtual displacement
field, by applying the above conversion rules the corresponding virtual work reads
σ : δε = σT δε. (D.6)
In plane problems, the general conversion rule for strains is
ε11
ε11 ε12
[ε] = −→ ε = ε22 . (D.7)
ε12 ε22
2ε12
ARRAY NOTATION FOR COMPUTATIONS WITH TENSORS 761
However, note that in elastoplasticity under plane strain, even though the total strain
component ε33 vanishes, the corresponding elastic and plastic components do not. Thus, we
adopt the following storage rule
e
e ε11
ε11 εe12 0 e
e ε22
[εe ] =
ε12 ε22 0 −→ ε = 2εe
e e (D.8)
e
12
0 0 ε33
εe33
for the elastic strain (and the plastic strain). Axisymmetric implementations follow the above
rule also for the array conversion of the total strain tensor, ε.
D = Dijkl ei ⊗ ej ⊗ ek ⊗ el . (D.9)
dσ = D : dε. (D.10)
Fourth-order tensors will be stored in two-dimensional arrays. In its array form, the compo-
nents of D in plane strain problems (i.e. with i, j = 1, 2) will be arranged as
D1111 D1122 D1112
D= D2211 D2222 D2212 ,
(D.11)
D1211 D1222 D1212
so that the tangential relation between the in-plane stress array can be represented as the
matrix product
dσ = D dε. (D.12)
That the above is equivalent to (D.10) is left as an exercise for the interested reader. Note that
in elasticity and associative plasticity problems the tensor D has the symmetries
and in three-dimensions,
D1111 D1122 D1133 D1112 D1123 D1113
D2211 D2222 D2233 D2212 D2223 D2213
D3311 D3322 D3333 D3312 D3323 D3313
D=
.
(D.15)
D1211 D1222 D1233 D1212 D1223 D1213
D D2313
2311 D2322 D2333 D2312 D2323
D1311 D1322 D1333 D1312 D1323 D1313
Note that, according to the above rule, the fourth-order symmetric identity tensor defined
by (2.108) (page 31) is represented in plane problems as
1 0 0
IS =
1 0 .
(D.16)
1
sym 2
In using the above representation in computations, account should be taken of the fact that
ε11
IS ε =
ε22
= ε. (D.17)
ε12
In three-dimensions, we have
1 0 0 0 0 0
1 0 0 0 0
1 0 0 0
IS = 1 . (D.19)
0 0
2
1
0
sym
2
1
2
ARRAY NOTATION FOR COMPUTATIONS WITH TENSORS 763
With the above notation, the integrand on the right-hand side of (C.25) has the representation
a : T : U = UT a T. (D.23)
Again, note that when the fourth-order tensor has the major symmetries
and
a1111 a1121 a1112 a1122 a1133
a2111 a2121 a2112 a2122 a2133
a = a1211 a1221 a1212 a1222 a1233 . (D.26)
a a2233
2211 a2221 a2212 a2222
a3311 a3321 a3312 a3322 a3333
An analogous conversion rule can be defined for three-dimensional problems.
REFERENCES
Alfano, G., de Angelis F. and Rosati, L. 2001. General Solution Procedures in Elasto/Viscoplasticity.
Comp. Meth. Appl. Mech. Engng, 190, 5123–5147.
Anderson, H. 2003. An Implicit Formulation of the Bodner–Partom Constitutive Equations. Comp.
Struct, 81, 1405–1414.
ANSI. 1978. American National Standard Programming Language FORTRAN. New York.
Aravas, N. 1987. On the Numerical Integration of a Class of Pressure-Dependent Plasticity Models. Int.
J. Num. Meth. Engng, 24, 1395–1416.
Argyris, J.H. and Kleiber, M. 1977. Incremental Formulation in Nonlinear Mechanics and Large Strain
Elasto-plasticity – Natural Approach. Part I. Comp. Meth. Appl. Mech. Engng, 11, 215–247.
Argyris, J.H. and Symeonidis, Sp. 1981. Nonlinear Finite Element Analysis of Elastic Systems Under
Nonconservative Loading – Natural Formulation. Part I. Quasistatic Problems. Comp. Meth. Appl.
Mech. Engng, 26, 75–123.
Argyris, J.H., Doltsinis, J.St. and Kleiber, M. 1978. Incremental Formulation in Nonlinear Mechanics
and Large Strain Elasto-plasticity – Natural Approach. Part II. Comp. Meth. Appl. Mech. Engng, 14,
259–294.
Armero, F. and Oller, S. 2000. A General Framework for Continuum Damage Models. I. Infinitesimal
Plastic Damage Models in Stress Space. Int. J. Solids Structs, 37, 7409–7436.
Armero, F. and Pérez-Foguet, A. 2002. On the Formulation of Closest-Point Projection Algorithms in
Elastoplasticity – Part I: The Variational Structure. Int. J. Numer. Meth. Engng, 53, 297–329.
Armstrong, P.J. and Frederick, C.O. 1966. A Mathematical Representation of the Multiaxial
Bauschinger Effect. CEGB: Report RD/B/N 731.
Arridge, R.G.C. 1985. An Introduction to Polymer Mechanics. London: Taylor and Francis.
Asaro, R.J. 1983. Micromechanics of Crystals and Polycrystals. Adv. Appl. Mech., 23, 1–115.
Asaro, R.J. and Needleman, A. 1985. Texture Development and Strain Hardening in Rate Dependent
Polycrystals. Acta Metall., 33(6), 923–953.
Ascher, U.M. and Petzold, L.R. 1998. Computer Methods for Ordinary Differential Equations and
Differential-Algebraic Equations. Philadelphia: SIAM.
Atluri, S.N. 1984. On Constitutive Relations at Finite Strains: Hypo-Elasticity and Elasto-Plasticity
with Isotropic or Kinematic Hardening. Comp. Meth. Appl. Mech. Engng, 43, 137–171.
Auricchio, F., Beirão da Veiga, L., Lovadina, C. and Reali, A. 2005. A Stability Study of Some Mixed
Finite Elements for Large Deformation Elasticity Problems. Comp. Meth. Appl. Mech. Engng, 194,
1075–1092.
Barlat, F. and Lian, J. 1989. Plastic Behavior and Stretchability of Sheet Metals. Part I: A Yield Function
for Orthotropic Sheets Under Plane Stress Condition. Int. J. Plast., 5, 51–66.
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
766 REFERENCES
Basista, M., Krajčinović, D. and Šumarac, D. 1992. Micromechanics, Phenomenology and Statistics of
Brittle Deformation. Pages 1479–1490 of: Owen, D.R.J., Oñate, E. and Hinton, E. (eds), Computational
Plasticity: Fundamentals and Applications – Proceedings of the Third International Conference held in
Barcelona, 6–10 April 1992. Swansea: Pineridge Press.
Bataille, J. and Kestin, J. 1979. Irreversible Processes and Physical Interpretation of Rational Thermo-
dynamics. J. Non-Equilib. Thermodyn., 4, 229–258.
Bathe, K.-J. 1996. Finite Element Procedures. Englewood Cliffs, New Jersey: Prentice-Hall.
Bazaraa, M. and Shetty, C. 1979. Non-linear Programming: Theory and Applications. New York: Wiley.
Beatty, M.F. 1987. Topics in Finite Elasticity: Hyperelasticity of Rubber, Elastomers, and Biological
Tissues – with Examples. Appl. Mech. Rev., 40, 1699–1734.
Belytschko, T. and Bindeman, L.P. 1991. Assumed Strain Stabilization of the 4-Node Quadrilateral with
1-Point Quadrature for Nonlinear Problems. Comp. Meth. Appl. Mech. Engng, 88, 311–340.
Belytschko, T. and Bindeman, L.P. 1993. Assumed Strain Stabilization of the Eight Node Hexahedral
Element. Comp. Meth. Appl. Mech. Engng, 105, 225–260.
Belytschko, T., Liu, W.-K. and Moran, B. 2000. Nonlinear Finite Elements for Continua and Structures.
Chichester: Wiley.
Belytschko, T., Ong, J.S.-J., Liu, W.K. and Kennedy, J.M. 1984. Hourglass Control in Linear and
Nonlinear Problems. Comp. Meth. Appl. Mech. Engng, 43, 251–276.
Benallal, A., Billardon, R. and Doghri, I. 1988. An Integration Algorithm and the Corresponding
Consistent Tangent Operator for Fully Coupled Elastoplastic and Damage Equations. Comm. Appl.
Num. Meth., 4, 731–740.
Benallal, A., Billardon, R. and Lemaitre, J. 1991. Continuum Damage Mechanics and Local Approach
to Fracture: Numerical Procedures. Comp. Meth. Appl. Mech. Engng, 92, 141–155.
Benallal, A., Billardon, R., Doghri, I. and Moret-Bailly, L. 1987. Crack Initiation and Propagation
Analyses Taking into Account Initial Strain Hardening and Damage Fields. Pages 337–351 of:
Luxmore, A.L., Owen, D.R.J., Rajapakse, Y.P.S. and Kanninen, M.F. (eds), Numerical Methods in
Fracture Mechanics: Proceedings of the Fourth International Conference held in San Antonio, Texas,
23–27 March 1987. Pineridge Press.
Billington, E.W. and Tate, A. 1981. The Physics of Deformation and Flow. New York: McGraw-Hill.
Blatz, P.J. and Ko, W.L. 1962. Application of Finite Elasticity to the Deformation of Rubbery Materials.
Trans. Soc. Rheol., 6, 223–251.
Bodner, S.R. and Partom, Y. 1975. Constitutive Equations for Elastic Viscoplastic Strain Hardening.
J. Appl. Mech., 42, 385–389.
Boisse, Ph., Bussy, P. and Ladevèze, P. 1990. A New Approach in Non-linear Mechanics: The Large
Time Increment Method. Int. J. Num. Meth. Engng, 29, 647–663.
Boisse, Ph., Ladevèze, P. and Rougee, P. 1989. A Large Time Increment Method for Elastoplastic
Problems. Eur. J. Mech. A/Solids, 8(4), 257–275.
Bonet, J. and Wood, R.D. 1997. Nonlinear Continuum Mechanics. Cambridge: Cambridge University
Press.
Borja, R. and Wren, J.R. 1993. Discrete Micromechanics of Elastoplastic Crystals. Int. J. Num. Meth.
Engng, 36, 3815–3840.
Boucard, P.-A., Ladevèze, P., Poss, M. and Rougée, P. 1997. A Nonincremental Approach for Large
Displacement Problems. Comput. Struct., 64(1–4), 499–508.
Boyle, J.T. and Spence, J. 1983. Stress Analysis for Creep. London: Butterworths.
REFERENCES 767
Brezzi, F. and Fortin, M. 1991. Mixed and Hybrid Finite Element Methods. New York: Springer-Verlag.
Bueche, F. 1960. Molecular Basis for the Mullins Effect. J. Appl. Poly. Sci., 4(10), 107–114.
Caddemi, S. and Martin, J.B. 1991. Convergence of the Newton–Raphson Algorithm in Elastic-plastic
Incremental Analysis. Int. J. Num. Meth. Engng, 31, 177–191.
Carlson, D.E. and Hoger, A. 1986. The Derivative of a Tensor-valued Function of a Tensor. Quart. Appl.
Math., 44(3), 409–423.
Cescotto, S. and Zhu, Y.Y. 1995. Modelling of Ductile Fracture Initiation During Bulk Forming. Pages
987–998 of: Owen, D.R.J., Oñate, E. and Hinton, E. (eds), Computational Plasticity: Fundamentals
and Applications – Proceedings of the Fourth International Conference held in Barcelona, 3–6 April
1995. Swansea: Pineridge Press.
Chaboche, J.L. 1978. Description thermodynamique et phénoménologique de la viscoplasticité cyclique
avec endommagement. Technical Report 1978–3. Office National d’Etudes et de Recherches Aérospa-
tiales.
Chaboche, J.L. 1981. Continuous Damage Mechanics – A Tool to Describe Phenomena Before Crack
Initiation. Nucl. Engng. Des., 64, 233–247.
Chaboche, J.L. 1984. Anisotropic Creep Damage in the Framework of Continuum Damage Mechanics.
Nucl. Engng. Des., 79, 309–319.
Chaboche, J.L. 1988. Continuum Damage Mechanics: Part I – General Concepts and Part II – Damage
Growth, Crack Initiation and Crack Growth. J. Appl. Mech., 55, 59–72.
Chaboche, J.L. and Cailletaud, G. 1996. Integration Methods for Complex Plastic Constitutive Equa-
tions. Comp. Meth. Appl. Mech. Engng, 133, 125–155.
Chadwick, P. and Ogden, R.W. 1971. A Theorem in Tensor Calculus and its Application to Isotropic
Elasticity. Arch. Rat. Mech. Anal., 44, 54–68.
Chakrabarty, J. 1987. Theory of Plasticity. London: McGraw-Hill.
Chen, W-F. 1975. Limit Analysis and Soil Plasticity. New York: Elsevier.
Chen, W-F. 1982. Plasticity in Reinforced Concrete. McGraw-Hill.
Chen, W-F. and Mizuno, E. 1990. Nonlinear Analysis in Soil Mechanics. Theory and Implementation.
New York: Elsevier.
Chorin, A.J., Hughes, T.J.R., McCracken, M.F. and Marsden, J.E. 1978. Product Formulas and
Numerical Algorithms. Comm. Pure and Appl. Math., 31, 205–256.
Ciarlet, P.G. 1988. Mathematical Elasticity. Volume I: Three-dimensional Elasticity. Amsterdam: North-
Holland.
Cognard, J.Y. and Ladèveze, P. 1993. A Large Time Increment Approach for Cyclic Viscoplasticity. Int.
J. Plast., 9(2), 141–157.
Cognard, J.Y., Ladèveze, P. and Talbot, P. 1999. A Large Time Increment Approach for Thermo-
Mechanical Problems. Adv. Engng. Software, 30(9–11), 583–593.
Coleman, B.D. and Gurtin, M.E. 1967. Thermodynamics with Internal State Variables. J. Chem. Phys.,
47(2), 597–613.
Comi, C., Corigliano, A. and Maier, G. 1991. Extremum Properties of Finite-step Solutions in
Elastoplasticity with Nonlinear Mixed Hardening. Int. J. Solids Struct., 27(8), 965–981.
Cordebois, J.P. and Sidoroff, F. 1979. Damage Induced Elastic Anisotropy. Pages 761–774 of: Euromech
115 – Villard de Lans – Juin 1979.
768 REFERENCES
Cordebois, J.P. and Sidoroff, F. 1982. Endomagement Anisotrope en Élasticité et Plasticité. Journal de
Mécanique Théorique et Appliquée, 45–60. Numéro spécial.
Cormeau, I.C. 1975. Numerical Stability in Quasi-Static Elasto/Visco-Plasticity. Int. J. Num. Meth.
Engng, 9, 109–127.
Cosserrat, E. and Cosserrat, F. 1909. Théorie des corps déformables. Paris: Hermann.
Cox, A.D., Eason, G. and Hopkins, H.G. 1961. Axially Symmetric Plastic Deformation in Soils. Trans.
R. Soc. Lond. A, 254, 1–45.
Crisfield, M.A. 1981. A Fast Incremental/Iterative Solution Procedure that Handles ‘Snap-Through’.
Comp. Structs, 13, 55–62.
Crisfield, M.A. 1983. An Arc-Length Method Including Line Searches and Accelerations. Int. J. Num.
Meth. Engng, 19, 1269–1289.
Crisfield, M.A. 1987. Plasticity Computations Using the Mohr–Coulomb Yield Criterion. Engng.
Comp., 4, 300–308.
Crisfield, M.A. 1991. Non-linear Finite Element Analysis of Solids and Structures. Volume 1: Essentials.
Chichester: John Wiley & Sons.
Crisfield, M.A. 1997. Non-linear Finite Element Analysis of Solids and Structures. Volume 2: Advanced
Topics. Chichester: John Wiley & Sons.
Crisfield, M.A. and Moita, G.F. 1996. A Co-rotational Formulation for 2-D Continua Including
Incompatible Modes. Int. J. Numer. Meth. Engng, 39, 2619–2633.
Crisfield, M.A., Moita, G.F., Jelenić, G. and Lyons, L.P.R. 1995. Enhanced Lower-Order Element
Formulations for Large Strains. Pages 293–320 of: Owen, D.R.J., Oñate, E. and Hinton, E. (eds),
Computational Plasticity: Fundamentals and Applications – Proceedings of the Fourth International
Conference held in Barcelona, 3–6 April 1995. Swansea: Pineridge Press.
Cuitiño, A. and Ortiz, M. 1992a. A Material-independent Method for Extending Stress Update
Algorithms from Small-strain Plasticity to Finite Plasticity with Multiplicative Kinematics. Engng.
Comp., 9, 437–451.
Cuitiño, A.M. and Ortiz, M. 1992b. Computational Modelling of Single Crystals. Modelling Simul.
Mater. Sci. Engng, 1, 225–263.
Dafalias, Y.F. 1984. The Plastic Spin Concept and a Simple Illustration of its Role in Finite Plastic
Transformations. Mech. Mater., 3, 223–233.
Dafalias, Y.F. 1985. The Plastic Spin. J. Appl. Mech., 52, 865–871.
Dashner, P.A. 1986. Invariance Considerations in Large Strain Elasto-Plasticity. J. Appl. Mech., 53,
55–60.
de Borst, R. 1987. Integration of Plasticity Equations for Singular Yield Functions. Comp. Structs, 26,
823–829.
de Borst, R. 1991. The Zero-normal-stress Condition in Plane-stress and Shell Elastoplasticity. Comm.
Appl. Numer. Meth., 7, 29–33.
de Borst, R. and Feenstra, P.H. 1990. Studies in Anisotropic Plasticity with Reference to the Hill
Criterion. Int. J. Numer. Meth. Engng, 29, 315–336.
de Souza Neto, E.A., Andrade Pires, F.M. and Owen, D.R.J. 2005. F-bar-based Linear Triangles
and Tetrahedra for Finite Strain Analysis of Nearly Incompressible Solids. Part I: Formulation and
Benchmarking. Int. J. Numer. Meth. Engng, 62(3), 353–383.
de Souza Neto, E.A. and Feng, Y.T. 1999. On the Determination of the Path Direction for Arc-length
Methods in the Presence of Bifurcations and ‘Snap-Backs’. Comp. Meth. Appl. Mech. Engng, 179,
81–89.
REFERENCES 769
de Souza Neto, E.A. and Perić, D. 1996. A Computational Framework for a Class of Models for Fully
Coupled Elastoplastic Damage at Finite Strains with Reference to the Linearization Aspects. Comp.
Meth. Appl. Mech. Engng, 130, 179–193.
de Souza Neto, E.A., Perić, D. and Owen, D.R.J. 1992. A Computational Model for Ductile Damage
at Finite Strains. Pages 1425–1441 of: Owen, D.R.J., Oñate, E. and Hinton, E. (eds), Computational
Plasticity: Fundamentals and Applications – Proceedings of the Third International Conference held in
Barcelona, 6–10 April 1992. Swansea: Pineridge Press.
de Souza Neto, E.A., Perić, D. and Owen, D.R.J. 1993. Some Aspects of Formulation and Implementa-
tion of Ductile Damage at Finite Strains. In: Proceedings of the ACME 93 Conference: Computational
Mechanics in the UK.
de Souza Neto, E.A., Perić, D. and Owen, D.R.J. 1994a. A Model for Elasto-plastic Damage at Finite
Strains: Computational Issues and Applications. Engng. Comp., 11(3), 257–281.
de Souza Neto, E.A., Perić, D. and Owen, D.R.J. 1994b. A Phenomenological Three-dimensional Rate-
independent Continuum Damage Model for Highly Filled Polymers: Formulation and Computational
Aspects. J. Mech. Phys. Solids, 42(10), 1533–1550.
de Souza Neto, E.A., Perić, D. and Owen, D.R.J. 1998. Continuum Modelling and Numerical
Simulation of Material Damage at Finite Strains. Arch. Comput. Meth. Engng, 5, 311–384.
de Souza Neto, E.A., Perić, D., Dutko, M. and Owen, D.R.J. 1996. Design of Simple Low Order Finite
Elements for Large Strain Analysis of Nearly Incompressible Solids. Int. J. Solids Structs, 33, 3277–
3296.
Desmorat, R. 2000. Dissymétrie de comportement élastique anisotrope couplé ou non à
l’endomagement. C. R. Acad. Sci. Paris, Série II b, 328, 445–450.
Dettmer, W. and Reese, S. 2004. On the Theoretical and Numerical Modelling of Armstrong–Friederick
Kinematic Hardening in the Finite Strain Regime. Comp. Meth. Appl. Mech. Engng, 192, 87–116.
Dienes, J.K. 1979. On the Analysis of Rotation and Stress Rates in Deforming Bodies. Acta Mech., 32,
217–232.
Dieter, G. 1986. Mechanical Metallurgy. McGraw-Hill.
Dodds, Jr., R.H. 1987. Numerical Techniques for Plasticity Computations in Finite Element Analysis.
Comput. Structs, 26(5), 767–779.
Doghri, I. 1995. Numerical Implementation and Analysis of a Class of Metal Plasticity Models Coupled
with Ductile Damage. Int. J. Numer. Meth. Engng, 38, 3403–3431.
Doll, S., Hauptmann, R., Schweizerhof, K. and Freischlager, C. 2000. On Volumetric Locking of Low-
order Solid and Solid-shell Elements for Finite Elastoviscoplastic Deformations and Selective Reduced
Integration. Engng. Comp., 17, 874–902.
Drucker, D.C. and Prager, W. 1952. Soil Mechanics and Plasticity Analysis of Limit Design. Quart. J.
Appl. Math., 10, 157–162.
Dureisseix, D., Ladevèze, P. and Schrefler, B.A. 2003. A LATIN Computational Strategy for Multi-
physics Problems: Application to Poroelasticity. Int. J. Numer. Meth. Engng, 56(10), 1489–1510.
Dutko, M., Perić, D. and Owen, D.R.J. 1993. Universal Anisotropic Yield Criterion Based on
Superquadric Functional Representation. Part 1: Algorithmic Issues and Accuracy Analysis. Comp.
Meth. Appl. Mech. Engng, 109, 73–93.
Duvaut, G. and Lions, J.L. 1976. Inequalities in Mechanics and Physics. Springer-Verlag: Berlin.
Engel, L. and Klingele, H. 1981. An Atlas of Metal Damage. Wolfe Science Books.
770 REFERENCES
Eterovic, A.L. and Bathe, K.-J. 1990. A Hyperelastic Based Large Strain Elasto-plastic Constitutive
Formulation with Combined Isotropic-kinematic Hardening Using the Logarithmic Stress and Strain
Measures. Int. J. Numer. Meth. Engng, 30, 1099–1114.
Eve, R.A., Reddy, B.D. and Rockafellar, R.T. 1990. An Internal Variable Theory of Elastoplasticity
based on the Maximum Plastic Work Inequality. Quart. Appl. Math., 48, 59–83.
Feijóo, R.A. and Zouain, N. 1988. Formulations in Rates and Increments for Elastic-plastic Analysis.
Int. J. Numer. Meth. Engng, 26, 2031–2048.
Feng, Y.T., Owen, D.R.J. and Perić, D. 1997. On the Sign of the Determinant of the Structural Stiffness
Matrix for Determination of Loading Increment in Arc-length Methods. Comm. Numer. Meth. Engng,
13, 47–49.
Feng, Y.T., Perić, D. and Owen, D.R.J. 1995. Determination of Travel Directions in Path-following
Methods. Math. and Comput. Modelling, 21, 43–59.
Feng, Y.T., Perić, D. and Owen, D.R.J. 1996. A New Criterion for Determination of Initial Loading
Parameter in Arc-length Methods. Comput. Structs, 58, 479–485.
Flanagan, D.P. and Taylor, L.M. 1987. An Accurate Numerical Algorithm for Stress Integration with
Finite Rotations. Comp. Meth. Appl. Mech. Engng, 62, 305–320.
Fletcher, R. 1980. Practical Methods of Optimization. Volume 1: Unconstrained Problems, and Volume
2: Constrained Problems. New York: Wiley.
Fonseka, G.U. and Krajčinović, D. 1981. The Continuous Damage Theory of Brittle Materials – Part 2:
Uniaxial and Plane Response Modes. J. Appl. Mech., 48, 816–824.
Fuschi, P., Perić, D. and Owen, D.R.J. 1992. Studies on Generalized Midpoint Integration in Rate-
Independent Plasticity with Reference to Plane Stress J2 -Flow Theory. Comput. Structs, 43, 1117–
1133.
Gear, C.W. 1971. Numerical Initial Value Problems in Ordinary Differential Equations. Englewood
Cliffs, New Jersey: Prentice-Hall.
Gelin, J.C. and Mrichcha, A. 1992. Computational Procedures for Finite Strain Elasto Plasticity with
Isotropic Damage. Pages 1401–1412 of: Owen, D.R.J., Oñate, E. and Hinton, E. (eds), Computational
Plasticity: Fundamentals and Applications – Proceedings of the Third International Conference held in
Barcelona, 6–10 April 1992. Swansea: Pineridge Press.
Gethin, D.T., Lewis, R.W. and Ransing, R.S. 1998. Compaction of Powder via a Deformable Discrete
Element Approach. Pages 45–50 of: Proceedings of the 1998 Powder Metallurgy World Congress and
Exhibition. Shrewsbury, UK: European Powder Metallurgy Association.
Glaser, S. and Armero, F. 1997. On the Formulation of Enhanced Strain Finite Element Methods in
Finite Deformations. Engng. Comp., 14(7), 759–791.
Govindjee, S. and Simo, J. 1991. A Micro-mechanically Based Continuum Damage Model for Carbon
Black-filled Rubbers Incorporating Mullins’ Effect. J. Mech. Phys. Solids, 39(1), 87–112.
Govindjee, S. and Simo, J.C. 1992. Mullins’ Effect and the Strain Amplitude Dependence of the Storage
Modulus. Int. J. Solids Structs, 20, 1737–1751.
Green, A.E. and Naghdi, P.M. 1965. A General Theory of an Elasto-plastic Continuum. Arch. Rat.
Mech. Anal., 18, 251–281.
Green, A.E. and Zerna, W. 1954. Theoretical Elasticity. Oxford University Press.
Green, A.P. 1953. The Plastic Yielding of Notched Bars Due to Bending. Quart. J. Mech. Appl. Math.,
6, 223–239.
Green, A.P. 1954. A Theory of the Plastic Yielding Due to Bending of Cantilevers and Fixed-ended
Beams. Part I. J. Mech. Phys. Solids, 3, 1–15.
REFERENCES 771
Gurson, A.L. 1977. Continuum Theory of Ductile Rupture by Void Nucleation and Growth – Part I:
Yield Criteria and Flow Rule for Porous Media. J. Engng. Mater. Tech., 99, 2–15.
Gurtin, M.E. 1972. The Linear Theory of Elasticity. Pages 1–295 of: Flügge, S. and Truesdell, C. (eds),
Handbuch der Physik, vol. VIa/2. Springer-Verlag.
Gurtin, M.E. 1981. An Introduction to Continuum Mechanics. New York: Academic Press.
Gurtin, M.E. and Francis, E.C. 1981. Simple Rate-Independent Model for Damage. J. Spacecraft, 18(3),
285–286.
Gurtin, M.E. and Martins, L.C. 1976. Cauchy’s Theorem in Classical Physics. Arch. Rat. Mech. Anal.,
60(4), 305–324.
Haftka, R.T., Gürdal, Z. and Kamat, M.P. 1990. Elements of Structural Optimization. 2nd edn. Kluwer
Academic Publishers.
Halphen, B. and Nguyen, Q.S. 1975. Sue les matériaux Standards généralisés. J. de Mécanique, 14,
39–63.
Han, W. and Reddy, B.D. 1999. Plasticity: Mathematical Theory and Numerical Analysis. Berlin:
Springer-Verlag.
Hancock, J.W. and Mackenzie, A.C. 1976. On the Mechanism of Ductile Fracture in High-strength
Steels Subjected to Multi-axial Stress-states. J. Mech. Phys. Solids, 24, 147–169.
Havner, K.S. 1992. Finite Plastic Deformation of Crystalline Solids. Cambridge: Cambridge University
Press.
Hencky, H. 1933. The Elastic Behavior of Vulcanized Rubber. J. Appl. Mech., 1, 45–53.
Hibbitt, H.D., Marcal, P.V. and Rice, J.R. 1970. A Finite Element Formulation for Problems of Large
Strain and Large Displacement. Int. J. Solids Structs, 6, 1069–1086.
Hill, R. 1948. A Theory of the Yielding and Plastic Flow in Anisotropic Metals. Proc. Roy. Soc. A, 193,
281–297.
Hill, R. 1950. The Mathematical Theory of Plasticity. London: Oxford University Press.
Hill, R. 1958. A General Theory of Uniqueness and Stability in Elastic-plastic Solids. J. Mech. Phys.
Solids, 6, 236–249.
Hill, R. 1978. Aspects of Invariance in Solid Mechanics. Adv. Appl. Mech., 18, 1–75.
Hinton, E. and Owen, D.R.J. 1977. Finite Element Programming. New York: Academic Press.
Hirsch, M.W. and Smale, S. 1974. Differential Equations, Dynamical Systems, and Linear Algebra.
New York: Academic Press.
Hoffman, O. 1967. The Brittle Strength of Orthotropic Materials. J. Composite Mater., 1, 200–206.
Holzapfel, G.A. 2000. Nonlinear Solid Mechanics. A Continuum Approach for Engineering. London:
John Wiley & Sons.
Horii, H. and Nemat-Nasser, S. 1983. Overall Moduli of Solids with Microcracks: Load Induced
Anisotropy. J. Mech. Phys. Solids, 31(2), 155–171.
Huerta, A., Pérez-Foguet, A. and Rodrígues-Ferran, A. 1999. Consistent Tangent Matrices for Complex
Problems: Substepping Schemes and Numerical Differentiation. In: Wunderlich, W. (ed), Proceedings
of the ECCM’99 – European Conference on Computational Mechanics, Munich 1999.
Hughes, T.J., Taylor, R.L. and Sackman, J.L. 1975. Finite Element Formulation and Solution of Contact-
impact Problems in Continuum Mechanics – III. Technical Report UC SESM 75-7. Department of Civil
Engineering, University of California, Berkeley.
772 REFERENCES
Hughes, T.J.R. 1980. Generalization of Selective Integration Procedures to Anisotropic and Nonlinear
Media. Int. J. Num. Meth. Engng, 15, 1413–1418.
Hughes, T.J.R. 1984. Numerical Implementation of Constitutive Models: Rate-independent Deviatoric
Plasticity. Pages 29–63 of: Nemat-Nasser et al. (Ed.), Theoretical Foundations for Large-scale Compu-
tations for Nonlinear Material Behavior. Dordrecht, The Netherlands: Martinus Nijhoff.
Hughes, T.J.R. 1987. The Finite Element Method. Linear Static and Dynamic Finite Element Analysis.
Englewood Cliffs, New Jersey: Prentice-Hall.
Hughes, T.J.R. and Pister, K. 1978. Consistent Linearization in Mechanics of Solids and Structures.
Comp. Structs, 8, 391–397.
Hughes, T.J.R. and Taylor, R.L. 1978. Unconditionally Stable Algorithms for Quasi-static Elasto/Visco-
plastic Finite Element Analysis. Comput. Structs, 8, 169–173.
Hughes, T.J.R. and Winget, J. 1980. Finite Rotation Effects in Numerical Integration of Rate Constitu-
tive Equations Arising in Large-deformation Analysis. Int. J. Numer. Meth. Engng, 15, 1862–1867.
Ibrahimbegović, A. 1994. Finite Elastoplastic Deformations of Space-Curved Membranes. Comp. Meth.
Appl. Mech. Engng, 119, 371–394.
Irons, B.M. 1970. A Frontal Solution Program for Finite Element Analysis. Int. J. Numer. Meth. Engng,
2, 5–32.
Janson, J. 1978. A Continuous Damage Approach to the Fatigue Process. Engng. Fract. Mech., 10,
651–657.
Jetteur, Ph. 1986. Implicit Integration Algorithm for Elastoplasticity in Plane Stress Analysis. Engng.
Comp., 3, 251–253.
Jirásek, M. and Bažant, Z.P. 2002. Inelastic Analysis of Structures. Chichester: Wiley.
Johnson, G. and Bammann, D.J. 1984. A Discussion of Stress Rates in Finite Deformation Problems.
Int. J. Solids Structs, 20, 725–737.
Kachanov, L.M. 1958. Time of the Rupture Process under Creep Condition. Izv. Akad. Nauk. SSSR, Otd.
Tekhn. Nauk., 8, 26–31.
Kachanov, L.M. 1977. Creep and Rupture under Complex Loading. Problemi Prochnosti, 6.
Kestin, J. and Bataille, J. 1977. Irreversible Thermodynamics of Continua and Internal Variables. Pages
39–67 of: Proceedings of the International Symposium on Continuum Models of Discrete Systems.
University of Waterloo Press.
Kocks, U.F. 1970. The Relation Between Polycrystal Deformation and Single-crystal Deformation.
Metall. Trans., 1, 1121–1143.
Koh, C.G., Owen, D.R.J. and Perić, D. 1995. Explicit Dynamic Analysis of Elasto-plastic Laminated
Composite Shells: Implementation of Non-iterative Stress Update Schemes for the Hoffman Yield
Criterion. Comput. Mech., 16, 307–314.
Koiter, W.T. 1953. Stress-strain Relations, Uniqueness and Variational Theorems for Elasto-plastic
Materials with Singular Yield Surface. Quart. Appl. Math., 11, 350–354.
Kojić, M. and Bathe, K.-J. 1987. Studies of Finite Element Procedures – Stress Solution of a Closed
Elastic Strain Path with Stretching and Shearing Using the Updated Lagrangian Jaumann Formulation.
Comp. Structs, 26, 175–179.
Kojić, M. and Bathe, K.J. 1987. The ‘Effective-stress-function’ for Thermo-elasto-plasticity and Creep.
Int. J. Numer. Meth. Engng, 24, 1509–1532.
Korelc, J. and Wriggers, P. 1996. Consistent Gradient Formulation for a Stable Enhanced Strain Method
for Large Deformations. Engng. Comp., 13(1), 103–123.
REFERENCES 773
Ladevèze, P. and Lemaitre, J. 1984. Damage Effective Stress in Quasi Unilateral Conditions. In: 16th
Int. Congress Theor. Appl. Mech.
Ladevèze, P. and Perego, U. 2000. Duality Preserving Discetization of the Large Time Increment
Method. Comp. Meth. Appl. Mech. Engng, 189(1), 205–232.
Leckie, F.A. and Hayhurst, D.R. 1974. Creep Rupture of Structures. Proc. Roy. Soc. Lond. A, 340,
323–347.
Leckie, F.A. and Onat, E.T. 1981. Tensorial Nature of Damage Measuring Internal Variables. Pages
140–155 of: Proceedings of the IUTAM Symposium on Physical Nonlinearities in Structures. Springer.
Lee, E.H. 1969. Elastic-plastic Deformation at Finite Strains. J. Appl. Mech., 36, 1–6.
Lee, E.H. and Liu, D.T. 1967. Finite Strain Elastic-plastic Theory with Application to Plane-wave
Analysis. J. Appl. Phys., 38, 19–27.
Lege, D.J., Barlat, F. and Brem, J.C. 1989. Characterization and Modeling of the Mechanical Behaviour
and Formability of a 2008-T4 Sheet Sample. Int. J. Mech. Sci., 31(7), 549–563.
Lemaitre, J. 1983. A Three-dimensional Ductile Damage Model Applied to Deep-drawing Forming
Limits. Pages 1047–1053 of: ICM 4 Stockholm, vol. 2.
Lemaitre, J. 1985a. A Continuous Damage Mechanics Model for Ductile Fracture. J. Engng. Mat. Tech.,
107, 83–89.
Lemaitre, J. 1985b. Coupled Elasto-plasticity and Damage Constitutive Equations. Comp. Meth. Appl.
Mech. Engng, 51, 31–49.
Lemaitre, J. 1987. Formulation and Identification of Damage Kinetic Constitutive Equations. Pages 37–
89 of: Krajčinović, D. and Lemaitre, J. (eds), Continuum Damage Mechanics: Theory and Applications.
Springer-Verlag.
Lemaitre, J. 1996. A Course on Damage Mechanics. 2nd edn. Springer.
Lemaitre, J. and Chaboche, J.L. 1990. Mechanics of Solid Materials. Cambridge University Press.
Lemaitre, J. and Dufailly, J. 1987. Damage Measurements. Engng. Fract. Mech., 28(8), 643–661.
774 REFERENCES
Lemaitre, J., Desmorat, R. and Sauzay, M. 2000. Anisotropic Damage Law of Evolution. Eur. J. Mech.
A/Solids, 19, 187–208.
Li, X. 1995. Large Strain Constitutive Modelling and Computation for Isotropic, Creep Elastoplastic
Damage Solids. Int. J. Numer. Meth. Engng, 38, 841–860.
Lion, A. 2000. Constitutive Modelling in Finite Thermoviscoplasticity: A Physical Approach Based on
Nonlinear Rheological Models. Int. J. Plast., 16, 469–494.
Lubarda, V.A. and Lee, E.H. 1981. A Correct Definition of Elastic and Plastic Deformation and its
Computational Significance. J. Appl. Mech., 48, 35–40.
Lubliner, J. 1990. Plasticity Theory. New York: Macmillan.
Luenberger, D.G. 1973. Introduction to Linear and Nonlinear Programming. Reading, Massachusetts:
Addison-Wesley.
Maier, G. 1970. A Matrix Structural Theory of Piece-wise Linear Plasticity with Interacting Yield
Planes. Meccanica, 6, 55–66.
Mandel, J. 1973. Equations constitutives et directeurs dans les milieux plastiques et viscoplastiques. Int.
J. Solids Structs, 9, 725–740.
Marques, J.M.M.C. 1984. Stress Computation is Elastoplasticity. Engng. Comp., 1, 42–51.
Marques, J.M.M.C. and Owen, D.R.J. 1983. Strain Hardening Representation for Implicit Quasistatic
Elasto-viscoplastic Algorithms. Comput. Structs, 17, 301–304.
Marquis, D. and Lemaitre, J. 1988. Constitutive Equations for the Coupling Between Elasto-plasticity
Damage and Ageing. Rev. Phys. Applic, 23, 615–624.
Marsden, J.E. and Hughes, T.J.R. 1983. Mathematical Foundations of Elasticity. Englewood Cliffs,
New Jersey: Prentice-Hall.
Martin, J.B. 1975. Plasticity: Fundamentals and General Results. Cambridge: MIT Press.
Martin, J.B., Reddy, B.D. and Griffin, T.B. 1987. Applications of Mathematical Programming Concepts
to Incremental Elastic-plastic Analysis. Engng. Structs, 9, 171–176.
Matthies, H. 1979. Existence Theorems in Thermo-plasticity. J. Mécanique, 18(4), 695–712.
Matthies, H. and Strang, G. 1979. The Solution of Nonlinear Finite Element Equations. Int. J. Numer.
Meth. Engng, 14, 1613–1626.
McMeeking, R.M. and Rice, J.R. 1975. Finite-element Formulations for Problems of Large Elastic-
plastic Deformation. Int. J. Solids Structs, 11, 601–616.
Meschke, G. and Liu, W.N. 1999. A Re-formulation of the Exponential Algorithm for Finite Strain
Plasticity in Terms of Cauchy Stresses. Comp. Meth. Appl. Mech. Engng, 173, 167–187.
Miehe, C. 1993. Computation of Isotropic Tensor Functions. Comm. Numer. Meth. Engng, 9, 889–896.
Miehe, C. 1995. Discontinuous and Continuous Damage Evolution in Ogden-type large-strain elastic
materials. Eur. J. Mech. A/Solids, 14(5), 697–720.
Miehe, C. 1996a. Exponential Map Algorithm for Stress Updates in Anisotropic Multiplicative
Elastoplasticity for Single Crystals. Int. J. Numer. Meth. Engng, 39, 3367–3390.
Miehe, C. 1996b. Multisurface Thermoplasticity for Single Crystals at Large Strains in Terms of
Eulerian Vector Updates. Int. J. Solids Structs, 33, 3103–3130.
Miehe, C. 1998a. Comparison of Two Algorithms for the Computation of Fourth-order Isotropic Tensor
Functions. Comp. Structs, 66, 37–43.
Miehe, C. 1998b. A Theoretical and Computational Model for Isotropic Elastoplastic Stress Analysis
in Shells at Large Strains. Comp. Meth. Appl. Mech. Engng, 155, 193–233.
REFERENCES 775
Nemat-Nasser, S. 1979. Decomposition of Strain Measures and their Rates in Finite Deformation
Elastoplasticity. Int. J. Solids Structs, 15, 155–166.
Nemat-Nasser, S. 1982. On Finite Deformation Elasto-plasticity. Int. J. Solids Structs, 18(10), 857–872.
Nemat-Nasser, S. 1999. Averaging Theorems in Finite Deformation Plasticity. Mech. Mater., 31, 493–
523.
Oñate, E., Rojek, J. and García Garino, C. 1995. NUMISTAMP: A Research Project for Assessment of
Finite-element Models for Stamping Processes. J. Mater. Proc. Technol., 50, 17–38.
Oden, J.T. 1972. Finite Elements of Nonlinear Continua. London: McGraw-Hill.
Oden, J.T. 1979. Applied Functional Analysis. Englewood Cliffs, New Jersey: Prentice-Hall.
Ogden, R.W. 1972. Large Deformation Isotropic Elasticity – On the Correlation of Theory and
Experiment for Incompressible Rubberlike Solids. Proc. R. Soc. Lond. A, 326, 565–584.
Ogden, R.W. 1984. Non-Linear Elastic Deformations. Chichester: Ellis Horwood.
Onat, E.T. 1986. Representation of Mechanical Behaviour in the Presence of Internal Damage. Engng.
Fract. Mech., 25, 605–614.
Onat, E.T. and Leckie, F.A. 1988. Representation of Mechanical Behavior in the Presence of Changing
Internal Structure. J. Appl. Mech., 55, 1–10.
Ortiz, M. and Martin, J.B. 1989. Symmetry-preserving Return Mapping Algorithms and Incrementally
Extremal Paths: A Unification of Concepts. Int. J. Numer. Meth. Engng, 28, 1839–1853.
Ortiz, M. and Popov, E.P. 1985. Accuracy and Stability of Integration Algorithms for Elastoplastic
Constitutive Relations. Int. J. Numer. Meth. Engng, 21, 1561–1576.
Ortiz, M. and Simo, J.C. 1986. An Analysis of a New Class of Integration Algorithms for Elastoplastic
Constitutive Relations. Int. J. Numer. Meth. Engng, 23, 353–366.
Owen, D.R.J. and Hinton, E. 1980. Finite Elements in Plasticity: Theory and Practice. Swansea:
Pineridge Press.
Owen, D.R.J., Souza Neto, E.A. de, Zhao, S.Y., Perić, D., and Loughran, J.G. 1998. Finite Element
Simulation of the Rolling and Extrusion of Multi-phase Materials: Application to the Rolling of
Prepared Sugar Cane. Comp. Meth. Appl. Mech. Engng, 151, 479–495.
Pankaj, and Bićanić, N. 1989. On Multivector Stress Returns in Mohr–Coulomb Plasticity. Pages
421–436 of: Owen, D.R.J., Hinton, E. and Oñate, E. (eds), Proceedings of the Second International
Conference on Computational Plasticity: Models, Software and Applications. Swansea: Pineridge Press.
Papadopoulos, P. and Taylor, R.L. 1994. On the Application of Multi-step Integration Methods to
Infinitesimal Elastoplasticity. Int. J. Numer. Meth. Engng, 37, 3169–3184.
Parisch, H. 1986. Efficient Non-linear Finite Element Shell Formulation Involving Large Strains.
Engng. Comp., 3, 121–128.
Peirce, D. 1983. Shear Band Bifurcation in Ductile Single Crystals. J. Mech. Phys. Solids, 31, 133–153.
Peirce, D., Asaro, R.J. and Needleman, A. 1982. An Analysis of Nonuniform and Localized Deforma-
tion in Ductile Single Crystals. Acta Metall., 30, 1087–1119.
Peirce, D., Asaro, R.J. and Needleman, A. 1983. Material Rate Dependence and Localized Deformation
in Crystalline Solids. Acta Metall., 31, 1951–1976.
Peirce, D., Shih, C.F. and Needleman, A. 1984. A Tangent Modulus Method for Rate Dependent Solids.
Comput. Sructs, 18, 875–887.
Pérez-Foguet, A. and Armero, F. 2002. On the Formulation of Closest-point Projection Algorithms in
Elastoplasticity – Part II: Globally Convergent Schemes. Int. J. Numer. Meth. Engng, 53, 331–374.
REFERENCES 777
Perić, D. 1992. On Consistent Stress Rates in Solid Mechanics: Computational Implications. Int. J.
Numer. Meth. Engng, 33, 799–817.
Perić, D. 1993. On a Class of Constitutive Equations in Viscoplasticity: Formulation and Computational
Issues. Int. J. Numer. Meth. Engng, 36, 1365–1393.
Perić, D. and de Souza Neto, E.A. 1999. A New Computational Model for Tresca Plasticity at Finite
Strains with an Optimal Parametrization in the Principal Space. Comp. Meth. Appl. Mech. Engng, 171,
463–489.
Perić, D. and Dettmer, W. 2003. A Computational Model for Generalized Inelastic Materials at Finite
Strains Combining Elastic, Viscoelastic and Plastic Material Behaviour. Engng. Comput., 20, 768–787.
Perić, D. and Owen, D.R.J. 1991. A Model for Large Deformation of Elasto-viscoplastic Solids at
Finite Strains: Computational Issues. Pages 299–312 of: Besdo, D. and Stein, E. (eds), Proceedings of
the IUTAM Symposium on Finite Inelastic Deformations – Theory and Applications. Berlin: Springer.
Perić, D. and Owen, D.R.J. 1992. Computational Model for 3-D Contact Problems with Friction Based
on the Penalty Method. Int. J. Numer. Meth. Engng, 35, 1289–1309.
Perić, D., Owen, D.R.J. and Honnor, M.E. 1992. A Model for Finite Strain Elasto-plasticity Based on
Logarithmic Strains: Computational Issues. Comp. Meth. Appl. Mech Engng, 94, 35–61.
Perić, D., Yu, J. and Owen, D.R.J. 1994. On Error Estimates and Adaptivity in Elastoplastic Solids:
Application to the Numerical Simulation of Strain Localization in Classical and Cosserat Continua. Int.
J. Numer. Meth. Engng, 37, 1351–1379.
Perzyna, P. 1963. The Constitutive Equations for Rate Sensitive Plastic Materials. Quart. Appl. Math.,
20, 321–332.
Perzyna, P. 1966. Fundamental Problems in Viscoplasticity. Advances in Applied Mechanics, vol. 9.
New York: Academic Press. Pages 243–377.
Perzyna, P. 1971. Thermodynamic Theory of Viscoplasticity. Advances in Applied Mechanics, vol. 11.
New York: Academic Press. Pages 313–354.
Pinsky, P.M., Ortiz, M. and Pister, K.S. 1983. Numerical Integration of Rate Constitutive Equations in
Finite Deformation Analysis. Comp. Meth. Appl. Mech. Engng, 40, 137–158.
Pires, F.M.A., de Souza Neto, E.A. and Owen, D.R.J. 2004. On the Finite Element Prediction of Damage
Growth and Fracture Initiation in Finitely Deforming Ductile Materials. Comp. Meth. Appl. Mech.
Engng, 193, 5223–5256.
Prager, W. 1959. An Introduction to Plasticity. London: Addison-Wesley.
Pramono, E. and Willam, K. 1989. Implicit Integration of Composite Yield Surfaces with Corners.
Engng. Comp., 6, 186–197.
Rabotnov, Y.N. 1963. On the Equations of State for Creep. Page 307 of: Progress in Applied Mechanics,
Prager Anniversary Volume. New York: Macmillan.
Ramm, E. 1981. Strategies for Tracing the Nonlinear Response Near Limit Point. Pages 63–89 of:
Wunderlich, W. Stein, E. and Bathe, K.-J. (eds), Nonlinear Finite Element Analysis in Structural
Mechanics. Berlin: Springer-Verlag.
Ramm, E. and Matzenmiller, A. 1987. Computational Aspects of Elasto-plasticity in Shell Analysis.
Pages 711–734 of: Owen, D.R.J., Hinton, E. and Oñate, E. (eds), Computational Plasticity: Models,
Software and Applications – Proceedings of the International Conference held in Barcelona, 6–10 April
1987. Swansea: Pineridge Press.
Ransing, R.S., Lewis, R.W., Gethin, D.T., Crook, A.J.L. and Dutko, M. 1998. Compaction of Granular
Materials using Deformable Discrete Elements. Pages 143–147 of: Huétink, J. and Baaijens, F.P.T.
(eds), Proceedings of the Sixth International Conference on Numerical Methods in Industrial Forming
778 REFERENCES
Schönauer, M., de Souza Neto, E.A. and Owen, D.R.J. 1995. Hencky Tensor Based Enhanced Large
Strain Element for Elasto-plastic Analysis. Pages 333–348 of: Owen, D.R.J., Oñate, E. and Hinton,
E. (eds), Computational Plasticity: Fundamentals and Applications – Proceedings of the Fourth
International Conference held in Barcelona, 3–6 April 1995. Swansea: Pineridge Press.
Schröder, J., Gruttmann, F. and Löblein, J. 2002. A Simple Orthotropic Finite Elasto-plasticity Model
based on Generalized Stress–strain Measures. Comp. Mech., 30, 48–64.
Schweizerhof, K. and Ramm, E. 1984. Displacement Dependent Pressure Loads in Nonlinear Finite
Element Analyses. Comput. Structs, 18, 1099–1114.
Seth, B.R. 1964. Generalized Strain Measure with Applications to Physical Problems. Pages 162–172
of: Reiner, M. and Abir, D. (eds), Second-order Effects in Elasticity, Plasticity and Fluid Dynamics.
Oxford: Pergamon Press.
Simo, J.C. 1985. On the Computational Significance of the Intermediate Configuration and Hyperelastic
Stress Relations in Finite Deformation Elastoplasticity. Mech. Mater., 4, 439–451.
Simo, J.C. 1987. On a Fully Three-dimensional Finite-strain Viscoelastic Damage Model: Formulation
and Computational Aspects. Comp. Meth. Appl. Mech. Engng, 60, 153–173.
Simo, J.C. 1992. Algorithms for Static and Dynamic Multiplicative Plasticity that Preserve the Classical
Return Mapping Schemes of the Infinitesimal Theory. Comp. Meth. Appl. Mech. Engng, 99, 61–112.
Simo, J.C. 1998. Numerical Analysis and Simulation of Plasticity. Pages 183–499 of: Ciarlet, P.G. and
Lions, J.L. (eds), Handbook of Numerical Analysis, Volume VI. Amsterdam: Elsevier.
Simo, J.C. and Armero, F. 1992. Geometrically Non-linear Enhanced Strain Mixed Methods and the
Method of Incompatible Modes. Int. J. Numer. Meth. Engng, 33, 1413–1449.
Simo, J.C., Armero, F. and Taylor, R.L. 1993. Improved Versions of Assumed Enhanced Strain Tri-
linear Elements for 3D Finite Deformation Problems. Comp. Meth. Appl. Mech. Engng, 110, 359–386.
Simo, J.C. and Govindjee, S. 1991. Non-linear B-stability and Symmetry Preserving Return Mapping
Algorithms for Plasticity and Viscoplasticity. Int. J. Numer. Meth. Engng, 31, 151–176.
Simo, J.C. and Hughes, T.J.R. 1986. On the Variational Foundations of Assumed Strain Methods.
J. Appl. Mech., 53, 1685–1695.
Simo, J.C. and Hughes, T.R.J. 1987. General Return Mapping Algorithms for Rate-independent
Plasticity. Pages 221–231 of: Desai, C.S. et al. (Ed.), Constitutive Laws for Engineering Materials:
Theory and Applications. Elsevier.
Simo, J.C. and Hughes, T.J.R. 1998. Computational Inelasticity. New York: Springer-Verlag.
Simo, J.C. and Ju, J.W. 1987. Strain- and Stress-based Continuum Damage Models – I. Formulation
and II. Computational Aspects. Int. J. Solids Structs, 23, 821–869.
Simo, J.C., Ju, J.-W., Pister, K.S. and Taylor, R.L. 1988a. Assessment of Cap Model: Consistent Return
Algorithms and Rate-dependent Extension. ASCE J. Engng. Mech., 114(2), 191–218.
Simo, J.C., Kennedy, J.G. and Govindjee, S. 1988b. Non-smooth Multisurface Plasticity and Viscoplas-
ticity. Loading/Unloading Conditions and Numerical Algorithms. Int. J. Numer. Meth. Engng, 26, 2161–
2185.
Simo, J.C. and Meschke, G. 1993. A New Class of Algorithms for Classical Plasticity Extended to
Finite Strains. Application to Geomaterials. Comp. Mech., 11, 253–278.
Simo, J.C. and Miehe, C. 1992. Associative Coupled Thermoplasticity at Finite Strains: Formulation,
Numerical Analysis and Implementation. Comp. Meth. Appl. Mech. Engng, 98, 41–104.
Simo, J.C. and Ortiz, M. 1985. A Unified Approach to Finite Deformation Plasticity Based on the Use
of Hyperelastic Constitutive Equations. Comp. Meth. Appl. Mech. Engng, 49, 221–245.
780 REFERENCES
Simo, J.C. and Pister, K.S. 1984. Remarks on the Rate Constitutive Equations for Finite Deformation
Problems: Computational Implications. Comp. Meth. Appl. Mech. Engng, 46, 201–215.
Simo, J.C. and Rifai, S. 1990. A Class of Mixed Assumed Strain Methods and the Method of
Incompatible Modes. Int. J. Numer. Meth. Engng, 29, 1595–1638.
Simo, J.C. and Taylor, R.L. 1985. Consistent Tangent Operators for Rate-Independent Elastoplasticity.
Comp. Meth. Appl. Mech. Engng, 48, 101–118.
Simo, J.C. and Taylor, R.L. 1986. A Return Mapping Algorithm for Plane Stress Elastoplasticity. Int. J.
Numer. Meth. Engng, 22, 649–670.
Simo, J.C. and Taylor, R.L. 1991. Quasi-incompressible Finite Elasticity in Principal Stretches.
Continuum Basis and Numerical Algorithms. Comp. Meth. Appl. Mech. Engng, 85, 273–310.
Simo, J.C., Taylor, R.L. and Pister, K.S. 1985. Variational and Projection Methods for the Volume
Constraint in Finite Deformation Elasto-plasticity. Comp. Meth. Appl. Mech. Engng, 51, 177–208.
Skrzypek, J.J. 1993. Plasticity and Creep. Theory, Examples and Problems. London: CRC Press.
Sloan, S.W. and Booker, J.R. 1992. Integration of Tresca and Mohr–Coulomb Constitutive Relations in
Plane Strain Elastoplasticity. Int. J. Numer. Meth. Engng, 33, 163–196.
Sowerby, R. and Chu, E. 1984. Rotations, Stress Rates and Strain Measures in Homogenous Deforma-
tion Processes. Int. J. Solids Structs, 20, 1037–1048.
Spencer, A.J.M. 1980. Continuum Mechanics. London: Longman.
Steinmann, P. and Stein, E. 1996. On the Numerical Treatment and Analysis of Finite Deformation
Ductile Single Crystal Plasticity. Comp. Meth. Appl. Mech. Engng, 129, 235–254.
Steinmann, P., Miehe, C. and Stein, E. 1994. Comparison of Different Finite Deformation Inelastic
Damage Models within Multiplicative Elastoplasticity for Ductile Metals. Comput. Mech., 13, 458–
474.
Stern, H.J. 1967. Rubber: Natural and Synthetic. London: McLaren and Sons.
Suquet, P. 1981. Sur les équations de la plasticité: existence et régularité des solutions. J. Mécanique,
20, 3–40.
Sussman, T. and Bathe, K.-J. 1987. A Finite Element Formulation for Nonlinear Incompressible Elastic
and Inelastic Analysis. Comp. Structs, 26, 357–409.
Taylor, R.L., Beresford, P.J. and Wilson, E.L. 1976. A Non-conforming Element for Stress Analysis.
Int. J. Numer. Meth. Engng, 10, 1211–1219.
Toupin, R.A. 1962. Elastic Materials with Couple-Stresses. Arch. Rat. Mech. Anal., 11, 385–414.
Treloar, L.R.G. 1944. Stress–strain Data for Vulcanized Rubber under Various Types of Deformation.
Trans. Faraday Soc., 40, 59–70.
Tresca, H. 1868. Mémoire sur l’écoulement des corps solides. Mém. pres. par div. sav., 18, 733–799.
Truesdell, C. 1969. Rational Thermodynamics. New York: McGraw-Hill.
Truesdell, C. and Noll, W. 1965. The Non-linear Field Theories of Mechanics. In: Flügge, S. (ed),
Handbuch der Physik, vol. III/3. Springer-Verlag.
Tsakmakis, C. 1996a. Kinematic Hardening Rules for Finite Plasticity. Part I: A Constitutive Approach.
Cont. Mech. Thermodyn., 8, 215–231.
Tsakmakis, C. 1996b. Kinematic Hardening Rules for Finite Plasticity. Part II: Some Examples. Cont.
Mech. Thermodyn., 8, 233–246.
Tvergaard, V. 1981. Influence of Voids on Shear Band Instabilities under Plane Strain Conditions. Int.
J. Fract., 17, 389–407.
REFERENCES 781
Tvergaard, V. 1982a. Material Failure by Void Coalescence in Localized Shear Bands. Int. J. Solids
Structs, 18, 659–672.
Tvergaard, V. 1982b. On Localization in Ductile Materials Containing Spherical Voids. Int. J. Fract.,
18, 237–252.
Tvergaard, V. and Needleman, A. 1984. Analysis of the Cup-cone Fracture in a Round Tensile Bar. Acta
Metall., 32, 157–169.
Vaz Jr., M. 1998. Computational Approaches to Simulation of Metal Cutting Processes. PhD thesis,
Department of Civil Engineering, University of Wales Swansea.
Vaz Jr., M. and Owen, D.R.J. 2001. Aspects of Ductile Fracture and Adaptive mesh Refinement in
Damaged Elasto-plastic Materials. Int. J. Numer. Meth. Engng, 50, 29–54.
von Mises, R. 1913. Mechanik der Festen Körper im Plastische-deformablen Zustand. Nachr. d.
Gesellsch d. Wissensch. zu Göttingen, Math.-phys. Klasse.
Wall, W.A., Bischoff, M. and Ramm, E. 2000. A Deformation Dependent Stabilization Technique
Exemplified by EAS Elements at Large Strains. Comp. Meth. Appl. Mech. Engng, 188, 859–871.
Wang, C.C. and Truesdell, C. 1973. Introduction to Rational Elasticity. Leyden: Noordhoff.
Washizu, K. 1968. Variational Methods in Elasticity and Plasticity. London: Pergamon Press.
Weber, G. and Anand, L. 1990. Finite Deformation Constitutive Equations and a Time Integration
Procedure for Isotropic, Hyperelastic-Viscoplastic Solids. Comp. Meth. Appl. Mech. Engng, 79, 173–
202.
Wempner, G.A. 1971. Discrete Approximations Related to the Nonlinear Theory of Solids. Int. J. Solids
Structs, 7, 1581–1599.
Wilkins, M.L. 1964. Calculation of Elasto-plastic Flow. In: Alder, B. et al. (Ed.), Methods of
Computational Physics, Vol. 3. New York: Academic Press.
Wolfram, S. 1991. Mathematica: A System for Doing Mathematics by Computer. New York: Addison-
Wesley.
Wriggers, P. 2001. Nichtlineare Finite-Element-Methoden. Berlin: Springer-Verlag.
Wriggers, P. and Reese, S. 1996. A Note on Enhanced Strain Methods for Large Deformations. Comp.
Meth. Appl. Mech. Engng, 135, 201–209.
Wriggers, P. and Taylor, R.L. 1990. A Fully Non-linear Axisymmetrical Membrane Element for
Rubber-like Materials. Engng. Comp., 7, 303–310.
Zienkiewicz, O.C. and Cormeau, I.C. 1974. Visco-plasticity – Plasticity and Creep in Elastic Solids - A
Unified Numerical Solution Approach. Int. J. Numer. Meth. Engng, 8, 821–845.
Zienkiewicz, O.C. and Taylor, R.L. 2000. The Finite Element Method. Volume II: Solid Mechanics. 5th
edn. Butterworth–Heinemann.
Zienkiewicz, O.C., Norris, V.A., Winnicki, L.A., Naylor, D.J. and Lewis, R.W. 1978. A Unified
Approach to the Soil Mechanics Problems of Offshore Foundations. In: Zienkiewicz, O.C., Lewis,
R.W. and Stagg, K.G. (eds), Numerical Methods in Offshore Engineering. John Wiley and Sons.
Zouain, N., Borges, L.A. and Hecke, M.B. 1988. A Force Method for Elastic-plastic Analysis of Frames
by Quadratic Optimization. Int. J. Solids Structs, 24, 211–221.
Zouain, N., Hecke, M.B. and Feijóo, R.A. 1992. Elastic-plastic Analysis with Potential Functionals
Using Internal Variables. Pages 163–174 of: Owen, D.R.J., Oñate, E. and Hinton, E. (eds), Computa-
tional Plasticity: Fundamentals and Applications – Proceedings of the Third International Conference
held in Barcelona, 6–10 April 1992. Swansea: Pineridge Press.
INDEX
Computational Methods for Plasticity: Theory and Applications EA de Souza Neto, D Perić and DRJ Owen
c 2008 John Wiley & Sons, Ltd
784 INDEX
MATICT, 134, 135, 235, 237, 569, 601 Incremental constitutive function, 95, 102,
MATIOR, 135 127, 133, 192, 229, 598
MATIRD, 132, 135 for elastoplasticity, 230
MATISU, 100, 128, 133, 135, 224, 227, for the viscoplastic von Mises-based
596 model, 464
MATISW, 120, 134, 135 for the von Mises model, 220, 223, 233,
ORDAMA, 486 260
ORVM, 135 Incremental constitutive problem
OUTPUT, 125 of finite strain multiplicative plasticity,
PLFUN, 228 592
RDDAMA, 486 of infinitesimal elastoplasticity, 194
RDDP, 337 Incremental displacement vector, 98
RDMC, 315 Incremental finite element equilibrium
RDOGD, 541 equations
RDTR, 283 at finite strains, 103
RDVM, 227 infinitesimal, 96
RSQ4, 129 Incremental objectivity, 625
RST3, 129 Incremental plastic multiplier, 195
RSTART, 118, 125 Incremental potential, 243
SFQ4, 130 Indicator function, 452
SFT3, 130 Infinitesimal deformation, 57
SHPFUN, 130 Infinitesimal strain tensor, 57
SPDEC2, 731, 735 Initial boundary value problem, 79
STFBA2, 129, 656 infinitesimal, 81
STSTD2, 100, 128, 129, 656 material, 80
SUDAMA, 486, 490 spatial, 79
SUDMEL, 509 Initial stiffness method, 99
SUDP, 324, 329–334, 337, 340, 342 Initial yield stress, 182
SUDPPN, 363 Inner product of tensors, 22
SUFAIL, 227 Inner product of vectors, 17
SUMC, 295, 303–310, 315, 316, 318, 319, Intermediate configuration, see Plastic
324, 334, 337 intermediate configuration
SUOGD, 528, 531, 536, 538, 541, 542, Internal force vector, 88, 192
567 Internal variables, 72
SUPDSC, 708 Interpolation function, see Shape function
SUPGD, 538 Interpolation matrix, 87
SUTR, 128, 273, 274, 276–279, 282, 283, Invertible tensor, 23
291, 294, 310, 315 Iso-error map, 214, 215
SUVM, 133, 224, 235, 261, 310, 364, 567, implicit Drucker–Prager model
569, 596 implementation, 337
SUVMMX, 257, 261 implicit Lemaitre model
SUVMPS, 133, 376, 378, 379 implementation, 483
SWDAMA, 486 implicit Mohr–Coulomb model
SWITCH, 124, 125 implementation, 315
SWOGD, 134 implicit Tresca model implementation,
SWVM, 134 283
UPCONF, 100, 109 viscoplastic von Mises-based model
implementation, 463
Identity tensor Isochoric deformation, 48
of fourth-order, 31 Isoparametric finite element, 90
symmetric, 31 Isotropic hardening, 178, 448
of second order, 19 Isotropic scalar function, 731
Incremental boundary value problem Isotropic solid, 71
at finite strains, 103 Isotropic tensor, 30
infinitesimal, 95 Isotropic tensor function, 287, 316, 733
788 INDEX