STRUCTURAL IMPACT LABORATORY, NTNU

Materials Mechanics
Part I
Odd Sture Hopperstad & Tore Børvik
1/4/2017
Contents

Chapter 1 Introduction

Chapter 2 Constitutive equations in one dimension

2.1 Governing equations 2-1

2.2 Elastic materials 2-4
2.3 Thermoelastic materials 2-6
2.4 Linear viscoelastic materials 2-8
2.4.1 Rheological models 2-8
2.4.2 Maxwell model 2-9
2.4.3 Kelvin-Voigt model 2-11
2.4.4 Standard model of viscoelasticity 2-11
2.4.5 Numerical solution 2-15
2.5 Elastic-plastic materials 2-17
2.5.1 Elastic-perfectly plastic materials 2-19
2.5.2 Elastic-plastic materials with isotropic hardening 2-24
2.5.3 Numerical solution 2-28
2.6 Elastic-viscoplastic materials 2-31
2.6.1 Elastic-viscoplastic material with isotropic hardening 2-33
2.6.2 Numerical solution 2-39
2.7 Kinematic hardening 2-40
2.7.1 Plasticity and viscoplasticity 2-41
2.7.2 Numerical solution 2-45
2.8 Large strains 2-47
2.9 References 2-52

I
Chapter 3 Vectors and tensors

3.1 Coordinate transformation 3-1

3.2 Linear operators (Tensors of 2nd order) 3-5
3.3 Tensors 3-10
3.4 Tensor fields 3-13
3.5 Isotropic functions 3-16
3.6 References 3-18

Chapter 4 Continuum mechanics

4.1 Displacement, strain and rotation 4-1

4.2 Stress and static equilibrium 4-8
4.3 Balance of mechanical energy 4-15
4.4 Principle of virtual work 4-18
4.5 Finite element method for material nonlinearity 4-20
4.5.1 Non-linear equilibrium equations 4-20
4.5.2 Solution method 4-25
4.6 References 4-31

Chapter 5 Elastic materials

5.1 Generalized Hooke’s Law 5-1

5.2 Strain energy function 5-2
5.3 Isotropic elasticity 5-4
5.4 Thermoelasticity 5-8
5.5 Anisotropic elasticity 5-9
5.5.1 Plane of material symmetry 5-9
5.5.2 Orthotropic symmetry 5-10
5.5.3 Isotropy 5-12
5.6 References 5-14
II
Chapter 6 Viscoelastic materials

6.1 Maxwell model 6-1

6.2 Kelvin-Voigt model 6-4
6.3 The standard linear solid 6-6
6.4 Maxwell-Wichert model 6-8
6.5 Numerical integration 6-10
6.5.1 Standard linear solid 6-10
6.5.2 Maxwell-Wichert model 6-11
6.6 References 6-12

Chapter 7 Elastic-plastic materials

7.1 Basic assumptions 7-1

7.2 Yield criteria 7-1
7.2.1 General considerations 7-1
7.2.2 Tresca yield criterion 7-8
7.2.3 von Mises yield criterion 7-9
7.2.4 High-exponent yield criterion 7-12
7.2.5 Quadratic yield criterion for anisotropic materials 7-13
7.2.6 Linear transformation based yield criteria for anisotropic materials 7-15
7.2.7 Drucker-Prager yield criterion 7-16
7.3 Strain decomposition and plastic dissipation 7-17
7.4 Plastic flow rule 7-19
7.4.1 The von Mises criterion 7-22
7.4.2 The Tresca criterion 7-23
7.5 Loading/unloading conditions 7-26
7.6 Work-hardening rules 7-28
7.6.1 Isotropic hardening 7-28
7.6.2 Extension to kinematic hardening 7-32

III
 7.7 Rate constitutive equations (tangent modulus) 7-36
7.8 Numerical integration 7-37
7.9 References 7-41

Chapter 8 Elastic-viscoplastic materials

8.1 Basic assumptions 8-1

8.2 General formulation 8-1
8.3 Special case: Rate-dependent J2 flow theory 8-4
8.3.1 Additive constitutive relation 8-4
8.3.2 Multiplicative constitutive relation 8-6
8.4 Viscoplasticity without a yield surface 8-11
8.5 Numerical integration 8-12
8.6 References 8-15

IV
1 Introduction
During the last decade computer technology has grown tremendously both in power and
capacity. As a result, numerical techniques have become commonplace within most branches
of industry and is currently used to simulate virtually all kinds of engineering problems. The
most used numerical technique within structural engineering is by far the finite element method
(FEM), which is a method for numerical solution of field problems. A field problem is normally
described by differential equations and associated boundary conditions, and the FEM provides
an approximate solution to the mathematical problem. In the FEM, the mathematical model
representing the real structure is discretized into a mesh of finite elements. This mesh is then
represented by a system of algebraic equations to be solved for unknowns at the nodes, where
the nodal unknowns are values of the field quantity, for instance displacements or temperature.
A large number of general-purpose FEM codes are today available on the commercial
marked, and the user-friendliness of these programmes has increased considerably over the
years. Even so, the analyst is still responsible for the input to the code. There are in general
three steps in a finite element analysis that cannot be solved without appropriate user
interference. The first step involves the modelling itself. Here the physical problem is idealized
into a mathematical model, and the geometry, boundary conditions and loads are defined. The
second step is the discretization, where the type and number of elements used to represent the
structure are selected. The third step is the selection of a constitutive model with appropriate
material parameters that is able to represent the behaviour of the material during loading. The
two first steps have been covered in various FEM and structural engineering courses. In this
course, the focus will be on the latter, where the main objective is to understand and predict the
response of materials when they are subjected to external loads. The different material models
covered in these lecture notes are elasticity, viscoelasticity, plasticity and viscoplasticity.
A material is said to be elastic if the deformation is reversible, i.e., the original shape is
recovered upon unloading, and there is a one-to-one relationship between the state of strain and
state of stress. The latter condition implies that the response of the material is path independent.
The elastic material is further rate independent, i.e., the rate at which the loading is applied has
no effect on the behaviour of the material. Usually it is further assumed that all energy used to
deform the material is stored and may be recovered; thus, there is no loss of mechanical energy
– or dissipation. This implies that a strain energy function exists for the material, which
represents the elastic energy stored in the material. Such materials are said to be hyperelastic.
Structural materials are approximately elastic if the deformations are sufficiently small.
Viscoelastic materials exhibit elastic and viscous behaviour. A viscous material behaves
like a fluid. The stress-strain response of a viscoelastic material is rate dependent and time
dependent, and there is dissipation of mechanical energy due to the viscosity. This type of
behaviour is typical for polymers. The behaviour of polymers is often strongly temperature
dependent. The characteristics of viscoelastic materials are most often determined from creep
tests, stress relaxation tests and cyclic tests. The latter tests are used to characterize the
hysteresis seen in the stress-strain curve, i.e., the energy loss (or dissipation) during each
loading cycle.
Structural materials like steel and aluminium behave linearly elastic for sufficiently small
deformations, but at a certain stress level, plastic yielding occurs and the behaviour becomes

1-1
nonlinear. Plastic deformations are irreversible, i.e., the original shape of the material is not
recovered upon unloading. The material behaviour is dissipative, but rate independent. The
theory of plasticity is applicable for a range of materials, like metals and alloys, concrete and
geo-materials, even if the physical mechanisms for the plastic deformations in these materials
are different. The theory is used to simulate a wide range of physical problems, such as collapse
of structures, crashworthiness and plastic forming. In these notes, both the phenomenological
theory of plasticity and the more physically based theory of crystal plasticity are described. The
latter theory is computational expensive and is not applicable to large-scale problems with
today’s computers, but it may be used to identify the parameters of phenomenological models
and to gain better understanding of elastic-plastic behaviour of metallic materials.
Viscoplasticity is the rate-dependent version of plasticity, which means that the
deformations are irreversible and dissipation occurs. The theory is used to model the rate
sensitivity of materials in the plastic domain, while the elastic domain is usually assumed to be
rate independent. The theory is applicable to creep and stress relaxation in metallic materials
and in simulations of impact problems and fast forming processes. Since the rate sensitivity of
metals increases with temperature, viscoplastic models are more important at elevated
temperatures.
Chapter 2 gives an overview of elasticity, viscoelasticity, plasticity and viscoplasticity in
one dimension, i.e., for a uniaxial stress state. The different constitutive models are presented
together with algorithms for numerical implementation. In Chapter 3, a brief introduction is
given to vectors and tensors. A 2nd order tensor is a quantity that provides a relationship between
two vectors and is thus independent of a particular choice of coordinate system. An example is
the stress tensor that provides a correspondence between the normal vector to a surface and the
traction vector acting on the surface. A 4th order tensor is a quantity that gives a relationship
between two 2nd order tensors, where an example is the elasticity tensor that relates the stress
and strain tensors in an elastic material. The stress and strain tensors are both 2nd order tensor.
The fact that vectors and tensor are quantities that are independent of a particular choice of
coordinate system is of utmost importance. Chapter 4 gives an introduction to continuum
mechanics for small deformations. Only the parts of continuum mechanics used in describing
the constitutive models of materials are included. This chapter describes topics that should be
known from previous courses in mechanics, but are repeated here using tensors. As already
mentioned, the constitutive equations are usually used in finite element simulations of various
mechanical problems. For this reason, Chapter 4 provides an introduction to the finite element
method for problems with material nonlinearity. Chapters 5–8 describe in turn the theories of
elasticity, viscoelasticity, plasticity and viscoplasticity for multi-axial stress states using tensor
notation. Small deformations are assumed throughout.
To enhance the readability of the lecture notes, there are relatively few references to
literature in the text. Instead, the most relevant books, chapters and articles are listed at the end
of each chapter for further reading. Sections marked with an asterisk * are not part of the
syllabus of the master’s course TKT4135 Materials mechanics.

1-2
2 Constitutive equations in one dimension
In this chapter, we will use a simple one-dimensional problem, namely a rod subjected to
tension/compression, to demonstrate various material models (or constitutive equations):
elasticity, viscoelasticity, plasticity and viscoplasticity. We start by defining the model problem
and continue by formulating the different constitutive equations. At the end, extension to large
deformations is briefly discussed.

2.1 Governing equations

Figure 2-1: One-dimensional rod subjected to loads.

We consider a one-dimensional rod of length L as shown in Figure 2-1. The rod is
subjected to a force FL  FL (t ) at the right end, while the left end is fixed. It is assumed that
the loading is time dependent and t denotes the time. Along the rod a body force b  b  x, t 
acts. This force is assumed to vary with time and place, and x is the coordinate in the
longitudinal direction of the rod in the initial configuration. The body force is given as force
per unit volume.
The displacement of the rod is
u  u  x, t  (2-1)
It is assumed that the displacement is infinitesimal, so that it is not necessary to distinguish
between the initial and deformed configurations of the rod. We further assume that the
displacement field is sufficiently smooth so that all the required mathematical operations are
legitimate. The velocity is then given by
u  x, t 
v  v  x, t    u  x, t  (2-2)
t
where the superposed dot signifies the material time derivative, i.e., the rate of change following
the motion of a particle. The strain is defined as

2-1
 u  x, t 
  x, t   (2-3)
x
This is a definition of the infinitesimal strain. It is seen that the strain equals the gradient to the
displacement field. It follows that the strain is zero if the rod is translated as a rigid body and
constant if the displacement of the rod varies linearly along the x coordinate. The strain rate is
obtained as
  u  x, t     u  x, t   u  x, t  v  x, t 
  x, t        (2-4)
t  x  x  t  x x
We note that the strain rate is nothing but the gradient of the velocity field. The strain rate is
important, since the response of a material to external loads often depends on the rate of the
deformation – and thus of the strain rate.
The Cauchy stress     x, t  in the rod is given by

N  x, t 
  x, t   (2-5)
A  x, t 

where A  A  x, t  is the current cross-section area of the rod and N  N  x, t  is the axial
force. It is assumed that the stress is uniform in the cross section. It is also possible to define
the stress with respect to the initial area A0  A0  x  , i.e.

N  x, t 
 e  x, t   (2-6)
A0  x 

The stress  e   e  x, t  is called the engineering (or nominal) stress. The Cauchy stress  is
frequently called the true stress, since it refers to the current area. We will intermittently use
 t to represent the true stress to clearly distinguish it from the engineering stress  e . Since the
displacement of the rod has been assumed to be infinitesimal, there is no need to distinguish
between the two stress measures. However, in Section 2.8 we will consider large deformations,
and then the difference between the true stress and the engineering stress becomes important.

2-2
Figure 2-2: Static equilibrium of an infinitesimal element of the rod.

It is assumed that the rod is loaded slowly so that inertia forces can be neglected. Static
equilibrium of an infinitesimal element of the rod at time t leads to (see Figure 2-2)

 A  bA  0 (2-7)
x
where N   A is the axial force acting in the rod. This is a first-order, linear differential
equation in the unknown stress field. If the area of the rod is uniform, the equilibrium equation
takes the simpler form

b  0 (2-8)
x
Solutions of the equilibrium equation are called equilibrium solutions. In general, exact
equilibrium solutions can only be found in simple cases, and it is therefore interesting to look
for approximate solution methods. Here we will focus only on the finite element (FE) method
as an approximate method.
Finally, we will derive the balance of mechanical power. Recall that power represents
work per unit time. The derivation for quasi-static conditions (i.e. the inertia forces are
neglected) is as follows. Equation (2-7) is multiplied with the velocity field v  v  x, t  and then
integrated over the length of the rod:

L L

0 x  A vdx  0 bAvdx  0 (2-9)

To proceed, integration by parts is applied to obtain the following relation

  v
L L L

0 x  Av  dx  0 x  A vdx  0  A x dx   Av0
L
(2-10)

We then get

2-3
 v
L L

 Av0    A dx   bAvdx  0
L
(2-11)
0
x 0

Since the velocity field should satisfy the boundary conditions, we have
v  0, t   0, v  L, t   vL  t  (2-12)
In addition, the mechanical boundary condition reads as
N  L, t     L, t  A  L, t   FL  t  (2-13)
Combining these equations, we arrive at the balance of mechanical power:
L L

  Adx  FLvL   bAvdx

0 0
(2-14)

The left-hand side is denoted the deformation power:

L L
Pd    Adx   d Adx (2-15)
0 0

It represents the power expended to deform the rod, while the right-hand side of Equation (2-14)
gives the power of the external forces. The deformation power (or stress power) per unit volume
has been introduced by
d   (2-16)
The equations derived up to now are all valid for any material. The only assumptions are
quasi-static conditions and infinitesimal deformations. To describe the relation between stress
and strain, we need to establish constitutive equations. In the following, we will establish
constitutive equations for elastic, viscoelastic, elastic-plastic and elastic-viscoplastic materials.

2.2 Elastic materials

A material is said to be elastic if, under isothermal conditions, the stress depends only on
the strain, i.e., there is a one-to-one relation between stress and strain. The elastic stress-strain
relation is written as
     (2-17)

where  is the uniaxial stress and  is the strain. The inverse relation is written
     (2-18)
Since there is a unique dependence between the stress and the strain, the material returns to its
original shape after the stress has been removed.
The elastic energy stored in the material per unit volume is obtained as

U 0        d (2-19)
0

2-4
where U 0    is the strain energy function. It emerges that the stress may be obtained by
applying U 0 as a potential, in the sense that

dU 0   
    (2-20)
d
The behaviour of the elastic material derived from the strain energy function is said to be
reversible and path independent, because the strains are recovered upon unloading and the
stored elastic energy (and thus the stress) is independent on the strain path.
Examples of linear and nonlinear elastic behaviour are illustrated in Figure 2-3. It is
important to notice that the same path is followed during loading and unloading of an elastic
material.

Linear
elasticity

Nonlinear
elasticity

Figure 2-3: Linear and nonlinear elastic behaviour.

Since we assume the strains to be infinitesimal, it is reasonable to assume a linear

relationship between the stress and the strain. We write
  E and    / E (2-21)
where E is Young’s modulus. An elastic material obeying Equation (2-21), which is called
Hooke’s law, is said to be a linear elastic material. Then the elastic energy stored in the material
per unit volume becomes
 
1 2
U 0         d    E d   E (2-22)
0 0
2

2-5
The elastic energy stored in linear and nonlinear elastic materials is illustrated in Figure 2-4.
The stored elastic energy of the body is U   U 0 dV , where V is the volume of the solid
V
body.

Linear Nonlinear
elasticity elasticity

Figure 2-4: Stored energy for linear and nonlinear elastic behaviour – the stored energy is
represented by the area under the stress-strain curve.

2.3 Thermoelastic materials

The mechanical behaviour of materials is sensitive to temperature variations. Accordingly,
if the temperature is allowed to vary, it is necessary to use a thermoelastic description of the
material behaviour. It is then assumed that the stress depends only on the strain and the change
of temperature. If it is further assumed that the strain is infinitesimal and the temperature
variation is small, it is reasonable to assume a linear relationship, namely
  E  T (2-23)
where T  T  T0 is the temperature variation, T0 is the reference temperature and  is the
thermal-stress coefficient. The minus sign stems from the fact that if straining is prohibited
(   0 ) under an increase of temperature, the resulting stress will be compressive and thus have
a negative sign according to our definition. Solving for the strain we get
   / E  T (2-24)
where    / E is the thermal expansion coefficient. Clearly, the strain has two contributions:
the first is due to the stress, while the second is a result of the temperature variation.
Also for the linear thermoelastic material it is possible to establish a potential function for
the stress

2-6
 1 2
U 0 ( , T )  E  T   0 (T ) (2-25)
2
where 0 is a function of temperature only. Using U 0 ( , T ) as a potential for the stress, we get
the desired result
U 0
  E  T (2-26)

Since the temperature variation is small per assumption, we may assume that E and  are
constants.

Example 2-1: Thermoelastic steel wire.

A steel wire with diameter D0  1 mm is heated to a temperature of T0  830 C before it
is stretched tight between two rigid walls (see Figure 2-5). The task is to calculate the tensile
stress that develops in the wire as it cools to T  20 C . For steel the Young’s modulus is
E  210000 MPa and the linear coefficient of thermal expansion is   1.2 105 / C .

Figure 2-5: Steel wire stretched tight between two rigid walls.

Let us first calculate the amount of shrinkage in the wire if it was allowed to cool freely.
From Equation (2-24) we may write
L  TL0  1.2 105  830  20  1300  12.6 mm
However, the wire is not permitted to shrink freely due to the fixture to the rigid walls. The
force that would be required to stretch the wire by this amount can be calculated as
F  LEA / L0  12.6  210000   12 / 4  /1300  1599 N

The tension stress developed in the wire due to the shrinkage becomes
  F / A  1599 /  12 / 4   2035 MPa
It should finally be remarked that this high stress level may well cause yielding of the steel wire.
If yielding occurs, the thermo-elastic relation is not valid anymore.
■

2-7
2.4 Linear viscoelastic materials
Metals and alloys, like steel and aluminium, behave as linear elastic materials for small
strains. Some materials, like polymers, exhibit rate-dependent behaviour already at small strains.
This type of behaviour is often modelled by linear viscoelasticity.

2.4.1 Rheological models

Viscoelastic materials are usually described by rheological models consisting of linear
springs and viscous dashpots (or dampers). The two basic mechanical elements used in
rheological models of linear viscoelastic materials are shown in Figure 2-6. To make the
rheological models more general, we will refer to  and  as the stress and strain, respectively,
which implies that these quantities are evaluated per unit basis. Similarly,  is referred to as
the strain rate. The constitutive relation for the linear spring reads as
  E (2-27)
where E is the spring constant. The stress is a function of the strain rate  for the viscous
dashpot, and thus
   (2-28)
where  is the viscosity constant. Equation (2-28) represents the behaviour of a viscous
material. The dashpot is introduced to enable modelling of rate dependence.
The two simplest rheological models for viscoelastic materials are the Maxwell model and
the Kelvin-Voigt model. These rheological models are shown in Figure 2-7.

Linear spring Linear viscous dashpot

Figure 2-6: Basic elements of rheological models: the linear spring and the linear viscous
dashpot.

2-8
2.4.2 Maxwell model
Using the relations for the basic elements, the stress-strain relation for the Maxwell model
is
 
 (2-29)
E 
For this model the total strain rate equals the sum of the strain rates in the linear spring and the
viscous dashpot, while the stress is the same in both elements. It is noted that the stress-strain
relation is established in rate form. The relation is alternatively re-written as
 
  e i, e  , i  (2-30)
E 
where the elastic and inelastic strain rates are identified. The inelastic contribution stems from
the viscous dashpot. From Equation (2-29) we see that for constant strain,   0 , the stress rate
is proportional to the stress and decreasing in magnitude. This is an example of stress relaxation.
For constant stress,   0 , Equation (2-29) predicts a constant strain rate and thus a linearly
increasing strain with time. This is an instance of creep.

Maxwell model Kelvin-Voigt model

Figure 2-7: Basic rheological models: the Maxwell model and the Kelvin-Voigt model.

The deformation power per unit volume of the Maxwell model is

d     e   i (2-31)

The elastic power  e is stored in the spring as elastic strain energy, while the viscous power,
which is related to the viscous dashpot, is dissipated and converted into heat:
Di   i   i i  0 (2-32)

where Di is the inelastic dissipation per unit volume. We note that the viscous power is non-
negative, since the viscosity is assumed to be positive. This implies that the dashpot will always
move in the direction of the applied stress. Non-negative dissipation is a requirement of the 2nd
law of thermodynamics [3].

2-9
 Assume that the stress is increased instantaneously from zero to a value  0 and then kept
constant with time. This is what is typical for a creep test. The Maxwell model will then give
an instantaneous elastic strain due to stretching of the spring equal to   0    0 / E . Using that
  0 and    0 , Equation (2-29) is solved to give
0 0   t
 t    t  0 1   (2-33)
E  E 
where    / E is called the relaxation time. It is seen that the strain will increase linearly with
time as long as the imposed stress is non-zero, see Figure 2-8.
In a stress relaxation test, the strain is increased instantaneously from zero to  0 and then
held constant. In this case, the instantaneous stress owing to stretching of the spring element is
  0   E 0 . Solution of the differential equation in Equation (2-29), assuming constant strain,
  0 , and initial condition,   0   E 0 , leads to

 t
  t   E 0 exp    (2-34)
 
It is seen that the stress relaxes towards zero as the time goes to infinity.

Figure 2-8: Comparison of creep behaviour predicted with the Maxwell model
( E  10000 MPa,   1000 s ), the Kelvin-Voigt model ( E  10000 MPa,   1000 s ) and the
standard model ( E0  E1  10000 MPa, 1  1000 s ) for  0  100 MPa .

2-10
2.4.3 Kelvin-Voigt model
The Kelvin-Voigt model gives the stress-strain relation
  E   (2-35)
Now the spring and dashpot are coupled in parallel and thus the stress equals the sum of their
individual stresses. The strain and strain rate are the same in both elements. In this model, we
may divide the stress into two parts, namely the elastic stress  e  E and an inelastic stress
 i   . The inelastic stress is linked to the viscous damper. From Equation (2-35) it is
apparent that at constant strain,   0 , the stress is constant and equal to   E  E 0 , which
is the contribution of the spring. It follows that the Kelvin-Voigt model is not capable of
describing stress relaxation. For constant stress,    0 , Equation (2-35) shows that the strain
rate (or creep rate) is proportional to the term  0  E , which represents exactly the stress in
the dashpot.
Let us consider a creep test in which the stress is constant in time and equal to  0 .
Applying the initial condition   0   0 and    0 , Equation (2-35) may be solved to give the
strain-time history as
0   t 
 t    1  exp    (2-36)
E   
It is seen that the strain approaches  0 / E as the time increases towards infinity, as shown in
Figure 2-8. This is contrary to the Maxwell model that predicts a linearly increasing strain as a
function of time in the creep test.

2.4.4 Standard model of viscoelasticity

The Maxwell and Kelvin-Voigt models are too simple to accurately represent the
behaviour of real viscoelastic solid materials. A more realistic model is the three-parameter
standard model of viscoelasticity (or the standard linear solid) shown in Figure 2-9.

Figure 2-9: The standard linear solid.

The rheological model consists of a spring with stiffness E0 in series with a parallel
spring-dashpot combination having stiffness E1 and viscosity 1 . Here, the spring with

2-11
stiffness E0 represents the elastic strain while the spring-dashpot combination  E1 ,1  defines
the inelastic strain.
The stress is assumed to act over both parts of the system, as illustrated in Figure 2-9, and
equilibrium requires that
  E0 e and   E1 i 1 i (2-37)
where the usual properties of springs and dampers were applied. It is seen that the strain can be
decomposed additively into an elastic part  e (i.e., the strain in the single spring) and an
inelastic part  i (i.e., the strain in the spring-dashpot combination), so that the decomposition
reads
  e i (2-38)
The inelastic strain  i is an internal variable introduced to describe the rate-dependent
behaviour of the viscoelastic material.
We then obtain the following constitutive relations defining the behaviour of the standard
model in linear viscoelasticity


E0
i and  i 
1
1
  E  
1
i
(2-39)

The material parameters E0 , E1 and 1 are temperature dependent. However, we will assume
isothermal conditions here, and thus any influences of temperature changes can be neglected.
If the stress history   t  or the strain history   t  is known, the linear viscoelastic equations
can be solved for given initial conditions (e.g.  i  0   0 ) either analytically or by a suitable
numerical method.
As an example of an analytical solution, we will again consider the creep test with the
stress constant in time and equal to  0 . Enforcing the initial condition  i  0   0 , the solution
of Equation (2-39)2 is

0   t 
 i t   1  exp     (2-40)
E1   1  
where 1  1 / 1 is the relaxation time. Combination of Equation (2-39) with    0 and
Equation (2-40) gives the solution for steady creep as

0 0   t 
 t    1  exp     (2-41)
E0 E1   1  
It is seen that the strain approaches a finite value for t   . The creep responses of the
Maxwell, Kevin-Voigt and standard models are compared in Figure 2-8.

2-12
Example 2-2: Tension test of a standard linear solid.
Consider a tensile specimen of a polymeric material that behaves according to the standard
linear model. The specimen is loaded with constant strain rate    0 and the task is to
determine the stress-time curve.
Since the initial value of the strain is zero, we have   t    0t in the tension test. Using
Equation (2-39), we get

i 
1
1
E  t E
0 0 0  E1   i   c0t   i  c1 i

where we have introduced the constants c0  E0 0 / 1 and c1   E0  E1  / 1 . Multiplying both

sides with the integrating factor ec1t , we get
ec1t c0t  ec1t  i  c1 i 

Thus

ec1t c0t 
dt
 e 
d i c1t

and integration gives

 i ec t  c0  ec t tdt  c2
1 1

where c2 is an integration constant. The solution is

c0
 i t   2  1
c t  1  c2e c1t
c1

Using the initial condition,  i  0   0 , we find c2  c0 / c12 , and finally

c0
 i t   2  1
 c t  1  e c1t 
c1
The stress-time curve is defined by
  t   E0  0t   i  t  
An example of the stress-strain curve obtained by the standard model is illustrated in
Figure 2-10 for given strain rate and model parameters.

2-13
Figure 2-10: Stress-strain curve in tension predicted with the standard model of viscoelasticity
( E0  10000 MPa , E1  1000 MPa , 1  1000 s ) for  0  0.0001s1 .

The standard model may be extended by adding more internal variables. This is obtained by
adding several spring-dashpot combinations in series, as illustrated in Figure 2-11. This model
is denoted the generalized Kelvin-Voigt model. The inelastic strain  i is then the sum of the
contributions from all spring-damper combinations and the viscoelastic constitutive equations
become

 n
   i ,  i    i (2-42)
E0  1

i 
1

  E   ,
 
i
  1, 2,..., n (2-43)

where the parameters E0 , E and  may depend on temperature and n is the number of
spring-damper combinations in the rheological model.

2-14
Figure 2-11: The generalized Kelvin-Voigt solid.
Assuming initial conditions i  0   0,   1, 2,..., n , the set of equations is solved either
analytically or by an appropriate numerical method for a given stress history   t  or strain
history   t  .
Note that it is also possible to extend the standard model by adding several Maxwell
elements in parallel, which gives a generalized Maxwell-type model.

2.4.5 Numerical solution

We will now devise an integration algorithm for the standard model in linear viscoelasticity.
To this end, we start by recalling the governing equations of the standard model, namely

  E0    i  ,  i    E  
1 i
(2-44)
1 1

We will here assume that the strain history   t  is given, and that the initial condition for the
inelastic strain is given as  i  0   0 . It is further assumed that at time tn , all the variables have
been determined using the numerical scheme that we will now define. A variable    at tn is
denoted   n and similarly the same variable is denoted   n1 at tn 1 . The time step or the
increment in time is designated tn1  tn1  tn . In the same fashion, the increment of the
variable    between tn and tn 1 is given as   n1    n1    n . What we have to do now
is to establish a way to determine all the variables at tn 1 given the strain  n 1 , and thus the
strain increment  n 1 , as well as all the variables at tn .
The backward-Euler method is used herein to integrate the constitutive equations. Given
the differential equation
dy  t 
 f t, y t  (2-45)
dt
with initial condition y  0   y0 , the backward-Euler algorithm is defined by

yn1  yn  f  tn1 , yn1  tn1 (2-46)

2-15
where tn1  tn1  tn is the time step. We note that the backward-Euler scheme is based on
expressing all equations using the variables at the end of the time step. The result is a very
robust and unconditionally stable algorithm with good accuracy.
Accordingly, the algorithmic version of Equation (2-44) is expressed as
  n 1 
 n 1  E0   n 1   ni 1   or  n 1    ni 1 
 E0 
(2-47)
tn 1
 ni 1   ni   ni 1 ,  ni 1 
1
 n 1  E1 ni 1 

where we have used:  ni 1   ni 1tn1 . Combining the equations in the algorithmic set-up, we
get
tn1
 ni 1 
1
E  0 n 1   E0  E1   ni   ni 1   (2-48)

This equation is readily solved to give an explicit equation in  ni 1 , and we find

E0 n 1   E0  E1   ni
 ni 1  (2-49)
1
 E0  E1
tn 1
When the strain history is known and suitable time increments chosen, the stress  n 1 and the
inelastic strain  ni 1 are obtained from Equations (2-47) and (2-49).
The algorithm described above is suitable for the finite element method, which is in a sense
strain driven, as well as for numerical solution of stress relaxation. For numerical solution of
the creep problem, we need an algorithm which starts from a known stress history   t  .
This algorithm is even simpler than the one just developed. Starting from the algorithmic
set-up, Equation (2-47)2, we obtain the incremental update of the inelastic strain as
 n 1  E1 ni
 ni 1  (2-50)
1
 E1
tn 1
Given the stress history and appropriate time increments, the strain  n 1 and the inelastic strain
 ni 1 are obtained from Equations (2-47) and (2-50).

2-16
2.5 Elastic-plastic materials
As already mentioned, metallic materials exhibit linear elastic behaviour for small stresses
(or strains), while at a certain stress, denoted the yield stress, the behaviour becomes elastic-
plastic. In the elastic-plastic domain, the material response is irreversible, which means that
parts of the strains are not recovered as the material is unloaded to zero stress. The strain is
therefore divided into a recoverable part, the elastic strain, and an irrecoverable part, the plastic
strain. Another characteristic of the elastic-plastic material is that the behaviour depends on the
straining or loading history. It is therefore generally required to express the constitutive
relations on incremental or rate form. There is no unique relationship between stress and strain
as in the theory of elasticity.
Typical idealized plastic behaviours are demonstrated in Figure 2-12. The behaviours of
an elastic-perfectly plastic material and an elastic-linear work-hardening material are shown. In
the case of linear-elastic perfectly plastic behaviour, the material behaves linearly elastic when
the absolute value of the stress is less than the yield stress. If the material is further strained,
this occurs without additional increase of the stress. If at a certain strain the material is unloaded,
the slope of the linear unloading path is defined by the elastic modulus E of the material. For
a material exhibiting elastic linear work-hardening behaviour the stress increases linearly with
the strain also after reaching the yield stress. The slope of the stress-strain curve in the elastic-
plastic domain is called the tangent modulus Et . The slope of the linear unloading path is given
by the elastic modulus. If the material is reloaded after unloading from an elastic-plastic state,
it follows the same path as during unloading and yielding commences at the stress at which
unloading took place.

1 1 1 1

Elastic-perfectly plastic Elastic-linear work-hardening

Figure 2-12: Idealized elastic-plastic behaviour.

To establish rheological models for elastic-plastic materials, we need to introduce another

basic rheological element, namely the friction element. This element is shown in Figure 2-13.
The stress increases without any sliding, i.e., the element is rigid, as long as the stress magnitude

2-17
is below  0 . Then sliding (or yielding) occurs at constant stress. During sliding d    0 ,
which implies that the sliding process is dissipative (i.e., the energy is lost as heat), and further
that  and  always have the same sign.

Figure 2-13: Basic elements of rheological models: the friction element.

The rheological model for the elastic-perfectly plastic material is shown in Figure 2-14 and
consists of a spring element with stiffness E and a friction element which slides when |  |
equals  0 . Obviously, the spring stiffness E corresponds to Young’s modulus and  0 to the
yield stress of the idealized material. It is seen that the strain may be divided into elastic and
plastic parts, so that
  e  p (2-51)
where the elastic strain  e is

e  (2-52)
E
The plastic strain  p represents the sliding of the friction element. To account for work-
hardening, we may assume that the resistance to sliding in the friction element increases with
plastic straining. If this dependence is assumed to be linear, the behaviour seen in Figure 2-12
for the elastic-linear work-hardening material is obtained.

Figure 2-14: Rheological model for elastic-perfectly plastic behaviour.

We will assume that the material is deformed under isothermal conditions, and thus any
effects of temperature variation can be neglected. The additive decomposition of the strain into
elastic and plastic parts is depicted in Figure 2-15. The elastic strain is reversible, while the
plastic strain is not (it is irreversible) and remains after unloading. Since the behaviour of
elastic-plastic materials is nonlinear, it is necessary to work with incremental quantities.

2-18
 The strain rate decomposes according to
d
  e  p (2-53)
dt
where, as indicated, the superimposed dot implies material time differentiation. If the material
is within the elastic domain, the plastic strain rate vanishes and the elastic strain rate equals the
total strain rate. The plastic domain is characterized by non-zero plastic strain rates. Multiplying
both sides of Equation (2-53) by the time increment dt leads to an incremental formulation of
the strain decomposition
d  d e  d p (2-54)
In most cases we will use the rate form, but it should be realized that the incremental form of
the constitutive relations is completely equivalent.
The elastic stress-strain response is still governed by Hooke’s law, but now using the
elastic strain instead of the total strain:
  E e  E (   p ) (2-55)
where isothermal conditions have been assumed. The rate form of Hooke’s law for the elastic-
plastic material reads
  E e  E (   p ) (2-56)
Note that these forms of Hooke’s law are equally valid in the elastic and plastic domains of the
material. It is assumed that the elastic modulus is independent of the plastic deformation of the
material, which is an accurate assumption for metals and alloys subjected to small strains.

2.5.1 Elastic-perfectly plastic materials

Assume now that the material is elastic-perfectly plastic, which implies that there is no
hardening (or increase of strength) of the material during plastic straining. The behaviour of the
elastic-perfectly plastic material is illustrated in more detail in Figure 2-15, where the
behaviours both in tension and compression are depicted. It is noted that whenever the absolute
value of the stress reaches the yield stress, whatever the straining history might be, the material
yields plastically without any work-hardening.

2-19
 1

Figure 2-15: Elastic-perfectly plastic material.

The yield condition is then given by
f  |  |  0  0 (2-57)

where f  f ( ) is the yield function, and  0 is the yield stress of the material. It is useful to
introduce an equivalent stress  eq that for uniaxial stress states is defined as

 eq  |  |   2 (2-58)
The yield condition implies that the material yields when the equivalent stress (or the stress
magnitude) reaches a critical value, namely the yield stress. If f  0 , the material is in the
elastic domain, while f  0 is inadmissible (not allowed). It follows that when the material is
in the plastic domain, the magnitude of the stress is always equal to the yield stress for the
elastic-perfectly plastic material. To illustrate the behaviour, a tension-compression loading
cycle is shown in Figure 2-16.
It is assumed that plasticity is a dissipative process, which means that the plastic work is
dissipated as heat. The plastic part of the mechanical work rate is lost as heat, while the elastic
part is stored in the material as strain energy. The deformation power per unit volume is defined
by
d     e   p (2-59)

The elastic power  e is stored, while the plastic power  p is dissipated. The incremental
work per unit volume reads
 d    d e   d p (2-60)
Integration of the incremental work over a closed strain cycle yields

2-20
   d    d   d 0
p e
, (2-61)

In a closed strain cycle the work done equals the plastic work, while the elastic work
vanishes. With reference to Figure 2-16, the plastic work per unit volume dissipated during the
tension-compression cycle is simply the dotted area inside the parallelogram.

1
1

Figure 2-16: Tension-compression loading cycle for the elastic-perfectly plastic material.
The assumption that the plastic power is dissipated implies that independent of the
direction of the loading process, the plastic power is non-negative, i.e.,
Di   p  0 (2-62)

where Di is the inelastic dissipation per unit volume. The very important implication of
Equation (2-62) is that the plastic strain rate must have the same sign as the stress. We may
then express the plastic strain rate as

 p  sgn      (2-63)
| |
where it is assumed that |  |  0 . The function sgn  x  is the sign of the variable x
( sgn  x   1 if x  0 , sgn  x   0 if x  0 , sgn  x   1 if x  0 ) 1 and   0 is the plastic
multiplier (a non-negative scalar). We note that plastic flow only occurs when the plastic

1
Any real number can be expressed as x  sgn  x  | x | , and thus for x  0 we have that sgn  x   x / | x | .

We also have the useful relations: sgn  x   1/ sgn  x  for x  0 and | x | x 2  x sgn  x  .

2-21
multiplier takes on values greater than zero, while a value equal to zero implies elastic
behaviour.
It is important to note that the partial derivative of the yield function with respect to the
stress is
f 
  sgn   (2-64)
 |  |
It follows that the plastic strain rate may be expressed as
f
p  (2-65)

This is the so-called associated flow rule in the theory of plasticity.
The plastic straining of the material is usually described in terms of the accumulated (or
equivalent) plastic strain p , which is defined to be energy conjugate (or more precisely power
conjugate) to the equivalent stress, in the sense that
  p   eq p (2-66)
Using the flow rule in Equation (2-63), as well as the definition of the equivalent stress,
Equation (2-58), we get
p (2-67)
Thus,
p   pdt    dt (2-68)

The plastic parameter  is non-negative, and thus the accumulated plastic strain p increases
monotonically during plastic straining independent of the plastic flow direction. In Section
2.5.2, we will use the accumulated plastic strain p as the internal variable governing the work-
hardening of the material, i.e., the strengthening of the material with plastic straining. We may
consider the accumulated plastic strain as a “plastic arc-length”, which measures the extent of
plastic deformation of the material. It is again emphasized that p is non-negative for both
tension and compression loading.
To determine whether or not a strain increment leads to plastic flow, we need to establish
loading/unloading conditions that define the value of the plastic parameter. If f  0 , the stress
is in the elastic domain and the plastic parameter vanishes (   0 ). If the yield condition is
fulfilled ( f  0 ), we have two possible cases: elastic unloading and plastic loading. For elastic
unloading   0 and the stress must move into the elastic domain making the yield function
negative, i.e.
f
f    sgn   E  0 (2-69)

It is seen that elastic unloading from a plastic state implies that the strain rate has the opposite
sign of the stress.

2-22
 For plastic loading   0 , and we must have f  0 to fulfil the yield condition at the end
of the strain increment. This consistency condition reads
f
f   0 (2-70)

Inserting Equations (2-64), (2-56) and (2-63) into Equation (2-70), we arrive at the following
expression for the plastic parameter during plastic loading
  sgn    (2-71)
It follows that in the plastic domain the elastic strain increment vanishes for the elastic-perfectly
plastic material and thus the plastic strain rate equals the total strain rate.
The loading-unloading conditions for the elastic-perfectly plastic material may be
summarized by the Kuhn-Tucker conditions
f  0,   0,  f  0 (2-72)
and the consistency condition in the theory of plasticity
f 0 (2-73)

The Kuhn-Tucker conditions imply that for elastic loading/unloading f  0 and   0 , so that
the product  f  0 . For plastic loading f  0 and   0 , so that once again we have  f  0 .
The consistency condition in the form provided by Equation (2-73) tells us that f must be
equal to zero when  is positive, i.e., for plastic loading, while for elastic loading/unloading
 equals zero and there is no restriction upon f .

2-23
2.5.2 Elastic-plastic materials with isotropic hardening

Figure 2-17: Elastic-plastic material with isotropic work-hardening.

Since materials usually work-hardens when plastically deformed, the elastic-perfectly

plastic model is seldom sufficiently accurate. The work-hardening of the materials leads to an
increase of the elastic domain as the material is plastically deformed, i.e., the flow stress (or
stress magnitude) increases with plastic working. The increase of the flow stress is caused by
rearrangements of the internal structure of the material which hinder the plastic flow. In metals
and alloys, the reason is an increase of the dislocation density of the material [1].
We represent the increase of the flow stress by the isotropic hardening variable R which
is taken to be a function of the accumulated plastic strain p . Thus, for the elastic-plastic
material with isotropic hardening the yield condition is rewritten as
f  , R   |  |   0  R   0 (2-74)
whereas the plastic flow is controlled by the associated flow rule as defined by Equation (2-65).
The isotropic work-hardening rule defines the evolution of the hardening variable R as a
function of the accumulated plastic strain p , and thus
R  R  p  or R  hR p (2-75)

2-24
where hR  hR  p  is the isotropic hardening modulus. The behaviour of the elastic-plastic
material with isotropic work-hardening is illustrated in Figure 2-17. In the figure, nonlinear
isotropic hardening is assumed, and the isotropic hardening variable R is identified. Since
p  0 in the elastic domain, work-hardening only occurs for plastic loading. Two frequently
used work-hardening rules are the power law

R  Kp n (2-76)
and the Voce rule
R  QR 1  exp  CR p   (2-77)

where ( K , n) and (QR , CR ) are hardening parameters. It is assumed that all the hardening
parameters are non-negative and further that n   0,1 .

Figure 2-18: Illustration of the Voce law ( QR  100 MPa , CR  50 ) and power law
( K  500 MPa , n  0.5 ) with yield stress  0  100 MPa .

These hardening rules are illustrated in Figure 2-18. It is seen that the Voce curve saturates
(i.e., the hardening rate tends to zero), while the power law curve increases indefinitely with
increasing strain. The hardening parameters are determined for instance from a tension test.
The isotropic hardening modulus hR is determined by differentiation of the hardening rules in
Equations (2-76) and (2-77).
It is noted that if the elastic-plastic material with isotropic hardening is loaded plastically in
tension (compression) so that the equivalent stress reaches the value  eq and then unloaded and
reloaded plastically in compression (tension), yielding commences when the equivalent stress

2-25
 eq reaches the same value  eq . The reason for this is that there is no plastic straining and thus
no work-hardening in the elastic domain. It is also noted that the plastic modulus hR has the
same value as it had at the point of unloading. This follows from the assumption that hR is a
function of the accumulated plastic strain p . The behaviour during unloading and reloading in
the opposite direction is illustrated in Figure 2-19. It is noted that the slope of the stress-strain
curve is defined by the tangent modulus Et , which will be shown below to depend on the elastic
modulus E and the plastic hardening modulus hR .

Figure 2-19: Response of the elastic-plastic material with isotropic hardening during plastic
loading in tension, elastic unloading and plastic reloading in compression.
The Kuhn-Tucker conditions and the consistency condition established for the elastic-
perfect plastic material holds also for the elastic-plastic material with hardening, but the rate of
the yield function takes the form
f f
f   R (2-78)
 R
where it is recognized that R is non-zero during plastic loading only. Combining the
consistency condition (Equation (2-73)) for   0 with Equations (2-64), (2-56), (2-63),
(2-74), (2-75) and (2-67), we obtain

 
f  sgn   E    sgn    hR  0 (2-79)

The plastic parameter then comes out as

2-26
 sgn   E
 (2-80)
E  hR
Inserting this expression into the rate form of Hooke’s law, Equation (2-56), and using Equation
(2-63), we arrive at the rate constitutive equation in the plastic domain (plastic loading)
EhR
  Et  , Et  (2-81)
E  hR
where Et is called the elastic-plastic tangent modulus. In the elastic domain (elastic
loading/unloading),  p  0 and the rate form of the stress-strain relation reads   E .

Example 2-3: Tension-compression test of an elastic-plastic material

A steel sample is first loaded in tension to a stress 1  200 MPa and then in compression
to a stress  2  250 MPa . The material is assumed to be elastic-plastic with isotropic
hardening and governed by the hardening rule
R  p   500 p 0.5

Young’s modulus is E  200000 MPa , and the yield stress is  0  150 MPa . Determine the
strains  1 and  2 corresponding to the stresses  1 and  2 , respectively.
At the two stress levels, the work hardening of the material is respectively
R  p1   50 MPa and R  p2   100 MPa . This gives the equivalent plastic strains as
1 1
 50  0.5  100  0.5
p1     0.01, p2     0.04
 500   500 
Since dp  | d p | , we obtain the plastic strains by
1p  p1  0.01,  2p  1p   p2  p1   0.02
The strains are then found by use of the strain decomposition and Hooke’s law, i.e.
1 200
1   1p   0.01  0.011
E 200000
 250
 2  2   2p   0.02  0.021
E 200000
■

2-27
2.5.3 Numerical solution
We consider the case of elastic-plasticity with linear hardening to make our numerical
recipe as simple as possible. We start by repeating all the equations governing the behaviour of
the material. The set of equations reads
  e  p

  E e  E     p 

 p   sgn   , p (2-82)

R  hR p (hR  const.)

f  |  |   0  R   0,   0,  f  0
There are numerous numerical schemes to solve this set of equations with unknowns  ,
R ,  p and p . It will be assumed that the strain history   t  is defined together with the
required initial conditions for the internal variables, e.g.  p  0   0 , p  0   0 and R  0   0 .
We will use the backward-Euler integration scheme here, since it is used almost without
exception in finite element implementation of elastic-plastic constitutive models.
As before, we assume that at time tn , all the variables have been determined using the
integration algorithm. We start by calculating an elastic trial state based on the variables at tn
and the strain increment  n 1 , assuming that the behaviour of the material is elastic during the
time step tn 1 , viz.

 ntr1   n  E n1 , Rntr1  Rn (2-83)

Note that the hardening variable is not assumed to be changing during the time step since the
behaviour is elastic per assumption.
If the trial state fulfils the yield condition at tn 1 , f  ntr1 , Rntr1   0 , our assumption was
correct and the trial state coincides with the real solution. In this case, we have
 n1   ntr1 , Rn1  Rntr1 ,  np1   np , pn1  pn (2-84)

On the contrary, if f  ntr1 , Rntr1   0 , the trial state is inadmissible (not allowed) and
accordingly plastic flow occurs during the time step – the time step involves plastic loading.
This implies that we have to find a method to re-establish consistency, f  n1 , Rn1   0 , in a
way that is in accordance with the constitutive equations defined in Equation (2-82). To this
end we write the algorithmic set-up of the constitutive equations for the backward-Euler method
as

2-28
  n 1   n  E   n 1   np1    ntr1  E  np1

 np1   np   np1 ,  np1  pn 1 sgn  n 1 

(2-85)
Rn 1  Rn  Rn 1 , Rn 1  hR pn 1

pn 1  pn  pn 1
This set of equations is subject to the nonlinear constraint defined by the yield condition at tn 1 ,
viz.
f  n1 , Rn1    n1   0  Rn1   0 (2-86)
The backward-Euler integration scheme is very efficient and it has good accuracy and excellent
robustness.
It is evident from the algorithmic set up that the only unknown quantity is pn 1 and when
pn 1 is determined all the variables at tn 1 can be directly calculated. This allows us to write
the yield condition at tn 1 as

f n1  f  pn1   0 (2-87)

To obtain the solution, we first note that
sgn  n1   sgn  ntr1  (2-88)

and thus
 n1  sgn  n1   n1  sgn  ntr1  n1 (2-89)

With the latter relation, the yield condition at tn 1 takes the form

f n1  sgn  ntr1  ntr1  E np1    0  Rn  hR pn1   0 (2-90)

Using that  np1  pn1 sgn  ntr1  , we obtain a closed-form solution for pn 1 as

 ntr1   0  Rn 
pn1  (2-91)
E  hR
With pn 1 determined the other variables at tn 1 are computed from Equation (2-85).
Since we have assumed linear hardening, we are able to obtain a solution without iteration.
If nonlinear hardening was assumed, i.e., hR  hR  p  , Equation (2-87) would have to be solved
by an iterative method. Using Newton’s method, we first express the nonlinear equation in
pn 1 in its residual form, viz.

  pn1   0 (2-92)
Next, the derivative of the residual function is approximated by

2-29
   pnI 11     pnI 1  0    pnI 1 
  pnI 1    (2-93)
pnI 11  pnI 1 pnI 11  pnI 1
We finally arrive at the iterative update scheme

I 1
  pnI 1 
p  pI

  pnI 1 
n 1 n 1

where I is the iteration counter and we have the initial value pn01  0 . The iteration scheme
stops when the residual is sufficiently close to zero. Newton’s method is very robust, and
usually it converges after a few iterations provided the strain increment is not too large.

Figure 2-20: Illustration of the backward-Euler algorithm, showing the initial state   n ,  n  at
tn , the elastic trial state  n 1 ,  ntr1  and the final state  n1 ,  n1  at tn 1 .

The backward-Euler algorithm is illustrated in Figure 2-20. In the trial step, the total strain
increment  n 1 equals the elastic strain increment  ne1 , and thus the trial stress is
 ntr1   n  E n1 . In the correction step, the total strain is constant and equal to  n 1 and at
the same time the plastic strain increment  np1 increases until consistency is re-established.
In this process, the stress is updated according to  n1   ntr1  E np1 and the hardening
variable develops according to its evolution equation. Consistency implies that
f  n1 , Rn1   0 , i.e., the yield condition is fulfilled by the state variables  n1 , Rn1  at tn 1 .

2-30
2.6 Elastic-viscoplastic materials
The flow stress of an elastic-plastic material usually depends on the rate of the plastic
straining. It is also observed that the rate sensitivity is typically higher at elevated temperatures
than at room temperature. The rate-dependent theory of plasticity is called viscoplasticity. This
theory has various applications where the rate dependence plays a role, such as creep and stress
relaxation, forming operations and structural impact problems.
Figure 2-21 illustrates the idealized behaviours of an elastic-perfectly viscoplastic
material and an elastic-viscoplastic material with linear hardening. As the figure indicates, the
flow stress increases with increasing strain rate. The yield stress at quasi-static (very low) strain
rate is  0 , whereas the viscous stress  v represents the increase of the yield stress due to
elevated strain rate. Usually,  v depends nonlinearly on the strain rate. This behaviour is not
accounted for in the rate-independent theory of plasticity described in Section 2.5. There neither
the yield stress nor the hardening rules depend on the strain rate, and thus the same stress-strain
behaviour will be obtained independently of the strain rate.

1 1

Elastic-viscoplastic with
Elastic-perfectly viscoplastic
linear work-hardening

Figure 2-21: Idealized elastic-viscoplastic behaviour.

To establish the rheological model for the elastic-viscoplastic material, we need another
basic rheological element, namely the nonlinear dashpot as shown in Figure 2-22. For this
element the stress is defined as       , where     is a nonlinear, positive function of the
strain rate.

2-31
Figure 2-22: Basic element of rheological models: the nonlinear viscous dashpot.

The rheological model for the elastic-perfectly viscoplastic material is illustrated in Figure
2-23. It consists of a spring element in series with a friction element and a nonlinear dashpot
coupled in parallel. The magnitude of the stress carried by the dashpot is the viscous stress
 v  0 that depends nonlinearly on the plastic strain rate p . The strain is divided into elastic
and plastic parts,    e   p , where the elastic strain is defined by Hooke’s law:  e   / E .
The plastic strain rate  p is then calculated as  p     / E . For stress magnitudes |  | less
than  0 , the response is elastic and defined by the spring with stiffness E , which represents
Young’s modulus of the idealized material. For stress magnitudes |  | greater than  0 , we get

|  |  0  v  p (2-94)
To include work-hardening in the rheological model, we may assume that resistance of the
friction element increases with the plastic strain. Linear work-hardening, as shown in Figure
2-21, is obtained by assuming this increase to be linear in the plastic strain.

Figure 2-23: Rheological model for elastic-perfectly viscoplastic behaviour.

The typical uniaxial creep and relaxation tests are illustrated in Figure 2-24 and Figure
2-25, respectively. In the creep test, the stress is increased to a level  c above the yield stress
 0 and then kept constant with time t . Under the action of the constant stress, the material
creeps, i.e., the strain increases with time in some nonlinear fashion, depending on the
properties of the actual material. If at time t A the material is unloaded to zero stress, the strain
becomes constant as indicated in the figure. In the relaxation test, the strain is increased to a
level  c that is higher than the strain at initial yielding  0   0 / E and then kept constant with

2-32
time. After the initial stress increase during straining towards  c , the stress decreases or relaxes
in a nonlinear manner. If after a time t A the strain is decreased to zero, the stress will decrease
abruptly, as indicated in the figure. If the stress yields in compression during the abrupt strain
reduction, the compressive stress will decrease nonlinearly with time, else it will remain
constant as indicated in the figure.

A A
B

Figure 2-24: Creep test on elastic-viscoplastic material (after [3]).

A
A

Figure 2-25: Relaxation test on elastic-viscoplastic material (after [3]).

2.6.1 Elastic-viscoplastic material with isotropic hardening

The main difference between the theories of plasticity and viscoplasticity is that in
viscoplasticity the stress is allowed to move outside the elastic domain, in the sense that the
yield function f is allowed to take values greater than zero. Roughly speaking, the plastic
strain rate depends in some nonlinear way on the value of f or, equivalently, on how far
outside the elastic domain the stress is located. It is now assumed that the material behaves
elastically when f  0 .

2-33
 For completeness we repeat some of the constitutive equations from the rate-independent
theory that are still valid, and we include isotropic hardening right from the start. As for the
rate-independent theory we will assume isothermal conditions, so that any influence of
temperature variation can be neglected.
The decomposition of strain into elastic and plastic parts is still valid, and reads
  e  p    e  p (2-95)
Hooke’s law relates the stress and the elastic strain
  E e  E    p     E e  E    p  (2-96)

where E is assumed constant.

Plastic flow occurs when the yield function f  f  , R  takes on positive values, where

f  |  |   0  R  (2-97)

Here,  0 is the yield stress and R is the isotropic hardening variable, while, as before, the
equivalent stress is defined by  eq  |  | . For values of f that are less or equal to zero, the
behaviour is elastic. Using the same reasoning as for the rate-independent theory, the associated
flow rule is used to define the plastic strain rate
f f
p  p (2-98)
 
where, as before, the plastic parameter equals the equivalent plastic strain rate (cf. Equation
(2-67)).
Since the yield function is allowed to take positive values, there is no consistency condition
in the rate-dependent theory. Instead, we are free to define p (or equivalently  ) as a function
of the state variables  and R . Formally this is written as
0 for f 0
p (2-99)
  , R  for f 0

where   0 is a function that has to be defined. Since  depends only on the state variables
 , R  , it is said to be a state function. The function should be so that p increases with
increasing magnitude of the stress, i.e., that the sensitivity to strain rate is positive. Sometimes
the constitutive relation is defined as
0 for f 0
p (2-100)
  f  for f 0

This means that it is explicitly the value of f that defines the equivalent plastic strain rate.
Recall that f depends on the state variables as shown in Equation (2-97).
To illustrate some typical constitutive relations in viscoplasticity, we will consider a
multiplicative and an additive constitutive relation. A generic multiplicative relation reads

2-34
 0 for f 0

  1

p    |  |  C  (2-101)
 p0    R   1 for f 0
  0  

where p0 and C are parameters governing the rate sensitivity of the material. Here p0 is often
called the reference strain rate. Observe that according to Equation (2-97) the expression inside
the brackets is zero for f  0 and positive for f  0 . Solving for  eq in the plastic domain, we
get
C
 p
 eq  |  |   0  R  1   (2-102)
 p0 
It is seen that the term contributing for rate sensitivity is multiplied with the term accounting
for work-hardening. Note also that f  0 as p  0 .
A basic additive constitutive relation is defined by
0 for f 0

p   f  , R    (2-103)
 p0  exp    1 for f 0
   S  
where p0 and S are parameters determining the rate sensitivity of the material. In the plastic
domain, the equivalent stress is obtained as
 p
 eq  |  |   0  R  S ln 1   (2-104)
 p0 
It becomes apparent that in this case the term accounting for the rate sensitivity is added to the
work-hardening terms. Again we note that f  0 as p  0 . The choice of constitutive
relation for a given material has usually to rely on observations from mechanical tests.

2-35
Figure 2-26: Flow stress  eq as a function of plastic strain rate p for the multiplicative
( C  0.01 , p0  0.001s1 ) and additive ( S  1 MPa , p0  0.001s1 ) constitutive relations for a
material without work-hardening ( R  0 ).

The multiplicative and additive constitutive relations are compared in Figure 2-26. The
flow stress  eq as a function of plastic strain rate p is shown for a material without work-
hardening. It is seen that for the actual sets of parameters the two models give similar rate
sensitivity. This is not always the case. For both models the influence of strain-rate increases
as the plastic strain rate p becomes greater than the reference strain rate p0 . Note in particular
that Equation (2-104) leads to a log-linear increase of the flow stress with strain rate in the range
p / p0  1 .
Finally, we may establish the rate constitutive equation for the elastic-viscoplastic material.
Combination of Equations (2-96), (2-98), (2-64) and (2-99) leads to

 E
 for f 0
 
 E     , R  sgn    for
(2-105)
 f 0

where both the elastic and plastic domains are covered. This equation needs some analysis:
 If the material is in the plastic domain ( f  0 ) and the stress rate is set to zero (   0 ),
the strain rate is not zero, but given by     , R  sgn   . This is creep under constant
stress. It will last until the material has work-hardened sufficiently to make f  0 . If
this is not possible, the creep will continue indefinitely. Since the stress rate is zero
during creep, Hooke’s law on rate form implies that the elastic strain rate is zero. This
means that the plastic strain rate equals the total strain rate.

2-36
  If the material is in the plastic domain ( f  0 ) and the strain rate is set to zero (   0 ),
the stress rate is not zero, but given by    E  , R  sgn   . This is stress relaxation
under constant strain. The stress will change until f becomes equal to zero. This occurs
because the magnitude of the stress decreases and at the same time the material work-
hardens. Since the strain rate is zero, the elastic and plastic strain rates have the same
magnitude but opposite sign under stress relaxation.
We conclude that the theory of elastic-viscoplasticity is indeed capable of describing the
important phenomena of general rate sensitivity of the stress-strain behaviour as well as creep
and stress relaxation.

Example 2-4: Creep response of perfectly viscoplastic material.

A perfectly viscoplastic material with constitutive relation
1
 | | C
p  p0  
 K  p  p n 
 0 
is considered. The material constants are p0  0.0001s1 , K  200 MPa , p0  0.002 ,
n  0.25 and C  0.05 . This material has no yield limit, and thus plastic flow takes place for
all non-zero stresses. In general, however, the plastic strain rate will be vanishingly small for
|  |  K  p0  p  . A test coupon of the material is first pulled in tension at quasi-static strain
n

rate, p  0.0001s1 , to a plastic strain of 0.02. The stress is then kept constant for 24 hours
before the test is stopped. The task is to calculate the plastic strain in the specimen when the
test terminates.
Since p  p0 during the loading phase, we have

|  |  K  p0  p 
n

With the given material constants and the plastic strain p equal to 0.02, this equation gives a
stress equal to  0.02  77.03 MPa . This stress is kept constant for 24 hour. The constitutive
relation during creep at constant tensile stress  0.02 reads
C
 p
 0.02  K  p0  p   
n

 p0 
and thus
1
  0.02  C
t p n

0 0  K 
p dt   0  p  p  C dp

0.02

where t is the creep time and p is the total plastic strain. Integration gives

2-37
 1
  0.02  C 1  n n
1 
   
1
p0   t  p  p C  p  0.02 C

C 1 
0 0
 K  n

so that
1
 n    0.02  C
n n
p t    
1 1
C
  1   0 p t  p0  0.02 C  p0
 C  K 
Inserting for the material constants, the creep stress,  0.02  77.03 MPa , and the creep time,
t  24 h , the total plastic strain after 24 hours comes out as: p (24 h)  0.078 . The creep
response p  p  t  is illustrated in Figure 2-27.

0,09

0,08

0,07

0,06
Plastic strain

0,05

0,04

0,03

0,02

0,01

0
0 4 8 12 16 20 24
Creep time (hours)

Figure 2-27: Creep response of the viscoplastic material without yield surface in Example 2-4.

2-38
2.6.2 Numerical solution
The integration algorithm for the viscoplastic equations is similar to the one for plasticity.
The set of equations governing the behaviour of the viscoplastic material with linear isotropic
hardening reads
  e  p

  E e  E     p 

 p   sgn   , p (2-106)

R  hR p,

0 for f  , R   0
p
  , R  for f  , R   0
The only difference between the set of governing equations for plasticity and viscoplasticity is
the last equation. Here the loading/unloading conditions of plasticity have been substituted with
the constitutive relation defining the equivalent plastic strain rate.
The trial step is exactly the same for plasticity and viscoplasticity, and is not repeated here.
The algorithmic set-up of the constitutive equations for the backward-Euler method is slightly
different, namely
 n 1   ntr1  E  np1

 np1   np   np1 ,  np1  pn 1 sgn  n 1 

Rn 1  Rn  Rn 1 , Rn 1  hR pn 1 (2-107)

0 for f  n 1 , Rn 1   0
pn 1  pn  pn 1 , pn 1  
  n 1 , Rn 1  tn 1 for f  n 1 , Rn 1   0

In the case of viscoplasticity, the constraint in the plastic domain is given by the last equation
in the algorithmic set-up. It is convenient to re-write the algorithmic version of the viscoplastic
constitutive relation in the form
n1    pn1     pn1  tn1  pn1  0 (2-108)

where n1 depends on pn 1 as the single variable, because  n 1 and Rn 1 are given by pn 1
and known quantities at tn . This nonlinear equation in pn 1 has to be solved by an iterative
method. Here we will indicate how Newton’s method can be used to solve Equation (2-108).
Taking the trial state as the initial condition to the iterative solution scheme, Newton’s
method gives

I 1 nI 1
 n 1  I
n 1  J d p
I
n 1
I
n 1  0 and d p I
n 1  I (2-109)
J n 1

2-39
where d pnI 1 is the iterative change of pn 1 in iteration I, and the Jacobean J nI 1 is defined
by


I

J nI1  (2-110)
p n 1

During the iteration we have pnI 11  pnI 1  d pnI 1 and pnI 11  pn  pnI 11 , where the initial
condition is pn01  0 . The iteration stops when d pnI 1 is less than a given tolerance.

2.7 Kinematic hardening

Figure 2-28: The Bauschinger effect.

For cyclic loading the assumption of isotropic hardening may not be sufficiently accurate,
since the so-called Bauschinger effect is not described. Let the initial yield stress of an elastic-
plastic, work-hardening material be denoted  0 . Assume that plastic yielding occurs at this
stress magnitude independent of the direction of loading. The material is then loaded (say in
tension) into the plastic domain to a stress  A . Since the material work-hardens, we have that
|  A |   0 . The material is then unloaded elastically from  A to zero stress and then reloaded
in the reverse direction. Assume that yielding in the reverse direction occurs at the stress  B .
It turns out that for most materials |  B |  |  A | and often we find that |  B |   0 . This
behaviour is shown in Figure 2-28 and is called the Bauschinger effect.

2-40
 This observation implies that in addition to the isotropic work-hardening discussed in
Section 2.5.2, where the strength increases by an equal amount in tension and compression due
to plastic working, we have a directional hardening, leading to a higher strength in the direction
of plastic loading. This is called kinematic hardening and can be looked upon as a translation
of the elastic domain in the direction of the plastic loading.

2.7.1 Plasticity and viscoplasticity

To account for kinematic hardening, the yield function is re-defined as
f    , R   |    |   0  R  (2-111)

Here  is denoted the kinematic hardening variable, which defines the translation of the elastic
domain in the direction of plastic loading. It is seen that    acts as an effective stress, and
it is the magnitude of this effective stress that determines whether or not a material yield
plastically for given values of yield stress  0 and isotropic hardening R .
The kinematic hardening variable  is often called the backstress, since it is attributed to
internal stresses caused by previous plastic deformations and assisting yielding in the reversed
direction.
The rheological models for elastic-plastic and elastic-viscoplastic materials with isotropic
and linear kinematic hardening are shown in Figure 2-29. The backstress  is the stress carried
by the spring with stiffness h . The strain is decomposed as    e   p , where the elastic strain
 e is the strain of the single spring and the plastic strain  p is the strain of the combined
element. Accordingly, the backstress may be expressed as
  h  p (2-112)

where h is denoted the kinematic hardening modulus and assumed here to be constant. The
condition for the friction element to start sliding reads: f  |    |   0  R   0 , which is
nothing but the yield function defined in Equation (2-111).
The rheological model shows that the stress power is decomposed as
   e   p   e   p       p (2-113)
where the first two terms in the last expression are stored in the two springs, whereas the third
term is dissipated as heat. It follows that the dissipation inequality reads as
Di       p  0 (2-114)
The inelastic dissipation takes place in the friction element and, in the case of viscoplasticity,
in the nonlinear dashpot.

2-41
 (a) Elastic-plastic material

(b) Elastic-viscoplastic material

Figure 2-29: Rheological model for elastic-plastic (a) and elastic-viscoplastic (b) material with
combined isotropic and kinematic hardening.

The implication of Equation (2-114) is that the plastic strain rate has the same sign as the
effective stress, i.e.
 p  sgn      (2-115)
The partial derivatives of the yield function with respect to the stress and the backstress are
obtained as
f f  
   sgn     (2-116)
    
and thus
f
p  (2-117)

The accumulated plastic strain p is defined to be energy conjugate to the magnitude of
the effective stress    , in the sense that
Di      p  |    | p (2-118)
Applying the flow rule, Equation (2-117), we once again arrive at Equations (2-67) and (2-68),
namely that p   and p    dt .

2-42
 Owing to the back stress, yielding will occur for a lower stress magnitude when the
material is reloaded in compression after first being loaded in tension, and vice versa. Hence,
to describe the Bauschinger effect,  should develop positive values in tension and negative
values in compression. Using Equations (2-112) and (2-115), the evolution equation for the
backstress is obtained as
  h  p  h sgn      (2-119)
It is evident that the rate of change of the kinematic hardening variable is in the direction of the
effective stress    .
Even if this is often a rather crude approximation to the real behaviour of materials, we will
limit our discussion here to models combining linear isotropic and linear kinematic hardening.
The work-hardening of the material is then defined by the two constants hR and h . While the
model with only isotropic hardening could be identified from a tension test, it is required to
carry out cyclic tests to determine the two constants hR and h of the model with combined
linear hardening. The behaviour predicted during a loading cycle with either linear isotropic
hardening or linear kinematic hardening is depicted in Figure 2-30, where the hardening moduli
hR and h are given the same value h0 .

Linear isotropic hardening Linear kinematic hardening

Figure 2-30: Loading cycle for materials with linear isotropic and kinematic hardening.
As clearly seen, there are large differences between the predicted behaviours during re-
loading first in compression and then in tension. However, the first excursion in tension is the
same for the two hardening rules when the hardening moduli are the same. It is further noticed

2-43
that the backstress  A at the point of unloading from the tensile stress  A is equal to  A   0 ,
which is seen to represent the centre of the elastic region along the stress axis.
If instead of the two extreme cases in Figure 2-30 we assume combined linear hardening
with hR   h0 and h  1    h0 , where  is a number between zero and unity, the behaviour
will also here be identical during the first excursion in tension. The behaviour during re-loading
first in compression and then in tension will be between the two limiting cases defined by   0
(linear kinematic hardening) and   1 (linear isotropic hardening).
For the elastic-plastic material with combined isotropic and kinematic hardening the
loading/unloading conditions defined by Equations (2-72) and (2-73) are valid, and the
consistency condition gives
f f f f f
f    R      R (2-120)
  R  R
where we have used that from Equation (2-116)
f f
 (2-121)
 
Making use of Equations (2-117), (2-56), (2-116), (2-111), (2-75) and (2-119), the consistency
condition with   0 gives

 
f  sgn     E    sgn       h  hR    0 (2-122)

and the plastic parameter is determined by

sgn     E
 (2-123)
E   h  hR 

Using the rate form of Hooke’s law, the rate constitutive equation for plastic loading is
obtained as
E  h  hR 
  Et  , Et  (2-124)
E   h  hR 

while for elastic loading/unloading we have   E , as before.

For the elastic-viscoplastic material, the constitutive relation is defined as
0 for f 0
p (2-125)
    , R  for f 0

which implies that to account for kinematic hardening the stress  is substituted with the
effective stress    as a variable in the function  defining the rate sensitivity of the
material. Apart from this substitution, we may use the same constitutive relations as described
in Section 2.6.1.

2-44
2.7.2 Numerical solution
In the following the numerical scheme derived in Section 2.5.3 for elastic-plastic materials
with isotropic hardening is extended to allow for combined linear isotropic and kinematic
hardening. The set of constitutive equations to be solved reads
  e  p

  E e  E     p 

 p   sgn     , p (2-126)

R  hR p,   h  p

f       0  R   0,   0,  f  0

The unknown variables are  , R ,  ,  p and p with initial conditions  p  0   0 , p  0   0 ,

R  0   0 and   0   0 . It is assumed that the strain history   t  is given and the backward-
Euler integration scheme is used to integrate the set of equations in time.
It is assumed that at time tn all variables have been determined using the integration
scheme. An elastic trial state based on the variables at tn and the strain increment  n 1 is
calculated, assuming that the behaviour of the material is elastic during the time step tn 1 :

 ntr1   n  E n1 , ntr1  n , Rntr1  Rn (2-127)

If f  ntr1   ntr1 , Rntr1   0 , the trial state coincides with the real solution, and we have

 n1   ntr1 , n1  ntr1 , Rn1  Rntr1 ,  np1   np , pn1  pn (2-128)

On the contrary, if f  ntr1   ntr1 , Rntr1   0 , the trial state is inadmissible, and consistency,
f  n1   n1 , Rn1   0 , has to be re-established using the set of constitutive equations defined
in Equation (2-126).
The algorithmic set-up of the constitutive equations for the backward-Euler method reads
as
 n 1   n  E   n 1   np1    ntr1  E  np1

 np1   np   np1 ,  np1  pn 1 sgn  n 1   n 1 

Rn 1  Rn  Rn 1 , Rn 1  hR pn 1 (2-129)

 n 1   n   n 1 ,  n 1  h  np1

pn 1  pn  pn 1

2-45
These equations are subject to the nonlinear constraint defined by the yield condition at tn 1 :

f  n1  n1 , Rn1    n1  n1   0  Rn1   0 (2-130)

It is apparent that pn 1 is the only unknown quantity and when it has been determined, all the
variables at tn 1 can be directly calculated. Thus, the yield condition at tn 1 can be written as

f n1  f  pn1   0 (2-131)

It may be shown that [4]
sgn  n1   n1   sgn  ntr1   n  (2-132)

and thus
f n 1  sgn  ntr1   n  ntr1  E  np1   n  h  np1 
(2-133)
  0  Rn  hR pn 1   0

Using that  np1  pn1 sgn  ntr1  n  , we obtain a closed-form solution for pn 1 as

 ntr1   n   0  Rn 
pn 1  (2-134)
E  h  hR
Since we have assumed linear hardening, we are able to obtain a solution without iteration. If
nonlinear hardening was assumed, Equation (2-131) would have to be solved by an iterative
method.
The numerical algorithm in Section 2.6.2 for the elastic-viscoplastic material with isotropic
hardening can be extended to allow for combined isotropic and kinematic hardening in a similar
way as described above.

2-46
2.8 Large strains

Figure 2-31: One-dimensional uniform rod subjected to tension loading.

In this section, we will consider the uniform rod in Figure 2-31, which is fixed at the left
end and loaded by a point load at the right. The rod is representative of a tension test, which is
the most common test for characterising the behaviour of materials. In this case, the rod
represents the uniform part of the tensile specimen, which is denoted the gauge length, i.e., the
region in which the strain is measured. We will assume that the displacement u L of the right
end is finite and thus that the geometrical changes are significant.
Assuming uniform conditions along the rod, the displacement field is given by
x
u  x, t   uL  t  (2-135)
L0

where L0 is the initial gauge length and uL  t  is the time-dependent displacement of the free
end. Note that x is the coordinate along the gauge length in the initial configuration. The
engineering strain  e is then defined by

u  x, t  uL  t 
 e t    (2-136)
x L0
It is seen that the engineering strain has the same definition as the infinitesimal strain that we
introduced for infinitesimal deformations. The increment of the engineering strain is
duL  t 
d e t   (2-137)
L0
which clearly demonstrates that the straining is referred to the initial configuration.
Since we now allow for large deformations, we have to distinguish between the coordinate
of the material point x in the initial configuration and its position x in the current
configuration. Or stated differently, we must account for the geometrical changes of the

2-47
specimen. With this in mind, it seems reasonable to define the strain increment with respect to
the current length of the gauge region, i.e.
duL  t 
d l t   (2-138)
L
Integration gives
 L
uL L
duL dL
l  
0
L

L0
L
 ln  
 L0 
(2-139)

where we have used that L  L0  uL and accordingly dL  duL . The strain  l obtained this
way is called the logarithmic (or true) strain. It is easily verified that
 L
 l  ln    ln 1   e  (2-140)
 L0 
The incremental work done by the external force over the gauge region is given as
FL duL
FL duL  AL   t d  l AL (2-141)
A L
or
FL duL
FL duL  A0 L0   e d  e A0 L0 (2-142)
A0 L0
Here the Cauchy stress (or the true stress) and the engineering stress have been introduced
respectively as
FL FL
t  and  e  (2-143)
A A0
We see that  t d  l and  e d  e are the incremental work per unit current and initial volume,
respectively, and thus the true stress is work conjugate with the logarithmic strain, whereas the
engineering stress is work conjugate with the engineering strain. It is important to notice that
in finite element codes the stress-strain data are by default given in terms of the true stress and
the logarithmic strain.
In a tensile test, the force FL and the elongation u L are continuously measured, and we
may then calculate the engineering stress and strain, provided the initial geometry ( L0 and A0 )
is known. To compute the true stress, we also need the current cross-sectional area. For metals
and alloys the elastic strains remain small (of the order 0.001 to 0.01) and in addition the plastic
deformation is volume preserving. The reason for this is that plastic strains occur by plastic slip.
Assuming volume constancy, we have the relation
A0 L0  AL (2-144)
Use of Equation (2-139), gives

2-48
 A  A0 exp   l  (2-145)
This leads to the following relations between the true stress and the engineering stress and
strain:
FL FL A0
t     e exp  l    e 1   e  (2-146)
A A0 A
To summarize: In a conventional tensile test, the engineering stress and strain are first
computed based on measurements of force and elongation as well as the initial geometry of the
specimen. Then the true stress and the logarithmic strain are computed using the established
relations to the engineering values.

Diffuse
necking

Yielding
Fracture

Figure 2-32: Engineering stress-strain curve in tension showing yielding, diffuse necking and
fracture.

In the derivations above, we have presupposed that the deformation of the specimen was
uniform over the length L . This is the case for increasing values of the force, and thus of the
engineering stress, but as the force reaches its maximum value the specimen starts to neck
down. This implies that the deformation localizes somewhere along the gauge length and in this
region the cross-sectional area decreases rapidly with increasing elongation. This process is
called diffuse necking and is an example of instability. The engineering stress-strain curve
from a tension test, showing yielding, diffuse necking and fracture, is depicted in Figure 2-32.
From Equation (2-146), the engineering stress is obtained as
 e   t exp   l  (2-147)
Applying the product rule, the incremental change of the engineering stress is
d e  d t exp   l    t exp   l  d l   d t   t d l  exp   l  (2-148)

2-49
The engineering stress reaches its maximum value as d e  0 , and thus diffuse necking occurs
when
d t
 t (2-149)
d l
The diffuse necking criterion is also known as the Considère criterion. This is shown
graphically in Figure 2-33. In conclusion, the straining of the specimen is uniform as long as
the slope of the true stress vs. logarithmic strain curve is greater than the value of the true stress.
When the slope equals the stress, diffuse necking occurs. It is extremely important to realize
that the relations derived in this section are only valid as long as the deformation is uniform.
Accordingly, under normal conditions, diffuse necking marks the end of the useful part of the
tensile test.

t  t vs  l

d t
 t
d l

d t
vs  l
d l
 lu
l

Figure 2-33: Graphical illustration of necking criterion, where  lu is the logarithmic strain at
necking.

2-50
Example 2-5: Necking criterion for the Voce hardening rule
For aluminium alloys it has been shown experimentally that the stress-strain relationship
is well described by the Voce hardening rule
 eq   0  QR (1  exp(CR p))
where  0 , QR and CR are parameters.

a) Use this stress-strain relationship and find the logarithmic strain  lu (expressed using
the parameters) when necking occurs under uniaxial tension. Elastic strains are
neglected.
b) Express the tensile strength,  eu , i.e., the engineering stress at maximum force in the
specimen, using the parameters  0 , QR and CR .

In uniaxial tension, we have  eq   t and p   lp   l before necking, and Equation

(2-149) leads to

QRCR e
 CR lu 
  0  QR 1  e   CR lu 

and thus
1 QR 1  CR 
 lu  ln
CR  0  QR
The true stress at necking is then obtained as


 tu   0  QR 1  e C  R lu 
   0  QR 
CR
1  CR
The tensile strength  eu is found from Equation (2-146) as
 tu
 eu 
1   eu 
where  eu is the engineering strain at necking. Utilizing Equation (2-140), we get
1
 Q 1  CR   CR
 eu  exp   lu   1   R  1
  0  QR 
Combination of these equations gives the final result as
CR
 0  QR 
1  CR
 eu  1
(2-150)
 QR 1  CR   CR

 
  0  QR 
■

2-51
2.9 References
[1] J. Lubliner. Plasticity Theory. Macmillan, 1990.
[2] L.E. Malvern. Introduction to the Mechanics of a Continuous Medium. Prentice-Hall,
1969
[3] J. Lemaitre, J.-L. Chaboche. Mechanics of Solid Materials. Cambridge University Press,
1990.
[4] N. Valoroso. Theory and implementation of plasticity and viscoplasticity. Class notes.
Instituto ITC-CNR, Rome, Italy. March 11, 2002.

2-52
3 Vectors and tensors
From statics and strength of materials we know that the body force b and the displacement
u are examples of physical vectors, and in mathematics we have learned vector algebra and
vector calculus. We have also been introduced to stress and strain. These physical quantities
are examples of 2nd order tensors, which transform a set of vectors into another set of vectors.
As an example, the stress tensor takes the unit normal vector to a surface and transforms it into
the traction vector acting on the surface. The aim of this Chapter is to give an introduction to
tensors which is sufficient for our main topic, namely the mechanical behaviour of materials.
The reader is referred to References [1]-[3] for more comprehensive treatments of tensors.

3.1 Coordinate transformation

We introduce a Cartesian coordinate system with coordinate axes  x1 , x2 , x3  and unit
vectors  e1 , e2 , e3  , as illustrated in Figure 3-1. The unit vectors are orthonormal, which implies
that
ei  e j   ij (3-1)

The symbol  ij is denoted the Kronecker delta and is defined by

1 if i j
 ij   ji   (3-2)
0 if i j

where i, j  1, 2,3 . The unit vectors  e1 , e2 , e3  are said to form an orthonormal basis. In the
following, we will use the notation  ei  for an orthonormal basis. As illustrated in Figure 3-1,
any vector v is then represented as the decomposition v   i 1 vi ei , where  v1 , v2 , v3  are the
3

components of the vector in the basis  ei  .

We will adopt Einstein’s summation convention:
If an index appears exactly twice in a term, summation over the number region of
this index is implied and the summation symbol may be omitted.

Thus, the decomposition of any vector takes the simple form v  vi ei , while the scalar product
between two vectors is u  v  uiei  e j v j  ui v jij  ui vi . The last equality holds because  ij
vanishes for i  j . The summation index is also called a dummy index. The reason for this is
that the summation index (e.g. i ) can be replaced by any index symbol (e.g. j, k , l ,... ) without
changing the meaning of the expression, as long as it is not the same as another index symbol
in the same term.

3-1
Figure 3-1: Cartesian coordinate system and unit vectors.

Another important operation is the projection of a vector u onto the unit vector ei , which
by use of the summation convention reads: ei  u  ei  e j u j  ij u j  ui . It is seen that the result
is the vector component along the coordinate axis xi .

Example 3-1: Some results for the Kronecker delta.

In the following, the summation convention is used to show some useful results concerning
the Kronecker delta  ij :

 ii  11   22   33  3
 ij ij  1 j1 j   2 j 2 j   3 j 3 j  3
 ij ik  jk  1 j1k  jk   2 j 2 k  jk   3 j 3k jk  3
 ij jk   i11k   i 2 2 k   i 3 3k   ik
■
One fundamental property of a vector is how it transforms under change of orthonormal
basis. To establish the transformation rule, we consider two Cartesian coordinate systems
defined by two sets of axes, xi and xi , and the associated two sets of unit vectors  ei  and
 e  . The direction cosines between the two sets of unit vectors are defined by

i

a  e  e  cos  x , x 
ij

i j

i j (3-3)

3-2
where cos  xi , x j  denotes the cosine of the angle between the coordinate axes xi and x j . The
direction cosines are illustrated in Figure 3-2. It is understood that aij are the components of
the unit vectors ei decomposed onto the unit vectors e j and similarly, a ji are the components
of the unit vectors ei decomposed onto the unit vectors ej . Thus, we have the relations

ei  aij e j and ei  a jiej (3-4)

As an example, we have (see also Figure 3-2): e2  a21e1  a22e2  a23e3 .

Figure 3-2: Coordinate transformation.

Since any vector can be expressed as u  uiei  uiei , the coordinate transformation rules
are obtained as
ui  aij u j and ui  a ji uj (3-5)
In expanded form, Equation (3-5) reads as
u1  a11u1  a12u2  a13u3 u1  a11u1  a21u2  a31u3
u2  a21u1  a22u2  a23u3 and u2  a12u1  a22u2  a32u3 (3-6)
u3  a31u1  a32u2  a33u3 u3  a13u1  a23u2  a33u3
The expanded form of Equation (3-4) is obtained in the same way. It is sometimes useful to
write the transformation rule in matrix format.

3-3
 We will use bold italic letters to indicate a matrix, whereas bold letters are
reserved for vectors and tensors.

With this notation, the coordinate transformation rules for the vector components are
u  au and u  aT u (3-7)
The transformation matrix a is defined as
cos  x1 , x1  cos  x1 , x2  cos  x1 , x3  
 
a  cos  x2 , x1  cos  x2 , x2  cos  x2 , x3  
   
(3-8)
 
cos  x3 , x1  cos  x3 , x2  cos  x3 , x3  
 
It can be shown that the transformation matrix a is a so-called orthogonal matrix, which is a
very important and useful characteristic. To show this, we use Equation (3-5) to write
ui  aij u j  aij akj uk and ui  a jiuj  a ji a jk uk (3-9)

The definition of the Kronecker delta  ij , given in Equation (3-2), implies that

ui   ik uk and ui   ik uk (3-10)

Since u is arbitrary, we get
a ji a jk  aij akj   ik (3-11)

We say that the matrix a with entries aij is orthogonal, i.e.

aT a  aaT  I (3-12)
where I with entries  ij is the identity matrix:

1 0 0 
I  0 1 0  (3-13)
0 0 1 

Example 3-2: Orthonormal base systems (after [4]).

Consider two right-handed, Cartesian coordinate systems defined by two sets of axes, xi
and xi , and the associated two sets of orthonormal base vectors  ei  and  ei  . Let the
transformation matrix a between the two systems be defined by
3 / 5 4 / 5 0 
a   0 0 1 
 a31 a32 a33 

Our task is to determine the three last entries in the transformation matrix a .

3-4
 Using Equation (3-4), we immediately find
e1  a1 j e j  53 e1  54 e2 , e2  a2 j e j  e3

The third base vector e3 of the starred system xi is obtained as

e1 e 2 e3
e3  e1  e2  3
5  54 0   54 e1  53 e2
0 0 1

This definition of e3 ensures that the basis  ei  is orthonormal and that the coordinate system
xi is right-handed. The last row of the transformation matrix is now given as a31   54 ,
a32   53 and a33  0 .
■

3.2 Linear operators (Tensors of 2nd order)

As already mentioned, force and displacement are examples of vectors, while stress and
strain are tensors of 2nd order. A tensor of 2nd order provides a linear operation between two
vectors. A vector will be defined below as a tensor of 1st order, while a scalar (e.g. temperature)
is a tensor of 0th order. From advanced strength of materials, we know that Hooke’s law gives
a linear relation between the stress and the strain of an elastic material. The elastic coefficients
involved in this operation between two tensors of 2nd order constitute a tensor of 4th order. More
generally speaking a tensor is a linear operator that transforms under coordinate transformation
according to a given rule. We start by introducing a 2nd order tensor as a linear operator between
two sets of vectors, and from the results achieved we will state the definition of a tensor of nth
order [2].
An operator L on the three-dimensional vector space is a vector-valued function of a
vector variable
v  Lu (3-14)
where u and v are any vectors. The operator L is said to be a linear operator if
L   au  bv   aL  u  bL  v (3-15)

where a and b are any real numbers. The Cartesian components Lij of any linear operator L
are obtained by projecting the vectors u and v onto the unit vectors
vi  ei  v  ei  L   e j u j    ei  L  e j  u j  Lij u j (3-16)

where per definition

Lij  ei  L  e j (3-17)

Since i, j  1, 2,3 , we understand that, in general, L has 9 independent components. For a

given basis  ei  , a 2nd order tensor can be represented as a 3×3 matrix. The matrix L with

3-5
entries Lij is the component matrix of the linear operator L with respect to the orthonormal
basis  ei 

v  Lu (3-18)
nd
A linear operator as defined above is said to be a tensor of 2 order.
The next question is how the components of L transforms with a change of the coordinate
system. To answer this question, consider two sets of unit vectors,  ei  and  ei  , so that
ei  aij e j . Using the definition of the Cartesian components of L and the relations between the
two sets of base vectors, we get
Lij  ei  L  ej  aik ek  L   a jl el   aik a jl  ek  L  el   aik a jl Lkl (3-19)

Lij  ei  L  e j  akiek  L   alj el   aki alj  ek  L  el   aki alj Lkl (3-20)

where it was made use of the assumed linearity of the operator L . Accordingly, the
transformation rules for the Cartesian components of L read as
Lij  aik a jl Lkl and Lij  aki alj Lkl (3-21)
In matrix form, we get
L  aLaT and L  aT La (3-22)
It is noted that L and L are the component matrices of the same tensor L expressed in the
two coordinate systems defined by the basis  ei  and the starred basis  ei  , respectively.

A linear operator having the same components in any coordinate system is said to be
isotropic. Thus, for isotropic operators we have per definition
Lij  Lij (3-23)
A property of the component matrix of a linear operator which is not changed by transformation
to a new basis is said to be invariant. An invariant property may be considered to be a property
of the linear operator rather than of the component matrix. The transpose is an invariant
property of the component matrix. The transpose of a tensor L is denoted LT and has the
property
LTij  L ji (3-24)
which means that the rows of the matrix L are equal to the columns of the transposed matrix
LT , and vice versa.
Another important property of a linear operator is that it can always be decomposed into
symmetric and anti-symmetric parts, namely
LSij  Lij   12  Lij  L ji  and LijA  Lij   12  Lij  L ji  (3-25)

3-6
where LSij  Lij  is the symmetric part and LijA  Lij  is the anti-symmetric part of the tensor. It
then follows that the tensor can be expressed as
Lij  Lij   Lij  (3-26)

In matrix format, we write

L  LS  LA (3-27)
It is noted that the symmetric part LS has 6 independent entries, while there are only 3
independent entries defining the anti-symmetric part LA .

Example 3-3: Coordinate transformation of a 2nd order tensor (after [4]).

Consider the symmetric tensor L with component matrix in the basis  ei  given by

 3 1 0 
L   1 3 0 
 0 0 1 

Let the starred basis  ei  be given by

e1  e3 , e2  1
2
e1  1
2
e2 , e3   1
2
e1  1
2
e2

The task is to determine the component matrix L in the starred basis.

Using Equation (3-4), we find the transformation matrix as
 0 0 1
 
a   12 1
2
0
 1 
  2
1
2
0 

Note that the rows of a are defined by the components of the starred base vectors ei in the
original basis  ei  . We may now apply Equation (3-22), and we get

3 1 0 0 1
2
1
2

1 3 0 0 1 1
2 2
1 0 0 
 0 2 0 
0 0 1 1 0 0
L  aLa T 
0 0 1 0 0 1 1 0 0
0 0 4 
1 1 2 2
2 2
0 2 2
0 0 2 0
1 1 4 4
2 2
0 2 2
0 0 0 4

It is seen that in the basis  ei  , the component matrix of the tensor L is diagonal, which
indicates that the starred basis has some special significance.
■

3-7
 Finally, we will consider the principal values and principal directions of symmetric
tensors of 2nd order. Consider any symmetric linear operator L with real components, and
define the vector v by v  L  n , where n is a unit vector that defines a direction in three-
dimensional space. If for a given direction n , the vector v is parallel to n , we have
Lij n j   ni (3-28)

Then, n is called a principal direction, or an eigenvector, and  is denoted a principal value,

or an eigenvalue, of the 2nd order tensor L . Using the definition of the Kronecker delta, we get

L ij   ij  n j  0 (3-29)

Or in matrix form:
 L  I  n  0 (3-30)
where I is the identity matrix. This linear system of equations represents 3 equations in the 4
unknowns, n j and  , associated with each principal direction. In addition, we have the
restriction that n is a unit vector  ni ni  1 , which gives us the fourth equation. Accordingly,
we have to solve the following system of equations
 L11    n1  L12 n2  L13n3  0
L12 n1   L22    n2  L23n3  0
(3-31)
L13n1  L23n2   L33    n3  0
n12  n22  n32  1  0

Here we have taken into account the symmetry of L , which implies that Lij  L ji .
Non-trivial solution of Equation (3-29) require
det  Lij   ij   0 (3-32)

where det    denotes the determinant of the matrix argument    . Expansion of Equation
(3-32) gives a cubic equation in  , namely
 3  I L 2  II L  III L  0 (3-33)

where I L , II L and III L are the first, second and third principal invariants of L , respectively,
given by
I L  Lii , II L  12  Lij Lij  I L2  , III L  det  Lij  (3-34)

Equation (3-33) is called the characteristic equation of the 2nd order tensor L . The solutions
1 , 2 and 3 to this equation are the principal values of L . Since L is assumed to be
symmetric with real components, the principal values are real, and if these values are distinct,
the three principal directions n1 , n 2 and n 3 are mutually orthogonal [1]. If 1  2 , any
direction orthogonal to the third principal direction n 3 is a principal direction. The result is that

3-8
any pair of unit vectors n1 and n 2 satisfying n1  n3 , n2  n3 and n1  n 2 are valid principal
directions and lead to the diagonal form of L . If all the three principal values are equal,
1  2  3 , any direction is a principal direction and any orthonormal basis gives the diagonal
form of the component matrix of the tensor L . It is customary to order the principal values
after their magnitude. The principal values are then denoted I , II and III , and are ordered
according to I  II  III .
The component matrix of a tensor L is diagonal in the basis defined by the eigenvectors
n i , i  1, 2,3 , where the components are the eigenvalues Li , i  1, 2,3 . This is established as
follows [2]. The transformation matrix between the two basis systems is given by the direction
cosines aij  ni  e j  n(ji ) , where n (ji ) are the components of the eigenvector n i . Then from the
transformation rule of 2nd order tensors, we obtain the relationship
Lij  aik a jl Lkl  nk(i ) nl( j ) Lkl (3-35)

where Lij are the components of L in the basis  ni  . The eigenvalue problem, Equation (3-29),
implies that
Lkl nl(i )  Li kl nl(i )  Li nk(i ) (no sum on i) (3-36)
Thus
Lij  Li nk(i ) nk( j )  Liij (no sum on i) (3-37)

where it was used that n i are orthonormal.

Example 3-4: Eigenvalues and eigenvectors of a symmetric 2nd order tensor (after [4]).
Consider once more the tensor L defined in Example 3-3. The task is now to show that
the diagonal elements of L are the eigenvalues of L and that the rows of a define the
components of the associated eigenvectors.
Using Equation (3-32) and L from Example 3-3, we get
3   1 0
det  Lij   ij   1 3    1     3     1  0
2
0
 
0 0 1 

The roots of this cubic equation are: 1  1 , 2  2 and 3  4 , and these are thus the principal
values of L . The principal values are seen to be the same as the diagonal elements of L in
Example 3-3.
Based on Equation (3-29), we understand that the principal directions are found by solving
the linear system of equations

3-9
 3   1 0   n1  0
 1 3   0   n2   0

 0 0 1     n3  0

after inserting, in turn, the principal values 1  1 , 2  2 and 3  4 , and applying the
constraint n12  n22  n32  1 . We then get

1  1  2n1  n2  0, n1  2n2  0

2  2  n1  n2  0, n1  n2  0, n3  0
3  4  n1  n2  0, n1  n2  0, 3n3  0
Solving for the eigenvectors, using n12  n22  n32  1 , gives
n1  e3 , n2   1
2
e1  1
2
e 2 , n3  1
2
e1  1
2
e2

We find that n1  e1 , n2  e2 and n3  e3 , and we have shown that the rows of a define
the components of the eigenvectors of L .
■

3.3 Tensors
A linear operator as defined above is an example of the more general notion of a tensor.
The general definition of a tensor is given in terms of its coordinate transformation rule [2]:
A tensor of nth order is a quantity T represented in a basis  ei  by the components
Tk1 ...kn and in another basis  ei  by the components Ti1...in , so that

Ti1...in  ai1k1 ...ainkn Tk1 ...kn (3-38)

where Einstein’s summation convention applies to the indices i1 ,..., in  1, 2,3 and
k1 ,..., kn  1, 2,3 .

From the definition, we see that a scalar is a tensor of 0th order, s*  s , a vector is a tensor of
1st order , vi  aij v j , and a linear operator is a tensor of 2nd order, Lij  aik a jl Lkl .

Henceforth, when it is apparent from the context that all tensor components are
referred to the same orthonormal basis, we will not always distinguish between a
tensor T and its array of components Tk1 ...kn in this basis.

3-10
There are three important results that may readily be verified using the general definition of a
tensor [2]:
 Tensor product. The product M i1 ...im Nk1 ...kn of two tensors M i1 ...im and N k1 ...kn of order m
and n , respectively, constitutes a tensor of order m  n . For instance, if ui and v j are
two vectors, then the product ui v j forms a 2nd order tensor, and if Lij and M kl are two
2nd order tensors, then the product Lij M kl constitutes a 4th order tensor.
 Tensor contraction. By setting any two of the free indices k1 , k2 ,..., kn of an nth order
tensor N k1 ...kn ( n  2 ) equal to each other, thus implying summation over these two
indices, a tensor of order n  2 is obtained. For example, starting from the 4th order
tensor Tijkl , we can construct several different 2nd order tensor by tensor contraction, e.g.
Tiikl , Tijjl and Tijkk .
 Quotient rule. If two tensors M i1 ...im and N k1 ...kn of order m and n , respectively, are
related by M i1 ...im  Ti1 ...imk1...kn Nk1...kn , then Ti1 ...imk1 ...kn is a tensor of order m  n . For
example, if two vectors ui and v j are related by ui  Lij v j , then Lij is a 2nd order
tensor. Similarly, if Lij and M kl are two 2nd order tensors related by Lij  Tijkl M kl , then
Tijkl is a 4th order tensor.
The components of a tensor of any order may clearly and concisely be represented by the
use of index notation. The rules for the index notation are given in the following list.

Rules for index notation

A letter index may occur either once or twice in a given term.
Unrepeated indices are known as free indices (no sum).
Correctly written tensor equations have the same letters as free indices in every
term.
Indices repeated twice are known as dummy indices and the resulting terms should
be summed.
Dummy indices may be replaced by any other letter not appearing as a free index
without changing the meaning of the term.
No index should occur more than twice in a properly written term.
A scalar has no indices attached, a 1st order tensor has one free index, a 2nd order
tensor has two free indices, and so on.

3-11
 The tensor product u  v of two vectors u and v is called a dyad, whereas the tensor
products of several vectors are called polyads. The tensor product is defined to obey all the
rules of ordinary algebra, with one exception: the tensor product is not commutative, i.e.,
u  v  v  u . For instance, the tensor product has the following properties
 u   v   w   u  w   v  w
u   v  w   u  v  w (3-39)
u  v   w   v  w  u
where  and  are real numbers and u , v and w are vectors. The linear combination of
dyads is called a dyadic. Using that u  ui ei and v  vi ei , we find that u  v  ui v j ei  e j ,
where the summation convention applies. In general, we may use the dyadic notation to express
a 2nd order tensor L by
L  Lkl ek  el (3-40)
As before, the components of L is obtained as
ei  L  e j  ei   Lkl ek  el   e j  Lkl  ei  ek   el  e j   Lklik lj  Lij (3-41)

The transformation rule in Equation (3-22) is now readily established by finding the
components of L in the starred basis ei  aij e j as

L*ij  e*i  L  e*j  e*i   Lkl ek  el   e*j  Lkl  e*i  ek  el  e*j   aik a jl Lkl (3-42)

In particular, in the basis defined by the eigenvectors n i , i  1, 2,3 , of the tensor L , the dyadic
notation gives
3
L   Li ni  ni (3-43)
i 1

where Li are the eigenvalues of L . As a rule, we will use index notation instead of the dyadic
notation, except in cases for which the dyadic notation is particularly favourable.
Examples of tensors of 2nd order that we will use frequently in the following chapters are
the stress tensor  ij and the strain tensor  ij . We will find that the stress tensor is a linear
operator giving the traction force ti (force per unit area) as ti   ij n j , where n j is the unit
normal vector to the surface upon which the traction acts. The elastic modulus tensor Cijkl is an
example of a 4th order tensor, which defines the stress-strain relation for a linear elastic material
as  ij  Cijkl  kl . Since 2nd order tensors generally have 9 independent components, the 4th order
tensor will in the general case have 81 independent components.

3-12
3.4 Tensor fields
In a body subjected to external forces, the displacement vector and the stress and strain
tensors depend on the position within the body. They are examples of tensor fields.

A tensor T of nth order whose components Tk1 ...kn depend on the Cartesian
coordinates  x1 , x2 , x3  is said to be a tensor field and denoted T  x  or Tk1 ...kn  x  ,
where x  xi ei is the position vector.

For example, given a basis  ei  , scalar, vector and 2nd order tensor fields are denoted   x  ,
vi  x  and Lij  x  . The partial derivatives of a tensor field T of nth order with respect to the
Cartesian coordinates are denoted Ti1 ...in / xin1 and represent the array of components of a
tensor field of order n  1 . The latter result follows from the quotient rule and the total
 
differential of the tensor field, dTi1 ...in  Ti1 ...in / xin1 dxin1 , where dTi1 ...in and dxin1 represent
th st
tensors of n and 1 order, respectively.
Consider the scalar field   x  and the vector field v  x  defined within a region V in
three-dimensional space, where the boundary of V is a closed surface S with outward unit
normal n . The total differentials of the fields  and v are then defined as
 vi
d  dxi and dv  dx j (3-44)
xi x j
The gradients to these fields are denoted  and v , and given by
 vi
  ei and v  ei  e j (3-45)
xi x j
where dyadic notation was used for clarity. We note that  and v are tensor fields of 1st
order and 2nd order, respectively. In terms of the gradients, the total differentials of the fields
may be written in the form
d    dx and dv  v  dx (3-46)
The gradient  to a scalar field  in two-dimensional space is illustrated in Figure 3-3. It is
notable that the gradient vector  is normal to the level surface   x   const .

3-13
Figure 3-3: The gradient   x  of a scalar field   x  in x1 x2 space.

The divergence of the vector field v is a scalar field and defined by

vi
div v    v  (3-47)
xi
where the del-operator  was introduced as

  ei (3-48)
xi
Note that in terms of the del-operator and the dyadic notation the gradient of a vector field v
can be defined as v     v  . In the same manner, the divergence of a 2nd order tensor field
T

L  x   Lij  x  ei  e j is a vector field defined by

 L
  L  ei   Lkl ek  el   kl el (3-49)
xi xk
An important theorem is Gauss’ theorem, which relates a volume integral over the domain
V to a surface integral over the closed surface S . The theorem states that [2]

V
 x i
dV   ni dS
S
(3-50)

where   x  is any differentiable tensor field (i.e., the theorem applies to scalar, vector and
tensor fields), and S is the boundary of V with outward unit normal n  ni ei . An important
special case is the divergence theorem, viz.

3-14
 vi
 x
V i
dV   ni vi dS
S
(3-51)

where vi  x  is any differentiable vector field.

Example 3-4: Level surface and gradient (after [5]).

Let a surface S in three-dimensional space be given by the equation
  x   const.

where   x  is a scalar field. In fact, this equation represents a family of surfaces that are called
the level surfaces of the scalar field   x  . We now define a curve C in space by

x  t   xi  t  ei

where t is any parameter, and require that C lies on the surface S . Thus, the functions xi  t 
must fulfil
 x  t   const.
Then, the chain rule gives

 xi    x  0
xi
where a superimposed dot implies derivation with respect to t . The vector
1
x  t   xi  t  ei  lim  x  t  t   x  t 
def

t 0 t 

is tangent to the curve C . We may now consider a family of curves on S passing through a
given point P in various directions. The tangents of these curves at P will lie in the same plane,
having a normal vector that is collinear with the gradient vector  at P . This plane is denoted
the tangent plane of the surface S at point P .
■

3-15
3.5 Isotropic functions*
Let Lij be a symmetric, 2nd order tensor with principal values L1 , L2 and L3 . The
principal invariants of Lij defined in Equation (3-34) can then be expressed in terms of the
principal values as
I L  Lkk  L1  L2  L3
II L  1
2 L
kl Lkl  I L2   1
2 L  L  L L  L  L  
2
1
2
2
2
3 1 2 3
2
(3-52)
III L  det  Lkl   L1L2 L3
Other combinations of these invariants are clearly also invariants. The following alternative
invariants are especially convenient for our purpose [1]
I1  I L  Lkk  L1  L2  L3
I 2  II L  12 I L2  12 Lkl Lkl  1
2 L 2
1  L22  L23  (3-53)
I 3  13 I L3  I L II L  III L  13 Lkl Llm Lmk  13  L13  L32  L33 

The two first of these relations follow immediately, while the last one is more tedious but quite
straightforward to derive. It is left as an exercise. We will also need the partial derivatives of
these alternative invariants. These are obtained as follows [6]:
I1
  ki kj   ij
Lij
I 2 1
 2  ki lj Lkl  12 Lkl ki lj  Lij (3-54)
Lij
I 3 1
  ki lj Llm Lmk  13 Lkl li mj Lmk  13 Lkl Llm mi kj  Lik Lkj
Lij 3

where the symmetry of Lij was utilized. The alternative invariants of the deviatoric part of the
tensor Lij , namely Lij  Lij  13 Lkk ij , are readily obtained as

I1  Lkk  0
I 2  12 Lkl Lkl  1
2  L  L  L 
1
2
2
2
3
2
(3-55)
I 3  13 Lkl Llm
 Lmk
   L  L  L 
1
3 1
3
2
3
3
3

where L1 , L2 and L3 are the principal values of Lij .

A scalar-valued function f of the tensor Lij is said to be isotropic if it is expressed in

terms of the invariants of the tensor, e.g.
f  f  I1 , I 2 , I 3  (3-56)

The partial derivatives to this function is obtained as

3-16
 f 3
f I f f f
   ij  Lij  Lik Lkj (3-57)
Lij  1 I Lij I1 I 2 I 3

If the isotropic function f depends only on the deviatoric part of the tensor Lij , we have
instead
f  f  I 2 , I 3  (3-58)

The partial derivatives of f with respect to Lij are then given by

f f I 2 f I 3 f f
   Lij  Lik Lkj (3-59)
Lij I 2 Lij I 3 Lij I 2 I 3

More often, we search for the partial derivatives of f with respect to Lij , and we get

f f Lkl f   Lkl  13 Lmm kl  f 1 f

     ij (3-60)
Lij Lkl Lij Lkl Lij Lij 3 Lkk
where we used that
  Lkl  13 Lmm kl    12  Lkl  Llk   13 Lmm kl 
  12  ik jl   il jk   13  ij kl (3-61)
Lij Lij

We note that f / Lij is a deviatoric tensor when f depends only on the deviatoric part of the
tensor Lij , while f / Lij is generally not deviatoric.

It follows from Equation (3-57) that the gradient f / Lij of the isotropic function f ,
which is a symmetric 2nd order tensor, has the same eigenvectors as the tensor Lij [6]. This is
shown as follows. Let ni be an eigenvector of the tensor Lij , so that Lij n j   ni , where  is a
corresponding eigenvalue (either L1 , L2 or L3 ). We then get
f f f f f f f
nj   ij n j  Lij n j  Lik Lkj n j  ni   ni   Lik nk
Lij I1 I 2 I 3 I1 I 2 I 3
(3-62)
f f f 2  f f f 2 
 ni   ni   ni       ni
I1 I 2 I 3  I1 I 2 I 3 

It follows that ni is also an eigenvector for the gradient to the isotropic function, while the
associated eigenvalue is the expression in the brackets.

3-17
3.6 References
[1] F. Irgens. Continuum Mechanics. Springer, 2008.
[2] J. Lubliner. Plasticity Theory. Macmillan, 1990.
[3] L.E. Malvern. Introduction to the Mechanics of a Continuous Medium. Prentice-Hall,
1969
[4] G.E. Mase. Continuum Mechanics. Schaum’s Outline Series. McGraw-Hill, 1970.
[5] E. Kreyszig. Advanced engineering mathematics. 5th edition. Wiley, 1983.

[6] N. Saabye Ottosen. Aspects of Constitutive Modelling. Report TVSM-3010. Division of

Structural Mechanics. Lund Institute of Technology, Sweden, 1987.

3-18
4 Continuum mechanics
In this chapter, we formulate the kinematics and kinetics of a continuum body loaded
quasi-statically by external forces. It is assumed that the deformations of the continuum body
are infinitesimal, and that there is no need to distinguish between its undeformed and deformed
configurations. This considerably simplifies the mathematical formulation of the continuum
mechanics. A short description of the finite element method for problems with material
nonlinearity is also provided at the end of the chapter.

4.1 Displacement, strain and rotation

Figure 4-1: Continuous body loaded with body forces bi and traction forces ti , where V is the
volume of the body and S is the surface.

We consider a body occupying the volume V in x1 x2 x3 space, where the boundary of V

is a closed surface S with outward normal n . A body is said to be a continuum if to any
configuration of the body every point of the volume V is occupied by a particle (material point).
The reference configuration of the body is taken as a reference to which displacements are
measured. Any configuration may be taken as the reference configuration.
We introduce a Cartesian coordinate system with coordinate axes  x1 , x2 , x3  and unit
vectors  e1 , e2 , e3  , and will express all tensor components in the orthonormal basis  ei  if
nothing else is said. Assuming that the loading process is quasi-static, the acceleration of the
body is negligible and thus inertia forces may be discarded. To further simplify the description,
we will assume infinitesimal deformations of the body.

4-1
 x3

u
x
e3
x
e2
e1
x2
x1

Figure 4-2: Position vector x in the reference configuration (dashed line), position vector x in
the deformed configuration (solid line) and displacement vector u .

Let x  xi ei be the position vector of a particle in the reference configuration of the body,
or the material coordinate. It is thus understood that x refers to a fixed particle or material point.
The position vector of the same particle after displacement of the body, or the spatial coordinate,
is denoted x  xi ei . The displacement field u  u  x,t  for the particle located at x in the
reference configuration is
ui  ui  x, t   xi  x, t   xi (4-1)

where t is the time. The position vectors in the reference and deformed configurations as well
as the displacement vector are defined in Figure 4-2.
It is assumed that the displacement field is sufficiently smooth, so that all the mathematical
operations are legitimate. The velocity field v  v  x,t  is then defined as the rate of change of
the displacement of a given particle, i.e.
ui  x, t 
vi  vi  x, t    ui  x, t  (4-2)
t
Since the displacements are assumed to be infinitesimal, it is not strictly necessary to distinguish
between the un-deformed and deformed configurations of the body.

4-2
Figure 4-3: Displacements of two points in close proximity with the same ordinate (after [5]).

The local deformation and rotation are described by the spatial variation in the
displacement field around a material particle, which is represented by the displacement gradient
u . This is illustrated in Figure 4-3, showing the displacements of two points having the same
coordinate x2 in the initial configuration. The relative displacement dui between two material
points at distance dx j is defined by

ui
dui  dx j (4-3)
x j
Expanding Equation (4-3), we get
u1 u u
du1  dx1  1 dx2  1 dx3
x1 x2 x3
u2 u u
du2  dx1  2 dx2  2 dx3 (4-4)
x1 x2 x3
u3 u u
du3  dx1  3 dx2  3 dx3
x1 x2 x3
To ensure small deformations of the body, it is necessary to assume that u should be
small as measured in some norm.

4-3
Figure 4-4: Illustration of the infinitesimal strains and rotations in the two-dimensional case.

From Equation (4-3), it is understood that u is a tensor of 2nd order (or a linear operator),
since it transforms the infinitesimal line segment dx (a vector) into the relative displacement
du (another vector) and the transformation is linear.
The infinitesimal strain tensor  ij and the infinitesimal rotation tensor ij are then defined
by

1  ui u j  1  u u j 
 ij     , ij   i   (4-5)

2  x j xi  2  x j xi 

It is noted that  ij is symmetric,  ij   ji , while ij is skew-symmetric, ij   ji , and that

ui
  ij  ij (4-6)
x j
An illustration of the components of the infinitesimal strain and rotation tensors is given in
Figure 4-4 for the two-dimensional case. It is noted that while the symmetric infinitesimal strain
tensor has 6 independent components, the skew-symmetric infinitesimal rotation tensor has
only 3 independent components. The 6 independent components of the infinitesimal strain
tensor are

u1 u2 u3

11   22   33 
x1 x2 x3
(4-7)
1  u u  1  u u  1  u u 
12   1  2   23   2  3   31   3  1 
2  x2 x1  2  x3 x2  2  x1 x3 
where the first row gives the normal strains and the second row the shear strains. Note that the
engineering shear strains are usually defined as  12  212 ,  23  2 23 and  31  2 31 . In the

4-4
absence of rotation, ij  0 , the relative displacements depend only on the strains and are
expressed as
du1  11dx1  12 dx2  13dx3
du2  12 dx1   22 dx2   23dx3 (4-8)
du3  13dx1   23 dx2   33dx3
Similarly, the non-zero components of the infinitesimal rotation tensor are given as
1  u1 u2  1  u2 u3  1  u3 u1 
12  21     , 23  32     , 31  13     (4-9)
2  x2 x1  2  x3 x2  2  x1 x3 

In the absence of strain,  ij  0 , the relative displacements consist only of rotation, and we have

du1  12 dx2  13dx3 , du2  12 dx1  23dx3 , du3  13dx1  23dx2 (4-10)
Since the strain tensor is a symmetric tensor of 2nd order, the principal values, denoted the
principal strains, 1 ,  2 ,  3 , and the principal invariants, I , II , III , are readily obtained as
described in Chapter 3.2.
The infinitesimal volumetric strain  v (or the dilatation) is defined (in a similar way as the
normal strain) as
dV  dV0
v  (4-11)
dV0
where dV0 and dV are the initial and final volume of the infinitesimal material element under
consideration. Assume that x1 , x2 , x3 are the principal axes of strain. A rectangular element with
edges of initial dimensions dx1 , dx2 , dx3 , has after the infinitesimal deformation edges of length
1  1  dx1 , 1   2  dx2 , 1   3  dx3 . Since the edges are along the principal direction of the
strain tensor, the edges are still mutually perpendicular, see Figure 4-5.

4-5
Figure 4-5: Volumetric deformation (or dilatation) of an infinitesimal parallelepiped cut out
from the solid body.

Using the definition in Equation (4-11), the volumetric strain is obtained by [3]
 v  1  1 1   2 1   3   1 (4-12)
and by neglecting terms involving products of strains (assuming infinitesimal strain
components), we get
 v  1   2   3  I   kk (4-13)

Since the result is an invariant, the volume strain  v equals the sum of the normal strains
 kk  11   22   33 in any system of axes.
The strain deviator  ij is a 2nd order tensor whose components are defined by

 ij   ij  13  v ij (4-14)
The strain deviator has the property that the sum of its normal components, or the dilatation, is
identically equal to zero:  ii  11   22
   33
   ii  13  vii  0 , because  ii   v and  ii  3 . This
implies that the strain deviator describes a state of strain with zero dilatation. The principal
invariants of the strain deviator will be denoted I , II , III , where the first invariant is
identically equal to zero: I  0 .
The behaviour of viscoelastic and elastic-viscoplastic materials depends not only on the
straining of the body, but also of the rate of straining. It is therefore necessary to define the
strain rate tensor  ij and rate of rotation (or spin) tensor ij . These tensors are readily obtained
by time differentiation of the strain and rotation tensors,  ij   ij (x, t ) and ij  ij (x, t ) , where

4-6
it is underlined that x refers to a fixed particle or material point. It is the coordinate of the
particle in the reference configuration of the body. From Equation (4-5), we accomplish
 ij1  v v j  ij 1  vi v j 
 ij    i   , ij      (4-15)
t 2  x j xi  t 2  x j xi 

where we have used that  / t  ui / x j    / x j  ui / t   vi / x j . In plasticity it is

convenient to work with an incremental formulation, since the relations between stress and
strain are nonlinear but independent of strain rate. We define the incremental strain tensor as
d ij   ij dt and the incremental rotation tensor as dij  ij dt , where dt is an infinitesimal
time increment. It is noted that even if the strain and rotation are assumed to be infinitesimal
their rates may indeed be finite. On the contrary, the strain and rotation increments are
infinitesimal quantities.

Example 4-1: Extension and simple shear.

Let the displacement field of a continuum body be defined by

u1  x, t     t   1 x1   (t ) x2 , u2  x, t    t  1 x2 , u3  x, t    t  1 x3

 12  12
 
where   l / l0 is the stretch ratio and  is the shear strain, both being time dependent
parameters, see Figure 4-6. The stretch ratio  is unity and the shear strain  is zero in the
undeformed configuration. It is assumed that  becomes only slightly greater than unity and
that  stays infinitesimal throughout the deformation process. This way the displacements
remain small. The assignment is to determine the strain and rotation tensors and the volumetric
strain, and then to interpret the prescribed deformation.
Using Equation (4-7), we obtain the strain components
11    1,  22   33     1, 12   / 2,  23   31  0
1
2

while the three independent components of the infinitesimal rotation tensor are (cf. Equation
(4-9))
12   / 2, 23  31  0
Recall the symmetry of the strain tensor,  ij   ji , and the skew-symmetry of the rotation
tensor, ij   ji . The volumetric strain is defined by Equation (4-13), and we get

 v    2  3 . By inspection we find that  v is approximately zero for values of  very
1
2

close to unity. This can be explained if we use instead Equation (4-12) to calculate the
volumetric strain, where 2nd order terms are accounted for. This gives
 v         1  0
1 1
2 2

If we account for the assumption that  remains close to unity, we may use a linear
approximation of  2 around unity. The linear approximation of a function f  x  around
1

4-7
x  a is given by the first two terms of a Taylor series, i.e. f  x   f  a   f   a  x  a  , and
thus  2  1  12    1 for  close to 1. Using this linear approximation, we obtain the
1

transverse normal strains as

 22   33   12    1
whereas the volumetric strain, as defined by Equation (4-13), becomes exactly equal zero.
We conclude that the prescribed deformation represents a volume-preserving (or isochoric)
axial elongation in the x1 direction coupled with simple shear in the x1 x2 plane. Applying this
deformation field to a square material element with initial dimensions l0  l0 , it deforms as
indicated in Figure 4-6. Note, however, that the thickness of the material element is reduced in
accordance with  33   12    1 during this deformation process.

Figure 4-6: Simple deformation field consisting of volume-preserving extension and shearing,
where   l / l0 is the stretch ratio and   tan  is the shear strain.

4.2 Stress and static equilibrium

The body is subjected to body forces b over its volume and traction forces t over its
surface S (see Figure 4-1). Static force equilibrium and moment equilibrium about the origin
requires that the resultant force and moment vectors are zero, i.e.,

 bdV   tdS  0
V S
(4-16)

 b  xdV   t  xdS  0
V S
(4-17)

The external forces lead to internal forces in the body. Consider an infinitesimal material
element which is cut out from the body and has edges parallel to the coordinate axes. Let t (i )

4-8
denote the three traction vectors acting on the three planes perpendicular to the coordinate axes
xk . These internal forces are required to keep the body in static equilibrium.

Figure 4-7: Illustration of the components of the stress tensor.

Let  ij denote the 9 components of the traction vectors t (i ) , so that

t (1)   11e1   12e2   13e3 


t (2)   21e1   22e2   23e3  or t (i )   ij e j (4-18)
t (3)   31e1   32e2   33e3 

The components  ij are positive numbers when the vector components of t (i ) have the positive
senses as illustrated in Figure 4-7. The vector components on the negative sides of the material
element have senses opposite to those on the associated positive sides.
It will be shown below that  ij are the components of a symmetric tensor σ of 2nd order,
which will be called the stress tensor. It is a linear operator that determines the traction vector
t (n )  σ  n acting on a surface with normal n . In component form, the relation reads
ti(n )   ij n j (4-19)
To prove Equation (4-19), we consider a free body in the form of a triangular pyramid (or
a tetrahedron) which is bounded by the three coordinate planes through the origin O , and an
inclined plane ABC , which is not passing through the origin, see Figure 4-8. The plane ABC

4-9
has outward unit normal n and is acted upon by a traction vector t (n ) . From the geometry of
the triangular pyramid, we find that the components of the unit vector n are given by
n1  cos   AON  , n2  cos   BON  , n3  cos   CON  (4-20)
From Figure 4-8 we obtain the following geometrical relation between the surface
elements S  12 AC  PB and S2  12 AC  PO :

S2  12 AC  PO  12 AC  PB cos  S cos  n2S (4-21)

where we have used that PO  PB cos and BON  BPO   . Similar relations can be
established for the other surface elements, so that
S1  n1S , S2  n2 S S3  n3S or Si  ni S (4-22)
The volume of the triangular pyramid can be expressed by the following four equivalent
expressions
V  13 hS  13 OA S1  13 OB S2  13 OCS3 (4-23)
Finally, the length h of the line segment ON is obtained as
h  OAn1  OB n2  OC n3 (4-24)

B B

O N A N A
O θ
C C P

Figure 4-8: Cauchy’s tetrahedron (after [8]).

4-10
 Static equilibrium requires that
3
0   bdV   tdS  bV  t (n ) S   t (i ) Si (4-25)
V S i 1

Using the geometric relations for the triangular pyramid, we obtain

3
1
3 hb  t (n )   t (i ) ni  t (i ) ni (4-26)
i 1

By letting h approach zero, we obtain the vector equation

t (n )  t (i ) ni   ij ni e j (4-27)
where Equation (4-18) was used to establish the latter equality. The corresponding algebraic
equations for the components of the traction vector t (n ) reads
t1(n )   11n1   21n2   31n3 

t2(n )   12 n1   22 n2   32 n3  or ti(n )   ji n j (4-28)
t3(n )   13n1   23n2   33n3 

This proves that  ij is a linear operator and thus a tensor of 2nd order.
Figure 4-9 shows the forces acting on an infinitesimal material element cut out of the
material. The edges of the infinitesimal element are parallel to the coordinate axes. Force
equilibrium gives the following three equations:
 11  21  31
   b1  0
x1 x2 x3
 12  22  32
   b2  0 (4-29)
x1 x2 x3
 13  23  33
   b3  0
x1 x2 x3
Further, it is found from Figure 4-9 that moment equilibrium implies that the shear stresses on
perpendicular planes are equal in magnitude:
12   21 ,  23   32 ,  31  13 (4-30)
This means that the stress tensor has to be symmetric. In summary, the force and moment
equilibrium equations for the body can be expressed as
 ij
 b j  0 and  ij   ji (4-31)
xi
These equations are called the local form of the equilibrium equations, while Equations (4-16)
and (4-17) are denoted the global form of the equilibrium equations.

4-11
 x

Figure 4-9: Static equilibrium of infinitesimal material element.

The symmetry of the stress tensor implies that Equation (4-28) can be re-written as
t1(n )   11n1   12 n2   13 n3 

t2(n )   12 n1   22 n2   23 n3  or ti(n )   ij n j (4-32)
t3(n )   13n1   23n2   33n3 

It is noted that the local form of the force equilibrium equations may readily be obtained
from Equation (4-16). Inserting the expression for the traction force from Equation (4-32), we
may write the global force equilibrium equation as

 b dV   
V
j
S
n dS  0
ij i (4-33)

Use of Gauss’ theorem (see Chapter 3.3), leads to

 ij   ij 
 b j dV  
V V
xi
dV   
V
xi
 b j  dV  0

(4-34)

Assuming that Equation (4-16) is valid for any part of the body, however small, we obtained
the desired local equilibrium equations (cf. Equation (4-29))
 ij
 bj  0 (4-35)
xi

4-12
Using a similar procedure, it is possible to start from Equation (4-17) and show that global
moment equilibrium implies symmetry of the stress tensor.
Now, since it has been shown that σ is a tensor of 2nd order, all the results derived in
Chapter 3.2 are valid. In particular, the principal values will be denoted 1 ,  2 ,  3 , while the
principal invariants are written I , II , III . Figure 4-10 shows the components of the stress
tensor when it is represented in the basis defined by the orthonormal eigenvectors (n1 , n2 , n3 ) .
In this particular basis, the shear stresses are all zero, and the normal stresses are equal to the
principal values, which are denoted the principal stresses.
It is important to realize that the transformation rules for the stresses under change of the
coordinate system are immediately obtained in the form
 ij  aik a jl kl (4-36)

where a is the transformation matrix defined in Chapter 3.1. Like for the strain, we also
introduce the stress deviator σ and the hydrostatic stress  H by
 ij   ij   H  ij and  H  13  kk  13 I (4-37)

Figure 4-10: Components of the stress tensor in the basis defined by the orthonormal
eigenvectors  n1 , n2 , n3  .

The hydrostatic stress defines the isotropic part of the stress tensor, namely  H  ij , and
P   H represents the pressure of the stress state. The first principal invariant of the stress
deviator I is identically equal to zero. The stress deviator plays an important role in metal
plasticity, because it turns out that yielding in metals is practically independent on the
hydrostatic stress (or equivalently on the pressure). Hence, the yield conditions and flow rules

4-13
in theories of metal plasticity most often depend only on the stress deviator. In matrix form, the
stress deviator reads
 11  12  13   11   H  12  13 
σ    12  22      12
  23
   22   H  23  (4-38)
 13  23
  33
    13  23  33   H 
It is noted that the shear stress components (off-diagonal entries) are the same for σ and σ ,
while the normal stress components are different in the two tensors. The principal invariants of
the deviatoric stress tensors, denoted I  J1  0 , II  J 2 and III  J 3 , will be frequently
used in Chapter 7 on metal plasticity. Since the first invariant of the stress deviator σ equals
zero, we have according to Equation 3-36:

J 2   ij ij , J 3  det  ij 

1
(4-39)
2
Two important results for the stress tensor are [1]:
 The maximum and minimum principal stresses are extremal values for the normal
stress:  max   I and  min   III , when it is assumed that  I   II   III .
 The maximum shear stress acts on planes that are inclined 45° with respect to the
principal directions associated with the maximum and minimum principal stresses, and
its magnitude is  max  12  max   min  .
The maximum shear stress is important in theory of plasticity, because it forms the basis
for the Tresca yield criterion, which states that yielding starts when  max reaches a critical value.

Example 4-2: Stresses in isochoric extension and simple shear of a linear elastic material.
In an isotropic, linear elastic material (cf. Chapter 5 for details), the relation between the
stress and strain tensors is defined by the generalized Hooke’s law in the form
 ij  K v ij  2G ij
where K and G are respectively the bulk and shear modulus. Assume that the loading
parameters in Example 4-1 have the values   1.005 and   0.004 , and that the material is a
high-strength steel with K  160 GPa and G  79.3 GPa . Calculate the stress components due
to the given deformation, and find the principal stress and the maximum shear stress.
Since we found in Example 4-1 that the volumetric strain was zero, only the strain deviator
contributes to the stress, which accordingly will be deviatoric. Thus, the stress components are
found to be (cf. Example 4-1)
11  2G    1 ,  22   33  2G  12    1 , 12  G ,  23   31  0
where the linear approximation has been used in the expression for the transverse normal strains.
With the given values, the non-zero stress are
11  793.0 MPa,  22   33  396.5 MPa, 12  317.2 MPa

4-14
 Using Equation (3-32), the principal stresses are determined by finding the three roots of
the cubic equation
 396.5     793.0    396.5     317.22   0
The principal stresses are  I  872.3 MPa ,  II  396.5 MPa and  III  475.8 MPa , while
the maximum shear stress is then given by  max  12  I   III   674.1 MPa .
■

4.3 Balance of mechanical energy

When the body is deformed under the action of external forces, work will be done by these
external forces over the displacements while the internal forces (or stresses) will perform work
over the strains. An important result, which we will prove in the following, is that in a quasi-
static deformation process, the work per unit time of the external forces equals the work per
unit time of the internal forces. This is a statement of the balance of mechanical energy. Work
per unit time is denoted power, and the work per unit time of the internal forces is herein called
the deformation power (or stress power).
Starting from Equation (4-31), multiplying both sides of the equation with the velocity,
and integrating over the volume V of the body, gives
 ij
 x
V i
v j dV   b j v j dV  0
V
(4-40)

We note that by the rules of differentiation

  ij v j   ij v j
 v j   ij (4-41)
xi xi xi
and accordingly Equation (4-40) may be re-written as
  ij v j  v j

V
xi
dV    ij
V
xi
dV   bi vi dV  0
V
(4-42)

Gauss’ theorem (see Chapter 3.3) gives us

  ij v j 

V
xi
dV    ij v j ni dS   t j v j dS
S S
(4-43)

where the definition of the stress tensor provided by Equation (4-19) was employed. Combining
(4-42) with (4-43), we obtain
v j
  ij
V
xi
dV   bi vi dV   ti vi dS
V S
(4-44)

Using the fact that  ij   ji (the symmetry of the stress tensor) in combination with (4-15), we
obtain the mechanical energy balance as

4-15
 Pd    ij ij dV   bi vi dV   ti vi dS (4-45)
V V S

The right-hand side of Equation (4-45) represents the power (or work per unit time) of the
external forces, while the left-hand side defines the deformation power (or the stress power) Pd ,
which is interpreted as the power of the internal forces. This is illustrated in Figure 4-11. It is
found that the work per unit time of the internal forces on the infinitesimal material element
reads as
 dx   dx   dx 
2 11  11 1  dx2 dx3  2 12  12 1  dx2 dx3  2 13  13 1  dx2 dx3 
 2   2   2 
 dx   dx   dx 
2 21   21 2  dx3 dx1  2 22   22 2  dx3 dx1  2 23  13 2  dx3 dx1  (4-46)
 2   2   2 
 dx   dx   dx 
2 31   31 3  dx1dx2  2 32   32 3  dx1dx2  2 33   33 3  dx1dx2   ij ij dV
 2   2   2 
The external forces do work on the body when Pd  0 (energy is absorbed by the body), while
the body does work on its surroundings when Pd  0 .
The stress power per unit volume is now defined by
d   ij ij (4-47)

so that Pd   d dV . It is sometimes useful to split d into distortional (deviatoric) and

V

volumetric parts, denoted d and do . To this end, we write

d   ij ij   ij   H  ij  ij  13  v ij    ij  ij   H  v (4-48)

where  ij is the deviatoric strain rate tensor and  v is the volumetric strain rate. The last
equality is established by applying that  ij ij   ij ij  0 per definition of the deviators and
further that  ii  3 . Accordingly, the stress power per unit volume can be re-written as

d   ij ij  d  do with d   ij  ij and do   H  v (4-49)
A deformation is said to be isochoric if the volume is conserved, so that the volumetric strain
rate and do are both zero.

4-16
Figure 4-11: Section through an infinitesimal element of the solid body to help in the
interpretation of the deformation power (after [5]).

In a reversible deformation process, the stress power integrated over a closed deformation
cycle vanishes. In this case, the integral
t2 t2

 d dt    ij ij dt
t1 t1
(4-50)

is path independent and should be a unique function of the integrated deformations (or strains).
The result is that there exists a strain energy function U 0  U 0 (ε) , defined so that
U 0  d   ij ij . This is the case for hyperelastic materials, and for these materials the stress
power is completely stored as elastic strain energy.
On the contrary, in a fully dissipative deformation process (e.g. as in rigid-perfect
plasticity) it is required that work has to be performed continually on the body for the
deformation to continue, i.e.,
d   ij ij  0 (4-51)
Equation (4-51) expresses the local dissipation, i.e., the loss of mechanical energy per unit time
and per unit volume. The lost energy is converted into heat, raising the temperature of the
system. In most cases, as for hardening elastic-plastic materials, a large part of the stress power
is dissipated and the rest stored in the material as elastic strain energy.

4-17
Example 4-3: Stress power and strain energy in isochoric extension and simple shear of a
linear elastic material.
We consider again the linear elastic material subjected to isochoric extension and simple
shear. We want to compute the stress power d and the strain energy function U 0 per unit
volume in terms of the stretch ratio  and the shear strain  , using the stress-strain relation
from Example 4-2 and the prescribed deformation mode from Example 4-1.
By means of the results from Example 4-1, the components of the strain rate tensor are
obtained as
11   ,  22   33   12  , 12   / 2,  23   31  0
where we have adopted the linear approximation in the expression for the transverse strain
components. Making use of Equation (4-47) and the stress components found in Example 4-2,
this gives
d   1111   22 22   33 33   1212   21 21
 
 2G    1   2  2G   12    1    12   2G    / 2 
 3G    1   G
It is important to remember that the shear stresses and strains operate in pairs, and thus we
must include both contributions  1212 and  21 21 in the stress power. The strain energy is
obtained by integration
t t
U 0   d dt   3G    1   G  dt
0 0

 

 3G     1 d   G   d  
 1 0 
 32G   2  2  1  G2  2

Inserting the values for the shear modulus, the stretch ratio and the shear strain given in
Example 4-2, we get U 0  3.61 MPa . In this case, 82% of the strain energy is due to the normal
strains and 18% due to the shear strains.
■

4.4 Principle of virtual work

The finite element method, which we will describe in the next section, is conveniently
derived based on the principle of virtual work. We will show in the following that the principle
of virtual work is equivalent to the equilibrium equations and the traction boundary conditions.
The principle of virtual work states that the internal virtual work equals the virtual work
of the external forces for a body in equilibrium, i.e.,

  
V
ij ij dV   bi ui dV   ti ui dS
V S
(4-52)

4-18
where  ij is the virtual strain tensor and  ui is the virtual displacement vector. The relation
between  ij and  ui is defined by

1   ui  u j 
 ij     (4-53)
 2  x j xi 
The virtual displacement field is assumed to fulfil the kinematic boundary condition
 ui  0 on Su (4-54)

where Su is the part of the surface S with prescribed displacement. It is assumed that the
boundary has two parts: Su where the displacement is prescribed and St where the traction is
prescribed, so that S  Su  St and Su  St   . Apart from this, the virtual quantities are
arbitrary.
To derive the equilibrium equations, we proceed as follows. Using the symmetry of the
Cauchy stress tensor and the integral theorems on the left-hand side integral in Equation (4-52)
leads to
 ui
  
V
ij ij dV    ij
V
x j
dV

 ij   ij ui 
   ui dV   dV (4-55)
V
x j V
x j
 ij
   ui dV    ij ui n j dS
V
x j S

Combining Equations (4-52) and (4-55) gives

  ij 
  x j   ui dV  S ti   ij n j   ui dS  0
V
  bi (4-56)

The virtual displacement  ui is arbitrary, except on Su where it vanishes. The results is that
 ij
 bi  0 in V , ti   ij n j on St (4-57)
x j

where ti is the prescribed traction on St . These two equations represent the equilibrium
equations and the traction boundary conditions.
All the arguments used here to derive the equilibrium equations and the traction boundary
conditions are reversible. It is thus possible to start out with the equilibrium equations and the
traction boundary conditions, multiply them with the virtual displacement, integrate them over
their respective domains, and then use the integral theorems to obtain the principle of virtual
work. In conclusion, the principle of virtual work is equivalent to the equilibrium equations and
the traction boundary conditions.

4-19
4.5 Finite element method for material nonlinearity
The finite element (FE) method is a numerical method for solving boundary value
problems (BVP), i.e., problems that can be formulated by a set of partial differential equations
and associated boundary conditions. The method is formulated from the weak form of the BVP,
which in our case is the principle of virtual work. Using interpolation, we may then derive the
semi-discrete form of the equilibrium equations from the principle of virtual work. These
equations are nonlinear, due to the nonlinear behaviour of the material, and may be solved by
the Newton-Raphson method, which is an incremental-iterative method. The mathematical
formulation of finite elements will not be covered here and the reader is referred to the textbooks
[6]‒[9].

4.5.1 Non-linear equilibrium equations

Continuous problem Discrete problem

6 5 4

(2) (4)

(1) (3)

1 2 3

(3)
1 2

Figure 4-12 Discretization of continuous mechanical problem into finite elements and nodes.

In the finite element method, the structure (or solid body) is first divided into elements and
nodes, as shown in Figure 4-12. This process is called discretization. The elements are
connected at the nodes, and continuity is usually assumed along the element borders. Let ne be
the number of elements, m the number of nodes per element, and nN the number of nodes in
the discretized structure [8]. We will here only consider elements where the number of degrees
of freedom at each node equals the space dimension nSD , which means that structural elements

4-20
like beams, plates and shells are not included. For a three-dimensional problem nSD  3 , for a
two-dimensional (or plane) problem nSD  2 and for a one-dimensional problem nSD  1 . In the
following, we will mainly assume nSD  3 , i.e., the three-dimensional problem is considered.
The displacements are interpolated within the elements as [8]
u  x, t   N e  x  ve  t  , x Ve (4-58)

where u  u1 u2 u3  is the displacement vector, x   x1 x3  is the position vector,

T T
x2
Ve is the element volume, v e is the nodal displacement vector of the element with dimension
nSD m1 , and N e is the matrix of interpolation (or shape) functions with dimension nSD  nSD m .
The interpolation functions are typically polynomials and are also called shape functions. The
relation between the degrees of freedom of an element and the degrees of freedom of the
structure is defined as
ve  t   ae r  t  , e  1, 2,..., ne (4-59)

where the global nodal displacement vector r has dimensions nSD nN 1 and the connectivity
matrix ae has dimensions nSD m  nSD nN , see Figure 4-13. The global displacement field is then
obtained by summing the contribution from each element, i.e.
ne
 ne 
u  x , t    N e  x  v e  t     N e  x  ae  r  t   N  x  r  t  , x  V (4-60)
e 1  e1 
where the global shape function matrix N has been introduced as
ne
N  x    N e  x  ae , x V (4-61)
e 1

The dimension of N is nSD  nSD nN . The shape function matrix N is called global because it
contains the shape functions for the entire structure. The shape functions for each and every
element of the structure are contained in N e . We will assume here that the element shape
functions fulfil continuity and completeness requirements which are necessary for convergent
solutions with the finite element method [6][7].

4-21
 6 5 4

(2) 3
(4)

(1) (3)
(3)
1 2

1 2 3

Figure 4-13 Connectivity between element nodes and structure nodes as defined by the
connectivity matrix.
The symmetry of the stress and strain tensors allows us to express them as column matrices
(or vectors) in Voigt notation as
 11  11 
   
 22   22 
 33   
σ    and ε   33  (4-62)
 23   23 
 31   31 
   
 12   12 

where  23  2 23 ,  31  2 31 and  12  212 are the engineering shear strains. Note that with this
definition the deformation power can be expressed as
Pd    ij ij dV   εT σdV (4-63)
V V

The relation between the strain and the displacement vectors is defined as
ε  u, x V (4-64)
where  is a differential operator:

4-22
  x1 0 0 
 

0 x2 0 
  
0 0 x3 
   
(4-65)
0
 x3 x2 
 0  
 x3 x1

 x 
0 
 2 x1

If we introduce the interpolation over the structure, we get

ε  x , t   B  x  r  t  , x V (4-66)
where the strain-displacement matrix has been defined as
B  x    N  x  , x V (4-67)

The dimension of B is 6  3nN for the three-dimensional problem.

In order to obtain the discrete equilibrium equations, we will apply the principle of virtual
work. To this end, we interpolate the virtual displacements  u in the same fashion as the real
displacements, i.e.
 u  x , t   N  x   r  t  , x V (4-68)

where  u   u1  u2  u3  and  r is the virtual global nodal displacement vector. The

T

relation between the virtual strains and the virtual displacements is

 ε   u  Β r, x V (4-69)
where  and B are given in Equations (4-65) and (4-67), respectively.

Let b = b1 b2 b3  and t = t1 t2 t3  be the body force and traction vectors,
T T

respectively. The principle of virtual work is then stated in the form

 ε σdV    uT bdV    uT tdS

T
(4-70)
V V St

where we have also used that  u  0 on Su and that t  t on St . Inserting the interpolated
displacement field, we get
 
 r T   BT σdV   N T bdV   N T tdS   0 (4-71)
V V St 
where it was used that the virtual nodal displacements  r depend on time only. Except for
nodes I located on Su , the entries of  r are arbitrary, and we arrive at the discrete form of the
equilibrium equations

4-23
  B σdV   N bdV   N T tdS  0 I  Su
T T
(4-72)
V V St

The nodal displacements that are prescribed by boundary conditions are excluded from these
equations. It is now useful to introduce the internal and external nodal forces respectively as
Rint  r    BT σdV , Rext   N T bdV   N T tdS (4-73)
V V St

where the dependence of Rint upon r is indicated. This dependence stems from the
constitutive relation that we will discuss below. The vectors Rint and Rext have dimension
nSD nN 1 . We may then express Equation (4-72) in the form

Rint  r   Rext (4-74)

which expresses the semi-discrete equilibrium equations. These equations are called semi-
discrete because they are discrete in space but continuous in time. In order to simplify the
notation, the time dependence is not shown explicitly in Equations (4-73) and (4-74). It is
sometimes more convenient to express the semi-discrete equilibrium equations in a residual
form, namely
G  Rint  Rext  0 (4-75)
where G is the residual nodal force vector; clearly, equilibrium requires G equal to zero at
any time.
The constitutive behaviour of the material only enters the problem through the stress tensor.
We will assume here that the stress rate σ depends on the stress σ , a collection of internal
variables, denoted ξ , and the strain rate ε . The state of the material is then defined by σ and
ξ , and we write the constitutive relation in the general form
σ  σ  σ, ξ , ε  (4-76)

The collection of internal variables ξ may include scalars and tensors, and could represent
phenomena like work hardening and damage. The formulation in Equation (4-76) is sufficiently
general to include viscoelastic, elastic-plastic and elastic-viscoplastic materials.

Remark 4-1: Linear-elastic material

If the material is linear elastic, the constitutive equation is expressed by Hooke’s law,
which on matrix form reads as
σ  Cε (4-77)
where C is the elasticity matrix with dimension 6  6 , containing the elastic moduli (see
Chapter 5 for details). Inserting the interpolation, we get
σ  CBr (4-78)
and the internal force vector is re-written as
Rint  r   Kr , K   BT CBdV (4-79)
V

4-24
where K is the global stiffness matrix, independent on r , with dimension 3nN  3nN . The
semi-discrete equilibrium equations are then reduced to the more familiar expression from
linear finite element methods
Kr  Rext (4-80)
Since this is a system of linear equations, it can be solved for the unknown degrees of freedom
r , when the boundary conditions are properly accounted for. See Bell [7] for details.
■

4.5.2 Solution method

We have now developed the equilibrium equation for the discretized structure. If the material
behaviour is not linear, as in the case of linear elasticity, the semi-discrete equilibrium equations
make up a system of non-linear equations. There are various ways to solve these equations, see
e.g. [11] (Chapter 16) for details. Here we will only discuss the Newton-Raphson method,
which is frequently used in nonlinear FE methods. This is an incremental-iterative method,
which requires linearization of the equilibrium equations.
Assume that the structure is in an equilibrium configuration at given external loads, i.e.,
the out-of-balance forces are all equal to zero. Let the time be given by tn . The residual form
of the equilibrium equations reads as
Gn  Rint  rn   Rnext  0 (4-81)

where rn  r  tn  and Rnext  Rext  tn  . At time tn1  tn  tn1 , the external forces have
increased to R  R  tn1  , and the task is to find rn1  r  tn1  so that the semi-discrete
ext
n 1
ext

equilibrium equations are satisfied. To this end, we define an iterative scheme in which the
global degrees of freedom are changed iteratively until equilibrium is re-established, i.e.,
GnI 11  Rint  rnI11   Rnext1  0 (4-82)

where rnI11 is the iterative estimate on rn1  r  tn1  and I is an iteration counter. A first-order
Taylor expansion of the semi-discrete equilibrium equations leads to
GnI 11  GnI 1  K t  rnI1  rnI11  rnI1   0 (4-83)

where the tangent stiffness matrix of the structure has been defined as
Rint  r 
K t r   (4-84)
r
Recall that the external loads Rext are fixed during the equilibrium iterations. The dimension
of K t is 3nN  3nN . Note that K t is not constant but varies from iteration to iteration, owing
to the variation of rnI 1 . The initial values in the iteration scheme are rn01  rn and
Gn01  Rint  rn01   Rnext1 . The Newton-Raphson method is illustrated in Figure 4-14.

4-25
 The global tangent stiffness matrix can be further developed by use of Equation (4-73)1,
and we get
Rint  r  σ  r 
K t r     BT dV (4-85)
r V
 r
where the dependence of the stress on the position vector x is not explicitly shown to simplify
the expressions. To proceed, we have to consider the material behaviour. Assume that σ n , ξ n
and εn are given values of the stress tensor, the collection of internal variables describing the
internal structure of the material, and the strain tensor at the last equilibrium configuration at
time tn , and that ε is the strain increment induced by the incremental change of the external
loads during the time increment t  t  tn . The stress σ at time t  tn  t is then defined
formally by a stress update algorithm and depends, in general, nonlinearly on the state of the
material at tn and the strain increment ε  ε  εn during the time step, viz.

σ  σ  σ n , ξ n , ε n , ε  εn  (4-86)

where σ represents the incremental stress response function [10]. The consistent (or
algorithmic) tangent operator C t is then obtained by linearization of the stress update algorithm
[9], i.e.,
σ
dσ  dε  C t dε (4-87)
ε
which means that C t is formally defined as
σ
Ct  (4-88)
ε
The consistent tangent operator at time tn 1 is then obtained as Cnt 1  C t  εn1  .
Proper computation of the consistent tangent operator is important for the convergence rate
of the Newton-Raphson solution method. Finding C t given the constitutive model and the
stress update algorithm is usually a rather tedious task, and detailed derivations are outside the
scope of these lecture notes. The reader is referred to the textbook of Belytschko et al. [9] on
nonlinear finite element methods for the details.

4-26
 Load

Displacement

Figure 4-14 Graphical illustration of the Newton-Raphson method.

With the consistent tangent operator, the global tangent stiffness matrix, as defined in
Equation (4-85), is expressed as
K t   BT C t BdV (4-89)
V

where we have used that

σ σ ε
  CtB (4-90)
r ε r
Now, as we have determined the expression for the global tangent stiffness matrix, we can
return to the iterative solution of the semi-discrete equilibrium equations. From Equation (4-83)
it follows that the new iterative estimate on rn1 is obtained by solving the linear equation
system
K t  rnI1  rnI11  rnI1   GnI 1 (4-91)

where rnI11 represent the only unknown quantities. Since Gn1 and K  rn1  are different for
I t I

each of the iterations, the Newton-Raphson procedure is computationally expensive, but it tends

4-27
to converge rapidly if the load increment is not too large. Equilibrium is assumed to be fulfilled
when the norm of the out-of-balance forces is less than a prescribed tolerance, e.g.

GnI 1  (GnI 1 )T GnI 1  tol. (4-92)

It is important to use a sufficiently strict tolerance to avoid drifting off the equilibrium path as
the solution proceeds.
The Newton-Raphson method implicitly assumes that K t is invertible, which implies that
det K t  0 at all times. If this is not the case, nontrivial solutions to the homogeneous equation
system, K t r  0 , exist, where  r represents an iterative change of r [11]. Put differently, r
may change without an associated change of the forces acting on the discretized structure. This
situation corresponds to the passing of a limit point at which a peak load is approached beyond
which the external forces may decreases with continuing deformation, see Figure 4-15 . Since
linear geometry is assumed here, thus excluding buckling phenomena, this situation can only
occur for materials exhibiting softening, i.e., the strength of the material decreases with
straining. If this is the case, the Newton-Raphson method will work only until the limit point is
reached and will then fail due to the singularity of the tangent stiffness matrix. To continue the
simulation in the post-peak domain, some kind of continuation method (or arc length method)
has to be utilized [9], but this is outside the scope of these lecture notes.

Limit point

Figure 4-15 Example of a limit point on the force-displacement curve.

4-28
Example 4-4: Rigid cantilever with nonlinear torsional spring.

Figure 4-16: Rigid cantilever with nonlinear torsional spring.

Figure 4-16 shows a rigid cantilever subjected to a concentrated force F at its tip. The
cantilever is supported by a torsional spring with nonlinear stiffness, k  k0 0    , where
n

k0 ,  0 and n are constants governing the response of the torsional spring. The angle 
describes the rotation of the rigid cantilever. The moment carried by the nonlinear spring is
M  k   . Given the spring constants k0  103 Nm/rad , 0  1rad and n  0.5 , the length
of the cantilever L  1 m and the force F  2.5 103 N , determine the angle  after
deformation using the Newton-Raphson method.
The relationship between the force and the angle of rotation is
k0
FL  k    F 0    
n

L
The residual G   reads

k0
G    0      F
n

L
The Newton-Raphson method is based on the algorithm
G   d   G    G   d  0
Thus
G  
d  
G  

where the tangent stiffness G   is derived as

k0 
G    0     n 0     
n n 1

L 
Inserting the given values, we obtain the residual and tangent stiffness as

4-29
 G    103 1      2.5 103
0.5

G    103  1     0.5 1    

0.5 0.5
 
We are now ready to start the iteration scheme, but we need an initial value for the angle
 . Here it is reasonable to take   0   0 as the start value, which corresponds to zero force.
The iteration scheme proceeds as follows:
 First iteration:

103 1  0   0  2.5 103

0.5

d    2.500 rad
103  1  0   0.5 1  0   0
0.5 0.5
 

 (1)   (0)  d  0  2.500  2.500 rad

 Second iteration:
103 1  2.500   2.500  2.5 103
0.5

d    0.857 rad
103  1  2.500   0.5 1  2.500   2.500
0.5 0.5
 

 (2)   (1)  d  2.500  (0.857)  1.643 rad

 Third iteration:
103 1  1.643 1.643  2.5 103
0.5

d    0.080 rad
103  1  1.643  0.5 1  1.643 1.643
0.5 0.5
 

 (3)   (2)  d  1.643  (0.080)  1.563 rad

 Fourth iteration:
103 1  1.563 1.563  2.5 103
0.5

d    0.001 rad
103  1  1.563  0.5 1  1.563 1.563
0.5 0.5
 

 (4)   (3)  d  1.563  (0.001)  1.562 rad

We stop after the fourth iteration because the iterative changed is now considered to be
sufficiently small. The value of the residual at this point is less than 0.0002 N, and we may
safely assume that we have the converged solution, namely   1.562 rad . The response curve
is shown in Figure 4-17. The reason for the overestimation of the angle, i.e., the convergence
is from above, is the increasing slope of the response curve.

4-30
Figure 4-17: Response of rigid cantilever supported by nonlinear torsional spring.

4.6 References
[1] F. Irgens. Continuum Mechanics. Springer, 2008.
[2] J. Lubliner. Plasticity Theory. Macmillan, 1990.
[3] L.E. Malvern. Introduction to the Mechanics of a Continuous Medium. Prentice-Hall,
1969
[4] G.E. Mase. Continuum Mechanics. Schaum’s Outline Series. McGraw-Hill, 1970.
[5] F.A. McClintock, A.S. Argon. Mechanical behaviour of materials. Addison-Wesley,
1966.
[6] R.D. Cook, D.S. Malkus, M.E. Plesha, R.J. Witt. Concepts and applications of finite
element analysis. Fourth edition. Wiley, 2002J. Lubliner. Plasticity Theory. Macmillan,
1990.
[7] K. Bell. An engineering approach to finite element analysis of linear structural mechanics
problems. Akademika Publishing, 2013.
[8] N.S. Ottosen, H. Petersson. Introduction to the finite element method. Prentice Hall,
1992.
[9] T. Belytschko, W. K. Liu, B. Moran. Nonlinear finite elements for continua and
structures, Wiley, 2000.
[10] J.C. Simo, R.L. Taylor. Consistent tangent operators for rate-independent elastoplasticity.
Computer Methods in Applied Mechanics and Engineering 48 (1985) 101–118.
[11] N. Saabye Ottosen, M. Ristinmaa. The Mechanics of Constitutive Modeling. Elsevier,
2005.

4-31
4-32
5 Elastic materials
This chapter deals with the elastic and thermoelastic behaviour of solid materials. The
emphasis is put on obtaining the constitutive equations for a given elastic material, starting
from the generalized Hooke’s law for a fully anisotropic elastic material and using the
appropriate material symmetries. Since we assume small deformations, it is sufficient to
consider linear elastic behaviour. We will see that while the fully anisotropic linear elastic
material has 21 independent elastic coefficients, materials with orthotropic symmetry (like
rolled steel plates and wood) have nine and isotropic materials have only two. It is important
to take advantage of the material symmetry to reduce the cost of identifying the elastic
coefficients from mechanical tests, but also to reduce the computational cost when analysing
elastic structures with the finite element method.

5.1 Generalized Hooke’s Law

Solid materials, like metals, wood and concrete, behave elastically when the stress (or
strain) is below a certain limit. The strain in this region is said to be elastic, and it consists
of reversible displacements of atoms and molecules. The material retains its original
configuration upon unloading of the stress. The limit strain at which the behaviour becomes
irreversible is typically in the order of 0.0010.005.
A material is said to be elastic if [1]
 a deformed body made of the material completely recovers its original
configuration upon removal of the forces causing the deformation, and
 a one-to-one relation exists between the state of stress and the state of strain of
the body at a given temperature.
These are the conditions of reversibility and path independence.
The material is said to be linear elastic if the relation between the state of stress and
state of strain is linear. Since we are dealing with infinitesimal deformations only, linear
elasticity will be assumed henceforth. The generalized Hooke’s law defines the most general
form of a linear elastic material
 ij  Cijkl  kl (5-1)

where Cijkl is the 4th order tensor of elastic constants. Writing these 9 constitutive equations
in expanded form, we get
 11  C111111  C111212   C1133 33
 12  C121111  C121212   C1233 33
(5-2)

 33  C331111  C331212   C3333 33

It is noted that owing to the symmetry of the stress and strain tensors, there are only 6
independent constitutive equations for the linear elastic material.
The elastic tensor Cijkl has 34  81 components. Owing to the symmetry of the stress
tensor,  ij   ji , it follows that

5-1
 Cijkl  C jikl (5-3)
This reduces the number of independent coefficients from 81 to 54. Since the strain tensor
is symmetric,  kl   lk , there is no loss of generality in assuming
Cijkl  Cijlk (5-4)
The number of independent coefficients is then reduced to 36. The symmetries implied in
Equations (5-3) and (5-4) are usually referred to as the minor symmetries.
Due to the symmetry of  ij and  ij , and the associated symmetries of Cijkl , it is
convenient to express the generalized Hooke’s law in matrix form
σ  Cε (5-5)
where
 11   11   C1111 C1122 C1133 C1123 C1131 C1112 
    C C2222 C2233 C2223 C2231 C2212 
 22   22   2211
    C C3322 C3333 C3323 C3331 C3312 
σ   33  , ε   33  , C   3311  (5-6)
 23   23  C2311 C2322 C2333 C2323 C2331 C2312 
 31   31  C3111 C3122 C3133 C3123 C3131 C3112 
     
 12   12   C1211 C1222 C1233 C1223 C1231 C1212 
Here σ and ε are column matrices representing the stress and strain tensors, while C is a
square matrix containing the 36 independent elastic coefficients. The engineering shear
strains have been introduced as  23  2 23 ,  31  2 31 and  12  212 .

5.2 Strain energy function

The material is said to be hyperelastic if there exists an elastic potential function (or
strain-energy function) U0  U0  ε  such that:

U 0
 ij  (5-7)
 ij

where U 0 is made symmetric in the variables  ij and  ji due to the symmetry of the stress
tensor. Then the stress power d equals the rate of change of the strain-energy function U 0 ,
because
U 0
U0   ij   ij ij  d (5-8)
 ij
We further have
t2 t2

  dt    
d ij ij dt  U 0  ε 2   U 0  ε1  (5-9)
t1 t1

where ε1 and ε 2 are the strain tensors at time t1 and t 2 , respectively. The integrals in
Equation (5-9) are clearly path independent, and thus the stress power integrated over a
closed cycle vanishes. We also conclude that for the hyperelastic material the stress power

5-2
is completely stored as strain energy, i.e., there is no dissipation in a hyperelastic material.
The elastic energy stored in the body is given by U   U 0 dV .
V

The generalized Hooke’s law (or the linear elastic solid) is obtained by assuming U 0
so that
U 0
 ij   Cijkl  kl  Cij1111  Cij1212   Cij 33 33 (5-10)
 ij
From Equation (5-10) it follows that
 2U 0  2U 0
Cijkl    Cklij (5-11)
 ij  kl  kl  ij

Thus, the assumption of an elastic potential function implies that the elastic tensor Cijkl
exhibits, in addition to the minor symmetries, the major symmetry
Cijkl  Cklij (5-12)
The outcome is that the number of independent elastic coefficients is reduced from 36 to
21, and the elastic coefficient matrix C in Equation (5-5) becomes a symmetric 6-by-6
matrix:
C1111 C1122 C1133 C1123 C1131 C1112 
C C2222 C2233 C2223 C2231 C2212 
 1122
C C2233 C3333 C3323 C3331 C3312 
C   1133  (5-13)
C1123 C2223 C3323 C2323 C2331 C2312 
C1131 C2231 C3331 C2331 C3131 C3112 
 
C1112 C2212 C3312 C2312 C3112 C1212 

The form of the strain energy function U 0 is obtained as follows. From Equation (5-10)
we have
U 0
 ij d ij  Cijkl  kl d ij  d ij  dU 0 (5-14)
 ij
Using the major symmetry of the elastic tensor Cijkl , which is expressed by Equation (5-12)
, we get [2]
2dU 0  Cijkl   kl d  ij   ij d  kl   Cijkl d   kl  ij  (5-15)

and thus
U 0  12 Cijkl  ij kl (5-16)
This shows that for the linear elastic solid the strain energy function is a quadratic function
of the strain tensor. Note that inserting Equation (5-10) back into Equation (5-16), gives
U 0  12  ij  ij (5-17)
To impose the restrictions due to material symmetry, to be presented in Chapter 5.5,
we need to establish the transformation rules for the elastic tensor Cijkl under change of the

5-3
orthonormal basis. To this end, consider the strain energy function U 0 expressed in the two
basis systems  ei  and  ei  , i.e.

U0  ε   12 Cijkl

 ij kl  12 Cmnrs mn rs (5-18)

where the relation between the two orthonormal bases is given by: ei  aij e j . The
transformation rule for 2nd order tensors gives  mn  aim a jn ij and  rs  akr als kl , and,
accordingly, Equation (5-18) leads to

C 
ijkl  aim a jn akr als Cmnrs   ij kl  0 (5-19)

Since ε is arbitrary, the transformation rules for Cijkl are derived as

 
Cijkl  aim a jn akr alsCmnrs and Cmnrs  aim a jn akr als Cijkl (5-20)

which (according to the definition in Chapter 3.3) shows that Cijkl is a tensor of 4th order.

5.3 Isotropic elasticity

If there are no preferred directions in an elastic material, the material is said to be
elastically isotropic. In this case, the elastic constants at a given particle must be the same
for all possible choices of Cartesian coordinate systems:

Cijkl  Cijkl (5-21)

where Cijkl and Cijkl are the elastic constants in the coordinate systems with basis  ei  and
ei  ,
respectively. The most general isotropic 4th order tensor with minor and major
symmetry can be shown to be [1]
Cijkl   ij kl    ik  jl   il jk  (5-22)

where  and  are the Lamé elastic constants. Using Equation (5-1), the generalized
Hooke’s law for the isotropic material is obtained as
 ij  Cijkl  kl   kk  ij  2 ij (5-23)
Alternatively, in expanded form, the stress-strain relations read
 11    11   22   33   211
 22    11   22   33   2 22
(5-24)
 33    11   22   33   2 33
 23  2 23 ,  31  2 31 ,  12  212
The relation between the Lamé constants and the more familiar Young’s modulus E ,
Poisson’s ratio  and shear modulus G will be derived below. Using Equations (5-16) and
(5-22), the strain-energy function becomes
U 0  ε   12  ij Cijkl  kl  12    kk    ij  ij
2
(5-25)
In expanded form, we get

5-4
 U0  ε   12  11   22   33    112  122  ...   332 
2
(5-26)

Applying Equation (3-34), the strain-energy function, expressed in terms of the principal
strain invariants, reads
U 0  ε   12  I 2    2 II   I 2  (5-27)

Since the material is isotropic, it should be possible to express the strain energy in terms of
invariants only, which is shown by Equation (5-27) to be the case.
To compare this relation to the more well-known relations for the elastic material that
we know from advanced strength of materials, it is convenient to establish the inverse
relation, i.e., the strains given as a linear function of the stresses. Putting i  j in the stress-
strain relation in Equation (5-23), we get:
 ii   3  2   ii (5-28)

Combining Equations (5-23) and (5-28), and solving for  ij , we obtain the inverse relation
in the form:
 1
 ij    kk ij   ij (5-29)
2  3  2  2
The relations between Lamé elastic constants and the more familiar Young’s modulus E ,
shear modulus G and Poisson’s ratio  , are obtained by looking at different simple stress
states.
In pure shear, say where only  12 takes non-zero values, Equation (5-29) gives
1
12  12 or  12  212   12 /  (5-30)
2
which implies that  equals the shear modulus G . In uniaxial tension (  11  0 ), we get
from Equation (5-29) the following non-zero strain components
 1 
11   11  11 ,  22   33    11 (5-31)
2  3  2  2 2  3  2 

But we know that in terms of E and  , these relations read 11  11 / E and
 22   33  11 / E . Solving for the Lamé elastic constants, we end up with the relations
E E
 G  and   (5-32)
2 1   1  1  2 
Inserting these relations back into Equation (5-29), we get the generalized Hooke’s law for
the isotropic material as
 1 
 ij    kk  ij   ij (5-33)
E E
In expanded form, this gives the well-known relations

5-5
 11 
1
E
 11   22   33  
 22   22   33   11  
1
E
(5-34)
 33   33   11   22  
1
E
1  1  1 
 23   23 ,  31   31 , 12   12
E E E
A particularly simple version of the linear elastic material law is obtained by
introducing the stress and strain deviators along with the pressure P   13  kk and the
volumetric strain  v   kk . Recall that the deviatoric stress and strain tensors were defined
by
 ij   ij  13  kk  ij and  ij   ij  13  kk  ij (5-35)
From Equation (5-33) (or alternatively from Equation (5-29)), we then arrive at
 ij  2G ij and P  K v (5-36)
where the bulk modulus is introduced as
2 E
K    (5-37)
3 3 1  2 
Use of Equation (5-36) gives us the following alternative form of the generalized Hooke’s
law for the isotropic material:
 ij   P ij   ij  K  v ij  2G ij (5-38)

Here the contribution to the stress is decomposed into hydrostatic and deviatoric parts,
which is convenient for example in formulation of elastic-plastic material models where
only the deviatoric state influences the yielding and plastic flow in metals. Similarly, the
strain energy function may be expressed as
U0  ε   G ij  ij  12 K v2 (5-39)
where Equations (5-17), (5-38) and (5-35)2 were used. It transpires that the strain energy is
additively decomposed into distortional (or deviatoric) and dilatational (or volumetric)
contributions.
It is reasonable to assume that the strain energy function should be a positive definite
function of the strain, which vanishes only when the strain equals zero. This implies that for
any small strain from an unstressed state, the stress must do positive work. Since the
deviatoric and volumetric strains are independent variables, we find from Equation (5-39)
that the necessary and sufficient conditions for positive definiteness are [1]
  G  0 and   23   K  0 (5-40)
Using the relations between the elastic coefficients, an equivalent set of conditions for
positive definiteness are
E  0 and  1    12 (5-41)

5-6
Note that the value   12 implies elastic incompressibility, since then 1/ K equals zero.
This type of behaviour is approximated by some rubber materials.

Example 5-1: Plane stress state of isotropic linear elastic material.

A plane stress state is defined by  33   23   31  0 , so that the only non-zero stress
components are  11 ,  22 and  12 in the x1 x2 plane. The assignment is to establish the
stress-strain relation for this plane stress state and to find the expression for the normal strain
 33 in the x3 direction. We shall also derive the expression for the strain energy function
U0 .
Using Equation (5-34), we readily find the non-zero strain components in the x1 x2
plane as
1 1 1 
11   11  22  ,  22   22  11  , 12   12
E E E
Solving for the stress components, we obtain
E E E
2  11
 11    22  ,  22  2  22
 11  ,  12   12
1  1  1 
Again applying Equation (5-34), the normal strain in the x3 direction is determined as

 33   11   22 
E
The expression for the strain energy function is found by use of Equation (5-17), and
the result is
1
U0   1111   22 22   1212   21 21 
2

2  11
 2  211 22   222    122   212 
E E

2 1   2  
1  

It is recommended not to introduce 12   21 in the expression for the strain energy function,
because this may lead to mistakes when using U 0 as a potential for the stress.
■

5.4 Thermoelasticity
We consider an elastic body under temperature variation, assuming both the strains and
the temperature variation to be small, so that linear stress-strain-temperature relations are
sufficiently accurate. The temperature variations are assumed so small that any changes in
the material parameters (e.g. the Young’s modulus) are negligible.
Assuming linear relationships, the generalization of Hooke’s law to thermoelasticity
reads as [1]
 ij   ij T  Cijkl  kl (5-42)

5-7
where T  T  T0 is the temperature variation (assumed to be small), T0 is the reference
temperature and  ij is the 2nd order, symmetric tensor of thermal stress coefficients. The
symmetry of the tensor  ij stems from the symmetry of the stress tensor. The minus sign in
front of the thermal contribution to the stress stems from the fact that an increase of the
temperature under constant strain will lead to compressive stresses in the body.
The constitutive relation is markedly simplified if isotropy is assumed, and in this case
the stress is given by
 ij   kk  ij  2 ij  T  ij where ij   ij (5-43)

Hence, the only extra material parameter for the isotropic thermo-elastic material is  ,
denoted the thermal stress coefficient. The stress deviator and pressure are
 ij  2G ij and P   K  v  T (5-44)

which shows that for an isotropic material a temperature change will only affect the pressure
or volumetric strain. Inverting the relations we have
1 P
 ij   ij and  v    3T (5-45)
2G K
where 3   / K and  is the linear coefficient of thermal expansion. The inverse relation
for the isotropic material reads
 1 
 ij    kk  ij   ij  T  ij (5-46)
E E
where the shear and bulk moduli have been expressed in terms of the Young’s modulus and
Poisson’s ratio.

5.5 Anisotropic elasticity

An elastic material is said to be isotropic if the form of the elastic potential function
U 0 and the values of the elastic coefficients Cijkl are independent of orientation. If the
material is not isotropic, it is said to be anisotropic. We have already seen that in the general
case of an anisotropic material (assuming a strain energy function to exist) the number of
independent elastic coefficient is 21, while for the isotropic material there are only two
independent coefficients.
Most materials have a structure with some kind of symmetry that reduces the number
of independent elastic coefficients. As an example, a rolled plate has typically three planes
of symmetry defined by the rolling direction, the transverse direction and the normal
direction. Such materials are said to have orthotropic symmetry, and the number of
independent elastic coefficients is reduced to 9. Since it is both time consuming and
expensive to determine the material constants in the laboratory through mechanical tests, it
is important to take advantage of the expected symmetry of the material when establishing
an experimental programme to determine the elastic constants.

5-8
5.5.1 Plane of material symmetry
Consider a plane S1 with normal e1 . A reflection with respect to S1 is defined by the
transformation (see Figure 5-1)
e1  e1 , e2  e 2 , e3  e3 (5-47)
A more compact representation of the transformation is given by
 1 0 0

e  aij e j
i with a   0 1 0 (5-48)
 0 0 1 
The plane S1 is said to be a plane of material symmetry if the constitutive equations of the
material, expressed in the basis  ei  , remain the same under the transformation defined by
the matrix a . The constitutive equations for the linear elastic material in the un-starred and
starred set of base vectors read as
 ij  Cijkl kl and  ij  Cijkl

 kl (5-49)

Figure 5-1: Reflection with respect to the plane S1 .


If S1 is a plane of material symmetry, it follows that the components of Cijkl and Cijkl
must be the same, and from Equation (5-20) we get
Cijkl  aima jn akr alsCmnrs (5-50)

Stated differently: the components of the tensor Cijkl are invariant under the transformation
defined by a . Note that this is only valid for the a defined by Equation (5-48) and other
planes of material symmetry the material may have.

5-9
 Next we need to establish the influence of the restriction given by Equation (5-50) on
the components of the tensor Cijkl . To this end, we note that a11  1 and a22  a33  1 ,
whereas all the other components are zero. Using Equation (5-50), we obtain for example
C1111  a1m a1n a1r a1s Cmnrs   1 1 1 1 C1111  C1111
(5-51)
C1131  a1m a1n a3r a1s Cmnrs   1 1 1 1 C1131  C1131

Here the first equation is trivial, while the second implies that C1131 must vanish. It turns out
that all Cijkl with an odd number of the subscript 1 have to vanish, i.e., the following eight
of the 21 independent coefficients are zero:
C1131  C1112  C2231  C2212  C3331  C3312  C2331  C2312  0 (5-52)
It follows that there are 13 non-zero and independent elastic coefficients, and the elastic
tensor Cijkl is thus reduced to

C1111 C1122 C1133 C1123 0 0 

C 0 
 1122 C2222 C2233 C2223 0
C C2233 C3333 C2333 0 0 
C   1133  (5-53)
C1123 C2223 C2333 C2323 0 0 
 0 0 0 0 C3131 C1231 
 
 0 0 0 0 C1231 C1212 
A linear elastic material having one plane of material symmetry is called a monoclinic
elastic material.
Using the tensor transformation rule, the components of the elasticity tensor C may be
obtained with respect to any coordinate system. In this connection it is important to realize
that if the plane of elastic symmetry is not a coordinate plane, there will be more than 13
non-zero elastic coefficients. There will, however, still be only 13 independent elastic
coefficients. It is usually convenient to work with a coordinate system having the coordinate
planes parallel to the planes of material symmetry.

5.5.2 Orthotropic symmetry

A material is said to exhibit orthotropic symmetry if there are three mutually orthogonal
planes of symmetry, e.g. the plane S1 with unit normal e1 , the plane S 2 with unit normal
e2 , and the plane S 3 with unit normal e3 . These are the three coordinate planes. For this
material the coefficients Cijkl must be invariant to the three transformations (reflections)

 1 0 0 1 0 0 1 0 0 

e  aij e j
i with a   0 1 0 , 0 1 0  , 0 1 0  (5-54)
 0 0 1  0 0 1  0 0 1

Using Equation (5-50), it can be established that besides Equation (5-52), we have
C1123  C2223  C2333  C1231  0 (5-55)

5-10
It then follows that there are 9 independent elastic coefficients for the material with
orthotropic symmetry, and, in the coordinate system defined by the three mutually
orthogonal symmetry planes, the elastic tensor Cijkl becomes

C1111 C1122 C1133 0 0 0 

C 
 1122 C2222 C2233 0 0 0 
C C2233 C3333 0 0 0 
C   1133  (5-56)
 0 0 0 C2323 0 0 
 0 0 0 0 C3131 0 
 
 0 0 0 0 0 C1212 

In expanded form, the elastic constitutive relations,  ij  Cijkl  kl , read as

 11  C111111  C1122 22  C1133 33

 22  C112211  C2222 22  C2233 33
(5-57)
 33  C113311  C2233 22  C3333 33
 23  2C2323 23 ,  31  2C3131 31 ,  12  2C1212 12
where the symmetry of the stress and strain tensors has been taken into account. If the stress
and strain components are defined with respect to other coordinate axes than those defined
by the three symmetry planes, the elastic tensor will take the form of Equation (5-13), but
there will still be only 9 independent elastic coefficients.
Since the orthotropic symmetry is exhibited by a number of important materials, e.g.
rolled plates, wood and fibre-reinforced polymers, it is a particularly important case of
elastic symmetry.

Example 5-2: Engineering elastic constants for the orthotropic material.

Determine the relation between the matrix of elastic compliances S , defined by the
relation ε  Sσ , and the engineering elastic constants, namely the Young’s moduli E1 , E2 ,
E3 , the shear moduli G23 , G31 , G12 , and the Poisson’s ratios  12 ,  21 , ...,  31 , for the linear
elastic material with orthotropic symmetry. It is assumed that the coordinate axes xi are
along the symmetry axes of the material. In this particular coordinate system, shear stresses
will not produce normal strains, and conversely, see Equation (5-56).
To obtain these relations, we start by considering a uniaxial tension test along the x1
axis in which only  11 is non-zero. The strains measured in this test are
 11  11  11
11  ,  22   1211   12 ,  33   1311   13
E1 E1 E1

Similarly, tensile tests along the x2 and x3 axes with non-zero stress  22 and  33 ,
respectively, give the strains
 22  22  22
11   21 22   21 ,  22  ,  33   23 22   23
E2 E2 E2

5-11
  33  33  33
11   31 33   31 ,  22   32 33   32 ,  33 
E3 E3 E3

Shear tests with in turn  23 ,  31 and  12 as the only non-zero stress components give,
respectively, the strains
 23  31  12
 23  2 23  ,  31  2 31  ,  12  212 
G23 G31 G12
By comparing the strains developing in these six tests with the definition of the matrix
of elastic compliances, we obtain
 1  E21  E31 0 0 0
 S11 S12 S13 0 0 0   E1 2 3

S S 22 S 23 0 0 0    E1
12 1 
 E323 0 0 0
 12
E2
  
S S 23 S33 0 0 0    E13  E23 1
0 0 0
S   13   E3
0   0 
1 2

0 0 0 S44 0 0 0 1
G23 0 0
0 0 0 0 S55 0   0 1 
   0 0 0 G31 0
 0 0 0 0 0 S66   1 
 0 0 0 0 0 G12 

where symmetry requires that

 12  21  13  31  23  32
 ,  , 
E1 E2 E1 E3 E2 E3
■

5.5.3 Isotropy
In an isotropic material, the elastic coefficients are invariant to the orientation of the
coordinate system. Accordingly, Equation (5-50) holds for any transformation ei  aij e j ,
representing a rigid rotation, reflection in a plane, or central inversion of the rectangular
Cartesian reference axes.
As we have already seen, isotropy implies that there are only two independent elastic
coefficients, e.g. Young’s modulus E and Poisson’s ratio  . To show this, we first note
that the isotropic material is also orthotropic and we can therefore start from Equation (5-57).
Then the restrictions induced by Equation (5-50) under rigid rotations about the coordinate
axes are considered.
Such rotations about the x1 axis, the x2 axis and the x3 axis are in turn defined by

1 0 0  cos  0  sin    cos  sin  0


a  0 cos  sin   ,  0 1 0  ,   sin  cos  0  (5-58)
0  sin  cos    sin  0 cos    0 0 1 

where  is the rotation angle. Applying first the rigid rotation about the x3 axis, Equation
(5-50) leads to
C1111  C2222 , C1133  C2233 , C2323  C3131 , C1212  12 C1111  C1122  (5-59)

5-12
Then, a rigid rotation about the x2 axis gives

C1111  C3333 , C3131  12  C1111  C1133  (5-60)

Finally, the rigid rotation about the x1 axis provides the relations

C3131  12  C3333  C1133  , C1122  C1133 , C2323  12 C2222  C2233  , C1133  C2233
(5-61)
The details of these derivations, which are quite tedious, are not shown here, but can be
found in Chung [3].
The resulting constitutive matrix for the isotropic elastic material comes out as
  2   0 0 0
    2  0 0 0 

     2 0 0 0
C  (5-62)
 0 0 0  0 0
 0 0 0 0  0
 
 0 0 0 0 0 
where  and  are the Lame coefficients, and further
C1111  C2222  C3333    2
C2323  C3131  C1212   (5-63)
C1122  C1133  C2233  
It is finally noted that Equation (5-5) with C defined from Equation (5-62) is fully
equivalent to Equation (5-23).

Example 5-3: From orthotropy to isotropy.

Start from the elastic compliance tensor S ijkl for the orthotropic material in Example 5-
2, and use simple reasoning to find S ijkl for the isotropic material.
Since the material is assumed to be isotropic, the same elastic coefficients should be
obtained from uniaxial tension tests along any of the three coordinate axes or shear tests in
any of the coordinate planes. This immediately implies that
E1  E2  E3  E, G23  G31  G12  G, 12   21    31  

Consider a tension test along the x1 axis in a starred coordinate system xi defined by
the transformation matrix
 cos  sin  0
a    sin  cos  0
 0 0 1 

where  is the angle between the x1 and x1 axes of the starred and un-starred coordinate
systems, respectively. The only non-zero stress component in this test is  11 , while in the
un-starred coordinate system the non-zero stress components are  11  cos 2    11 ,

5-13
 22  sin 2    11 and  12  sin  cos    11 according to the transformation rule in Equation
(3-22). Similarly, the shear strain component in the un-starred coordinate system is obtained
from coordinate transformation as 12  sin  cos   11   22 
 , where, due to isotropy, the
only non-zero strain components in the starred coordinate system are 11   11 / E and
 22

  33

  11 / E . Thus, we get

 12  11  11 E
212   2sin  cos  1    sin  cos   G
G E G 2 1  

The matrix of elastic compliances S is then given by

 E1  E  E 0 0 0
  1
 E 0 0 0 
 E E
  E  E 1
0 0 0
S 
E
21 
 0 0 0 E 0 0
 21  
 0 0 0 0 E 0
 21  
 0 0 0 0 0 E 

5.6 References
[1] L.E. Malvern. Introduction to the Mechanics of a Continuous Medium. Prentice-Hall,
1969
[2] W.M. Lai, D. Rubin, E. Krempl. Introduction to continuum mechanics. Third edition.
Butterworth-Heinemann, 1993.
[3] T.J. Chung. Continuum Mechanics. Prentice-Hall, 1988.
[4] F. Irgens. Continuum Mechanics. Springer, 2008.
[5] J. Lubliner. Plasticity Theory. Macmillan, 1990.
[6] G.E. Mase. Continuum Mechanics. Schaum’s Outline Series. McGraw-Hill, 1970.

5-14
6 Viscoelastic materials
Some materials, like polymers, have time-dependent elastic properties, which call for
models based on the theory of viscoelasticity. We will restrict our formulations to simple
viscoelastic models under the assumption of isothermal conditions. We further assume the
material to be isotropic. The Maxwell and Kelvin-Voigt models of viscoelasticity are first
extended to complete stress states, and then we extend the standard linear solid of viscoelasticity
along the same lines. The one-dimensional versions of these models were introduced in Chapter
2. Thereafter, the Maxwell-Wichert model, which can be considered as a generalization of the
standard linear solid, is presented. Finally, integration algorithms for the rate constitutive
equations are developed. The reader is referred to the textbooks [1]–[3] for more comprehensive
descriptions of viscoelasticity.

6.1 Maxwell model

Figure 6-1: Rheological representation of the Maxwell model.

The rheological representation of the Maxwell model is illustrated in Figure 6-1. With
reference to Chapter 2.4.2, we start by decomposing the strain tensor ε additively into elastic
and inelastic parts
 ij   ije   iji (6-1)

where  ije is the elastic strain and  iji is the inelastic strain. This is the natural decomposition
since the Maxwell model consists of spring and dashpot elements in series, see Figure 6-1.
Similarly, the strain rate decomposition reads as
 ij   ije   iji (6-2)
The stress power is then defined by
d   ij ij   ij ije   ij iji (6-3)

where  ij ije is the reversible part, whereas  ij iji  0 represents the intrinsic dissipation due to
viscoelasticity. Note that the intrinsic dissipation is restricted to be non-negative by the second
law of thermodynamics.
As already mention in the introduction to this chapter, we will assume that the material is
isotropic. We will further assume that the volumetric response is elastic, so that it is only the

6-1
deviatoric response that leads to dissipation [3]. It is noted that these assumptions are introduced
mainly to simplify the theory and they have to be verified experimentally for the material at
hand.
With these assumptions, the elastic deviatoric strain and the volumetric strain are obtained
from Equation (5-36) as
1 P
 ije   ij and  v   (6-4)
2G K
It is noted here that  v represents the total volumetric strain of the material, whereas  ije is the
elastic deviatoric strain. The inelastic strain rate is defined by
1
 iji   ij (6-5)
2
where   0 is the viscosity constant. Note that ε i is a deviator per definition. It is then
immediately verified that the dissipation is non-negative

 ij ij   ij  13  kk ij  ij   ij ij  0

1 1 1
Di   ij iji  (6-6)
2 2 2
where it was used that  ij  ij  0 per definition of the deviator. Thus, our model is fulfilling the
requirement posed by the second law of thermodynamics.
Using Equations (6-4) and (6-5), the deviatoric part of the strain rate is obtained as
1 1
 ij   ije   iji   ij   ij (6-7)
2G 2
or
1  1 
 ij    ij   ij  (6-8)
2G   
where    / G is denoted the relaxation time. It is noted that a Maxwell body will never reach
equilibrium for finite stresses – the deviatoric strain rate is non-zero as long as the deviatoric
stress is non-zero. The Maxwell model is in this sense a fluid model.
Consider a relaxation test in which the strain deviator is constant in time. Let this constant
value be denoted  ij0 . Imposing the initial condition  ij  2G ij0 at t  0 , the solution to
Equation (6-8) is found as
 t
 ij  t   2G ij0 exp    (6-9)
 
Next we consider a steady creep test in which the stress deviator equals  ij0 . The instantaneous
response in the spring leads to a deviatoric strain equal to  ij0 / 2G , whereas the steady creep
response predicted by the Maxwell model is

6-2
 1 1 1  t
 ij  t    ij0   ij0t   ij0 1   (6-10)
2G 2 2G  
The relaxation test and the steady creep test can be used to determine the constants G and
   / G of the Maxwell model.

Example 6-1: Creep of thin-walled polymeric tube – Part I.

Figure 6-2: Polymeric tube subjected to tension force and torque.

Consider a thin-walled, cylindrical tube made of a polymeric material. The outer diameter
of the tube is 200 mm, the wall thickness is 10 mm and the length is 2000 mm, see Figure 6-2.
Assume that the material behaves according to the Maxwell model with G  1 GPa ,   106 s
and K  1.67 GPa . The tube is loaded statically by an axial force and a torque which induce
the stresses 11  9 MPa and 12  6 MPa , while all other stress components are zero.
Determine the deviatoric and hydrostatic stress state of the tube, and use this to find the strains
in the tube after 100 days of this creep loading.
Since  11 is the only non-zero normal stress, the pressure in the tube wall is given as
P   13 11  3 MPa . The deviatoric stress components are
11  11  P  6 MPa,  22
   33
  P  3 MPa, 12  12  6 MPa
The volumetric response is elastic, and thus
P
v    0.0018
K
The deviatoric strains are defined by Equation (6-10) and by inserting t  tmax with
tmax  100  24  60  60  8.64 106 s , we get

 11  
tmax
11  tmax   1    0.0289
2G   
  t 
  tmax    33
 22   tmax   22 1  max   0.0145
2G   
  t 
12  tmax   12 1  max   0.0289
2G   

6-3
The total strains  ij   ij  13  v ij in the tube after 100 days of creep are then given by
11  0.0295 ,  22   33  0.0139 and 12  0.0289 .
■

6.2 Kelvin-Voigt model

Figure 6-3: Rheological representation of the Kelvin-Voigt model.

Figure 6-3 illustrates the rheological representation of the Kelvin-Voigt model of linear
viscoelasticity. According to Chapter 2.4.3, we decompose the stress additively into elastic and
inelastic parts
 ij   ije   iji (6-11)

where  ije is the “elastic” stress and  iji is the “inelastic” stress. The background for this
decomposition is that the Kelvin-Voigt model consists of a spring and a dashpot in parallel. The
stress power is defined by
d   ij ij   ije ij   iji  ij (6-12)

where  ije ij is the reversible part and  iji  ij the intrinsic dissipation due to viscoelasticity.
As for the Maxwell model, we will assume isotropy and that the dissipation stems from
the deviatoric response. The elastic deviatoric stress and the pressure are then given by Equation
(5-36) as
 ije  2G ij and P   K v (6-13)
The inelastic stress is defined by
 iji  2 ij (6-14)
It is seen that the inelastic stress is deviatoric per definition. The dissipation is non-negative
since

6-4
 Di   iji  ij  2 ij  ij  2 ij  ij  13  kk ij   2 ij ij  0 (6-15)

where we used that  ij ij  0 per definition of the deviator.

The deviatoric stress tensor in the Kelvin-Voigt solid is then defined by
 ij   ije   iji  2G ij  2 ij (6-16)
or
 ij  2G  ij   ij  (6-17)

In a creep test, the stress deviator is constant in time and equal to  ij0 . Enforcing the initial
condition  ij  0 at t  0 , Equation (6-17) is solved to give

 ij0   t 
 ij  t    1  exp     (6-18)
2G    
This explicit relation enables the identification of the constants G and    / G from
laboratory tests.

Example 6-2: Creep of thin-walled polymeric tube – Part II.

Reconsider the thin-walled, cylindrical polymeric tube in Example 6-1, but assume that
the material response is governed by the Kelvin-Voigt model. The material constants and the
loading are the same as in Example 6-1. Calculate the strains in the tube after 100 days.
The volumetric response remains the same, while the deviatoric strains are now defined
by Equation (6-18) inserting t  tmax , viz.

 11   tmax  
11  tmax   1  exp      0.0030
2G    
 
 22  tmax  
  tmax    22
 22   tmax   1  exp      0.0015
2G    
   t 
12  tmax   12 1  exp   max    0.0030
2G    
The total strains  ij   ij  13  v ij in the tube after 100 days of creep are thus equal to
11  0.0036 ,  22   33  0.0009 and 12  0.0030 .
■

6-5
6.3 The standard linear solid
The extension of the standard linear solid to full stress states follows along the same lines
as above. We assume isotropy and that only the deviatoric response leads to dissipation. The
rheological representation of the standard linear solid is illustrated in Figure 6-4.
The constitutive relations defining the behaviour of the standard model in linear
viscoelasticity are with these assumptions given in the form (see Chapter 2.4.4)
1 P
 ij   ij   iji ,  v   (6-19)
2G0 K

 iji 
1
21
 ij  2G1 iji  (6-20)

where G0 , G1 , 1 are the constants defining the deviatoric response. These constants are all
assumed positive. The volumetric response is elastic and defined by the bulk modulus K .

Figure 6-4: Rheological representation of the linear standard solid.

The inelastic dissipation is the power expended by the dashpot and is given as
Di   iji  iji  21 iji  iji  0 (6-21)

where the inelastic stress  iji carried by the dashpot follows from Equation (6-20) as

 iji   ij  2G1 iji (6-22)

We once again consider a creep test with the stress deviator constant in time and equal to
 ij . Enforcing the initial condition  iji  0 at t  0 , we may solve Equation (6-20) to find
 0

 ij0   t 
 t  
i
1  exp     (6-23)
 1  
ij
2G1 

6-6
Combination of Equation (6-19) with  ij   ij0 and Equation (6-23) makes it possible to
experimentally determine the constants G0 , G1 and 1  1 / G1 . Alternatively, the stress
relaxation tests or the tension test at constant strain rate can be used to identify the model
constants.
The extended constitutive equations for the generalized Kelvin-Voigt solid are obtained in
a similar way, but this is outside the scope of these lecture notes.

Example 6-3: Creep of thin-walled polymeric tube – Part III.

The thin-walled, cylindrical polymeric tube from Examples 6-1 is once again considered,
but now the material behaviour is determined by the linear standard model. The material
constants are G0  G1  1 GPa and 1  106 s , while the loading is the same as in Example 6-1.
Find the strains in the tube after 100 days of creep loading.
The deviatoric strains are obtained from Equations (6-19) and (6-23) by inserting t  tmax , and
we get

 11  11   tmax 

11  tmax    1  exp      0.0060
2G0 2G1   1 

 22  
 22  tmax 
  tmax    33
 22   tmax    1  exp      0.0030
2G0 2G1   1 
 12  12   tmax 
12  tmax    1  exp      0.0060
2G0 2G1   1 
The total strains  ij   ij  13  v ij in the tube after 100 days of creep are thus equal to
11  0.0066 ,  22   33  0.0024 and 12  0.0060 .
The creep response of the tube as described by the Maxwell, Kelvin-Voigt and linear
standard models are illustrated in Figure 6-5 in terms of the evolution of the normal strain along
the x1 axis versus time. It is noted that the creep strains computed with the three models are
completely different.

6-7
 Maxwell model Kelvin-Voigt model Standard model
0,035

0,030

0,025

11 0,020

0,015

0,010

0,005

0,000
0 20 40 60 80 100
t (days)

Figure 6-5: Creep strain 11 versus time t for the polymeric tube as described by the Maxwell
model, the Kelvin-Voigt model and the linear standard model.

6.4 Maxwell–Wichert model

Figure 6-6: Rheological illustration of the Maxwell–Wichert model with N internal variables.

The rheological illustration of the Maxwell–Wichert viscoelastic model is presented in

Figure 6-6. It consists of one elastic spring and N spring-dashpot combinations in parallel, and
represents an extension of the standard linear model of viscoelasticity. We will assume, as
before, that the material is isotropic and that only the deviatoric part of the stress-strain response
is viscoelastic.

6-8
 To describe the behaviour of the rheological model in Figure 6-6, we need to introduce N
internal variables that represent the inelastic strains in the N dashpots. We will denote these
inelastic strains qij , where   1, 2,..., N , and the internal stresses in the N chains will be
called Qij . The inelastic strain tensors qij and the internal stress tensors Qij are deviatoric
tensors per assumption.
From the rheological model, the deviatoric stress is obtained as
N
 ij  2G0 ij   Qij (6-24)
 1

where the internal stresses are given by

Qij  2G  ij  qij  (6-25)

Here, G0 and G are the shear moduli of the N  1 elastic springs of the Maxwell–Wichert.
The evolution equations for the internal variables qij follow from the rheological model as

qij 
1  G
2
Qij 

  ij  qij  (6-26)

where  is the viscosity of dashpot  . Combination of Equations (6-25) and (6-26)1 leads to
a set of linear differential equations in the internal stresses
G
Qij  Qij  2G  ij ,   1, 2,..., N (6-27)

The stress power expended by the dashpots is dissipated. The inelastic dissipation is
accordingly defined by
2G2
  qij  ij  qij   0
N N
Din   Qij qij  
 
(6-28)
 1  1  ij

It is evident that non-negative dissipation is ensured because the viscosities  are all assumed
positive from physical arguments.
The total stress tensor is obtained by adding the hydrostatic and deviatoric stresses and the
result is
N
 ij   P ij   ij  K  kk ij  2G0 ij   Qij (6-29)
 1

where K is the bulk modulus.

6-9
6.5 Numerical integration*
Since the constitutive equations for viscoelastic materials are expressed in rate form, we
need to integrate the equations in time. For the viscoelastic models the integration is done with
generalized Euler methods, which are rather simple methods [2].

6.5.1 Standard linear solid

We consider first the standard model in linear viscoelasticity, which is defined by the set
of equations (see Equations (6-19) and (6-20))
1 1
 ij   ij   iji ,  v   P
2G0 K
(6-30)
 i 1
21
ij   ij  2G1 iji 

Here G0 , G1 , 1 are the constants defining the deviatoric response, whereas the volumetric
response is elastic and defined by the bulk modulus K . Remember that ε i is deviatoric per
definition.
It is the evolution of the inelastic strain tensor that has to be integrated in time. We assume
that the state of the material is known as time tn , so that σn , Pn and ε in are known quantities.
In addition, the strain tensor ε n1 at time tn 1 is given. The time increment is tn1  tn1  tn .
As already mentioned, we adopt a generalized Euler method to integrate the constitutive
equations. Then, the algorithmic set-up for the viscoelastic material is
 ij , n1  2G0   ij , n1   iji , n1  , Pn1   K  v , n1
(6-31)
 iji , n1   iji , n 1       2G  i      2G  i
tn 1

21
 ij , n 1 ij , n   ij , n1 1 ij , n1 
21
where  is a parameter between 0 and 1. The forward-Euler method is obtained for   0
(explicit method), whereas   1 represents the backward-Euler method (implicit method). The
Crank-Nicholson method is obtained for   12 [2]. Combining the two equations in the
algorithmic set-up, the inelastic strain at tn 1 is found by the explicit expression

 i t 1    t  
 iji , n 1 
t 
1
  ij , n  n1
21
 ij , n  2G1 iji , n   n1 G0 ij , n 1  (6-32)
1
1  n 1  G0  G1   
1
Inserting Equation (6-32) into Equation (6-31)1, the deviatoric stress at tn 1 is found. We note
that the algorithm yields the state variables at tn 1 in one step without any iteration.

6-10
6.5.2 Maxwell–Wichert model
We will now derive an efficient and accurate integration algorithm for the Maxwell-
Wichert model. As before, we assume that the state of the material is known as time tn , so that
σn , Pn and Qn ,   1, 2,..., N , are known quantities. In addition, the strain tensor ε n1 at time
tn 1 is given. The time increment is tn1  tn1  tn .

The uncoupled, linear differential equations for the internal stresses Qij may be solved
analytically within the time step by assuming the strain rate to be constant and then using an
integrating factor to make the left-hand side of Equation (6-27) a total differential. To this end,
Equation (6-27) is multiplied by et / , viz.
1
et / Qij  et / Qij  2G  ij et / (6-33)

where     / G are the relaxation times of the Maxwell chains. The left-hand side is now a
total differential, namely

dt

d t / 

e Qij  2G  ij et / (6-34)

Assuming that ε is constant during the time step, integration from tn to tn 1 gives


etn1 / Qij,n1  etn / Qij,n  2G   ij etn1 /  etn /  (6-35)

Multiplication by etn1 / gives after some simple operations, the internal stresses at tn 1 by the
explicit equations
 ij ,n1

Qij,n1  etn1 / Qij,n  2G  1  etn1 /  tn1
,   1, 2,..., N (6-36)

With the internal stresses determined, the stress at tn 1 is obtained by use of Equation (6-29),
namely
N
 ij ,n1  K  kk ,n1 ij  2G0 ij ,n1   Qij,n1 (6-37)
 1

Again we obtain a stress update algorithm that is explicit in the sense that the stresses at tn 1
are obtained without iterations.

6-11
6.6 References
[1] F. Irgens. Continuum Mechanics. Springer, 2008.
[2] J. Lubliner. Plasticity Theory. Macmillan, 1990.
[3] L.E. Malvern. Introduction to the Mechanics of a Continuous Medium. Prentice-Hall,
1969.

6-12
7 Elastic-plastic materials
7.1 Basic assumptions
The theory of plasticity described in this chapter is restricted to small deformations and
isothermal conditions. The description is phenomenological, i.e., it is based on experimental
observations. It is further assumed that the behaviour of the material is rate-independent, and
thus the stress-strain behaviour is the same at all strain rates. This implies that the theory cannot
be used to describe phenomena like creep and stress relaxation. These are viscous phenomena,
depending on the rate of deformation (or the loading rate). The theory described in this chapter
is developed for quasi-static processes, where the rate of loading plays little role.
The theory of plasticity has three main ingredients: the yield criterion, the flow rule and
the work-hardening rule. We will describe these elements of the theory in that order. In
additions, we will establish the loading/unloading conditions in the theory of plasticity and
discuss the aspect of plastic dissipation.
For more comprehensive descriptions of the theory of plasticity the reader is referred to
various textbooks (e.g. [1]-[5]).

7.2 Yield criteria

7.2.1 General considerations
When a body is subjected to forces, the resulting deformations are first elastic. For metals
the response is typically linear, i.e., there is a linear relationship between stresses and strains.
As the stress level is increased, the material reaches a limit at which irreversible or plastic
deformations start to appear. This limit is called the yield limit. The transition between the
elastic domain, in which only elastic deformations occur, and the elastic-plastic (or simply
plastic) domain, where also the plastic deformations are significant, may be abrupt or more
gradual, depending on the actual material. Some mild steels exhibit a sharp yield point, while
other materials like most aluminium alloys and many high-strength steel materials show a
gradual transition into the plastic domain.
Mathematically the yield limit is described by the yield criterion, which is stated as
f σ   0 (7-1)

where f is the yield function. The function f is assumed to be a continuous function of the
stress tensor σ , which takes on negative values while the material is in the elastic domain.
Accordingly, the elastic domain is defined by the inequality
f σ   0 (7-2)
When this inequality is fulfilled, the material will only undergo elastic deformations. It is
further assumed that f  σ   0 is inadmissible (not allowed), which implies that plastic
deformation may take place only for f  σ   0 .

7-1
 Yield surface:
Inadmissible region:

Elastic domain:

Figure 7-1: Elastic domain, yield surface and inadmissible region.

Geometrically we can consider f  σ   0 to define a surface in stress space, i.e., the vector
space defined by the components of the stress tensor σ , and we denote this surface the yield
surface. The elastic domain is then enclosed by the yield surface, i.e., it is the part of the stress
space that is inside the yield surface. The plastic domain is the surface itself. Figure 7-1
illustrates the elastic domain, the yield surface and the inadmissible region for a two-
dimensional stress state. In the figure, σ1 , σ 2 , σ 3 represent three different stress states in the
elastic domain, on the yield surface and in the inadmissible region, respectively.
It is convenient to write the yield function in the form
f σ    σ   Y (7-3)

where  eq    σ  is the equivalent stress, measuring the magnitude of the stress state to which
the material is subjected, and  Y is the yield stress of the material. The yield stress is a material
property, which is determined by mechanical tests. The equivalent stress (and thus the function
 ) is assumed here to be a positive homogeneous function of order one of the stress, which
means that
  aσ   a  σ  (7-4)

where a is a non-negative scalar. By Euler’s theorem for homogeneous functions 1 , it then

follows that

1
See e.g. http://mathworld.wolfram.com

7-2
   σ 
 ij   σ  (7-5)
 ij
This is a very useful property of the equivalent stress.
As the material is strained plastically, work-hardening occurs, making the material
stronger. This reflects on the yield function, since the elastic domain evolves during work-
hardening. One way to account for work-hardening is to let the yield stress  Y depend on the
plastic straining. This is called isotropic hardening. Another way is to let the elastic domain
translate in stress space in direction of the plastic straining. This is called kinematic hardening,
and is important especially for cyclic loading (loading, unloading and reloading in the reverse
direction) or non-proportional (not following a straight line in stress space) loading paths. To
describe work-hardening, we need to introduce internal variables with evolution equations,
which reflect the changes of the microstructure of the material as a result of plastic straining.
We will come back to descriptions of work-hardening in Chapter 7.6. Now we will assume that
the structure of the material stays unchanged, so that the yield function only depends on the
stress state, i.e., there are no internal variables included and, hence, no work-hardening.
For some materials, like most metals and alloys, the yield criterion can be assumed to
depend only on the deviatoric stress state. Such materials are said to be pressure insensitive.
Examples of pressure-sensitive materials are friction materials (like concrete, soils, rocks) and
porous materials. The reason for the pressure insensitivity in metals and alloys is that the plastic
deformation to a large extent takes place by plastic slip, which is a shear-driven deformation
mode. For pressure-independent materials, we may express the yield function in the form
f  σ   0 (7-6)

where σ is the stress deviator. Similarly, for pressure dependent materials it is sometimes
practical to underline the pressure sensitivity by writing the yield function in terms of the stress
deviator σ and the hydrostatic stress  H :

f  σ,  H   0 (7-7)
It is recalled that the hydrostatic stress is equal to the pressure in magnitude but has the opposite
sign.

7-3
Figure 7-2: Illustration of the yield surfaces for an isotropic, pressure-dependent material (solid
line) and an isotropic, pressure-independent material (dashed line), where the dependence on
the pressure and the deviatoric stresses is represented by I and J 2 , respectively.
The plastic yielding of materials can often be assumed to be isotropic, meaning that the
yield function is independent of the direction of the loading within the material. This is for
example the case for polycrystalline metals and alloys with random texture, i.e., the grains have
arbitrary orientation in space. For isotropic materials the yield function should depend only on
either the principal stresses ( 1 ,  2 ,  3 ) or the principal invariants of the stress tensor
( I , II , III ), and thus we have

f 1 ,  2 ,  3   0 or f  I , II , III   0 (7-8)

If the yield function is expressed in terms of the principal stresses,  k ( k  1, 2,3 ), the
dependence should be symmetric, i.e., invariant to changes in the index k .
For isotropic and pressure independent materials, the yield function is most conveniently
written in terms of the principal invariants of the stress deviator, namely
f  J 2 , J3   0 (7-9)

where it is recalled that J1  0 by definition of the stress deviator. If instead the material is
isotropic and pressure dependent, we may express the yield criterion as
f  I , J 2 , J 3   0 (7-10)

where I   ii , J 2  12  ij ij and J 3  det σ . Examples of yield surfaces for an isotropic,
pressure-dependent material and an isotropic, pressure-independent material are illustrated in
Figure 7-2. The yielding of these materials is assumed to depend on the pressure through I
and the deviatoric stresses through J 2 . Note that I and J 3 take on both positive and negative
values, whereas J 2 is a non-negative measure.

7-4
Figure 7-3: Illustration of the principal stress space,  -plane and hydrostatic axis (after [4]).

For isotropic materials it is often useful to represent the yield surface graphically in the
principal stress space—also called the Haigh-Westergaard space. This space is illustrated in
Figure 7-3. The position vector OP of an arbitrary point 1 ,  2 ,  3  in the principal stress
space is decomposed into the vector OA along the line OZ and the vector OB in the  -plane
which, per definition, is perpendicular to OZ and passes through the origin O [4]. The line OZ
is called the hydrostatic axis, and it makes equal angles with the coordinate axes. Along the line
OZ , all three principal stresses are equal, 1   2   3 , and thus the vector OA represents the
hydrostatic part of the stress state. We will use the Haigh-Westergaard space for plotting of
isotropic yield functions and for deriving explicit expressions for the principal stresses in terms
of the stress invariants.
Let the base vectors  e1 , e2 , e3  of the coordinate system be aligned with the principal stress
directions, as shown in Figure 7-4 (a). The stress tensor σ is then represented in this basis by
the vector σ  1  2  3  , where we use the superimposed arrow to indicate vector
T

quantities—and thus to differentiate them from the underlying tensors.

7-5
 P

O P
A

(a) (b)
Figure 7-4: Stress representation in Haigh-Westergaard space: (a) principal stress space and (b)
deviatoric plane through point A.

Using Figure 7-4 (a), we get

OA   σ  n  n  1
3
1   2   3  n  3 H n (7-11)

where n  1 1 1 is the unit vector along the hydrostatic axis. Then

1 T
3

AP  OP  OA  σ  3 H n  σ (7-12)

where σ  1  2  3  is the deviatoric stress vector lying in a plane normal to the
T

hydrostatic axis. This plane is called the deviatoric plane. As an illustration, the deviatoric plane
through point A is depicted in Figure 7-4 (b), which also defines the unit vectors m i along the
projections of the base vectors ei onto the deviatoric plane. The deviatoric plane containing the
origin O is the  -plane, and it follows from the above that the equation for the  -plane is
1   2   3  0 (7-13)

Accordingly, every point 1 ,  2 ,  3  lying in the  -plane represents a purely deviatoric stress
state. The  -plane is plotted in Figure 7-5, as it is seen if one is looking along the hydrostatic
axis towards the origin. In this plot, the principal axes are projected onto the plane and appear
symmetrically located 120° apart.

7-6
 Tresca

Von Mises

Figure 7-5 Deviatoric plane (or  -plane) with projections of the Tresca and von Mises yield
surfaces. The dot is the projection of the stress point onto the deviatoric plane, where it has
been assumed that the stress point is located on the von Mises yield surface.

The distance r in the deviatoric plane to the projection of the stress point is simply the
magnitude of the deviatoric stress vector and thus

r  σ  σ  12   2 2   32  2 J 2 (7-14)

For instance, if we assume uniaxial tension, e.g.  1  0 and  2   3  0 , then the distance
along the  1 axis to the stress point equals

r  12   2 2   32   23 1     13 1     13 1  | 1 |
2 2 2
 2
3
(7-15)

The unit vectors m i in Figure 7-4 (b) are obtained by projecting the base vectors ei onto
the deviatoric plane, viz.
ei   ei  n  n
mi  , i  1, 2,3 (7-16)
ei   ei  n  n

7-7
where v  v  v is the magnitude of the vector v . The unit vectors are readily calculated
from Equation (7-16) as m1  2  1  1 , m2   1 2  1 and m3   1  1 2
1 T 1 T 1 T
6 6 6
.
Using Figure 7-4 (b), we get the relations
σ  m1  r cos L , σ  m2  r cos  23   L  , σ  m3  r cos  23   L  (7-17)

where  L is the so-called Lode angle. But we also have

σ  m1  3
2 1, σ  m2  3
2  2 , σ  m3  3
2  3 (7-18)
By combining these two equations, we get
1  2
3
J 2 cos L   2  2
3
J 2 cos  23   L    3  2
3
J 2 cos  23   L  (7-19)

where 0   L   / 3 . The third principal invariant of the stress deviator is obtained as

J 3  det  σ  1 2 3  3 83 J 23/2 cos L cos  23   L  cos  23   L  (7-20)

After some struggling with trigonometric identities, we finally arrive at

3 3 J3 
cos  3 L   3/2
, 0  L  (7-21)
2 J2 3

The relations between the principal stresses 1 ,  2 ,  3  and the principal stress invariants
 I , J 2 , J 3  are finally obtain as
1   H  2
3
J 2 cos  L
2  H  2
3
J 2 cos  23   L  (7-22)
3   H  2
3
J 2 cos  23   L 

where 0   L   / 3 and 1   2   3 . These relations are very useful to determine the

principal stresses when the stress components  ij are known. They will also prove useful in
finding the gradient f /  ij of a function f that takes the principal stresses as its
arguments—such as a yield function for an isotropic material.
In the following, we will describe some of the more common yield criteria used in
modelling of solid materials. We start by looking at pressure-independent, isotropic criteria,
then we move to an anisotropic criterion, and finally to a pressure-sensitive, isotropic criterion.

7.2.2 Tresca yield criterion

The Tresca criterion is the oldest yield criterion. It is an isotropic and pressure-independent
criterion. The assumption is that yielding starts when the maximum shear stress  max reaches a
critical value  Y , i.e.

7-8
  max  12  max   min    Y (7-23)

where  max and  min are the maximum and minimum principal stresses, respectively. Since the
uniaxial tension test is usually used to characterize the stress-strain behaviour of materials, it is
convenient to establish the relation between the yield stress and the maximum shear stress in
uniaxial tension. In uniaxial tension we have  max   and  min  0 , and it follows that at
yielding
Y
 max  (7-24)
2
where  Y is the uniaxial yield stress.
Combination of Equations (7-23) and (7-24) leads to
f 1 ,  2 ,  3    max   min    Y  0 (7-25)
To clearly express the yield function in the principal stresses, we may write
f 1 ,  2 ,  3   max  1   2 ,  2   3 ,  3  1    Y  0 (7-26)

This implies that the yield surface consists of the six planes
 1   3   Y  0,  2   3   Y  0
 2   1   Y  0,  3   1   Y  0 (7-27)
 3   2   Y  0,  1   2   Y  0
The elastic domain is the region in stress space which is enclosed by these six planes. The
projection of the Tresca yield surface onto the  -plane is shown in Figure 7-5.
The yield criterion consists of six hyper-planes in the stress space, and the projection onto
the  -plane of the yield surface makes a regular hexagon with vertices on the projections of
the positive and negative principal stress axes. In the 3D principal stress space, the Tresca yield
surface is a regular hexagonal prism with generator parallel to the hydrostatic axis.
By considering all combinations represented by Equation (7-26), it is rather
straightforward to show that the Tresca criterion alternatively could be expressed as

f  1 ,  2 ,  3  
1
2
 1   2   2   3   3  1    Y  0 (7-28)

which is believed to be the most convenient of several possible representations of this yield
criterion.

7.2.3 von Mises yield criterion

The von Mises criterion is based on the assumption of isotropy and pressure insensitivity
of the material. It is assumed that yielding occurs when the second principal invariant of the
stress deviator J 2 reaches a critical value k 2 , namely

J2  k 2 or J2  k (7-29)

7-9
This assumption is based on experimental evidence. Using the definition of the stress deviator,
 ij   ij  13  kk ij , we may express J 2 in terms of the stress components  ij as

J 2  12  ij ij  12  ij  13  kk ij  ij  13  mm ij   12  ij ij  16  kk mm

(7-30)
 12  112   22   332   122   21   23   322   132   312   16  11   22   33 
2 2 2 2

The uniaxial tension test is often used to determine the yield strength of a material. In
uniaxial tension, 11   is the only non-zero stress component, and at yielding    Y . Thus,
at yielding in uniaxial tension, we have

J 2  13  Y2  k 2 or k  1
3 Y (7-31)
By combining these equations, the von Mises yield function is obtained in the two
alternative forms
f  σ   3J 2   Y  0 or f σ   3
2  ij ij   Y  0 (7-32)

where the argument 3J 2 is expressed in terms of the stress components as

3J 2  32  ij ij   112   22

2
  332   11 22   22 33   33 11
(7-33)
 32  122   21
2
  23
2
  322   132   312 

The two versions are used depending on what is found most convenient in the actual
application. It is noted that plasticity theories based on the von Mises criterion is frequently
called J 2 flow theories, since the yield criterion depends on this invariant only.
Finally, we may express the von Mises yield criterion in terms of the principal stresses.
The result is

f 1 ,  2 ,  3   12   22   32  1 2   2 3   31   Y  0 (7-34)

or equivalently

f  1 ,  2 ,  3  
1
2

1   2    2   3    3  1    Y  0
2 2 2
 (7-35)

These forms are particularly useful for analytical solutions of problems with proportional
loading where the principal axes are fixed throughout the loading process.
The Tresca and von Mises criteria are compared in the  -plane in Figure 7-5. It is evident that
in the  -plane the projection of the von Mises yield surface represents a circle, while in the
3D principal stress space it is a right circular cylinder having generator parallel to the
hydrostatic axis. Figure 7-6 shows the geometric representation of the Tresca and von Mises
yield loci for plane stress conditions with  3  0 .

7-10
 von Mises

Tresca

Figure 7-6: Geometric representation of the Tresca and von Mises yield loci for plane stress
conditions with  3  0 .

Example 7-1: Tresca and von Mises criteria in plane stress states.
Consider a plane stress state with  3  0 . Show that the geometrical representations of the
Tresca and von Mises loci are as illustrated by Figure 7-6.

Starting from Equation (7-27) and setting  3  0 , we find that the Tresca yield locus is
defined by the lines
1   Y ,  2   Y ,  2  1   Y
These are exactly the lines used to construct the locus in Figure 7-6. For the von Mises yield
locus, we insert  3  0 in Equation (7-34), and the result is

12   22  1 2   Y2
This is equivalent to

 2   1    2   1  
2 2
 1  1

    1
2 2

 2 Y   2
3 Y 

which is just the equation for the ellipse drawn in Figure 7-6. The denominators 2 Y and
2
3  Y are correspondingly the semi-major axis and semi-minor axes of the ellipse.
■

7-11
7.2.4 High-exponent yield criterion
A generalized isotropic and pressure-independent yield criterion is obtained as a
generalization of Equation (7-35). This criterion gives a yield surface that lies in-between the
Tresca and von Mises surfaces. The high-exponent yield criterion (also called the Hershey
criterion) reads as [6][7]
1


1
2
m m
 m m
f  1 ,  2 ,  3     1   2   2   3   3   1    Y  0

(7-36)

where m  1 determines the curvature of the yield surface. For body centred cubic (bcc)
materials m  6 and for face centred cubic (fcc) materials m  8 have been proposed based on
calculations using polycrystal plasticity [8]. Ferrite steels are examples of bcc materials, while
aluminium alloys are examples of fcc materials.
It is noted that the criterion is equal to the von Mises criterion for m  2 , and becomes
more and more similar to the Tresca criterion as m increases. Too high values of m could lead
to numerical problems, and should be avoided. Also m  1 gives the Tresca criterion. The high-
exponent yield criterion is illustrated in Figure 7-7 for different values of the exponent m in
the case of plane stress states.

1.5

0.5

22
 0
0

-0.5

-1

-1.5
-1.5 -1 -0.5 0 0.5 1 1.5
11

0
Figure 7-7: Geometric representation of the high-exponent yield criterion for plane stress
conditions with  12  0 and m equal to 2 (outermost locus), 6, 8, 12, 20 and 60 (innermost
locus). Note that m  2 gives the von Mises criterion, whereas m  60 gives a yield criterion
practically indistinguishable from the Tresca criterion.

7-12
7.2.5 Quadratic yield criterion for anisotropic materials
The criteria presented above are all valid for isotropic materials only. However, it is
important to establish yield criteria that can be used also for anisotropic materials. A particularly
important class of anisotropy is the orthotropic symmetry that we find in rolled plates and
extruded profiles, and in many other materials. A yield criterion frequently used for orthotropic
metals is the Hill criterion [9].
Here we first define a quadratic yield criterion for anisotropic materials in a rather general
form, namely [2]
f (σ)  3
2 Pijkl ij kl   Y  0 (7-37)

where P is a 4th order tensor with the same symmetry as the elasticity tensor and  Y is a
reference yield stress. The symmetries of P are
Pijkl  Pjikl  Pklij (7-38)
which means that the number of independent constants of anisotropy is 21 (like a symmetric
6  6 matrix). If the yield criterion is further assumed to be pressure independent, it follows
that [2]
Pijkk  Piikl  0 (7-39)

This is realized by looking at a purely hydrostatic stress state defined by  ij   H  ij . If P has

the property defined by Equation (7-39), we have
Pijkl kl   H Pijkl kl   H Pijkk  0 (7-40)
Hence, only the deviatoric part of the stress tensor will contribute to yielding, and the criterion
is pressure independent. The number of independent constants of anisotropy is now reduced to
15.
Hill [9] considered the special case of orthotropic symmetry (i.e., three mutually
perpendicular planes of symmetry), assuming mean stress independent yielding. If the
Cartesian basis is chosen so that the coordinate planes are parallel to the symmetry planes, then
in this basis the components of P coupling normal stresses with shear stresses are zero. We
may then write the criterion as [2]

 eq2  32 Pijkl ij kl  F  22   33   G  33   11   H  11   22 

2 2 2

(7-41)
 L  23
2
  322   M  132   312   N  122   21
2

where F , G,..., N are constants. It is left as an exercise for the reader to determine the relations
between Pijkl and the constants F , G,..., N .
To determine the coefficients of the Hill criterion, it is necessary to perform mechanical
tests in different material directions to elucidate the plastic anisotropy. As an example, the three
Hill yield loci presented in Figure 7-8 are determined based on the two yield stresses in uniaxial

7-13
tension ( 11  0,  22  12  0 and  22  0, 11  12  0 ) and the yield stress in equi-biaxial
tension ( 11   22  0, 12  0 ).

1.5

0.5

22
 0
0

-0.5

-1

-1.5
-1.5 -1 -0.5 0 0.5 1 1.5
11

0
Figure 7-8: Geometric representation of the quadratic yield criterion for anisotropic materials
(or the Hill criterion) assuming plane stress conditions with  12  0 . Three examples based on
different values of the two yield stresses in uniaxial tension ( 11  0,  22  12  0 and
 22  0, 11  12  0 ) and the yield stress in equi-biaxial tension ( 11   22  0, 12  0 ).

Example 7-2: Calibration of the Hill yield criterion.

Uniaxial tensile tests along the three coordinate axes and shear tests in the three coordinate
planes have been conducted for a given material. The resulting tensile yield stresses are  Y 11 ,
 Y 22 and  Y 33 , whereas the yield stresses in shear are  Y 23 ,  Y 31 and  Y 12 . Use this to
determine the coefficients F , G,..., N of the Hill yield criterion.
We consider first the three uniaxial tension tests along the coordinate axes. In the test along
the x1 axis, the only non-zero stress component is  11 and at yielding its value is  Y 11 . Using
similar reasoning also for the tensile tests along the two other coordinate axes, Equations (7-37)
and (7-41) give
G  H  Y211   Y2 ,  F  H  Y222   Y2 ,  F  G  Y233   Y2
Solving these equations for the anisotropy coefficients, we get

7-14
  Y2  1 1 1   Y2  1 1 1   Y2  1 1 1 
F  2   2 
, G   2    , H   2  2  2 
2   Y 22  Y 33  Y 11 
2
2   Y 33  Y 11  Y 22 
2 2
2   Y 11  Y 22  Y 33 

If we take the x1 axis as the reference direction in our calibration, we get  Y 11   Y and it
follows that G  H  1 .
Next we apply the three shear tests in the three coordinate planes. In the shear test in the
x1 x2 plane,  12 and  21 are the two non-zero stress components and at yielding these
components have the value  Y 12 . Similar reasoning holds as well for the shear tests in the two
other coordinate planes. Using Equations (7-37) and (7-41), we immediately arrive at
 Y2  Y2  Y2
L , M , N
2 Y2 23 2 Y231 2 Y212
It should be noted that it is possible, and sometimes more convenient experimentally, to
identify the coefficients of the Hill criterion from uniaxial tension tests along six or more
material directions. Stress transformation is then required to find the expressions for the
coefficients F , G,..., N . Let  11 be the tensile stress (and the only non-zero stress component)
in one of the tensile tests expressed in the coordinate system xi with the x1 axis along the
tension direction. Then the stress components in the xi system are obtained from the tensor
transformation:  ij  a1i a1 j11 , where aij is the transformation matrix between the starred and
un-starred coordinate systems.
■

7.2.6 Linear transformation based yield criteria for anisotropic materials

Yield criteria for anisotropic materials can be constructed from isotropic yield functions
by using linear transformations of the stress tensor. This method is described in detail by Barlat
et al. [10], and the brief description below follows closely this reference.
Consider the following linear transformation of the stress tensor
 ijM  M ijkl kl (7-42)

where σ M is the transformed stress tensor and M is a 4th order tensor defining the anisotropy
of the material. The tensors are expressed in a basis associated with the material symmetry axes.
Assuming the transformed stress tensor σ M to be symmetric, M exhibits the minor
symmetries (see Chapter 5)
M ijkl  M jikl , M ijkl  M ijlk (7-43)

while the major symmetry is not required, i.e., generally M ijkl  M klij . If yielding is independent
of the hydrostatic stress, then M should fulfil the constraint [2]
M ijkk  0 (7-44)

7-15
The anisotropy coefficients contained in M are usually determined from experiments probing
the direction dependence of the mechanical behaviour. With M determined, the yield function
is expressed as
f (σ)  F 1M ,  2M ,  3M   0 (7-45)

where F is an isotropic function of σ M and, accordingly, expressed in terms of the principal

values 1M ,  2M ,  3M of σ M . It is noted that even if F is an isotropic function of its arguments,
f represents an anisotropic yield function due to the linear transformation.
An interesting yield criterion for metals and alloys is obtained by adopting the Hershey
isotropic yield function for pressure independent materials to define the isotropic function F .
The result is a high-exponent yield criterion for anisotropic materials with relatively few
independent anisotropy coefficients. More advanced yield functions may be constructed by
using two or more linear transformations. The reader is referred to Barlat et al. [10] for more
detailed discussions of linear transformation based yield criteria.

7.2.7 Drucker-Prager yield criterion

The Drucker-Prager yield criterion [11] is valid for isotropic and pressure-sensitive
materials. It has applications in modelling of frictional materials like concrete, soils and rocks.
Another possible application is polymers, which also often exhibit pressure sensitivity.
The pressure-dependent criterion is defined in terms of the stress invariants as
3 J 2   I
f σ    Y  0 (7-46)
1 
where  governs the pressure sensitivity of the material. The parameters  Y and  are
determined from tension and compression tests. Recall that the first stress invariant is defined
as I   kk  3P , where P is the pressure. Note that the von Mises criterion is obtained by
setting the parameter  equal to zero.
The Drucker-Prager criterion is shown graphically in Figure 7-9, where it is represented in
 
the space I / 3, 3J 2 defined by the two stress invariants. In the principal stress space, the
Drucker-Prager criterion represents a right circular cone about the hydrostatic axis.
It is noted that for uniaxial tension (  1  0 and  2   3  0 ), we have 3J 2   1 and
I   1 , and yielding ( f  0 ) occurs as 1   Y . In uniaxial compression (  3  0 and
1   2  0 ), 3J 2   3 and I   3 , and yielding starts for  3  11  Y . This explains the
graphical representation of the Drucker-Prager criterion in Figure 7-9. In the figure, the von
Mises criterion is represented by the bold dashed line, having the equation f  3J 2   Y  0 .
The von Mises criterion is a horizontal line in this diagram simply because it does not depend
on the pressure (or the hydrostatic stress).

7-16
 Uniaxial Uniaxial
compression tension

Figure 7-9: Geometric representation of the Drucker-Prager yield criterion in the space
 
I / 3, 3J 2 defined by the stress invariants.

7.3 Strain decomposition and plastic dissipation

In theories of plasticity for small deformations it is generally accepted to decompose the
strain tensor additively into elastic and plastic parts
 ij   ije   ijp (7-47)

where the elastic strain  ije is reversible and the plastic strain  ijp is irreversible. In metals and
alloys (crystalline materials), the elastic strain deforms the crystal lattice, and represents
reversible deformations that vanish when the material is unloaded. The plastic strain on the
other hand represents irreversible deformations in the form of plastic slip which leaves the
crystal lattice unaltered. The additive strain decomposition is illustrated in Figure 2-15 and
Figure 2-17 for an elastic-perfectly plastic material and an elastic-plastic material with work-
hardening, respectively.
We assume that the stress may be derived from an elastic potential, and we define the
elastic strain energy density as
1
U 0   ijeCijkl  kle (7-48)
2
Thus, the elastic relation (generalized Hooke’s law) becomes

7-17
 U 0
 ij   Cijkl  kle (7-49)
 ij
e

where Cijkl exhibits the major and minor symmetries and for elastic anisotropy there are 21
independent elastic coefficients. The stress rate is then given by the Hooke’s law on rate form,
namely
 2U 0 e
 ij   kl  Cijkl  kle (7-50)
 ij  kl
e e

The deformation power for the elastic-plastic material reads

Pd    ij ij dV    ij ije dV    ij ijp dV (7-51)
V V V

where time differentiation of the additive decomposition of the strain tensor was applied to
obtain
 ij   ije   ijp (7-52)
Use of Equations (7-48) and (7-49) then gives the following expression for the elastic
deformation power

  dV   U 0 dV  U
e
ij ij (7-53)
V V

where U   U 0 dV is the elastic strain energy of the body. Hence, the rate of change of the
V

elastic strain energy U is equal to the elastic deformation power. U is uniquely defined by the
elastic strain field of the body, and the elastic deformations are reversible and path independent.
It is understood that U is positive when loads are imposed on the body and negative when the
body is unloaded.
For dissipative deformation processes, like plastic deformations, it is assumed that work
has to be done to the body all the time for the deformation to continue, and thus we impose the
following restriction to the plastic deformation power

  dV  0
p
ij ij (7-54)
V

In view of Equations (4-45) and (7-51), the restriction defined by Equation (7-54) implies that

  dV   bi vi dV   ti vi dS
e
ij ij (7-55)
V V S

Thus, the elastic energy stored in the body per unit time is always smaller than or equal to the
power of the external forces.
If we assume that Equation (7-54) holds for any part of the body, however small, we arrive
at the inequality
D p   ij ijp  0 (7-56)

7-18
where D p expresses the plastic dissipation per unit volume, which is assumed to be dissipated
as heat.

7.4 Plastic flow rule

The plastic flow rule defines the plastic strain rate tensor ε p in a way that ensures non-
negative dissipation. In the most general case considered here, the plastic flow rule is defined
by
g
 ijp   (7-57)
 ij

where g  g  σ   0 is the plastic potential function and  is a non-negative scalar (   0 )

denoted the plastic parameter (or the plastic multiplier). Non-negative dissipation leads to the
following restriction on the plastic potential function
g
 ij 0 (7-58)
 ij

The plastic parameter  is determined based on the consistency condition in the theory of
plasticity, i.e., the stress must be located on the yield surface during continuous plastic flow.
This will be discussed in the next chapter.
If we assume that g is a positive homogeneous function of order one, we have from
Euler’s theorem for homogeneous functions that
g
 ij g (7-59)
 ij
With this assumption, the dissipation inequality is fulfilled [13], viz.
g
D p   ij ijp   ij  g  0 (7-60)
 ij
It is noted that one particular choice is to assume that the plastic potential function g is
defined by the yield function f , in the sense that
f
 ijp   (7-61)
 ij
This is called the associated flow rule, since the plastic potential function is associated with the
yield function, while flow rules in the form by Equation (7-57) is said to be non-associated.
Using Equations (7-3) and (7-5), we see straight away that non-negative dissipation is achieved
f 
D p   ij ijp   ij   ij    0 (7-62)
 ij  ij

where the equivalent stress  eq    σ  is non-negative by definition. If we consider the

symmetric tensors as vectors in a six-dimensional vector space, the associated flow rule implies

7-19
that the plastic strain increment vector dε p  ε p dt is parallel to the gradient of the yield surface
at σ and thus directed along the outward normal of the surface at this point. The associated
flow rule is therefore also called the normality rule (cf. Figure 3-3).
This property is visualized for the Tresca and von Mises yield surfaces in Figure 7-10. It
is important to note that the normality rule implies that the shape of the yield surface not only
determines the stress state at which yielding initiates, but also the direction of the plastic flow.
In some applications of the theory of plasticity, such as metal forming, this is of the utmost
importance.
Another possibility is to assume that g is a convex function of the stress. Let the function
g be sufficiently smooth. Then g is convex on a domain in the stress space if and only if
g
g  ij   g  ij    ij   ij  (7-63)
 ij

for all  ij and  ij , where the partial derivative is evaluated at  ij [2]. We further assume that
g is non-negative, g  0 , and contains the origin, g  0   0 [14]. Then by selecting  ij  0 ,
we get
g
 ij  g  ij   0 (7-64)
 ij
It immediately follows that the plastic dissipation is non-negative for such a convex plastic
potential function.
The normality rule may be derived from a stronger assumption than the dissipation
inequality defined by Equation (7-56). This stronger assumption is denoted the principle of
maximum plastic dissipation, which is stated as

 ij   ij   ijp  0 or  ij ijp   ij ijp (7-65)

where σ is the actual stress at which the plastic dissipation takes place and σ  is any stress that
fulfils the yield condition, in the sense that f  σ   0 .

7-20
Figure 7-10: Geometric representation of the associated flow rule for Tresca and von Mises
yield surfaces. The associated flow rule implies that the incremental plastic strain vector is
normal to the yield locus.

To establish the implications of the principle of maximum plastic dissipation, we follow

the argumentation in [2]. We first consider a smooth yield surface so that the tangent hyperplane
and normal exist at every point. It is then realized from the illustration in Figure 7-11 (left) that
the only way to fulfil Equation (7-65) for all σ  to the inward side of the tangent to the yield
surface, is to assume that ε p is directed along the outward normal at σ . It is further seen that
for σ  lying to the outward side of the tangent, as in Figure 7-11 (right), the inequality defined
by Equation (7-65) is violated. This entails that the elastic region must lie to one side of the
tangent and accordingly the yield surface must be convex. If the surface has corners at which
the normal is not unique, then ε p has to lie within the cone defined by the normal vectors
meeting at the corner, see Figure 7-11 (right).
We have thus established that the implications of the principle of maximum plastic
dissipation are the normality of the plastic flow and the convexity of the yield surface. The reader
is referred to [2] for a more in-depth description of the principle of maximum plastic dissipation.

7-21
Figure 7-11: Consequences of the principle of maximum plastic dissipation: normality of the
plastic flow and convexity of the yield surface.
As two examples, we will derive the plastic strain rate tensor ε p predicted by the associated
flow rule for the Tresca and the von Mises yield criteria.

7.4.1 The von Mises criterion

We start with the von Mises criterion as defined by Equations (7-32) and (7-33). Applying
Equation (7-61), we get the components of the plastic strain rate tensor as
11p  
2 eq  2 11   22   33  12p  3
2 eq  12 13p  3
2 eq  13
 21p  3
2 eq  21  22p  
2 eq  2 22   33   11   23p  3
2 eq  23 (7-66)
 31p  3
2 eq  31  32p  3
2 eq  32  33p  
2 eq  2 33   11   22 
where  eq  3J 2 is the equivalent stress. By inspection, we recognize that this is nothing but

3 3
 ijp 
2 eq
 ij  13  kk ij  
2 eq
 ij (7-67)

This equation states that the plastic strain rate tensor is proportional to the stress deviator and
is itself a deviatoric tensor. The implication is that  vp   iip  0 , i.e., the plastic volumetric
strain rate is zero. There is in other words no change of volume due to the plastic deformations
of the material, according to the von Mises yield criterion and the associated flow rule.
We are now in the position to calculate the plastic dissipation for this model, and the result
is

7-22
 3 3
D p   ij ijp 
2 eq
 ij  13  kk ij   ij 
2 eq
 ij ij   eq   0 (7-68)

It is clear that the dissipation is non-negative, since  is non-negative per assumption. From
this we conclude that  may be interpreted as an equivalent plastic strain rate, since it is the
product of  eq and  that is equal to the plastic dissipation.
Based on this, we define the equivalent plastic strain as
t t
p   pdt    dt (7-69)
0 0

so that per construction

D p   ij ijp   eq p (7-70)
We note that the equivalent plastic strain (which is sometimes called the accumulated plastic
strain) is monotonically increasing function of time. Its rate is obviously zero in the elastic
region where the plastic strain rate vanishes.
Combining Equation (7-68) with Equation (7-67) and using the relation p   , we may
readily show that

p 2
3  ijp ijp (7-71)

which is often the most convenient expression of the equivalent plastic strain rate for the von
Mises criterion and the associated flow rule.

7.4.2 The Tresca criterion

Before we look into the associated flow rule for the Tresca criterion, we need to define
carefully the derivative of the absolute value function y  x . We have

dy d
 x  sgn x (7-72)
dx dx
where we define the signum function as
1 for x0
sgn x   (7-73)
1 for x0
It is here assumed that sgn x can have any value between 1 and 1 for x  0 [2]. We note
that it is more common to define sgn 0  0 , as we did in Chapter 2.
Starting from the definition of the Tresca criterion given by Equation (7-28) and using
Equation (7-61), we obtain [2]

7-23
 1p  12   sgn  1   2   sgn  1   3  
 2p  12   sgn  2   1   sgn  2   3   (7-74)
 3p  12   sgn  3   2   sgn  3   1  
If we, for example, assume that 1   2   3 (i.e., the principal stresses are ordered and
distinct), the result is
1p   ,  2p  0,  3p   (7-75)

In the case of two equal principal stresses, say 1   2   3 , we get [2]

1p  12  1    ,  2p  12  1    ,  3p   (7-76)

where  is an arbitrary real number between 1 and 1 . This particular case is illustrated in
Figure 7-10. Similar expressions can be found for other combinations of the principal stresses.
We note that  vp  1p   2p   3p  0 and it follows that the plastic deformations conserve the
volume.
The plastic dissipation for the Tresca criterion is
D p   ij ijp  11p   2 2p   3 3p   eq   0 (7-77)

where the equivalent stress  eq is defined by Equation (7-28) as

 eq 
1
2
 1   2   2   3   3  1  (7-78)

The equivalent plastic strain is again defined by Equation (7-69). We may use Equation
(7-74) to show that for the Tresca criterion and the associated flow rule, the following relation
holds [2]

p 
1 p
2

1   2p   3p  (7-79)

We may use the associated flow rule in the same way for the other yield criteria defined in
Chapter 7.2 to establish the plastic strain rate tensor and further the expressions for the plastic
dissipation and the equivalent plastic strain. This is left as an exercise.

7-24
Example 7-3: Plane-strain tension test.

Figure 7-12: Plane-strain tension specimen for sheet material.

The plane-strain tension test is important in sheet metal forming for determining the
formability of the sheet material. For sheets plane stress conditions may be assumed, and thus
the stress components  33 ,  23 and  31 are all zero. We have here assumed that the x1 x2 plane
coincides with the plane of the sheet metal, while the x3 axis is along the normal direction of
the sheet.
Plane strain conditions are obtained by testing a notched specimen in tension, as shown in
Figure 7-12. By properly designing the geometry of the test specimen, the notch will constrain
the deformations in the transverse direction and the transverse strain will thus be approximately
equal to zero. Let the tension be along the x1 axis, so that the x2 axis is parallel the transverse
direction. Assume the von Mises yield criterion and the associated flow rule, and determine
the stress components in plane-strain tension at yielding.
We start by finding the expression for the von Mises criterion valid for plane stress states
with  12  0 . To this end, we insert  33   23   31  0 and  12  0 in Equations (7-32) and
(7-33), and the result is

f  11 ,  22    112   11 22   222   Y  0

 eq

where the expression for the equivalent stress  eq is identified explicitly. The associated flow
rule, Equation (7-61), gives according to Equation (7-66):

7-25
   
11p   211   22  ,  22p   2 22  11  ,  33p   11   22 
2 eq 2 eq 2 eq
Zero strain in the transverse direction implies
 22p  0   22  12 11
Inserting this into the yield criterion, we obtain the stress components at yielding in plane-stress
tension as

 11 
 Y
2

f  ,  11    11   Y  0  
1 3 3

 22  Y
11 2 2 1
 3

Hence, the yield stress in plane-strain tension will according to the von Mises criterion be 15.5%
higher than in uniaxial tension.
It should finally be noted that the strain in the transverse direction in a notched tension
specimen will seldom be exactly zero and further the strain field will not be homogeneous over
the notch region.
■

Remark 7.1: If the material is isotropic, the plastic potential is an isotropic function of the
stress, i.e., it is only a function of the stress invariants. By the result obtained in Chapter 3.4, it
then follows that the stress tensor σ and the plastic strain rate tensor ε p have the same principal
directions. In situations where the principal directions are unchanged during the loading process,
it is then possible to work with the principal stresses and principal plastic strains, which is useful
in analytical solutions of elastic-plastic problems.

7.5 Loading/unloading conditions

To distinguish between elastic loading/unloading and plastic loading, loading/unloading
conditions have to be established. These loading/unloading conditions will also provide a
method for determining the plastic parameter  in the case of plastic loading. The yield surface
is assumed to be stationary.
The definition of the yield condition entails that the plastic flow vanishes (i.e., ε p  0 )
within the elastic domain (i.e., for f  σ   0 ), and thus

f 0   0 (7-80)

Plastic flow may occur (i.e., ε p  0 ) only when the stress is situated on the yield surface (i.e.,
for f  σ   0 ), and therefore

f 0   0 (7-81)

The situation f  0 and   0 occurs during elastic unloading from a plastic state, i.e., the
stress is located on the yield surface but the stress increment is directed into the elastic domain.

7-26
 Based on Equations (7-80) and (7-81), the loading/unloading conditions may then be
expressed by the Kuhn-Tucker conditions
f  0,   0,  f  0 (7-82)
which constitute the constraints determining when plastic flow may occur in the elastic-plastic
material.
To determine the plastic parameter, we first note that during plastic flow (i.e., for   0 )
the yield function remains equal to zero: f  0 . This implies that f  0 at any time   0 ,
while f may take any value for elastic loading (   0 ). We have then arrived at the
consistency condition in the theory of plasticity:
f 0 (7-83)

Using the consistency condition for plastic loading (   0 ), we obtain in the case of a
stationary yield function
f
f   ij  0 (7-84)
 ij
Inserting Equations (7-50), (7-52) and (7-57) into Equation (7-84), we get
f  g 
f  Cijkl   kl   0 (7-85)
 ij   kl 
Solving for the plastic parameter, we arrive at
f
Cijkl  kl
 ij
 0 (7-86)
f g
Cijkl
 ij  kl
where we have assumed that
f f g
Cijkl  kl  0 and Cijkl 0 (7-87)
 ij  ij  kl
The loading/unloading conditions are illustrated in Figure 7-13.

7-27
 Yield Plastic
surface loading

Elastic
unloading

Elastic
domain

Figure 7-13: Loading/unloading conditions in the theory of plasticity.

7.6 Work-hardening rules

We have up to now assumed that the yield function only depends on the stress, but this is
normally not the case. In general, materials work-harden when they are deformed plastically,
and thus the strength increases.
There are different ways to account for the work-hardening of the material, and here we
will consider the two most common work-hardening rules. These are isotropic hardening and
kinematic hardening. Isotropic hardening implies that the elastic region expands (while keeping
its shape) in stress space during plastic deformation, while kinematic hardening involves rigid
translation of the elastic domain in stress space. Kinematic hardening is required to account for
the Bauschinger effect (see Chapter 2.7), and introduces a strain-induced anisotropy into the
material.

7.6.1 Isotropic hardening

To account for isotropic hardening, we introduce the isotropic hardening variable R ,
which represents the expansion of the elastic domain, see Figure 7-14.
The yield function in the presence of isotropic hardening takes the form
f  σ, R     σ    Y  R   0 (7-88)

where  eq    σ  is the equivalent stress,  Y   0  R is the flow stress of the material and
 0 is the initial yield stress. As the material is plastically deformed, the isotropic hardening

7-28
variable R and thus the flow stress  Y are both increasing, and as a result the elastic region is
enlarged in stress space.
It is usually assumed that the work-hardening depends on the equivalent plastic strain
p   pdt experienced by the material. The equivalent plastic strain p is defined as the energy
conjugate variable to the equivalent stress, i.e.
 ij ijp
D p   ij ijp   eq p or p (7-89)
 eq
For the associated flow rule, the equivalent plastic strain rate equals the plastic parameter,
namely
p (7-90)
while for non-associated flow rules Equation (7-89) should be used.

Figure 7-14: Isotropic hardening: the yield surface expands in stress space as the materials
work-hardens.

The isotropic hardening rule is defined quite generally in the form

R  hR  (7-91)

where it is assumed that the hardening modulus hR depends somehow on the state of the
material. Two frequently used isotropic hardening rules are the power law and the Voce rule,
where the work-hardening is uniquely defined by the equivalent plastic strain.
The power law is here formulated as

7-29
 R  p   Kp n (7-92)

where K and n are hardening parameters determined from material tests. Assuming the
associated flow rule (so that p   ), the hardening modulus is

hR  Knp n1 (7-93)

The Voce rule is define by
R  p   QR 1  exp  CR p   (7-94)

Here, QR and CR are hardening parameters. It is noted that QR represents the saturated value
of R , i.e., the maximal value R can take. Again assuming associated flow, we get for the
hardening modulus
hR  CRQR exp  CR p   CR  QR  R  (7-95)
There are numerous isotropic hardening rules proposed in the literature and the choice of
a particular law depends on the actual material and the applications.
The next step is to determine the plastic parameter. The loading/unloading conditions
remain the same as for the case of perfect plasticity, but in the case of plastic loading, the
consistency condition now takes the form
f f
f   ij  R0 (7-96)
 ij R
Using in turn Equations (7-50), (7-52), (7-57), (7-88), (7-91) and (7-90), we obtain
f  g 
f  Cijkl   kl     hR   0 (7-97)
 ij   kl 
Solving for the plastic parameter, we have
f
Cijkl  kl
 ij 1  f 
   Cijkl  kl   0 (7-98)
f g  
Cijkl  hR H   ij 
 ij  kl
where we have assumed that
f f g
Cijkl  kl  0 and H  Cijkl  hR  0 (7-99)
 ij  ij  kl
We recognize again the inequality to the left as the criterion for plastic loading, while the
inequality to the right is required to avoid singularities.

7-30
Example 7-4: Temperature increase in adiabatic processes.
A metallic material obeying the Voce hardening rule is tested in tension at constant, high
strain rate in a high-speed testing machine and the material constants are identified based on
these high-rate tests. Adiabatic conditions are assumed, which means that the deformation
process is so fast that the heat generated due to plastic work has no time to escape. Under these
conditions, the rate of temperature increase is defined by
 c pT  TQ ij ijp
where  is the density, c p is the specific heat capacity, T is the temperature and TQ is the
Taylor-Quinney coefficient. The coefficient TQ defines the percentage of the plastic power
that is dissipated as heat, while (1  TQ ) % is stored in the material due to structural
rearrangements. We will here assume that TQ is constant, while this is generally not the case,
even if the value TQ  0.9 is often used in applications. Establish the expression for the
temperature as a function of the equivalent plastic strain for this material under adiabatic
loading conditions. It is assumed that the initial temperature is T0 , while the equivalent plastic
strain is initially zero.
Using Equation (7-89), we have before necking
 ij ijp   eq p   0  R  p
where the latter equality follows from the yield criterion, namely: f   eq  ( 0  R)  0 .
Inserting the Voce rule for R , according to Equation (7-94), we obtain the rate of temperature
increase in the form
TQ
T   QR 1  exp  CR p   p
cp  0
Integration of this expression, using the given initial values, the temperature as function of
the equivalent plastic strain is obtained as
TQ  
T  p   T0   0  QR  p 
cp 
QR
 exp  CR p   1
CR 
The temperature evolution is illustrated in Figure 7-15 using values representative for high-
strength steel. The temperature increase is only important at large plastic strains, since steels
with a melting temperature of about 1500°C are usually not very sensitive to temperature
increases below say 100°C.

7-31
Figure 7-15: Temperature evolution during high-rate deformation of a high-strength steel
assuming adiabatic conditions:  0  400 MPa , QR  400 MPa , CR  10 , TQ  0.9 ,
c p  490 J/kg K and   7850 kg/m3 .
■

7.6.2 Extension to kinematic hardening*

As explained in Chapter 2.7, kinematic hardening is required for cyclic or non-proportional
loading conditions. The backstress tensor χ is introduced to describe kinematic hardening of
the material. The yield function is now defined by assuming that the backstress tensor χ defines
the centre of the elastic region in stress space, see Figure 7-16. It acts like a residual stress, and
lowers the yield strength under reversed loading. In this way, the Bauschinger effect is
described. The yield function is now defined by
f  σ  χ, R     σ  χ    Y  R   0 (7-100)

where   σ  χ  is the equivalent stress with respect to χ , and σ  χ is the effective stress
tensor.
With reference to Chapter 2.7, the equivalent plastic strain p is defined as the energy
conjugate variable to the equivalent stress with respect to χ , i.e.,

  ij   ijp
 ij  ij      σ  χ  p or
p
ij p
ij

 σ  χ 
(7-101)

If the associated flow rule is adopted, we have p   , according to Euler’s theorem for
homogeneous functions.

7-32
Figure 7-16: Kinematic hardening: the yield surface translates in stress space as the materials
work-hardens.

The evolution equation for the backstress tensor is generally defined by

ij  hij  (7-102)

where the second-order tensor h  depends on the state of the material through σ , χ and R .
Linear kinematic hardening is obtained by assuming that the rate of the backstress tensor is
proportional to the plastic strain rate tensor ε p . The result is Prager’s linear kinematic
hardening rule (see e.g. [5])
ij   ijp (7-103)
where  is a constant. An alternative linear kinematic hardening rule was later proposed by
Ziegler (see e.g. [5], [12]) in the form
 ij  ij
ij   p (7-104)
 σ  χ 
where  is a constant. It is noted that the tensor σ  χ points from the centre of the elastic
region to the stress point on the yield surface. Several nonlinear hardening rules have been
proposed to better describe the behaviour of materials subjected to cyclic loading or non-
proportional straining. Armstrong-Fredericks’ nonlinear kinematic hardening rule is defined as
(see [14] for more details)
ij  1 ijp   2 ij p (7-105)

where 1 and  2 are parameters. For  2  0 we retrieve Prager’s linear rule. We note that 1
determines the initial kinematic hardening rate for χ  0 , but as χ grows larger, the second

7-33
contribution governed by  2 becomes increasingly important, leading to a diminishing
kinematic hardening rate—also called dynamic recovery. An alternative version of Armstrong-
Fredericks’ nonlinear kinematic hardening rule reads [12]
 ij  ij
ij  1 p   2 ij p (7-106)
 σ  χ 
where 1 and  2 are parameters. For  2  0 we retrieve Ziegler’s linear rule. The parameters
1 and  2 are usually determined from cyclic tests.
It should be noted that for von Mises plasticity with the associated flow rule Prager’s and
Ziegler’s hardening rules are equivalent in three-dimensional cases [5]. The same holds for the
two versions of the Armstrong-Fredericks’ rule.

Example 7-5: Non-linear kinematic hardening in uniaxial stress states.

Consider a metallic material that is loaded cyclically in uniaxial tension-compression. The
material obeys the associated flow rule and exhibits combined non-linear isotropic and
kinematic hardening according to the Voce and Armstrong-Fredericks hardening rules,
respectively. Express the constitutive equations for the material using the notation from Chapter
2, and obtain the closed form solution for the stress-plastic strain curve within a half-cycle.
Assume that at the beginning of the half-cycle the plastic strain is  0p , the equivalent plastic
strain is p0 and the backstress equals  0 .
In a uniaxial stress state, the yield criterion given by Equation (7-100) reduces to
f    , R   |    |   0  R   0
where the isotropic and kinematic hardening rules are defined by Equation (7-94) and the
reduced form of Equation (7-106), namely
  1   2   p    1   2    p

where we have introduced the notation   sgn     and used the relation p   p , which
holds in the uniaxial case (see Chapter 2.7). Integration of this differential equation for the
backstress  in terms of the plastic strain  p , assuming initial values  0 and  0p , gives

1   
  p       0   1  exp   2  p   0p  
2  2 
It has been assumed that 1 and  2 are constants. The yield criterion then gives the stress as a
function of the plastic strain within the half-cycle as
    0  R  p     p 

where |    |       was used, and within the half-cycle we have the following relation
between the equivalent plastic strain p and the plastic strain  p

7-34
 p  p0    p   0p 

The stress is seen to be a unique function of the plastic strain within the half-cycle. The
behaviour of the material is illustrated in Figure 7-17. The relations developed in this example
are useful in calibration of the material constants using cyclic tension-compression tests.

Figure 7-17: Uniaxial tension-compression behaviour of a metallic material during cyclic

loading:  0  100 MPa , QR  100 MPa , CR  10 , QR  100 MPa , 1  5000 MPa and
 2  50 .
■

The final step is to establish the expression for the plastic parameter. We will show this
derivation assuming the non-associated flow rule and combined isotropic and kinematic
hardening. The consistency condition then takes the form
f f f
f   ij  ij  R0 (7-107)
 ij ij R
Using Equations (7-50), (7-52), (7-100), (7-57), (7-91) and (7-102), we obtain
f  g  f 
f  Cijkl   kl    hij   hR   0 (7-108)
 ij   kl   ij

where we have used the relation f / ij  f /  ij which follows from the definition of the
yield function in Equation (7-100). Solving for the plastic parameter, we readily obtain

7-35
 f
Cijkl  kl
 ij 1  f 
   Cijkl  kl  (7-109)
f g f   
Cijkl  hij  hR H   ij 
 ij  kl  ij
where we have assumed that
f f g f 
Cijkl  kl  0 and H  Cijkl  hij  hR  0 (7-110)
 ij  ij  kl  ij
As before, we identify the inequality to the left as the criterion for plastic loading, while
the inequality to the right is required to avoid singularities.

7.7 Rate constitutive equations (tangent modulus)

It is sometimes important to establish the explicit relation between the stress rate and the
strain rate for the elastic-plastic material. This will define the tangent modulus tensor for the
elastic-plastic material. We will in the following show the derivation for the elastic-plastic
material with non-associated flow.
To this end, we start from the rate form of the generalized Hooke’s law (Equation (7-50)),
taking into account the decomposition of the strain rate (Equation (7-52)). We have
 ij  Cijkl  kl   klp  (7-111)

Assuming plastic loading, we may apply the flow rule (Equation (7-57)) and the expression
for the plastic parameter (Equation (7-98) for isotropic hardening or (7-109) for combined
hardening) to get
 1  f  g 
 ij  Cijkl   kl   Cmnpq pq   (7-112)
 H   mn   kl 
Or alternatively (after some struggle with dummy indices), we arrive at
 1 g   f 
 ij   Cijkl   Cijpq   Cmnkl    kl (7-113)
  pq  
 H    mn
This expression is valid for plastic loading (   0 ), while for elastic loading/unloading (   0 )
the relation is  ij  Cijkl  kl .
The rate equation for the elastic-plastic material is therefore given by
 ij  Cijkl
t
 kl (7-114)
where
Cijkl for elastic loading/unloading

 1 g   f 
t
Cijkl (7-115)
C 
 ijkl H  Cijpq    Cmnkl  for plastic loading
  pq    mn 

7-36
 t
The 4th order tensor Cijkl is called the tangent modulus tensor of the elastic-plastic model.
t
If the associated flow rule is assumed, Cijkl has the major symmetry: Cijklt
 Cklij
t
. This has
important consequences for implementation of elastic-plastic constitutive models into the finite
element method. Non-associated models lead to non-symmetric tangent stiffness matrices and
thus make the solution procedure more complex.

7.8 Numerical integration*

For the elastic-plastic material models return mapping algorithms are normally applied to
integrate the constitutive equations in time [15]. These algorithms are accurate, robust and
efficient—and are used today in most FE codes.
The constitutive relations for the elastic-plastic material are now summarized by the set of
equations:
 ij  Cijkl   ij   ijp 

 ijp   gij  σ, ξ  ,    h  σ, ξ  (7-116)

f  σ, ξ   0,   0,  f  σ, ξ   0

where we have introduced the notation gij  g /  ij to simplify the expressions. The vector
of internal variables  with   1, 2,..., N includes the isotropic hardening variable R , the
kinematic hardening variable  ij and possibly other internal variables. The state functions h
define the evolution of the internal variables  . The number of internal variables is denoted
N , where each component of tensor variables is counted as one internal variable. As an
example, if the only internal variables are R and  ij , then N  1  6  7 , where the symmetry
of  ij has been accounted for.
For the elastic-plastic model in Equation (7-116) a return mapping algorithm will be
employed [15]. Assume that at time tn the stress, strain and internal variables are known and
given by  ij , n ,  ij , n and  , n . Then during the time increment tn1  tn1  tn an incremental
change  ij , n1 in the strain tensor occurs. The assignment is then to find the stress and the
internal variables at tn 1 so that the constitutive relations are fulfilled.
To this end, an elastic trial stress is calculated by assuming that the strain increment is
purely elastic, i.e., there is no plastic flow during the time interval tn 1 . This leads to a trial
state, defined by  ijtr, n1   ij , n  Cijkl  kl , n1 and tr, n1   , n , which is the correct state at tn 1
provided
f  σtrn1 , ξtrn1   0 (7-117)

If this inequality is violated, the stress is situated outside the yield surface and it has to be
brought back to the surface in a way that is consistent with the flow and hardening rules.

7-37
Generalized mid-point rule
We first consider the generalized mid-point rule [16], which has the semi-implicit method
and the backward-Euler method as two important cases. The equations describing this return
mapping to the yield surface are then given by
 ij , n 1   ijtr, n 1  n 1Cijkl g kl  σ n  , ξ n  
 ,n 1   ,n  n 1h  σ n  , ξ n   (7-118)

f  σ n 1 , ξ n 1   0

where n 1 is the “incremental” plastic parameter. The state variables σ n  and ξ n  at time
tn   tn1  1    tn are defined as

σ n    σ n 1  1    σ n
(7-119)
ξ n    ξ n 1  1    ξ n

The scalar  takes values between zero and unity. Note that the yield condition is enforced at
the end of the time step. Equation (7-118) defines a set of three nonlinear equations in the
unknown quantities σ n1 , ξ n1 and n 1 , which may be solved by a suitable iterative solution
method, e.g. Newton’s method. Figure 7-18 illustrates the generalized mid-point rule for
elastic-plastic material models.
If  is taken equal to zero, a semi-implicit algorithm is obtained [15]. According to
Equation (7-118) with   0 , the state variables σ n1 and ξ n1 depend only on n 1 and the
known quantities at tn . The results is a nonlinear equation in n 1 , defined by the yield
condition as
f  ijtr, n1  n1Cijkl gkl  σ n , ξ n  ,  , n  n1h  σ n , ξ n    0 (7-120)

This equation is readily solved by some iterative technique, e.g. Newton’s method or the secant
method.
For all other choices of the parameter  implicit algorithms are obtained and we have to
solve the system of nonlinear equations given by Equation (7-118). It should be noted that   1
leads to the backward-Euler method [15], which is very robust for large time steps, and most
often used in FE codes.

7-38
Figure 7-18 Illustration of generalized mid-point rule for elastic-plastic material models.

To illustrate the use of Newton’s method, we rewrite the system of equations defined by
Equation (7-118) in the residual form:
 ij , n 1   ijtr, n 1  n 1Cijkl g kl  σ n  , ξ n    0

 , n 1   ,n  n 1h  σ n  , ξ n    0 (7-121)

f  σ n 1 , ξ n 1   0
This equation system may formally be written in an abstract form as
y  x  0 (7-122)

where y represents the vector of residuals in Equation (7-121) and x stands for the vector of
unknown quantities comprising σ n1 , ξ n1 and n 1 . An iterative solution scheme is obtained
by linearization of the nonlinear vector residual around the state x I and taking the approximate
expression to be zero, i.e.
y  xI 1   y  xI   J I  xI 1  xI   0 (7-123)

where I is an iteration counter ( I  0,1, 2,3,... ) and the so-called Jacobean matrix reads
y  x 
J (7-124)
x
The elastic trial state is used to define the initial values of the unknown quantities. Solving
for the new state x I 1 , we get

7-39
 xI 1  xI  J I1 y  xI  (7-125)
The iteration stops when the norm of the residual is less than a prescribed tolerance, i.e.

y  xI 1   yT  xI 1  y  xI 1   tol. (7-126)

It is not absolutely necessary to re-calculate the Jacobean matrix for all iterations in order
to obtain a converged solution [2]. This will reduce the computation cost of each of the
iterations, but the number of iterations used to get a converged solution will normally increase.
It is noted that for the semi-implicit scheme the solution only involves one nonlinear
equation, whereas for implicit schemes (like the backward-Euler method) a system of nonlinear
equation has to be solved. While the implicit methods are more computationally expensive,
they are also usually more accurate and robust, allowing for larger strain increments.

Cutting plane method

An alternative method which is well suited for rate-independent plasticity is the cutting
plane method [17]. The advantage of the cutting plane algorithm is that only one nonlinear
equation has to be solved within each of the iterations.
Assume that a trial stress state has been computed, and that the yield condition is violated,
i.e., f  σtrn1 , ξtrn1   0 . Consider the iterative update scheme

 ijI ,n11   ijI , n 1  nI1Cijkl g klI ,n 1

I ,n11  I ,n 1  nI1hI ,n 1 (7-127)

f nI11  0

where I is an iteration counter ( I  0,1, 2,... ), nI1 is the iterative change of the incremental
plastic parameter n 1 , and further

gijI ,n 1  gij  σ nI 1 , ξ nI 1 

I ,n 1    σ nI 1 , ξ nI 1  (7-128)

f nI1  f  σ nI 1 , ξ nI 1 

The initial state I  0 is defined by the trial state, i.e., σ 0n1  σtrn1 and ξ 0n1  ξtrn1 . A Taylor
expansion of the yield function after I iterations at time tn 1 gives
I
f f
I
f I 1
n 1  f I
n 1   I
ij , n 1  I ,n1  0 (7-129)
 ij n 1
 n 1

where the summation convention is applied for sub-indices i, j, k , l  1, 2,3 and   1, 2,..., N .
Using Equation (7-127), we find that the iterative changes of the state variables are

7-40
  ijI ,n1  nI1Cijkl gklI , n1 , I ,n1  nI1hI ,n1 (7-130)
Combining Equations (7-129) and (7-130), and solving for the iterative change of the
plastic parameter  , we get
f nI1
nI1  (7-131)
I
f f
I
Cijkl g klI , n 1  hI , n 1
 ij n 1
 n 1

The plastic parameter is updated as nI11  nI1  nI1 , whereas the state variables σ and ξ
are updated iteratively according to Equation (7-127). The iteration procedure stops when the
yield condition is fulfilled to within a given tolerance, i.e., f  σ nI 1 , ξ nI 1   tol.

The reader is referred to the textbook of Simo and Hughes [18] for a comprehensive
description and analysis of return mapping algorithms in computational inelasticity.

7.9 References
[1] F. Irgens. Continuum Mechanics. Springer, 2008.
[2] J. Lubliner. Plasticity Theory. Macmillan, 1990.
[3] L.E. Malvern. Introduction to the Mechanics of a Continuous Medium. Prentice-Hall,
1969
[4] G.E. Mase. Continuum Mechanics. Schaum’s Outline Series. McGraw-Hill, 1970.
[5] A.S. Khan, S. Huang. Continuum Theory of Plasticity. Wiley, 1995.
[6] A.V. Hershey. The plasticity of an isotropic aggregate of anisotropic face centred cubic
crystals. Journal of Applied Mechanics 21(1954) 241–249.
[7] W.F. Hosford. A generalized isotropic yield criterion. Transactions of the ASME. Series
E. Journal of Applied Mechancis 39 (1972) 607–609.
[8] W.F. Hosford. The Mechanics of Crystal and Textured Polycrystals. Oxford University
Press, 1993.
[9] R. Hill. A theory of the yielding and plastic flow of anisotropic metals. Proc. Roy. Soc.
London, 193 (1948) 281–297.
[10] F. Barlat, J.W. Yoon, O. Cazacu. On linear transformations of stress tensors for the
description of plastic anisotropy. International Journal of Plasticity 23 (2007) 876–896.
[11] D.C. Drucker, W. Prager. Soil mechanics and plastic analysis for limit design. Quarterly
of Applied Mathematics 10 (1952) 157–165.
[12] ABAQUS Theory manual. Version 5.8. Hibbitt, Karlsson & Sorensen, Inc. 1998.
[13] N. Saabye Ottosen, M. Ristinmaa. The Mechanics of Constitutive Modeling. Elsevier,
2005.
[14] J. Lemaitre, J.-L. Chaboche. Mechanics of Solid Materials. Cambridge University Press,
1990.

7-41
[15] T. Belytschko, W. K. Liu, B. Moran. Nonlinear finite elements for continua and
structures, Wiley, 2000.
[16] M. Ortiz, E.P. Popov. Accuracy and stability of integration algorithms for elastoplastic
constitutive relations. International Journal for Numerical Methods in Engineering 21
(1561-1576) 1985.
[17] M. Ortiz, J.C. Simo. Analysis of a New Class of Integration Algorithms for Elastoplastic
Constitutive Relations. International Journal for Numerical Methods in Engineering 23
(1986) 353–366.
[18] J.C. Simo, T.J.R Hughes. Computational Inelasticity. Springer, 1998.

7-42
8 Elastic-viscoplastic materials
8.1 Basic assumptions
In the previous chapter, we formulated the theory for rate-independent plastic
deformation of materials. To take into account the rate-sensitivity of plastic behaviour, we
have to extend the theory. The rate-dependent theory of plasticity is denoted viscoplasticity
[1]-[3]. This theory is applicable to describe phenomena like creep and stress relaxation as
well as the behaviour of materials at elevated strain rates, which is important in simulations of
e.g. plastic forming, crashworthiness and structural impact. Since the rate dependence of
materials tend to increase with temperature, the theory of viscoplasticity is even more
important at elevated temperatures than at room temperature.
Note that both viscoelasticity and viscoplasticity describe rate-dependent material
behaviour, and can thus be used to describe e.g. creep and stress relaxation. The difference is
that in viscoplasticity the material is usually assumed to be rate-independent in the elastic
domain. This is a reasonable assumption for metals but not always for polymers.

8.2 General formulation

Since the deformations are assumed to be infinitesimal, the strain tensor is additively
decomposed into elastic and plastic parts:
 ij   ije   ijp (8-1)

where the elastic strain  ije is reversible and the plastic strain  ijp is irreversible. The linear
relationship between stress and elastic strain is defined by the generalized Hooke’s law as
 ij  Cijkl  kle (8-2)

where Cijkl is the 4th order tensor of elastic coefficients.

We include only isotropic hardening in the formulation, and define the yield function as
f  σ, R     σ    0  R  (8-3)

where  eq    σ  is the equivalent stress,  0 is the yield stress and R  R  p  is the isotropic
hardening variable which depends on the equivalent plastic strain p . In the theory of
viscoplasticity, only elastic strains are assumed to develop while the stress is inside or on the
yield surface ( f  0 ). Only when the stress moves outside the yield surface, plastic strains are
assumed to evolve, see Figure 8-1. The plastic strain rate is further assumed to increase
monotonically as the stress moves away from the yield surface. In this way the rate sensitivity
of the material is accounted for, i.e., the flow stress increases with increasing plastic strain
rate.

8-1
 Elastic domain

Figure 8-1: The elastic domain, the yield surface ( f  0 ) and the stress situated outside the
yield surface giving rise to plastic flow.

As in the theory of plasticity, the plastic flow rule defines the plastic strain rate tensor by
g
 ijp   (8-4)
 ij

where g  g  σ  is the plastic potential function and   0 is the plastic parameter. If the
plastic potential g is taken as the yield function f , we obtain the associated flow rule
f
 ijp   (8-5)
 ij
As before, the equivalent plastic strain p is defined as
 ij ijp  ij g
p   pdt , p   (8-6)
 eq  eq  ij

If g  g  σ  is a positive homogeneous function of order one, we have  ij g /  ij  g , and

thus
g
p (8-7)


8-2
where  eq    σ  was used. In the case of the associated flow rule, g   , which gives, as
before,
p (8-8)
Since the stress is allowed to move outside the yield surface, the consistency condition is
not applicable to viscoplastic materials. Instead, the plastic parameter is determined based on
a constitutive relation depending on the state of the material as defined by the stress and the
isotropic hardening variable:

0 for f  σ, R   0
 (8-9)
  σ, R  for
 f  σ, R   0

The constitutive relation in Equation (8-9) determines the rate sensitivity of the elastic-
viscoplastic material. A special case of Equation (8-9) is

0 for f  σ, R   0
 (8-10)
  f  σ, R   for
 f  σ, R   0

where the dependence of the yield function f is explicitly displayed. The simplest choice is
to assume   f  σ, R  /  , where  is constant and represents the viscosity of the material.
By viscosity we mean the resistance of the material to plastic flow under deviatoric stresses.
It is noted that the extension of the formulation to include kinematic hardening is readily
obtained along the same lines as shown for the rate-independent theory.
Assuming that the elastic coefficients are constant (isothermal conditions), the rate
constitutive equations for the elastic-viscoplastic material read as
 g 
 ij  Cijkl  kle  Cijkl  kl   klp   Cijkl   kl    (8-11)
  kl 
Two important cases are considered: creep and stress relaxation. In the case of creep, the
stress is kept constant,  ij  0 , and thus the creep strain becomes

g
 kl   klp   (8-12)
 kl
The creep stops if and only if the elastic domain has expanded/translated by hardening so that
the stress is located on the yield surface. Stress relaxation implies that the total strain rate is
zero,  kl  0 , and then the stress rate is
g
 ij  Cijkl  kle  Cijkl (8-13)
 kl
whereas the relation between the elastic and plastic strain rates is
g
 kle   klp   (8-14)
 kl

8-3
The relaxation process stops when the stress has relaxed to the expanding/translating yield
surface. We conclude that the elastic-viscoplastic model is capable of modelling the important
processes of creep and stress relaxation that occur within the plastic domain of the material.

8.3 Special case: Rate-dependent J2 flow theory

The rate-dependent J2 flow theory is valid for elastically and plastically isotropic
materials, and the onset of yielding is described by the von Mises yield criterion. The rate
sensitivity is described either by an additive or a multiplicative constitutive relation. Only
isotropic hardening is included.
The stress is defined by the isotropic Hooke’s law in the form (Chapter 5.3)
 ij  Cijkl  kle  e kke ij  2e ije (8-15)

where e and e are the Lamé constants. The yield function is defined by (Chapter 7.2.3)

f  σ, R   3
2  ij ij   0  R  or f  σ, R    eq  σ    0  R  (8-16)

where  eq  3
2  ij ij is the equivalent stress. We will assume here that R is a user-defined
monotonically increasing function of the equivalent plastic strain, i.e., R  R  p  , which is a
usual assumption in nonlinear finite element codes.
The associated flow rule is adopted, and thus according to Chapter 7.4.1, we have
3 ij
 ijp   (8-17)
2 eq
The equivalent plastic strain for the associated J2 flow theory is defined by
p    dt (8-18)

It then remains to define the constitutive relation, i.e., the relation for  in terms of the state
variables σ and R .

8.3.1 Additive constitutive relation

An additive constitutive relation defines the equivalent stress as a sum of the yield stress
 0 , the strain hardening R and the strain-rate hardening  v as
 eq  σ    0  R  p    v  p  or f  σ, R    v  p  (8-19)

Here  v  0 is a monotonically increasing function of the plastic strain rate and is denoted the
viscous stress. It represents the strengthening of the material caused by the elevated strain rate.
In the limit of vanishing plastic strain rate, the viscous stress tends towards zero. Hence, in the
plastic domain  v  0 . Solving Equation (8-19) for p (   ), we get a relation in the form


0 for f  σ, R   0
p (8-20)
  σ, R  for
 f  σ, R   0

8-4
 If we assume that the viscous stress is defined by a power law (similar to the so-called
Cowper-Symonds model), we get
 v   v 0  p / p0 
C
(8-21)
and further
0 for f  σ, R   0
 1
p    f  σ, R   C (8-22)
 p0   for f  σ, R   0
   v 0 
Here  v 0 , C and p0 define the rate sensitivity of the material, which may be determined
from creep tests, relaxation tests or tensile tests at different strain rates. It is seen that the
plastic strain rate increases with the value of the yield function when it is greater than zero.
The constant  v 0 is the value of the viscous stress when the plastic strain rate p equals the
reference strain rate p0 . The behaviour is illustrated in Figure 8-2.

0.001/s 1/s 1000/s

250

200
Stress (MPa)

150

100

50

0
0 0,01 0,02 0,03 0,04 0,05
Plastic strain

Figure 8-2: Work-hardening curves at plastic strain rates 0.001, 1 and 1000 s 1 calculated with
the additive constitutive relation, Equations (8-19) and (8-21), and the Voce work-hardening
rule, Equation (7-84). The material parameters were taken as  0  100 MPa , QR  100 MPa ,
CR  50 ,  v 0  100 MPa , C  0.1 and p0  106 s1 .

8-5
8.3.2 Multiplicative constitutive relation
The multiplicative constitutive relation is here defined as

 eq  σ    0  R  p   1  p  , p  p / p0
C
(8-23)

where p0 and C define the strain-rate dependence of the material. These parameters may be
determined from creep tests, relaxation tests and/or tension tests at a range of different strain
rate. Also in this case, we can solve for the plastic strain rate and we get
0 for f  σ, R   0

  1

p      eq  σ   C  (8-24)
 p0     R p   1 for f  σ, R   0
   0   
 
We understand that the plastic strain rate increases as the stress moves outside the yield
surface, because this means that  eq becomes greater than  0  R which represent the
strength of the material at quasi-static strain rate ( p  0 ). With respect to Equation (8-24),
we observe that
 eq  σ 
f  σ, R   0   eq  σ    0  R  p   1 (8-25)
0  R  p
where  0  0 and R  0 are assumed.
0,001/s 1/s 1000/s
250

200
Stress (MPa)

150

100

50

0
0 0,01 0,02 0,03 0,04 0,05
Plastic strain

Figure 8-3: Work-hardening curves at plastic strain rates 0.001, 1 and 1000 s 1 calculated with
the multiplicative constitutive relation, Equation (8-23), and the Voce work-hardening rule,
Equation (7-84). The material parameters were taken as  0  100 MPa , QR  100 MPa ,
CR  50 , C  0.01 and p0  0.001s1 .

8-6
 The stress-strain behaviour obtained with the multiplicative constitutive relation is shown
in Figure 8-3. It is noted that the distance between the curves increases with plastic straining
for the multiplicative constitutive relation, while it is constant for the additive constitutive
relation in Figure 8-3.

Remark 8-1:
It is possible to rewrite Equation (8-23) as


 eq  σ    0  R  p    0  R  p   1  p   1
C
 (8-26)

We then identify the viscous stress as


 v  p, p    0  R  p   1  p   1
C
 (8-27)

where it is seen that  v is positive in the plastic domain. Hence, the constitutive relation takes
an alternative “additive form”, namely
 eq  σ    0  R  p    v  p, p  or f  σ, R    v  p, p  (8-28)
The difference between this relation and the one in Equation (8-19) is that now the viscous
stress depends on the work-hardening of the material in addition to the plastic strain rate.
■

The choice of constitutive relation depends on the material at hand, and the examples
given above are but two possible formulations. There are several other relations proposed in
the literature [3]. It is finally noted that the material parameters entering into the constitutive
relation are in general temperature dependent.

Remark 8-2:
The rate dependence of metals is strongly influenced by the temperature. The general
observation is that the rate sensitivity is higher at elevated temperatures than at room
temperature. It is therefore often appropriate to include temperature dependence in the theory
of viscoplasticity.
A rather general thermoelastic-thermoviscoplastic constitutive model is obtained by the
following set of constitutive equations:
 ij   ije   ijp
 ij   ij T  T  Cijkl T   kle
f    σ    0  T   R  p, T  (8-29)
g  σ   ij  ijp
 
p
; p
ij
 ij  σ 
0 for f 0

  σ, R, T  for f 0

8-7
where it is recalled that  eq    σ  . The thermoelastic parameters Cijkl and  ij , the yield
stress  0 , the hardening variable R , and the function  defining the rate sensitivity are all
assumed to depend on the temperature T . The exact form of the temperature dependence has
to be determined through mechanical testing at a wide range of strain rates and temperatures.
■

Example 8-1: Creep of thin-walled metal tube.

Figure 8-4: Metal tube subjected to tension force and torque.

Figure 8-4 shows a thin-walled, cylindrical metal tube. The tube is made of an isotropic
material and obeys the rate-sensitive J2 flow theory with associated flow and isotropic
hardening. The constitutive relation is given as:
C
 p
 eq  A  p0  p   
n

 p0 
The material constants are defined by: A  600 MPa , p0  0.002 , n  0.2 , C  0.01 and
p0  103 s1 . As illustrated in Figure 8-4, the tube is subjected to constant axial force and
torque giving the stresses 11  240 MPa and 12  100 MPa . All other stress components
are zero. The task is to determine the equivalent plastic strain p as a function of time t , and
to find the components of the plastic strain tensor after 100 hours creep loading of the tube.
According to the given constitutive relation, the equivalent plastic strain rate may be
expressed as
1
  eq C
p  p0  n 
 A  p0  p  
Integration of the differential equation gives
1
  eq  C
p n t

 p
0
0  p C dp   p0 
0  A 
 dt

8-8
The equivalent plastic strain is then expressed as a function of time
1

 n    eq  C
n
n
1
p t  
1
C
  1   0 p t  p0
C
 p0
 C  A 

where  eq is constant for the prescribed loading situation and equal to

 eq   112  32 122   212   2402  32 1002  1002   296 MPa

After 100 hours loading, the equivalent plastic strain equals

1
 0.2  296  0.01
0.2
0.2 1

p 100 h    10   3600 100   0.002

1
3
 1  0.002  0.051
0.01 0.01


 0.01  600 
The deviatoric stress components are
11  23  240  160 MPa,  22
   33
   13  240  80 MPa, 12   21
  100 MPa
Since the loading is proportional, the associated flow rule gives the plastic strain
components as functions of time as
3 11 3 160
11p  t   p t   p t   0.811 p t 
2 eq 2  296

3 22 3  (80)
 22p  t   p t   p t   0.405 p t 
2 eq 2  296
3 12 3 100
12p  t   p t   p t   0.507 p t 
2 eq 2  296

where p    dt was used. Inserting t  100 h into this set of equations, we get the final
result:
11p 100 h   0.041,  22p 100 h    33p 100 h   0.021, 12p 100 h    21p 100 h   0.026
■

Example 8-2: Viscoelastic-viscoplastic constitutive model for polymers.

Polymers often exhibit rate dependence both in the elastic and plastic domains. To
capture this behaviour, we have to combine viscoelasticity and viscoplasticity into the same
constitutive model. Consider an isotropic polymeric material and establish a viscoelastic-
viscoplastic model based on the rheological representation in Figure 8-5.
In the rheological model, E0 and E1 are the elastic moduli of the two linear elastic
springs, 1 is the viscosity of the linear dashpot,  Y is the flow stress of the friction element,
and 2 ( p) is the viscosity of the non-linear dashpot.

8-9
Figure 8-5: Rheological representation of a viscoelastic-viscoplastic constitutive model.

For the viscoelastic part it may be assumed that only the deviatoric part of the response is
rate dependent, whereas the J2 flow theory, the associated flow rule, isotropic hardening, and
the multiplicative constitutive relation from Section 8.3.2 should be utilized for the
viscoplastic part. Finally, derive the rate equation for the deviatoric response of the
viscoelastic-viscoplastic polymer.
From the rheological model in Figure 8-5 it is clear that the strain decomposes into three
parts: elastic, inelastic and plastic, and thus
 ij   ije   iji   ijp

where  iji and  ijp are deviatoric tensors per assumption. The volumetric strain  v is elastic
and defined by Equation (6-19) as
P
v  
K
where P is the pressure and K the bulk modulus. The deviatoric strain  ij then has three
parts as follows:
3 
 ije 
1
2G0
 ij ,  iji 
1
21
 ij  2G1 iji  ,  ijp  p ij
2 eq
where Equations (6-19) and (6-20) were used to obtain the first two relations, whereas the
latter follows from Equations (8-17) and (8-8). The plastic parameter is here defined by
Equation (8-24) as
1/ C
 32  ij ij 
p  p0   1
 0  R  p 
 

where the Macaulay bracket, a  max  a,0  for any scalar a , has been employed. The rate
equation for the deviatoric response of the viscoelastic-viscoplastic polymeric material is
subsequently obtained as

8-10
   32  ij ij 
1/ C

3 ij 
 ij  2G0  ij 

1
21
 ij  2G1 iji   p0 
 0  R  p 
 1
2 eq 
   
This differential tensor equation may be integrated in time with an appropriate numerical
scheme given the initial values of the state variables.
■

8.4 Viscoplasticity without yield surface

Consider again the constitutive relation in Example 8-1. It is seen that if p  0 also the
equivalent stress  eq tends to zero. Or put differently, plastic straining occurs for all non-zero
equivalent stresses. This is an example of a model of viscoplasticity without an explicit yield
surface.
A rather general form of the constitutive equations for this class of elastic-viscoplastic
models are summarized as
 ij   ije   ijp
 ij  Cijkl  kle
g  σ   ij ijp (8-30)
 
p
; p
ij
 ij  σ 
    σ, R 

To complete the model, we need to select the functions  eq    σ  , g  σ  and   σ, R  . If we

take g   , the model is associated, in the sense that the function defining the equivalent
stress is also used as a plastic potential. In this case, Equation (8-30) is simplified to read
  σ 
 ij   ije   ijp ,  ij  Cijkl kle ,  ijp  p , p    σ, R  (8-31)
 ij
A typical example of a model of viscoplasticity without a yield surface is the Norton
creep law [3]. The scalar function   σ, R  is then defined as
1
  σ  C
  σ, R   p0   (8-32)
 0  R 
where the constants p0 and C define the strain-rate dependence of the material. Since C is
rather small for most materials, there will be no significant plastic flow as long as
 eq   0  R , even if the model has no explicit yield criterion.
Models of viscoplasticity without a yield surface are often used in the rate-dependent
theory of crystal plasticity.

8-11
8.5 Numerical integration*
The stress update algorithms for the elastic-viscoplastic materials are developed in the
following. Two types of algorithms are considered here, namely return mapping algorithms
(semi-implicit and implicit versions) and the explicit rate tangent modulus method. We
develop the first for viscoplasticity with a yield surface and the latter for viscoplasticity
without a yield surface.

Return mapping algorithms

The constitutive equations of the elastic-viscoplastic material with a yield surface are
summarized as
 ij  Cijkl   kl   klp 

 ijp   gij  σ, ξ  ,    h  σ, ξ  (8-33)

 0 for f  σ, ξ   0

  σ, ξ  for f  σ, ξ   0

where gij  g /  ij . The vector of internal variables  (   1, 2,..., N ) includes isotropic

and kinematic hardening variables and, if necessary, other internal variables. The number of
internal variables is denoted N , where each component of the tensor variables are counted as
one internal variable.
As before, we assume that at time tn the stress, strain and internal variables are given by
σ n , ε n and ξ n , and subsequently an incremental change ε n1 in the strain tensor occurs
during the time increment tn1  tn1  tn . Our job is to determine the stress and the internal
variables at tn 1 so that the constitutive relations are fulfilled.
As for the elastic-plastic materials, we calculate an elastic trial state defined by
 tr
  ij , n  Cijkl  kl , n1 and tr, n1   , n , which is the correct state at tn 1 provided
ij , n 1

f ntr1  f  σtrn1 , ξtrn1   0 (8-34)

On the contrary, if f ntr1  0 , plastic flow occurs, and we express the return map (i.e., the
generalized mid-point rule [4]) due to plasticity as
 ij , n 1   ijtr, n 1  n 1Cijkl g kl ,n 
 ,n 1   ,n  n 1h ,n  (8-35)

n 1  tn 1n 1

where n 1 is the “incremental” plastic parameter. We have introduced the convenient
notation gkl ,n  gkl  σ n , ξ n  , h ,n  h  σ n , ξ n  and n1    σ n1 , ξ n1  , and the state
variables σ n  and ξ n  at time tn   tn1  1    tn are defined as

8-12
 σ n    σ n 1  1    σ n
(8-36)
ξ n    ξ n 1  1    ξ n

If   0 , the algorithm is semi-implicit [5]. The state variables σ n1 and ξ n1 depend only
on n 1 and the known quantities at tn . The result is a nonlinear equation in n 1 defined by

n1  tn1  ijtr, n1  n1Cijkl gkl ,n ,  , n  n1h ,n   0 (8-37)

This equation is readily solved by some iterative technique, e.g. Newton’s method or the
secant method. This is an efficient algorithm, but small time steps should be taken to maintain
stability and sufficient accuracy.
For other values of the parameter  , typically   1 (i.e., the backward-Euler method),
we have to solve the system of nonlinear equations given by Equation (8-35). The residual
form of the return map reads as
 ij , n 1   ijtr, n 1  n 1Cijkl g kl ,n   0

 , n 1   ,n  n 1h ,n   0 (8-38)

n 1  tn 1n 1  0

This set of equations is solved by Newton’s method as described for the rate-independent
material in Chapter 7.8.
Again, the reader is referred to Simo and Hughes [7] for more details on return mapping
algorithms in computational inelasticity.
It should be noted that for elastic-viscoplastic models there exist several other integration
algorithms that could be used instead of the return mapping algorithm presented above, e.g.
the Runge-Kutta methods and the rate tangent modulus method. We will briefly describe the
latter method in the following, owing to its simple implementation.

Rate tangent modulus method

We now consider elastic-viscoplasticity without a yield surface. The constitutive
equations are then compiled as
 ij  Cijkl   kl   klp 

 ijp   gij  σ, ξ  ,    h  σ, ξ  (8-39)

    σ, ξ 
where all variables have the same meaning as in Equation (8-33). The rate tangent modulus
method is a one-step, explicit solution method [5][6]. The algorithmic set-up reads as

8-13
  ij , n 1   ije, n 1   ijp, n 1

 ij , n 1  Cijkl   kl , n 1   klp , n 1 

 ijp, n 1   ijp, n  n 1 gij ,n (8-40)

 , n 1   , n  n 1h ,n

n 1  tn 1  1    n  n 1 

where  is a parameter between 0 and 1, and we have introduced the short forms
gij ,n  gij  σ n , ξ n  , h ,n  h  σ n , ξ n  and n    σ n , ξ n  . It is seen that the set-up is explicit in
the plastic flow and hardening, whereas a generalized trapezoidal rule is used to determine
n 1 . Linearization of n1    σ n1 , ξ n1  around the state  σ n , ξ n  at tn gives

 
n1  n   ij , n 1   , n 1 (8-41)
 ij n  n

where the summation convention is applied for sub-indices i, j, k , l  1, 2,3 and   1, 2,..., N .
Combination of Equation (8-40)5 with Equation (8-41) gives
   
n 1  tn 1n  tn 1   ij , n 1   , n 1  (8-42)
  ij n  
 n 
The increments in stress and internal variables are found from Equation (8-40) as
 ij , n 1   ij , n 1   ij , n  Cijkl   kl , n1  n1 g kl ,n 
(8-43)
 , n 1   , n 1   , n  n 1h ,n
Putting everything together, we get


n   Cijkl  kl , n 1
 ij n
n 1  (8-44)
1    
  Cijkl g kl ,n  h ,n 
tn 1   ij n  
 n 
Using Equation (8-44) together with Equations (8-40)1-4, the state variables at tn 1 are readily
determined.
Since the rate tangent modulus method is explicit and thus only conditionally stable,
small time steps should be used.

8-14
8.6 References
[1] F. Irgens. Continuum Mechanics. Springer, 2008.
[2] J. Lubliner. Plasticity Theory. Macmillan, 1990.
[3] J. Lemaitre, J.-L. Chaboche. Mechanics of Solid Materials. Cambridge University
Press, 1990.
[4] M. Ortiz, E.P. Popov. Accuracy and stability of integration algorithms for elastoplastic
constitutive relations. International Journal for Numerical Methods in Engineering 21
(1561-1576) 1985.
[5] T. Belytschko, W. K. Liu, B. Moran. Nonlinear finite elements for continua and
structures, Wiley, 2000.
[6] D. Peirce, C.F. Shih, A. Needleman. A tangent modulus method for rate dependent
solids. Computers & Structures 18 (1984)
[7] J.C. Simo, T.J.R Hughes. Computational Inelasticity. Springer, 1998.

8-15
8-16