Lecture Notes

Advanced Finite Element Methods

Part II: Material Nonlinearity

Univ. Prof. Dr. techn. G. Meschke

Oktober 2011

Ruhr University Bochum

Institute for Structural Mechanics
Lecture Notes

Advanced Finite Element Methods

Part II: Material Nonlinearity

Univ. Prof. Dr. techn. G. Meschke

Oktober 2011

Ruhr University Bochum

Institute for Structural Mechanics
Universitätsstraße 150 IA 6/126
D-44780 Bochum
Telefon: +49 (0) 234 / 32 29055
Telefax: +49 (0) 234 / 32 14149
www: http://www.sd.ruhr-uni-bochum.de
Contents

1 Introduction 2

2 Elasto-plastic models 3
2.1 Basics and governing equations of plasticity theory . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.1 Governing equations of infinitesimal plasticity theory . . . . . . . . . . . . . . . . . . 3
2.1.2 A 1D-think model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.3 Consistency parameter and elastoplastic tangent . . . . . . . . . . . . . . . . . . . . 7
2.1.4 Flow rule for a non-smooth yield surface . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.5 Decomposition of the stress and strain tensor . . . . . . . . . . . . . . . . . . . . . . 9
2.2 A basic elastoplastic model: von Mises model . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1 Von Mises yield condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.2 Hardening law: isotropic hardening . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3 Damage models 17
3.1 Mechanical Representation of Damage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Isotropic Damage Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 A Model for Concrete under Uniaxial Tension . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2 Introduction

Chapter 1
Introduction

Nonlinear finite element analyses based on realistic material laws are gaining increasing attention
by the engineering community in civil as well as mechanical engineering, considering the steady
increase of computer power accompanied by the decrease of hardware costs.
As far as the realistic mathematical description of inelastic engineering materials is concerned, the
theory of elastoplasticity and damage are used in the majority of applications. The reason for the
attractiveness of these theories is their relative conceptual simplicity and the easy calibration of
model parameter to experimental results, such as yield- or failure strength obtained from standard
material tests.
The scope of the lecture covers the fundamentals of plasticity and damage theory, which can be
easily applied to a large variety of material models. Typical applications include not only (rein-
forced) concrete, soil mechanics and tunnelling but also numerical simulations of forming processes
of materials, such as forging, extrusion, rolling, stretch forming and deep drawing.
The mathematical structure of plasticity and damage theory in the context of the geometrically
linear theory is fairly well understood. The restriction to geometrically linear configuration requires
that displacements, rotations as well as the strains must be small.
It should be noted, that in some engineering applications, not only geotechnical situation but also
in simulations of forming problems of automobile crash tests, the restriction to small deformations
may be prohibitive. However, for the interest of limitation of the overall scope of the lectures, all
formulations described within the present lecture notes are restricted to the geometrically linear
theory.
 Chapter 2
Elasto-plastic models

For the realistic mathematical description of inelastic materials, especially metallic materials, the
theory of elastoplasticity is frequently used.
In general, plasticity theory is applied in cases, when deformations caused by external loads are
not fully recoverable in unloading processes. Such inelastic deformations are always accompanied
by dissipation of energy.
The mathemtical structure of (visco-) plasticity theory in the context of the geometrically linear
theory is fairly well understood. A comprehensive review of plasticity theory formulated in the
context of the geometrically linear theory (often referred to as ”linearized or infinitesimal plasticity
theory”) is contained in the work of Simo and Taylor (1985) [9], [10], Simo, Kennedy and Govindjee
(1988) [7], Pramono and Willam (1989) [4], Lubliner (1990) [3] and Simo und Hughes (1996) [6],
among many others. The state of the art of the linearized theory of plastictiy and its algorithmic
formulation is characterized by

1. a thermodynamical basis,

2. a clear mathemtaical structure,

3. numerical efficiency of the integration algorithms in the context of the incremental-iterative

strategy of the finite element method,

4. quadratic rate of asymptotic convergence of the global iteration procedure based on the
Newton scheme.

2.1 Basics and governing equations of plasticity theory

2.1.1 Governing equations of infinitesimal plasticity theory

This subsection summarizes the governing equations of classical plasticity theory in the context
of the geometrically linear theory. A more elaborated discussion of this topic is contained in [6].
Details on multisurface plasticity are found in [7].
4 2.1: Basics and governing equations of plasticity theory

The mathematical structure of the constitutive equations, its thermodynamical basis and the in-
tegration of the evolution equations are addressed in this Section. The foundations of classical
plasticity is a necessary prerequisite to understand the extensions of the theory to the large strain
regime [6, 5, 8].
Noting that the geometrical linear theory is adopted in this section, no difference between reference
configuration and deformed configuration has to be made. For simplicity, only single surface plas-
ticity, characterized by a single yield surface related to a single hardening mechanism, is considered.
The classical (infinitesimal) theory of plasticity is based on a set of equations, which determine the
elastic material behavior and the evolution of the inelastic deformations:

(1) Additive Split of the strain tensor:

ε = εe + εp . (2.1)

(2) Hyperelastic Constitutive Law:

s L o a d in g
E

U n lo a d in g

e
p e
e e
e

Figure 2.1: Elastoplastic Material Behavior: Elastic and Plastic Portions of Strain

∂W (εe )
σ= , (2.2)
∂εe
where W is the function of free energy. A frequently used example for W is the expression
1 1
W = λ (tr(εe ))2 + 2G tr(εe2 ) , (2.3)
2 2
with λ denoting the Lamé-parameter and G is the Shear modulus. Inserting (2.3) into (2.2) yields
Hooke’s Law:
∂W
σ= = λ tr(εe ) 11 + 2G εe = C : εe . (2.4)
∂εe

(3) Elastic domain in the stress space; Yield condition:

The elastic domain in the stress space is defined by

I = {(σ, ζ) ∈ SS × IR kf (σij , ζ) ≤ 0} ,
E (2.5)
where SS denotes the space of symmetric rank 2 tensors. In (2.5)

f (σ, ζ) ≤ 0 (2.6)

is the so-called yield function and ζ is a stress-like hardening parameter. The hardening parameter ζ
determines the “size” of the yield function in the process of continuing plasticication of the material.
f (σ, ζ) = 0 defines the boundary of the elastic domain E I,

I = {σ : f (σ, ζ) = 0}.
∂E (2.7)

States (σ, ζ) outside E

I are non-admissible.

s 2

¦ = 0
s y

s 1
¦ > 0
¦ < 0

Figure 2.2: Elastic Domain and Yield Function in Stress Space

(4) Flow rule:

In the case of associative plasticity theory, the evolution of the tensor of plastic strains εp is
controlled by the gradient of the yield function with respect to the stress tensor, ∂f (σ, ζ)/ ∂σ, and
by a factor γ̇, denoted as the so-called consistency parameter:

ε̇p = γ̇ ∂f (σ, ζ)/ ∂σ. (2.8)

∂f (σ, ζ)/ ∂σ defines the orientation and γ̇ determines the magnitude of ε̇p .
In the more general case of non-associative plasticity theory, the evolution equations have the form

ε̇p = γ̇ r(σ, ζ) . (2.9)

Frequently r is defined from a potential function g(σ, ζ) 6= f (σ, ζ) as

∂g(σ, ζ)
r= . (2.10)
∂σ

(5) Hardening Rule:

6 2.1: Basics and governing equations of plasticity theory

In the case of associative plasticity, the evolution of the internal plastic variable α is governed by
the gradient of the yield function with respect to its conjugate variable ζ and by the consistency
parameter γ as
∂f (σ, ζ)
α̇ = γ̇ . (2.11)
∂ζ
(2.11) defines the so-called associative hardening rule. In the general case of non-associative plas-
ticity, tha rate of α is defined in the form

α̇ = γ̇ h(σ, ζ), (2.12)

where h(σ, ζ) is a suitable hardening function.

The hardening parameter ζ(α) depends on the plastic internal variable α. The rate of ζ is obtained
as

ζ̇ = −H α̇, (2.13)
dζ
with H = − dα as the so-called hardening modulus.

(6) Loading/Unloading Criteria (Kuhn-Tucker-Condtions):

The criteria for plastic loading and unloading, respectively, can be written in the compact form

γ̇ ≥ 0 , f (σ, ζ) ≤ 0 , γ̇ f (σ, ζ) = 0. (2.14)

Remark: First consider the case in which {σ, ζ} ∈ int(EI σ ), e.g. in which f (σ, ζ) < 0. It then follows
from (2.14)3 that γ̇ = 0. Hence, ε̇p = 0. The process is a purely elastic one.
Now consider the case, that γ > 0. It follows from (2.14)3 that f = 0. This implies, that plastici-
cation of the material in any point of the body requires that σ ∈ ∂E
I.

2.1.2 A 1D-think model

For the purpose of illustrating the basic theory for hardening elasto-plastic materials, a simple
one-dimensional rehelogical model is considered (Figure 2.3). It consists of two linear springs and
a friction element. For this one-dimensional example, the governing equations are summarized as:

Figure 2.3: Rheological Model of hardening elasto-plastic materials

1. Yield function:

f (σ, ζ(α)) = σ − σy + ζ(α) ≤ 0 (2.15)

 2. Evolution equations:
∂f ∂f
ε̇p = γ̇ = γ̇ · 1, α̇ = γ̇ = γ̇ · 1 (2.16)
∂σ ∂ζ
3. Kuhn-Tucker conditions:
f (σ, ζ) ≤ 0, γ̇ ≥ 0, f · γ̇ = 0 (2.17)
For plastic loading state, γ̇ ≥ 0, f = 0; for unloading situation, γ̇ = 0, f < 0

2.1.3 Consistency parameter and elastoplastic tangent

In what follows, the continuum theory of isotropic plasticity is summarized shortly. Basis for the
derivations are the state and evolution equations
ε= εe + εp , σ = C : (ε − εp ), ζ = ζ(α)
∂f ∂f
ε̇p = γ̇ , α̇ = γ̇ ,
∂σ ∂ζ
f (σ, ζ) ≤ 0, γ̇ ≥ 0, γ̇f (σ, ζ) = 0. (2.18)

Assuming loading (i.e. γ > 0), the consistency condition requires, that f = at t and t + dt. Hence,
f˙(σ, ζ) = 0. (2.19)
From (2.19) follow
∂f ∂f
f˙(σ, ζ) = dσ + dζ = 0. (2.20)
∂σ ∂ζ
Inserting the rate form of (2.18)2 into (2.20) and considering the hardening law (2.18)3 yields
∂f ∂f dζ
f˙(σ, ζ) = (C : (dε − dεp )) + dα = 0. (2.21)
∂σ ∂ζ dα
With the hardening modulus defined as
dζ
H=− , (2.22)
dα
and using the evoultion equations (2.18)4 and (2.18)5 , the consistency parameter γ̇ is obtained from
(2.21) as
df
dσ Cdε
γ̇ = ∂f
. (2.23)
∂σ :C : ∂∂f ∂f ∂f
σ + ∂ζ H ∂ζ
Inserting of (2.23) into the rate form of the elastic law (2.18)2 provides the differential relation
between strains and stresses and, therefore, the continuum tangent:
 
df
dσ = C dε − γ̇
dσ
" # !
df
dσ Cdε df
= C dε − ∂f ∂f ∂f ∂f
.
: C : dσ
∂σ ∂ σ + ∂ζ H ∂ζ
T
h i
C ddfσ ⊗ C ddfσ
= Cdε − ∂f ∂f ∂f ∂f
δε. (2.24)
∂ σ : C : ∂ σ + ∂ζ H ∂ζ
8 2.1: Basics and governing equations of plasticity theory

Hence, the elastoplastic tangent Cep is given as

T
h i
C ddfσ ⊗ C ddfσ
Cep = C− ∂f
. (2.25)
∂σ : C : ∂∂f
σ+
∂f ∂f
∂ζ H ∂ζ

2.1.4 Flow rule for a non-smooth yield surface

If a yield surface consists of several partial surfaces and the intersections of these partial surfaces
form edges or corners, then the gradient of the yield surface and, hence, the associated flow rule is
not uniqely defined in such intersection points (see Fig. 2.4). A non-smooth yield surface consists of
m partial surfaces, where each partial surface is described by an equation fk (σ, ζ) = 0, k = 1, ..., m,
and the elastic region of the stress space is defined by conditions fk (σ, ζ) < 0 for all k = 1, ..., m.
σ is located on the yield surface if at least for one partial surface fk = 0 applies and it is located

¶ ¦ i/ ¶ s

¦ i( s , q i) = 0 d e p

Figure 2.4: Example for a non-smooth yield surface

at an edge or in a corner if for at least two partial surfaces fk = 0 applies. The associated flow
rule for those points on a non-smooth yield surface can be obtained after Koiter by means of an
extension of Eq. 2.8 to

m
∂fk (σ, ζ)
ε̇pij
X
= γ̇k . (2.26)
∂σij
k=1

Note that for all fk < 0, from the Kuhn-Tucker conditions (Eq. 2.14), γ̇k = 0 is deduced. At an edge
that is formed by the intersection of twi partial yield surfaces, in general, two consistency parameters
are to be determined. For this, two additional equations are available, since the conditions fk = 0
and f˙k = 0 are to be considered for two partial yield surfaces.
The fulfillment of the postulate of maximum plastic dissipation yields a rate law for the additional
internal variables of the form
m
X ∂fk (σ, ζ)
α̇i = γ̇k , k = 1, ..., m. (2.27)
∂ζi
k=1

Eq. 2.27 states an extension of Eq. 2.11 that is analogous to Koiter’s flow rule.
2.1.5 Decomposition of the stress and strain tensor

It is often advantageous for the formulation of yield functions to split the stress tensor σ into a
hydrostatic and a deviatoric part:
1
σij = I1 δij + sij . (2.28)
3
In Eq. 2.28,

I1 = σkk = σkl δkl (2.29)

denotes the first invariant of the stress tensor, sij denotes the deviatoric stress tensor and δij
is denoted as Kronecker delta. The first term on the right hand side of Eq. 2.28 represents the
hydrostatic (shear-free) stress state; the quantity
1
p = I1 (2.30)
3
is denoted as hydrostatic pressure. Solving Eq. 2.28 for sij yields
1
sij = σij − [σkl δkl ]δij
 3 
1
= δik δjl − [δijδkl σkl
3
dev dev 1
= Iijkl σkl with Iijkl = δik δjl − δij δkl . (2.31)
3
The hydrostatic stress can be rewritten as
1 vol vol 1
I1 δij = Iijkl σkl Iijkl = δij δkl . (2.32)
3 3
Applying the deviatoric operator Idev to σ yields the deviatoric stress tensor s. By the application
of the volumetric operator Ivol to σ the hydrostatic stress tensor 13 I1 1 is obtained. The application
of the deviatoric operator to a deviatoric tensor does not change the deviatoric tensor, e.g.

Idev s = s. (2.33)

Analogous to the decomposition of the stress tensor, also the strain tensor can be split into a
volumetric and a deviatoric part:
1
εij = I¯1 δij + eij . (2.34)
3
In Eq. 2.34

I¯1 = εkl δkl (2.35)

denotes the first invariant of the strain tensor and eij denotes the deviatoric strain tensor. The
forst term on the right hand side of Eq. 2.34 represents the volumetric part of the strais that causes
a volume change but no change in shape. Analogous to Eqns. 2.31 and 2.32 the deviatoric strain
tensor is obtained as
dev 1¯ vol
eij = Iijkl εkl I1 δij = Iijkl εkl . (2.36)
3
10 2.1: Basics and governing equations of plasticity theory

Inserting Eq. 2.34) in the generalised Hookean Law

σij = Cijkl εkl (2.37)

with the elasticity tensor for an isotropic material

Cijkl = λδij δkl + G (δik δjl + δil δjk ) , (2.38)

where λ and G denote the Lamé parameters, yields

σij = K I¯1 δij + 2Geij . (2.39)

The comparison of Eq. 2.39 with Eq. 2.28 yields

I1 = 3K I¯1 , sij = 2Geij . (2.40)

In Eq. 2.39 K = λ + 23 G is denoted as modulus of compression and G as shear modulus. In the case
of isotripic material it is hence possible to completely separate volumetric and deviatoric behaviour.
If we replace λ by K and G in Eq. 2.38, we obtain
 
1 1 vol dev
Cijkl = Kδij δkl + 2G (δik δjl + δil δjk ) − δij δkl = 3KIijkl + 2GIijkl (2.41)
2 3
dev is now defined by
for the elasticity tensor. In contrast to Eq. 2.312 , Iijkl

dev 1 1
Iijkl = (δik δjl + δil δjk ) − δij δkl . (2.42)
2 3
dev according to Eq. 2.42 does not violate Eq. 2.31 due to the symmetry of the
The definition of Iijkl
dev follows from the isotropic tensor C
stress and strain tensors. Yet it has the advantage that Iijkl
(Eq. 2.38) and therefore also states an isotropic tensor. Note that the cartesian components of an
isotropic tensor are not changed during a coordinate transformation.
The principal directions of the deviatoric stress (and strain) tensor s (e) coincide with those of σ
(ε). The principal stress axes are normal to the planes of vanishing shear stresses and on subtraction
of the hydrostatic part of the stresses the shear stresses remain unchanged. This statement can be
verified by means of Mohr’s circle. The subtraction of the hydrostatic part from the stresses only
causes a shift in the stress circles but does not change their radii.
For isotropic materials, the yield function may only depend on the stresses by means of the three
principal stresses σ1 , σ2 , σ3 or by means of the invariants of the stress tensor I1 , I2 , I3 . This is
because on the one hand the directions of principal stresses are not relevant and on the other hand
the exclusive dependency on invariants ensures the independence from coordinate transformations.
In the case of isotropic material behaviour, if the dependency of the yield function on σ is replaced
by a dependency on I1 and s, the yield function has to be representable as a function of I1 and
the second and third invariant of the deviatoric stress tensor, J2 and J3 . This does not affect other
dependencies (e.g. from further internal variables). From the definition of s, J1 = 0 can be deduced.
The decomposition of σ into a hydrostatic and a deviatoric part can be geometrically interpreted
(see Fig. 2.5). A stress point P can be represented in a coordinate system whose axes are the
principal directions of the stress state σ1 , σ2 , σ3 . Two stress stated that are only different by the
directions of their principal stresses but not by different values of their principal stresses will be
represented by the same point in the stress space. This geometrical interpretation is therefore only
of concern in the case of isotropic materials. The hydrostatic axis in the stress space is defined
 s d e v
1
s 1

h y d r o s t a t is c h e
s A c h se
J

Ö 2 J 2 = ||s || d e v ia t o r is c h e
E b e n e

0 s
I1 / Ö 3 3

s 2

Figure 2.5: Decomposition of the stress vector in principal stress space

by σ1 = σ2 = σ3 . Stress points that are located on this axis are free of shear stresses. The planes
that are normal to the hydrostatic axis are denoted as deviatoric planes. The deviatoric plane that
contains the origin 0 (σ1 + σ2 + σ3 = 0) is called π-plane. Stress points that are located in the
π plane only have a deviatoric part but no hydrostatic part. The length of the component of the
stress vector depicted in Fig. 2.5 that points into the hydrostatic√direction σ = b σ1 σ2 σ3 c yields,
in combination with the unit vector of the same direction e = 1/ 3b 1 1 1 c the following relation:

1
σ · e = √ I1 . (2.43)
3
Considering Eq. 2.30, the vector of the hydrostatic part of the stress tensor can be written as
   
 1   1 
1 I1    
√ I1 e =  1  = −p  1  . (2.44)
   
3 3    
   
1 1

The deviatoric principal stress vector b s1 s2 s3 c with s1 , s2 , s3 as deviatoric principal stresses

becomes
     
 s1   σ1   1 
    I  
1 
 s2  =  σ2  −  1  . (2.45)
    
    3  
     
s3 σ3 1

Geometrically, s is obtained by projection of the stress point P onto the π plane. This projection
is denoted as P 0 . The length of the component of σ, that matches the distance between the stress
point P or the projection P 0 and the hydrostatic axis can be obtained as
q
s21 + s22 + s23 . (2.46)
12 2.2: A basic elastoplastic model: von Mises model

s 3 s 3 s 3

s 1 = s 2 = s 3

s 2
s 2

s 1
s 2

s 1 s 1
(a ) (b ) (c )

Figure 2.6: Examples for yield surfaces in the principal stress space: (a) f = f (I1 , J2 ), (b) f = f (J2 ),
(c) Intersection of a yield surface f = f (I1 , J2 , J3 ) with a deviatoric plane

This length equals the norm of the deviatoric stress tensor in any coordinate system; in particular
also in those that are not transformed to principal stress directions.
√
ksk = sij sij (2.47)

and because of
1 1
(σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ3 − σ1 )2

J2 = sij sij = (2.48)
2 6
√
proportional to J2 .
It can be summarised that I1 determines which deviatoric plane a stress point is located in and
√ the distance between the stress point and the hydrostatic axis equals ksk and is proportional
that
to J2 . If a yield surface depends on I1 and J2 but not on J3 an intersection of the yield surface
with an arbitrary deviatoric plane yields a circle. For a specific deviatoric plane J2 = const applies
(see Fig. 2.6(a)). If the yield surface only depends on J2 , the yield surface has generatrices that are
parallel to the hydrostatic axis. In this case, J2 = const applies to all points on the yield surface
(see Fig. 2.6(b)). If the yield function also depends on the third invariant of the stress deviator

J3 = det(sij ) = s1 s2 s3 , (2.49)

with sA = σA − 13 tr(σ), A = 1, 2, 3, intersections of the yield surface with a deviatoric plane yield
a non-circular, closed curve that is symmetric with respect to the projection of the coordinate axes
on this plane (see Fig. 2.6(c)).

2.2 A basic elastoplastic model: von Mises model

In what follows, one basic model for ductile materials: von Mises-model, also denoted as J2 -theory,
is described.
Using the volumetric-deviatoric split of the stress tensor, the yield function f (σ, ξ) can be formu-
lated as f (I1 , s, ξ). Hence, the associated flow rule (2.8) is obtained as
 
p ∂f ∂f ∂I1 ∂f ∂skl
ε̇ij = γ̇ = γ̇ + . (2.50)
∂σij ∂I1 ∂σij ∂skl ∂σij
Considering that
1
I1 = σkk , sij = [δik δjl − δij δkl ]σij , (2.51)
3
substituting
∂I1 ∂skl 1 dev
= δij , = δik δjl − δij δkl = Iijkl (2.52)
∂σij ∂σij 3

into (2.50) yield the rate of the plastic strains

1
ε̇pij = I¯˙1p δij + ėpij (2.53)
3
with
∂f ∂f dev
I¯˙1p = 3γ̇ ėpij = γ̇ I (2.54)
∂I1 ∂skl ijkl
The first term o the right hand side of (2.53) corresponds to the volumetric part of the plastic
strains, associated only with the inelastic volumetric deformation of the material. The second part
corresponds to the deviatoric part of the plastic strains, associated to the change of the shape of
the material.

2.2.1 Von Mises yield condition

Originally, plasticity theory has been developed for the description of the ductile, elastoplastic
behavior of metals. Experimental evidence shows, that for metals the hydrostatic pressure has no
influence on the state of yielding. Furthermore, no volumetric inelastic deformations are observed.
Hence, metals behave plastically incompressible.
In analogy to the split of the plastic strain tensor according to (2.53), also the strain energy density
can be decomposed into a volumetric part and a deviatoric part associated with the energy required
to change the shape of the material. The von Mises yield surface is based on the hypothesis, that
yielding occurs, when the specific deviatoric part of the strain energy for any three-dimensional
state of stress
G  G
Wdev = (σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ3 − σ1 )2 = J2 (2.55)
12 2
is equal to the strain energy invested in a uniaxial test at the onset of yielding, i.e. when the stress
σ1 reaches the yield stress σy :

G
2(σy )2

Wdev = (2.56)
12
In (2.55), the definition of the second invariant of the deviatoric stress tensor according to (2.48)
has been used:
1 1
(σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ3 − σ1 )2 .

J2 = sij sij = (2.57)
2 6
Setting (2.55) equal to (2.56) yields

3J2 − σy2 = 0 (2.58)

14 2.2: A basic elastoplastic model: von Mises model

(2.58) defines the yield locus in the stress space

1
f (σ) = f (J2 ) = J2 − σF2 , (2.59)
3
with σy as the yield stress determined from a uniaxial tenisle test. Alternatively, the yield function
f (σ) can be expressed as
p σy
f (J2 ) = J2 − √ . (2.60)
3
The advantage of this formulation is, that the function value f has the dimension of stress. Hence,
the gradient of f appearing in the associative flow rule (2.8) and, consequently, the consisteny
parameter γ̇ is dimensionless. Considering (2.47) and (2.48) one obtaines from (2.60)
r
2
f (s) = ksk − σy = 0. (2.61)
3
This formulation of the Mises-yield function allows a geometrical interpretation. For f = 0 the
norm s corresponds to the distance of a point σ located
p on the yield surface from the hydrostatic
axis. This distance is constant and has the value 2/3σy . Hence, the Mises-yield surface p de-
scribes a cylinder centered around the hydrostatic axis in the stress space with the radius 2/3σy .
(Figure 2.6(b)).
Considering now a stress state leading to yielding, characterized by pure shear (s12 = s21 = τy ,
all other components of s are zero) in (2.60), one obtains the critical shear stress at which the
materials start to yield as
√
σy = 3τy . (2.62)

Inserting into (2.60) yields the alternative formulation of the yield function
p
f (J2 ) = J2 − τy = 0. (2.63)

Since, in general, the yield strength σy is known from uniaxial tests, the format according to
equations (2.60) or (2.61) are preferred compared to (2.63).
The rate of the plastic strains are otained from inserting of (2.61) into (2.50) considering (2.33) as
sij s
ε̇pij = ėpij = γ̇ √ or ε̇p = ėp = γ̇ . (2.64)
skl skl ksk

Since the yield function does not depend on I1 , the plastic strain rate is purely deviatoric. This
assumption of plastic incompressibility is in accordance with experimental results from metallic
materials.

2.2.2 Hardening law: isotropic hardening

The yield functions that have been covered in the previous section are limited to ideally plastic
material behaviour. Since, in this case, f only depends on the stress state, the yield surface is
immutable in the stress space. In the following it will be shown how hardening laws influence the
formulation of yield surfaces. Isotropic hardening is characterised by an expanding yield surface in
the stress space. The orientation of the yield surface, however, is not changed (Fig. 2.7). Now, the
yield stress σy (or τy ) is not a constant quantity anymore but a function of an internal variable
that is usually denoted as α.
 s 2 s
¦ (s 3, k 3) = 0
s F(k 2)
¦ (s 2, k 2) = 0 2
0
s
¦ (s 1, k 1) = 0 F

2 s F(k 2)

s 1
e

- s 0 2 s F(k 2)
F s F(k 2)
2
3

Figure 2.7: Isotropic hardining: (a) for biaxial stress states, (b) for uniaxial stress states

Considering isotropic hardening, the von Mises yield function for ideally plastic material behaviour
(Eq. 2.61) becomes
r
2
f (s, ζ(α)) = ksk − (σy − ζ(α)) (2.65)
3
with
r
∂f 2
α̇ = γ̇ = γ̇ . (2.66)
∂ζ 3

From (2.50) and (2.64), the rate of plastic strains is obtained as:
s
ε̇p = γ̇ = γ̇n. (2.67)
ksk

Taking the norm of ε̇p

√
ε̇p : ε̇p = γ̇ (2.68)

and inserting into (2.66) yields

r
2 p p
α̇ = ε̇ ε̇ (2.69)
3 ij ij

Inserting Eq. 2.64 in Eq. 2.66 yields

r
2
α̇ = γ̇ . (2.70)
3

A very general approach to ζ(α) is

ζ(α) = −Hlin α − ∆σy∞ (1 − e−δα ). (2.71)

16 2.2: A basic elastoplastic model: von Mises model

- z

D s y e x p ( -d a )
D s y
H lin
a

Figure 2.8: Definition of an isotropic hardening law

In Eq. 2.71, H denotes the linear isotropic hardening modulus. Furthermore this approach requires
two additional material parameters: ∆σy∞ and δ. The function ζ(α) is visualised in Fig. 2.8. The
respective rate law and the isotropic hardening modulus H is obtained as

dζ(α)
ζ̇(α) = Hi α̇, Hi = − . (2.72)
dα

The consideration of isotropic hardening is rather simple yet, in general, dies not allow for a realistic
description of metallic materials. This can be easily seen from the uniaxial case. If a yield stress
σF (ᾱ) > σF0 is reached by isotropic hardening under uniaxial tension, yielding will not occur in
a subsequent compressive loading until the same absolute value of σF (ᾱ) has been reached (see
Fig. 2.7). This means, that hardening in the tensional region automatically causes hardening in
the compressive region and vice versa. This, however, is in contrast to the real material behaviour
that is characterised by an approximately constant distance of the tensile and the compressive yield
stress of approx. 2σF0 (Bauschinger effect). Since in isotropic hardening the yield surface expands
but its location remains unchanged, an isotropic yield surface remains isotropic even after plastic
strains have occurred.
 Chapter 3
Damage models

All real materials deform when loaded. The deformation may be elastic or inelastic. It may be time
dependent or independent. Occasionally rupture occurs, being either ductile or brittle.
The deformation characteristics of the material are determined by its internal structure and also
by the way of loading. Crystalline and amorphous materials have different properties as have single
crystals versus polycrystalline aggregates. Their behaviours are quite different at high and low
loading rates. Complete understanding of the type of deformation of a specific material would
require detailed knowledge of the atomic structure of that material in addition to a huge amount
of computational effort.
The deformation properties of a certain kind of material can be described by constitutive equations,
either derived from micromechanical or statistical considerations, or, more often, postulated to fit
experimental measurements.
In general, constitutive equations treat the material as a continuum, i.e. a material without atomic
structure. The deformation is described by a field variable, the strain. The distribution of internal
forces in the material is described by another field variable, the stress. These concepts have been
found extremely useful in analysing the behaviour of load carrying structures in spite of the fact
that they do not take into account the discrete structure in real materials.
Under certain kind of loading, the material’s internal structure may change. Tiny cracks may form
between crystal grains or across them. Voids and other forms of small cavities may appear in highly
stressed parts. Such deterioration weakens the material, lowers its load carrying capacity. These
defects are discrete entities. An accurate analysis of their influence would require to consider them
as discrete disturbances of the material continuum, a prohibitive task.
A pioneering paper [1] proposed to describe the collective effect of such deterioration by a field
variable, the continuity. Hence an inherently discrete process was modelled by a continuous vari-
able. The loss in accuracy of modelling the deterioration was then compensated by computational
simplicity.
A similar idea lies behind the concepts of stress within a material continuum. Only when the
discrete nature of the material was disregarded was it possible to develop what became the theory
of elasticity. The success of that theory and its later development into general continuum mechanics
guarantees a further pursuit of Kachanov’s basic concept.
18 Damage models

The state of the material due to deterioration was characterised by a dimensionless, scalar field
variable ψ representing the continuity. To an original material free of defect was ascribed the
condition ψ = 1, whereas ψ = 0 was defined to characterise a completely destroyed material with
no remaining load carrying capacity at all. Therefore, the complementary quantity d ≡ 1 − ψ is
obviously a measure of the extent of deterioration or damage. For the initial material d = 0, while
d = 1 corresponds to a state of complete loss of integrity of the internal structure. The parameter
d as a field variable describing the degree of material damage has recently come into widespread
use and will be used in the sequel of the lecture notes.
Unlike Kachanov(1958) assumed ψ to be a scalar field variable, later researches have led to the
study of tensorial quantities to describe damage.
Although the Kachanov model was entirely phenomenological, later micromechanical studies of
deformation in metals at elevated temperatures have lent support to this model. Such results
create increasing interest in damage analyses based on mechanical principles. The term Continuum
Damage Mechanics (CDM) has come into the sight. The aim of CDM is to develop methods for the
prediction of the load carrying capacity of structures subjected to material damage evolution. It is a
contrast to Fracture Mechanics (FM), which deals with structures containing one or several cracks of
finite size. In FM the cracks are usually assumed to be embedded in a non-deteriorating material.
Studies have also been made where a crack of finite size is embedded in a material undergoing
damage growth. The methods of FM and CDM may then be combined to predict the resulting
decrease of load carrying capacity.
Continuum Damage Mechanics describes, on a macroscopic level, the degradation of materials
caused by the formation and growth of microcracks, microvoids or other similar damaging mecha-
nisms acting on the micro-, meso- and, eventually, in case if cracks are fully open, on the macro-level.
According to this class of material models, degradation is manifested on the macro-level by the
reduction of the macroscopic material stiffness without giving rise to permanent deformations.
The microstructural damage processes are, similar to the way in plasticity theory, described by
means of internal strain like variables. Furthermore, in analogy to plasticity theory, the following
ingredients are required to set up a continuum damage model:

• a loading function specifying the elastic domain characterised by the absence of damage
growth,

• an evolution law for the internal damage variables

• and a relation between the damage variable and the degradation of the apparent stiffness.

Depending on the material, different mechanisms of damage have to be distinguished (see Fig-
ures 3.1 and 3.2):

• Brittle damage manifested by the opening of cracks at the mesoscale without significant
plastic deformations. This type of damage is typically observed in quasi-brittle materials
such as ceramics, mortar, concrete and bricks.

• Ductile damage typically observed in metallic materials accompanied by considerable plastic

strains. It is caused by the growth and coalescence of microvoids.

• Creep damage related to the time-dependent growth of damage in the presence of viscous de-
formations. Various sources of creep damage exist in different materials. For example, creep in
cementitious materials is associated with the presence of moisture within the nanopores of the
 materials, while in metals it results from intergranular decohesions at elevated temperatures
[2].

• Low cycle fatigue damage in metals is caused by the nucleation and growth of microvoids and
microcracks in the presence of large plastic strains during cyclic loading.

• High cycle fatigue damage in metals is caused by the propagation of microcracks during cyclic
loading at relatively small stress levels without significant plastic strains.

a ) b ) c )

Figure 3.1: Damage in cementitious materials: a) diffuse damage resulting from increased porosity
due to calcium dissolution, b) shrinkage cracks, c) ASR-induced cracks in concrete

a ) b )

Figure 3.2: Damage in metallic materials: a) ductile damage in highly plasticised steel, b) low cycle
fatigue damage

In what follows, the fundamental idea of continuum damage theory together with the concept of
effective stress is first explained in a general context (section 3.1). Subsequently, a simple isotropic
model suitable for the modelling of concrete cracking is described in section 3.2.

3.1 Mechanical Representation of Damage

Considering a block named “representative volume element (RVE)” at a point x inside a damaged
material (Figure 3.3), a differential area (“unit” area) of the cross section is dax , having the normal
direction defined by a vector n. The destroyed proportion of the area, which cannot transfer any
stress, is denoted by dax,d . In general, the growth of microcracks is an anisotropic phenomenon,
20 3.2: Isotropic Damage Model

n
d a x ,d s d a x ,d
n
x d a x

0
d a x

a ) b )

Figure 3.3: a) Mesoscopic representation of damage; b) effective stress σ̄

which implies, that dax,d varies with the direction n. Continuum Damage Mechanics is based upon
the definition of a continuous field variable d, defined as
dax,d
d(x, n) = . (3.1)
dax
representing the ratio of damage. Obviously, if d = 0, the material is original; if d = 1, it is
completely damaged. Adopting the assumption of isotropy also for damaged materials, the damage
parameter d in (Equation 3.1) is independent of the spatial orientation n:
dax,d
d(x) = . (3.2)
dax
Assuming an ideal distribution of microdamage within the RVE, the area fraction (3.2) is identical
to the respective volume fraction. One can also write
dvx,d
d(x) = . (3.3)
dvx
The force f acting on the differential area dax is given as

f = tdax = σ · ndax , (3.4)

with σ the effective stress acting on the area element dax . The restriction of the stress transfer
within the preserved part of the area yields

f = σ̄ · n(dax − dax,d ), (3.5)

with σ̄ denoting the nominal stress. Setting (3.5) equal to (3.4) and taking into account (3.2) gives
the definition of the effective stress of damaged materials subjected to tension:
dax 1
σ̄ = σ =σ . (3.6)
dax − dax,d 1−d

3.2 Isotropic Damage Model

The simplest damage model is the isotropic model using one single damage parameter d, which
is based on the assumption that the Poisson’s ratio is not affected by damage. Similar to the
 Plasticity Damage
σ σ

σy σy

E E E E0

εp εe ε ε
ε ε

Figure 3.4: 1D illustration of comparison of Plasticity and Damage theories

elastoplasticity theory, the damage model is based on the definition of free energy in the following
form:
1
Ψ(ε, d) = (1 − d) ε : C0 : ε, (3.7)
2
with C0 denoting the elastic stiffness of the original material. By differentiation of the free energy
with respect to the strain, one reachs
∂Ψ
σ= = C : ε = (1 − d)C0 : ε. (3.8)
∂ε
The equation above gives this fundamental information: the damage variable d reduces the elastic
stiffness of the material, providing the secant stiffness tensor:

C0 = (1 − d)C0 . (3.9)

The Figure 3.4 may offer a direct impression of the difference between plasticity and damage
theories.
Analogous to the plasticity theory, a loading function f is defined, which specifies the elastic domain
and the states, in which damage grows. In damage mechanics, this loading function is, in general,
defined in the strain space. It depends on the state of strain and on an internal strain-like variable
which describes the evolution of the elastic domain as damage grows:

f (ε, κ) ≤ 0. (3.10)

The Kuhn-Tucker-form of the loading/unloading conditions reads

f (ε, κ) ≤ 0, κ̇ ≥ 0, κ̇f (ε, κ) = 0. (3.11)

The internal variable κ governs the damage parameter d:

d = d(κ). (3.12)

The loading function has, in general, a simple form:

f (ε, κ) = ε̃κ ≤ 0. (3.13)

22 3.3: A Model for Concrete under Uniaxial Tension

with ε̃ the so-called equivalent strain. This form of loading function implies a simple evolution law
for the internal parameter κ:

 κold IF ε̃ ≤ κold , i.e. unloading,

κ= (3.14)

 ε̃ IF ε̃ > κold , i.e. loading.

The expression above can be simply interpreted as: the internal parameter κ equals to the maximum
equivalent strain ε̃ in the loading history.
The allowable equivalent strain defines the shape of the elastic domain. Hence, it has to be chosen
and defined such that the characteristics of the specific material is accounted for. Various suggestions
for ε̃ exist. The simplest choice is to take the Euclidean norm of the strain tensor
√
ε̃ = ||ε|| = ε : ε. (3.15)

As is illustrated in Figure 3.5a, the formula (3.15) does not distinguish tensile from compressive
strains. However, for some materials, microcracks only grow under tension, therefore, only the
positive part of the strain tensor should be considered, which leads to the modified definition
p
ε̃ = ||hεi|| = hεi : hεi, (3.16)

see Figure 3.5b. In (3.16), the McAuley brackets h•i means “the non-negative of”, e.g. hxi =
(x + |x|)/2.
As a matter of fact, for cementitious materials, the Rankine (maximum stress)-failure criterion
has been proven to be appropriate one. The relevant definition of the equivalent strain has the form

1
ε̃ = max(σi ), i = 1, 2, 3. (3.17)
E
with σi the principal stresses. The representation of this definition is shown in Figure 3.5c.
The difference between the shapes of strain surfaces and stress surfaces is the effect of Poisson’s
ratio.
A great advantage of damage models is their explicit nature. In other words, the models have a
“strain-driven” scheme, and the stress evaluation algorithm is usually explicit, without the need
of an iterative solution for a set of nonlinear equations, such as the Return Map Algorithm in
elastoplastic models. Once a new strain, computed from the given displacement field determined
by means of the stepwise solution of the discrete equilibrium equations at the structural level, is
obtained at one integraton point of a certain element, only the loading function f (ε, κ) needs to
be evaluated and, in case of loading, the internal parameter κ has to be updated. This property
greatly enhances the robustness of failure analyses of structures.

3.3 A Model for Concrete under Uniaxial Tension

As already mentioned above, concrete – as well as other similar quasi-brittle materials – suffers
damage and failure mainly due to tensile loading conditions. Therefore, ignoring the nonlinear
behaviour under compression, the Rankine criterion becomes an appropriate choice for the concrete
model.
The experimentally observed behaviour of concrete under uniaxial tension is almost linear up to
the peak (cracking), followed by the nonlinear branch of the stress-strain diagram, which can be

n o r m a liz e d p r in c ip a l s t r a in s n o r m a liz e d p r in c ip a l s t r e s s e s
1 .5 1 .5

1 .0 1 .0

0 .5 0 .5

0 .0 0 .0

- 0 .5 - 0 .5

- 1 .0 - 1 .0

- 1 .5 - 1 .5
- 1 .5 - 1 .0 - 0 .5 0 .0 0 .5 1 .0 1 .5 - 1 .5 - 1 .0 - 0 .5 0 .0 0 .5 1 .0 1 .5

a )
n o r m a liz e d p r in c ip a l s t r a in s n o r m a liz e d p r in c ip a l s t r e s s e s
2 2
1 1
0 0
- 1 - 1

- 2 - 2
- 3 - 3
- 4 - 4
- 5 - 5
- 5 - 4 - 3 - 2 - 1 0 1 2 - 5 - 4 - 3 - 2 - 1 0 1 2

b )
n o r m a liz e d p r in c ip a l s t r a in s n o r m a liz e d p r in c ip a l s t r e s s e s
2 2
1 1
0 0
- 1 - 1

- 2 - 2
- 3 - 3
- 4 - 4
- 5 - 5
- 5 - 4 - 3 - 2 - 1 0 1 2 - 5 - 4 - 3 - 2 - 1 0 1 2
c )
Figure 3.5: 2D illustration of loading surfaces in the strain space and their resulting shapes in the
stress space, using different definitions of the equivalent strain ε̃: a) Acc. to (3.15), b) acc. to (3.16),
c) acc. to (3.17)
.
24 3.3: A Model for Concrete under Uniaxial Tension

ft

ε0 εf ε

Figure 3.6: Rankine type exponential law for concrete under tension

well described by an exponential law:

 
ε − ε0
σ(ε) = ft exp − , (3.18)
εf

where ε is the uniaxial tensile strain, which is the driving variable of the model, according to (3.17).
And these are material constants: ε0 the strain correponding to the elastic limit state, ft = Eε0 the
uniaxial tensile strength (i.e. the elastic limit stress) and εf is a parameter controlling the post-peak
shape of the tensile stress-strain curve.
Recalling the definitions of damage and damage parameter in the previous section, one can write:
 
0 σ ε − ε0 1
E = = ft exp − = (1 − d)E, (3.19)
ε εf ε

where E 0 represents the “damaged elasticity modulus” (secant stiffness). From the formula above,
the expression of the damage parameter with respect to the internal parameter is obtained:

ε0 κ − ε0
d(κ) = 1 − exp(− ). (3.20)
κ εf

Now, one can derive the tangent stiffness E tan for the loading situation. Starting with

σ[d[κ(ε)], ε] = (1 − d)Eε, (3.21)

the differentiation of the stress gives

∂σ ∂σ
∆σ = ∆d + ∆ε = (−Eε)∆d + (1 − d)E∆ε, (3.22)
∂d ∂ε
and considering
 
∂d ∂κ ε0 κ − ε0 1 1
∆d = ∆ε = exp(− ) + ∆ε, (3.23)
∂κ |{z}
∂ε κ εf κ εf
=1
one can rewrite (3.22) as
 
 
(1 − d)E − Eε ε0 exp(− κ − ε0 ) 1 1 

∆σ = +  ∆ε
| {zκ} εf κ εf 
| {z }
E0
ft
 
 
E − ft exp(− κ − ε0 ) 1 1 
 0
= +  ∆ε
 εf κ εf 
| {z }
σ
  
0 1 1
= E −σ + ∆ε. (3.24)
κ εf
| {z }
E tan

from which it is obvious

 
∂σ 1 1
E tan
= = E0 − σ + . (3.25)
∂ε κ εf

With the tangent stiffness E tan on hand, recalling the isotropic assumption, the tangent stiffness
tensor Ctan is obtained according to

Ctan (E tan , ν) = 2µIsym + λ1 ⊗ 1, (3.26)

with µ and λ Lamé parameters, Isym the symmetric fourth-order unit tensor and 1 ⊗ 1 the dyadic
product of the second order unit tensor. Within the scheme of Finite Element Method, the tangent
stiffness tensor can be used for the computation of the element tangent stiffness matrix:
Z
Ke = BT Ctan BdV. (3.27)
Ω

Similar to the Elastoplasticity Theory, the numerical procedure of the isotropic damage model, for
concrete material under uniaxial strain-driven tensile loading, which is essentially a 1D case, can
be summarised as the following:

1. Initially, one has all the state variables on hand: Young’s modulus E, stress state σ = 0,
internal parameter κ = ε0 , damage parameter d = 0, strain state ε = 0 and the applied total
tensile strain εtot . The loading procedure is equally divided into m steps.

2. At the beginning of the load step n + 1 (n = 0, 1, 2, ..., m − 1), one knows all the state values
and results of the last step: En0 , σn , κn , dn and εn . An increment of the load ∆ε = εtot /m is
applied, therefore the strain at the end of this step is simply εn+1 = εn + ∆ε.

3. Check the loading function fn+1 = εn+1 − κn .

• IF f ≤ 0, it is an unloading step. Assign the state variables directly with the old values:
0
En+1 ← En0 , κn+1 ← κn , dn+1 ← dn , calculate σn+1 = En+10 εn+1 and finish the step.
• ELSEIF f > 0, it is a loading step. Update the internal parameter as κn+1 = εn+1 ,
calculate the damage parameter dn+1 using (3.20), the secant stiffness En+1 0 = (1 −
0
dn+1 )E, the stress σn+1 = En+1 εn+1 and the tangent stiffness E tan by (3.25).
26 3.3: A Model for Concrete under Uniaxial Tension

4. GOTO 2, in order to start a new step, until the whole loading procedure is completed.

Compared with the case in Elastoplasticity Theory, it is obvious that the algorithms of the damage
model described above does not require Prediction-Return Mapping iterations. One may get better
understanding about the explicit nature of Damage Models.
Bibliography

[1] L.M. Kachanov. Time rupture process under creep conditions. IVZ Acad Nauk SSR Otd Tech
Nauk, 8:26–31, 1958. 17

[2] J. Lemaitre. A Course on Damage Mechanics. Springer, Berlin, New York, 1996. 19

[3] J. Lubliner. Plasticity Theory. Macmillan Publishing Company, New York, 1990. 3

[4] E. Pramono and K. Willam. Implicit integration of composite yield surfaces with corners.
Engineering Computations, 6:186–197, 1989. 3

[5] J.C. Simo. Algorithms for static and dynamic multiplicative plasticity that preserve the clas-
sical return mapping schemes of the infinitesimal theory. Computer Methods in Applied Me-
chanics and Engineering, 99:61–112, 1992. 4

[6] J.C. Simo and T.J.R. Hughes. Computational inelasticity. Springer, Berlin, 1998. 3, 4

[7] J.C. Simo, J.G. Kennedy, and S. Govindjee. Non-smooth multisurface plasticity and vis-
coplasticity. Loading/unloading conditions and numerical algorithms. International Journal
for Numerical Methods in Engineering, 26:2161–2185, 1988. 3

[8] J.C. Simo and G. Meschke. A new class of algorithms for classical plasticity extended to finite
strains. application to geomaterials. Computational Mechanics, 11:253–278, 1993. 4

[9] J.C. Simo and R.L. Taylor. Consistent tangent operators for rate independent elasto–plasticity.
Computer Methods in Applied Mechanics and Engineering, 48:101–118, 1985. 3

[10] J.C. Simo and R.L. Taylor. A return mapping algorithm for plane stress elastoplasticity.
International Journal of Numerical Methods in Engineering, 22(3):649–670, 1986. 3