You are on page 1of 23

1 Damage Mechanics

Notes on Continuum
Damage Models
Notes on Continuum Damage Mechanics
E.W.V CHAVES and J. OLIVER (2013)

1.1 Introduction
The term Continuum Damage Mechanics has been used to models materials which are
characterized by loss of stiffness, i.e. by a decrease in their stiffness modulus. Damage
models have also been used to simulate different materials (fragile and ductile), which are
fundamentally characterized by irreversible material degradation. Physically speaking, we
can describe the degradation of mechanical material properties as processes in which the
initiation and growth (propagation) of micro-defects such as micro pores and microcracks
take place.
In the pioneering work of Kachanov (1958) the concept of effective stress was introduced,
and by using continuum damage he solved problems related to creep in metals. Rabotnov
(1963) gave the problem physical meaning by suggesting we measure how the sectional area
has reduced by means of the damage parameter. Nowadays, Continuum Damage
Mechanics has become an important tool and is a consistent theory based on irreversible
thermodynamic processes (the Clausius-Duhem inequality). Thermodynamic formalism
was developed by Lemaitre&Chaboche (1985) and among important contributors to our
knowledge about damage mechanics we can include: Mazars (1986), Mazars&PijaudierCabot (1985), Chaboche (1979), Simo&Ju (1987 a,b), Ju(1989), Oliver et al. (1990) and Oller
et al. (1990).
The continuum damage models, from a computational point of view, are very attractive
since these present simple algorithms and are satisfactory for solving large problems.
In this chapter we will present some basic damage models used to study the failure
mechanism after which we can develop more complex ones.

NOTES ON CONTINUUM DAMAGE MODELS

1.2 The Isotropic Damage Model in a Small


Deformation Regime
Continuum damage models have been widely accepted for simulating the behavior of
materials whose mechanical properties are degrading due to the presence of small cracks
that propagate during loading. To fully describe this phenomenon, we will first use a onedimensional model (1D) which we will then extrapolate to three dimensional ones (3D).
With regard to continuum kinematics, our study in this section will be carry out in a small
deformation regime, and will be based on the lecture notes of Prof. Javier Oliver,
Universitat Politcnica de Catalunya.

1.2.1 Description of the Isotropic Damage Model in Uniaxial


Cases
Let us now suppose that a material point is subjected to the stress state as shown in Figure
11.1, whose apparent stress ( ) acts on the section s and due to the presence of faults
(microcracks), only the undamaged region will be considered, i.e. the effective section ( s )
on which the effective stress ( ) acts.

material point

micro
crack

- effective stress
- apparent stress

Figure 11.1: Continuum with microcracks.


Then, if we consider the force balance in Figure 11.1, we obtain:
s = s

(11.1)

The equation (11.1) can also be rewritten without altering its outcome as follows:
=

s
s
ss

s s

=
+ 1 = 1
= 1 d
s
s
s
s

(11.2)

where s d is the damaged section.


Note, the expression

sd
represents the amount of the original section which is corrupted,
s

which in extreme cases, assumes the follows values:

NOTES ON CONTINUUM DAMAGE MODELS

sd = 0

sd
=0=
s

sd = s

sd
= 1 = 0 - The section is completely damaged.
s

- The section is not damaged;

The amount s d depends on the stress state or indirectly on . The dimensionless ratio
s
sd
represents the damage variable and is denoted by d = d . Then, the equation in (11.2)
s
s

can be written as:

= (1 d )

0 d 1

(11.3)

where is the effective stress.


1.2.1.1 The Constitutive Equation
The effective stress and strain, in the undamaged area element, are interrelated by
Hookes law as:
(11.4)

=E

where E is Youngs modulus. Then, by substituting (11.3) into (11.4) we can obtain the
constitutive equation for stress in the one-dimensional isotropic damage model:
= (1 d ) E

0 d 1

(11.5)

We can now verify that as the damage variable evolves, the state no longer returns to its
original value. Physically speaking, we can interpret this as once the material has suffered
damage this will be permanent. Hence, we can conclude that d 0 , which characterizes an
irreversible process. Now, the equation in (11.5) can still be written as:
= E sec_d

E sec_d = (1 d ) E

with

(11.6)

where E sec_d is the damage secant stiffness modulus with which we can observe that the damage
variable can be interpreted as a measure of the loss of stiffness modulus of the material.
In general, materials have a yield stress that separates the elastic (reversible process) from
the inelastic zone (irreversible process). In the strain space, we can represent the elastic
limit by the variable 0 , (see Figure 11.2), in which the damage process has not yet begun,
i.e.:
d =0

if

< 0

(11.7)

In the elastic region, the following is satisfied:


d = 0 ;

s d = 0

0 where 0 is the threshold that defines the elastic region.

In a representative stress-strain curve, (see Figure 11.2), during an unloading process


(d = 0) , the secant modulus is given by E sec_d = (1 d ) E and after unloading is complete,
there is no residual (permanent) strain, (see Figure 11.2), although the material has suffered
some internal damaged.
We can now summarize the basic features of the one-dimensional damage model as
follows:

NOTES ON CONTINUUM DAMAGE MODELS

= (1 d ) E

(0 d 1)

if

d =0

(11.8)

< 0

Now, by starting from the above equation we can obtain the energy equation in the system
as follows:
Y = = (1 d ) E
1
2

1
2

Y = (1 d )Y e

(11.9)

=Y e

where Y e is the elastic strain energy.

Loading(damage)

Elastic limit

1
0

E sec_d = (1 d ) E

Unloading / elastic loading

Figure 11.2: Stress-strain curve.

1.2.2 The Three-Dimensional Isotropic Damage Model


The basis of this damage model is to define a transformation between physical (real) and
fictitious spaces (effective space) in which the material is undamaged, (see Figure 11.3).
Physical space (real)

Effective space (fictitious)

effective stress

= (1 d )

Figure 11.3: Real and fictitious spaces.


NOTE: As described above, this model depends on a single variable: the damage parameter
d , which means that we are assuming a mechanical behavior in which the degradation is
independent of the orientation, and because of this, this model is referred to as the Isotropic
Damage Model. As a result of this, the fourth-order damaged elasticity tensor remains an
isotropic tensor.
1.2.2.1 Helmholtz Free Energy
Let us now consider the Helmholtz free energy function Y = Y ( F , T , a k ) , or simply free
energy, which is a function of the deformation gradient ( F ), temperature ( T ) and the set

NOTES ON CONTINUUM DAMAGE MODELS

of internal variables ( a k ). Let us also consider there is a process independent of


temperature, and the internal variable associated with the problem is characterized by the
damage variable d . Furthermore, as seen in previous chapters, as the Helmholtz free
energy must satisfy the principle of objectivity (see Chapter 6), we can express Y in terms
of the Green-Lagrange strain tensor ( E ), which in turn collapses with the infinitesimal
strain tensor in a small deformation regime, i.e. E . Then, if we consider all of the
above, the Helmholtz free energy can be expressed in terms of:
Y = Y (, d )

(11.10)

or explicitly as follows:

Y = (1 d ) Y e = (1 d ) : C e :
1
2

Helmholtz free energy for


isotropic damage model

(11.11)

where Y e ( ) is the elastic strain energy density, which is a function of strain only, and C e
is the elasticity tensor (or elastic stiffness tensor).
1.2.2.2 Internal Energy Dissipation and the Constitutive Equations
The damage model has thermodynamic consistency, and so, entropy inequality is fulfilled.
One way to express this entropy inequality is by means of the alternative form of the
Clausius-Planck inequality, (see Chapter 5), which is expressed by:

J
m3

Dint = : D h T + Y 0

(11.12)

Note that the terms : D , h T , Y have the unit of energy per unit volume (density
energy). In a small deformation regime D holds, and by considering the isothermal
process we have T = 0 , so, the equation in (11.12) becomes:

Dint = : Y 0

(11.13)

Y Y
: +
d
Y (, d ) =

(11.14)

Then, the rate of change of the free energy Y = Y (, d ) can be evaluated as follows:

Next, by substituting (11.14) into the internal energy dissipation given in (11.13) we obtain:

Dint = : Y (, d ) = :

Y Y
Y Y
d 0
d =
:
:
d

d

(11.15)

Note that the above inequality must hold for any admissible thermodynamic process, so, let
Y

us assume there is one where d = 0 . Here, we obtain Dint =


: 0 , which in turn

must also be true for any process. Additionally, if we have a process such that , the
Y
only way for the entropy inequality to be satisfied is when =
holds with which we

obtain the constitutive equation for stress. Thus, the entropy inequality becomes:
Y
Y
Y
d 0
d =
Dint = :
d
d

=0

(11.16)

NOTES ON CONTINUUM DAMAGE MODELS

Now, if we consider the energy equation Y = (1 d ) Y e , we obtain

Y
= Y e , thus
d

Dint = Y e d 0

(11.17)

d 0

(11.18)

where by definition Y e 0 . Then, to satisfy the inequality (11.17), the rate of change of the
damage parameter must satisfy:
Then, by means of thermodynamic considerations we can draw the conclusion that:
=

d 0

(11.19)

We can also express the rate of change of the Helmholtz free energy by means of the
equation in (11.11), i.e.:
Y = Y e (1 d ) d Y e = (1 d ) : C e : d Y e = : d Y e

(11.20)

Next, the rate of change of the elastic strain energy, Y e = 12 : C e : , was obtained as
follows:
1
Y e = ( : C e : + : C e : + : C e : )

(11.21)

where C e = 0 , since C e is constant, and as the elasticity tensor features major symmetry
e
( C eklij = C ijkl
), the equation in (11.21) becomes:
1
1
e
e
e
kl )
kl + ij C ijkl
kl ) = ( kl C eklij ij + ij C ijkl
Y e = ( ij C ijkl
2

e
= ij C ijkl
kl = : C e : = : =

(11.22)

1
:
(1 d )

Note that due to the major symmetry of C e , : C e = C e : is fulfilled.


Then, starting from the equation in (11.11) we can obtain the stress by taking the derivative
of the strain energy with respect to strain, i.e.:
ij =

Y ( , d )

(1 d ) 1 pq C epqkl kl = (1 d ) 1 C epqkl pq kl
=

ij
ij
ij
2
2

pq

1
= (1 d ) C epqkl kl
+ pq kl

ij
ij
2

1 pq + qp
[1 ( kl + lk )]
1
= (1 d ) C epqkl kl 2
+ pq 2

2
ij
ij

1
1
1

= (1 d ) C epqkl kl d pi d qj + d qi d pj + pq d ki d lj + d li d kj
2
2
2

1 1
1
1

e
e
= (1 d ) kl C ijkl
+ C ejikl + pq C epqij + C epqji = (1 d ) kl C ijkl
+ pq C epqij
2 2
2
2

[(

)]

(11.23)

where we have taken into account the minor symmetry of the elasticity tensor, i.e.
e
C ijkl
= C ejikl , C epqij = C epqji . Note also that the indexes p , q are dummy indexes, so we can

NOTES ON CONTINUUM DAMAGE MODELS

exchange them for k and l without altering the expression. Additionally, by taking into
e
account the major symmetry of the elasticity tensor, C ijkl
= C eklij , we obtain:
e
ij = (1 d ) C ijkl
kl

(11.24)

which in tensorial notation becomes:


=

Y ( , d )
= (1 d ) C e : = (1 d )

The constitutive equations for isotropic


damage model

(11.25)

where is the effective Cauchy stress tensor and is defined as:


= Ce :

The effective Cauchy stress tensor

(11.26)

and C e is the elasticity tensor (fourth-order definite positive tensor) which contains the
elastic mechanical properties. Remember that C e can be represented in terms of the Lam
constants ( l , m ) as follows:
C e = l1 1 + 2 m I

e
C ijkl
= ld ij d kl + m d ik d jl + d il d jk

(11.27)

where 1 is the second-order unit tensor, and I I sym is the symmetric fourth-order unit
tensor, whose components are expressed in terms of the Kronecker delta ( d ij ) as follows:
1
(1) ij = d ij =
0

if
if

i= j
i j

sym
I ijkl
(I) ijkl = ijkl =

1
d ik d jl + d il d jk
2

(11.28)

Then, by analyzing the constitutive equation in (11.25) we can put in evidence the
following sentences:
Since the damage parameter is a scalar, the stiffness degradation is isotropic;
We can calculate the stress immediately once we know the current values of
(strain) and d (internal variable);
We can interpret the equation in (11.25) as the sum of elastic and inelastic parts, i.e.:
e
e
= (1 d ) C e : = C
: = e i
: d
C

elastic

inelastic

(11.29)

The Elastic-Damage Secant Stiffness Tensor

We can then define the elastic-damage secant stiffness tensor for the isotropic damage
model as:
C sec_d = (1 d ) C e

The elastic-damage secant stiffness tensor

(11.30)

Let us now consider a uniaxial case, (see Figure 11.4), where the material is loaded until the
stress state reaches the point P represented in Figure 11.4, after which unloading occurs,
with the unloading path being that indicated by the slope E sec_d = (1 d ) E defined in
Figure 11.4.
1.2.2.3 Ingredients of the Damage Model
The damage constitutive model is completely determined when the damage variable d t is
known at each time step t of the loading/unloading process. Then, we can define the
following elements of the constitutive equation:

NOTES ON CONTINUUM DAMAGE MODELS

The energy norm of the stress (or strain) tensor;

The damage surface and damage criterion. The damage surface defines the elastic
limit, and the damage criterion establishes when the material is in a loading or in a
elastic process, and;

A set of evolution laws for internal variables.

Dissipated energy

E
1

E sec_d = (1 d ) E

Figure 11.4: Stress-strain curve.


The Energy Norm in the Stress/Strain Space

The norm is a measure of distance and so is a scalar. Next, we will define a simple norm in
the stress space denoted by t (equivalent stress), and in the strain space denoted by t .
The latter is also known as the equivalent strain:
1
e
e
: Ce :
;
e 1 =
= Ce = : C : = 2
C

t =

t = (1 d )t

(11.31)

Note that t and t are surface equations (ellipsoids) that characterize the stress state at
the current point (see Figure 11.5). The proof of (11.31) now follows:

t = : C e : = (1 d ) : = (1 d )2 : = (1 d ) :
1

t = (1 d )t (11.32)

t = :C : = :
e

In order to better describe material behavior, others norms will be introduced (see
subsection 11.2.4).
The Damage Criterion

Next we will define the damage criterion in the stress and strain space:
F (t , q ) = t q (r ) 0 and G(t , r ) = t r 0

stress space

strain space

(11.33)

where r is an internal variable (current damage threshold), and q is a stress-like


hardening/softening variable which is a function of r . Note that each material in its
undamaged state is characterized by the initial value of r which is denoted by r0 (the
material parameter), which defines the initial yield in the strain space. Then, the material

NOTES ON CONTINUUM DAMAGE MODELS

starts to fail (initial damage) when the energy norm exceeds the value r0 . Later we will
relate the variables r and q to the damage variable.
2

t = (1 d )t

G(t , r ) = 0

t
F (t , q ) = 0

1
3

b) Stress principal space

a) Strain principal space

Figure 11.5: Strain and stress state in the principal space.


The damage criterion requires that the current stress state must be on or inside the damage
surface. When the stress state lies inside said damage surface, the material shows elastic
behavior, which can be elastic loading or unloading.
Then we can define the admissible strain space as follows:

E := { G(t , r ) 0 }

(11.34)

E := { F (t , q) 0 }

(11.35)

and the admissible stress space as:

When it holds that F (t , q) = 0 , in the stress space, the stress state is on the surface as
indicated in Figure 11.5(b).
The stress space ( E ), (see Eq. (11.35)), can be decomposed into the inner domain
int (E ) (when the stress state is inside the surface), and other by the surface itself, E .
We can define then the elastic region in strain and stress respectively as:
intE := { G(t , r ) < 0 }

intE := { F (t , q ) 0 }

(11.36)

E := { G(t , r ) = 0 }

E := { F (t , q ) = 0 }

(11.37)

and the elastic limit (damage surface):


where it holds that:

E = int (E ) E

int (E ) E =

(11.38)

Note that int (E ) is the same as F (t , q) < 0 which describes the elastic region, and E
is the damage surface. Note that when the stress state is at a point inside of the space E it
will also be inside the space E , and when the stress state is on the surface E it will also
be on the surface E . Hence, we can use either the stress or strain space to describe how
the damage evolves, proof of which follows:

t q(r ) 0

(1 d )t (1 d )r 0

t r 0

(11.39)

NOTES ON CONTINUUM DAMAGE MODELS

10

Said damage evolves when the norm t exceeds the maximum value reached by r . Then,
considering (11.33) and (11.31) we can also conclude that:
q (r ) = (1 d )r

(11.40)

In uniaxial cases, damage starts when t exceeds the first damage threshold value r0 .
Then, from the equation in (11.31) and by means of Figure 11.2, we can obtain:

t = : C e : uniaxial

t = 0 E 0 = 0 E =
t r0 = 0

r0 =

Y
E

E=

Y
E

(11.41)

Y
E

where Y is the yield stress (obtained in the laboratory). Then, r0 ( Y , E ) can be


interpreted as a material mechanical property also obtained in the laboratory.

2( r ( 2 ) r0 )

3( r (3) t )
4( r ( 4 ) t
)

5( r (5) r ( 4 )
)

t
2

r0

6
5

2
1

t
=0
>0 r
r

3
2

4,

r0

t
r0

r3

r4 = r5

Figure 11.6: The evolution of t and r over time t .


The Internal Variable Evolution Law. The Kuhn-Tucker and Consistency Conditions

NOTES ON CONTINUUM DAMAGE MODELS

11

The constitutive equation described above uses three types of variables, namely: the free
variable { } ; the internal variable {r}; the dependent variables {Y (, r ), (, d ), d (r )} .
Now, to establish how the internal variable r evolves, let us take the example described in
Figure 11.6. As we can observe, the discretized r between points 2-3 and 3-4 are positive
and between points 1-2 and 4-5 are equal to zero, so we can conclude that r is a
monotonically increasing function, i.e.:
(11.42)

r 0

Graphically, we can see in Figure 11.6 how the variables r and t evolve. Furthermore,
we can also verify that in the range between the points 4-6 G(t , r ) = t r < 0 holds, i.e.
there is an elastic regime.
Thus, we can establish that at time t , r t is given by the following equation:

r t = max r0 , max
t s
s( ,t ]

(11.43)

As we saw, the Helmholtz free energy is a function of Y = Y (, d (r ) ) , where now the


damage variable is in terms of the variable r (the internal variable), thus:
Y Y d (r ) r
Y (, d (r ) ) = [1 d (r )] Y e ( ) Y (, d (r ) ) =
: +
d r t

d (r )
Y
r
=
: Y e
r

where we have considered that


energy dissipation becomes:

(11.44)

Y
= Y e . Then, by using the above equation, the internal
d

d (r )
Y
: + Y e
r 0

r
d (r )
d (r )
Y

: + Y e
r = Y e
r 0
=

r
r

Dint = : Y (, r ) = :

(11.45)

=0

If we compare the above inequality with the one obtained in (11.17) we can conclude that:
d (r )
d =
r = H(t , d ) r
r

(11.46)

where H is the continuum hardening/softening modulus. The evolution laws for r and
for d (damage variable) are then given by:
r = z (, r )

d = z H(t , d )

(11.47)

where z 0 is the damage parameter consistency (damage multiplier).


Let us now consider a time t in which the process is characterized by t and r . Then, we
can observe the following: if at any time both G(t , r ) t = 0 and G(t , r ) > 0 hold, this
implies that G(t , r ) t + t > 0 , which thereby violates the condition {G(t , r ) t 0 t }, so
z > 0 G = 0 must be satisfied. Another possible situation is when the current state is
inside the damage surface, i.e. G(t , r ) t < 0 , and if in the next loading step G(t , r ) t + t < 0

NOTES ON CONTINUUM DAMAGE MODELS

12

is satisfied, this implies that G < 0 r = z (, r ) = 0 . We can gather these previous


conditions by means of the loading/unloading condition, also called the Kuhn-Tucker conditions:
z 0

G(t , r ) 0

z G(t , r ) = 0 The Kuhn-Tucker conditions

(11.48)

and by the consistency (persistency) condition:


z G(t , r ) = 0 The consistency condition

(11.49)

If we are undergoing loading, this implies that z > 0 , then by means of the Kuhn-Tucker
conditions G(t , r ) = 0 must be fulfilled. Here, the value of z can be obtained by means
of the consistency condition:
G(t , r ) = G(t , r ) = 0

t = r

(11.50)

Schematically, we can summarize the above loading/unloading states as follows:


G < 0

G < 0

G = 0

G = 0

z = 0

d = 0

(elastic)

z = 0

d = 0

(unloading)

d = 0

(neutral loading)

d > 0

(loading)

z = 0

z > 0

(11.51)

NOTE: If the parameter H(t , d ) , given in (11.46), is not a function of d , we can

express it by means of H(t ) =

G (t )
, where we have introduced the scalar function G
t

which is a monotonically increasing function, which has proven to be a convenient way to


express the damage criteria:
G (t , r ) = G (t ) G (r ) 0

F (t , q ) = F (t ) F (q ) 0

t0

;
;

t0

(11.52)

Here the loading/unloading condition becomes:


r = z (, r )

z 0

G (t , r ) 0

G (t , r )
d = z
r

(11.53)

z G (t , r ) = 0 The Kuhn-Tucker conditions

(11.54)

z G (t , r ) = 0 The consistency condition

(11.55)

The Damage Variable

The parameter q is the stress-like hardening/softening parameter, and is defined in terms


of r as follows:
q (r ) = (1 d )r

d (r ) = 1

q(r )
r

Now, by using the equations in (11.56) and (11.25) we can obtain:

(11.56)

NOTES ON CONTINUUM DAMAGE MODELS

13

q(r )

(11.57)

in which the following holds:


r [r0 , ]

0 d 1

(11.58)

Note that with the new definition of the damage parameter given in (11.56), we can
restructure the equation in (11.46) as follows:
q(r )
r
q (r )

q (r ) r
(
)
(
)
d
r
q
r

d =
r = 1
r =
r
r
r
r
r2

(11.59)

d (r ) = 1

q(r ) H d (r )
d =
r
r2

where we have defined a new parameter H d (r ) =


parameter.

(11.60)

q (r )
, which is the hardening/softening
r

1.2.2.4 The Hardening/Softening Law


The expression
q = H d (r ) r ;

q (r )
defines the hardening/softening parameter, thus:
r
q [r0 , a ] ;

r [r0 (d = 0), (d = 1) ) ;

q 0 = r0 =

Y
E

(11.61)

where H d is the continuum hardening/softening parameter and which is characterized by:


Damage with Hardening H d (r ) > 0
Perfect Damage
H d (r ) = 0
Damage with Softening H d (r ) < 0

(11.62)

Here, we will consider the relationship between q and r to be linear or exponential.


The Linear Hardening/Softening Law

Now, by assuming that q varies linearly with r , we have:


r0

r0

q(r ) =
r + H d (r r ) </ 0
0
0

(11.63)
r > r0

Then, taking into account the equation in (11.56) we can still state that:
0
q
d =1 =
r
1

r r0
r0
r

H 1
d

r0
r

) >/ 1

(11.64)
r > r0

q (r )
Hd > 0
r0

Hd < 0

NOTES ON CONTINUUM DAMAGE MODELS

14

Figure 11.7: The linear hardening/softening law.


The Exponential Hardening/Softening Law

The exponential law is described by Figure 11.8. Then we can express q(r ) as follows:
q (r ) = q (q r0 )exp

r
A 1
r0

(11.65)

A>0

with

in addition to this we have:

(q r ) A1 r
q (r )
= A 0 exp 0
r0
r

(11.66)

q (r )
q
q > r0
r0
q < r0
r

r0

Figure 11.8: The exponential hardening/softening law.


Table 11.1: Summary of the Isotropic Damage Model in a small deformation regime
described in the strain space.
ISOTROPIC DAMAGE MODEL IN A SMALL DEFORMATION REGIME

Helmholtz free energy

Y (, r ) = [1 d (r )] Y e

Damage parameter

d (r ) = 1

The constitutive equations

Evolution law

q
r

with

Ye =

q [r0 , a ], a ;

Y
= (1 d ) = (1 d ) C e :

r [r0 , )

r = z
Y

r0 = r t =0 =
E

Damage criterion

G(, r ) = t r = : C e : r

Hardening Law

q = H d (r ) r

(H

= q (r ) 0

1
: Ce :
2
d [0,1]

(11.67)
(11.68)
(11.69)
(11.70)
(11.71)
(11.72)

NOTES ON CONTINUUM DAMAGE MODELS

Loading/unloading condition

G<0 ;
z G = 0

Consistency condition

z 0

15

z G = 0

(11.73)
(11.74)

1.2.3 The Elastic-Damage Tangent Stiffness Tensor


Next, we will obtain the elastic-damage tangent stiffness tensor, which gives us an
advantage, from a computational point of view, when we are dealing with the incrementaliterative solution procedures and as a result of this, convergence is improved considerably.
The relationship between and give us this tensor denoted by C tan_d , i.e. = C tan_d : .
Now, by considering the equation in (11.25), = (1 d ) C e : , we can obtain the rate of
change of the stress as follows:


(, d ) =
: +
d = (1 d ) C e : C e : d = (1 d ) C e : d

d
= (1 d ) C e : d

(11.75)

in which there is the following:


a)

A process with elastic loading or unloading

z = 0 d = 0 , thus the equation in (11.75) becomes (, d ) = (1 d ) C e : , with which the


elastic-damage tangent stiffness tensor coincides with the elastic-damage secant stiffness
tensor when we are dealing with elastic loading:
C sec_d = C tan_d = (1 d ) C e = x C e

b)

(11.76)

x = (1 d )

where

A process with damage loading

t = r t = r , and the rate of change of the damage parameter d = d (r ) becomes:


d
d r d
t
=
r =
d =
t
r t r

(11.77)

where the rate of change of t can be evaluated as follows:

t = : C e :
t

=
=

1
: Ce :
2
1
: Ce :

1
2

) ( : C

: + : C e :

: C e : =

: C e : =

(11.78)

Now, by substituting (11.78) into the equation in (11.77) we obtain:


d 1
:
d =
t t

(11.79)

Then, taking into account the equations (11.79) and (11.75), we can find the relationship
between the rates of stress and strain change:
d 1
= (1 d ) C e : d = (1 d ) C e :
:
t t

d 1
( ) :
= (1 d ) C e
t t

which thus defines the elastic-damage tangent stiffness tensor:

(11.80)

NOTES ON CONTINUUM DAMAGE MODELS

16

d 1
( )
C tan_d = (1 d ) C e
t t

(11.81)

and by considering that in a loading process t = r holds we then find:


d 1
d 1 q (r ) H d r 1 q (r ) H d r
=
=
=

t t
r r
r2
r3
r

where we have taken into account the equation in (11.61),

(11.82)

d q (r ) H d r
.
=
r
r2

Then, by substituting the equation in (11.82) into that in (11.81) we can obtain C tan_d in
terms of q and r :

q (r ) H d r e
e
C
C tan_d = (1 d ) C e
:
:C

(11.83)

Now, the general equation for the elastic-damage stiffness tensor C tan_d (symmetric
fourth-order tensor) is given by:
elastic with (d = 0) The elastic-damage stiffness
C e
C tan_d =
tensor for isotropic damage (11.84)
x C e K C e : : C e loading unloading

model

where, K =

r =0K=0

q(r ) H d r
and x = (1 d ) .
r3

1.2.4 The Energy Norms


Next, we will define some energy norms, which together with the damage criteria, play an
important role in defining the yield damage surface.
In order to adequately represent the materials different norms will need to be defined so as
to describe how these materials really behave. For example, in a simple model for concrete,
if we only want to simulate the process of failure caused by tension, the tension-only damage
model is used which means that it cannot capture the other type of failure caused by
compression. Next, we will define some models used in the isotropic damage process.
1.2.4.1 The Symmetrical Damage Model (Tension-Compression)
Model I
This type of model shows when the material behave the same both with tension or and
compression. The energy norm of this model is then represented by:
1

t I = : C e : = (1 d ) :

(11.85)

We can also define the energy norm of the strain tensor (also known as the equivalent
strain), proposed by Simo&Ju(1987), (see equation (11.31)):

t I = : C e : = 2Y e

(11.86)

NOTES ON CONTINUUM DAMAGE MODELS

17

To better illustrate this model, let us consider the state of plane stress ( i 3 = 0 ). In this
case, the yield surface is represented by an ellipse, (see Figure 11.9), where Y > 0 is the
stress limit for tension and compression and the damage surface evolves symmetrically.
2

t = r0

E
1
Elastic
region

Y
b) Stress-strain curve

a) Norm in the principal stress space-2D.

Figure 11.9: Damage surface in 2D and the uniaxial stress-strain curve for model I.
1.2.4.2 The Tension-Only Damage Model Model II
The tension-only damage model does not take into account failure by compression, i.e. the
material can only fail by tension and here we can define the following stress field:
+ = =
def

where =

+
2

(11.87)

is the Macaulay bracket whose graphical representation can be

appreciated in Figure 11.10.


x

0
x =
x

if x < 0
if x 0

x
Figure 11.10: Ramp function.
Now, by means of spectral representation, we can represent the stress tensor in terms of
eigenvalues (principal stresses) and eigenvectors as follows:
=

( a ) n ( a )

an

(11.88)

a =1

thus:
+ =

n ( a ) n ( a )

(11.89)

a =1

Note, the relationship between the real and effective stress remains valid, i.e.:
+ = (1 d ) +

(11.90)

NOTES ON CONTINUUM DAMAGE MODELS

18

Then, the norm for the isotropic damage model defined previously becomes:
(11.91)

t = 2Y e = : C e : = :
Next, in the tension-only damage model + , it follows that:
1
1
1
1
+ : Ce : =
+ : Ce :
2
(1 d )
(1 d )

t II = + : = + : C e : =

(11.92)

Then, if we consider the equation in (11.31), we can conclude that:


(11.93)

t II = + : C e :

Finally, in Figure 11.11 we can visualize the damage surface for two-dimensional cases
(2D).
1.2.4.3 The Non-Symmetrical Damage Model Model III
The non-symmetrical damage model is useful to simulate materials, such as concrete,
whose tension domain differs with respect to compression. This model uses the following
norm:

t III = q +

1 q

: Ce :

(11.94)

where the parameter q is the weight factor dependant on the stress state which is given
by:
3

q=

i =1
3

(11.95)

i =1

The parameter n is defined by means of the ratio of the compression elastic limit Yc to
the tension elastic limit Yt , i.e.:
n=

Yc

(11.96)

Yt

In the case of concrete n is approximately equal to n 10 .

t = r0

E
1
Elastic region

a) Norm in the principal stress space-2D.

b) Stress-strain curve.

NOTES ON CONTINUUM DAMAGE MODELS

19

Figure 11.11: Damage surface in 2D and the uniaxial stress-strain curve for model II.

Yt

t = r0

Yt
Yc

Yt

Elastic
region

Yc
a) Norm in the principal stress space-2D.

Yc = n Yt
b) Stress-strain curve.

Figure 11.12: Damage surface in 2D and the uniaxial stress-strain curve for model III.

1.3 The Generalized Isotropic Damage Model


Note that the elasticity tensor C e can be written in terms of the following sets of
mechanical parameters (l, m) , ( E , ) , ( , G ) :
C e = l1 1 + 2 m I =

1
E
E

1 + 2 m I 1 1
I=
1
11 +

3
(1 + )
(1 + )(1 2)
volumetric

part

(11.97)

isochoric part

where (E ) =Youngs modulus, () =Poissons ratio, (l, m) =Lam constants, () =bulk


modulus, and G = m is the shear modulus.
In the isotropic damage model the elastic-damage secant stiffness tensor can be
represented as follows:
C sec_d = (1 d )C e =

(1 d ) E
(1 d ) E
E sec_d
E sec_d
11 +
I=
11 +
I
(1 + )
(1 + )(1 2)
(1 + )
(1 + )(1 2)

Note that, in this model the damage variable affects only one of the mechanical parameters,
namely, the Youngs modulus. We can also verify that the same damage parameter equally
affects both the spherical and deviatoric part:
1

C sec_d = (1 d )C e = (1 d ) 1 1 + (1 d )2 m I 1 1
3

(11.98)

Another model described by Carol et al. (1998) generalizes the isotropic damage model by
considering independent degradation of the spherical and deviatoric parts and because of
this the model requires two independent damage variables.

NOTES ON CONTINUUM DAMAGE MODELS

20

Now, the elasticity tensor components can be expressed by means of their spherical and
deviatoric parts as follows:

1
1

e
C ijkl
= d ij d kl + 2 m d ik d jl + d il d jk d ij d kl
3
2

1
3

V
D
Then, with Pijkl
= d ij d kl and Pijkl
=

(11.99)

1
V
, the above equation becomes:
d ik d jl + d il d jk Pijkl
2

e
V
D
C ijkl
= 3Pijkl
+ 2 m Pijkl

C e = 3P V + 2 m P D

(11.100)

Let us now consider that the material parameters and m can be degraded by means of
the variables d V and d D , respectively, and according to the following equations:
= (1 d V ) 0

(11.101)

; m = (1 d D )m 0

with which the elastic-damage secant stiffness tensor becomes:


sec_d
e _V
e_D
V
D
= 3(1 d V ) 0 Pijkl
+ 2 (1 d D )m 0 Pijkl
= (1 d V )C ijkl
+ (1 d D )C ijkl
C ijkl

(11.102)

where we have introduced:


e _V
V
C ijkl
= 3 0 Pijkl

= 0 d ij d kl

e_D
D
C ijkl
= 2m 0 Pijkl

;
1

1
= 2m 0 d ik d jl + d il d jk d ij d kl

(11.103)

1.3.1 The Strain Energy Function


Now, if we consider (11.100), the elastic strain energy function can be rewritten as follows:
Y e = : C e : = : (3P V + 2 m P D ): = : (3P V ): + : (2 m P D ):
1
2

=Y

1
2

e _ vol

+Y

1
2

1
2

(11.104)

e _ dev

where we have introduced:


Y e _ vol = : (3P V ): = : C e _ V :

2
2
1

e _ vol

e
e _ vol
+ Y e _ dev
Y ( ) = Y
1
1
D
e_D
:
= : 2m P : = :C

2
2

(11.105)

after which it becomes:


1
2

1
2

Y (, d V , d D ) = : C sec_d : = : (1 d V )C e _ V + (1 d D )C e _ D :
1
1
= (1 d V ) : C e _ V : + (1 d D ) : C e _ D :
2
2
V
e _ vol
D
e _ dev
= (1 d )Y
+ (1 d )Y
= Y vol (, d V ) + Y dev (, d D )

=Y vol

Additionally, the following holds:

=Y dev

(11.106)

NOTES ON CONTINUUM DAMAGE MODELS

21

1
1
2
2
1
1
= (1 d V ) sph : C e _ V : sph + (1 d D ) dev : C e _ D : dev
2
2
V
e _ vol
D
e _ dev
vol
= (1 d )Y
+ (1 d )Y
= Y ( sph , d V ) + Y dev ( dev , d D )

Y (, d V , d D ) = (1 d V ) : C e _ V : + (1 d D ) : C e _ D :

=Y vol

(11.107)

=Y dev

1.3.2 Spherical and Deviatoric Effective Stress


Note that the following equations hold:
= C sec_d : = (1 d V )C e _ V : + (1 d D )C e _ D : = (1 d V ) sph + (1 d D ) dev

(11.108)

where sph , dev are the spherical and deviatoric effective stresses, respectively and where
the following is valid:
sph = (1 d V ) sph
= sph + dev
dev
D
dev
= (1 d )

(11.109)

It is noteworthy that the following equations hold:


= (1 d V )C e _ V : + (1 d D )C e _ D :

= (1 d V )C e _ V : sph + dev + (1 d D )C e _ D : sph + dev


= (1 d )C
V

e _V

sph

+ (1 d )C
D

e_D

(11.110)

dev

Then, the relationship between stress and strain in rate is given by:
= C tan _ d :

sph

dev

=C

tan _ d

=C

tan _ d

sph

sph

+C

dev

tan _ d

sph
= C tan _ d : sph

dev

= C tan _ d : dev

dev
:

(11.111)

where C tan _ d is the elastic-damage tangent stiffness tensor.

1.3.3 Thermodynamic Considerations


In a small deformation regime D holds and in isothermal processes T = 0 is satisfied,
so, it then follows that the expression for internal energy dissipation given in (11.13)
becomes:
D = : Y 0
(11.112)
int

Then, by evaluating the rate of change of the strain energy function given in (11.106),
Y = (1 d V )Y e _ V + (1 d D )Y e _ D , we can obtain:
Y = Y e _ V (1 d V ) Y e _ V d V + Y e _ D (1 d D ) Y e _ D d D
= Y e _ V (1 d V ) + Y e _ D (1 d D ) Y e _ V d V Y e _ D d D

(11.113)

and by using the stress equation given in (11.108) we have:

: = (1 d V )C e _ V : + (1 d D )C e _ D : :
= (1 d V ) : C e _ V : + (1 d D ) : C e _ D :

(11.114)

NOTES ON CONTINUUM DAMAGE MODELS

22

Note that Y e _ V = : C e _ V : and Y e _ D = : C e _ D : , thus:


: = (1 d V ) : C e _ V : + (1 d D ) : C e _ D :
= (1 d V )Y e _ V + (1 d D )Y e _ D

(11.115)

Then, together the equations (11.115), (11.113), and the internal energy dissipation given in
(11.112), yields:

Dint = : Y 0

= (1 d V )Y e _ V + (1 d D )Y e _ D Y e _ V (1 d V ) Y e _ D (1 d D ) + Y e _ V d V + Y e _ D d D 0
= Y e _ V d V + Y e _ D d D 0

(11.116)

Since (11.116) must be satisfied for any admissible thermodynamic process, it follows that:
d V 0

d D 0

(11.117)

where we have taken into account that Y e _ V 0 and Y e _ D 0 .

1.3.4 The Elastic-Damage Tangent Stiffness Tensor


Initially we adopt the following norms:

t V = 2Y e _ V = sph : = sph : sph = sph : C e _ V : sph

(11.118)

t D = 2Y e _ D = dev : = dev : dev = dev : C e _ D : dev

(11.119)

where the following holds:

t V =

t D =

sph

t D

:C

e _V

dev

sph

sph

: C e _ V : sph =

sph

: sph =

sph

(11.120)
(11.121)

Next, we obtain the rate of change of the Cauchy stress tensor:


V
D sph dev

D
(, d V , d D ) =
: + V d +
d =
: ( + ) + V d V +
d
D

d
d
d
d D

D
(11.122)
sph
dev
=
: + V d V +
: +
d
D
d
d

sph
dev
=
+

where the following holds, (see equation (11.109)):

= sph
V
d

= dev
D
d

(11.123)

d V r V d V V d V V
d V = V
t
= V r =
t V
r t
r

d D r D d D D d D D
d D = D
t
= D r =
t
t D
r
r

(11.124)

and

Then, we can express the rates of change sph and dev as follows:

NOTES ON CONTINUUM DAMAGE MODELS

23

sph

d V V
t
: + V d V = (1 d V )C e _ V : sph
sph =

d
t V

d V 1
= (1 d V )C e _ V : sph
sph :
t V t V

d V 1
= (1 d V )C e _ V
sph sph :
V
V
t t

(11.125)

sph
d V 1
sph
sph
= (1 d V )C e _ V

:
t V t V

and
D
dev
D
D
e_D
dev d
dev =
: +
d
=
(
1

d
)
C
:

t D
D
D

d
t

= (1 d D )C e _ D : dev

d D
= (1 d D )C e _ D
t D

d D
= (1 d D )C e _ D
t D

d D 1
dev :
t D t D
1

t D
1

t D

dev

dev

(11.126)

dev dev : dev

with which we can define the following equation:

d D 1
d V 1
dev
dev

sph sph :
= (1 d D )C e _ D + (1 d V )C e _ V

D
D
V
V
t t
t t

(11.127)

and by comparing the above with (11.111), we can conclude that:


C tan _ d = (1 d D )C e _ D + (1 d V )C e _ V

d D 1
d V 1
dev
dev

sph sph
t D t D
t V t V

(11.128)

You might also like