You are on page 1of 31

Cohesive zone models

towards a robust implementation of irreversible behaviour


Rene Kregting
23rd February 2005
MT05.11

Supervision:
dr.ir. O. van der Sluis (Philips Applied Technologies)
dr.ir. R.H.J. Peerlings
prof.dr.ir. M.G.D. Geers

Contents
1 Introduction

2 Cohesive zone models

2.1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2

General theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3

MSC.Marc user subroutines . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3 Implementation issues

10

3.1

Numerical integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10

3.2

History dependency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

13

3.2.1

Uncoupled irreversibility . . . . . . . . . . . . . . . . . . . . . . . . . .

14

3.2.2

Coupled irreversibility . . . . . . . . . . . . . . . . . . . . . . . . . . .

16

4 Benchmark test

20

5 Conclusions and recommendations

24

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A Derivation of tangent stiffnesses

26
28

Chapter 1

Introduction
Deformation and fracture are complex processes, which exhibit the interactions between the
different phases in the material and between micro- and macrocracks, which may initiate and
propagate simultaneously. Philips Applied Technologies and more specifically the process
modelling group (PPM) investigates failure behaviour of ceramic matrix composites for
optimization purposes. These composites consist of a ceramic matrix with metal inclusions.
For this reason, these composites can exhibit a combination of brittle and ductile behaviour.
Fracture in these heterogeneous materials consists of different stages: initiation of voids
and microcracks in random sites throughout the body, the subsequent growth, interaction,
clustering, coalescence. This leads to the formation of initial macrocracks, their growth and,
finally, propagation of one of the cracks up to complete failure. Initial stresses, present in
the composite due to processing, have a large influence on the resulting behaviour of the
composite, (van der Sluis, 2004).
Over the last decades several numerical methods have been developed to simulate failure
mechanisms in materials. The Finite Element Method (FEM) provides a way to predict the
failure behaviour of materials, in order to optimize the microstructure of these composites.
For this purpose, appropriate continuum mechanical failure models have to be used to
describe microstructural failure mechanisms. Several failure models are available to this end.
Fracture mechanics, continuum damage mechanics and XFEM are possible methods to
simulate failure behaviour in materials. The scale at which fracture occurs is equal to
the scale at which the microstructure is evaluated, therefore fracture will to be simulated
explicitly. Also, fracture can occur at physical interfaces. A method like continuum damage
mechanics is therefore not very suitable. Since the behaviour can be quasi-brittle, the use
of fracture mechanics is also not recommended. A different technique to simulate crack
initiation and crack growth is the use of so-called cohesive laws implemented in finite
elements. The basis for cohesive zone models can be traced back to the works of Dugdale
(1960) and Barenblatt (1962). The implementation of these cohesive zone models is rather
straightforward in commercially available finite element packages. Calculations on crack
initiation and crack growth are possible for both ductile and brittle materials. However,
there are still a number of issues that need to be solved before this approach can be used for
realistic simulations.
2

This report focusses on a few numerical issues which still affect the behaviour of cohesive
zones. The goal is to develop a robust cohesive zone implementation which includes history
dependent behaviour. An implementation in MSC.Marc which has already been developed by
Marco van den Bosch (2004) will be used as the starting point. Also, previous research has
shown that spurious oscillations of computed stresses can occur under certain circumstances
which may lead to undesirable results. The occurrence of these oscillations will be discussed, as
well as some remedies. Finally a benchmark test will be simulated to determine if the resulting
cohesive zone behaviour (including history dependent behaviour) is robust. The benchmark
test will also be used to verify the computed stresses .

ip2
1
3

lus = 2000 [MPa]


h = 10 [mm]
l = 22.5 [mm]
hnotch = 2 [mm]
= 0.3
n
t
0max
eff < max
max

max

eff = max
1max
t
n
max

max

Chapter 2

Cohesive zone models


2.1

Introduction

eff < max


eff = max The viewpoint from which cohesive zone models originate regards fracture as a gradual phe0max nomenon in which separation takes place across an extended crack tip, or cohesive zone,
1max and is resisted by cohesive tractions (Ortiz and Pandolfi, 1999). Thus cohesive zone elements
n do not represent any physical material, but describe the cohesive forces which occur when
t material elements (such as grains) are being pulled apart. Therefore cohesive zone elements
p
2 2 + 2
=

ff
n are placed between continuum (bulk) elements, as shown in Figure 2.1.
t
eff > max,old
yes
no
Loading
Unloading
nts
Tn = eq(2.3)
eleme
Bulk
ents
Tt = eq(2.4)
elem
zone
e
v
Ti
i
s
Cohe
i = eq(B.1-B.4)
nts
Tn = eq(3.11)
eleme
Bulk
Tt = eq(3.12)
= eq(B.16-B.19)
Figure 2.1: Application of cohesive zone elements along bulk element boundaries
When damage growth occurs these cohesive zone elements open in order to simulate crack
initiation or crack growth. Since the crack path can only follow these elements, the direction
of crack propagation strongly depends on the presence (or absence) of cohesive zone elements,
implying the crack path is mesh dependent. However, refining the mesh reduces this problem.
In two dimensions, tractions can occur in the normal and the shear direction. The description
of the failure behaviour is defined by traction-separation laws. These relations describe the
tractions as a function of separations and determine the constitutive behaviour of cohesive
zone models. There is a great variety in traction separation laws (Chandra et al., 2002) but
they all exhibit the same global behaviour. As the cohesive surfaces separate, the traction
first increases until a maximum is reached, and subsequently the traction decreases to zero,
4

max

traction T

eff < max


eff = max
0max
which results in complete
(local) separation. This holds for both the normal and the shear
1max
direction. A schematic
example of a traction-separation curve is shown in Figure 2.2.
n
t
p
2
2
eff = t + 2n
eff > max,old
yes
no
Loading
Unloading
Tn = eq(2.3)

Tt = eq(2.4)
Ti
i = eq(B.1-B.4)
Tn = eq(3.11)
Tt = eq(3.12)
Ti
displacement
i = eq(B.16-B.19)
Figure 2.2: Example of traction-separation
curve
There are a number of factors which play an important role in the resulting failure behaviour.
For instance, the area under the traction separation curve corresponds with the energy needed
for separation (). The initial stiffness of the cohesive zone model has a large influence on
the overall elastic deformation and should be very high in order to obtain realistic results.
It is shown by Chandra et al. (2002) that the form of the traction-separation relations plays
an important role in the macroscopic mechanical response of the system. The cohesive zone
model which is used in the following is the exponential model by Xu and Needleman (1994).
This model provides smooth traction-separation curves and may therefore be more stable
than discontinuous models, such as the bilinear model.

2.2

General theory

The traction vector T acting at the cohesive surface is derived from an interfacial potential
(Xu and Needleman, 1994) with normal and tangential components Tn and Tt , respectively:
T=

()
()

(2.1)

with
= (n , t )
The potential can be written as:



 




n
n 1 q
r q n
2t
(n , t ) = n +n exp
(2.2)
q+
1r+
exp 2
n
n r 1
r 1 n
t
5

=0
= +1
ip1
ip2

=0
= +1
ip1
ip2

1
3

1
3

T /

Tn / max

max

Wheren and t represent characteristic separations, in such a way that Tn (n ) = max and
ulus = 2000 [MPa]
E-modulus = 2000 [MPa]
T ( / 2) = max . max and max represent the maximum values of the normal traction and
h = 10 [mm]t t
h = 10 [mm]
0
0
the shear traction respectively. Furthermore, q = t /n and r = n /n , where n is the
l = 22.5 [mm]
l = 22.5 [mm]
value of n when complete shear separation has taken place without resulting in normal
hnotch = 2 [mm]
hnotch = 2 [mm]
tension (Tn = 0). q will be taken equal to one and r equal to zero. The resulting equations for
= 0.3
= 0.3
the normal and shear tractions are derived by
combining (2.1) and (2.2) with exp(...) = e (...)
n
n
as:
t
t








0
2
0max

n
n
t max1 q
n
2t
n
exp
exp eff <2 max
(2.3)
Tn =
1 exp 2
+
r
eff < max
n
n
n
n
t maxr 1
t
max










2t
r q n
n t
max n
max

q+
(2.4)
exp
exp 2
Tt = 2

r 1 n

t2
t
eff = max
eff = max n
1max
1max
t
t

and

are
the
areas
under
the
normal
traction-separation
curve and the shear traction
n
t
n
n
max
max
curve respectively. They represent the amount
of work needed for complete separation. This

can be seen for q = 1, r = 0, and assumingthat


max Tn = Tn (n , t = 0), Tt = Tt (n = 0, t ),
max

for which
tractions
are obtained. Using T n (n ) = max and
case the so-called uncoupled
eff < max

<

max
eff
Tt (t / 2) = max the following relations for n and t can then be obtained:
eff = max
= max
p eff
0max
0max
max t
n = max exp(1) n ,
t = exp(1)/2
(2.5)
1max
1max
The
normalized traction curves for uncouplednormal
separation and shear separation are
n
n

shown
in
Figure
2.3
and
Figure
2.4.
In
these
figures
Tn /max and Tt /max represent the
t
t
p
p
eff = 2 2t + 2n
eff = 2 2t + 2n
1
1
eff > max,old
eff > max,old
yes 0.5
yes 0.8
0.6
no
no
0
Loading
Loading 0.4
Unloading 0.5
Unloading 0.2
0
Tn = eq(2.3)
Tn = eq(2.3)
1
Tt = eq(2.4)
Tt = eq(2.4) 0.2
0.4
Ti
T
i
1.5
i = eq(B.1-B.4)
i = eq(B.1-B.4)
0.6
Tn = eq(3.11) 2
Tn = eq(3.11)
0.8
Tt = eq(3.12)
Tt = eq(3.12)
1
2.5
Ti
T
4
3
2
1
0
1
2
3
4
2
0
2
4
6 i
8
10
/
/
i = eq(B.16-B.19)
i = eq(B.16-B.19)
n

Figure 2.3: Normal traction curve for the


uncoupled modelling

Figure 2.4: Shear traction curve for the


uncoupled modelling

dimensionless normal and shear tractions. n /n and t /t represent the dimensionless


normal and shear openings. The normal traction-separation curve shows that starting from
an opening of zero and increasing the separation also increases the traction until a maximum
value is reached at n . After that the cohesive force decreases until the cohesive zone no
longer has any stiffness in normal direction. When the cohesive zone is given a separation
displacement in negative direction the traction rapidly becomes more negative in order to
6

n
t

eff

eff

n
t

0max
< max

eff

max

prevent
max

0max
< max

max

separation
max

penetration. The shear traction


curve does not show such a behaviour for
negative separations. Separations in negative direction merely lead to shear tractions in the
= max
eff = max
negative direction, which are opposite to
those1 for a positive t .
1max
max
t
t
was
already
mentioned,
figures
2.3
and
2.4
illustrate
the uncoupled relations. Next, the
As

n
n
max

max it holds that:


coupling
effect
will
be
illustrated,
as
for
this
case,

max

max
Tt = Tt (n , t )
(2.6)
eff < max
eff < max
eff = max
= max
The relations (2.3) and (2.4) correspondeffwith surfaces,
which are shown in figures 2.5 and 2.6.
0max
0max
Figure
2.5
shows
that
a
non-zero
value
of

results
in
a lower curve for the Tn (n ) relation.
t
1max
1
On the other hand, Figure 2.6 shows that a max
non-zero value of n results in a lower curve
n

for the Tt (t ) relation. For now, this potential nand its derivatives can only be used for two
t
t
p
p
eff = 2 2t + 2n
eff = 2 2t + 2n
eff > max,old
eff > max,old
1
1
yes
yes
0
no
no 0.5
Loading
Loading
1
0
Unloading
Unloading
Tn = eq(2.3) 2
Tn = eq(2.3) 0.5
Tt = eq(2.4) 3
Tt = eq(2.4)
Ti
Ti
1
4
6
i = eq(B.1-B.4)
i = eq(B.1-B.4)
2
4
Tn = eq(3.11)
Tn = eq(3.11)
0
2
6
3
2
4
0
2
Tt = eq(3.12)
T
t = eq(3.12)
1
2
0
1
0
2 3
2
4 2
/
Ti
Ti

/
/
/
i = eq(B.16-B.19)
i = eq(B.16-B.19)
Tt /max

Tn /max

Tn = Tn (n , t ) and

t t

n n

n n

t t

Figure 2.5: Normal traction surface

Figure 2.6: Shear traction surface

dimensions. This can be extended to three dimensions by adding a new tangential traction
to the set of equations. This new traction will be directed perpendicularly to the other two.
However, this will not be discussed here. For information regarding a three dimensional
implementation see Goncalves et al. (2000).

2.3

MSC.Marc user subroutines

The cohesive zone element has been implemented as a user element in the commercial finite
element package MSC.Marc. The user subroutines have been written by van den Bosch et al.
(2004). The cohesive zone element is defined as a four node element with two integration
points, lying on the line A-B. The location of the integration points depends on which
numerical integration scheme is chosen, see Figure 2.7. The definition of the element nodes
is important. The nodes must be assigned in counterclockwise direction, the last node (4)
must be opposite to the first node (1) and the third node must be opposite to the second node.

eff = max
1max
t
n
max

max

eff < max


eff = max
0max
1max
n
t
p
2 2t + 2n
ff =
eff > max,old
yes
no
Loading
Unloading
Tn = eq(2.3)
Tt = eq(2.4)
Ti
i = eq(B.1-B.4)
ip1
Tn = eq(3.11) A
Tt = eq(3.12) A
= eq(B.16-B.19) = 1

3
~ey
~en

4
A

~ex

~et

1
3

ip2

B
B

=0

Gauss
NewtonCotes

= +1

Figure 2.7: Cohesive zone element, shown with nodes and length AB and the location of
integration points in the Gauss and Newton-Cotes schemes
The main subroutine is uselem, this subroutine is called by MSC.Marc every time the element stiffness matrix and the internal force column is needed for the calculation. For this
purpose uselem exports the current nodal coordinates and the total nodal displacements to
czbehav. With these quantities czbehav calculates the separations n and t . First the current coordinates are used to calculate the length and the orientation of the center line AB,
see Figure 2.7. Next, the locations of the integration points are determined with respect to
A-B. After that interpolation of the nodal displacements is used to determine n and t at
the integration points. The element stiffness matrix is calculated in uselem using
Z
l +1 T
e
K =
N DNd
(2.7)
2 1
with D for the uncoupled case

Tt
0
t

D=
Tn
0
n

(2.8)

and the internal force column is calculated using


Z
l +1 T
e
N Td
fint
=
2 1
8

(2.9)

with
T=

Tt
Tn

(2.10)

D is the consistent tangent stiffness matrix, l is the length (width) of the element and N
is the matrix of shape functions. T is the column of calculated tractions. The tractions are
calculated according to equations (2.3) and (2.4).
The following subroutines are used by MSC.Marc
uselem : This is the main subroutine, the element stiffness matrix and internal forces
are calculated here.
czbehav : The constitutive behaviour of the cohesive zone element is implemented in this
subroutine. It is called by uselem for the actual calculation of the stiffness matrix and the
internal force column. This subroutine contains the traction separation characteristics
and stiffnesses as well as nodal transformations and a declaration of the integration
points.
plotv : This subroutine writes the status of the cohesive zone elements to the marc
postprocessing file, it can also be used to write different quantities to file if needed.
elevar : This subroutine writes the openings n and t and the tractions Tn and Tt
to a file as well as desired additional quantities.

= 1
=0
= +1
ip1
ip2

= 1
=0
= +1
ip1
ip2

ulus = 2000 [MPa]


h = 10 [mm]
l = 22.5 [mm]
hnotch = 2 [mm]
= 0.3
n
t
0max
eff < max

E-modulus = 2000 [MPa]


h = 10 [mm]
l = 22.5 [mm]
hnotch = 2 [mm]
= 0.3
n
t
0max
eff < max

1
3

1
3

Chapter 3

Implementation issues

max

max

max

max

eff = max
eff = max
1max
1max
3.1 Numerical integration
t
t
n
n
A high initial stiffness of the cohesive zone elements
max
max is necessary when trying to obtain a phys

ically realistic model. When the initial stiffness


is too low additional deformation will occur,
max
max

caused by the decreased overall (bulk) stiffness.


Therefore the stiffness must be sufficiently
eff < max
eff <inso-called
max
high. However, a high stiffness can result
spurious oscillations. This means that
eff = max

max
eff
the traction profile exhibits an oscillatory behaviour that has no physical meaning. Research
0max
0max
by Schellekens and de Borst (1993) has shown
that these oscillations are caused by the com1max
1max
bination of a high traction gradient within one
element and a Gauss integration scheme.
This

n
can be illustrated by means of a test case nwhich was also devised by Schellekens and
t
t
p
p
eff = 2 2t + 2n
eff = 2 2t + 2n
eff > max,old
eff > max,old
yes
yes
no
no
Loading
Loading
Unloading
Unloading
Tn = eq(2.3)
Tn = eq(2.3)
cohesive zone elements
Tt = eq(2.4)
Tt = eq(2.4)
h
Ti
Ti
=
eq(B.1-B.4)
=
eq(B.1-B.4)
i
i
Tn = eq(3.11)
Tn = eq(3.11)
d
Tt = eq(3.12)
Tt = eq(3.12)
Ti
Ti
i = eq(B.16-B.19)
i = eq(B.16-B.19)
L
Y

Figure 3.1: Test case set up

Figure 3.2: Test case mesh

de Borst (1993), see Figure 3.1 and Figure 3.2. The beam has a length L of 25[mm] and
a height h of 5[mm]. An initial crack is inserted at the bottom of the beam with length
d = 1[mm]. The cohesive zones are placed directly above this crack. The bulk material has a
Youngs modulus of 2000[MPa] and a Poissons ratio of 0.3. Five nodes which lie in the middle
directly above the centerline are given a very small displacement downward, unlike the test
case by Schellekens and de Borst (1993) in which a downward pressure was applied. This
10

height [mm]

height [mm]

eff = max
eff = max
1max
1max
t
t
n
n
max
max
displacement is kept small since the normal
opening
of the cohesive zone elements needs to

max

max
be
kept
very
small
(

)
in
order
to
investigate
the
influence of the initial stiffness. Two
n t


situations are simulated, using cohesive zone elements with a low stiffness (max = 1 [MPa],
eff < max
eff < max
n = t = 0.1[mm]) and with a high stiffness
(max = 1000 [MPa], n = t = 1E-4[mm]). Paeff = max
eff = max
rameters q and r are taken one and zero respectively.
Figures 3.3 and 3.4 show the computed
0max
0
normal tractions in the cohesive zones. Figure max
3.3 shows no oscillations in the traction profile,
1
1
max
max
but Figure 3.4 does show oscillations.
n
n

t
t
p
p
eff = 2 2t + 2n
eff = 2 2t + 2n
5
5
eff > max,old
eff > max,old
yes 4.5
yes 4.5
4
4
no
no
Loading 3.5
Loading 3.5
Unloading
Unloading
3
3
Tn = eq(2.3)
Tn = eq(2.3)
Tt = eq(2.4) 2.5
Tt = eq(2.4) 2.5
Ti
Ti
2
2
i = eq(B.1-B.4)
i = eq(B.1-B.4)
Tn = eq(3.11) 1.5
Tn = eq(3.11) 1.5
Tt = eq(3.12)
Tt = eq(3.12)
1
1
Ti
T0.05
0.1 0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
0.2
0.15
0.1
0.05
0
0.1
0.15
i
T [MPa]
T [MPa]
i = eq(B.16-B.19)
i = eq(B.16-B.19)
n

Figure 3.3: Traction curve using Gauss integration and max = 1 [MPa]

Figure 3.4: Traction curve using Gauss integration and max = 1000 [MPa]

To plot the traction profiles use was made of the integration point values, to avoid possible
inaccuracies due to the extrapolation of integration point values to nodal point values. A
possible way to remove the spurious oscillations is by using a different integration scheme
instead of Gauss. Schellekens and de Borst (1993) have shown that the Newton-Cotes
integration scheme does not suffer from this problem. This integration scheme has a first
order accuracy instead of the second order accuracy which is obtained when using the Gauss
scheme. However, if the occurrence of spurious oscillations has a significant (long term)
effect, it may be wise to nevertheless use the Newton-Cotes scheme instead of the Gauss
scheme. The only parameters that need to be changed in the subroutine are the locations of
the integration points since the weight factors are the same for two point Gauss integration
and Newton-Cotes integration. Figures 3.5 and 3.6 show the results for the same simulation
but using the Newton-Cotes integration scheme. These figures show that the use of the
Newton-Cotes integration scheme does not result in spurious oscillations in the traction
profile.

11

eff

5
4.5
4
3.5
3
2.5
2
1.5
1
0.2

0.15

0.1

0.05
Tn

max

ip2
0max
1
1max3
E-modulus = 2000 [MPa]
n
h = 10 [mm]
t
p
[mm]
eff = l = 222.5
2t +
2n
hnotch
[mm]
2max,old
eff >=
= yes
0.3

non

Loadingt
0max
Unloading
max
Tneff=<eq(2.3)
max
Tt = eq(2.4)

Ti
max
=
eq(B.1-B.4)

Tn=
eq(3.11)
=

max
eff
Tt = eq(3.12)
1max
T0.05
0
0.1
0.15
i
[MPa]
t
i = eq(B.16-B.19)
n

5
4.5
4

height [mm]

max

ip2
0max
1
1max3
ulus = 2000 [MPa]
n
h = 10 [mm]
t
p
[mm]
eff = l = 222.5
2t +
2n
hnotch
[mm]
2max,old
eff >=
= yes
0.3

non

Loadingt
0max
Unloading
max
Tneff=<eq(2.3)
max
Tt = eq(2.4)

Ti
max
=
eq(B.1-B.4)

Tn=
eq(3.11)
=

max
eff
Tt = eq(3.12)
1max
Ti
t
i = eq(B.16-B.19)
n

height [mm]

eff

3.5
3
2.5
2
1.5
1
0.15

0.1

0.05

0.05
0.1
Tn [MPa]

0.15

0.2

0.25

0.3

max
max

Figure
3.5:
Traction
curve
using
Newtonmax Figure 3.6: Traction curve using Newton max

Cotes integration and


Cotes integration and
max = 1 [MPa]
max = 1000 [MPa]
eff < max
eff < max

height [mm]

height [mm]

eff = max
eff = max
0
Another
way
to
avoid
the
oscillations
is to use
mesh refinement, since this inherently lowers
max
0max
1
1
the
traction
gradient
over
one
element
when
applying
the same load. The result is shown
max
max
Figure 3.7 using twice the amount of cohesive
in
n zones. This figure shows that the spurious
n
oscillations
are
not
present
when
mesh
refinement
t
t is used.
p
p
2
2
2
2
2
eff = t + n 10
eff = t + 2n 10
eff > max,old
eff > max,old
9
9
yes
yes
no 8
no 8
Loading 7
Loading 7
Unloading 6
Unloading 6
Tn = eq(2.3)
Tn = eq(2.3)
5
Tt = eq(2.4)
Tt = eq(2.4) 5
Ti
Ti
4
4
i = eq(B.1-B.4)
i = eq(B.1-B.4)
Tn = eq(3.11) 3
Tn = eq(3.11) 3
Tt = eq(3.12) 2
Tt = eq(3.12) 2
Ti
4
2
0
2
4
6 Ti 8
10
12
4
2
0
2
4
6
8
10
12
=
eq(B.16-B.19)
=
eq(B.16-B.19)
Tn [MPa]
Tn [MPa]
i
i
Figure 3.7: Traction profile using a fine
mesh, Gauss integration and max = 1000
[MPa]

Figure 3.8: Traction profile using a


fine mesh, Newton-Cotes integration and
max = 1000 [MPa]

The cause of the spurious oscillations seems to lie in the eigenmodes which an element
has when using Gauss integration as opposed to Newton-Cotes integration (Schellekens and
de Borst, 1993). The different eigenmodes (except for the three rigid body modes and the
zero-energy mode which represents extension of the cohesive zone along its length) for the
Newton-Cotes integration scheme are shown in Figure 3.9. These eigenmodes have been obtained by computing the eigenvalues and eigenvectors of the element stiffness matrix. This is
12


max

max

eff < max


eff = max
0max
1max
n
t
p
eff = 2 2t + 2n
eff > max,old
yes
no
Loading
Unloading
Tn = eq(2.3)
Tt = eq(2.4)
Ti
=
eq(B.1-B.4)
i
Tn = eq(3.11)
Tt = eq(3.12)
Ti
i = eq(B.16-B.19)

3
4
1
2

1.

= 2.3316

3.

= 4.6632

eff < max


eff = max
0max
1max
n
t
p
eff = 2 2t + 2n
eff > max,old
yes
no
2. = 2.3316
Loading
Unloading
Tn = eq(2.3)
Tt = eq(2.4)
Ti
=
eq(B.1-B.4)
i
Tn = eq(3.11)
Tt = eq(3.12)
Ti
eq(B.16-B.19)
4.
i ==4.6632

Figure 3.9: Eigenmodes and eigenvalues


with Newton-Cotes integration

4
3
2
1
1.

= 0.7772

2.

= 1.5544

3.

= 2.3316

4.

= 4.6632

Figure 3.10: Eigenmodes and eigenvalues


with Gauss integration

done for a given combination of n and t , using Gauss integration and Newton-Cotes integration. The problem lies in the coupling between degrees-of-freedom of the different nodes.
This is illustrated by Figure 3.10, which shows the eigenmodes and eigenvalues using a Gauss
integration scheme. When using Gauss integration a single nodal degree-of-freedom can be activated in different eigenmodes, this is not the case when using the Newton-Cotes integration
scheme. However it is not completely clear why this would result in spurious oscillations. An
alternative explanation, which is also not completely rigorous is that for Gauss integration
there is an interaction between the nodes at the left end of the element and those at the
right end, whereas for the Newton-Cotes integration the openings at both ends can be varied
independently.

3.2

History dependency

The current implementation does not exhibit history dependent behaviour or irreversible behaviour. When unloading, the same traction curve is followed as during loading. This is shown
in Figure 3.11. This implies that to achieve unloading one must increase the traction. This is
not realistic since damage is regarded as an irreversible process. It is assumed that in order
to achieve realistic irreversible behaviour, unloading should occur in a linear way to the origin.
To this end, one or more history parameters will be introduced in the model. This can be
done for both the normal and the shear direction or for the both of them in a coupled fashion.
In the following, three different forms of history dependency are implemented and tested.
First an uncoupled history dependent model will be implemented and tested. Following
the same framework, coupled history dependent behaviour will be implemented. Finally a
different approach will be used to achieve coupled history dependent behaviour.
13

1
0.9
0.8

max

0.7

Tn /

eff = max
0max
1max
n
t
p
eff = 2 2t + 2n
eff > max,old
yes
no
Loading
Unloading
Tn = eq(2.3)
Tt = eq(2.4)
Ti
=
eq(B.1-B.4)
i
Tn = eq(3.11)
Tt = eq(3.12)
Ti
=
eq(B.16-B.19)
i

0.6
0.5
0.4
0.3
0.2
0.1
0
0

n / n

Figure 3.11: Current implementation, subsequent loading and unloading shows reversibility

3.2.1

Uncoupled irreversibility

To obtain history dependent behaviour a history parameter needs to be introduced. This


parameter is chosen as
i = max{ i ( ) | 0 t}

with i = n,t

(3.1)

and
n =

hn i
,
nc

t =

|t |
tc

hn i =

0
if
n if

n 0
n > 0

(3.2)

(3.3)

n is defined as the history parameter for separations in normal direction and t for separations in tangential direction. nc and tc are critical separations in normal and shear direction
respectively. These separations are chosen as the separations where the corresponding
tractions are almost zero. Since the traction curves approach zero very slowly for larger
separations, the exact value of these parameters is not very important.
Damage evolution is already accounted for in the free energy potential . Since the tractions
are derived from this potential, damage evolution is also accounted for in the traction relations. This damage evolution can be seen in the traction curves as softening behaviour. There
is no need to degrade the tractions with some sort of history dependent damage parameter
14

max

max

t
n

t
n

max

max

max

max

for continued loading. It is only necessary


whether the cohesive zone is loading
eff < max
eff <todetermine
max
or
unloading
and
in
what
direction
this
takes
place.
Since
the cohesive zone element should
eff = max
eff = max
0
0
experience a different behaviour for unloading
than for loading, the traction-separation equamax
max
1
1
tions
should
be
extended
with
an
unloading
mode.
max
max This extension is made for the normal and
responses separately and should result
tangential
n in the following behaviour. When loading
n

t
p
eff = 2 2t + 2n
eff > max,old
yes
no
Loading
Unloading
Tn = eq(2.3)
Tt = eq(2.4)
Ti
i = eq(B.1-B.4)
Tn = eq(3.11)
Tt = eq(3.12)
Ti
i = eq(B.16-B.19)

t
t =

|t |
ct

yes

t > t,old
no
Unloading
t = t,old
Tt = t t
Tt
t = t

t
p
eff = 2 2t + 2n
eff > max,old
yes
no
Loading
Loading
Unloading
t = t
Tt = eq(2.4)
Tn = eq(2.3)
Tt
t = eq(B.4)
Tt = eq(2.4)
Tt
t =
T
i
t
i = eq(B.1-B.4)
Tn = eq(3.11)
Tt = eq(3.12)
Ti
i = eq(B.16-B.19)

Figure 3.12: Flowchart for tangential loading

n
n =

Loading
n = n
Tn = eq(2.3)
Tn
n = eq(B.1)
Tn
n =
n

yes

hn i
cn

n > n,old
no
Unloading
n = n,old
Tn = n n
Tn
n

= n

Figure 3.13: Flowchart for normal loading

takes place, the original relations, (2.3) and (2.4), should be used to calculate the tractions.
When unloading takes place in normal or tangential direction the last calculation on the curve
during the loading process for this direction determines the unloading stiffness n for normal
direction and t for tangential direction. When the element is subsequently loaded, these
unloading relations should then be used until the previous maximum value is again reached,
this is the reload path.
i =

Ti,max
i,max

with i = n,t

(3.4)

The traction-separation relations in case of unloading are hereby simplified to


Tn = n n

Tt = t t

(3.5)

This is illustrated in Figure 3.12 for the tangential direction and in Figure 3.13 for the
normal direction. Note again that the two directions are treated in a completely separate way.
The simulations are displacement controlled. The test case consists of two material elements
with one cohesive zone element in between. The results are obtained by respectively loading,
unloading and after that increased loading, see Figure 3.14. The same has been done for the
shear direction which was respectively loaded, unloaded, loaded in opposite direction, and
unloaded, see Figure 3.15. The results are shown in figures 3.16 and 3.17.

15

0max
=
0.3
n
1max
nt

0max
t
p
2
<2t
eff = eff
+ max
2n
max
eff > max,old

max
yes

no
eff = max
Loading
1max
Unloading
t
Tn = eq(2.3)
n
max
Tt = eq(2.4)

Ti
i = eq(B.1-B.4)
max

Tn = eq(3.11)
eff < max
Tt =
eq(3.12)
eff = max 1. Normal loading/
Ti
unloading
i = eq(B.16-B.19)
0

0max
=
0.3
n
1max
nt

0max
t
p
2
<2t
eff = eff
+ max
2n
max
eff > max,old

max
yes

no
eff = max
Loading
1max
Unloading
t
Tn = eq(2.3)
n
max
Tt = eq(2.4)

Ti
i = eq(B.1-B.4)
max

Tn = eq(3.11)
eff < max
Tt =
eq(3.12)
eff = max 1.Shear loading/
2. Normal
loading
Ti
unloading
i = eq(B.16-B.19)
0

max

max

2.Shear loading/
unloading

1max
1max
n Figure 3.14: Prescribed normal displace- n Figure 3.15: Prescribed shear displacet ments
t ments
p
p
eff = 2 2t + 2n
eff = 2 2t + 2n
eff > max,old 60
eff > max,old 6000

yes 50
yes 4000
T

no 40
no
Loading
Loading 2000
Unloading 30
Unloading
0
Tn = eq(2.3) 20
Tn = eq(2.3)
Tt = eq(2.4)
Tt = eq(2.4) 2000
Ti
Ti
10
i = eq(B.1-B.4)
i = eq(B.1-B.4)

Tn = eq(3.11) 0
Tn = eq(3.11) 4000
Tt = eq(3.12)
Tt = eq(3.12) 6000
10
2
1.5
1
0.5
0
0.5
1
Ti
0.01
0
0.01
0.02 Ti
0.03
0.04
= eq(B.16-B.19)
displacement X
displacement Y i
x 10
i = eq(B.16-B.19)
max

n,max

comp 22 of stress

comp 11 of stress

max

Figure 3.16: Uncoupled irreversibility:


Normal traction curve

3.2.2

Figure 3.17: Uncoupled irreversibility:


Shear traction curve

Coupled irreversibility

Introduction
The previous implementation of history dependent behaviour was done in an uncoupled way,
i.e. the evolution of the history parameter n did not have an influence on the traction behaviour in tangential direction and vice versa. This implies that a cohesive zone element which
is loaded in normal direction until it is damaged, and subsequently unloaded, still has the
initial stiffness in tangential direction. This is not very realistic for most materials and we
therefore attempt to couple the history dependent traction-separation behaviour. To achieve
the coupling, the complete, coupled traction-separation relations need to be used for loading
and the tangential stiffness matrix should also be extended. The original tangent stiffness
matrix only contained the diagonal terms. The new consistent tangent stiffness matrix D for

16

loading also contains the cross-derivatives:


Tt Tt
D=

Tn
t

Tn
n

(3.6)

The expressions in D are given in appendix B en equations (A.1)-(A.4).


One effective history parameter
Based on the work of Ortiz and Pandolfi (1999), irreversible behaviour is implemented using
a single history parameter max instead of n and t . This parameter is defined as
max = max{ eff ( ) | 0 t}

(3.7)

with eff the effective opening displacement


q
eff = 2 2t + 2n
with = (0, )

(3.8)

eff 0, and unloading (or reloading) when


Loading takes place when eff = max and
eff < max . When loading occurs, the tractions from equations (2.3) and (2.4) and the
corresponding stiffnesses are used. For unloading, the following relations for T n and Tt are
used
Tn =

Tn,max
Tn,max
n +
t
n,max
t,max

(3.9)

Tt =

Tt,max
Tt,max
n +
t
n,max
t,max

(3.10)

The unloading stiffnesses used in equations (3.9) and (3.10), Ti,max /j,max where i,j = n,t,
are evaluated for the maximum values of n and t reached during the previous loading.
However, when a cohesive zone element is subjected to mixed mode loading, the tractions
during reloading are not calculated correctly. If this reloading takes place in same other
direction than the unloading, however, these stiffnesses do not correspond with the undamaged
loading response in this direction. As a consequence, a jump in the tractions may occur at
the transition from the unloading/reloading relations to the loading relations (2.3)-(2.4), see
Figure 3.18. This does not seem very realistic; it seems more realistic that the normal traction
follows the dashed line instead. For this reason, a different approach is needed to calculate
the tractions during unloading and reloading.
To calculate the tractions during reloading correctly, the reloading stiffnesses should be take
into account the new (re-)loading direction and the loading response in this direction. For
this purpose the tractions are interpolated between the origin and the tractions at which the
loading surface eff = max is reached in the new loading direction. This is shown in Figure
3.19. This figure shows the damage surface given by eff = max in the displacement space.
This surface is somewhat similar to the yield surface for plasticity in a stress space. The surface
eff > 0. The idea is now that for
of max can only grow, this occurs when eff = max and
a given (n , t ) for which eff < max the tractions are linearly interpolated between the
17

n
max

max

t
max

40
35
30

25
T

eff < max


eff = max
0max
1max
n
t
p
eff = 2 2t + 2n
eff > max,old
yes
no
Loading
Unloading
Tn = eq(2.3)
Tt = eq(2.4)
Ti
=
eq(B.1-B.4)
i
Tn = eq(3.11)
Tt = eq(3.12)
Ti
i = eq(B.16-B.19)

20
15
10
5
0
0

0.5

1
n

n
t
p
eff = 2 2t + 2n
eff > max,old
yes
no
Loading
Unloading
Tn = eq(2.3)
Tt = eq(2.4)
Ti
=
eq(B.1-B.4)
i
Tn = eq(3.11)
Tt = eq(3.12)
2
Ti 1.5
x 10
i = eq(B.16-B.19)
3

Figure 3.18: Current mixed mode loading


result

eff = max

eff < max


0max

1max

max

Figure 3.19: Damage surface in displacement space

origin and the tractions at which the damage surface will be reached for continued loading in
the same direction (in the n t space). This effectively means that the openings n and
t are first scaled by a factor max /eff , the tractions associated with these scaled openings
are computed, and these tractions are scaled back by multiplying them by eff /max . The
resulting tractions can thus be written as


eff
max
max
unl
(3.11)
Tn =
Tn
n ,
t
max
eff
eff


eff
max
max
unl
Tt =
(3.12)
Tt
n ,
t
max
eff
eff
where Tn (n , t ) and Tt (n , t ) are the traction-separation relations used for loading as
given by (2.3) and (2.4). Note that this approach also works properly for nonuniform n t
paths, since the intersection with the loading surface is automatically updated for such paths.
This implementation basically works the same as the uncoupled implementation, with max
as history parameter instead of i (i = n or t). Figure 3.20 shows the resulting flowchart for this
approach. The tangent stiffnesses associated with (3.11) and (3.12), which are to be used in
unloading/reloading, are given in Appendix B, equations (A.14)-(A.17). This implementation
has been tested for mixed mode loading, for which the prescribed displacements are shown
in Figure 3.21. The element is first loaded and unloaded in both normal and shear direction,
with the rate of normal displacement equal to the rate of shear displacement. Subsequently, a
mixed mode load is applied with the rate of shear displacement being larger than the rate of
normal displacement. The resulting tractions are shown in figures 3.22 and 3.23. The traction
jump which is seen in Figure 3.18 is absent, the traction curves appear to have been improved.

18


Tn > tmax
nn

Tn = t
n
max

= 1
eff < =
0
max
eff =
max
=
+1
0max
ip1
1max
ip2

Tn > tmax
nn

Tn = t
n
max

= 1
eff < =
0
max
eff =
max
=
+1
0max
ip1
1max
ip2
1
1n

3
3
t
p
ulus = 2000 [MPa]
E-modulus
=
2000
[MPa]
n
t
2 2 + 2
eff = h=
n 6 x 10
t [mm]
h = 10 [mm]
10

>

eff
max,old
p
l = 22.5 [mm]
l = 22.5 [mm]
2 2
2
yes 5
hnotch = 2 [mm] eff = t + n
hnotch = 2 [mm]
no
= 0.3
= 0.3
4

Loading
n
n

Unloading
no
eff > max,old
Unloading
t
t 3
Tn = eq(3.11)
Tn = eq(2.3)
0max
0max
yes
Tt = eq(3.12)
T
=
eq(2.4)
T
eff < max
teff < max 2
Teq(B.16-B.19)
=
i
Loading
max
max
=
eq(B.1-B.4)
i

1
T
=
eq(2.3)
n
max
Tn = eq(3.11)
max

Tt = eq(2.4)
Tt = eq(3.12) 0
eff = max
eff = max 0
T
500
1000
1500
Ti
= eq(B.1-B.4)

increment []
i = eq(B.16-B.19)
1max
1max
t
t
n Figure 3.20: Flowchart for coupled imple- n Figure 3.21: Prescribed displacements for
max
max
mentation using one history parameter
mixed mode loading
max
max

displacement [mm]

i
i

i
i

35
30

25
20
15
10
5
0
0

0.5

1
n

eff < max


eff = max
0max
1max
n
t
p
eff = 2 2t + 2n
eff > max,old
yes
no
Loading
Unloading
Tn = eq(2.3)
Tt = eq(2.4)
Ti
i = eq(B.1-B.4)
Tn = eq(3.11)
Tt = eq(3.12)
2
Ti 1.5
x 10
i = eq(B.16-B.19)
3

Figure 3.22: Normal tractions for mixed


mode loading

120

100

80

Tt

40

eff < max


eff = max
0max
1max
n
t
p
eff = 2 2t + 2n
eff > max,old
yes
no
Loading
Unloading
Tn = eq(2.3)
Tt = eq(2.4)
Ti
i = eq(B.1-B.4)
Tn = eq(3.11)
Tt = eq(3.12)
Ti
i = eq(B.16-B.19)

60

40

20

0
0

5
3

x 10

Figure 3.23: Shear tractions for mixed


mode loading

19

Tn = t n
= 1
=0
= +1
ip1
ip2

Tn = t n
= 1
=0
= +1
ip1
ip2

ulus = 2000 [MPa]


h = 10 [mm]
l = 22.5 [mm]
hnotch = 2 [mm]
= 0.3
n
t
0max
eff < max

E-modulus = 2000 [MPa]


h = 10 [mm]
l = 22.5 [mm]
hnotch = 2 [mm]
= 0.3
n
t
0max
eff < max

1
3

1
3

Chapter 4

Benchmark test

max

max

max

max

eff = max
eff = max
In order to obtain quantitative data on the
occurring stresses during the delamination
1max
1max
process a benchmark test has been developed by Davies (2002). This benchmark test
t
t
involves a double cantilever beam (DCB) which is pulled apart in mode I. The benchmark
n
n
problem is shown in Figure 4.1. The ends ofmaxthe beam are given a displacement u/2 in
max

the y-direction, while the other end of themax


beam
is entirely fixed in x and y direction.
max

The cohesive zones are placed on the dotted line in this figure. When loading the beam
eff < max
eff <
in this mode the top and bottom parts
ofmax
the beam are first bent over the 30[mm]
eff = max

max
eff
long initial crack. Subsequently, the cohesive zones start to open, which results in a
0max
0max
decrease of the stiffness. When the cohesive
zones are all completely opened the beam
1
max
1maxlength of 100[mm]. The mesh of the final
starts to bend again, this time over the total
result
n are available to verify the numerical results.
n
is shown in Figure 4.2. Analytical relations

t
t
p
p
eff = 2 2t + 2n
eff = 2 2t + 2n
eff > max,old
eff > max,old
yes
yes
no
no
Loading
Loading
Unloading
Unloading
Tn = eq(2.3)
Tn = eq(2.3)
Tt = eq(2.4)
Tt = eq(2.4)
Ti
Ti
i = eq(B.1-B.4)
i = eq(B.1-B.4)
depth = 30 mm
Tn = eq(3.11)u/2
Tn = eq(3.11)
Tt = eq(3.12)
Tt =3 mm
eq(3.12)
u/2
30 mm
Ti
Ti
100 mm
i = eq(B.16-B.19)
i = eq(B.16-B.19)
Inc: 100

Time: 1.000e+000

lcase1

Figure 4.1: Benchmark test set up

Figure 4.2: Benchmark deformed mesh

The analytical relations are obtained by combining the linear elastic beam theory with linear
elastic fracture mechanics. The results are
3uEI
for initial bending
(4.1)
P =
2a30
20

Tn
n

= n
t 6= 0
Tn > t n
Tn = t n
s = 1
3/2
2(bGIc=
EI)
0
P =
for delamination
3
2a+1
=
0
ip1
with
ip2
1 on beam edge
P = vertical force
3

(4.2)

E-modulus = 2000 [MPa]


b = beam thickness
h = 10 [mm]
GIc =
release rate for mode I
l =energy
22.5 [mm]
h
=
2
[mm]
E =notch
elasticity modulus
= 0.3
I = moment ofinertia
n
t
a0 = length of
initial
crack
0max
eff < that
The two equations
max determine the resulting force-displacement curve are equations (4.1)
and (4.2). Equationmax
(4.1) is derived from classic beam theory and relates the (perpendicular)

max
displacement (u)to
the bending force (P ) . A similar relation for the final state of the

complete opening
can be obtained by replacing a0 by the length of the beam. Equation (4.2)
eff =
max
relates the energyrelease
rate to the forces needed for delamination. This energy release
1
max
rate corresponds with
tn and is therefore a parameter in the constitutive behaviour of the
cohesive zone elements.
n This relation describes the decreasing force when the cohesive zone
max
elements are opening.

P [N]

max

The load P as a function of the displacement u/2 is shown in Figure 4.3. The dashed lines
eff < max
correspond with the analytical solutions as given above. The numerical solution follows the
eff = max
analytical relations
quite closely. Figure 4.3 shows that the beam regains some stiffness
0max
when the opening
exceeds
8[mm], this point corresponds with the beam bending purely over
1
max
100[mm] (full length). The results obtained by Alfano and Crisfield (2001) show a less than
n
t 140
p
eff = 2 2t + 2n
eff > max,old 120
yes 100
no
Loading 80
Unloading
Tn = eq(2.3) 60
Tt = eq(2.4) 40
Ti
i = eq(B.1-B.4)
Tn = eq(3.11) 20
Tt = eq(3.12)
0
Ti
0
5
10
15
u/2 [mm]
i = eq(B.16-B.19)
Figure 4.3: Numerical and analytical results for
benchmark test
smooth curve for the delamination part with many oscillations. This may be the result of the
21

max

eff = max
1max
t
use of a bilinear constitutive
model. The oscillations were reduced by making the cohesive
n
max is achieved by lowering
zone more ductile, this
max (while keeping the energy GIc the same).

max

The coupled
irreversible
eff
< max implementation should be able to describe the opening and closing
of the beam.
effHowever,
= max during the initial phase of the opening, there are areas which are
compressed. This can
be seen in Figure 4.4, with the dark grey areas representing a downward
0max
1
displacement and the
light grey areas an upward displacement. These areas consist of multiple
max
n
t
p
eff = 2 2t + 2n
eff > max,old
yes
no
Loading
Unloading
Tn = eq(2.3)
Tt = eq(2.4)
Ti
=
eq(B.1-B.4)
i
Tn = eq(3.11)
Tt = eq(3.12)
Ti
=
eq(B.16-B.19)
i

Figure 4.4: Areas of compression during the initial opening


cohesive zone elements. They are not a numerical artifact, but reflect a wave-like theoretical
solution, cf. a beam on an elastic foundation. Because the elements are compressed before
they are being pulled apart, eff = 0 at some point and max > 0. This leads to numerical
difficulties, the tractions and the stiffnesses cannot be calculated for eff = 0. An extra
statement has been added to the unloading implementation to make sure that for eff 0,
the tractions are zero and the stiffnesses are calculated using the relations for loading. The
results of subsequently opening and closing the double cantilever beam are shown in Figure
4.5. The figure shows that the unloading also takes place in a linear fashion to the origin,
which seems realistic.

22

h = 10 [mm]
l = 22.5 [mm]
hnotch = 2 [mm]
= 0.3
n
t
0max
eff < max
max

max

eff = max
1max
t
n
max

max

100
90
80
70
60
P [N]

eff < max


eff = max
0max
1max
n
t
p
eff = 2 2t + 2n
eff > max,old
yes
no
Loading
Unloading
Tn = eq(2.3)
Tt = eq(2.4)
Ti
=
eq(B.1-B.4)
i
Tn = eq(3.11)
Tt = eq(3.12)
Ti
=
eq(B.16-B.19)
i

50
40
30
20
10
0
0

u/2 [mm]

Figure 4.5: Benchmark results for subsequent opening and closing


of the double cantilever beam

23

Chapter 5

Conclusions and recommendations


Conclusions
Spurious oscillations: The investigation into the spurious oscillations did not have the
desired depth. It is not yet clear what the exact cause of these oscillations is, other
than it has to do with coupled eigenmodes. However, this behaviour does not need to
be a problem, since it can be avoided by using mesh refinement or a Newton-Cotes
integration scheme.
Irreversibility: The uncoupled and the first attempt at coupled implementation of irreversibility have their shortcomings. The first implementation exhibits no coupling
between the normal and the tangential direction. This implementation is therefore not
very useful when coupled loads are applied, such as a mixed mode load. Neither is it
very useful when subsequent loading is applied in different directions. This model has
been modified to show coupled behaviour, by rigorously coupling the history parameters
to each other. This implementation is able to describe subsequent loading, but it is too
instable to apply a complex load or to apply a load to multiple cohesive zone elements,
such as the benchmark problem. This is due to the extensive use of several criteria
and statements. The final coupled implementation is able to describe both subsequent
and mixed mode loading, because of the interpolation between the maximum effective
opening and the origin. This results in a model in which the unloading path can vary,
but the tractions and the stiffnesses can still be calculated.
Benchmark test: The model seems to describe the analytical solution quite well. The
computed force-displacement curve lies very close to the analytical solution. Subsequently opening and closing of the beam leads to the desired result, since unloading is
linear to the origin.

Recommendations
There are some issues that still deserve attention regarding the final implementation, as one
can not yet call the model robust. For instance, it is debatable whether damage should grow

24

for compression as it is assumed here. To avoid this, the effective opening can be made independent of n when the cohesive zone is compressed. Furthermore, a three dimensional
implementation of irreversible behaviour would be useful. This will allow more realistic simulations than is possible at the moment. Regarding a three dimensional implementation, it
is even more important that the model is able to describe irreversible behaviour in a robust
way, since the mesh becomes increasingly complex. Finally, it may be useful to investigate to
possibility to develop different constitutive relations. Since the current model already exhibits
damage evolution, and therefore damage can not be described directly, it may be worthwhile
to try to develop a model which is not based on the Xu and Needleman model. Such a model
could use an energy potential to describe the elastic response alongside another potential
which describes damage growth, similar to continuum damage mechanics. Furthermore, the
influence of the element size may be investigated, the element width should be less than one
tenth of the characteristic opening, according to Tomar et al. (2004).

25

Bibliography
Alfano, G. and Crisfield, M.A. (2001). Finite element interface models for the delamination analysis of laminated composites: mechanical and computational issues. International
Journal for Numerical Methods in Engineering, 50, 17011736.
Barenblatt, G. (1962). The mathematical theory of equilibrium cracks in brittle fracture.
Advances in Applied Mechanics, 7, 55129.
van den Bosch, M.J., Schreurs, P.J.G. and Geers, M.G.D. (2004). Background and implementation of a cohesive zone element. Internal report.
Chandra, N., Li, H., Shet, C. and Ghonem, H. (2002). Some issues in the application of
cohesive zone models for metal-ceramic interfaces. International Journal of Solids and
Structures, 39, 28272855.
Davies, G.A.O. (2002).
R0084 .

Benchmarks for composite delamination.

NAFEMS publication

Dugdale, D. (1960). Yielding of steel sheets containing slits. Journal of the Mechanics and
Physics of Solids, 8, 100104.
Goncalves, J., de Moura, M., de Castro, P. and Marques, A. (2000). Interface element including point-to-surface constraints for three-dimensional problems with damage propagation.
Engineering Computations, 17, 2847.
Ortiz, M. and Pandolfi, A. (1999). Finite-deformation irreversible cohesive elements for threedimensional crack-propagation analysis. International Journal for Numerical Methods in
Engineering, 44, 12671282.
Schellekens, J.C.J. and de Borst, R. (1993). On the numerical integration of interface elements.
International Journal for Numerical Methods in Engineering, 36, 4366.
van der Sluis, O. (2004). Continuum mechanical failure models a literature overview . Philips
CFT, CTB591-04-2422, Internal Philips report.
Tomar, V., Zhai, J. and Zhou, M. (2004). Bounds for element size in a variable stiffness
cohesive finite element model. International Journal for Numerical Methods in Engineering,
61, 18941920.
Xu, X. and Needleman, A. (1994). Numerical simulations of fast crack growth in brittle solids.
Journal of the Mechanics and Physics of Solids, 42, 13971434.
26

Acknowledgements
I would like to thank Olaf van der Sluis from Philips Applied Technologies (or PAT, formerly
known as Philips CFT) for his supervision and for helping me out with the project many
times during my internship there. Furthermore I would like to thank Ron Peerlings from the
university for his supervision and critical comment during my internship and for afterwards
helping me with the finishing touch on the final implementation.

27

Appendix A

Derivation of tangent stiffnesses


loading stiffnesses
The tangents for loading follow by straightforward differentiation of equations (2.3) and (2.4)
as
Tn
n

Tn
t


 



1
n
n
n
2t
exp
2 exp 2 +
n
n
n
n
t
!!



2t
1
n
r
1q
1 exp 2
+ 2

+
r1
n
n
n
t

(


n
n
2t
n t
exp
exp 2 +
2
n
n
n t2
t
)



2t
n
1 q t
exp

r
+2
2
2
r 1 t
n
t

(A.1)

(A.2)

Tt
n

!






n t
n
2t
rq
n
= 2 2
exp
exp 2
q
1
n
n
r1
n
t t

(A.3)

Tt
t

!
!




r q n
22t
n
2t
n
q+
1 2 exp 2
= 2 2 exp
n
r 1 n
t
t
t

(A.4)

28

coupled stiffnesses for unloading


To obtain the stiffnesses in case of unloading using the coupled implementation with scaling
the following unloading tractions are used
Tnunl =
Ttunl =

eff
Tn (n , t )
max
eff
Tt (n , t )
max

(A.5)
(A.6)

with
max
n
eff
max
t
eff

n =
t =

(A.7)
(A.8)

Tnunl and Ttunl are the unloading tractions in the normal and tangential direction respectively.
The variation of eff is needed for the differentiation of Tnunl and Ttunl and can be derived as
2eff = 2n + 2 2t

(A.9)
2

2eff eff = 2n n + 2 t t
t
n
n + 2
t
eff =
eff
eff

(A.10)
(A.11)

The variation of the traction in normal direction, Tnunl , can be derived as


Tnunl =

=
=

!
max
max n
n
eff +
eff
2eff
!
max t
eff Tn max
eff
t
+
max t eff
2eff
!
1
Tn
t Tn
Tn
n Tn
eff +
t

n +
Tn

max
eff n eff t
n
t
!
!
1
n
t Tn
n Tn
2 t

Tn
n +
t +
max
eff n eff t
eff
eff
1

eff Tn
eff Tn +
max
max n

+
=

Tn
Tn
n +
t
n
t

!
2 T

n
n
n
t
n
n +

Tn + 2 2t
max eff
eff n
2eff t
!
2 T

T
t
n t
n
n
Tn + 2n
+ 2
2 2
t
max eff
eff t
eff n

29

(A.12)

The variation of the traction in shear direction:


Ttunl

!
max n
max
n
eff +
eff
2eff
!
max t
eff Tt max
eff
t
+
max t eff
2eff
!
!
n Tt

n
1
t Tt
t
Tt
n + 2
t +

max
eff n eff t
eff
eff

eff Tt
eff Tt +
max
max n
1

+
=

Tt
Tt
t
n +

n
t

!
2
n t Tt
n
2 t Tt

Tt + 2
n +
max eff
eff n
2eff t
!
2n Tt
t
2 n t Tt
2

Tt + 2
t
+
max eff
eff t
2eff n

(A.13)

The tangent stiffnesses for unloading then become


n
2 Tn
n t Tn
Tn + 2 2t

max eff
eff n
2eff t

Tnunl
n

Tnunl
t

= 2

Ttunl
n

Ttunl
t

= 2

n t Tn
t
2 Tn
2 2
Tn + 2n

max eff
eff t
eff n

n t Tt
n
2 Tt

Tt + 2 2t
max eff
eff n
2eff t
n t Tt
t
2 Tt
2 2
Tt + 2n

max eff
eff t
eff n

30

(A.14)

(A.15)

(A.16)

(A.17)

You might also like