You are on page 1of 21

INTERNATIONAL JOURNAL FOR NUMERICAL AND ANALYTICAL METHODS IN GEOMECHANICS

Int. J. Numer. Anal. Meth. Geomech., 2007; 31:239259


Published online: 7 November 2006 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/nag.560

Crack propagation criteria in the framework of X-FEM-based


structural analyses
Peter Dumstorz and Gunther Meschke*,y
Institute for Structural Mechanics, Ruhr University Bochum, 44780 Bochum, Germany

SUMMARY
The extended nite element method (X-FEM) has proven to be capable of simulating cracking and crack
propagation in quasi-brittle materials, such as cement paste or concrete, without the need for re-meshing.
In the framework of the X-FEM cracks are represented as surfaces of discontinuous displacements
continuously propagating through nite elements. Since crack path continuity is required in X-FEM-based
analyses, the reliability of numerical analyses of cracked structures crucially depends on the correct
prediction of the crack path and, consequently, on the criterion used for the determination of the crack
propagation direction. In this paper four dierent crack propagation criteria proposed in the literature are
investigated including two local and two global criteria. The two local criteria include an averaged stress
criterion and the maximum circumferential stress criterion based on the linear elastic fracture mechanics.
The two global criteria include a global tracking criterion proposed by Oliver and Huespe (Online
Proceedings of the Fifth World Congress on Computational Mechanics, 2002) and an energy based X-FEM
formulation recently proposed in (Computational Plasticity 2005. CIMNE: Barcelona, 2005; 565568;
Comput. Methods Appl. Mech. Eng. 2006, in press). Representative numerical benchmark examples,
characterized by mode-I dominant fracture as well as by mixed-mode fracture, are used to study the
performance and the robustness of the dierent crack propagation criteria. Copyright # 2006 John Wiley
& Sons, Ltd.
Received 2 December 2005; Revised 11 August 2006; Accepted 14 August 2006
KEY WORDS:

extended nite element method; cracking; cohesive cracks; crack growth; concrete structures

1. INTRODUCTION
The process of cracking in quasi-brittle materials such as cement paste, brickwork or concrete is
characterized by the formation of microcracks which eventually coalesce and form a
propagating macrocrack. The realistic modelling of the process of crack opening and
*Correspondence to: Gunther Meschke, Institute for Structural Mechanics, Ruhr University Bochum, 44780 Bochum,
Germany.
y
E-mail: guenther.meschke@sd.rub.de
z
E-mail: peter.dumstor@siemens.com

Copyright # 2006 John Wiley & Sons, Ltd.

240

P. DUMSTORFF AND G. MESCHKE

propagation is a prerequisite for reliable prognoses of the safety and the durability of concrete,
reinforced concrete and masonry structures. Numerical modelling of cracks and structural
analysis of cracked structures dates back to the early 1970s. Up to the middle of the 1990s, one
focus of research in computational failure analysis was laid on the development of continuumbased models (see the reviews contained, e.g. in [14]). At the same time another focus was laid
on the development of numerical models for a discrete representation of fracture using linear
elastic fracture mechanics (LEFM) and, partially, cohesive crack models (see, e.g. [5, 6]). More
recently the goal of a discrete representation of cracks within a continuum setting of the nite
element method has resulted in approaches which allow for the representation of cracks as
embedded discontinuities within nite elements, circumventing the need for re-meshing as
cracks evolve.
These formulations can generally be categorized into element-based formulations, generally
denoted as embedded crack models (see [711], among others) and nodal-based formulations,
e.g. the Extended Finite Element Method (X-FEM) (see [1214]). For a comparative assessment
of both approaches we refer to [15, 16]. With the exception of the rotating crack formulation of
the strong discontinuity approach [11], the topology of crack segments is held xed once they
are signalled to open. Furthermore, in the framework of the extended nite element method
crack path continuity is required. Hence, the correct prediction of the direction of new crack
segments is crucial for the reliability as well as for the robustness of the numerical analysis. If the
predicted propagation direction is incorrect, locking occurs, which generally leads to
unreasonable results and eventually may cause failure of the analysis.
The main focus of this paper lies on the numerical assessment of crack propagation criteria
and their eects on the predicted crack path by means of comparative numerical analyses based
on the X-FEM. Four dierent crack propagation criteria proposed in the literature are
investigated: a principal stress criterion based on the averaged stresses around the crack tip [14],
a maximum circumferential stress criterion based on the stress intensity factors determined
according to linear elastic fracture mechanics [13, 17, 18] and a global tracking algorithm
[19, 20]. In addition to these three criteria a global energy-based crack propagation criterion
recently proposed in [2124] is investigated. In this variational formulation, the orientation as
well as the length of new crack segments are included as an additional global unknowns in the
variational formulation. These additional degrees of freedom are solved for simultaneously with
the regular degrees of freedom.
The X-FEM is characterized by the incorporation of discontinuous shape functions into the
nite element approximation, as has been suggested rst in the pioneering work of Dvorkin
et al. [25], by exploiting the partition of unity property of the nite element shape functions
[12, 26, 27]. Cracks are not limited to element boundaries but can be located arbitrarily in the
nite element mesh. In the present model, cracks are allowed to continuously propagate through
elements. To this end, a crack tip function is introduced to enhance the resolution of the
displacement eld approximation in the vicinity of the crack tip.
In zones of predominantly tensile stresses in quasi-brittle materials such as concrete or cement
paste, microcracks start to form within a process zone in the vicinity of the crack tip. In contrast
to perfectly brittle materials, the length of this process zone is usually not negligibly small
compared to the size of typical structures. Hence, the transfer of residual stresses parallel as well
as orthogonal to the crack faces after crack initiation must be represented by an adequate
traction separation law considering interactions between shear and normal components of crack
displacements and interface tractions in mixed-mode situations. In contrast to LEFM the stress
Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

CRACK PROPAGATION CRITERIA FOR X-FEM ANALYSES

241

eld in the vicinity of the tip of cohesive cracks does not exhibit a singularity. Consequently, in
contrast to crack tip functions used in LEFM [12], modied crack tip functions are used in this
paper which do not exhibit stress singularities but yield bounded values of stresses at the
crack tip.
The remainder of the paper is organized as follows: Section 2 contains an overview
of the X-FEM. In Section 3 the four investigated crack growth criteria are described. The
performance of the dierent crack growth criteria is investigated in Section 4 by means of two
representative numerical examples characterized by mode-I and mixed-mode fracture,
respectively.

2. EXTENDED FINITE ELEMENT METHOD


This section provides a concise description of the X-FEM. Using enhancement functions
for cracks fully penetrating elements and for the crack tip located within elements, cracks
may continuously propagate through elements and are not restricted to elements edges. This
enables modelling of curved cracks within elements. Since this paper is focused on the
investigation of dierent crack growth criteria, the presentation of the method is restricted to a
general overview. For a more detailed and complete description of the method we refer to
[12, 14, 28].
2.1. Kinematics
We consider a body B whose domain O is separated into two parts O and O by means of a
localization surface @s O: The displacement eld u of this body can be decomposed into a
continuous part u% and a discontinuous part u$
ux u% x u$ x 8x 2 O

with

u$ x Ss x#ux

where u% and u# are continuous functions in the domain O and Ss is the Sign function
dened as
8
1
8x 2 O
>
>
<
2
Ss x 0
8x 2 @s O
>
>
:
1 8x 2 O
The Sign function Ss can be expressed in terms of the Heaviside function Hs0:5 centred on the
localization surface @s O
Ss x 2Hs0:5 x  1

Hence the amplitude of the displacement jump at the discontinuity is given as


[u] 2#u

8x 2 @s O

The geometrically linear strain eld e is obtained by taking the symmetric gradient of u
according to Equation (1)
eu =s u =s u% Ss =s u# 2#u  ns s ds
|{z} |{z}
regular

Copyright # 2006 John Wiley & Sons, Ltd.

singular

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

242

P. DUMSTORFF AND G. MESCHKE

The singular part of Equation (5) is expressed in terms of the Dirac-delta distribution using the
classical result for the gradient of the Heaviside function =Hs0:5 ns ds ; where ns is the unit
normal vector on the surface of discontinuity.
2.2. Enhanced displacement approximation
The X-FEM is based on the partition of unity method. The partition of unity method [27] oers
the possibility to include arbitrary enhancement functions into the nite element approximation.
In the framework of the X-FEM the nite element approximation of the displacement eld is
locally enhanced by discontinuous functions where a crack has opened. In this paper two
dierent categories of enhancement functions are used. The Sign function Ss is chosen to
enhance the approximation of the displacement eld across open cracks. Since the
determination of the crack propagation direction strongly depends on the quality of the
approximation in the vicinity of the crack tip additional crack tip enhancement functions Fi for
cohesive cracks illustrated in Figure 1 are used
 
 
y
y
F1 r; y r cos
siny; F2 r; y r sin
siny
2
2
6
 
 
y
y
; F4 r; y r sin
F3 r; y r cos
2
2
y and r denote the local polar coordinates at the crack tip. The crack tip enhancement functions
Fi are similar to the functions used in [12]. However, the polynomial degree in radial direction is
modied to avoid stress singularities at the crack tip. This is consistent with the cohesive crack
model proposed in Section 2.4. Using these crack tip enhancement functions, crack tips of
cohesive cracks do not have to be located on element boundaries, but can be located arbitrarily
in the nite element mesh. From Equation (6) and Figure 1 it is obvious that only the function
F4 is discontinuous across the localization surface @s O: In analogy to the discontinuous part of
the displacement eld the function F4 is decomposed into the product of the Sign function Ss
and the continuous function F* 4
F* 4 Ss F4

with Ss Ss 1

Using standard nite element shape functions as a partition of unity, a nite element
approximation which includes all three parts, the standard nite element approximation, the

Figure 1. Crack tip enhancement functions used for cohesive cracks.


Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

CRACK PROPAGATION CRITERIA FOR X-FEM ANALYSES

243

Sign function Ss and the crack tip enhancement functions Fi ; can be written as
u% 

NN
X

Ni u% ri

X
i2N

3
X

Fj u% tij

j1

i2N

Ni u# si

[u]  2

Ni

i1

u# 

X
X

Ni F* 4 u# ti4

i2N

Ni u# si 2

i2Ns

Ni F* 4 u# ti4

8x 2 @s O

i2Nt

where u% ri are the regular degrees of freedom, u# si are the enhanced degrees of freedom associated
with the Sign function Ss and u% ti ; u# ti are the enhanced degrees of freedom associated with the
crack tip enhancement functions Fi : A node is enhanced by the Sign function if the support of
the respective node is crossed by a crack and the crack tip segment is not located within this area
(see Figure 2). The set of nodes which is enhanced by the Sign function is denoted as Ns : A node
is enhanced by the crack tip enhancement functions if the support of the respective node is
located within a circle of radius re centred in the root of the crack tip segment (see Figure 2). The
set of nodes enhanced by the crack tip enhancement function is denoted as Nt : This concept
allows to keep the nodal enrichments unchanged during equilibrium iterations when the global
energy criterion is used for crack propagation (see Section 3.4).
2.3. Finite element formulation
Starting with the local equilibrium condition
div r 0
the standard principle of virtual work without body forces is written as
Z
Z
=s du : r dV
dut$ dA
O

10

@s O

where du are admissible test functions, r is the stress eld and t$ are the prescribed Neumann
boundary conditions. Inserting the strain eld given by Equation (5) and integrating the

Figure 2. Concept of nodal enhancement in X-FEM-based crack analyses.


Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

244

P. DUMSTORFF AND G. MESCHKE

Dirac-delta distribution over the domain O the virtual work equation results in
Z
Z
Z
Z
s
s
= du% : r dV Ss = d#u : r dV
d[u] t dA
du% t$ dA
O

@s O

11

@s O

where t rns are the traction forces acting on the discontinuity. To simplify matters, the
enhanced part of the displacement eld is assumed to be zero where boundary conditions are
imposed. This avoids the consideration of d#u on the right-hand side of Equation (11) and,
consequently, the formulation of complex procedures to apply boundary conditions. The
admissible displacement variations du% ; d#u and d[u] are constructed in analogy to Equation (8).
2.4. Interface constitutive law
This subsection is concerned with the formulation of a traction-separation law for cracks in
quasi-brittle materials subjected to general mixed-mode conditions. To this end a model
introduced by Camacho and Ortiz [29] and later employed by Mariani and Perego [30], among
others, is used as the starting point. The model accounts for dissipative mechanisms in mode-I
and mode-II conditions using equivalent crack tractions and crack openings.
The proposed model is based upon a relation between the equivalent stresses teq and the
equivalent displacement jump [u]eq
teq T[u]eq T e  T d [u]eq
with [u]eq dened as
[u]eq

q
2
2
[u]n b2 [u]s

12

13

In (12) T e denotes the initial stiness and T d is a scalar damage variable governing the
degradation of the equivalent interface stiness T: The parameter b allows to control the shear
strength and the shear stiness in mode-II fracture. It depends on the microstructure of the
material and the tortuosity of the crack, respectively, and has to be adjusted to the specic
investigated material.
The range of elastic interface material behaviour is characterized by means of the damage
criterion
f[u]eq ; a [u]eq  a40

14

where a denotes a history variable associated with the maximum equivalent displacement jump
ever reached during the loading history. In the space of normal and shear tractions, Equation
(14) describes an ellipse (see Figure 3(b)). The width-to-height ratio of the elliptical
damage criterion is determined by means of the variable b: In Figure 3(b) the damage
criterion}transformed to the stress space}is illustrated for b 2:
The equivalent damage stiness T d is dened as an exponentially increasing function of the
history variable a (Figure 3(a))



a0
ftu
d
e
T T 1  exp  a  a0
15
a
Gf
ftu is the tensile strength, Gf is the fracture energy and a0 is the initial value of a dened as
ftu
a0 e
16
T
Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

CRACK PROPAGATION CRITERIA FOR X-FEM ANALYSES

(a)

245

(b)

Figure 3. Damage-based interface model for cohesive cracks: (a) evolution of the equivalent traction; and
(b) evolution of the damage criterion in the stress space for b=2.

The equivalent traction teq is dened via the work equivalence condition
eq tn [u]
n ts [u]
s
teq [u]

17

which provides a relation between teq and the components tn and tt of the traction vector t:
Taking the time derivative of Equation (13) and multiplying with teq
eq teq [u] [u]
n b2 teq [u] [u]
s
teq [u]
18
n
[u]eq
[u]eq s
the components tn and ts can be determined from a comparison of the coecients of Equation
(17) and (18) as
teq
teq
tn
[u]n ; ts b2
[u]s
19
[u]eq
[u]eq
From the damage criterion (14) together with the consistency condition and the KuhnTucker
condition follows
eq
a [u]eq ; a [u]
20
After time integration within an interval tn ; tn1  the increment of the equivalent traction is
obtained as
!
@T d
d
[u]eq D[u]eq
21
Dteq T  T 
@[u]eq
Finally the tangential stiness matrix Ttan is determined from Equations (19) and (21)
2
3
2
2
Gf ftu [u]eq
[u]n [u]s
[u]s [u]n
6 2 
7
Gf ftu [u]eq 7
b Gf ftu [u]eq
b2 ftu T se 6
6
7 D[u]
Dt 
[u]eq 6
Gf ftu [u]eq b2 [u]2s [u]2n 7
4
5
[u]s [u]n

Gf
Gf ftu [u]eq
ftu [u]eq
|{z}

22

Ttan

Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

246

P. DUMSTORFF AND G. MESCHKE

where T se denotes the secant stiness dened as




a0
ftu
se
d
T T  T T exp  a  a0
a
Gf

23

For elastic loading or unloading conditions the evolution of the traction vector t is given as
Dteq T  T d D[u]eq
resulting in the elastic tangential stiness matrix
"
#


Dtn
se 1 0
T
D[u]
0 b2
Dts
|{z}

24

25

Ttan

3. CRACK GROWTH MODELS


The numerical analysis of crack propagation can be decomposed into two steps: In a rst step,
at the end of each time interval, it has to be determined, whether an existing crack is going to
propagate or not. If crack propagation is signalled, the direction and length of the new crack
segment have to be determined in a second step. In this paper, the focus lies upon crack
propagation rather than on crack initiation. Therefore, a small initial crack is prescribed to
initialize the crack propagation process.
In this section four dierent criteria for the determination of the crack growth direction are
investigated. Two criteria are based on local information from the vicinity of the individual
crack tip and two criteria have a global character: a principal stress criterion based on the
averaged stresses around the crack tip [14], a maximum circumferential stress criterion based on
the stress intensity factors determined according to LEFM [13, 17, 18], a global tracking
algorithm [19, 20] and a global energy-based crack propagation criterion.
The main purpose of this paper is the investigation of the four criteria for the determination of the
crack growth direction. Therefore, the same crack propagation criterion is used for all crack growth
models investigated in Section 4. To this end, an energy-based crack propagation criterion is used for
all analyses. According to this criterion the crack propagation condition is formulated as follows:
8
>
> > 0 ? no crack propagation
@C <
26
0 ? stationarity condition
@Acr >
>
:
50 ? crack propagation
where Acr is the area of a new crack segment displayed in Figure 7. More information on this crack
propagation criterion is given in References [6, 31].
The algorithm used for the present comparative analyses is contained in Figure 4. After each
converged load increment the crack propagation criterion (26) is checked. If no crack growth is
signalled, the current crack conguration remains unchanged and the next loading step is applied.
In case that crack growth is indicated, the crack growth direction is determined according to the
specic crack propagation criterion and a new crack segment with a xed, predened length is
inserted. Based on this updated crack conguration, global equilibrium is restored without
increasing the load. The crack is further extended as long as, for the given load level, no further
crack growth is signalled. Subsequently, the analysis continues with the next loading step.
Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

CRACK PROPAGATION CRITERIA FOR X-FEM ANALYSES

247

Figure 4. Algorithm used for crack propagation analyses.

3.1. Averaged stress criterion


According to the averaged stress criterion proposed in [14], the principal axis of the averaged
stress tensor rm ; corresponding to the maximum principal stress, is taken as the normal vector ns
of the crack extension. The motivation for using a non-local stress quantity instead of the local
stresses is to improve the reliability of the computed stresses in the vicinity of the crack. To this
end, an averaged stress tensor rm is computed
Z
rm wr dV
27
using a Gaussian weight function w (see Figure 5) dened as
 2
1
r
w
exp  2
3=2 3
2l
2p l

28

In (28) r is the distance from the crack tip and l is the parameter which determines how fast the
weight function decays from the crack tip. In the analyses described in Section 4 the parameter l
is taken as 1:5 times the internal element length lc
p
lc A e
29
where Ae is the area of an average nite element.
Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

248

P. DUMSTORFF AND G. MESCHKE

Figure 5. Averaged stress criterion: nite element discretization, crack conguration


and Gaussian weight function.

3.2. Linear elastic fracture mechanics


Although cohesive crack models are generally not compatible with the underlying assumptions
of LEFM, a maximum circumferential stress criterion based upon the evaluation of the stress
intensity factors KI and KII has been used in combination with cohesive crack models by various
authors [13, 17, 18]. The use of this criterion is motivated by the assumption that the size of the
investigated structure has a minor inuence on the crack path. This assumption is corroborated
by experimental investigations in [32, 33]. Therefore, this approach has been included in the
present comparative evaluation in order to assess its suitability in comparison with other criteria
specically designed for cohesive cracks.
The numerical analyses using the LEFM-based criterion are characterized by two separate
computational steps: In a rst step the crack path is calculated assuming linear elasticity and
traction-free cracks. The second step is characterized by the incorporation of the cohesive crack
model, i.e. the consideration of traction forces along the open crack as well as the new crack
segment, using the crack topology obtained from LEFM. Using this approach the inuence of
cohesive forces on the crack path is obviously disregarded.
In the present analyses the two computational steps described above are performed
simultaneously. Therefore an additional set of degrees of freedom is introduced at each node.
The rst set of degrees of freedom is used for the approximation of the linear elastic structural
system with traction-free cracks and the additional set of degrees of freedom is used for the
analysis of the structural system incorporating the cohesive crack model.
In this paper the crack is assumed to grow in the direction of the maximum circumferential
stress. This direction is determined by means of the stress intensity factors KI and KII : The stress
intensity factors are calculated by evaluating the interaction integral around the crack tip
assuming linear elasticity and traction free cracks. Since the maximum circumferential stress is a
principal stress, the propagation direction can be determined analytically by setting the shear
stress to zero


1
1
1
cosy=2 KI siny KII 3 cosy  1 0
sry
2pr
2
2
Copyright # 2006 John Wiley & Sons, Ltd.

30

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

CRACK PROPAGATION CRITERIA FOR X-FEM ANALYSES

249

where y and r are the local polar co-ordinates at the crack tip. This leads to the propagation
direction in terms of the stress intensity factors
0 0
s11
 2
1 KI
KI

8AA
31
ycr 2 arctan@ @
4 KII
KII
According to this formula the maximum kinking angle is limited to 70:538 for pure mode-II
fracture.
3.3. Global Tracking Algorithm
The so-called Global Tracking Algorithm has been proposed by Oliver et al. [19, 20] in the
context of the strong discontinuity approach and has been modied recently by Feist and
Hofstetter [34]. Nevertheless, the implementation into the X-FEM is straightforward. In this
paper only the basic idea of the method is presented. For a more detailed description we refer to
[19, 20]. In contrast to the previously described criteria the Global Tracking Algorithm does not
need to be evaluated for each individual crack but traces all possible discontinuity paths at once.
The basic idea of this method is to construct a function W whose iso-lines run perpendicular to
the directions of the principal stresses (i.e. the directions of potential crack normal directions) in
all integration points of the investigated structure. To this end a stationary anisotropic heat
conduction-like problem is solved with the anisotropic conductivity being dened by the
principal axes of the maximum stresses. The isothermal lines of the calculated temperature eld
represent all possible crack paths.
If crack growth is indicated, the value of W is calculated at the crack tip. The crack is then
extended following the isoline corresponding to this value of W: An isoline Si of the function W is
dened as the collection of points in the structure having the same value of W
Si fx 2 OjWx Wsi g

32

For the construction of a suitable function W a stationary anisotropic heat conduction-like


problem, characterized by the partial dierential equation and the respective boundary
conditions
div q 0;

q q$

Wx W$

8x 2 @q O;

8x 2 @W O

33

with q as the heat ux and @q O and @W O as the Neumann and Dirichlet boundaries, respectively,
has to be solved. The weak form of Equation (33)
Z
Z
d=WK=W dV
dWq$ dA
34
O

@q O

where K denotes the thermal conductivity tensor, is solved using the nite element method. The
anisotropic thermal conductivity tensor K is constructed using the following condition:
ts  =W 0

35

where ts is the unit vector normal to the vector in the direction of the maximum principal stress.
Thus, Equation (35) states that the ux in the direction perpendicular to ns vanishes. The
anisotropic thermal conductivity tensor K in each integration point is dened by the dyadic
product
ts  ts =W K=W 0
Copyright # 2006 John Wiley & Sons, Ltd.

36

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

250

P. DUMSTORFF AND G. MESCHKE

To avoid ill-posedness of the problem a perturbation term zI is used


K ts  ts zI

37

where I is the unit tensor and z is a perturbation factor large enough to avoid the singularity of
the global system of equations. As Dirichlet boundary conditions two dierent temperatures are
prescribed at two nodes in the nite element mesh. On the Neumann boundary zero heat ux
conditions are prescribed. Since only the iso-lines are relevant, the absolute values of the two
prescribed temperatures may be arbitrary. Even their location is irrelevant as long as they are
not imposed on the same iso-line and as long as the inuence of the nite element mesh can be
neglected. As an illustration the maximum principal stress directions and the respective iso-lines
obtained for the numerical example investigated in Subsection 4.1 are shown in Figure 6.
3.4. Minimum of total energy
In this subsection a global energy-based criterion for cohesive crack models recently proposed in
[2224] is investigated. Inspired by energy considerations on which LEFM is based, the crack
propagation angle ycr of a new cohesive crack segment relative to the existing crack (see
Figure 7) is determined by minimizing the total energy of the body. A similar procedure has
been proposed by Shen and Stephansson [35] within the framework of the boundary element
method and by Peters et al. [36] in the context of LEFM using X-FEM.
The total energy of a cracked body consists of the internal energy U; the external work Wb
and Wt of the body forces b and the surface tractions t$ and the surface energy of the crack Ws
Cu; ycr U  Wb  Wt Ws
with
U

Z Z
O


re de dV;

Wb

Z Z

ub dV;
O

Wt

38
Z

Z
@s O

ut$ dA

39

The functional C depends on the displacement eld u and the crack angle ycr : Consequently, ycr
is introduced as an additional degree of freedom in the discretized structural model. It should be
noted, that for the sake of comparison, for the energy-based X-FEM model only the crack angle
is included as an additional unknown variable. In the nal implementation of the model as
proposed in [2224] also the length of the new crack segment is an unknown variable in addition
to the unknown displacement eld and the unknown crack direction of the new segment, which

(a)

(b)

Figure 6. Global Tracking Algorithm: (a) vector plot of the maximum principal stress;
and (b) isothermal lines.
Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

CRACK PROPAGATION CRITERIA FOR X-FEM ANALYSES

(a)

251

(b)

Figure 7. Scheme of crack propagation in a cracked body.

is simultaneously solved for within the iterative procedure to nd a global energy minimum.
This additional degree of freedom ycr and the unknowns of the regular and the enhanced
displacement eld}u% and u# }are solved for simultaneously in analogy to multield problems.
For cohesive cracks, the surface energy Ws is given as
Z Z [u]
Ws
ts [u]d[u] dA
40
@s O

with the separation-dependent residual tractions ts [u] acting along the process zone of the
crack (see Figure 8). Among all possible deformed congurations of the body containing one (or
more) crack(s) extended by one (or more) new crack segment(s), the actual one, associated with
a safe equilibrium of the cracked body will lead to a minimum of Cu; ycr [37]. This is equivalent
to the stationarity condition
dCu; ycr

@C
@C
du
dycr 0
@u
@ycr

41

As mentioned above the consideration of body forces is neglected in the present


formulation. Inserting Equation (38) into (41) and considering that the enhanced displacements
are assumed to be zero where boundary conditions are imposed leads to the following
conditions


@U @Wt @Ws


du 0
@u
@u
@u
42


@U @Ws

dycr 0
@ycr @ycr
The extended weak form associated with the stationarity condition (41) and its linearization are
given as
"
# "" u # " $u ##
r
du
r
dP


y
dycr
0
r
43
"
# " uu
#
"
#
kuy
k
Du
du
DdP

Dycr
dycr
kyu kyy
Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

252

P. DUMSTORFF AND G. MESCHKE

Figure 8. Body containing cohesive cracks.

where the superscript


identied from (42) as

stands for the external forces. The vectors of internal forces are

@U @Ws

@u
@u
@U
@W
s
ry

@ycr @ycr

ru

44

The global degree of freedom ycr is introduced through the enhancement function Ss which is
formulated in terms of ycr : To this end, the Sign function Ss at the crack tip segment is
re-formulated in the form
Ss x; ycr signdx; j

with dx; ycr ns ycr x  xcr

45

where xcr are the co-ordinates of the root of the crack tip segment (see Figure 2) which are
invariant with respect to ycr : In the present implementation of the crack propagation models
described at the beginning of Section 3 the length of the new crack segment is held constant
during the Newton-iteration. During the iterative procedure it may also be possible that element
boundaries are crossed. In order to enhance the robustness of numerical analyses, all degrees of
freedom associated with the crack tip enhancement function (see Figure 1) within a circular area
in the vicinity of the crack tip (see Figure 2) are coupled. This coupling assures a linear crack
opening function independent of the fact whether an element boundary is crossed or not.
Without this coupling the crack opening function would not be independent of the nite element
mesh. Consequently, discontinuities would be induced in the functional dependence of the total
energy from the crack angle. These discontinuities would lead to convergence problems in the
iterative solution procedure. The derivatives required in (42) and (43) are evaluated numerically.
Details of a consistent numerical implementation of the energy-based X-FEM model,
characterized by taking the crack propagation angle ycr as well as the crack length lcr as
unknowns in the variational formulation, together with the computation of the extended
tangent stiness matrix (43) are contained in [22, 24].
Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

CRACK PROPAGATION CRITERIA FOR X-FEM ANALYSES

253

4. COMPARATIVE NUMERICAL ANALYSES


In this section the performance of the four crack growth criteria described above is investigated
by means of two representative numerical examples. The rst example is concerned with
re-analyses of a mixed-mode fracture benchmark test [32], while the second benchmark test, a
L-shaped concrete panel tested by Winkler [38], is dominated by mode-I fracture. For all
analyses, the interface law as described in Section 2.4 has been employed. For the
approximation of the regular displacement eld, bi-quadratic shape functions are used. This
improves the quality of approximation of the stress eld and enhances the reliability of the crack
propagation models. In contrast, bilinear shape functions are used as partition of unity to
reduce possible linear dependencies between regular and enhanced shape functions.

4.1. Mixed-mode fracture test


A square shaped double edge notched specimen made of mortar which has been tested by
Nooru-Mohamed [32] is used as a benchmark for the assessment of the dierent crack
propagation criteria in a mixed-mode dominated situation. In [32] mortar and concrete
specimens of dierent sizes were tested under dierent loading conditions. Only one of these
experiments (load path 4b, specimen size 200  200  50) is investigated in this paper. The
geometry, the loading and boundary conditions of the specimen as well as the material
parameters used for the interface law described in Section 2.4 are contained in Figure 9. The
uniaxial tensile strength and the fracture energy were estimated using the experimentally
determined values of the tensile splitting strength and the compressive strength. The load
sequence is dened as follows: rst the specimen is subjected to a shear force which is increased
up to Fs 10 kN: Subsequently, a displacement-controlled tensile axial load Fn is applied.
A relatively coarse nite element mesh consisting of 435 elements assuming plane stress
conditions as shown in Figure 10(a) was used.
Figure 10 shows a comparison of the experimentally obtained crack path and the numerical
results obtained from the four investigated crack growth criteria. The range of crack paths
observed in the experiments is indicated in light grey colour in Figures 10(c)(f). It should be

Figure 9. Mixed-mode fracture test: geometry and material parameters.


Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

254

P. DUMSTORFF AND G. MESCHKE

(a)

(b)

(c)

(d)

(e)

(f)

Figure 10. Numerical analysis of the mixed-mode fracture test: (a) nite element discretization; (b) load
displacement curves; and (c)(f) crack paths based on: (c) averaged stress criterion, (d) LEFM, (e)
minimum of total energy, (f) Global Tracking Algorithm.

noted, that, as was shown in [24, 39], the chosen degree of polynomial approximation of the
regular displacement eld has only a minor inuence on the crack trajectories.
The computed crack paths using the averaged stress criterion (Figure 10(c)), the Global
Tracking Algorithm (Figure 10(f)) and the maximum global strain energy release rate (Figure
10(e)) are located well within the experimentally determined range. Only the crack path based on
LEFM (Figure 10(d)) is not located within the range of experimental observations. The
curvature of the crack observed in the experiments is perfectly reproduced by all numerical
simulations independent of the crack growth criterion.
The load displacement curves computed using the four dierent crack growth direction
criteria are shown in Figure 10(b). With the exception of the LEFM criterion, all load
Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

CRACK PROPAGATION CRITERIA FOR X-FEM ANALYSES

255

displacement curves are more or less identical. The relatively large deviation of the numerical
results obtained from the LEFM-based criterion is associated with the misprediction of the
crack trajectory using this criterion. It should be noted, that all calculated load displacement
curves deviate from the experimental observations. This is a consequence of the obvious
overestimation of the fracture energy and of the tensile strength as also reported in [40].
4.2. L-shaped panel
The second numerical example is concerned with a L-shaped panel made of concrete which was
investigated experimentally by Winkler [38]. The geometry, the loading and boundary
conditions of the panel as well as the material parameters are contained in Figure 11. The
load F is applied incrementally via prescribed displacements. Figure 12(a) contains the chosen
nite element discretization which consists of 261 elements assuming plane stress conditions. In
a numerical study [39], the inuence of the spatial discretization on the numerical results of this
benchmark problem has been investigated. It has been found, that the computed crack
trajectory is almost independent of the mesh size.
In Figure 12(c) the crack path calculated using the averaged stress criterion is shown. The
crack follows a path located below the experimental range. Shortly after the peak load is
reached a sharp bending of the crack path is observed. This pathological behaviour can be
attributed to the state of almost equal biaxial tensile stresses existing in the vicinity of the crack
tip. The numerical results after bending of the crack are no longer reasonable. A similar
phenomenon was also observed in numerical simulations of three point bending tests using the
averaged stress criterion [39].
The crack path computed by means of the crack growth criterion based on LEFM
(Figure 12(d)) is located perfectly within the experimentally determined range. Obviously, for
crack propagation problems characterized primarily by mode-I fracture, such as the L-shaped
panel, LEFM seem to result in reasonable predictions of the crack path even for quasi-brittle
materials like concrete.

Figure 11. Numerical analysis of a L-shaped panel: geometry and material parameter.
Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

256

P. DUMSTORFF AND G. MESCHKE

(a)

(b)

(c)

(d)

(e)

(f)

Figure 12. Numerical analyses of the L-shaped panel: (a) nite element discretization; (b) load
displacement curves; and (c)(e) crack paths based on: (c) averaged stress criterion, (d) LEFM,
(e) minimum of total energy, (f) iso-lines based on Global Tracking Algorithm.

The crack path obtained by means of the criterion based upon the global energy criterion,
illustrated in Figure 12(e), is also located perfectly within the experimentally determined range.
It should be noted that the specic choice of parameters used for the interface law has an
inuence on the predicted crack path, since the energy dissipation along the crack faces
predicted by the specic model directly enters the crack propagation criterion. In particular, the
parameter b used for the interface law described in Section 2.4 considerably inuences the crack
path. For a more detailed investigation of the global energy criterion we refer to [22, 24], where
the inuence of b regarding the crack path is investigated.
The isothermal lines determined at the beginning of the analysis using the Global Tracking
Algorithm are shown in Figure 12(f). This gure shows, that the isothermal lines are almost
circular in the vicinity of the initial crack tip. Hence, no pronounced crack direction could be
Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

CRACK PROPAGATION CRITERIA FOR X-FEM ANALYSES

257

determined for this example without using additional criteria. Obviously, the present
implementation of the Global Tracking Algorithm does not seem to be applicable to this
benchmark example. It should be noted, that a modication of this algorithm [34] seem to work
also for the present situation. Furthermore, consideration of a continuum dissipation before a
discrete crack opens also improves the performance of the Global Tracking Algorithm in
analyses of the present benchmark problem [41].
The calculated load displacement curves are displayed in Figure 12(b). All calculated load
displacement curves show a good agreement with the experimental results. Obviously the exact
crack topology seems to have a minor inuence on the computed load displacement curve.

5. CONCLUSIONS
In this paper the X-FEM is used for 2D plane stress analyses of the propagation of cohesive
cracks within structures made of quasi-brittle materials. A crack tip function compatible with
the proposed cohesive crack model was introduced to enhance the displacement approximation
in the vicinity of the crack tip. Two local and two global crack growth criteria have been
investigated by means of comparative numerical analyses: a principal stress criterion based on
the averaged stresses around the crack tip [14], a maximum circumferential stress criterion based
on LEFM [13, 17, 18], the Global Tracking Algorithm proposed by Oliver and Huespe [42] and
a global energy criterion based on the minimization of the total energy within the structure
[22, 24]. An interface law considering dissipative coupling within cracks in mixed-mode
problems has been formulated in the framework of damage mechanics and employed in all
analyses.
The performance of the dierent criteria has been investigated numerically by means of two
representative numerical examples characterized by mode-I and mixed-mode dominated
fracture. In the mode-I fracture test the Global Tracking Algorithm in the original form as
proposed in [42] and the averaged stress criterion failed and therefore do not seem to be
generally applicable without further enhancements. It should be noted, that for the Global
Tracking Algorithm, such enhancements exist [40, 41].
The criterion based on the LEFM seem to perform well in mode-I dominated fracture in
quasi-brittle materials. For the mixed-mode problem, a more or less signicant deviation from
the range of experimental results has been observed. It is concluded, that the applicability of
LEFM for cohesive crack problems depends on the problem and cannot be generalized.
The criterion based on the minimum of the total energy yields satisfactory results for both
examples. This criterion does not need any material-specic assumptions regarding the
determination of the crack direction. It is, in contrast to the LEFM-based criterion, fully
consistent with the assumption of cohesive cracks. The inuence of the specic material is
reected by the specic choice of the interface law for open cracks and the parameters used
therein. A detailed investigation of the inuence of the material parameter b on the calculated
crack path is provided in [24].

REFERENCES
1. Hofstetter G, Mang HA. Computational Mechanics of Reinforced and Prestressed Concrete Structures. Vieweg:
Braunschweig, 1995.
Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

258

P. DUMSTORFF AND G. MESCHKE

2. de Borst R, Bicanic N, Mang MA, Meschke G (eds). Computational modelling of concrete structures. Proceedings of
the EURO-C 1998 Conference. Balkema: Rotterdam, Badgastein, 1998.
3. de Borst R. Some recent issues in computational failure mechanics. International Journal for Numerical Methods in
Engineering 2002; 52:6395.
4. Mang HA, Meschke G, Lackner R, Mosler J. Computational modelling of concrete structures. In Comprehensive
Structural Integrity, Milne I, Ritchie RO, Karihaloo B (eds), vol. 3, Chapter 9. Elsevier: Amsterdam, 2003; 167.
5. Ingraea AR, Saouma V. Numerical modelling of discrete crack propagation in reinforced and plain concrete. In
Fracture Mechanics of Concrete: Structural Application and Numerical Calculation, Sih G, DiTommaso A (eds).
Martinus Nijho: Dordrecht, 1985; 171225.
6. Xie M, Gerstle WH. Energy-based cohesive crack propagation modeling. Journal of Engineering Mechanics (ASCE)
1995; 121(12):13491358.
7. Simo JC, Oliver J, Armero F. An analysis of strong discontinuities induced by strain-softening in rate-independent
inelastic solids. Computational Mechanics 1993; 12:277296.
8. Oliver J. Modelling strong discontinuities in solid mechanics via strain softening constitutive equations. Part 1:
Fundamentals, Part 2: Numerical simulation. International Journal for Numerical Methods in Engineering 1996;
39:35753623.
9. Jiras ek M, Zimmermann T. Embedded crack model: Part 1: Basic formulation. Part 2: Combination with smeared
cracks. International Journal for Numerical Methods in Engineering 2001; 50:12691305.
10. Armero F, Garikipati K. An analysis of strong discontinuities in multiplicative nite strain plasticity and their
relation with the numerical simulation of strain localization in solids. International Journal of Solids and Structures
1996; 33(20):28632885.
11. Mosler J, Meschke G. 3D modeling of strong discontinuities in elastoplastic solids: xed and rotating localization
formulations. International Journal for Numerical Methods in Engineering 2003 57:15531576.
12. Moes N, Dolbow JE, Belytschko T. A nite element method for crack growth without remeshing. International
Journal for Numerical Methods in Engineering 1999; 46:131150.
13. Moes M, Belytschko T. Extended nite element method for cohesive crack growth. Engineering Fracture Mechanics
2002; 69:813833.
14. Wells GN, Sluys LJ. A new method for modelling cohesive cracks using nite elements. International Journal for
Numerical Methods in Engineering 2001; 50:26672682.
15. Jiras ek M, Belytschko T. Computational resolution of strong discontinuities. In Fifth World Congress on
Computational Mechanics (WCCM V), Mang HA, Rammerstorfer FG, Eberhardsteiner J (eds). Online publication,
2002.
16. Dumstor P, Mosler J, Meschke G. Advanced discretization methods for cracked structures: the strong
discontinuity approach vs. the extended nite element method. Computational Plasticity 2003, CIMNE: Barcelona,
CD-ROM, 2003.
17. Cendon DA, Galvez JC, Elices M, Planas J. Modelling the fracture of concrete under mixed loading. International
Journal of Fracture 2000; 103:293310.
18. Zi G, Belytschko T. New crack-tip elements for X-FEM and applications to cohesive cracks. International Journal
for Numerical Methods in Engineering 2003; 57:22212240.
19. Oliver J, Huespe AE, Samaniego E, Chaves EWV. On strategies for tracking strong discontinuities in computational
failure mechanics. In Fifth World Congress on Computational Mechanics (WCCM V), Mang HA, Rammerstorfer
FG, Eberhardsteiner J (eds). Online publication, July 2002.
20. Oliver J, Huespe AE, Samaniego E, Chaves EWV. Continuum approach to the numerical simulation of
material failure in concrete. International Journal for Numerical and Analytical Methods in Geomechanics 2004;
28:609632.
21. Dumstor P, Meschke G. Investigation of crack growth criteria in the context of the extended nite element method.
In IV European Congress on Computational Methods in Applied Sciences and Engineering (ECCOMAS 2004),
Neitaanmaki P, Bossi T, Konotov S, Onate E, Periaux J, Knorzer D (eds). Jyvaskyla, Finland, 2004. CD-ROM.
22. Dumstor P, Meschke G. Modelling of cohesive and non-cohesive cracks via X-FEM based on global energy
criteria. In Computational Plasticity 2005, (Complas VIII), Owen DRJ, Onate E, Suarez B (eds). CIMNE: Barcelona,
2005; 565568.
23. Meschke G, Dumstor P, Fleming W, Jox S. Numerical analysis of crack propagation in concrete using X-FEM. In
Computational Modelling of Concrete Structures, Meschke G, deBorst R, Mang HA, Bicanic N (eds). A.A. Balkema:
Leidon/London/New York/Philadelphia/Singapore, 2006; 157166.
24. Meschke G, Dumstor P. Energy-based modeling of cohesive and cohesionless cracks via X-FEM. Computer
Methods in Applied Mechanics and Engineering 2006, in press.
25. Dvorkin E, Cuitino AM, Gioia G. Finite elements with displacement interpolated embedded localization lines
insensitive to mesh size and mesh distortions. International Journal for Numerical Methods in Engineering 1990;
30:541564.
26. Melenk JM, Babus ka I. Partition of unity nite element method: basic theory and applications. Computer Methods
in Applied Mechanics and Engineering 1996; 139(14):289314.
Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

CRACK PROPAGATION CRITERIA FOR X-FEM ANALYSES

259

27. Babus ka I, Melenk JM. The partition of unity method. International Journal for Numerical Methods in Engineering
1997; 40:727758.
28. Dolbow JE. An extended nite element method with discontinuous enrichment for applied mechanics. Ph.D. Thesis,
Northwestern University, 1999.
29. Camacho GT, Ortiz M. Computational modelling of impact damage in brittle materials. International Journal of
Solids and Structures 1996; 33:28992938.
30. Mariani S, Perego U. Extended nite element method for quasi-brittle fracture. International Journal for Numerical
Methods in Engineering 2003; 58:103126.
31. Bazant ZP, Planas J. Fracture and Size Eect in Concrete and Other Quasibrittle Materials. CRC Press: Boca Raton,
FL, 1998.
32. Nooru-Mohamed MB. Mixed-mode fracture of concrete: an experimental approach. Ph.D. Thesis, Technische
Universiteit Delft, 1992.
33. Schlangen E. Experimental and numerical analysis of fracture processes in concrete. Ph.D. Thesis, Technical
University Delft, The Netherlands, 1993.
34. Feist C, Hofstetter G. Mesh-insensitive strong discontinuity approach for fracture simulations of concrete. In
Numerical Methods in Continuum Mechanics (NMCM 2003), Kompis V, Sladek J, Zmindak M (eds). Society and
Engineering: Zilina, Slovak Republic, 2003.
35. Shen B, Stephansson O. Modication of the g-criterion for crack propagation subjected to compression. Engineering
Fracture Mechanics 1994; 47:177189.
36. Peters M, Hoppe U, Hackl K. Simulation of crack propagation using embedded discontinuities. In 2nd International
Conference on Lifetime Oriented Concepts, Stangenberg F, Bruhns OT, Hartmann D, Meschke G (eds). Bochum:
Germany, 2004, 141148.
37. Le KC. Variational principles of non-linear theory of brittle fracture mechanics. Journal of Applied Mathematics and
Mechanics (PMM) 1990; 54:543549.
38. Winkler BJ. Traglastuntersuchungen von unbewehrten und bewehrten Betonstrukturen auf der Grundlage eines
objektiven Werkstogesetzes fur Beton. Ph.D. Thesis, Universitat Innsbruck, 2001.
39. Dumstor P. Modellierung und numerische Simulation von Rissfortschritt in sproden und quasi-sproden
Materialien auf Basis der Extended Finite Element Method. Ph.D. Thesis, Lehrstuhl fur Statik und Dynamik,
Institut fur konstruktiven Ingenieurbau, Ruhr-Universitat Bochum, 2005.
40. Feist C. A numerical model for cracking of plain concrete based on the strong discontinuity approach. Ph.D. Thesis,
Institut fur Baustatik, Festigkeitslehre und Tragwerkslehre, Universitat Innsbruck, 2004.
41. Oliver J. Private Communication. 2004.
42. Oliver J, Huespe AE. On strategies for tracking strong discontinuities in computational failure mechanics. Online
Proceedings of the Fifth World Congress on Computational Mechanics (WCCM V), 2002. wccm.tuwien.ac.at

Copyright # 2006 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2007; 31:239259


DOI: 10.1002/nag

You might also like