You are on page 1of 15

Int J Fract (2009) 159:151165

DOI 10.1007/s10704-009-9391-y

ORIGINAL PAPER

Computational modeling of mixed-mode fatigue crack


growth using extended finite element methods
Yangjian Xu Huang Yuan

Received: 13 August 2008 / Accepted: 4 August 2009 / Published online: 30 August 2009
Springer Science+Business Media B.V. 2009

Abstract The extended finite element method


(XFEM) combined with a cyclic cohesive zone model
(CCZM) is discussed and implemented for analysis of
fatigue crack propagation under mixed-mode loading
conditions. Fatigue damage in elastic-plastic materials is described by a damage evolution equation in the
cohesive zone model. Both the computational implementation and the CCZM are investigated based on
the modified boundary layer formulation under mixedmode loading conditions. Computational results confirm that the maximum principal stress criterion gives
accurate predictions of crack direction in comparison
with known experiments. Further popular multi-axial
fatigue criteria are compared and discussed. Computations show that the Findley criterion agrees with tensile
stress dominant failure and deviates from experiments
for shear failure. Furthermore, the crack propagation
rate under mixed mode loading has been investigated
systematically. It is confirmed that the CCZM can agree
with experiments.
Keywords Mixed-mode fatigue crack propagation
Extended finite element method Cyclic cohesive zone
model Boundary layer formulation

Y. Xu H. Yuan (B)
Department of Mechanical Engineering,
University of Wuppertal, 42097 Wuppertal, Germany
e-mail: h.yuan@uni-wuppertal.de

1 Introduction
Since damage mechanics models confront difficulties
in computational convergence as well as in strain localizations, the cohesive zone model (CZM) has been
popular in simulating material failure (Scheider et al.
2006; Tan et al. 2005; Yuan and Cornec 1990; Yuan
et al. 1996), which provides an alternative way to
describe crack propagation. In these works the crack
propagations are only modeled for simple loading history, i.e. fracture under monotonic loading conditions.
The idea in monotonic fracture is that the material fails
as soon as the critical fracture energy is exceeded, while
fatigue cracks follow accumulative damage. The cohesive zone model has been extended to model fatigue
in recent years (Nguyen et al. 2001; Siegmund 2004;
Yang et al. 2001). Compared to the crack simulation
under simple loading history, the development of the
cyclic cohesive zone model (CCZM) is still in the early
stages. Both cyclic damage accumulation and mixedmode failure are involved in controversial discussions.
A cyclic cohesive model for engineering applications
needs substantial further understanding and validation
with experiments.
An additional limitation for the cohesive zone model
is related to the finite element method. In most of
the published works the crack path was pre-assumed
(Scheider et al. 2006; Tan et al. 2005; Yuan and Cornec
1990; Yuan et al. 1996). The conventional finite element method does not allow new crack surfaces into
an element. Therefore, investigations of cohesive zone

123

152

models are mainly limited to the cohesive tractionseparation law with zero threshold under mode I loading. However, in engineering applications a fatigue
crack initiates under the shear loading condition (Stage
I) and propagate further under the mode I condition
(Stage II) (Socie and Marquis 2000). In this sense
fatigue crack propagation is generally mixed-mode.
Especially in nonproportional multi-axial fatigue, the
crack direction can vary with crack growth, and in these
cases the crack path cannot be pre-assumed. Additionally, the cohesive traction-separation law calculated
from atomistic simulations implies a finite threshold
value for the cohesive zone (Krull and Yuan 2009). As
discussed in Zhai et al. (2004), it is difficult to simulate the curve crack propagation, even impracticable
for 3D mixed-mode crack problems using FEM (Chen
et al. 2005).
Recently several promising computational methods
for considering discontinuity (Oden et al. 1998; Zi
et al. 2005), generally referred as extended finite element methods (XFEM), have been developed based on
Melenk and Babuskas work on the approach of partition of unity (Melenk and Babska 1996). In these
type of methods, the discontinuity is introduced by
means of additional degrees of freedom (DOFs) on
those elements where the discontinuity crosses. The
most appealing feature is that the XFEM inherits the
finite element framework and its advantageous properties, such as sparsity and symmetry. Combining with
the XFEM, the cohesive zone model can be naturally incorporated to simulate mixed-mode crack propagation and has found wide application (Jirasek and
Zimmermann 2001; Mergheim et al. 2005; Meschke
and Dumstorff 2007; Wells and Sluys 2001; Simone
et al. 2003; Zi and Belytschko 2003). Using the XFEM,
one can consider crack initiation at an arbitrary material
point and crack propagation in an arbitrary direction,
without adding extra nodes and elements. Additionally,
it allows multi-cracks nucleation, growth and coalescence without remeshing.
Under mixed-mode loading conditions crack propagation depends on loading mode mixity and intensity.
The loading mode mixity affects crack direction, while
the intensity determines the crack growth rate. In the
past decades many different fatigue criteria were suggested based on experimental observations and fracture mechanics considerations (Qian and Fatemi 1996;
Richard 1984, 1985; Socie and Marquis 2000). A popular criterion was developed based on maximal principal

123

Y. Xu, H. Yuan

stress around the crack tip, which was applied in brittle


fracture analysis (Richard 1984, 1985). More experimental observations reveal that both fatigue and fracture of a cracked specimen share the same crack angle,
at least for finite/high cycle fatigue (Richard 1989;
Sander and Richard 2006). That is, one may use the
conventional fracture criterion to predict fatigue crack
direction. On the other hand, multi-axial fatigue investigation in uncracked specimens generally suggests
that fatigue failure is related to the shear stress/strain
amplitude and the normal stress on the shear plane.
A systematic review of different fatigue models can
be found in Karolczuk and Macha (2005), Socie and
Marquis (2000). As Findley (1956) model introduced
[ f = (/2 + kn )max ], the damage indicator for
multi-axial fatigue is based on shear stress/strain amplitude and normal stress. The index max denotes that
the damage indicator f reaches its maximum, i.e. the
critical plane, which gives the direction of material failure under mixed-mode condition. Obviously, the Findleys model differs from the maximal principal stress
criterion used in fracture mechanics.
In the present study, a modified boundary layer
model is used to investigate the correlation between
external loading mixity and crack propagation based
on the XFEM in combination with the CCZM. Crack
extensions under both simple and cyclic mixed-mode
loading are considered. Different fracture criteria are
evaluated. The feasibility of predicting fatigue crack
propagation behavior under mixed-mode loading based
on the CCZM is validated by known experiments.

2 Review of the extended finite element method


(XFEM)
2.1 The discontinuity in finite elements
Comprehensive overviews of discontinuity methods
in conjunction with partition of unity property have
been given by numerous publications (De Borst 2006;
Belytschko et al. 1988; Jirasek and Zimmermann 2001;
Mergheim et al. 2005). The XFEM, based on the
approach of partition of unity, is an efficient way to
reduce mesh dependency when it is used to analyze
crack growth. The key lies in that the discontinuity
field space can be directly approximated by introducing the enriched degrees of freedom and the enriched
basis function j . By combining the function j with

Using extended finite element methods

153

the standard basis function i , a good, approximate


property can be inherited from the standard basis function, and the specific discontinuous property is from
the enriched function j as well.
Note that the approximation is only locally enriched
in the region of interest such as the damage zone.
According to the approach of partition of unity, if the
standard basis function i fulfills the partition of unity
n
i = 1), the enriched field approximation
(i.e. i=1
can be expressed as following:

n
m


u=
(1)
i a i +
j bi j ,
i=1

j=1

where n is the number of nodes associated with the


standard basis function i , and j is the enriched function with m terms. ai is the vector of the conventional
degrees of freedom on node i, and bi j is the vector of
the enhanced degrees of freedom associated with node
i. Since DOFs are enriched on the existing nodes, modification of topology can be avoided.
In the finite element notation, the element shape
function N fulfils the partition of unity. If the discontinuity is contained in the finite element and the enriched
basis is taken as Heaviside function Hd , the approximation of displacement field can be written as
u = N(a + Hd b) = 

Na + Hd Nb,
 

standard

(2)

The whole FEM domain consists of two parts: + and


 , with the discontinuity d , as shown in Fig. 1.
In the formulation, only the nodes associated with
the discontinuity d have to be enriched. That is, the
extra enriched term in Eq. (2) will be considered for
these nodes. Otherwise, the standard shape function is
retrieved. By the enriched degrees of freedom, the discontinuity is introduced directly. Effectively, the standard DOFs, a are used to represent the continuum part,
while the enriched DOFs b, stand for a displacement
jump across the border d , expressed as
(4)

In a nonlinear analysis of the finite elements with the


discontinuity, the nodal displacements are computed
iteratively from the global equilibrium equations.

u
Fig. 1 An FEM domain crossed by the discontinuity d and
restrained by the applied boundary conditions. d does not exist
before the material failure starts

2.2 Governing equations


We consider a domain, , with a discontinuity, d ,
as shown in Fig. 1. Without taking body forces into
account, the equilibrium equations can be expressed as
following,
=0

in ,

(5)

where is the gradient operator and is the Cauchy


stress tensor. The essential and natural boundary conditions are, respectively provided as
m = t on d ,
u = u on u .

enriched

where a and b represent the standard and enriched nodal


degrees of freedom, respectively. The Heaviside step
function Hd is defined as

0 if x  ,
Hd =
(3)
1 if x + .

[u] = Nb |d .

n = t on t ,
(6)

Herein m is the inward unit vector perpendicular to the


discontinuity d and n is the outward normal vector
of the boundary. Correspondingly, t is the inner traction in d and t is the prescribed traction vector at the
boundary t . u is the prescribed displacement at the
boundary u .
Equation (5) can be converted into the governing
equation of the XFEM by using the Galerkin formulation. If the discontinuity d is not traction-free, due to
the cohesive zone for example, we may introduce an
additional function to connect + and  as
d .
= T(Nb)|
t = T[u]

(7)

The function T describes behavior of the discontinuity


d . Generally speaking, d is determined during computation and contains nonlinear features so that all of
the finite element equations are to be solved incrementally. Using the Galerkin method and considering the
nonlinear behavior of the discontinuity curve, the linearized governing equations of the XFEM (Wells and
Sluys 2001) can be obtained as

123

154

Kaa Kab
Kba Kbb




a
f
f
= a,ext a,int
fb,ext
fb,int
b

(8)

Monotonic Loading

with

0.8

fa,int =

Cyclic Loading (one cycle)

BT d,


fb,int =

BT d +
+

fa,ext =

NT td,

fb,ext =

NT td,

-0.2
0.0


BT DBd,

+

T
Kba = Kab
=

BT DBd,
+

BT DBd +
+

NT TNd.
d

In the linearized equation system (8), a and b


denote the increment of nodal displacements in an
incremental step. In addition, D is the material stiffness matrix; T is the cohesive stiffness matrix and B is
the strain matrix.
In the present work the crack tip is located at the
end of the cohesive zone so that the stress at the crack
tip becomes non-singular. One gets the same numerical accuracy by the XFEM without using special shape
functions.

3 Cyclic cohesive zone models for fatigue crack


growth
3.1 The cohesive law for fracture
Through experiments it has been proven that the crack
propagation is dominated by the mode I loading case in
both brittle and ductile material for cracked specimens
(Richard 1985, 1989). Based on this consideration, the
failure is dominated by the normal traction, Tn (Fig. 2).

123

1.0

2.0

3.0
n

4.0

5.0

6.0

Fig. 2 The cyclic cohesive zone model under pure normal separation. Under the simple loading condition the cohesive law
is identical with the model of Xu and Needleman (1994), as
0 = 0 in (11). Under unloading and reloading the traction
decreases/increases linearly with separation. The material damage is characterized by diminishing tensile strength Tn together
with decreasing material stiffness

BT DBd,

Kbb =

0.4

Kab =

Degradation

0.2

NT td,

t+

Kaa =

0.6

d

t

and

T n / max,0

Y. Xu, H. Yuan

Therefore, the cohesive strength is considered to be


an exponential function of the separation as suggested
by Xu and Needleman (1994), while the shear traction
Tt is assumed to be linearly related to the tangential
separation. Under monotonic loading conditions, the
relationship between tractions and separations in the
cohesive law is
 written as 
n
n
t 2
exp 2 ,
(9)
Tn =
0
0
0
t
Tt = G t ,
(10)
0
where
n = n + 0

(11)

denotes the effective normal separation. n is the normal separation from the FEM computation. 0 stands
for a model parameter to consider the threshold value
of the cohesive traction. For 0 = 0 the model recurs
to the original version of the model of Xu and Needleman (1994). G t is the shear stiffness and t is the tangential separation. The characteristic cohesive length is
denoted by 0 . The fracture energy, n , resulting from
normal separation for failure is calculated as
n = emax 0 ,

(12)

wherein the Eulers number e = exp(1), and max is


the cohesive normal strength of material. In the crack

Using extended finite element methods

simulation under simple monotonic loading conditions,


the parameter max is a model parameter to fit experimental curves, together with 0 (Scheider et al. 2006).
Under cyclic loading conditions the material may fail
at a significantly lower loading level than the material strength. This implies that max decreases with the
cyclic loading history.

3.2 The cyclic fatigue damage evolution


A CCZM should be able to describe both fracture
under monotonic loading and fatigue failure, i.e. due
to accumulative damage (Nguyen et al. 2001; Roe and
Siegmund 2003; Siegmund 2004). The accumulative
damage will be described by an additional damage
parameter, Dc , which is determined by a damage evolution equation for the cyclic loading process, as suggested by Roe and Siegmund (2003) and Siegmund
(2004)


f
Tn
| n |
H( 0 )

(13)
D c =
max
max,0

| n | dt. In the above
with D c 0 and =
expression H represents the Heaviside function which
prescribes that the damage accumulation starts once
the accumulated material separation is greater than
the characteristic length 0 . The cohesive strength is
affected by Dc , which can be written as
max = (1 Dc )max,0
with max,0 as the initial cohesive strength. Obviously,
this consideration is a direct extension of continuum
damage mechanics (Lemaitre 1996). Additionally, f
is the cohesive zone endurance limit, and is the accumulated cohesive length which scales the increment of
material separation n .
In this investigation, the unloading and reloading
are assumed to proceed along a current stiffness, kn =
max,0 (1 dc )e/0 , which is equal to the slope of the
current cohesive curve at zero separation. This assumption will lead to the presence of a residual separation.
The material damage is presumed to be accumulating
during the whole process of unloading and reloading,
except for the case of compression, i.e. n < 0. As for
the potential compressing case, from a viewpoint of
numerical treatment, the material penetration at the discontinuity is reduced by enhancing the material compression stiffness.

155

4 Implementation of the XFEM into ABAQUS


The XFEM (8) combined with the cyclic cohesive
model (9)(13) has been implemented within ABAQUS via the user element subroutine (UEL) (ABAQUS
2006). The subroutine UEL allows the user to define a
new element in which all variables, e.g. strains, stresses
and residual forces as well as the element stiffness
matrix, have to be passed in and updated for a new
iteration. In the present work, the degrees of freedom
at enriched nodes are covered by some additional variables to include displacement discontinuity within an
element. In the present UEL subroutine the main following functions are achieved: (1) updating strains and
stresses; (2) generating an element stiffness matrix; (3)
determining tractions and separations in accordance
with cohesive law; (4) defining crack initiation and
propagation; (5) creating data for post-processing. For
simplicity only two kinds of discontinuity patterns are
implemented, as shown in Fig. 3. That is, a cohesive zone can go through two opposite element edges
(Fig. 3a) or through two adjacent edges of one element
(Fig. 3b). All other crack paths can be represented by
these two patterns.
The integration scheme is crucial in the implementation of the XFEM. Here, an intersected element is
divided into 8 triangular zones and integrated based
on 24 integration points, together with two integration
points for the discontinuity curve, as shown in Fig. 3.
Note that the partition of the element is merely for
integration based on iso-parametric interpolations over
all the triangles, while no additional node is needed.
Additionally, to reduce computational efforts, only the
potential damage zone is covered with user-defined elements, otherwise the conventional isotropic elements
are adopted. Connections between two zones are realized without special algorithms.
In the XFEM programme, another two built-in
ABAQUS subroutines are adopted. Using subroutine
UEXTERNALDB, one can exchange data between
external databases and user subroutines at the required
steps. Based on the subroutine, initial user-defined
parameters can be passed into subroutine UEL, and the
results obtained from the user-defined elements can be
written to the output file. The other important builtin subroutine is URDFIL by which the results of the
general elements from ABAQUS (e.g. element type
CPE4) can be explored. With these two subroutines,
the user-defined elements and ABAQUS elements can

123

156

Y. Xu, H. Yuan

(a)

(b)

Fig. 3 Two types of intersected elements implemented in ABAQUS. The middle line with two dots denotes the discontinuity in
the XFEM-element. a The discontinuity divides a quadrilateral
into two sub-quadrilaterals. b The discontinuity separates an element into one triangle and another pentagon. In both cases the

enhanced elements are divided into 8 sub-triangles for integration. Each sub-triangle possesses three integration points (open
circles). The discontinuity curve is integrated using two integration points (solid circles)

be coupled to calculate and export results. Since the


post-processor of ABAQUS does not support the userdefined elements, the output database (ODB) has to be
modified by the user in order to visualize the results.
In the present work, a Python script is used to produce an ODB file which is compatible with the ABAQUS/Viewer.

inner crack will not be significantly affected so that one


may correlate crack propagation with external loading parameters K I and K I I . The modified boundary
formulation method states that the desired load case
with respect to different combination of K I and K I I
fields can be directly applied on the external boundary.
The displacement-controlled loads can be obtained by
superposing mode I and mode II displacement fields
from fracture mechanics, as

5 Computational models

The XFEM and CCZM discussed previously have been


applied to the investigations of three-point and fourpoint bending cases. The results are in good agreement with the experimental observations (Xu and Yuan
2009a,b). In order to understand further the fracture and
fatigue mechanisms, the modified boundary layer formulation is investigated. In fracture analysis, the modified boundary layer formulation is popular to identify
the relation between loading mixity and crack tip field
(Anderson 1995; Larsson and Carlsson 1973; Yuan
2002). In this case both loading mixity and intensity
can be characterized by K I and K I I in the remote
field. In the present work the modified boundary layer
formulation is used to predict crack angles and crack
growth rate under the given mixed-mode loading. For
a small amount of crack growth the loading mixity of

123

the displacements from mode I:




r
KI
cos
G(1 +  ) 2
2

,
1  + (1 +  ) sin2
2


r
KI
sin
=

G(1 + ) 2
2



2
2 (1 + ) cos
;
2

uIx =

uIy

the displacements from mode II:


uI Ix


r
KI I

=
sin
G(1 +  ) 2
2

,
2 + (1 +  ) cos2
2

(14)

Using extended finite element methods

uI Iy

6 Results and discussion




r
cos
1 + 
2
2

.
+(1 +  ) sin2
2

KI I
=
G(1 +  )

157

(15)

Wherein, G is shear modulus and  = /(1 ) with


as Poissons ratio. The applied displacements, u x and
u y , are obtained by superposing u I with u I I like so:
ux = u I x + u I I x ,
uy = uIy + uI Iy

(16)

at
 the remote boundary with the radius of r0 =
x02 + y02 and the polar angle = tan1 (y0 /x0 ). The
initial crack tip is located at the origin of the geometry (r = 0). The cyclic loading is imposed by varying
the stress intensity factors, K I and K I I . For simplicity we only consider pulsating load with the loading
ratio R = K min /K max = 0 so that K = K max . The
range of stress intensity factor K is correlated with
the incremental energy release rate G as
G =

(1 2 )K 2
.
E

(17)

The global finite element mesh and the boundary conditions are plotted in Fig. 4. To reduce the artificial
effect of K fields on the crack, crack propagation is
confined to a small region, and its area is negligible in
comparison to the whole specimen (less than 1%). The
radius of the crack extension region, l, is smaller than
5%r0 , where r0 represents the radius of the boundary
layer model. As shown in Fig. 4b, the mesh is highly
refined in the crack extension region.
Computations were performed under the plane strain
condition in ABAQUS, in which the CPE4-type element was employed, together with the user-defined
elements discussed in the previous sections. To save
computational time, only those elements near the crack
path were defined as the user-defined elements.
The material properties of the specimen are normalized by yield stress 0 which is not needed in elastic
materials. Accordingly, elastic modulus E = 3000 ,
and Poissons ratio = 0.3. For the cohesive model,
the cohesive strength and length are taken as max,0 =
6.70 and 0 = 0.0153 mm, respectively. In this investigation, in accordance with a suggestion from Roe and
Siegmund (2004), the parameters f and are set to
0.25max,0 and 40 , respectively.

6.1 Crack kinking angles


It is a common postulate that a crack in brittle materials propagates toward mode I, i.e. in the direction
following K I I = 0. Among known fracture criteria,
the maximum tangential stress (MTS) criterion and the
maximum energy release rate (MERR) criterion are
most popular, and have been widely applied in fracture mechanics analysis (Anderson 1995). Since the
philosophy of fracture mechanics differs from that of
fatigue assessment, a criterion for a fatigue crack could
be distinctive. Under multi-axial loading conditions,
for instance, the shear stress amplitude is responsible
for fatigue failure and the fatigue crack follows the socalled critical plane of the damage indicator (Socie and
Marquis 2000).
In the present section we are computing fatigue
crack growth based on various criteria with verification
of experimental data from literature. Richard (1985)
and Sander and Richard (2006) carried out a number of
tests for the mixed-mode crack extension under the simple or cyclic loading cases. Most tests were based on the
Compact Tension Shear (CTS)-specimen which was
put forward first by Richard (1984, 1985, 1989). With
this specimen the mode-mixity in fracture mechanics
can be effectively and conveniently calculated, making
it easy to evaluate the crack extension by the fracture
mechanics method. Different loading mode mixities
can be approached by varying the loading angle () of
the specimen in 15 -steps, from = 0 to 90 . Pure
Mode I ( = 0 ), pure Mode II ( = 90 ) or mixedmode( = 15 , . . . , 75 ) situations can be generated
by the use of just a uniaxial testing system. The details
can be obtained from Richard (1984, 1985, 1989).
6.1.1 Maximum tangential stress criterion
The MTS criterion is one of the most popular fracture
models for material strength. It was proposed by Erdogan and Sih (1963) as



3
( )|=0 = K I 3 cos + cos
2
2



3
3K I I sin + sin
(18)
2
2
max
for the angular function of the circumferential stress
component . According to this criterion, the crack

123

158

Y. Xu, H. Yuan

Fig. 4 The finite element


mesh for the modified
boundary layer formulation
(MBLF). The remote
boundary (r = r0 ) is loaded
by the given displacements
according to the plane strain
K -fields in fracture
mechanics. a The whole
mesh. b The refined mesh
around initial crack tip

will advance along the radial direction = 0 , in which


the circumferential stress becomes maximum, i.e.
/ = 0. It follows





 
1
KI 2
1 KI
0

=
+ 8.
(19)
tan
2
4 KI I
4
KI I
In this criterion, K I and K I I are loading parameters and
are directly used to determine the cracking orientation.
This criterion has been verified by many experimental
results.

cohesive zone tip and within a prescribed radius. A


weighted Gaussian function (Mergheim et al. 2005) is
defined as
w(r
)
with
w(r ) = n gp
i Ai
i=1 w


1
r 2
.
(20)
w(r
) = exp
[2]2
2

6.1.2 Nonlocal maximum principal stress criterion

A nonlocal treatment becomes significant if material


failure is related to a certain material volume, but not
affected by the stress state at a single mathematical
point. From discussions on nonlocal damage mechanics we learn that the local damage assumption may
lead to a paradox in computations. A nonlocal criterion accounts for a weighted averaging of the damage
over a certain area, which can reduce effects of the
local stress distribution. That is, the nonlocal criterion
may diminish the influence of the finite element size.
For this reason, the maximum principal stress criterion
based on the averaging method is becoming popular in
the computational mechanics community (Mergheim
et al. 2005; De Borst 2006). In the present work we use
the nonlocal maximum principal stress criterion to predict mixed-mode crack growth, as suggested by Wells
and Sluys (2001).
The nonlocal stress is computed as a weighted average of the stress over n gp Gaussian points around the

123

The nonlocal stress is calculated by the following


expression,
n gp


i wi Ai .

(21)

i=1

In the above equations r is the distance of the Gaussian point to the cohesive zone tip and denotes the
radius of the nonlocal zone. The nonlocal stress tensor
results from the sum of the local stresses at the Gaussian points i, the weight function wi and the associated
area Ai . In the 2-dimensional case the orientation of
maximum principal stress can be calculated from the
nonlocal stresses as
2x y
tan(20 ) =
.
(22)
x y
In the XFEM computations the maximum principal
stress criterion includes two aspects: propagation of
the cohesive zone, and direction of the cohesive zone
propagation. During computations the maximum principal stress ahead of the cohesive zone tip is calculated
and checked against the fatigue limit of the material.
Once it exceeds the fatigue limit, the enhanced degrees
of freedom in this element are activated. That is, the

Using extended finite element methods

cohesive zone propagates into this element. As regards


the direction of crack extension, the criterion states that
the crack will grow in the perpendicular direction to that
of nonlocal maximum principal stress.
The nonlocal maximum principal stress differs from
the local maximum principal stress just due to averaging over a certain region. In XFEM the NMPS is easier
to realize in comparison with the MTS criterion.
In Fig. 5 the crack paths of two mixed-mode loading cases are plotted, and they are predicted based on
the NMPS criterion. The loading ratios, K I I /(K I +
K I I ), for these two cases are 0.22 and 0.47, respectively, which correspond to the loading orientations of
the CTS-specimen (Richard 1984, 1985), = 30 and
= 60 . Whereas K I I /(K I + K I I ) = 0.22 shows
a straight crack path, the higher mode II, K I I /(K I +
K I I ) = 0.47, causes varying kinking angles after crack
initiation. With crack growth the angle increases from
0 to 1 and finally reaches a steady state.
Figure 6a summarizes the steady crack angles as
a function of the loading ratio. The computations are
based on three different criteria: MTS, NMPS and
MERR. The experimental results were taken from the
works of Richard (1984, 1985, 1989). All criteria fall
in a small scatter band in the whole loading ratio and
agree with experimental results (Richard 1984, 1985,
1989). The predictions based on the MTS and MERR
criteria come from the analytical solution (Anderson
1995). The results of the nonlocal maximum principal
stress criterion (NMPS) were acquired from XFEM so
that the crack path can be characterized in detail.
The predictions from MTS and MERR assume a
straight crack, while the XFEM computational analysis
shows a curved crack path. In Fig. 6b the crack initiation
angles 0 calculated from both XFEM computation and
analytical solution are summarized, as well as the stabilized crack angles 1 from the XFEM computation,
and the experimental results. After a crack initiates with
kinking, the stress field around the crack tip changes.
It follows that the crack no longer propagates in the
initial direction. The computations show that the crack
initiation angles are generally significantly smaller than
the experimental data, while the stabilized angles agree
with experimental observation. Additionally, as shown
in Fig. 5b, the transient phase of crack initiation in
mixed-mode cracking is too small to be measured in
experiments. However, it can be observed from our
numerical XFEM calculation. To validate that the initial
kinking is not induced by numerical errors, the analyt-

159

ical solutions are further derived from the mixed-mode


analytical crack stress fields and maximum principal
stress criterion. The calculations indicate that the analytical solutions are in good agreement with the XFEM
results, as shown in Fig. 6b.
6.1.3 The modified Findleys multi-axial fatigue
criterion
Multi-axial fatigue criteria such as Findley (1956),
Fatemi-Socie and Marquis (2000) and McDiarmid
(1994), etc. have been proposed for predicting fatigue
failure. Combined with the critical plane concept, the
multi-axial fatigue criteria try to predict crack propagation direction (Karolczuk and Macha 2005; Socie
and Marquis 2000). Some experimental observations
in multi-axial fatigue seem, however, not to agree with
fracture mechanics predictions.
Development of multi-axial fatigue life assessment
relies mainly on extensive experimental observations
(Karolczuk and Macha 2005; Socie and Marquis 2000).
Life models are trying to give a prediction of component durability in service, without detailed information about crack initiation and crack propagation.
Crack growth is caused by cyclic loading and will not
be described explicitly by fracture parameters, but by
the macro-stresses and strains. Therefore, fatigue differs from fracture significantly. Based on experimental
observations Sines (Socie and Marquis 2000) predicted
that the multi-axial fatigue failure is related to the shear
stress amplitude and the total normal stress, but is independent of total shear stress and less dependent on the
normal stress amplitude. It turns out that the fatigue
damage indicator is a function of the shear stress amplitude,  , and the normal stress at the shear plane, n .
Using the critical plane concept one can predict the failure direction (Socie and Marquis 2000). Such considerations have been further extended to both high cycle
fatigue and low cycle fatigue. This concept seems not to
be compatible with the mixed-mode fracture mechanics. In this section we are implementing the Findleys
model to re-analyze crack propagation under mixedmode loading conditions.
Findley took the idea of Sines (Socie and Marquis
2000) and assumed that the material failure is characterized by the following parameter (Findley 1956),



+ n
.
(23)
f =
2
max

123

160

Y. Xu, H. Yuan

Fig. 5 Crack paths


predicted by the NMPS
criterion: a
K I /(K I + K I I ) = 0.22
corresponds to the loading
orientation = 30 ; b
K I /(K I + K I I ) = 0.47
corresponds to = 60 .
The crack angle changes
from the initiation angle 0
to the steady-state angle 1

(a)

(b)
O
O

70

60

60

50

50

()

70

()

(b) 80

40
30

40
30

Experimental value
MPS steady comp. solu.
MTS analy. solu.
MERR analy. solu.

20
10
0

(a) 80

0.2

0.4

0.6

0.8

Experimental Value
MPS initial comp. solu.
MPS steady comp. solu.
MPS analy. solu.

20
10

0
0

0.2

0.4

KII/ (KI+KII)

0.6

0.8

KII/ (KI+KII)

Fig. 6 Crack propagation angles under different mixed-mode


loading conditions, predicted by different criteria: MTS, NMPS
and MERR. Experimental data are taken from the works of
Richard (1984, 1985). a The stabilized angle as a function of

the loading mixity. b Comparisons between the initiation angle


0 and the steady-state angle 1 based on analytical and computational analyses using nonlocal MPS criterion

In Findleys multi-axial fatigue model the normal stress


is introduced in order to consider the effects of the mean
stress, primarily for smooth specimens. Usually is
significantly smaller than 0.5 which implies that fatigue
is dominated by the shear stress amplitude. The criterion is effective for a proportional combination of
tension and torsion loading configurations (Socie and
Marquis 2000). The damage occurs at the plane with
the biggest value of f , the critical plane. According to
Findleys model (Socie and Marquis 2000) the critical
plane depends on the direction of the shear stress amplitude, the mean normal stress and the coefficient . For
ductile materials is small and the position of the critical plane approaches the maximum shear stress. For
brittle materials is large and the critical plane turns
perpendicular to the maximum principal stress.

To use the Findleys model for cracked specimens,


one has to reformulate the definition of the damage
indicator (23). To consider normal stress and shear
stress simultaneously, we re-write Eq. (23) into

123

f = (an + b ) max .

(24)

Both a and b are defined as material coefficients so that


the effects of both shear stress amplitude and normal
stress can be adjusted simultaneously. The expression
identifies a critical plane for fatigue crack initiation and
propagation depending on both alternated shear stress
and maximum normal stress at that plane. The crack
will propagate along the critical plane on which the
resultant value of these two terms in Eq. (24) is maximal and surpasses the failure limit f . Observing this
expression, one finds that this fatigue criterion tends

Using extended finite element methods

161

80
Criteria: (a n +b )max= f

60
40

()

20
0
Experiment
a=0.0,b=1.0
a=0.3,b=0.7
a=0.5,b=0.5
a=0.7,b=0.3
a=1.0,b=0.0

-20
-40
-60

0.2

0.4

0.6

0.8

KII/ (KI+KII)
Fig. 7 Crack angle as a function of the loading mixity predicted
by the modified Findleys model (24)

to approach the maximum principal stress criterion in


fracture mechanics as b goes to zero.
Figure 7 shows crack angle as a function of the
loading mixity computed using the XFEM with various values for a and b. The results confirm that the
crack angle is sensitive to both parameters a and b. As b
increases the effect of shear stress rises. That means the
crack angle will kink even under mode I loading conditions, which seems not to coincide with the known
experimental observations in a pre-cracked specimen
(Richard 1989). On the other side, effects of a and b
on the predicted crack angle are nonlinear. For b < 0.5
and a > 0.5 the crack angle does not deviate from the
known experimental results significantly.
6.2 Fatigue crack propagation rates
Siegmund (2004) has used the cyclic cohesive model
for fatigue crack growth in various specimens and load
cases. Their extensive investigations indicate the feasibility of CCZM for life prediction, and the computations reproduce Paris law in cracked specimens based
on the incremental energy release rate G. However,
their discussions are limited to some special cases due
to the limitations of FEM mentioned in the Sect. 1.
Based on the current proposed method, the computational prediction can be extended to more general
mixed-mode cases. In this section, the fatigue crack
propagation rate will be further investigated via the
modified boundary layer formulation.

Firstly, the pure mode I loading is considered, illustrating the detailed procedure of fatigue life prediction
and investigating the validity in terms of applying the
XFEM combined with CCZM in fatigue life assessment. In Fig. 8a, the approximately linear relationship between the crack extension and loading cycle
number in the double-logarithmic coordinate system
is depicted. The slopes of these curves (i.e. crack propagation rates) depend on K I . As K I increases the
crack growth rate rises. On the basis of these data, the
crack propagation rates can be expressed as a function
of G/n according to Expression (17). As shown in
Fig. 8b, it can be seen that d(a/0 )/d N is linearly
dependent on G/n in double-logarithmic coordinates, which can be curve-fitted in Paris law


G m
d(a/0 )
=C
(25)
dN
n
with C = 67.35 and m = 2.16. Paris law only
describes stationary fatigue crack growth, whereas the
CCZM also can predict fatigue crack growth in low and
high loading amplitudes. By introducing the threshold,
f in the damage evolution Eq. (13), one may shift
the fatigue limit in the life curve. That is, the fatigue
damage predicted in the XFEM approaches a threshold
below which the material degradation will not occur.
On the other hand, the normalized energy release rate
range possesses an upper limit which is characterized
by the general cohesive law without cyclic damage.
The traction degradation processes on three material
points ahead of the initial crack tip under K I loading
conditions are recorded in Fig. 9, which shows different damage processes in a cracked specimen. All
three points are located on the crack path along the
initial crack direction, as shown in Fig. 4. Point A is
at the initial crack tip. Point C is 10 elements away
from Point A, i.e. approximately, 0.3 l (l denotes the
length of the refined region). B is located in the middle
of A and C. Since A is located at the initial crack tip
where the stress concentration occurs initially and the
stress level is highest, the cohesive zone goes through
A earlier than the other points. Once the criterion of
cohesive zone growth has been satisfied, the discontinuity will immediately be introduced into these elements ahead of the crack tip. Thereafter, with the cyclic
damage accumulation and the cohesive zone propagation, the loading capacity of the material point begins
to sink and the material degrades gradually. As shown
in Fig. 9, Point A experiences the highest peak traction,

123

162

Y. Xu, H. Yuan

(b) -0.5

(a) 35
K I=2.00
KI=2

25

KI=2.5
K I=2.50

20

KI=2.75
K I=2.75

d (a / 0 )
G
= 67.35
dN
n

-0.7

log(d(a/ d 0 )/dN)

a/ d 0

-0.6

KI=2.25
K I=2.25

15

30

2.16

-0.8
-0.9
-1
KII=0
KII =0

10

-1.1
5
0

-1.2
0

50

100
150
N (cycles)

-1.3
-1.5

250

200

-1.4

-1.3
-1.2
log(G/f n)

-1.1

-1

Fig. 8 Correlation between crack growth and mode I loading: a variation of crack growth with loading amplitude K I ; b dependence
of crack growth rate on G

(a) 0.8

(b) 1

0.7

point A
point B

0.6

0.8

Dc at B

Dc at C
0.8

Tn at A

point C

Tn / max,0

0.4
0.3
0.2

Tn at B

0.6

Tn at C

0.4

0.4

0.2

0.2

Dc

0.6

0.5

Tn / max,0

Dc at A

0.1
0

0
-0.1
0.17

-0.2
0.27

0.37

0.47

0.57

n / 0

20

40

60

80

100

-0.2

Cycles (N)

Fig. 9 Evolution of material damage predicted by XFEM: a the predicted traction-separation responses for three points A, B, C along
the crack path; b the relationships between traction and damage evolution of the three points

Tn /max,0 0.75, which is reached in the first cycle,


and subsequently it decreases gradually with loading
cycles, till complete failure.
The material separation processes at B and C are
distinctly different from that at A. Both Fig. 9a, b indicate that the material damage process at B is similar to
that at C. Their traction first experiences the hardening
process, then drops as they form the envelope curves.
In both cases the peaks of traction are nearly the same,
Tn /max,0 0.5, and the shapes of the envelope curves

123

are also similar, except that the process of Point C is


postponed. The similarity of the traction losing process
for Points B and C indicates that the damage process
tends to be steady after a small crack growth phase.
These results are in good agreement with experimental
observation, and the steady state propagation of fatigue
crack indicates the features of Paris law.
Figure 9 additionally indicates that the damage indicator Dc grows over-proportionally due to the interaction of traction and separation, which is mathematically

Using extended finite element methods

163

(a) -0.2

(b) 1.2

-0.4

1.2

KI /KII =2

KI /KII =1

KI /KII =0.5
1

-0.8
-1
-1.2

-1.6
-1.8
-1.6

-1.5

-1.4

-1.3

-1.2

-1.1

0.8

KII =0
0.6

0.6

0.4

0.4

KII=0
KI=0
KI/KII=1
KI/KII=2
KI/KII=0.5

-1.4

KI =0

0.8

|max
MTS
Analy. Solu.
Crack growth rate

0.2

-1

-0.9

0.2

log (G/fn)

0.4
0.6
KII/ (KI+KII)

Scaled crack growth rate

Scaled s |max

log(d(a/d0)/dN)

-0.6

0.2

0.8

Fig. 10 Computational correlations under mixed-mode fatigue


conditions: a the fatigue crack growth rate versus the applied
energy release rate range; b the scaled |max obtained from

crack tip field according to the MTS criterion versus the scaled
crack growth rates at log(G/n ) = 1.25

described in Eq. (13). Furthermore, comparing the variations of Dc over Points A and C, it can be found that
when the material point A is just completely damaged,
i.e. Dc = 1, Dc on C is just setting out, meaning that
the cohesive zone is just now embedded in this element.
This result signifies that the length of the cohesive zone
in this investigation is the distance between A and C.
Based on experimental observations (Richard 1984,
1989; Sander and Richard 2006), the mixed-mode
fatigue crack in brittle materials or high cycle fatigue
is searching in the direction of mode I and the crack
propagation is dominated by the mode I stress field.
In this sense, understanding of the mode I crack is
essential for understanding of mixed-mode cracks. To
examine the validity of the proposed method for mixedmode fatigue, various combinations of K I and K I I
have been considered. In Fig. 10a, fatigue crack growth
rate, d(a/0 )/d N , is plotted with respect to the normalized energy release rate range (G/n ), under various mixed-mode loadings. For the mixed-mode case,
G = G I + G I I is introduced.
We find a linear correlation between log[d(a/0 )/
d N ] and log(G/n ) for all cases analyzed with a
unique exponent m of Paris law (25), regardless of
mode-mixity. That is, even for the mixed-mode fatigue
cracking there exists a Paris law as, d(a/0 )/d N =
C(G/n )m , in which m is independent of mode-mixity, but the parameter C varies with K I I /(K I + K I I ).

The computations show that the crack growth rate in


terms of pure K I I is evidently higher than that of pure
K I . The maximum rate is reached under mixed-mode
loading. This is rather unexpected, because one would
usually assume the maximum rate under the mode I
loading case, due to normal stress failure assumed in
the model.
To clarify this variation, the crack growth rate for
a given loading intensity, log(G/n ) = 1.25, is
plotted as a function of loading mixity in Fig. 10b.
For comparison, the maximum circumferential stress
|max in accordance with the MTS criterion is calculated and plotted in Fig. 10b. Note both curves are
scaled by their maximum values and show similar variations with K I I /(K I + K I I ). If one accepts the MTS
assumption for the mixed-mode, the highest circumferential stress under the given loading intensity appears
at around K I I /(K I + K I I ) 0.5, and so does the
cracking velocity. From the curves one may conclude
that the maximum crack propagation rate has to occur
under a mixed-mode condition.

7 Conclusions
In the present paper the formulation of the XFEM
and the CCZM have been discussed and applied for
the analysis of mixed-mode fatigue cracking. Both

123

164

implementation and verification of CCZM in combination with XFEM in the frame of ABAQUS have been
reported. Application of XFEM allows displacement
discontinuities within finite elements so that the curved
crack propagation can be simulated with less effort.
The combination of XFEM and CCZM possesses an
appealing ability to predict the fatigue crack curvilinear extension under mixed-mode loading.
Based on the modified boundary layer formulation, a
detailed investigation has been conducted regarding the
crack propagation direction under mixed mode loading. Compared with known experimental results, the
nonlocal stress based maximum principal stress criterion of the XFEM accurately predicts crack propagation behavior in brittle materials. Moreover, the
computation shows that the initial crack kinking angle
differs from the stabilized crack direction. The transient region, however, is very small. These computations agree with known experimental data.
Additional computations based on a modified Findleys multi-axial fatigue criterion reveal that Findleys
model approaches experimental observations only with
very small shear stress contribution. This implies that
this kind of criterion is only suitable for considering
shear stress contribution, for fatigue crack propagation
in ductile materials.
In fatigue analysis the crack propagation rate is
another important issue. In this context the fatigue crack
rate prediction under mixed mode loading in the framework of XFEM combined with CCZM has been investigated as well. The numerical simulation shows that
the computational results match the known form of
Paris law under mixed mode condition. Additionally,
simulations verify the proposed method is able to simulate the crack initiation and steady propagation process for very low cycle problems. The computational
results in fatigue simulation coincide with the experimental records in higher loading levels. For finite cycle
fatigue the evolution of damage accumulation predicts over-proportional increment of damage indicator.
This implies that the damage evolution equation needs
further modifications.
Although the results obtained so far are important and many experimental phenomena can be
simulated properly, the main difficulties for the
application of the CCZM are in the formulation of
the cyclic damage evolution equation as well as in the
description of the shear stress failure. Especially in ductile materials the shear stress becomes significant in

123

Y. Xu, H. Yuan

fatigue and cannot be neglected. Additionally, the high


cycle fatigue problem needs a robust method beyond
cycle-by-cycle computations. For engineering applications more detailed experimental and computational
verifications are necessary.
Acknowledgments The present work is financed by the
German Science Foundation (DFG) under the contract number
YU119/5-1.

References
ABAQUS (2006) Theory manual. Version 6.6. ABAQUS Inc.
Providence, R. I.
Anderson TL (1995) Fracture mechanicsfundamentals and
applications, 2nd edn. CRC Press, Boca Raton
Belytschko T, Fish J, Engelmann BE (1988) A finite element
with embedded localization zones. Comp Meth Appl Mech
Eng 70:5989
Chen CR, Kolednik O, Heerens J, Fischer FD (2005) Threedimensional modelling of ductile crack growth: cohesive
zone parameters and crack tip triaxiality. Eng Fract Mech
72:20722094
De Borst R (2006) Modern domain-based discretization methods for damage and fracture. Int J Fract 138:241262
Erdogen F, Sih GC (1963) On the crack extension in plates under
plane loading and transverse shear. J Bas Eng ASME Trans
85:519525
Findley WN (1956) Modified theories of fatigue failure under
combined stress. Proc Soc Exp Stress Anal 14:3546
Jirasek M, Zimmermann T (2001) Embedded crack model. Part
I: basic formulation. Int J Num Meth Eng 50:12691290
Karolczuk A, Macha E (2005) A review of critical plane orientations in multiaxial fatigue failure criteria of metallic
materials. Int J Fract 134:267304
Krull H, Yuan H (2009) Suggestions to cohesive models based
on atomistic simulations. Eng Fract Mech. Submitted for
publication
Larsson SG, Carlsson AJ (1973) Influence of non-singular stress
terms and specimen geometry on small-scale yielding at
crack tips in elastic-plastic materials. J Mech Phy Solids
21:263277
Lemaitre J (1996) A course on damage mechanics. Springer,
Berlin
McDiarmid DL (1994) A shear stress based critical-plane criterion of multiaxial fatigue failure for design and life prediction. Fat Fract Eng Mat Struct. 14751485
Melenk JM, Babska I (1996) The partition of unity finite element method: basic theory and applications. Comp Meth
Appl Mech Eng 139:289314
Meschke G, Dumstorff P (2007) Energy-based modeling of
cohesive and cohesionless cracks via X-FEM. Comp Meth
Appl Mech Eng 196:23382357
Mergheim J, Kuhl E, Steinmann P (2005) A finite element
method for the computational modelling of cohesive cracks.
Int J Num Meth Eng 63:276289
Nguyen O, Repetto EA, Ortiz M, Radovitzky RA (2001) A cohesive model of fatigue crack growth. Int J Fract 110:351369

Using extended finite element methods


Oden JT, Duarte CA, Zienkiewicz OC (1998) A new cloudbased hp finite element method. Comp Meth Appl Mech
Eng 153:117126
Qian J, Fatemi A (1996) Mixed mode fatigue crack growth: a
literature survey. Eng Fract Mech 55:969990
Richard HA (1984) Some theoretical and experimental aspects
of mixed mode fracture. In: Valluri SR, Taplin DMR, Rama
Rao P et al., (eds) Advances in fracture research. Pergamon
Press, Oxford
Richard HA (1985) Bruchvorhersage bei berlagerter Normalund Schubbeanspruchung von Rissen. VDI, Dsseldorf
Richard HA (1989) Specimens for investigating biaxial fracture
and fatigue processes. In: Brown MW, Miller KJ (eds) Biaxial and multiaxial fatigue. Mechanical Engineering Publications, London pp 227229
Roe KL, Siegmund T (2003) An irreversible cohesive zone
model for interface fatigue crack growth simulation. Eng
Fract Mech 70:209232
Sander M, Richard HA (2006) Experimental and numerical
investigations on the influence of the loading direction on
the fatigue crack growth. Int J Fatigue 28:583591
Scheider I, Schdel M, Brocks W, Schnfeld W (2006) Crack
propagation analyses with CTOA and cohesive model:
comparison and experimental validation. Eng Fract Mech
73:252263
Siegmund T (2004) A numerical study of transient fatigue crack
growth by use of an irreversible cohesive zone model. Int J
Fatigue 9:929939
Simone A, Wells GN, Sluys LJ (2003) From continuous to discontinuous failure in a gradient-enhanced continuum damage model. Comp Meth Appl Mech Eng 192:45814607
Socie DF, Marquis GB (2000) Multiaxial fatigue. SAE International, Warrendala
Tan H, Liu C, Huang Y, Geubelle PH (2005) The cohesive law
for the particle/matrix inter-faces in high explosives. J Mech
Phy Solids 53:18921917

165
Wells GN, Sluys LJ (2001) A new method for modelling cohesive cracks using finite elements. Int J Num Meth Eng
50:26672682
Xu XP, Needleman A (1994) Numerical simulations of fast
crack growth in brittle solids. J Mech Phy Solids 42:1397
1434
Xu YJ, Yuan H (2009a) Computational analysis of mixedmode fatigue crack growth in quasi-brittle materials using
extended finite element methods. Eng Fract Mech 76:
165181
Xu YJ, Yuan H (2009b) On damage accumulations in the cyclic
cohesive zone model for XFEM analysis of mixed-mode
fatigue crack growth. Comp Mat Sci. (in print)
Yuan H (2002) Numerical assessments of cracks in elastic-plastic materials. Springer, Berlin
Yuan H, Cornec A (1990) Application of cohesive zone model
in investigation of elastic-plastic crack growth. In: Zarka
J et al., (ed) STRUCENG & FEMCAD: structural engineering and optimization. IITT International, Gournay pp
317323
Yuan H, Lin G, Cornec A (1996) Applications of cohesive zone
model for assessment of ductile fracture processes. Trans
ASME: J Eng Mat Tech 118:192200
Yang B, Mall S, Ravi-Chandar K (2001) A cohesive zone model
for fatigue crack growth in quasibrittle materials. Int J Solids Struct 38:39273944
Zhai J, Tomar V, Zhou M (2004) Micromechanical simulation of
dynamic fracture using the cohesive finite element method.
J Eng Mater Tech 126:179191
Zi G, Belytschko T (2003) New crack-tip elements for XFEM
and applications to cohesive cracks. Int J Numer Methods
Eng 57:22212240
Zi G, Chen H, Xu J, Belytschko T (2005) The extended finite
element method for dynamic fractures. Shock Vibr 12:
923

123

You might also like