Sie sind auf Seite 1von 32

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING

Int. J. Numer. Meth. Engng 2011; 85:1017–1048


Published online 23 August 2010 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/nme.3002

Numerical simulation of multiple crack growth in brittle materials


with adaptive remeshing

H. Azadi and A. R. Khoei∗, †


Center of Excellence in Structures and Earthquake Engineering, Department of Civil Engineering,
Sharif University of Technology, P.O. Box 11365-9313 Tehran, Iran

SUMMARY
In this paper, an automated adaptive remeshing procedure is presented for simulation of arbitrary shape
crack growth in multiple cracked bodies. The Zienkiewicz–Zhu error estimator is employed in conjunction
with the modified superconvergent patch recovery (SPR) technique to obtain more accurate estimation
of error. A stability analysis is performed to determine active cracks from a set of competitive cracks.
Various crack growth criteria together with the respective crack trajectory prediction are compared. Several
numerical examples are illustrated to demonstrate the efficiency, robustness and accuracy of computational
algorithm in the simulation of multiple crack growth. Copyright 䉷 2010 John Wiley & Sons, Ltd.

Received 31 May 2009; Revised 17 June 2010; Accepted 18 June 2010

KEY WORDS: multiple cracks; crack propagation criteria; mixed-mode fracture; adaptive remeshing;
error control; brittle material

1. INTRODUCTION

The study of structures containing many interacting cracks is important for the understanding of
damage and failure processes. The existence of bifurcations and multiple post-bifurcation paths is an
important characteristic of multiple crack systems. Analytical solutions for structures with random
distributions of cracks were reported by Freji-Ayoub et al. [1], Rubinstein [2–4] and Kachanov
[5–7]. Fractal methods for multiple crack systems were investigated by Rybaczuk and Stoppel
[8]. Chen [9] developed singular integral equations method for regular and random distribution
of cracks. Boundary element techniques were applied to multiple crack problems by Carpinteri
and Monetto [10]. Dena and Dong [11] developed a complex variable approach in modeling of
straight multiple center and edge crack problems using BEM. The extended finite element method
(XFEM) was applied by Dolbow et al. [12] in the propagation of a single crack, in which the crack
representation is embedded in the local finite element approximants. A method was presented by
Budyn et al. [13] for modeling the growth of multiple cracks in the framework of X-FEM. The
element-free Galerkin method (EFGM) has been applied to two- and three-dimensional LEFM
problems, both in statics [14] and dynamics [15, 16] with materially non-linear cohesive crack
and plastic fracture models. Rashid [17] developed the arbitrary local mesh replacement method,
which is an alternative to the remeshing technique. In this method, two distinct mesh refinements
are used; one that surrounds the propagating crack front and moves with crack, and the other that
fills the rest of the domain. However in multiple crack problems, the number of mesh refinements
increases as the number of cracks increases and the automatic propagation of cracks will be more

∗ Correspondence to: A. R. Khoei, Center of Excellence in Structures and Earthquake Engineering, Department of
Civil Engineering, Sharif University of Technology, P.O. Box 11365-9313, Tehran, Iran.
† E-mail: arkhoei@sharif.edu

Copyright 䉷 2010 John Wiley & Sons, Ltd.


1018 H. AZADI AND A. R. KHOEI

difficult. In the present study, a global adaptive remeshing procedure is developed for the modeling
of structures with multiple crack propagation.
Remeshing strategy toward the crack growth simulation allows modeling arbitrary shape crack
propagation as well as adapting the FE mesh in such a way that the error due to FE discretization
remains in a prescribed range. In order to improve the rate of convergence of FE solution and
the accuracy of stress intensity factors, the collapsed quarter-point singular elements proposed by
Barsoum [18] and Henshell and Shaw [19] are used at crack tips. The well-known Zienkiewicz–
Zhu [20, 21] error estimator with a recovery procedure based on the modified superconvergent
patch recovery (SPR) technique [22] is used, which provides more accurate error estimation for
fracture problems. As described by Khoei et al. [22], several modifications must be taken for the
recovery process when using the standard SPR technique in conjunction with singular elements.
The implementation of a polynomial with the same order of shape function for smoothing procedure
cannot describe the behavior of stresses in the vicinity of a crack tip. Although the derivatives
of singular element shape functions represent an appropriate order of singularity due to singular
mapping between the natural and parametric coordinates, the polynomial shape functions are not
able to represent such a feature and therefore the recovery process loses its efficiency near the
crack tip. To overcome this problem, the modified SPR technique was presented by Khoei et al.
[22], in which the analytical solution for crack tip fields was incorporated into the recovery process
as the smoothing function at the crack tip region. Considerable improvements in the accuracy of
the error estimator were observed using this modified patch recovery procedure.
The stress intensity factors are computed by means of path independent J -integral. The procedure
of FE solution is parameterized by the crack length control. A load parameter is adjusted such
that for the most critical crack tip, the crack driving force reaches the crack resistance. Based on
this constraint, the problem remains static and we can capture the snap-back instabilities in global
force–displacement response. The identification of correct post-bifurcation path in the problem
with competing cracks, which corresponds to the most stable crack evolution configuration, is
performed through the stability analysis. The variation of energy release rate at one crack tip due
to the growth of other cracks is evaluated to determine the strength of crack interaction. In this
study, the generalized form of virtual crack extension method proposed by Hwang et al. [23–25]
is employed to calculate the variation of energy release rate. The most significant feature of the
method is that the derivative of energy release rate for a multiple cracked body is computed in a
single analysis.
There are a number of first-order LEFM crack kinking theories developed in the literature based
on the singular terms of the local asymptotic crack front fields. These theories have been proposed
and tested for different types of materials. In the absence of a widely accepted criterion for the
mixed-mode crack growth, three of the most widely used criteria, i.e. the maximum hoop stress
[26], the maximum energy release rate [26, 27], and the minimum strain energy density theory
[28], are discussed and their corresponding predicted crack trajectories are computed. The outline
of the paper is as follows: in Section 2, the procedure for the computation of stress intensity
factors and the derivative of energy release rate with respect to the crack length are described. In
Section 3, various fracture initiation and crack kinking theories are presented. The error estimation
and adaptive mesh refinement procedure are described in Section 4. In Section 5, the algorithm
for multiple crack growth in the framework of crack length control is presented. Finally, several
numerical examples are presented in Section 6 to assess the robustness and accuracy of the proposed
computational procedures.

2. COMPUTATION OF FRACTURE PARAMETERS

2.1. Stress intensity factors


The stress intensity factors are extracted from the FE solution using the J -integral technique. Under
linear elastic material assumptions, the J -integral can be interpreted as being equivalent to the
energy release rate. The direct evaluation of J -integral along a contour in the finite element mesh

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1019

x2 x1

Figure 1. The local coordinate system at the crack tip together with the rosette
of collapsed quarter-point singular elements.

x2 2

q=0 C2 q=0

q=1 q=3/4
x1
q=1
x1
C1
S S

(a) (b)

Figure 2. The domain of integration for an equivalent domain evaluation of the J -integral together with
the weighting function q; (a) the annular domain, (b) the rosette of CQPE elements.

rarely exhibits path independence and other procedures must be adopted to obtain an objective
value for J . Li et al. [29] transformed the contour J -integral to an equivalent area integral. This
technique is much simpler to implement in a finite element context. Banks-Sills and Sherman [30]
demonstrated that this area integral provides objective values for J with respect to the domain of
integration. Consider a crack coordinate system, as shown in Figure 1, where x 1 aligns with the
crack axis and x2 is perpendicular to the crack, the area form of J -integral is defined as
  
*u i *q
J= ij − W 1 j dS (1)
S *x1 *x j
where  is the Kronecker delta and q is a weighting function defined over the domain of integration.
The domain of integration is an annular region S, around the crack tip which encloses no other
crack or void, as depicted in Figure 2(a). The function q has a value of one on the inner contour
of the domain C1 , and has a value of zero on the outer contour of the domain C2 , as shown in
this figure. In any other point, q is interpolated over the elements in the domain using the standard
shape functions as

Node
q= Ni qi (2)
i=1

and its derivatives


 *Ni
*q Node
= qi (3)
*x j i=1 *x j
Li et al. [29] and Blank-Sills and Sherman [30] showed that if the inner contour of the domain
C1 shrinks onto the crack tip, Equation (1) will still converge to the value of J -integral. In this
study, the crack tip singular elements rosette is used as the domain of integration. Although it
is recommended to choose the domain of integration as far as possible from the singularity, the
incorporation of singular finite elements in conjunction with the adaptive mesh refinement improves
the accuracy of FE solution near the crack tip considerably and thus, the results are noticeably
insensitive to the domain of integration [22]. On the other hand, we note that for a curvilinear

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1020 H. AZADI AND A. R. KHOEI

crack the domain of integration must only surround the actual crack edges and therefore it cannot
be selected arbitrarily far away from the crack tip. Hence, it seems that the selection of crack tip
singular elements is reasonable as the domain of integration. In this case, the nodal values for q
must be taken as one at the crack tip, 0.75 at the quarter-point nodes, and zero at all other nodes,
as shown in Figure 2(b).
In order to obtain the contribution of different fracture modes to the J -integral, a mode separation
procedure introduced by Ishikawa [31] and Bui [32] is implemented. In this technique, the near-
tip displacement and stress fields are decomposed into symmetric and anti-symmetric parts with
respect to the crack axis. The mode-separated J -integral values can then be computed by evaluating
Equation (1) based on the decomposed displacement and stress fields. Having computed the mode-
separated J -integral values, the stress intensity factors can easily be obtained as
 
K I = E  J I and K II = E  J I I (4)
in which E  = E for plane stress and E  = E/(1−2 ) for plane strain problems.

2.2. Variation of energy release rate


In multiple crack systems, a stability analysis must be performed to obtain the most stable crack
growth pattern. An important requirement of the stability analysis is to evaluate accurately the
derivatives of the energy release rate of each crack tip with respect to the growth of other cracks.
For this purpose, we use the generalized virtual crack extension method developed by Hwang
et al. [23–25]. The energy release rate G i at crack tip i is defined as the variation of the potential
energy of the cracked body with respect to the length of the crack i. Based on this definition, the
derivative of G i is calculated with respect to the length of any other crack j. In this derivation, the
variations of the structural stiffness matrix, nodal displacement vector and the applied nodal force
vector with respect to the length of crack j must be calculated. The variation of the displacement
vector can be obtained from the global equilibrium equation in terms of the variations of the
stiffness matrix and nodal force vector. If the crack faces are traction-free and there are no body
forces and thermal loadings, the contribution of global load variations to the rate of energy release
rate will vanish. Thus, the variation of stiffness matrix due to a virtual extension of crack j must
only be computed. The element stiffness variations calculated based on the method presented by
Lin and Abel [33] are assembled to produce the global stiffness variations. The imposed virtual
displacement field has the value of unity on the crack tip j and zero on a contour surrounding the
same crack tip, which does not intersect any other crack. The equations used for calculating the
rate of energy release rate are presented in Appendix A.
In this study, the first and second rings of crack tip elements are perturbed as a result of virtual
crack extension, as shown in Figure 3. Hwang et al. [23] expressed that the use of non-singular

x2

1.00
0.90
=0 0.80
0.70
x1
0.60
0.50
=1 0.40
0.30
0.20
0.10
0.00

Figure 3. The perturbation of the first and second layers of crack tip elements in
the generalized virtual crack extension technique ( represents the perturbation
in nodal coordinates in the direction of crack axis).

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1021

elements in the mesh perturbation improves the accuracy of solution for the higher order rates of
the energy release rate, since they contain additional higher order terms. By perturbing additional
layer of non-singular elements, the higher order terms can be included in the computation, and
the accuracy of the energy release rate variations can be considerably enhanced. As illustrated in
Figure 3, the variation of nodal coordinates in the direction of crack is equal to one at the crack tip
and zero at the exterior edge of second ring of crack tip elements, with a linear spatial distribution.
It must be noted that in the developments reported here, it is assumed that the elements influenced
by the virtual crack extension at each crack tip, comprise disjoint sets. This requirement must be
considered in the remeshing process.

3. PREDICTION OF CRACK TRAJECTORY IN MIXED-MODE LOADING

In the practical engineering problems, the cracked structures are generally subjected to mixed-mode
loading and the values of K I and K II are both non-zero. Nevertheless, the fracture toughness of
material is usually measured in the case of pure mode I loading condition noted by K IC . Thus,
a fracture initiation criterion is required to determine a critical combination of stress intensity
factors that leads to fracture initiation. Moreover, the prediction of crack trajectory is important
in the analysis of potential structural failure. Thus, a criterion is necessary to determine the angle
of incipient crack propagation with respect to the crack axis. Various criteria have been proposed
in the literature and tested for different materials. Three of the most widely used criteria; i.e. the
maximum circumferential tensile stress criterion, the maximum energy release rate criterion, and
the minimum strain energy density criterion, are discussed here and their algorithmic implemen-
tation is presented.

3.1. Maximum circumferential tensile stress criterion (MCS)


The maximum tensile hoop stress theory was first presented by Erdogan and Sih [26]. The theory
is based on the knowledge of stress state near the crack tip, in which the crack extension starts in a
radial direction perpendicular to the direction of greatest tension. Moreover, the fracture initiation
occurs when this maximum hoop stress reaches its critical value. The circumferential stress 
reaches its maximum on a plane in which the shear stress r  vanishes. Therefore, the crack kinking
angle can be expressed as
⎛ ⎞

2
K 1 K
0 = 2 tan−1 ⎝ +8⎠ for K II >0
I I
− (5)
4K II 4 K II
⎛ ⎞

2
KI 1 KI
0 = 2 tan−1 ⎝ + +8⎠ for K II <0 (6)
4K II 4 K II

On the plane of crack propagation, the hoop stress is expressed as





KI 0 2 0 K II 3 0 3 30
−max = √ cos 1−sin +√ − sin − sin (7)
2r 2 2 2r 4 2 4 2

The fracture initiation occurs if this value reaches the critical value of K IC / 2r . Thus, an
equivalent stress intensity factor K eq can be defined for the mixed-mode problems as
0 3 0
K eq = K I cos3 − K II cos sin 0 (8)
2 2 2
In the finite element implementation, the stress intensity factors extracted by the J -integral are
more accurate, and hence the crack tip stress fields are obtained using the above equations. In this
way, we avoid less accurate values of stresses obtained directly from the FE solution, which may
introduce the error in the prediction of crack growth direction.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1022 H. AZADI AND A. R. KHOEI

K I , K II

δ
θ K I (θ) , K II (θ)

Figure 4. The crack with an infinitesimal kink at angle .

3.2. Maximum energy release rate criterion (MER)


Based on the Griffith energy theory, Erdogan and Sih [26] presented that the crack grows in the
direction along which the elastic energy release per unit crack extension is a maximum, when
this energy reaches a critical value. Hence, an accurate extraction of the energy release rate for
different crack kinking angles is fundamental to the application of this criterion. It must be noted
that the J -integral technique described earlier, provides the energy release rate only for the case
of self-similar crack growth. Hussain et al. [27] obtained K I () and K II () for the stress intensity
factors of a major crack with an infinitesimal kink at an angle , in terms of the stress intensity
factors of the original crack K I and K II as (Figure 4)
⎛ ⎞ 2 ⎧ ⎫
 
1−  ⎪ ⎪
3
⎨ K I − K II sin ⎪
⎪ ⎬
K I () 4 ⎜  ⎟ 2
= ⎝ ⎠ (9)
3+cos2   ⎪
⎪ ⎪
⎩ K II + K I sin ⎪
K II () 1
1+ ⎭
 2
Substituting the stress intensity factors of the kinked crack into the Irwin’s generalized expression
for the energy release rate, we arrive at
1
G() = (K 2 ()+ K II2 ()) (10)
E I
where E  = E for plane stress and E  = E/(1−2 ) for plane strain problems. The above expression
yields
⎛  ⎞


2 1−
4 1 ⎜  ⎟ [(1+3 cos2 )K 2 −8 sin  cos K K +(9−5 cos2 )K 2 ]
G()= ⎝ ⎠ I II (11)
E 3+ cos2   I II
1+

The angle of crack propagation 0 can be obtained by maximizing G() as
*G()
=0
*
2
and −75.2◦ 75.2◦ (12)
* G()
<0
*2
The above limit angles correspond to the pure mode II loading conditions. The crack starts to
propagate when the energy release rate reaches the critical value of G C = K IC2 /E  . The numerical

implementation of Equation (12) requires the linearization of the derivative of G() and the solution
of a nonlinear equation. For the simplicity of computation, we directly evaluate G() for  lying
in the interval [−75.2◦ , +75.2◦ ] with 100 equal increments of . Subsequently, the angle 0 with
the maximum value of G(), which corresponds to a tensile hoop stress is selected as the angle of
crack propagation.

3.3. Minimum strain energy density criterion (MSD)


The minimum strain energy density criterion was proposed by Sih [28]. This theory postulates that
a crack starts to propagate in a direction along which the strain energy density at a critical distance

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1023

is a minimum, when this minimum reaches a critical value. For two-dimensional problems the
strain energy density, i.e. the strain energy dW per unit volume dV , can be expressed as
 
dW 1 +1
= (x + y ) −2(x  y −x y )
2 2
(13)
dV 4 4
where  is the shear modulus, = 3−4 for the plane strain problem and = (3−)/(1+) for
the plane stress problem. For brittle materials, the strain energy density near the crack tip can be
obtained by substituting the Westergaard solution (Broek [34]) for the stress field near the crack
tip into Equation (13), which yields
dW 1 S()
= (c1 K I2 +2c2 K I K II +c3 K II2 ) = (14)
dV r0 r0
where S() is the strain energy density factor, which needs to be minimized according to the
criterion. The coefficients c1 , c2 and c3 are defined as
1
c1 = [(1+cos )( −cos )] (15)
16
sin 
c2 = [2 cos − +1] (16)
16
1
c3 = [( +1)(1−cos )+(1+cos )(3 cos −1)] (17)
16
In Equation (14), r0 is a material characteristic parameter related to the depth of plastic zone,
in which S() is minimized at this distance from the crack tip. However in the derivation of
Equation (14), it is assumed that the behavior of material is brittle and r0 is therefore negligible,
which allows to use only the singular terms of the solution for the stress field. Consequently, the
value of S() in Equation (14) is independent of the parameter r0 as it appears on both the sides
of equation.
The angle of crack propagation can be obtained by minimizing S() as
2
*S * S
=0 and >0 (18)
* *2
The fracture initiation occurs when S−min reaches the critical value of Scr which can be obtained
by setting Scr = Smin for pure mode I loading and the collinear crack growth with critical stress
intensity factor defined as
( −1) 2
Scr = K (19)
8 IC
The numerical implementation of this criterion is similar to the MER criterion described in the
previous section.

4. ERROR ESTIMATION AND ADAPTIVE MESH REFINEMENT

In the simulation of crack propagation, the remeshing procedure plays two main roles. First, it must
account for the topological and geometrical changes of the model due to the crack growth. This
technique supports arbitrary shape crack growth by modifying the mesh to conform to evolutionary
geometry of the crack. Second, the adaptive mesh refinement improves the accuracy of finite
element solution prior to the growth of crack, which is used in the prediction of fracture initiation
and subsequent direction of crack propagation. The error estimation procedure employed here is
based on the Zienkiewicz–Zhu error estimator [20, 21] with the modified SPR technique developed

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1024 H. AZADI AND A. R. KHOEI

by Khoei et al. [22]. In what follows, we briefly discuss the main issues related to the error
estimation and mesh refinement.

4.1. Error estimation using modified SPR technique


The error estimator developed by Zienkiewicz and Zhu [20], has proved to be economical and
effective in the estimation of error in the FE solution due to domain discretization. The main idea
is to obtain an improved solution for the stress field and then approximate the stress error as the
difference between the recovered values and those obtained directly from FE analysis, in a global
measure such as the energy norm. The recovery procedure consists of obtaining improved values
for the nodal stresses, and then interpolating these values to other points using element shape
functions. The nodal smoothing procedure is performed based on the SPR technique, which results
in a higher rate of convergence for quadratic triangular elements compared to other projection
techniques [35].
Although it is ideal to use the full-superconvergence points in the recovery process, such points
do not exist for general meshes of triangular elements. In fact, the existence of superconvergence
points is very sensitive to the mesh pattern, element distortion and the presence of singularities.
Babuška et al. [36, 37] and Zhang et al. [38] introduced an approximate superconvergence property
which generalizes the classical idea of superconvergence to general meshes. They presented a
computer-based approach to determine the optimal sampling points. This procedure requires a
local periodization of the finite element grid and performing a numerical optimization to obtain robust
sampling points. However, for practical engineering applications of adaptive remeshing, such a process
makes the computational algorithm more complicated and uneconomical. Zienkiewicz and Taylor
[39] have shown that the stresses of Gauss–Legendre integration points are still of excellent quality
and these points are suggested to be used in the recovery process. Accordingly, these Gauss integration
points have been used here as optimal sampling points. It has been demonstrated that the ZZ error
estimator performs very well even if the sampling points are not superconvergence points [40].
Based on this technique, we solve a patch-wise projection at each node of the finite element
mesh from optimal sampling points to the considered nodal point. Solving these small patch-wise
problems, a continuous stress field can be obtained in an efficient way. It is assumed that the
nodal values ¯ ∗p belong to a polynomial expansion ∗p of the same complete order p of that shape
functions, which is valid over an element patch surrounding the particular assembly node. A typical
patch of elements is shown in Figure 5(a). Thus, we have

∗p = a0 +a1 x +a2 y +· · ·+an y n

= [1, x, y, . . . , y n ][a0 , a1 , a2 , . . . , an ]T
= Pa (20)

(a) (b)

Figure 5. The patch of elements: (a) the standard SPR technique (•—nodes, ◦—patch assembly point,
—Gauss quadrature points) and (b) the modified SPR technique for the crack tip region.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1025

The unknown vector a is determined by performing a least-square fit of ∗p to the existing data of
finite element solution at the Gauss quadrature points of elements patch as
 −1
 T
NG  T
NG
a= P (xk , yk )P(xk , yk ) [P (xk , yk )h (xk , yk )] (21)
k=1 k=1

where h is the finite elements solution for the stress component at kth Gauss point, with its
coordinates (xk , yk ) and NG represents the total number of Gauss points in the elements patch.
However, when the quarter-point singular elements are used at the crack tip, the implementation
of a polynomial with the same order of shape functions for smoothing procedure over the crack
tip patch, leads to the over-estimation of errors [22]. The shape functions of these elements are
not able to represent the singular behavior of the stress field in the vicinity of the crack tip, and
they are not a suitable choice for the recovery purpose over the crack tip patch. As proposed by
Khoei et al. [22], the analytical solution for crack tip fields can be used to solve this problem using
the smoothing function at the crack tip region. Here, the first three terms of Williams expansion
[41] are used for each stress component as the smoothing function, as described in Appendix A.
The crack tip patch of elements comprises the first and second layers of elements surrounding
the crack tip, as shown in Figure 5(b). The incorporation of higher order terms in the smoothing
functions and application of a crack tip patch described above were shown to be useful in the
automatic adjustment of singular elements size through the mesh adaptivity [22]. This special
recovery process is only used at the crack tip region, while in other areas the conventional SPR
technique is performed.
Having computed the improved stress field, the error can be approximated as
e ≈ e∗ = r∗ −rh (22)
where e is the exact error, e∗ the estimated error, and r∗ the recovered solution. The estimated
error based on the energy norm can be obtained as

 1/2
ēu  = e∗T
 D−1 ∗
e  d (23)


Similarly, the energy norm of solution can be written as



 1/2
ū = rhT D−1 rh d (24)


and the relative error norm can be calculated as


ēu 


¯ = (25)
ū
where is an empirical correction factor. This relative error norm can be used later in the mesh
refinement procedure. The accuracy or the quality of a posteriori error estimator is measured by
the effectivity index , which is defined as the ratio of predicted error to the exact error value, i.e.

¯
= (26)

exact
If the recovered solution converges at a higher rate than the finite element solution we shall always
have asymptotically exact error estimation and  should converge to unity as the mesh is refined.

4.2. Adaptive mesh refinement


The objective of adaptive remeshing is to achieve a new FE mesh with a relative error norm
less
than a prescribed value
aim . The process of new FE mesh generation aims toward the achievement
of an optimal mesh for which the error norm is equally distributed among the elements. Hence,

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1026 H. AZADI AND A. R. KHOEI

the allowable relative error norm for an element i can be obtained as


(ēu i )aim = √aim ū (27)


m
where
aim is the prescribed target percentage error and m is the total number of elements. It is
assumed that the rate of convergence of local error is O(h ), with h denoting the size of element
and the minimum of shape function order and singularity order. Thus, the new element size can
be calculated in an h-refinement adaptive procedure as
h old
h new = 1/
(28)
i
where  = min( p, ) and
ēu i
i = (29)
(ēu i )aim
in which = 0.5 for the crack problems and p is the order of shape functions. In order to obtain
the nodal element size, a simple averaging method is used among the elements connected at a
nodal point. Subsequently, the domain is completely remeshed using an automatic advancing-front
mesh generator.

5. SOLUTION ALGORITHM FOR MULTIPLE CRACK PROBLEMS

In the simulation of quasi-static crack propagation, the load must be applied such that the crack
driving force remains less than or equal to the crack resistance; otherwise the problem turns into
a dynamic fracture process. In this section, we describe the ‘crack length control’ scheme imple-
mented into the code, in which the loading process is controlled by an increasing function of the
total cracks length. A load parameter is adjusted to satisfy the fracture initiation criterion. Appli-
cation of this constraint is an appropriate scheme of handling snap-back instabilities. Furthermore,
an algorithm is described to select active cracks from a set of competing cracks throughout the
stability analysis. It is shown that the proposed crack propagation algorithm provides an automatic
and an efficient way for modeling of multiple growing cracks in the framework of remeshing
technique. Finally, a flowchart of the algorithm is presented to illustrate the required sequences
processed by the multiple crack growth simulation code.

5.1. Determination of crack increment and load parameter


In the most FE computer algorithms applied for the simulation of crack growth, the crack increments
are usually assigned after the determination of stress intensity factors and the direction of crack
growth. In the presence of fatigue crack growth, the cracks are driven by the empirical Paris
law [42] while the computational procedure for predicting the amount of crack growth is rarely
reported. Nguyen et al. [43] proposed a coupled displacement–crack length rate analysis based on
the J -integral method to calculate the second derivatives of potential energy for the evaluation of
crack growth stability. Fortino and Bilotta [44] presented an algorithm to determine the amount of
crack growth in 2D problems with stable rectilinear crack growth. Formica et al. [45] developed
a crack growth formulation based on the so-called Destuynder–Djaoua method [46], in which the
crack length increment is related to the given load parameter. However, more developments are
necessary for the general case of curvilinear crack growth in multiple cracked bodies since the
crack growth path is not known a-priori.
The solution algorithm proposed here is based on the ‘crack length control’ scheme, in which
the loading process is controlled by a monotonically increasing function of the total crack length,
i.e. the sum of all crack lengths. Bocca et al. [47] presented that the crack length control assures
the uniqueness of the solution being an elegant way of handling the snap-through and snap-back

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1027

instabilities. In this technique, the total amount of crack growth is set equal to a fixed value of
ltot at each step of simulation. It must be noted that this algorithm is explicit, in which the crack
growth increments are set and not calculated at the beginning of each step. The value of ltot
is simply distributed equally among all active cracks [13]. The amount of crack growth for each
active crack is then evaluated by
ltot
li = (30)
n act
where n act is the number of active cracks propagated at the beginning of time step n +1, and
determined by a stability analysis at time step n. As the algorithm is explicit and the equation of
equilibrium and the Griffith condition are implemented in an operator-split manner, the convergence
of the solution to the physical response of structure is conditionally stable. This means that ltot
must be chosen sufficiently small. As we are tracing the unique stable path of the structure response,
a small perturbation in the crack growth increments will not affect the convergence of the solution.
Thus, if ltot is reduced, the results of the simulation converge to the actual behavior of the
structure regardless of the distribution given by Equation (30). If a growing crack intersects the
free boundary, the value of li is adjusted properly, and the increment of other active cracks is
increased to keep ltot constant. At the beginning of each step, the total load is applied to the
system and the FE analysis is performed. According to the selected fracture criterion, the crack
driving force F and the crack resistance R are calculated for each crack tip. The most critical
crack tip, i.e. the crack tip with maximum value of F/R, is then determined. The dimensionless
load parameter is chosen so that the crack driving force for the most critical crack tip attains the
crack resistance. The parameter for the linear elastic material is defined as
n tip
R
i
= min for the MCS criterion (31)
i=1 Fi
and


n tip
Ri
= min for the MER and MSD criteria (32)
i=1 Fi
Based on this technique, the crack driving force for all crack tips remains less than or equal to the
corresponding crack resistance, and therefore the dynamic drops in the global force–displacement
curve are avoided.
In the case that several cracks are close to the critical state of propagation, i.e. Fi is close
to Ri , a very small error in the calculation of stress intensity factors may prevent a crack from
propagating, which would result in an incorrect prediction of crack propagation pattern [22]. Thus,
we accept a tolerance in the evaluation of fracture initiation criterion. The set of competitive cracks
with the potential of propagation is denoted by Ncomp and defined as
  
 Ri − Fi
Ncomp = crack tip i ∈ {1, 2, . . . , n tip }  ε (33)
Ri
The tolerance ε is selected based on the accuracy of computation of the stress intensity factors.
As the J -integral method is used in the calculation of stress intensity factors, ε is set to 1% in the
applications reported herein. The growing cracks are determined from a set of competitive cracks
by using the stability analysis described in the following section.

5.2. Stability analysis and identification of growing cracks


The selection of correct post-bifurcation equilibrium path in multiple crack systems requires a
stability analysis of crack growth to be performed. Bazant and Cedolin [48] presented that for irre-
versible systems, such as structures with plasticity, fracture or damage effects, most post-bifurcation
branches can consist of stable states. Thus, in such systems more general thermodynamic based
criteria must be used to determine the stable path. They showed that for isothermal conditions and

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1028 H. AZADI AND A. R. KHOEI

for states infinitely close to thermodynamic equilibrium we have


W = −T (S)in (34)
where W is the work increment that needs to be supplied to the system to extend the cracks,
T denotes the temperature and (S)in is the increment of internal entropy of the system. According
to the second law of thermodynamics, the system approaches the equilibrium state that maximizes
Sin , and consequently minimizes W . Hence, the post-bifurcation path that actually occurs is
the one for which W becomes minimum. For a linear elastic body with a system of cracks, we
have the following expression for W [49]:
 ai

n
W =U (a1 , a2 , . . . , an , )+ 2 i dai (35)
i=1 0

where U denotes the elastic strain energy of the body, ai the crack length, the loading parameter,
and i the surface energy of the material for crack tip i. The function W can be expressed as

W = W +2 W +· · · (36)
As a result of equilibrium, the first-order variation W must vanish for any admissible change of
crack lengths. Therefore, we can approximate W with the second-order variation 2 W , neglecting
the effect of other higher order terms. If the crack tips i = 1, 2, . . . , m extend and the crack tips
i = m +1, m +2, . . . , n remain stationary, the second variation of W can be expressed as
2
1 n  n * U m *
i
2
W ≈  W = ai a j + (ai )2 (37)
2 i=1 j=1 *ai *a j i=1 *ai

The crack propagation modeling is performed in a discrete form, i.e. we first solve the equilibrium
equation at the fixed crack configuration and then the cracks are propagated at the fixed load
parameter. Hence, in the derivation of Equation (37) the load parameter is assumed to remain
constant. For the case of homogeneous material, the second term of the above expression vanishes.
The strain energy release rate is defined as
*U
Gi = − (38)
*ai
and the second-order variation of W can be then rewritten as
1 n  n
2 W = − G i, j ai a j (39)
2 i=1 j=1
The matrix G i, j is constructed for a set of competitive cracks Ncomp using the generalized virtual
crack extension technique, described in Section 2.2. Moreover, for a given subset of Ncomp the
corresponding values of a can be obtained from the algorithm presented in Section 5.1. For all
possible combinations of growing crack tips, which are subsets of Ncomp , the second variation
2 W is calculated from Equation (39). The crack evolution configuration is the subset for which
2 W becomes minimum. This configuration is the one that gives the smallest rising slope in the
global force–displacement response, the steepest softening slope or the most drastic snap-back. In
a special case that the values of 2 W are equal for some post-bifurcation branches and at the same
time, are the minimum values among all branches, one must consider the higher order term in the
work expansion expression (Equation (36)). However, for symmetric structures the branch that is
followed is decided by random imperfections.

5.3. Algorithm of crack propagation simulation


The crack growth simulation is an incremental process, where a number of steps are required for
the progression of the model. Each increment of the simulation relies on the computed results from
previous step and represents one crack configuration. Based on the mechanics of crack growth, the

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1029

Face 2

Face 1

Figure 6. Solid modeling with the boundary representation technique.

Table I. The adjacency queries.


Query Description
Get VV Given a vertex, find all adjacent vertices connected to
the given vertex through an edge
Get EV Given an edge, find its defining vertices
Get FV Given a face, find the vertices along its boundary
Get VF Given a vertex, find all adjacent faces
Get EF Given an edge, find its adjacent faces
Get FF Given a face, find all adjacent faces
Get VE Given a vertex, find all adjacent edges
Get EE Given an edge, find adjacent edges
Get FE Given a face, find all edges along its boundary

Table II. The Euler operators.


Operator Description
AFV Add face, vertex
RFV Remove face, vertex
AEV Add edge, vertex
REV Remove edge, vertex
AFE Add face, edge
RFE Remove face, edge
SEAV Split edge, add vertex
JERV Join edge, remove vertex

geometry and topology of the model are modified at each step to reflect the current configuration of
cracks. In this study, modeling of the structure is accomplished by storing its boundary information
in a boundary database using topological elements, such as vertices, oriented edges, non-oriented
edges and faces (Figure 6). This database includes the geometrical information of actual vertex
locations. Query routines are implemented in the FE code to provide efficient retrieval of adjacency
information from the boundary database. These adjacency relationships which can simplify modular
programming are listed in Table I [50]. Euler operators listed in Table II are implemented in
modification routines used to update the boundary database and to render the crack propagation
process [51]. The boundary condition information is defined based on the boundary database and
therefore is independent of finite element mesh. For each crack tip that satisfies the crack growth
criterion, the crack tip vertex is then duplicated and the new crack tip vertex along with new edges
is added to the boundary database. The crack tip rosette configuration has been adjusted here by
adding the new non-oriented boundaries around the crack-tip.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1030 H. AZADI AND A. R. KHOEI

t1
t2
t3
t4
t5

(a) (b)

Figure 7. The junction of a crack with another crack: (a) predicting the potential coalescence
point using the ray casting technique and (b) merging cracks by modifying the boundary
database using appropriate Euler operators.

Crack

A
C

B
Free boundary

Figure 8. The crack approaching the free boundary; AB denotes the direction of propagation
predicted by the fracture criteria and, AC indicates the perpendicular direction to free
boundary, in which the path AC is followed.

Multi-crack simulation algorithms must deal with several situations, such as junction of cracks
and growth to a free boundary. We detect any impending intersection by using a simple ray casting
technique [52]. At each step of the simulation for every growing crack tip, we consider a ray which
is a straight line defined in a parameterized form based on the coordinates of crack tip (x t , yt )
and the normalized direction vector of propagation (ex , e y ). In this form, the points on the ray are
ordered via a single parameter t>0. For each value of t, the corresponding point (x, y) on the
ray is given by

x = xt +t.ex (40)

y = yt +t.e y (41)

For each boundary edge, the ray and the edge equations are simultaneously solved to obtain the inter-
section point. If the intersection point is within the edge’s bounds, the parameter t corresponding to
intersection point is stored. The ascending ordered list of ray-edge intersection parameters denotes
the points where the ray exits and enters the structure, as shown in Figure 7(a). The first component
of this list tmin indicates the potential coalescence point. The crack is allowed to intersect the
boundary if tmin is less than the crack increment li plus the radius of singular elements rosette.
This prevents the presence of singular elements in a region very close to the free boundary, which
would result in severe gradients and excessive mesh refinement. Having obtained the coalescence
point, the boundary database is updated accordingly using appropriate Euler operators, as shown in
Figure 7(b). The orientation of newly created edges is assigned such that at each vertex, one edge
points inward and other points outward. On the other hand, if the predicted li from Equation (30)
is greater than the distance between the crack tip and free boundary but it is still less than the
required value to reach the crack outside of the boundary in the direction of crack propagation,
it seems more reasonable that the crack intersects the free boundary by changing the direction of
propagation, as shown in Figure 8. In Figure 9, a flowchart of required sequences processed by
the crack propagation simulation code is presented.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1031

Figure 9. The process sequences of the multiple crack growth simulation.

6. NUMERICAL SIMULATION RESULTS

In order to illustrate the computational efficiency of the proposed approach in the modeling of
crack growth in multiple crack systems, several examples are investigated. The computer code
SUT–DAM [53], which was developed for the simulation of brittle and ductile fracture problems
[22, 54–56], has been extended to automatic adaptive analysis of multiple crack systems. In the
first example, the accuracy of calculation of the stress intensity factors and the derivative of energy
release rate are investigated. The next examples are chosen to demonstrate the robustness of

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1032 H. AZADI AND A. R. KHOEI

Figure 10. The finite crack in an infinite plate under mixed-mode remote loading.

adaptive remeshing technique in the simulation of fracture process in multiple cracked bodies with
complicated geometries. Various fracture criteria are employed and the corresponding predictions
of crack trajectories are compared. It has been shown that the error estimation and adaptive mesh
refinement process are efficient in the reduction of errors. In all examples, the three-point Gauss
quadrature rule is used for the standard elements and the four-point reduced integration rule for
collapsed quarter-point singular elements. Each crack tip is surrounded by eight singular elements,
refined adaptively only in the radial direction. For all examples, the plane strain and isothermal
conditions are assumed.

6.1. Accuracy of the computation of fracture parameters


The first example is of an infinite plate with a finite crack of length 2a subjected to far field
loading, in which 22 = ∞ , 11 = 0 and 12 = ∞ , as shown in Figure 10. The analytical solution
is available for this mixed-mode problem and used for comparison [57]. The stress components
are as follows:
11 = Re( +2 )− yIm( + ) (42)
22 = Re( )+ yIm( + )+∞ (43)
12 = −yRe( + )−Im( )+∞ (44)
where
 
 z
 = −i∞ √ −1 (45)
z 2 −a 2
 
z
  = ∞ √ −1 (46)
z 2 −a 2

and z = x +i y.√The analytical values of stress intensity factors for this problem are K I = ∞ a
and K II = ∞ a. The analytical expression for the derivative of energy release rate is


*G 2 *K I *K II
= KI + K II (47)
*a E  *a *a
The finite element model is constructed for one half of the plate, as shown in Figure 10. Tractions
obtained from the exact stress field are applied on the boundary of the plate with ∞ = 1 MPa
and ∞ = 0.5 MPa. The displacement components of vertices A and B of the plate are prescribed
into the analytical values to prevent the rigid body modes. The material properties are chosen as
follows; E = 6.4×105 MPa and  = 0.2. The accuracy is set to 2%. Both uniform and adaptive
mesh refinements have been carried out. Table III summarizes the results of the analysis for each

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
Table III. The accuracy of computation of the fracture parameters for a mixed-mode problem with different mesh refinement strategies.

Copyright 䉷 2010 John Wiley & Sons, Ltd.



-estimation
-exact KI-FEM Relative error KII-FEM Relative error dG/da-FEM Relative error
Mesh DOF (%) (%)  (Mpa.mm1/2 ) (%) (Mpa.mm1/2 ) (%) (Mpa) (%)
Uniform 1 226 10.76 9.48 1.1340 4.33279 9.32 2.13273 7.62 5.15254E-06 12.53
Uniform 2 698 8.20 7.92 1.0354 4.04222 1.99 2.03375 2.63 5.37294E-06 8.79
Uniform 3 1630 6.46 7.19 0.8989 4.00011 0.93 1.98327 0.08 5.88893E-06 0.03
Adaptive 1 146 10.43 9.48 1.1000 4.34716 9.68 2.10641 6.30 4.50884E-06 23.46
Adaptive 2 608 2.36 2.22 1.0624 4.02230 1.49 1.97178 0.50 5.70149E-06 3.21
Adaptive 3 846 1.65 1.76 0.9385 3.95800 0.13 1.97936 0.12 5.84149E-06 0.83
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH

DOI: 10.1002/nme
Int. J. Numer. Meth. Engng 2011; 85:1017–1048
1033
1034 H. AZADI AND A. R. KHOEI

1.15

normalized fracture parameter


1.1
1.05
1
0.95
KI-uniform
0.9 KII-uniform
0.85 dG/da-uniform
0.8 KI-adaptive
KII-adaptive
0.75
dG/da-adaptive
0.7
0 200 400 600 800 1000 1200 1400 1600 1800
DOF

Figure 11. The convergence rate of fracture parameters normalized by their analytical counterparts using
the uniform and adaptive mesh refinements.

0.1000

0.0100 1
log (||e||)

0.25

1
1
0.0010
adaptive refinement
uniform refinement
aim error
0.0001
100 1000 10000
log (DOF)

Figure 12. The convergence rate for the uniform and adaptive mesh refinements in energy norm.

mesh refinement sequence. The values of effectivity index are close to unity, which demonstrate
how well the error estimator is performing. The convergence of stress intensity factors and the
derivative of energy release rate which are normalized by their analytical values are shown in
Figure 11. It can be observed that by using the adaptive mesh refinement, the fracture parameters
rapidly converge to their exact values. After second adaptive mesh refinement, a relative error of
0.1% for the stress intensity factors and 0.8% for the derivative of energy release rate are achieved
which demonstrates the accuracy and reliability of the implemented computational algorithms. It
can be seen in Figure 12 that the convergence rate for uniform mesh refinement is dominated by
the intensity of singularity, i.e. /2 = 0.25, while for adaptive mesh refinement the optimal rate of
convergence is preserved.

6.2. A center-cracked tension plate


The second example is chosen to verify the stability analysis demonstrated in Section 5.2 for a
center-cracked tensile plate, originally performed by Bazant and Tabbara [58]. A plate of 2 in
width, unit thickness, and 7 in length is considered with a centric crack of length 0.2 in, as shown
in Figure 13. The plate is modeled using an initial FE mesh of 420 elements and 896 nodes
with the total crack length increment of 0.05 in. The plate is loaded at its top and bottom edges
with a fixed √tensile stress. The material properties are as follows; E = 348 076 psi,  = 0.35 and
K IC = 910 psi in. The target error in adaptive mesh refinement is set to 5%.
Based on the analytical solution, the determinant of derivative of the energy release rate is equal
to zero for the symmetric crack growth path and thus, each point on this path is a bifurcation
point. The stability analysis demonstrates that the second-order work 2 W for the secondary non-
symmetric growth path is smaller than that for the main symmetric path. On the non-symmetric

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1035

7 in
σ
2 in

Figure 13. A center-cracked tension plate: the problem statement.

0.18
0.16
0.14
nominal stress (psi)

0.12
0.1
0.08
0.06
0.04
non-symmetric
0.02
symmetric
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
nominal strain x10-3

Figure 14. The nominal stress–nominal strain curve for a center-cracked plate loaded with tensile stress.

path, only one crack tip grows until the ligament on one side is completely ruptured. Then, the
remaining edge crack grows toward the opposite edge until percolation occurs. The nominal stress–
nominal strain curve for this example is plotted in Figure 14. It can be seen that the non-symmetric
crack growth path results in a more drastic snap-back compared to the symmetric path. The
breakdown of symmetry and localization of deformation into a single crack tip was experimentally
demonstrated by Cedolin et al. [59]. In the finite element analysis, the second-order work 2 W is
slightly different for the two possible secondary paths due to the asymmetry of the mesh (either
the left crack tip grows, or the right one). This small numerical error produced in the calculation
of derivative of the energy release rate switches the solution process to one of the secondary
branches. However, as noted in Section 5.2 for such a symmetric structure, random imperfections
will determine the secondary branch that must be followed. In Figure 15, the contours of von-Mises
equivalent stress are shown as the crack grows.

6.3. A plate with a central inclined crack


The next example demonstrates the effect of fracture criteria on the prediction of crack trajectory
and the overall response of system in a plate with a central inclined crack. Consider a square plate
of 10 in width containing an inclined crack of 2 in length, as shown in Figure 16(a). The lower
edge of the plate is fixed in both directions, while the upper edge is only fixed in the x-direction.
The plate is √made of PMMA with the mechanical properties of E = 348 076 psi,  = 0.35 and
K IC = 910 psi in. The target error in the adaptive mesh refinement process is set to 5% and the
total crack increment is 0.5 in.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1036 H. AZADI AND A. R. KHOEI

Figure 15. The evolution of the von-Mises equivalent stress field with crack growth (psi).

2 in
10 in

(a) 10 in

MCS
MER
MSD
(b)

Figure 16. The plate with a central inclined crack under tensile load: (a) the problem statement and (b) the
comparison of crack trajectories predicted by various fracture criteria.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1037

(a) (e) (i)

(b) (f) (j)

(c) (g) (k)

(d) (h) (l)

Figure 17. The adaptive mesh evolution sequences for various fracture criteria: (a–d) the MCS criterion
with 476, 646, 613, 890 elements, (e–h) the MER criterion with 516, 568, 891, 984 elements, (i–l) The
MSD criterion with 495, 802, 852, 941 elements.

The simulation is performed using various fracture criteria and their corresponding crack trajec-
tories are shown in Figure 16(b). The agreement between the trajectory predictions of different
criteria depends on the mode mixity ratio of K II /K I . For small values of mode mixity ratio, the
agreement between the predictions is quite good; however, as the effect of mode II increases, the
disparity increases as well. For this example, the agreement between the predictions of maximum
circumferential stress and maximum energy release rate criteria is good, nevertheless the path
predicted by the minimum strain energy density criterion deviates from the other two criteria. The
adaptive mesh evolution sequence for each criterion is shown in Figure 17. The variation of vertical
reaction with crack length is depicted in Figure 18. The global relative error in energy norm is

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1038 H. AZADI AND A. R. KHOEI

600 MSD
MER
500 MCS

Force (lbf)
400

300

200

100

0
2 3 4 5 6 7 8 9 10 11
Crack length (in)

Figure 18. The force–crack length curve for the inclined crack problem using various fracture criteria.

6
Estimated error %

3
MCS
2
MER
1 aim error
MSD
0
2 3 4 5 6 7 8 9 10
Crack length (in)

Figure 19. The variation of estimated error during the simulation of inclined crack
growth using various fracture criteria.

shown in Figure 19 during the adaptive analysis. It can be observed that at final steps of analysis
the estimated error is larger than the allowable error. This is due to the fact that the specimen is
close to complete failure and the crack tip is close to the free boundary, which necessitates the
use of a very dense mesh at the crack tip region. In this case, the convergence rate of FE solution
decreases that makes the adaptive refinement process less efficient.

6.4. A plate with two holes and multiple cracks


The next example presents a rectangular plate with two off-center circular holes and two edge
cracks, as shown in Figure 20. This example was simulated by Bouchard et al. [60, 61] to present
the applicability of their remeshing technique in the modeling of crack propagation. They compared
the crack path predictions of various fracture criteria for this
√ example. The material properties are
as follows; E = 2×105 MPa,  = 0.3 and K IC = 1300 MPa mm. The total crack increment at each
step is 1 mm and the target error is set to 5%.
The resulting crack trajectories are shown in Figure 21(a) using various fracture criteria. It must
be noted that although the problem is symmetric, the correct crack path is asymmetric. This can be
easily observed by comparing the curves of force versus fracture displacement for symmetric and
asymmetric crack growth patterns (Figure 22). The fracture displacement is the displacement of
load application point due to crack growth only. It can be seen that the asymmetric path gives rise
to a steeper softening slope and more drastic snap-back. Thus, if both cracks grow simultaneously
in a symmetric configuration, the path is unstable. According to Figure 21(a), we observe that the
trajectory of the left crack, which is dominated by mode I, is similar for various fracture criteria.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1039

3 mm

3 mm

10 mm
2m
1 mm

m
2.85 mm

20 mm

Figure 20. A plate with two holes and multiple cracks: the problem statement.

MCS
MSD
MER

(a)

(b)

Figure 21. The crack trajectory for a plate with two holes: (a) the crack paths obtained in the present
study and (b) the crack paths reported by Bouchard et al. [58].

However, as the propagation of the right crack is affected by mode II, some discrepancies are
observed among the predicted trajectories. It can be observed from Figure 21(b) that the crack
paths given in reference [61] show more disparity between the predictions of different fracture
criteria, when compared to those reported here. This is due to some differences in the numerical
implementation of these criteria. In reference [61], the finite element results are directly incorporated
in the process of crack path prediction to avoid the calculation of stress intensity factors and the
results are therefore affected by numerical errors. However, in the analysis performed here, the
stress intensity factors are first evaluated accurately and, the analytical crack tip fields, which are
more accurate than the finite element ones, are then used in the application of fracture criteria.
The successive mesh refinements during the crack growth simulation are shown in Figure 23.
The variation of estimated error with crack propagation is plotted in Figure 24 for adaptive and
uniform FE meshes. It can be observed that near the complete failure of the structure, the estimated
error of the uniform meshes increases, while for adaptive meshes this error is controlled by mesh
refinement strategy. However as explained in previous example, in such an extreme situation the
adaptive mesh refinement process is less efficient. Figure 25 shows the evolution of the von-Mises
equivalent stress field at different stages of crack growth. Figure 26 represents the evolution of
vertical reaction force with the crack length obtained in this analysis. The points at which the

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1040 H. AZADI AND A. R. KHOEI

16
symmetric
14 non-symmetric

12

10

Force (KN)
8

0
0 0.01 0.02 0.03 0.04 0.05 0.06
Fracture Displacement (mm)

Figure 22. The force–fracture displacement curve for a plate with two holes, comparison
of stable path and symmetric path.

(a) (b)

(c) (d)

Figure 23. The crack propagation sequences using the adaptive remeshing and MCS criterion: (a) 425
elements; (b) 462 elements; (c) 646 elements; and (d) 998 elements.

snapshots of Figure 25 were taken are indicated in this diagram. It must be noted that although
the crack paths are similar for the MCS and MER criteria, the corresponding force–crack length
curves are different. In fact, with the same mode mixity ratios that control the crack path, the
individual values of K I and K II which result in fracture initiation are different for each criterion
and thus, the resulting force–crack length curves are different.

6.5. A plate with random distribution of cracks


The last example is a plate with random distribution of cracks, which was originally presented by
Budyn et al. [13] in the simulation of multiple crack growth using the X-FEM technique. A square
plate of 2 in width is considered containing 10 initial cracks subjected to a tension load, as shown

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1041

25
uniform mesh
adaptive mesh
20

Estimated error %
aim error

15

10

0
0 2 4 6 8 10 12 14 16 18 20 22
Crack length (mm)

Figure 24. The variation of estimated error with total crack length for adaptive and uniform FE meshes.

Figure 25. The evolution of the von-Mises equivalent stress field at various crack growth (×10 MPa).

16
MSD
14 MER
(a) MCS
12
Force (KN)

10
8
6 (b)
4
2
(c) (d)
0
2 4 6 8 10 12 14 16 18 20 22 24
Crack length (mm)

Figure 26. The force–crack length curve for a plate with two holes at various fracture criteria.

in Figure 27(a). The material properties are similar to those of a glass given in the first example.
The target error is 5% and the crack increment is set to 0.07 in. The initial FE mesh contains 2626
nodes and 1241 elements, and the simulation is performed using the MCS criterion.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1042 H. AZADI AND A. R. KHOEI

2 in
2 2 2 2
1 1 3
1 1

(a) (b)
2 in

Figure 27. A plate with random distribution of cracks: (a) the problem statement; (b) the final crack path
obtained by the adaptive remeshing technique.

(a) (b)

Figure 28. A plate with random distribution of cracks: (a) the final mesh in the simulation of adaptive
remeshing technique (4362 elements and 9028 nodes) and (b) the final configuration reported by
Budyn et al. [13] using the X-FEM technique (5228 elements and 10 653 nodes).

The final crack configuration is given in Figure 27(b). It can be observed from the figure that
three middle cracks propagate while the other cracks remain stationary. The coalescence of cracks
and the intersection with free boundary are automatically handled in the boundary update process.
In Figure 28(a), the adaptive mesh refinement is shown at the final configuration. The results are
in good agreement with those reported by Budyn et al. [13] in Figure 28(b). The nominal stress–
nominal strain curve for this example is shown in Figure 29. In this figure, three points at which
the crack coalescence occurs are indicated. At point (a), the crack tip 1 from crack 2 connects to
crack 1; at point (b), the crack tip 1 from crack 1 connects to the left edge of the plate; and at
point (c), the crack tip 2 from crack 2 connects to crack 3. Point (d) describes the final state of
the specimen just before percolation. Finally, the contours of the von-Mises equivalent stress are
shown in Figure 30 at the final adaptive mesh refinement.

7. CONCLUSION

In this paper, an adaptive remeshing technique was applied to the automatic simulation of multiple
crack growth in brittle materials. The error estimation procedure was based on Zienkiewicz–Zhu
error estimator and a modified superconvergent patch recovery technique was employed. In order
to improve the rate of convergence of FE solution and the accuracy of fracture parameters, the
collapsed quarter-point singular elements were used at the crack tip regions. In order to maintain

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1043

1200

1000

nominal stress (psi)


800

600 (a)

400

200 (b)
(d)
(c)
0
0 0.01 0.02 0.03 0.04 0.05 0.06
nominal strain

Figure 29. The nominal stress–nominal strain curve for a plate with random
distribution of cracks before percolation.

Figure 30. The contours of von-Mises equivalent stress at various crack growth (psi).

the static crack growth condition and satisfy the Griffith criterion, a load parameter was introduced
in the computation process. The solution method was parameterized by the crack length control,
which is able to capture the snap-back behavior. The intersection of cracks with free boundaries
and the crack coalescence were dealt with in an automatic boundary update procedure and thus,
the simulation can be run until the complete failure of structure. A stability analysis of crack
growth was performed to determine the growing cracks in a multiple cracked body. Different

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1044 H. AZADI AND A. R. KHOEI

fracture criteria were employed in the simulation process and the corresponding predictions of
crack trajectory and the global response of structure were compared. The proposed computational
algorithm was validated through the simulation of several numerical examples.
In the present study, significant improvements are made to the remeshing technique developed
by Bouchard et al. [60, 61]. The use of stability analysis to detect growing cracks, incorporation of
ZZ error estimator with the modified SPR technique and mesh adaptivity to improve the accuracy
of the results, the capability to capture snap-backs and the enhancement of the algorithm to
automatically model intersecting cracks are some of these improvements. The main advantage of
the presented remeshing technique in comparison to the X-FEM technique developed by Budyn
et al. [13] is the error estimation and the adaptive mesh refinement, as can easily be seen by
comparing Figures 28(a) and (b). Budyn et al. [13] used a fine mesh over the entire domain to
achieve accurate results, which is computationally expensive. However, the adaptive refinement
process attempts to achieve the desired accuracy while keeping the number of degrees-of-freedom
as few as possible. It must be noted that the error estimation, the boundary update and generation
of a new mesh is a costly process. In addition, generation of a mesh on a domain with very complex
boundaries and prescribed element size distribution requires a very powerful mesh generator. In
this sense, the X-FEM approach is very helpful, as it eliminates repeated application of domain
discretization step. However, substantial calculations are required to identify the elements to be
enriched, perform the necessary coordinate transformations along each crack segment within each
enriched element, discretize enriched elements into regular subcells for numerical quadrature and
to perform higher order quadrature on each subcell. According to these issues and based on the
numerical simulation results presented in this paper, the developed adaptive remeshing technique
seems to be a robust computational framework to efficiently model crack propagation process in
multiple crack systems.

APPENDIX A: CALCULATION OF THE DERIVATIVE OF ENERGY RELEASE RATE

The variation of energy release rate G i at the crack tip i with respect to the growth of any other
crack j with i
= j can be obtained as



G i T K −1 f K f K T −1 f
= −u K − u + − u K (A1)
a j ai a j a j a j a j ai
where ai is the length of crack i and u, K and f are the nodal displacement vector, the stiffness
matrix of system and the applied nodal force vector, respectively. For the case of i = j we have



G i T K −1 f K 1 T 2 K f K T −T  f 2
T f
= −u K − u − u u+ − u K +u (A2)
ai ai ai ai 2 ai2 ai ai ai ai2

in which the element stiffness variations k/a and 2 k/a 2 are defined as [30]
  T 
k B B
= DB+BT D +Tr(ε̃)BT DB dV (A3)
ai V ai ai
  2 T
T 
2 k  B BT B  2
B B B
= DB+2 D +BT D 2 +2|ε̃|BT DB+2Tr(ε̃) DB+BT D dV
ai2 V ai2 ai ai ai ai ai

(A4)
where B is the strain–nodal displacements matrix, ε̃ is the virtual strain-like matrix and D is the
elastic constitutive matrix. Note that there is no summation over index i in Equation (9). The
matrix ε̃ is defined as
 
x̄i Ni,x  ȳi Ni,x
ε̃ = (A5)
x̄i Ni,y  ȳi Ni,y

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1045

where x̄i and  ȳi are the perturbations in nodal coordinates due to virtual crack extension and
Ni are the shape functions. This equation can be rewritten using the Jacobian matrix as
   
Ni, ε̃11 ε̃12
ε̃ = J−1 [x̄i  ȳi ] = (A6)
Ni,
ε̃21 ε̃22

where and
are the local coordinates of element. The variations of the B matrix in Equations (10)
and (11) are defined as
⎡ ⎤
⎡ ⎤ N j,x 0
ε̃11 ε̃12 0 0 ⎢ ⎥
B j ⎢ ⎥⎢ N j,y 0 ⎥
=−⎣ 0 0 ε̃21 ε̃22 ⎦ ⎢


⎥ (A7)
ai ⎣ 0 N j,x ⎦
ε̃21 ε̃22 ε̃11 ε̃12
0 N j,y

The components B j /ai are assembled into a row matrix to form B/ai . The second-order
variation of B is expressed as
⎡ ⎤
⎡ ⎤ N j,x 0
ε̃  11 ε̃  12 0 0 ⎢ ⎥
2 B j ⎢ ⎥⎢N
 ⎥ ⎢ j,y
0 ⎥
=⎢
⎣ 0 0 ε̃ 
21 ε̃ 22 ⎦ ⎢ ⎥
⎥ (A8)
ai2 ⎣ 0 N j,x ⎦
ε̃  21 ε̃ 
22 ε̃ 
11 ε̃  12
0 N j,y

where ε˜ is defined as ε̃  = 2ε̃ 2 . The variations of element stiffness in Equations (10) and (11) can
be then assembled to produce the global stiffness variations K/a and 2 K/a 2 .

APPENDIX B: SMOOTHING FUNCTIONS AT THE CRACK TIP REGION

It is well known that the rate of convergence for the standard finite element solution of singular
problems, such as crack, is low. The implementation of crack tip singular elements can improve
the convergence rate considerably. Although the derivatives of singular element shape functions
represent an appropriate order of singularity due to singular mapping between the natural and
parametric coordinates, the polynomial shape functions are not able to represent such a feature.
Furthermore, the interpolation of nodal values to interior points through the element shape functions
produces the error in recovery procedure due to several gradients in the exact solution. Thus,
some modifications must be taken into the recovery process when using the SPR technique in
conjunction with singular elements. For this reason, the analytical solution of crack tip fields is
incorporated in the SPR technique as the smoothing function at the crack tip region. Smoothing
functions at the crack tip region for mode I are defined as

   
1 3  1 5 3√ I 5  1 3
x = √ a1I cos + cos +4a2I + ra3 cos − cos (B1)
2 r 2 2 2 2 2 2 2 2 2
   
1 5  1 5 3√ I 3  1 3
 y = √ a1I cos + cos + ra2 cos − cos −4ra3I cos  (B2)
2 r 2 2 2 2 2 2 2 2 2
   
1 1 5 1  3√ I 1 3 1 
x y = √ a1I sin − sin + ra2 − sin − sin −8ra3I sin  (B3)
2 r 2 2 2 2 2 2 2 2 2

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1046 H. AZADI AND A. R. KHOEI

and for mode II as


   
1 7  1 5 3 √ II 9  1 3
x = − √ a1II − sin − sin − ra2 sin + sin −8ra3II sin  (B4)
2 r 2 2 2 2 2 2 2 2 2
 
1 II 1  1 5
 y = − √ a1 − sin + sin
2 r 2 2 2 2
   
3 √ II 1  1 3 15 3/2 II 3 
+ ra2 sin + sin + r a3 sin +sin (B5)
2 2 2 2 2 4 2 2
 
1 1 5 3 
xy = √ a1II − cos − cos
2 r 2 2 2 2
   
3 √ II 1 3 5  5 3/2 II 3  7 3
+ ra2 cos − cos + r a3 cos − cos (B6)
2 2 2 2 2 2 2 2 2 2
where aiI and aiII are unknowns to be computed in the SPR technique. For mixed-mode problems,
these equations are superposed.
In order to implement the above relations into the SPR technique, the terms of above equations
are employed for each stress component as a smoothing function with three unknowns of a1I , a2I
and a3I for the first mode of fracture and three a1II , a2II and a3II for second mode of fracture. These
unknowns are then obtained through a least-square fit of selected functions to the finite element
solution at the Gauss quadrature points of crack tip patch. After solving six equations for six
unknowns aiI and aiII , the improved stresses are calculated at the interior points of crack tip rosette
directly from the smoothing function. This technique is only used for the crack tip region, while
the standard SPR technique is performed for other areas.

ACKNOWLEDGEMENTS
The authors are grateful for the research support of the Iran National Science Foundation (INSF).

REFERENCES
1. Freij-Ayoub R, Dyskin AV, Galybin AN. The dislocation approximation for calculating crack interaction.
International Journal of Fracture 1997; 86:57–62.
2. Rubinstein A. Macrocrack interaction with semi-infinite microcrack array. International Journal of Fracture 1985;
27:113–119.
3. Rubinstein A. Macrocrack–microdefect interaction. Journal of Applied Mechanics 1996; 53:505–510.
4. Rubinstein A. Crackpath effect on material toughness. Journal of Applied Mechanics 1990; 57:97–103.
5. Kachanov M. Elastic solids with many cracks: a simple method of analysis. International Journal of Solids and
Structures 1987; 23:23–43.
6. Kachanov M. A simple technique of stress analysis in elastic solids with many cracks. International Journal of
Fracture 1985; 28:11–19.
7. Kachanov M. Macrocrack interaction. International Journal of Fracture 1986; 30:65–72.
8. Rybaczuk M, Stoppel P. The fractal growth of fatigue defects in materials. International Journal of Fracture
2000; 103:71–94.
9. Chen YZ. General case of multiple cracks problems in an infinite plate. Engineering Fracture Mechanics 1984;
20:591–597.
10. Carpinteri A, Monetto I. Snap-back analysis of fracture evolution in multi-cracked solids using boundary element
method. International Journal of Fracture 1990; 98:225–241.
11. Denda M, Dong YF. Complex variable approach to the BEM for multiple crack problems. Computer Methods
in Applied Mechanics and Engineering 1997; 141:247–264.
12. Dolbow J, Moës N, Belytschko T. An extended finite element method for modeling crack growth with frictional
contact. Computer Methods in Applied Mechanics and Engineering 2001; 190:6825–6846.
13. Budyn E, Zi G, Moës N, Belytschko T. A method for multiple crack growth in brittle materials without remeshing.
International Journal for Numerical Methods in Engineering 2004; 61:1741–1770.
14. Belytschko T, Lu YY, Gu L. Crack propagation by element-free Galerkin methods. Engineering Fracture
Mechanics 1995; 51:295–315.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
NUMERICAL SIMULATION OF MULTIPLE CRACK GROWTH 1047

15. Lu YY, Belytschko T, Tabbara M. Element-free Galerkin method for wave propagation and dynamic fracture.
Computer Methods in Applied Mechanics and Engineering 1995; 126:131–153.
16. Belytschko T, Organ D, Gerlach C. Element-free Galerkin methods for dynamic fracture in concrete. Computer
Methods in Applied Mechanics and Engineering 2000; 187:385–399.
17. Rashid MM. The arbitrary local mesh replacement method: an alternative to remeshing for crack propagation
analysis. Computer Methods in Applied Mechanics and Engineering 1998; 154:133–150.
18. Barsoum RS. On the use of isoparametric finite elements in linear fracture mechanics. International Journal for
Numerical Methods in Engineering 1976; 10:25–37.
19. Henshell RD, Shaw KG. Crack-tip finite elements are unnecessary. International Journal for Numerical Methods
in Engineering 1975; 9:495–507.
20. Zienkiewicz OC, Zhu JZ. A simple error estimator and adaptive procedure for practical engineering analysis.
International Journal for Numerical Methods in Engineering 1987; 24:337–357.
21. Zienkiewicz OC, Zhu JZ. The superconvergent patch recovery (SPR) and a posteriori error estimates. Part 2:
error estimates and adaptivity. International Journal for Numerical Methods in Engineering 1992; 33:1365–1382.
22. Khoei AR, Azadi H, Moslemi H. Modeling of crack propagation via an automatic adaptive mesh refinement based
on modified superconvergent patch recovery technique. Engineering Fracture Mechanics 2008; 75:2921–2945.
23. Hwang CG, Warwzynek PA, Tayebi AK, Ingraffea AR. On the virtual crack extension method for calculating
rates of energy release rate. Engineering Fracture Mechanics 1998; 59:521–42.
24. Hwang CG, Wawrzynek PA, Ingraffea AR. On the calculation of derivatives of stress intensity factors for multiple
cracks. Engineering Fracture Mechanics 2005; 72:1171–1196.
25. Hwang CG, Ingraffea AR. Virtual crack extension method for calculating the second order derivatives of energy
release rates for multiply cracked systems. Engineering Fracture Mechanics 2007; 74:1468–1487.
26. Erdogan F, Sih GC. On the extension of plates under plane loading and transverse shear. Journal of Basic
Engineering 1963; 4:519–427.
27. Hussain M, Pu S, Underwood J. Strain energy release rate for a crack under combined mode I and mode II.
ASTM 1974; STP 560:2–28.
28. Sih G. Strain energy factors applied to mixed mode crack problems. International Journal of Fracture 1974;
10:305.
29. Li FZ, Shih CF, Needleman A. A comparison of methods for calculating energy release rates. Engineering
Fracture Mechanics 1985; 21:405–421.
30. Banks-Sills L, Sherman D. On the computation of stress intensity factors for three dimensional geometries by
means of the stiffness derivative and J-integral methods. International Journal of Fracture 1992; 53:1–20.
31. Ishikawa H. A finite element analysis of stress intensity factors for combined tensile and shear loading by only
a virtual crack extension. International Journal of Fracture 1980; 16:243–246.
32. Bui HD. Associated path independent J -integral for separating mixed modes. Journal of Mechanics and Physics
of Solids 1983; 31:439–448.
33. Lin SC, Abel JF. A variational approach for a new direct-integration form of the virtual crack extension methods.
International Journal of Fracture 1988; 28:217–235.
34. Broek D. Elementary Engineering Fracture Mechanics. Kluwer Academic Publishers: Dordrecht, 1986.
35. Zienkiewicz OC, Zhu JZ. The superconvergent patch recovery and a posteriori error estimates. Part I: the recovery
technique. International Journal for Numerical Methods in Engineering 1992; 33:1331–1364.
36. Babuška I, Strouboulis T, Upadhyay CS.
%—superconvergence of finite element approximations in the interior
of general meshes of triangles. Computer Methods in Applied Mechanics and Engineering 1995; 122:273–305.
37. Babuška I, Strouboulis T, Upadhyay CS, Gangaraj SK. Computer-based proof of the existence of superconvergence
points in the finite element method; superconvergence of the derivatives in finite element solutions of Laplace’s,
Poisson’s and the elasticity equations. Numerical Methods for Partial Differential Equations 1996; 12:347–392.
38. Zhang L, Strouboulis T, Babuška I.
%—superconvergence of finite element solutions and error estimators.
Advances in Computational Mathematics 2001; 15:393–404.
39. Zienkiewicz OC, Taylor RL. The Finite Element Method, Vol. 1: The Basis (5th edn). Butterworth-Heinemann:
London, 2000.
40. Babuška I, Strouboulis T, Upadhyay CS, Gangaraj SK, Copps K. Validation of a posteriori error estimator by
numerical approach. International Journal for Numerical Methods in Engineering 1994; 37:1073–1123.
41. Williams ML. On the stress distribution at the base of a stationary crack. ASME Transaction, Journal of Applied
Mechanics 1957; 24:109–114.
42. Paris P, Erdogan F. A critical analysis of crack propagation laws. Journal of Basic Engineering 1963; 85:528–534.
43. Nguyen QS, Stolz C, Debruyne G. Energy methods in fracture mechanics: stability, bifurcation and second
variations. European Journal of Mechanics A/Solids 1990; 9:157–173.
44. Fortino S, Bilotta A. Evaluation of the amount of crack growth in 2D LEFM problems. Engineering Fracture
Mechanics 2004; 71:1403–1419.
45. Formica G, Fortino S, Lyly M. A ϑ method-based numerical simulation of crack growth in linear elastic fracture.
Engineering Fracture Mechanics 2007; 74:1727–1738.
46. Destuynder P, Djaoua M. Sur une interpre’tation mathe’matique de l’inte’grale de Rice en The’orie de la rupture
fragile. Mathematical Methods in the Applied Sciences 1981; 3:70–87.
47. Bocca P, Carpinteri A, Valente S. Mixed mode fracture of concrete. International Journal of Solids and Structures
1991; 27:1139–1153.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme
1048 H. AZADI AND A. R. KHOEI

48. Bazant ZP, Cedolin L. Stability of Structures. Oxford University Press: New York, 1991.
49. Bazant ZP, Ohtsubo H. Stability conditions for propagation of a system of cracks in a brittle solid. Mechanics
Research Communications 1977; 4:353–366.
50. Wawrzynek PA, Ingraffea AR. An edge-based data structure for two-dimensional finite element analysis.
Engineering with Computers 1987; 3:13–20.
51. Mantyla M, Sulonen R. A solid modeler with Euler operators. IEEE Computer Graphics Applications 1982;
2:17–31.
52. Roth SD. Ray casting for modeling solids. Computer Graphics and Image Processing 1982; 18:109–144.
53. Khoei AR, Gharehbaghi SA, Azami AR, Tabarraie AR. SUT-DAM: an integrated software environment for
multi-disciplinary geotechnical engineering. Advances in Engineering Software 2006; 37:728–753.
54. Khoei AR, Moslemi H, Majd Ardakany K, Barani OR, Azadi H. Modeling of cohesive crack growth using an
adaptive mesh refinement via the modified-SPR technique. International Journal of Fracture 2009; 159:21–41.
55. Moslemi H, Khoei AR. 3D adaptive finite element modeling of non-planar curved crack growth using the
weighted superconvergent patch recovery method. Engineering Fracture Mechanics 2009; 76:1703–1728.
56. Moslemi H, Khoei AR. 3D modeling of damage growth and crack initiation using adaptive finite element
technique, Scientia Iranica, Transaction A. Journal of Civil Engineering 2010; in press.
57. Anderson TL. Fracture Mechanics: Fundamentals and Applications (2nd edn). CRC Press: Boca Raton, FL,
1994.
58. Bazant ZP, Tabbara M. Bifurcation and stability of structures with interacting propagating cracks. International
Journal of Fracture 1992; 53:273–289.
59. Cedolin L, Dei Poli S, Ion I. Experimental determination of the fracture process zone in concrete. Cement and
Concrete Research 1983; 13:557–567.
60. Bouchard PO, Bay F, Chastel Y, Tovena I. Crack propagation modeling using an advanced remeshing technique.
Computer Methods in Applied Mechanics and Engineering 2000; 189:723–742.
61. Bouchard PO, Bay F, Chastel Y. Numerical modeling of crack propagation: automatic remeshing and comparison
of different criteria. Computer Methods in Applied Mechanics and Engineering 2003; 192:3887–3908.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2011; 85:1017–1048
DOI: 10.1002/nme

Das könnte Ihnen auch gefallen