Computational Materials Science 50 (2011) 1834–1845

Contents lists available at ScienceDirect

Computational Materials Science

journal homepage: www.elsevier.com/locate/commatsci

A study of finite size effects on cracked 2-D piezoelectric media using

extended finite element method
R.R. Bhargava, Kuldeep Sharma ⇑
Department of Mathematics, Indian Institute of Technology Roorkee, Roorkee 247 667, India

a r t i c l e i n f o a b s t r a c t

Article history: The extended finite element method is applied on the two-dimensional (2-D) finite piezoelectric media
Received 29 September 2010 weakened by a crack. The fourfold standard enrichment functions are taken in conjugation with the inter-
Received in revised form 31 December 2010 action integral to evaluate the intensity factors. Four sequence of analysis, namely crack–mesh alignment,
Accepted 17 January 2011
aspect ratio, mesh with local refinement and domain independency is done on the center and edge crack
Available online 12 February 2011
problems. These four analyses when combined together give an optimum result to study the finite spec-
imen. It is observed that for smaller values of strip width to crack length ratio the finiteness of the spec-
Keywords:
imen size affects the intensity factors.
Crack
Electric-displacement intensity factor
Ó 2011 Elsevier B.V. All rights reserved.
Extended finite element method
Interaction integral
Piezoelectric ceramics
Stroh formalism

1. Introduction impermeable, permeable and semi-permeable cracks. Kuna [7,8]

gave a detailed survey of different techniques to solve piezoelectric
Piezoelectric ceramics have been widely used due to the prop- fracture mechanics problems and present status of the work done
erty of mechanical–electrical coupling, as key components in till date in piezoelectrics.
sensors/actuators/transducers. However, piezoelectric ceramics The extended finite element method (X-FEM) proposed by Moës
are very brittle in nature, therefore fracture analysis becomes et al. [9], Belytschko and Black [10] have proven to be a very effi-
imperative for such materials. cient tool for the numerical modeling of cracks in LEFM. In compar-
The mathematical methods used in linear elastic fracture ison to the standard FEM, the X-FEM provides significant benefits
mechanics (LEFM) such as complex function theory (Lekhnitski, in the numerically modeling of crack propagation. The main advan-
Stroh), integral transformation and singular integral equations tages are that the finite element mesh needs not to conform to the
have been applied/extended to piezoelectric problems using linear crack boundaries to account for the geometric discontinuity, and
approximation. Many two-dimensional standard crack configura- furthermore, mesh regeneration is not needed in crack growth
tions in infinite domains with different possible electric conditions simulations. The X-FEM has also been applied to various fields in
at the crack surface have been treated successfully. engineering [11,12].
However for practical applications and analysis of fracture test More recently X-FEM [13] has been applied to investigate
specimens, the bounded geometry, the complex electromechanical enrichment schemes and their convergence for semi-infinite crack
boundary conditions and material non-linearity requires the and Griffth–Irwin crack weakening a piezoelectric material. The
numerical analysis as finite elements method (FEM) or boundary problem investigated using analytic calculation based on Lekhnit-
elements method (BEM). Allik and Hughes [1] were the first to ski formalism. It has been concluded that fourfold enrichment
apply FEM to investigate vibrations in piezoelectric ceramics. functions used in isotropic materials is almost efficient, concerning
Kumar and Singh [2,3], investigated crack propagation and energy accuracy both in energy and intensity factors (IFs). Implementation
release rate under combined mechanical and electrical loadings. of using these fourfold enrichment functions becomes simpler and
While Kuna et al. [4–6] applied FEM analysis to develop electrome- also involves less computation.
chanical J-integral, an equivalent domain integral for a crack piezo- To the best of the author’s knowledge, there is paucity in liter-
electric 2-D and 3-D problems. They considered the case of ature on X-FEM solutions for finite piezoelectric domains, particu-
larly those encountered in practical situations.
⇑ Corresponding author. Tel.: +91 9456318765. We attempt this paucity, to investigate the effect of finite
E-mail address: [email protected] (K. Sharma). dimension of specimen on a cracked piezoelectric material.

0927-0256/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.commatsci.2011.01.026
 R.R. Bhargava, K. Sharma / Computational Materials Science 50 (2011) 1834–1845 1835

X-FEM has been applied using fourfold enrichment functions under 3. Crack tip fields in homogeneous piezoelectric media
in-plane tensile and electric loading. Two benchmark problems:
center and edge crack have been discussed in this paper. The open- For cracks in homogeneous piezoelectric media the asymptotic
ing mode stress intensity factor (KI) and electric-displacement behavior of the field quantities has been given by Sosa [14] and Pak
intensity factor (KIV) have been evaluated using an interaction inte- [15]. The electromechanical stress and electrical displacement
gral technique [7,8] and the crack tip behavior has been given by fields can be written in polar coordinates (r, h) (with the origin at
Stroh formalism [14,15]. To reduce the computational cost a mesh the crack tip) as
with local refinement has been taken and also the results of KI and X
1
KIV are compared with those of the structured mesh. rij ðr; hÞ ¼ pffiffiffiffiffiffiffiffiffi K N fijN ðhÞ; ð9Þ
2pr N
1 X
Di ðr; hÞ ¼ pffiffiffiffiffiffiffiffiffi K N g Ni ðhÞ; ð10Þ
2. Basic equations for piezoelectric media 2pr N

As is well-known the governing equations and the boundary and the near tip displacement field and electrical potential can be
conditions which form the foundation of piezoelectric media are obtained from
given below. rffiffiffiffiffi
2r X N
ui ðr; hÞ ¼ K N di ðhÞ; ð11Þ
p
2.1. Field equations rffiffiffiffiffi
N

2r X
/ðr; hÞ ¼ K N mN ðhÞ; ð12Þ
In a fixed rectangular coordinate system xj (j = 1, 2, 3), the field p N
equations for a linear piezoelectric medium subjected to electro-
mechanical loads in the absence of body forces and charges are where i, j = 1,2, and the summation over N = {II, I, III, IV} comprises
the fracture opening mode intensity factors KII and KIII denote
 Constitutive equations mode-II, mode-III stress intensity factor, respectively. In case of
two-dimensional piezoelectric structure, KIII = 0. The functions
N
rij ¼ C ijks eks  esij Es ; Di ¼ eiks eks þ jis Es ; ð1Þ fijN ðhÞ; g Ni ðhÞ; di ðhÞ, and mN(h), are the standard angular functions for
a crack in a homogeneous piezoelectric–elastic medium, which de-
 Kinematic equations
pend only on the material properties and can be determined by
1
eij ¼ ðui;j þ uj;i Þ; Ei ¼ /;i ; ð2Þ means of the extended Stroh formalism and semi-analytical calcu-
2
lations. They can be expressed in terms of complex material eigen-
 Equilibrium equations values pa, eigenvectors AMa and matrices MMa and NaN (Kuna [7,8];
Rao and Kuna [16]) as
rij;j ¼ 0; Di;i ¼ 0; ð3Þ
( )
X
4
M ia NaN pa
where rij, eij, Di and Ei denote the components of the stress, strain, fi1N ¼  Re pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ;
electric displacement, and electric field; Cijks and eiks denote the a¼1 cos h þ pa sin h
elastic and piezoelectric constants, respectively; jis denotes the ( )
X4
M ia N aN
dielectric permitivities. In Eqs. (2) and (3), comma denotes partial fi2N ¼ Re pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; ð13Þ
differentiation with respect to argument following it; ui is the com- a¼1 cos h þ pa sin h
ponent of the elastic displacement vector u; / is the electric poten- ( )
tial; where i, j, k and s = 1, 2, 3. X
4
M 4a N aN pa
g N1 ¼ Re p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ;
a¼1 cos h þ pa sin h
2.2. Boundary conditions ( )
X4
M 4a N a N
g N2 ¼ Re pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; ð14Þ
Consider a piezoelectric medium occupying the space X en- a¼1 cos h þ pa sin h
closed by surface S. On the boundaries Sr and SD, the resultant of
stresses and electric displacements are respectively X
4  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
N
di ¼ Re Aia NaN cos h þ pa sin h ;
a¼1
rij nj ¼ t0j ; on Sr ; ð4Þ  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X
4
Dj nj ¼ x0 ; on SD ; ð5Þ mN ¼ Re A4a NaN cos h þ pa sin h ; ð15Þ
a¼1
where t0j is prescribed on surface Sr and x0 is given on surface SD. n where Re{. . .} denote the real part, of the quantity in brackets.
is the outward drawn unit normal vector on S. The displacement The four conjugate pairs of eigenvalues pa can be obtained by
vector u is prescribed on surface Su and the electric potential / is solving the following
prescribed on surface S/ as   
C i1k1 ei11 C i2k1 þ C i1k2 ei21 þ ei12
þ p
uj ¼ u0j ; on Su ; ð6Þ e1k1 j11 e2k1 þ e1k2 j12  j21
   
/ ¼ /0 ; on S/ ð7Þ C i2k2 ei22 Ai
þ p2 ¼ 0: ð16Þ
e2k2 j22 A4

2.3. Impermeable crack conditions Only the four eigenvalues having positive imaginary part and the
corresponding eigenvectors are used as a column vectors in the ma-
In case of impermeable crack conditions on the upper and lower trix A. The matrices MMa and NaN are calculated by
crack faces Sc,  
ðC i2k1 þ C i2k2 pa ÞAka ðe1i2 þ e2i2 pa ÞA4a
N1
aN ¼ M M a ¼ : ð17Þ
Dj nj ¼ 0; on Sc : ð8Þ ðe2k1 þ e2k2 pa ÞAka ðj21  j22 pa ÞA4a
1836 R.R. Bhargava, K. Sharma / Computational Materials Science 50 (2011) 1834–1845

4. Extended finite element approximation

Modeling of cracks using FEM is cumbersome in 2-D for com-

plex structures or crack geometries. But X-FEM models crack(s)
geometry independent of the mesh leading to a simplification in
mesh generation and avoids remeshing as the crack grows.
The X-FEM exploits the partition of unity property of FEM first
identified in [9], which allows local enrichment functions to be
easily incorporated into a finite element approximation. A standard
approximation is thus enriched in a region of interest by the local
functions in conjunction with additional degrees of freedom. For
crack problems the enrichment functions are the near tip asymp-
totic fields and a discontinuous function to represent the jump in
the displacement across the crack boundary.

4.1. Description of crack geometry Fig. 1. The set of nodes elected for enrichment.

To represent a crack, Level Set Method (LSM) has been applied

which is a numerical method proposed by Osher and Sethian parameters. Fk(r, h) is the basis for the Westergaard field for the
[17]. The principle of the method is to represent an interface by crack tip, which are defined as
the zero of a function, called level set function, and to update this  
pffiffiffi h pffiffiffi h pffiffiffi h pffiffiffi h
function with Hamilton–Jacobi equations knowing the speed of the F k ðr; hÞ ¼ r sin ; r cos ; r sin sin h; r cos sin h ;
2 2 2 2
interface in the direction normal to this interface.
In two-dimensions, the crack is an open curve bounded by the where r is the kx  xTIPk; N the set of all nodes in the discretization;
crack tip(s) and in three-dimensions, it is an open surface bounded NTIP the set of all nodes that are connected to elements containing
by the crack front. An extension of the LSM is proposed by crack tip(s); and Ncr is the set of nodes that are connected to ele-
Stolarska et al. [18] to represent open boundaries using two level ments containing the crack but not in NTIP.
set functions The set of nodes elected for enrichment are shown in Fig. 1.
Substituting the approximate displacement from Eq. (18) and the
 One the normal level set function, w1, that is the signed distance electric potential from Eq. (19) into the weak form illustrated in
to the union of the crack and the tangent extension from its [19], the standard discrete system of equations is obtained
front,
K s d ¼ f ext ð21Þ
 And other the tangent level set function, w2, that is the signed
distance function to a surface that passes by the crack boundary where fext is the vector of external nodal forces and Ks the stiffness
and that is normal to the crack. matrix.
The elementary stiffness matrix for an enriched element is de-
The crack surface is defined as the subset of the zero level set of fined as
w1 where w2 is negative. The crack front is defined as the intersec- 2 UU Ua Ub 3
kij kij kij
tion of the two zero level sets. 6 aU
e aa ab 7
kij ¼ 6
4 kij kij kij 7
5 ð22Þ
4.2. Enriched approximation bU
kij
ba
kij
bb
kij

Considering the impermeable boundary conditions defined in e UU

Section 2.3 at the crack surface, the extended finite element and in case of non-enriched element kij ¼ kij : ð23Þ
approximation for the displacement and electric potential can be While the element contribution to the global element force vector
written as follows: fext is
X X
uh ðxÞ ¼ NI ðxÞuI þ NI ðxÞðHðf h ðxÞÞ  HðfI ÞÞaI
T
fie ¼ fiU fia fib and fie ¼ fiU for the non-enriched element: ð24Þ
I2N I2Ncr

X X
4
k
where
þ NI ðxÞ ðF k ðr; hÞ  F k ðxI ÞÞbI ; ð18Þ
k k
I2N TIP k¼1 U ¼ fu/gT ; a ¼ faI cI gT ; b ¼ fbI dI g and
Z
X X rs  r T s 
/h ðxÞ ¼ NI ðxÞ/I þ NI ðxÞðHðf h ðxÞÞ  HðfI ÞÞcI kij ¼ Bi C Bj dX ðr; s ¼ U; a; bÞ; i; j ¼ 1; 2; 3; 4; ð25Þ
Xe
I2N cr
I2N
Z Z
X X
4
þ NI ðxÞ
k
ðF k ðr; hÞ  F k ðxI ÞÞdI ; ð19Þ fiU ¼ NitdC þ Ni fdC;
I2NTIP k¼1 Z @ Xe Xe
Z
where H(f(x)) is a modified Heaviside step function fia ¼ Ni ðHðf h ðxÞÞ  Hðfi ÞÞtdC þ Ni ðHðf h ðxÞÞ  Hðfi ÞÞfdC;
@ Xe Xe
  Z Z
1 if y < 0 fib ¼ N i ðF k ðxÞ  F k ðxi ÞÞtdC þ Ni ðF k ðxÞ  F k ðxi ÞÞfdC;
HðyÞ ¼ : ð20Þ
þ1 if y > 0 @ Xe Xe
ð26Þ
And the shape functions, NI(x), are isoparametric linear quadrilat-
eral element shape functions that construct the partition of unity. Ni is the standard finite element shape function defined at
The column matrices uI and /I are the nodal displacements and node i, and B0i s are the nodal matrices of the shape derivatives.
k k
electric potential respectively, and aI, bI and cI, dI are the additional The detailed of this method is given in the following flow chart
 R.R. Bhargava, K. Sharma / Computational Materials Science 50 (2011) 1834–1845 1837

ðIFsX-FEM  IFsAnalytically Þ
Percentage error in IFs ¼  100; ð30Þ
IFsAnalytically
and
IFsX-FEM
Normalized IFs ¼ ; ð31Þ
IFsAnalytically
respectively.
In this section we present numerical results for the computation
of the intensity factors in piezoelectric material using interaction
integral. The interaction integral method is an effective tool for cal-
culating the SIFs and EDIF in the homogeneous piezoelectric mate-
rials [16,22].
The path independent electromechanical J-integral for a homo-
geneous piezoelectric cracked body is given by Cherepanov [23].
Z  
@ui @/
J¼ Wd1j  rij  Dj nj dC; ð32Þ
C @x1 @x1
R R
where W = rijdeij  DidEi is the electric enthalpy density, nj is the
jth component of the outward unit vector normal to an arbitrary
contour C enclosing the crack tip and dij is the Kronecker delta.
For linear piezoelectric material, an equivalent domain form is gi-
ven as
Z  
@ui @/ @q
J¼ rij þ Dj  Wd1j dA; ð33Þ
A @x1 @x1 @xj
where A is the area inside the contour C and q is a smooth weight
function chosen such that it has a value of unity at the crack tip,
zero along the boundary of the domain C, and smoothly interpo-
lated in between.
Consider two independent equilibrium states of the cracked
body: let state 1 correspond to the actual state for the given bound-
ary conditions, and let state 2 correspond to an auxiliary state,
which can be near tip electromechanical fields of any of the frac-
ture opening modes I, II, III and IV. Superposition of these two
states leads to another equilibrium state called (state s) for which
the domain form of the J-integral is
0 
Z  @ uið1Þ þ uið2Þ
J ðsÞ
¼ @ r þr
ð1Þ ð2Þ
ij ij
A @x1
5. Interaction integral !
 ð1Þ
þ /ð2Þ Þ
ð1Þ ð2Þ @ð/ ðsÞ @q
þ Dj þ Dj  W d1j dA; ð34Þ
For an infinite domain problem in piezoelectric fracture @x1 @xj
mechanics under tensile stress r1 and electric-displacement D1
on the remote boundary, the mode-I stress intensity factor KI and where superscript i = 1, 2, and s indicate fields and quantities asso-
electric-displacement intensity factor (EDIF) KIV at the crack tip ciated with state i and
pffiffiffiffiffiffi pffiffiffiffiffiffi n    o.
are r1 pa and D1 pa respectively. In case of a finite dimensional
W ðsÞ ¼ rð1Þ ð2Þ
ij þ rij eð1Þ ð2Þ
ij þ eij
ð1Þ ð2Þ
 Dj þ Dj
ð1Þ ð2Þ
Ej þ Ej 2:
specimen, a dimensionless correction factor [20,21] that depends
on the ratio wa is introduced for the IFs as
pffiffiffiffiffiffi pffiffiffiffiffiffi By expanding Eq. (34), we may write
K I ¼ C r1 pa and K IV ¼ CD1 pa: ð27Þ
J ðsÞ ¼ J ð1Þ þ J ð2Þ þ Ið1;2Þ ;
where
where J(1) and J(2) are the electromechanical J-integrals for states 1
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pa ffi a 2 a 4  and 2 respectively, and interaction integral I(1,2) is
C ¼ sec 1  0:025 þ 0:06 ; ð28Þ !
2w w w Z ð2Þ ð2Þ ð1Þ ð1Þ
@ui ð1Þ @/ ð2Þ @ui ð2Þ @/ @q
Ið1;2Þ ¼ rð1Þ
ij þ D j r þ D  W ð1;2Þ
d1j dA;
a is the semi-crack length and w is the semi-width of the specimen A @x1 @x1 ij @x1 j
@x1 @xj
for the center crack problem and ð35Þ
a a 2 a 3 a 4 
ð1;2Þ ð1Þ ð2Þ ð2Þ ð1Þ ð1Þ ð2Þ ð2Þ ð1Þ
C ¼ 1:12  0:231 þ 10:55  21:72 þ 30:39 ; In Eq. (35), W ¼ r e þr e 
ij ij ij ij Dj Ej  Dj Ej =2:
w w w w
ð29Þ As is well-known for piezoelectric solids under mixed-mode
loading conditions; the electromechanical J-integral can be written
a is the edge crack length and w is the width of the specimen in case (using Kuna [7]) as
of edge crack problem.
The percentage error and normalized value in IFs are calculated 1 T
J¼ K YK; ð36Þ
by 2
1838 R.R. Bhargava, K. Sharma / Computational Materials Science 50 (2011) 1834–1845

where K ¼ fK II K I K III K IV gT is the vector of the intensity factors, and Y ryy = 10 MPa, electric displacement Dy = 0.001 C/m2. The material
is the 4  4 generalized Irwin matrix, which depends on the elastic, considered for the analysis in this paper is PZT-5H and for which
piezoelectric, and dielectric material constants and is given by material constants are given in Table 1 taken from [24]. The IFs
are calculated on a domain radius equal to 0.8a.
Y MN ¼ ImfAMa NaN g: ð37Þ
Therefore, the interaction integral for two-dimensional case is given
6.1.1. Crack–mesh alignment vs. Non-alignment
by
The analysis of crack–mesh alignment vs. non-alignment has
ð1Þ ð2Þ ð1Þ ð2Þ ð1Þ ð2Þ
Ið1;2Þ ¼ K II K II Y 11 þ K I K I Y 22 þ K IV K IV Y 44 been done on the structured mesh Figs. 4 and 5. The effect of
  crack–mesh alignment is considered for two possible cases: full
ð1Þ ð2Þ ð1Þ ð2Þ ð1Þ ð2Þ ð1Þ ð2Þ
þ K I K II þ K II K I Y 12 þ K II K IV þ K IV K II Y 14 alignment and non-alignment. The elements used in the mesh con-

ð1Þ ð2Þ ð1Þ ð2Þ
þ K I K IV þ K IV K I Y 24 : ð38Þ

The individual IFs for the actual state can be obtained by judi-
ciously choosing the auxiliary state (state 2). If state 2 is taken to
be

(i) For fracture opening mode-I

ð2Þ ð2Þ ð2Þ

then, K I ¼ 0; K II ¼ 1; and K IV ¼ 0:
Hence, Eq. (38) can be reduced to
ð1Þ ð1Þ ð1Þ
Ið1;IÞ ¼ K I Y 22 þ K II Y 12 þ K IV Y 24 : ð39Þ
(ii) For fracture opening mode-II
then Eq. (38) reduces to
ð1Þ ð1Þ ð1Þ
Ið1;IIÞ ¼ K I Y 12 þ K II Y 11 þ K IV Y 14 ; ð40Þ
and
(iii) For fracture opening mode-IV

ð1Þ ð1Þ ð1Þ

Ið1;IVÞ ¼ K I Y 24 þ K II Y 14 þ K IV Y 44 ; ð41Þ
respectively.

ð1Þ ð1Þ
Solving the simultaneous Eqs. (39)–(41), the IFs K I ; K II and
ð1Þ
K IVare obtained. The interaction integrals I(1,I), I(1,II) and I(1,IV) are
evaluated from Eq. (35).

6. Numerical applications
Fig. 2. Model geometry of the center crack.

Two benchmarks problems in fracture mechanics of piezoelec-

tric materials are considered with the following objectives

 Comparison /trends of IFs in case of crack line and tip alignment

(full alignment) to the mesh with neither the crack line nor the
crack tip align to the mesh (non-alignment).
 Influence of aspect ratio on IFs accuracy.
 To show the accuracy obtainable on mesh with local refinement
having full alignment case.
 To study the domain independence in the IFs computations.
 Explore finite size specimen effects and trends of the IFs.

6.1. Center-crack in-plane tension and electric loading

First case considered is of a center-cracked test specimen under

in-plane tension and electrical loading. The model geometry
shown in Fig. 2 consists of a homogeneous poled piezoelectric plate
of length L = 2h and width 2w. The plate is weakened by a center
crack occupies the interval [a, a]. The crack is perpendicular to
the polarization direction. The plate is subjected to tensile stress
r1 or ryy and electric-displacement D1 or Dy on the remote
boundary.
Due to symmetry, only half the specimen is considered as
shown in Fig. 3 with appropriate displacement boundary
conditions to remove rigid-body motion and the edge singularity.
The plain strain case has been discussed with tensile loading Fig. 3. Half plate symmetry for center crack problem.
 R.R. Bhargava, K. Sharma / Computational Materials Science 50 (2011) 1834–1845 1839

Table 1
Material constants of PZT-5H.

Properties PZT-5H
Elastic constants c11 = 126 GPa c12 = 55 GPa
c13 = 53 GPa c33 = 117 GPa
c44 = 35.3 GPa
Piezoelectric constants e15 = 17 C/m2 e31 = 6.5 C/m2
e33 = 23.3 C/m2
Permittivity j11 = 15.1 nC/V m j33 = 13.1 nC/V m

Fig. 6. Enrichment scheme in full alignment case.

Fig. 4. Structured mesh with full alignment.

Fig. 7. Enrichment scheme in non-alignment case.

9 show the convergence of IFs with respect to the number of ele-

ments per unit length. The results of IFs are presented for both
the cases, i.e. full alignment and non-alignment to the mesh using
X-FEM, and further these results are compared with the FEM or
without enrichment. The results of FEM are calculated on the same
structured mesh and number of elements. It clearly shows that the
results of X-FEM are better than the conventional FEM results. The
Fig. 5. Structured mesh with non-alignment. convergence in case of full alignment and non-alignment has been
obtained for 1/he > 50. It is observed that behavior of convergence
in KI and KIV is the same vis-a-vis for all the three cases.
sist of uniform four noded quadrilateral isoparametric element. It is also observed that the results with the non-alignment case
The enrichment scheme used in both the cases is shown in Figs. are the best one and as the mesh density increases the results of
6 and 7. Plate dimensions of semi-width w = 2 and length L = 4, full alignment case converges to non-alignment case in X-FEM.
respectively, and a half-crack length a = 0.4 are used. Figs. 8 and The trends of KI in both the cases are in support of the work [25].
1840 R.R. Bhargava, K. Sharma / Computational Materials Science 50 (2011) 1834–1845

for the same number of elements. It is observed that the percent-

age of increment in Dofs is very small in comparison to the total
number of elements employed in the mesh. Fig. 10 shows the com-
putational time taken by the CPU to run the Matlab code for X-FEM
and FEM with respect to the mesh density. It is found that full
alignment case takes more computational time than the FEM and
this difference increases further with the mesh density. Even using
the less number of elements in non-alignment case takes almost
the same computational time as in the case of FEM. But still the dif-
ference is not more than 10% for higher mesh density. The accuracy

Fig. 8. Convergence study of KI w.r.t 1/he for center crack problem.

Fig. 10. Computational comparison of X-FEM and FEM w.r.t 1/he for center crack
problem.

Fig. 9. Convergence study of KIV w.r.t 1/he for center crack problem.

The results of IFs are tabulated in Table 2 for different specimen

sizes and for a constant value of a/w = 0.2. It is found that the
trends of IFs are independent of the specimen size. The results
are only deviating with respect to the values of he/a. This may allow
the authors to work on a higher dimensions with variation in the
values of he/a.
As the application of X-FEM involves additional degree of free-
doms due to enrichment, so it takes more computational cost than
the FEM on the same structured mesh. The comparison of degree of
freedoms (Dofs) used in X-FEM and FEM are tabulated in Table 3 Fig. 11. Influence of aspect ratio in the center crack problem.

Table 2
Comparison of IFs w.r.t no. of elements for different specimen sizes in the center crack problem.

No. of elements he/a w = 2, L = 4, a = 0.4 w = 5, L = 10, a = 1.0 w = 8, L = 16, a = 1.6

he % Error in KI % Error in KIV he % Error in KI % Error in KIV he % Error in KI % Error in KIV
20  40 0.25 0.1 0.32 2.08 0.25 0.41 2.13 0.4 0.36 2.11
40  80 0.125 0.05 0.43 2.24 0.125 0.43 2.19 0.2 0.43 2.25
60  120 0.0833 0.033 0.46 2.34 0.0833 0.46 2.33 0.1333 0.46 2.34
80  160 0.0625 0.025 0.47 2.36 0.0625 0.47 2.36 0.1 0.47 2.36
100  200 0.0250 0.01 0.48 2.41 0.05 0.48 2.39 0.08 0.48 2.41
 R.R. Bhargava, K. Sharma / Computational Materials Science 50 (2011) 1834–1845 1841

Fig. 12. Mesh with local refinement.

Fig. 14. Structured mesh and mesh with local refinement vs. w for the center crack
problem.

Fig. 13. Comparison between structured mesh and MESH with local refinement for
the center crack problem.

of the results with X-FEM on the structured mesh in comparison to

the FEM is always preferable. This will also avoid the computa-
tional cost involved in using higher order elements and the proper Fig. 15. Computational comparison between the structured mesh and mesh with
meshing techniques in conventional FEM. local refinement.

The size of the specimen and crack length is taken as w = 5,

6.1.2. Aspect ratio analysis L = 10 and a = 1.0. The results depicted in Fig. 13 show good accu-
Influence of the plate aspect ratio (i.e. h/w), on the IFs calcula- racy and agreement of IFs in both type meshes. Fig. 14 depicts the
tions is investigated. The semi-width of the plate is taken as variation of KI and KIV with respect to w for structured mesh and
w = 5, the half-crack length is a = 1 and an element length near mesh with local refinement considering he = 0.1. It is observed in
the tip (he) equal to 0.25. The semi-length of the specimen is then this case also that the results for each of KI and KIV in both the cases
varied from 8 to 30. Fig. 11 shows the percentage error in KI and KIV are almost identical. Graphs are plotted in Fig. 15 for the total
with respect to aspect ratio and is calculated on the structured number of elements required in structured mesh and mesh with
mesh for full alignment case. It is observed in Fig. 11, a dramatic local refinement with respect to 1/he. It is amply clear from the
reduction in the measured error can be seen around h/w = 1.0, be- graphs drawn that mesh with local refinement needs almost one
yond it the percentage error stabilizes to 0.8% in case of KI and third of the elements as compared to that in case of structured
1.0% in case of KIV. The deviation of the X-FEM results from the mesh. Consequently the computation cost is drastically reduced
theory at low aspect ratio is most likely due to the fact that the ap- with use of mesh with local refinement. And this is one of the
plied loads can no longer be considered as far-field loadings. The objectives of present work. In further computational analysis for
behavior of KI is validated by the work of McNary [25]. center crack the mesh with local refinement has been used.

6.1.3. Structured mesh and Mesh with local refinement

Results for KI and KIV are compared for structured mesh shown 6.1.4. Domain independency
in Fig. 4 with the mesh having local refinement shown in Fig. 12. The accuracy of interaction integral on numerical solution for
The case considered for the crack and its tip align with the mesh. different domain sizes has been tabulated in Table 4. The sample
1842 R.R. Bhargava, K. Sharma / Computational Materials Science 50 (2011) 1834–1845

Table 3  For the study of finite specimen size, the width of the specimen
Comparison of degree of freedoms used in X-FEM and FEM. is the only varying parameter. For a fixed length of an element
No. of Dofs Increment No. of Dofs near the crack tip, the higher values of w require more number
elements in Dofs due to elements of elements in case of structured mesh. Therefore mesh with
enrichment local refinement is preferred.
(in %)
 Domain independency analysis suggested us to take the domain
Full- FEM Non-alignment
alignment radius equal to 0.8a for better results of far off crack tip
elements.
20  40 2700 2595 4.04 19  39 2466
40  80 10,092 9987 1.05 39  79 9690
60  120 22,284 22,179 0.47 59  119 21,714 The IFs KI and KIV have been tabulated in Table 5. The normal-
80  160 39,276 39,171 0.27 79  159 38,538 ized values of KI and KIV are calculated with the help of Eqs. (27),
100  200 61,068 60,963 0.17 99  199 60,162 (28), and (31). As expected, with an increase in w/a, the numeri-
cally computed results obtained for both KI and KIV are collectively
best and approximate to the exact solution with less than 0.23% er-
Table 4 ror. For w/a P 20, finite specimen effects are negligible, and hence
Domain independence study for center crack. specimen dimensions of that order or larger can be used with con-
he/a Radius (rd) Normalized KI % Error Normalized KIV % Error in KIV fidence to model the infinite domain problem.
in KI
0.25 0.5 1.0028 0.28 1.0226 2.26 6.2. Edge crack in-plane tension and electric loading
0.6 1.0030 0.30 1.0229 2.29
0.7 1.0031 0.31 1.0238 2.38 In case of edge crack problem, a specimen of width w, length L
0.8 1.0032 0.32 1.0235 2.35
and a crack of length a is taken. A tensile loading r1 = ryy = 10 Mpa
0.9 1.0032 0.32 1.0235 2.35
and an in-plane electric loading D1 = Dy = 0.001 C/m2 is prescribed.
0.1 0.4 1.0042 0.42 1.0232 2.32
The model geometry and the load specified are shown in Fig. 16.
0.5 1.0045 0.45 1.0234 2.34
0.6 1.0046 0.46 1.0237 2.37 The plain strain case has been considered. The domain, mesh and
0.7 1.0044 0.44 1.0235 2.35 the data for the analysis of objectives mentioned in Section 6 are
0.8 1.0044 0.44 1.0235 2.35 the same as in the center crack problem, respectively. The IFs are
calculated on a domain radius equal to 0.8a. The only change is
size chosen is the same as in Section 6.1.3 with full alignment case. in the symmetric boundary conditions and the geometric correc-
Results are listed for two different values of he/a = 0.25 and 0.1. The tion factor [21].
numerically calculated IFs KI and KIV are normalized by using Eqs. The analysis in this case is almost parallel to that for the center
(27), (28), and (31). The domain for computation of interaction crack when considered half-plate symmetry. Here too study for
integral is selected as the domain that falls within a ball of radius crack–mesh alignment, aspect ratio, mesh with local refinement
rd. In Table 4, the results of the domain independence study and domain independency has been carried out.
are listed. It has been observed that the normalized IFs are
approximately the same, consequently domain independency for 6.2.1. Crack–mesh alignment vs. Non-alignment
rd > 0.6 (in case he = 0.25) and rd > 0.4 (in case of he = 0.1) is To study the convergence of IFs a specimen having width w = 2,
established. length L = 4 and an edge crack length a = 0.4 has been considered.

6.1.5. Finite size specimen effects

Effect of finite size of specimen is carried out including.

 The length of the element near the tip is kept fixed which
requires the full alignment in case of structured mesh.
 As is observed in the aspect ratio analysis and for optimal com-
putational cost, a square specimen gives the best result.

Table 5
Finite specimen effects for center crack.

he/a w/a Normalized KI % Error Normalized KIV % Error

in KI in KIV
0.25 rd = 0.8 5 1.0032 0.32 1.0235 2.35
7 1.0004 0.04 1.0125 1.25
10 0.9991 0.09 1.0061 0.61
20 0.9980 0.20 1.0011 0.11
30 0.9977 0.23 1.0001 0.01
40 0.9977 0.23 1.0001 0.01
0.1 rd = 0.8 5 1.0044 0.44 1.0235 2.35
10 1.0009 0.09 1.0057 0.57
20 0.9997 0.03 1.0005 0.05
30 0.9997 0.03 1.0005 0.05
0.05 rd = 0.8 5 1.0045 0.45 1.0238 2.38
10 1.0011 0.11 1.0061 0.61
Fig. 16. Model geometry of edge crack.
 R.R. Bhargava, K. Sharma / Computational Materials Science 50 (2011) 1834–1845 1843

Figs. 17 and 18 depict the variation of percentage error of KI and KIV It is also found that the values of IFs in full alignment case and
with respect to the number of elements per unit length in full non-alignment case are not differing significantly vis-a-vis each
alignment and non-alignment case. The results are also compared other. The values of KI are better in full alignment case than the
with the FEM or without enrichment on the same structured mesh non-alignment case. This is not observed in the center crack prob-
and number of elements. From the graphs as could be observed lem. The trends of KI for edge crack problem are in agreement with
that the values of KI show a good convergence rate whereas KIV [25].
has almost the constant values and have a very low convergence Table 6 shows the results of IFs with respect to the number of
rate in respect to all the cases for analysis. This is somewhat differ- elements per unit length for different specimen sizes. It is observed
ent from the center crack problem. that as in the case of center crack here too the trends of IFs are
same and they are independent of the specimen size. The only var-
iation in IFs are found with respect to he/a. So, in further sections a
structured mesh with full alignment case having specimen size of
w = 5, L = 10 and an edge crack length a = 1.0 has been taken for the
analysis.
Fig. 19 shows the computational time taken by the CPU to run
the Matlab code for X-FEM and FEM with respect to the number
of elements per unit length. It is found that in case of edge crack
problem the computational time taken is less than the center
crack. But the behavior of computational time taken in all the cases
is similar to the center crack problem. Again the key issue is the
accuracy of the results obtained by X-FEM on a structured
mesh.

6.2.2. Aspect ratio analysis

Fig. 20 illustrates the influence of aspect ratio on the IFs KI and
KIV. It is observed that for aspect ratio, h/w P 1, the variation in the
results for KI and KIV is almost constant. It is noted that this behav-
ior matches with that of the center crack case. Again, the more per-
centage error at low aspect ratios is most likely due to the
decreased validity of far-field loadings. It has been also supported
by the analysis done in [25].
Fig. 17. Convergence study of KI w.r.t 1/he for edge crack problem.

6.2.3. Structured mesh and mesh with local refinement

Figs. 21 and 22 show the behavior of structured mesh and mesh
with local refinement with respect to mesh density and width of
the specimen, respectively. The results for each of KI and KIV are
in good agreement for structured and mesh with local refinement.
And as the mesh density is increased the results for two cases
match with each other. It has been concluded that mesh with local
refinement is independent of the nature of the problems
considered.

6.2.4. Domain independency

Normalized values of IFs KI and KIV are shown in Table 7. Two
cases of he/a are considered. Results show that for rd > 0.5 (in case
of he = 0.25) and rd > 0.4(in case of he = 0.1) are very near to each
other and therefore may be considered to be domain independent.
The domain independence was also observed in the center crack
case.

6.2.5. Finite size specimen effects

As is done in the case of center crack case, here too the first four
objectives implemented for edge crack problem have been applied
Fig. 18. Convergence study of KIv w.r.t 1/he for edge crack problem. on the finite specimen effect analysis. The results are given in Table

Table 6
Comparison of IFs w.r.t no. of elements for different specimen sizes in the edge crack problem.

No. of elements he/a w = 2, L = 4, a = 0.4 w = 5, L = 10, a = 1.0 w = 8, L = 16, a = 1.6

he % Error in KI % Error in KIV he % Error in KI % Error in KIV he % Error in KI % Error in KIV
20  40 0.25 0.1 0.43 45.15 0.25 0.51 44.77 0.4 0.43 45.15
40  80 0.125 0.05 0.21 45.87 0.125 0.26 45.7 0.2 0.21 45.87
60  120 0.0833 0.033 0.18 45.99 0.0833 0.18 45.99 0.1333 0.18 45.99
80  160 0.0625 0.025 0.17 46.00 0.0625 0.17 46.00 0.1 0.17 46.01
100  200 0.0250 0.01 0.14 46.12 0.05 0.14 46.12 0.08 0.14 46.13
1844 R.R. Bhargava, K. Sharma / Computational Materials Science 50 (2011) 1834–1845

Fig. 19. Computational comparison of X-FEM and FEM w.r.t 1/he for the edge crack Fig. 22. Structured mesh and mesh with local refinement vs. w for the edge crack
problem. problem.

8. In case of edge crack the results are quite interesting. For he

= 0.25, with the increase in the values of w/a, results of IFs KI and
KIV approximate to the exact solution for infinite domain. As the
length of the element near the tip is decreased, the percentage er-
ror in KI increases and in KIV decreases with the increasing values
w/a. It is to be noted that the amount of increment in KI is very less
than the decrement in KIV. It may be remembered that for piezo-
electric ceramics the collective behaviors of KI and KIV is very
important. It is also to be noted that for w/a > 20, the percentage
error decreases with increasing values of w/a, for both IF’s KI and
KIV. In nutshell, it may be emphasized that the results are
collectively better as w/a ratio increased. For still higher values
of w/a P 20, the domain of the specimen may be considered as
an infinite one.
As the value of percentage error in KIV is higher than the center
crack case, so the results of X-FEM are also compared with the
Wang and Mai [26] for different electrical loadings and a/w ratio.
The specimen size and crack length has been taken as in Section
6.2.1. A structured mesh with full alignment case and an element
Fig. 20. Influence of aspect ratio in the edge crack.
length he = 0.02 is also considered for the X-FEM analysis. Figs.
23 and 24 ensure that the results of X-FEM are similar to the re-
sults of [26] for all the applied loadings. Fig. 23 and Table 8 show
that the normalized values of KIV increases with the increasing val-
ues of a/w and the applied electrical loadings. It is amply clear from
Fig. 24 that KI is independent of the applied electric loadings. It
may also conclude that in edge crack problem KIV has more affect
of finite specimen and applied electrical loadings than the center
crack.

Table 7
Domain independence study for edge crack.

he/a Radius (rd) Normalized KI % Error Normalized KIV % Error

in KI in KIV
0.25 0.5 0.9871 1.29 1.4293 42.93
0.6 0.9869 1.31 1.4265 42.65
0.7 0.9872 1.28 1.4275 42.75
0.8 0.9875 1.25 1.4288 42.88
0.9 0.9875 1.25 1.4288 42.88
0.1 0.4 0.9964 0.36 1.4547 45.47
0.5 0.9965 0.35 1.4550 45.50
0.6 0.9965 0.35 1.4549 45.49
0.7 0.9964 0.36 1.4548 45.48
Fig. 21. Comparison between structured mesh and mesh with local refinement for 0.8 0.9964 0.36 1.4548 45.48
the edge crack problem.
 R.R. Bhargava, K. Sharma / Computational Materials Science 50 (2011) 1834–1845 1845

Table 8 7. Conclusions
Finite specimen effects for edge crack.

he/a w/a Normalized KI % Error Normalized KIV % Error The conclusions are
in KI in KIV
0.25 rd = 0.8 5 0.9875 1.25 1.4288 42.88  It is noted that the results obtained with mesh having local
10 0.9982 0.18 1.2700 27.00 refinement can predict the behavior extremely well and having
20 1.0014 0.14 1.2155 21.55
a good computational advantage.
30 1.0009 0.09 1.2025 20.25
40 1.0003 0.03 1.1972 19.72  A finite specimen effect is observed in both the center and edge
50 0.9998 0.02 1.1944 19.44 crack problems. Even the latter case has a stronger effect than
0.1 rd = 0.8 5 0.9964 0.36 1.4548 45.48 the center crack. From our computation we observe that
10 1.0047 0.47 1.2897 28.97 (i) For center cracked specimen a ratio w/a P 10 gives a good
20 1.0072 0.72 1.2333 23.33 accuracy.
30 1.0066 0.66 1.2202 22.02 (ii) For edge cracked specimen a ratio w/a P 20 gives a good
0.05 rd = 0.8 5 0.9980 0.20 1.4596 45.96 accuracy.
10 1.0059 0.59 1.2936 29.36  The percentage error in KIV is higher than the percentage error
in KI in case of edge crack problem. This behavior is also vali-
dated with Kuna [13]. The results presented for an edge crack
using X-FEM are also similar to the Wang and Mai [26].
 Finally, our analysis confirms the statement of Kuna [13] that
the fourfold enrichment functions used in the X-FEM are effi-
cient and accurate to provide the IFs and energy release rate
for piezoelectric materials.

Acknowledgements

The authors are grateful to Prof. R.D. Bhargava (Senior Professor

and Head, Retd., Indian Institute of Technology Bombay, Mumbai,
India) for the encouragement throughout the course of this work.

References

[1] H. Allik, T.J.R. Hughes, International Journal for Numerical Methods in

Engineering 2 (1970) 151–157.
[2] S. Kumar, R.N. Singh, Acta Materialia 44 (1996) 173–200.
[3] S. Kumar, R.N. Singh, Acta Materialia 45 (1997) 849–868.
[4] M. Abendroth, U. Groh, M. Kuna, A. Ricoeur, International Journal of Fracture
114 (2002) 359–378.
[5] K. Wippler, A. Ricoeur, M. Kuna, Engineering Fracture Mechanics 71 (2004)
2567–2587.
[6] A. Ricoeur, M. Enderlein, M. Kuna, Archive of Applied Mechanics 74 (2005)
Fig. 23. Variation of normalized value of KIV w.r.t a/w for different electric loading 536–549.
conditions in case of edge crack problem. [7] M. Kuna, Archives in Applied Mechanics 76 (2006) 725–745.
[8] M. Kuna, Engineering Fracture Mechanics 77 (2010) 309–326.
[9] N. Moës, J. Dolbow, T. Belytschko, International Journal for Numerical Methods
in Engineering 46 (1999) 131–150.
[10] T. Belytschko, T. Black, International Journal for Numerical Methods in
Engineering 45 (1999) 601–620.
[11] A. Yazid, N. Abdelkader, H. Abdelmadjid, Applied Mathematical Modelling 33
(2009) 4269–4282.
[12] T. Belytschko, R. Gracie, G. Ventura, Modelling and Simulation in Materials
Science and Engineering 17 (2009) 043001.
[13] E. Bechet, M. Scherzer, M. Kuna, International Journal for Numerical Methods
in Engineering 77 (2009) 1535–1565.
[14] H. Sosa, International Journal of Solids and Structures 29 (1992) 2613–
2622.
[15] Y.E. Pak, International Journal of Fracture 54 (1992) 79–100.
[16] B.N. Rao, M. Kuna, International Journal of Solids and Structures 45 (2008)
5237–5257.
[17] S. Osher, J.A. Sethian, Journal of Computational Physics 79 (1988) 12–49.
[18] M. Stolarska, D.L. Chopp, N. Moës, T. Belytschko, International Journal for
Numerical Methods in Engineering 51 (2001) 943–960.
[19] V. Piefort, A. Preumont, Finite Element Modeling of Piezoelectric Structures,
Active Structures Laboratory, ULB-CP 165/42.
[20] Y. Mukamai, Stress Intensity Factors Handbook, Pergamon Press, 1987.
[21] H. Ewalds, R. Wanhill, Fracture Mechanics, Edward Arnold, 1984.
[22] M. Enderlein, A. Ricoeur, M. Kuna, International Journal of Fracture 134 (2005)
191–208.
[23] G.P. Cherepanov, Journal of Applied Mathematics and Mechanics 41 (1977)
397–410.
[24] T.J.C. Liu, Theoretical and Applied Fracture Mechanics 51 (2009)
102–110.
[25] M.J. McNary, Implementation of the XFEM in the ABAQUS Software Package,
M.Sc. Thesis, Georgia Institute of Technology, 2009.
Fig. 24. Variation of normalized value of KI w.r.t a/w for different electric loading [26] B.L. Wang, Y.-W. Mai, International Journal of Solids and Structures 39 (2002)
conditions in case of edge crack problem. 4501–4524.