A novel variable power singular element in G space with strain smoothing

for bi-material fracture analyses

L. Chen
a,b,n
, G.R. Liu
c
, K. Zeng
b
, J. Zhang
d
a
Institute of High Performance Computing, Singapore 138632, Singapore
b
Department of Mechanical Engineering, National University of Singapore, 9 Engineering Drive 1, Singapore 117576, Singapore
c
University of Cincinnati, Cincinnati, OH 45221-0070, USA
d
Department of Civil and Environmental Engineering, National University of Singapore, 1 Engineering Drive 2, Singapore 117576, Singapore
a r t i c l e i n f o
Article history:
Received 14 January 2011
Accepted 14 June 2011
Keywords:
Singular element
Numerical method
G space
Strain smoothing
Variable singularity
M-integral
Stress intensity factor
Bi-material
a b s t r a c t
This paper aims to formulate a triangular ve-node (T5) singular crack-tip element in G space with
strain smoothing to simulate an r
l1
(0olo1) stress singularity for bi-material fracture analyses. In
the present formulation, a direct point interpolation with a proper fractional order of extra basis
functions is specially employed to construct variable power type singular shape functions that are in a
G
1
space. Within strain smoothing, the singular terms of functions as well as mapping procedures are
no longer necessary to compute the stiffness matrix. Furthermore, thanks to the point interpolation, the
proposed singular element eliminates the need to shift the position of the side nodes adjacent to the
crack-tip, and is thus quite straightforward and easily implemented in existing codes. The effectiveness
of the present singular element is demonstrated via numerical examples of a wide range of material
combinations and boundary conditions.
& 2011 Elsevier Ltd. All rights reserved.
1. Introduction
With the increasing demands on multi-functional needs (wear,
corrosion, thermal resistance and toughness) in mechanical, aero-
space and biomedical applications, the development of layered
material systems has come to the forefront. In the layered
systems, the overall mechanical behaviors or responses hinge
mainly on the cracks occurring near the interfaces between two
dissimilar materials [1]. Therefore, the development of a robust
and effective simulation tool to characterize the singular behavior
of these cracks is crucial to a better understanding of the
inuence of the mismatch in properties and their effects on crack
growth.
With regard to the cracks associated with the bi-material
interfaces, the problem of an interface crack and that of a crack
normal (and impinging) to the interface are of technical impor-
tance. It is well known that the order of singularity for the stress
eld is 1=

r
_
(where r is the radial distance from the crack-tip) in
typical linear elastic fracture mechanics. However, for a crack
lying along a bi-material interface, the stress singularity in the
vicinity of the crack-tip is oscillatory in nature, along with the
presence of an inverse

r
_
singularity, i.e., s
ij
r
1/2ie
[24],
where the oscillatory term r
ie
depends on the mismatch in elastic
properties of two bonded materials. Moreover, for the case of a
crack perpendicularly terminating at the interface, Zak and
Williams [5], and Cook and Erdogan [6] showed that the near-
tip stress eld is in the form of s
ij
r
l1
(0olo1), where the
exponent l is the lowest root of the characteristic equation
depending on the Dundurs parameters that are related to the
mechanical properties of two materials [7,8]. To model such a
behavior accurately, efforts have been made to embed arbitrary
order singularity (r
l1
, 0olo1) in the vicinity of the crack-tip
[915]. Abdi and Valentin [16] generalized the idea of quarter-
point elements [17] for modeling an r
l1
stress singularity, and
the optimal position of side nodes adjacent to the crack-tip for
quadratic and cubic isoparametric elements was obtained using a
least-square method. Wu [18] extended this idea for collapsed
triangular isoparametric elements and showed that they give
better results. Lim and Kim [19] proposed a simpler method for
calculating the optimal position of the side nodes adjacent to the
crack-tip. A major disadvantage of these quadratic crack-tip
elements is that the entire domain has to be, in principle,
modeled by quadratic elements of the same type to ensure
compatibility. Otherwise, the transition elements are needed to
bridge the crack-tip elements to the standard elements [20].
Recently, Belytschko and Moes developed so-called extended
nite element method (XFEM) to model arbitrary discontinuities
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/enganabound
Engineering Analysis with Boundary Elements
0955-7997/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.enganabound.2011.06.007
n
Corresponding author at: Department of Mechanical Engineering, National
University of Singapore, 9 Engineering Drive 1, Singapore 117576, Singapore.
Tel.: 65 65164797.
E-mail address: g0700888@nus.edu.sg (L. Chen).
Engineering Analysis with Boundary Elements 35 (2011) 13031317
in meshes [21,22]. This extension exploits the partition of unity
property of nite elements, which allows local enrichment func-
tions to be easily incorporated into a nite element approxima-
tion while preserving the classical displacement variational
setting [23,24]. Nevertheless, as the complete asymptotic displa-
cement elds are too complicated for the cracks related to the
interfaces, special treatments [25,26] should be used to derive
enrichment functions characterizing the local behavior with
variable order singularity.
On another front of computational mechanics, a strain smooth-
ing technique [27,28] was introduced by Chen et al. [27] for spatial
stabilization of nodally integrated meshfree methods, and later
extended by Yoo and Moran to the natural element method (NEM)
[28]. More recently, Liu [29] has generalized this gradient smooth-
ing technique in order to weaken the consistence requirement for
the eld functions, allowing the use of certain types of discontin-
uous displacement functions. Based on this generalization, a G
space theory and a generalized smoothed Galerkin (GS-Galerkin)
weak form have been developed [30], leading to the so-called
weakened weak (W
2
) foundations of a family of numerical meth-
ods. Among them, a cell-based smoothed nite element method
(SFEM or CS-FEM) [31] was rst formulated by introducing the
gradient smoothing technique to (compatible) FEM settings. In such
SFEM, the elements are divided into smoothing domains (SDs) over
which the strain is smoothed. A node-based smoothed nite
element (NS-FEM) [32] was also formulated using smoothing
domains associated with nodes in FEM settings. However, similar
to other nodal integrated methods [33], the NS-FEM suffers from
the temporal instability due to its overly soft feature rooted at the
use of a relatively small number of SDs in relation to the nodes. Liu
et al. [34] formulated then an edge-based smoothed nite element
method (ES-FEM) to eliminate this temporal instability. The ES-FEM
works very well with triangular linear elements that can be
generated automatically for problems even with complicated geo-
metry, and exhibits super convergence properties, in addition to its
ultra accuracy. The ES-FEM was also found computationally very
efcient, and known as the star performer among all the linear
numerical models [35].
Most importantly, for the numerical models based on the G
space theory, domain integration in the W
2
formulation is trans-
formed into boundary integration. Thus, the stiffness matrix calcu-
lation requires only evaluating the shape functions values (and not
the derivatives) on the boundaries of the strain smoothing domains
(SDs). Making use of this signicant property, Liu et al. [36,37] has
further developed a singular crack-tip element for fracture analyses
in a homogeneous media by enriching displacement elds with a
desired order of

r
_
using the simple and nonmapping point
interpolation method (PIM) [38]. As a result, an r
1/2
stress
singularity can be properly produced in the vicinity of the crack-
tip, without using any mapping procedure. However, the displace-
ment elds within this singular element were just easily assumed
as linear in the circumferential direction. Consequently, it did not
adequately account for variation along the angular direction near
the crack-tip. In addition, this singular element was only limited to
the simple r
1/2
stress singularity, and did not consider a variable
order stress singularity.
In this paper, a novel triangular ve-node (T5) singular crack-tip
element is formulated in the framework of G space theory with
strain smoothing to simulate a variable order stress singularity. In
the radial direction, the displacement elds with designed power
singularity are constructed via a direct point interpolation with a
proper fractional order of extra basis functions. Also, a quadratic
polynomial function is utilized to reect the crack-tip variation of
displacements in the circumferential direction. In addition, a basic
mesh of linear T3 elements with one layer of T5 singular elements is
used to produce the stress singularity behavior in the vicinity of the
crack-tip. Because of the use of point interpolation, the proposed
singular element eliminates the need to shift the position of the side
nodes adjacent to the crack-tip, and is thus quite general and
straightforward. Numerical examples with existing simplied solu-
tions showthat the proposed singular elements yield accurate results
for a wide range of material combinations and boundary conditions.
2. Variable power singularity with bi-material interfaces
Consider two dissimilar homogeneous materials perfectly
bonded along an interface shown in Fig. 1, two classic categories
of cracks are of technical importance in practice: (1) an interface
crack as illustrated in Fig. 1(a); (2) a crack perpendicularly
terminating at the interface as shown in Fig. 1(b). The regions
on both sides of the crack (occupied by material 1 with the shear
modulus and Poissons ratio of m
1
and v
1
) are denoted by O
1
,
while the region occupied by material 2 with corresponding
properties of m
2
and v
2
is denoted by O
2
. A local coordinate
system (r,y) originating at the crack-tip o is introduced.
In linear elastic interfacial fracture mechanics [4], the complex
stress intensity factor K=K
I
iK
II
is adopted. The in-plane traction
vector at a distance r ahead of the crack takes the form [4]
(s
yy
it
xy
)
y = 0
=
Kr
ie

2pr
_ (1)
Fig. 1. (a) A crack along the bi-material interface; and (b) a crack perpendicularly terminating at the bi-material interface.
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1304
where i =

1
_
, and e is the bi-material constant which is a
function of b, one of the two Dundurs bi-material parameters
(denoted by a and b) that are both given in the following
equations for later use[7]:
a =
m
1
(k
2
1)m
2
(k
1
1)
m
1
(k
2
1)m
2
(k
1
1)
, b =
m
1
(k
2
1)m
2
(k
1
1)
m
1
(k
2
1)m
2
(k
1
1)
(2)
e =
1
2p
log
1b
1b
_ _
(3)
k
p
=
3vp
1vp
(plane stress)
34v
p
(plane strain)
(p =1,2)
_
_
_
(4)
where m
p
, v
p
and k
p
are the shear modulus, Poissons ratio and the
Kolosov constant, respectively, of material p (p=1,2).
For the case of a crack orthogonally terminating at the inter-
face, Cook and Erdogan [6] showed in a theoretical study that for
the stress singularity in the form of s
ij
r
l1
, the exponent l is
the lowest root of the following characteristic equation:
2l
2
(ab)(b1)ab
2
(1b
2
)cos(lp) =0 (5)
where a and b are the Dundurs bi-material parameters that are
dened in Eq. (2). It is noted that 0olo0.5 when cracked material
1 is stiffer than material 2, and 0.5olo1 when the opposite case
occurs. Particularly, when the crack-tip is embedded in a homo-
geneous media, l equals 0.5.
3. Brief on G space theory
To effectively construct the displacement elds in the vicinity
of the crack-tip, we shall allow the use of certain types of
discontinuous functions. A G space of a set of discrete functions
was established in the recent work by Liu [30] for this purpose.
This space was originally dened to include some continuous/
discontinuous functions allowing the use of more types of
methods/techniques to produce shape functions for numerical
models. Here, we briey introduce the denition and its differ-
ences with Hilbert or H spaces in the following section. More
details on the properties of a G space and detailed proofs can be
found in Ref. [30].
Consider a d-dimensional problem domain O discretized into
N
e
non-overlapping subdomains with a set of N
n
nodes. Recall
that the denition of H spaces, taking H
1
as an example, can be
expressed as:
H
1
(O) ={v9vAL
2
(O),@v=@x
i
AL
2
(O), (i =1,. . .,d)] (6)
In the recent work by Liu [30], a subspace H
1
h
in the H
1
space
created using interpolation techniques that ensure compatibility,
is also dened as
H
1
h
(O) ={vAH
1
(O)9v(x) =U
h
(x)d, dAR
Nn
] (7)
where d =
d
1
d
2
d
Nn
_ _
T
is the vector of nodal function
values, and U
h
is the vector of all the (compatible) nodal shape
functions that are linearly independent and can be written as
U
h
(x) =[f
h
1
(x) f
h
2
(x) f
h
Nn
(x)] (8)
The domain is then, in a basically independent way, divided into a
set of N
s
non-overlapping smoothing domains bounded by
G
s
k
(k =1,. . .,N
s
) that do not share any nite portion of the line-
segments on which the function is not square integrable. The G
spaces, taking G
1
h
as an example as well, can be then dened as
G
1
h
(O) =
v9v(x) =

Nn
n = 1
f
h
n
(x)d
n
=U
h
(x)d, dAR
Nn
vAL
2
(O) oo

N
S
k = 1
_
G
s
k
v(B)n
i
dB
_ _
2
40, \va0, (i =1,. . .,d)
_

_
_

_
(9)
where f
h
n
(x) is the nodal shape function for node n, and these N
n
shape functions form the basis of the G
1
h
space. It is observed that
the G
1
h
space is of nite dimension (denoted by subscript h) and
with a set of functions that are square integrable in O formed by
point interpolation using a basis. These nodal shape functions are
created by selected nodes based on elements/cells using the
standard FEM or meshfree procedures, and hence they are contin-
uous at all these nodes and on these boundaries G
s
k
(k =1,:::,N
s
). The
continuity on all these boundaries G
s
k
allows a unique evaluation of
the generalized smoothed gradient of the functions over the
smoothing domains, so that the variation of the functions over O
can be captured in a local averaged fashion. How the basis is
created is not restricted, as long as these nodal shape functions are
linearly independent over O and hence are capable of forming
a basis.
By comparing Eq. (9) and Eq. (7), the major differences
between a G
1
h
space and the corresponding Hilbert space or H
1
h
space are as follows:
(1) The H
1
h
space requires the functions and the rst derivatives
of the functions to be all square integrable, but in the G
1
h
space, only the functions themselves are required to be square
integrable;
(2) The requirement on function in a G
1
h
space is weakened upon
the already weakened requirement for functions in an H
1
h
space, and hence the G
1
h
space can be viewed as a space of
functions with weakened weak (W
2
) continuity;
(3) The rst derivatives of functions in an H
1
h
space need to be
bounded from above
_
O
(@v=@x
i
)
2
dOoo. This is because the
energy in the weak formulation needs to be bounded. We do
not worry about the possibility of
_
O
(@v=@x
i
)
2
dO=0, because
it will never happen as long as the function is not zero
everywhere in terms of the well-known PoincareFriedrichs
inequality. On the other hand, for functions in a G
1
h
space,
however, the positivity is needed to bound the non-constant
functions from below

N
S
k = 1
_
G
s
k
v(B)n
i
dB
_ _
2
40, ensuring the
stability of our W
2
formulations. The energy in the weakened
weak formulation is automatically bounded from above,
because the functions themselves are square integrable,
ensuring the convergence of the discrete model to be created;
(4) A member in an H
1
h
space is a member in the corresponding
G
1
h
space (using the same mesh with proper smoothing
domains), we shall have H
1
h
G
1
h
.
4. The formulation of singular elements
4.1. Variable power singularity modeling
The most fundamental issue for bi-material fracture analyses
using the FEM or any other numerical methods is to simulate
properly the variable order stress singularity in the vicinity of the
crack-tip while maintaining a certain order of consistency. When
the polynomial basis shape functions are used in the conventional
nite elements, consistency can be easily achieved, but the strain
singularity cannot be produced, for which we have to resort to the
so-called singular elements [9,13,1619]. Currently, the most
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1305
widely used singular elements in the standard FEM setting are
collapsed eight-node (T8) or six-node (T6) quadratic isoparametric
elements as shown in Fig. 2. These collapsed singular elements shift
the mid-side node adjacent to the crack-tip to an optimal position
that depends on the properties of two materials. The singularity is
usually achieved nicely by the well-known isoparametric mapping
procedure. However, to ensure all of these quadratic singular
elements are compatible with other standard elements, the entire
domain has to be, in principle, modeled by quadratic elements of
the same type. Otherwise, transition elements [20] are needed to
bridge between the crack-tip elements and the standard elements.
On the other hand, the numerical models in G space rely on
three-node linear T3 meshes that can be generated automatically
for problems with complicated geometry. Based on this simple
mesh setting, we specially formulate a ve-node singular crack-tip
(T5) element by a simple PIM with extra basis functions of a proper
fractional order. In addition, one layer of these T5 singular elements
is incorporated in a basic mesh of T3 elements to produce the
desired stress singularity in the vicinity of the crack-tip as illu-
strated in Fig. 3(a). In our singular elements, we add in a node to
each edge of T3 elements that are connected to the crack-tip (called
the crack-tip edge). Therefore, the crack-tip element has a total of
ve-nodes, and is termed as the T5 singular element (see Fig. 3(b)).
Our previous study [36] has demonstrated that the location of the
added node in this T5 singular element can be arbitrary since the
position of the added node does not signicantly affect on the
result. Accordingly, the added node is selected herein at the one
quarter length of the edge from the crack-tip for simplicity (see
Fig. 3(b)). The creation of a singular stress eld consists of two
steps: (1) displacement interpolation in a radial line from crack-tip;
(2) displacement eld creation within the T5 singular element.
The displacement eld at any point of interest x on a radial line
from the crack-tip as illustrated in Fig. 3(c) is chosen to have the form
u
h
(x) =abr cr
l
(10)
at a xed value of y. Where r is the radial coordinate originated at
the crack-tip for the interpolation point x, the crack-tip displace-
ments are represented by a, the b term indicates the regular stress
elds and the c term represents the variable power order stress
singularity. It is clear that the displacement elds are (complete)
linear in r and enriched with a special basis r
l
. By differentiation
of the assumed displacements, the strain distributions in the
r-direction can be expressed as
e
r
=
@u
h
(x)
@r
=bclr
l1
(11)
and the desired strain (hence stress) singularity eld of an order
r
l1
along r-direction is therefore produced.
l
op
l
l
op
l crack tip crack tip
Fig. 2. Schematic of collapsed eight-node and six-node quadratic singular ele-
ments. The optimal position of side nodes adjacent to the crack-tip, l
op
, depends on
the material properties.
crack tip
x
y
r
P
2
P
3
x
crack tip
l
/
4
l

Fig. 3. Crack-tip conguration using a basic mesh of triangular elements: (a) one layer of ve-node crack-tip (T5) singular elements with one additional node added on
each edge leading to the crack-tip (called the crack-tip edge); (b) the location of the added node is the one quarter length of the edge from the crack-tip; (c) displacement
interpolation within a T5 crack-tip element (12345): the displacement varies with r via the enriched form of Eq. (10) in the radial direction and the variation in the
tangential direction is quadratic.
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1306
To perform the point interpolation in such a T5 singular
element, in addition to the enriched form of Eq. (10) in the radial
direction, a quadratic polynomial function is utilized to reect the
crack-tip variation of singularity in the tangential direction. The
objective of this operation is to reduce the circumferential effects
of the linear variation assumption in the previous work [36,37]. At
the same time, a linear dependence on y is chosen for the regular
stress elds. Based on the above analysis, the complete displace-
ment functions within a T5 singular element can be approximated
using the following enriched PIM:
u
h
x
=a
1
r(b
1
b
2
y) r
l
(c
1
c
2
yc
3
y
2
) (12)
u
h
y
=a
2
r(b
1
b
2
y) r
l
(c
4
c
5
yc
6
y
2
) (13)
where the generalized coefcients a
i
, b
i
, c
i
can be determined by
enforcing equations to exactly pass through nodal displacement
values of the ve nodes this T5 element hosts. The result can be
written in the following matrix form:
u
e
=P
0
G (14)
where
G =[a
1
, a
2
, b
1
, b
2
, c
1
, c
2
, c
6
]
T
(15)
u
e
=[u
x
1
, u
x
2
, u
x
5
, u
y
1
, u
y
2
, u
y
5
]
T
(16)
P
0
=
1 1 r
1
r
1
y
1
r
l
1
r
l
1
y r
l
1
y
2
r
l
1
r
l
1
y r
l
1
y
2

1 1 r
5
r
5
y
5
r
l
5
r
l
5
y r
l
5
y
2
r
l
5
r
l
5
y r
l
5
y
2
1 1 r
1
r
1
y
1
r
l
1
r
l
1
y r
l
1
y
2
r
l
1
r
l
1
y r
l
1
y
2

1 1 r
5
r
5
y
5
r
l
5
r
l
5
y r
l
5
y
2
r
l
5
r
l
5
y r
l
5
y
2
_

_
_

_
(17)
where P
0
is a constant matrix that is only constructed by
coordinates of the nodal points. From Eq. (14), we have
G =P
1
0
u
e
(18)
Substituting Eqs. (12) and (13) into Eq. (14), we have
u
h
(x) =U(x)u
e
(19)
in which the shape function U(x) is dened by
U(x) =p
T
(x)P
1
0
(20)
where p(x) is built by combining Eq. (12) into Eq. (13), and is
given by
p
T
(x) = 1 1 r ry r
l
r
l
y r
l
y
2
r
l
r
l
y r
l
y
2
_ _
(21)
Remark: (consistency)
As illustrated in Fig. 3(a), displacement compatibility is guar-
anteed on the crack-tip edges between T5 singular elements,
while there is incompatibility on the interfaces of singular (T5)-
standard (T3) elements. Hence, the PIM shape functions (thus the
displacement elds) in such an interpolation are not in an H
1
h
space. On the other hand, the displacement elds are at least
(complete) linearly consistent and square integrable: the displa-
cements in T5 singular elements constructed using the above
basis are on top of the completely linear eld, and the displace-
ment elds in T3 elements are linearly independent. Therefore,
the displacement elds throughout the entire problem domain
are in a G
1
h
space, ensuring the stability and the convergence of
the solution in our T5 singular elements.
4.2. Edge-based strain smoothing
Our previous works [35] have shown by several in-depth
studies that, among the family of numerical methods in G space
with strain smoothing, the ES-FEM is found particularly superior
in convergence, accuracy, computational efciency and stability.
Therefore, it is chosen as a platform for our newly designed T5
singular elements. Consider the domain O discretized into N
e
non-
overlapping and non-gap triangular elements and N
n
nodes, the
local smoothing domains in the ES-FEM are constructed with
respect to the edges of triangular elements such that O= C
Ns
k = 1
O
s
k
and O
s
i
O
s
j
=0, \i aj, in which N
s
is the number of smoothing
domains. As shown in Fig. 4, the smoothing domain correspond-
ing to the inner edge k, O
s
k
, is formed by connecting two end
points of edge k and two centroids of the adjacent triangular
elements. In addition, the smoothing domain for the boundary
edge m, O
s
m
, is just one-third region of triangular element which
contains the edge m.
In the ES-FEM, however, the strain is written as the divergence
of a spatial average of the standard (compatible) strain given by
e
h
=V
s
u
h
(22)
in which V
s
u
h
is the symmetric gradient of the displacement
eld: V
s
=1=2( VV
T
). For example, the strain smoothed over
the local smoothing domains, O
s
k
, associated with the edge k can
be expressed as
e
k
=
1
A
s
k
_
O
s
k
V
s
u
h
(x)dO (23)
where A
s
k
=
_
O
s
k
dO is the area of the smoothing domain O
s
k
. Using
GaussLegendre integration along the segments of boundary G
s
k
,
we have
e
k
=
1
A
s
k
_
O
s
k
V
s
u
h
(x)dO=
1
A
s
k
_
G
s
k
L
n
u
h
(x)dG (24)
in which L
n
is the outward unit normal matrix which can be
expressed as
L
n
=
n
x
0
0 n
y
n
y
n
x
_

_
_

_ (25)
In the above setting of smoothing domains, it can be under-
stood that the conventional smoothing domains associated with
the crack-tip edges cannot adequately capture the singularity of
Fig. 4. Construction of the edge-based strain smoothing domains. Note that shaded
area indicates one layer of smoothing domains associated with crack-tip edges.
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1307
the stresses as shown in Fig. 4. In order to ensure accuracy, a
higher integration point density should be used close to the crack-
tip. Therefore, we divide such smoothing domains into several
(n
sc
) polygonal sub-cells (SC) as illustrated in Fig. 5, and then
perform the numerical integration cell by cell.
(i) n
sc
=1: no sub-division is performed in such crack-tip
smoothing domains as shown in Fig. 5(a).
(ii) n
sc
=2: each of crack-tip smoothing domains is partitioned
into two sub-cells. The rules involved in this division are: the
sub-cell far away the crack-tip is constructed by connecting
(1) the centroid of one adjacent singular elements of the
edge; (2) the no-tip-edge-point; (3) the centroid of another
adjacent singular elements; (4) the 1/4-centroid-point of the
second adjacent singular elements; (5) the 1/4-edge-point;
(6) the 1/4-centroid-point of the rst adjacent singular
elements, and back to (1) the centroid of the rst adjacent
singular elements. For example, the SC
(1)
as shown in
Fig. 5(b) is created by connecting sequentially #B, #3, #A,
#C, #2, #D and #B.
For the sub-cell connected directly to the crack-tip, its region
is the left part of the region of the original crack-tip
smoothing domain subtracting the region of the correspond-
ing SC far away the crack-tip dened above. For instance, the
SC
(2)
is constructed by connecting sequentially #1, #D, #2 #C
and #1, as shown in Fig. 5(b).
(iii) n
sc
=3: each of crack-tip smoothing domains is partitioned
into three sub-cells. Compared to the sub-division scheme
with n
sc
=2, the sub-cell far away the crack-tip SC
(1)
is kept
the same. The sub-cell connected to the crack-tip, SC
(2)
is
then split into two smaller sub-cells (SC
(2)
and SC
(3)
) follow-
ing the similar rule used in the n
sc
=2 scheme. As shown in
Fig. 5(c), the SC
(2)
is formed by connecting sequentially #D,
#2, #C, #E, #G, #F, and #D, and the SC
(3)
is constructed by
connecting sequentially #1, #F, #G, #E, and #1.
Another important issue affecting integration point density is
the number of Gauss points on a boundary segment of smoothing
cells. Remind that in the formulations of ES-FEM, to compute the
smoothed strain gradient matrix, only the shape function values
at the Gauss points along the boundary segments are needed. This
is also performed similarly to the smoothing domains or
smoothing sub-cells associated with the crack-tip edges. In the
standard ES-FEM, the shape function used is always linear
compatible along any boundary segments. Therefore, only one
Gauss point is needed on each boundary segment. However, in
the crack-tip smoothing sub-cells, more Gauss points should be
used for each boundary segment due to the singularity behavior
in the vicinity of the crack-tip. Numerical experiments demon-
strate that ve Gauss points on a boundary segment of smoothing
cells (n
gau
=5) are sufcient (details can be found in Section 7).
4.3. Weak form and discrete equations
Substituting Eq. (10) into Eq. (24), the smoothed strain can be
written in the following matrix form of nodal displacements:
e
k
=

i An
s
k
B
i
(x
k
)d
i
(26)
where d
i
is the vector of nodal displacements and n
s
k
is the set of
nodes associated the smoothing domain O
s
k
. B
i
(x
k
) is termed as
the smoothed strain gradient matrix that is calculated by
B
i
(x
k
) =
b
ix
(x
k
) 0
0 b
iy
(x
k
)
b
iy
(x
k
) b
ix
(x
k
)
_

_
_

_
(27)
where b
ih
(x
k
), h=x, y, is computed by
b
ih
(x
k
) =
1
A
s
k
_
G
s
k
n
h
(x)f
i
(x)dG (28)
where f
i
(x) is a shape function corresponding to node i in the
node set n
s
k
.
Using Gauss integration along the segments of boundary G
s
k
,
we have
b
ih
=
1
A
s
k

Nseg
m = 1

Ngau
n = 1
w
m,n
f
i
(x
m,n
)n
h
(x
m,n
)
_ _
(h =x,y) (29)
where N
seg
is the number of segments of the boundary G
s
k
, N
gau
is
the number of Gauss points used in each segment, w
m,n
is the
corresponding weight of Gauss points, n
h
is the outward unit
normal corresponding to each segment on the smoothing domain
boundary and x
m,n
is the nth Gaussian point on the mth segment
of the boundary G
s
k
.
SC
(1)
SC
(3)
SC
(2)
SC
(1)
SC
(1)
SC
(2)
7
6
1
4
B
A
2
A
6
1
F
4 D 2
C
B
5 3 5 3
7
6
1
2 4
D
B
five Gauss points in a
boundary
5 3
C
A
7
E
G
Fig. 5. Division of a smoothing domain associated with the crack-tip edge into sub-smoothing cells: (a) n
sc
=1; (b) n
sc
=2 and; (c) n
sc
=3.
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1308
The numerical models in G space are variationally consistent
as proven in Ref. [30], so the assumed displacement u
h
and the
smoothed strains e satisfy the smoothed Galerkin weak form
_
O
d(e(u
h
))
T
D(e(u
h
))dO
_
O
(du
h
)
T
bdO
_
Gt
(du
h
)
T
t
G
dG=0 (30)
Substituting the approximated displacements from Eq. (10)
and the smoothed strains in Eq. (26) into the smoothed Galerkin
weak form yields the following system of equations:
Kd =f (31)
where f is the nodal force vector that can be obtained by
f =
_
O
U
T
(x)bdO
_
Gt
U
T
(x)t
G
dG (32)
The stiffness matrix Kis then assembled by
K
ij
=

Ns
k = 1
K
s
ij,k
=

Ns
k = 1
_
O
s
k
B
T
i
DB
j
dO=

Ns
k = 1
B
T
i
DB
j
A
s
k
(33)
4.4. Advantages over collapsed quadratic singular elements
The T5 singular element in G space that we propose in this
paper introduces a generalized strain smoothing technique, cor-
responding to the smoothed bilinear form [29] with smoothing
domains (edge-based), and thus possesses the following distinc-
tive advantages over collapsed quadratic singular elements:
(1) The use of such mesh setting: one layer of T5 singular
elements together with linear T3 elements away from singu-
lar zone (see Fig. 3(a)), eliminates the requirement of transi-
tion elements and allows modeling arbitrary geometries;
(2) The collapsed singular elements shift the mid-side node
adjacent to the crack-tip to an optimal position for the
purpose of variable order singularity, which is, however,
directly embodied in the approximation of our T5 singular
elements. Consequently, the shift of the mid-side node adja-
cent to the crack-tip is not necessary;
(3) The smoothed bilinear form is used to formulate a smoothed
Galerkin weakform, which weakens the consistence require-
ment for the eld functions [29,30];
(4) With the strain smoothing, the derivatives of the shape
functions are no longer necessary. Hence, when singularity
is considered in the approximation, the 1=

r
_
term does not
appear, so that the need of integrating the singular functions
present in the stiffness matrix of collapsed quadratic singular
elements can be avoided;
(5) The absence of isoparametric mapping is enabled by the
strain smoothing operation, decreasing its sensibility to mesh
distortion.
Based on the above analysis, the proposed T5 singular element
is quite straightforward and there is a good deal of generality in
the numerical formulation. As a consequence, it can be easily
implemented in existing codes.
5. M-integral for stress intensity factors
The general form of J-integral, which is identical to energy
release rate G, for a 2D cracked body can be written as [39]
J =G =
_
G
1
2
s
ik
e
ik
d
xj
s
ij
u
i,x
_ _
n
j
dG, i =x or y ; j =x or y (34)
where s
ij
and u
i
are the Cartesian components of the stress tensor
and the displacement vector, respectively, n
j
are the Cartesian
components of outward unit vector normal to G. The path of
integration, G, is dened as a counterclockwise contour encircling
and shrinking onto the tip O as shown in Fig. 6.
In the M-integral method [15,22,25], two states of a cracked
body are used to evaluate the stress intensity factors. State 1,
(s
(1)
ij
,e
(1)
ij
,u
(1)
i
), corresponds to the present state and state 2,
(s
(2)
ij
,e
(2)
ij
,u
(2)
i
), is an auxiliary state. On summing the J-integral of
two states, we can obtain the M-integral
M=
_
G
(s
(1)
ik
e
(2)
ik
d
xj
s
(1)
ij
u
(2)
i,x
s
(2)
ij
u
(1)
i,x
)n
j
dG, k =x or y (35)
5.1. Interface crack
The J-integral remains globally path independent for an inter-
face crack between bi-material plates when there exists no
material inhomogeneity in the direction parallel to the crack
[40]. Its value can be related to the stress intensity factor
amplitude through the following relation [4]:
J =
1
E
n
9K9
2
cosh
2
(pe)
, 9K9
2
=KK=K
2
I
K
2
II
(36)
where K=K
I
iK
II
is the complex stress intensity factor, and
2
E
n
=
1
E
1

1
E
2
, E
p
=
E
p
(plane stress)
Ep
1v
2
p
(plane stress)
_
_
_
(37)
After some manipulations, the M-integral is related to the SIFs
through the relation
M=
2
E
n
K
(1)
I
K
(2)
I
K
(1)
II
K
(2)
II
cosh
2
(pe)
(38)
Making the judicious choice of state 2 (auxiliary) as the pure
Mode I asymptotic elds, i.e., setting K
(2)
I
=1, K
(2)
I
=0 and evalu-
ating M=M
I
, we can compute K
I
and we proceed in an analogous
manner to evaluate K
II
K
I
=
E
n
cosh
2
(pe)
2
M
I
K
II
=
E
n
cosh
2
(pe)
2
M
II
(39)
In which we use the asymptotic elds in the vicinity of
interfacial crack-tip for the auxiliary elds that are provided in
Ref. [41]. For completeness, we report these asymptotic elds in
Appendix A.
5.2. Crack orthogonally terminating at the interface
For the case of a crack orthogonally terminating at the bi-
material interface; however, path independence of J-integral is no
longer available [42]. To investigate the behavior of J-integral for
this kind of cracked laminated plate, we consider a properly
chosen circular contour with center at the tip and of arbitrarily
small radius within K-dominant region (say, r). Substituting the
asymptotic elds provided in Ref. [43], into Eq. (34) and taking
integration along this path counterclockwisely, we then have
J =
1
E
1
r
2l1
(aK
2
I
bK
2
II
) (40)
where a and b are the Dundurs bi-material parameters that are
dened in Eq. (2). It is observed that, in addition to K
I
and K
II
, an
extra term r
2l1
is contained in Eq. (40). With the presence of this
extra term, we anticipate that the value of J-integral either
vanishes (when cracked material 1 is softer than material 2, with
0.5olo1) or approaches innity (when material 1 is stiffer than
material 2, with 0olo0.5) as the limiting condition r-0 by
denition.
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1309
From Eq. (40), the M-integral is related to the SIFs through the
relation
K
I
=
E
1
2a
r
12l
M
I
(41)
K
II
=
E
1
2b
r
12l
M
II
(42)
where the auxiliary elds are chosen as the asymptotic terms of
the crack-tip elds as listed in the Appendix B.
The contour integral in Eq. (35), however, is not the best form
suited for numerical calculations. We therefore recast the integral
into an equivalent domain form by multiplying the integrand by a
sufciently smooth weighting function q which takes a value of
unity on an open set containing the crack-tip and vanishes on an
outer prescribed contour C
0
as shown in Fig. 6(a). Assuming the
crack faces are traction free, the interaction integral may be
written as
I =
_
C
s
(1)
ik
e
(2)
ik
d
xj
s
(1)
ij
@u
(2)
i
@x
s
(2)
ij
@u
(1)
i
@x
_ _
qm
j
dG (43)
where the contour C=GC

C
0
and m is the unit outward
normal to the contour C. Now using the divergence theorem and
passing to the limit as the contour G is shrunk to the crack-tip, we
can give the following equation for the interaction integral in
domain form:
I =
_
O
d
s
(1)
ik
e
(2)
ik
d
xj
s
(1)
ij
@u
(2)
i
@x
s
(2)
ij
@u
(1)
i
@x
_ _
@q
@x
j
dA (44)
where we have used the relations m
j
=n
j
on G and m
j
=n
j
on C

,
C

and C
0
.
For the numerical evaluation of the above integral, as shown in
Fig. 6(b), the domain O
d
is then set to be the collection of all
elements which have a node within a radius of r
d
=r
k
h
e
and this
elements set is denoted as N
d
. h
e
is the characteristic length of an
element touched by the crack-tip and the quantity is calculated as
the square root of the element area.
The weighting function q that appears in the domain form of the
interaction integral is set: if a node n
i
that is contained in the
element eAN
d
lies outside O
d
, then q
i
=0; if node n
i
lies in O
d
, then
q
i
=1. Since the gradient of q appears in Eq. (44), the elements set
N
d
in
with all the nodes inside O
d
as shown in Fig. 6(b) contributes
nothing to the interaction integral, and non-zero contribution to the
integral is obtained only for elements set N
d
eff
with an edge that
intersects the boundary qO
d
. Therefore, Eq. (44) can be given by
I =

N
d
eff
m = 1
_
O
d
eff ,m
s
(1)
ik
e
(2)
ik
d
xj
s
(1)
ij
@u
(2)
i
@x
s
(2)
ij
@u
(1)
i
@x
_ _
@q
@x
j
dA (45)
where O
d
eff ,m
is the domain of the mth element in the elements set
N
d
eff
.
It is noted that each triangular element domain hosts three
sub-parts of smoothing domains associated with three edges, e.g.,
for element domain O
d
eff ,m
, three sub-parts O
s
1
,O
s
2
and O
s
3
are
involved as shown in Fig. 6(c). The strains are smoothed and thus
constant in each parts belonging to three different smoothing
domains. Therefore, the integration in Eq. (45) for one element
(e.g., O
d
eff ,m
) is conducted by the summation of integration for

C
0
C+
C
x
y
n
m

d
o
d
r
d
in
N
d
eff
N
d
out
N
d

eff,m
1
s

2
s

3
s

Fig. 6. (a) Conventions at crack-tip. Domain O

d
is enclosed by G, C

,C

and C
0
.Unit normal m
j
=n
j
on G and m
j
=n
j
on C

, C

and C
0
; (b) different types of elements at
the crack-tip for calculation of the interaction integral; (c) each triangular element domain hosts three sub-parts of the smoothing domains associated with three edges,
e.g., for element domain O
d
eff ,m
, three sub-parts O
s
1
, O
s
2
and O
s
3
are involved.
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1310
three sub-parts (O
s
1
,O
s
2
and O
s
3
)
_
O
d
eff ,m
W
@q
@x
j
dA =

3
n = 1
_
O
s
n
W
@q
@x
j
dA (46)
where
W =s
(1)
ik
e
(2)
ik
d
xj
s
(1)
ij
@u
(2)
i
@x
s
(2)
ij
@u
(1)
i
@x
(47)
6. Numerical implementation
The numerical procedure for the proposed T5 singular element
incorporating ES-FEM is outlined as follows:
(1) Divide the problem domain into a set of elements and obtain
information on node coordinates and element connectivity;
(2) Create the smoothing domains using the rule given in Section
4.2, and sub-divide the smoothing domains associated with
the crack-tip edges into several (n
sc
) sub-cells;
(3) Loop over smoothing domains
a. Determine the outward unit normal, and the proper number
of Gauss points for each boundary segment of the smoothing
domains;
b. Calculate the shape functions of Gauss points, especially
for the crack-tip smoothing domain by using Eq. (20);
c. Compute the smoothed strain gradient matrix B
i
(x
k
) by
using Eqs. (27)(29).
d. Evaluate the smoothed stiffness matrix K
s
ij,k
and load
vector of the current smoothing domain;
e. Assemble the contribution of the current smoothing domain
to form the system stiffness matrix K and force vector;
(4) Implement essential boundary conditions;
(5) Solve the linear system of equations to obtain the nodal
displacements;
(6) Evaluate strains and stresses at locations of interest;
(7) Calculate the fracture parameters including the J-integral (energy
release rate), and the stress intensity factors of K
I
as well as K
II
.
7. Numerical examples
Two sets of numerical example problems are presented in the
following two subsections. In the rst subsection, an interface
cracked bi-material plate subjected to pure tension remote load-
ing is investigated for verication of the proposed T5 singular
element. In the second subsection, we consider a crack terminat-
ing normally at the bi-material interface, and numerical results
are given for the variable order singularity with a broad range of
material combination.
7.1. Crack along the bi-material interface
The problem of a crack lying on the interface in a bi-material
plate is rst studied. The exact solution to this problem under
remote traction t =s
o
yy
it
o
xy
was obtained by Rice and Sih [4]. The
solution for K
I
and K
II
at the right crack-tip is [3,4]
K=K
I
iK
II
=(s
o
yy
it
o
xy
)(12ie)

pa
_
(2a)
ie
(48)
We rst consider the case of pure tension remote loading with
s
o
yy
=1:0MPa. The crack dimension is selected as 2a=2 mm. In
the computation, only half of the specimen is considered with the
appropriate displacement constraint due to symmetry (see Fig. 7).
The right edge is constrained in x direction to remove the edge
singularity [4]. The factors K
0
and G
0
are used to normalize the
stress intensity factors and the energy release rate, respectively
K
0
=s
o
yy

pa
_
=1:7725MPa

mm
_
G
0
=
(s
o
yy
)
2
a
E
1
=10
3
MPa mm
(49)
The material constants used in the numerical computation are:
E
1
=110
3
MPa, E
2
/E
1
=22, v
1
=0.3 and v
2
=0.2571 and plane
strain conditions are assumed. The exact solutions from Eq. (48) are
K
I
K
0
=1:008
K
II
K
0
=0:1097
G
G
0
=1:4358 (50)
Since the exact solution is for the innite domain problem, the
sample size W/a=30 is used in all models to avoid the effect of
nite size. Five structured meshes are employed by successively
mesh renement, with the size of elements in the vicinity of
crack-tip ranging with a/h
c
=3.0, 4.0, 6.0, 8.0 and 10.0, where h
c
is
the mesh spacing in the vicinity of the crack-tip (a representation
mesh with a/h
c
=8.0 is shown in Fig. 8). All the studies are
conducted using the path independence radius parameter r
k
=5
unless stated otherwise. The strain energy and the error in energy
norm are, respectively, dened as
E
(O)
=
1
2
_
O
e
T
DedO (51)
e
e
= E
num
(O)
E
ref
(O)

1=2
(52)
The relative error of fracture parameters is given by
e =
FP
num
FP
ref
FP
ref
100% (53)
where the superscript ref denotes the exact or reference solution,
numerical denotes numerical solution obtained using a numerical
method and FP can be the stress intensity factors (e
k
) or the
energy release rate (e
g
).
7.1.1. Inuence of the number of smoothing cells and Gauss points
As mentioned in Section 4.2, for a crack-tip smoothing domain, a
sufcient number of smoothing cells in a smoothing domain, n
sc
,
and Gauss points along a boundary segment of the smoothing cell,
n
gau
should be used to ensure accuracy. In order to x these two
parameters, a study of their inuences on the energy release rate
G and strain energy is rst conducted. In this study, a mesh with
W
yy

a
W
W
u
x
= 0
u
x
= 0
Fig. 7. Crack along the bi-material interface under remote tension (half model).
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1311
a/h
c
=6.0 is used, and the parameters are varied until the difference
between two consecutive computations is less than a specied
tolerance. When varying n
sc
, the n
gau
is kept constant, and likewise
for n
gau
. The dependence of G and strain energy on the number of
smoothing cells and Gauss points is given in Table 1. It is seen that
with the increase in the number of n
gau
, the normalized G and strain
energy initially decrease and reach a constant value beyond 5 Gauss
points along a boundary segment. On the other hand, beyond
3 smoothing cells in a smoothing domain, the normalized G and
strain energy remain almost constant. Thus, n
sc
=3 and n
gau
=5 are
adopted for all the models discussed later.
7.1.2. Accuracy study
To investigate the accuracy, ve triangular models aforemen-
tioned are computed using the standard FEM and ES-FEM without
any singular elements, the FEM with T6 quadratic singular
elements: FEM-sin-T6, and the ES-FEM with our T5 singular
elements: ES-FEM-sin-T5. Note that the six-node triangular
meshes are used for the FEM-sin-T6, and the degrees of freedom
(DOF) of these quadratic meshes are the same as those of the
linear meshes other methods use for the purpose of fair compar-
ison. The comparisons of the normalized SIFs and energy release
rate G using different numerical methods are performed, and the
results are tabulated in Table 2. It can be found that the ES-FEM
with our T5 singular elements improves signicantly the accuracy
in comparison with the standard FEM and ES-FEM without
singular elements. In addition, compared to the FEM-sin-T6, the
K
I
, K
II
and G obtained by our T5 singular elements are slightly
better. Most importantly, the relative error is around a few
percent even for the coarsest meshes, and decreases with mesh
renement. In other words, with mesh renement, all numerical
methods converge to the reference solutions.
7.1.3. Convergence rate study
Fig. 9 compares the convergence rate, R, in energy norm with
respect to the meshing distance, h, for different numerical
methods. It can be observed that the proposed ES-FEM-sin-T5
(R=0.74) outperforms the standard FEM (R=0.50) and ES-FEM
(R=0.56), while the FEM-T6 (R=0.91) converges faster than our
ES-FEM-sin-T5 (R=0.74). However, we should keep in mind that
the optimal convergence rate for T6 quadratic elements is
R
op
=2.0, while the optimal value for T3 linear elements is
R
op
=1.0. In that case, it is easily concluded that the proposed
ES-FEM-sin-T5 (R/R
op
=0.74) produces the most optimal conver-
gence rate compared to the standard FEM (R/R
op
=0.50), ES-FEM
(R/R
op
=0.56) and the FEM-sin-T6 (R/R
op
=0.46). Note also that the
suboptimal convergence rate obtained by the standard FEM and
FEM-sin-T6 for a singular problem is quite similar to the results
available in the literature [44].
7.1.4. Computational efciency study
In this study, the computational efciency in terms of the error
in energy norm against computation time (unit: s) is compared
for different numerical methods, and plotted in Fig. 10. It is
evident that the ES-FEM with our T5 singular elements always
produces higher computational efciency, i.e., accuracy to com-
putational time ratio, compared to other numerical methods. In
particular, at a computation time of about 10
0.6
, the ES-FEM-sin-
T5 leads to an error approximately 10
3.34
, FEM-sin-T6 10
3.03
,
ES-FEM 10
2.56
, and FEM 10
2.38
. All of these conrm numeri-
cally the superiority of our T5 singular elements in convergence
rate, accuracy and computational efciency.
a
interface
Fig. 8. Crack along the bi-material interface: meshes in the vicinity of the crack-
tip with a/h
c
=8.0 (a=1, W=30).
Table 1
Crack along the bi-material interface: study of the effects of the number of
smoothing cells in a crack-tip smoothing domain, n
sc
, and Gauss points along a
boundary segment of the smoothing cell, n
gau
(unit for E
(O)
: 10
4
J/m
2
).
n
sc
E
(O)
G/G
0
(e
g
) N
gau
E
(O)
G/G
0
(e
g
)
1 1.050235 1.4512 (1.1) 1 1.049016 1.4306 (0.3)
2 1.049179 1.4331 (0.1) 3 1.048783 1.4297 (0.4)
3 1.048779 1.4297 (0.4) 5 1.048779 1.4297 (0.4)
4 1.048768 1.4295 (0.4) 7 1.048778 1.4297 (0.4)
Table 2
Crack along the bi-material interface: comparison of the accuracy in the stress intensity factors and energy release rate using different numerical methods.
Exact solution Mesh (a/h
c
) 3.0 (e
k
) 4.0 (e
k
) 6.0 (e
k
) 8.0 (e
k
) 10.0 (e
k
)
K
I
/K
0
=1.008 FEM 0.9740 (3.4) 0.9834 (2.4) 0.9903 (1.8) 0.9939 (1.4) 0.9959 (1.2)
FEM-sin-T6 1.0030 (0.5) 1.0046 (0.4) 1.0057 (0.3) 1.0059 (0.2) 1.0062 (0.2)
ES-FEM 0.9944 (1.3) 0.9989 (0.9) 1.0020 (0.6) 1.0033 (0.5) 1.0041 (0.4)
ES-FEM-sin-T5 1.0038 (0.4) 1.0051 (0.3) 1.0059 (0.2) 1.0061 (0.2) 1.0063 (0.2)
K
II
/K
0
=0.1097 FEM 0.1244 (13.4) 0.1192 (8.6) 0.1141(4.6) 0.1122 (2.3) 0.1111 (1.3)
FEM-sin-T6 0.1119 (2.0) 0.1106 (0.9) 0.1105 (0.9) 0.1101 (0.3) 0.1099 (0.2)
ES-FEM 0.1165 (6.2) 0.1134 (3.4) 0.1118 (1.0) 0.1108 (1.0) 0.1104 (0.6)
ES-FEM-sin-T5 0.1114 (1.6) 0.1104 (0.7) 0.1103 (0.6) 0.1100 (0.3) 0.1099 (0.2)
G/G
0
=1.4358 FEM 1.3459 (6.3) 1.3699 (4.6) 1.3872 (3.4) 1.3964 (2.7) 1.4018 (2.4)
FEM-sin-T6 1.4219 (1.0) 1.4261 (0.7) 1.4292 (0.5) 1.4296 (0.4) 1.4304 (0.4)
ES-FEM 1.3993 (2.5) 1.4109 (1.7) 1.4188 (1.2) 1.4223 (0.9) 1.4244 (0.8)
ES-FEM-sin-T5 1.4242 (0.8) 1.4273 (0.6) 1.4297 (0.4) 1.4301 (0.4) 1.4306 (0.3)
Note: the FEM-sin-T6 uses the six-node triangular meshes. However, the comparison in every column is conducted under the meshes with the same DOFs for the purpose
of justice.
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1312
7.1.5. Material combinations study
To study the performance of the ES-FEM with our T5 singular
elements on the interface cracks for different material property
pairs, we varied the ratio E
1
/E
2
from 2 to 1000 with the constant
Poison ratios: v
1
=0.3 and v
2
=0.2571. It is observed that the
results are also accurate to within a few percent relative errors as
listed in Table 3.
7.2. Crack terminating normally at the bi-material interface
Next, we consider a benchmark problem solved in Refs. [6,9],
where a crack terminates normally at the interface between a
plain strain bi-material specimen as shown in Fig. 11. In the
calculation, the ES-FEM with our T5 singular elements is used to
demonstrate its validity for variable order singularity problems.
The model dimensions are: a=1.0 mm and h=b=9a=9.0 mm.
The bi-material (material
1
material
2
) is an epoxyaluminum
combination. For epoxy, Youngs modulus E=1 148 MPa, Pois-
sons ratio v=0.35; and for aluminum, E=26 498 MPa, v=0.3.
From Eq. (5), when epoxy is the cracked material, the value of the
order of stress singularity order is l1=0.66191=0.3381,
and when aluminum is the cracked material, the corresponding
singular order is l1=0.17521=0.8248.
The crack is subjected to a uniform pressure p=1.0 MPa, and
terminates normally at the interface of aluminum and epoxy. In
the computation, the uniform tensions s
1
and s
2
are applied,
respectively, at the material 1 and material 2 boundaries in order
to describe the behavior due to the pressure on the crack. To
ensure constant strain along the boundaries under plane strain, s
1
is taken as p, and s
2
is taken as s
1
(1v
2
1
)E
2
=(1v
2
2
)E
1
. The
specimen is analyzed by using the mesh representation as shown
in Fig. 12. The discretized model is progressively rened as the
elements approach the crack-tip, and the ratio of crack length to
element size at the crack-tip is taken as a/h
c
=100.
-2.6 -2.4 -2.2 -2 -1.8 -1.6 -1.4 -1.2 -1 -0.8 -0.6
-4
-3.5
-3
-2.5
-2
-1.5
L
o
g
1
0

(
e
e
)
Log
10
(h)
FEM R = 0.50
FEM-sin-T6 R = 0.91
ES-FEM R = 0.56
ES-FEM-sin-T5 R = 0.74
Fig. 9. Convergence rate in terms of energy norm for the problem of a crack along
the bi-material interface under remote tension.
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2
-4
-3.5
-3
-2.5
-2
-1.5
L
o
g
1
0

(
e
e
)
FEM
FEM-sin-T6
ES-FEM
ES-FEM-sin-T5
-2.38
-2.56
-3.03
-3.34
Log
10
(t)
Fig. 10. Comparison of computational efciency in terms of energy norm for the
problem of a crack along the bi-material interface under remote tension.
Table 3
Crack along the bi-material interface: material mismatch study.
E
1
/E
2
e ES-FEM-sin-T5 Exact solution
K
I
/K
0
(e
k
) K
II
/K
0
(e
k
) G/G
0
(e
g
) K
I
/K
0
K
II
/K
0
G/G
0
2 0.0249 0.9995 (0.12) 0.0327 (0.64) 2.1497 (0.25) 1.0007 0.0325 2.1551
4 0.0516 1.0016 (0.14) 0.0678 (0.72) 1.7633 (0.29) 1.0030 0.0673 1.7684
8 0.0699 1.0039 (0.17) 0.0915 (0.36) 1.5620 (0.35) 1.0056 0.0912 1.5675
20 0.0833 1.0060 (0.19) 0.1088 (0.16) 1.4375 (0.41) 1.0079 0.1086 1.4434
40 0.0883 1.0072 (0.17) 0.1151 (0.03) 1.3959 (0.39) 1.0089 0.1151 1.4013
100 0.0914 1.0081 (0.14) 0.1192 (0.11) 1.3712 (0.33) 1.0096 0.1191 1.3758
1000 0.0933 1.0089 (0.11) 0.1217 (0.05) 1.3569 (0.26) 1.0100 0.1216 1.3604
2a
h
h
b b
h = b = 9a
x
y
p = 1.0MPa
Material 1 Material 2

1
, v
1

2
, v
2
Fig. 11. Crack terminates normally at the interface under uniform pressure on the
crack faces.
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1313
7.2.1. Stress singularity
Fig. 13 depicts the logarithmic stress distributions along the
radius path of y=451 for a material
1
material
2
pair of epoxy
aluminum. We can notice that the s
rr
and s
yy
stress distributions
are almost linear and parallel to each other by omitting numerical
errors. A real singular order as l1=0.3303 is obtained in the
r-direction, and the corresponding value in the y-direction is
l1=0.3279. It is clear that the results obtained by our ES-FEM-
sin-T5 are in close agreement with the analytical result 0.3381
from Eq. (5), with the deviations to be 2.31%, and 3.01%, respec-
tively. In addition, Fig. 14 shows logarithmic stress distributions
for the material
1
material
2
pair of aluminumepoxy. The singular
order is found as l1=0.8307 and 0.8124 in the r-direction
and y-direction, respectively. Compared to the analytical result
0.8248, the relative errors are both under 3%
7.3. Energy release rate
In this study, the behavior of energy release rate G or J-integral
with respect to various normalized domain radius r
d
/a (see Fig. 6)
is examined, and the results are listed in Table 4 and Table 5 for
the two material combinations aforementioned. Note that, the
values of r
d
/a in the calculation are taken small enough to be
inside the zone of dominance of the asymptotic solution. It is
observed that path independence of J-integral breaks down: for
the epoxyaluminum combination, the J-integral decreases with
the increase of domain radius r
d
/a in Table 4. On the contrary, for
the aluminumepoxy pair, the J-integral increases with r
d
/a
increasing in Table 5. However, as long as we normalize the
J-integral with a parameter (r
d
=a)
2l1
, the path-independence (in
a modied sense) can be observed for J-integral as shown in the
fourth column of Tables 4 and 5, regardless of material combina-
tions. These phenomena are well consistent with the founding in
Ref. [42].With this property of path-independence (in a modied
sense) for J-integral (thus M-integral), we can then calculate the
stress intensity factors by using the M-integral method.
7.3.1. Stress intensity factors
In addition to the energy release rate G (J-integral),
Tables 4 and 5 also list the stress intensity factor K
I
obtained by
the ES-FEM with our T5 singular elements for two material
combinations. The factor K
0
=p

p
_
a
1l
is used to normalize the
stress intensity factors, and the results obtained by Cook and
Erdogan [6] are also given for comparison purposes. It can be
found that the results of normalized K
I
/K
0
appear to be almost
2a
interface
Fig. 12. Crack terminates normally at the interface: (a) meshes of the whole discretized model and (b) meshes in the vicinity of the crack-tip with a/h
c
=100.
-2 -1.8 -1.6 -1.4 -1.2 -1 -0.8
0.4
0.5
0.6
0.7
0.8
0.9
L
o
g
1
0

(

i
j
/
p
)

rr

Slope: - 0.3224
Slope: - 0.3303
Log
10
(r/a)
Fig. 13. Crack terminates normally at the interface with material
1
material
2
of
epoxyaluminum: logarithmic stress distributions along the radius path of y=451.
-2 -1.8 -1.6 -1.4 -1.2 -1 -0.8
-0.2
0
0.2
0.4
0.6
0.8
L
o
g
1
0

(

i
j
/
p
)

rr

Slope: - 0.8124
Slope: - 0.8307
Log
10
(r/a)
Fig. 14. Crack terminates normally at the interface with material
1
material
2
of
aluminumepoxy: logarithmic stress distributions along the radius path of y=451.
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1314
invariant with respect to different selections of r
d
/a. Furthermore,
the average normalized K
I
/K
0
values obtained by our ES-FEM-sin-
T5 agree well with the results computed by [6], with deviation to
remain under 2.5%, as anticipated.
7.3.2. Mixed-mode loads study
To further investigate the capability of our T5 singular ele-
ments in characterizing the variable order singular behavior, we
reconsider the same specimen and impose the mixed loads
(s, t)=(2.76,1.38)MPa as shown in Fig. 15. The results of
mixed-mode normalized K
I
/K
0
and K
II
/K
0
with respect to different
material combinations are listed in Table 6. Also listed in the table
is the corresponding ratio of the two SIFs K
II
/K
I
.
Although there is no analytical form for direct verication of
the above calculation, the validity the mixed-mode SIFs can be
demonstrated by addressing the following two points. First, the
numerical results of K
I
and K
II
by our ES-FEM-sin-T5 appear to be
invariant with respect to different selections of M-integral radius
r
d
/a, as anticipated by Eq. (40). Second, it is observed that the
absolute value for K
II
/K
I
decreases as the relative shear modulus
m
1
/m
2
decreases. This indicates that the contribution from mode II
stress component is more signicant than mode I when the
cracked material (material
1
) is relatively softer, and vice versa.
This founding is quite similar to the results available in the
literature [15].
8. Conclusion
In this work, a novel triangular ve-node (T5) singular crack-
tip element is formulated in G space with strain smoothing to
simulate a variable order stress singularity. In the radial direction,
the displacement elds are enriched via a direct point interpola-
tion with a proper fractional order of extra basis functions (r
l
,
0olo1). Also, a quadratic polynomial function is utilized to
model the crack-tip variation of displacements in the circumfer-
ential direction. In addition, a basic mesh of linear T3 elements
with one layer of T5 singular elements is used to produce the
(r
l1
, 0olo1) singularity near the crack-tip. Through the for-
mulation, we conclude theoretically that:
+ The enriched displacements in the T5 element are on top of the
completely linear led. Therefore, the linear consistency property
and the square integrable property are all ensured throughout the
entire problem domain, ensuring the stability and the conver-
gence of the solution.
+ In the T5 singular element, the singular terms of functions as well
as mapping procedures are no longer necessary to compute the
stiffness matrix due to the use of strain smoothing.
+ The T5 singular element eliminates the need to shift the position
of the side nodes adjacent to the crack-tip, and is thus quite
straightforward and easily implemented in existing codes
Intensive studies including two classic categories of cracks
associated with the bi-material interfaces have been then con-
ducted, and we conrmed numerically that:
+ The T5 singular element together with ES-FEM outperforms in
convergence rate, accuracy and computational efciency in
comparison with the standard FEM, ES-FEM and even the FEM
with T6 singular elements.
+ The numerical results obtained by the T5 singular element are
in excellent agreement with the corresponding analytical or
Table 4
Crack terminates normally at the interface with material
1
material
2
of epoxy
aluminum: energy release rate and stress intensity factors (unit for J-integral:
MPa mm).
No. r
d
/a G (J-integral)
J=(r
d
=a)
2l1
K
I
/K
0
ES-FEM-sin-T5 Ref. [6]
1 0.05 1.088410
3
2871.289 2.83
2 0.10 1.324010
3
2790.878 2.79
3 0.15 1.488210
3
2750.672 2.77
4 0.20 1.621710
3
2730.847 2.76
Ave. 2790.536 2.79 2.85
Note: r
d
is the domain radius in J-integral or M-integral as shown in Fig. 6.
Table 5
Crack terminates normally at the interface with material
1
-material
2
of aluminum-
epoxy: energy release rate and stress intensity factors (unit for J-integral:
MPa mm).
No. r
d
/a G (J-integral)
J=(r
d
=a)
2l1
K
I
/K
0
ES-FEM-sin-T5 Ref. [6]
1 0.05 3.349210
2
48.8403 0.115
2 0.10 2.020610
2
46.7781 0.110
3 0.15 1.487710
2
44.8815 0.108
4 0.20 1.196010
2
43.3433 0.106
Ave. 44.6358 0.110 0.112
2 a
h
h
b b
h = b = 9a
x
y
Material 1 Material 2
= 2.76MPa

1
, v
1

2
, v
2

1
.
3
8
M
P
a
Fig. 15. Crack terminates normally at the interface under mixed-mode loads.
Table 6
Crack along the bi-material interface: normalized stress intensity factors under
mixed-mode loads.
material
1
material
2
epoxyaluminum
(m
1
/m
2
=23.08)
aluminumepoxy
(m
1
/m
2
=0.043)
r
d
/a K
I
/K
0
K
II
/K
0
K
I
/K
0
K
II
/K
0
0.05 1.8722 2.3622 0.6655 0.1543
0.1 1.8850 2.3420 0.6861 0.1478
0.2 1.8737 2.3361 0.6651 0.1425
0.4 1.8373 2.3282 0.6550 0.1405
Ave. 1.8673 2.3421 0.6679 0.1463
9K
II
/K
I
9 1.254 0.219
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1315
reference solutions for variable order singularity for a wide
range of material combinations, with a fraction of deviation.
Finally, we would like to mention that it seems that the
proposed singular elements could be extended and serve as a
competitive means to solve complex problems in the context of
3D as well, which will be a topic of forthcoming communications.
Appendix A
The following gives the near-tip asymptotic displacement elds
for a crack lying on the interface of a bi-material plate. These
asymptotic elds, i.e., auxiliary displacement elds in the local xy
crack-tip coordinate system as shown in Fig. 1 (a) are given by
u
x
(r,y) =
1
4m
i
cosh(pe)

r
2p
_
(f
I
x
K
I
f
II
x
K
II
)
u
y
(r,y) =
1
4m
i
cosh(pe)

r
2p
_
(f
I
y
K
I
f
II
y
K
II
) (A:1)
where e is the bi-material constant that are dened in Eq. (3), and m
i
is the shear modulus of material i(i=1,2). (r,y) corresponds to a
local polar coordinate system (see Fig. 1(a)). The terms f
I
i
and f
II
i
i=x, y are written as
f
I
x
=D
i
2d
i
sinysinj, f
II
x
=C
i
2d
i
sinycosj (A:2)
f
I
y
=C
i
2d
i
sinycosj, f
II
y
=D
i
2d
i
sinysinj (A:3)
where
d
1
=e
(py)e
d
2
=e
(py)e
j=elogr
y
2
(A:4)
C
i
=b
/
g
i
cos
y
2
bg
/
i
sin
y
2
, D=bg
i
cos
y
2
b
/
g
/
i
sin
y
2
(A:5)
b =
0:5cos(elogr)esin(elogr)
0:25e
2
, b
/
=
0:5sin(elogr)ecos(elogr)
0:25e
2
(A:6)
g
i
=k
i
d
i

1
d
i
, g
/
i
=k
i
d
i

1
d
i
(A:7)
in which k
i
are the Kolosov constant of material i. The auxiliary
strain and stress components can be obtained from the auxiliary
displacement elds using Hookes law.
Appendix B
In this appendix, we list the auxiliary displacement elds for a
crack terminating normally at the interface between a bi-material
plate in the local crack-tip coordinate system as shown in Fig. 1
(b) used in M-integral computations. In the following equations,
l1 is the order of singularity, m
i
and k
i
are the shear modulus and
the Kolosov constant, respectively, of material i(i=1,2), and a and b
are the Dundurs bi-material parameters that are dened in Eq. (2)
u
x
(r,y) =r
l
(f
I
x
K
I
f
II
x
K
II
)
u
y
(r,y) =r
l
(f
I
y
K
I
f
II
y
K
II
) (B:1)
where
f
x
=C
1
(b
2
1)
cos
sin
l(py) y
_ _
l
cos
sin
l(py)y
_ _
(k
1
l)
cos
sin
{l(py) y]
_ _ _
(ab)(b1)(2l81) (k
1
l)
cos
sin
{(l1)y](l71)
cos
sin
{(l1)y]
_ _ _
(B:2)
fy =C
1
7(b
2
1)
sin
cos
l(py) y
_ _
l
sin
cos
l(py)y
_ _
(k
1
l)
sin
cos
l(py) y]
_
_ _
7(ab)(b1)(2l81) (k
1
l)
sin
cos
{(l1)y](l71)
sin
cos
(l1)y]
_
_ _
(B:3)
for the domain O
1u
(0ryrp/2), and
f
x
=C
2
(b1)(k
2
l)
cos
sin
{(l1)(py)]{b(3l81) (l81)]
cos
sin
(l1)(py)]
_
_
(B:4)
f
y
= 8C
2
(b1)(k
2
l)
sin
cos
(l1)(py)]{b(3l81)
_
_
(l81)]
cos
sin
(l1)(py)]
_
(B:5)
for the domain O
2
(p/2ryr3p/2), and
f
x
= 7C
1
(b
2
1)
cos
sin
l(3py)y
_ _
l
cos
sin
l(yp)y]
_
_ _
(k
1
l)
cos
sin
l(yp)y]
_
(ab)(b1)(2l81)
(k
1
l)
cos
sin
{(l1)(2py)](l71)
cos
sin
(l1)(2py)]
_
_ _
(B:6)
f
y
= 8C
1
7(b
2
1)
sin
cos
l(3py)y
_ _
l
sin
cos
l(yp)y]
_
_ _
(k
1
l)
sin
cos
l(yp)y]
_
7(ab)(b1)(2l81)
(k
1
l)
sin
cos
{(l1)(2py)](l71)
sin
cos
(l1)(2py)]
_
_ _
(B:7)
for the domain O
1d
(3p/2ryr2p). In the above equations, the
upper signs of
//
7
//
,
//
7
//
,
cos
sin
, and
sin
cos
are chosen for f
I
x
and f
I
y
, and the lower signs for f
II
x
and f
II
y
.
References
[1] Hutchinson JW, Suo Z. Mixed mode cracking in layered materials. In:
Hutchinson JW, Wu TY editors. Advances in applied mechanics; 1992.
p. 63191.
[2] Williams ML. The stress around a fault or crack in dissimilar media. Bull
Seism Soc Am 1959;49:199204.
[3] Rice JR, Sih GC. Plane problems of cracks in dissimilar media. J Appl Mech
1965;32:41823.
[4] Rice JR. Elastic fracture mechanics concepts for interfacial cracks. J Appl Mech
1988;55:98103.
[5] Zak AR, Williams ML. Crack point singularities at a bi-material interface.
J Appl Mech 1963;30:1423.
[6] Cook TS, Erdogan F. Stresses in bonded materials with a crack perpendicular
to the interface. Int J Eng Sci. 1972;10:67797.
[7] Dundurs J. Edge-bonded dissimilar orthogonal elastic wedges. J Appl Mech
1969;36:6502.
[8] Brusselaarsa N, Mogilevskayab SG, Crouchb SL. A semi-analytical solution for
multiple circular inhomogeneities in one of two joined isotropic elastic half-
planes. Eng Anal Boundary Elem 2007;31:692705.
[9] Tracey DM, Cook TS. Analysis of power type singularities using nite
elements. Int J Numer Methods Eng 1977;11:122533.
[10] Stern M. Families of consistent elements with singular derivative elds. Int J
Numer Methods Eng 1979;14:40922.
[11] Hughes TJR, Stern M. Techniques for developing special nite element shape
functions with particular references to singularities. Int J Numer Methods
Eng 1980;15:73351.
[12] Gao YL, Tan CL, Selvadurai APS. Stress intensity factors for cracks around or
penetrating an elliptic inclusion using the boundary element method. Eng
Anal Boundary Elem 1992;10:5968.
[13] Lim WK, Lee CS. Evaluation of stress intensity factors for a crack normal to a
bi-material interface using isoparamteric nite elements. Eng Fract Mech
1995;62:6570.
[14] Pan E, Amadeib B. Boundary element analysis of fracture mechanics in
anisotropic bimaterials. Eng Anal Boundary Elem 1999;23:68391.
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1316
[15] Chang JH, Wu DJ. Calculation of mixed-mode stress intensity factors for a
crack normal to a bimaterial interface using contour integrals. Eng Fract
Mech 2003;70:167595.
[16] Abdi RE, Valentin G. Isoparametric elements for crack normal to the
bimaterial interface. Comput Struct 1989;33:2418.
[17] Barsoum RS. On the use of isoparametric nite elements in linear fracture
mechanics. Int J Numer Methods Eng 1976;10:55164.
[18] Wu YL. Collapsed isoparametric element as a singular element for a crack
normal to the bimaterial interface. Comput Struct 1993;47:93943.
[19] Lim WK, Kim SC. Further study to obtain a variable power singularity using
quadratic isoparametric elements. Eng Fract Mech 1994;47:2238.
[20] Yavari A, Sarkani S, Moyer ET. On quadratic isoparametric transition ele-
ments for a crack normal to a bimaterial interface. Int J Numer Methods Eng
1999;46:45769.
[21] Belytschko T, Black T. Elastic crack growth in nite elements with minimal
remeshing. Int J Numer Methods Eng 1999;45:60120.
[22] Moes N, Dolbow J, Belytschko T. A nite element method for crack growth
without remeshing. Int J Numer Methods Eng 1999;46:13150.
[23] Belytschko T, Moes N, Usui S, Parimi C. Arbitrary discontinuities in nite
elements. Int J Numer Methods Eng 2001;50:9931013.
[24] Rabczuk T, Zi G. A meshfree method based on the local partition of unity for
cohesive cracks. Comput Mech 2007;39:74360.
[25] Sukumar N, Huang ZY, Pre vost JH, Suo Z. Partition of unity enrichment for bi-
material interface cracks. Int J Numer Methods Eng 2004;59:1075102.
[26] Menk A, Bordas S. Numerically determined enrichment functions for the
extended nite element method and applications to bi-material anisotropic
fracture and polycrystals. Int J Numer Methods Eng 2010;83:80528.
[27] Chen JS, Wu CT, Yoon Y. A stabilized conforming nodal integration for
Galerkin mesh-free methods. Int J Numer Methods Eng 2001;50:43566.
[28] Yoo JW, Moran B, Chen JS. Stabilized conforming nodal integration in the
natural-element method. Int J Numer Methods Eng 2004;60:86190.
[29] Liu GR. A generalized gradient smoothing technique and the smoothed
bilinear form for Galerkin formulation of a wide class of computational
methods. Int J Comput Methods 2008;5:199236.
[30] Liu GR. A G space theory and weakened weak (W
2
) form for a unied
formulation of compatible and incompatible methods: part I Theory. Int J
Numer Methods Eng 2009;81:1093126.
[31] Liu GR, Dai KY, Nguyen TT. A smoothed nite element method for mechanics
problems. Comput Mech 2007;39:85977.
[32] Liu GR, Nguyen-Thoi T, Lam KY. A node-based smoothed nite element
method for upper bound solution to solid problems (NS-FEM). Comput Struct
2008;87:1426.
[33] Dohrmann CR, Heinstein MW, Jung J, Key SW, Witkowski WR. Node-based
uniform strain elements for three-node triangular and four-node tetrahedral
meshes. Int J Numer Methods Eng 2000;47:154968.
[34] Liu GR, Nguyen-Thoi T, Lam KY. An edge-based smoothed nite element
method (ES-FEM) for static, free and forced vibration analysis. J Sound Vib
2009;320:110030.
[35] Chen L, Nguyen-Xuan H, Nguyen-Thoi T, Zeng KY, Wu Sc. Assessment of
smoothed point interpolation methods for elastic mechanics. Commun
Numer Methods Eng 2010;32:163555.
[36] Liu GR, Nourbakkhsh-Nia N, Chen L, Zhang YW. A general construction of
singular stress eld in the ES-FEM method for analysis of fracture problems
of mixed modes. Int J Comput Methods 2010;7:191214.
[37] Liu GR, Chen L, Nguyen-Thoi T, Zeng KY, Zhang GY. A novel singular node-
based smoothed nite element method (NS-FEM) for upper bound solutions
of fracture problems. Int J Numer Methods Eng 2010;83:146697.
[38] Liu GR, Gu YT. A point interpolation method for two-dimensional solids. Int J
Numer Methods Eng 2001;50:93751.
[39] Moran B, Shih CF. Crack tip and associated domain integrals from momentum
and energy balance. Eng Fract Mech 1987;27:61541.
[40] Li FZ, Shih CF, Needleman A. A comparison of methods for calculating energy
release rates. Eng Fract Mech 1985;21:40521.
[41] Matos PPL, McMeeking RM, Charalambides PG, Drory MD. A method for
calculating stress intensities in bimaterial fracture. Int J Fract 1989;40:
23554.
[42] Chiang CR, Chen WH, Chen KT. An integral JR0 for a crack perpendicular to
the interface of a bonded composite. Eng Fract Mech 1987;28:3017.
[43] Chen DH. A crack normal to and terminating at a bimaterial interface. Eng
Fract Mech 1994;49:51732.
[44] Strang G, Fix G. An analysis of the nite element method. Englewood, Cliffs,
NJ: Prentice-Hall; 1973.
L. Chen et al. / Engineering Analysis with Boundary Elements 35 (2011) 13031317 1317