Está en la página 1de 16

JID: APM

ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

Applied Mathematical Modelling 0 0 0 (2016) 116

Contents lists available at ScienceDirect

Applied Mathematical Modelling


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

Edge-based smoothed extended nite element method for


dynamic fracture analysis
L. Wu a,, P. Liu a, C. Shi a, Z. Zhang b, Tinh Quoc Bui c, D. Jiao d
a
College of Civil Engineering, Hunan University, Changsha, Hunan 410082, China
b
Centre for Future Materials, The University of Southern Queensland, Toowoomba, QLD 4350, Australia
c
Department of Civil and Environmental Engineering, Tokyo Institute of Technology, O-okayama Meguro, Tokyo 152-8552, Japan
d
State Key Laboratory of Green Building Materials, China Building Materials Academy, Beijing 100024, China

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

Article history: In this paper, the edge-based smoothed extended nite element method (ES-XFEM) is ap-
Received 12 November 2015 plied to the dynamic fracture analysis of two-dimensional elastic solids. The dynamic stress
Revised 9 May 2016
intensity factor (DSIF) is deduced from J-integral with the train smoothing method. With
Accepted 19 May 2016
the deduced DSIF, the implementation of numerical integration and sub-smoothing regions
Available online xxx
are numerically veried in dynamic fracture behaviors of linear elastic solids. The analysis
Keywords: shows that ES-XFEM is an ecient numerical approach to simulate the dynamic fracture
ES-XFEM problems.
Dynamic fracture mechanics 2016 Published by Elsevier Inc.
Dynamic stress intensity factor
Strain smoothing

1. Introduction

The modeling of fracture mechanics is important to describe the singular elds near the crack-tip of a solid. However,
numerical simulation of dynamic fracture problems remains a challenge in many applications, and is more complex com-
pared to the numerical simulations under static loading conditions. The studies of dynamic fracture mechanics usually give
consideration to the responses of time-dependent loads with inertial effects. However, due to the mathematical complexity,
the deriving analytical models/approaches are hard to be used in solving the fracture problems in practice regarding general
dynamic element analysis. Numerical methods may solve the limitation of analytical model. A number of numerical meth-
ods, such as extended nite element method (XFEM) [1] and Meshfree method [24], have been applied to solve fracture
problems in the last decades. The XFEM has become the most widely used approach to study the arbitrary crack growth.
However, the computation accuracy of XFEM is dependent on the mesh regulation as it is based on the level set method
(LSM) [5] and partition of unity method (PUM) [6], which are similar to classical FEM.
To improve the computation accuracy, Liu [8] proposed a cell-based nite element method (CS-FEM), which is based
on strain smoothing technique [7]. Following this, smoothed nite element method (SFEM) was developed and up-grated
to node-based smoothed nite element method (NS-FEM) [9], edge-based smoothed nite element method FEM (ES-FEM)
[10], cell-based smoothed point interpolation method (CS-PIM) [11] and face-based smoothed nite element method FEM
(FS-FEM) [12] by the same team. Because the integral calculation of the stiffness matrix by SFEM is carried out only at
smoothed region boundary, therefore there is a signicant reduction of the requirement of the quality of mesh assemble but


Corresponding author. Tel.: +86 18019672688.
E-mail addresses: wulinmei_163_com@163.com (L. Wu), cshi@hnu.edu.cn (C. Shi).

http://dx.doi.org/10.1016/j.apm.2016.05.027
S0307-904X(16)30277-3/ 2016 Published by Elsevier Inc.

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

2 L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116

Nomenclature

nj outward normal vector


ij Cauchy stress tensor
bi physical
ui , ti displacement and force on boundary, respectively
u , t , density
ij , ij stress and strain vectors, respectively
Cijkl elastic matrix
x reference point
x point on crack which is nearest to x point
en outward normal vector of crack element for x
Ni , Nj , Nk shape function for the nodes i, j, k
ui nodal displacement vector for parts
aj , b k strengthen variables for the node
T, U, Wd , Ws system kinetic energy, strain energy, damping energy and potential energy force, respectively
strain vector
stress vector
C elastic matrix
Fb , Fs , F enriched force, surface force and concentrated force, respectively
K stiffness matrix
M mass matrix
R load vectors
A(i) area of smoothing boundary domain
NI I-shape function node
n(i) (x) outward normal vector matrix of the boundary
M total number of the boundary segment  (j i )
xGP
j
period of midpoint coordinates (Gauss point) in the border  (j i )
n(jhi ) l (j i ) outward normal vector and length of  (j i ) , respectively
(i )
Ne smoothing area
I all nodes of the k set in smoothing zone
K , K enriched node set respectively by Heaviside function and crack-tip function in a smoothing domain
Bui , Baj , Bb k boundary smoothed strain matrix in conventional smoothing domain, Heaviside function and enriched
smoothed strain by crack-tip function, respectively
Ask kth smoothing domain
h separated into x, y
Nseg boundary number of the smoothing domain
Ngau Gaussian integral points on the boundary
,x , ,y need to calculate the crack-tip under local coordinate system
,x1 , ,x2 perform transformation between the whole and local coordinates
0 contour around the crack-tip
Amef f
area of the mth element in the equivalent integral domain Nedf f
Asn area of the mth element in the nth smoothing edge
Kdyn , Kexact normalized numerical solution and analytical solution of the dynamic stress intensity factor,
respectively

it still keeps high accuracy. In brief, the advantages of SFEM include (1) singular integration will not occur in the calculation
of crack-tip because no derivation of the shape function is used; (2) integration calculation is focused on smoothed region
boundary, instead of a throughout mapping; and (3) insensitive to mesh distortion, which is suitable for the analysis of
fracture problems under large deformation. Because of these advantages, smoothed XFEMs are extensively studied recently,
particularly on the fracture problems. Bordas et al. [13] combined CS-FEM with XFEM to contrive cell-based nite element
method (CS-XFEM) and applied it to the analysis of the fracture problem of thin plate. The results by the authors showed
that the stress intensity factor was affected by the mesh distortion and the convergence was better than using conventional
XFEM. Chen et al. [14] deduced the edge-based smoothed extended nite element method (ES-XFEM) and applied it to
linear elastic crack propagation, and they suggested that the precision and convergence of ES-XFEM were better than CS-
XFEM. Jiang et al. [15] applied ES-XFEM to orthotropic material fracture mechanism, which proved again that ES-XFEM was
more accurate than XFEM. However, the effectiveness and eciency of applying ES-XFEM to the analysis of dynamic fracture
problems are rarely reported, remaining unknown.

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116 3

Fig. 1. Scheme of crack problem.

The aim of this study is to examine the effectiveness and eciency of ES-XFEM for the analysis of dynamic fracture
problems in 2D elastic solids. To validate the accuracy of the method, numerical results from DSIFs are compared with the
analytical results and other numerical solutions.

2. Solution of dynamic fracture

2.1. Problem description

The diagram of a plane body with a crack is shown in Fig. 1. The boundary of  is composed of the displacement
boundary  u , stress boundary  t and the crack  c . The whole domain is recorded as .
The motion and boundary conditions are given as follows:

i j, j + bi = ui in , (1)

i j n j = ti on t , (2)

i j n j = 0 on c , (3)

ui = ui on u , (4)

where nj is outward normal vector; ij is the Cauchy stress tensor; bi is the body force; ui and ti are the displacement and
stress vectors on boundary  u and  t , respectively; is the density,  is the domain of the body.
Considering small deformation, the geometric equation is expressed as follows:
1 
i j = ui, j + u j,i . (5)
2
The constitutive equation is:

i j = Ci jkl kl , (6)

where ij , ij are the stress and strain vectors, respectively; Cijkl are components of the elastic tensor.

2.2. Displacement mode

Compared with classical nite element method, the displacement mode of XFEM takes more functions to describe the
models of the discontinuous region. Displacement discontinuity equation of crack elements is expressed demonstrated with
enriched functions, such as the Heaviside function H(x) [1]. Above the crack, H(x) is positively 1; while below the crack it is
negative 1:

1 ( x x ) en > 0
H (x ) = . (7)
1(x x ) en < 0

In Eq. (7), x means reference point; x represents a point on crack which is next to x; en is the outward normal vector of
crack element of x.
In order to capture the singularity and discontinuity near crack-tip, the crack-tip enriched functions are used. For an
isotropic elastic body, the crack-tip enriched functions usually use the branch crack-tip (x) (see [1] and references

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

4 L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116

therein), which is expressed as follows:


         
{ (r, )}4=1 = r sin 2 r cos 2 r sin 2 sin ( ) r cos 2 sin ( ) , (8)

where (r, ) is the crack-tip local coordinates in Eq. (8), and the rst term r sin( 2 ) describes the discontinuity of crack
surface, while the other three terms are employed to improve the singularity of crack-tip (see [1,16] and references therein)
In this way, XFEM displacement mode can be expressed in the following form:

   
4
uh ( x ) = Ni (x )ui + N j (x )H (x )a j + Nk (x ) (x )bk , (9)
iI jK kK =1

where Ni , Nj , Nk Ni , Nj and Nk are the shape functions for nodes i, j and k, respectively; I is the total node set in the mesh;
K and K are Heaviside Functions and crack-tip enriched node sets, respectively; ui are nodal displacement vectors of
different parts; aj and b k are two strengthen variables of nodes.
According to the above XFEM displacement mode, it is not convenient to impose Dirichlet boundary in the calculation of
an unknown quantity . To solve this problem, the shifted-basis enriched functions are proposed [17]:

     
4
uh ( x ) = Ni (x )ui + N j (x ) H (x ) a j + Nk (x ) ( (x ) )bk , (10)
iI jK kK =1

where H (x ) = H (x ) H (x j ); (x ) = (x ) (xk ).

2.3. The description of dynamic fracture

According to Hamiltons law of variational principle [17], the governing equation of the XFEM dynamics can be derived
as follows Firstly, Lagrange functional system is expressed as:

L = T U Wd Ws , (11)

where T, U, Wd and Ws represent the system kinetic energy, strain energy, damping energy and potential energy of external
force, respectively.
The kinetic energy T is:

1  T
T = d , (12)
 2

where and are displacement and velocity vector, respectively.


Strain energy U can be calculated by:
 
1 1
U= ( )T d = ( )T Cd, (13)
 2  2

where is strain vector, is the stress vector and C is the elastic matrix.
The damping potential energy Wd is:

1  T
Wd = c d , (14)
 2

where c is the viscous damping coecient.


Potential energy of external force Ws is expressed by:
 
Ws = ( )T Fb d + ( )T Fs d + ( )T F , (15)
 

where Fb , Fs and F are dened as enriched force, surface force and concentrated force vector, respectively.
According to Hamiltons principle, the conservative system is expressed by:
 t2
Ldt = 0. (16)
t1

Neglecting of damping, and combining Eqs. (11)(15) with the XFEM displacement mode substitution Eq. (16), the equa-
tion of dynamic XFEM is expressed as:

M + K = R, (17)

where = [ u a b ]T ; K is stiffness matrix; M is the mass matrix and R is load vector.

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116 5

Fig. 2. Scheme of edge-based smoothing nite element method.

2.4. Solution of the dynamic equation

The methods to solve dynamic equation can be classied as (1) implicit integration (Newmark method, Wilson-method,
etc.) and (2) explicit integration (center difference method, etc.) [18]. In this works Newmark method was adopted to resolve
Eq. (17). At the nth time step, discrete simulation equations can be expressed as:

  t 2
M + t K n = F K n1 + t n1 + (1 2 )
2 n1 , (18)
2

n = n1 + (1 )n1 + n t, (19)

 1  
n = n1 + n1 t + n1 + n t 2 , (20)
2
where t is the time increment; when 0.5 and 0.25( +0.5 )2 , the Newmark method is unconditionally stable. This
study is to analyze the dynamic fracture of crack under the shock load conditions, the effect of damping is therefore not
considered.

3. Edge-based smoothed extended nite element method (ES-XFEM) for crack problems

3.1. Smoothing domain of ES-FEM

Comparing with conventional nite element, the integral stiffness matrix of SFEM is based on the smoothing region,
rather than an element. Smoothing domain of ES-XFEM is based on edge of an element (as shown in Fig. 2). The entire
smoothing domain is expressed by  = (1 ) (2 ) (Ns ) , and (i) (j) = , i = j, where Ns represents the number
Ns edge. The ABCD ((k) ) and EFG ((m) ) are the smoothing domains based on edge BD ( (k) ) and EF ( (k) ), respectively.
According to the smoothing operation [11], the boundary strain i of the element i can be obtained from the compatible
strain (x) in the region of the smoothing domain (i) :
 
i = (x )i (x )d = s u(x )i (x )d. (21)
 (i )  (i )

Smoothing function i (x) meets:



i (x )d = 1. (22)
 (i )

The smoothing function can be expressed as follows:


 1
A (i )
x  (i )
i = , (23)
0 /  (i )
x

where A(i) is area of smoothing boundary domain (i) [24].


For smoothing area (i) , the stiffness matrix is:

(i ) T
k = bI CbJ d. (24)
 (i )

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

6 L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116

Under the bifurcation theory [24], the smoothed strain matrix bI of the node I in domain (i) can be expressed as:

b Ix 0
b Iy ,
BI = 0 (25)

b Iy b Ix
where

1
b Ih = (i ) NI (x )n(i ) (x )d (h = x, y ) , (26)
A
 (i )

and NI is I-shape function node; n(i) (x) is outward normal vector matrix of the boundary  (i) , which is represented as
follows:

nx(i ) 0
ny(i )
n (i ) ( x ) = 0 . (27)
(i ) (i )
ny nx

As linear displacement eld is used at  (i) in (i) , a Gaussian point is applied at  (i) . Eq. (26) can be further expressed
in the algebra form:

1 
M
(i ) (i )
b Ih = (i ) NI (xGP
j )n jh l j (h = x, y ) , (28)
A
j=1

where M is the total number of the boundary segment  (j i ) ; xGP


j
means the period of midpoint coordinates (Gauss point) in
the border  (j i ) ; n(jhi ) and l (j i ) are the outward normal vector and length of  (j i ) , respectively. Smoothing region of the stiffness
matrix can be further expressed as follows:
(i ) T
k = bI CbJ A(i ) . (29)

From Eq. (26), the stiffness matrix of the shape function does not require partial derivatives. The whole stiffness matrix
K assembling each region stiffness matrix has been set, which is:
(i )

Ne
(i )
K= k , (30)
i=1

where Ne(i ) is the smoothing area.

3.2. Equations of ES-XFEM

ES-XFEM is derived from Eq. (17), the assembled mass matrix and load vector for ES-XFEM do not involve strain, therefore
dynamic extension of nite element method (FEM) is needed.
Similar to XFEM, the smoothed strain of the displacement mode of ES-XFEM in the smoothing domain can be written
by:

  
4
k = Bui ui + Baj a j + Bb k b k , (31)
iI jK kK =1

where I means all nodes of the set k in the smoothing zone; K and K mean enriched node set by Heaviside function
and crack-tip function in a smoothing domain, respectively; Bui , Baj and Bb k denote boundary smoothed strain matrix in
conventional smoothing domain, Heaviside function and enriched smoothed strain by crack-tip function, respectively [14].
To enrich the matrix, its specic expressions are as follows:

brix 0
0 briy
Bi =
r
r = u, a, b, (32)
briy brix

where

1
buih = n (x )Ni (x )d , (33a)
Ask ks h

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116 7


1
baih = n (x )Ni (x )H (x )d , (33b)
Ask ks h

1
bbih = s n (x )Ni (x ) (x )d = 1, 2, 3, 4, (33c)
Ak ks h

where Ask is the number k smoothing domain, h is separated into x, y.


Eq. (33) can be presented in numerical forms:
 
Nseg Ngau
1  
buih = s wm,n Ni (xm,n )nh (xm,n ) , (34a)
Ak
m=1 n=1
 
Nseg Ngau
1  
baih = s wm,m Ni (xm,n )H (xm,m )nh (xm,n ) , (34b)
Ak
m=1 n=1
 
Nseg Ngau
1  
bbih = s wm,m Ni (xm,n ) (xm,m )nh (xm,n ) , (34c)
Ak
m=1 n=1

where Nseg is the boundary number of the smoothing domain; Ngau represents Gaussian integral point on the boundary.
The stiffness matrix ki j is derived from the smoothed strain matrix:
  u T u   u T a   u T b
sk Bi CB j d sk Bi CB j d sk Bi CB j d
  T u   a T   a T
ki j = s Bai CB j d CB j d
a u
k sk Bi CB j d sk Bi . (35)
  b T u   b T a   u T u
s Bi CB j d
k
sk Bi CB j d  sk Bi CB j d 
Due to smoothed strain matrix in smoothing domain is constant, Eq. (35) can be written as,
 T u  u T a  u T b

u
Bi CB j Ask Bi CB j Ask Bi CB j Ask
 T  a T  a T
ki j = Bai CB j Ask
u a b
CB j Ask Bi CB j Ask Bi . (36)
 b T u  b T a  b T b
Bi CB j Ask Bi CB j Ask Bi CB j Ask
Using me to express the element mass matrix set:

muu
ij
mua
ij
mub
ij

mei j = mau
ij
maa
ij
mau
i j , (37)

mbu
ij
mba
ij
mbb
ij

where

muu
ij = (Ni )T N j d, (38a)
e
  
mua
ij = (Ni )T H N j d, (38b)
 e

  
mub
ij = (Ni )T N j d, ( = 1, 2, 3, 4 ) , (38c)
 e

  T  
maa
ij = H Ni H Ni d, (38d)
e
  T  
mab
ij = H Ni N j d, ( = 1, 2, 3, 4 ) , (38e)
e
  
mbb
ij = ( Ni )T N j d, ( = 1, 2, 3, 4 ) , (38f)
e
Using rie to express the element nodal load vector components:
 
rei = rui rai rb1
i
rb2
i
rb3
i
rb4
i
, (39)

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

8 L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116

Fig. 3. Schematic diagram of edge-based smoothed extended nite element method.

Fig. 4. Partition of split smoothing domains in the ES-XFEM [14].

where
u  
ri = e Ni Fb d +  Ni Fs d + Ni F
ra = e Ni H Fb d +  Ni H Fs d + Ni H F . (40)
ib  
ri = e Ni Fb d +  Ni Fs d + Ni F ( = 1, 2, 3, 4)

4. Integration implementation of ES-XFEM

Because the stiffness matrix ki j is derived from integration on the smoothing boundary, the conventional XFEM integral
solution is no longer applicable. A new integral formula has been proposed in [9], which is expressed as follows:

(1) As shown in Fig. 3, each boundary employs a Gaussian point in the conventional smoothing domain (such as ABCD).
(2) Each boundary employs a Gaussian point in crack-free enhanced Heaviside function domain.
(3) Each boundary adopts 35 Gaussian points in crack-free and crack-tip enriched smoothing domain.
(4) In the smoothing domain where the crack run through, it can be divided into two smoothing subdomain and one
Gaussian point is used on the each smoothing boundary; As shown in Fig. 4, irregular smoothing subdomain is usually
divided into a triangle sub-domain in order to facilitate programming.

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116 9

Fig. 5. Partition of crack-tip smoothing domains in the ES-XFEM.

Fig. 6. Scheme of Gaussian point in split and crack-tip element.

(5) As shown in Fig. 5, crack is extended to smoothing boundary of the region and then divides the crack-tip domain
into multiple sub-domains, generally three to ve ones. The boundary of the smoothing subdomain near the crack-tip
needs 5 Gaussian points, and other boundary needs 13 points.
(6) As shown in Fig. 6, in the integral of enriched function, the partition rule of smoothing sub-domain is same as above,
but Gaussian points are not taken on the smoothing sub-domain boundary and while adopt the interior points of
the smoothing sub-domain. This is because if the smoothing domain on the boundary points is taken as a Gaussian
point, boundary AB, CD, EF, GH are coincidence with crack surface and then cannot satisfy the accuracy of calculation.
In general, the crack-tip items of enriched function utilize about 35 Gaussian points, otherwise enhanced Heaviside
takes one Gaussian point. Sub-domain integration method is employed to calculate both Mass matrix and stiffness
matrix in conventional XFEM. It is needed to employ one Gaussian point in the internal unit of unenhanced nodes,
57 points in crack-tip sub-domains and 3 points in other domains.

5. Dynamic stress intensity factor

Stress intensity factor can be calculated by displacement extrapolation method [19], virtual crack extension method [20],
virtual crack closure legal [21], and interaction integral method [22]. In this work, interaction integral method has been
used. Considering the two dimensional crack, crack is described as a line, while a local coordinate is set up at crack-tip. As
shown in Fig. 7,  0 means a contour around the crack-tip.

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

10 L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116

Fig. 7. Scheme of J-integral domain.

Contour integrals J near the crack-tip can be expressed as:


     
J = lim W 1i i j u j,1 ni d = lim W 1i i j u j,1 mi d , (43)
 0 0  0  0 0  0

where W = 12 i j i j , which is the strain energy density [18].


Eq. (43) cannot be directly applied to numerical calculation and then adopted by a closed domain  s =  1 +  + +  +
 0 . The set weight factor q is 1 in  0 , and is 0 in  1 , Eq. (43) can be rewritten as:
 
J= i j u j,1 W 1i qmi d + 2 j u j,1 d . (44)
s  + +
Using the divergence theorem, the integral form of the J integral is,
 
J= i j u j,1 W 1i q,i dA+ i j u j,1i W,1 qdA. (45)
A A

By incorporating the balance equations and geometric equation:


 
J= i j u j,1 W 1i q,i dA+ uu j,1 ui u ,1 qdA. (46)
A A

Choose two independent equilibrium states: state 1 (i(j1 ) , i(j1 ) , ui(1 ) ) as the real state, state 2 (i(j2 ) , i(j2 ) , ui(2 ) ) as the
auxiliary condition. The third equilibrium is achieved by linear superposition of the two equilibrium, with J integral can be
rewritten by J(1, 2) :
J ( 1,2 ) = J ( 1 ) + J ( 2 ) + I ( 1,2 ) , (47)
where I (1,2) means the interaction integral of state 1 and state 2:
   
I ( 1,2 ) = (1) u(j,21 ) + (2) u(j,11) W (1,2) 1i q,i dA + u(1) u(j,21) qdA, (48)
A A

where W (1,2) = i(j2 ) i(j1 ) = i(j1 ) i(j2 ) is the interaction of strain energy density.
The two-dimensional equation of energy release rate can be expressed as
KI2 K2
J= + II , (49)
E E
with

E planestress
E = E .
1 v 2
planestrain

For the third equilibrium, Eq. (49) can be reformed as follows:


 1 2 
2 KI( ) KII( ) + KI( ) KII( )
2 1
J ( 1,2 ) = J (1 ) + J (2 ) + . (50)
E
Comparing with Eqs. (48) and (50), it is expressed as:
 1 2 
2 KI( ) KII( ) + KI( ) KII( )
2 1
I ( 1,2 ) = . (51)
E

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116 11

Fig. 8. Geometry of a semi-innite crack.

State 2 is chosen as asymptotic eld of mode I and II, then:


E
KI( ) = I (1, mod eI) KII( ) = E2 I (1, mod eII) .
1 1
(52)
2
As shown in Fig. 7(b), the interaction integrals are based on the smoothing boundary, no longer based on the element
integrals for ES-XFEM, so Eq. (49) is rewritten as:
ef f  ef f 
Nd Nd
   
I ( 1,2 ) = i(j1) u(j,21) + i(j2) u(j,11) i(j1) i(j2) 1i q,i dA + u(1) u(j,21) qdA, (53)
m m
m=1 Ae f f m=1 Ae f f

where Amef f
is area of the number m element in the equivalent integral domain Nedf f .
In Eq. (53), the rst part of the right side is related to the displacement derivative which is based on integral of the
smoothing domain; the second item of right side is independent of the displacement derivative term, which is based on the
element integral. Eq. (53) is depicted as:
Nd

ef f 3 
   ef f 
Nd

(1 ) (2 ) (2 ) (1 ) (1 ) (2 )
I ( 1,2 ) = i j u j,1 + i j u j,1 i j i j 1i q,i dA + u(1) u(j,21) qdA, (54)
Asn Am
m=1 n=1 m=1 ef f

where Asn is area of the mth element in the nth smoothing edge.

6. Numerical examples

6.1. Example 1: dynamic fracture of edge crack in the semi-innite domain

Because of computational limits, a limited domain is modeled for analysis, as shown in Fig. 8. The size of the limited
plate are L = 10 m, H = 2 m, crack length a = 5 m; the elastic constants of materials are: E = 210 GPa, v = 0.3, = 80 0 0 kg/m3 .
0 =500 MPa. The top of the plate is subject to a distributed step load 0 (t) = 0 H(t). When t > 0, H(t) = 1. Computation
time is 1 103 s, time step is 2 105 s and the number of iterations is 50 times. The regular meshes use 65 25 triangular
elements, as shown in Fig. 9(a). In order to conrm rough requirement of mesh in the smoothed expansion nite element
method, different irregular meshes are adopted. The irregular mesh is generated randomly by [8]:

x = x + x rd ir
, (55)
y = y + y rd ir
where x, y respectively represents node coordinates of initial rules mesh, as shown in Fig. 9(a); Accordingly, x, y is the
length of element respectively in the x, y direction; rd is random numbers in the domain[1, 1]; ir is irregular factors.
Regular mesh (ir = 0.00) and irregular mesh (ir = 0.12) are shown is shown in Fig. 9.
Since reection occurs when the stress wave meets the crack-tip, calculation process does not consider the im-
 of reection, and the dynamic stress intensity factor analytic formula is valid just as time t 3tc (tc = H/cd , cd =
pact
E (1 v )/((1 + v )(1 2v ) )).
Regardless of the crack propagation, dynamic stress intensity factor analytic is expressed as [23]:

2 0 cd t ( 1 2 v )
KIdyn (t ) = . (56)
1v
The relative error of normalized dynamic stress intensity factor is dened as follows:
 
 Kdyn Kexact 
err =   100%,
 (57)
Kexact
where Kdyn and Kexact are normalized numerical solution and analytical solution of the dynamic stress intensity factor,
respectively.
In the following simulation gures, KI /(0 H1/2 ), t/tc are the dynamic stress intensity factor of the normalized mode I
dyn

and the normalized time which the dimension is 1, respectively.

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

12 L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116

Fig. 9. Computational meshes.

Fig. 10. (a) Normalized dynamic stress intensity factor, (b) relative error of dynamic stress intensity factor at various meshes.

6.1.1. Convergence analysis


It is researched on convergence of different computational meshes. As can be seen from Fig. 10., when t 2.0tc , relative
error of dynamic stress intensity factor by 43 19 meshes decreased better than 34 14 meshes. 53 21 meshes and 65 25
meshes have similar performance with convergence. The following analysis is based on 65 25 meshes.

6.1.2. Selection of crack-tip smoothing subdomain


As the crack-tip region is divided into different number of smoothing subdomain and the inuence of subdomains
number on the dynamic stress intensity factor is analyzed. As can be seen from Fig. 11(a), when number of smoothing sub-
domain nssd 2, the difference of dynamic stress intensity factor in normalization mode I is not obvious. However, as shown
in Fig. 11(b), when nssd 3, relative error of the normalized dynamic stress intensity factor in mode I tends to be stable.
Therefore, in order to improve the computation eciency and guarantee the calculation accuracy, in the crack-tip smoothing
domain sequence number should be used as nssd = 3 and employs ve Gaussian points at each boundary. The smoothing
domain with crack across do not need partition with one Gaussian point is taken; other area also take one Gaussian point.

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116 13

Fig. 11. (a) Normalized dynamic stress intensity factor, (b) relative error of dynamic stress intensity factor with different smoothing subdomain.

Fig. 12. (a) Normalized dynamic stress intensity factor, (b) relative error of dynamic stress intensity factor comparing with different method.

6.1.3. Accuracy and mesh distortion


As shown in Fig. 12, it is compared the ES-XFEM results with the analytical solution, nite element method of singular
boundary smoothing [24] (sES-FEM) and extended nite element method (XFEM). It is demonstrated that when t 1.5tc ,
numerical error of three methods are all relative large; when t > 1.5tc , the results of three kinds of numerical methods are
very close to the analytical solution; when t = tc , the error of the numerical solution is the largest and the maximum error
is XFEM.
With classical nite element method, mesh distortion may affect the computational accuracy. ES-XFEM is no longer re-
stricted, while Fig. 13 shows the calculation results of different mesh irregular factors. It is illustrated that the ES-XFEM
results affected by the quality of the mesh are very small. When the irregular factor takes 0.12, the mesh is extremely
irregular. There are 74 angles greater than 120o , 372 angles less than 30o and 39 angles less than 20o (as shown in
Fig. 9(b)). In this case, the ES-XFEM can still get a stable result as the time 1.8tc t 3.0tc , the maximum relative er-
ror of the dynamic stress intensity factor in the normalized mode I is less than 3%.

6.2. Example 2: dynamic fracture of the bevel edge crack plate

In order to prove the eciency and stability of ES-XFEM for dynamic fracture problems, we further investigative the
mixed-mode dynamic fracture. An inline edge crack of plate is shown in Fig. 14, its geometry sizes are L = 44 mm, H =
32 mm, D = 16 mm, crack length is a = 22.63 mm, and the incline angle of crack = 45. Material parameters are set as:
modulus of elasticity E = 29.4 GPa, Poissons ratio v = 0.286, the density = 2450 kg/m3 . The left, top and bottom of plate
bear restriction of chain support in normal direction, the right end bears a step load distribution 0 (t) = 0 H(t). Computa-
tion time is 3 105 s, time step is 6 107 s and the number of iterations is 50 times. Two kinds of computational meshes
(66 48) with different quality (regular and irregular mesh) are revealed in Fig. 15.
1/2 1/2
Calculation results in the gure below shows that, KI /(0 ( a ) ), KII /(0 ( a ) ) are the normalized dynamic stress
dyn dyn

intensity factor of mode I and II, respectively.

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

14 L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116

Fig. 13. (a) Normalized dynamic stress intensity factor, (b) relative error of dynamic stress intensity factor with different irregular factor.

Fig. 14. Geometry of a rectangular plate with an inclined crack.

Fig. 15. Computational meshes.

In Fig. 16, analyzes and compares the normalization of the dynamic stress intensity factor under different numerical
methods with the regular mesh and the results of three methods are consistent and the peak value of dynamic stress
intensity factor in normalized mode I is higher than that in normalized mode II.
Fig. 14 presents ES-XFEM results under different mesh distortion. When the irregular factor takes 0.2, maximum of the
element angle is 154.3, minimum is 10.6. There are 172 elements with maximum angles greater than 120, 485 elements
with the minimum angles less than 30 and 30 elements with the minimum angles less than 20. As shown in Fig. 17, mesh
distortion effected on the dynamic stress intensity factor is small. It is demonstrated again that ES-XFEM is not sensitive to
the mesh quality, which will help to analyze large deformation of crack in propagation process.

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116 15

Fig. 16. Comparison of the normalized mixed-mode DSIFs for a rectangular plate with a slanted edge crack.

Fig. 17. The normalized mixed-mode DSIFs for a rectangular plate with a slanted edge crack under different mesh irregular factor.

7. Conclusions

This paper applies the edge based smoothed extended nite element (ES-XFEM) to solve stationary dynamic fracture
problems subjected to impact loads. Dynamic stress intensity factor DSIF is computed with the domain-form of the inter-
action integral in terms of the strain smoothing method. Two case studies including the mode I and mixed crack prove
that the dynamic ES-XFEM is highly accurate, and insensitive to mesh distortion. It is also proved that when smoothing
subdomain nssd 3, the relative error of DSIF become stable.

Acknowledgments

This work was supported by the Graduate Students innovative research projects in hunan province (No. CX2016B106)
and National Science Foundation of China (NSFC) (No. 51078249).

References

[1] X.D. Zhang, T.Q. Bui, A ctitious crack XFEM with two new solution algorithms for cohesive crack growth modeling in concrete structures, Eng. Comput.
32 (2015) 473497.
[2] N.T. Nguyen, T.Q. Bui, Ch. Zhang, T.T. Truong, Crack growth modeling in elastic solids by the extended meshfree Galerkin radial interpolation method,
Eng. Anal. Bound. Elem. 44 (2014) 8797.
[3] S. Tanaka, H. Suzuki, S. Sadamoto, M. Imachi, T.Q. Bui, Analysis of cracked shear deformable plates by an effective meshfree plate formulation, Eng.
Fract. Mech. 144 (2015) 142157.
[4] T.Q. Bui, M.N. Nguyen, A moving Kriging interpolation-based meshfree method for free vibration analysis of Kirchhoff plates, Comput. Struct. 89 (2011)
380394.
[5] S. Osher, J.A. Sethian, Fronts propagating with curvature-dependent speed: algorithms based on HamiltonJacobi formulations, J. Comput. Phys. 79
(1988) 1249.
[6] C.A. Duarte, J.T. Oden, An h-p adaptive method using clouds, Comput. Methods Appl. Mech. Eng. 139 (1996) 237262.
[7] J.-S. Chen, C.-T. Wu, S. Yoon, Y. You, A stabilized conforming nodal integration for Galerkin mesh-free methods, Int. J. Numer. Methods Eng. 50 (2001)
435466.

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027
JID: APM
ARTICLE IN PRESS [m3Gsc;June 1, 2016;15:55]

16 L. Wu et al. / Applied Mathematical Modelling 000 (2016) 116

[8] G.R. Liu, A generalized gradient smoothed technique and the smoothed bilinear form for Galerkin formulation of a wide class of computational
methods, Int. J. Comput. Methods 05 (2008) 199.
[9] Z.C. He, G.Y. Zhang, L. Deng, Eric Li, G.R. Liu, Topology optimization using node-based smoothed nite element method, Int. J. Appl. Mech. 07 (2015)
1550085.
[10] N. Nourbakhshnia, G.R. Liu, Fatigue analysis using the singular ES-FEM, Int. J. Fatigue 40 (2012) 105111.
[11] G.R. Liu, G.Y. Zhang, A normed G space and weakened weak (W2) formulation of a cell-based smoothed point interpolation method, Int. J. Comput.
Meth. 6 (2011) 147179.
[12] T. Nguyen-Thoi, G. Liu, K. Lam, G. Zhang, A face-based smoothed nite element method (FS-FEM) for 3D linear and geometrically non-linear solid
mechanics problems using 4-node tetrahedral elements, Int. J. Numer. Methods Eng. 78 (2009) 324.
[13] S.P. Bordas, T. Rabczuk, N.-X. Hung, V.P. Nguyen, S. Natarajan, T. Bog, et al., Strain smoothing in FEM and XFEM, Comput. Struct. 88 (2010) 14191443.
[14] L. Chen, T. Rabczuk, S.P.A. Bordas, G. Liu, K. Zeng, P. Kerfriden, Extended nite element method with edge-based strain smoothing (ESm-XFEM) for
linear elastic crack growth, Comput. Methods Appl. Mech. Eng. 209 (2012) 250265.
[15] Y. Jiang, T. Tay, L. Chen, X. Sun, An edge-based smoothed XFEM for fracture in composite materials, Int. J. Fract. 179 (2013) 179199.
[16] T.Q. Bui, Ch. Zhang, Extended nite element simulation of stationary dynamic cracks in piezoelectric solids under impact loading, Comput. Mater. Sci.
62 (2013) 243257.
[17] P. Stapr
, Application of XFEM with shifted-basis approximation to computation of stress intensity factors, Arch. Mech. Eng. 58 (2011) 447483.
[18] T.J. Hughes, The Finite Element Method: Linear Static And Dynamic Finite Element Analysis, Courier Corporation, 2012.
[19] C. Shih, H.D. Lorenzi, M. German, Crack extension modeling with singular quadratic isoparametric elements, Int. J. Fract. 12 (1976) 647651.
[20] D.M. Parks, A stiffness derivative nite element technique for determination of crack tip stress intensity factors, Int. J. Fract. 10 (1974) 487502.
[21] E.F. Rybicki, M. Kanninen, A nite element calculation of stress intensity factors by a modied crack closure integral, Eng. Fract. Mech. 9 (1977)
931938.
[22] Z. Kang, T.Q. Bui, D.D. Nguyen, T. Saitoh, S. Hirose, An extended consecutive-interpolation quadrilateral element (XCQ4) applied to linear elastic fracture
mechanics, Acta Mech. 226 (2015) 39914015.
[23] T. Menouillard, T. Belytschko, Dynamic fracture with meshfree enriched XFEM, Acta Mech. 213 (2010) 5369.
[24] P. Liu, T. Bui, C. Zhang, T. Yu, G. Liu, M. Golub, The singular edge-based smoothed nite element method for stationary dynamic crack problems in 2D
elastic solids, Comput. Methods Appl. Mech. Eng. 233 (2012) 6880.

Please cite this article as: L. Wu et al., Edge-based smoothed extended nite element method for dynamic fracture analysis,
Applied Mathematical Modelling (2016), http://dx.doi.org/10.1016/j.apm.2016.05.027

También podría gustarte