Accepted Manuscript

Stochastic investigation of the facture problem in functionally graded materials

with uncertain mechanical properties and an arbitrarily oriented crack

Haizhu Pan, Tianshu Song

PII: S0167-8442(17)30103-9
DOI: http://dx.doi.org/10.1016/j.tafmec.2017.07.002
Reference: TAFMEC 1907

To appear in: Theoretical and Applied Fracture Mechanics

Received Date: 1 March 2017

Revised Date: 28 June 2017
Accepted Date: 3 July 2017

Please cite this article as: H. Pan, T. Song, Stochastic investigation of the facture problem in functionally graded
materials with uncertain mechanical properties and an arbitrarily oriented crack, Theoretical and Applied Fracture
Mechanics (2017), doi: http://dx.doi.org/10.1016/j.tafmec.2017.07.002

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 Stochastic investigation of the facture problem in
functionally graded materials with uncertain mechanical
properties and an arbitrarily oriented crack
Haizhu Pan a, b, Tianshu Song a, 

a
College of Aerospace and Civil Engineering, Harbin Engineering University, Harbin
150001, PR China
b
College of Computer and Control Engineering, Qiqihar University, Qiqihar 161006,
PR China

Abstract: The macro and micro analysis methods of the fracture mechanics are
adopted to investigate the fracture mechanics problem of functionally graded
materials (FGMs) with uncertain mechanical properties and an arbitrarily oriented
crack. In this paper, the piecewise-exponential model and the Mori-Tanaka model are
combined to solve the mixed-mode crack problem for every sample analytically. Then
according to the probability theory, the mixed-mode SIFs for the samples are obtained.
So the probabilistic attributes of the mixed-mode SIFs in FGMs can be obtained. As a
result, the probabilistic density function of SIFs can be provided by means of the
graphic method. The method based on macro and micro mechanics can be efficiently
applied to carry out the related research of the influences of the crack angles and
lengths on the stochastic characteristics of the mixed-mode SIFs.
Keywords：Arbitrarily oriented crack; Functionally graded materials; Stochastic
characteristics
1 Introduction
Functionally graded materials (FGMs) are two- or multi-phase particulate
composites in which material composition and microstructure vary spatially in the
macroscopic length scale to meet a desired functional performance. This is achieved
by gradually varying volume fraction of the constituent materials, which can be
created by controlling over the manufacturing. This characteristic inevitably creates
uncertainties in the effective material properties of a FGM [1]. The randomness in
material properties and the random characteristics in fracture problems of FGMs have
attracted great attention from the international research community.
During the last few decades, the mechanical properties of FGMs are usually
considered as deterministic values in various theoretical and computational studies on
fracture behavior of FGMs. Many published papers assumed the effective properties
to be particular functions (e.g. exponential functions) [2-5]. Guo et al. [2,3] proposed
the early analytical models for static and dynamic crack problems of orthotropic
FGMs and FGM coating-substrate structures. Gu and Asaro [4] analyzed a


Corresponding author.
E-mail address: pan_haizhu@163.com (Haizhu Pan), 593552258@qq.com (Tianshu Song)
semi-infinite crack in a strip of an isotropic FGM under edge loading and in-plane
deformation conditions. Guo and Noda [5] gave the dynamic analytical solution of a
functionally graded layered structure with a crack crossing the interface and depicted
the variation of the dynamic SIFs with the nonhomogeneous parameters when the
crack moves from one layer into another layer. Their work overcame basic fracture
problems of FGMs with general mechanical properties. In the past decades, most of
the analytical models for the crack problems of FGMs assumed the mechanical
properties to be very particular functions (mainly exponential functions) so that the
crack problems can be solved analytically. However, this assumption may not
coincide with practice. Having considered great difference from the previous
analytical models, Guo and his coauthors [6-8] proposed the piecewise-exponential
model (PE model) to solve the static and dynamic crack problems of FGMs with
general mechanical properties, while the PE model is based on the analytical model of
a single layer with exponential properties. Hundreds of documents based on analytical
models were published in the past decades on the crack problems of FGMs. The
analytical models can be unified and proved to be significant with the idea of the PE
model as they form a strong base for establishing a general analytical fracture model
of FGMs with arbitrary mechanical properties. Guo and Noda [9] proposed an
analytical thought combining with the PE model and a perturbation method to
investigate the thermal shock crack problems of FGMs with general
thermomechanical properties. Pan et al. [10,11] and Wang et al. [12,13] expanded the
PE model to the fracture problems of FGMs with general thermomechanical or
viscoelastic properties and collinear cracks. Chalivendra [14] developed a quasi-static
mixed-mode stress fields for a crack in orthotropic inhomogeneous medium. However,
there is substantial plenty of randomness in the composition of FGMs. Many
investigations do not provide any measures of the stochastic behavior of FGMs [15].
To enable a better understanding and characterisation of the actual behaviour of
FGMs, it is obviously of prime importance that the inherent randomness in system
parameters be incorporated in the analysis [16]. The phase volume fractions in
practical FGMs show random fluctuations around the average trend, which leads to
stochastic features of the mechanical properties of FGMs. Thus, the corresponding
fracture analyses of FGMs with stochastic mechanical properties will become very
complex [17]. Unfortunately, relatively little efforts have been made in the past by the
researchers and investigators on research in analytical fracture problems of FGMs
with stochastic properties. Shahabian and Hosseini [1] studied the dynamic response
of a functionally graded cylinder subjected to mechanical shock loading with
randomness in material properties by using the Monte Carlo simulation. Second-order
statistics of the static response-mean and variance of transverse deflection of
functionally graded plates with multiple randomness in material properties, volume
fraction index and lateral loading were obtained by Yang et al. [16]. A stochastic
fracture mechanics model was proposed for predicting probabilistic characteristics of
SIF for a crack problem of FGMs by Guo et al. [17]. Ferrante and Graham-Brady [18]
presented three multiscale models for fracture analysis of a crack in a two-phase,
functionally graded composite. The statistics of temperature and thermal stress were
analytically obtained in FGM plates with uncertainties in the thermal conductivity and
coefficient of linear thermal expansion by Chiba and Sugano [19]. Besides, some
researchers have given attention to the uncertain properties of FGMs. Lal et al. [20]
provided a probabilistic tool for incorporating and handling the structural material
uncertainties in the analysis of the structures. Hosseini and Shahabian [21]
stochastically studied the transient thermo-elastic waves and coupled thermo-elasticity
without energy dissipation based on Green–Naghdi theory in functionally graded
thick hollow cylinder with Gaussian uncertainty in constitutive mechanical properties.
And the stochastic hybrid method presented in Ref. [22] can be considered as an
innovation for stochastic analysis with large numbers of samples. In Ref. [23], the
material properties of FGMs and piezo-electric materials with volume fraction
exponent were modeled as independent random variables. A numerical example
involving an edge-cracked functionally graded specimen under a mixed-mode
deformation was analyzed by the various multiscale and microscale models examined
in Ref. [24]. In Refs. [25-27] new interaction integral methods were developed for
solving the static and dynamic crack problems of nonhomogeneous materials with
complex interfaces. Their contribution lies in: the domain-independence of the new
interaction integral methods still stands in nonhomogeneous materials even when the
integral domain intersects the interfaces under thermal loading or dynamic loading. It
will provide significant support for the crack propagation simulation of FGMs or
nonhomogeneous materials with complex interfaces. Then the new interaction integral
methods were expanded to the nanoscale crack propagation problems [28,29]. Apart
from the SIFs, the T-stress is another key parameter in fracture mechanics because it
affects the crack growth direction, crack-tip constraint and fracture toughness greatly.
A domain-independent interaction integral for extracting the T-stress was proposed for
nonhomogeneous materials with interfaces [30]. Then it is set in the extended finite
element method (XFEM) so that the T-stress can be solved with high accuracy and
efficiency. If the damage and crack propagation in nonhomogeneous materials need to
considered, the damage evolution method [31] can be applied.
In previous investigations, only a few of them analyzed the crack problems of
FGMs considering the stochastic mechanical properties. The analytical model for
these problems was developed by Guo et al. [17]. But only mode-I crack problem was
considered in it. Up to now, no analytical solution of the crack problems aims to the
both uncertain mechanical properties and the arbitrarily oriented crack. Usually the
direction of the crack in FGMs is assumed to be particular, such as being parallel to
the variation direction of the material property. In this paper, taking into account
microstructural details of uncertain properties, a macro analysis method of the fracture
mechanics will be applied to the fracture mechanics problem of FGMs with uncertain
mechanical properties and an arbitrarily oriented crack. And the probabilistic
characteristics of the mixed-mode SIFs of FGMs with stochastic mechanical
properties will be studied analytically. Especially the influences of the different angles
and lengths of the crack on the randomness of the mixed-mode SIFs are discussed.
2 Stochastic micromechanics description
2.1 Problem definition
 Consider a two-phase FGM body with domain D  3 and a schematic illustration
of its microstructure, as shown in Fig. 1. It illustrates three disjoint FGM regions in
3
the macroscopic length scale (x) with subdomains D1, D2 and D3，where D  i 1
Di .

The microstructure includes two distinct material phases: phase A (black), phase B
(white), either of which represents an isotropic and linear-elastic material. As shown
in Fig.1, the particle–matrix region D1 comprises particles from phase B material
(white) embedded in the matrix material which is phase A (black). The particle–matrix
role reverses in region D3, where particles and matrix are phase A (black) and phase B
(white) material, respectively. In the transition region D2, there are forming
intertwined clusters of particle and matrix materials.
A representative volume element at an arbitrary point x( x  D  3 ) characterizes
material heterogeneity in the microscopic length scale [15,17]. The volume fractions
of particle and matrix materials are respectively denoted by  p ( x ) and m ( x ) , each

of which is constrained in the scope of 0 to 1 and meets therestrictive condition

p ( x)  m ( x)  1 (1)

The volume fractions are stochastic. This implies that  p ( x ) and m ( x ) both are

random variables. According to Eq. (1), m ( x ) is only dependent to  p ( x ) . Thus, only

the probabilistic characteristics of  p ( x ) should be considered. The mean  ( x ) , p

the standard deviation   ( x ) and the distribution function F ( z, x ) of  p ( x ) can

p p

be given by
 ( x )   tf (t, x)dt (2)
p 

[ p (x)]2 =  [t  p ( x )]2 f (t, x )dt (3)



 
z
Fp ( z, x )  p p ( x )  z   f (t , x )dt (4)


where f (t, x) expresses the probability density function of  p ( x ) .

2.2 Probabilistic description of FGMs properties by random variables

A two-dissimilar FGM strip with an arbitrarily oriented crack will be considered at
this work as shown in Fig.1. The material’s physical behavior varies along the
direction of the x axis. The effective shear modulus G(x) is related to  p ( x ) , thus G(x)

is also a random variable. So G(x) can be expressed as

G( x) H [ G G , p ( x ) ]
p , m (5)

where Gp and Gm represent shear moduli of the particle and matrix constituents,
respectively. Once the particle and matrix materials are selected, Gp and Gm will be
determined. Thus G(x) is only related to  p ( x ) . So we have

G( x)  H [p ( x)] (6)

Furthermore, Employed to probability theory, the distribution function of G(x) can be

obtained as
F ( z, x )  p H [p ( x)]  z (7)

Similarly, the probability density function, the mean and the standard deviation of G(x)
can all be solved by the related expression of  p ( x ) .

In the following examples, the volume fraction  p ( x ) is assumed to be a

one-dimensional Beta random field adopted in Refs. [15,17]. It can be expressed as

 1
 t 1 1 (1  t )2 1 ,0  t  1
f (t , x )   B ( ,
1 2 ) (8)
0，
 otherwise

where 1 and  2 refer distribution parameters, respectively, and

B(1, 2 )  (1 )(2 ) / (1  2 ) (9)


( )  e x p(  1)d  (10)
0

Mean p ( x ) and standard deviation  p ( x ) can be written as

 ( x )   j 0 aij x j
5
p
(11)

  ( x )   j 0 bij x j
5
p
(12)

where aij and bij indicate polynomial coefficients, respectively. p ( x ) and

  ( x ) can be directly given by the following expression [15,17]

p

 ( x)  0.109( x / l )  4.25( x / l )2  9.762( x / l )3  8.629( x / l )4  2.748( x / l )5 (13)

p

  ( x)    ( x)
p 0
and   ( x)  0.178( x / l )  0.309( x / l )2 +0.155( x / l )3
0
(14)

While 1 and  2 can be solved respectively based on Eqs. (2), (3), (13) and (14).

Employed to the Mori-Tanaka Model, the effective shear modulus G( x ) of FGM

can be given by
 G( x)  Gm  (Gp  Gm )p ( x) / {1  4[1  p ( x)]Gp (Gp  Gm )} (15)

where
Gp  3(2Gm  km ) / [10Gm (4Gm  3km )] (16)

Here k m refers the bulk modulus of the matrix. Combining Eq. (15) with Eq. (6)，we

obtain
H [p ( x)]=Gm  (Gp  Gm )p ( x) / {1  4[1  p ( x)]Gp (Gp  Gm )} (17)

Substitution of Eq. (17) into Eq. (7), the distribution function F ( z, x) of G( x ) can be

given by


F ( z, x )  p Gm  (G p  Gm )p ( x ) / {1  4[(1  p ( x)]G p (G p  Gm )}  z  (18)

So the probability density function, the mean and the variance of the effective shear
modulus G( x ) can expressed as follows

H 1 ( z ) 1
F ( z, x )   t 1 ( x )1 (1  t )2 ( x )1 (19)
 B(1 ( x ), 2 ( x ))

 1 (G  Gm )1 ( x )1[1  4G p (G p  Gm )]1 ( x ) (G p  G )2 ( x )1

 , Gm  G  G p
g (G, x )   B[1 ( x ), 2 ( x )] (G p  Gm )1 ( x )2 ( x )1[1  4G p (G  Gm )]1 ( x )2 ( x )

0, else
(20)

G ( x)   g (G, x)GdG （21）



[ G ( x)]2   (G  G ) g (G, x)dG (22)


Eq.(7) can be written by

F ( z, x)  p G( x)  z (23)

According to Eq.(23), we have

z  F 1 ( p, x ) （24）

where p expresses the probability of G( x)  z , while F 1 ( p, x) implies the inverse

function of F ( z, x) . Choice on the suitable samples G M ( x)( M  1,2,...) may be

performed from appropriate samples zM of z, while the appropriate samples zM can be

determined by selecting appropriate samples pM of p. Let
M
pM  , M  1,2,..., N (25)
N
where N indicates the sample number of G(x). The sample number should be large
enough so as to maintain the convergence of the probability density function of the
SIFs for obtaining satisfied results of the SIFs. Referred to other document of
different sample number (100, 125, 150 and 200) by Guo et al. [17], the results
display satisfied probabilistic characteristics on the SIFs when the sample size is no
less than 100.
3 A macro fracture mechanics model for an arbitrarily oriented crack of FGMs
with stochastic properties
A cracked FGM strip with stochastic mechanical properties varying along x-axis is
shown in Fig. 2. And the coordinate systems, (x-y) and (x0 -y0) are used as shown in
Fig.2. The thickness of the infinitely long FGM strip is l. There is an arbitrarily
oriented crack with length c and angle θ in the FGM strip. a and b denote
x0-coordinates of both crack-tips, respectively.
3.1 The layering model and stochastic properties
For each sample of the shear modulus G(x), the crack problem of the FGM strip
can be studied by expanding the piecewise-exponential model of Guo et al. [6,7] for
mode-I crack problem to the present stochastic mixed-mode crack problems of FGMs.
The model provides a nonhomogeneously-layered thought that introduces the
continuity of material properties between sublayers. Compared with the
homogeneously-layered method found in Ref[32, 33], in which The FGMs are
divided into sublayers with piecewise constant material properties, the
nonhomogeneously-layered method may make it possible that the influences of local
material nonhomogeneity on the crack-tip fields can be adequately allowed for. In Fig.
3, the graded material is separated into M nonhomogeneous layers along x-axis, and
subscript n (n = 1, 2,…, M) is used to mark each layer. Counting from the top surface
of the structure, the n-th layer is positioned between the region x = ln-1 and x = ln,while
let l0 = 0.
The real shear modulus of the strip can be expressed as
G( x )    x  （26）

where   x  is a known real function.We assume the shear modulus of each layer to

be
Gn ( x)  Gn 0en x ， n  1,2, ,M (27)

here Gn 0 and  n are nonhomogeneous parameters of each layer. In the coordinate

system x0-y0, we have

Gn ( x0 , y0 )  Gn 0en x  Gn 0en ( x0 cos  y0 sin ) l0  x  lM , n  1,2,..., M （28）

Using the real material properties on both surfaces of each layer, we can obtain
Gn (ln 1 )    ln 1   Gn 0e nln 1

 , n  1, 2, ,M (29)
Gn (ln )    ln   Gn 0e
 n ln

From Eq.(25), the material coefficients in Eq.(28) corresponding to each layer can be
derived as follows
 ln  (ln ) / (ln 1 ) 
 n 
 ln  ln 1 , n  1,2, ,M (30)
G  (l )e  nln
 n0 n

According to the principle of superposition, the solution of this crack problem can
be divided into two parts. One is associated with an infinite strip containing the
crack on the x0-axis, while the second is an uncracked strip. The total stresses and
displacements in the x-y coordinate system are denoted by  nxx ,  nyy ,  nxy , un and vn (n

represents the number of layers, n  1,2, , M ). For each layer, the governing
equations and constitutive equations are expressed as follows
 nij , j ( x, y )  0 (i, j  x, y ) , n  1,2, ,M (31)

 un 
 
 nxx   x 
1   n ( x ) 3   n ( x ) 0 
  Gn ( x )    vn 
 =
  ( x)  1  3   ( x ) 1   ( x ) 0   n  1,2, ,M
nyy n n  y
 nxy 
 
n
 0 0  n ( x )  1  
 un vn 
  
 y x 
(32)
where  n ( x)  3  4 for plane stain,  n ( x)  (3   ) / (1   ) for plane stress, and

 is the Poisson’s ratio. The boundary conditions and the continuity conditions of the
FGM strip are expressed as follows
 1 xx  0, y   0
  y  
， (33)
 1 xy  0, y   0

 Mxx  l , y   0
  y  
， (34)
 Mxy  l , y   0

 ny0 y0 ( x0 ,0)  p1 ( x0 )

 ，a  x0  b (35)
 nx0 y0 ( x0 ,0)  p2 ( x0 )


 nxx  ln , y     n 1 xx  ln , y 

 ，n  1,2,...M  1 (36)

 nxy  ln , y     n 1 xy  ln , y 

un  ln , y   un 1  ln , y 
 ，n  1,2,...M  1 (37)
 vn  ln , y   vn 1  ln , y 
According to the superposition principle, for each layer the solutions of Eqs. (31) -
(37) can be divided into two parts. Let  nxI 0 x0 ( x0 , y0 ) ,  nyI 0 y0 ( x0 , y0 ) ,  nxI 0 y0 ( x0 , y0 ) ,
unI ( x0 , y0 ) and vnI ( x0 , y0 ) denote the stress components and displacement components

of the first part. They will be solved in the local coordinate system x0-y0. Let
 nxx
II
( x, y ) ,  nyy
II
( x, y ) ,  nxy
II
( x, y ) , unII ( x, y ) and vnII ( x, y ) denote the stress components

and displacement components of the second part. They can be solved in the global
coordinate system x-y. In global coordinate system x-y, the total stresses of the
mixed-mode crack problem for the n-th layer can be written as
 nxx ( x, y )   nxx ( x, y )  cos2  sin 2  2cos  sin    nx0 x0 ( x0 , y0 ) 
II I

   II   2  
 nyy ( x, y )    nyy ( x, y )   sin  cos2  2cos sin    nyI 0 y0 ( x0 , y0 )  ，
 nxy ( x, y )   nxy

II
( x, y )  cos  sin   cos sin  cos2   sin 2    nxI y ( x0 , y0 ) 
 00 
n  1,2, , M (38)
Similarly, in the coordinate system x0- y0, the total stresses can be expressed as
 nx0 x0 ( x0 , y0 )   nxI 0 x0 ( x0 , y0 )  cos2  sin 2  2cos sin    nxx II
( x, y ) 
   I   2   II 
 ny0 y0 ( x0 , y0 )    ny0 y0 ( x0 , y0 )   sin  cos2  2cos  sin    nyy ( x, y )  ,
   I    cos sin   cos sin  cos2   sin 2    nxy ( x, y ) 
 nx0 y0 ( x0 , y0 )   nx0 y0 ( x0 , y0 )  
II

n  1,2, ,M (39)
where x  x0 cos  y0 sin  and y  x0 sin   y0 cos . The total displacement

components of the n-th layer can be written as

un ( x, y )  unII ( x, y )  cos  sin   unI ( x0 , y0 ) 
 v ( x, y )    II   , n  1,2, ,M (40)
 n   vn ( x, y )  sin  cos   vnI ( x0 , y0 ) 

3.2 Solutions of the stress and displacement fields

The attention will be focused on the first part, the infinite strip containing the crack
in the local coordinate system x0-y0. The governing equations and constitutive
equations can be expressed as
 nijI , j ( x0 , y0 )  0 (i, j  x0 , y0 ) , n  1,2, ,M (41)

 unI ( x0 , y0 ) 
 x0 
 nxI 0 x0  1   3 0   
 I  Gn ( x0 , y0 )   vn ( x0 , y0 )
I 
 ny0 y0  = 3 1 0  ,
 I   1  
 y0 
 0 0   1
 nx0 y0   u I ( x , y ) v I ( x , y ) 
 n 0 0  n 0 0 
 y0 x0 
n  1,2, , M (42)
From Eq. (41) and Eq. (42), applying Fourier transform to the x0 variable, solving the
characteristic equation and using inverse Fourier transform, the expressions of the
displacement components can be obtained as follows
 I 1  4

 E
n 1 j ( s ) y0 isx0
un ( x0 , y0 )  2 
n1 j ( s ) An1 j ( s)e ds
 j 1
 4
， n  1,2, ,M (43)

v I ( x , y )  1
 F
n 1 j ( s ) y0 isx0
( s ) An1 j ( s )e ds


n 0 0
2 
j 1
n1 j

From Eq. (39) and Eq. (40), the stress components can be obtained as follows
 I e n ( x0 cos  y0 sin  )  4

 ( s ) y isx

 nx0 x0 0 0 ( x , y ) 
2  
Bn1 j ( s ) An1 j ( s )e n1 j 0 0 ds
 j 1

 I 
en 0 ( x cos   y sin  )
 4 ， n  1,2, (44)
 ,M
0
 ( s ) y isx
 ny0 y0 ( x0 , y0 ) 
2   
Cn1 j ( s ) An1 j ( s )e n1 j 0 0 ds
 j 1

 e  n ( x0 cos   y0 sin  )
 4

 ( s ) y isx
 nxI 0 y0 ( x0 , y0 ) 
 2 
j 1
Dn1 j ( s ) An1 j ( s )e n1 j 0 0 ds

Here, n1 j are the roots of the characteristic equation and s is the Fourier variable.

En1 j ( s ) , Fn1 j ( s) , Bn1 j ( s) , Cn1 j ( s ) and Dn1 j ( s) are known expressions. An1 j ( s) are

unknowns to be solved.
Now let f1 ( x0 ) and f 2 ( x0 ) denote new unknown auxiliary functions at the crack

faces

  
 f1 ( x0 )  x  ylim unI ( x0 , y0 )  lim unI ( x0 , y0 ) 
 0 y 0 

0
, n  1, 2, ,M (45)
 f 2 ( x0 )    lim vnI ( x0 , y0 )  lim vnI ( x0 , y0 ) 
 x0  y 0 y 0  

From Eqs. (43) - (45), using the boundary conditions and the continuity conditions
with the Fourier integral transform method, An1 j ( s) (n= 1, 2, …, M; j= 1, …, 4) can

be obtained as follows
 Wn11 Wn12 
 
1  Wn 21 Wn 22   F1 ( s ) 
( An11 An12 An13 An14 ) 
T
 , n  1,2, ,M (46)
Wn  Wn 31 Wn 32   F2 ( s ) 
 
 Wn 41 Wn 42 
b
where the superscript T denotes transpose, F1 ( s)   f1 ( x0 )eisx dx0 and 0

b
F2 ( s)   f 2 ( x0 )eisx0 dx0 . Wn ( n  1,2,..., M ) and Wnjk ( n  1,2,..., M , k  1, 2 ,
a

j  1, 2,..., 4 ) are known expressions.

Consider the second part, the uncracked strip in the global coordinate system x-y.
Using the similar procedure for solving the first part, the displacement components
and the stress components of the second part can be obtained as follows
 II 1  4

 E
n 2 j ( ) x i y
un ( x, y )  2 n2 j ( ) An 2 j ( )e d

 j 1
 4
, n  1,2, ,M (47)

v II ( x, y )  1
 F
n 2 j ( ) y0 i y
( ) An 2 j ( )e d


n
2 
j 1
n2 j

 II e n x  4

 ( ) x i y

 nxx ( x , y ) 
2  
Bn 2 j ( ) An 2 j ( )e n 2 j d
 j 1

 II e n x  4 , n  1,2, (48)
 
 
 ( ) x i y
Cn 2 j ( ) An 2 j ( )e n 2 j d
,M
 nyy ( x , y )
 2  j 1

 e n x
 4

 ( ) x i y
 nxy ( x, y )   Dn 2 j ( ) An 2 j ( )e n 2 j d
II

 2 
j 1

where n 2 j are the roots of the corresponding characteristic equation and  is the

Fourier variable. En 2 j , Fn 2 j , Bn 2 j , Cn 2 j and Dn 2 j are known expressions. An 2 j are

unknown functions to be solved.

From the stress, displace continuity conditions and the former expressions，after
lengthy formula derivation An 2 j ( n  1,2,..., M , j  1, 2,..., 4 ) can be obtained as

follows
 f1 (u) 
[ An 2 ]4 M 1  [( )]41M 4 M    (u, , x)
b
  du (49)
a 4 M 2
 f 2 (u ) 
where
[ X 1 ( ,0)] 0 0 0 0 0 
[ X ( , l )] e( 2 1 ) l1 [ X ( , l )] 0 0 0 0 
 1 1 2 1 
 [ Z1 ( , l1 )] [ Z 2 (l1 )] 0 0 0 0 
 
( )4 M 4 M  
 0 0 0 0 [ X M 1 ( , lM 1 )] e( M  M 1 ) lM 1 [ X M ( , lM 1 )]
 
 0 0 0 0 [ Z M 1 ( , lM 1 )] [ Z M (lM 1 )] 
 0 0 0 0 0 [ X M ( , lM )] 

(50)
  [  01 (u,  , x )] 
 [ 1 (u,  , l1 )]  e (  2 1 ) l1
[  2 (u,  , l1 )] 
 
 [T1 (u,  , l1 )]  [T2 (u,  , l1 )] 
 
 
 [  n 1 (u,  , ln 1 )]  e( n  n 1 ) ln 1 [  n (u,  , ln 1 )] 
  (u, , x )4 M 2   , n  1,2, , M (51)
 [Tn 1 (u,  , ln 1 )]  [Tn (u,  , ln 1 )] 
 
 (  M  M 1 ) lM 1 
[  M 1 (u,  , lM 1 )]  e [  M (u,  , lM 1 )]
 [TM 1 (u,  , lM 1 )]  [TM (u,  , lM 1 )] 
 
 [  02 (u,  , lM )] 
Here the coefficients in Eqs. (50) - (51) are knowns.
3.3 Singular integral equations and numerical solutions
Substituting Eqs. (44), (48) into Eq. (39), we can obtain the total stress expressions.
Then applying boundary conditions (32) of the crack faces, the following equations
can be derived as
 lim [ nyI y   nxx
II
sin 2    nyy
II
cos2   2 nxy
II
sin  cos ]  p1 ( x0 )
 y0 0
, a  x0  b (52)
0 0


 I
 y0 0 nx0 y0
lim [   II
nxx sin  cos    II
nyy sin  cos    II
nxy (cos 2
  sin 2
 )]  p2 ( x0 )

Through tedious derivation, the singular integral Equations can be obtained as follows
 b b 4Gn 0
 a [ K1n1 (u, x0 )T1n1 (u, x0 )] f1 (u )du  a [ K1n 2 (u, x0 )T1n 2 (u,x0 ) (1 )(u x ) ] f 2 (u )du 2 e
 n x0
p1 ( x0 )
 0
 b
 [ K2 n1 (u, x0 )T2 n1 (u, x0 ) 4Gn 0 b



a (1 )(u x0 )
] f1 (u )du   [ K2 n 2 (u, x0 )T2 n 2 (u,x0 )] f2 (u )du 2 e  n x0 p2 ( x0 )
a

a  x0  b (53)

where n   n cos , and Kinj (u, x0 ) and Tinj (u, x0 ) (i, j1,2; n1,2,...M ) are obtainable.

The method of Erdogan and Gupta [34] is applied to solve numerically the singular
integral Eq.(53). Then Eq. (53) can be expressed as follows
 m 4Gn 0 2 4  n ( b2a rp b2a ) p ( r )
g1 (tk ) wk [ K1n1 (tk ,rp )T1n1 (tk ,rp )] g2 (tk ) wk [ K1n 2 (tk ,rp )T1n 2 (tk ,rp ) 1 (ba )(tk rp ) ]=ba e
m


 1 p

 mk 1 k 1
m 4  n ( 2 rp  2 ) p ( r )
b  a b  a
g1 (tk ) wk [ K2 n1 (tk ,rp )T2 n1 (tk ,rp )4Gn 0  2 ] g2 (tk ) wk [ K2 n 2 (tk ,rp )T2 n 2 (tk ,rp )]=ba e 2 p

 k 1 1  ( ba )( tk rp ) k 1

(54)
where

ba ba ba ba

u t and x0  r (55)
2 2 2 2

1
w(t )  (56)
1  t2
 (k  1)
tk  cos[ ], k  1,..., m (57)
m 1

(2 p  1)
rp  cos[ ], p  1,..., m  1 (58)
2m  2

 
w 1  wm  and wr  , r  2,..., m  1 (59)
2(m  1) m 1
Combining the two additional equations as follows
m m

 g1 (tk )wk  0 and

k 1
 g (t )w
k 1
2 k k 0 (60)

the mixed-mode SIFs can be expressed as

2 2Gn 0 na g2 (a )
K I (a )  lim 2(a  x0 ) y0 y0 ( x0 ,0)   e (61)
x0 a  1 ba

2 2Gn 0 nb g2 (b)

K I (b)  lim 2( x0  b) y0 y0 ( x0 ,0)  e (62)
x0 b  1 ba

2 2Gn 0 na g1 (a )
K II (a )  lim 2(a  x0 ) x0 y0 ( x0 ,0)   e (63)
x0 a  1 ba

2 2Gn 0 nb g1 (b)

K II (b)  lim 2( x0  b) x0 y0 ( x0 ,0)  e (64)
x0 b  1 ba

where g1 (a ) , g1 (b) , g2 (a ) and g2 (b) can be obtained from Eq. (54) for each sample.

Therefore, g1 (a ) , g1 (b) , g2 (a ) and g2 (b) are also related to those of the shear modulus.

In order to make the probabilistic characteristic of the mixed-mode stress fields, the
samples should be chosen efficiently, the mean of the SIFs can be
N N
1 1
K I (a) 
N
K
j 1
( j)
I (a ) and K I ( b ) 
N
K
j 1
( j)
I ( b) (65)

N N
1 1
K II (a) 
N
 K II( j ) (a) and KII (b) 
j 1 N
K
j 1
( j)
II ( b) (66)

Here, the superscript j denotes the j-th sample. Finally, after running the program on
the computer to complete the calculation of each sample, the probability
characteristics of the SIFs can be obtained by developing histograms.
4 Analysis and discussion of typical examples
Now the probabilistic characteristics of the SIFs will be discussed. An arbitrarily
oriented crack under the fixed-grip loading condition is considered. In this case,
p1 ( x0 ) and p2 ( x0 ) in Eq. (35) can be expressed as follows

p1 ( x0 )  2(1   ) cos2  (67)

p2 ( x0 )  2(1   )cos sin  (68)

The SIFs are normalized by k0  2(1   ) (b  a) / 2 . Poisson’s ratio is assumed to

be 0.3. We can obtain the mixed-mode SIFs of different samples by the method in
section 3. It should be mentioned that the shear modulus G( x ) is still a random

variable because the volume fraction  p ( x ) is a random variable. The previous

research (Guo et al., 2012) showed that the mean and standard variance of G( x ) can

vary with the variation of the ratio of the shear modulus of the matrix Gm and the

shear modulus of the particle G p . At the same time, there exists the influences of the

standard variance  p ( x ) of  p ( x ) on G( x ) .Thus we will illustrate the

probabilistic characteristics of the SIFs for adequate sample number under the
conditions of the certain ratios of G p / Gm and  p / 0 by sample distribution

graphics and probability density function diagram.

First we will discuss the influences of random characteristics of the material
properties with the different ratios of  p / 0 on the stochastic characteristics of the

SIFs of the crack-tip fields. Figs. 4-7 show the variation of the stochastic
characteristics of mixed-mode SIFs with the different ratios of  p / 0 when

Gp / Gm  10 and   0.25 .

For the mode-I SIFs, from Figs. 4-5, it can be found that
(1) when  p   0 the fluctuations of each curve corresponding to different length

of the crack are all not evident. It indicates that for the same crack length the SIFs
vary slightly with the number of the sample. But the means of mode-I SIFs increase
with the increasing length of the crack.
(2) when  p = 3 0 the sample curves of SIFs corresponding to the different

lengths of the crack fluctuate obviously. And the fluctuation is more noticeable when
c / l  0.4 . It indicates that the normalized SIFs are of the stochastic features.
(3) By further comparative observation, with the increasing standard variance
  ( x ) of  p ( x ) , the random characteristics of the material properties can have
p

noticeable effect on the stochastic characteristics of the mode-I SIFs at the crack tip of
the crack with   0.25 . And the effect can be accumulated with the increasing
crack length.
For the mode-II SIFs, from Figs. 6-7, we can find the similar varying pattern with
that of Figs. 4-5. The difference between two groups of graphs of the random
characteristics of the mode-II SIFs is more obvious than that of the mode-I SIFs.
Next, the influences of random characteristics of the material properties with the
different ratios of G p / Gm on the stochastic characteristics of the SIFs will be

discussed. Figs. 8-11 show the variation of the stochastic characteristics of the mixed-
mode SIFs with the different ratios of G p / Gm when  p = 2 0 and   0.25 .

We have
(1) For the mode-I SIFs, shown in Figs. 8-9, when Gp / Gm  20 the fluctuation

range of the sample curves of mode-I SIFs corresponding to the same crack length is
getting slightly larger than that when G p / Gm =5 .

(2) For the mode-II SIFs, shown in Fig. 10 and Fig. 11, the similar phenomena can
be observed.
As described above, the random behaviors of the mixedmode SIFs are more
obviously with the increasing ratio G p / Gm and the increasing crack length. Table 1

may also help to illustrate the above view. When Gp / Gm  20 , the standard

deviations of K / K0 for the same crack length are larger than that when G p / Gm =5 .

And when   0.25 , the standard deviations of K / K0 with c / l  0.4 are much

larger under the same material properties.

At least, taking the crack-tipa as an example, the influence of random properties on
the stochastic characteristics of the mixed-mode SIFs with the different crack angle
are investigated. Figs. 12-13 show the probability density function of the mixed-mode
SIFs at the crack-tip a corresponding to the different crack angle with c / l  0.3 and
the random properties ratio Gp / Gm  20 and  p = 2 0 . It can be found that

(1)when   0.1 , the corresponding mean and standard variance of K I (a ) / K0

are smaller than those when   0.25 , as shown in Fig.12.

(2) when   0.1 , the corresponding mean and standard variance of K II (a ) / K0
are larger than those when   0.25 . But with the variation of the crack angle from
  0.15 to   0.25 , the decrease in the standard variance of K II (a ) / K0
shows a negligible change in Figs.13-b, 13-c and 13-d.
(3)Through further comparison, when c / l  0.3 the probabilistic characteristics
of the mode-I SIFs of crack-tip a are more noticeable than those of the mode-II SIFs
with the same angle.

5 conclusions
The probability characteristics of SIFs in FGMs with an arbitrarily oriented crack
and stochastic material properties are analyzed by means of integrating the macro
fracture mechanics model, i.e. nonhomogeneously-layered method, and the micro
material characteristics model, i.e. Mori-Tanaka model. The probability density
functions and samples distribution cases of the mixed-mode SIFs are studied
analytically. And the effects of the material stochastic properties with geometric
variation of crack on the probability characteristics of the mixed-mode SIFs are
illustrated. At present similar investigations of this kind of issues are rare. The results
in this paper may provide theoretical basis for actual FGMs. The layered model has
been use to analyze the crack problems of FGMs with general mechanical properties
in our previous researches and has been proved to be effective in studying the macro
fracture problem. Other micro material characteristics models also can be considered
to replace the Mori-Tanaka model. This will be our work to complete next step.

6 Acknowledgements
The authors are very thankful to Dr. Zhihai Wang for his helpful discussion and
suggestions.
The research reported in this paper was supported by Heilongjiang Provincial
Education Department Surface Scientific Research (135109239).

References
[1] F. Shahabian, S.M. Hosseini
Stochastic dynamic analysis of a functionally graded thick hollow cylinder with
uncertain material properties subjected to shock loading
Mater. Design, 31(2) (2010), pp. 894–901
[2] L.C. Guo, L.Z. Wu, T. Zeng, L. Ma
Mode I crack problem for a functionally graded orthotropic strip
Eur. J. Mech. - A/Solids, 23(2) (2004), pp. 219–234
[3] L.C. Guo, L.Z. Wu, T. Zeng, L. Ma
The dynamic fracture behavior of a functionally graded coating-substrate system
Compos. Struct., 64(3-4) (2004), pp. 433–441
[4] P. Gu, R.J. Asaro
Cracks in functionally graded materials
Int. J. Solids Struct., 34(1)(1997), pp. 1–17
[5] L.C. Guo, N. Noda
Dynamic investigation of a functionally graded layered structure with a crack
crossing the interface
Int. J. Solids Struct., 45(1)(2008), pp. 336–357
[6] L.C. Guo, N. Noda
Modeling method for a crack problem of functionally graded materials with
arbitrary properties—piecewise-exponential model
Int. J. Solids Struct., 44(2007), pp. 6768–6790
[7] L.C. Guo, N. Noda, L.Z. Wu
Thermal fracture model for a functionally graded plate with a crack normal to
the surfaces and arbitrary thermomechanical properties
Compos. Sci. Technol., 68(3-4)(2008), pp. 1034–1041
[8] X.M. Bai, L.C. Guo, Z.H. Wang, S.Y. Zhong
A dynamic piecewise-exponential model for transient crack problems of
functionally graded materials with arbitrary mechanical properties
Theor. Appl. Fract. Mec., 66(2013), pp. 41–51
[9] L.C. Guo, N. Noda
Investigation methods for thermal shock crack problems of functionally graded
materials–Part I: Analytical method
J. Therm. Stresses, 37(3) (2014), pp. 292–324
[10] H.Z. Pan, T.S. Song, Z.H. Wang
An analytical model for collinear cracks in functionally graded materials with
general mechanical properties
Compos. Struct., 132(2015), pp. 359–371
[11] H.Z. Pan, T.S. Song, Z.H. Wang
Thermal fracture model for a functionally graded material with general
thermomechanical properties and collinear cracks
J. Therm. Stresses, 39(7) (2016), pp. 820–834
[12] Z.H. Wang, L. Zhang, L.C. Guo
A viscoelastic fracture mechanics model for a functionally graded materials strip
with general mechanical properties
Eur. J. Mech. - A/Solids, 44(2014), pp. 75–81
[13] Z.H. Wang, L.C. Guo, L. Zhang
A general modelling method for functionally graded materials with an arbitrarily
oriented crack
Philos. Mag., 94(8)(2014), pp. 764–791
[14] V.B. Chalivendra
Mixed-mode crack-tip stress fields for orthotropic functionally graded materials
Acta Mech., 204(1–2)(2009), pp. 51–60
[15] S. Rahman, A. Chakraborty
A stochastic micromechanical model for elastic properties of functionally graded
materials
Mech. Mater., 39(2007), pp. 548–563
[16] J. Yang, K.M. Liew, S. Kitipornchai
 Stochastic analysis of compositionally graded plates with system randomness
under static loading
Int. J. Mech. Sci., 47(10)( 2005), pp. 1519–1541
[17] L.C. Guo, Z.H. Wang, N. Noda
A fracture mechanics model for a crack problem of functionally graded materials
with stochastic mechanical properties
Proc. R. Soc. A-Math. Phys. Eng. Sci., 468(2146) (2012), pp. 2939–2961
[18] F.J. Ferrante, L.L. Graham-Brady
Stochastic simulation of non-Gaussian/non-stationary properties in a
functionally graded plate
Comput. Methods Appl. Mech. Engrg., 194(12–16) (2005), 1675–1692.
[19] R. Chiba, Y. Sugano
Stochastic analysis of a thermoelastic problem in functionally graded plates with
uncertain material properties
Arch. Appl. Mech., 78(10)(2008), pp. 749–764
[20] A. Lal, H.N. Singh, N.L. Shegokar
FEM model for stochastic mechanical and thermal postbuckling response of
functionally graded material plates applied to panels with circular and square
holes having material randomness
Int. J. Mech. Sci., 62(1)(2012), pp. 18–33
[21] S.M. Hosseini, F. Shahabian
Stochastic assessment of thermo-elastic wave propagation in functionally graded
materials(FGMs) with Gaussian uncertainty in constitutive mechanical
properties J. Therm. Stresses, 34(10)(2011), pp. 1071–1099
[22] S.M. Hosseini, F. Shahabian
Stochastic hybrid numerical method for transient analysis of stress field in
functionally graded thick hollow cylinders subjected to shock loading
J. Mech. Sci. Technol., 27(5)(2013), pp. 1373–1384
[23] N.L. Shegokar, A. Lal
Stochastic finite element nonlinear free vibration analysis of piezoelectric
functionally graded materials beam subjected to thermo-piezoelectric loadings
with material uncertainties
Meccanica, 49(5)(2014), pp. 1039–1068
[24] A. Chakraborty, S. Rahman
Stochastic multiscale models for fracture analysis of functionally graded
materials
Eng. Fract. Mech., 75(8)(2008), pp. 2062–2086
[25] L.C. Guo, F.N. Guo, H.J. Yu, L. Zhang
An interaction energy integral method for nonhomogeneous materials with
interfaces under thermal loading
Int. J. Solids Struct., 49(2)(2012), pp. 355–365
[26] K. Huang, L.C. Guo, H.J. Yu, P.F. Jia, T. Kitamura
A domain-independent interaction integral method for evaluating the dynamic
stress intensity factors of an interface crack in nonhomogeneous materials
 Int. J. Solids Struct., 100(2016), pp. 547–557
[27] F.N. Guo, L.C. Guo, H.J. Yu, L. Zhang
Thermal fracture analysis of nonhomogeneous piezoelectric materials using an
interaction energy integral method
Int. J. Solids Struct., 51(3–4)(2014), pp. 910–921
[28] L.C. Guo, T. Kitamura, Y.B. Yan, T. Sumigawa, K. Huang
Fracture mechanics investigation on crack propagation in the nano-multilayered
materials
Int. J. Solids Struct., 64-65(2015), 208–220
[29] K. Huang, L.C. Guo, Y.B. Yan, T. Kitamura
Investigation on the competitive fracture behavior in nano-multilayered
structures
Int. J. Solids Struct., 92-93(2016), pp. 45–53
[30] F.N. Guo, L.C. Guo, K. Huang, X.M. Bai, S.Y. Zhong, H.J. Yu
An interaction energy integral method for T-stress evaluation in
nonhomogeneous materials under thermal loading
Mech. Mater., 83(2015), pp. 30–39
[31] S.Y. Zhong, L.C. Guo, G. Liu, L. Zhang, S.D. Pan
A random waveness model for the stiffness and strength evaluation of 3D woven
composites
Compos. Struct., 152(2016) , pp. 1024–1032
[32] G.H. Paulino, Z.H. Jin
Transient thermal stress analysis of an edge crack in a functionally graded
material
Int. J. Fract., 107(2001), pp. 73-98
[33] J.J. Mao, L.L. Ke, S. Kitipornchai, Y.S. Wang
Thermoelastic instability of functionally graded coating with arbitrarily varying
properties considering contact resistance and frictional heat
Appl. Math. Model., 43(2017), pp. 521-537
[34] F. Erdogan, G.D. Gupta
On the numerical solution of singular integral equations
Quart. Appl. Math., 29(1972), pp. 525–534
 D1

D2

D3

Fig. 1. Schematic of the microstructure of FGM strip.

y0
a

l o y
b
θ
xx x0

Fig. 2. The geometry of the FGM strip with an arbitrarily oriented crack and stochastic
properties.
 y0 l1 l2
a ln

… …
l o y
b
lM-2
θ
xx x0 lM-1
lM

Fig. 3. The Schematic diagram of layering method.

Fig. 4. Distribution of mode I SIFs for different samples and different crack lengths with

 p   0 .
Fig. 5. Distribution of mode I SIFs for different samples and different crack lengths with

  = 3  .
p 0

Fig. 6. Distribution of mode II SIFs for different samples and different crack lengths with

 = .
p 0
Fig. 7. Distribution of mode II SIFs for different samples and different crack lengths with

  = 3  .
p 0

Fig. 8. Distribution of mode I SIFs for different samples and different crack lengths with

G p / Gm  5 .
Fig. 9. Distribution of mode I SIFs for different samples and different crack lengths with

Gp / Gm  2 0.

Fig. 10. Distribution of mode II SIFs for different samples and different crack lengths with

G p / Gm  5 .
 Fig. 11. Distribution of mode II SIFs for different samples and different crack lengths with

Gp / Gm  2 0.

Fig. 12. Probability density function of mode I SIFs at crack-tip a for different crack angles.
Fig. 13. Probability density function of mode II SIFs at crack-tip a for different crack angles.
 Table 1.
Comparison between the present SIFs and the corresponding analytical results of Guo and Noda
[6].
Normalized Guo and Noda's results [6] Present results
crack length
KI(a)/K0 KI(b)/K0 KI(a)/K0 KI(b)/K0
c/l=0.1 1.022 0.976 1.029 0.987
c/l=0.2 1.071 0.979 1.074 0.989
c/l=0.3 1.141 0.998 1.142 1.009

Table 2.

The standard deviations of K/K0 when  p = 2 0 and   0.25 .

standard
G p / Gm  5 Gp / Gm  20
deviation
of K/K0 c/l = 0.2 c/l = 0.3 c/l = 0.4 c/l = 0.2 c/l = 0.3 c/l = 0.4

 K ( a )/ K
I 0
0.0010 0.0017 0.0020 0.0012 0.0020 0.0024

 K ( b )/ K
I 0
0.0005 0.0009 0.0012 0.0009 0.0017 0.0024

K II ( a )/ K0 0.0007 0.0015 0.0021 0.0009 0.0021 0.0029

K II ( b )/ K0 0.0004 0.0007 0.0008 0.0006 0.0011 0.0014

Highlights
 The method based on macro and micro mechanics can be efficiently applied.
 The probabilistic attributes of the mixed-mode SIFs in FGMs can be obtained.
 The material stochastic properties can affect the probability characteristics of
SIFs.