Engineering Analysis with Boundary Elements ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

Engineering Analysis with Boundary Elements

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

The singular boundary method: Mathematical background

and application in orthotropic elastic problems
Yan Gu a,b, Wen Chen a,n, Zhuo-Jia Fu a, Bo Zhang c
a
Department of Engineering Mechanics, College of Mechanics and Materials, Hohai University, Nanjing 210098, PR China
b
Department of Mechanical Engineering and Materials Science, Duke University, NC 27708, USA
c
Center for Research in Extreme Scale Technologies, Indiana University, Bloomington, IN 47408, USA

art ic l e i nf o a b s t r a c t

Article history: The singular boundary method (SBM) is a recent strong-form boundary discretization numerical
Received 20 October 2013 technique and can be viewed as one kind of modified method of fundamental solutions (MFSs).
Accepted 4 February 2014 Although the method has been successfully used in many fields of engineering analysis, there has been
no attempt yet to present a work discussing the mathematical background of the method. This paper fills
Keywords: this gap in the SBM and documents the first attempt to apply the method to the solution of orthotropic
Singular boundary method elastic problems. Three benchmark numerical problems are tested to demonstrate the feasibility and
Meshless boundary collocation method accuracy of the proposed method through detailed comparisons with the MFS and the boundary element
Method of fundamental solutions method (BEM).
Origin intensity factor
& 2014 Elsevier Ltd. All rights reserved.
Orthotropic elastic problems

1. Introduction evaluation, will be discussed in detail. This paper also documents

the first attempt to extend the method to general orthotropic
The singular boundary method (SBM) has emerged as an effective elastic problems. The study of boundary value problems for
meshless boundary collocation method for the solution of certain orthotropic elastic materials has received considerable attention
boundary value problems. This method, based on the notion of in recent years [13,14]. This interest is partly related to the
boundary element method (BEM) [1–3] and MFS [4–8], fully inherits extensive use of composite materials in various engineering
the merits of both and in the meantime possessing its unique applications, like wood, reinforced concrete, and all materials that
advantages. First, it does not involve integration which could be are reinforced with fibers.
otherwise troublesome and expensive as in the BEM-based methods. A brief outline of the rest of this paper is as follows. In Section
Second, it sidesteps the fictitious boundary issue associated with the 2, we describe the traditional MFS formulation for the solution of
traditional MFS by means of a concept of the origin intensity factor, a 2D orthotropic elastic problems. The SBM formulation and its
numerical strategy that isolates singularities of the fundamental mathematical background are discussed in details in Section 3.
solutions and allows the source and collocation points coincide on And then in Section 4 the method is tested successfully over three
the real boundary. The ease of implementation and its low computa- benchmark numerical problems. Finally, the conclusions and
tional cost make the method a competitive alternative for certain remarks are provided in Section 5.
boundary value problems. Comprehensive reviews on the SBM for
various applications can be found in Refs. [9–12]. Although the
method has been successfully used in many fields of engineering
analysis, the mathematical background of the method has not, as yet,
2. MFS formulation for orthotropic elastic problems
been discussed.
One of the aims of the present paper is to fill the gap
For the assumption of plane stress distribution in an ortho-
mentioned above and present a relatively detailed theoretical
tropic material, Hooke's law takes the form (matrix representa-
background of this methodology. Related issues in the SBM, such
tion)
as the existence of the origin intensity factors and their numerical
2
εx 3 2 S11 S12 0
32
sx
3
6 εy 7 6 S 0 54 s y 7
7 6
4 5 ¼ 4 21 S22 5
n
Corresponding author. Tel.: þ 86 15261889263; fax: þ 86 2583736860. γ xy 0 0 S66 τxy
E-mail address: [email protected] (W. Chen).

http://dx.doi.org/10.1016/j.enganabound.2014.02.001
0955-7997 & 2014 Elsevier Ltd. All rights reserved.

Please cite this article as: Gu Y, et al. The singular boundary method: Mathematical background and application in orthotropic
elastic problems. Eng. Anal. Boundary Elem. (2014), http://dx.doi.org/10.1016/j.enganabound.2014.02.001i
2 Y. Gu et al. / Engineering Analysis with Boundary Elements ∎ (∎∎∎∎) ∎∎∎–∎∎∎

2 32 3
1=E1  v12 =E1 0 sx α2 A1 pffiffiffiffiffiffi pffiffiffiffiffiffi
6 T 21 ðy; xÞ ¼ K ½ α2 ðx1  y1 Þn2 ðyÞ  ðx2  y2 Þn1 ðyÞ= α2 
¼ 4  v12 =E1 1=E2 0 7 6 7
54 s y 5 ð1Þ
2
r2
0 0 1=G12 τxy α1 A2 pffiffiffiffiffiffi pffiffiffiffiffiffi
K 2
½ α1 ðx1  y1 Þn2 ðyÞ  ðx2  y2 Þn1 ðyÞ= α1 ; ð5cÞ
r1
where the stress sij and strain εij are mean values taken through
" pffiffiffiffiffiffi pffiffiffiffiffiffi#
the thickness of the material; Sij(i,j¼ 1,2) and S66 are flexibility A2 α 2 A1 α 1
coefficients; E1 and E2 are Young's moduli in the directions of x1 T 22 ðy; xÞ ¼ K½ðx1  y1 Þn1 ðyÞ þ ðx2  y2 Þn2 ðyÞ  ð5dÞ
r 22 r 21
and x2 axes; G12 denotes the shear modulus for planes parallel to
the x1 x2 plane; v12 is Poisson's ratio characterizing the contrac- By employing the radial basis functions (RBFs) technique [16–19],
tion in the direction of the x2 axis when tension is applied in the the displacements and stresses can be approximated by linear
direction of the x1 axis. combinations of fundamental solutions with respect to different
The Navier–Cauchy equations for plane orthotropic materials, source points x as follows:
in the absence of body forces, referring to displacements u1 and u2 N N
are [15] ui ðym Þ ¼ ∑ aj ðxn ÞU ij ðym ; xn Þ ¼ ∑ ½a1 ðxn ÞU i1 ðym ; xn Þ
n¼1 n¼1
∂2 u1 ðxÞ ∂2 u2 ðxÞ ∂2 u1 ðxÞ þ a2 ðxn ÞU i2 ðym ; xn Þ; ð6aÞ
C 11 þðC 12 þ C 66 Þ þ C 66 ¼ 0; ð2aÞ
∂x12 ∂x 1 ∂x2 ∂x22
N N
t i ðym Þ ¼ ∑ aj ðxn ÞT ij ðym ; xn Þ ¼ ∑ ½a1 ðxn ÞT i1 ðym ; xn Þ
∂2 u2 ðxÞ ∂2 u1 ðxÞ ∂2 u2 ðxÞ n¼1 n¼1
C 22 þðC 12 þ C 66 Þ þ C 66 ¼ 0; ð2bÞ
∂x22 ∂x1 ∂x2 ∂x21 þ a2 ðxn ÞT i2 ðym ; xn Þ ð6bÞ
in which C11 ¼S22/D, C12 ¼  S12/D, C22 ¼S11/D, C66 ¼1/S66, where i,j¼ 1,2, N is the specified number of sources, y A Ω ¼ Ω [ m

D ¼ S11 S22  S212 . ∂Ω is the mth collocation points, xn is the nth source point,
These are subject to the boundary conditions fa1 ðxn ÞgN n N
n ¼ 1 and fa2 ðx Þgn ¼ 1 denote the unknown coefficients.
ui ðxÞ ¼ ui ; x A Γ u ðDirichlet boundary conditionsÞ; ð3aÞ In the traditional MFS, a fictitious boundary slightly outside the
problem domain is required in order to place the source points
t i ðxÞ ¼ t i ; x A Γ t ðNeumann boundary conditionsÞ; ð3bÞ and avoid the singularity of the fundamental solutions. These
source points are either pre-assigned or taken to be part of the
where ti(x) denotes the component of boundary traction in the ith
unknowns of the problem along with the coefficients faj ðxn ÞgN n ¼ 1.
coordinate direction, Γu and Γt construct the whole boundary of In either case, the unknowns are determined so that the approx-
the domain Ω which we shall assume to be piecewise smooth, ui
imations (6) satisfy, in some sense, the boundary conditions (3) as
and t i represent the prescribed displacements and tractions,
close as possible [20,21]. In the early applications of the MFS, the
respectively.
locations of the source points were determined by a non-linear
Employing indicial notation for the coordinates of points
system of the equations that can be solved using a non-linear
x ¼(x1,x2) and y ¼(y1,y2), respectively, the Kelvin fundamental
least-squares minimization software. This approach, however, has
solutions of the systems (2) and (3) can be expressed as [15]
attracted limited attention primarily because of its high computa-
pffiffiffiffiffiffi pffiffiffiffiffiffi
U 11 ðy; xÞ ¼ K½ α1 A22 ln r 1  α2 A21 ln r 2  ð4aÞ tional costs and the criticism that a linear boundary value problem
is converted to a non-linear discrete problem. In the more
 
x2  y2 x2  y2 established approach these days the source points are pre-
U 12 ðy; xÞ ¼  KA1 A2 arctanpffiffiffiffiffiffi  arctanpffiffiffiffiffiffi ;
α1 ðx1  y1 Þ α2 ðx1 y1 Þ assigned, collocation simply leads to a linear system of M equa-
ð4bÞ tions in N unknowns which can be solved by a least-squares solver.
However, despite many years of focused research, the pre-
U 21 ðy; xÞ ¼ U 12 ðy; xÞ ð4cÞ determination of the fictitious boundary is largely based on
experiences and therefore often troublesome, especially for com-
pffiffiffiffiffiffi pffiffiffiffiffiffi
U 22 ðy; xÞ ¼ K½A22 ln r 2 = α2  A21 ln r 1 = α1  ð4dÞ plicated higher dimensional domain problems [22–27]. This draw-
back severely downplays the applicability of the MFS to real-world
where α1 and α2 satisfy
applications.
α1 þ α2 ¼ ð2S12 þ S66 Þ=S22 ; α1 α2 ¼ S11 =S22 ;
and
3. Singular boundary method: mathematical background
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
K ¼ 1=½2π ðα1  α2 ÞS22 ; Ai ¼ S12  αi S22 ; r i ¼ αi ðx1  y1 Þ2 þ ðx2  y2 Þ2 : and numerical implementation

The fundamental solution Uij(y,x) described above indicates the The basic idea of the SBM is to introduce a concept of the origin
displacement produced at point y by a concentrated unit body intensity factor to isolate the singularity of the fundamental
force applied at point x, in which the first subscript (i) denotes the solutions, so that the source points can be directly placed on the
direction of the displacement whereas the second one (j) denotes real boundary [10]. With this idea in mind the SBM interpolation
the direction of the unit force. The fundamental solution of the can be expressed as
tractions can be obtained by first calculating the fundamental N
solutions of strains and then applying Hooke's law ui ðym Þ ¼ ∑ aj ðxn ÞU ij ðym ; xn Þ þ aj ðxm ÞAij ðxm Þ; ð7Þ
" pffiffiffiffiffiffi pffiffiffiffiffiffi# n¼1
A2 α1 A1 α2 ma n
T 11 ðy; xÞ ¼ K½ðx1  y1 Þn1 ðyÞ þ ðx2 y2 Þn2 ðyÞ  ; ð5aÞ
r 21 r 22
N
t i ðym Þ ¼ ∑ aj ðxn ÞT ij ðym ; xn Þ þaj ðxm ÞBij ðxm Þ; ð8Þ
A2 pffiffiffiffiffiffi pffiffiffiffiffiffi
T 12 ðy; xÞ ¼ K 2 ½ α2 ðx1 y1 Þn2 ðyÞ  ðx2  y2 Þn1 ðyÞ= α2  n¼1
r2 m an

A1 pffiffiffiffiffiffi pffiffiffiffiffiffi where y A Γ, Aij ðxm Þ and Bij ðxm Þ are defined as the origin intensity
m
 K 2 ½ α1 ðx1  y1 Þn2 ðyÞ  ðx2  y2 Þn1 ðyÞ= α1 ; ð5bÞ
r1 factors, i.e., the diagonal and sub-diagonal elements of the SBM

Please cite this article as: Gu Y, et al. The singular boundary method: Mathematical background and application in orthotropic
elastic problems. Eng. Anal. Boundary Elem. (2014), http://dx.doi.org/10.1016/j.enganabound.2014.02.001i
 Y. Gu et al. / Engineering Analysis with Boundary Elements ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3

Z Z
interpolation matrix. In the following, some related issues, such as ðxk  yk Þ3 ðxk  x^ k Þ3
the existence of the origin intensity factors and their numerical lim ½ψ ðxÞ  ψ ðxÞd
^ Γx ¼ ½ψ ðxÞ  ψ ðxÞd
^ Γx ;
y-x^ Γ x jx  yj
2
Γ x jx  x^ j2
evaluations, will be discussed in detail. ð12bÞ

3.1. Regularized boundary integral equations Z

ðxk  yk Þ2 ðxl  yl Þ
lim   ½ψ ðxÞ  ψ ðx^ ÞdΓ x
y-x^ Γ x x  y 2
From the mathematical point of view, both the SBM and MFS Z
can be viewed as a discrete version of the indirect BEM. Thus the ðxk  x^ k Þ2 ðxl  x^ l Þ
¼   ½ψ ðxÞ  ψ ðxÞd
^ Γx: ð12cÞ
indirect BEM can provide us useful information for determining Γx x  x^ 2
the singular terms of the influence matrices (coefficient matrices).
Bsased on elasticity theory, we have the following indirect BEM
Proof. For any ε 40, the difference between the left- and right-
formulation in the form of a single-layer potential to present the
hand sides of Eq. (12a) is
solution of displacement field
Z Z " #
xk  yk xk  x^ k
ui ðyÞ ¼ aj ðxÞU ij ðy; xÞ dΓ x ; ð9Þ Δ¼  
2   ½ψ ðxÞ  ψ ðxÞd
^ Γx
Γx Γ x x  y x  x^ 2

and the corresponding traction field is given by

Z
t i ðyÞ ¼ aj ðxÞT ij ðy; xÞdΓ x ð10Þ Dividing integral Δ into two parts: Δ1 over the segment
Γx
Γ δ ¼ fx A Γ jjx  x^ j r δg and Δ2 over ΓC ¼ Γ  Γδ. We first examine
It is noticed that due to the properties of the singular-layer Δ1 r ε. Using jyk  x^ k j r K 1 jy  xj, we find that
potential, the displacement field (9) is continuous throughout the  
 x y xk  x^ k  ðx1  x^ 1 Þ2 þ ðx2  x^ 2 Þ2
 k 1
domain as well as across the boundary. The surface traction (10),  k
 r þ  
however, undergoes a discontinuity across the boundary [28,29]. x  y2 x  x^ 2    
x yx  x^ 2 x  x^ 
On the internal side of Γ, indicated by the direction of the normal    
ðx1  x^ 1 Þðx1 y þ y  x^ 1 Þ ðx2  x^ 2 Þðx2  y þ y  x^ 2 Þ 1
þ  
1 1 2 2
vector, ¼    þ   
x  yx  x^ 2 x  yx  x^ 2 x  x^ 
Z
1 þ
t i ðyÞ ¼ ai ðyÞ þ aj ðxÞT ij ðy; xÞ dΓ x ; y-Γ ; ð11aÞ      
ðx1  x^ 1 Þðx1  y1 Þ ðx1  x^ 1 Þðy1  x^ 1 Þ ðx2  x^ 2 Þðx2  y2 Þ
2 Γx
r   2 þ   2 þ   2
whereas on the external side of Γ, x  yx  x^  x yx  x^  x  yx  x^ 
Z  
1 ðx2  x^ 2 Þðy2  x^ 2 Þ 1

t i ðyÞ ¼  ai ðyÞ þ aj ðxÞT ij ðy; xÞ dΓ x ; y-Γ ð11bÞ þ    þ  
2 Γx x  yx  x^ 2 x  x^ 

In the integrand of Eqs. (11) appears the term Tij(y,x) which K2

r   ðK 2 is a constantÞ
diverges like 1/|y x| as y-x and must therefore be treated x  x^ 
judiciously when employing numerical quadrature. Reducing the    
order of the singularity prior discretization is an effective approach
α
Note that ψ(x) AC0, (Γ), i.e., ψ ðxÞ  ψ ðxÞ ^  rK 3 x  x^ α (K3 is a
and results in improved accuracy regardless of the choice of constant), we have
quadrature scheme. We first give the following limiting theorem Z
  1
Δ1  r K 2 K 3 dΓ x
to be used in the remainder of the paper. ^ 1α
Γ δ jx  xj
 
Definition 1. A function u: E A R3-R is said to be uniformly Let t ¼ x  x^  and notes that t is the chordal length of the arc xx^ .
Hölder on E with exponent α A (0, 1) if Since Γ be a piecewise smooth curve, it is obvious that the values
  of t and the length of arc xx^ are both bounded, i.e., |dΓx| rK4 dt
uðxÞ  uðyÞ
sup α o 1: (K4 is a constant). Therefore, we have
x;y A E;x a y ‖x  y‖
Z δ
  1 α
Δ1  r K 2 K 3 K 4 dt rK 5 δ ðK 5 is a constantÞ
1α
If the above inequality holds for α ¼1, u is called uniformly 0 t

Lipschitz on E. By choosing sufficiently small δ, one has |Δ1| r ε.

l,α For the integral Δ2, we have
Definition 2. If lZ 0 is an integer and α A(0, 1], the space C (Ω) is
formed by those functions in Cl(Ω) whose derivatives of order l are Z 
   xk  yk
Δ2  r 
uniformly Hölder (uniformly Lipschitz for α ¼ 1) with exponent α Γ c ðx1  y1 Þ þ ðx2  y2 Þ
2 2
on every compact subset of Ω. 
xk  x^ k  

Lemma 1. Let Γ be a piecewise smooth curve, x^ be any point on Γ,   ψ ðxÞ  ψ ðx^ Þ dΓ x
ðx1  x^ 1 Þ þ ðx2  x^ 2 Þ 
2 2
and ψ ðdÞ A C 0;α ðΓ Þ be a well-behaved function. Suppose
    Noting that δ has  been fixed and x does not belong to ΓC, therefore
^
h ¼ y  x^ ; d ¼ inf y  x
xAΓ by taking y  x^  sufficiently small there holds |Δ2| r ε. Eqs. (12b)
then if and (12c) can be proved in a similar way.
Using the subtraction technique, Eq. (10) can be rewritten in
h=d rK 1 ðK 1 is a constantÞ the following equivalent form
there holds ðk; l ¼ 1; 2Þ Z Z
Z Z t i ðyÞ ¼ ½aj ðxÞ  aj ðx^ ÞT ij ðy; xÞdΓ x þ aj ðx^ Þ ½T ij ðy; xÞ þ T ij ðx; yÞdΓ x
xk  yk xk  x^ k Γx Γx
lim ½ ψ ðxÞ  ψ ð ^ Γx ¼
xÞd ½ψ ðxÞ  ψ ðx^ ÞdΓ x ; : Z
y-x^ Γ x jx  yj2 Γx  x
jx ^ j2
 aj ðx^ Þ T ij ðx; yÞdΓ x ; ð13Þ
ð12aÞ Γx

Please cite this article as: Gu Y, et al. The singular boundary method: Mathematical background and application in orthotropic
elastic problems. Eng. Anal. Boundary Elem. (2014), http://dx.doi.org/10.1016/j.enganabound.2014.02.001i
4 Y. Gu et al. / Engineering Analysis with Boundary Elements ∎ (∎∎∎∎) ∎∎∎–∎∎∎

where x^ A Γ be any point on the boundary and elements. One is about the geometry of the boundary, while the other
Z ( has to do with the variation of the unknown boundary quantity over
 δij ; y A Ω;
T ij ðx; yÞdΓ x ¼ ð14Þ the element. Here supposing the values of the boundary quantity aj
yAΩ :
C
Γx 0; are constant over each segment and equal to their value at the
The detailed derivations of Eq. (14) are provided in Appendix A. It collocation/source point ym. Then considering a discretized approach
is noted that the kernel [Tij(y,x)þTij(x,y)] contains the term [nj(y) of the above integral Eq. (19) one gets
N
nj(x)] (see Eq. (7)) which tends to zero as y-x.
t i ðym Þ ¼ τai ðym Þ þ ∑ aj ðxn ÞT ij ðym ; xn ÞLn
Taking the limit as the point y approaches x^ A Γ and using n¼1
Lemma 1, Eq. (13) can be rewritten as n am
Z N
t i ðx^ Þ ¼ ai ðx^ Þ þ ½aj ðxÞ  aj ðx^ ÞT ij ðx^ ; xÞdΓ x þ aj ðym Þ ∑ T ij ðxn ; ym ÞLn þ aj ðym ÞI m ; ð20Þ
ZΓ x n¼1
n am
þ aj ðxÞ ^ ½T ij ðx^ ; xÞ þ T ij ðx; x^ ÞdΓ x ; ð15Þ
Γx
in which
for interior problems and Z
Z Z Im ¼ ½T ij ðym ; xÞ þ T ij ðx; ym ÞdΓ m ; ð21Þ
t i ðx^ Þ ¼ ½aj ðxÞ  aj ðx^ ÞT ij ðx^ ; xÞ dΓ x þ aj ðx^ Þ ½T ij ðx^ ; xÞ þ T ij ðx; x^ ÞdΓ x ; Γm
Γx Γx
xn is the nodal point of the nth element, and Ln denotes the length of
ð16Þ
the element Γn. In practical computation, Ln can be approximated
for exterior problems. using the half distance between the (n 1)th and (nþ 1)th nodes. It is
Eqs. (15) and (16) constitute a compatibility relation between noticed that the above integral Im is regular and can be accurately
the boundary values of ti and the unknown density aj. At the same calculated by using the standard Gaussian quadrature. In this paper,
time, these two equations can be viewed as the integral equations we use the Trapezoidal quadrature formula to approximate the
on the boundary Γ, that is the boundary integral equations (BIEs), integral Im, which leads to a truly integration-free SBM.
with the unknown densities which need to be determined by Using the procedure described above, we have
2
prescribed boundary conditions.
We can observe from Eqs. (15) and (16) that the 1=jx^  xj diver- N 6
t i ðym Þ ¼ ∑ aj ðxn ÞT ij ðym ; xn ÞLn þ aj ðym Þ6
4τδij þ I m
gence of the kernel is now removed or smoothed out by the n¼1
relative quantity ½aj ðxÞ  aj ðx^ Þ and ½nj ðxÞ
^  nj ðxÞ. The whole integral, na m
therefore, is more amenable for numerical integration and can be 3
accurately calculated by using the standard Gaussian quadrature. N 7
While this reduction technique is similar to these proposed in þ ∑ T ij ðxn ; ym ÞLn 7
5; ð22Þ
n¼1
Refs. [30–32], the resulting form of the integrand is quite different. n am
The most striking difference is the use of the limit theorem
(Lemma 1) as oppose to use the jump property of the traction Combining integration weights Ln and values of aj into a new set of
fundamental solutions. aj, we have
N
t i ðym Þ ¼ ∑ aj ðxn ÞT ij ðym ; xn Þ þ aj ðym ÞBij ðxm Þ; ð23Þ
3.2. SBM formulation and its numerical implementation n¼1
na m
According to the above analysis, the regularized indirect BIEs where
for boundary tractions can be written as 2 3
Z Z
16 N 7
6τδ þ I m þ ∑ T ij ðxn ; ym ÞLn 7;
t i ðyÞ ¼ τai ðyÞ þ ½aj ðxÞ  aj ðyÞT ij ðy; xÞ dΓ þ aj ðyÞ ½T ij ðy; xÞ þ T ij ðx; yÞdΓ ; Bij ðxm Þ ¼ ð24Þ
Γ Γ Lm 4 ij n¼1
5
ð17Þ n am

where x,yA Γ and is the aforementioned origin intensity factor Bij(xm) for traction
( boundary equation (8).
1; for interior problems;
τ¼ ð18Þ When y-x, the origin intensity factor Aij has a weak singularity
0; for exterior problems
of order ðln rÞ. The calculation of Aij can be directly set as an
In the above equations we have written y instead of x^ , and for average value of the fundamental solution over a portion of the
convenience this notation will be used from here on. boundary. Algorithms relating to the direct evaluation of weakly
Eq. (17) can be split into singular integrals are widely available and more details can be
Z found, e.g., in Refs. [22,33,34]. Alternatively, the origin intensity
t i ðym Þ ¼ τai ðym Þ þ ½aj ðxÞ  aj ðym ÞT ij ðym ; xÞdΓ factor Aij can be calculated indirectly using the so-called “inverse
Z Γ  Γm
interpolation method” proposed in Ref. [35].
þ aj ðym Þ ½T ij ðym ; xÞ þ T ij ðx; ym ÞdΓ With the proposed treatment for singular terms, we can
Γ  Γm
Z Z develop a modification of the MFS for orthotropic elastic problems,
þ ½aj ðxÞ  aj ðym ÞT ij ðym ; xÞdΓ þ aj ðym Þ ½T ij ðym ; xÞ þ T ij ðx; ym ÞdΓ which can be summarized in the following steps:
Γm Γm
ð19Þ  Step 1. Distribute a number of N source points over the
where Γm is the segment on which the mth collocation point ym is boundary.
located, i.e., the segment that the fundamental solution will  Step 2. When xay, the corresponding elements of the coeffi-
present singularity. cient matrix are computed directly in a way similar to the MFS
The quintessence of the BEM is to discretize the boundary into a (see Eq. (23)).
finite number of segments, not necessarily equal, which are called  Step 3. When x ¼y, use the origin intensity factors described
boundary elements. Two approximations are made over each of these above to approximate the original singular terms.

Please cite this article as: Gu Y, et al. The singular boundary method: Mathematical background and application in orthotropic
elastic problems. Eng. Anal. Boundary Elem. (2014), http://dx.doi.org/10.1016/j.enganabound.2014.02.001i
 Y. Gu et al. / Engineering Analysis with Boundary Elements ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 5

For a well-posed boundary value problem, the unknown coeffi- where I knumerical and I kexact denote the numerical and analytical
cients fai ðxn ÞgN
n ¼ 1 can be determined by collocating N observation solutions at the kth calculation point, respectively.
points on the boundary conditions from Eq. (3a) for Dirichlet
problems and Eq. (3b) for Neumann problems. Once all coefficients
are computed, the displacements and stresses at any point inside the 4.1. An infinite plate with a circular hole
domain can be obtained directly from Eqs. (6a) and (25):
An infinite orthotropic plate with a circular hole subjected to the
N N
sij ðym Þ ¼ ∑ ak ðxn ÞDijk ðym ; xn Þ ¼ ∑ ½a1 ðxn ÞDij1 ðym ; xn Þ uniform tensile forces p¼ 1 at infinity is studied first, as illustrated in
j¼1 j¼1 Fig. 1. The radius of the circular hole is r¼1. The analytical solution
n m n
þ a2 ðx ÞDij2 ðy ; x Þ; ð25Þ corresponding to this problem can be found in Ref. [36]. The geometry
of the problem and the distribution of the source and collocation
where points for the SBM and MFS are given in Fig. 1(a) and (b).
pffiffiffiffiffiffi pffiffiffiffiffiffi!
A2 α1 A1 α2
D111 ðy; xÞ ¼ Kðx1  y1 Þ  ;
r 21 r 22
3.5
!
A1 A2
D112 ðy; xÞ ¼ Kðx2  y2 Þ pffiffiffiffiffiffi 2  pffiffiffiffiffiffi 2 ; ð26aÞ Material I: Analytical solution
α1 r 1 α2 r 2 Material I: SBM result
2.5
Material II: Analytical solution
pffiffiffiffiffiffi pffiffiffiffiffiffi! Material II: SBM result
A2 α1 A1 α2

Tangential stresses
D121 ðy; xÞ ¼ Kðx2  y2 Þ  ; Isotropic material
r 21 r 22 1.5
pffiffiffiffiffiffi pffiffiffiffiffiffi !
A2 α2 A1 α1
D122 ðy; xÞ ¼ Kðx1  y1 Þ  ; ð26bÞ
r 22 r 21 0.5

pffiffiffiffiffiffi pffiffiffiffiffiffi!
A1 α2 α2 A2 α1 α1 Material I: S11=1.25E-6, S22=2E-6
D221 ðy; xÞ ¼ Kðx1  y1 Þ  ; -0.5 S12=-1.25E-7, S66=3.69E-6
r 22 r 21 Material II: S11=2E-6, S22=1.25E-6
pffiffiffiffiffiffi pffiffiffiffiffiffi ! S12=-2E-7, S66=3.69E-6
A2 α2 A1 α1
D222 ðy; xÞ ¼ Kðx2  y2 Þ 2
 ð26cÞ -1.5
r2 r 21 0 10 20 30 40 50 60 70 80 90
Angle of boundary nodes (degree)
are stress fundamental solutions.
Fig. 2. Tangential stress sθ distribution along the edge of the hole.

4. Numerical results and discussions

Table 1
Results of tangential stress sθ along the line x2 ¼0.
To verify the methods developed above, three benchmark
numerical examples are studied in which the SBM solutions are x1 Exact SBM Relative errors of MFS
compared with the analytical solutions and/or results obtained
SBM results Relative error d ¼0.02 d ¼ 0.05
using the BEM and MFS. In the MFS, for simplicity the locations of
the source points are pre-assigned, taken to be a curve similar to 1.5 1.4979 1.4973 3.9506E  4 1.0444E  2 3.8736E 6
the real boundary, and assume that the number of source points is 2 1.2064 1.2064 1.8959E  5 7.8588E  3 2.6362E  6
equal to that of collocation points. The relative errors used is 2.5 1.1123 1.1125 9.9140E  5 5.6505E  3 1.0504E  6
defined by 3 1.0709 1.0710 1.0373E  4 4.1545E  3 1.0832E  6
3.5 1.0490 1.0491 9.2030E  5 3.1520E  3 9.5624E 6
2 !2 31=2 4 1.0360 1.0361 7.8395E  5 2.4617E  3 8.1261E  6
k k
1 M I  I 6.6260E  5 1.9709E  3 6.8587E  6
Relative error ¼ 4 ∑ numerical exact 5 ; ð27Þ 4.5 1.0276 1.0277
Mk¼1 I kexact 5 1.0219 1.0220 5.6178E  5 1.6112E  3 5.8100E  6

p =1 p =1

x2 x2

o x1 o x1

Fig. 1. The geometry of the problem and the distribution of the source ( ) and collocation ( ) points for (a) SBM and (b) MFS.

Please cite this article as: Gu Y, et al. The singular boundary method: Mathematical background and application in orthotropic
elastic problems. Eng. Anal. Boundary Elem. (2014), http://dx.doi.org/10.1016/j.enganabound.2014.02.001i
6 Y. Gu et al. / Engineering Analysis with Boundary Elements ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Here the stress of interest is tangential stress sθ near the edge edge of the hole for two different orthotropic materials. For the
of the hole, where as a number of solved problems show, it is the numerical implementation, N ¼200 evenly distributed source
greatest. Fig. 2 illustrates the variation of the stress sθ along the points are chosen along the boundary. As Fig. 2 shows that the
results predicted by the proposed SBM are in quite good agree-
2.5 ment with the analytical solutions. It can be also seen that, for all
3 possible tangential stresses the greatest one occurs in the case that
the tension is applied in a direction for which Young's modulus is
1.5 2.5 maximum (Sii is minimum), which is quite different from the
isotropic material. In such case the maximum stress in an
2 orthotropic plate (smax ¼ 3:27) is greater than that in a similar
isotropic plate (smax ¼ 3). Furthermore, for all cases discussed here
0.5
1.5 the most severely stressed regions, as would be expected, are near
the point (1,0) (the angular distance from this point does not
1 exceed 100).
-0.5 Table 1 illustrates tangential stress sθ distributions along the
0.5 line x2 ¼0 for various values of x1, with S11 ¼2E  6, S22 ¼ 1.25E  6,
S12 ¼  2E 7, S66 ¼3.69E  6, and N ¼ 200. It can be observed
-1.5 from this table that the SBM results agree pretty well with the
0
analytical solution, with the largest relative error less than 4E  4.
We also observe that the solution accuracy of the MFS with the
-0.5
fictitious boundary d ¼0.02 dramatically deteriorates compared
-2.5 with d ¼0.05, where d denotes the distance between the fictitious
-2.5 -1.5 -0.5 0.5 1.5 2.5
boundary and the real boundary. This shows that the appropriate
Fig. 3. Tangential stress sθ distribution inside the domain. or optimal placement of the fictitious boundary in the MFS
can result in a very accurate solution. However, the determining
0 of optimal fictitious boundary is a difficult task in real-world
10
applications. Fig. 3 presents the tangential stress distribution
SBM
MFS (d=0.05)
inside the domain, where the result curves were yielded at
10
-2
MFS (d=0.1) 60  60 calculation points uniformly-spaced over the square
Relative error

-4
10

Table 2
10
-6 Tangential stresses at interior points distributed along the circular with radius
R¼ 1.1 and center at the origin.

-8 Angle (deg) Exact solution SBM results Relative errors

10
0 1.0087 1.0107 1.9480E  3
15 0.9370 0.9389 1.9504E  3
-10
10 30 0.7343 0.7357 1.9638E  3
40 80 120 160 200
45 0.6069 0.6081 2.0023E  3
Number of boundary nodes 60 0.7341 0.7356 2.0430E  3
75 0.9383 0.9402 2.0585E  3
Fig. 4. Relative error curves of tangential stress sθ at interior points along the circle 90 0.1006 0.1008 2.0613E  3
r¼ 1.5.

x2

1.5
Stresses
Geometry
1

0.5

0
o
x1
-0.5

-1

-1.5
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Fig. 5. An infinite plate with a circular hole subject to normal pressure: (a) the geometry of the problem; (b) tangential stresss sθ distribution along the edge of the hole.

Please cite this article as: Gu Y, et al. The singular boundary method: Mathematical background and application in orthotropic
elastic problems. Eng. Anal. Boundary Elem. (2014), http://dx.doi.org/10.1016/j.enganabound.2014.02.001i
 Y. Gu et al. / Engineering Analysis with Boundary Elements ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 7

[  2.5,2.5]  [ 2.5,2.5]. As the number of boundary nodes 4.2. Normal pressure distributed uniformly along the edge of a hole
increases, Fig. 4 shows the relative error curves of the tangential
stresses sθ calculated at M ¼100 interior points along the circle As a further illustration we consider an infinite plate with a
r ¼ 1.5. hole subject to normal pressure p ¼1 distributed uniformly along

2 1.6 2 1

1.5 1.4 1.5 0.9

0.8
1 1.2 1
0.7
0.5 1 0.5
0.6
0 0.8 0
0.5
-0.5 -0.5 0.4
0.6

-1 -1 0.3
0.4

-1.5 0.2
-1.5
0.2
0.1
-2 -2
-2 -1 0.0 1.0 2.0 -2 -1 0 1 2

Fig. 6. Tangential stress sθ inside the domain for (a) orthotropic plate; and (b) isotropic plate.

x2

1.5

0.5
1
0.4

0.5 0.3

0.2
0
0.1
o
x1 0
-0.5

-0.1
-1
-0.2

-1.5 -0.3
-1.5 -1 -0.5 0 0.5 1 1.5

3 10
1.5
8
1
2
6

0.8 4
1 1
2
0.6
0 0

0.4 -2
-1 0.5
-4
0.2 -6
-2
-8
0
-3 -10 0
-3 -2 -1 0 1 2 3 -10 -5 0 5 10

Fig. 7. The geometry of the problem (a) and tangential stress distribution inside the domain (b–d), where the ratios are rb/ra ¼ 1.5 for (b), rb/ra ¼ 3 for (c), and rb/ra ¼ 10 for (d),
respectively. (a) The geometry of the problem, (b) rb/ra=1.5, (c) rb/ra=3 and (d) rb/ra=10

Please cite this article as: Gu Y, et al. The singular boundary method: Mathematical background and application in orthotropic
elastic problems. Eng. Anal. Boundary Elem. (2014), http://dx.doi.org/10.1016/j.enganabound.2014.02.001i
8 Y. Gu et al. / Engineering Analysis with Boundary Elements ∎ (∎∎∎∎) ∎∎∎–∎∎∎

the edge of the hole, as illustrated in Fig. 5(a). The radius of the computationally far more efficient, easier-to-program, and math-
circular hole is r ¼1. The analytical solution corresponding to this ematically simpler than the BEM. In this case, the average CPU-
problem can be found in Ref. [36]. times are 0.12 s for the SBM and 0.23 s for the BEM.
Table 2 shows the numerical results of tangential stress sθ at
interior points distributed along the circular with radius R ¼1.1
and center at the origin. The elastic constants are S11 ¼1.61E  6, 5. Conclusions
S22 ¼1.76E  6, S12 ¼  1.61E  7, S66 ¼8.33E  6 and the number of
boundary nodes is taken to be N ¼300. We can observe from this This paper discusses the mathematical background of the SBM
table that the SBM results are in good agreement with the and documents the first attempt to apply the method to plane
analytical solution, with the largest relative error less than 2.1E  3. orthotropic elastic problems. A limiting theory, based on the bound-
Fig. 5(b) plots the variation of tangential stresses around the ary elements formulation, is presented to guarantee the existence
boundary. The intensity of the stress is represented by distances and uniqueness of the origin intensity factors. The method is still
outward from the circular hole along lines through the center of truly meshless and easy to implement. Some key features of the
the hole. Note that the maximum stress occurs at regions near the proposed SBM in comparison with the FEM, BEM and MFS can be
direction of x1 and x2 axes, which agrees well with the distribution summarized as
illustrated in Ref. [36], (Fig. 53). In addition, the tangential stress
predictions inside the domain [ 2,2]  [ 2,2] are provided in (i) It is meshless which means that no mesh, but a mere
Fig. 6(a). The results for an isotropic plate are also given for the collection of “nodes” is required. The nodes need not have
purpose of comparison, see Fig. 6(b). It can be observed that the any connectivity, and the trial functions are built entirely in
maximum stress in an orthotropic plate is, in general, greater than terms of nodes.
that in a similar isotropic plate. (ii) It requires no numerical integration, and the troublesome
singular integrals can be avoided.
(iii) It is a boundary method which means that it shares all the
4.3. A thick-walled cylinder advantages that the BEM has over domain discretization
methods, such as the FEM and FDM.
Finally, consider the problem of a circular orthotropic ring, fixed (iv) It overcomes the fictitious boundary issue associated with the
on its outer boundary, subjected to a uniform normal pressure p¼ 1 traditional MFS.
on its inner boundary, as illustrated in Fig. 7(a). The inner and outer
radii of the cylinder are ra ¼1 and rb, respectively. Here, each Overall, it can be concluded that the proposed scheme is
boundary is divided into 150 equal intervals. The material constants computational efficient, robust, accurate and in comparison with
are the same as those used in example 2. existing methods for solving numerically certain boundary value
Since analytical solutions for this problem are not available, the problems, it could be considered as a competitive alternative.
problem is solved for an increasing sequence of values of the outer
radius with a final ratio rb/ra ¼ 2000. The final ratio approximates an
infinite plate with a circular hole subjected to uniform normal Acknowledgments
pressure which is solved in Section 4.2. The distribution of tangential
stress sθ inside the domain is studied and three of the distributions The work described in this paper was supported by the
are shown in Fig. 7(b–d). It can be seen from these figures that, as the National Basic Research Program of China (973 Project no.
ratio rb/ra increases, the stress distribution gradually approaches the 2010CB832702), the National Science Funds for Distinguished
limiting case, i.e., Fig. 6(a). Young Scholars of China (11125208), the R & D Special Fund for
The results of tangential stress sθ from the present method are Public Welfare Industry (Hydrodynamics, Project no. 201101014),
compared in Fig. 8 with the results obtained by the BEM [30], the 111 project under grant B12032, Jiangsu Province Graduate
where the results curves were yielded at interior points distrib- Students Research and Innovation Plan (No. CXZZ11_0424), and
uted along the circles R ¼1.1 and R¼1.2 with rb/ra ¼ 2000. It can be the project from Sandia National Laboratories (Project no. DE-
seen that the results predicted by the proposed method are in AC04-94AL85000).
close agreement with the BEM solutions. It is worth noting that
the proposed SBM is inherently free of mesh and integration and is
Appendix A. Detailed derivations of Eq. (14)
1.05
SBM (R=1.1)
BEM (R=1.1) (Ref. [29])
According to the direct boundary integral equation, we have
1
SBM (R=1.2)
Z
0.95 BEM (R=1.2) (Ref. [29]) τuj ðym Þ ¼ ½U ij ðx; ym Þt i ðxÞ T ij ðx; ym Þui ðxÞdΓ ðxÞ; ðA1Þ
Γ
Tangential stresses

0.9
where
0.85
i; j ¼ 1; 2
0.8
and
(
0.75
1; y A Ω;
τ¼ ðA2Þ
yAΩ :
C
0.7 0;
0.65
The simplest solution to the problem is to choose rigid-body
0.6 displacements of the whole body in the directions of the coordi-
0 10 20 30 40 50 60 70 80 90
nate axes, such as u1(x) ¼1, u2(x)¼0, t1(x)¼ t2(x)¼0. After the
application of this elementary solution, Eq. (A.1) becomes
Angle of boundary nodes (degree) Z Z
Fig. 8. Results of tangential stress sθ at interior points along the circles R¼ 1.1 and T 11 ðx; ym Þ dΓ ðxÞ ¼  τ; T 12 ðx; ym Þ dΓ ðxÞ ¼ 0; ðA3Þ
R¼ 1.2, calculated using the SBM and BEM, respectively. Γ Γ

Please cite this article as: Gu Y, et al. The singular boundary method: Mathematical background and application in orthotropic
elastic problems. Eng. Anal. Boundary Elem. (2014), http://dx.doi.org/10.1016/j.enganabound.2014.02.001i
 Y. Gu et al. / Engineering Analysis with Boundary Elements ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 9

In a similar way, let u1(x)¼ 0, u2(x)¼1 and t1(x)¼t2(x) ¼0, [15] Rizzo FJ, Shippy DJ. A method for stress determination in plane anisotropic
Eq. (A1) becomes elastic bodies. J Compos Mater 1970;4(1):36–61.
Z Z [16] Young DL, Jane SJ, Fan CM, Murugesan K, Tsai CC. The method of fundamental
solutions for 2D and 3D Stokes problems. J Comput Phys 2006;211(1):1–8.
T 21 ðx; ym ÞdΓ ðxÞ ¼ 0; T 22 ðx; ym ÞdΓ ðxÞ ¼  τ ðA4Þ [17] Karageorghis A, Fairweather G. The method of fundamental solutions for the
Γ Γ
numerical solution of the biharmonic equation. J Comput Phys 1987;69(2):
We can rewrite Eqs. (A3) and (A4) as follows: 434–59.
Z [18] Li Z-C, Mathon R. Boundary approximation methods for solving elliptic
T ij ðx; ym ÞdΓ ðxÞ ¼  δij τ; ðA5Þ problems on unbounded domains. J Comput Phys 1990;89(2):414–31.
Γ [19] Šarler B, Liu QG. Non-singular method of fundamental solutions for two-
dimensional isotropic elasticity problems. Comput Model Eng Sci 2013;91(4):
which is Eq. (14) in Section 3.1. 235–66.
The rigid-body motion has many advantages over other possi- [20] Karageorghis A, Lesnic D. The method of fundamental solutions for the inverse
ble choices. No stresses or tractions are generated in the body, thus conductivity problem. Inverse Probl Sci Eng 2010;18(4):567–83.
[21] Johansson BT, Lesnic D, Reeve T. A method of fundamental solutions for the
no integrals of Uij occur in the boundary integral equation (A.1).
radially symmetric inverse heat conduction problem. Int Commun Heat Mass
Moreover, due to the fact that only one displacement component Transf 2012;39(7):887–95.
is non-zero in each elementary solution, only the sum of one Tij [22] Sarler B. Solution of potential flow problems by the modified method of
appears in (A1). The rigid-body motion approach was first pro- fundamental solutions: formulations with the single layer and the double
layer fundamental solutions. Eng Anal Bound Elem 2009;33(12):1374–82.
posed by Lachat and Watson in their advanced BEM implementa-
[23] Liu YJ. A new boundary meshfree method with distributed sources. Eng Anal
tion [37]. Some remarks on the correct application of such Bound Elem 2010;34(11):914–9.
approach in case of integral equations can be found in Refs. [24] Li Z-C. Combinations of method of fundamental solutions for Laplace's
[38,39]. equation with singularities. Eng Anal Bound Elem 2008;32(10):856–69.
[25] Marin L. Regularized method of fundamental solutions for boundary identi-
fication in two-dimensional isotropic linear elasticity. Int J Solids Struct
References 2010;47(24):3326–40.
[26] Chen KH, Chen JT, Kao JH. Regularized meshless method for solving acoustic
eigenproblem with multiply-connected domain. Comput Model Eng Sci
[1] Cheng AHD, Cheng DT. Heritage and early history of the boundary element
2006;16(1):27–39.
method. Eng Anal Bound Elem 2005;29(3):268–302.
[27] Perne M, Šarler B, Gabrovšek F. Calculating transport of water from a conduit
[2] Cheng AHD, Chen CS, Golberg MA, Rashed YF. BEM for theomoelasticity and
to the porous matrix by boundary distributed source method. Eng Anal Bound
elasticity with body force — a revisit. Eng Anal Bound Elem 2001;25(4–5):
377–87. Elem 2012;36(11):1649–59.
[3] Gu Y, Chen W, Zhang C. Stress analysis for thin multilayered coating systems [28] Banerjee PK. The boundary element methods in engineering. London:
using a sinh transformed boundary element method. Int J Solids Struct McGraw-Hill Book Company Europe; 1994.
2013;50(20–21):3460–71. [29] Brebbia CA, Telles JCF, Wrobel LC. Boundary element techniques: theory and
[4] Fairweather G, Karageorghis A. The method of fundamental solutions for applications in engineering. New York: Springer-Verlag; 1984.
elliptic boundary value problems. Adv Comput Math 1998;9(1):69–95. [30] Guiggiani M, Krishnasamy G, Rudolphi TJ, Rizzo FJ. A general algorithm for the
[5] Chen CS, Golberg MA, Hon YC. The method of fundamental solutions and numerical solution of hypersingular boundary integral equations. J Appl Mech
quasi-Monte-Carlo method for diffusion equations. Int J Numer Methods Eng 1992;59(3):604–14.
1998;43(8):1421–35. [31] Sladek V, Sladek J. Non-singular boundary integral representation of stresses.
[6] Karageorghis A, Fairweather G. The method of fundamental solutions for Int J Numer Methods Eng 1992;33(7):1481–99.
axisymmetric elasticity problems. Comput Mech 2000;25(6):524–32. [32] Karami G, Derakhshan D. An efficient method to evaluate hypersingular and
[7] Young DL, Chen KH, Lee CW. Novel meshless method for solving the potential supersingular integrals in boundary integral equations analysis. Eng Anal
problems with arbitrary domain. J Comput Phys 2005;209(1):290–321. Bound Elem 1999;23(4):317–26.
[8] Gáspár C. Some variants of the method of fundamental solutions: regulariza- [33] Watson JO. Advanced implementation of the boundary element method for
tion using radial and nearly radial basis functions. Cent Eur J Math 2013;11(8): two- and three-dimensional elastostatics. Developments in boundary element
1429–40. methods – 1. London: Elsevier Applied Science Publishers; 1979; 31–63.
[9] Gu Y, Chen W, Zhang C-Z. Singular boundary method for solving plane strain [34] Cruse TA. An improved boundary-integral equation method for three dimen-
elastostatic problems. Int J Solids Struct 2011;48(18):2549–56. sional elastic stress analysis. Comput Struct 1974;4(4):741–54.
[10] Gu Y, Chen W, He X-Q. Singular boundary method for steady-state heat [35] Gu Y, Chen W, He XQ. Domain-decomposition singular boundary method for
conduction in three dimensional general anisotropic media. Int J Heat Mass stress analysis in multi-layered elastic materials. CMC: Comput Mater Contin
Transf 2012;55(17–18):4837–48. 2012;29(2):129–54.
[11] Gu Y, Chen W. Infinite domain potential problems by a new formulation of [36] Lekhnitskii SG. Theory of elasticity of an anisotropic elastic body. San
singular boundary method. Appl Math Model 2013;37(4):1638–51.
Francisco: Holden-Day Series in Mathematical Physics; 1963.
[12] Chen W, Gu Y. An improved formulation of singular boundary method. Adv
[37] Lachat JC, Watson JO. Effective numerical treatment of boundary integral
Appl Math Mech 2012;4(5):543–58.
equations: a formulation for three-dimensional elastostatics. Int J Numer
[13] Azis MI, Clements DL. A boundary element method for anisotropic inhomo-
Methods Eng 1976;10(5):991–1005.
geneous elasticity. Int J Solids Struct 2001;38(32–33):5747–63.
[38] Guiggiani M, Casalini P. Rigid-body translation with curved boundary ele-
[14] Benitez FG, Wideberg J. The boundary element method based on the three-
dimensional elastostatic fundamental solution for the orthotropic multi- ments. Appl Math Model 1989;13(6):365–8.
layered space: application to composite materials. Comput Mech 1996;18(1): [39] Guiggiani M. The evaluation of cauchy principal value integrals in the boundary
24–45. element method—a review. Math Comput Model 1991;15(3–5):175–84.

Please cite this article as: Gu Y, et al. The singular boundary method: Mathematical background and application in orthotropic
elastic problems. Eng. Anal. Boundary Elem. (2014), http://dx.doi.org/10.1016/j.enganabound.2014.02.001i