You are on page 1of 14

ON THE BEM ALGORITHMS FOR THREE-DIMENSIONAL

ELASTOSTATIC PROBLEMS

Alisson P. Chaves
Gabriel O. Ribeiro
Departamento de Engenharia de Estruturas – Universidade Federal de Minas Gerais
Av. do Contorno, 842, 2o andar, Centro – 30110 060 – Belo Horizonte – MG, Brazil
apchaves@mail.com ; gabriel@dees.ufmg.br

Abstract. The choice for simple and efficient algorithms must be a concern when a boundary
element method (BEM) program is being developed. The integration algorithms should use
techniques which provide accurate results with minimum computational effort. This work
deals with some techniques used for the evaluation of integrals involved in three-dimensional
BEM for elastostatics. In the classical BE implementation, Cauchy principal value (CPV)
integrals have been evaluated through the use of well-known rigid body motion concept. The
remaining integrals, which are regular or weakly-singular have been performed by either one
analytical or two numerical algorithms. One of the numerical approaches makes use of the
Hammer’s scheme associated with the technique of element subdivision, and the other adopts
a degenerate mapping of a triangular domain into a square one, followed by the standard
Gaussian quadrature. Furthermore, a non-singular BEM algorithm that considers the source
points placed outside the solid domain has been implemented. Displacements at internal
points are evaluated using the classical form and a self-regular form of the Somigliana’s
identity. The last one is supposed to eliminate nearly-singular behavior from integrands.
Numerical examples are presented in order to check the accuracy and to compare the
efficiency of the implemented algorithms.

Keywords: Boundary element method, Three-dimensional elastostatics, Self-regular


formulation, external collocation point.
1. INTRODUCTION

The boundary element method (BEM) has been successfully applied for the analyses of a
wide range of engineering problems, including structural ones. This paper is concerned with
the BEM for three-dimensional elastostatic problems in solid mechanics. In elastostatic
problems, the standard boundary integral equation (BIE) is usually derived from the
Somigliana displacement identity (SDI), using the fundamental solution due to Kelvin. This
fundamental solution for displacements (Uji) and the corresponding fundamental traction
solution (Tji), are the kernel functions of the BIE integrals. In three-dimensional elastostatic,
Kelvin’s solution are proportional to the singular functions 1/r and 1/r², respectively.
Therefore, the integrals involved may be regular, weakly-singular or strongly-singular, which
should be understood in the sense of Cauchy principal value (CPV).
There are many algorithms capable to evaluate singular integrals, with different levels of
accuracy and computational efficiency. An analytical integration approach, available for flat
linear triangular elements, was first presented by Cruse (1974), and later by Li et al. (1985)
and Milroy et al. (1997). However, the most used and investigated approaches are the
numerical ones, which are based on Gaussian quadratures. Transformation techniques, using
series expansion and subtraction of singularity, have been demonstrated to be the most
efficient approach to deal with CPV integrals, and even with the hypersingular ones, as done
by Guiggiani (1998) and Doblaré & Gracia (1997). Some other proposed methods for CPV
integrals evaluation do not work as the authors have claimed. Huang & Cruse (1993) have
pointed out some flaws in the algorithms presented by Cerrolaza & Alarcon (1989) and Li et
al. (1992). Nevertheless, for the elastostatic problems, the direct evaluation of CPV integrals
can be avoided by using the rigid body motion concept, proposed by Cruse (1974). Hence,
only the remaining regular and weakly-singular integrals have to be calculated. Standard
Gaussian quadrature do not provide accurate results for weakly-singular integrals evaluation,
thus, special algorithms must be used. An usual and efficient procedure is to reduce the order
of the singularity, and then perform the integration over the regularized function, as Li et al.
(1985).
Regular integrals can be evaluated accurately through Gaussian methods, as for instance
Hammer’s method for triangular domains. This method provide accurate results if the
collocation point is far enough from the element, otherwise, nearly-singular behavior may
appear, and special techniques are needed to deal with this problem. A common device
presented by Lachat & Watson (1976) uses the element subdivision to improve the integration
results, known as h-method. However, when the collocation point is close to the element, (e.g.
internal point close to the boundary) this strategy is not efficient, due to the nearly-singular
behavior of the integrand. The use of cumbersome techniques to perform the nearly-singular
integrals, in the evaluation of displacements at internal points, may be dismissed, if the self-
regularized form of the Somigliana displacement identity is used.
An alternative BEM algorithm may be applied in order to deal only with regular integrals.
It consists in setting collocation points outside the boundary of the problem. This collocation
points must be placed keeping away from the boundary, so that the nearly-singular integrals
will not occur. On the other hand, if this collocation point is too far from the boundary,
problems related to numerical condition of the system may arise.
A program has been implemented in order to evaluate some aforementioned algorithms.
Rigid body motion concept has been used to determine the CPV integrals indirectly.
Hammer’s method, as well as a degenerate mapping algorithm have been applied in the
evaluation of regular and weakly-singular integrals. The alternative algorithm based on BIE
using external collocation points has also been implemented. Displacements at internal points
have been determined in a post-processing stage, using the classical form and the self-regular
form of the Somigliana’s identity. Numerical examples have been processed in order to check
and compare the accuracy and efficiency of the algorithms.

2. BEM APPROACHES FOR 3-D ELASTOSTATIC PROBLEMS

The boundary integral equation for elasticity problems, based on Somigliana’s identity
for displacements is classically written as (Cruse, 1974):

C ji ( P ) u i ( P ) + ∫T
<Γ >
ji ( P , Q ) u i (Q ) dΓ (Q ) =
∫U
Γ
ji ( P , Q ) t i (Q ) dΓ (Q ) (1)

where the integral on the left may be understood as a CPV integral. The term r(P,Q) denotes
the distance of an evaluation point Q(x) from the collocation point P(x). Fundamental solution
for displacement and traction are represented by Uji(P,Q) and Tji(P,Q), respectively. The
matrix Cji is derived from discontinuous integral terms by using limit process when the
internal point p(x) is taken to the boundary.
Numerical treatments of a problem require the boundary to be represented by surface
elements mesh. The boundary data (geometry, displacements and traction) are represented, by
shape functions Nα(ξ), and nodal values ( xiα , u iα and t iα ) in the elements. Each integral in Eq.
(1) is the sum of integrals over the boundary elements. The discrete form of Eq. (1) may be
written as:

NE n
C ji ( P ) u i ( P ) + ∑∑u
β =1 α =1
α
i (Q ) ∫T ji ( P , Q ) N
α
J A dξ β ( Q ) =
< ξβ >
NE n
(2)
= ∑∑
β =1 α =1
t iα (Q ) ∫U ji ( P , Q )
α
N J A dξ β (Q )
ξβ

where JA is the Jacobian relating the global coordinate system to the homogeneous one. The
integrals in Eq. (2) may be regular, weakly-singular and strongly-singular. Following usual
procedures (Brebbia et al.,1984), integrals on the left and the free-term in Eq.(2) are grouped
in a global matrix H, and the integrals on the right, form coefficients of the global matrix G.
Nodal values of displacements and tractions are written as vectors u and t, respectively. Thus,
Eq. (2) may be represented by the equation H u = G t. Through a proper permutation of
terms, the unknown values are grouped in a vector y and its coefficients in a matrix A.
Similarly, the known boundary data multiplied by related coefficients (from G or H) are
grouped in a vector f. Therefore, a system of linear equations is derived (A y = f), from which,
the unknown boundary values y can be achieved.

2.1 Evaluating the integrals

The strongly-singular integrals and the free-term Cji can be easily determined by using
the rigid body motion concept (Cruse, 1974). These contributions form the coefficients of the
main diagonal submatrices of the matrix H, which can be expressed as the sum of the
coefficients of the off-diagonal submatrices of H. These off-diagonal submatrices have
regular and weakly-singular terms, which have to be evaluated accurately.
Triangular linear and quadratic elements have been considered herein. To perform the
integration over these elements, three different approaches have been used. Analytical
integration has been used to perform weakly-singular integrals over triangular linear flat
elements, according to Li et al. (1985) paper. The weakly-singular integrals have also been
performed numerically, for elements with linear and quadratic approach. This numerical
technique makes use of a coordinate transformation, named triangle polar coordinates (Li et
al., 1985). Through this transformation (degenerate mapping) a triangular domain is mapped
onto a square of unit side-length (Fig. 1). The originally weakly-singular function becomes
regular, so that the integration can be performed accurately by means of Gauss-Legendre’s
quadrature.

Figure 1- Triangular polar coordinates (Degenerate Mapping).

This technique may also be used for quadratic isoparametric elements, as shown by Li
et al. (1985). The sequence of mappings used herein starts with a transformation of a curved
triangular space (xi) into a plane triangular domain (ξi). The singular node must be a corner
node, otherwise, if the singular node is a mid-node, the flat triangle is divided into two new
triangles which have a singular node at its corners. Following the procedures for flat linear
elements, the flat triangle is mapped onto a square (ρi), which is re-mapped onto a Gauss-
Legendre’s domain (ηi ∈[-1,1]), where the integration is performed. Figure 2 illustrates this
sequence.

Figure 2- Sequence of mapping.

Most of the integrals in a BEM analysis are fully regular. In such case, standard Gaussian
integration quadratures are capable to evaluate accurately these integrals. In this paper the
well known Gaussian scheme adapted for triangular domains, called Hammer’s quadrature,
has been applied. When the collocation point is near to the element under integration, more
integration points are required in order to preserve the accuracy (Lachat & Watson, 1976). An
efficient way to do so is to adopt an element subdivision algorithm. The algorithm used
herein, divide the triangular element into NS equal triangular subelements (Fig. 3), over
which, the integrals are performed. The relation between the local element coordinate system
and the subelement coordinate systems is given by (Nakaguma, 1979).
 Q (ξ1 )   ξ1nod 1 ξ1nod 2 ξ1nod 3   Q(η1 ) 
   nod 1 nod 2 nod 3   
 Q (ξ 2 )  =  ξ2 ξ2 ξ2  ⋅  Q (η 2 )  (3)
 Q (ξ 3 )   ξ 3nod 1 ξ 3nod 2 ξ 3nod 3   Q(η3 ) 
    

Figure 3- Element subdivision (Subelements).

An alternative approach to perform regular integrals based on the aforesaid degenerate


mapping technique has also been implemented. More integration points have been used when
the collocation point was near to the element under integration. The point A(xi) – singular
point in Fig. 1 – has been assumed to be the nearest element node from the collocation point,
in such way that integration points are concentrated towards this node, as shown in Fig. 4.

a) Linear triangular element (singular point b) Quadratic triangular element (singular point
near to a corner node). near to a mid node).

Figure 4- Integration points clustered towards collocation point.

2.2 External Collocation Point Algorithm

When collocation points are placed outside boundary domain the free-term Cji is zero.
Therefore, the BIE (Eq. 1) is formed only by the integral terms, which are fully regular. These
integrals have been evaluated either using Hammer’s quadrature with element subdivision or
the degenerate mapping technique combined with Gauss-Legendre’s quadrature.
The position of the external point is a concern in this algorithm, in order to avoid nearly-
singular behavior and other possible computational troubles related to complex boundaries.
Herein, the collocation points, corresponding to each boundary node, have been placed in the
direction of the average normal of the elements sharing the node. The distance from the
boundary is given by a parameter α (Fig. 5), which multiplies the average of the lengths of the
straight edges connecting the element vertices ( L ).
Figure 5- External collocation point position.

2.3 Post-processing stage

The stresses components at boundary nodes can be readily evaluated by applying the
approach described in Brebbia et al. (1984). These stresses components are calculated in a
local coordinate system, using the displacements and tractions previously determined. No
further integrals need to be evaluated.
Displacements at internal points are evaluated using the classical Somigliana’s identity
(SDI) in a post-processing stage. The involved integrals are regular, what means that
Hammer’s quadrature or the degenerate mapping technique can be used. Nevertheless, when
the internal point is close to the boundary, the integrals may present a nearly-singular
behavior. Either of the used integration algorithms may fail when dealing with these integrals.
The lack of accuracy observed when the collocation point p is moved towards the boundary is
shown by Cruse & Aithal (1993). The nearly-singular behavior of an integrand may be
observed in Fig. 6 below, that shows a square domain resulting from the degenerate mapping
of a triangular element, for a singular 1/r² kernel function. Standard Gaussian quadrature is
unsuitable to perform accurately this type of function.

3 5
0.4 3 4
1 2 1 1 3 1
2
0.2 0.8 2
1 0.8 1 0.8 0.8
0 1
0.6 0 0.6 0 0.6 0 0.6
0 0 0 0
0.2 0.4 0.2 0.4 0.2 0.4 0.4
0.2
0.4 0.4 0.4 0.4
0.6 0.2 0.2 0.2
0.6 0.6 0.6 0.2
0.8 0.8 0.8 0.8
10 10 10 10
r(p,Q) = 10 L r(p,Q) = 1.0 L r(p,Q) = 0.25 L r(p,Q) = 0.125 L

Figure 6- Nearly-singular behavior arising when p approach the element centroid.

In order to evaluate the displacements at points close to the boundary, using ordinary
Gaussian quadrature algorithms, a self-regularized form of the SDI may be used. This
formulation is supposed to eliminate the nearly-singular behavior from the integrals, what
obviates the need for sophisticated or cumbersome integration methods. The self-regular form
of the Somigliana’s identity for displacements, obtained by applying a simple solution –
corresponding to a rigid body motion – to the ordinary SDI, is given in Eq. 4 (Cruse &
Richardson, 1996). The point P in Eq. 4 has been taken as the nearest boundary node from the
internal point p.


u i ( p) − u i ( P) = − T ji ( p, Q ) [u i (Q ) − u i ( P)] dΓ(Q ) + U ji ( p, Q ) t i (Q ) dΓ(Q )
Γ

Γ
(4)
3. NUMERICAL EXAMPLES

Two examples were processed using the implemented code. Thirteen integration points
have been used to perform the integrals by means of Hammer’s quadrature, while the number
of subelements (NS) varies according to the distance (d) from the collocation point to the
nearest element node or centroid. The number of subelements is given by:

 3 L ≥ d , NS = (NDIV)2  
 ; NDIV =  round 10  ; α=d; NS ≤ 256 (5)
 3 L < d , NS = 1  α  L

The amount of integration points used to perform the integration by means of the
degenerate mapping approach combined with the Gauss-Legendre quadrature varies
according to the function:

 d ≥ 10 L , NP = (0.5 NIP )
2

 10 L > d ≥ 3L , (NP = NIP )2 ; α=
d
; NP ≤ (48)²
 (6)
 3L > d ≥ L , (NP = 1.2 NIP )2 L
 L > d , (NP = (1.20 NIP) / α )2

where NIP (herein fixed as eight) is a parameter which denotes the number of integration
point used to perform the weakly-singular integrals.
These criteria, determined experimentally, are supposed to keep the integration errors
under 0.001% when r(P,Q) is larger than 0.3 L . Figure 7 shows the number of integration
points required in order to achieve results with this level of accuracy, when the point p is
moved towards the centroid in the normal direction. These results were calculated for a
specific triangular element (vertices (0,0,0);(1,0.5,0);(0,1,0)), in which the integrals had the
indicated kernel functions.

500 Hammer's Quadrature with Element 1000 Degenerate Mapping combined with Gauss-
450 Subdivision - error < 0,001% 900 Legendre's Quadrature
400 800 error < 0,001%
Integration Points

Integration Points

350 Used Criterion U11 700


300 U13 T11 600 Used Criterion U11
250 T13 D111 U13 T11
500
D223 S111 T13 D111
200 400
S223 D223 S111
150 300
S223
100 200
50 100
0 0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
(Dist. from centroid) / (average of edges) (Dist. from centroid) / (average of edges)

Figure 7- Number of integration points.

The main objective of these criteria is to set the number of integration point, in order to
limit the numerical integration error below a specific value, although better critera (less
conservative) may be found in Lachat & Watson (1976) and Bu & Davies (1985).
During the evaluation of displacements and tractions at the boundary, the computations
have been done using three alternatives, i.e. integration using analytical algorithm to perform
weakly-singular integrals and Hammer’s quadrature with element subdivision (HQES) to
perform the regular ones; integration using HQES to perform all integrals; or degenerate
mapping approach, combined with Gauss-Legendre’s quadrature to perform all integrals.

3.1 Fracture Mechanics Test Specimen

A standard fracture mechanics compact test specimen has been considered. The specimen
is stressed by a static load which simulate the well-known mode I (opening mode). Due to the
symmetry of the problem, it is possible to use ordinary BEM formulations for the analysis of
this problem (Cruse & Vanburen, 1971). The dimensions of the specimen are shown in Fig. 8,
and the material properties and load as well.

Dimensions: Material: Load:

W = 100 E = 205,000 P = 3.0 x 105


H = 60 ν = 0.3 σxy = P/TH = 100
T = 50
a = 50

Figure 8- Standard fracture mechanics specimen.

Two usual fracture mechanics parameters, the crack opening displacement (COD) and the
stress intensity factor (KI), are analyzed, both of interest in order to evaluate the behavior of
cracked (Aliabadi & Rooke, 1991). The computation of stress intensity factor can be done by
means of linear extrapolation technique, according to Eq. (7) shown in Aliabadi & Rooke,
1991:

Figure 9- Parameters for the evaluation of KI.

KI =
G 2π b
κ +1 r
[ ]
u y (θ = π) − u cy (θ = − π) , (7)

where G is the shear modulus; κ = (3-4ν) for plane strain and r is the distance of the points (b
and c) from the crack tip (adopted 10% of the crack length a). Figure 10 illustrates the used
BE model. The load P has been simulated by means of a uniform distribution of tangential
load σxy.
Figure 10- BEM symmetric model.

In order to simulate plane strain in the analyses, the displacements uz have been restrained
at the planes z = T/2 and z = -T/2. Numerical results of COD and KI are shown in Tab. 1
beside the reference answer presented by Tada et al. (2000). Such results are in good
agreement with the reference values, especially for refined meshes. Numerical integration
have been performed by means of the degenerate mapping combined with Gauss-Legendre’s
quadrature.

Table 1. Values of COD and KI for plane strain

Quadratic Quadratic
Plane Strain Linear Elements Linear Elements
Elements Elements
Paramet 833 1009 Error 634 Error 974 Error
(TADA, et al.) Error %
er nodes nodes % nodes % nodes %
COD 0.98517 1.01342 2.87 0.98661 0.15 0.97920 -0.61 0.98176 -0.35
KI 5795.45 4832.41 -16.62 5742.18 -0.92 5673.77 -2.10 5779.95 -0.27

Reference values for an effective three-dimensional analysis have been evaluated using
FRANC3D code, which is based on BEM algorithm and fracture mechanics procedures
according to Lutz et al. (1992). The reference value for stress intensity factor, obtained by
using triangular elements, is KI = 6356 at the plane z = 0. The numerical results obtained in
this work, are shown in Tab. 2, where one can note the good agreement between the answers
and the reference value. The algorithm using Hammer’s quadrature with subelements for the
evaluation of all integrals provides the less accurate results.

Table 2. Stress intensity factor

Num. Analytical Integr. of Degenerate mapping Hammer’s quadrat. with


Elements nodes weakly-sing. integrals Gauss-Legendre subelements
KI difer. % KI difer. % KI difer. %
Linear 859 6039.1 -4.99 6039.1 -4.99 6028.7 -5.99
Linear 1408 6273.4 -1.30 6273.4 -1.30 6266.4 -1.41
Quadratic 822 - - 6359.4 -0.05 6327.9 -0.44
Quadratic 1862 - - 6355.9 -0.002 6339.6 -0.26
Reference values (FRANC3D): KI = 6356

The distribution of KI along the crack front (-T/2 < z < T/2) is presented in Fig. 11 for
most refined meshes. These degenerate mapping combined with Gauss-Legendre quadrature
have been employed to evaluate all integrals. The numerical answers of this paper are in close
agreement with the numerical results obtained with FRANC3D, mainly the answers achieved
using quadratic triangular elements.

Stress Intensity Factor


KI along the Crack Front 6400

6300

6200

6100

6000
KI

FRANC3D (5224 Triangular elements) 5900


Quadratic Triangular Elements 5800
Linear Triangular Elements
FRANC3D (Using Symmetry) 5700

5600
-25 -20 -15 -10 -5 0 5 10 15 20 25
Z co-ordinate

Figure 11- Stress Intensity Factor along the crack front.

Displacements at nodes placed over the crack surface along the x-axis have also been
achieved using the BEM algorithm which considers external collocation points. Different
values for the parameter α, which determines the distance from the collocation point to the
correspondent boundary node, have been considered in different analyses. Hammer’s
quadrature with element subdivision has been applied to evaluate all integrals and the
numerical results are shown in Fig. 12.

Displacements Uy along the crack symmetry plane Displacements Uy along the crack symmetry plane
1 Linear Elements 1.0 Quadratic Elements

0.8
α = 0.1 0.8
0.6 α = 0.3 α = 0.1
Displacements u y

α = 0.5
Displacement u y

0.6
0.4 α = 0.3
α = 0.7 α = 0.5
0.2 0.4
α = 0.9
α = 0.7
0
0.2 α = 0.9
0 25 50 75 100
-0.2
0.0
-0.4
0 25 50 75 100
-0.6 -0.2
X co-ordinate X co-ordinate

Figure 12- Displacements along the crack-surface symmetry axis (external collocation point).

When the degenerate mapping algorithm is used to perform the integrals, similar results
are obtained. Some poor answers for displacements have been observed in Fig. 12 for some
values of α, either using linear elements or quadratic ones. Critical values of α for linear
elements are not necessarily the same for quadratic elements.
The capability of the BEM to represent high variation of σyy (singular stress field) at the
region near to the crack front may be noted in Fig. 13. Results for stress σyy, using the
implemented code can be seen for different algorithms. The results obtained from FRANC3D
analysis differ slightly from the results obtained from the implemented code, possibly due to
the use of discontinuous elements at the crack front, by FRANC3D. Despite this, both are
properly representing the stress concentration.
σ yy at the crack surface (y=0;z=0) σ yy at the crack surface (y=0;z=0)
1500 Linear elements 1500 Quadratic elements

0 0

-1500 -1500
-3000

σ yy
σ yy

-3000
Degenerate mapping integr. approach
-4500 Analytical Integration
-4500 Hammer's quadrature with elem. subdivision
Hammer's quadrature with elem. subdivision
-6000 Extenal collocation point (α = 0.1)
Extenal collocation point (α = 0.1) -6000 FRANC3D (Linear Elements, 2102 nodes)
-7500 FRANC3D (Linear Elements, 2102 nodes)
-7500
-9000
50 60 70 80 90 100
50 60 70 80 90 100
X co-ordinate X co-ordinate

Figure 13- Stress concentration at the crack tip.

3.2 Cantilever Beam

A prismatic cantilever beam subject to a transversal shear load at its end is analyzed in
this example. Young modulus (E), Poisson ration (ν), the beam dimensions, and the load
distribution are shown at Fig. 14.

Figure 14- Cantilever beam and shear load distribution.

Reference values have been obtained from a detailed numerical finite element analysis
(FEA), using quadratic solid brick elements and successive mesh refinements. The final
answers have been achieved with 215,424 degrees of freedom (DF).
The results obtained for the displacement δ using the implemented BEM algorithm with
collocation points on the boundary are close to the reference value when refined meshes are
considered. More accurate results are achieved with quadratic element meshes, applying the
degenerate mapping approach to perform the integrals. The use of Hammer’s quadrature with
element subdivision to perform all integrals leads to the less accurate results.
The accuracy of the displacement answers δ obtained by using external collocation point
algorithm varies widely according to the value of α. Figure 15 shows results for the stress σxx
component at A (Fig. 14) obtained by applying such algorithm, in different computations,
using distinct values for the parameter α.
190 LINEAR TRIANGULAR ELEMENTS +58.3 QUADRATIC TRIANGULAR ELEMENTS
210 + 75

170 +41.6 180 + 50

150 + 25 150

Erro %
130 120

Erro %
110 90
90 - 25 60 - 50

70 -41.6 30 - 75

0 -100
50 -58.3
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 0.00 0.25 0.50 0.75 1.00 1.25 1.50
α parameter α parameter
Stress σ xx at A FEA - 215,424 D.F. Stress σ xx at A FEA (215,424 D.F.)

Figure 14- Stress σxx for analyses with different values of α.

Linear and quadratic elements have been used in a mesh with 74 nodes. Maximum
relative errors of 30% have been observed at the displacement δ results, as the values of α
vary between 0⋅01 and 1⋅5, in different analyses. On the other hand, relative errors in the
stress componente σxx at A have presented a severe variation, with relative errors from –270%
to 1854%. Similar variations are observed when the degenerate mapping integration approach
is used to perform the integrals. These variations change with different mesh refinements, and
different kind of elements (linear, quadratic). An increase in the condition number
[ cond ( A ) = A A − 1 ] of the linear equation system can be noted (Fig. 14) as the
distances between the collocation points and the boundary increase. There is no apparent
correlation between the critical values of α and the peaks observed in Fig. 16.
1.E+11

1.E+10
Condition number

1.E+09

1.E+08

1.E+07

Linear Triangular Elements


1.E+06
Quadratic Triangular Elements

1.E+05
0 0.25 0.5 0.75 1 1.25 1.5
α parameter
Figure 16- Condition number in analyses with different values of α.

Displacements at internal points have been calculated using the results achieved by
means of an ordinary BEM analysis (1,154 nodes). Boundary displacements and tractions
achieved by using the degenerate mapping approach have been adopted. The displacements at
internal points close to the boundary, placed on the straight line linking the internal point
(70;0;0) to the corner point (80;10;10), are plotted at Fig. 17. Integrations have been
performed using either Hammer’s quadrature with element subdivision (HQES) or the
degenerate mapping combined with Gauss-Legendre’s quadrature (DMGQ).
One can note the lack of accuracy when the standard Somigliana’s identity has been used
without any special treatment, due to the nearly-singular behavior of the integrals. However,
when the self-regular form has been used the results accuracy is preserved, no matter how
close to the boundary the internal point is. The average of the edges of the elements in the
mesh are 2⋅845 units for linear elements and 5⋅690 units for quadratic ones.

20 20
LINEAR triangular elements QUADRATIC triangular elements
Displacement u y at internal points Displacement u y at internal points 15
15
10
10

RELATIVE ERROR (%)

RELATIVE ERROR (%)


5
5 0
0 -5
-10
Standard SDI - HQES -5 Standard SDI - HQES
-15
Self-Regular. SDI - HQES -10 Self-Regular. SDI - HQES
-20
Standard SDI - DMGQ Standard SDI - DMGQ
-15 -25
Self-Regular. SDI - DMGQ Self-Regular. SDI - DMGQ
-20 -30
100 10 1 0.1 0.01 0.001 100 10 1 0.1 0.01 0.001
DISTANCE FROM THE CORNER POINT (80,10,10) DISTANCE FOM THE CORNER POINT (80,10,10)

Figure 17- Displacements uy at internal points close to a corner boundary.

4. FINAL REMARKS AND CONCLUSIONS

Some BEM algorithms have been evaluated in this paper. The evaluation of weakly-
singular integrals, by means of the degenerate mapping combined with Gauss-Legendre
quadrature, present in BEM for 3-D elastostatic problems, have been shown to be very
efficient. The accuracy of the results obtained using such quadrature, in this paper, have the
same level of accuracy of the results obtained using analytical integration scheme.
The evaluation of regular integrals may be accurately performed by using either
Hammer’s quadrature or the degenerate mapping approach. However, when the singular point
is nearby the element being integrated, more integration points are needed, in both schemes.
An alternative to do so is to apply the element subdivision technique. Hammer’s quadrature
has its performance improved when associated with this techinique, as more integration points
are provided. The amount of integration points may be increased directly in the degenerate
mapping integration algorithm. Nevertheless, if the singular point is very close to the element,
both numerical approaches are unsuitable to perform accurately the arising nearly-singular
integrals.
The algorithm based on external collocation points seems difficult to be applied in a
reliable way, as it was implemented. The determination of an ideal value for the parameter (α)
is the major concern. More researches are desirable in order to turns this algorithm reliable,
once this is a simple approach and accurate results were achieved using some specific values
of α.
The best approach for the evaluation of displacements at internal points seems to be the
self-regularized formulation. The internal points may be placed as close to the boundary as
wished, without any problem related to nearly-singular behavior of the integrals.

Acknowledgements

This work has been supported by the Brazilian Agencies CAPES (Coordenação de
Aperfeiçoamento de Pessoal de Nível Superior), CNPq (Conselho Nacional de
Desenvolvimento Científico e Tecnológico) and FAPEMIG (Fundação de Amparo à Pesquisa
do Estado de Minas Gerais).
REFERENCES

Aliabadi, M. H.& Rooke, D. P., 1991, Numerical Fracture Mechanics, Solid Mechanics and
its Applications, Kluwer Academic Publishers, Dordrecht, Netherlands.
Brebbia, C. A., Telles, J. C. F. and Wrobel, L. C., 1984, Boundary Element Techniques -
Theory and Applications in Engineering, Springer-Verlag, Berlin.
Bu, S. & Davies, T. G., 1995, Effective Evaluation of Non-Singular Integrals in 3D BEM,
Advances in Engineering Software, (Research Note), vol. 23, pp. 121-128.
Cerrolaza, M. & Alarcon, E., 1989, A Bi-Cubic Transformation for the Numerical Evaluation
of the Cauchy Principal Value Integrals in Boundary Methods, Int. J. Numer. Meth.
Engng, vol. 28, pp. 987-999.
Cruse, T. A., 1974, An Improved Boundary-Integral Equation Method for Three Dimensional
Elastic Stress Analysis, Computers & Structures, vol. 4, pp. 741-754.
Cruse, T. A. & Aithal, R., 1993, Non-Singular Boundary Integral Equation Implementation,
Int. J. Numer. Meth. Engng, vol. 36, pp. 237-254.
Cruse, T. A. & Richardson, J. D., 1996, Non-Singular Somigliana Stress Identities in
Elasticity, Int. J. Numer. Meth. Engng, vol. 39, pp. 3273-3304.
Doblaré, M. & Gracia, L., 1997, On Non-Singular Transformations for the Integration of
Weakly-Singular and Cauchy Principal Value Integrals, Int. J. Numer. Meth. Engng, vol.
40, pp. 3325-3358.
FRANC3D, A Three Dimensional Fracture Analysis Code, version 1.15 (2000), Cornell
University Fracture Group (CFG), Cornell University, Ithaca, New York,
<http://www.cfg.cornell.edu/>.
Guiggiani, M., 1998, Formulation and Numerical Treatment of Boundary Integral Equations
with Hypersingular Kernels, in Singular Integrals in B. E. Methods, eds V. Sladek, and J.
Sladek.
Huang, Q. & Cruse, T. A., 1993, Some Notes on Singular Integral Techniques in Boundary
Element Analysis, Int. J. Numer. Meth. Engng, vol. 36, pp. 2643-2659.
Lachat, J. C. & Watson, J. O., 1976, Effective Numerical Treatment of Boundary Integral
Equations: A Formulation for Three-Dimensional Elastostatics, Int. J. Numer. Meth.
Engng, vol. 10, pp. 991-1005.
Li, H-B., Han, G-M. and Mang, H.A., 1985, A New Method for Evaluating Singular Integrals
in Stress Analysis of Solids by the Direct Boundary Element Method, Int. J. Numer.
Meth. Engng, vol. 21, pp. 2071-2098.
Li, Y-S., Obata, T., Koguchi, H. and Yada, T., 1992, Some Improvements of Accuracy and
Efficiency in Three Dimensional Direct Boundary Element Method, Int. J. Numer. Meth.
Engng, vol. 33, pp. 1451-1464.
Lutz, E., Ingraffea, A. R. and Gray, L., 1992, Use of ‘Simple Solutions’ for Boundary Integral
Methods in Elasticity and Fracture Analysis, Int. J. Numer. Meth. Engng, vol. 35, pp.
1737-1751.
Milroy, J., Hinduja, S. and Davey, K., 1997, The Elastostatic Three-Dimensional Boundary
Element Method: Analytical Integration for Linear Isoparametric Triangular Elements,
Applied Math. Modelling, vol. 21, pp. 763-782.
Nakaguma, R. K., 1979, Three Dimensional Elastostatics Using The Boundary Element
Method, Ph.D. diss., University of Southampton, Southampton, UK.
Tada, H., Paris, P. C. and Irwin, G. R., 2000, The Stress Analysis of Cracks Handbook,
ASME Press, 3rd edition, New York.

You might also like