Documente Academic
Documente Profesional
Documente Cultură
AdvisorJeng-Tzong Chen
Undergraduate StudentCheng-Chan Hsieh
Keywords: Dual boundary integral equations, Stiffness, Flexibility, Rigid body mode, Singular value
decomposition, Laplace problem
Abstract
In this paper, the dual boundary integral formulation is used to determine the stiffness and flexibility for
rods and beams by using the direct and indirect methods. The stiffness and flexibility matrices derived by the
dual boundary integral equations (DBIEs) are compared well with those derived by the direct stiffness and
flexibility methods after considering the sign convention. Since any two boundary integral equations can be
chosen for the beam problem, six options by choosing two from the four equations in dual formulation can be
considered. It is found that only two options, either displacement-slope (single-layer and double-layer) or
displacement-moment (single-layer and triple-layer) formulations in the direct (indirect) method can yield the
stiffness matrix except the degenerate scale and a special fundamental solution. The rank deficiency is
examined for the influence matrices. Not only rigid body mode in physics but also spurious mode in numerical
implementation are found in the formulation by using SVD updating term and document, respectively.
1
successful experiences, SVD updating technique where t ( s ) = du ( s ) / ds , and the kernels are
will be employed to study the mathematical defined as
structure of the influence matrices derived by U ( x , s )
using dual formulation. T ( x, s ) = , (7)
In this report, rank deficiency for the n x
influence matrices is also our concern. The rigid U ( x , s )
body mode and spurious mode in the dual L ( x, s ) = , (8)
ns
formulation will be examined through SVD
U ( x , s )
technique. The relation between zero singular M ( x, s ) = . (9)
values of updating matrices (updating terms and nx ns
updating document) and nontrivial modes (rigid and degenerate kernels are shown in Table 1.
body mode and spurious mode) will be constructed.
Both the rod and beam structures are considered as Table 1 Degenerate kernels for rod problem.
illustrative examples. Kernels
U ( x, s ) T ( x, s ) L ( x, s ) M ( x, s )
Domain
2 Dual boundary integral 1 1 1
x>s ( x s) 0
formulation for rod problems 2 2 2
Let us consider the rod problem as shown in 1 1 1
Fig. 1. The governing equation for a rod is
x<s ( s x) 0
2 2 2
u (0) u ( L)
u ( s ) = [u ( x)U ( x, s ) u ( x)U ( x, s ) ]
x=L
x =0
. (4) For the degree of freedom of generalized
displacements (d. o. f.) and generalized forces, we
By differentiating with respect to the source point have
s , with to Eq. (4), the dual boundary integral u (0) 1 0 u0 u
equations as shown below = = Tru 0 , (11)
u ( L ) 0 1 u L u
u ( s ) = [T ( x, s )u ( x) U ( x, s )t ( x) ]
x=L
, (5) L
x =0
t (0) 1 0 t0 t
t ( s ) = [ M ( x, s )u ( x ) L( x, s )t ( x) ]
x=L = 1 = Trt 0 . (12)
, (6) t ( L) EA 0 1 tL t
x =0
L
2
By substituting Eqs. (11) and (12) into Eq.(10), 1 1
and the relation between generalized displacement 2
and generalized force is shown below [] = 2 (21)
1 1
u p
[ AT ] u0 = [ BT ] p0 , (13) 2 . 2
L L The [ A] matrix can be expressed as
where r
[ A]Tru = [ AT ] , [ A] = [ui ][ i ][vi ]
T
(14) . (22)
i =1
[ B ]Trt = [ BT ] . (15)
where [ i ] is the singular value, [ui ] and [ vi ]
The stiffness matrices are defined as
are the left and right unitary vectors, respectively.
t (0) u (0) The inverse of the [ A] matrix is
t ( L ) = [ K B ] u ( L ) , (16)
r
A1 = [ vi ][ i ] [ui ]
1 T
p0 u0
= [KF ] . (17) i =1
pL u L 1
It is found that Eq. (6) fail in constructing the 2 1
1 1
stiffness matrix, and the stiffness matrix can be = .
1 1 11 2 2 1 2
expressed as the same form of that derived by (23)
FEM as shown in Table 2. 2 21
1 1
2
2.2 The flexibility matrix of rods = 2
The flexibility matrix can not be obtained, 1 1
because the [ A] matrix is singular in Table 2. We 2 2
1 The flexibility matrices are defined as
utilize the SVD technique to calculate [ A] and
u (0) t (0)
try to get the flexibility matrix of the rod. By u ( L) = [ FB ] t ( L) , (24)
employing SVD technique, we have
u0 p0
[ A] = [ ][ ][ ] = [ FF ]
T
, (18) (25)
u
L pL .
where [ ] and [ ] are the right and left It is found that Eq. (6) fail in constructing the
unitary matrices, and [] is a diagonal matrix flexibility matrix, and the flexibility matrix can be
expressed as the same form of that derived by
composed of singular value. It is found that FEM as shown in Table 3.
1 1
[ ] = 2 2 , (19) 3 Dual boundary integral
1 1 formulation for beam problems
2 2
Based on the successful experience of
0 0
[] = , (20)
deriving the stiffness for a rod using BEM, we
0 1 extend the one-dimensional Laplace equation to
Table 2 Stiffness matrix for rod problems using dual BEM.
Eq. A B K B K F
1 1 1
2 0 2 L
2 1 1 1 EA 1 1
Rank ( A) = 1 Rank ( B) = 2 Rank ( K B ) = 1 Rank ( K F ) = 1
L 1 1 L 1 1
(6)
1 1 1
2
2 2 L 0
1 1
2 2 Rank ( B) = 1
(7) 0 0 NA NA
Rank ( A) = 0 1
1
0 0 2 2
3
x=L
biharmonic equation for a beam. Let us consider
u ( s ) = u ( x)
3U ( x, s)
u ( x)
2U ( x, s )
+ u ( x)
U ( x, s )
u ( x)U ( x, s)
x 2 x 2 x
. (31)
the Euler beam problems as shown Fig. 3 . x =0
m(0) m( L )
(0) 3.1 Direct method
( L)
By rewriting the displacement field, we have
x=L
u ( s ) = U ( x, s )u ( x) + ( x, s )u ( x) M ( x, s )u ( x) + V ( x, s )u ( x)
(32) x =0
v(0) u (0) u ( L ) v ( L) By differentiating the displacement, the slope,
Fig. 3 Generalized displacement and force d.o.f.
moment and shear force fields can be obtained
x= L
d 4 u ( x) u ( s) = U m ( x, s )u ( x) + m ( x, s )u ( x) M m ( x, s )u ( x) + Vm ( x, s)u ( x)
x=L
, (34)
=0. (26) x =0
dx 4 u ( s) = U v ( x, s )u ( x) + v ( x, s )u ( x) M v ( x, s)u ( x) + Vv ( x, s)u ( x)
(35)
x= L
.
x =0
where L is the length of the beam, u ( x ) is the where u ( s ) is the deflection, ( s ) is the slope,
lateral displacement, D is the domain between of
m( s ) is the moment and v( s ) is the shear force,
0 < x < L . By introducing one auxiliary system of
the fundamental solution respectively, and the relations of the sixteen
kernels are shown in Fig. 4. Degenerate kernels of
4U ( x, s )
= ( x s ) , < x < , (27) the sixteen kernels in a one-dimensional
x 4 biharmonic problem are shown in Table 4. Any
where is Dirac-delta function, x is field two boundary integral equations can be chosen, six
point, and s is the source point. For simplicity, options can be considered. We utilize the
the fundamental solution is selected as degenerate kernel expansion and substitute them
1 3
into the two boundary integral equations which are
U ( x, s ) = xs , (28) chosen. By approaching s to 0+ and L , we
12
and can be expressed in terms of degenerate kernel have the matrix form as follows
as u (0) u (0)
u (0)
1
( x s )3 , x > s [ A] = [ B ] u (0) , (36)
12 u ( L) u ( L)
U ( x, s ) = . (29)
u ( L) u ( L)
1
( s x) , x < s
3
12
By multiplying the auxiliary system in Eq. (28)
where [ A] and [ B] are obtained through six
with respect to the governing equation and formulation ( u , u m , u v , m , v ,
integrating by parts, we have the boundary integral m v ) as shown in Table 5.
equation as
L 4U ( x, s ) 4 u ( x) . 3.1.1 The stiffness matrix of the Euler beam
u ( s ) = u ( x ) U ( x, s ) dx (30)
0
x 4
x 4 We utilize a simple structure in the sign
The boundary integral equation is derived as convention to define the notations of generalized
Table 3 Flexibility matrix for rod problems using the dual BEM.
Eq. [ A] [B] [ FB ] [ FF ]
1 1 1
2 0 L
2 2 L 1 1
L 1 1
1 1
(6)
2
1 41 1 4 EA 1 1
2 2 L 0
Rank ( FB ) = 1 Rank ( FF ) = 1
Rank ( A) = 1 Rank ( B) = 2
1 1
2 2
0 0
(7) 1 1 NA NA
0 0
2 2
Rank ( A) = 0
Rank ( B) = 1
4
u (0) u (0)
x x x u (0)
U ( x, s ) ( x, s ) M ( x , s ) V ( x , s ) = [ K ] u (0) , (42)
u ( L) B
u ( L)
s
u ( L) u ( L)
U ( x, s ) ( x, s ) M ( x, s ) V ( x, s )
v0 u0
s
0 = [ K ] 0 .
m
(43)
U m ( x, s ) m ( x, s ) M m ( x, s ) Vm ( x, s ) vL F
u L
s
mL L
U v ( x, s ) v ( x, s ) M v ( x, s ) Vv ( x, s ) It is found that only two combinations of Eqs. (33),
Fig. 4 Differential operators for the sixteen (34) and Eqs. (33), (35) can construct the stiffness
kernels of the Euler beam. matrix, and the stiffness matrix can be expressed
as the same form of that derived by FEM as shown
Table 4 in Table 5.
Kernels
U ( x, s ) U ( x, s ) U m ( x, s ) U v ( x, s )
Domain 3.2 Indirect Method
( x s)
3
( x s)
2
xs 1 Instead of choosing two equations from the
x>s
12 4 2 2 dual formulation in the direct BEM, we can also
sx adopt two potentials from single, double, triple and
( x s)
2
( x s)
3
1
x<s guadrupole potentials as denoted by U - ,
12 4 2 2
U M , U V , M , V , and M V
formulations.
displacement and generalized force to connect the
FEM notations. For the degree of freedom of (1)Single and double layer approach ( U - )
generalized displacements and generalized forces, u ( s ) = U (0, s )0 + U ( L, s )L + (0, s ) 0 + ( L, s ) L . (44)
we have (2)Single and triple layer approach ( U M )
u (0) 1 0 0 0 u0 u0 u ( s ) = U (0, s )0 + U ( L, s )L + M (0, s ) 0 + M ( L, s ) L . (45)
u (0) 0 1 0
0 0 (3)Single and guadrupole layer approach ( U V )
= = T 0
u ( L) 0 0 1 0 uL
bu u (37)
L u ( s ) = U (0, s )0 + U ( L, s )L + V (0, s ) 0 + V ( L, s ) L . (46)
0
0 1 L
u ( L) 0 L , (4)Double and triple layer approach ( M )
u (0) 1 0 0 v0
0 v0 u ( s ) = (0, s )0 + ( L, s )L + M (0, s ) 0 + M ( L, s ) L . (47)
u (0) m0 m , (5)Double and guadrupole layer approach ( V )
= 1 0 1 0 0
=T 0
u ( L) EI 0 v bt v (38)
0 1 0
L L u ( s ) = (0, s )0 + ( L, s )L + V (0, s ) 0 + V ( L, s ) L . (48)
0
u ( L) 0 0 1 mL mL (6) Triple and guadrupole layer approach ( M V )
since u ( x ) is defined downward. u ( s ) = M (0, s )0 + M ( L, s )L + V (0, s ) 0 + V ( L, s ) L . (49)
By substituting Eqs. (37) and (38) into Eqs.(36), By approaching s to L and to 0+ , and The
and the relation between generalized displacement unknown fictitious densities ( , ) can be
and generalized force is shown below as obtained by
u0 v0 0 u (0) 0 u (0)
(0)
m
[ A] L = L = [ A] ,
1 (0)
[ AT ] 0 = [ BT ] 0 , (39) u ( L) u ( L)
(50)
uL vL
0
0
L ( L) L ( L)
L mL 0 v(0) 0 v (0)
m(0) m(0) .
1 L
where
[ B ] =
L [ B ][ A] =
0 v ( L)
(51)
[ A]Tbu = AT , (40) 0
v ( L)
L m( L) L m( L)
[ B ]Tbt = BT . (41) where [ A] and [ B] are obtained through six
The stiffness matrices are defined as formulation as shown in Table 6. It is found that
the stiffness matrix can be obtained by selecting
U and U M formulations.
5
4 Discussion of the rigid body If the rigid body term, c , and the linear,
mode and spurious mode quadratic and cubic terms, ax , bx 2 and dx 3 are
If the rigid body term, c , and the linear term, superimposed in the fundamental solution, we
ax , are superimposed in the fundamental solution, have U b ( x, s ) = U ( x, s ) + ax + bx 2 + dx3 + c . By
we have U r ( x, s ) = U ( x, s ) + ax + c . By substituting the auxiliary system U b ( x, s ) into
substituting the auxiliary system U ( x, s ) into Eq. Eqs. (31) for the u formulation, and setting
(4), and setting EA = 1, L = 1 , we have EI = 1, L = 1 , we have the
1 1 1 1 1 1
2 + a 2 a u (0) c a c 2 + 6d 2b 6d + 2b + 6d
2 u (0)
2 2
u (0)
= . (52) 1 + 6d 2b + 6d
1 1
1 + a 1 a u (1) 1 + c u (1) 2b 6d
a c 2 u (0)
2 2
2 2
2 =
0 1
0
1 u (1)
The [ B1 ] matrix for a rod is singular when 2 2 u (1)
1 1
0
(1 + 2a) = 4c . This results in the degenerate 2
0
2
1 1
(54)
scale problem. According to the Fredholm c a c a b d
12 4
+ a + 2b + 3d
u (0)
alternative theorem, the degenerate scale depends 1 +c 1
a c a b d a + 2b + 3d
12 u (0)
on the rigid body term. When a = 0 and
4
.
0 1
1 u (1)
c = 1/ 4 , [ B1 ] matrix is not invertible and
0
4 2 u (1)
1 1
results in a degenerate scale. By employing the 0 0
4 2
SVD technique with respect to the influence The [ B1 ] matrix for a beam is singular when
matrix, for [ A1 ] and [ B1 ] matrices. The spurious
(1 + 12d ) 24a = 48c . This results in the
mode [ ] satisfies degenerate scale problem. When a = 0 , b = 0 ,
AT c = 1/ 48 and d = 0 , [ B1 ] matrix is not
[ ] = 0 , (53)
BT invertible and results in a degenerate scale. By
employing the SVD technique with respect to the
where the spurious mode [ ] and the rigid body influence matrix in the u formulation, for
mode [ ] are shown in Table 7. [ A1 ] , [ A2 ] and [ B1 ] matrices.
Table 5 Stiffness matrix for the Euler beams by using the direct method.
Eqs. A B K B K F
1 1 L 1 1
u- 2 0 0 0
2 2 12 4L
12 6L 126L 12 6 L 12 6 L
Eqs. (40) and (41)
1 L
0 1 0
1 1
0 2 L2 2
2 3 12 1 6 L 4 L 6L EI 6 L 4 L 6 L 2 L
2 2
2 2 4L
L
1 1 1 1
3
L 12 6L 12 6 L L 12 6 L 12 6 L
3
0 0 2 2
0 0
2 2 4L 2L 6 L 2 L2 6 L 4 L2 6 L 2 L 6 L 4 L
2
1 1 1 1
0 0 0 0
2 2 4 L 2 L2
Rank ( A) = 2 Rank ( B ) = 4 Rank ( K B ) = 2 Rank ( K F ) = 2
1 1
0 0
u-m 1
1 L 12 4L
2 0
2 2 1
12 6L 12 6 L 12 6 L 12 6 L
0
1
Eqs. (40) and (42)
1 L 3 12
0
1 6 L 4 L
2
6 L 2 L2 2 2
EI 6 L 4 L 6 L 2 L
0
1 4L
L
2 2 2 1 1 1 L 12
3
6L 12 6 L L 12 6 L 12 6 L
3
0 2
0 0 0 0 2 L3 2 L2 2L
3
6 L 2 L2 6 L 4 L2 6 L 2 L 6 L 4 L
2
0 0 0 0 1 1 1
2 0
2 L 2 L3 2 L3
Rank ( A) = 2 Rank ( B ) = 4 Rank ( K B ) = 2 Rank ( K F ) = 2
6
Table 6 Stiffness matrix for the Euler beam by using the indirect method.
Portential [ A] [ B] [ K ] = [ B ][ A]
1
L3 L2
U - 0
0 1 1
12 4 2 2 0 0
single and double layer
2
0 L L
0 L 1 1 12 6L 12 6 L
2 0 2 2 2
4
3 2 1 6 L 4 L
2
6 L 2 L2
L L 1 1
12 0 0 0 0
L 12
3
6L 12 6 L
4 2 2
L2 6 L 2 L2 6 L 4 L2
L L 0 1 1
0 0
4 2 2 2 2
Rank ( A) = 4 Rank ( B ) = 2 Rank ( K ) = 3
L3 L
U-M 0 0
12 2 1 1
2 2 0 0
single and triple layer
0
2
L 1 1
2 2
1 12 6L 12 6L
4 0 2 0 0
3
1 6 L 4 L
2
6L 2 L2
L L
12 0 0 1 1 0 0
L 12
3
6L 12 6L
2
2 2
L2 6 L 2 L2 6 L 4 L2
1 1 1 0
0 0 0
4 2 2 2
Rank ( A) = 4 Rank ( B ) = 2 Rank ( K ) = 3
According to Fredholm alternative theorem [11], SVD updating term and document technique. It is
the spurious mode [ ] satisfies found that rigid body mode and spurious mode are
imbedded in the right and left unitary vectors of
AT the influence matrices through SVD.
[ ] = 0 , (55)
BT
and the rigid body mode [ ] satisfies 6 References
[1] Gere, J. M., Mechanics of Materials, Thomson
A1 Learning, Inc., California (2002).
[ ] = 0 . (56) [2] , , , ,
A
2 (1992).
The spurious mode [ ] and the rigid body mode [3] Chen, J. T., W. C. Chen, S. R. Lin and I. L. Chen,
Rigid Body Mode and Spurious Mode in the Dual
[ ] are shown in Table 7. The mathematical Boundary Element Formulation for The Laplace
framework of [ A] and [ B ] is shown in Fig. 5. Problems, Computers and Structures, Vol. 81, pp.
1395-1404 (2003).
[4] Felippa, C. A., K. C. Park and M. R. Justino Filho,
The Construction of Free-Free Flexibility Matrices
5 Conclusions as Generalized Stiffness Inverses, Computers and
Dual boundary integral equations were Structures, Vol. 68, pp. 411-418 (1998).
employed to derive the stiffness and flexibility of [5] Dumont, Ney. A., Generalized Inverse Matrices
the rod and beam which match well with those of and Structural Analysis, Numerical Methods in
FEM. Not only the direct method but also the Continuum Mechanics (2003).
indirect method was used. It is found that [6] Pozrikidis, C., A Note on The Relation Between
u The Boundary- and Finite-Element Method with
displacement-slope ( ) and
Application to Laplaces Equation in Two
displacement-moment ( u m ) formulations in the Dimensions, Engineering Analysis with Boundary
direct method can construct the stiffness matrix. Elements, Vol. 30, pp. 143-147 (2006).
Similarly, the single-double layer approach [7] Banerjee, P. K. and R. Butterfield, Boundary
( U ) and single-triple layer approach ( U M ) Element Methods in Engineering Science,
work for the constructing of stiffness matrix in the McGraw-Hill, London (1981).
indirect method. For choosing a special [8] Hong, H.-K. and J. T. Chen, Derivation of Integral
Equations in Elasticity. Journal of Engineering
fundamental solution, the stiffness matrix can not
Mechanics, ASCE, Vol. 114, pp. 1028-1044
be obtained for the degenerate scale. Rigid body (1988).
mode and spurious mode were studied by using the
7
Table 7 Spurious modes and the rigid body modes for a rod and a beam in BEM.
Rod Beam
0.632 0.316
1 1
Spurious mode
2 2
(generalized force)
1
0.632
= 2
1 0.632 = 0.632 0.316
0.316
2
0.316
u '(1) = 0.258
u '(0) = 0.258
u (1) = 0.516
0.774
Rigid body mode u(0) = 0.774 A 0.258
=
1 1 0.516
(generalized displacement) u (1) =
u (0) = 0.258
2 2
1 0
u '(1) = 0.577
A
= 2 A 0.577
1 =
0.577
2
0.577
u (1) = 0.577
u '(0) = 0.577
u(0) = 0
A1 = B1
spurious mode
-0.632 T
" " " 0 " " " -0.774
0 " " -0.632
" " " 0
" " "
A = -0.632 " " " " 0 " " 0.258 0.577 " " B = -0.632 " " " " " " " T
1 1
-0.316 " " " " " " " -0.516 0.577 " " -0.316 " " " " " " "
0.316
" " " " " " " 0.258 0.577 " "
[ A1 , B1 ]
0.316 " " " "
" " "
[9] Chen, J. T. and H.-K. Hong, Review of Dual [11] Chen, J. T., I. L. Chen and K. H. Chen, A Unified
Boundary Element Methods with Emphasis on Formulation For The Spurious and Fictitious
Hypersingular Integrals and Divergent Series, Frequencies in Acoustics Using The Singular Value
Applied Mechanics Reviews, ASME, Vol.52, No.1, Decomposition and Fredholm Alternative
pp.17-33. Theorem, J. Comp. Acoustics, (2006).(Acepted)
[10] Chen, J. T., C. F. Lee and S. Y. Lin, A New Point [12] Chen, J. T., L. W. Liu and H.-K. Hong, Spurious
of View For The Polar Decomposition Using and True Eigensolutions of Helmholtz BIEs and
Singular Value Decomposition, Int. J. Comp. BEMs for a Multiply-Connected Problem, Proc.
Numer. Anal. Appl, Vol. 2, No. 3, pp. 257-264 Royal Society London Series A, Vol. 459, pp.
(2002). 1891-1925 (2003).
8
9