Professional Documents
Culture Documents
Boundary Element Techniques in Computer-Aided Engineering
Boundary Element Techniques in Computer-Aided Engineering
Aided Engineering
NATO ASI Series
Advanced Science Institutes Series
C.A. Brebbia
Institute for Computational Mechanics
Ashurst, Southampton S04 2AA
UK
Distributors for the United States and Canada: Kluwer Academic Publishers, Inc.,
190 Old Derby Street, Hingham, MA 02043, USA
Distributors for the UK and Ireland: Kluwer Academic Publishers, MTP Press Ltd,
Falcon House, Queen Square, Lancaster LA 1 1RN, UK
Distributors for all other countries: Kluwer Academic Publishers Group, Distribution
Center, P.O. Box 322, 3300 AH Dordrecht, The Netherlands
PREFACE XI
1.1. Introduction
1.2. Basic Definition
1.3. Approximate Solutions
1.4. Method of Weighted Residuals
1.5. Weak Formulations
1.6. The Inverse Problem
1.7. Conclusions
References
v
Chapter 4. Scalar and Vector Potential Theory
by G.T. Symm
5.1. Introduction
5.2. Flow past an Obstacle
5.3. Discretisation
5.4. Green's Boundary Formula
5.5. Applications
5.6. Boundary Singularities
5.7. Composite Domains
5.8. Conclusion
References
6.1. Introduction
6.2. The Newtonian Potential
6.3. Discretisation
6.4. General Domain
6.5. Axisymmetric Problems
6.6. Conclusion
References
VI
Chapter 7. Heat Transfer Applications
7.1. Introduction
7.2. Integral Equations associated with Steady Heat
Conduction Problems
7.3. Nume.rical Solution of the Integral Equations
7.4. Poisson's Equation
7.5. Non-homogeneous Bodies; Method·of Subregions
7.6. Anisotropic Bodies
References
8.1. Introduction
8.2. Isoparametr1c Elements
8.3. Numerical Integration
References
9.1. Introduction
9.2. The Boundary Element Method
9.3. Advantages and Disadvantages of the BEM compared
to FEM
9.4. Introduction to BEASY
9.5. Examples
9.6. Conclusions
References
10.1 Introduction
10.2. Pre- and Post Processing
10.3. C.A.D. Coupling
10.4. Installation on Different Computers
10.5. Recommendations for BEM use
VII
Chapter 11. Electrostatics Problems
11.1. Introduction
11. 2. Theoretical Basis
U.3. Boundary Elements
11.4. Applications
U.S. Conclusions
References
12.1. Introduction
12.2. Theoretical Development
12.3. Boundary Conditions
12.4. Numerical Implementation
12.5. Velocities and Pressures
12.6. Identification of Areas in Shadow
12.7. Test Example
12.8. Conclusions
References
13.1. Introduction
13.2. Governing Equations
13.3. Boundary Integral Formulation
13.4. Two Dimensional Elasticity Problems
13.5. Three Dimensional Elasticity Problems
13.6. Axisymmetric Elasticity Problems
References
Appendix
14.1. Introduction
14.2. Transformation to Boundary Integrals
14.3. 2D Body Forces
14.4. 3D Body Forces
14.5. Axisymmetric Body Forces
References
Appendix
VIII
ChaEter 15. Time DeEendent Problems
l5.I. Introduction
15.2. Time Dependent Diffusion
15.3. The Scalar Wave Equation
15.4. Transient Elastodynamics
15.5. Mass Matrix Representation
15.6. Conclusions
References
l6.I. Introduction
16.2. Integral Formulation of Heat Conduction Problems
16.3. Numerical Solution of the Integral Equations
16.4. Conclusions
References
by M. Stern
IX
Chapter 20. Fracture Mechanics Stress Analysis, I.
20.1 Introduction
20.2. Stress Intensity Factors
20.3. Integral Equation Methods for Crack Tip
Stress Analysis
References
21.1. Introduction
21.2. Invariant Integral based on the Energy
Momentum Tensor
21.3. Invariant Integrals deduced from Betti's
Reciprocal Theorem
21.4. Some Numerical Results for a Nocht Problem
21.5. A Problem of Debond Stress Analysis
References
22.1. Introduction
22.2. Notation and some Basic Ideas
22.3. BEM applied to the Interaction between
Structures and the Supporting Ground
22.4. Inhomogeneity, Zoning and Layering
22.5. E1astop1asticity
22.6. Concluding Remarks
References
x
PREFACE
The Institute was held under the auspices of the newly launched
"Double Jump Programme" which aims to bring together academics and
industrial scientists. Consequently the programme was more industr-
ially based than other NATO ASI meetings, achieving an excellent
combination of theoretical and practical aspects of the newly
developed Boundary Element Method.
CARLOS A. BREBBIA
Editor
Chapter 1
C. Brebbia
1. INTRODUCTION
1
2
2. BASIC DEFINITIONS
LCu) = b in fl (1)
Xc ) (2)
<
...u -+
w>=<u
-+ ...
w> (3)
<
...u -+
w> =
r -+ -+
• w dfl (4)
J u
One can now use the idea of the inner product to define
the 'transposed' form of (5) whic~ gives the adjoint of ~( )
operator, which will be called ~ ( ).
< J:(u), w > < u ,X* (w) > + Boundary terms (6)
The boundary terms on the right hand side are valid over the
r boundary of the ~ region. They can be written in terms of
an inner product or integrals on the boundary as,
- < * (w),
G S(u) >r (7)
S, G and S ,G
,~ *
are in general differential operators,
associated with the boundary conditions of the problem as
will be seen in the next examples.
Example
x(u) ~ ~ + du + U in x (a)
dx' dx
(where 0 < x < 1). The inner product of ;( (u) with a function
w is
1
r
< {(u), w > {d U + du + u} w dx
-2- (b)
J dx' dx
0
(dull 1
)
+ rw {dU + uJ -
1- dx u'dx) o J
4
Notice that the first integral on the R.H.S. gives the adjoint
operator i.e.
du
G(u) = - + U S(u) = u
dx
(e)
dw
G* (w) = dx S* (w) w
Example 2
in x (a)
1 1
f L(u) w dx f ~:~ u dx + [w ~:~I
o o (b)
and similarly for SI* ' G*1 etc. Notice that in a beam
application the S conditions are called kinematic, displace-
ments or essential conditions and the G's are usually named
as mechanical, forces or natural boundary conditions.
5
3. APPROXIMATE SOLUTIONS
~(u ) b in (l (8)
o
(9)
n
u = L OI.k <Pk + 01. 0 (10)
k=l
E = (u) - b .f 0 in (l
S(u) - s .f 0 ( 11)
G(u) - g .f 0
Example 3
u
-0.12 o
/"----- ....... , ..: -0.6
/
/' "-
// .-" '-'_.'," ,t-1. 0
-0.10 \
-0.08 - \ I
/ .. /
AU"',
....
'.,
\
\ ,
/'
- -0.8
"+, II / / " \ +'
-0.06
I..
\,/. "" \ /
A\
~i -0.6
-0.04
i/\E l:\\ ~-0.4
/" '. ,. '-\\ i
-0.02 I/ "~ /~"
..... :f \~ -0.2
6' , " ,
O.O~__~~__~__·~~~-~··~__~··~~~~__~__~_·~___ x
0.0 0.2 0.4 0.6 0.8 1.0
0.8
0.7 L.14
+-0.12
0.6
0.5
i -0.10
0.4
-0.08
0.3
-0.06
0.2
-0.04
0.1 -0.02
o (b)
u ~ U : a sin ~x (c)
o
Then the residual function E in x (notice that E1 E2 - 0)
is
E : - a~2 - b (e)
a : - b /~2
One can now compute the values of the two functions, i.e.
the error and the approximation to u o ' i.e.
b
u 1fT sin ~x
E : b sin ~x - b
(12)
(13)
I E W dn o (14)
Example 4
1
J EW dx o in 0 < x < 1 (c)
o
with a residual,
d2 u
E = dx' + U + x ~ (-2+x-x 2 )U 1 + (2-6x+x'-x 3 )U 2 + x (d)
(e)
(f)
(g)
x (1-x)
u = 217 (42 + 40x) (h)
and
1
E = 2T7 (- 4 + 19x - 2x 2 - 4Ox3) (i)
+ '" + ao (16)
( 17)
Example 5
u (b)
o
1 1
r
0
f £ <PI dx 0 I
J
0
£ <P2 dx 0 (c)
0.8
e:
0.7
0.6
0.5
0.4 0.08
_ 0.06
0.3
0.2 0.04
0.1 0.02
"-.
=t.:..
0.08
.'..f.. -0.02
-0.04
\.
\.
\ -0.06
-0.08
1
J [x + (-2+x-x 2 )a 1 + (2-6x+x'-x 3 )a 2 ]x 2 (1-x)dx 0
o
[ ] )"') L:l
3 3
10 20 (f)
3 13
20 105 la 2
_ 71
a 1 - 369 (g)
u =
71 + liT
x(1-x) { 369 7x}
£ =~
369
(-2+x-x 2 ) + ~ (2-6x+X 2 _X 3 ) + x
41
5. WEAK FORMU~ATIONS
a)
f df
dx
b)
~ x
~
; I
df
f dx
c)
d2 u
X(u) - b = dx 2 + U + x o<x < 1 (16)
J {~:~ + u + x} w dx o (17)
o
it is clear that a different order of continuity is required for
u than for w. To define this continuity requirement we need to
introduce a classification for the degree of continuity of a
function. Let us assume that a function f is discontinuous at
a discrete point but is finite throughout the region (figure 4a)
its now defining the condition
f f2 dx < 00 (18)
(20)
1 1 1
f {~:~ + u + X}W dx - f {~~ ::} dx + f (u+x)w dx
o o 0
1 (21)
+ [q w10
r 1
+ [q wl1 _ lu dwl (22)
0
L dxl• 0
Natural conditions q q
(23)
Essential conditions u u
- IlU -I
dwl
+
1- dxj'
dw 1
u = 0 (24)
dx. x=l . x=o
I {~:~
1
i.e. one has not returned to the same expression as (17) but to,
1
Example 5
d'u
= dx' + U + x u < x < 1 (a)
u ax' + (q -2x)x
(b)
aq, + qx
a ; - O.5208(q + 1) (e)
(g)
(h)
o
where (i)
1
r idu
J [Six
dw l
dx - (u+x)wJ dx (j)
o
We can substitute u and w into (j) which gives
r
J [(2a 1x + a 2 )(26 1x + 6 2 ) - (a 1x 2 + a 2x + x)(6 1x2 + 6 2x ))dx
(k)
u 2nd solution ~
...---) i 0
0.5
0.4
0.3
0.2 _
0.1
x
0.0 0.2 0.4 0.6 0.8 1.0
~1 0.9859 + 1.9864q
(m)
~2 -0.4319 - 0.43229q
d2 u (27)
dx' + U + x
1
{ddx2u
2
} - dwUX=1
f
o
+ U + x w dx + [(u-u) dx x=o = 0 (28)
1 1
x=1 [- dWr=l _
J {ddx2W
2 + w} u dx rx
J
w dx + [q w]x=o - ~ dx - o (29)
x=o
0 0
d2 w
dx 2 + W =0 (30)
w ~1 cos x + S2 sin x
(31)
dw
dx = - ~1 sin x + S2 cos x
1
Jx w dx + [q w] x=1 - [q w]x=o = 0 (32)
o
Expanding this expression one can write
1
Jx(Sl cos x + S2 sin x)dx + ql(SI cos 1 + S2 sin 1)
o
o (33)
21
qo =~-1
=_1_ _ 1 (34)
sin 1 ql sin 1
1 1
J (d'
i
dx'
w
+ W
)
u dx
J /':,.
1
u dx u.
1
(37)
o o
7. CONCLUSIONS
REFERENCES
L(u) = b (1)
L( ) a ~ + a ~+ a 2 ( ) (2)
o dx 2 1 dx
(3)
or an integro operator
t
L( ) JV(t-T). ( )dT (4)
o
L( ) = v(t)*( ) (5)
23
24
<~ v >
J u.v dO
0
[L(u)]v = *
[L (v)]u+ ( 10)
It follows that
(e)
p(u,v) du d
a v - - u - (a v) + aluv
o dx dx 0
Example 2
Let US consider
d2 {b 0 ~} +d- {b dx
L(u) = dX2
dx 2 dx l dU} + b 2u (a)
12 12 d2 u d 2 v du dv
v L(u)dx [h dx 2 dx 2 hI dx dx + b 2uv)dx
J J 0 -
xI xI
(h)
+ r{d
LV - dx
(
b 0d-
2
dx-
u) + b -dU} + dv
2 I dx dx {- h 0
d 2 Upx2
dx 2 I
-'xI
< L(u), v > < u, L* (v) > + < R(u,v)-R(v,u), 1 > (d)
(e)
S* (v)
( 13)
G*(v)
Example 3
_G(u) = {~(b
dx 0
dZu) + b du
dx 2 1 dx - b d2U}
o -dx 2
(a)
Essential
u or d (b d2 u) b du_
dx 0 dx 2 + 1 dx - gl
and
du
dx or
3. VARIATIONAL FORMULATION
L(u) + b = 0 in 0 (17)
E.(u) = s on r 1
(18)
Q(u) = ~ on r 2
d( )
(21)
(IX
i r-T
To interpret (19) as a stationary requirement for a functional,
we define v as the variation of u, i.e.
v = OU (22)
29
-S(u) -s -
= (28)
~(ou) = 0 on r 1 (29)
C(u) = ~ (30)
with ~(eu) = Q on r 1 •
L(u) + b 0 in Q
(36)
~(u) = s on r 1
~(u) =~ on r 2
Example 4
_S(u) = dU}
{u , dx
_G (u) = f· A-
dx
(b
0
d 2 u) + b 1 ·du
dx 2 dx -b -d
o dx 2
2U}
32
H(u) - 1 b 1
(d 2 UJ 2 - -1 b 1-
(duJ 2 + _1 b u 2
-"2 0 (dx- 2 2 1 (dx Z Z
For the case where F(u) is specified on fl' the expanded form
is
I f [H(u) + bujdn
~ [(u-s ){~
1 dx
(b
0
d 2 U) + b dU} + (du _ s )(_ b d2 u)j
dx 2 1 dx dx Z 0 dx 2 fl
where
EI L(u) + b in n
El ~(u) - s on fl (37)
EZ = Q.(u) -~ on f2
o (38)
w u
Q(w) (39)
- ~(w)
+ J [£ -'Q(u)] E(w)dr 0
rZ
If we require u to satisfy the boundary conditions, the
boundary integrals vanish. If, in addition, we select w from
the basis functions contained in u, i.e.
w $. (41)
~
+f [~.~(w)-i(u)·~(w)jdr 0
r2
L(w) = 0 (43)
L(w) = - 6(; - ; )
p
(44)
+ < *
S (w) -l!. - *
~(u) -G (w) >r o
2
o (48)
*
+ <[~(a) - ~J, wI + ~ (w), £(a) - ~(a), G (w) * >r
1
(50)
Comparing (47) with (50) shows that both forms are identical
when we select ~1 and ~2 as follows:
~1
= G* (w)
(51)
~2
- s* (w)
36
E:1 = u - u (54)
E: 2 q - q
where q = au/an. Bar superscripts indicate these values are
prescribed. Notice that here
~(u) = u ,!!.=u
(55)
G(u) 3. q q
f (\l 2 u + b)w dO + f
(56)
n r1
+ f (q-q)w dr =0
r2
Applying Green's formula to w\l2u transforms (56) to
< b, w > + < u, L(w) > + < ~(w). G(u) - ~(u).£(w) > =0
(a)
6. FUNDAMENTAL SOLUTIONS
fJ2 W + 1I (~ - ~ ) = 0 (58)
';p
J aw
an
dr (59)
Also,by definition,
r !J.(~ -
I:
for x'P in
~ )dfl
J p
for x p outside (60)
fI
J
r e;
- 1 (61)
1
w = 4'Jfr ='41r (62)
lit - itpl
awl
ar .2'Jfe: - 1 (a)
r=e;
and obtain
x - x->- I
w = --1 In r = --1 In 1->- (63)
2'Jf 2'Jf P
r ->-->-
J f[-t(w - wp)]df!
f! (a)
- f 1->- ->- ;: - f
x=xp P
(
+ J (wpq - u ~ w )dr
an p
= 0
r2
We utilize (b4) in two ways. Firstly, we can generate the
solution at a set of interior points once the function and its
normal derivative are known on the boundary. Secondly, by
taking P at various points on the boundary, and introducing
approximations for u and au/an in terms of a finite number
of parameters, we can establish a system of algebraic equations
for the unknown parameters. This set of equations can be
solved and the distribution of u and au/an on the boundary can
be evaluated, allowing one then to evaluate the interior values
of u.
(a)
a -1
an wp 41Tr2
Then,
(
u ..l. w dr lim r.
L
-1 1l
[21TE2]
J an P E .... 0
U •
41T ETJ
rE
1 1
2 ul po~nt
. P -2 up
40
Ilni:l,
f
r
~,
l' an
au dr lim
e: ... 0
o (65)
e:
The same result is obtained for the two dimensional case.
Therefore, when the weight ing function corresponds to a unit
delta -function (referred to as a source) applied at a point
p on the boundary, and the bound ry is smooth, the weighted
residual expression has the following form
u = u
(a)
au
an q
- J....
211
In r
(a)
"interior"
Figure 3
and
r -ll
f
a
r
u an wp dr lim
e: ... 0
t· 211E:]
[ (1I+a)f:]
(67)
e:
1 0.
- "2 (1 + - )u p
11
( au
J wp an dr 0
r
e:
41
(68)
q - Cp)U p +
(
a
J u an
r
WF dr -
I
r
wpq dr I
n
b Wp ,Ut (69)
(0)
q = 1)1. Q.
-J -J
Figure 5
where: U. contains the values
-J of u associated with
discrete points in
segment j and 1j are
the corresponding interpolation functions; Qj and 1. are
the corresponding terms for the normal derivative, J q •
(71)
(72)
51 - 5*
(73)
+ 5
*
52
For this case,
a = HU jl + u j2 )
0
(74)
al
= 25* (u j2 - u jl )
and
u.
-J
{u jl ; u j2 }
5 5
cpo
-J
{HI - 5* HI + - ) }
5*
rj rj 0 (75)
44
o - CF)u P + J UI f·
J=l
r.
a w
J an P
u.
dr J -J - [I r.
) 1J
w 1jJ. dr o.
P "'J -'-J
J J
Jb w dn (76)
P
n
N
o - C)u + I F'
p.
(77)
P Pi j=l 1
i = 1,2, ••• ,N + N
u q
u
(78)
"
v
AX = F
N 6
N 3 (u 4 ,u 5 ,u6 prescribed)
u
N
q
2 (Q1,Q2,q3,q4 prescribed)
u {u 1 ,u 2 ,u 3 }
{Q4,Q5 }
Fig. 8
!!.1
N =6
N = S
u
N =' 2
q
U = {u 1 ,u 2 '.·· ,uS}
.9. = {qS,q6}
Figure 9
ul e 1ement J. = 4>. U.
-J-J
(SO)
ql e1 ement J. = 4>. Q.
-J-J
[J Wp 4>.
-J
drJQ.
-J
]
r.
J
(S1)
[[ fr.i j [J
r.
J J
(83)
Note that $£* does not contain the contribution from CPi; only
the uPi/2 term is included.
48
~ = f= 0
(85)
<jJ= a{ 1,1 •••. ,1
We need to modify ~3
Fig. 10
IJ
-[ d1
d2
.J (87)
d.
1. 2 ?t:.
1.J
j=l ,2, •••
Mid-element node
Y Nodal value
Nodal value
u or q
y ~ dx
x
or q r
r = (x,y)
H F (88)
<1>1 s«;-1)/2
<1>2 s(s+1)/2
<1>3 (1-1;)(1+0
{ (:~J 2 + (:p 2} i
f = x or y or u or q
I
r
J
g dr = g IJI dE; (91)
r. -1
J
+1
r
I J h{E;)dE;
-1
n i E;i w.
l.
(linear) 0 2
2 {cubic} + 1/13 +1
2 - 1/13 +1 (92)
3 (quintic) 0 8/9
2 +1:6 5/9
3 -r.6 5/9
w = ;; In I~I (a)
52
->-
n. Noting that w is axisymmetric,
J the normal derivative on
segment j is given by
Figure 14
(dx -7
1
ldf ~-71
4-
t.
J TJT 1
+
dE; J J (94)
4-
n.
=m [ddE;Y -7
1
1
dx -7
+ dE; J
J
J
j=l
L [I
>l.
(95)
J
Two dimensional Gaussian quadrature can be applied for the
element integrations. See Figure 17.
53
---
global coordinates
boundary element
Internal element
on r
'fl2 U + b = 0 b constant
u=;ionr 1 q
as
o (96)
(97)
BIBLIOGRAPHY
loc:£.
global coordinates
coordinates
(x~.y~;O.1)
1. SIMPLE-LAYER FORMULATIONS
According to the Dirichlet existence-uniqueness theorem, there
exists a unique harmonic function ~ in Bi which assumes
prescribed continuous boundary values on a Liapunov surface aBo
To construct ~ in Bi, we write
~(e) = f g(.E.,.9)a(g)dq ; .E.
Co Bi , (l)
aB _.
where a appears as a hypothetical Holder-contin~ous source
density to be determined. In principle a = - 1TI (~i + f~),
but neither ~i nor fe are known ab initio. TI
An effective way forward is to note~oth sides of (1) remain
continuous at aB, i.e. both ~ and its representation remain
continuous at aB, so yielding the boundary relation
aB
I ~i (.9.)dq = o. (7)
I g(£'.9.)i~(9)dq
aB
- 271~(£) = 0 ; £ Co aB, (10)
I ~(p)~i(£)dp = I ~i(£)dp =
aB aB
0, (11)
I
aB
gi (e.,.9.) A(.9.)dq - 271A(£) = 0; £ Co aBo (12)
o = 0 + kA ( 13)
o
I
aB
g(£,.9.)oo(.9.)dq + k I
aB
g(£,.9.)A(.9.)dq, (14)
i.e.
I
aB
g(£,.9.)oo(.9.)dq + k; .E. Co Bi + aB, (15)
J ~~(£)dp = -211 J J
o(.9.)dq - 211 o(£)dp = -411Jo(£)dP•
aB aB aB (20)
aB
The same procedure applied to (8) gives the expected result
since
IaBg{£,S)A{j~)dp = J g(9.,£)A(~)dp
aB
= 1 ; .9.,£ GaB.
2. DOUBLE-LAYER FORMULATIONS
The representation
~(£) = J g{£,.9.)i].l{.9.)dq ; E. Co Bi (22)
aB
provides a classically preferred alternative to (1) fo~.
*
Dirichlet problems, where ].I appears as a hypothetical Holder-
continuous source density to be determined. In principle
].I = r~-f), but of course f is not available ab initio.
An effective way forward is to note that ~ remains continuous at
aB, whereas the integral jumps by -2~].I{£) at E. caB, so
yielding the boundary relation
62
f g(e.,.9.)i ll (g)dq
+ 2n"ll(e.)= <p(e.) ; .e. c. aBo (23)
aB
This may be viewed as a Fredholm integral equation of the second
kind for II in terms of <p , to which classical Fredholm theory
applies: the homogeneous equation
f g(.e.,.9.)i ll (.9.)dq
+ 2lfll(.e.) = 0 ; .e. c. aB (24)
aB
is mathematically equivalent to (18) and therefore has no non-
trivial solution; consequently equation (23) has a unique
solution, which generates <p in Bi via (22).
The representation
<p(.e.) = f g(.e.'.9.)~Il(.9.)dq ; .e. C Be (25)
aB
aB
63
~ = 0(r- 1
) as r ~ 00. The difficulty may be overcome by
writing
(29)
f A(.e.)
r
~~ (.e.) - cl dp = f A(.e.) ~ (.e.)dp - cf
lElJ A(p)
1.e.1 dp= O. (31)
aB L aB aB
i.e. J g(.e.,.9.)~
]Jo(.9.)dq p Co B
- e
(33)
aB
for insertion into the representation (29).
3. DIRECT FORMULATIONS
Both the Dirichlet and Neumann problems may be formulated
directly through Green's boundary formula. Thus, given ~ on
aB (interior Dirichlet problem), it becomes a Fredholm
integral equation of the first kind for ~i in terms of ~, i.e.
aB aB
This has a unique solution as already noted in connection with
equation (2). With ~ ,~i both known on aB, we may generate ~
throughout Bi via Green's formula. It requires proof,
however, that ~i satisfies the Gauss condition (7) : operating
with f ... A(.e.)ap upon both sides of equation (34), and
aB
interchanging the order of integration where appropriate, we
obta in
J ~i(.9.)dq f g(.e.'.9.)A(p)dP=I~(g)dq J g(.e.'.9.)iA(.e.)dP-21TI~(.e.)A(E.)dP'
aB aB aB aB aB
64
i. e.
jel>i(g)d q f g(g,E)A(e)dp = f 4>(g)dq f gi(g,E)A(e)dP- 2Tff4>(e)A(e)dP,
aB aB aB
f A(.e) {f l
g(.e,g)4>i (g) dq dp = 0,
dB aB J
i.e. f 4>i (g)dq f g(.e,g)A(.e)dp = f 4>i (g)dq f g(.e,g)A(.e)dp
aB dB aB dB
aB
I g(e.'.9.)~~(9)dq I g(e.'.9.)~~(.9.)dq
=
aB
- 21T~(E) ; .E. c aB,(41)
aB
I ~~(.9.)dqaB
= -41T I A(.E.)~(.E.)dp, (42)
aB
I g(.E.,.9.)~~(.9.)dq I
- 21T~(.E.) = g(.E.'.9.)~~(.9.)dq ; .E. c aB,(43)
aB
i.e. a Fredholm equation of the second kind for ~ in terms of
~e. This always has a unique solution as follows by reference
to equation (18).
The Dirichlet and Neumann conditions are particular cases of a
prescribed linear relation
a~ + S~' = Y; ~' ~~
1
or ~'e (44)
at each point of aBo Thus,
a = l,S = 0, Y given (continuous) on aB (45)
I gi(.E.·~)As(~)dq
aB
- 2TIA S (E)=0 ; s = 1.2 ••.• 6
.E. c: aB • (51)
IaB~(E)As(E)dp = Icr(q)lls(~)dq
aB
s = 1.2 ••.. 6 (52)
67
in parallel with (21).
Kupradze was the first to propose that the vector interior
Neumann problem could be formulated by the vector integral
equation (8). with the associated homogeneous systems (9).
(10). Naively applying classical Fredholm theory to this
system. it follows that a solution only exists if
I <p~(£)].Is'-E)dp =
aB
0; s = 1.2 •.•. 6. (53)
(54)
I
aB
g(£ •.9)oo(g)dq + ks I
aB
g(£ •.9.)As(g)dq ; £C Bi + aBo
i.e.
pcB.1
-
+ aB 1.• (55)
.I ...
regular exterior displacement field (19). Operating with
].1 (p)dp
s -
upon both sides of (16). we find
aB
I <p~(£)].Is(£)dp = I
aB
-4n
aB
cr(£)].Is(£)dp; s = 1.2 •..• 6.
i.e.
aB
I <P~(£hs(£)dp I = -4n
aB
<p (£) As (£)dp (56)
from (52).
68
I ~<'e)As<'e)dp
aB
= 0; s = 1,2, ••• 6 (57)
~ = ~o + L ks~s
1
' (58)
I g<'e.'.9.)~~o<,g)dq
aB
+
6
~ ks J g<'e.,.9.)~~s(.9.)dq;.E. c Be ,(59)
aB
I g(.E.'.9.)~~o(.9.)dq
i.e.
; .E. c Be + aBo (60)
aB
Condition (57) stems essentially from the fact that the vector
representation (25) has only 0(r- 2 ) behaviour at infinity,
whereas a regular exterior ~ has in general 0(r- 1 ) behaviour
at infinity.
Following (29), we extend the vector representation (25) by
writing
~(.E.) = J g(.E.,.9.)~~(.9.)dq + ~·g(.9.·.E.)q=O
aB -
+ E. AV. 9(.9.,.E.).9.=0 ; .E. cBe (61)
b_a__ b_a_>
1 aq2 2 aql
3
71
72
for any p,q C dB. The strongest class of Holder-continuous
functions 1s defined by a=l, since the condition (i.e. the
Lipschitz condition)
1cr (e)-cr (g) 1 < 01.£.-.9.1 ; 0 > 1 (4)
(6)
where ~'~q are the unit normal vectors at any .£.,g C dB. This
is a Liapunov surface. If dB has the local equatlon z = z(x,y)
at any point p C dB, with the z-axis pointing in the normal
direction at p, then (6) implies that the partial derivatives
ZX,Zy are Holaer-continuous at p. However, zxx etc., do not
necessarily exist at p, i.e. the Liapunov surface has a
continuously varying normal direction - but not necessarily a
curvature - at every point. A simple example is provided by
the circular cylinder capped by hemisphere at each end, since
the curvature jumps on passing from the cylinder to the
hemisphere whilst the normal direction remains continuous.
Subject to (3), (6) above, the main properties of V may be
summarised as follows.
(1) V is continuous and differentiable to any order in Bi,Be'
I
Also
---ap;:- = ~~l (.£.,.9.)cr(.9.)dq (7)
dB
etc., where .£. = (Pl'P2'P).
(2) V satisfies Laplace's equation in Bi,Be i.e.
as -1
p. ~p - ; -1
p. Co B·l
1
(10)
as .E.e + £ ; .E.e c. Be
-- =
aV(£) J atag (£,.9.)o(g)dq ; £ c. dB, (11)
dt aB
where a/at denotes differentiation along any tangential
direction to dB at £.
(6) V has two formally distinct normal derivatives V~,Vi
at p caB pointing into Be,Bi respectively. These
may~e constructed by writing
it follows that
showing that
lim 9~(R'~) as q + P = lim(-4a 2sin ~)-l as e + O,etc.
- - (16)
J g~ (R,g)dq = -
aB
I
aB
gi (R,~)dq -21T; R c. aB (17)
J ar
aB
age
I
aB
agi dq = 0
ar
( 18)
from which
(20)
(25)
£e -+- £
assuming that W(£) is defined everywhere by (24).
(5) Even if 11 is Holder.-continuous, the normal derivatives Wi,
We are not necessarily finite though they satisfy the relation
limWi(£i) + limW~(£e) = 0; £ Co aBo (29)
£i ->- £ £e ->- £
(6) The separate limits in (29) exist if 11 is Holder-
continuously differentiable, in which case the tangential
derivative of Walso exists :
aw(p)
at - =
f 1ta (£·~)il1(g)dq
; £ c. aB (30)
aB
as in (11). Also ~ satisfies the limiting relations
. aW(£i) aW(£) al1(£)
11m ~ ar-- + 21T ar-- }
£i -+- £
(31)
aW(£e) aW(£) ajl(p)
1im -at-- ar-- - 21T at -
e
£e ->- £
where a/ati' a/ate denote differentiations parallel to a/at
at £i'£e respectively.
77
W(e) = ~o J
g(e,.9.)i dq = llOrl (32)
aB
where rI denotes the solid angle subtended at p by aB. This
interpretation immediately yields the result -
J g(~'.9.) i dq
= 411 ; £ c Bi (33)
aB
for aB closed, which may be viewed as the Gauss flux theorem
applied to a unit simple source at £. It follows from (28)
that
J g(E.>.9.)i~(g)dq, (39)
aB
and the simple-layer potential
- J g(£'.9)~~(9)dq, (40)
aB
which have the properties associated with W,V respectively for
Holder-continuous boundary data on a Liapunov surface aBo
Superposing (39) and (40) yields the identity
J g(£'.9.)i~(.9.)dq
- J g(£'.9.)~i (.9.)dq = 0 ; £ Co Be (43)
aB aB
This may be proved directly by noting that g(£,.9.) is a harmonic
function of .9. in Bi for any fixed £ C Be : also .p(.9.) is a
79
harmonic function of q in Bi' and any two harmonic functions
~,~ in Bi satisfy the reciprocal relation
f (~~i f
- ~~i)dq = (~V2~ _~V2~ )dq = 0 (44)
aB Bi
as follows directly from the Gauss divergence theorem. Starting
from (43), and reversing our steps, we successively recover
(42), (43). Of course, this procedure hinges upon the validity
of the boundary jumps (28), which can only be justified by the
same kind of limiting analysis as would be involved in the
direct proof of (41).
Green's formula may be readily adapted to a regular harmonic
function f in Be' which assumes boundary values f(q), and
boundary normal derivatives fe(q), as q runs over -aBo
Corresponding with (41), (42), T43) we-nave
g' (P1qZ)
g' (P2qZ) (55)
g' (P3 qZ)
where g'(Paqn) signifies the traction component in the a-
direction at p generated by a unit point-force acting in the
n-direction at~. Clearly column 1 defines the traction vector
at p generated oy a unit point-force acting in the l-direction
at S, etc. Finally, corresponding with g(£,~)', we construct
the traction dyadic
g(P1q2) ,
g(pzqz)' (56)
g(P3 qZ)'
where row 1 defines the traction vector at ~ generated by a
unit point-force acting in the l-direction at £, etc. It may
be readily proved that column 1 defines a singular displacement
vector at p, i.e. that generated by a unit traction-source
associateC-with the l-direction at q, etc., in line with the
fact that g(p,q)' may function as a-unit dipole-potential
generated aCq-:- By the same token, row 1 of (55) defines
the singular aisplacement vector at g generated by a unit
traction-source associated with the T-direction at p. Any
individual component of (55) or (56) carries two possible
interpretations : either the traction component generated by
a unit point-force or the displacement component generated
by a unit traction-source. The interpretation will always
be clear from the context. We note that g'(p,q) stands for
ge(£,g~ or gi(£,~) as the case may be, and sTmllarly for
g~£,~J .
1
independent vectors
~l <1,0,0> , 11 2 = <0,1,0>, ~3 = <0,0,1>
(60)
~4 <1,0,0>/\ E.'~5= <0,1,0>1\E.,11 6 = <0,0,1> AE.J
which provide the six vector double-layer identities
=0 £ c Be (63)
George T. Symm
National Physical Laboratory, Teddington, U.K.
1 INTRODUCTION
(1)
0(p) = Ic a- (q)loglq-pldq, p e D, (2 )
85
86
Cp = 1 - (lVI/lUI) 2 , (4)
']t- = - x, (5)
so that
(7)
where
V = - V(<<I + ~) (8 )
for 0-'. This equation has a unique solution and, integrating the
equation with respect to p, noting that, by virtue of the
Cauchy-Riemann equations (describing the contour C in a clockwise
direction) ,
whence
where
and
If we can solve equation (11) for 0-', then we may evaluate these
derivatives and hence obtain the pressure coefficient C from
equation (9). p
88
3 DISCRETISATION
where J. J
denotes integration over the interval I. of C.
J
Solution of this set of simultaneous linear algebraic
equations leads to the approximations
N
0 x (qi) = L
j=1 J
0-. J.IOg I q. -q I dq +
J x 1
7r °ix ' (qi ) (22)
and N
0 y (qi) L
j=1 J
0-: J.Iog I q. -q I dq +
J Y 1
'7t~y' (qi) (23)
(24)
In the example which follows, each interval of the boundary
is approximated by a chord and the nodal points are defined as
the mid-points of these chords, so that
(26)
89
(27)
Hj
and
by the transformation
from which it may be shown that C varies between 1.0 and - 0.876
over the surface of the aerofoi1. P
For any such problem, either III or III' may be eliminated from
equation (3) at each point q E C by means of the boundary
conditions. This yields, in general, a system of coupled integral
equations for those boundary values of III and III' which are not
eliminated. By solving these integral equations and substituting
their solution, complemented by the boundary conditions, back
into Green's formula:
91
P E D, (36)
Ic o~(q)loglq-pldq = 0, P ~ CO'
DISCRETISATION
fl = £a - E.
a fl' (40)
when a is greater in magnitude than b at the nodal point and
fl' = £ - .! fl (41)
b b
otherwise. We thus obtain a system of N simultaneous linear
algebraic equations (in N unknowns) whose solution provides, via
equation (36) and the boundary data, the approximation
N N
-
fl(p)
1
= 2'7t' l/j Jloglq-pldq -
\" 1
2'7t'
\"
j~/j flOg' Iq-pldq (42)
Sjloglqj-qldq = Iqj-qj_I(loglqj-qj_1 - 1) +
Iqj-qj+ I (loglqj-qj+1 - 1) (44)
for the case when i = j in equations (39).
93
Ic o~(q)dq = 0, (45)
I 6 'hJ
j=1
j o =0 (47)
94
5 APPLICATIONS
The method described above has been successfully applied to
a variety of practical problems including measurement of crack
lengths via electrical potentials (McCartney et al., 1977),
capaCitance calculations (Symm, 1981) and determination of the
temperature distribution in a turbine blade (Symm, 1982). The
first two of these problems involved mixed Dirichlet and Neumann
boundary conditions, while the third application required the
solution of a Robin problem in a multiply-connected domain.
In the example which follows, the method is applied to a
typical heat conduction problem - a mixed boundary value problem
in a simply-connected domain. This problem is chosen particularly
to illustrate the effect of a boundary singularity.
R.----------------------------,Q
D c
S'---------I---------'p~
o x
Figure 2. Domain of Example 2
BOUNDARY SINGULARITIES
p e D = D + C. (53)
u' =0 on OP,
u = '000 - a,r'/2cos(9/2) - a2r3/2cos(39/2) on PQ,
u' t
= a,r-'/2 sin (9/2) - i a 2r'/2sin (9/2) on QR, (54)
u' = - ~ a,r-'/2coS(9/2) - i a 2r'/2cos(9/2) on RS,
u = 500 on so.
u, = 500, uN = 0, (55)
in the first and last intervals of the boundary C, starting from
the origin and describing the contour in an anti-clockwise
direction. The N equations may then be solved and an
approximation U to u, analogous to the approximation (42) to 0,
obtained at any point of the domain. The required solution 0 is
then approximated by
7 COMPOSITE DOMAINS
(57)
where
(58)
8 CONCLUSION
REFERENCES
George T. Symm
National Physioal Laboratory, Teddington, U.K.
1 INTRODUCTION
(1)
101
102
!6(p) = Js
cr(q)
- - dq, P E D,
Iq-pl
(2)
1(]"'
S Iq-pl
(q) dq --
!6(p), pES. (3)
3 DISCRETISATION
-
N
e(p) = L~ Jj lq-pl-1 dq , P E D,
j=1 J
4 GENERAL DOMAIN
(8)
where A is the area of the triangle and the sides of the triangle
have lengths a, band c, with the latter opposite to the vertex
p. Thus, if P is at the centre of a square element Sj of side h,
for example, it follows that
(10)
C= Is 0- (q)dq, ( 11)
I
S
0- (q)
- - dq
Iq-pl
= 1, P € S. (12)
to the capacitance C.
Table 1 shows results obtained by Laskar (1974) using this
method, which is essentially the same as the "method of subareas"
of Reitan and Higgins (1951).
N n C
150 6 0.6538
294 10 0.6568
486 15 0.6582
726 21 0.6590
1014 28 0.6595
1350 36 0.6598
1734 45 0.6600
2166 55 0.6602
2646 66 0.6603
105
5 AXISYMMETRIC PROBLEMS
(15)
Defining
k 2 = 4rR/{(z_Z)2 + (r+R)2} (16)
2 2 2 1/2 2 2 2 1/2
a 10g[b+(a +b +d) ] + b 10g[a+(a +b +d) ]
(a 2+d 2 ) 1/2 (b2+d2) 1/2
-1 [ ab ] (21)
- d tan 2 2 2 112 •
d(a +b +d )
2 2 1/2 2 2 1/2
4{a log[b+(a :b) ] + b log[a+(a ~b) ]} (22)
Ie(p) = J
'rr/2
'Tr/2-e
(1 - k
2 2
sin u)-
112
du (23)
(24 )
0.5 log x
c = fL cr (q)dq, (27)
1
c = - fL ~ (q)dq. (28)
o
To obtain the charge distribution ~ on either conductor, we
must consider the potential problem for the infinite line as a
whole. However, for practical reasons, we truncate this line at
some distance T beyond each end of L and let So and S1 denote the
outer and inner conductors of the truncated line of length
L + 2T.
C+ = L OJ + Aj' (29)
N NO N1 C+ C- C
CONCLUSION
REFERENCES
H.L.G. Pina
Department of Mechanical Engineering, Instituto Superior
Tecnico, Lisbon, Portugal
1. INTRODUCTION
pc
au
at = v.g + r (la)
111
112
heat flow by conduction into the unit volume and the second
term in the right hand side represents the heat generated by
sources in the unit volume. The above equation must be
supplemented with a constitutive relation for the heat flux q,
which in engineering applications is the well known Fourier's
law,
(lb)
pc
au
at = v. (~ v u) + r. (2)
on rl ' (4)
where u is a function giving the prescribed temperature and r l
denotes the part of the boundary r of the domain n where this
temperature is enforced.
(5)
(6a)
on r3, (6b)
(7)
u * (~,~') =- 211
1 tn r, in the two dimensional case, (lOa)
*
u (~,~')
1
411r ' in the three dimensional case, (lOb)
(12)
I I
u= OI.(u * £)dOl.+ r<u *q-uq * )dr. (13)
u = r (u *q)dr -
J Jr (uq
- *)dr. (15)
(lla)
q= q on r. (17b)
This function q cannot be arbitrary specified. To verify this
116
(18)
(19)
Jf u df = 0 • (20)
J J
cu + f (q *u)df = f (u *-q)df. (21)
in n, (22)
u= u on f land q = q on r 2 (23a,b)
for ~ € r 1 (25a)
(25b)
in n, (26a)
u + aq = ~ on r. (26b)
= c (~-a.q) . (28)
118
In this equation the unknown is the heat flux q. Collecting all
terms involving q in the left hand side we get
- For ~£ f I:
- For ~£ f 2:
(34a)
(34b)
(35a)
120
{b} (35b)
4. POISSON EQUATION
When heat sources are present the differential equation
governing steady heat conduction is
(36)
The heat sources term f gives rise to an integral in the domain
n, as equation (13) shows. This integral presents a
complication and we now review the methods to deal with this
added complexity.
o in n, (37a)
u=u in r. (37b)
122
The particular solution v obeys, by definition,
in o. (38)
w=u-v, (39)
in 0, (40a)
(41)
J
-. (u *f)dO=
.0
fr (v -afan - f ~~)dr. (43)
v =-...!:...
8'1[
r2 (R.nr-l) for the two dimensional case, (44a)
1
v -r for the three dimensional case. (44b)
411
Then if the heat sources term happens to be harmonic the
123
(46a)
(46b)
124
where the subscript a refers .to subregion 0a' and [Hl2 1. for
instance. is the submatrix obtained by collocations it nodes
of r l • Similarly. for subregion ~ we can write
and (48a,b)
Note the change of sign in the last equation due to the fact
that the exterior normals at r 2 of the two subregions point to
opposite directions. Introducing equations (48) in (46) and
(47) and rearranging conveniently we can present the final
system of equations in partioned matrix form as follows
Hll H12 _c l2 1 b al
a 0
a ua
a
H2l H22 _c 22 2 b a2
a a 0
a ua
(49)
0 ~2 ~3 Cb23 ub3 b2
b
32 2
0 ~2 ~3 Cb qa bb3
We notice that now the final system (49) exhibits some sparsity
which can be taken into account when programming this method.
6. ANISOTROPIC BODIES
7. REFERENCES
Y.D. CHANG, C.S. KANG and D.J. CHEN, The Use of Fundamental
Green's Functions for the Solution of Problems of Heat
Conduction in Anisotropic Media, Int. J. Heat Mass Transfer,
16, 1905-1918 (1973).
S. CHRISTIANSEN, Integral Equation without a Unique Solution
can be made Useful for Solving some Plane Harmonic Problems,
J. Inst. Maths. Applies, ~, 143-159 (1975).
L.C. WROBEL and C.A. BREBBIA, The Boundary Element Method for
Steady State and Transient Heat Conduction, in Numerical
Methods in Thermal Problems, R.W. Lewis & K. Morgan (eds.),
Proceedings of the First Conference, Pineridge Press (1979).
Chapter 8
H.L.G. Pina
1. INTRODUCTION
2. ISOPARAMETRIC ELEMENTS
~. (a.) = 6 •• , (2)
~ -J ~J
127
128
~~--~----~--x
-1 o
Linear element
U1
Quadratic element
- =2
N2 (x) 1 (l+x).
- (5b)
(6)
(7)
(8a)
N2 = (l-x)(l+x), (8b)
1 -(
N3=2 -) •
x l+x (8c)
for x e: T • (11)
- e
(12a)
(12b)
(12c)
2
2 )(,
linear triangle
X2
5 5
6 6
, 3
2 3 X,
~
Quadratic triangle
and
(14)
The quadratic triangle - This element has six nodes and its
shape functions are second order polynomials, as follows,
(15a)
(lSb)
(15c)
(15d)
(lSe)
x'" r
6
i=l
N. a.
~-~
(16)
u= r
6
i"'l
u. N.,
~ ~
(17)
3. NUMERICAL INTEGRATION
(18)
J
E
1= I
e=l
I
e with I
e
= r f(x)dr •
- e
(19)
e
x= cP (x). (20)
- -e
equation (19) becomes
I
e
=Jfe
r (x)dr with f e (x) =f(cp e (x»Je (x). (21)
elements r e .
(23)
It can be proved (see e.g. Krylov, 1962) that Gauss type rules
give the best accuracy for a given number of integration points.
In fact, a Gauss integration rule with P integration points is
exact for all polynomials of degree up to 2P-l. Experience
shows the following rules to be a good compromise between
accuracy and speed:
fT
E
I= I I
e
wit:'h I
e
.. f(x)dT •
- e
(24)
e=l
e
(25)
and thus
I =
e
fT f (g)dT
- with f e (x)
-
= f (41 (x»J (it).
-e - e-
(26)
136
[ a~i
- (g) ] , (27a)
ax.
J
with
(27b)
I = f
R.n r (x)
ere
f e (x)dr. (32a)
where
R.n I-=:-
r (x)I Ir I
x = R,n -=:- + R.nlxl.
(x)
(33)
x x
The first term on the right hand side can be seen to be
bounded when x tends to zero. Thus the integral (32a) splits
into a regular part and a singular part written in a form
suitable for application of integration rules with logarithmic
Kernel.
or
I
e
= fn ~
r\x)
1
f(x)dn •
- e
(34b)
e
dn rd8 (35)
e
we conclude easily that both singularities in equations (34)
disappear. We can then employ the standard Gauss type
integration rules. Another possibility is to employ the general
technique developed for isoparametric elements in the following
way.
4. REFERENCES
1 INTRODUCTION
The start of Boundary Element applications in an
environment that was mainly oriented towards the
Finite Element Method is shown on the example of
introducing the BEASY-System at Brown Boveri in
Switzerland.
The CAD-programs and the pre- and postprocessors
used in connection with the Finite-Element-programs
could be adapted with little effort to the genera-
tion of geometry and the plotting of results for
the Boundary Element Method (BEM).
The great advantages of the BEM are shown on seve-
ral practical applications, especially in connection
with an existing program for contact calculations.
Also comparisons with results of Finite Element cal-
culations are presented.
141
142
FINITE-ELEMENT GENERIERUNG
BEA5Y ADINA
FIG. 1
143
FINITE-ELEMENT AUSWERTUNG
F1G.2
144
BEASY SYSTEM
i SEASYG
Preprocessor
BEASY SYSTEM
BEAXPP BEAXTE
BE2DPP BE2DTE
BE30PP BE.3DTE
BEASYP
ostprocessor
FIG. 3
146
STRESS
FILE
_try
Boundary
BEASY6 Boundary
SctIlUiftiiCIIIII Solution
Internol
Solution Solution
FIG. 4
147
Advantages:
- Only boundary to discretize (1 dim less than FEM)
- Not so much experience necessary as in FEM to decide for
the optimium element size
- Coupling to Computer Aided Design programs much easier
- Accuracy much better with the same computer time. especial-
ly for stress concentrations
- Infinite problems easy to solve
- Numbering of nodes and elements not important
- Simple interpretation of results
- Results at internal points after the main calculation
as requested by the user
Di sadvantages:
- For nonlinear problems the inner region must be discretized
- Not economic if surface/volume large
- Material properties constant in zones
- No internal forces except gravity, centrifugal forces and
steady-state temperature
- No advantages for beams, plates and shells
Conclusions:
The BEM will certainly not replace the FEM for structural
analysis, but it is a valuable supplement for certain types
of problems:
- 20. axisymmetric and 3D. if ratio of volume/surface is
large
- stress concentrations
- contact problems
- infinite problems
4 INTRODUCTION OF BEASY
It is not so easy to introduce a new method in an industri-
al environment which is already oriented towards the FEM.
148
LO
(!)
LL
o
o
o
o
o
o
.,o
~
D~
(OtDlDCDalCDCOOOj
(Im
0::
I
(I
I ~
O~
Ii
S.300E·02 l LRST ~rHa"l
I ---~
S.2S0E1"02 ,1- J 1I
5.200E·02
I I 1...-
5.150E·02 1--- 1----
5.100E·02 I- --
J
-
5.050E·02
/
s.oaOf:tC2 /
/
~.950E·oa _L
1/
~. 900E'02
Il.BSOE+O
10 II 13 I~
I
la 15 17 19 21 23 25 27 29 KHOTE~
HD-RRRD ....
.j::o
FIG. 6 '"
TEMP RK BESCH
150
"
""
""
"
z
~
%
...."
.."
~
o
'-
<=
'"c
'"
"o
'"
('(")0
"'"
~ __~________~l
152
I
i I I
~
~
0
I
i I
I CO
I I I <.':i
i I I
iL
\\ fr I ~
~
0
\\ I
I
r
I
I
I, r I \~ t
,, I , .~
i
,~ I I
!
\1
V //,
-
I
/ m
! !)
m
.... -....! ~ 0
_/
m
0
,
W
-IJiII N
e l-
./' ,/ir. cr:
I-
-~
/' on
-.l
0
"
:::J
... (f)
~ 0
0 W
a::
: I
I"~ ".
"-
, I
a:
Il- :lC
'" en
; I
i 0
0
cr:
I
(Y)
I m
e
0
153
5 EXAMPLES
Three typical examples for the application of the BEM are
shown and comparisons to FEM-calculations are presented:
Temperature calculation for a part of a turbine rotor
The temperature calculation with heat transfer boundary
condition from node 8 to 65 (Fig. 5) showed very good co-
incidence with a FE-calculation. Linear boundary elew~nts
were used and the temperature was also calculated at 14
internal pOints.
Fig. 6 shows the temperature distribution from node 8 to 30.
Contact calculation
Fig.ll shows the FE-discretization for a three-dinlensional
contact-calculation of a blade root. This very coarse mesh
could not give the right results at the notches, so addi-
tional calculations for the notches were necessary.
The same contact-program (4) as used with the finite ele-
ments can be used for boundary element calculations, be-
cause this program was already changed before from using
contact forces to using contact pressures as unknowns for
the contact calculation.
154
1
,
;;; ;;I oj
~ ~ d
~
~ ~ ~ 0
~ ~ on
on on
'" on
" '"
i
'f 1 m
1 (!)
LL
:c
u
(J)
I-
a:
4 a:
0
a:
I
1..
~o
0-
1 I
:c"':
u -
0
0
0
1
(J)
~g
1-':
a:-
a:
Cl
0:%
1
~~
o~
N:5
:;:
I-
..:r ~~ =~ 0::z _ ru
= ~
-:;:
lL!
1
,(n'(l't1'To fn' (rT =r :Co
Lo
~'"
'"
lL!
~
1 l.l..o
1
11
0
:;::~
>--
(J)o
1-0
0'"
a:
1"'
]
I
:l:~
a~
1 -:
'"
~ l!! ~ .. .. d
I
iii I ,.
155
- I
i I I ~
I
N
I Ii
I
I \ I i
I \ I I
iT '"
~
! \ 1 I
I
I \11 I I
~
I
7-
i
. ~
I
U
(f)
t
I
\\
I
I
f-
I a:
I :\l
a:
\ I 0
I ~~\ di ,.. a:
! I '"
::l
i\\
0
I
iI
I I
I r 1\
, \ I
II
I
t In
'"
l \« I
t
u
f ..;i en
'"'"
~
I i
~/~
~
f-
I !
~-
~~
a:
a:
o
~ , ~ a:
~~
, ::l
-~ r o
/'
/"
7( I 1 C\j
-- -~ ::;;\ '-';
".
l(J
, '\~, ,
, , T
I
'..,. r 'W
I i :i!: LL
I I ~' I
T
,
I I t ;;
I "'"
156
6 CONCLUS IONS
The introduction of a Boundary Element program like BEASY
in an industrial environment is most easily done, if the
boundary elements are treated like simple finite elements
and the existing FE pre- and postprocessors can be used.
After the FEM-user's have tried examples and made compa-
risons with FEM-results, they soon will be convinced of
the advantages of the BEM for certain types of problems.
Especially the easy preparation of input data and inter-
pretation of results will help the spreading of the BEM
in the future.
With the steadily increasing use of Computer Aided" Design
programs another great advantage of the BEM will "get more
and more importance. One can use the basic geometry, ge-
nerated by CAD-programs, directly as input to the BEASY-
System and has only to add the boundary conditions for
completing the input data.
In general one can say that the BEM will certainly not
replace the FEM for temperature and stress analysis, but
it will be a valuable supplement for certain types of pro-
blems, like stress concentrations, infinite problems and
contact problems.
REFERENCES
1. Brebbia, C.A. The Boundary Element Method for Engineers.
Pentech Press. 1978
2. Zienkiewicz. O.C •• The Finite Element Method in Engineer-
ing Science. McGraw Hill. 1971
3. BEPSY User's Manual, Computational Mechanics. 1982
German version by FEMCAD AG, Oberrohrdorf. Switzerland
4. Kuich G•• Berechnung allgemeiner Schrumpf- und Kontakt-
probleme, FEM-Kongress Baden-Baden. 1975
157
Lastelement
FIG.11
158
2-dim. Boundary
FIG. 12
Chapter 10
Gero I<uich
1 INTRODUCTION
159
160
(!)
......
LL
..
"'><
--
IJ)
m
0
W
..J
w b
..J
3: it
~
~
162
o
163
After program name and title one defines general
data, in this case the symmetry of the problem
with respect to the planes X=O, Y=O and Z=O by
(SX,SY,SZ). On symmetry-planes no elements .have
to be defined.
3 CAD-coup I ing
UJ
....J
~ ~
~======-___-----.J~
165
166
It is further possible to delete lines, change the
direction in which the line is defined and define
the number of elements per line. One can also cre-
ate new points or lines and define the boundary con-
ditions. So just with an CAD-File all necessary data
can be completed interactively and the calculation
may be started with the final geometry (Fig.6).
Pr"eci si on
& .J
I In
I
I
I
(!)
....
I LL.
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
- I
I
I
I
I
I
I
I
I
oJ
• I
I
I
I
I
.... ..
..J
-
(D
....Q..
IT)
I w 0
I
I
"
(J)
I ...... I-
I
w
CD
9a.
I '"I
I
- .1.----------1-
~~
168
1.& 13 12
I to
I
I (!)
.
I .....
I lL.
~
I
t
I
I
I
I
I
+
I
I
I
N I
I
tI
I
I
I
I
t
I
I
" I
I
. III TI
I
I
I
--
to
I
. I
I
.....
....J CD
I ....
(I)
UJ
.. I a..
d
I
I ....
U)
...
• I
UJ
III
9Go
I ~
I
- ---::---:-- ... --------"1-
~
169
There was a pressure load applied between points 3
and 5 and zero displacements prescribed between 5
and 7, so the stresses should be zero in the notch.
The following table compares the displacement and
stress in point 19 as calculated on different com-
puters and with different precision.
-----------------_ ..... _-----------------_._------------
I Computer I Precision I Displacement I Stress
_._-------------------------------------------------
I PDP I single I -0.2853 I -12.8~4 I
I VAX I single I -0.2853 I -16.487 I
I CDC I single I -0.2842 I 0.434 I
I IBM I single I -0.2821 I -148.22 I
I IBM I double I -0.2842 I 0.434 I
Computer dependence
r_ -,
I 1
, ".
....
(!)
LL
!
I
I I
I
I
, I
I
I
~
I
I
I
I
I I
I
I I
I I
I I
(()
I
I I
I I
I I
,
+
I
I
~
•
~ I
I
I
~
I
I I
I I
I
.I I lJJ
I I -l
I I 0
:J:
I I
I I :J:
I
I
I
~
II l-
--
I
...
10
ct:
<
:::> t-
o(1) 0
~
~~
I
I
--
171
2
m
..-- ,......
~ ..
,.-- ..--
iG
..
-..--
fa
V
III
(
---
'A
-- ----
II
r--. -!!I •
....
.-
~:. •
=
~
..-- v--- -= ...
W
...J
~
----
0
.- :!
::I:
(
::I:
I-
-- -- --
......
-!!I • :a
- --..-
w
I-
!1 <
...J
r--
-= ..
a... GI
W
I"- ct:
< ....
r-. ::J
2 a(f) 0
I I I I I I., I'#
~
-
01 ... .: II
of
172
Element subdivision
- ::. ::.
01
!!!
....(!)
lL.
~
!!! -+
...
!l
-
~
N ::
51
I-
" (f)
1 --I
w
I-
>-
la an
.
0 Cl)
(l)
I d
a
......
....
t::) t-
o
..
a::: -'
a..
., '" . ~
174
o
I
I (!)
.
I .....
I b...
I
I
I
I
I
I
I
!
..
.g
II
I
I
§
I I:
o
I
,
I ~
I
I§
,,, C0-
o
,,
w fi
~§
:::l I:
0"tl
I (J) 5
I
... / <
:t
3j u.
Of
>.
I
,
I
• I-~
I I
I Z (II
I < ~
j/
/ I 1--
(J) (II
/
/ ~'!;
/ U I:
1.-/ zI ..0
//// 8~
/// §
.
(J)
// 0:: Co-
~ ~
§
01
" Q
X Error
-- - - .- -
-1.0lI0 L LN,-- ----
-- -------
V "...
1-""- LN2 ........--
~~
;'
;'
... -- ",'"
"'~
~~
--------
~~'"
-2.0lI0
/ "'~
....
/ /'
/ .
I ,/ ..'
-1.000
I /
I ,/
I /
I /
-.4.0lI0
/ I
/
I
I /
I / I
-5.000
4 12 111 20
• NuIb. . of> unknowne
/ '\
rI ~ -
-1.SIlEtII
\
\
2 8 4 5 8 7 • I W U 12 18
,-18_____
12
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
.1 ___________ _
L
I
ELECTROSTATICS PROBLEMS
R.A. Adey
1 INTRODUCTION
The solution of electrostatic problems which obey the Laplace
(Potential) equation has been accomplished over a number of
years using techniques such as finite element, finite differ-
ences etc. However the boundary element method has recently
proven proven itself [1] an extremely powerful technique for
the solution of this type of problem because it can easily and
accurately model the singularities which commonly occur in this
type of problem. Secondly and perhaps most importantly it
easily models infinite regions.
2 THEORETICAL BASIS
kV' 2 v = 0
where
k conductivity
v voltage
i - kV'v (2)
or expanding
177
178
i k dv i k av etc. (3)
x dx y ay
i heat flux
v temperature
3 BOUNDARY ELEMENTS
c(x)v(x) +
J i *(x,y)v(y)df y = Jv *(x,Y)i(y)dfy (4)
f f
where x
source point coordinates
=
field point coordinates
y
c(x) constant depending upon the position of the
source point e.g. on a smooth boundary c(x) .5
*
i*(x,y) = current density fundamental solution
v (x,y) = voltage fundamental solutmn
vex) voltage at the source point
v(y) voltage at the field point
iCy) current density at the field point
df surface area at field point
y
Applying the element approach
J .* (x,y)v(y)df .* (x,y)v(y)df e
i.e.
f
1- = L
Elements J
f
1- (5)
(6)
.* T
c(x)u(x) + L
elements J 1- (x,y)cp
~
df
y
vn(y)
~e (7)
f
e
L v * (x,y)cp T df
elements Jf ~ y
e
179
HU GP (8)
where
U contains the nodal potentials vn
P contains the nodal current densities in
APPLICATIONS
C Magnet
An interesting example of the application of BE2DPP occurred
in the field of electromagnetism. The author would like to
thank the Rutherford Laboratory for allowing us to publish
details of the problem. Although the actual problem would in
real life be somewhat more complicated it serves to illustrate
the capabilities of the module.
Vx Vx A = J (a)
/J
where /J is the permeability
A is the magnetic vector potential
and J is the current density
In two dimensions A has only one component (the out-of-plane
component) so by making use of the vector identity V x (VxA)
V(V.A)-V'A we have that for the case of constant permeability
/J Eq: (4)-reduces to the scalar potential equation
While BEASY would not be able to solve Eq. (a), the two
dimensional form, Eq. (b), may be solved using BE2DPP for
constant values of current density J.
The geometry of the problem is shown in Figure la.
The p'roblem was divided into four regions. In region 1 the
permeability was unity and the current density 1257. This region
represents a copper conductor carrying current. Region 3 re-
presents a similar copper conductor carrying a current equal in
magnitude but opposite in sign. Region 2 represents the iron of
the magnet and has a permeability of 10000. Region 4 represents
the air surrounding the magnet and its boundaries are at infin-
ity. Figure lb shows a plot of the flux across the air gap.
180
v
.. , ,
I
I
= I
I
... ++ ... +++++J.
- :
. ., •
. .
(J)
I-
C'<") z:
IJ.J
=f j
lij
...J
IJ.J
- ...z:
l!l
N d
J:
(J)
.. i I-
-:
0
1
11 ...J
l>-
I
t;J
z:
l!l
~
I'
b
u
...
..J
~
181
/
=
/
~
·
/
·
/ ·
P-
a:
V
l!l
De:
a:
(f)
·
/
(f)
~
u
a:
x
·
/
::J
...J
LL
LL
D
l-
"
I 0
...J
P-
I
· t;j
z:
l!l
I:I
.· U
~
~I
182
Corrosion Protection
A major application area of boundary elements has been in the
field of the analysis of corrosion protection problems. The
method has proved ideal for the analysis of structures or
bodies in the sea where the boundary of the problem is the
surface of the structure in contact with the sea and the con-
ducting medium is the sea itself. To illustrate this point
further the reader is encouraged to think of the task of sub-
dividing the water between the one thousand members of a com-
plex offshore jacket type structure into finite elements.
-_'5t.ETJIO
( b)
2 4 6 8 10 12 14
~---------------------------1
I I
I i 10: 1
h
I :
I Enlargement
I .
I----------~----------------
I
200:1
Enlargement
on cathode
v = v + a In i (9)
o
where v initial voltage
o
Substituting into the matrix equation
HU = G P ( 10)
H
__U0 + H A G P ( 11)
where U
-0
vector of nodal v
0
[}. vector of nodal a In i
l' vector of nodal i
II-
------ ""
1':0 =1=
~ ~ r--f-
r--I-
-I-
<:;;: ts :$ ts
D-
Special Elements
For structures consisting of tubes where the diameter is small
compared with the overall dimension of the structure, special
elements have been developed. These tube elements assume that
the current density and voltage is constant about the circum-
ference of the tube but has constant, linear or quadratic vari-
able along the length of the tube. Fig. 6.
~ x L q x 211aL
e
The results are compared (Fig. 9) for both using the actual
line sources and the equivalent tube sources and good agreement
is observed.
d 1- t--I-()-
Figure 8
6 CONCLUSIONS
'10
Line Source
~
10
2 4 6 10
X=Y, Z=O
Figure 9
188
I
o
Line Source
Tube Source
12
2 4 6 8 10
X=Y, 2=0
Figure 9 (Conrd)
189
REFERENCES
D J Danson
SUMMARY
1 INTRODUCTION
2 THEORETICAL DEVELOPMENT
Governing Equations
The governing equation is the 3D wave equation
191
192
- 7;2 G + 6(r-r
_ _ 0 )6 (t-t0 ) o (2)
J J G'i/ u - G ~2 2 U - u'i/ 2 G + u ~2 G dr dt
t=-oo Q
(3)
u 6(r-r
_ _ 0 )o(t-t 0 ) - Gy(r,t)dr dt
t=-oo
J J
Q
J J G'i/u.n - u'i/G.n dr dt
t=-oo r
r
c, u(r ,t )
1. _0 0 I
t=-co
J
(l
Gy(r,t)dr dt (4)
where c,
1.
, for r_0 inside (l
R
and t =t
ret 0 c
y(:s,t ret )
c, u(r ,t )
1. _0 0 41fR
s
+
Jr 41fR [:~J t dr
ret
R.n R.n
f
J 41fR2
u(r,t
_ ret )dr_ +
J 41fcR u(:,tret)d: (6)
r r
r-r
and R .:::....£
R
194
Equation (6) is the basic equation for BEREPOT. Since the pot-
ential u(r ,t ) is expressed solely in quantities at previous
_0 0
time steps, this method is explicit and requires no matrix
solution. Note furthermore that neither time nor volume-
integration has to be performed. The only integration is over
the boundary r of the problem domain.
3 BOUNDARY CONDITIONS
.au au
ii) Radiation + 0 (8)
an c at
4 NUMERICAL IMPLEMENTATION
c,u(r ,t )
~ _0 0 ~ 4lTR
</>T dr
elements
re
~
R.n
+
elements
f 4lTR2
r
e
R.n
+ L fr 4lTcR
dr (9)
elements
e
v = 'i'u (10)
'i'u(r ,t )
_0 0
3(R.n)R-n r 3(R.n)R-n
+
I
r
41TR 3 u(:,tret)d: +
J
r
41TCR' u(:,tret)d:
I
(R.n)R R
+
Jr 41TC'R
u(r,t
_ ret )dr_ + 41TR' [~~L dr +
r ret
R
+ r ---
} 41TCR [~~t dr (11 )
r ret
r - r
where R ~
_s R
s
au tp v' o
p at + + p (12)
7 TEST EXAMPLE
x = 0 y o z = 0
x = 3 y 4 z =5
y(t) = 0 t < 0
y(t) J!.)'
-le
0 < t <
8 CONCLUSIONS
REFERENCES
2'1 ~o
.).'1 30 3,
31 3?
3l 3~ lit
~J 'l..S
.l.3 ~'t 1$
:LS :l.b
IS I .. 17
;2.'::. :l.1t
'" '"3 IS
, ~ 'I
I' :l.$
.:2.'1
:u 31
I~ ~ S & 1'+ l'l.
10 ,2.
"
1'3 lit
'f ,'I .w
IS I~
2.7
t 1
0.15
...."
.~
..
."
o
'"t- 0.1
....
...."o
<II
>
0.05
-0.05~1--------~------~--------~------~--------4--------r--------~------~-------+------___
o 10 15 20 25 30 35 40 45 50
Time t
ELASTICITY PROBLEMS
,D.J. Danson
1• INTRODUCTION
2. GOVERNING EQUATIONS
where
201
202
0 ..
~J
o i " j
0 .. i
~J
a .. [ujn.(x) (4)
~J J
E •• Hu .. + U •• ) (5)
~J ~.J J,~
J a . . (y)dfl
1,1 y
= J. n.1 (y)a.1 (y)drY (6)
fI r
For the theorem to be valid the vector ai(y) must be contin-
uously differentiable in fI and on r. Let us consider two
arbitrary continuously differentiable displacement fields
ai(y) and ai(y) and their corresponding stress fields aij[a]
and a . . [a].
1J
Applying the divergence theorem to the product ai(y) a .. [a]
gives 1J
J {a. 1.
(y)a .. . [a] + a. . (y)a . • [a]}dfl
1.J,] 1.,J 1.J Y J a.(y)a .. [a]n.(y)dr Y
1. 1.J J
fI r (7)
= Ha 1.,J
.. (y) + . . (y)}a 1.J
aJ,1. . • [a] (8)
a1.,]
.. (y)a 1.J
•• [a] E• •
1J
[a]a • . [a"]
1.J
(9)
J{a.(y)a
1
. . . [a)
1J , J
+ E
iJ
.[a)a .. [S]}dfl
1J Y
= Ja.(y)a
1
.. [a)n.(y)dr
1J J Y
fI r (11)
J E 1J
•• [llja .. [ajdQ
1J Y= J E ••
1J
[aja .. [ajdQ
1J Y
Q Q
r
oj {a.(y)a .• • [aj - a.(y)a . . . [a]}dQ
1 1J,] 1 1.J,J Y
Q
J {a.(y)a .. [aj
1 1J
- a.(y)a .. [a]}n.(y)dr
1. 1J J Y
(13)
r
(14)
J6(xm-ym)dQy
a11
space
f
a11
f(y)6(x -y )dQ
m m y
f(x) ( 15)
space
205
J Uk·(x,y)t.[u]dry
r
~ ~
- fr u. (y)Sk"
~ ~J
(x,y)n. (y)dr
J Y
Now
206
Ju,(y)ok,6(x
l. l.
-y)dQ
m m y J uk(y)6(xm-ym)dQy (18)
And putting
Sk,,(x,y)n,(y) = Tk,(x,y) (19)
l.J J l.
we have
J Uk,(x,y)t,[u]dr
l. l.
- Tk,(x,y)u, (y)dr
Y J l. l. Y
r r
Tk , (x,y)u, (y)dr
1. 1. Y
r (22)
= J Uk,(x,y)t,[u]dr Y + J Uk,(x,y)b,(y)dQY
1. 1. 1. 1.
r Q
(23)
T n
Uk,(x,y)$ dr y_1.
t,(y) Uk' (x,y)b, (y)dQ
1. _ 1. 1. Y
HU=GP+B (25)
numerically.
This last domain integral may be evaluated numerically by
dividing the body into integration cells. In anyone equation
either the traction or the displacement will be known (assuming
the problem has been correctly formulated). Hence each
equation in (25) contains one and only one unknown and thus
(25) may be rearranged as a system of simultaneous linear
equations,
A x = F (26)
(28)
(29)
(30)
E' _ v'
E = 1-v'2 and (32)
- 1-v'
(33)
Fundamental Solution
The stress-displacement relations are (from Equation (2»
(35)
+00 00
I' (37)
I' .
,J
. x3 Jco _2y.
_J
G r'
y.dy. y.dy.
I' = - 2 J~ - 2 J __J ___J = - In r' + A (38)
r' YkYk
l+v
41TECl-v) {(3-4v)1n r <Ski + rkr i }
1
Uki(x,y) (41)
Tki (x,y)
(42)
Boundary Points
We now have all the kernel functions in Equation (22) but we
still need to calculate Cki. We start with Equation (20) which
we know is valid when x is an internal point,
n,r. = 1 (46)
J J
rkn i - ri~ = '0 (47)
and r = £
214
--L
dX 3 -:r
':2
so that
(49)
Moreover,
cos 8 and r 2 sin 8 (50)
Hence
82+1T /2
x
41T (l-v)
(1-2v)
f8 - 1T /2d8+2 I8 -1T/2cos 2 8d8 2
f8 - 1T /2sin8cos8d8
1 __1 __ 1
-8;+;;:/2
- .... 82+1T72- - 82+1T 72- - - -
I
2
I sin8cos8d8
8 1- 1T /2
(1-2v)
J
8 1- 1T /2
d8+2
J
sin 2 8d8
8 1- 1T /2
(52)
And evaluating the integrals gives
81T ( l-v)
+sin28 2 - sin28 1
(53)
92+11 /2
l '
l.m
£-+-o
{ l+v
411EU-v) f [(3-4v)ln(f) tiki
9 _11/2
1
(54)
+ I
Q
Uk'l. (x,y)b,l. (y)dQy (55)
Internal Points
Once the boundary values are known we can calculate the dis-
placements at any internal point from Equation (20). Differ-
entiation of Equation (20) and substitution into the stress
displacement relations (34) enables us to calculate stresses
at internal points. Thus
Dk,·(x,y)bk(y)dQ (56)
l.J Y
where
217
and
A6 .. Tmk + p(T· k . + T· k .) (58)
~J ,m OIl oJ J, ~
r. (59)
~
y. - x.
~ ~
r. (60)
~ .r
Differentiating
6 ij 1
r .. - - - - r . (y. - x.) (61)
~,J r r2, J ~ ~
r.r. - 6 .•
~ J ~J
(62)
r ..
~oJ r
(63)
U: k .
~ ,J
218
(l+v)Cl-2v)rk
Umk,m = 2wE(1-v)r (66)
Dk · • (67)
1.J
(70)
(70
Tk . (x,y)u. (y)dr
L L Y f Uk·(x,y)t.(y)dr Y
L L
r
+fUk.(X,Y)b.(Y)dn (72)
L L Y
n
where Cki = 0ki when x is an internal point and Cki = lOki
when x LS on a smooth boundary.
(J ••
1.J
(x) f Dkij(x,y)~[u]dry - J Skij(x,y)~(y)dry
r r
+ f Dkij(x,y)bk(y)dOy (73)
o
where
Sk1.·J· = ~O 1.J,m
.. Tmk + ~(T·k • + T· k .)
1. oJ J ,1.
(75)
Dkij(x,y)
(76)
and
Sk .. (x,y)
1.J
- (l-4v)ao .. ] (77)
K 1.J
221
When the Galerkin vector has only one non-zero component then
that non-zero component is known as Love's ~rain function.
Fundamental Solution
For axisymmetric elasticity analysis the point load x is
generalized to a ring load as shown in Figure 3. Note that the
z total load acting
round the ring is
unity (Le. the
loading is not unity/
unit length). There
are, of course, two
cases: one with the
ring load acting in
the radial (r)
direction and the
other with it acting
in the axial (z)
direction.
r
It turns out
that the Galerkin
vector for the rad-
ial load has no
axial component and
the Galerkin vector
for the axial load
Fig. 3 Axisymmetric Ring Load has no radial com-
ponent. Thus call-
ing these Galerkin vectors Gr and GZ respectively we have that
G e
r ~r
and
223
where Fr and F Z are the body force terms describing the unit
ring loads in~the radial and axial directions respectively.
F e (86)
Z ~Z
where V2 _ 32 +.! ~ + ~2
- 3r2 r dr dZ
6(R-r,Z-z)
Thus F F (88)
r Z 21TR
Note that Rand Z are the coordinates of the load point x, and
rand Z are the .coordinates of the field point y.
}
F 0
r
R, Z # r, z (89)
F 0
z
J
all
F
r
drl
J
all
F
z
drl
space space
(Z-z) 2 + (R-r) 2
where y 1 +
2Rr
U 1 {(1-2V)l(V2G _ Gr ) + (l2G r }
rr 2( I-v) r r2 dZ 2
(91)
(l2G r
I dG r
I
U - 2Tf=") -- + ---
rz drdz r dZ
1
a2 Gz
U
zr - ZTT=VT drdz
uzz = ~
1 { (1-2v)V 2 G + d 2 GZ + 1
- aG z }
hI-v) z ar 2 r dr
(l-y 2) d 2 w _ 2y dw + {V(V+l)
dy2 dy
-~}
l-y2
w = 0 (92)
(93)
dQv(Y)
(y2-1) - - - '"
v(v+1)v' y 2-1 Q-1 (94)
dy . v
And hence
(95)
dQ !(y)
- 4v' y 2-1 ----7"--
dy
U
rz
_ _-'Z=--..::.z'---_ { Q;! _ [Y _ *) d~+y! }
1611 2 (1-v)~rv'Rr
(96)
Q_~(Y) = kK(k)
kE(k)
- 2(y-1)
{K(k) _ yE(k)
y-l
}
,fz( l+y)
k = 11:y (98)
Traction Kernels
Traction kernels are computed in the usual way. The displace-
ment kernels are substituted into the stress-displacement
relations and the inner product of the result with the unit
normal at the field point is taken. Expressing the traction
kernels in terms of the displacement kernels and their deri-
vatives gives
au J
T
rz
2 [{ I-v
Il 1-2v az +
rz v
1-2v
r,..E
u au 1~
r + ---E.!".J
ar
1 (au rr
z + zlaz + -ar-J
1
au rz
nr
(99)
1
ilLr{
2 I-v au zr
ar
v zr zz[U
au)} (au
az
nr + .!l~
au 1
2 az + ~J
T 1-2v + 1-2v -r- + n
zr ar zJ
227
T
zz
2J.l[{~
1-2\1
au zz + _.~ (U zr + aUzrJ)}n + ..!.(aU zr + au zz ) n
az 1-2\1 r ar z 2 az ar r
J
Boundary Formulation
Equation (78) represents the displacement at an interior point
in terms of the displacements and tractions on the boundary.
Since these are not known initially we must take the point x
to the boundary in the usual way in order to set up a system
of simultaneous equations which when solved will give the
complete boundary solution. First, however, we convert the
surface integrals in Equation (7~ to contour integrals by
integrating with respect to e analytically. Since the axi-
symmetric nature of the problem ensures that the integrands
are constant with respect to e this is trivial and yields
2~r Uk.(x,y)b.(y)dA
~ ~ Y
s A
( 101)
£
Lim
+ 0
e2
f 2~r£ Tki de (102)
e1
Internal Points
After obtaining the boundary solution in the usual way
Equation (78) may be used to obtain the displacements at inter-
nal points. Differentiation of Equation (7S) and substitution
into the stress displacement relations enables stresses at
internal points to be calculated. This generates a quite
unbelievable quantity of algebra but is perfectly possible.
REFERENCES
APPENDIX
(J
rr (1+v)~7-2v) {(l-V) aR
au +
v
(u
lR: aw)}
+ az
E
(J
zz
m aw
(1+v) U-2v) { (l-v) az
+
v{i + ~~)}
E
m raw + au)
(Jre = 2Ti"+VT aR az
Q = kK
-~
dQ_~ kE
--cry = - 2TY=1T
23y2 + 9
8(y-D
[K -~)
y-l
230
U l+v {(3-4V)Q
(Z-Z)2
+--- d Q+ 1}
rr 87T 2 E (l-v)1Rr +! Rr dy
m
+ (Z-Z)2 [d 2Q +! ~ _ 2. d Q+!'}
Rr dy2 ar 2r dy J
d 2Q dQ
+ -Z-z { (Z-z) -- +! -ay - 2 - +!J]
2 -
Rr dy az dy
( l+v)(Z-z)
U
rz
_ [ER _Y)dQ+dy ~ - Q+ ~ ]
-2-_
U
zr
(l+v)(Z-z) {[E _ )dQ_~ _ Q_! }
81T'Em(1-v)r,!Rr R y dy 2
au
---ar;
zr (l+v)(Z-z) {[r 2
- - y ) () y d Q_12
81T'E (l-v)r1Rr
m
R ~ ar
au
a~z ; l+v
81T'E (l-v)1Rr
{(3-4V) (ir.
ar
dQ_! _
dy
Q_~)
zr
m
232
au
~rRr =
o
1+v
{" dQ
. (3-4v) ray ~ _
87T 2E (l-v)1Rr" . aR dy
Q
2R
-.:!:i)
m
d2 Q dQ
+ (Z-Z)2 ray ~ _ 2- ~)}
Rr aR dy2 2R dy
d 2Q dQ
+! + 2 - +l}
+ -Z-z { (Z-z) -ay - - - ]
Rr az dy2 dy
-y)--+1
au rz l+v d2Q
[ r ay
az-= 87T 2E (l-v)rlRr"
(Z-z){[-
R az dy'
m
233
au d2 Q
zr (1+v)(Z-z) {[~ _ )~---=l
--aR =
8TI2E (1-v)rfu
R Y aR d
y
2
+ ..!.[l.
2 R
-3 lr _ 3rJ'" dQ_!
aR R2 dy
+ Q_! }
4R
+ [~-
R
yJd Q-! - Q_! ]
dy 2
+~-
(Z-Z)2
Rr
1+v
~0-4v) ay dQ_!
3Z--;r-y
STI 2 E (1-v)fu
m
- ~
Z
Rr
{
(z-z) -
ay d 2 Q_! dQ_~
-- + 2 --
az dy2 dy_
1
}-I
+
[aaRar
2y 3b 1 oy 1 oy~ J
- 2" r aR + R ar
d2 Q
~ +! 9
dQ
+ 4Rr dY }]
+~
234
+ {CZ-Z) £.:L
3Zaz
+ ~ _~} {[E. _ Jd2Q+~ _.!.
az az R y dy2 2
dQ+!}l
dy J
au rr
[az au rz )
3R + ----ar-
r d2 Q
1+v I (3-4v) {~~ ~ + l(a 2y 1 3Y) dQ+! }
8rr 2 E (l-v)1Rr L 3R 3z dy2 3RdZ - 2R 3z dy
m
d 2 Q+~ )
dY2
235
3 oy oy 3r oy 5 oy)
- "2 oR ar - 2R2ar - 2R oR
_ 1 02 Y
"2
15 3y )
oRor + 4R2 - 4Rr
+ 3Q+! }
BRr
J
+ [* - y)
oy oy d l Q+!
az ar ~ + "2 2r az -
1 [3 ay 02y ) dQ+! }
nar dy
236
y [2~ l~)
r ) o2y 5 ay oy
+ {[R - y aRar - "2 aR +"2 r aR ar +
R ar ar
3r ay
- 2R2
lOY} d2 Q
-! {9 ay 3 02 Y
3 oy 3y
- 2R aR ~ + 4r oR - "2 oRar + 4R ar - 4Rr
d'Q
ay oy -!
az ar CiY"'
3
+ -2 [2-lr _~J
2r az azar
d Q_!} +
dy
[E.R _ y) oy
ar
2
d Q_!
dy2
dQ_! + 3Q_! ]
+"21 [3r - 3 ~
ar - ..!.J
Y
R dy 4r
02U
{ay ay d 2 Q_! 1 ay ) d Q_!}
zz
aRaz
1 +v
81T2E (l-v)1Rr
1LO-4V) aR 3z ~y2[a 2 y
+ aR3z - 2R dy az
m
+ z-z{z-z
r R
[r2- ~ _
[zR oz
a2 y
aRaiJ
l -----=l
d2
dy2
Q
_~aR ir -----=lJ
d'
az dy'
Q
+ 1 [2 ~ 2
d Q_! _2 d Q-!1} ]
R aR dy2 R dy)
237
a2 u d2Q dQ
.zz 1+\1 -i [a 2 y
aRaz = 87T2E (1-\1)&
[
(3-4\1) { aR az
a y ay
CfY2 + aRaz - 2R dy az
1 aY)----=-i}
a2 uzz
azaz
+ ~ dQ_! ]
Rr dy
~ [aU zr au)
zz
ar-
<:J
aR a;- +
dQ : d'
-! Jaya y la 2 y
3)
+ 2R2 dY + rZ-z {(r
li - y aR az --cryr
Q_! [Er
+ . i - YJ ClRClz
+ z-z
R
[Ila {l r~ aR
ay + 1 ClY} _ a 2 y
i. Clr
ldQ_! _ ay ay d 3 Q_i
ClRClil dy ClR Clr dy'
__9 dQ-']ll
4Rr dy LI
238
3U I
+--
3r
ZZJ
1-2v 3y dQ_!
- -r - -3Z- -
dy+
r
Z-Z {aazy a;
+ -r-
3y [rr l ~Q-!
lJ - YJ~ 5
~Q
-!)
- 2"""'dy2 + dZ3z
32 y [Ir l ~Q-!
lJ: - YJ""'dy2
_ ~
2
d Q_!) + ~
dy R
(~
r
dQ_! _
dy
2.h
3r
d 2Q _!)+
dy2
Z-Z
R
(r3l.3r
3y
az
Chapter 14
D.J. Danson
1. INTRODUCTION
J Uki(x,y)ti(y)df y - J Tki(x,y)u(y)df y
+ Jr Uk' (x,y)b.(y)dn (1)
1. 1. Y
n
For non-zero body forces (b.(y) F 0) Equation (1) contains a
domain integral which must Be evaluated to complete the bound-
ary integral formulation. This appears to be something of a
nuisance if we have to divide the domain into integration cells
in order to evaluate this domain integral since the data pre-
paration necessary to run the problem will be greatly increased.
Fortunately Cruse [1,2] and Danson [3] have indicated how for
some commonly encountered body forces this domain integral may
be replaced by boundary integrals. We shall consider three
commonly encountered body forces, gravitational (i.e. self
weight), rotational inertia and steady state thermal loading.
In all cases we are concerned with reducing the last term of
Equation (1), Bk(x) to boundary integral terms where
r
I
J
Uk' (x,y)b. (y)dQ
1. 1. Y
(2)
n
The Galerkin Tensor
In order to transform the domain integral (2) to an equivalent
boundary integral it is useful to define a tensor ~i(x,y) (the
Galerkin tensor) such that the displacement kernel -(or funda-
mental solution) is given by
239
240
Gk , ,,(x,y)
Uk' (x,y)
~
= Gk , ,,(x,y) -
~ oJ J
~(r')
-v
(3)
l(Gk' ,J,,(x,y)
Gk , "CX,y))
C )
J ~ J
J,J~
2(1-v) b i y dQ y (4)
(5)
dr y (6)
Rotational Inertia
Let us consider a body of uniform mass density p rotating about
some axis with angular velocity w,. We may without loss of
generali ty consider the axis of r6tation as passing through the
241
(8)
b. (10)
~
where g .. ( 11)
~J
d(l (12)
Y
Similarly
i-
Ym
{y. Gk . (x,y)}
J m,l
= y. Gk
J
. (x,y) + Gk . . (x,y)
m,ml J,l
(14)
g ..
~J
J -a~ {YoJ
y
Gk · (x,y)} - Gk · .(x,y)
~ ,m ~ oJ
n m (15)
+_I_[G ( ) a 1
.(x,y)}J dn
2(I-v) L:kj,i X,y - ay
m
{Yo Gk
J m,~ Y
Thus Bk(x) = g ..
~J
J y.J Gk · ,m (x,y)nm -
~
Gk · (x,y)n.
~ J
r (16)
( 17)
dr
y
= J c,1J,[Cll{cr,1J,un
kEe[B10"}
+ 1J dn (20)
1-2\1 "y
n
we find that the boundary integral expression is for the thermo-
elastic case
~(x) Jr Uk,(x,y)t,[uldr
1 1 Y
- JTk,(x,y)u,(y)drY
1 1
r r
+ -kE
1
2
- \I
J Uk" 1,1
(x,y)e(y)dnY (21)
n
Differentiating Equation (3) with respect to Yi gives
Gk , .. , (x,y)
(
U ' ,x,y ) () J,J 11
k 1,L = Gk'L,1JJ
'" X,Y - 2(1-\1) (22)
(23)
Bk(x) = -kE
2
1
- \I
J Uk" 1,1
(x,y)9(y)dflY
n
(24)
244
kE
Bk(X) ; 2(1-v) J e(y)Gk , ' "
l.,l.JJ
(x,y) - Gk , ,(x,y) e" (y)dil
1,1. ,JJ Y
(25)
il
3, 2D BODY FORCES
(28)
( _ l+v
Bk x) - 4nE (29)
r
245
~(x) J Uk·(x,y)t.[u]dr
L L Y
- JTk' (x,y)u. (y)dr Y
L L
r r
+ J Pk(x,y)dr y (30)
r
S~.
LJ
1
= -8
11
rL2nmr m(b.Lr.J + b. r.) +1 11 f \)0 ..(2n r b r
J L '1'=V1. LJ m m s s
(37)
s~.
l.J
1
= -8 f2n r (y g . r. + y g . r.)
11 L m m s Sl. J S SJ 1.
,-M
+ 1
( va.l.J·l2nmr mr s g sqy q + flL - 2 In ill ry g n -n g r rl J1
-v r ~mms s mms s:J
1-2V[
+ -2- 1 - 2 In -
r
l)l(ymgml..n.+y g .n.-[n g .r.+n g .r.Jr)
J m mJ 1. m ml. J m mJ 1.
- r g y (n.r.+n.r.)ll (38)
m ms s 1. J J 1. fJ
Pk(x,y) k ( 1+v ) ( 1 1) }
41T( I-v) In -r - -2 nk - nmrmrk
(41)
r r(ln..!.-..!.)
k r 2
+ Jr s*1.J
.. (x,y) 6(y)dr y - f V*1J•• (x,y)6 ,m (y)nm(y)dry
r r
kE
- -1
2 6(x)o .. (43)
- v 1.J
where
* (x,y) kE
Sij {nmm
r [Ol iJ2' - 2r.r.) + n.r. + n.r.}
-v 1.J 1.J J1.
(44)
and
4. 3D BODY FORCES
(l+v)R 0ki
Gki(x,y) = 4'1lE (45)
(46)
(47)
(48)
1
+ -1- {vo .. (n r b r - b n ) (49)
-v ~J m m ssm m
~
- -21 (b r [n.r. + n.r.] + [1-2v][b.n. + b.n']}1
mm~J J~ ~J J~.J
where
(51)
S~.(x,y)
1.J
1 [{n
= -8
'IT S
r Ym + 2 1(-1 2V ) n } (g. r. + g. r.)
s R -v m lorn J Jm 1.
- Y2 m
R (r g [n.r. + n.r.] + [1-2v][n.g: + n.g. ])}Jl
s sm 1. J J 1. 1.. Jm J lorn
(52)
and (53)
*
V •• (x,y)
1J
250
r r z
B(x) =J u ' (x,y).b(y)dQ (55)
- Y
and U U
rr rz
(57)
U
zr
uzz
V(V.Gr,z)
Ur,z V2 Gr,z (58)
Hi-v)
B (59)
b = [0 , b 1 = constant (60)
to get
B r~
J 2(1-v)
V(V.Gr,z).b - V x (V x Gr,z).b dQ
~ ~
(62)
Q
251
(63 )
and V.{(V x Gr,z) x b} V x (V x Gr,z).b - (V x Gr,z).(V x b)
B = J~
2(1-v)
V.{(V.Gr,z)b}
__
- V.{(V x Gr,z) x b} dQ (64)
Q
B - r
~ (V.Gr,z) (b.n) - {(V x Gr,z) x bLn dr
- J 2 (1-v)
(65)
r
Noting that Gr G 0
r
GZ 0 G (66)
z
b 0 b ]
B = b 1-2v aG z aG dr (67)
z
2(1-v) az nz + -- n
ar r
r
Rotational Inertia
b = P w2 f (68)
r V(V.G r . z )
B pw 2
J
(V2G r ,z).f -
2(1-v)
.f dQ (70)
n
252
B pw 2
_ - 2T'f=V) J (71)
Q
Now
r
- f.{n x (V x Gr,z)}dr
(72)
Proof
The last term on the R.H.S. is a triple scalar product. Thus
we may rewrite the R.H.S. as
(V2G r ,z).f dQ
R.H.S.
J
Q
Q.E.D.
{V x (V x f) }.f dO x (V x Gr,z)}dr
Similarly
J
Q
Jr f.{n (73)
Proof
The last term on the R.H.S. is a triple scalar product. Thus
we may rewrite the R.H.S. as
B pw 2 J 21-2v
( 1-v )
{
(~ • ~)(\1 • ~
r,z r z
) - (G ' •n)(\1 • f) }
r
Noting that Gr [G
r o1
GZ = 0 , Gzl
f [ r o1 (7
n [n
r nz 1
\1.f 2
B pw 2
j 1-2v
2 (1-v)
aG
[r aGz _ 2G n ] _ r - -zn
Clz z z Clr z
dr (75)
B =
~
~ Je
4-2v.
\1. ur ,z dO
~
(76)
Q
Replacing Ur,z with the right hand side of expression (58) this
becomes "
254
(77)
(80)
~
kE
= 2(1-v) f 8V(V.~r,z ).~ - r Z
(V.G ' )(V8.n)dr (81)
r
Writing this equation as
~ =f : 8 dr - f Q V8.n dr (82)
r r
we have that
[aarazGz r a2 G
z]z
2
kE
Pz = 2(1-v) --n + - 2- n
az
(83)
kE [ aGr + Gr )
Qr = zn=vy ar r
kE aG z
Qz = zn=vy az
In order to calculate a corresponding boundary integral term
for calculating the stress equation one simply differentiates
255
Equations (67), (75) and (82) with respect to Rand Z and sub-
stitute in the Lam~ equations. The necessary expressions for
Gr , Gz and their derivatives are listed in the appendix.
REFERENCES
APPENDIX
aG r (l+v)iRr Q+ i
az- = 47T 2 E
G
z
aG z (l+v)~ Q-l ay
az- = 47T 2 E az
257
l(Q+!2
aG
r (l+v)1Rr ay + y2-1 d Q+!)
aR 21[2E aR 3R dy
aG r (l+v)1Rr Q+! ay
az 41[2E az
a2_Gr (l+v)fu nay ay + y2- 1) d Q+!
aRar = 41[2 E UaR ar 3Rr dy
+ n~~r + i {~ ~~ + ~ ~~)} Q+ i ]
1 [rhl
+ 2R Carl
2
-
y2-1]}
Jr'2
dQ
CiY
+1
+
{a 3y
oRClr 2 +
1 Cl2y
raRor
258
~G
r (l+v)..'Rr
aRaraz = 41[2 E
[b:.?r. b:. ~
aR ar az
~Q
dy2
+ {ll ay II ay
aRar az
+
aRaz ar
ay [a 2 y 1 ay ) 1 ay a y } dQ+!
+ aR araz + 2r az
+ 2R ar ~ rz
aG z (l+v)..'Rr Q_! dY
az = 41[2E az
a2 G
__
z =
azar
259
32y 3y 3 2y 3y
{
dZ3r az + aZ3zar
+ ay [a' y 3Y)'} d _! Q
az ar3z + zr1 az CfY
Chapter 15
C. A. Brebbia
University of Southampton
INTRODUCTION
261
262
problem. The main advantage of the new procedure is that the
integrals need to be computed only once as they are frequency
independent. Hence the approach is extremely economic for
free vibrations when compared to the ones previously presented.
The technique also allows for the general elastodynamics case
to be solved in time rather than in the transform domain
employing the simple fundamental solution of elastostatics.
The results obtained are very accurate as can be seen in some
of the examples presented in this chapter.
u =u on rl
q =q on r2 (2)
2
a'i/ u
t fI
o
(3)
2
Integrating twice by parts the terms in 'i/ and once with
respect to time, the au/at term, applying the causality
principle and taking into consideration the character of the
fundamental solution, one obtains the following inverse
statement:-
au*
at )udfldt +
(4)
(5)
tF
ci ui +
Jf rUq dr dt
•
t0 r
(6)
200., - - Analytical
• F.D.H.
a Linear F.E.H.
v Parabolic F.E.H.
i>. cubic F.E.H"
100
o~
::s
100
- Analytical
• Constant B.E.H •.
o Linear B.E.H.
~.C)'
~
.. _.~~c_.1i i ~
.~ hu-kc;FC' . ~
~
Figure 1 Surface Temperature of'Plane Plate
266
discretisation is simply the decomposition of the total
integral with respect to time into a summation of boundary
integrals.
Example 1
This example studies a plane plate, initially at OOc,
surrounded by a medium at 1000c. Its cross section is
0.1 x O.lm and the values of the thermal conductivity, heat
capacity and heat transfer coefficients are lBkcal/(hmOc),
912 kcal/(m30c~ and 5000 kcal/(hm20 c), respectively.
Example 2
For problems with regions extending to infinity, BEM solutions
are· much more economical than finite element ones. In order
to demonstrate this, the case of a circular opening in an
infinite plane region, initially at zero temperature is
267
o
........... _*_ ...\1.
I
1
BE Mesh
10.
FE Meshes'
u
h=0.2
u
h=l.O
.
__ Analyt ical
B.E.M •
h=S.O
-10
'0 t 10
Example 3
This example is more practical, the temperature variation of
an actual turbine disc is studied here (Figure 3). The initial
temperature of the turbine disc is 295.l oK and the values of
the thermal conductivity, density and specific heat of the
material are l5W/ (mOK), 8221 kg/m3 and 550 J I (kgOK)·,
respectively. There are 18 different zones along the boundary,
each of which with a different set of prescribed values for the
heat transfer coefficient and the temperature of the surround-
ing gas. The time variation at one of such boundary zones is
shown in Figure 4.
in fl (7)
- -q)u"" dr dt -
(8)
- -
1 a2 u ",,]
--- udfldt +
c2 at 2
t fl
o
271
2500' 1000
I II
II II
0';< II I II
II ~ I IL
N
i.c:
I II O~ I II
I II ::- I II
I II II II
I II II II
I II II II
.1 II II II
II II II II
II II II II
0' 0
0 100 900 1000 1100 '2500 0 100 900 1000 1100 2500
t(sec) t(sec)
~
E H(t-O).
Y'r q=O p
~L IU~
E
bI
q-O x
/1 t
a
PH (t-O) Eq
r~'-~'
tF ~
i iU'.dr dt - i i~'dr dt
t r t r
0 0
1
-2
c
j~u*u -~ U*] dO
at at
0
. t
Q 0 (9)
- uq * )dr dt (11)
t
o
r
Analytical\
BEMfor--S = 0.6\
Eu
Pa
a 4a
ct
-- Analytical,
Y1 .' ........... BEM for e= 0.6
I
a 2a:
[\.
3a 4a Sa 6a 7a
1\
Sa
i
ct,
but has the advantage, from the users point of view that it
does not require any internal points. The examples presented
here are based on a linear, spatial interpolation for the
potentials and constant for the fluxes, to better represent
certain types of shock waves.
Example 4
Results obtained using the two-di~nsional boundary element
program were compared against analytical results for a one-
dimensional rod under a Heaviside type forcing function. The
boundary element solution .considered a rectangular domain with
sides of length a and b (b = a/2) as indicated in Figure Sa.
The u-function represents the displacements which are assumed
to be zero at x = a and their normal derivative q is taken as
null at y = 0 and y = b for any time. At x = 0 and t = 0 a
load Ep is suddenly applied and kept constant until the end of
the analysis -E is the Young's modulus of the material. Due
to the topology and boundary conditions the problem is actually
one-dimensional and its analytical solution can be easily found.
Example 5
This example analyses a square membrane with an initial
velocity Vo = c prescribed allover the boundary.
l', --."
, j
t .... t
a' u -= 0" k.'_ '-....
~
5
u = 0:
a
o.os
t , ----- Analytical
y'
for a = 0.6
,000001) BEM
0.04
\ aIG~
x t---I
,a'
-0.04
-o.osL,-L.--J..-,-J----1---1---1--:-l7::-7.L--L-;2~2::'~
0.4a a ct 1.6a__________________________
"~
• ,a '
O.SO; yl - - Analytical
e=
o
.
18
a
40, 1
IIA B,
'~
a--' x
0.2
TRANSIENT ELASTODYNAMICS
The governing equations for transient elastodynamics for a
homogeneous, isotropic linear elastic body can be expressed
in terms of accelerations and velocities by,
in n (12)
c d =~3'" (13)
---'"
p
-~)u
2
s J,~~
2 . . . +f.
. . . +cu
s ~,JJ J - U'lu~'
~ ~J
dO dt
t 0
o
277
~
*
- tr. )p .. df! dt
~J
(15)
2 2 *
(cd - cs)uik,kj + csuij ,kk - Uij
2 * *
= -7 1I(~)II(~)
&••
(16)
c .. u. +
~J J
jtFjp~.U.drdt j~ju~.p.drdt
~J J
= ~J J
t
o
r t
0
r
(17)
,~
10. TIME t m .ee)
A I XI
I
b
~
•
I
~ Il.
I
t
L..
.
. 180 ~ TRANSMITTING
BOUNDARIES
It
Figure 7(b) Finite difference mesh for half-plane
279
-G
65'
'~l 30'
• ITRIIS OR DIPl.ACrUENT
~PAl'IIEO WITH TUN.
.!! -6
h
TD 1-,5
-TSENG
••• lEW
O~O~~~40~----~607---~eo~----looL-----I20L-·
tI msecJ
Example 6
The first elastodynamics example is the solution of a half-plane
under a continuous prescribed stress distribution (Mansur and
Brebbia, 1984). In this application the technique discussed
above is compared with the finite-difference model implemented
by Tsenget al (1975), who developed a transmitting boundary
condition together with a generalised lumped parameter model.
- TSENG
:E -3
....Z
~ -2
~
..J
CL
on
2i
-I
P WAVE ARRfYES J
OL-__-L____~~~~--~----~--~----~--
o 40 60 eo 100 120 140 160
t (m ••eJ
] -1.0
,·.s
!II: ~""'l..!.'!'.!.~.~.!'..l'!.!'~'..............,......... • ••••••••••
Ii -0.5 YSENG
. . . . BEM ome SOWTION
Example 7
This application consists of a semi-infinite beam simply
supported along its edge (see Figure 8a) and subjected to a
suddenly applied bending movement,
Mo = M H(t - 0) (a)
The Poisson ratio for this plane stress problem was taken
to be 1/3. The boundary element mesh consisted of 36 equal
elements as shown in Figure 8b and B = Cd ~t/~x was taken to
be 0.5.
5r
t
c (b)
o
·'~~II- i
----x-.
+
Figure 8(a) Geometry and Boundary conditions of the semi-infinite
beam
A
·~Ii
~
+.5
0
~
-u,]u*"
~ ~J dO
r2 rl (18)
* k'
I[co! - c2s )u'k ~,J
2
+ csuij,kk
* -u~} ~J ~
dll
. jP~.u.
~J J
dr ju~.P .
~J J
dr
r r (19)
286
(cd2 - c s2 )u*
.. k. + c s2 u*
~J, J
.. kk = 6~J
~J,
.. 6(0 (20)
c .. u. +
~J J
Ip;;u; dr lu~.p.
~J J
dr - Iu~.u.
~J J
dO (21)
r r 0
u. a~ fk
~ ~
and
u. /j~ fk (22)
~ ~
i u ;;.; dO ki k·
= /ji f u ij dO
0 0 (23)
One can now consider that the functions fk are a source .dist-
ribution and consequently each can be associated. with a pseudo
displacement field Wand ies corresponding stresses Tand
surface tractions nO:; The fk functions can be taken to' be very
simple functions such as the distance between nodes and hence
can be easily integrated to yield the new ps.eudo-state. This
i
means that the mass term can be transformed as. follows,
i u~.u. dO
~J J
+ u*~J
.. n.k~ ..n dr - ip;A, df]
o r r (24)
therefore,
(26)
HU - GP -Mil (27)
where
M = (I.!P - ~!}) [~J
HU + MU = GP (29)
Example 8
Figure 9 shows an application of the above technique to find
the eigenvalues of a shear wall with four openings. The
boundary element model consists of 29 quadratic elements with
58 nodes, as shown in the figure. The finite element mesh
comprising of 559 nodes and used for comparison purposes is
shown in the same figure. The results for the free vibration
periods of the first eight natural nodes are given in Table 1.
In spite of the complicated geometry and the rather small
288
o
o
o
r3.0-t- 3.0-+--4.8--1
BE model FE model
58 nodes 559 nodes
pit) A
[f1]
0
N
E/p = 10'
v =0.2
0
N B.E. Model
"'k n
2.0 2.0 2.0
I' -I' 'I' -I
F.E. Model
10.0
_ t
0.1
!MODE 1 2 3 4 5 6 7 8
i
Example 9
The application of transient analysis is confined here to the
case of fixed vibrations caused by external, time dependent
tractions, with no support movements. Equation (29) can then
be solved using a direct integration scheme. In this case,
the Houbolt method was chosen because of its unconditional
stability; a number of other stable methods could have been
used as well.
CONCLUSIONS
REFERENCES
uH(A)
~- present method IHoubolt. Jl.t = 0.005)
-- - - finite element method (Wilson 6. At ;;;; 0.005)
~
'--~
... 'l'/.----"~
-7
-7 ;/ ~
~ ~ ~
~ ~ ~
t7
t7 ~
7 17 "
7
H.L.G. Pina
Department of Mechanical Engineering, Instituto Superior
Tecnico, Lisbon, Portugal.
I." INTRODUCTION
Rizzo and Shippy, 1970, appears to be the first paper where the
time dependent heat equation is solved by numerical methods
based on integral equations. The differential equations is
transformed via Laplace transformation for the time variable
and the resulting integral equation is solved numerically. The
subsequent inversion"of the Laplace transform is also done
numerically.
(1)
293
294
(2)
(3a)
(3b)
(3c)
in Ox T. (4)
I d u*(~,t;~',t') da(x) =
-
I d
u*(x,t;x' ,t') da(x')
- - -
= 1 (7b)
R R
I: Ia ;t' (uu*)da(~')dt' .. Ia
o
I: ;t' (uu*)dt'da(~')=
0
(llb)
I: o
In ;t' (uu*)dn(~')dt' = c(~)u(~,t) - Inuou~ dn(x'). (12)
In r
au
an
I au*
(u*v 2 u-uV 2 u*)dn(x') = (u* - - u - )dr(x').
an
(13)
cu= InUou~dn+ Ir I: o
(u*q-uq*) dt'dr. (14)
(16)
But, by definition,
(17)
(18)
u u* dg =J (u 21_ ~
JgOO aUo)df. (19)
f 0 an an
1 -
-
r dr
d~
d (r-)= 1
dr 4 II I<: (t-t: o )
exp [.- r2
4 I<: (t-t )
1
• (20a)
0
(20b)
(22)
(23)
au =k V2u in Ox T, (24a)
at
u =u on r l x T, (24b)
q q on r 2 x T, (24c)
cu = Ir!u 0
u*
0
(25)
<j{,u =
2 r
I It 2 to
q*u
2
dt'dr, (26a)
l',ql=II
'17 r t
t
u*qldt'dr, (26b)
o
b=- dt'dr + I
r!
u u* dr!.
0 0
(26c)
cu = - 'd€,u 2 + ~ ql + b. (27)
(28b)
(29a)
300
and
b=b-c 1 ii (29b)
(30)
- -t1
where the symbols 1:1 and 1:2 stand for summation over the
nodes of f1 and of f2' respectively. The parameters U2i~ and
q1i~ represent the nodal values of u2 and q1 at node i on r
and node ~ on T. Since by definition the 1/Ii and ~~ are
interpolating functions, at the nodes ~i and t~ we have that
(34)
(35a)
(35b)
(36)
which, once solved, yields the nodal values u2ia and q1ia'
where lit is the constant time step (see Figure 2). Therefore
the interpolation function $a is defined by
(38)
o otherwise,
Choosing the colocation points to be coincident with nodes
(x' ,tl) on r x (to,tl) allows one to obtain from (36) the
following system of linear equations in matrix notation
(39)
where
[H~~J
J l. =H'l'l
J l. ' [G~~]
Jl. =G'l'l
J l. and
-1
J
-
{b,} = {b'l}.
J
(AOa)
IH~~]=H'l'l=
Jl. J l. (~tjJ.$
l. a ) -J l )
(x.,t
. '1 ar I y, matrl.X
Sl.ml. , [G11} requl.res
. .
l.ntegra 1s o
f th e f orm
304
I
J---4
~
I
I
I
,I
to t1 t2 to-1 to "t
=J
-
ft u*(x.,tl;x',t')
-J -
~.(x')
1. -
dt'dr(x')
-
(41)
rl to
-1
To obtain {b } we need to compute boundary integrals that have
the same form as (40,41) plus an integral over the domain.
1
r.n ~. u*dt'dr(~'), (42a)
2 (t -t') - - 1.
1
2
exp[- r ]dt'dr(X')
4K(t l -t') -
(42b)
2 2
s=_...;r:...-.__ , r r
t = t 1-- s=--,sl="" (43)
4K(t l -t') 41<:s o 41<:t.t
41<: 41<: 2
=2 exp(-so) =2 exp(- _r_) (44)
r r 4Kllt
[Hli1 =-~
2IT
f-
r2
~ E'~
r
2
exp(- _r_)1jJ. dr(x')
4ITt.t 1. -
(45)
306
This integral has to be evaluated numerically, ghoosing an
appropriate quadrature rule in each element of r2 and adding
up the contributions from all the elements. It can be shown
that the integrand is regular when r + 0, therefore Gauss-
-Legendre quadrature rules are suitable. These rules however
must be carefully selected. To choose too many integration
points is time consuming but too few may yield poor accuracy.
Also it is important to notice that for elements far away from
the collocation point the integrand on the right hand-side of
(45) varies little so we could use there fewer integration
points than for elements near or conta"ining the collocation
point or singularity. This could lead to adaptive quadrature
but so far nobody seems to have used them, due perhaps to the
increased programming complexity. This subject of numerical
integration was treated in another lecture (Numerical
Integration and Other Computational Techniques).
[G~~]
J1
0; I
r
I1 t
t
1
4nK(t -t')
1 0 1
(46)
.!. exp(-s)ds
s
(47)
r
where
11 _ 1 I r
2 ,
[G··1 -- _ E l ( - ) tjI.dr(x ). (49)
J1 4ITK rl 4Kdt 1 -
IQ0
u (x') u*(x.,t ;x',t )dQ(x')
- 0 -J 1 - 0 -
= -1- J_ 2
uo(~') exp(- _r_)dQ(x') (51)
4ITKLlt Q 4KLlt
Procedure I
Procedure II
(5Z)
(1. 53)
which, once solved, delivers {u~} and {q~}. This procedure
requires the formation of a triangular array of matrices that
can be depicted as
a=l [ H111........
a=Z 1HZl] ......... [H ZZ1 . . ......
a=3 [H 3lJ ... ..... [H3Z1 [H33 J
... .... .... ....
a=n [Hnl] [HnZl [Hn3] [Hnl
NOTATION
Y.P. CHANG, C.S. KANG and D.J. CHEN, The Use of Fundamental
Green's Functions for the Solution of Problems of Heat
Conduction in Anisotroplc Media, Int. J. Heat Mass Transfer,
16, 1905-1918 (1973).
Morris Stern
1 PRELIMINARIES
q/D in n (1)
315
316
~----~L----------4'------X
FIGURE 1
u(u,w) DI
-2 {V 2 uV 2w - (l-v)[u, w,
xx yy
+u w
'yy 'xx
Ii
- 2u, w, l} da {J)
xy xy
317
which is exactly the strain energy of the plate when both u and
ware identified with the deflection. Now Equation 3 can be
integrated by parts formally twice (say in u), and the resulting
boundary term involving tangential derivatives integrated
between corners. After some tedious algebra the result is:
d a + sini3 ..l
dn cosi3 ar r ao
a2 1 a 1 a2
t.l 2- r dr 2
(5)
at r ae 2
t.2
1.(1.. _d_)
ar r2 ae
and
D(l-V) + l+v V2 }
M - 2 - {cos2i3t.l + 2sin2i3t.2 I-v
D(l-v)
T - 2 - {sin2i3t.l - 2cosi3t. 2 } (6)
v· -D .!V 2 - {sini3
a _ cosi3 a + (osi3
1 a
dn ar r ae r R)ai3 }T
th
k corner of the plate boundary. The final s·tep is to
recognize that U(w,u) - U(u,w) vanishes identically, which then
leads to the reciprocal work rela~ion
u *f =.-!.../
81T0 lnr (8)
N*f 1
81TO(l + 21nr)rcosB
* ~(l+lnr) I-v
M
41T BiTcos 2B
f
(9)
F
*f(k)
~-r---L--------~~----x
FIGURE 2
(10)
+ -
boundary less rand r , the portion within the deleted circle.
We have also added two new corners at l+ and l-, and if P is
indeed a corner we should delete any contribution from there.
Now we apply the reciprocal work identity, Equation 7, to the
region indicated in Figure 3, identifying u with the concentrated
load solution, Equations 8 and 9. Again, as p ~ 0 the only
surviving contribution from the integral on rp is the term
~wlp and there is no contribution from the new corners at l+ and
l-. Finally, we note that the integrand can become poorly
behaved near P, but an analysis shows that the sum of the
contributions from r+ and r- goes to zero with p. Thus we
arrive at the boundary integral equation
K * (11)
+ l:: {F f(k)wl
k=l ~
f
rp
{'}ds = lim f
~O *
{'}ds
r
for the Cauchy Principal Value.
FIGURE 3
1
u* 2lTDr1nrcos¢ (13)
m
N*m 1
2lTD{cos¢cosS + Inrcos(¢ + S)}
*
F m(k) -I-v -
2lT
sin¢rr
r
- cos 2D]]
P j
-K
322
w - wi (15)
p
where
\!
c 2 [cos2y - cos2(c~ - y)] (17)
n
K
• • • }ds + E { • • • } Jum*qda (18)
k=l
n
323
FIGURE 4
(19)
K *
+ p{U*ml}dS + Z{u I}
~m
rp
(20)
K *
c- {Nlp+ - Nlp-} + p{U*m2}dS + Z{u I}
k=l m
rp
where
4 CONCLUDING REMARKS
ACKNOWLEDGEMENT
REFERENCES
Morris Stern
1 INTRODUCTORY REMARKS
327
328
second example (plane stress), we obtain a very useful alterna-
tive formulation of the boundary integral equations which leads
to a new representation for the normal stress component on the
boundary in the direction tangent to the boundary.
in ~ (1)
f{U dn
dV - V dU}ds = 0
dn
(3)
r d·
where r c a~ is the boundary of ~ and dn denotes the directional
derivative along the outward normal on r. Now suppose P is an
interior point of ~ and T* is a function harmonic in IT-p. We
construct the simply connected region ~ by deleting from ~ a
circular region centered at P and of raaius £ (smaller than the
distance from P to r) and then connecting the circular boundary
r£ to r by a cut with edges labelled r+ and r- as indicated
in Figure 1. Applying Green's second identity, Equation 3, to
the region ~£ and identifying the harmonic functions with the
solution of the boundary value problem T and the. auxiliary
function T*:, produces the result
in IT - P
329
FIGURE 1
(ii) lim
e+O
f T*ds = 0 so that the first term of the integrand
r£
in Equation 4 contributes nothing to the integral over re as
e+O.
(iii) lim
e+O
f dT*
dn
d
s
= cp r 0 so that the limiting contribution
re
of the second term onre is a non-zero multiple of Tip (the
value of the function T at the point p).
(note ~
dn
Ir+ = - ~
dn
I r- ) •
Thus if T* is a fundamental solution for P an interior
point of n, upon taking £+0 in Equation 4, we obtain the
representation
330
I
T pE n=cI
-:-
p
f{T* -dndT - T -}ds
dT*
dn
(5)
r
for values of the solution T in the interior in terms of the
boundary values of T and dT •
dn
As is well known, any constant multiple of In r, where r
is the distance from P to the argument point, is a suitable
fundamental solution for Laplaces equation in the plane; indeed
with this identification for T* Equation 5 is commonly referred
to as Green's third identity:
I I
T pEn = 27f f {T d (In r) - In r dn
dn dT} ds (6)
r
It should be noted that since r is bounded away from zero on r
(for P in the interior of Q) the representation Equation 6,
considered a function of P, is differentiable as often as
desired.
(it) lim
£-+-O J T*d S = 0 (note however, that in the present case
r£
r£ is not a complete circle.)
331
FIGURE 2
lim
(iii) e;+0
rp
The requirement (iv') is much less restrictive than (iv)"
and yields a larger choice from which to select fundamental
solutions associated with boundary points. In particular, if
332
T*(r,8) = A In r + B8 ln r (9)
8=rr!2
8=-rr /2
Then with A/O the boundary integral equation, Equation 8, takes
the form
f {[In r + !L
A
8 In r 1 dT
dn Q
rp
[cosS + ~ (8 cos S + In r sin S) IT}ds
r Ar
o (12)
(13)
and therefore
333
FIGURE 3
-.!
Ar
In r sin [3(->- +" B In r + 0(1))
Ar
is not integrable in the normal sense, however, the integral
converges in the principal value sense and the boundary integral
equation, Equation 12, is a valid representation for all values
of the constant B including, of course, B~O.
dr
dsp = sin 8
d8 cos 8 1
·dsp = - - r - - R
dS cos 8
(15)
dsp = r
FIGURE 4
335
to Q
FIGURE 5
in iil (17)
'NQ
~..,
= ~{ua.,..,0 + u Q + l2V u 8 o}
.."a. -v PoP a..., (18)
(19)
f
r
{t*u - t u*} ds
a. a. a. a.
=0 (20)
obtained from
(21)
(22)
r£ + r+ + r- + r
lim
(H)
£+0 fr£ ua.... ds 0
(iii)
lim
£+0 f t~ ds c P with (CP )2 + (CP
a 1 2 )2 '" 0
r£
Bl B2
o~ = -r cos 8 + -r sin 8 (26)
BI B2
T*
r8
= -r sin 8 - -r cos 8
(27)
I-v
B = --A
a
(28)
a 2
This is essentially the plane analog of Kelvin's solution, and is
used as well when P is a boundary point in virtually all conven-
tional representations of plane elastostatic boundary integral
equations derived by the direct method; see for example,
Hartmann, 1981, or Brebbia and Walker, 1980 for a discussion
of the properties of this conventional representation.
2 { I-v
u*(p) - -E 0ap In r + -2- e ap e
a 1f
l+v I (30)
+ --2- eaSe pcr r'Sr,cr
The components of the traction vector at Q then take the form
u
p IpE r f ft*(p)u
a a
t u*(p)
a a Ids (32)
rp
with n*Cp) and t*(p) given by Equations 30 and 31.
a a
As in the heat conduction example, we can examine the Be-
havior of the integrand of Equation 32 as the integration point
Q approaches the origin point P. The only difficulty arises
340
FIGURE 6
~
Cly
du 2 1:. [0n -
- ds p = E vo1;]
but at the point P
341
t. Q
n to Q
/ /',s
R
FIGURE 7
so that
dU 2
ern Ip = -E ds p + \! tIl p (33)
tJ {ca.u a. - t u }ds
a. a.
(34)
dr _ sinq, sinS
dsp -
so that
cos6 1 v
rrEr {(I-v) - (l+v)cos26}- rr;R
sin6 { (I-V)}
2rrEr (1+v)cos26 (36)
1
- - 2 {cosScos36 + cos6cos(26-S)}
rrr
1
- - 2 {2cosScos6sin26 + (1-cos26) sin(S-6)}
rrr
4 CONCLUDING REMARKS
ACKNOWLEDGEMENT
REFERENCES
Morris Stern
345
346
FIGURE 1
w(r,S;A)
HI (A) . 0)
r {b 1 s1n(A+l)S+b 2 cos(A+l)S
(3)
M(r,S) -DrA-l[(A+l)(l+V)AF(S,A)+F"(S,A)]
(4)
VCr,S) = -Dr A- 2{[(A+l)2+(1-V)A(A-l)]F'(S;A)+F"'(S;A)}
sin2A7r = 0 (5)
+r 5 / 2 {d [sin 58 _ 5(1-V)]sin ~
1 2 9-v 2
1/2 . 8 I-v
u*(r,8) r {BI[s~n 2 + 5+3v sin 38]
2
+ I-v cos T]
8 38
+B 2 [COS
2 7+v
-1/2 8
+r DI[sin + I-v sin 58
2 3+5V T
8 I-V 5
+D 2 [ -cos 2 + §=\j cos T]} (7)
Boundary values of the normal slope and of the moment and shear
resultants for the asymptotic singular deflection fields,
Equations 6 and 7 associated with the base of a through crack
are calculated from the formulas given in Lecture 22 and are
348
f u*qda (8)
Q*
w (9)
(10)
FIGURE 2
---
I
/ /
./ "-
"
\
\j
I .J.:
I );
\
\
""- -
FIGURE 3
r
J(u~',u') J{V*U'-M*N'+N*M'-U*V'}dS + L {F*u'-F'u*}
rl + -
Ll,L l
f[V*U'-M*N'+N*M'-U*V'ldS + L {F"'u'-F'u*}
r2 + -
L2 ,L2
r
J{'}dS + L {.} (12)
r -e+,r
is a path independent integral, where r is any smooth simple arc
connecting the edges and lying inside the plate. (If r has
additional corners the result still holds provided that the jump
terms are correspondingly augmented.)
A*+A'
J(u*,u') Cr
351
I
8=27T
Jp = {v*w -M*N +N*M -u*V }rd8+lim{F*w -F u*}
s s s s 0 s s £+ £-
8=0 p+,
(5+3v) U+V)
(14)
247TD(1-v) (3+v)
352
rp
-2a+J{w-w u * }ds +
x 0' m 1 k=l
~
{w-w,u * }
0 m1
(17)
(18)
3 CONCLUDING REMARKS
ACKNOWLEDGEMENTS
REFERENCES
1. INTRODUCTION
There are certain features of the stress analysis of cracked
solids that distinguish these problems from others to which
B.I.E. methods have been applied. The most significant
feature is that for a slit crack in a homogeneous elastic
medium the stress field at such a sharp crack tip may be
singular (the stress going to infinity like r- with r
measured from the crack tip). Accurate boundary element
modelling is required to obtain reliable numerical results.
More complicated stress and displacement fields occur in
composite bodies when cracks meet interfaces between different
materials or lie along such interfaces. Some discussion of
these singularities can be found in the review Atkinson (1979)
where reference to a large literature is also given. A recent
account of B.I.E. and its application to fracture mechanics
can be found in Atkinson (1983).
In addition to the need for accurate element modelling of
crack tip stress and displacement fields there is a difficulty
encountered when applying standard B.I.E. procedures to
fracture models in which a crack is modelled as having a
planform of zero thickness (e.g. a line crack in two dimen-
sions). If the direct B.I.E. formulation of a thin ellipse
is used to represent a crack when the ellipse degenerates to a
line, it has been pointed out (e.g. Cruse (1977)) that the
standard formulation is indeterminate. This difficulty and
ways of overcoming it will be described below. It is also
worth noting that information about 'stress-intensity factors'
(notation to be explained in section 2) can be obtained from
stress and displacement fields far from the crack tip by
judicious use of certain paJh independent integrals so that
a less accurate model of a crack can still be put to good use.
(An account ~f these methods will be given in lecture 2.)
Finally it should be noted that a natural representation for
a crack is by continuous distributions of dislocations.
355
356
where the tensors Ok" and Sk" are determined from U.. and
T .. respectively, by1differentiation with respect to i~e
pdlnt p(~) and using the stress-strain relations.
The above results give the stress and displacement at an
interior point. The well known boundary integral equation
is obtained by taking the limiting form of equation (1)
(sometimes known as Somigliana's identity for the displace-
ment vector) and letting the interior point p(x~ tend to a
boundary point P(~). Taking due account of the disconti-
nuity of the second integral on the right hand side of
357
(4)
The kernels U.. and T.. are defined in equation (A8) and have
the property t~at Jl
and
u.(p) ~
)
f
s)
U.1..(p,Q)t.(Q)dS(Q) -
1. S)1.
f
T .. (p,Q)u.(Q)dS(Q)
1.
- fr)1.
T .. (p,Q)6u.(Q)dS(Q)
1.
+ f
r
u .. (p,Q)
)1.
where
(8)
(9)
358
+
p
Figure 1
Definition of crack modelling terms used
in deriving equation (4)
In general, situations are encountered where the crack is
loaded by equal and opposite tractions such that Et .(Q) = O.
The usual next step is to let p(~~ * P(~) a boundaty point in
equation (7) in order to obtain the boundary integral equation.
However, when P(~) is a point on the plane r (not at the edge
of r) the equation becomes
u.(p) - 6u.(P)/2 + f
T .• (P,Q)u.(Q)dS(Q)
J J S J~ ~
where the usual term for the jump in the boundary integral
has been taken and the integral over r is a principal value
integral. Equation (10) is deficient in two critical respects.
In the first place while a single surface r is being treated,
two variables are unknown. These are 6u. and Eu. =2u.(P) -
fiu.(P) . The second deficiency is thatJequatio~ (101 is the
sa~e for any set of equal and opposite crack boundary tractions
(i.e. such that Et.(Q) = 0 where Q lies on r). Thus this
integral equation ~annot deal with situations where the crack
is loaded internally.
As a consequence of the above mentioned deficiencies inherent
359
in a straightforward application of the boundary integral
equation (10) to flat crack modelling, a variety of methods
have been suggested which each deal with special cases. No
all purpose method applicable to a variety of unsymmetric crack
plan forms seems to be available at the present time. We
indicate a possible method in Section 3 and discuss the
variety of methods which have been suggested for special
situations. Note, however, that the above-mentioned diffi-
culties are all specific to the sharp crack model of fracture
mechanics. Useful numerical results might still be obtained
by modelling the crack as an elliptical hole to give an
accurate stress field far from the crack and then use invariant
integrals to calculate the (sharp) near crack tip stress
field. A discussion of such invariant integrals will be made
in lecture 2.
2. STRESS INTENSITY FACTORS
As discussed in the introduction there are certain features of
the stress analysis of theoretical fracture models that distin-
guish these problems from others to which B.I.E. methods have
been applied. We include here a brief description of notations
and results pertaining to a crack in a homogeneous elastic
body. For situations where cracks meet interfaces between
different elastic media or lie along a bimaterial interface
different singularities are encountered, for a review of some
of these situations see Atkinson (1979).
Fracture modes and stress intensit¥ factors
A key feature 1n the stress analys1s of the crack models
considered here is the singular stress field at the crack tip.
For a crack in a homogeneous medium an eigenvalue analysis
(see for example Williams (1957)) gives a result of the form
-1/2
a ~ KF(S)r + non-singular terms (11 )
2 (0) 2 (b)
Figure 2
Fracture modes. 2(a) opening mode; 2(b) sliding
mode; 2(c) anti-plane shearing mode
A standard notation is to use KI for the opening (mode I), KII
for in-plane sliding (mode II), K for anti-plane sliding
(mode III) modes of relative disp{!tement of the crack surfaces.
The following results are standard for isotropic bodies in
plane strain, plane stress or anti-plane strain. For the pure
opening mode (Figure 2(a)), we have
I - Since!2)SinC3e!z)}
cosCe!Z) { sinCe!Z)cos(3e!Z) (12)
I + sin(e!Z)sinC3e!2)
l Ull
.
Uz
=
K
-.l c..!..) ~
z~ z~
f cos(e!z) [K-I+Z
sinCe!2) [K+I
(13 )
t
-siu(e!2) [2 + COS(O!2)COS(3S!Z)I}
cos(S!2)[1 - sio(0/2)s10(30/2)1 (14 )
sio(O!2)cos(O/2)cos(30!2)
361
1
and
j -since/2)l
l cosC e/2)
(16)
and
K
2 -!!!
II
C~)1/2
211'
sinCe/2) (17)
cr - E
( 18)
where E is Young's ·modulus. The Griffith fracture criterion
then says that fracture occurs when G = 2y (y.being the. surface
energy of each of the newly created surfaces). Thus when a
single mode is operative the criteria of critical stress
intensity factor and critical energy release rate are equiva-
lent. However, in the mixed mode situation for a crack
growing in its own plane the energy release rate is
2 K 2
G "' (Kr 2 + Kr/> (1~\I > + r~~ (19)
,.-:=.0_ '----~~---Xl
Figure 3
A crack positioned unsymmetrically in a body
When there is no symmetry in the cracked specimen it is still
possible to get a formulation like that obtained in the last
section by subdividing the body into regions. We illustrate
this approach by considering the configuration shown in
Figure 3. For convenience we denote the part of the boundary
described by ABCDA as r and the part described by ADEFA as
r 2 . Then applying the lntegral equation (21) to both regions
(1) and (ii) gives the equations
(1) (1)
[u. (Q) - u. (P»)T •• (P,Q)dS(Q)
1 1 J1
(22)
(1)
t. (Q)U •• (P,Q)dS(Q)
1 J1
and
366
f [u.(2)(Q)
L - u. (2)( P)JT •• (P,Q)dS(Q)
r2 L JL
(23)
(2)
- rJ t.
1
(Q)U •• (p ,Q)dS(Q)
JL
2
Along the common boundary AD we have the boundary conditions:-
(i) Continuity of stress, i.e. t.(1) = _t.(2) for all
points on AD. 1 1
(24)
t
f(t) in the form
.!
II
a~~)(x)
1.J -
• _(4
W
I)
IC+
J
-1
f(f;}K •• (x,f;)df;
1.J -
where
2 2 2
(IC-l) In r - 4x2 /r (26)
f;-x 1 4x 2 (f;-x 1 )
-2(1C+l)arctan ( - - ) - 2
x2 r
and
t. (l)(P) + t. (2)(p) = t.
(p), P e c.1 t
1 1 1
where u.(P)
1
and t.(P)
1
are the specified boundary displacements
369
and tractions of the original problem
Also for IXll < 1, x2 ~ 0
(29)
(32)
Figure 4
A Somigliana dislocation
A natural extension to the approach of the last section is to
model the crack as a mosaic of infinitesimal dislocations of
area dS 1 , normal nk and a discontinuity vector b. which is
effectively constant allover the small area dS1~ The displace-
ment produced by one of these elementary dislocations can be
written
du.( r) (34)
J -
371
where p~~)(r - r') is the stress produced at a point r' on Sl
by a polnt rorce of unit magnitude at r parallel to the x.
axis and n is the normal to S at r'. A Somigliana dislJc-
ation at Sk (see Figure 4) is tharacterised by the vector
2 whichis\he displacement on the arrow side of Sl defined
by n, minus the value on the tail side of the arrow. Thus
in an infinite linear elastic medium the displacement field
due to the Somigliana dislocation can be shown to be
(35)
Note that as far as the above formulation is concerned the
surface 51 can take any shape, so in principle a crack of
any shape can be represented in this way. However, the
numerical solution of the integral equation which results
when expression (35) is differentiated to obtain the stress
field requires further investigation. A proof of (35) which
also verified its properties for an anisotropic medium was
given by Burgers (1939). A derivation of (35) by means of
the reciprocal theorem is attributed to Eshelby in the book
by Nabarro (1967). We shall give in the next section a
derivation based on Betti's reciprocal theorem.
In principle the integral equation deduced from (35) by
computing the stresses at r and then lettingr tend to 5 can
be coupled with a boundary integral formulation of the Jxternal
boundary in a similar manner to that of the last section. How-
ever, the numerical implementation of such a procedure has yet
to be made. It should be noted nevertheless that this formu-
lation can be used for any crack profile.
The special case of a flat crack in an infinite medium has been
tackled by Weaver (1977) using a formulation similar to that
discussed above. He reduces the singular nature of the
integral equations by first dOing an integration by parts.
3.6 General formulation
We discuss here a general formulation of the crack problem in
which the crack profile is modelled by a distribution of
Somigliana dislocations. We begin with the formulation of the
situation shown in Figure 1. If in equation (7) the external
boundary 5 is allowed to tend to infinity and the boundary
displacement and stresses are such that the integrals over S
tend to zero, then since Et.(Q) = 0 on r the following result
is obtained:- 1
u . Cp)
] = - fr T .. Cp,Q)4u.CQ)dSCQ)
]1 1
(36)
(37)
- f T.. (P,Q)du.CQ)dS(Q)
r 1.J J
where du.(Q) +u.(Q+) : u.(Q-) and we have used the fact that
rt~(Q) =Jti(Q ) ~ tj(Q ) J 0 across th~ crack. Differentiating
(31) with respect, to the co-ordinates of p to obtain the
stresses gives
- rf Sk··(P,Q)duk(Q)dS(Q)
1.J
(38)
and
t 1.*(p,q) a - !r {dr [(1-2v)6
dn ij
+ 2r .r .J -
,1 ,J
(A2)
(l-2v)[n.r .-n.r .J}e./4w(1-v)
J,l 1,] J.
r .
ar (A3)
,~ 3x.
1
lim
e:-+O
f ti*dS - 6 •. e.
lJ J
(M)
and
t .*(p,q) - - ~ {~[(1-2v)6 .• + 3r .r .] -
1 r2 dn lJ ,I ,J
(A6)
(1-2v)(n.r .-n.r .)}e./8w(1-v)
J ,I 1 oJ J
such that the first index in U.. (p,q) and T.. (p,q) corresponds
to the direction of the pOint ~6ad and the ~~cond index refers
to the component of the respective displacements and tractions.
Chapter 21
1. INTRODUCTION
In the previous lecture we have outlined the difficulties
associated with developing a B.I.E. formulation for slit cracks
of a general planform and reviewed various methods of circum-
venting these difficulties. In addition to the question of
which integral equation formulation to use in a given situation
there remains the question of appropriate element modelling to
represent stress singularities. In section 4 different methods
for stress singularity modelling will be outlined and compared
on a model problem, that of a crack under longitudinal shear
(anti-plane strain) where the governing equation is Laplace's
equation. The methods suggested there should however be
applicable to more complex situations. In sections 2 and 3
we describe invariant integrals which can be useful in deducing
values of stress intensity factors from numerical information
far from the crack tip. These are used in the model problem
of section 4 also.
Finally in section 5, we describe a problem of debond stress
analysis and indicate how the methods outlined here together
with other analytic information in certain cases could be used
to give an accurate description of a debonding event.
2. INVARIANT INTEGRALS BASED ON THE ENERGY MOMENTUM TENSOR
We have already indicated briefly in the last lecture how
certain invariant integrals can be used as tools for deducing
numerical values of stress intensity factors. Here a simple
deduction of integrals which can be derived from a (perhaps
pseudo) energy momentum tensor is made. The prescription
used is similar to that described by Eshelby (1970) although
equivalent results can be obtained by a formal application of
Noether's theorem (see Knowles and Sternberg (1972) and for
somewhat equivalent earlier work by Gunther (1962).
377
378
1
c •• c -2 (u • • + u •• )
1.J 1.,J J,1.
(7)
0 ....
1.J
p G.• k. &k. - P ~ .•
1.J.'. 1.J
e ( anisotropic)
(k •• /To)lf .... p m
2- If + p 2-
•.• -c ••
1J ,1J 1J 1J (8)
+ 1 k 1 --2
2ToP I.J
" e,1.. e,J. + -2 P m e ( 10)
where (11 )
k ..
(t .. - pe~
I.J
.. )~.
I.J L, t
+ ~a
TOP • i "
e,. - Lo lj (13 )
p •• dS. (14 )
.. J J
t 1J
.. u.1, R.. (15)
which is the expression for the Energy momentum tensor for an
elastic medium. In this case for a plane crack, with tip
pointing in the 1 direction, the crack extension force or
energy release rate (G) can be defined as
G = Fl = f Pl' dS. (16)
s J J
where S is a surface enclosing the crack tip dS. = n.dS where
n. is the outward normal (in the plane strain case tHis is a
cylinder with generators parallel to the x axis so the
integral in (16) is effectively a line int~gral in the (xl'
x2 ) plane). For a stress free crack lying on the x, axis
t.? = 0 so P3? = 0 from (15) and the integral (16) lS zero
w~en taken along the face of the crack.
2.2 Land M integrals in elastostatics
In the above we have discussed the integrals F~ associated with
the tensor PR..' paying particular attention to the integral Fl
which can be ~hownto be related to the energy release rate G.
Under certain conditions other invariant integrals can be
derived from P~. and are sometimes useful. We describe these
briefly. In aoaition to the F~ integrals obtained in section
2.1, Gunther (1962) obtained for elastostatics the path-
independent integrals
Li Z Eikt ! (xkP tj + UkP tj ) dS j
(17)
and
c.
Figure 1
383
u -uo
6 t
~ ~
4\.1 (~)~{[-(2K+l)sin~
2n 2 + Sl.n
. T 30]K
1
-[(2K+l) cos 2e - 30 ~
3 cos :r ]K rr } + oCr )
1 ~
{( 3 cos 2
0 + cos :r
30) Kl - ( 3Sl.n
' 2 e + 3 sl.n:r
. 3e)K II }
4(2nr) 1-
+ oCr-'l,
1
--::....-,-~
{('
Sl.n 2e . T
+ Sl.n 30) Kl + ( cos 2e + 3 cos T
30) KII }
4( 2nr) _.1
+ o(r 2) (22)
where
o
u and
0
ue arO the radial and tangential components of
the displacement ~ of the crack tip and
KI = lim (211r)~ooI0=0
r-+{)
(23)
Kn lim ( 211 r) ~o rO I 0 =0
r-+O
384
are the conventional stress intensity factors. The remainder
terms are of the order indicated in distance from the crack
tip.
Thus when C is a small circular contour close to the crack
tip, ds = rae, and the dominant contribution from the traction
T involves the stress intensity factors multiplying spatially
varying terms proportional to r-. Hence if the auxiliary
displacement u ~an be chosen to have dominant behaviour pro-
portional to f-~ as r tends to zero the product of T.u"with ds
will have a n~n-zero finite limit as € tends to ze~~'2 Similarly
the traction T which will now be proportional to r as r
tends to"zero-~ill lead to the required 1/r behaviour in the
product 1.(u-u). The only stipulation is that u and T must
be equilibrium solutions and moreover must satisfy the-stress
free crack boundary conditions. For elastic problems an
exact solution of the field equations satisfying stress free
boundary conditions can be found by an eigenfunction approach.
The auxiliary elastic field required for the above problem
has been given by Stern et al (1976) as:-
ar =
1
~ {[(2K+l)cos :r
36
-3 cos Z6 IC l +
2(211r) (1+1<)
36 6
[(2K-l)cos-
2 - cos zlc 2 } (24)
~
~
2(2~r3)~(1+K)
{(7 cos :r
36
-3 6
cos -Ic +
2 1
7 . 36 .
[ s1n:r - S1n 26 JC 2
}
[cos ~ + cos 0 J e 2 l
2 2
385
where c, and c2 are arbitrary constants.
Now on the inner circular boundary the evaluation of the
contour integral in terms of the traction and displacement
(relative to that of the crack tip) components takes the form
I ..
L
~
an = Vu(Q) . _n(Q) .
Hr-____________________-i"G
PI,,;)
/
/
,/
0/(8
A I-------------=L J'__ a
t I'---------'F
Figure 2
Region with a slit AO.
The boundary conditions are taken to be
387
u = 1000 on AE
u = 500 on BO (28)
~~ = 0 on OA, EF and FB
where
f j ( r, e) - r
= (2j+l)/2 nn
. { (2j+l)e)
2
!~
intensity factor (SIF). Note that, for the model problem we
have
u
°1
r -
t = (33)
l'.=1
388
where ~he coefficients ut ~nd ttl are dete~mined so that with u
and t 1n (27) replaced by u and t. respect1vely. the(i~tegral
e~M,tion is satisfied exactly at m boundary points P •...•
P
Thus. one obtains
k • l ••.•• m (34)
vhere
Table 1b
Values of u at r 3/4; 'exact' value = 634.45 (SYMM (1973)}
Method 2 3 4
N q =0 q 2 q =4
2 596.96 634.29 634.15 634.68 634.56 634.50
3 611. 49 634.40 634.77 634.50 634.46 634.45
4 621.29 634.44 634.88 634.43 634.45 634.45
5 627.70 634.45 634.85 634.45 634.45 634.45
Calculation of an approximation a to a
For Method 4, above, we can take ,8 ~ aO. By whatever method u
and t are'computed, one can use thg for~ulae obtained by sub-
stituting u and t into the r.h.s. of (31) and (32), respectively,
to obtain approximations to a O:
• (u) - -1/2 - -
ao = a 0 (u) = lim r [u(r,~) - u(O)] (31)
r->O a
and
• (t) ~ 1/2-
aO =a 0 (t) = lim Zr t(r,O). (32)
a
r->O
Note that if u and f are obtained from
(i) Method 4, then
o(u) (~) = a(t)(£T= -
o 0 laO·
391
(ii) Method 1, then
d(~)(u1 = a(~)(t) = O.
Nevertheless, in case (ii), one can obtain useful approximations
to aO from u and t using the approximations to (31) and (32):
-(u) - -1/2 -
ao (u,ru ) - ru [u(ru'w) - ~(O)l
and
-et) - 1/2-
a 0 (t,r t ) = 2rt t(r~,O) (32)b
with r ,r sufficiently small. Formulae (31) and (32) can,
of cou~se,tbe used to compute approximations t8 ao howevbr u and
t are calculated. For Method 1, experiments showed that there
exist constants a, v, a*, v* such that, .with
r = a!v
r - a* !v*
u '1 '
i~b,r~ ! 1 -.length of element n = 7/N),
= leng!~t~f_element
a (u,r) and a (t,r ) approxlmate closely a. The attempt
to deterMine such const~nts was prompted by theOobservation of
Schatz and Wahlbin (1978) that, f8~ the ~13ndard finite element
method, a good choice of r for a is h where h is the
maximum diameter of all elements. O The fact that we were able
to find cO~~rants a, v, aft)v* so that ag is closely approxi-
mated by a (u,r) and a (t,r) is pr bably significant.
Additional ~pproxYmations eo a ~an be obtained from formulae
deduced from certain path-indeBendent (invariant) integrals,
the F- and M-integrals, evaluated along a circuit surrounding
the point where the singularity occurs (see section 2). These
integrals are defined by
F (. = f P •• n. dS, (36)
S J .. J
M = f
S
x P. n. dS ,
t Jt J (37)
N 2 3 4 5
(F)
..... Cl 0 (u,t) 155.52 154.22 153.57 153.18
"tl (a)
o (M)
-!'iCl 0 (u,t) 152.29 152.07 151.96 151.89
'"
:i:
N (u)C) Cl(~)(t) (n - -
Cl 0 (u,u)
(M) - -
Cl 0 (u,t)
Cl 0 u
Table 2 (continued)
N q ao a1 a2 a3 a4
Figure 3
Typical debond pattern observed in the pull
out test (A) Tip debond (B) Surface debond.
At a later stage in the loading process debonding initiated at
the surface and propagated down the sides of the rod (Figure
3 (B)) completing pullout. In order to explain these debonding
events numerical evaluation of the energy release rate for both
surface debond (Figure 4) and rod tip de bond (Figure 5) was made.
The numerical results were obtained by finite elements, the
polyurethane matrix was assumed incompressible and the glass
rod rigid.
-
"~O.30
'.
0.35 • TOTA(ENERdi"
,. MODE I CONtRIBUTION
o MODE II CONTRIBUTION
---- DISPLACEMENT METHOD
~
:§.O.25
....UJ
~ o.zo
UJ·
~
w
0.15
-'
w
a: 0.10
>-
------
"ffiz O.O~
w
0.1 o.z 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 I.Z
• CRACK TIP LOCATION A,/B
Figure 4
Numerical evaluation of the energy release
rate for a surface debond including mode
separation.
395
0.15 • TOTAl.lNERCY
N: & MODI I CONTRIBUTION
o MODE II CONTRIBUTION
~.. 0.12
- NUMERICAL RESULTS
--_. ANALYTICAL RES&A.TI
:§.
w
!:( 0.09
a:
w
II>
~ 0.06
a:
>-
~ 0.03
~
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
CRACK TIP L99~J;!Q.N··l.Cl(Bt~1
Figure 5
Numerical and analytical energy release
rates for a tip debond including mode
separation.
It is worth noting that even in the absence of debonding, the
stress analysis of the embedded rod has stress singularities
at the rod base and where the rod leaves the free surface.
Thus axially symmetric B.I.E. methods could be used but require
accurate modelling of these stress singularities. Furthermore,
modelling the debonding process itself can lead to difficulties
if the rod is elastic and the matrix compressible. In this
case oscillatory singularities may occur which lead to results
which imply interpenetration of the crack faces near the tip.
Recently contact zone models have been suggested to remove this
anomaly ~ijt the contact zone length can be very small e.g. of
order 10 times the crack length (Comninou (1977) and Atkinson
(1982)). Nevertheless, it ought to be possible to do B.I.E.
analysis of the ordinary crack model (which gives the oscilla-
tory stress and displacement field) and derive the near crack
tip behaviour using a far field integral such as described in
section 3 (Hong and Stern (1978)). Once this 'pseudo' crack
tip behaviour is determined it acts as an 'outer solution',
in the sense of matched expansion, to the problem rescaled on
the contact zone length. this new problem has been solved
analytically by Atkinson (1982 (a), (b)) and the singular
perturbation procedure outlined in the second of those papers.
Some features of the experimental results for the pullout
problem can be explained using the results of Figures 4 and 5.
The reader is referred to Atkinson et al (1982) for more
detailed discussion of the experimental results and the
analytical modelling used to explain them.
396
REFERENCES
Atkinson, C. and Smelser, R.E., (1982) Invariant integrals of
thermo viscoelasticity. Int. J. Solids Structure 18, 533-549.
Atkinson, C., Xanthis, L.S. and Bernal, M.J.M., (1981) Boundary
integral equation crack-tip analysis and applications to elastic
media with spatially varying elastic properties, Compo Methods
in Appl. Mechs. and Engng. 29, 35-49.
Atkinson, C. (1982) The interface crack with a contact zone.
(An analytical treaement) Int. J. Frac. 18, 161-177.
Atkinson, C. (1982)h The interface crack with a contact zone
(The finite crack problem) Int. J. Frac. 19, 131-138.
Atkinson, C., Avila J., Betz, E. and Smelser, R.E., (1982) The
rod pullout problem, theory and experiment. J. Mech.Phys.Solids
30, 97-120.
Comninou, M. (1977) The interface crack, ASME JAM 44, 631-636.
Eshelby, J.D. (1-970) Energy relations and the energy-momentum
tensor in continuum mechanics, in Inelastic Behaviour of Solids
77-115 (ed. M.F. Kanninen et al.) McGraw-Hill, New York.
Gunther, W. \1962) Uber einige Randintegrale der Elastomechanik,
Abh. Braunschw. wiss. Ges. 14, 54-63.
Hong, C-C. and Stern, M. (1978) The computation of stress
intensity factors in dissimilar materials, J. Elasticity
8 21-34.
Knowles, J.K. and Sternberg, E. (1972) On a class of conservation
laws in linearised and finite elastostatics. Arch. Rat. Mech.
Anal. 44, 187-211.
Schatz, A.H. and Wahlbin, L.B. (1978) Maximum norm estimates
in the finite element method on plane polygonal domains,
part I. Math. Comput. 32 73-109.
Soni, M.L. and Stern, M. (1976) On the computation of stress
intensity factors in fiber composite media using a contour
integral method, Int. J. Frac. 12, 331-344.
Stern, M., Becker, E.B. and Dunham, R.S. (1976) A contour
integral computation of mixed-mode stress intensity factors.
Int. J. Frac. 12 359-368.
Stern, M. and Soni, M.L. (1976) On the computation of stress
intensities at fixed-free corners. Int. J. Solids Structures
331-337.
~7
B.E.M. IN GEOMECHANICS
R. Butterfield
1. INTRODUCTION
399
400
where G(x,O = __ 1
21TK
tn 1..£.1
r'
o
is the basic singular solution for the potential generated at
point x. by a point source applied at the point ~. (i = 1,2)
in an 1 infinite two-dimensional region of permehbility K and
if ni (x) defines a unit vector at Xi then,
(2)
u(x)
JS F(x,~) ~(~) dS(~) + J F(x,z) ~(z) dA(z)
A (4)
( a)
(b)
FIGURE 1
404
X2'~2.Z2 _______
,/'
,
(
I
\,
~'---
FIGURE 2 /
T
/
(a)
FIGURE 3
405
,~
A, Au Au u,
f-
Bu I -B., Uu
r-
B"
'---
2 ~ Ull
I-
--
A2 A.3 A2' u.
I~~
B.3 3 -B32 Un
t-
B24
L...---
4
~ ..... u••
~-=CI,I-CJI A3 U.
0- Cu - C:u
I-
o 0-~U-CH B•• S -B•• 03
I-
~==CI.. -C.2
P. 0- Cu- C..,
A. u.
0. r-:
D,
Du
D"
-D2'
-D3'
.,
D2
D13 -D32
•• (3-66)
D•• -D.2 .3
DJ
DJ' -D. J
"'.
D.
FIGURE 4
409
(13)
(ii) since, from the B.E.M. viewpoint, the boundary of the plate
is merely its edge the above algorithm illustrates how
coupling can be achieved between the internal cells of
one body with the boundary elements (half space surface)
of another.
410
(14)
(15)
(16)
(17)
(18)
411
which will, for example, yield all the fluxes if {PI} and {P2}
are specified or (after re-arrangement) the unknown variables
for a mixed boundary value problem. A further four zone
example is shown in Fig. 3(b) and the corresponding solution
matrix arrays in Fig. 4 from which the general procedure for
assembling matrices for multi-zone problems should be clear.
(19)
can be transposed to
(20)
(21)
(22)
For algebraic convenience we shall now drop the body force term
and note that from the recurrence relation established by
equation (22) we can now relate
{S~} ~ {:;} to {S;+l },the top surface of the (N+l) th layer via
(23)
l
11
ICXY"'O I
1-+ 2
l~ 3
~~
l-f N
~\
rr
Q. 0.
t,! u.,
2
L-:\
t 2'
b ~.
a.
FIGURE 5
413
(24)
5. ELASTO-PLASTICITY
6. CONCLUDING REMARKS
REFERENCES
W. L. WENDLAND
INTRODUCTION
417
418
(I. I ) Au =f •
(I. 4)
or, equivalently,
N
(1.5) (A~,1Jk)J' L y~(A1J~,1Jk)'J
R.=I
= (f,llk)'
J
k=I, ... ,N.
This is a quadratia rinea~ system of equations for YI""'YN
and the crucial property for the asymptotic error analysis is
existence and stability of the family of the so called
Gare~kin p~ojeations
. -I
(1.6) Gh(A,J)u := (PhAP h ) PhAu ~
Let wh E ~ be defined by
PhDwh = PhDw •
Then G~rding's inequality and (v,z). ,;; c Ilv II. liz II·
yield J J-a J+a
2
Ko Ilwh IIj(a';; I (Dwh,wh)jI = I (PhDwh,wh)jI = I (Dw,wh)jl
yielding
-I , -I
[I-Gh(D,j)D CJu h = Gh(D,J)rI-D CJu,
C : HJ' +", + Hj -'" ~s
, compact an d G ( ') ' J ' +'"
h D,J + I ~n H There-
fore
lim sup IIGh(D,j)D-Icv - D-ICv II, = 0
h+O Ilv Ii, =1 J+'"
J+'"
[7, Hilfssatz 3J. Moreover [I-D-ICJ- I = A-ID exists.
Therefore the inverse of
All the terms of this formal expansion must also satisfy some
growth conditions. Then A is called a pseudo-differentiat
operator of order 2a (for details see e.g. [47J). Clearly, k
and a depend on the choice of the local representation !(t)
of r ~
10 -x2
t(x) = M(X)~ where M(x) .. ( 0 1 xl
o 0 0
such that ~(x) - t(x) = 0(1) and a!. (~(x) - t(x» = o(Tir)
for Ixl"'''', (In 112 cancel the lastJ three columns and the
third row.)
423
(2.3) +
u(x) = E J +
{E(x,y)T[u](y) +
- T(x,y)u(y)}ds +
+ M(x)w
r y
where E = I and ~ = 0 for interior and E = -I for ex-
terior problems. The kernels are given by
A + 3lJ A+lJ I + + + + T
E(x,y) = 41f(n-I)(A+2lJ) {y(x,y)I + H3lJ ---n (x-y)(x-y) }
Ix-yl
where y os -log Ix-yl for n = 2 and y = Ix-yl-I for n=3,
lJ n(A+lJ) + + + + T a
T(x,y) = 21f(n-I)(A+2lJ){(I + 2 (x-y)(x-y) )av- y(x,y)
lJlx-yl y
___1_ «;'y>"~T (y) _ «;'y)~T (y» T)} .
Ix_yin
A + 3lJ '-I I (
~ 8(A+2p) N (x) ~ -K~~~2
I ~ 12+K~22,
I
'~:K~l
-K~I ~2
12 2 ,
, 0)0 N(x)
,I ~ 12
for n" 3 ,
(see [13], [34])
(2.5)
fr MT(y) t(y)ds
y
=0 .
Order 2y /
0 • Principal symbols:
E , -1 Y
~)
. TIT
(2.6) = ( .
1Y TIT '
~
E for n=2
Order 2a = I .
for n = 2
N(x)
for n =3
Further boundary integral equations with corresponding prin-
cipal symbols can be found for problems in electromagnetic
fields [1],[34J, acoustics [10],[45],[51] , viscous flows [13],
[25] and will be collected in [14].
for all X € Hh
Then
427
HP (sUPP(~k»
where ~~k'~' ~ ,fk denote the numerical weights. In order
to combine Inequalities (3.5) with the stability of the inverses
of Equations (3.1) we now require the family of meshes on r
to be regular, i.e., we further require from ~ an inverse
assumption:
(3.6) for all X E ~
(4.2) Ju:=
... f'"uds, ...
JIJ.u:= I·
N ...
0ku(~)
r k=1
where ok denote the weights of the trapezoidal rule.
(4.3)
1
-I for where
ER,(t) t € [tR"t HI ) h H1 =t H1 -tR, •
:H1
elsewhere
const.
I I
p:t>ovided 2a ~ r ~ t ~ m+ 1 , r < m+"2 ' 2a +"2 < t in case of
the smoothZy graded meshes and 2a ~ r ~ a+j ~ t ~ m+1 , 2a < m
in case of odd m and arbit~y famiZies II = b of meshes
Concluding remarks: From the results by Prossdorf and by
Schmidt [40],[44] for uniform meshes it follows that the strong
ellipticity condition for A is not only sufficient but also
necessary for the optimal order convergence of spline collo-
cation on the above a as well as spline Galerkin methods if
n=2 • Schmidt also modifies the choice of collocation points
to approximate a much wider class of operators A.
REFERENCES
35. Nedelec, J.C. (1976) Curved finite element methods for the
solution of singular integral equations on surfaces in
R3. - Compo Math. Appl. Mech. Eng., 8: 61-80.
36. Nedelec, J.C. (1977) Approximation par potentiel de double
cuche du probleme de Neumann exterieur. C.R. Acad. Sci.
Paris, Ser. A 286: 616-619.
37. Nedelec, J.C. (1977) Approximation des Equations Integrales
en Mechanique et en Physique. Lecture Notes, Centre de Math.
Appliquees, Ecole Poly technique, Palaiseau, France.
38. Nede1ec, J.C. (1982) Integral equations with nonintegrab1e
kernels. Integral Equations and Operator Theory, 5:
502-572.
39. Nede1ec, J.C. and P1anchard, J. (1973) Une methode
variationne11e d'e1ements finis pour 1a resolution
numerique d'un prob1eme exterieur dans lR 3 • R.A.I.R.O.
7, R3: 105-129.
40. Prossdorf, S. and Schmidt, G. (1981) A finite element
collocation method for singular integral equations.
Math. Nachr. 100: 33-60.
41. Richter, G.R. (1978) Superconvergence of piecewise poly-
nomial Galerkin approximations for Fredholm integral
equations of the second kind. Numer. Math., 3]: 63-70.
42. Rizzo, F.J. (1967) An integral equation approach to
boundary value problems of classical elastostatics.
Quart. Appl. Math.,25: 83-95.
43. Saranen, J. , Wendland, W.L. (]982) On the asymptotic
convergence of collocation methods with spline functions
of even degree. Mathematics of Computation. (Preprint
700, FB. Math., Technische Hochschule Darmstadt, 1982):
To appear.
44. Schmidt, G. (1983) The convergence of Galerkin and collo-
cation methods with splines for pseudodifferential
equations on closed curves. Zeitschrift f. Angew. Analysis:
In print.
45. Stephan, E. (1982) Solution procedures for interface
problems in acoustics and e1ectromagnetics. In: Theoretical
Acoustics and Numerical Techniques (ed. P. Filippi.
Summer School Udine 1982) Springer, Lecture Notes in
Physics: In print.
46. Stephan, E., Wendland, W.L. (1976) Remarks to Galerkin
and least squares methods with finite elements for general
elliptic problems. Springer Lecture Notes Math. 564:
461-471 and Manuscripta Geodaetica J: 93-123.
47. Taylor, M. (1981) Pseudodifferential Operators. Princeton:
University Press.
48. Watson, J.O. (1979) Advanced implementation of the boundary
element method' for two- and three-dimensional 'elastostatics.
In: Developments in Boundary Element Methods] (ed. P.K.
Banerjee and R. Butterfield) London: Appl. Science Pub1.
LTD: 31-63.
49. Wendland, W.L. (1980) On Galerkin methods for integral
equations of elliptic boundary value problems. In: Numerical
Treatment of Integral Equations (ed. J. Albrecht and
L. Collatz), Intern. Ser. Num. Math. 53, Birkh§user Basel:
244-275.
50. Wendland, W.L. (1981) Asymptotic accuracy and convergence.
In: Progress in Boundary Element Methods (ed. C.A. Brebbia),
London • Plymouth, Pentech Press: 289-313:
51. Wendland, W.L. (1982) Bounda~y element methods and their
asymptotic convergence. In: Theoretical Acoustics and
Numerical Techniques. (Ed. P. Filippi, Summer School
Udine 1982), Springer, Lecture Notes in Physics, (Preprint
690, FB. Math., Technische Hochschule Darmstadt, 1982):
In print.
Subject Index
Anisotropy 124
Approximate solutions 5
Boundary conditions 4, 26
heat transfer 112
non-linear 184
wave equation 194
Boundary singularities 96
Collocation method 9
Dislocations 370
Elasticity problems
axisymmetric problems 221
body forces 239
thermal loading 242
three dimensional problems 218
two dimensional problems 209
Elastodynamics 276
Elastoplasticity 413
438
Elastostatics problems 177
Flow problems 86
Functional analysis 2, 23
Functions, continuity 14
Fundamental solutions
choice of 327
for cracks 345
heat conduction 328
Laplace's equation 37
plane e1astostatics 335
plate bending 318
two dimensional elasticity 210
Geomechanics 397
Inverse problems 19
Postprocessing 160
Potential problems 85
axisymmetric 101, 105
three dimensional 101
two dimensional 85
Preprocessing 160
Somig1iana's formula 82
Variational formulation 28
Weak formulations 12
Weighted residual formulations 1, 7, 32