Professional Documents
Culture Documents
I
1
Q"
a
IQ
(z) r?I (ln r)Qf
IQ
(0)#w(r, 0, z) (1)
where S*0 is an integer which is zero for most practical problems, except for special cases,
:
I>
*:
I
are called edge eigenvalues, a
IQ
(z) are analytic in z called ESIFs, and can become very
large as they approach one of the vertices, and f
IQ
(0) are analytic in 0, called eigenfunctions. The
vector function w(r, 0, z) belongs to the Sobolev space [HO(E)]`, where q depends on K, and can
be made as high as required providing that K is large enough. We shall assume that S"0,
therefore, (1) becomes
u(r, 0, z)"
)
I
a
I
(z)r?If
I
(0)#w(r, 0, z) (2)
u"(u
V
u
W
u
X
)' is the displacement vector, with u
V
(r, 0, z), u
W
(r, 0, z) and u
X
(r, 0, z) being
its components in the x, y and z directions, respectively. We denote the tractions on
the boundaries by T"(
V
X
)' and the Cartesian stress tensor by
+
, and in vector form
by "(o
V
o
W
o
X
t
VW
t
WX
t
VX
)'. In the vicinity of the edge we assume that no body forces are
present.
A sub-domain of E
B 0
is considered, which is bounded by the surfaces z"0 and z"A, and by
the radii r"R* and r"R. This domain, denoted by D
*
0
, is shown in Figure 3.
NME 276
NUMERICAL ANALYSIS OF EDGE SINGULARITIES 4613
1997 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng., 40, 46114632 (1997)
Figure 2. The edge neighborhood E
B 0
(A
`
)
Figure 3. The modied Steklov domain D
*
0
The elastostatic problem in 3-D can be cast in terms of stresses over the domain D
*
0
:
c
V
o
V
#c
W
t
VW
#c
X
t
VX
"0
c
V
t
VW
#c
W
o
W
#c
X
t
WX
"0
c
V
t
VX
#c
W
t
WX
#c
X
o
X
"0
in D
*
0
, where
c
V
"
c
cx
c
W
"
c
cy
c
X
"
c
cz
(3)
NME 276
4614 Z. YOSIBASH
Int. J. Numer. Meth. Engng., 40, 46114632 (1997) 1997 John Wiley & Sons, Ltd.
while on the boundaries of 0"0 and 0"c
GH
homogeneous traction boundary conditions are
assumed (T"0). This assumption is only for simplifying the presentation, and the presented
methods are equally applicable to homogeneous displacements as well as homogeneous mixed
boundary conditions. Boundary conditions on the other four surface areas are yet to be specied.
The stressdisplacements relationship through the constitutive material law (Hookes law) is
given by
"[E][D]u (4)
where [D] is the dierential operator:
[D]
"
c
V
0 0
0 c
W
0
0 0 c
X
c
W
c
V
0
0 c
X
c
W
c
X
0 c
V
"
cos 0 0 0
0 sin 0 0
0 0 0
sin 0 cos 0 0
0 0 sin 0
0 0 cos 0
c
P
#
!sin 0 0 0
0 cos 0 0
0 0 0
cos 0 !sin 0 0
0 0 cos 0
0 0 !sin 0
GFFFHFFFI
[A
P
]
GFFFFFHFFFFFI
[A
F
]
c
F
r
#
0 0 0
0 0 0
0 0 1
0 0 0
0 1 0
1 0 0
c
X
(5)
GFHFI
[A
X
]
and [E] is the 6;6 symmetric material matrix which is given for an isotropic material by
[E]"
E
(1#v) (1!2v)
(1!v) v v 0 0 0
(1!v) v 0 0 0
(1!v) 0 0 0
(1!2v)/2 0 0
(1!2v)/2 0
(1!2v)/2
"
0
*
0
S`
"
([D]v)'[E][D]ur d0 dr dz (7)
F(v)
"
cD*
0
(v)' T dA (8)
If homogeneous displacement boundary conditions are prescribed on the faces 0"0, c
`
, then
the weak form (6) remains unchanged except for the spaces in which u and v lie.
2.1. Notations and mathematical relationships
Some new notations and mathematical relationships are derived to simplify the manipulations
required for deriving the modied Steklov eigenformulation. In view of (2), we seek for functions
u in D
0
*
(respectively, v) which can be represented by pairs of : and f(0) as follows:
u
"A(z) r?
f
V
(0)
f
W
(0)
f
X
(0)
"A(z)r?f (0), v
"B(z) r?f(0) (9)
We also denote the in-plane variation of the displacements as follows:
u` (r, 0)
"u/A(z) v` (r, 0)
"v/B(z) (10)
Applying (5) (9) one obtains
[D]u"
A(z)
r
(:[A
P
]#[A
F
]c
F
) r?f#A' (z) [A
X
]r?f (11)
The relationship between tractions and the stress on a face having an outer normal vector n is
given by
T"
+
n"
n
V
0 0 n
W
0 n
X
0 n
W
0 n
V
n
X
0
0 0 n
X
0 n
W
n
V
(12)
GFFFFHFFFFI
[n]
Using (11) and (4) in conjunction with (12) one obtains
T"r?[n][E]
A(z)
r
(:[A
P
]f#[A
F
]f' )#A' (z) [A
X
]f
(13)
The relationship (13) can also be stated in terms of u` :
T"[n][E]
A(z)
r
(:[A
P
]u` #[A
F
]c
F
u` )#A' (z) [A
X
]u`
(14)
NME 276
4616 Z. YOSIBASH
Int. J. Numer. Meth. Engng., 40, 46114632 (1997) 1997 John Wiley & Sons, Ltd.
2.2. Modied Steklov eigenformulation
The right-hand side in (6) is being considered. F(v) is split into three parts; the rst part, F
, is
dened on the two straight faces of D
0
*
, z"0 where n'"(0 0!1), and z"A where
n'"(0 0 1):
F
(v)"
0
0
*
S`
"
(B(z)r?f ') r?[n][E]
A(z)
r
(:[A
P
]f#[A
F
]f' )#A' (z) [A
X
]f
X
#(B(z) r?f') r?[n][E]
A(z)
r
(:[A
P
]f#[A
F
]f' )#A' (z) [A
X
]f
X"
r dr d0.
Notice that on the face z"0, [n]"![A
X
]', and on the face z"A, [n]"[A
X
]', then F
be-
comes
F
(v)"A(z)B(z)
X
X"
r`?>
2:#1
P0
P0
*
S`
"
f'[A
X
]'[E](:[A
P
]f#[A
F
]f' ) d0 (15)
#A' (z)B(z)
X
X"
r`?>`
2:#2
P0
P0
*
S`
"
f'[A
X
]'[E] [A
X
]f d0 (16)
The second part, F
`
, is dened on the two cylindrical faces of D
0
*
, r"R where n'"(cos 0
sin 0 0), and r"R* where n'"!(cos 0 sin 0 0):
F
`
(v)"
"
S`
"
B(z) v` '[n][E]
A(z)
r
(:[A
P
]u` #[A
F
]c
F
u` )#A' (z)[A
X
]u`
P0
#B(z)v` '[n][E]
A(z)
r
(:[A
P
]u` #[A
F
]c
F
u` )#A' (z)[A
X
]u`
P0
*
dz d0
On the cylindrical face r"R*, [n]"![A
P
]', and on the face r"R, [n]"[A
P
]', then F
`
(v)
becomes
F
`
(v)"
"
A(z) B(z) dz
:
S`
"
v` '[A
P
]'[E] [A
P
]u`
P0
P0
*
d0 (17)
#
S`
"
v` '[A
P
]'[E][A
F
]c
F
u`
P0
P0
*
d0
(18)
#
"
A' (z) B(z) dz
S`
"
rv` '[A
P
]'[E][A
X
]u`
P0
P0
*
d0 (19)
The third part of the linear formF, dened of the straight faces 0"0 and 0"c
`
, vanishes. This
is because we assume T"0 on these faces.
NME 276
NUMERICAL ANALYSIS OF EDGE SINGULARITIES 4617
1997 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng., 40, 46114632 (1997)
The bi-linear form in (6) is being considered. Substitute (11) into (7), we may split B(u, v) into
three parts B
"B
#B
`
#B
`
, given by
B
"
"
A(z)B(z) dz
0
0
*
S`
"
[A
P
]c
P
#[A
F
]
c
F
r
v`
'
[E]
[A
P
]c
P
#[A
F
]
c
F
r
u`
r dr d0
(20)
B
`
"
"
A(z) B' (z) dz
r`?>
2:#1
P0
P0
*
S`
"
f'[A
X
]'[E](:[A
P
]f#[A
F
]f' ) d0
#
"
A' (z) B(z) dz
r`?>
2:#1
P0
P0
*
S`
"
( f':[A
P
]'#f''[A
F
]')[E][A
X
]f d0 (21)
B
`
"
"
A' (z)B' (z) dz
0
0
*
S`
"
v` '[A
X
]'[E][A
X
]u` r dr d0 (22)
Examining the bi-linear form B
`
it can be shown that for any material matrix,
f'[A
X
]'[E](:[A
P
] f#[A
F
]f' )"( f':[A
P
]'#f''[A
F
]') [E][A
X
]f (23)
and integrating by parts one obtains
"
A(z) B' (z) dz#
"
A' (z)B(z) dz"A(z) B(z)
X
X"
(24)
As a result, substituting (23) and (24) into (21), B
`
becomes
B
`
"A(z)B(z)
X
X"
r`?>
2:#1
P0
P0
*
S`
"
f'[A
X
]'[E](:[A
P
]f#[A
F
]f' ) d0 (25)
We are now in the stage where we can gather all parts that contribute to the weak formulation.
Substituting (15)(22), into (6), and noting that (25) is identical to the rst part of F
given in (15)
one obtains
"
A(z)B(z) dz
0
0
*
S`
"
[A
P
]c
P
#[A
F
]
c
F
r
v`
'
[E]
[A
P
]c
P
#[A
F
]
c
F
r
u`
r d0 dr
#
"
A' (z)B' (z) dz
0
0
*
S`
"
v` ' [A
X
]'[E][A
X
]uJ r dr d0
"
"
A(z)B(z) dz
:
S`
"
v` '[A
P
]'[E][A
P
]u`
P0
P0
*
d0
#
S`
"
vJ'[A
P
]'[E][A
F
]c
F
u`
P0
P0
*
d0
#
NME 276
4618 Z. YOSIBASH
Int. J. Numer. Meth. Engng., 40, 46114632 (1997) 1997 John Wiley & Sons, Ltd.
#A' (z)B(z)
X
X"
0
0
*
S`
"
vJ'[A
X
]'[E][A
X
]u` r d0 dr
#
"
A' (z)B(z) dz
S`
"
rv` '[A
P
]'[E][A
X
]u`
P0
P0
*
d0 (26)
Rearranging (26), using integration by parts, and dividing by j
"
A(z) B(z) dz, (26) becomes
0
0
*
S`
"
[A
P
]c
P
#[A
F
]
c
F
r
v`
'
[E]
[A
P
]c
P
#[A
F
]
c
F
r
u`
r d0 dr
!
S`
"
v` '[A
P
]'[E][A
F
]c
F
u`
P0
P0
*
d0
":
S`
"
v` '[A
P
]'[E][A
P
]u`
P0
P0
*
d0
#
"
A'' (z)B(z) dz
"
A(z)B(z) dz
0
0
*
S`
"
v` '[A
X
]'[E][A
X
]uJ r d0 dr
#
"
A' (z)B(z) dz
"
A(z) B(z) dz
S`
"
rv` '[A
P
]'[E][A
X
]u`
P0
P0
*
d0 (27)
The obtained formulation is an eigen-problem cast in the weak sense. However, the edge eigen
pairs are independent of the coordinate z, i.e. are independent of the edge stress intensity factors
A(z) or their derivatives. Thus, the two last expressions in (27) must vanish, obtaining the
Modied Steklov formulation for the edge eigenpairs
Seek :3C, 0Ou` 3[H(D
*
0
) ]`, such that, v` 3[H(D
*
0
)]`
BI (u` , v` )![N
0
(u` , v` )!N
0
*
(u` , v` ) ]":[M
0
(u` , v` )!M
0
*
(u` , v` ) ] (28)
D
*
0
is the two-dimensional domain which is the at surface bounded by 0)0)c
`
and
R*)r)R and
BI (u` , v` )
"
0
0
*
S`
"
[A
P
]c
P
#[A
F
]
c
F
r
v`
'
[E]
[A
P
]c
P
#[A
F
]
c
F
r
u`
r d0 dr (29)
N
0
(u` , v` )
"
S`
"
v` '[A
P
]'[E][A
F
]c
F
u`
P0
d0 (30)
M
0
(u` , v` )
"
S`
"
v` '[A
P
]'[E][A
P
]u`
P0
d0 (31)
NME 276
NUMERICAL ANALYSIS OF EDGE SINGULARITIES 4619
1997 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng., 40, 46114632 (1997)
Figure 4. A typical nite element mesh in the domain D
*
0
Remark 1. Although the test and trial functions have three components, the domain over
which the weak eigenformulation is dened is two-dimensional, and excludes any singular points.
Therefore the application of the p-version of the FEM for solving (28) is expected to be very
ecient.
Remark 2. The bilinar forms N
0
and N
0
*
are non-symmetric with respect to u` and v` , thus the
problem formulated in the weak sense loses its self-adjoint property. As a consequence, the
minimax principle does not hold, therefore the approximated eigenvalues (obtained by a series of
hierarchical family of trial functions) cannot be considered as an upper bound of the exact ones
and the monotonic behaviour of the error is lost as well. Nevertheless, convergence is assured
(with a very high rate as will be shown by the numerical examples) under a general proof provided
in Reference 22.
Remark 3. Note that the weak eigenformulation (28) is not limiting the domain D
*
0
to be
isotropic, and in fact (28) can be applied to multi-material anisotropic interface, as will be
demonstrated by numerical examples.
3. NUMERICAL TREATMENT BY THE FINITE ELEMENT METHOD
In the following, the weak eigenformulation (28) is discretized by considering a nite dimensional
sub-space of [H(D
*
0
)]`, employing the p-version of the nite element method.
Assume that the domain D
*
0
consists of three dierent materials as shown in Figure 4.
We divided D
*
0
into, let us say 3 nite elements, through a meshing process. Let us consider
a typical element, element number 1, shown in Figure 4, bounded by 0
)0)0
`
. A standard
element in the , p plane such that !1((1, !1(p(1 is considered, over which
the polynomial basis and trial functions are dened. These standard elements are then mapped
by appropriate mapping functions onto the real elements (for details see Reference 21,
Chapters, 5, 6). The functions uJ
V
, uJ
W
, uJ
X
are expressed in terms of the basis functions u
G
(, p) in the
NME 276
4620 Z. YOSIBASH
Int. J. Numer. Meth. Engng., 40, 46114632 (1997) 1997 John Wiley & Sons, Ltd.
standard plane:
uJ
V
(, p)"
,
G
a
G
u
G
(, p)
uJ
W
(, p)"
,
G
a
,>G
u
G
(, p)
uJ
X
(, p)"
,
G
a
`,>G
u
G
(, p)
(32)
or
u` "
. . . u
,
0 . . . 0 0 . . . 0
0 . . . 0 u
. . . u
,
0 . . . 0
0 . . . 0 0 . . . 0 u
. . . u
,
a
a
`,
"[u]a (33)
where a
G
are the amplitudes of the basis functions, and u
G
are products of integrals of Legendre
polynomials in and p. u` and v` lie in the same space, therefore, we dene similarly v`
"[u]b.
Using (33) the unconstrained stiness matrix corresponding to BI (u` , v` ) on the typical element is
given by
[K]
"
0
0
*
F`
F
[A
P
]c
P
#[A
F
]
c
F
r
[u]
'
[E]
[A
P
]c
P
#[A
F
]
c
F
r
[u]
r d0 dr (34)
We concentrate our discussion now on N(u` , v` ). We start by evaluating the required expressions
involved in the computation. The mapping of , p"!1 (side 1 of the standard element) onto
I
`
is given by the following:
0"
0
`
!0
2
#
0
`
#0
2
(35)
so that d0"
`
(0
`
!0
()"(1!)/2, P
`
()"(1#)/2 (see, for details, Reference
21, Chapters 3, 6). By dening the following matrix:
[cP]
"
!P'
sin 0 . . . !P'
,
sin 0 0 . . . 0 0 . . . 0
0 . . . 0 P'
cos 0 . . . P'
,
cos 0 0 . . . 0
0 . . . 0 0 . . . 0 0 . . . 0
P'
cos 0 . . . P'
,
cos 0 !P'
sin 0 . . . !P'
,
sin 0 0 . . . 0
0 . . . 0 0 . . . 0 P'
cos 0 . . . P'
,
cos 0
0 . . . 0 0 . . . 0 !P'
sin 0 . . . !P'
,
sin 0
NME 276
NUMERICAL ANALYSIS OF EDGE SINGULARITIES 4621
1997 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng., 40, 46114632 (1997)
(36) becomes
[A
F
]c
K
u` d"[cP]a d (37)
We also dene the matrix
[PI]
"
cos 0 . . . P
,
cos 0 0 . . . 0 0 . . . 0
0 . . . 0 P
sin 0 . . . P
,
sin 0 0 . . . 0
0 . . . 0 0 . . . 0 0 . . . 0
P
sin 0 . . . P
,
sin 0 P
cos 0 . . . P
,
cos 0 0 . . . 0
0 . . . 0 0 . . . 0 P
sin 0 . . . P
,
sin 0
0 . . . 0 0 . . . 0 P
cos 0 . . . P
,
cos 0
[PI]'[E][cP]
E
d
a
"b'[N
0
]a (39)
The entries of [N
0
] are computed using the Gauss quadrature
(N
0
)
GH
"
+
K
'
J I
PI
JG
(
K
)E
JI
cP
IH
(
K
) (40)
where
K
and
K
are the weights and abscissas of the Gauss quadrature points, respectively.
Using the same arguments as above, the expression M
0
(u` , v` ) in (31) is evaluated
M
0
(u` , v` )"b'
0
`
!0
[PI]'[E][PI ]
E
d
a
"b'[M
0
]a (41)
and
(M
0
)
GH
"
0
`
!0
2
+
K
'
J I
PI
JG
(
K
)E
JI
PI
IH
(
K
) (42)
The matrices [N
0
*
] and [M
0
*
] have same values as those of [N
0
] and [M
0
], but of opposite sign.
This is because the shape functions on the artical boundaries I
`
and I
"
are the same (except for
some sign changes), and so is the mapping to the standard plane. Denoting the set of amplitudes
of the basis functions associated with the articial boundary I
`
by a
0
, and those associated with
the articial boundary I
"
by a
0
*
, the eigenpairs can be obtained by solving the generalized matrix
eigen problem:
[K]a!( [N
0
]a
0
![N
0
*
]a
0
*
)":([M
0
]a
0
![M
0
*
]a
0
*
) (43)
NME 276
4622 Z. YOSIBASH
Int. J. Numer. Meth. Engng., 40, 46114632 (1997) 1997 John Wiley & Sons, Ltd.
Augmenting the coecients of the basis functions associated with I
`
with those associated with
I
"
, and denoting them by the vector a
00
*
, (43) becomes
[K]a![N
00
*
]a
00
*
":[M
00
*
]a
00
*
(44)
We assemble the left-hand side of (44). The vector which represents the total number of nodal
values in D
*
0
may be divided into two vectors such that one contains the coecients a
00
*, and the
other contains the remaining coecients: a'"a'
00
*
, a'
[K
00
*
]![N
00
*
]
[K
00
*
]
[K
00
*
]
[K
]
a
00
*
a
":
[M
00
*
]
[0]
[0]
[0]
a
00
*
a
(45)
The relation in (45) can be used to eliminate a
][K
][K
00
*
]
It is possible to eliminate the unknowns a
":
`
":
`
"05, and in linear elastic fracture mechanics terminology they are associated with
deformation modes I, II and III. This example problem has been used to demonstrate the
accuracy which is typically achieved with the proposed numerical method.
A four-element mesh has been used as shown in Figure 6, and over each element the
polynomial degree of the shape functions has been increased from 1 to 8. In all examples R"1.
As will be shown in the following the magnitude of R* has virtually no inuence on the accuracy
of the obtained eigenpairs, and as R* P1 the accuracy of the results slightly improves, therefore
the value R*"099 is used in all computations. The relative error (%) in the ith eigenvalue is
dened by
e
?G
(%)"100
:''
G
!:
G
:
G
(47)
NME 276
4624 Z. YOSIBASH
Int. J. Numer. Meth. Engng., 40, 46114632 (1997) 1997 John Wiley & Sons, Ltd.
R Computations performed on a SGI Indigo` machine, with a R4400 200 MHz processor, Specfp92"131
Figure 7. Convergence of approximated eigenvalues for plane crack edge in isotropic domain
Table I. Relative error (%) in the rst 3 eigenvaluesPlane crack edge in isotropic domain
p-level p"1 p"2 p"3 p"4 p"5 p"6 p"7 p"8
DOF 30 69 108 159 222 297 384 483
CPU (s) 017 048 24 22 39 64 108 164
e-val C
e
?
006 !0199 !0004 !000007 !00000030 00000011 !0000000019 !0000000039
e
?`
197 !00977 !000078 !0000016 00000021 !000000057 !0000000031 0000000032
e
?`
257 00256 00001 000000028 !00000000015 00000000097 0000000005 !00000000035
We summarize the relative error (%) in the rst 3 eigenvalues, the number of degrees of freedom
before performing static condensation, and the CPU elapsed timeR required for the computation
in Table I. The rate of convergence of the eigenvalues (reported in Table I) is clearly visible when
plotted on a loglog scale as shown in Figure 7. This shows the rapid but non-monotonic
convergence rate as expected. To demonstrate the inuence of R* on the obtained results, Figure
8 presents the absolute relative error (%) in the rst three eigenvalues at p"8 as R* varies from
05 to 099.
4.2. Plane crack at bi-material interface
Consider a bi-material interface which is composed of two homogeneous materials, with
continuity of tractions and displacements across interface maintained. The two materials are
NME 276
NUMERICAL ANALYSIS OF EDGE SINGULARITIES 4625
1997 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng., 40, 46114632 (1997)
Figure 8. R* inuence on eigenvalues accuracy (plane crack edge in isotropic domain)
Figure 9. Plane crack at bi-material interface
isotropic, both having Poisson ratio 03, the upper material having E"10 and the lower E"1.
We are interested in plane cracks at the interface of the two materials as shown in Figure 9. This
example problem has been chosen to demonstrate the methods performance for cases where
complex eigenpairs arise. The exact rst three eigenpairs for this example problem are
:
`
"05$i007581177769 and :
`
"05. In linear elastic fracture mechanics terminology :
and :
`
are associated with deformation in the xy plane (where mode I and mode II are coupled
in this case), and :
`
is the out-of-plane mode.
NME 276
4626 Z. YOSIBASH
Int. J. Numer. Meth. Engng., 40, 46114632 (1997) 1997 John Wiley & Sons, Ltd.
Figure 10. Cross-ply anisotropic laminate
Table II. Relative error (%) in the rst eigenvaluesplane crack edge at bi-material interface
p-level p"1 p"2 p"3 p"4 p"5 p"6 p"7 p"8
e-val C
e
R
? `
103 !0196 !0011 !000050 !0000014 52e!11 !00000000028 !00000000025
e
I
? `
!95 !155 !1118 !00425 00010 !0000017 0000000013 0000000079
e
?`
259 00256 00001 000000028 !00000000015 00000000097 0000000005 !00000000035
The same four-element mesh as shown in Figure 6 has been used for the computations. The
relative error (%) in the rst two eigenvalues is split in two: one dening the relative error in the
real part and being denoted by e
R
?`
and the other dening the relative error in the imaginary
part e
I
? `
e
R
? `
(%)"100
R:''
!R:
R:
, e
I
? `
(%)"100
Im:''
!Im:
Im:
(48)
Since we are using exactly the same mesh as in the previous example problem, the number of
degrees of freedomand the CPU elapsed time for computations remain as reported in Table I. We
summarize the relative error (%) in the rst 3 eigenvalues in Table II. The method provides
excellent results for complex eigenpairs as well, with no deterioration in the performance.
4.3. Two cross-ply anisotropic laminate
We study edge singularities associated with a two cross-ply anisotropic laminate. Consider
a composite laminate with ply properties typical of a high modulus graphiteepoxy system, as
NME 276
NUMERICAL ANALYSIS OF EDGE SINGULARITIES 4627
1997 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng., 40, 46114632 (1997)
Figure 11. Finite element mesh used for the cross-ply anisotropic laminate
Table III. Relative error (%) in the rst 5 non-integer eigenvaluesCross-ply anisotropic laminate
p-level p"1 p"2 p"3 p"4 p"5 p"6 p"7 p"8
DOF 18 39 60 87 120 159 204 255
CPU (s) 012 023 048 092 154 241 419 608
e-val C
e
?
!75 !044 !038 0056 !0018 !00067 !0000825 !000006
e
R
?` `
!304 !1521 !0298 !0228 !0053 !00064 !00022 !000033
e
I
?` `
** ** !1571 !3196 0112 0106 00077 !000004
e
R
?" `
!775 611 405 0315 !0187 !0122 !0013 !00034
e
I
?" `
** ** !1852 !1143 !113 !0041 0073 00135
**No imaginary part
shown in Figure 10. The orientation of bres diers from layer to layer. Referring to the principle
direction of the bres, we dene
E
*
"20;10' psi E
2
"E
X
"2.1;10' psi
G
*2
"G
*X
"G
2X
"085;10' psi v
*2
"v
*X
"v
2X
"021
where the subscripts , , z refer to bre, transverse and thickness directions of an individual ply,
respectively. The material matrix [E] for a ply with bres orientation rotated by an angle [ about
the y-axis is given by
[E]"[([) ]'[E
"
][([) ]
NME 276
4628 Z. YOSIBASH
Int. J. Numer. Meth. Engng., 40, 46114632 (1997) 1997 John Wiley & Sons, Ltd.
Figure 12. Convergence of approximated eigenvalues for the cross-ply anisotropic laminate
where
[([)]"
s` 0 c` 0 0 c s
0 1 0 0 0 0
c` 0 s` 0 0 !c s
0 0 0 s c 0
0 0 0 !c s 0
!2c s 0 2c s 0 0 s`!c`
, c
"cos([), s
"sin([)
[E
"
]"
(1!v
2X
v
X2
) E
*
(v
*2
#v
*X
v
X2
) E
2
(v
X*
#v
X2
v
2*
)E
*
0 0 0
(1!v
*X
v
X*
)E
*
(v
X2
#v
*2
v
X*
) E
2
0 0 0
(1!v
*2
v
2*
)E
X
0 0 0
G
2X
0 0
G
*X
0
G
*2
"(1!v
*2
v
2*
!v
2X
v
X2
!v
*X
v
X*
!2v
*2
v
2X
v
X*
)
v
2*
"v
*2
E
2
E
*
, v
X2
"v
2X
E
X
E
2
, v
X*
"v
*X
E
X
E
*
NME 276
NUMERICAL ANALYSIS OF EDGE SINGULARITIES 4629
1997 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng., 40, 46114632 (1997)
Figure 13. Eigenfunctions associated with :
"0974424342,
:
` `
"188147184$i023400947, :
" `
"25115263$i079281732, . . . .
We use in our computation the two-element mesh shown in Figure 11. We summarize the
relative error (%) in the rst 5 non-integer eigenvalues, the number of degrees of freedom before
performing static condensation, and the CPU elapsed time required for the computation in Table
III. The rate of convergence of the eigenvalues (reported in Table III) is clearly visible when
plotted on a loglog scale as shown in Figure 12. Again one obtains a rapid convergence rate. The
three-dimensional eigenfunctions (displacement elds) associated with :