Yoshibash 1997 - Numerical Analysis of Edge Singularities

*Correspondence to: Zohar Yosibash, Ben-Gurion University of the Negev, Mechanical Engineering Department,
P.O. Box 653, Beer-Sheva 84105, Israel

CCC 00295981/97/24461122$17.50 Received 24 November 1996
1997 John Wiley & Sons, Ltd. Revised 27 April 1997
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING, VOL. 40, 46114632 (1997)
NUMERICAL ANALYSIS OF EDGE SINGULARITIES IN
THREE-DIMENSIONAL ELASTICITY
ZOHAR YOSIBASH*
Pearlstone Center for Aeronautical Engineering Studies, Department of Mechanical Engineering,
Ben-Gurion niversity of the Negev, Beer-Sheva 84105, Israel
ABSTRACT
A numerical method is described for the computation of eigenpairs which characterize the exact solution of
linear elastostatic problems in three-dimensions in the vicinity of edge singularities. These may be caused by
re-entrant corners, abrupt changes in boundary conditions or material properties.
Such singularities are of great interest from the point of view of failure initiation: The eigenpairs
characterize the straining modes and their amplitudes quantify the amount of energy residing in particular
straining modes. For this reason, failure theories directly or indirectly involve the eigenpairs and their
amplitudes.
This paper addresses the problem of determining the edge eigenpairs numerically on the basis of the
modied Steklov formulation (presented in Reference 1 in a 2-D framework) in conjunction with the
p-version of the nite element method. Numerical results are presented for several cases including isotropic
as well as anisotropic multi-material interfaces. 1997 John Wiley & Sons, Ltd.
Int. J. Numer. Meth. Engng., 40, 46114632 (1997)
No. of Figures: 13. No. of Tables: 5. No. of References: 24.
KEY WORDS: three-dimensional elasticity; singularities; nite element methods; p-version; Steklov method; fail-
ure/fracture analysis; multi-material interfaces; delamination
1. INTRODUCTION
Realistic mechanical response of structural objects undergoing small displacements is governed
by the equations of three-dimensional linear elasticity. Many three-dimensional objects have
surface boundaries which intersect at curves called edges, and some of these edges intersect at
points called vertices. The behaviour of the elastic solution in the neighbourhood of edges has
been a subject of intensive research during the past 30 years especially with regard to interface
edge cracks and interlaminar stresses in composite laminates. Renewed interest in the solution of
the three-dimensional linear elastic problem at edges occurred due to increasing interest in
laminated composites and electronic devices. In the vicinity of these edges the stress tensor
exhibits singular behaviour, i.e. tends to innity as the distance from the edge tends to zero. The
displacement solution (associated with the singular stress tensor) is uniquely characterized by
a sequence of discrete eigenpairs and their coecients (in the neighbourhood of edges). These are
NME 276
of great interest in structural mechanics because they provide a basis for predicting failure events
in the vicinity of edges.
Due to the complex treatment of three-dimensional edge eigenpairs, most of the research on
singular stress elds has focused on two-dimensional domains under the assumption of plane
stress or plane strain. The reader is referred to the list of publications" (not exhaustive by any
means), and the references therein which address the analytical as well as numerical computation
of eigen-pairs in two-dimensions.
The fully three-dimensional edge singularities have been less investigated, especially when
associated with anisotropic materials and multi-material interfaces. Analytical methods as in
Reference 15 provide the means for computing the eigenpairs for a two-material interface,
however, requires extensive mathematics. Several numerical methods, mainly based on the
h-version of the nite element method have been suggested lately. Among them are Reference 12
in which an excellent reference list to the subject is provided, and Reference 16. These methods
provide good results, however, they require the solution of a quadratic eigenproblem (which is an
expensive task), do not provide an easy adaptive scheme for assuring the convergence of the
computed values, and in Reference 16 the method presented does not provide information on the
performance for cases where complex eigenpairs appear. It is felt that the methods in References
12 and 16 fail to indicate these cases which give rise to logarithmic stress singularities.
Herein a numerical procedure based on the modied Steklov method, presented in Reference 1 in
a two-dimensional setting, is developed. This method computes eciently and reliably approxi-
mations for the edge singular solutions when used in conjunction with the p-version of the nite
element method. It is general, that is, applicable to edge singularities associated with corners,
non-isotropic multi-material interfaces and abrupt changes in boundary conditions.
In Section 2, an explicit formulation of the modied Steklov method for edge singularities is
presented. Numerical treatment of the weak modied Steklov formulation by the p-version of the
nite element method is presented in Section 3. In Section 4, numerical examples are provided.
These include edge crack singularities in isotropic and at a bi-material interface, and free edge
eects in a two cross-ply anisotropic laminate. The obtained eigenpairs are compared to the exact
values, demonstrating the eciency, accuracy and robustness of the method. We conclude with
a summary in Section 5.
2. FORMULATING THE MODIFIED STEKLOV EIGEN-PROBLEM
The elastostatic displacements eld in three dimensions, in the vicinity of an edge (which is
suciently away from a vertex) can be decomposed in terms of edge eigenpairs and edge stress
intensity functions (ESIFs). Mathematical details on the decomposition can be found e.g. in
References 1720 and the references therein. A representative three-dimensional domain denoted
by D, which contains typical 3-D singularities is shown in Figure 1. Edge singularities arise in the
neighbourhood of the edges A
GH
, vertex singularities arise in the neighbourhood of vertices A
G
, and
vertex-edge singularities arise close to vertex/edge intersections. Herein, only edge singularities
are treated, whereas the vertex singularities, which are also of major engineering importance, will
be addressed in a future work. It shall be assumed that curved edges which intersect at vertices do
not exist, and that crack faces, if any, lie in a at plane.
The edges A
GH
, connecting the vertices, A
G
and A
H
, are examined. Moving away from the vertex
a distance o/2 (o'0), and creating a cylindrical domain of radius r"R having the edge A
GH
as its
axis, we dene a sub-domain in the vicinity of the edge denoted by E
B 0
(A
GH
). Figure 2 shows the
NME 276
4612 Z. YOSIBASH
Int. J. Numer. Meth. Engng., 40, 46114632 (1997) 1997 John Wiley & Sons, Ltd.
Figure 1. Typical 3-D singularities
edge singularity sub-domain E
B 0
(A
`
). The displacements in E
B 0
can be decomposed as follows:
u(r, 0, z)"
)
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
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
where E is Youngs modulus and v is the Poisson ratio.

Multiplying (3) by v'"(v
V
v
W
v
X
)3[H(D
*
0
)]`, integrating over the domain D
*
0
shown in Figure
3, then using Greens theorem, and following the steps presented in Reference 21, Chapter 13, one
obtains the weak form for the elastostatic problem:
Seek u3[H(D
*
0
) ]` such that
(6)
B(u, v)"F(v), v3[H(D
*
0
) ]`
NME 276
where
B(u, v)

"
"

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
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
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
#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
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
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
) d, and the expression [A

F
]c
F
u` d0 becomes
[A
F
]c
F
u` d0"[A
F
]c
K
u` d (36)
On side 1 the basis and trial functions u
G
(, p"!1) are nothing more than integrals of Legendre
polynomials P
G
() for i'3, and P
()"(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
(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
so that the expression v` '[A

P
]' becomes
v` '[A
P
]'"b'[PI]' (38)
Substituting (37) and (38) into (30) we obtain the expression for N
0
(uJ, vJ)
N
0
(u` , v` )"b'
[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
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'
. By partitioning [K], we can write the

eigenproblem (44) in the form
[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
by static condensation, thus obtaining the reduced

eigenproblem
[K
1
]a
00
*
":[M
00
*
]a
00
*
(46)
where
[K
1
]"([K
00
*
]![N
00
*
])![K
00
*
][K
][K
00
*
]
It is possible to eliminate the unknowns a
from the matrix [K], because the relevant equations

do not involve the as yet unknown eigenvalues :.
For the solution of the eigenproblem (46), it is important to note that [K
1
] is, in general, a full
matrix. However, since the order of the matrices is relatively small, the solution (using Cholesky
factorization to compute [K
]) is not expensive. Details on the implementation issues for

solving the generalized eigenvalue problem can be found in Reference 23.
Remark 4. There is the possibility that an eigenvalue is repeated m times with less than
m corresponding eigenvectors (the algebraic multiplicity is higher than the geometric multipli-
city). This is associated with the special cases when the asymptotic expansion contains logarith-
mic terms, and this behavior triggers the existence of ln(r) terms.
Remark 5. Although we derived our matrices as if only one nite element exists along the
boundary I
`
and I
"
, the formulation for multiple nite elements is identical, and the matrices
[K], [N
0
] and [M
0
] are obtained by an assembly procedure.
4. NUMERICAL INVESTIGATION
The performance of the modied Steklov weak formulation is demonstrated in the following
subsections on several test cases for which the exact eigenvalues are known. We rst consider
a plane crack in an isotropic material, and at the interface of a bi-material interface, followed by
the computation of eigenpairs associated with free edge eects in two cross-ply anisotropic
laminate. These three examples demonstrate the robustness, eciency and accuracy of the
proposed numerical methods for extracting eigenpairs.
4.1. Plane crack in isotropic domains
Consider a plane crack in an isotropic material as shown in Figure 5. The material properties
are E"1 and v"03. The exact rst three eigenpairs for this example problem are
NME 276
Figure 6. Finite element mesh used for the plane crack in isotropic domain
Figure 5. A typical plane crack in isotropic domains
:
":
`
":
`
"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
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
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
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
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
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
Figure 13. Eigenfunctions associated with :
for the cross-ply anisotropic laminate

We rst investigate the eigenpairs associated with the singularities near the junction of the free
edge and the interface, edge A in Figure 10, for a commonly used [$[] angle-ply composite. Of
course, the eigenpairs depend on [ and we chose ["45 for which the rst 12 exact non-integer
eigenpairs are reported in [15] with 8 decimal signicant digits: :
"0974424342,
:
` `
"188147184$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 :
obtained at p"8 are

illustrated in Figure 13.
The variation of the eigenvalues for dierent [$[] cross-ply laminate is investigated in the
following. We use the same mesh presented in Figure 11, with dierent material properties which
reect a laminate with bers rotated at an angle $[ about the y-axis. The obtained eigen values
reported in Table IV are accurate up to the 5 decimal digit shown. This accuracy is guaranteed
because these digits have not been changed while increasing the polynomial level over the nite
element mesh.
Delamination. Once a crack starts propagating between two lamina, the behaviour of the
singular stress tensor changes dramatically, and the eigenpairs are dierent. We consider a plane-
crack with a tip at the edge denoted by edge B in Figure 10. Again we consider a [$45] cross-ply
NME 276
4630 Z. YOSIBASH
Table IV. Eigenvalues for dierent [$[] bre orientation cross-
ply laminate
[$[] 15 30 45 60 75
:
099936 098834 097442 097665 099105

Table V. First 3 eigen-values for a crack between two laminas in a cross-ply anisotropic laminate
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
R:''
`
0517095 0498988 0500189 04999305 0499995 0500005 0500002 0500000
Im:''
`
*
100 3472695 3464185 3518766 3467100 3444288 3437041 3435015 3434454
:''
`
0516157 0510878 0499144 0499133 0499847 0500039 0500020 0500001
laminate, and use a four-element mesh (as the one in Figure 6) for the computations. The rst
three eigenvalues and the number of DOFs at each p-level are summarized in Table V. Observing
the obtained results we may conclude with a high degree of condence that :
`
"0500000$
i00343 and :
`
"050000.
5. SUMMARY AND CONCLUSION
We have described a numerical method, based on the modied Steklov formulation, for the
reliable computation of eigenpairs resulting from three-dimensional edge singularities due to
reentrant-corners, abrupt changes in material properties or boundary conditions. The formula-
tion has been solved by means of the p-version of the nite element method, and is suitable for
implementation in nite element programs. Numerical experiments for isotropic as well as
anisotropic multi-material interfaces indicate that the computed values converge strongly, are
accurate and inexpensive from the points of view of human time needed for input data prepara-
tion, and required CPU time. As demonstrated, the convergence behaviour is non-monotonic,
however, using p-extensions excellent convergence rates were obtained.
The modied Steklov method also provides eigenpairs corresponding to innite strain energy
(i.e. :
G
(0), which may be required for extracting stress intensity functions by the dual singular
method. This is achieved because the singular point is excluded from the solution domain. This
also allows a smaller domain to be analyzed and higher eciency, due to the fact that radial
renements towards the singular edge are not required.
Although not demonstrated, the robustness of the method is maintained also when power-
logarithmic stress singularities arise: the two computed distinct but adjacent eigenvalues and their
corresponding eigenfunctions collapse into one as the number of degrees of freedom is increased,
indicating the presence of these kind of singularities.
Computations of edge eigenpairs is important because they provides a rigorous quantitative
basis for investigating failure events, such as delamination of composite materials, and failure in
electronic devices. Having obtained the eigenpairs, a research project is underway for computing
NME 276
the edge stress intensity functions by the complementary energy principle, as described in
Reference 24 in a two-dimensional setting.
REFERENCES
1. Z. Yosibash and B. A. Szabo , Numerical analysis of singularities in two-dimensions. Part 1: computation of
eigenpairs, Int. J. Numer. Meth. Engng., 38, 20552082 (1995).
2. M. L. Williams, Stress singularities resulting from various boundary conditions inangular corners of plates in
extension, rans. ASME J. Appl. Mech., 19, 526528 (1952).
3. D. B. Bogy and K. C. Wang, Stress singularities at interface corners in bounded dissimilar isotropic elastic materials,
Int. J. Solids Struct., 7, 9931005 (1971).
4. V. L. Hein and F. Erdogan, Stress singularities in a two-material wedge, Int. J. Fracture, 7, 317330 (1971).
5. J. P. Dempsey and G. B. Sinclair, On the stress singularities in the plane elasticity of the composite edge, J. Elasticity,
9, 373391 (1979).
6. J. P. Dempsey and G. B. Sinclair, On the singular behavior at the vertex of a bi-material wedge, J. Elasticity, 11,
317327 (1981).
7. X. Ying, A reliable root solver for automatic computation with application to stress analysis of a composite plane
wedige, Ph.D. hesis, Washington University, St. Louis, Missouri, USA, 1986.
8. T. C. T. Ting, Explicit solution and invariance of the singularities at an interface crack in anisotropic composities,
Int. J. Solids Struct., 22, 965983 (1986).
9. D. Leguillon and E. Sanchez-Palencia, Computation of Singular Solutions in Elliptic Problems and Elasticity, Wiley,
New York, NY, 1987.
10. R. Barsoum, Theoretical basis of the nite element iterative method for the eigenvalue problem in stationary cracks,
Int. J. Numer. Meth. Engng., 26, 531539 (1988).
11. M. Costabel and M. Dauge, Computation of corner singularities in linear elasticity, in M. Costabel, M. Dauge and
S. Nicaise (eds.), Boundary alue Problems and Integral Equations in Nonsmooth Domains, Marcel Dekker, New York,
1995, pp. 5968.
12. L. Gu and T. Belytschko, A numerical study of stress singularities in a two-material wedge, Int. J. Solids Struct., 31,
865889 (1994).
13. P. Papadakis and I. Babus\ka, A numerical procedure for the determination of certain quantities related to stress
intensity factors in two-dimensional elasticity, Comput. Methods Appl. Mech. Engng., 122, 6992 (1995).
14. S. S. Pageau, P. F. Joseph and S. B. Jr. Biggers, Finite element analysis of anisotropic materials with singular inplane
stress elds, Int. J. Solids Struct., 32, 571591 (1995).
15. S. S. Wang and I. Choi, Boundary layer eects in composite laminates: Part 1free edge stress singularities, rans.
ASME J. Appl. Mech., 49, 541548 (1982).
16. S. S. Pageau and S. B. Jr. Biggers, A nite element approach to three dimensional singular stress states in anisotropic
multi-material wedges and junctions, Int. J. Solids Struct., 33, 3347 (1996).
17. M. Dauge, Elliptic Boundary alue Problems in Corner DomainsSmoothness and Asymptotics of Solutions, Lecture
Notes in Mathematics vol. 1341, Springer, Heidelberg, 1988.
18. P. Grisvard, Singularities in Boundary alue Problems, Masson, France, 1992.
19. S. Nicaise and A.-M. Sa ndig, General interface problems I & II, Math. Methods Appl. Sci. 17, 395450 (1994).
20. B. Andersson, U. Falk, I. Babus\ka and T. Von-Petersdor, Reliable stress and fracture mechanics analysis of complex
components using a h-p version of FEM, Int. J. Numer. Meth. Engng., 38, 21352163 (1995).
21. B. A. Szabo and I. Babus\ka, Finite Element Analysis, Wiley, New York, 1991.
22. I. Babus\ka and A. K. Aziz, Survey lectures on the mathematical foundations of nite element method, in A. K. Aziz
(ed.), he Mathematical Foundations of the Finite Element Method with Applications to Partial Dierential Equations,
Academic Press, New York, NY, 1972, pp. 3343.
23. Z. Yosibash, Numerical analysis of singularities and rst derivatives for elliptic boundary value problems in
two-dimensions, D.Sc. hesis, Sever Institute of Technology, Washington University, St. Louis, Missouri, USA, 1994.
24. B. A. Szabo and Z. Yosibash, Numerical analysis of singularities in two-dimensions. Part 2: Computation of the
generalized ux/stress intensity factors, Int. J. Numer. Meth. Engng., 39, 409434 (1996).
.
NME 276
4632 Z. YOSIBASH

Yoshibash 1997 - Numerical Analysis of Edge Singularities

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Yoshibash 1997 - Numerical Analysis of Edge Singularities

Uploaded by

Copyright:

Available Formats

*Correspondence to: Zohar Yosibash, Ben-Gurion University of the Negev, Mechanical Engineering Department,

P.O. Box 653, Beer-Sheva 84105, Israel

where E is Youngs modulus and v is the Poisson ratio.

) d, and the expression [A

so that the expression v` '[A

. By partitioning [K], we can write the

by static condensation, thus obtaining the reduced

from the matrix [K], because the relevant equations

]) is not expensive. Details on the implementation issues for

for the cross-ply anisotropic laminate

obtained at p"8 are

099936 098834 097442 097665 099105

You might also like