0045-7Y49/89 53.00 + 0.

00
0 I988 Pergamon Press plc

A FINITE ELEMENT ITERATIVE APPROACH FOR

ANALYSIS OF LAMINATED COMPOSITE
STRUCTURAL ELEMENTS
W. C. HWANG~ and C. T. SUNI
tChung-Shan Institute of Science and Technology, Taiwan, Republic of China
JCollege of Engineering, University of Florida, Gainesville, FL 32611, U.S.A.

Abstract-A continuous mixed field iterative scheme based on a three-dimensional finite element
displacement method is presented. This method is very powerful in the determination of stress
distributions for problems with either material and/or geometric discontinuities. For laminated composite
materials this method is reliable in stress evaluation at locations away from the optimal Gauss points such
as free edges near a notch or a hole. It is also useful in the determination of interlaminar stresses at an
interface of laminated composites.
Numerical results of two problems are presented. The first problem is to evaluate interlaminar stresses
of laminated composite beams and plates under in-plane loads. The second problem is to obtain stress
distributions (also stress concentration factor) of finite unidirectional composite plates with a circular hole
at the center of the plate.
The proposed method needs lower computer storage and less CPU time than most of the other schemes.
Ease of implementation is another advantage of this method. Since there are no changes to the finite
element solutions, the iterative part can be implemented as a post-processing subroutine in any existing
program for anisotropic materials without altering the main body of the code.

1. INTRODUXION technique has been shown to yield reasonable (and

perhaps less accurate) results in some cases for a
For more than a decade, the finite element method straight free edge [l], its usefulness for curved edges
has been a powerful technique in stress and displace- has not been shown [4].
ment analysis of aerospace structures. A detailed Furthermore, in a region of high stress concentra-
literature review in the development and application tion, this region still needs to be modeled by ordinary
of finite element methods in structural analysis is elements [3].
beyond the scope of this paper. In the following a For the bending analysis of isotropic and aniso-
brief review in the state-of-the-art of using three- tropic plates, a variety of analytical and discrete
dimensional finite element methods in stress analysis mathematical models, based on different theories,
of composite materials will be presented. have been proposed, for example by Mindlin [5] and
The finite element analysis of complex three- by Yang et al. [6].
dimensional multilayer composite laminates under There have been relatively few analyses of plates
multiaxial and flexural loading conditions is difficult. based on three-dimensional elasticity theory. This is
In general, each ply must be modeled separately in particularly true for laminated composite plates.
order to obtain the correct values for the peel (inter- Pagan0 [7] presented an analytical three-dimensional
laminar normal) and interlaminar shear stresses at finite element model for the analysis of simply
the interface. The literature contains little informa- supported laminated plates. Three-dimensional finite
tion on the use of three-dimensional finite element element models for analysis of laminated composites
models, based on three-dimensional elasticity, for the have been reported [8,9]. The 24-node isoparametric
analysis of laminated composites. The reason is the brick element [8], which has cubic displacement
large number of degrees of freedom (dof) pertaining patterns in the x and y directions and a linear
to the formulation scheme. Wang and Crossman [l] variation in z direction, has been widely used in static
reported a super-element scheme (by substructuring), analysis. It is costly to use this element because the
which is based on the classical lamination theory sizes and band-widths of the matrices which result
(CLT). Teh [2] and Jones et al. [3] extended this from the modeling of a realistic fiber composite
scheme to three-dimensional elasticity theory. This structure are large. Putcha and Reddy [9] stated that
scheme can reduce the number of elements required an accurate prediction of the interlaminar stresses is
by “smearing” two or more laminae into a super- possible only when a three-dimensional theory is
element through its thickness. Hence, it reduces the used. They used trilinear eight-node isoparametric
number of elements required. However, one should elements to analyse layered composite plates. Nu-
not smear two adjacent layers if the stresses on a merical results are compared with the exact solution
particular interface are to be calculated. While this obtained by Pagan0 [7]. It converged toward the

55
56 W. C. HWANGand C. T. SUN

exact solution as the mesh was refined. However, above observation, a continuous nodal mixed field is
further mesh refinements were not done due to the defined, represented by
limitations on computational resources. The use of
three-dimensional finite element models, based on the {a,> = {err%I 00 rs,r 7m rd. (3)
three-dimensional elasticity theory for the analysis of
layered composite laminates, needs further study. The nodal mixed field vector (a,} at each node is
In order to be able to handle any type of boundary obtained by averaging the mixed stress contribution
geometry, the aim of this paper is to develop a of all elements connected to that node. Therefore,
three-dimensional, finite element iterative approach interface mixed field continuity can be satisfied in
based on a continuous mixed field to analyse multi- multilayer laminates. Let {a$} be the average nodal
layer composite laminates. mixed field vector in the reference axes. Within an
element, we choose different shape functions for (6~)
2. ITERATIVEFINITE ELEMENT SCHEME
and {u}

2.1. Finite element formulation {SR} = [N]{6U’}, {u} = [N*]{ V} (4)

A three-dimensional body v bounded by surface s
is loaded by body forcesfon v, concentrated forces where {v’} are the nodal displacements, [N] is the
F on m boundary nodes and surface tractions p on classical shape function matrix, and [m*] are selected
surface s,. The principle of virtual displacement is so that mixed field components are continuous on
stated as a sum of integrations over the volumes v, element boundaries.
and areas S, of all finite elements; i.e. [lo] We substitute (2) and (4) into (1), and by applying
the constitutive equations, we obtain

+c @u)?~Pj ds,+ 5 @U%?F)i (1) For an arbitrary {SUe}T, whence we obtain

.5 Sk i=l

where (a}, {6u}, and (6U’) are the vectors of virtual [xl(U) = {R) (5)
strains, displacements, and nodal displacements, re-
where
spectively. The strains, {e}, within an element can be
expressed as a function of a displacement vector {a},
by using the strain-displacement relation,

{e> = [Lliu) (2) and

where [L] is the strain-displacement matrix.

Most commercial finite element packages are coded {R) =x(1 [Nl~~}dv.+~~,~[NI1PldS.)+{P)
e &
by using the standard displacement solution method.
This method gives good answers for displacements (7)
but poor answers for stresses. For an accurate stress
analysis of multilayer laminates, we make a new and where [d] is the matrix of elastic properties in the
formulation. If the orthogonal curvilinear coordi- constitutive law. Our problem now is to solve (5).
nates are r, s, and t, then the state of stress and strain
2.2. Iterative procedures
vectors of a solid is {Q,,u,, u,, r,,, rt,, r,] and
(e,, er, e,, yS,,y,,, y,}, respectively. Assume that a coor- Equation (5) can be solved iteratively by writing it
dinate surface t = constant separates two different as
materials. The {e,} is a unit vector that is normal to
the interface between the materials. In addition to the
displacement continuity conditions at interlayer
boundaries, we recall that the interface traction con- or
tinuity between adjacent layers must be satisfied in
hybrid stress formulation. Therefore, interlayer conti- [J4{AU}i= {R}- {R} (8)
nuity conditions on stress components el, T,,, and t,,
are imposed at the interface. Moreover, we propose where {AU}, = {U}i+I - {U}, (7if = [K]{ U}, and
another interface inplane strain continuity; strain [K] is the classical system stiffness matrix. If after the
components c,, es, and Y,~ are continuous at the ith iteration {U}j represents the exact solution of (5)
interface too, but the stresses cr,, u,, and T,~ may not then in (8) {AU}i must vanish, i.e. {R}i= {R}. {R},
be continuous across the interface. In view of the represents the load vector at the ith iteration. This
 Analysis of laminated composite structural elements

vector {Rji can be identified as the nodal force vector

by summing over all elements at the iteration i. Let “I”
a
fY
c
V¶O W
fQ ), be such a vector. From the element stress-strain wro WX
relations we can compute {6*} and (Z*) from {@gf t
and let (a) and {c) be interpolated from the {a*} or
{e*} with a modified shape function [m which may
or may not be the same as [N]. Thus

{a) = [Rj{B*}, {El = [RI@*& (9) L t .I

L

Next, we define the nodal force vector at the iteration Fig. 1. Cantilever laminated composite beam.
i, i.e.

iterative process repeats until convergence. From (15)

the convergence will occur when (R) - (Q>,._ I = 0.
In eqns (14) or (15), [K] is formed and constructed
only once. Thus, the time and effort of reconstruction
= T ~eWVW due (10) and triangularization of the stiffness matrix after each
iteration can be saved. Normally, the iterative scheme
converges very quickly in the first few iterations. The
where [B] = [L][N] is the usual straindisplacement
criterion of convergence for Euclidian norms of
matrix. If at the iteration i the exact solution for (6)
relative values is achieved if
and {r) is obtained, {Q}, should balance the entire
load vector; otherwise (R) - (Q), is a load imbalance
indicator. Therefore
(16)
W]W% = @) - (Q>P 01)
where e,,” is the specified displacement convergence
Comparing (8) and (1 1), we see that (R > is equivalent tolerance.
to {Q}i as the starting approximation,

3. STRESS ANALYSIS OF LAMINATED COMPOSITE

{U), = PT’IRI (12)
BEAM AND PLATE

and then we determine {c) and {af from element In this section the interlaminar stresses and the
nodal displacements {UC), which come from (VI,. in-plane stresses of laminated plates will be evaluated
using the iterative finite element scheme as described
163 = Pl(V, ~~~[~]~~I~~~. in Sec. 2. Numerical examples of laminated com-
posite beams under multiaxial loading (see Fig. 1)
The first load imbalance indicator is {R) - (Qf,. and simply-supported laminated plates (Fig. 2) under
Then, the new displacement vector is sinusoidal loading are presented to illustrate the

1% = (Vi + {Au),

= VW1C8R1- IQ},, (13)

etc.
Hence, the general formula is

(ut, = Kl-‘(i(R) - fQh - IQ12-. *’ - tQ),-,I

= [K]-‘(Rjj (14)

or

{rr>i=lu>i-,+[lul-‘(fR}-{Q},=,, (15)

where i is the iteration number; (i.e., the ith iteration)

the new load vector {R}, is considered as the ith
different load case in the equilibrium equation. The Fig. 2. Simply-supported O”/900/00square laminated plate.

C.A.S. 3,/i--~
58 W. C. HWANG and C. T. SUN

proposed scheme. In this problem, the lamina proper- element is used to model the present problems in
ties are assumed to be testing the convergence. The free-edge effect is not
considered here but is emphasized in Sec. 4. The
results will be compared to those obtained by using
E, = 172.5 Gpa (25 x lo6 lb/in2)
the 20-node isoparametric element and the CLT.
E2 = E, = 6.9 Gpa ( lo6 lb/in2)
3.1. Laminated composite beams under multiaxial
loading
G,2 = G1, = 3.45 Gpa (0.5 x lo6 lb/in2)
A symmetric three-layered cantilever cross-ply
G2, = 1.4 Gpa (0.2 x lo6 lb/in2) (O/90/0) solid beam subjected to a multiaxial loading
appears in Fig. 1. The external loads at the free end
vn = v,~ = vZJ= 0.25 are u = 6.9 Kpa (1 lb/in2), P = 0.7 N/mm (4 lb/in),
(17)
and M = 2260 N-mm (20 lb-in). The laminates are
constructed with width W of 25.4 mm (1 in), length L
where E,, E,, and E, are Young’s moduli in 1, 2, and of 508 mm (20 in.), and a total thickness h of 10.2 mm
3 directions respectively, vij is Poisson’s ratio for (0.4 in.) and 20.4 mm (0.8 in.), respectively. The hori-
transverse strains in the j direction, when stressed in zontal stresses Q, and the interlaminar shear stresses
the i direection, G,2, Gn, and G,, are shear moduli in T,, are examined at x = 0.375L, y = 0.5 W. The nor-
the l-2, l-3, and 2-3 planes, respectively. malized quantities are expressed as
Depending on the geometry and loading, these
problems may be represented as a beam, plate, or
tiX= a,/6 (18)
three-dimensional solid. Note that there is a wide
variety of beam and plate theories with different
f*, = T;,/U. (19)
levels of accuracy [1 1, 121. Many engineering lamin-
ates are thin and are adequately analysed by CLT,
which can provide reasonable predictions only for Comparisons between 6,, as determined by the itera-
relatively thin laminated plates having typical thick- tive mixed field method with the values as determined
ness to span ratio, h/L, as low as 0.05 [13]. The shear from the CLT and 20-node mesh without iteration,
deformation theory, which is an extension of the are shown in Table 1. The finite element models are
Mindlin plate theory, allows for transverse shear 8 x 1 x 1 mesh of eight-node elements in each layer
deformation [14]. It includes the assumption of a and 8 x 2 x 2 mesh of 20-node elements in each layer.
simple shear profile, or an artificial shear correction Figure 3 shows the results of interlaminar shear
factor, which requires sophisticated finite element stresses fZ,*.
analyses [14]. It can provide accurate predictions for Numerical convergence and accuracy characteris-
thickness to span ratio, h/L. = 0.1 or lower [13]. A tics show a distinct improvement of stress distribu-
more accurate representation is considered by using tions obtained by using the iterative mixed field
the three-dimensional element, since each ply must be method with low-order element meshes in laminated
modeled separately in order to obtain the correct beam problems.
values for the interlaminar shear stresses and peel
stress [3]. However, this results in a large numerical 3.2. Simply-supported laminated plate with sinusoidal
model. In order to reduce the total number of dof, loading
low-order elements are selected for use in the iterative In the problem of a composite square plate with
scheme. A trilinear eight-node isoparametric brick three-layered O/90/0 equal thickness lay up, fibers in

Table 1. Convergence study of o, at x = 0.37X, y = O.SW for cantilever O/90/0 beam with multiaxial
loading (8 x 1 x 1 mesh of &node elements in each layer)
h(mm) i z =h/2 z = h+/4 s = h-/4 .z =h-/4 .z = h-14 z = -h/2
10.2 1 1844 923 36.5 - 36.9 -919 - 1840
2 2547 1284 51.1 - 50.9 - 1280 -2543
3 2827 1434 56.9 -56.7 - 1430 - 2824
4 2923 1488 59.0 -58.9 - 1486 -2919
5 2944 1500 59.5 - 59.3 - 1498 - 2940
CLT 2988 1496 59.0 - 58.6 - 1487 - 2978
FEM* 2984 1494 58.5 -58.3 - 1490 -2980
20.4 1 647 325 12.8 - 12.6 -321 -643
2 733 370 14.5 - 14.4 -365 -729
3 744 376 14.9 - 14.6 -372 -740
CLT 748 375 14.8 - 14.6 -371 - 743
FEM* 747 375 14.6 - 14.5 -371 -744
*Results of 8 x 2 x 2 mesh of 20-node isoparametric elements in each layer without iteration.
 Analysis of laminated composite structural elements 59

symmetry considerations require only a quadrant of

- h = 1'0.16mm - the plate to be modeled. A normal traction
--0-h = 20.36 mm _

a, = q. sin z sin lflI (20)

a b

is applied on the upper surface, while the lower

surface is traction-free. In order to compare the exact
solutions obtained in Pagan0 [7], we prescribe the
boundary conditions as shown in eqn (18). The
nondimensional values of stresses are defined with
respect to the data, S = a/h, where h is total thickness
of the plate.

ITERATlOii NUMBER OF EIGHT-NODE MESH (6X:,,

5.VQj) = (Q,, GyprXJ/%~* (21)
Fig. 3. Convergence diagram of T._~for cantilever solid
laminated composite beam. (f&J = (7xn7&l& (22)

Results are compared to the exact solutions obtained

the outer layers of the laminate run parallel to by elasticity [7] in Figs 4 and 5 for a, and 7_.
the x-axis, while those in the middle layer run parallel Convergence characteristics are demonstrated by
to the y-axis. For the simply-supported plates, comparing results obtained using Mesh 1 (6 x 6 x 1

s
z.55
.%
g$
2 g.45 +
I
zg
9: 35 a: Full lntogntton +r,rl,ml
2&’ b: Raducod tntogrrtton Aa,O2,m2
01: ElghbNodr Ob,r2,m2
02: 2woda xa,o2,m3
ix ml: Moshl qb,rZ,m3
g a.25 mz: Yah2
2s m% Maah
5
715
0 2 4 B 8
ITERATION NUMBER
Fig. 4. Convergence diagrams of 6, for a simply-supported square laminated plate (S = 100).

+
;~.25~ I: Full lntogntion +a,al,ml
aC0 b: Roducrd tntogrrtion Ar,a2,m2
._ . 20 - al: Eight-No& ob,rl,ml
rZ: 20.Uodr xr,r2,m3
*b;.15- ml: M&l n b&m3
:A m2: MDsh2
m3: Mwh3
sky0 -

.05 ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ 1

ITERATION NUMBER
Fig. 5. Convergence diagrams of fzx for a simply-supported square laminated plate (S = 100).
60 W. C. HWANGand C. T. SUN

uniform mesh for eight-node elements in each layer), To carry out a solution for the stress concentration
Mesh 2 (4 x 4 x 1 uniform mesh of 20-node elements factors around the notch, four-node isoparametric
per layer) and Mesh 3 (3 x 3 x 1 uniform mesh for bilinear elements will be used. Linear element meshes
20-node elements per layer). were chosen after testing several different mesh sys-
tems. As prototypes, coarse meshes of Mesh 1 and
Mesh 2, in the iterative case (Fig. 7), are used.
4. STRESS CONCENTRATIONOF UNIDIRECTION
For comparison, finer meshes of Mesh 3 and Mesh
LAMINATED COMPOSITES
4 (Fig. 8) are also computed based on the classical
In this section the stress concentration of a fiber analysis (without iteration).
reinforced finite unidirectional composite plate with In general, the element aspect ratio S should not
a circular hole at the center is investigated. exceed roughly seven for good displacement results
The rectangular panel (Fig. 6) under uniform ten- and roughly three for good stress results in a classical
sion co in the x direction, with a traction-free center analysis. For an element of radial size AS, around the
hole, is used as a simple test model. Material proper- hole (see Fig. 7), a special aspect ratio S(k) is defined
ties used for the present analysis are T300/934 below,
graphite epoxy
S(k) = A&/AS,. (24)
E, = 138 Gpa (20 x lo6 lb/in*),
The stress concentration factor (SCF) in an infinite
E2 = E3 = 10.4 Gpa (1.5 x lo6 lb/in*), plate is theoretically obtained [15] and plotted in
Fig. 9. Numerical calculations are tested using
G,* = G,, = 6.6 Gpa (0.95 x lo6 lb/in*), different S(k) since the resulting SCF are dependent
on S(k) and meshes.
Cl3 = 2.6 Gpa (0.37 x lo6 lb/in*),

v,* = v,, = 0.21, v*j = 0.47 5. CONCLUDINGREMARKS AND DISCUSSION

(23)
From the numerical results in Sec. 3 one can
and for an isotropic case, E = 138 Gpa, v = 0.21. The conclude for laminated composites that Mesh 2 and
panel is 152.4mm x 152.4mm with a 10 mm diame- Mesh 3 with 20-node elements converge faster but
ter center hole. This can be considered as a “plane
stress” problem. Numerical results are obtained for
composites with fiber direction 0,= 0” and 90” and
for an isotropic material. Since the plate is symmetric
about the x- and y-axis, a quarter of the plate will be
utilized in the numerical model (Fig. 6).

09 Fiber Dlrectlon

w=F 4Y
2L

(b)
k&-l

Fig. 6. Geometry and boundary conditions for (a) a rectan-

gular plate with a center hole under tension, (b) numerical
model for composites with fiber direction O,= O”, 90” and Fig. 7. Finite element coarse meshes. (a) Mesh 1 (4 x 4). (b)
isotropic material. Mesh 2 (5 x 6 in I and 0 directions).
 Analysis of laminated composite structural elements 61

As is expected with the use of iterative schemes for

coarse mesh systems, the results are improved over
classical results. Due to the idealization error, which
is an error associated with modeling a curved
boundary by straight-sided elements, some results are
(a) a little larger or smaller than the theoretical values.
Mesh refinement around the curved edge would
expect SCF to converge to values that are exact for
the curved free edge. Hence, Mesh 2 is better than
Mesh 1, but it needs more dof (see Fig. 10). Even a
very fine mesh (Mesh 3 or Mesh 4) (see Fig. 11) needs
a good S(k) value to improve the results.
For an isotropic case, the very coarse mesh gives
excellent results with only one more (i = 2) iteration,
because SCF values are not large. All figures show a
very important result about the convergence rule of
this proposed scheme. When S(k) = 0.3-0.6, the con-
vergence of the SCF becomes faster. The same con-
(b) clusion is obtained from Mesh 5 (Fig. 12).
Tests of multilayered laminates with a center hole,
using three-dimensional elements, give similar results.
To avoid the idealization error in the classical finite
element method, a good aspect ratio is needed as in
low-order elements with finer meshes or high-order
elements with coarse meshes. When we use the high-
order element and select the modified shape func-
X
tions, which are the same as classical shape functions
Fig. 8. Finite element finer meshes (12 x 18). (a) Mesh 3. in the iterative case, the convergence rate becomes
(b) Mesh 4. slower. The results are not presented here. A better
modified shape function needs to be investigated
Mesh 1 with eight-node elements still converges
with lower speed. Convergence characteristics
OrthotropicPanel, Fiber Anglo = 90
are faster for thicker laminated plates than for thin-
ner laminated plates. For thin laminates, the assump-
tion of independent rotations in each ply is not
ii Xf---~go-
necessary. It is costly because element of grows with
the number of layers. Similar conclusions can be
drawn for stress concentration problems as shown in
Sec. 4. ~rkl.0
9 8
G-2.0
Analytlcal Sol. of an lnflnlte Plane zz
bv G -3.0
Sun et-al. 1986 =O’
I I I 1 I I I I -4.0
Orthotrplo A&&lATlO 1.0S (K,.l
1.4 TO
1.64) OF-MESH’
22 24 1
OrthotroploPmrl Flbrr Anglo = 90’bgg.

i: ::;y

.6 1.2 1.6 2.6 24

0 26 46 60 80 106 120 14g 166 180 ASPKik0 S (Kpt TO 4) OF MESH i
ANGLE OF THETA (DEGREES)
Fig. 10. Convergence diagram of SCF at 0 = 0” and 90” for
Fig. 9. The theoretical stress concentration factors. the orthotropic case (0,= 90”). (a) Mesh 1. (b) Mesh 2.
 62 W. C. HWANGand C. T. SUN

W r I8otroplo C888

ASPECTRAT10S&&2 TO 1.2)OF MESH 3 &4

ASPECT RATIO S&=02 TO 4) OF lvlE!SH5
Orthotropic Panel, Flbar Angk = 0
Fig. 12. SCF at 0 = 90” for the orthotropic case, 0,= 0”.

3. R. Jones, R. Callinan, K. K. Teh and K. C. Brown,

Analysis of multilayer laminates using three-dimen-
sional super-elements. Int. J. Numer. M%h. Engng. 20,
583-588 (1984).
4. W. M. L&king, S. V. Hoa and T. S. Sankar, The effect
of geometry on interlaminar stress of (0/9O)s composite
laminates with circular hole. J. Composite Mater. 17,
188-198 (1984).
R. D. Mindlin, Infiuence of rotatory inertia and shear
on flexual motions of isotropic, elastic plates. J. appl
Meek 18, 31-38. (1951).
Fig. 11. SCF for finer meshes (Mesh 3 and Mesh 4). (a) 6. P. C. Yang, C. H. Norris and Y. Stavsky, Elastic wave
Isotropic case. (b) Orthotropic case (@,= D”). propagation in heterogeneous plates. ht. J. Solids
Struet. 2, 665-684 (1966).
7. N. J. Pagano, Exact solution for rectangular bidirec-
tional composite and sandwich plates. J. Composite
when high-order elements are used. Fortunately, Mater. 4, 20-34 (1970).
the iterative results of a low-order element mesh *. R. M. Barker, F. T. Lin and J. R. Dana, Three
dimensional finite element analysis of laminated com-
that needs fewer dof converge very fast. The degree posite. Comput. Struct. 2, 1913-1929 (1972).
of accuracy which can be obtained is dependent 9. N. S. Putcha and J. N. Reddy, Three dimensional finite
_.
on the mesh refinement around the curved free element analysts of Iavered comoosite
. .olates. Advances
edge. in Aerospace Structures and Materials. ASME Winter
Annual Meeting, Phoenix, Arizona (1982).
In conclusion, based upon the numerical examples Io. K. J. Bathe, Finite Element Procedures in Engineering
presented in Sets 3 and 4, the proposed finite element Analysis. Prentice-Hall, New York (1982).
method seems to provide a more powerful technique 11. E. Hinton and D. R. J. Owen, An Introduction to Finite
for stress evaluation than other existing finite element Element Computations. Pineridge Press, Swansea (1979).
12. M. M. Hrabok and T. M. Hrudey, A review and
methods. In addition to the accurate results, time
catalogue of plate bending finite elements. Compur.
saving in computation is another advantage of using Struck 19, 475495 (1984).
this method. 13. R. L. Spilker and B. E. Engelmann, Hybrid stress
element for thin and thick mul&ayer laminated plates.
ht. J. Numer. Meth. Ennnn. 20. 573-587 (19861.
REFERENCES 14. J. N. Reddy, A simple higher-order theor; for’lamin-
ated composite plates. J. appl. Mech. 51, 745-752
I. A. S. D. Wang and F. W. Grossman, Calculation of (1984).
edge stresses in multilayer laminates by substructuring. 15. C. T. Sun and S. A. Khalil, Stress concentration factor
J. Composite Mater. 21, 76-83 (1979). and strength reduction factor of unidirectional com-
2. K. K. ?eh, A three-dimensional analysis of laminated posite weakened by an elliptical hole. Proc. int. Symp.
composite ptates using numerical methods. Ph.D. on Composite ~aterjals and Struciures, pp. 825-831.
Thesis, Melbourne University, Australia (1983). Beijing, China.