Professional Documents
Culture Documents
1994
Copyright Q 1994 Elswier Scii Ltd
Pergamon Printedin Great Britain.All rights rcwved
co45-7949/w 56.00 + 0.00
Alxstracf-A very simple way to obtain a quadrilateral shell element is to write membrane and bending
stiffness matrices for a flat geometry and add them. This basic membrane-plus-bending element can be
modified to incorporate membrane&ending coupling and to accommodate a geometry in which the four
nodes are not coplanar. The present paper develops a 24 degree of freedom quadrilateral shell element
and concurrently provides a status report on the formulation method by pointing out what works and
what does not. Numerical results are good: the element is not the best available four-node shell element
in all test cases but is far from the worst.
A finite element for shell analysis can be obtained by Local coordinate system
combining two flat elements, one for plane elasticity
and the other for bending of flat plates. Early triangu- The first step in element formulation is to place the
lar elements were of this type. They did not work element in a local coordinate system xyz that is
well in some problems because of poor membrane oriented with reference to the element and is defined
response and the absence of coupling between mem- by using the global XYZ coordinates of element
brane and bending actions. Also, ridge-like intersec- nodes (Fig. 1). Element properties are formulated in
tions were awkward to model because there were but the local coordinate system. The last step before
five degrees of freedom (DOF) per node. More assembly of elements is transformation of nodal DOF
recently, a membranebending coupling device has from local to global directions. Use of a local coordi-
been devised [l, 21 and the advent of ‘drilling’ ro- nate system is convenient and also guarantees that the
tational DOF [3,4] has lessened the ridge intersection assembled finite element structure will display geo-
difficulty. Unfortunately, drilling DOF make mem- metric invariance, regardless of the nature of the
brane locking more likely, i.e. the model may become element stiffness formulation.
excessively stiff because rotational DOF associated Details of how the local coordinate system are
with bending activate membrane strains as well as established are not important. The procedure used
bending strains. Locking, however, can usually be here is similar to that described in [7]. The result is a
reduced or even removed. As a result of all these local system xyz with xy the reference plane (Fig. 2).
efforts some very good triangular shell elements are If the element is flat, its local nodal coordinates zi
now available [2, 5,6]. are all zero. If the element is warped, the zi are
Similar procedures can be used to produce a alternately + H and -H units above and below the
four-node quadrilateral shell element having six DOF reference plane xy. The x and y axes are parallel to
per node. However, the quadrilateral presents some sides of a rectangular element, i.e. x and y coincide
difficulties that do not appear in the triangle, so that with lines connecting midpoints of opposite sides. If
the quadrilateral is not a direct extension of the the element is nonrectangular, x and y are approxi-
triangle. What follows is a report on the quadrilat- mately parallel to the lines connecting midpoints,
eral, with attention to procedures that succeed and with the longer of the two lines more heavily
those that fail. The principal shortcoming of the weighted.
quadrilateral appears to be a tendency toward mem- Except for membranebending coupling terms and
brane locking, which however is not great, so that a the correction for a warped element, stiffness matrix
good element is nevertheless produced. formulation proceeds as if the element were flat, using
549
550 R. D. COOK
(3)
Control of a mechanism
+ 6,~ Ns - x23 Ns F’z, + 623 h’s Like all other elements having drilling DOF, a
quadrilateral element has the instability in which all
- XY NT N3 + h N, - x41 % Pz4 (lb) drilling DOF Bzihave the same value. A previously
suggested control device [8] can be applied directly to
where the first four Ni are the standard bilinear shape the quadrilateral, but this leads to a very slight failure
functions [e.g. Nr = (1 - <)(l - r/)/4], xi/ = Xi - Xj
and y, = yi - yj are comer coordinate differences, and
of the patch test unless /? = 0 or /I = 1 in eqn (4). The elevation z may be taken as a parabola between nodes
q~d~laterai passes the patch test for all values of /? i and j with a mid-edge elevation hi relative to line g
if instead the device is applied to the four overIapping (Fig. 2). Thus eqn (8) yields
subtriangles 123, 234, 341, and 412, and the four
results are added. For the subtriangle spanning
nodes 1, 2, and 3 for example, we add to the
(c,), = y + 2. (ei - e,). (9)
I i
R = x23
4A 4A
Yz3 1 2 Y31 1 -xJZ ---YJZ
---
3 4A 4A 3 4A 4A
1
3
1(6) relationships of the form
which operates on DOF [u, v, 0,, u2 u, 6,, u3 vj where ss is a column matrix of three edge-parallel
e,,]‘. The factor of 10e6 is rather arbitrary. A factor strains, the Cartesian strains are
as large as lop2 changes element behavior very
little [8]. The value 10m6 has almost no effect other
than suppression of the mechanism.
C contains sines and cosines of edge orientations (pi ,
Bending stiffness 8 contains six in-plane nodal rotation DOF, and S
Plate-bending stiffness resists the transverse dis- contains the information of eqn (9) for each edge
placement DOF wi and the in-plane rotation DOF 0, along with sines and cosines of &. The expression
and f?,. For plate action we choose the well-known L = C’S8 defines a constant strain state over the
DKT element [l 1, 121, arranged as a set of four triangle. Terms of C’S are entered into the mem-
overlapping subtriangles to avoid directional bias, brane strain-displacement matrix and make it poss-
and with each subtriangle having half the flexural ible for inextensional bending states to exist, thus
rigidity of the actual material. This choice is made removing a source of membrane locking.
merely because the DKT element works well and is Unfortunately the foregoing procedure for a tri-
conveniently available. A good quadrilateral plate angle is not directly transferrable to a quadrilateral.
element could be used instead, probably with very A least squares procedure that uses four edges to
similar results. determine the three strains in L fails for a rectangle
because it provides no info~ation about shear strain
membrane-bending c~upl~g
yXy.This defect can be cured by adding diagonals of
The coupling procedure is applied in various but the quadrilateral to the least squares procedure, but
similar forms in [l, 2,6]. The form used here is the then a cylindrical shell element does not respond
very simple form described in 161,so it is presented properly to moment vectors parallel to the axis of the
~ompac~y here. Along a typical element edge of cylinder. Both problems are solved by using the
length LI between nodes i and j, let s be an edge- scheme of four overlapping triangles for all three
tangent coordinate and u, be the s-direction displace- strains in c = C’S& as described by yXyalone in eqn
ment, where u, varies linearly with s between nodal (3). Here the weighting by triangle areas A, is not
values u, and Use.Let the lateral displacement w along necessary but provides a definite improvement in
edge 0 be performance.
The foregoing subtriangle procedure requires el-
L,-s evations hi at midpoints of both diagonals of the
-%Wj+ q$ (0, - e,>, (7)
w=L,wi , I quadrilateral. They are computed as follows. Let h, ,
h2, h,, and hg be the hi at midpoints of sides 12, 23,
where Bi and 13~are nodal rotations in the xy plane, 34, and 41. ~esumably these hi are available as input
positive outward from the edge, and expressible in data. Let h, be hi at the midpoint of diagonal 24
terms of fJIXi
and f?,,and the angle $i between the x axis
and direction ij. The average strain along edge v is
taken as
Warping correction
a rank-two matrix that operates on only the drilling
Thus far the element has been formulated using its
DOF. The latter matrix is the sole cause of membrane
projection, for which all nodes lie in the xy reference
locking in shells, but must be present when elements
plane. If global coordinates do not place all element
are coplanar. This matrix can be multiplied by a
nodes in the same plane, element behavior is much
factor that is unity for coplanar elements but ap-
too stiff in many problems unless modified by a
proaches zero as the curvature of the element in-
warping correction. The correction used here is taken
creases. Thus it is easy to avoid membrane locking in
from eqns (9b), (SC), (lOc), and (11) of [7l. The
the triangular shell element.
correction is embodied in a matrix T such that the
The corresponding remedy for the quadrilateral
transformation TrIcT produces a z-direction stiffness
would use /_I= 1 in eqn (4) for coplanar elements and
from membrane stiffness coefficients while leaving
arrange for /I to approach zero as elements become
existing membrane stiffness coefficients and bending
more highly curved. However, edge tractions on
stiffness coefficients unchanged. Matrix T is gener-
elements produce nodal torques corresponding to the
ated entirely from nodal coordinate data in the local
0,, when /I = 1, but zero torques when #I = 0. Hence
xyz system.
adjacent elements having different values of /I would
be incompatible (with the triangular element, similar
Membrane locking tendency adjustments of the second stabilization matrix pre-
sent no such difficulty). At this time the writer does
The origin of membrane locking is suggested by
not know of a satisfactory cure, but fortunately the
Fig. 3, where a nodal rotation 0, is associated with
quadrilateral element does not have a great propen-
bending deformation in element 2 but has a drilling
sity for membrane locking.
component 6, normal to the base plane of element 1;
It may be noted that use of one-point integration
0, is associated with membrane action in element 1.
for only the membrane terms associated with the tlli,
Accordingly eb is in part resisted by membrane
with four-point integration for other terms, produces
stiffness, which is typically far greater than bending
elements that either fail the patch test or behave badly
stiffness and hence causes bending deformations to be
otherwise.
much too small. This ‘locking’ tendency vanishes as
elements become coplanar.
The same difficulty appears in a triangular shell
element [6]. The element of [6] contains a form of
constant-strain triangle and requires two stabilization
matrices: one is described by eqn (6) and the other is
1.0
.A__
OF I
1.0
I- “.”
Diaphragm
lW .I
L I.1 1.9 3.1 3.9 5.1 I.0
J
X
+F
Free placement at C is 5256 x 10e6 according to theory.
Fig. 7. Hemispherical shell test base. One quadrant with a distorted mesh, fl = 0: w/w,, = 0.051
4 x 4 mesh is shown. Radius = 10.0, thickness = 0.04,
E = 6.825 x lo’, v = 0.3, F = 1.0 for the quadrant modeled. distorted mesh, /I = 1: w/w, = 0.897.
Displacement at A is 0.0924 according to theory.
Table 1. Shell test case results reported as the ratio of computed deflection to accepted
theoretical deflection
Element formulation
Hemisphere, Fig. 7
1x1 0.117 0.056 1.533 0.057
2x2 0.852 0.228 0.654 0.168
4x4 1.002 0.444 0.586 0.606
8x8 1.001 0.948 0.971 0.962
Table 2. Shell test case results using the ‘curved softened’ TRIANGULAR element of [6]
Mesh Shell roof Pinched cylinder Hemisphere Twisted strip
IX1 moo 0.121 0.213
2x2 1.406 0.959 0.873 0.775
4x4 0.880 0.974 0.978 0.962t
8x8 0.939 0.999 1.007 0.992$
t 2 x 4 mesh.
$2 x 8 mesh.
7. B. P. Naganarayana and G. Prathap, Force and mo- Me& (Edited by J. P. Den Hartog and H. Peters),
ment corrections for the warped four-node quadrilateral pp. 93-101. John Wiley, New York (1939).
plane shell element. Comput. Struct. 33, 1107-1115 14. M. Iura and S. N. Atluri, Formulation of a membrane
(1989). finite element with drilling degrees of freedom. Comput.
8. R. H..MacNeal and R. L. Harder, A refined four-noded Mech. 9, 417-428 (1992): _
membrane element with rotational degrees of freedom. 15. T. Belytschko, B. L. Wong and H. Stolarski, Assumed
Comput. Struct. 28, 75-84 (1988). strain stabilization procedure for the 9-node Lagrange
9. R. D. Cook, D. S. Malkus and M. E. Plesha, Concepts shell element. Int. J. Numer. Meth. Engng 28, 385414
and Applications of Finite Element Analysis, 3rd Edn. (1989).
John Wiley, New York (1989). 16. T. J. R. Hughes, M. Cohen and M. Haroun, Reduced
10. R. D. Cook, Avoidance of parasitic shear in plane and selective integration techniques in the finite element
element. AXE J. Struct. Div. 101, 1239-1253 (1975). analysis of plates. Nuclear Engng Design 46, 203-222
Il. J. A. Stricklin, W. E. Haisler, P. R. Tisdale and R. (1978).
Gunderson, A rapidly converging triangular plate ele- 17. W. K. Liu, E. S. Law, D. Lam and T. Belytschko,
ment. AIAA Jnl7, 180-181 (1969). Resultant-stress degenerated-shell element. Com-
12. C. Jeyachandrabose, J. Kirkhope and C. R. Babu, An put. Meth. Appl. Mech. Engng 55, 259-300
alternative formulation for the DKT plate-bending ele- (1986).
ment. ht. J. Numer. Meth. Engng 21, 1289-1293 18. K. Y. Sxe and C. L. Chow, A mixed formulation of a
(1985). four-node Mindlin shell/plate with interpolated covari-
13. K. Marguerre, Zur theorie der gekriimmten plate ant transverse shear strains. Comput. Struct. 40,
grosser formlnderung. Proc. Fifih Int. Cong. for Appl. 775-784 (1991).