Finite Elements in Analysis and Design 30 (1998) 175—186

Perspective on finite elements for shell analysis

Richard H. MacNeal
The MacNeal-Schwendler Corporation, 815 West Colorado Avenue, Los Angeles, CA 90041, USA

Abstract

The history of the development of finite elements for shell analysis is reviewed with emphasis on innovations that
improved quality. Three major events are noted: the introduction of the first plate bending element in 1961, the
introduction of Mindlin shell elements beginning in 1969, and the introduction of higher order p elements, which is recent
and ongoing. The paper concludes with an examination of the effects of transverse shear flexibility on the accuracy of
p elements. ( 1998 Elsevier Science B.V. All rights reserved.

Keywords: Plate elements; Shell elements; Kirchhoff elements; Mindlin elements; NASTRAN; MSC/NASTRAN; Re-
duced integration; Shear locking; Membrane locking; Transverse shear locking; Transverse shear; Assumed strain
method; Finite element history

1. Introduction

Although the first plate bending element was introduced in 1961 [1], elements which are
adequate for general shell analysis only became available in the 1970s. Table 1 outlines the
chronology of plate and shell elements development. Needless to say, research on the design of
plate and shell elements continues to this day.
A simple shell problem which has become a standard for benchmarking shell elements
is the Scordelis-Lo roof [2], shown in Fig. 1. Results for this example with pre-1970 and
post-1970 elements are plotted in Fig. 2. The improvements with time are evident and indicate
a growing understanding of accuracy issues. For example, a comparison of the Ahmad
shell element with and without reduced integration indicates that an issue of profound importance
is operative here.

2. Kirchhoff elements, 1961—1970

Coordinate directions and displacement definitions for a flat plate are shown in Fig. 3. Finite
elements for plates and shells employ the fundamental assumptions of first-order plate theory
0168-874X/98/$19.00 ( 1998 Elsevier Science B.V. All rights reserved
PII S 0 1 6 8 - 8 7 4 X ( 9 8 ) 0 0 0 0 5 - 5
176 R.H. MacNeal / Finite Elements in Analysis and Design 30 (1998) 175—186

Table 1
Chronology of plate and shell element development

Element development Date

First membrane element 1956

First plate bending element 1961
Kirchhoff plate elements 1961—1970
Discrete Kirchhoff plate and shell elements 1969
First Mindlin shell element (eight-node quadrilateral) 1969
Reduced integration 1971
Four-node quadrilateral, three-node triangle 1976—1982
Nine-node quadrilateral 1985—1990
Higher-order shell elements 1998

Fig. 1. Shell roof under gravity load (the Scordelis—Lo Roof).

which are that inplane displacements and normal stress satisfy

u"um(x, y)!za(x, y), (1)
v"vm(x, y)!zb(x, y), (2)
p "0. (3)
z
The membrane part of displacement (um, vm) is separable, for flat plates, from the bending part
(w, a, b). Finite elements to represent the membrane part have been available since 1956 [3]. Credit
for the earliest plate bending element goes to Adini and Clough who published in 1961 [1]. Theirs
was a rectangular element with four corner nodes and three degrees of freedom (w, a, b) per node. It
employed the Kirchhoff hypothesis, namely that the transverse shear strains
c "w !a, c "w !b (4)
xz ,x yz ,y
 R.H. MacNeal / Finite Elements in Analysis and Design 30 (1998) 175—186 177

Fig. 2. Scordelis—Lo Roof: Performance of pre-1970 elements (n, h, L) versus post-1970 elements (m, j, v).

Fig. 3. Coordinate definitions for a plate.

are null. Long before the invention of finite elements, analysts used the Kirchhoff hypothesis to
reduce the plate bending equations to a single fourth-order partial differential equation in the
lateral displacement, w. The virtue of the Kirchhoff hypothesis for the design of plate elements is
that the designer needs to specify only w as a function of position since a and b can be computed by
Eq. (4).
In their element, Adini and Clough specified w as a twelve-term polynomial which allowed
independent cubic variation of w on each edge. Specifically,
w"(1, y)(1, x, x2, x3)#(1, x)(1, y, y2, y3) (5)
where all distinct product terms have independent coefficients. The twelve independent coefficients
were then evaluated in terms of the twelve corner displacements. The remaining work to make
178 R.H. MacNeal / Finite Elements in Analysis and Design 30 (1998) 175—186

a finite element was to compute a stiffness matrix from the functional form of the strain energy
implied by the assumed variation of w. See Ref. [4] for example.
Early finite element designers also tried to develop a corresponding three-node triangle with nine
degrees of freedom. They were confronted by the fact that a complete cubic has ten terms
1, x, y, x2, xy, y2, x3, x2y, xy2, y3 (6)
so that they had to eliminate one term. Unfortunately, they could find no combination of terms
which retained constant bending (the quadratic terms) and retained isotropy with respect to the
element’s geometry.
Developers of Kirchhoff elements also soon discovered that a completely satisfactory solution
could not be achieved for any element with three degrees of freedom per node. Specifically, Irons
and Draper [5], 1965, discovered that an expression for w which assures uniqueness of the bending
curvatures (s , s , s ) over the surface cannot assure slope continuity (continuity of w and w )
xx yy xy ,x ,y
along the common edges of adjacent elements when only w, w , w are prescribed at nodes.1
,x ,y
Continuity of slope is important because without it a field of elements cannot correctly represent
a state of constant bending curvature.
At this point, three alternative courses presented themselves to the designers of Kirchhoff plate
elements. They were:
1. Accept the non-uniqueness of curvature within an element in order to assure slope continuity.
2. Accept discontinuity of slope in order to assure uniqueness of curvature.
3. Add higher order derivatives of w, such as w , as nodal degrees of freedom.
,xy
All three courses of action have been used by element designers. It is clear from its description
that the Adini—Clough rectangle uses the second alternative. The Clough—Tocher triangle [6],
1965, which appeared in NASTRAN as the TRIA2 element, uses the first alternative. It consists of
three subtriangles (see Fig. 4) with nine displacement terms each, and enough constraints to
eliminate the degrees of freedom at the common center node and to ensure internal slope
continuity. The Fraeijs de Veubeke quadrilateral [7], 1968, which Irons called the prettiest such
element, also uses the first alternative. It has four subtriangles and preserves slope continuity.
The most interesting option is to add the second derivatives of w as nodal degrees of freedom so
as to assure slope continuity and uniqueness of bending curvature. A three-node bending triangle
constructed according to this option will have six degrees of freedom (w, w , w , w , w , w ) per
,x ,y ,xx ,xy ,yy
node for a total of eighteen. Since a complete quintic in x and y has twenty-one terms, we can
design an element by deleting three terms from the complete quintic for w or by adding three
degrees of freedom (such as normal slope at the midpoints of edges) to the element. 1968 and 1969
saw the publication of seven independent papers which described either the first approach [8,9], or
the second [10—12], or both [13,14].
Theoretically the element just described has a tremendous degree of accuracy, but today it is
mainly a curiosity. The reason for its lack of acceptance is the use of the second derivatives of w as
nodal variables. Since they physically correspond to curvatures, how are they to be loaded and

1 An elementary proof will be found in Ref. [4], pp. 381, 382.

 R.H. MacNeal / Finite Elements in Analysis and Design 30 (1998) 175—186 179

Fig. 4. The Clough-Tocher triangle.

how are boundary conditions to be applied to them? They also prevent discontinuities in bending
curvature at element corners such as would result from applied moments or changes in material
properties.
Elements, such as those just described, which apply the Kirchhoff hypothesis á priori in the
selection of a displacement field, have long since passed from favor. In discrete Kirchhoff elements,
separate fields are assumed for a and b and the Kirchhoff constraints (a"w , b"w ) are applied
,x ,y
at discrete points to eliminate excess nodal variables. Early successful elements of this type include
both triangles [15] and quadrilaterals [16] with minimum node counts and, most prominently,
Irons’ semiloof element [17], 1976.

3. Lower-order Mindlin elements, 1969—1990

While the designers of Kirchhoff elements were attempting to cope with fundamental limitations,
progress was about to come from a different direction—which today we would call a paradigm
shift — namely, extension of the isoparametric mapping concept to shell elements. Introduced with
the Taig quadrilateral membrane element [18] in 1961, isoparametric mapping maps the interior of
an irregular element into a standard shape (e.g., a curved quadrilateral into a square) and
automatically preserves interelement displacement continuity in the process.2 In 1966, Irons [19]
extended the concept generally to elements of all dimensions and any number of edge nodes.
An important three-dimensional element made possible by Irons’ work is the twenty-node solid
brick which has nodes at the eight corners and at the midpoints of the twelve edges. In 1969,
Ahmad [20] published a “degenerate” eight-node isoparametric shell element which was derived
from the twenty-node brick by the procedure illustrated in Fig. 5. The essential step, which
corresponds precisely to the assumptions of plate/shell theory, is to replace the six translational
degrees of freedom at corresponding points on the top and bottom surfaces of the solid by three
translations (u, v, w) and two rotations (a, b) at their midpoint. Each of these five new displacement
components is then interpolated from nodes to internal points by the same shape functions,

2 In isoparametric mapping, the same basis functions are used to interpolate positions and displacements from nodal
values.
180 R.H. MacNeal / Finite Elements in Analysis and Design 30 (1998) 175—186

Fig. 5. Degeneration of a twenty-node brick element into an eight-node shell element.

expressed as functions of m, g, the parametric coordinates of points on the midsurface. Membrane

strains, transverse shear strains, and bending curvatures are then computed from the displacement
shapes in a manner entirely analogous to that used in the earlier plate elements but rendered more
complicated by the curvature. Transverse shear strains are not set to zero as in the Kirchhoff
elements. This occasioned the application of a new name — Mindlin or Reissner — Mindlin — to the
theoretical basis of the new element.
The new Ahmad element represented a giant step backward with respect to the representation of
transverse displacement. Since a and b are now independent of w, the spatial representation of w in
the eight-node element contains only eight terms rather than twenty-four. Nevertheless, Ahmad’s
element has the very important advantages that it can include transverse shear deformation and
that it fits neatly within the isoparametric formalism.
The new element also has problems of its own which soon revealed themselves. Refering to
Fig. 2, we note that the Ahmad shell element produces results which are not much better than those
for the lower order Clough—Tocher triangle described earlier. Fig. 2 also shows a dramatic
 R.H. MacNeal / Finite Elements in Analysis and Design 30 (1998) 175—186 181

improvement in results when “reduced order integration” is added to the Ahmad shell element.
This result for the Ahmad element was published by Zienkiewicz et al., in 1971 [21] without
a fundamental explanation, but it proved to be fundamentally important for the future develop-
ment of plate and shell elements.
One would not expect that changing the integration points for the Ahmad element from a 3]3
mesh of Gauss points to a 2]2 mesh would make all that difference. The explanation of the efficacy
of reduced integration for the Ahmad curved shell element is not easy — it had to wait until 1982
[22,23] — but reduced integration had already been understood and applied in 1969 [24] to
improve the accuracy of the isoparametric membrane quadrilateral. At bottom, the difficulty is that
the element’s shape or basis functions are unable to correctly interpolate higher order displace-
ments states from nodal values. For example, the basis functions of the four node isoparametric
quadrilateral, which are (1, m, g, mg), are unable to represent a pure inplane bending state,
u"xy, v"!x2/2, even for a rectangular element where x"am, y"bg. In fact, for this case, v is
represented by its alias, v"!a2/2, which has the correct value at nodes but not elsewhere. The
only point at which the inplane shear strain, c "u #v , is correct is at the center, m"g"0.
xy ,y ,x
This point happens to be the reduced-order Gauss integration point for the four-node quadrilat-
eral. The other two strain components, e and e , are measured at the 2]2 Gauss points, giving rise
x y
to the term selective reduced-order integration.
If all three strain components in the last example are measured at the 2]2 Gauss points,
the measured shear strains will be incorrect (they should be zero in the example). If the aspect ratio
of the element is large, aAb, the effect will be to greatly increase the strain energy, giving rise
thereby to shear locking, a term which characterizes the greatly reduced displacements that will
result.
Shear locking also appeared when shell element developers tried to apply the isoparametric
formulation pioneered by Ahmad to three-node and four-node plate bending elements. Locking
occurs because the lowest order bending state, w"x2, is not representable by the basis set of the
three-node element (1, x, y) or by the basis set of the four-node element (1, m, g, mg). The first paper
which proposed a remedy, 1977 [25] for the quadrilateral case, used a single integration point for
transverse shear strain. This unfortunately gave rise to spurious strain-free modes because the
number of independent strain evaluations was insufficient to restrain these modes.
It was also soon discovered that the only places where the transverse shear strains are correctly
computed for three-node and four-node elements when w"x2 are at the midpoints of the edges.
This discovery led to a new wrinkle in finite element theory — the direct assumed strain method —
whereby transverse shear strains were computed at the midpoints of edges and interpolated to
integration points. Between 1976 and 1982, Hughes [26,27] and MacNeal [22,28] independently
published three- and four-node elements based on the direct assumed strain method. These
elements have become the workhorses of finite element plate and shell analysis. Fig. 2 shows that
these elements, TRIA3 and QUAD4, have respectable accuracy.
The original Ahmad eight-node shell element avoids transverse shear locking in its most virulent
form because its basis set can model quadratic terms. It suffers, however, from another form of
locking known as membrane locking [23] because, when an element is curved, pure bending
involves inplane displacements, measured in a Cartesian coordinate system, which are cubic
functions of position. The remedy, reduced integration, works because the values of the membrane
strains are correct at 2]2 Gauss integration points.
182 R.H. MacNeal / Finite Elements in Analysis and Design 30 (1998) 175—186

Attempts to apply reduced integration to the nine-node shell element fail because the additional
five degrees of freedom at the center produce spurious modes. The period 1985—1990 saw the
development of several approaches toward the elimination of locking and spurious modes for the
nine-node element. While the nine-node shell element has some minor advantages over the
eight-node element [29], the chief motivation was probably just the challenge of a difficult task.
One successful approach was extension of the direct assumed strain method from four to nine
nodes by Park and Stanley [30], 1986. Another was the development of the assumed strain hybrid
method in which a strain field with carefully selected low-order terms is assumed in addition to
a displacement field. The two fields are correlated using mixed variational principles [31,32] or
constrained variational principles [33,34]. In another approach, mode stabilization [35,36], high
and low terms in the displacement field are separated and the low-order terms are treated exactly
while the high-order terms are treated approximately. In its crudest form, mode stabilization gives
just enough stiffness to the high-order terms to avoid the appearance of spurious modes. Mode
stabilization is a preferred method for nonlinear analysis because the restriction of accurate
integration to the low-order terms reduces computer time.

4. Higher order elements, 1988

The extension of conventional element technology beyond bi-quadratic displacement shapes has
been rare. Rhiu, et al., [37,38] 1989, 1990, have developed third-order degenerated shell elements
with sixteen nodes which feature an assumed strain field to eliminate locking. The accuracy of these
elements appears to be excellent.
An important recent development has been the extension of p technology to shell elements. With
the p method [39], element displacements are designed to have a variable polynomial degree, p,
which is selected at execution time. Conventional nodes with displacement degrees of freedom are
placed at element corners. All higher degrees of freedom for p*2 are defined as non nodal
hierarchical variables along edges and in the element’s interior. The designation hierarchical refers
to the fact that such degrees of freedom can be added progressively without changing existing parts
of the stiffness matrix.
The chief virtue of p elements is that they allow easy implementation of adaptive mesh re-
finement by increasing the polynomial degree of elements rather than by increasing the number
of elements. It then becomes possible for the user to specify a desired level of accuracy and to
let the computer program set the p level for the whole mesh or selectively for individual
elements.
In perhaps the earliest paper on the application of p elements to shell analysis, Szabó and
Sahrmann [40], 1988, employed solid elements with p"1 in the thickness direction and prescribed
a zero value for Poisson’s ratio to avoid locking. Their analysis of the Scordelis-Lo roof (Fig. 1),
demonstrated that reasonable convergence of displacements is achieved for a single element model
with p"6 in the directions of the midsurface.
The appearance of shell p elements in the technical literature has been slow. Leino et al. at the
Helsinki University of Technology have published reports [41,42] which show, among other
things, that reduced integration produces significantly higher accuracy out to p"4. Shell p
elements are known to exist in two commercial finite element programs, Mechanica and
 R.H. MacNeal / Finite Elements in Analysis and Design 30 (1998) 175—186 183

Fig. 6. Clamped square plate (point load at center).

MSC/NASTRAN. The development of practical shell p elements is labor intensive, which perhaps
explains the slowness of their appearance in the open technical literature.
It may be worthwhile to comment on some of the features of the shell p elements in
MSC/NASTRAN. There it is recognized that uniform accuracy does not require that all elements
have the same p level, so that p is allowed to be different in every element and different in the m and
g directions of a quadrilateral element. Along any common edge, however, the p values must be the
same in adjacent elements to preserve displacement continuity.
The formulation of stiffness in p elements has generally followed the standard isoparametric
formalism3 without special treatment to avoid locking. It has, in effect, been argued that lock-
ing is only a problem for p"1 and p"2 and that these low levels are unimportant and can be
discarded. In MSC/NASTRAN we have taken the opposing view that p"1 and p"2 are
important because many examples exist where accuracy requirements can be easily satisfied with
low order elements over large regions of the structure. We have, accordingly, included variations of
the antilocking devices described earlier for p"1 and p"2. This becomes complicated when one
realizes that p may be different on all four edges of an element. We also employ reduced integration
for p*2.
One of the standard test problems for MSC/Nastran elements is the square plate with a clamped
boundary condition shown in Fig. 6. The standard thickness-to-length ratio for this surprisingly
difficult test problem is t/a"0.0001, chosen so small to emphasize any tendency toward transverse
shear locking. Table 2 records the lateral deflection at the center for N, p"1,2,3,4, normalized to
the correct solution. The results are quite poor, indicating a strong presence of locking even though
several antilocking devices are used.
The results improve dramatically, as shown in Table 3, when the thickness-to-length ratio is
increased. This is remarkable because, even at t/a"0.01, the ratio of transverse shear flexibility to
bending flexibility is exceedingly small. These results can be explained by noting that the Kirchhoff

3 Actually subparametric in MSC/NASTRAN because only cubic edge geometry is allowed for p*3.
184 R.H. MacNeal / Finite Elements in Analysis and Design 30 (1998) 175—186

Table 2
Lateral deflection at center of a clamped square plate with a point load at center

N p

1 2 3 4

1 0.00000 1.11607 0.00001 0.38940

2 0.00001 0.38308 0.71086 0.98198
3 0.00005 0.19744 0.85764 0.99394
4 0.00029 0.12684 0.91864 0.99767

Table 3
Accuracy versus thickness of central deflection of a clamped square plate with point load at
center normalized to correct value

p N t/a"0.0001 t/a"0.001 t/a"0.01

1 3 0.00005 0.00541 0.33340

2 1 1.11607 1.11608 1.11663
2 2 0.38308 0.38432 0.48697
2 3 0.19744 0.24164 0.86304
2 4 0.12684 0.43630 0.97361
3 1 0.00001 0.00115 0.10203
3 2 0.71086 0.74251 0.93456
3 3 0.85746 0.91375 0.99154
4 1 0.38940 0.39590 0.57622

condition is effectively imposed at integration points by the small thickness, thereby imposing
substantial distortion on higher modes which, because of aliasing, cannot satisfy the Kirchhoff
condition. Relaxation of the Kirchhoff constraint, even to a slight degree, significantly relaxes
distortion of the displacement shape.
Another conclusion which can be drawn from this example is that a minimum amount of
transverse shear flexibility should be included in the elements, even for higher p values, to combat
transverse shear locking. The older MSC/NASTRAN shell elements have, for a long time, included
a minimum amount of transverse shear flexibility, called the residual bending flexibility [43], which
was originally introduced to reduce discretization error. The example shows, however, that residual
bending flexibility is, and always has been, more valuable as an antilocking device.

5. Concluding remarks

The long history of gradual improvements in the design of finite shell elements has been
reviewed. Two major events have been noted: the change from Kirchhoff elements to Mindlin
 R.H. MacNeal / Finite Elements in Analysis and Design 30 (1998) 175—186 185

elements occurring about 1970, and the introduction of p technology which is ongoing. The first
event was occasioned by the discovery of fundamental limitations in Kirchhoff elements. The
second is occurring in response to the need to make structural analysis accessible to a wider user
community through automation.
Along the way, progress has occurred by the process of detecting failure modes, discovering the
reasons for failure, and devising fixes. Users of finite elements should be at least marginally aware of
the failure modes in order to avoid disappointment.

References

[1] A. Adini, R.W. Clough, Analysis of plate bending by the finite element method and report to Natl. Sci.
Foundation/USA, G7337, 1961.
[2] A.C. Scordelis, K.S. Lo, Computer analysis of cylindrical shells, J. Am. Concr. Inst. 61 (1961) 539—561.
[3] M.J. Turner, R.W. Clough, H.C. Martin, L.J. Topp, Stiffness and deflection analysis of complex structures,
J. Aeronaut. Sci. 23 (1956) 803—823.
[4] R.H. MacNeal, Finite Elements: Their Design and Performance, Chap. 2, Marcel Dekker, New York, 1993.
[5] B.M. Irons, J.K. Draper, Inadequacy of Nodal connections in a stiffness solution for plate bending J. AIAA, 3
(1965) 5.
[6] R.W. Clough, J.L. Tocher, Finite element stiffness matrices for analysis of plates in bending, Proc. Conf. Matrix
Methods in Struct. Mech., Air Force Inst. Tech., Wright—Patterson AFB, Ohio, 1965.
[7] B. Fraeijs de Veubeke, A conforming finite element for plate bending, Int. J. Solids Struct. 4 (1968) 95—108.
[8] G.A. Butlin, R. Ford, A compatible plate bending element, Univ. of Leicester Eng. Dept. Report, 1968, pp. 68—115.
[9] G.R. Cowper, E. Kosko, G.M. Lindberg, M.D. Olson, Formulation of a new triangular plate bending element,
trans. Canad. Aero-Space Inst., 1, pp. 86—90, 1968 (see also N.R.C. Aero Report LR514, 1968).
[10] B.M. Irons, A conforming quartic triangular element for plate bending, Int. J. Numer. Methods Eng. 1 (1969) 29—46.
[11] W. Bosshard, Ein Neues Vollverträgliches Endliches Element Für Plattenbiegung, Mt. Assoc. Bridge Struct. Eng.
Bull. 28 (1968) 27—40.
[12] W. Visser, The finite element method in deformation and heat conduction problems, Dr. W. Dissertation, T.H.,
Delft, 1968.
[13] K. Bell, A refined triangular plate bending element, Int. J. Numer. Methods Eng. 1 (1969) 101—122.
[14] J.H. Argyris, I. Fried, D.W. Scharpf, The TUBA family of plate elements for the matrix displacement method,
Aeronaut. J.R. Ae. S. 72 (1968) 701—709.
[15] J.A. Stricklin, W. Haisler, P. Tisdale, R. Gunderson, A rapidly converging triangular plate element, J. AIAA,
7 (1969) 180—181.
[16] J.L. Batoz, M. Ben Tahar, Formulation et Evaluation d’un Nouvel Elément Quadrilatéral á 12 D.L. pour la Flexion
des Plaques Minces, Département de Génie Mécanique, Université de Technologie, Compiégne, France.
[17] B.M. Irons, The semiloof shell element, Finite Elem. Thin Shells Curved Members 11 (1976) 197—222.
[18] I.C. Taig, Structural analysis by the matrix displacement method, Engl. Electric Aviation Report No. 5017, 1961.
[19] B.M. Irons, Engineering application of numerical integration in stiffness methods, J. AIAA, 14 (1966) 2035—2037.
[20] S. Ahmad, Curved finite elements in the analysis of solid, shell, and plate structures, Ph.D. Thesis, University of
Wales, Swansea, 1969.
[21] O.C. Zienkiewicz, J. Too, R.L. Taylor, Reduced integration technique in general analysis of plates and shells, Int.
J. Numer. Methods Eng. 3 (1971) 275—290.
[22] R.H. MacNeal, Derivation of element stiffness matrices by assumed strain distributions, Nucl. Eng. Design 70
(1982) 3—12.
[23] H. Stolarski, T. Belytschko, Shear and membrane locking in curved C0 elements, Comput. Methods Appl. Mech.
Eng. 41 (1983) 279—296.
[24] W.P Doherty, E.L. Wilson, R.L. Taylor, Stress analysis of axisymmetric solids using higher order quadrilateral
finite elements, Univ. of Calif. Berkeley, Struct. Eng. Lab. Report SESM 69—73, 1969.
186 R.H. MacNeal / Finite Elements in Analysis and Design 30 (1998) 175—186

[25] T.J.R. Hughes, R.L. Taylor, W. Kanoknukulchai, A simple and efficient element for plate bending, Int. J. Numer.
Methods Eng. 11 (1977) 1529—1543.
[26] T.J.R. Hughes, T.E. Tezduyar, Finite elements based upon Mindlin plate theory with particular reference to the
four-node bilinear Isoparametric element, J. Appl. Mech. (1981) 587—596.
[27] T.J.R. Hughes, R.L. Taylor, The linear triangular bending element, in: Math. Finite Elem. Appl. IV, MAFELAP
1981, 1982, pp. 127—142.
[28] R.H. MacNeal, The TRIA3 plate element, MacNeal-Schwendler Corp. Memo RHM-37, 1976.
[29] R.H. MacNeal, R.L. Harder, Eight nodes or nine?, Int. J. Numer. Methods Eng. 33 (1992) 1049—1058.
[30] K.C. Park, G.M. Stanley, A curved C0 shell element based on assumed natural-coordinate strains, J. Appl. Mech. 53
(1986) 278—290.
[31] J.J. Rhiu, S.W. Lee, A new efficient mixed formulation for thin shell finite element models, Int. J. Numer. Methods
Eng. 24 (1987) 581—604.
[32] T.Y. Chang, A.F. Saleeb, W. Graf, On the mixed formulation of a nine-node Lagrange shell element, Comput.
Methods Appl. Mech. Eng. 73 (1989) 259—281.
[33] H.C. Huang, E. Hinton, A new nine-node degenerated shell element with enhanced membrane and shear
interpolation, Int. J. Numer. Methods Eng. 22 (1986) 73—92.
[34] J. Jang, P.M. Pinsky, An assumed covariant strain based nine-node shell element, Int. J. Numer. Methods Eng. 24
(1987) 2389—2411.
[35] T. Belytschko, W.K. Liu, S.-J. Ong, D. Lam, Implementation and application of a nine-node Lagrange shell element
with spurious mode control, Comput. Struct. 20 (1985) 121—128.
[36] D.W. White, J.F. Abel, Accurate and efficient nonlinear formulation of a nine-node shell element with spurious
mode control, Comput. Struct. 35 (1990) 621—641.
[37] J.J. Rhiu, S.W. Lee, A sixteen-node shell element with a matrix stabilization scheme, Comput. Mech. 3 (1988)
99—113.
[38] J.J. Rhiu, R.M. Russell, S.W. Lee, Two higher-order shell finite elements with stabilization matrix, J. AIAA 28 (1990)
1517—1524.
[39] B.A. Szabó, A.K. Mehta, p-Convergent finite element approximations in fracture mechanics, Int. J. Numer.
Methods Eng. 12 (1978) 551—560.
[40] B.A. Szabó, G.J. Sahrmann, Hierarchic plate and shell models based on p-Extension, Int. J. Numer. Methods Eng.
26 (1988) 1858—1881.
[41] Y. Leino, J. Pitkäranta, On the membrane locking of h—p finite elements in a cylindrical shell problem, Helsinki
Univ. of Technology, Inst. Math. Res. Reports A311, June 1992.
[42] H. Hakula, Y. Leino, J. Pitkäranta, Pinched shell of revolution: experiments on high order FEM, Helsinki Univ.
Tech., Inst. Math. Res. Reports, A346, April 1995.
[43] R.H. MacNeal, A simple quadrilateral shell element, Comput. Struct. 8 (1978) 175—183.