INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING, VOL.

11, 1529-1543 (1977)

A SIMPLE AND EFFICIENT FINITE ELEMENT FOR

PLATE BENDING
THOMAS J. R. HUGHES*
Division of Engineeringand Applied Science, CaliforniaInstitute of Technology,Pasadena, California, U.S.A.

ROBERT L. TAYLOR? AND WORSAK KANOKNUKULCHAIS

Division of Siructural Engineering and Structural Mechanics, Department of Civil Engineering, University of California,
Berkeley, California, U.S.A.

SUMMARY
A simple and efficient finite element is introduced for plate bending applications. Bilinear displacement
and rotation functions are employed in conjunction with selective reduced integration. Numerical
examples indicate that, despite its simplicity, the element is surprisingly accurate.

INTRODUCTION
An enormous amount of effort has been devoted to the development of finite elements for the
bending of plates. The literature is extensive and we will not make an attempt to review it here.
(The interested reader may consult any of the standard texts for reference^.'-^) Most of this
effort has been oriented towards linear problems; in particular, to the classical Poisson-
Kirchhoff theory of bending. The C’-continuity requirement imposed by this theory on
‘displacement’ finite element models precludes the development of simple and natural ele-
m e n t ~ Because
.~ of this, incompatible elemen@ are often resorted to, since they involve
simpler programming than the rather complicated C1-continuous elements6-’ and are competi-
tive from an accuracy standpoint.
Accurate higher-order C’-elements have also been de~eloped,~-” but they too are quite
complicated and involve nodal derivative degrees-of-freedom of order greater than one, which
complicates the specification of boundary conditions.
The assumed stress hybrid bending elements of Pian and his associates have often proved to
be accurate, but they have some drawbacks and thus are not widely used.
Another approach to the development of bending elements for thin plates involves the
so-called ‘discrete Kirchhoff hyp~thesis.”~-’~ In this approach the classical equations are
abandoned in favour of a bending theory which includes shear deformations. The result is that
only Co-continuity is required of the shape functions. To capture the behaviour of thin plate
theory, the constraint of zero shear strains is imposed at a discrete number of points. The method
is effective, but implementations tend to be somewhat complicated. Recent improvements and
variants on this theme have been proposed by

* Assistant Professor of Structural Mechanics.

t Professor of Civil Engineering.
$ Graduate Student.
Received May 1976
Revised 12 October 1976
@ 1977 by John Wiley & Sons, Ltd.

1529
1530 T. J. R. HUGHES, R. L. TAYLOR AND W. KANOKNUKULACHAI

An accurate quadrilateral element for thick and thin plates has been developed by Zien-
kiewicz et aL’* This element possesses eight nodes-four corner and four midside-with the
basic three degrees-of-freedom per node. The transverse displacement and rotation shape
functions are selected from the ‘serendipity’ family.’ Two-by-two Gaussian quadrature is an
essential requirement for the good performance of the element.
In summarizing these developments one can confidently assert that for linear problems of
plate bending many adequate elements exist. The choice is more a matter of taste as no single
element is clearly superior to the rest in all cases.
Many users of finite element computer programs find a ‘basic’ four-node quadrilateral
element particularly appealing due to its simplicity. It is our feeling that this appeal will become
even greater when non-linear applications are undertaken. In the non-linear regime-and
especially in non-linear dynamics-computational cost is the prime concern. Due to frequent
reformulations of tangent stiffnesses, complicated element routines can lead to exorbitant
computational expenditures and may actually preclude non-linear analysis. A simpler element
of competitive accuracy becomes quite desirable under such circumstances. Other factors in
non-linear analysis support this viewpoint. For example the accuracy level attainable in
non-linear problems is often severely limited due to the uncertainty of non-linear material
characterizations. Thus it makes little sense to engender significant computational cost for
complicated bending elements which are only marginally more accurate than simpler elements,
since the confidence level of the overall analysis may be affected only negligibly. Unfortunately,
heretofore, no really simple alternative has existed.
In this paper we attempt to remedy this situation. We develop what we believe is the
simplest effective plate bending element yet proposed. The element is a four-node quadrilateral
with the basic three degrees-of-freedom per node. The element shape functions are bilinear for
transverse displacement and rotations. The shear ‘locking’ associated with such low-order
functions in application to thin plates is alleviated by splitting the shear and bending energies
and using one-point quadrature on the shear term. The simplicity of the element lends itself to
concise and efficient computer implementation.
To develop the theory in its simplest setting, we consider in Section 2 ‘Example: linear beam
element’ a beam element involving linear displacement and rotation shape functions. We show
how exact integration (two-point Gaussian quadrature) of the element stiffness matrix leads to
an overly stiff element and we present an heuristic argument why this is the case. We then show
how employing one-point quadrature on the shear term lessens the stiffness. The concept is
identical for the plate bending element which is developed in Section 3 ‘Bilinear plate bending
element’. The effectiveness of the element in thin plate bending is demonstrated in Section 4
‘Numerical examples: thin plates’. A simple computing strategy for dealing with the numerically
sensitive case of extremely thin plates is presented in Section 5 ‘Numerical sensitivity due to
extreme thinness’. Following this we consider applications to thick plates. It is shown that the
element is still effective for moderately thick plates. However, for very thick plates, in which tne
thicknessesof individual elements exceed their characteristic lengths, a slight modification of the
shear quadrature need be employed.

EXAMPLE: LINEAR BEAM ELEMENT

In this section we will describe the formulation of a beam element stiffness for which
displacement and rotation are assumed to be independent linear functions. Exact integration of
the element stiff ness can be facilitated by two-point Gaussian quadrature, whereas one-point
integration exactly integrates the bending contribution, but ‘underintegrates’ the shear con-
 A SIMPLE AND EFFICIENT FINITE ELEMENT FOR PLATE BENDING 1531

tribution. For the case of thin beams we view the shear term as a constraint which attempts to
enforce the condition of negligible shear strains. We shall show that one-point quadrature has a
decisively positive effect on the accuracy of the element; two-point quadrature leading to
worthless numerical results. The upshot of all this is that by appropriately underintegrating
troublesome terms, good bending behaviour can be attained by the simplest shape functions.
The equations of a rectangular cross-section beam, including shear deformation effects,
emanate from the following expression for strain energy:

dx + ~sl~~(
’ - 8) dx] ,
where w is the transverse displacement of the centre-line, 8 is the rotation of the cross-section, E
is Young’s modulus, G is the shear modulus, K is the shear correction factor (throughout we
employ K = 5/6), t is the depth, L is the length and x is the axial co-ordinate. The first term on
the right-hand side of (1) is the bending energy and the second is the shear energy. With
independent expansions for w and 8, (1) can be employed to derive beam element stiffness
matrices. The case we are interested in is when both w and 8 are assumed to behave linearly over
an element. This leads to a four-degree-of-freedom element in which displacement and rotation
are the nodal degrees-of-freedom. By virtue of the fact d8/dx is constant within this element,
the bending energy may be exactly evaluated by one-point Gaussian quadrature. On the other
hand, two-point Gaussian quadrature is required to exactly integrate the shear energy term due
to the explicit presence of 8, which is linear within the element. Employing one-point quadrature
on the shear energy term ‘underintegrates’ the element and it is our prime concern here to
ascertain the effect of this procedure. (See also Gallagher’ pp. 364-367.)
A series of test computations were performed to determine the behaviour of the element. A
cantilever beam subjected to an end load (see Figure 1)was analysed for various descretizations.
The first example is for a relatively deep beam. The data are:
E=1000 G=375 t=l L=4

x - SECT I0N

Figure 1.Cantilever beam subjected to end load

Tip displacement results for several discretizations are presented in Table I. As is evident, the
one-point quadrature results are vastly superior to the two-point results. A more severe test for
linear elements is bending governed by Bernoulli-Euler theory. In this case shear strains are to be
equal to zero. Such a situation can be brought about in the present theory if depth is taken very
small compared with element length. Alternatively, a very large fictitious value of G can be
specified. In the second example we attempt to ascertain the behaviour of the linear element
when the assumptions of the Bernoulli-Euler theory apply. The data of the previous example
are employed with the exception of G which is set here to 375 X lo5.Results are listed in Table
1532 T. J. R. HUGHES, R. L. TAYLOR AND W. KANOKNUKULACHAI

Table I. Normalized tip displacement for a deep cantilever beam

Number of Tip displacement- Tip displacement-

elements one-point quadrature two-point quadrature

1 0.762 0.416 x lo-'

2 0.940 0.445
4 0.985 0.762
8 0.996 0.927
16 0.999 0.98 1

11. The one-point quadrature results are quite accurate whereas the two-point results are in
error by approximately three orders of magnitude. Early attempts at developing bending
elements with simple shape functions were abandoned because of results like those for the
two-point quadrature presented here.

Table 11.Normalized tip displacement for a thin cantilever beam

Number of Tip displacement- Tip displacement-

elements one-point quadrature two-point quadrature

1 0.750 0.200 x 10-4

2 0.938 0.800 x
4 0.984 0-320x
8 0.996 0.128~ 10-3
16 0.999 0.512 x

We shall now proceed to give a heuristic argument why two-point quadrature causes such an
overly stiff element. Consider a cantilever beam discretized into N elements. In the assembled
stiffness matrix there are 2N degrees-of-freedom-two degrees-of-freedom per element. The
shear contribution to the stiffness represents a constraint on the shear strains. If one-point
quadrature is employed, one constraint is imposed upon the element, whereas if two-point
quadrature is employed, two constraints are imposed upon the element. In the latter case the
number of constraints per element equals the number of degrees-of-freedom per element,'and
the result is that the mesh 'locks'.
This can be seen more precisely by looking at the stiffness contributions of the bending and
shear terms. We assume the nodal degrees-of-freedom are ordered as follows: wl,el, w2,02;
and h is the element length. The stiffnesses are:

1 h/2 -1 h/2
h / 2 h 2 / 4 -h/2 h2/4
-1 -h/2 1 -h/2
h / 2 h 2 / 4 -h/2 h2/4
 A SIMPLE AND EFFICIENT FINITE ELEMENT FOR PLATE BENDING 1533

r
k:2) = -
KG~
h -1
1
h/2
h/2
h2/3
-h/2
-1
-h/2
1
h/2 1
h2/6
-h/2
1
lh/2 h2/6 -h/2 h2/3 J

where kb is the bending stiffness, and k:’) and k!2’ are the one-point and two-point quadrature
shear stiffnesses, respectively. It is easily verified that the rank of k:” is one and the rank of k!” is
two. In the latter case, kb is completely dominated by the shear stiffness, as the following simple
example illustrates.
Consider the case of a one-element cantilever beam, subjected to an end load P and moment
M.

(i) One-point quadrature

Combining (2a) and (2b), eliminating appropriate rows and columns, and solving for the tip
displacement and rotation yields
w = (h2/4a +pP1)P+hM/2a, (34
8 = (hP/2+M)/a, (3b)
where
a = Et3/12h, (34
p = KGt/h. (34
In the thin beam limit (i.e., when p >>a),
(3a) becomes
w = h(hP/2+M)/2a, (4)
and (3b) remains unchanged. Thus we are left solely with the deformation due to bending as is
right.

(ii) Two-point quadrature

Carrying out the same steps as in case (i), with (2c) in place of (2b), yields

w=(
+
py )
h2p/3 P+ hM/2y,

8 = (hP/2+M)/y,
where
y=a+h2P/12. (5c)
In the thin beam limit (5a) and (5b) become
w = (4P + 6M/h)/P, (64
8 = 6(hP+2M)/h2P, (6b)
respectively. In this case only deformations due to shear are in evidence and (6a) and (6b) are
o(tP2) in error.
In passing we note that there are some circumstances in which the present element may have
some practical value. For example, an axisymmetric shell version might be useful for shell
1534 T. J. R. HUGHES, R. L. TAYLOR AND W. KANOKNUKULACHAI

covered solids in which bilinear elements are used to model the solid. The fact that only one
quadrature point is involved may lead to more economical computations in non-linear analysis.

BILINEAR PLATE BENDING ELEMENT

The strain energy for an isotropic, linear elastic plate, including shear deformation, is
UW, el, 0), =

A
where x1 and x 2 are Cartesian co-ordinates, w is the transverse displacement, 8, and 0' are the
rotations about the x1 and x 2 axes, respectively, E is Young's modulus, v is Poisson's ratio, K is
the shear correction factor, t is the plate thickness and A is its area. The first integral in (7)
represents the bending energy and the second represents the shear energy. We consider a
four-node quadrilateral element and assume the displacement and rotations are expanded in
independent bilinear shape functions. The isoparametric concept is employed.' This results in
three degrees-of-freedom-one displacement and two rotations-at each of the corners.
For very thick plates two-by-two Gaussian quadrature leads to acceptable results, however,
for thin plates it causes 'locking' as indicated for the beam in the previous section. In this case we
use two-by-two Gaussian quadrature on the bending energy term and one-point Gaussian
quadrature on the shear energy term. This results in one constraint per element. In large meshes
there are approximately three equations per element, thus there is no danger of the mesh
'locking'. As is apparent, the proposed element is extremely simple, and easily and concisely
coded. We are certain that the element routines are faster than any other plate bending element
yet proposed. In the next section we will show that the element is also surprisingly accurate.

NUMERICAL EXAMPLES: THIN PLATES

In this section we present several numerical examples which have become more or less standard
ones for evaluating plate elements. All computations were performed on a CDC 6400 computer
in single precision. (A single precision word consists of 60 bits on the CDC 6400.)

Square plate
The data for this example consists of the following (see Figure 2):
E=10-92X105 v=0*3 t=0*1 L=10
Both simply supported and clamped boundary conditions were considered as well as concen-
trated and uniformly distributed loadings. Results are presented in Tables I11 and IV for K
values of 1000 and 5/6. The former value is set to artificially maintain the Poisson-Kirchhoff
constraint. Due to the fact that the plate is rather thin (L/t = loo),there is little difference in the
results for the two values of K . In fact, the bending moments are identical. In practical situations
there seems no point in exceeding the 'natural' value of K = 5 / 6 .
 A SIMPLE AND EFFICIENT FINITE ELEMENT FOR PLATE BENDING 1535

Figure 2. Square plate. Due to symmetry only one quandrant is discretized

Table 111.Normalized centre displacement and bending moment for a simply supported square plate

Number of Displacement- Displacement- Moment-

elements concentrated load uniform load uniform load

4 0.9922 0.9770 0-851

16 0.9948 0.9947 0.963
64 0.9982 0.9987 0-991

(a) K = 1000

Number of Displacement- Displacement- Moment-

elements concentrated load uniform load uniform load

4 0.9957 0.9782 0.851

16 0.9991 0.9960 0.963
64 1.0034 0.9997 0.99 1

(b) K =:

Moment and shear resultants, and displacements are plotted in Figure 3 along the line x1 = 0,
for the clamped, concentrated load case in which K = 5/6. Along x1 = 0 and x2 = 0, the x1 and x2
components of the moment are equal, as are the x1 and x2 components of the shear.
The simply supported concentrated load case has, it seems, taken on the role of the
pre-eminent comparison problem for bending elements. In Figure 4 the present element, with
K = 1000,is compared with data taken from Gallagher.’ The good convergence of the element is
evident.

Clamped circular plate

The data for this example is the same as for the previous problem except (see Figure 5 ) :
R=5 t10.1
1536 T. J. R. HUGHES, R. L. TAYLOR AND W. KANOKNUKULACHAI

Table IV. Normalized centre displacement and bending moment for a clamped square plate

Number of Displacement- Displacement- Moment-

elements concentrated load uniform load uniform load

4 0.8652 0.9535 0,822

16 0.9650 0.9850 0.955
64 0.9920 0.9937 0.986
(a) K = 1000
~ ~~~~

Number of Displacement- Displacement- Moment-

elements concentrated load uniform load uniform load

4 0.8720 0.9575 0.822

16 0.9748 0.9890 0.955
64 1.0034 0.9976 0.986

(b)K=z

t
o 4 ELEMENTS
x
-64
16 ELEMENTS
ELEMENTS /
M 0.1 -

0 -)-I
1.0

0
2.0

-
1
- 3.0 4.0
I *
5.0
XI

-0.1 -

Figure 3. Moment ( M ) , shear resultant (Q) and displacement (w)versus edge co-ordinate (XJ for a
clamped square plate subjected to a concentrated load
 A SIMPLE AND EFFICIENT FINITE ELEMENT FOR PLATE BENDING 1537

.:-I
I

CONFORMING
QUADRILATERAL . MIXED FORMULATION [251

'
SUBDOMAIN
I I FORMUL! \TION [ e l LINEAR DISPLACEMENT,

?I-
0 CONSTANT MOMENTS
-16-TERM POLYNOMIAL [20] .6-TERM (QUADRATIC)
-21-TERM (QUINTIC) POLYNOMlA L 1211
Cn 30 POLYNOMIAL [lo] . MIXED FORMULATION
n PRESENT ELEMENT LINEAR MOMENTS
AND DISPLACEMENT [Is]
r NONCONFORMING [6]
' 12-TERM POLYNOMIAL [5]
MIXED FORMULATION
QUADRATIC MOMENTS
AND DISPLACEMENT [I91

W
0
Z 10
- *OI
.I \
\
\
EQUILIBRIUM STRESS
FIELD FORMULATION 1261
(LINEAR STRESS
FUNCTION FIELDS)
MIXED FORMULATION
QUADRATIC DISPLACEMENT,
LINEAR MOMENTS [ZA
LT
0 10-TERM (CUBIC) PLUS
CORRECTION MATRIX 1241
EW o 10-TERM (CUBIC) PLUS
CONSTRAINTS 1221

I- ASSUMED STRESS HYBRID

LINEAR DISPLACEMENT,
6
0
-10 CONSTANT MOMENTS C281
COMPATIBLE DISPLACEMENT
[r FORMULATION c71
W (9-TERM POLYNOMIAL
PER SUBDOMAIN)
a--20 CONFORMING [6]
RAZZAQUE A-9 [23]
NUMBER OF ELEMENTS
Figure 4. Simply supported square plate subjected to a concentrated load. Comparison of centre
displacement for various bending elements

t - - - - - - R F
3 ELEMENT 12 E L E M E N T 48 ELEMENT

Figure 5. Finite element meshes for clamped circular plate. Due to symmetry only one quadrant is
discretized
1538 T. J. R. HUGHES, R. L. TAYLOR AND W. KANOKNUKULACHAI

Results are presented in Table V for concentrated and uniform loadings, and K values of 1000
and 5/6.
Again, due to the thinness of the plate, there is little difference in the displacement results for
the two values of K, and the moment results are identical.

Table V. Normalized centre displacement and bending moment for a clamped circular plate

Number of Displacement- Displacement- Moment-

elements concentrated load uniform load uniform load

3 0.9197 0-8587 0-827

12 0.9579 0.9535 0.957
48 0.9883 0.9888 0.990

(a) K = 1000

Number pf Displacement- Displacement- Moment-

elements concentrated load uniform load uniform load
~~ ~ ~~

3 0.9267 0.8621 0.827

12 0.9674 0.9570 0.957
48 1.0005 0.9925 0-990

NUMERICAL SENSITIVITY DUE TO EXTREME THINNESS

The results of the previous section indicate that despite the simplicityof the present element it is
quite accurate. However, precaution must be taken when employing elements derived from the
reduced integration concept. This admonition stems from the observation that the shear stiffness
is O((h/t)*) times the oending stiffness. (In the case of a quadrilateral bending element h may be
taken to be the length of the longest side.) For fixed h, as t + 0, it is only a matter of time before
the effect of the bending stiffness vanishes completely due to the finite computer word length.
Results of this kind can be seen for the beam element in Figure 6 and for the plate element in
Figure 7. The plateaus represent the appropriate solutions for the meshes in question in the
‘thin’limit. Deterioration of the numerical solution begins to occur at L / t = lo4in the case of the
beam, and L / t = lo5for the plate; this corresponds to element aspect ratios (i.e., h / t ) of 104/16
and 105/8, respectively. It is unlikely that aspect ratios larger than these values will be met in
practice. However, on computers with shorter single precision word length, deterioration will
commence at smaller aspect ratios. Here it is important to employ a strategy which circumvents
these difficulties.This can be done as follows: Determine the maximum element aspect ratio for
which good results are obtained by plotting graphs similar to Figures 6 and 7. Before combining
the shear and bending contributions of the element stiffness, test the aspect ratio. If it is less than
the value for which good results are obtained, combine in the usual way. Otherwise, multiply the
shear stiffness by (t/h)’ times the square of the maximum element aspect ratio allowed, then
combine. This will reduce the disparity between the bending and shear term to an acceptable
level, yet maintain thin element behaviour. Numerical results illustrating these ideas are
depicted in Figures 6 and 7. The maximum allowable aspect ratios were determined to be
104/16 for the beam element and 105/8 for the plate element, from Figures 6 and 7,
 A SIMPLE AND EFFICIENT FINITE ELEMENT FOR PLATE BENDING 1539

1.04

1.03

1.02
I- \
z \
!$ 1.01 - '\
W Q
V
a
J 1.00
a
E
0.99
5 I
I-
I
$ 0.98
N
-I
Q: 0.97
I
U
0
0.96 - i
I
I
0.95 I I I I I I II
4 lo2 lo4 lo6 lo8
L/t

Figure 6.16 element cantileverbeam subjected to tip load. Normalized tip displacement versus aspect ratio

1 . 1 00

1.075
?
I
I
I I
W 1.050 I
u I
a I
J
a 1.025 I
?! ! 4,
0
1.000
W
I-
z
W
0.975
u
\
0 0.950 I
W
N I
J 0.925 I
a I
5 I
5 0.900 I
I
z I
b

10 50 lo3 lo5 10'

L/t

Figure 7.16 element model of simply supported square plate subjected to uniform load. Normalized centre
displacement versus aspect ratio
1540 T. J. R. HUGHES. R. L. TAYLOR AND W. KANOKNUKULACHAI

respectively. Employing these values in the procedure described above enables the plateaus to
be extended to the higher aspect ratios, as illustrated in Figures 6 and 7. This process enables the
reduced integration procedure to be applied successfully in cases involving arbitrarily large
aspect ratios.
Of course, another way to avoid difficulties is to work in double precision on short word
cbmputers.

APPLICATION TO THICK PLATES

One-point Gaussian quadrature on the shear term has a decidedly beneficial effect in application
to thin plates. However, the opposite is true for very thick plates. The difficulties do not always
manifest themselves. For example, results for the uniformly loaded clamped circular plate are
acceptable (See Figure 8; 4 refers to the magnitude of the load). For t / R = 0.4 the results are of
about the same level of accuracy as those in Reference 18, where a higher-order element is
employed. On the other hand, results for the same plate subjected to a concentrated load tend to
oscillate about the exact solution (see Figure 9). This problem is a trying one as the exact
Reissner's theory solution consists of an infinite displacement under the load, viz.

bending shear
PR r
p=-[-ln;] R
41rD R

- EXACT SOLUTION
( R E I S S N E R ' S THEORY 1
X O N E - P O I N T SHEAR
INTEGRATION

64D
W-
qR4
Figure 8. Clamped circular plate subjected to uniform load (48 element model). Comparison of underin-
tegrated shear element with exact solution
 A SIMPLE AND EFFICIENT FINITE ELEMENT FOR PLATE BENDING 1541

+
r/R

t/R =0.4 Oe10

I/ h C 3.2
0.20
-(RE
EXACT SOLUTION
ISSNERS THEORY)

o MODIFIED ONE-POINT
16rD SHEAR INTEGRATION 4sD
W‘m D TWO-BY-TWOSHEAR
[NTEGRATION
p,,

t /R = 0.2
t / h z 1.6

0. I
t / R = 0.08
I.o t / h ~ 0 . 6 4 0.2

0.3

0.5 1.0
/1 r/R

I / R = 0.02
t/h ~ 0 . 1 6
0.3
EXACT
(THIN PLATE)
1.0 0.4 EXACT
(THIN PLATE 1

Figure 9. Clamped circular plate subjected to concentrated load (48 element model). Comparison of
underintegrated shear elements with exact solution

where P is the concentrated force and D is the bending rigidity. As the plate thickness is reduced
the oscillations are lessened. From these results we conclude that when the t/h ratio exceeds
unity, the one-point Gaussian quadrature of the shear term should be abandoned in favour of
the following scheme: Two-by-two Gaussian quadrature should be used on the (aw/ax,)’ and
( I ~ w / & ~contributions
)’ to the shear energy. The remaining terms in the shear energy should be
integrated as usual by one-point Gaussian quadrature. We refer to this element as the ‘modified’
one-point shear element.
1542 T. J. R. HUGHES, R. L. TAYLOR AND W. KANOKNUKULACHAI

However, we note that even in the unmodified case the stresses are very good and the values of
the displacements at the element midsides (equivalently the node-to-node averages) are also
very good.
A spectral analysis of the element stiffness, when one-point Gaussian quadrature is employed
on the shear term, reveals that there are five zero eigenvalues-two more than the usual three
rigid body modes. Thus the element by itself is rank deficient, but this only manifests itself in
problems for very thick plates and here only in certain cases. The two additional zero-energy
modes are illustrated in Figure 10. The first mode consists of fI1 = f12 = 0 and an 'hourglass'
pattern for w . * ~Modifying the one-point shear integration as indicated above removes the
hourglass mode and leads to good results for very thick plates as evidenced by Figure 9. The
second mode consists of w = 0 and an in-plane twisting of the plate. In a mesh in which the rigid
body modes are removed, this pattern cannot persist and thus causes no further rank deficiency.

f x2

'XI

W'XIX2 w=o
8,=8,=0 8, = - x p , 8 2 ' X I
MODE I HOURGLASS MODE MODE 2 IN-PLANE T W I S T MODE

Figure 1O.Zero-energymodes of underintegrated plate bending element

CONCLUSIONS
We have presented an element for the bending of thin and moderately thick plates which
involves minimal programming, is highly efficient and competitively accurate. Due to these
attributes the element offers an attractive basis for non-linear developments. Numerical
sensitivity in applications involving extremely thin elements has been shown to be avoidable by
employing a simple computational strategy. Very thick plates may be successfully analyzed by a
slight modification to the element.

ACKNOWLEDGEMENT

The authors would like to thank T. Shugar of the Civil Engineering Laboratory, Port Hueneme,
California, L. Ovenshire of the National Highway Tr&c Safety Administration of the Depart-
ment of Transportation and K. Saczalski of the office of Naval Research, for continued support
and interest.
 A SIMPLE AND EFFICIENT FINITE ELEMENT FOR PLATE BENDING 1543

REFERENCES
1. 0.C. Zienkiewiu, 7’he Finite EZement Method in Engineering Science, McGraw-Hill, New York, 1971.
2. R. H. Gallagher, Finite Element Analysis Fundamentals, Prentice-Hall, Englewood Cliffs, N.J., 1975.
3. R. D. Cook, Concepts and Applications of Finite Element Analysis, Wiley, New York, 1974.
4. B. M. Irons and K. J. Draper, ‘Inadequacyof nodal connections in a stiffnesssolution for plate bending’, AZAAJ., 4,
961 (1965).
5. R. J. Melosh, ‘Basis for derivation of matrices for the direct stiffness method’, AZAA J. 1, 1631-1637 (1963).
6. G. P. Bazely, Y. K. Cheung, B. M. Irons and 0.C. Zienkiewicz, ‘Triangularelements in plate bending--conforming
and non-conforming solutions’, Roc.Conf. Matrix Meth. Smct. Mech., Wright-Patterson A.F.B., Ohio, 547-576
(1965).
7. R. W. Clough and J. L. Tocher, ‘Finite element stiffness matrices for analysis of plate bending’, Roc. Conf. Matrix
Merh. Struct. Mech., Wright-Patterson A.F.B., Ohio, 515-545 (1965).
8. B. Fraeijs de Veubeke, ‘Aconformingfinite element for plate bending’, Znt. J. Solidsand Struct.,4,95-108 (1968).
9. K. Bell, ‘A refined triangular plate bending finite element’, Znt. J. num.Meth. Eng., 1, 101-122 (1969).
10. G. R. Cowper, E. Kosko, G. Lindberg and M. Olson, ‘Staticand dynamic applications of a high precision triangular
plate bending elemedt’, AZAA J., 7 , 1957-1965 (1969).
11. G. Butlin and R. Ford, ‘A compatible triangular plate bending finite element’, Znr. J. Solids and Struct. 6,323-332
(1970).
12. T. H. H. Pian, ‘Element stiffness matrices for boundary compatibility and for prescribed boundary stresses’, Roc.
Conf. Matrix Meth. Struct. Mech., Wright-Patterson A.F.B., Ohio, 457-478 (1965).
13. G. Wempner, J. T. Oden and D. Kross, ‘Finite element analysis of thin shells’, Roc.ASCE, J. Eng. Mech. Div. 94,
1273-1294 (1968).
14. J. A. Stricklin, W. E. Haisler, P. R. Tisdale and R. Gunderson, ‘A rapidly converging triangular plate bending
element’, AZAA J., 7 , 180-181 (1969).
15. I. Fried, ‘Shear in and C’plate bending elements’, Int. J. Solids and Smct. 9,449-460 (1973).
16. I. Fried, ‘Residual energy balancing technique in the generation of plate bending finite elements’, Computers and
Structures,771-778 (1974).
17. I. Fried and S.K. Yang, ‘Triangular, nine-degree-of-freedom, Co plate bending element of quadratic accuracy’,
Quart. Appl. Math., 31,303-312 (1973).
18. 0. C. Zienkiewicz, R. L. Taylor and J. M. Too, ‘Reduced integration technique in general analysis of plates and
shells’, Znt. J. num. Meth. Eng., 3, 575-586 (1971).
19. 3. Bron and G. Dhatt, ‘Mixed quadrilateral elements for bending’, AZAA J., 10, 1359-1361 (1972).
20. F. K. Bogner, R. L. Fox and L. A. Schmit, ‘The generation of interelement, compatible stiffness and mass matrices
by the use of interpolation formulas’, Roc.1st. Conf.Matrix Meth. Struct. Mech., Wright-Patterson A.F.B., Ohio,
397-443 (1965).
21. L. S. D. Morley, ‘The constant-moment plate bending element’, J. Strain Analysis, 6,20-24 (1971).
22. J. W. Harvey and S. Kelsey, ‘Triangular plate bending elements with enforced compatibility’, AZAA J., 9,
1023- 1026 (1971).
23. A. Q. Razzaque, ‘Program for triangular elements with derivative smoothing’, Znt. J. num. Meth. Engng., 6,
333-344 (1973).
24. F. Kikuchi and Y. Ando, ‘Some finite element solutions for plate bending problems by simplified hybrid
displacement method’, Nuc. Eng. Design, 23, 155-178 (1972).
25. L. R. Herrmann, ‘Finite element bending analysis of plates’, J. Eng. Mech. D i n 94, 13-25 (1968).
26. Z. M. Elias, ‘Duality in finite element methods’, J. Eng. Mech. Div. 94,931-946 (1968).
27. W. Visser, ‘A refined mixed-type plate bending element’, AZAA J.,7 , 1801-1802 (1969).
28. R. D. Cook, ‘Some elements for analysis of plate bending’, J. Eng. Mech. Div. 98, 1452-1470 (1972).
29. S. W. Key, ‘A finite element procedure for the large deformation dynamic response of axisymmetric solids’,
ComputerMeth. in Appl. Mech. and Engng., 4, 195-218 (1974).