INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING, VOL. 1.

101-122 (1969)

A REFINED TRIANGULAR PLATE BENDING FINITE

ELEMENT
KOLBEIN l3EL.L
Division of Striictitral Mechanics.
The Technical University of Norway. Trondheim, Norway

SUMMARY
The derivation of the stiffness matrix for a refined, fully compatible triangular plate bending finite
element is presented. The Kirchhoff plate bending theory is assumed. Six parameters or degrees of
freedom are introduced at each of the three corner nodes resulting in a n 18 degree of freedom element.
This refined element is found to give better results for displacements and particularly for internal
moments than any plate bending element, regardless of shape, previously reported in the literature.

INTRODUCTION
The finite element method originally developed by Turner and others’ has proved to be a
convenient and powerful technique for the approximate analysis of problems of continuum
mechanics. A comprehensive presentation of the method and its many applications has been
given by Zienkiewicz.’
The finite element concept is essentially a generalization of standard structural procedures to
two- and three-dimensional structures. In a plate bending problem the plate structure is idealized
as an assemblage of a finite number of appropriately shaped elements interconnected at a finite
number of nodal points or nodes. As in the case of simple structures, the analysis involves the
evaluation of the relationship between element nodal displacements and corresponding element
nodal forces. This relationship is conveniently expressed in terms of the element stiffness matrix,
the evaluation of which is the key step of the finite element displacement method.
Of the various shapes of plate bending finite elements the triangle is the most attractive because
it makes it possible to treat plates with irregular boundaries. With triangular elements it is easy
to vary the element size in the vicinity of stress concentrations, and it is possible to approximate
the geometry of arbitrary doubly curved shell structures.
Many investigators have concerned themselves with the problem of determining suitable
stiffness characteristics for triangular plate bending elements. An excellent summary has been
given by Clough and Toche? who also present the stiffness matrix for a compatible 9 degree
of freedom element. Zienkiewicz and others4 using a somewhat different approach also construct
stiffness matrices for compatible triangular plate bending elements with 9 degrees of freedom.
With these elements the displacements can in most cases be determined with satisfactory accuracy
when fine mesh systems are used. However, the accuracy of the results obtained for the internal
moments is more doubtful.
Hellan’ and Herrmand have developed a simplified method, which is not a stiffness method,
for the analysis of plate bending by triangular elements. This method appears to give reasonably
good results for fine mesh idealizations.
Results obtained with a refined rectangular plate bending element,’ as well as those obtained
with a refined triangular plane stress elernent,’~~ seem to indicate that there is much to be said
for more refined element stiffness matrices.
Received I5 June 1968
101
102 KOLBElN BELL

The purpose of this paper is to present the development of the stiffness matrix for a refined
triangular plate bending element which ensures displacement and slope compatibility along the
edges of adjacent elements. In addition to the deflection and the two rotation components, the
three curvature components are introduced as parameters at each of the three corner nodes,
thus resulting in an 18 degree of freedom element. This element is obtained by eliminating three
degrees of freedom associated with three midside nodes of a 21 degree of freedom element which
in turn is based on a displacement field assumed as a complete fifth degree polynomial in x and y.

THIN PLATE THEORY

The notation and sign conventions used in this paper are shown in Figure 1, where all forces
and moments shown act in the positive direction. It should be noted that the lateral displacement
w is taken as positive in the downward direction, i.e. in the direction of negative z.

Figure 1. (a) Plate coordinate system and displacement notation;

(b) Notation for stress resultants and loading

The classical theory of thin elastic plates (Kirchhoff's theory) is well known" and need not
be presented in any detail. It is based on assumptions which imply that the deformed state of
the plate can be described entirely by the lateral displacement w of the middle surface.
Introducing the notation

- -- w , x a2
dW
-
a x 2 - w,xx
azw
-- etc.
ax
9 9
axay - w,xy
 A REFINED TRIANGULAR PLATE BENDING FINITE ELEMENT 103

the moment-curvature relation may be expressed as

where
Eh3
D=
12(1- v2)

is the flexural rigidity of the plate. E is the modulus of elasticity and v is Poisson’s ratio.
With matrix symbols* equation (la) may be written as

m=-Dc

ELEMENT STIFFNESS ANALYSIS

The element stiffness relation may be expressed as

where k is the element stiffness matrix. The vector v contains a set of selected displacement
parameters defined at the nodal points of the element (they may be regarded as kinematic degrees
of freedom), and the vector S contains the nodal forces corresponding with v.
Methods of calculating element stiffness matrices have been discussed by a number of
authors.”.” However, in order to establish the notation used in this paper and also for the
sake of completeness, an outline of the standard element stiffness analysis procedure will be
presented.
Assuming the Kirchhoff plate bending theory to be applicable the deformed state of the plate
element is completely defined by the lateral displacement, which in terms of assumed displacement
patterns may be expressed as

or

where 4i = 44(x, y) (i = I , 2, 3, . . .) are assumed displacement patterns having amplitudes

qi which may be regarded as generalized displacement parameters. The superscript T denotes
the transpose of a vector or a matrix.

* Vectors and matrices are represented by bold-type characters.

104 KOLBEIN BELL

The curvatures, defining the element strains, are

"=Pq
c = [ 2w9, (4)

where the matrix P is obtained by appropriate differentiation of +.

The internal moments are found by substituting equation (4) into equation (lb).

m= - DPq

Let the element be subjected to a virtual displacement field

Equating the external and the internal work done during this displacement yields

4'Q = - JETmdA
A
or, by use of equations (4) and (5),

4'Q = eTJPTDP dA q
A

where the vector Q contains a.set of generalized forces corresponding with the displacements q.
A denotes the area of the element. The above equation is valid for any q. Hence

Q = k,q
in which
k, = [P'DP dA
i
is the generalized element stiffness matrix.
The relationship between the nodal displacement parameters and the generalized displacements
may be written as
v = Aq (8)
where the matrix A is obtained by merely substituting the coordinates of the nodal points into
the displacement functions (equation 3) and their appropriate derivatives.
From equation (8)

where

The desired element stiffness matrix is now defined by the following transformation

k = BTkqB
 A REFINED TRIANGULAR PLATE BENDING FINITE ELEMENT 105

As the matrix D is a symmetrical matrix (equation la) it follows from equations (7) and (11)
that k is a symmetrical matrix.
From equation (10) it follows that A must be a square, non-singular matrix. This implies that
the number of assumed displacement patterns must be equal to the number of nodal displacement
parameters (number of nodal degrees of freedom), and also that the functions bi(i = I , 2, 3 . . .)
are linearly independent of each other.
It is evident that the accuracy of the finite element solution is directly dependent upon the
extent to which the assumed displacement patterns or functions can approximate the true
distortions of the plate, and the deformations will not necessarily converge to the correct values
as the element mesh size is successively reduced unless the displacement functions are properly
chosen.
The basic requirement which the assumed displacement functions must satisfy is what might
be called the completeness criterion, i.e. they must be capable of describing all possible rigid
body displacements of the elements as well as representing constant strain conditions (constant
curvatures) within the elements.’ In the case of polynomial displacement functions the complete-
ness criterion requires a complete quadratic polynomial to be present in the assumed expression
for w. If the assumed displacement functions in addition to satisfying the completeness criterion
also provide for complete displacement and slope compatibility, i.e. single-valued deflection
and slopes within the elements as well as along the boundaries between adjacent elements, the
finite element solution will provide a lower bound to the strain energy of the system and the
results will converge toward the true state of deformation as the mesh size is redu~ed.’~~’
The finite element solution should be invariant, i.e. independent of the position of the external
(global) reference system. The displacement functions should therefore be independent of
the orientation of the element with respect to the coordinate axes to which the functions are
referred.

COORDINATE SYSTEMS AND INTEGRATION FORMULAE FOR A

PLANE TRIANGLE
In recent works dealing with triangular finite elements, whether plate bending or plane stress
elements, the so-called area or triangular coordinates have been extensively used, partly in
order to avoid the two major difficulties encountered in cartesian coordinates, which are
1. Derivation in Cartesian coordinates leads very often to a lack of invariance.
2. Evaluation of the integral in equation (7) for all elements of the generalized stiffness matrix
can be an extremely tedious operation in Cartesian coordinates.
These difficulties can, however, be overcome without introducing a different coordinate system.
Invariance is automatically satisfied if a complete polynomial expansion in x and y is used to
describe the lateral displacement, and it will be shown that very simple integration formulae can
be found by making use of the corresponding formulae in area coordinates.

Cartesian coordinates
- - -
A plane triangle 1-2-3 lies in the 2-7 plane of a global Cartesian coordinate system x , y , z as
shown in Figure 2. The coordinates of the corners are GI, vl), (Fz,F,) and (& y3).
A ‘local global’ Cartesian system x, y , z is defined with the axes parallel to the corresponding
global axes 2,7, Z and with the origin located at the centroid of the triangle area.
106 KOLBEIN BELL

Figure 2. Global and local artesian system

The global coordinates of the centroid are

The local coordinates of the triangle corners may be expressed in terms of the global coordinates
as
-
xi = xi - xo
-
i = 1,2,3 (13)
Yi = Yi - FO
Hence
+ +
XI x2

Y,+Yz+Y3=O
x3 =0
I
The area of the triangle may be expressed in terms of the corner coordinates by the following
third-order determinant

(15)

Area coordinates
--
Any point P(x, y ) within the triangle 1-2-3 divides it into three sub-triangles as shown in
Figure 3. Let A , , A , and A , be the areas of these sub-triangles, the index designatingthe opposite
corner number.
The three area coordinates of P are defined as

Ai
=-
Wi i = 1, 2, 3
A
 A REFINED TRIANGULAR PLATE BENDING FINITE ELEMENT 107
-
Y b

Figure 3. Area coordinates

where A = A , + A , + A , is the total area of the triangle. It follows that

0 1 + +
0, 0 3 =1 (17)
The equation oi= constant represents a line parallel to the side opposite corner ‘i’. The
corner ‘i’ has the area coordinates w, = 1 and oi= = 0 (i # j # k).

Relationship bet ween Cartesian and area coordinates

The Cartesian coordinates are related to the area coordinates by the following matrix equation,

[i] =
1
[xl
Yl
1
x2
Yz
1

Y,
[z:)
0 3
(18)

where x and y are the Cartesian coordinates and w l ,w2 and w , the area coordinates of any
point P within the triangle or on the triangle sides. xi, yi (i = I, 2, 3) are the coordinates of the
triangle corners. Equation (18) is valid for any position of the Cartesian coordinates system.

Integration formulae
The integration over the triangle area of polynomial terms in area coordinates is independent
of the shape of the triangle and may be expressed as a fraction of the area, or
[oYfcoy’ 07 dA = p A (19)
A
where the indices i, j , and k represent any permutation of I, 2 and 3, and where the factor p
is

Consider the local Cartesian coordinate system x, y with the origin located at the centroid of
the triangle, Figure 2, and let
Pra= JYy’ dA (21)
A
108 KOLBEIN BELL

From equation (18) it follows that

For given values of r and s the right-hand side of this equation may be evaluated by the use of
equations (19), (20) and (14). The evaluation of Prs is straightforward but tedious and will not
+
be shown here. Formulae for P,, for orders n = r s = I to 6 are listed in Table I. It is seen
that these formulae are extremely simple for polynomial terms up to and including the 5th order.
Unfortunately the formulae for the 6th order terms cannot be expressed in the same simple
form. This inconvenience is, however, hardly noticeable when the formulae are programmed
for a digital computer.

Table I. Integration Formulae for a Triangle

Order P,, = X'Y' dA

n=r+s
A

I P,. = 0

Pz4,P15,and Po, are found by interchangingx and y in the expressions for P,,, P,, and Pro
respectively.

Note. x and y refer to a Cartesian coordinate system with the origin located at the centroid of the triangle.
 A REFINED TRIANGULAR PLATE BENDING FINITE ELEMENT 109

It should be noted that the formulae would become considerably more complicated for a
different position of the coordinate system.

REFINED ELEMENT STIFFNESS MATRIX

The refined element is based on a set of assumed displacement functions which form a complete
5th order polynomial expansion in x and y. This expansion contains the following 21 terms:
I
X Y
xz xy y’
x3 x2y xy2 y3
x4 x3y x’y’xy3 y4
xs x4y x3yz xzy3 xy4 y5
Consequently the element must have 21 degrees of nodal point freedom. A nodal point system
as shown in Figure 4 is chosen.

ty

Figure 4. Nodal point system and nodal parameters

In addition to the deflection w and the slopes w , and

~ w , the
~ three curvatures w , ~ w, , and
~
ww have been introduced as parameters at each of the three corner nodes (I, 2, 3). This gives a
total of eighteen nodal parameters or degrees of freedom, and the remaining three are the edge
normal slope w,, at each of three mid-side nodes (4, 5, 6). These side nodes, necessary at this
stage, are undesirable and will be eliminated later.
The assumed displacement functions together with the selected nodal parameters provide for
displacement as well as slope compatibility along the boundaries between adjacent elements.
Consider for instance two elements with a common edge having an arbitrary direction t. The
displacement w describes a quintic curve along the edge, and the edge normal slope w,, follows
a quartic curve. The nodal parameters, which are common to both elements, provide for six
‘boundary conditions’ for w, three (w, w,,and w,J at each end node. The displacement w along
the edge is therefore the same for both elements as a quintic curve is uniquely defined by six
constants. For the edge normal slope w,, the nodal parameters provide for five ‘boundary
conditions’, two (w,,and w,,J at each end node and one (w,”) at the midside node, which is
sufficient to completely define a quartic curve.
110 KOLBEIN BELL

In addition to complete displacement and slope compatibility being obtained, the curvatures
and consequently the internal moments are continuous between elements at the corner nodes
if the thickness is continuous.
Following the standard element stiffness analysis procedure, equation (3) becomes

w(x,y) = ( I x y x 2 . . . 2 y 3 xy4 y 5 ) (“i

921
= 4;lq21 (23)

where x and y refer to a ‘local global’ coordinate system with the origin at the centroid of the
element, as in Figure 4. It is evident that the completeness criterion is satisfied.
The matrix P of equation (4) is

... 2y3 o o
0 0 0 0.. . 6x2y 12xy2 20y3]
0 0 0 0.. . 12xy2 8y3 0

The generalized element stiffness matrix is defined by equation (7) as

kq21 = j-[pT,D p211 dA

A

The integration is carried out by integrating each element of the matrix [P:, D PZ1]. With the
formulae in Table I this is a straightforward operation, and explicit expressions for all elements
of kqzl are easily found. The expressions for some typical elements of the generalized stiffness
matrix read
(k,ZA 13 = (4P02 + 4vP20)D
(kq21)8 18 = [12y13 + (24 - ~OV)P~IID
2, = ~OOPO&
(kq21)~1

where the indices i and j of (kq2& denote row and column number respectively. There are
altogether 153 non-zero elements on and above the main diagonal of kqzl.
Equation (8) becomes
vz1= A21q21

The nodal displacement parameters are listed in the vector vZ1as

v21= where vi = and v j = { w , . } ~

j = 4, 5, 6

i = 1, 2, 3

in which the indices i a n d j denote local nodal point numbers, see Figure 4. Equation (26) may
be written as
 A REFINED TRIANGULAR PLATE BENDING FINITE ELEMENT 111

The sub-matrices Al, A2 and A3 are found by introducing the coordinates of the corner nodes
into the displacement functions and their appropriate derivatives. Hence

Ai = ‘ I xi yi X: .. . x i2y i3 X~Y: y’
0 I 0 2xi. .. 2xiy: yf 0
0 0 I 0 ... 3x;y: 4xiy; 5y:
0 0 0 2 ... 2y: 0 0
0 0 0 0 . .. 6xiy; 4y: 0
-0 0 0 0 . .. 6xTy, 12xiy: 20y;

In order to establish the three sub-matrices A4, AS and As (row-matrices) an expression for
the edge normal slope w,, must be found.
Consider first a rotation of the coordinate axes through an angle a about the z-axis, and let
the new axes be denoted by t and n as shown in Figure 5.

t’

Figure 5. Rotation of axes

The relationship between the two systems may be expressed as

+ sy
where the notation
t =

n = -sx
cx

+ cy
or
x = ct - sn

y = st + cn I
1
s = sina

c = cosa

has been introduced. Differentiation with respect to t and n gives

112 KOLBEIN BELL

and

Hence
a a a

and

From these two equations it follows that

C2

(34)

s2 - 2sc c2 -

Next, consider any one of the three sides of an arbitrary element. The axis normal to this
side must be given a positive direction. With reference to Figure 6, positive direction of the
axis nr ( j = 4, 5, 6) is defined by

0s
which implies that

7l 311
O S a j < 2- or -2S a j < 2 n

pos. fl

X X

Figure 6. Angle notation and sign convention

 A REFINED TRIANGULAR PLATE BENDING FINITE ELEMENT 113

The slope in the n-direction may, according to equation (33b), be expressed as

IV,, = -s w,z +c (35)

The last three sub-matrices of equation (28) are now easily determined. The transpose of Aj
( j = 4,5, 6) reads

A; = 0
- sj
ci
- 2s,xj

where sj = sin a,, cj = cos a, and (x,, y,) are the coordinates of the mid-side nodes (see Figure 6).
With the transformation matrix A l l established the desired element stiffness matrix kzl is
found by use of equations (11) and (10).
The element nodal forces S represent the actual loading on the element. In order to make
these concentrated forces statically equivalent to a distributed transverse load p(x,y) acting on
the element, a virtual displacement field

is imposed on the element. Equating the virtual work done by the distributed loadp(x,y) and the
virtual work done by the equivalent concentrated nodal forces S during the virtual displacement
gives

ST? = J'p(x,y) G(x,y) dA

A

= [p(x,y) cbT dA Bf
d

This equation is valid for any virtual displacement field. Hence

or in terms of the 21 degree of freedom element

A physical interpretation of the nodal forces is not called for and will not be attempted. It should
be noted that the load p(x,y) need not be uniformly distributed over the area. With the integra-
tion formulae in Table I it is also possible to treat a load having a hydrostatic distribution.
114 KOLBEIN BELL

In general side nodes are undesirable. They have a severe disadvantage in that they cause a
significant increase of the width of the band in the banded complete structure stiffness matrix.
The side nodes, each of which is common to only two elements, also have a disproportionately
large effect on the dimension of the complete structure stiffness matrix as compared to the corner
nodes.
In the case of the 21 degree of freedom element, which will be designated T-21, only three
degrees of freedom are associated with the mid-side nodes. A far more convenient 18 degree of
freedom element, designated T-18, may therefore be obtained without much loss of refinement
by eliminating the mid-side nodes of T-21. In order to maintain complete displacement and slope
compatibility the mid-side nodes are eliminated by imposing a cubic variation on the normal
slope w,, along the element edges. This is obtained by expressing w,. at the mid-side nodes in
terms of the corner parameters.

I m i

Figure 7. Variation of the normal slope along an arbitrary edge

The edge normal slope having a cubic variation along an arbitrarily oriented element edge
i-j, as shown in Figure 7,may be expressed as

w,,(t) = a, + a,t + a2t2 + ad3 (39)

Substituting the boundary conditions

into equation (39) yields

 A REFINED TRIANGULAR PLATE BENDING FINITE ELEMENT 115

Hence the normal slope at the mid-point 'm' may be expressed as

The nodal parameters at the corner nodes are, however, expressed in a local global coordinate
system x,y.

Let

where , s = sin a, and c, = cos a, (see Figure 6), and I, is the length of the element side on
which node number m lies (m = 4, 5, 6).
In the case of a cubically varying edge normal slope the value of w,,, at the midside node 'm'
may now, by use of equations (a), (34), (35) and (41), be expressed in terms of the parameters
at corner nodes 'i' and 'j' (local numbers) as

Positive direction of the axis n is defined in Figure 6. Hence equation (42) is valid only if xj > xi.
If xi < xi the signs associated with the curvature terms must be changed.
The relationship between the side node parameters v, and the corner node parameters v,
may be written as
v,= Hv, (43)
in which

where V~ (i = I, 2, 3) is defined by equation (27).

The transformation matrix H may be expressed in partitioned form as
116 KOLBEIN BELL

If x2 > x3 > x, then

It should be noted that care must be taken in connection with the signs of the K ~ ( " ' )and K ~ ( " ' )
terms. If for instance xl = x2 the signs in equation (46) are correct only if y , > y,. If x1 = x2
and y2 > y , the signs associated with the K ~ ( and
~ ) K ~ ( terms
~ ) must be changed (see Figures 6
and 7 and equation 40).
For T-2 1 the element stiffness relation already established is

s21 = k2lVZl (47)

The corresponding element stiffness relation for T-18 is

(48)
in which
(49)

The nodal forces S18are defined in such a way that

From equations (43) and (49) it follows that in the case of the edge normal slope varying cubically
along the element edges

Equations (50) and (51) imply that

and

where 118 is the 18 by 18 identity matrix.

Alternatively k , may be obtained directly from the generalized element stiffness matrix
kqzl.From equations (26) and (51) it follows that
 A REFINED TRIANGULAR PLATE BENDING FINITE ELEMENT 117

Hence

and k18 = GTk,,,G (54)

in which
(55)

kI8 is the element stiffness matrix for a fully compatible triangular plate bending element with
nodal points at the corners only and with 3 x 6 = 18 nodal degrees of freedom.
The assembly process, i.e., the Formation of the complete structure stiffness relation, and the
application of the boundary conditions have been discussed extensively el~ewhere,~.'~ and will.
not be dealt with here.
It should be noted that once the solution of the nodal displacement parameters has been
obtained the internal moments at the nodal points may be determined directly by use of
equation (la).
NUMERICAL RESULTS
Results obtained with a computer programme based on the refined 18 degree of freedom element

FK]
(T-18) for two illustrative examples are presented below. In the first example results obtained
with the 21 degree of freedom element (T-21) are included for comparison.

m Poisson's ratio: u = 0.3

-___
"2 3 C U
0 d

EDGES simply supported (SS) or clamped (C)

LOADING: uniform (U) or concentrated 01 centre (C)

A-1 A -2 A-3

Figure 8. Schedule of square

plate analysis and finite element
idealizations
A-4 B
118 KOLBEIN BELL

Example 1
Square plate of uniform thickness. Three combinations of support and loading conditions
were considered :
Case 1. All edges simply supported-uniformly distributed load p.
Case 2. All edges simply supported-concentrated load P at centre of plate.
Case 3. All edges clamped-uniformly distributed load p.
Poisson's ratio was taken to be v = 0.3.
The plate geometry and the finite element idealizations are shown in Figure 8. Because of the
double symmetry of the plate, only one quarter of it (a-b-c-d) was considered in the analysis.

Some representative values obtained with the refined elements T-18 and T-21 for deflection
and internal moments, are shown in Tables 11, 111, and 1V for Cases 1, 2, and 3 respectively.
The analytical solutions shown in Table I1 (Case 1) were obtained by a computer programme
based on a Navier solution, while those shown in Tables 111 and IV are given by Timoshenko."

Table 11. Square Plate Case 1

Deflection Bending moment Twisting moment

Mesh W M. = M , M*
at centre (pt. c.) at centre (pt. c) at corner (pt. a)
T-18 T-21 T-18 T-21 T-18 T-21
/

A- 1 4092 4057 4889 4916 3082 3167

A-2 4063 4062 4791 4795 3204 3232
A-3 4063 4062 4789 4790 3228 3241
A-4 4062 4788 3236
Analytical 4062 4789 3246
solution
Multiplier pP/D 10-5 pi= -10-6 pl'

Table 111. Square Plate Case 2

Deflection Twisting moment
Mesh W M,
at centre (pt. c) at corner (pt. a)
T-18 T-21 T-18 T-21
~

A- 1 1128 1138 649 1 5904

A-2 1153 1155 6152 6089
A-3 1157 1158 6104 6093
A-4 1158 6097
B 1160 608 1
Analytical 1160
solution
Multiplier PPlD -10-5 P

It is doubtful whether the analytical values in Table IV should be considered as exact values.
The results are largely self-explanatory. It is seen that the loss of accuracy resulting from the
elimination of the midside nodes of T-21 is very small.
 A REFINED TRIANGULAR PLATE BENDING FINITE ELEMENT 119

Table 1V. Square Plate Case 3

Deflection Bending moment Bending moment

Mesh W M,= M, M, at point b
at centre (pt. c) at centre (pt. c) (M, at point rf)
T-18 T-21 T-18 T-21 T-18 T-2 1

A- 1 1 14.9 124-6 225.7 265.7 392.6 567.9

A-2 126.4 126.5 2293 230.3 496.5 517-1
A-3 1263 126.5 229.1 229.3 510.2 513.9
A-4 126.5 229.1 512.3
Analytical 126 23 1 513
solution
Multiplier p14/D 10-4 pi2 -lo-‘ p1=

Example 2
Equilateral triangular plate of uniform thickness. The plate is simply supported along all
edges, and it carries a uniformly distributed load of intensity p. Poisson’s ratio was again taken
to be v = 0-3.
The plate geometry and the finite element idealizations are shown in Figure 9. Because of the
symmetry of the plate, only half of it was considered in the analysis.

IY
Edges simply supported
Uniform load p
Poisson’s ratio v = 0.3

- x

M-1 M-2
Figure 9. Plate geometry and finite element idealizations
120 KOLBEIN BELL

Results obtained with the 18 degree of freedom element for the deflection and the bending
moment (M, = Mu)at the centre of gravity of the plate (point ‘0’ in Figure 9) are shown in
Table V compared with the analytical solution given by Timoshenko.”

Table V. Equilateral Triangular Plate Analysed by T-18

Deflection Bending moment

Mesh W M, = M,
at point o at point a
M- I 10288 24068
M -2 10288 24074
Analytical 10288 24074
solution
Multiplier lo-’ pa4/D 10-6 pa2

From the results in Table V it is tempting to conclude that the analytical solution in this
particular case must be contained in the finite element solution. This is, however, not the case.
The analytical solution for the deflection as given in reference 10 reads

w=-[
64aD
x3
4
- 3y2x - a(x2 + y2) + - a 3
27
] 4
(9.2 - x2 - y2

Examining this expression it is seen that the slope w,, for example, describes a 4th order curve
in the y-direction. In the finite element solution, however, a cubic variation has been imposed
on the edge normal slope along all element boundaries. Theoretically speaking the two solutions
are therefore not identical.

CONCLUSIONS
The derivation of the stiffness matrix for a refined, fully compatible triangular plate bending
finite element has been presented. Six nodal parameters or degrees of freedom have been intro-
duced at each of three corner nodes, resulting in an 18 degree of freedom element.
Results obtained with this refined element show that it gives extremely good accuracy even
for very coarse mesh idealizations. This is true for both deflection and stresses. Perhaps the
most important feature of the refined element is the ease and accuracy with which it permits the
internal moments to be determined. This is achieved by the introduction of the curvatures as
nodal parameters. Another advantage obtained by having the cuwatures among the nodal
parameters is that they permit the boundary conditions to be satisfied more closely.
Very good convergence towards the correct solution is observed. In assessing the monotonic
convergence it is important to recognize that it is the strain energy of the structure which should
be examined. In Example 1, Case 2, the deflection under the concentrated load gives the strain
energy, and Table 111 indicates a monotonic convergence (from the stiff side). A strict assessment
of the monotonic convergence should omit the results obtained with meshes A-3 and B.
The derivation of the stiffness matrix follows standard procedures, and the development of
very simple formulae for the integral of polynomial terms over the triangle area has made it
possible to write explicit expressions for all elements of the generalized element stiffness matrix.
However, the evaluation of the refined element stiffness matrix does require the inversion of a
21 by 21 transformation matrix, which is a straightforward but time-consuming operation on
a digital computer. Although this inversion is a drawback, it does not appear to be a serious
 A REFINED TRIANGULAR PLATE BENDING FINITE ELEMENT 121

one. In terms of computer time approximately 2.7 seconds are needed to evaluate the element
stiffness matrix and the element nodal forces and to add them into the complete structure
stiffness relation. This time refers to a FORTRAN IV programme on a UNIVAC-I 107 computer.

ACKNOWLEDGMENTS
The work presented in this paper is based on the di~sertation'~ submitted by the writer to the
Department of Civil Engineering of the Technical University of Norway, in partial fulfillment of
the requirements for the degree of Licentiatus Technicae. The investigation was made possible
by a grant from the Technical University of Norway, and was conducted under the supervision
of Professor I. Holand.
APPENDIX
Notation
The following symbols are used in this paper:
1. Scalars
A = area of triangle.
a = plate dimension (Figure 9).
a, = constant (equation 39).
c = abbreviation for cos a.
D = flexural rigidity of the plate.
E = modulus of elasticity (Young's modulus).
h = plate thickness.
k = element of the stiffness matrix.
I
plate dimension (Figure 8).
=
I,,
= length of element side m.
M,,Mu, M,, = moment components in the x-y system.
n = direction normal to an element side.
P concentrated load.
prs J x' 4 dA,
A
intensity of distributed transverse load.
generalized displacement parameter.
abbreviation for sin u.
direction of an element side.
lateral displacement.
Cartesian coordinates relative to centroid.
global Cartesian coordinates.
angles (Figure 6).
parameter defined by equation (41).
area coordinates defined by equation (16).
factor defined by equation (20).
Poisson's ratio, and
assumed displacement pattern.
122 KOLBEIN BELL

2. Vectors and Matrices

A = transformation matrix defined by equation (8).
6 = inverse of A.
c = curvature vector defined by equation (4).
D = moment-curvature matrix defined by equation (1).
G = transformation matrix defined by equation (55).
H = transformation matrix defined by equation (43).
I = identity matrix.
k = element stiffness matrix.
m = moment vector defined by equation (1).
P = matrix defined by equation (4).
Q = generalized forces.
q - generalized displacement parameters.
s = element nodal forces.
v = element nodal displacement parameters, and
4 = vector containing assumed displacement functions.

REFERENCES
1. M. J. Turner, R. W. Clough, H. C. Martin and L. J. Topp, ‘Stiffness and deflection analysis of complex struc-
tures’, J. aeronaut. Sci., 23, 9 (1956).
2. 0. C. Zienkiewicz, The Finite Element Method in Structural and Continuum Mechanics, McGraw-Hill, London,
1967.
3. R. W. Clough and J. L. Tocher, ‘Finite element stiffness matrices for analysis of plate bending’, Proc. Conf.
Matrix Meth. Struct. Merh., Wright-Patterson Air Force Base, Dayton, Ohio (1965).
4. G. P.Bazeley, Y. K. Cheung, B. M. Irons and 0.C. Zienkiewia, ‘Triangular elements in plate bending4on-
forming and nonconforming solutions’, Proc. Con5 Matrix Meth. Struct. Mech.. Wright-Patterson Air Force
Base, Dayton, Ohio (1965).
5. K. Hellan, ‘Analysis of elastic plates in flexure by a simplified finite element method’, Acta polytech. scandi., Ci
46, Trondheim, Norway (1967).
6. L. R. Herrmann, ‘Finite element bending analysis for plates’, J. Engng. Mech. Div. Am. SOC.civ. Engrs,
ASCE, 93 (1967).
7. H. Hansteen, ‘Finite element displacement analysis of plate bending based on rectangular elements.’ International
Symposium on the Use of Electronic Digital Computers in Structural Engineering, Newcastle upon Tyne (1966).
8. C. A. Felippa, Refined Finite Element Analysis of Linear and Nonlinear Two-Dimensional Structures’, SESM
Report No. 66-22, University of California, Berkeley, California, 1966.
9. P. G. Bergan, Plane Stress Analysis Using the Finite Element Method. Triangular Element with 6 Parameters
in Each Node. Division of Structural Mechanics, The Technical University of Norway, Trondheim, 1967.
10. S. Timoshenko and S. Woinowsky-Krieger, Theory ofPlatesandShells, 2ndedn., McGraw-Hill, New York, 1959.
11. R. W. Clough, ‘The finite element method in structural mechanics’, in Stress Analysis (Ed. 0.C . Zienkiewia
and G. S. Holister), Wiley, London, 1965, Chap. 7.
12. B. Fraeijs de Veubeke, ‘Displacement and equilibrium models in the finite element method’, in Stress Analysis,
Chap. 9, cited above.”
13. R. G . Anderson, B. M. Irons and 0. C . Zienkiewicz, Vibration and Stability of Plates, Research Report No.
C/R/78/67, University of Wales, Swansea, 1967.
14. K. Bell, Analysis of Thin Plates in Bending Using Triangular Finite Elements, Division of Structural Mechanics,
The Technical University of Norway, Trondheim, 1968.

Note-Since the acceptance of this paper the following publication, which deals with an element basically the
same as the 21 degree of freedom element described in this paper, has come to our notice: ‘Ein neues,
vollvertragliches endliches Element fur Plattenbiegung,’ by W. Bosshard. Abhandlungen IVBH, Zurich. 1968,
pp. 1-14.