You are on page 1of 16

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING, VOL.

3, 275-290 (1971)

REDUCED INTEGRATION TECHNIQUE IN GENERAL


ANALYSIS OF PLATES AND SHELLS
0.c. ZIENKIEWICZ*
University of Wales, Swansea
R . L. TAYLOR?
University of California, Berkeley, California
AND

J. M. TOO$
University of Wales, Swansea

SUMMARY
The solution of plate and shell problenis by an independent specification of slopes and middle surface
displacements is attractive due to its simplicity and ability of reproducing shear deformation.
Unfortunately elements of this type are much too stiff when thickness is reduced.
In an earlier paper a derivation of such an element was presented’ which proved very successful in
‘thick’ situations. Here a very simple extension is made which allows the element to be economically
used in all situations.
The improved flexibility is achieved simply by reducing the order of numerical integration applied
to certain terms without sacrificing convergence properties. The process is of very wide applicability
in improvement of element properties.

INTRODUCTION
The conventional treatment of thin plate and shell structures based on applying the Kirchhoff
hypothesis defines fully the displacement pattern by the middle surface displacements. Great
difficulties arise in satisfying the necessary continuity of slopes at interfaces2 and in the inability
of such formulations to account for shear deformation.
To overcome such problems a fairly obvious artifice of avoiding the normality (Kirchhoff)
hypothesis and prescribing independently the middle surface displacements and rotations of the
normal could be adopted. Such elements have been proposed and used effectively in thick shell
situations by Melosh? Utku4 and others.*,”10 It was soon discovered, however, that in thin shell
and plate situations the new approach gave such a large stiffness as to make the use of such
elements quite uneconomical, and several devices for improvement were suggested and used.
Utku4 and Martin,5 for instance, use in this context a substitution of constants arrived at by an
intuitive reasoning. Such procedures are not easy to generalize and may indeed lead to non-
convergent results. A more acceptable alternative is that of constraining the element to obey
Kirchhoff conditions at a discrete number of points. This idea introduced by Wempner and
co-workers8,9has been elaborated further by others.l03l1 While now convergence is achieved and

* Professor of Civil Engineering, Dean of Applied Science.


t Associate Professor of Civil Engineering. Visiting Professor, University of Wales, Swansea.
Research Associate.
Received I October 1970
Q 1971 by John Wiley & Sons, Ltd.

215
276 0. C. ZIENKIEWICZ, R. L. TAYLOR AND .I.
M. TOO

the process can be generalized to any element its practical application is difficult. In this paper a
new alternative is presented which while possessing complete generality is much simpler to
implement. Its practical success is such as to make elements so formulated amongst the most
economical to use in practice f o r both thick and thin shell or plate structures.
The basis of the new development is in numerical integration applied to the evaluation of
stiffness matrices. That this is economical and indeed necessary in a variety of elements is now
well known. What perhaps is less widely known is the fact that
I . The stiffness of displacement elements is reduced as the order of numerical integration
decreases.
2. The convergence of numerically integrated elements is always guaranteed providing the
integration order allows sufficient accuracy for an exact evaluation of the element volume12
in isoparametric formulation (although the bounded nature of the solution is no longer
available).
If now in a plate or shell element numerical integration is adopted and its order reduced
generally or for particular stress components which may be ‘parasitic’ an improvement of its
behaviour will be achieved. Such a selective reduction of integration order was first proposed by
Doherty and co-workers13 in the context of a plane quadrilateral element. (See Figure 1.)
Noting that with the linear displacement field the element behaves badly in pure bending due
to ‘parasitic’ shear, the authors reduce the numerical integration for shear to that of one central
point and obtain excellent results. While convergence may only be proved in this case for
rectangular element forms the practical application of this development has been great.

_ - - - _-
,
r
*-

-
\
.
\
<‘ ,
/--
I.

( ,x-,.----
-_..\<< , ‘ ) ‘*
(b)
Figure 1. ‘Parasitic’ shear stresses induced in a linear element under bending mode. (a) Constrained
mode; (b) true mode

In the plates and shells of the class discussed here a very similar parasitic shear occurs
particularly in large aspect ratios (left hand side of Figure I ) assumed to represent now a plate
section. Hence reduction of the order of shear integration is imperative. Other similar reductions
and their effects will be discussed further.

THE BASIC ELEMENT FOR SHELLS AND PLATES


The element discussed here is that derived from a full three-dimensional isoparametric form by
prescription of linear displacement variation across the thickness and suppression of strain
energy due to stresses normal to its middle plane (Figure 2). Full details of derivation of such
elements are presented in a previous publication’ and will not be repeated here. The final element
has as its degrees of freedom three displacements of the middle plane and two rotations of the
normal about axes parallel to it at each node.
REDUCED INTEGRATION TECHNIQUE 277

Figure 2. (a) Parabolic isoparametric hexahedron; (b) degeneration to a shell element

The stiffness of the elements derived in Reference 1 is obtained by numerical integration of the
usual form

[see equations (7)-(16) of Reference I].


This integration was carried out originally using two Gauss points in the transverse direction 5
and either 3 x 3 or 4 x 4 mesh of Gauss points in the middle plane depending on whether parabolic
or cubic type of element was used. In this original form excellent results are obtainable for thick
plates and shells (i.e. when the aspect ratio of the elements is small relative to thickness) but
complete failure occurs in bending when thin sections are considered. Figure 3(a) shows a direct
application of elements of this type to a square plate. With fairly thick plate situations t / a = 0.1
the deformation represents well the results of Reissner theory (as will be seen later), but as the
thickness decreases the results do not tend as should be expected to the classical ones of thin plate
theory. Indeed, as the thickness is reduced progressively the stiffness increases-apparently
without a limit-and results are completely unreliable when t / a is below 0.01. At that stage the
aspect ratio of the element is some 1 : 12.5.

REDUCED INTEGRATION FOR TRANSVERSE SHEAR STRESSES


Clearly, unless subdivision is reduced to uneconomical limits, the element is not applicable to a
general plate situation. Inspecting the reasons for the ‘odd’ behaviour it is immediately realized
that the problem is of similar nature to that of ‘parasitic’ shear stresses shown in Figure 1. A
reduced integration order was thus applied selectively to the transverse shear component. In
Figure 3(b) results are shown for the reduction of such an integration to 2 x 2 Gauss points for a
parabolic element (8 nodes). Now, not only is a slightly larger deformation obtained for thick
situations but complete convergence to the thin plate result occurs as plate thickness is reduced
[Figure 3(c)]. This convergence occurs irrespective of the thickness reduction and indeed the
element appears to be exceedingly good as a thin plate element, while preserving its capacity to
render thick plate s01utions.l~Little improvement is achieved by finer subdivisions for any t / a
ratios.
278 0.C. ZIENKIEWICZ, R. L. TAYLOR AND J. M. TOO

Thin plate--
exact

A 0 2 4 6 10
2x/a
P

= Central deflection for thin plate theory


= 0 004062 qo
&-+
I = Plate thickness

D = Plate rigidity

3005 1 ,,,Transverse stgar


‘94-+-~ . reduced
Thin plote
.-
---,a
3 004
;
1

;-.Q

Original’elementq \

m4
40 0
0 003 I
-
Original element \ \
Reference 1 s\ (Divergence)
\

3 002

0 001 + r i

0’
I 1

I
4x4Mesh

I
I I I I I

0 10 20 40 70 100 200 400

Figure 3. A simply supported square plate under uniform load qo. (a) Plot of central deflection
for element of Reference 1 ; (b) with reduced transverse shear integration; (c) displacement at
centre versus thickness parameter
REDUCED INTEGRATION TECHNIQUE 279

The bending moments once again converge even for this coarse mesh to the results of thin plate
theory (Figure 4). (It is of interest to remark here that a better representation can be obtained if
these are evaluated at Gauss points rather than at nodes as shown).

,-Thin plate
,’’ theory
10 h

I
OO 10 2VC

0 01
o = Calculated at nodes
A = Calculated at Gauss points

Figure 4. Bending moment for example of Figure 3

Figures 5 and 6 show a further example of application of the new element designed to assess
its accuracy. In Figure 5(a) the case of a circular plate is presented under a uniform load solved
with a very coarse mesh of three curved elements as shown. Once again the wrong results given by
the unmodified element of Reference 1 are indicated. In Figure 6 a finer subdivision is used for
0 0.2 0-4 0.6 0-8 1.0
w W O
PO2

Reduced snear
intearation
0
0
I !. e
@ / ,,wxact (Reference 14)

OQ
2000-

Figure 5. Deflections of circular plate with c1amped;boundary under uniform load (three elements)
280 0. C. ZIENKIEWICZ, R. L. TAYLOR AND J. M. TOO

a concentrated load and compared with results of ReissnerI5 and KriegerZ2theories. The element
reproduces almost exactly the behaviour of Reissner’s theory which near the concentrated load
neglects the true ‘pinching’ effect taken into account by the three-dimensional solution of
Krieger.

-/A Reduced shear


121
138 A t / r =002
1 81 / / r = O 20
2 00 t / r = 0 2 Krieger

t / r = 0 4 Krieger P = 4x10
- - t / r = 0 4 Reissner

c- tx

Figure 6 . Deflections in circular plate with clamped boundary ander concentrated load at centre
of the plate (twelve elements)

SHELLS-FURTHER REDUCED INTEGRATION FOR ‘IN PLANE’ STRESSES


The improved performance of the element in bending described already was anticipated to be
of value in general shell problems. However, it was soon discovered that this improvement was
minor and could be overshadowed by a reduced integration for the other stress components.
As a test case a cylindrical shell roof was chosen as here both membrane and bending stresses
are of importance. The particular example originally solved by Scordelis and co-workers16has been
used frequently since for assessment of finite element perf~rmancel’-~~ and hence will be useful
here for comparison purposes. Figure 7 shows the geometrical and physical characteristics of the
problem and the meshes used in subdivision.
In Figure 8 some results for deflections under the action of self-weight are presented for
parabolic elements. ‘Exact’ results are those in Reference 16.
In Figure 8 the results of a direct application of the element of Reference I are shown together
with corresponding results obtained by a reduced order of integration applied to transverse shear
terms. Convergence of both sets towards the ‘exact’ solution is evident (as indeed it must be for
this element), but this is slow and approximation is poor. Improvement due to reduced shear
integration is, as expected, present but of a minor character.
REDUCED INTEGRATION TECHNIQUE 28 1

E = 3xIO3 KSI
v - 0-0
9' 0-09 K/Sq ft

edge

Meshes used

d 272 448
Figure 7. Test shell example of Reference 16. Self-weight behaviour
r
b

01
w (ft)

I O2
1
03

Reference 1 shear reduced -

C Y 0' ct suppcrt

Figure 8. Displacement (parabolic element), cylindrical shell roof


282 0. C. ZIENKIEWICZ, R. L. TAYLOR AND J. M. TOO

As the shape of the elements has a low aspect ratio presence of ‘parasitic’ in plane shears was
not anticipated to be serious initially-however, as a further experiment, all stresses were integrated
using a 2 x 2 set of Gauss points. The results are shown in Figure 8(b) and exceeded all reasonable
expectation. Even with one element deflections are of ‘engineering accuracy’ and for further
subdivision the results are indistinguishable from the ‘exact’ ones.
This development has a great advantage over the selective reduction of particular stresses
computationally. The program is maintained in its original form and the alteration simply results
in an over-all economy of four-ninths in stiffness computation time. Further, the element is
independent of orientation, and isotropic as well as anisotropic properties can be dealt with.
Total numbers degrees of freedom
100 200 300 400 500 600
I I I I I I

‘Exact’ (Reference (16)

A
m

Reference Element type


17 Flat plate
a 18 Curved
A 21 Curved
V 19 Curved
0 20 Flat plate
Isopararnatric elements
8 Original parabolic (Reference 1)
Parabolic Clement
@ with modified transverse shears
I
8 Parobolic element with modified
transverse and in plane shears
0 Simple 2 x 2 integration
m Original cubic (Reference 1)
~ Cubic element with modified
transverse shears
o Simple 3 x 3 integration
Figure 9. Cylindrical shell roof problem, convergence of deflection versus degrees of freedom

The same pattern of tests was applied to the cubic form of the element with the original 4 x 4
Gauss point integration reduced first for transverse shear and then for all stresses to a 3 x 3
pattern. The results are again essentially similar to those of the parabolic case although the
differences are not so dramatic as the element even in its original form gives more reasonable
results. In Figure 9 the convergence of both the elements each with the three variants is plotted
against the total degrees of freedom. For comparison the convergence of other element^^^-^^ used
in the analysis of the same problem is shown. With the all-round reduction of integration order
the present elements show a more rapid convergence and better accuracy than any others so far
used (at least for this problem).
It is of interest to note that the convergence is not now achieved ‘from below’ and is not mono-
tonic as indeed was anticipated as the ‘bounded’ nature of solution is lost.
REDUCED INTEGRATION TECHNIQUE 283

0 0

( K I P/ft)
.'/v

T"80
35 -8~ 325 0 025 y/c +
47
80
05

8 8 87 1

t
I 1 mwpn
Original element M&.
(a)
Simple 2x2

9
t- -1
40 30 20 10
(C)

40 30
(d)

Figure 10. Cylindrical shell roof. Resultant comparisons, parabolic element. (a) longitudinal
force (N,)at central section, y = 0 (interpolated values from Gauss points); (b) longitudinal
force (N,)at free edge, 4 = 40" (interpolated values from Gauss points); (c) transverse moment,
(interpolated values from Gauss points); (d) longitudinal moment, ML (interpolated
values from Gauss points)
284 0.C. ZIENKIEWICZ, R. L. TAYLOR AND J. M. TOO

Comparison of some stress resultants shown in Figure 10 gives the same type of improvement
as is demonstrated in deflection.
Two further points require attention:
1. Does the over-all reduction of integration order achieve comparable results with the
selective reduction in plate bending situations ?
2. Is the improvement evident when elements are distorted from their rectangular form?
Both are tested in repeating the solution of Figure 5 where indeed little or no difference is
observed from the previous reported results. Similar tests carried out on doubly curved shells
which will be reported elsewhere confirm the findings.

CONCLUDING REMARKS
By the simple device of reducing the order of integration for all stress components the elements of
Reference 1 have been shown to yield accurate results for both thick and thin plate and shell
situations. The aspect ratio of the elements can be increased almost without limit without
introducing excessive stiffness effects. Indeed the element thus achieved is amongst the most accurate
ones known and by being able to render correctly shear deformations may be termed ‘universal’.
By deriving its properties directly from three-dimensional concepts it is the simplest for use in
non-linear problems whether the non-linearity is due to material or geometric properties.
Extension to anisotropy or to sandwich construction is readily achieved and details will be
presented elsewhere.
With the importance of this element family likely to increase for practical purposes attention
must be given to reduction of the computation times for its formation. In part such a reduction is
already accomplished by reduced integration order-but further detail improvements are possible.
Two such factors are discussed in the Appendix.
The complete reasons for the success of the ‘reduced integration’ concept are not discussed in
this paper. The two facts, i.e. that convergence can still be proved and that approximation is
excellent, can stand up on their own. It is unlikely that a satisfying mathematical proof for
reasons of improvement can be presented for a general case but heuristic arguments can readily
be produced for various cases. To limit the length of the baper these are not here presented.
However, an obvious inference of the argumehts to all numerically integrated elements leads to
a speculation for a general improvement of their performance by the paradoxical reduction of
accuracy in integration. For instance, tests on simple plane stress problems already carried out
bear out this point.
AUTHORS’ NOTE

Since undertaking this investigation the authors have learned of an independent investigation
reporting similar findings. This was carried out under the direction of Professor R. H. Clough by
Dr. S. F. Pawsley in The Analysis of Moderately Thick to Thin Shells by the Finite Element Method,
SESM Report 70/12, University of California, Civil Engineering Department, Berkeley, 1970.
As the actual processes adopted differ in detail we are looking forward to seeing this published.

APPENDIX
Some economics in element ,formulation
With numerically integrated elements certain penalty has to be paid in computation time if
elements are complex. Thus, in programming, care has to be taken to utilize all possible short
cuts and reduce the computing time as much as possible.
REDUCED INTEGRATION TECHNtQUE 285

The new shell element is such that its accuracy is superior to any other currently available
elements and is, therefore, likely to be adopted as standard. Here we discuss some economics
which can be achieved in the appropriate programs in addition to the obvious economy
introduced by a reduction of the number of integrating points. To avoid repetition readers must
consult Reference I for appropriate equations.

Jacobian inverse [J]-l and direction cosines [el


The computation of strains, as defined by equations (10) and (14) in Reference 1 , involves the
evaluation of the inverse Jacobian matrix. This matrix is defined by equation (1 I ) of Reference 1.

where x5 = h x / h ( , etc.
By definition, we note that this can in fact be written in terms of three vectors

B
(3)

+
The first two vectors are tangential to the reference surface while V3 is the direction of the normal
to it (if I can be considered as normal to midside surface).
The inverse can now be written explicitly as
-+ - + + ++ -+
[J]-l = [T x V,, V, x S, S x TI t 11 J 11 * (4)
The first two columns are again vectors tangent to the reference.plane and the third normal
to it.
The direction cosine matrix is expressed as
PI = [Ti,02,V3I
where vector T3 has the same sense as the third vector in [J]-’.
As a consequence of normal condition of vectors we have

[A] = [elT.[J]-l

* I! Jll is equal to the determinant [J].


286 0.C. ZIENKIEWICZ, R. L. TAYLOR AND J. M. TOO

and by utilizing this special form of A, equation (14) in Reference 1 is then uncoupled into two
parts:

[e;ll,= A,bg vc wgl [el


By doing so, we can avoid the multiplication of zero terms and thus reduce the number of
operations in the program to save time.

Energy formulation and stifness matrix of an element


The energy expressions for an isotropic material are related to equations (6)-(9) of Reference 1.
Note that the elastic properties [D‘]and strain vector (E’} can also be decomposed into two parts
(in plane and transverse), i.e.

DI, j 0
_____ ___

Thus the strain energy can be expressed as


ID’] =
[ 0 iD:]

du = $[{&IT [D;,J{ELI+ {E:I~ID:] I}:E{ (8)


where the strain vector {E’} is obtained from equation (14) of Reference 1 and is taken the
following form:
a
- o o
axf
0 -
a 0.
aY‘

{El) = a -
- a o (9)
ay‘ ax!

-
a a
azt O z
a a
0 - -
ayt
azr

Transforming the local displacement vector to global frame we have

this equation is then expanded by using equation (3) of Reference 1 and equation (7):
REDUCED INTEGRATION TECHNIQUE 287

- - -
B, 0 0 0 0 0
0 B, 0 0 0 0
.[Bi] = B, Bl 0 ; [CJ= 0 0 0
0 0 B, c, 0 0
0 0 B,_ L 0 c, 0

To simplify expressions we will denote hereafter [vli,


-6,J as [&I.
By minimizing the strain energy of the element domain with respect to nodal displacement
parameters we can obtain the element stiffness matrix:

In constructing equation (14), two special features as can be seen in equations (11) and (12)
(i.e. identical terms and sparseness of the matrices) have been taken into account in performing
matrix multiplication.

Explicit integration
Approximate explicit integration through the shell thickness can be performed if we neglect the
variation of [el with respect to 5 at each integration point. This approximation, in fact, was
adopted by Love in the shell theory. But here we are using it to save computation time for forming
an element stiffness matrix.
Note that the element stiffness matrix derived from energy expression (8) contains components
as listed below:
288 0. C. ZIENKIEWICZ, R. L. TAYLOR AND J. M. TOO

0
0
+
0 0 Bli Du Blj B24D55Bzj
0 0
P i D, Cjl = 0 0
B1i D44 c3j B2i D56 c31

C3i DMC3j 0 0
[Ci D, Cj] = 0 C3iD55C3j 0
0 0 0

Instead of using numerical integration in the expressions (1 5a)-(15k) for all three directions,
~ 5, we may apply explicit integration with respect to 5 and leave the other two directions
8 , and
numerically integrated. It should be noted that expressions (1 5c), (1 5f), (1 5i) and (1 3) consist

L
+1
of integral [d 5 which is in fact zero, therefore these expressions are deleted from the stiffness
matrix before integration is carried out. The remaining terms after explicit integration have the
REDUCED INTEGRATION TECHNIQUE 289

following forms.

In the above expressions, it is clear that the integration through 5 direction has been taken care
of by multiplying of appropriate factors in front of each expression as shown above. By doing SO,
we have eliminated half of the usual numerical integration operations. Thus computation time
for element stiffness matrix has been reduced by fifty per cent.
While this approximation is plausible, the .users should be aware of two important facts, i.e.
1. For plate problems this approximation is as accurate as numerical integration over all
three directions.
2. Erroneous results may occur when dealing with very thick shell problems. This is because
the variation of [el with 5 is no longer negligible.

REFERENCES
1. S. Ahmad, B. M. Irons and 0. C. Zienkiewicz, ‘Analysis of thick and thin shell structures by curved finite
elements’, Int. J . num. Meth. Engng, 2, 419-451 (1970).
2. 0. C. Zienkiewicz, The Finite Element Method in Structural and Continuum Mechanics, McGraw-Hill,
1967.
3. R. J. Melosh, ‘A flat triangular shell element stiffness matrix’, Proc. Conf. Matrix Meth. Srruct. Mech.,
Air Force Institute of Technology, Wright-Patterson Air Force Base, Ohio, 1965.
4. S . Utku, ‘Stiffness matrices for thin triangular elements of non-zero Gaussian curvature’, AIAA Jnl, 5,
1659-1667 (1967).
5 . H. C. Martin, Stiffness Matrix for Triangular Sandwich Element in Bending, TP-32-1158, Jet Propulsion
Laboratory, Pasadena, California, 1968.
6. S. W. Key and Z. E. Beisinger, ‘The analysis of thin shells with transverse shear strains by the finite element
method’, Proc. Con]: Matrix Meth. Srruct. Mech., Air Force Institute of Technology, Wright-Patterson Air
Force Base, Ohio, 1965.
7. S. Ahmad, B. M.Irons and 0. C. Zienkiewicz, ‘Curved thick shell and membrane elements with particular
reference to axi-symmetric problem’, Proc. Conf. Matrix Meth. Struct. Mech., Air Force Institute of
Technology, Wright-Patterson Air Force Base, Ohio, 1965.
8. G . A. Wempner, J. T.Oden and D. A. Kross, ‘Finite element analysis of thin shells’, Proc. Am. SOC.ciu.
Engng EM6. 1273--1294 (1968).
9. G. A. Wempner, ‘Finite elements, finite rotations and small strains of flexible shells’, Int. J . Solids Struct. 5 ,
117-153 (1969).
290 0. C. ZIENKIEWICZ, R. L. TAYLOR AND J. M. TOO

10. J. A. Stricklin, W. E. Haisler, P. R. Tisdale and R. Ganderston, ‘A rapidly converging triangular plate
element’, AIAA Jnl, 7, 180-181 (1969).
1 1. S. W. Key and Z. E. Beisinger, The Analysis of Thin Shells by the Finite Element Method, Symposium on high
speed computing for elastic structures, IUTAM, Liege, 1970.
12. B. M. Irons, Finite Element Techniques in Structural Mechanics, Discussion, H . Tottenham and C. Brebbia
(Eds.), Southampton University Press, 1970, pp. 328-331,
13. W. P. Doherty, E. L. Wilson and R. L. Taylor, Stress Analysis of Axisymmetric Solids Utilizing Higher
Order Quadrilaieral Finite Elements, Structural Engineering Laboratory, University of California,
Berkeley, California, 1969.
14. S. Timoshenko, Strength of Materials, Part 11, Advanced Theory and Problems, 3rd edn., McGraw-Hill,
New York, 1956, pp. 100-101.
15. E. Rsissner, ‘On bending of elastic plates’, Quart. Appl. Math. 5, 55 (1947).
16. A. C. Scordelis and K. S. Lo, ‘Computer analysis of cylindrical shells’, ACI Jnl, 61, 539-561 (1964).
17. R. W. Clough and R. J. Johnson, ‘A finite element approximation for the analysis of thin shells’, Znr. J .
Solids Struct. 4, 43-60 (1 968).
18. G. R. Couper, G. M. Lindberg and M. D. Olson, ‘A shallow shell finite element of triangular shape’, fnt.
J. Solids Struct. 6, 1133-1156 (1970).
19. K . Forsberg, An Evaluation of Finite Difference and Finite Element Techniquesfor Analysis of General Shells,
Symposium on high speed computing for elastic structures, IUTAM, Mechanics, Litge, 1970.
20. G. B. Bazeley, Y. K. Cheung, B. M. Irons and 0. C.Zienkiewicz, ‘Triangular elements in plate bending-
conforming and non-conforming solutions’, Proc. Conf. Matrix Meth. Srrucr. Mech., Wright-Patterson
Air Force Base, Ohio, 1965.
21. G. E. Strickland and W. A. Loden, ‘A doubly-curved triangular shell element’, Proc. Conf. Matrix Meth.
Struct. Mech., Wright-Patterson Air Force Base, Ohio, 1968.
22. S. Woinowsky-Krieger, ‘Der Spannungszustand in dicken elastischen Platten’, Ingenieur-Archiv, 4, 305
(1933).

You might also like