Documente Academic
Documente Profesional
Documente Cultură
(This paper was delivered at the FEMSA’92 Symposium, 11th Symposium on Finite Element
Methods in South Africa, 15-17 January 1992, organised by FRD/UCT Center for Research in
Computational and Applied Mechanics, University of Cape Town)
The CST was the first element, which was developed for finite element analysis (FEA) and thirty
years ago it served its purpose well. In the meantime more accurate elements have been
created and these should be used to replace the CST.
A simply supported beam loaded vertically with a uniformly distributed load and modeled with
eight CST elements gives zero stress where it should have its maximum. This is a 100% error
and shows that such elements are useless and very dangerous in the hands of inexperienced
users.
Some finite element programs, which are being marketed worldwide, have still only the CST in
their element library for plane stress and three-dimensional shells.
Over the past seven years, the author has discussed the problems associated with the CST with
a number of lecturers who claim to be finite element experts. Some of these erudite gentlemen
still consider three-noded plane stress elements to be better than four-noded ones! In many
lecture notes, textbooks and pamphlets on FEA only triangular elements are shown.
Furthermore, some automatic mesh generators can create only triangular elements, while others
create meshes with unnecessary triangular elements.
In considering that, in the future, under-qualified people will be using FEA, this paper is intended
to create an awareness of the problems associated with the CST and eventually to have these
elements removed from element libraries for the sake of safer structures and thus saving time,
money and probably human lives.
1. THE SHOCK
A concrete beam, 8m long, 2m high and 0,2m thick is loaded with a vertical u.d.l. of 100 kN/m
at the top (see fig. 1). The material properties are: Young’s modulus 3,0 x 107 kPa and
Poisson’s ratio v = 0,2.
The horizontal stresses at the centre of the span can easily be calculated manually using the
bending theory. They are –6000 kPa (compression) at the top and +6000 kPa (tension) at the
bottom.
Let us model this two-dimensional problem with finite elements, using eight three-noded plane
stress elements with a mesh layout according to fig. 2a.
The results of the finite element analysis look strange (see table 1). At the bottom centre of
the beam we only get +1347 kPa compared to the theoretical value of +6000 and at the top
centre, where we should have the maximum compression of –6000 kPa we get zero stress.
The error, which has been defined as the maximum of the errors for the two stress columns,
is, therefore, 100%.
Now the results look much better. We are getting values of +5250 kPa at the bottom and
–5250 kPa at the top, an error of 13%.
Various other finite element models were analysed. The numerical results can be found in
Table 1 and the plots for the Sxx stresses for line numbers 2 to 9 are displayed in figs. 3a-3h.
In the past seven years the author has asked several hundred people to estimate the stresses
for the two above mentioned finite element meshes. Some of these people had no finite
element background at all; others were “experienced” users of FEA and a number of them
were even university lecturers, teaching FEA to students.
The author first asked for an error estimate for the mesh with the four-noded elements. Some
people had such faith in FEA that they first did not understand the question and then said that
there would be no error at all in such a simple problem. When they were told the error was
13%, they were surprised. When they were asked for an error estimate for the mesh with the
three-noded elements, almost everybody, including the academics, estimated lower errors
than before (typically 3%). When confronted with the facts they could firstly not believe it and
then they were truly shocked.
The author first discovered the misbehaviour of the three-noded triangular plane stress
elements about eleven years ago although he had already five years of experience with FEA
and had analysed and designed in that time major structures using the finite element method.
2. THE EXPLANATION
(2.1) Consider a 3-noded plane stress element in the xy-plane with node points 1, 2 and
3. The x-deflections are u1, u2, u3 and the y-deflections v1, v2 and v3, six values
together.
The displacement function then has the following form (using six constants
i to describe the behaviour of the element):
u= 1 + 2x + 3 y
v= 4 + 5x + 6 y
∂u
εx = = α2
∂x
∂v
εy = = α6
∂y
The strains in such an element are constants. We know, however, that in a beam
as in fig. 1 we have compression at the top and tension at the bottom. Our single
element is, therefore, not capable of modelling bending behaviour of a beam, it
cannot model anything at all.
(2.3) The results for configuration 2c (for four-noded elements) from programmes
COSMOS *** (incompatible elements), ALGOR and ANSYS are identical with the
FESDEC results.
It is confusing for the novice user that COSMOS ** (full integration) is inferior to
COSMOS * (reduced integration) and that the best of all, COSMOS *** is an
“incompatible element”.
It should be noted that one row of 6- and 8-noded elements (fig. 2e and 2f)
performs equally well as two rows of four-noded elements (fig. 2g).
The results from the boundary programme BEASY are excellent and one should
also consider that for plane stress and plane strain problems no meshes are
required in boundary element programmes. The whole interior is defined by the
boundary only.
3. THE REMEDY
u= 1 + 2 x+ 3 y+ 4 xy
v= 5 + 6 x+ 7 y+ 8 xy
∂u
εx = = α2 +α4 y
∂x
∂v
εy = = α7 + α 8 X
∂y
(3.2) Use triangular elements which also include the rotational degree of freedom about
the z-axis normal to the xy-plane.
The idea was implemented by D. Neille [1] in his programme PST. The error for a
mesh configuration 2a is reduced from 100% (CST) to 56%. Using 2 layers of
elements according to 2h the error is 27% compared with 72% for the CST.
(3.3) Many CST elements are required to give satisfactory results (with say an error of
less than 10%). These elements should, therefore, only be permitted as fillers
between four-noded elements (6- and 8-noded elements are preferred).
Finite element programmes that have CST elements only should be prohibited by
a Code of Practice for Finite Elements.
Triangular elements for plane stress and shells should be banned from textbooks
and courses and should only be mentioned as bad examples.
5. REFERENCES
[1] D Neille, The Finite Element for Infinite Applications, NCL Stewart Scott,
Consulting Engineers
6. ACKNOWLEDGEMENTS
The author would like to thank the following persons for carrying out test runs:
Table 1