Sie sind auf Seite 1von 15

A Mixed 3D Finite Element for Modelling Thick Plates

Mark Asch

Laboratoire dAnalyse Numerique,


Universite Paris XI, 91405 Orsay , FRANCE.
Michel Bercovier
Bertold Badler Chair of Computer Science,
Hebrew University of Jerusalem, Givat-Ram, ISRAEL.
May, 1993 (revised version)
Abstract
Based on the Hellinger-Reissner variational principle, we formulate a mixed 3-d nite element for
plate bending. This element is used to model thick plates and alleviates the problem of shear-locking in
plates with large length/thickness ratios. The computer code which was used here, is available.
1 Introduction
The approach of [4] to the plate-bending problem has no a priori assumptions of the small thickness type.
It is an asymptotic theory based on the mixed variational principle of Hellinger and Reissner [10]. This
enables one to treat the plate-bending problem as a fully three-dimensinal problem in linear elasticity.
In this paper, we express the above theory in a computational form to which we apply a nite element
method. This is a mixed method in which we approximate both displacement and stress on brick type
elements. We carefully choose and explain the spaces of polynomials used for the displacements and stresses.
We then present the results of numerical simulations and draw conclusions from these.
We begin with a brief survey of the plate theories currently used in the development of nite elements.
More details can be found in [8].
2 Plate theories
2.1 Historical introduction
In 1828, Poisson and Cauchy developed the rst thin plate theory. Starting from the general equations of
elasticity, and assuming that all quantities can be expanded in powers of the distance from the mid-surface,
they arrived at the equations of equilibrium for the displacement normal to the plane of the plate. Kircho,
in 1850, improved and corrected this theory based on two assumptions:
1. linear laments of the plate, initially normal to the mid-surface, remain straight and normal to the
mid-surface after strain.
2. all elements of the mid-surface remain unstretched

Currently: URA 0743, Modellisation Statistique et Stochastique, Universite Paris XI.


1
2.2 Plate theories for nite elements
We present here the theory of Reissner-Mindlin for thin plates. We state the hypotheses. This then leads
to a discussion of the problem of shear-locking related to the instability of plate elements as their thickness
tends to zero. The reduced-selective integration methods are used to overcome this problem. Our approach
has some similarities with these methods, but is fundamentally dierent.
2.2.1 The main hypotheses of plate theory
H1. the structure is thin in one dimension, usually taken as the z-direction.
H2. stress perpendicular to the mid-surface is zero This is the plane-stress hypothesis.
H3. material particles that are originally on a straight line normal to the mid-surface, remain on a straight
line during deformations i.e. plane sections remain plain.
H3a. in Mindlin theory, shear deformations are included, so a line originally normal does not in general
remain normal to the mid-surface during deformations.
H4. the transverse displacement does not vary through the thickness.
In contrast, our mixed method (see below), will not rely on any a priori hypotheses.
2.2.2 Finite element formulation for a plate
Based on the mathematical form of the hypotheses presented above, we write down the boudary value
problem of elasticity and its variational form. For the nite element approximation, we use (for example)
serendipity elements with three degrees of freedom per node: w,
1
and
2
. This yields the stiness matrix
and load vectors of the nite element approximation.
2.3 Stability of thin plates
Shear locking, a phenomenon, which leads to overly sti elements, occurs as the thickness of the plate tends
to zero. Due to the hypotheses of the plate theories, we obtain an overly constrained problem since there
are more shear strain constraints than equilibrium equations and unreliable results.
To avoid this shear locking, two methods are widely used: uniform reduced integration and selective
reduced integration. In the former, both the shear and bending terms are integrated with the same rule of
an order lower than what should be used for full accuracy. In the selective integration, the bending term is
integrated with the normal rule, whereas the shear term is integrated with a lower order rule usually one
order lower than the normal. Both these methods introduce spurious zero-energy modes and the resulting
element may be unstable.
Special elements, possessing correct rank, have been developed, but they tend to be more expensive in
terms of computations. We introduce a 3-d mixed element that does have a rank deciency similar to that
of selective-reduced integration plate elements but which exhibits a stable behaviour when the thickness
tends to zero. By using a mixed formulation, we actually construct an element which diers from the
selective-reduced one due to the choice of the projection spaces.
We note that there are other approaches to the construction of Reissner-Mindlin elements using known
results on the approximation of Stokes problems. For details and some error analysis see [6].
3 3D Mixed formulation for the thick plate problem
3.1 The 3-D linear elasticity boundary value problem
The equilibrium equations for three dimensional elasticity are given by
1

ij,j
+ f
i
= 0 (1)
1
Lower case indices take values in the set 1,2,3; upper-case in the set 1,2,. . . ,6. A repeated index implies summation. A
comma implies partial dierentiation.
2
The constitutive relation is written as

ij
= b
ijkl

kl
=
1 +
E

ij


E

kk

ij
(2)
where

ij
=
1
2
(u
i,j
+ u
j,i
)
To these we must still add the boundary conditions. We write (1), (2) in matrix form as
AX = F (3)
where
A =
_
0 T
T
T B
_
, X =
_
u

_
, F =
_
f
0
_
and
T =
_
_
_
_
_
_
_
_
/x
1
0 0
0 /x
2
0
0 0 /x
3
0 /x
3
/x
2
/x
3
0 /x
1
/x
2
/x
1
0
_
_
_
_
_
_
_
_
, =
_
_
_
_
_
_
_
_

11

22

33

23

13

12
_
_
_
_
_
_
_
_
f = f
i
=
_
_
f
1
f
2
f
3
_
_
, u = u
i
=
_
_
u
1
u
2
u
3
_
_
.
Multiplying out (3), we obtain
T
T
= f (4)
Tu B = 0 (5)
which are the equilibrium and constitutive equations respectively.
3.2 The Hellinger-Reissner variational formulation
The Hellinger-Reissner principle of minimum potential energy, requires the mininization of

R
=
_

ij

ij

1
2

ij
b
ijkl

kl
f
i
u
i
d (6)
The stationarity conditions are obtained by dierentiating this expression with respect to u and . We get

=
_

ij
b
ijkl

kl
d = 0 (7)
and after an integration by parts of (6),

R
u
=
_

ij,j
+ f
i
d = 0 (8)
These imply (2) and (1) respectively.
The weak formulation based on the Hellinger-Reissner principle is obtained as follows. First we dene
two Hilbert spaces,
S =
_
= (
ij
)
_
L
2
()
_
9
,
ij
=
ji
_
(9)
3
and
V =
_
v = (v
i
)
_
H
1
()
_
3
, v
i
|

= 0
_
(10)
S is a space of symmetric second-order tensors with norm
||

=
_
_

i,j
|
ij
|
2

_
_
1/2
and V is a space of three-dimensional Euclidean vectors, vanishing on the boundary, with product norm
_

i
||v
i
||
2
1,
_
1/2
where
||v
i
||
2
1,
= |v
i
|
2

+
3

j=1

v
i
x
j

, R
3
and | | is the usual L
2
norm.
Multiplying (7) and (8) by suitable test functions from the spaces (9) and (10) respectively, we obtain
the desired weak formulation:
Find an element (, u) S V which satises
S ,
_

ij

ij
(u) b
ijkl

kl

ij
d = 0 (11)
v V ,
_

ij

ij
(v) + f
i
v
i
d = 0 (12)
The existence and uniqueness of the solution to (11-12) is proved using the Brezzi-Babuska condition and
Korns inequality. Details can be found in [4].
4 The nite element discretization
If we take a typical 3-d brick element,
e
, then the nite element approximation of the weak formulation
(11-12) is:
Find an element (
h
, u
h
) S
h
V
h
which satises

h
S
h
,
_

e
h
Tu
e
h
(
e
h
)
T
B
e
h
d
e
= 0 (13)
v
h
V
h
,
_

e
(Tv
e
h
)
T

e
h
+ (v
e
h
)
T
f d
e
= 0 (14)
where S
h
and V
h
are nite dimensional subspaces of S and V respectively. The true displacements are
approximated by
u
e
h
=
N
e

j=1

e
j
u
e
j
(15)
where
4
u
j
are the unknown nodal displacements

j
are shape functions for element e
v
e
h
=
e
Similarly, for the stresses

e
h
=
M

K=1

e
K

e
K
(16)
where

K
are the unknown element stresses

e
K
are basis functions for element e

e
h
=
e
Then for a typical element, we rewrite (13-14) as
_
m
e
p
e
(p
e
)
T
0
__

e
u
e
_
=
_
0
f
e
_
(17)
where
m
e
=
_

e
[
e
]
T
B
e
d
e
p
e
=
_

e
[
e
]
T
T
e
d
e
(p
e
)
T
=
_

e
[T
e
]
T

e
d
e
f
e
=
_

e
[
e
]
T
f d
e
Note that this formulation is equivalent to saying that starting from a continuous displacement eld, we
have projected the stresses on some specic (lower order or not) polynomial spaces.
5 The algorithm
In this section we describe the numerical method used to solve the plate-bending problem. We start with
the formulation of the mixed nite element stiness matrix. We then detail the interpolations which were
found to be optimal for displacements and stresses. In the Appendix we describe the many options included
in the code, which enable one to study the problems of stability and excess shear rigidity in thin elements.
5.1 Stiness matrix formulation
To obtain the stiness matrix of our mixed formulation, we must write out (17). We have
M + Pu = 0 (18)
P
T
= f (19)
We can solve for in the rst equation,
= M
1
Pu (20)
5
and substitute in the second, to obtain
P
T
M
1
P u = f (21)
Then dening the stiness matrix, K, as
K
.
= P
T
M
1
P
we obtain the equivalent system to (21) :
Ku = f (22)
This can be written in terms of the nite element approximation of the previous subsection, as
__

(T)
T

_ __

T
B
_
1
__

T
T
_
u = f (23)
5.2 Interpolation of displacements and stresses
Following the approach of [3], we choose the interpolants and (for u and respectively) from the following
polynomial spaces:
P
k
, k = 0, 1 the space of complete polynomials in 3 variables, of degree k
Q
1
the space of polynomials in 3 variables, of degree 1 in each variable
Other possibilities (not used here) are presented in the appendix.
6 Results of numerical investigations
6.1 Standard plate-bending problem
We take a standard plate-bending problem from the literature, see [7], whose analytical solution is known
from [9]. Many other examples of test problems can be found in [11]. We condider two types of boundary
conditions:
(i) clamped edges,
(ii) simply-supported edges;
two load cases:
(i) uniform, distributed transverse load, q
(ii) concentrated transverse load at the centre of the plate, P;
and two mesh types:
(i) a uniform, square mesh,
(ii) an irregular skew mesh.
We compute the maximum transverse displacement at the centre of the plate and normalize it with respect
to the analytical solution of [9].
The dierent interpolations of displacements and stresses that were considered are presented in Table 1.
Note that in all cases the diplacements are Q
1
. There are three dierent stress interpolations. We start from
all six components of the stress tensor in the space P
1
and end with three direct stresses in P
1
and the two
transverse shear stresses in P
0
. The in-plane shear stress component must remain in P
1
in order to have a
consistent approximation. We will present here, only the results of the best interpolations. In the report [2]
results for all cases are given.
6
Case Direct Shear
u
11

22

33

23

13

12
A Q
1
P
1
P
1
P
1
P
1
P
1
P
1
B Q
1
P
1
P
1
P
1
P
0
P
1
P
1
C Q
1
P
1
P
1
P
1
P
0
P
0
P
1
Table 1: Interpolation for displacement and stress.
6.1.1 Uniform mesh plate problem.
In Tables 24 we present the results of the numerical computations. These are performed on a one-quarter
symmetric model of the square plate with N denoting the number of elements on a quarter side. The quarter
plate has length equal to one unit. We set the distributed load equal to one and the point load equal to 0.25
(giving one on the whole plate)
2
. Poissons ratio is taken as 0.3 and Youngs modulus is set equal to one.
We compare our results with computations performed using a standard nite element code which employs a
displacement formulation and uses an eight-node brick element with three degrees of freedom at each node.
We denote these results by the subscript dsp. The results of the mixed nite element computations are
denoted mix. In the case of the uniform mesh, we present results for three mesh renements:
A very coarse mesh: 2 2 on the symmetric part, which implies a mesh of 16 elements on the entire
plate denoted N = 2.
A less coarse mesh: 4 4, giving a 64-element mesh denoted N = 4.
The nest mesh: 16 16, yielding a 1024-element mesh denoted N = 16.
Case h L/h e
dsp
e
mix
N = 2 N = 4 N = 16 N = 2 N = 4 N = 16
unif-2 0.25 8 0.76 1.073 1.308 1.083
unif-3 0.1 20 0.18 0.456 1.056 1.036
unif-4 0.05 40 0.05 0.162 0.731 1.021 0.999 0.994
unif-5 0.02 100 0.008 0.029 0.304 1.011 0.974 0.983
unif-6 0.01 200 0.002 0.006 0.097 1.010 0.986 0.982
unif-7 0.002 1000 7.10
5
2.10
4
0.0014 1.009 0.986 0.981
Table 2: Results of clamped plate-bending problem for uniform mesh with N = 2 ,N = 4 and N = 8
In the tables, h denotes the thickness of the plate, L/h denotes the aspect ratio of the entire plate
the quantity in parentheses denoting the aspect ratio of a single element. The results, e
dsp
and e
mix
are the
normalized transverse displacements,
e
dsp
=
w
dsp
w
theory
, e
mix
=
w
mix
w
theory
2
The results for this load case are reported in [2]
7
Case h L/h e
dsp
e
mix
N = 2 N = 4 N = 2 N = 4
unif-2 0.25 8 1.225 0.999 1.419 1.169
unif-3 0.1 20 0.858 0.901 1.052 1.068
unif-4 0.05 40 0.803 0.959 1.005 1.035
unif-5 0.02 100 1.091

0.959 0.993 1.035


unif-6 0.01 200 0.238 1.052

0.991 1.030
unif-7 0.002 1000 0.004 0.0004 0.990 1.028
unif-8 0.001 2000 7.10
5
3.10
4
0.991 1.028
Table 3: Results of simply-supported plate-bending problem for uniform mesh with N = 2 and N = 4 (an
asterisk implies zero diagonals in stiness matrix)
where w
theory
is the theoretical displacement, as given in [9]. A value of 1.0 for the normalized displacement
implies an exact result. If the ratio is greater than one, then the computation was more exible than the
theory and if the ratio is smaller than one, then the results were stier. A shear correction factor of 5/6 see
[11] is used in the clamped case in order to match the thin plate theory. However, for the simply-supported
plate, we do not need a correction of this kind since we can simulate precisely the correct 2-dimensional
boundary conditions. In fact we impose the boundary condition on the bottom surface of the brick only,
setting the x, y and z displacements equal to zero there. The top surface is completely unconstrained and
we do not even impose the symmetry boundary conditions there. This is justied by the numerical results,
which show in-plane displacements (on the top surface) which are two orders of magnitude smaller than
the transverse displacements and also seems to agree with the observations in [5] concerning the correct
boundary conditions for their asymptotic approach.
We can make the following observations in the uniform mesh case with clamped boundary conditions:
with full interpolation (case A not presented here), the mixed model gives essentially identical results
to the displacement model this serves as a verication of the correctness of the mixed formulation;
in the mixed model, as we successively under-interpolate the transverse shear stress components
(making them P
0
instead of P
1
), the plate becomes more exible and the error diminishes;
as the length over thickness ratio increases, the mixed results pass from slightly more exible than the
theory, to slightly stier than the theory the displacement fem results become stier and stier;
even with an aspect ratio of 40 for an element (1000 on the whole plate), the mixed results are within
2% of the theory, whereas the displacement fem results are completely worthless;
for the N = 2 case (a 4x4 mesh), we observe monotone convergence to the thin plate limit, with high
accuracy for h = 0.002 the error is less than 1%;
In the simply supported case, the trends are similar, but we would emphasize:
with the mesh N = 2, we obtain high accuracy (less than 1% error), and a very cheap element in terms
of the computational cost;
the displacement results become completely unreliable, with negative values appearing on the diagonal
of the stiness matrix, starting from thickness h = 0.02;
8
6.1.2 Irregular mesh plate problem.
The irregular skew mesh (see Figure 1) is 4 4, giving 16 lements on a one-quarter model. This standard
test for plate elements shows up any sensitivity to distorted elements.
0.9 0.7 0.4
1.0 0.7 0.3 0.1
Figure 1: Skew mesh plate-bending problem.
The skew mesh results show (see Table 4):
an overall deterioration of the results when compared with the uniform mesh
this deterioration is far more marked in the displacement results
the mixed results (clamped) maintain acceptable accuracy (10%) up to an aspect ratio of 200
In the simply supported case, the trends are similar to what we observed in the uniform mesh. We remark
the good accuracy obtained in the mixed interpolation with an error of 3.8% only.
The curves in Figures 2 6 show the above-mentioned trends clearly. We plot the results for each N as a
function of the plate length-to-thickness ratio. This enables us to examine the convergence to the thin plate
theory (as plate thickness goes to zero) note the good performance of the case N = 2 in both clamped and
simply-supported cases.
9
Case h L/h e
dsp
e
mix
clamped simp.-supp. clamped simp.-supp.
unif-3 0.1 20(2.5) 0.435 0.906 1.030 1.127
unif-4 0.05 40(5) 0.157 0.719 0.983 1.077
unif-5 0.02 100(12.5) 0.029 0.460 0.955 1.044
unif-6 0.01 200(25) 0.008 0.245 0.924 1.039
unif-7 0.002 1000(125) 0.0003 0.0

0.850 1.038
unif-8 0.001 2000(250) 0.00007 0.0

0.844 1.038
Table 4: Results of clamped and simply-supported plate-bending problems for skew mesh with N = 4 (an
asterisk implies negative diagonals in stiness matrix)
6.1.3 Stability analysis of the mixed 3-D element
An important consideration in the development of nite elements with reduced integration is the appearance
of a rank deciency. It is thus necessary to check the rank of the mixed element for each of the interpolations
of Table 1. These checks were performed numerically. We used Gaussian elimination with full pivoting
to determine the rank of the element stiness matrices. In the case A (all stresses P
1
) we obtained a
rank of 15. The element has 24 degrees of freedom 8 nodes, 3 displacements per node. Since there are
6 rigid body modes (3 translations, 3 rotations), the rank deciency is 3. These are the twisting modes
corresponding to the cross-terms xy, xz and yz which do not appear in a P
1
interpolation. Each time we
under-interpolate a transverse shear term (P
0
instead of P
1
) we pick up an additional spurious zero-mode.
Thus the B-interpolation has a deciency of 4 and the C-interpolation a deciency of 5.
If we look at the assumed behaviour of a Naghdi-Reissner plate element eld as a 3-d displacement eld
[4], such a eld has a rank deciency of 9 (including the 6 rigid body modes). To avoid shear locking we must
lower the order of the projection for the shear terms which brings us to a rank deciency of 11 (two more
than before). This can be compared to the 4-node selective-reduced integration plate element which has a
rank deciency of 1 under the Naghdi-Reissner eld assumption. Noting that in a 3-d model the degrees of
freedom are twice present in the thickness, this would clearly double the expected rank deciency.
This last point is overlooked in the classic 3-d selective-reduced approach and the corresponding element
necessitates a mesh which is dependent on the thickness. This is clearly not the case in our approach.
7 Conclusions
(i) the use of a 3-d brick element with an original mixed interpolation (displacement-stress with cross-
terms omitted in the polynomial spaces) gives accurate results for the solution of a classical plate
problem in which the displacement model nite element method fails dismally;
(ii) under-interpolation of transverse shear stress components enables investigation and alleviation of shear-
locking in plate bending;
(iii) the mixed 3-d brick element gives very good results on a very coarse mesh and may provide a cheap
element for certain problems where we do not want to use higher order plate bending elements for
example in dynamic problems such as crash simulations;
(iv) the mixed 3-d element performs acceptably well on distorted meshes and is thickness independent.
10
0
0.2
0.4
0.6
0.8
1
1.2
1.4
10 100 1000
w
fem
w
theory
L/h
displ 3
3
3
3
3 3 3
mixed + +
+
+
+ + +
Figure 2: Normalized centre displacement for clamped plate as function of aspect ratio (L/h): uniform mesh
N = 2.
(v) the mixed 3-d element used here has a rank deciency similar to that found in lower-order optimal
plate elements.
Appendix The computer code MIX3D
The computer code developed for this paper, Mix3D is available as is from the authors. Please send
requests via electronic mail to:
berco@cs.huji.ac.il
We emphasize that this is a research code and that it contains many more capabilities for higher order
elements than those used in this paper. We present here the details of these options.
A Interpolation of displacements and stresses
Following the approach of [3], we choose the interpolants and (for u and respectively) from the following
polynomial spaces:
P
k
, k = 0, 1, 2 the space of complete polynomials in n variables, of degree k
Q
k
, k = 0, 1, 2 the space of polynomials in n variables, of degree k in each variable
where n = 2, 3 depending on whether we use two- or three-dimensional elements. The interpolations and
the elements available are presented in Table 5.
B Interpolation and integration options
In order to study phenomena associated with shear locking in thin plates, we have provided numerous
options.
11
0
0.2
0.4
0.6
0.8
1
1.2
1.4
10 100 1000
w
fem
w
theory
L/h
displ 3
3
3
3
3
3 3
mixed +
+
+
+
+
+ +
Figure 3: Normalized centre displacement for clamped plate as function of aspect ratio (L/h): uniform mesh
N = 4.
1. the interpolation for stress can be selected from any of the spaces P
i
, Q
i
, i = 0, 1, 2
2. the displacements can be approximated by variable 4-to-9 node elements in 2-d, and 8-to-21 node in
3-d
3. each component of the stress tensor can be interpolated dierently from the others
4. within a specic polynomial space, a subset of basis functions can be used in the interpolation of the
stresses
5. the interpolation for stress can be taken in either the local or global coordinate systems
6. the order of numerical integration can be changed in any of the coordinate directions
The rst and second options let us interpolate stress and displacement dierently. The variable node
elements provide exibility in the displacement approximation. The third option allows us to under-
interpolate the shear stress components of the stress tensor by taking them in a space of one order less
than the direct stresses e.g. we take shear in P
0
and direct components in P
1
. The fourth option enables
us to investigate the optimality of the projection of the unknowns onto the polynomial spaces. This is
investigated in more detail in [1]. The fth option is included following the experience of [3] who found
distinct advantages when stresses are approximated in the global space. With the last option we can achieve
the eects of reduced/selective integration see [12].
C Stability Analysis
Built in to the code is the possibility to determine the rank (and hence the number of zero energy modes)
of any element stiness matrix.
12
0
0.2
0.4
0.6
0.8
1
1.2
1.4
10 100 1000
w
fem
w
theory
L/h
displ 3
3
3
3
3
3
3
mixed +
+
+
+
+ + +
Figure 4: Normalized centre displacement for simply-supported plate as function of aspect ratio (L/h):
uniform mesh N = 2.
D Conclusions
We see that the code provides a versatile research tool. It has been used to examine
shear-locking in slender beams
shear-locking in thin plates
behaviour of multi-layered plates
In this paper we dealt with the second item only.
References
[1] M. Asch and M. Bercovier. Mixed nite elements with optimal order projections. In preparation.
[2] M. Asch and M. Bercovier. Mix3d a mixed nite element code for modelling thick plates. Technical
Report 6 (February), Leibniz Institute of Computer Science, Hebrew University of Jerusalem, Givat-
Ram, Israel, 1993.
[3] M. Bercovier, M. Engelman, P. Gresho, and R. Sani. Consistent vs. reduced integration penalty methods
for incompressible media using several old and new elements. Int. J. for Numerical Methods in Fluids,
2, 1981.
[4] P. G. Ciarlet and Ph. Destuynder. A justication of the two-dimensional linear plate model. J.
Mecanique, 18:315344, 1979.
[5] P. G. Ciarlet and Ph. Destuynder. A justication of a nonlinear model in plate theory. Comp. Methods
Appl. Mech. Engrg., 26:145172, 1981.
13
0
0.2
0.4
0.6
0.8
1
1.2
1.4
10 100 1000
w
fem
w
theory
L/h
displ 3
3 3
3
3
3
3
mixed +
+ +
+
+ + +
Figure 5: Normalized centre displacement for simply-supported plate as function of aspect ratio (L/h):
uniform mesh N = 4.
[6] K.J. Bathe F. Brezzi and M. Fortin. Mixed-interpolated elements for Reissner/Mindlin plates. Int.
Journal for Numerical Methods in Engineering, 28:17871801, 1989.
[7] R. L. Harder. Review of the MacNeil-Harder linear static test problems. Finite Elements in Analysis
and Design, 1(1), 1985.
[8] T. J. R. Hughes. The nite element method: Linear static and dynamic nite element analysis. Prentice-
Hall, New York, 1987.
[9] S. Timoshenko and S. Woinowsky-Krieger. Theory of Plates and Shells. McGraw-Hill, New York, 1959.
[10] K. Washizu. Variational Methods in Elasticity and Plasticity. Pergamon Press, New York, 1974.
[11] S. L. Weissman and R. L. Taylor. Mixed formulations for plate bending elements. Computer Methods
in Applied Mechanical Engineering, 94:391427, 1992.
[12] O.C. Zienkiewicz. The Finite Element Method, 3rd edition. Mc Graw-Hill, New York, 1977.
14
0
0.2
0.4
0.6
0.8
1
1.2
1.4
10 100 1000
w
fem
w
theory
L/h
displ 3
3
3
3
3
3 3
mixed +
+
+
+ + + +
Figure 6: Normalized centre displacement for simply-supported plate as function of aspect ratio (L/h): skew
mesh N = 4.
Unknown Polynomial Element
Space 2-D 3-D

i
Q
1
bilinear, 4-node quad trilinear, 8-node
isoparametric brick
u

i
Q
2
biquadratic, 9-node triquadratic, 27-node
quad isoparametric brick

i
P
0
, P
1
4-node quad 8-node brick

i
P
1
, P
2
9-node quad 27-node brick
Table 5: Interpolations and elements.
15

Das könnte Ihnen auch gefallen