Sie sind auf Seite 1von 37

Theory of Plates

Part III: Finite elements for plates in bending

Lecture Notes

Winter Semester

Prof. Dr.-Ing. Kai-Uwe Bletzinger

Lehrstuhl fr Statik
Technische Universitt Mnchen

kub@bv.tum.de

http://www.statik.bauwesen.tu-muenchen.de/
Many parts and figures of the present manuscript are taken from the German
lecture notes on Platten by Prof. E. Ramm [9], University of Stuttgart.

Special reference is also made to the books of O.C. Zienkiewicz and R.L. Taylor.

Lehrstuhl fr Statik
Technische Universitt Mnchen
80290 Mnchen

October 2000

2
0 REFERENCES ..................................................................................................................... 5

2 PLATES IN BENDING (CONTD.):...................................................................................... 7

2.8 FINITE E LEMENT FORMULATION ........................................................................................ 9


2.8.1 MATRIX FORMULATION OF THE GOVERNING EQUATIONS, THICK AND THIN PLATES .................9
2.8.2 VIRTUAL WORK, WEAK FORM OF EQUILIBRIUM ...................................................................12
2.8.3 DISCRETIZATION ..............................................................................................................13
2.8.4 KIRCHHOFF PLATE ELEMENTS............................................................................................16
2.8.4.1 Continuity conditions for shape functions ......................................................................16
2.8.4.2 Conforming Kirchhoff elements....................................................................................19
2.8.4.3 Non-conforming Kirchhoff elements, the patch test........................................................22
2.8.4.4 Other Kirchhoff elements .............................................................................................23
2.8.5 REISSNER /MINDLIN PLATE ELEMENTS................................................................................23
2.8.5.1 Continuity requirements ...............................................................................................23
2.8.5.2 Shear locking ...............................................................................................................24
2.8.5.3 Reduced integration .....................................................................................................26
2.8.5.4 Zero energy modes.......................................................................................................27
2.8.5.5 Comparison of several Reissner/Mindlin plate elements .................................................29
2.8.6 DISCRETE KIRCHHOFF PLATE ELEMENTS............................................................................35
2.8.7 MIXED AND HYBRID PLATE ELEMENTS ...............................................................................36

3
0 References
Books in English Language:
[1] Gould, Philipp L.: Analysis of Shells and Plates. Springer Verlag New York, 1988.
[2] Pilkey, W.D., Wunderlich, W.: Mechanics of Structure: Variational and Computational
Methods. CRC Press, 1994.
[3] Reddy, J. N.:Theory and Analysis of Elastic Plates. Taylor and Francis, London, 1999.
[4] Szilard, R.: Theory and Analysis of Plates. Prentice-Hall, Englewood Cliffs, 1974.
[5] Timoshenko, S.P., Woinoswski-Krieger, S.: Theory of Plates and Shells. McGraw-Hill,
1987. (2. Aufl.)
[6] Zienkiewicz, O.C., Taylor, R.L.: The Finite Element Method. Vol. 1: Basis, Vol. 2:
Solid Mechanics, Vol.3 Fluid Dynamics. 5. Auflage. Butterworth and Heinemann,
2000.

Plate theory (in German):


[7] Girkmann, K.: Flchentragwerke. 6. Auflage. Springer-Verlag, Wien, 1963
[8] Marguerre, K., Woernle, H.T.: Elastische Platten, BI Wissenschaftsverlag, Mannheim,
1975.
[9] Ramm, E.: Flchentragwerke: Platten. Vorlesungsmanuskript, Institut fr Baustatik,
Universitt Stuttgart, 1995.
[10] Hake, E. und Meskouris, K.: Statik der Flchentragwerke, Springer, 2001.

Finite Element Method (in German):


[11] Werkle, H.: Finite Elemente in der Baustatik. Vieweg Verlag, Wiesbaden, 1995.
[12] Bathe, K.-J.: Finite-Element-Methoden. Springer-Verlag, 1986.
[13] Link, M.: Finite Elemente in der Statik und Dynamik. Teubner-Verlag, Stuttgart, 1984.
[14] Ramm, E.: Finite Elemente fr Tragwerksberechnungen. Vorlesungsmanuskript, Institut
fr Baustatik, Universitt Stuttgart, 1999.

Tables (in German):


[15] Czerny, F.: Tafeln fr vierseitig und dreiseitig gelagerte Rechteckplatten. Betonkalen-
der, 1987 1990, 1993 I. Teil drillsteife Platten mit Gleichlast und linear vernderli-
cher Last
[16] Pieper, K., Martens, P.: Durchlaufende vierseitig gesttzte Platten im Hochbau. Beton-
und Stahlbetonbau (1966) 6, S. 158-162, Beton- und Stahlbetonbau (1967) 6, S. 150-
151.
[17] Pucher, A.: Einflufelder elastischer Platten. 2. Auflage. Springer-Verlag, Wien, 1958.

5
[18] Schneider, K.-J.: Bautabellen. 7. Auflage. Werner-Verlag, Dsseldorf, 1986.
[19] Schleeh, W.: Bauteile mit zweiachsigem Spannungszustand (Scheiben), Betonkalender
1978 (T2), Ernst & Sohn, Berlin.
[20] Stiglat, K., Wippel, H.: Platten. 2. Auflage. Ernst & Sohn, Berlin 1973

Concrete Design (in German):


[21] Leonhardt, F.: Vorlesungen ber Massivbau, Teil 2: Sonderflle der Bemessung. Sprin-
ger-Verlag, Berlin, 1975.
[22] Leonhardt, F.: Vorlesungen ber Massivbau, 3. Teil: Grundlagen zum Bewehren im
Stahlbetonbau. Springer-Verlag, 1974.
[23] Schlaich, J. und Schfer, K.: Konstruieren im Stahlbetonbau. Betonkalender, 1993.

6
2 Plates in bending (contd.):
Finite Element Formulation

7
8
2.8 Finite Element Formulation
The subject of bending of plates was one of the first to which the finite element method was
applied in the early 1960s. At that time the various difficulties that were to be encountered
were not fully appreciated and for this reason the topic remains one in which research is ac-
tive to the present day.

Plates are but a particular form of a three-dimensional solid. However, the thickness of such
structures is very small which motivated to introduce several assumptions to ease solution (as
demonstrated in the previous chapters) and, finally, resulted in the theories of Kirchhoff, Re-
issner/Mindlin and others. The finite element method is additionally applied to any of those
theories. The numerical results of a finite element analysis, therefore, are affected by the chain
of assumptions and simplifications from the real structure to the finite element model.

mathematical model
real structure finite element numerical solution
(Kirchhoff,
formulation
Reissner/Mindlin)
modelling discretization solution
errors errors errors

Fig. 2.71: : Modeling and dicretization errors of the finite element method

It turns out that the thick plate theory (Reissner/Mindlin) is simpler to implement the finite
element method, although for analytical treatment it presents more difficulties.

2.8.1 Matrix formulation of the governing equations, thick and thin plates
Again, we start from the shear deformable theory of Reissner and Mindlin and the related
kinematical definitions:

x, u
P

x
w
w(x) = w0(x)
xz x

z, w

u(x) = z x(x)
Fig. 2.72: Plate deformation (section parallel x z plane)

w(x,y,z) = w0 (x,y)

u(x,y,z) = z x (x,y)

v(x,y,z) = z y (x,y)
9
The strains may be separated into bending (in-plane components) and transverse shear groups
and can be collected in the following matrices:


0
x x
x
= y = z 0 = zL = z
xy y y


y x

and

w +
xz x
x

= = = w +
yz w
+ y
y

The stress resultants can also be grouped into bending and shear. For isotropic elasticity we
can write:

mx

m = my = Db L
m xy

q
q = x = Ds
q y

where, assuming plane stress behavior in each layer of the plate

1 0 E t3
D b = K 1 0 ; K=
0 0 (1 ) / 2 12(1 2 )

and

D s = G t 10 01

Again, E, , G, and are Youngs modulus, Poissons ratio, shear modulus, and the shear
correction factor, respectively.

The constitutive relations can be simply generalized to anisotropic or inhomogeneous behav-


ior such as can be manifested if several layers of materials are assembled to form a composite.
10
The only difference is the structure of the constitutive matrices Db and Ds, which always can
be found by integration over the thickness.

The equilibrium conditions state the balance of vertical forces and moments:

q x
+ p = q + p = 0
T
x
y q y


x 0 m
y x q x
m = LTm q = 0
0 y q y
m
y x xy

These are the equations which are valid for thin as well as for thick plates. For thin plates the
shear deformations are suppressed and the Kirchhoff constraint applies:

w + = 0

The strain-displacement relations now become:

2w

x 2
2w
= z L w = z = z
y 2
2w
2
x y

where is the matrix of curvatures and twisting in the plate.

x

x x
y
= y = = L
xy + yx y
y
y + x
x

Finally, the Kirchhoff PDE which represents the balance of vertical forces is written in matrix
form as:

(L)T (Db L w ) p = 0

11
stress resultants constitutive equations strains

mx 1 0 x

m y = K 1 0 y
m xy 1 2
mx 0 0 xy x
q x 2 x
m y q y
m xy y m Db 2 xy y

m q
q x 1 0 x
q = G t 0 1
y y
q Ds

equilibrium geometry
q x
x y q y + p = 0 L: differential operator 0
LT : adjoint operator x x
T
q +p= 0 0 x =
y y y
2 xy
x 0 y m x q x x
my = 0 = y x
0 m q y
y x xy y L =

w + =
LT m q =0

load displacement and


p rotations

w x
y
w
Table 2.1. Tonti diagram of shear deformable plate.

2.8.2 Virtual work, weak form of equilibrium


The finite element equations are usually obtained from the application of the virtual work
equation or, in other words, from a weak form of the equilibrium equations. The virtual work
of internal forces, the stress resultants, and the external forces states as:

W = Wint + Wext = 0

Wint = (m x x + 2 m xy xy + m y y + q x x + q y y ) dA
A

Wext = p w dA + boundary terms


A

The internal virtual work written in matrix notation and separated into bending and shear
terms:
12
Wint = m dA + Tq dA
A A

= D b dA + T Ds dA
T

A A

= (L ) D b L dA + ( w + ) D s ( w + ) dA
T T

A A

where the kinematical unknowns are the displacement field w(x,y) and the rotations fields
x (x,y) and y (x,y).

The above equation further reduces to the case of thin plates if the Kirchhoff constraint
= w + = 0 is applied:

Wint = m dA
A

= T D b dA = (L ) D b L dA
T

A A

= (L w ) D b L w dA
T

where
T
2 2
(L ) = 2 , 2
, 2
x y 2 x y

It remains the displacement w(x,y) as the only unknown field.

For linear isotropic material the principle of virtual work is identical to the minimization of
the potential energy:

=
1
(L )T D b L dA + 1 (w + )T Gt (w + ) dA w p dA + bt minimum
2A 2A A

Obviously, the thin case can be interpreted as the solution of the general principle with the
Kirchhoff constraint being imposed in a penalty manner with the shear rigidity Gt as penalty
parameter. As is physically evident, the thin plate formulation is simply a limiting case of a
general plate analysis where the thickness approaches zero. The penalty form can yield a sat-
isfactory solution only when discretization of the corresponding mixed formulation satisfies
the necessary convergence criteria.

2.8.3 Discretization

Formally, the discretization for plate analysis follows the usual paths of the finite element
technique. Each of the unknown fields w, x , and y is approximated by a set of discrete val-
ues and shape functions which are related to the nodes of a finite element discretization:

13
n
w( x , y) = Ni ( x , y ) w i ; n = number of element nodes
i =1
n
( x , y) = N i ( x , y ) i ; = x, y
i =1

Or, in a more formal notation:

w1
x1

y1
w = [N1 0 0 ... N n 0 0] ...
w n
xn

yn
w= Nw v

w1
x 1

x 0 N1 0 ... 0 N n 0 y1
= = ...
y 0 0 N1 ... 0 0 Nn w n
xn

yn
= N v

where v is the vector of nodal displacement and rotation values, Nw and N the matrices of
shape functions, arranged for the interpolation of the approximated displacement w(x,y) and
rotations x and y , respectively.

Thick plate, Reissner/Mindlin theory:

Inserting the discretization of displacement and rotation fields into the thick plate virtual work
expression yields:

Wint = v T (L N ) Db L N v dA + v T (N w + N ) Ds (N w + N ) v dA
T T

A A

Defining the strain-displacement matrices

Bb = L N
and B s = Nw + N

we can derive a familiar expression for the element stiffness matrix:

k = k b + k s = B Tb Db B b dA + B Ts Ds Bs dA
A A

which obviously consists of a bending and shear related part.


14
Thin plate, Kirchhoff theory:

Only the bending term remains in the virtual work expression:

(~ T
)~
Wint = v T L N D b L N v dA
A

where the shape functions are modified to respect the Kirchhoff constraint:
~
= N v = w = N w v

leading to

Wint = v T (L N w ) Db L Nw v dA
T

and second order derivatives of the shape functions Nw in the work functional.

The stiffness matrix reduces to:

k = k b = BTb Db Bb dA
A

The external load vector f is derived from the external virtual work:

Wext = w p dA
A

= v T N Tw p dA
A

f = N Tw p dA
A

So far, the finite element technique is applied straight forward to the analysis of plates. Spe-
cial problems, however arise from the special structures of the virtual work functionals with
respect to the mixture of shape functions and their first derivatives in the Reissner/Mindlin
thick plate case (ref. to Bs) and with respect to shape function continuity requirements because
of the second order derivatives in the case of the Kirchhoff thin plate formulation.

Finally, there is an important difference between thin and thick plates when point loads are
involved. In the thin plate case the displacement w remains finite at locations where a point
load is applied; however, for thick plates the presence of shear deformation leads to an infinite
displacement. In finite element approximations of thick plates one always predict a finite dis-
placement at point locations with the magnitude increasing without limit as a mesh is refined
near the loads. Thus, it is meaningless to compare the deflections at point load locations for
different element formulations. It is, however, possible to compare the total strain energy for
such situations and here we immediately observe that for cases in which a single point load is
involved the displacement provides a direct measure for this quantity.

15
2.8.4 Kirchhoff plate elements

2.8.4.1 Continuity conditions for shape functions


Shape functions have to satisfy certain conditions of continuity to be able to generate numeri-
cal solutions of the considered problem. A look at the Bernoulli beam gives a first insight and
motivates the further arguments for the Kirchhoff plate.

The virtual work expression for the Bernoulli beam is:

W = ( M w p) dx = ( EI w p ) dx = (( w ) EI ( w) w p) dx = 0
L L L

where w is the lateral displacement, L and EI are the length and the bending rigidity of the
beam, respectively.

x p(x)

EI(x)
w(x)

Fig. 2.73: Bernoulli beam.

Twice integrating by parts transforms the virtual work expression into:

W = (w M ) 0 + (w Q ) 0 + (EI w p ) w dx
L L

where the boundary terms represent the natural boundary conditions at the beams start and
end points indicating either vanishing rotations and displacements or moments or shear forces.

Applied to two adjacent finite beam elements the terms at the common boundary have to van-
ish otherwise they would introduce additional spurious energy:

BT = (w M ) left + (w Q ) left = 0
right right

In the absence of point loads moment and shear force are continuous, i.e. the same on both
sides of the boundary. The boundary terms can be transformed into:

BT = (w right w left ) M bt + (w right w left )Q bt = 0

which indicates the continuity conditions

w left = w right
and w left = w right
16
The same conditions apply for the displacement function w itself. Since the function as well
as its first derivative have to be continuous, i.e. the same across the common boundary of ad-
jacent elements, a function of that type is called C1 -continuous. Obviously, a cubic function
does the job. A finite element which satisfies the continuity conditions is also called con-
forming or compatible.

Now, let us take a look at the Kirchhoff plate. The virtual internal work is:

Wint = m dA = w T LT Db L w dA
A A

w w 2w
2 2
where (L ) w = +2 +
x 2 x y y 2

Obviously, the highest derivative of w is of order 2, that means, proper shape functions for the
Kirchhoff plate have to be at least C1 -continuous. However, since the plate is a two-
dimensional structure and the displacement function w is a function of two co-ordinates (x, y)
the requirements for the shape functions are somewhat more complicated. This is preferably
explained by an example.

Incompatible rectangular element with 12 degrees of freedom (Adini-Clough-Melosh):

Nodal parameters:
4 3
at each node: wk w,x k w,y k

shape functions consist of the following polynomials:


y
1
1 2 x y
x x xy y 2
2

x3 x 2 y x y2 y3
x3 y x y3

Investigation of compatibility properties.

Consider the common edge 1-2 of two adjacent elements.


The elements are compatible if all edge information can be
described by parameters which both elements have in
4 3 common, as e.g. all parameters of nodes 1 and 2.
1
The displacement is at y = const. a cubic function in x
y
which must be described by 4 independent parameters.
1 x 2
w w Along edge 1-2 these are the nodal variables w1 , w, x 1, w2 ,

w , x w, x and w,x 2, i.e. w is continuous along 1-2 with respect to the
w , y w, y function itself and the tangential derivative.
1 2 2

5 6
The cross derivative w,y along 1-2 is also a cubic function
in x. However, there are only 2 parameters, w, y 1 and w,y 2,
available, which both elements have in common. In gen-
eral, the ACM-element is, therefore, not compatible.

17
The following figure shows the displacement field generated by the given data:

node x y w w,x w,y


1 0 0 0 1 0
2 1 0 0 1 0
3 1 1 0 0 1
4 0 1 0 0 -1
5 0 -1 0 0 -1
6 1 -1 0 0 1

discontinuous cross derivative


along edge 1-2.

continuous cross derivative


along edges in y-direction

Fig. 2.74: Example of discontinuous cross derivatives of the ACM-element.

Compatible rectangular element with 16 degrees of freedom (Bogner-Fox-Schmidt, Schfer):

Nodal parameters:

at each node: wk w,x k w,y k w,xy k


4 3
shape functions are bi-cubic and consist of the following
polynomials.:

y 1
x y
1 2
x 2 xy y2
x x3 x 2 y x y2 y3
x 3y x 2 y 2 x y 3
x 3 y 2 x 2 y3
x 3 y3

18
Investigation of compatibility properties.
There are now 4 parameters on the edge available to generate a cubic function for the cross
derivative, e.g. edge 1-2: w,y 1 w,x y 1 w,y 2 w,x y 2 .The twisting degree of freedom is difficult
to handle.

The following figure shows the displacement field generated by the data:

node x y w w,x w,y w,xy


1 0 0 0 1 0 1
2 1 0 0 1 0 1
3 1 1 0 0 1 1
4 0 1 0 0 -1 1
5 0 -1 0 0 -1 1
6 1 -1 0 0 1 1

Fig. 2.75: Example of continuous cross derivatives of the BFS-element.

2.8.4.2 Conforming Kirchhoff elements


The rectangular 4-node ACM- and BFS-elements have already been introduced in the previ-
ous section. Besides those there exists the full variety of all other plane element geometries, as
there are triangular, parallelogram, quadrilateral, and iso-parametric elements.

Fig. 2.76: Schematic overview on Kirchhoff plate element types.

19
The problem, however, are the continuity conditions which in general are difficult to be satis-
fied for conforming, compatible elements. The BFS-element is the exception. It is, however,
restricted to a rectangular geometry. It is, therefore, not very much used in practice. The con-
tinuity conditions very often lead to different variables at the nodes. This is indicated in the
above figure by the different node symbols (bullets and circles). An example for that is the 21
degree-of-freedom triangle, which is independently described by Argyris et al., Bell, Cowper
et al., and Irons during 1968 and 1969. Mention of the element shape functions was made
earlier by Withum and Felippa.

The 21 degree-of-freedom triangle

Referring to the BFS-element, we saw that a conforming element can be derived by prescrib-
ing the following data at each node: the displacement, both first derivatives in (x,y) direction,
and the second mixed derivatives, indicating the twist of displacement. With respect to an
arbitrarily orientated edge, now, the first and second derivatives have to be prescribed normal
and tangential to the edge. Compare with the following figure.

t
n

1 x
2
Fig. 2.77: C1 -continuity at edge 2-3 of triangular element.

The first and second mixed derivatives at edge 2-3 are evaluated by the chain rule of differen-
tiation:

w w dx w dy
= +
n x dn y dn
w w dx w dy
= +
t x dt y dt
2w 2 w dx dx 2 w dx dy dx dy 2w dy dy
= + + +
n t x 2 dn dt x y dn dt dt dn y 2 dn dt

It follows that all second derivatives with respect to the global coordinates (x,y) have to be
supplied at the nodes. At any node the following data has to be prescribed: w w, x w,y w,xx
w,xy w,yy . Together at all three corners of the triangle that are in total 3x6 = 18 degrees of
freedom.

20
1 1 deg ree of freedom
x y 3
2 2
x xy y 6
3 2 2 3
x x y xy y 10
4 3 2 2 3 4
x x y x y xy y 15
5 4 3 2 2 3 4 5
x x y x y x y x y y 21

Fig. 2.78: Pascals triangle up to polynomial of 5th order.

A quick glance at Pascals triangle shows, however, that an ideal, complete polynomial as
shape function with at least 18 free parameters must be of 5th order or more. As additional
variables which fill up the remaining 3 parameters up to 21 the normal derivatives at the mid-
sides are chosen. Finally, we end up with an 6 node triangle where at the vertices : w w,x w,y
w,xx w,xy w,yy are prescribed and at mid-sides the cross derivative w, n .

degrees of freedom

w w 2 w 2w 2w
w
x y x 2 x y y2 k

w

n k

Fig. 2.79: Nodal variables of the 21 degree-of-freedom element.

The existence of the mid-side nodes with their single degree of freedom is an inconvenience.
It is possible to constrain these by allowing only a cubic variation of the normal slope along
each triangle side giving an element with three corner nodes and in total 18 degrees of free-
dom. Both elements perform well.

General remarks

As we saw by the example of the ACM-element, it is impossible to devise a simple polyno-


mial function with only three nodal degrees of freedom that will be able to satisfy slope conti-
nuity requirements at all locations along element boundaries. The alternative is to impose cur-
vature parameters at nodes. This has basically three disadvantages: (i) difficulties to impose
boundary conditions for higher order derivatives, (ii) nodes with different number and types
of parameters, and (iii) imposing excessive conditions of continuity. The latter involves in-
consistency when a discontinuous variation of material properties occurs. Then moments are
continuous but curvatures are discontinuous. The engineer may still feel a justified preference
for the more intuitive formulation involving displacements and slopes only, despite the fact
that very good accuracy is demonstrated for higher-order, quartic and quintic elements.

21
2.8.4.3 Non-conforming Kirchhoff elements, the patch test
Continuity of shape functions is a severe condition for the convergence of the finite element
method. Conforming elements are designed such that these conditions are identically satisfied
at any location on the element boundary. However, since the finite element method approxi-
mates the solution it is enough to satisfy the continuity conditions in a weak sense. That is the
theoretical basis for the formulation of suitable non-conforming elements. The idea is to relax
the conditions such that the continuity conditions have to be satisfied by a patch of elements.
As the finite element approximation converges to the correct solution the size of elements
tend to zero and one can assume that the moment is constant in the element. As the element is
part of a patch of elements the continuity condition is satisfied in the mean if the integral
along the element boundary of the virtual work done by internal forces acting at the disconti-
nuity is zero (see section 2.8.4.1). Assume the tractions acting on an element interface are mn
mnt and qn . The difference of corresponding virtual displacements left and right of the com-
mon edge of adjacent elements are:

w w
n = , t = , w
n t

Then the weak form of the continuity conditions is:

m n n d + m ntt d + q n w d = 0
e e e

For constant moments the shear force qn is zero and the last term of the above equation van-
ishes. The remaining expression is satisfied if at each straight side e of the element the follow-
ing holds:

n d = 0 and d = 0
t
e e

where the second integral usually is identically satisfied. This is the motivation for the formu-
lation of several non-conforming elements which converge to the correct solution and indeed
show to be very effective. Preferable are elements which are defined in terms of nodal displace-
ment and rotational parameters.

t
mn
n

qn mnt

Fig. 2.80: Patch of elements Fig. 2.81: Boundary tractions.

22
Refined analysis of the non-conforming ACM-element.

Consider the displayed patch of three elements. Assuming


constant moment my element 2 contributes to the virtual
x 4 1 x 3 work of my as:
y 4 4 3 y 3
2w
a b y 2 dy dx = a ( y1 y 2 + y 3 + y 4 )
a b

I=
x 1 2 x 2

2b
y 1 1 2 y 2 which contributes to the edge integrals 1-2 and 4-3 after in-
tegration by parts:
3
a w a w
y
I = dx dx = a (3 + 4 ) a (1 + 2 )
2a 1424 3 1424 3
x a y y =b a y y= b edge 4 3 edge 12

Obviously, the inspected edge integrals 1-2 and 4-3 of element 2 are functions of parameters
which are defined on the related edge. The same argument applies to elements 1 and 3 which
means that the difference of common edge integrals of adjacent elements just cancels out. The
relaxed continuity conditions for the ACM-element are satisfied.

2.8.4.4 Other Kirchhoff elements


Besides the presented types of elements there exist e.g. quasi-conforming elements and ro-
tation free elements where other arguments are used to satisfy the continuity conditions.
Discrete Kirchhoff elements are formulated via the refined Reissner/Mindlin theory where
the Kirchhoff constraint is used to eliminate the shear effect. Insofar they have a position
between Kirchhoff and Reissner/Mindlin elements which will be discussed next. In all cases
the constant moment patch test has to be satisfied to guarantee convergence.

2.8.5 Reissner/Mindlin plate elements

2.8.5.1 Continuity requirements


The Reissner/Mindlin theory allows shear deformation, i.e. rotations and displacement deriva-
tives are not directly coupled and differ by the shear deformation. Consequently, rotation
fields x (x,y) and y (x,y) as well as displacement field w(x,y) are independently introduced
into the virtual work expression:

Wint = m dA + Tq dA
A A

= (L ) Db L dA + ( w + )T Ds ( w + ) dA
T

A A

By inspection we see that the highest derivatives of displacements and rotations are of first
order. Finite element approximations must be not more than C0 -continuous. That is an impor-
tant simplification compared to the Kirchhoff theory and was one of the major motivations to
develop Reissner/Mindlin plate elements.

23
Because of that there exist no real restrictions for the kind of element geometry used for the
implementation. Iso-parametric elements allowing for curved boundaries and arbitrarily com-
posed patches are standard. Further more, the Reissner/Mindlin theory is the basis for up-to-
date shell elements on curved surfaces.

several iso-parametric elements used for Reissner/Mindlin plates


bi-linear linear

bi-quadratic quadratic
8-node serendipity 9-node Lagrange

bi-cubic cubic
12-node 16-node Lagrange

2.8.5.2 Shear locking


As the plate becomes thinner shear deformation becomes unimportant and the Reiss-
ner/Mindlin theory meets in the limit t = 0 the Kirchhoff theory. A measurement of thick-
ness is the ratio (L/t), span over thickness. Repeated analysis with decreasing thickness
should, therefore, generate a series of results which converge to the Kirchhoff solution. How-
ever, if the Reissner/Mindlin theory is implemented straight forward as sketched in section
2.8.3 the element behaves far too stiff, for some element formulations the plate even does not
deflect. The reason why this effect is called shear locking. The behavior of the quadratic
serendipity element is dramatically influenced by shear locking if the element is not modified,
Fig. 2.82. Other elements, particularly higher order Lagrange elements are rather insensible.
The 16-node Lagrange behaves almost perfect in the classical displacement formulation for
the price of comparatively high numerical effort. The insensibility of higher order elements is
the motivation for elements of extreme high order polynomials, the so called p-adaptive ele-
ments, up to orders of 8 and more. The idea is to adapt the polynomial order just to meet an
optimal behavior as compromise between exactness and effort.

24
0.0044

0.0043
0.0042
exact thin plate
wcK / qL 4

0.0041 solution 0.00406

0.0040
0.0039

0.0038
0.0037 simply soft supported square plate
1
10 10 2 3
10 10 constant area load, 4x4 subdivisions in a quarter
4

L/t serendipity elements, fully integrated

Fig. 2.82: Performance of quadratic serendipity elements with varying span-to-thickness L/t
ratios, shear locking for L/t > 102

The reason of shear locking becomes obvious if we take a look at the shear part of the virtual
internal work:

Wint = (L ) D b L dA + ( w + ) D s ( w + ) dA
T T

14442444
A
3 A144444244444 3
bending shear

Recall, that the bending and shear parts are characterized by K = Et3 /12(1- 2 ) and GtL2 , re-
spectively. As the thickness t becomes smaller GtL2 becomes larger than the bending rigidity
K and the shear deformation dominates the internal work. In theory this tendency is counter-
balanced by activating the Kirchhoff constraint w + = 0. Therefore, the shear part fades
out of the internal work. If the finite element technique would properly work this effect
should, of course, also take place after discretization. Introducing shape functions and nodal
displacement parameters v the virtual work expression displays as:

Wint = v T (L N ) Db L N v dA + v T (N w + N ) Ds (N w + N ) v dA
T T

A A

where the discretized Kirchhoff constraint is:

( N w + N )v = B s v = 0

N1 N n
0 0 ... 0 0
x x 0 N 0 ... 0 N n 0
or + 0 01 v=0
N1 N n N n
0 0 N1 ... 0 0
0 0 ...
y y

As displacement w and rotations x y are interpolated by the same shape functions we see
that the discretized Kirchhoff constraint implies that the linear combination of polynomials of
different order (N i and its derivatives) must identically vanish as t approaches zero. That is in
general impossible for polynomial shape functions. As a consequence, even at the Kirchhoff
limit t = 0, some shear energy remains, which dominates the bending energy and causes shear
locking. The numerical effect is more severe for lower order functions.
25
Two general lines of modification have evolved to overcome the above deficiency: (i) re-
placement of Bs by some modification Bs which indeed is known as B-bar-method, or (ii)
to modify the integration of the shear strain energy such that the constraint is weakly satisfied.
The latter approach leads to the concept of reduced integration. So called assumed natural
(ANS) or, enhanced assumed (EAS) strain concepts modify the strain fields and lead at the
end also to modified Bs-matrices.

2.8.5.3 Reduced integration


The concept is briefly sketched for the example of a Timoshenko beam where the identical
problem of shear locking arises. Because of the aspects of convergence we can restrict our-
selves to the case of a beam subjected to constant moment loading. The beam shall be discre-
tized by a two node, linear element:

approximation by 2-node linear interpolation


real structure
of displacement w and rotation
M M

M M -L-
1 = 2 =
x
0 w(x) = 0
w1 = 0 w2 = 0
2 =

z, w +
x
1 = = + 2
L

FEM approximation of shear strain


shear strain
dw x
0 h = + = + 2 0
dx L

Obviously, the shear strain approximation h is not identical zero along the whole beam as it
should be with regard to theory. There is just one location at the beam center x = L/2 where
this is the case. If this location is used as integration point or Gauss point (GP) of a numeri-
cal integration of the shear strain energy the finite element approximation would be correct.
For the integration of the shear energy we now use one instead of two integration points. We
speak of reduced integration instead of full integration. If at the same time the bending
term is fully integrated the scheme is called selective reduced integration. The bending
terms are sometimes also reduced integrated although this is not necessary with respect to
shear locking. The same procedure applies analogously for higher order beams and plates.

integration order of some selective reduced integrated plate elements


element bending term shear term
4-node bi-linear 2x2 1x1
8-node bi-quadratic 3x3 2x2
9-node bi-quadratic 3x3 2x2

Note, that reduced integration may introduce additional internal zero energy modes (physi-
cally meaningless displacement modes without strain energy) and the elements may also not
be completely locking free.

26
2.8.5.4 Zero energy modes
Zero energy modes shall be explained by the example of a rectangular 4-node plate element.

bi-linear interpolated approximations of displacement and rotations:


y
w( x , y ) = [1 x y xy ] a where aT = [a1 a2 a3 a4 ]
x ( x , y ) = [1 x xy ] b where b = [b1 b4 ]
x T
y b2 b3
y ( x , y ) = [1 x y xy ] c where c = [c1 c 2
T
c3 c4 ]

The strains are determined as:

x
xx = = [0 1 0 y] b
x

yy = y = [0 0 1 x ] c
y
1 x y 1
xy = + = ([0 0 1 x ] b + [0 1 0 y] c )
2 y x 2
w
x = + x = [0 1 0 y ] a + [1 x y xy ] b
x
w
y = + y = [0 0 1 x ] a + [1 x y xy ] c
y

Bending and shear terms shall be reduced integrated, i.e. consider 1 integration point at the
element center (x,y) = (0,0) instead of 2x2 points. At the center we get the strains:

xx = b 2
yy = c3

xy =
1
(b 3 + c 2 )
2
x = a 2 + b1
y = a 3 + c1

Balance of unknowns and governing equations:

Number of unknown parameters 12


governing equations:
restrain rigid body modes,
statically determined supports 3 -8
number of strains at integration point 5
number of independent unknowns 12 8 = 4

There are 4 parameters which cannot be determined and can be chosen arbitrarily. They are
related to 4 displacement modes of zero strains and internal energy although they are no rigid
body modes. Therefore, they are called zero energy modes (ZEM). On the other hand, that
means, that the (reduced integrated) stiffness matrix is not definite, i.e. det(k) = 0:

27
int
all

point. s
= Tk = T BT D B dA = T BTi DiBi = 0
i =1

A

where represents the zero energy modes. They are the solution of the eigenvalue problem
(k I ) = 0 with zero eigenvalues .
When using reduced integration to prevent shear locking there exists always the possibility of
zero energy modes.

The minimum number of zero energy modes can easily be evaluated. E.g., consider a reduced
(2x2) integrated 8-node serendipity element:

Number of degrees of freedom: 8x3 = 24


5 strain equations at each integration point 5x(2x2) = 20
3 rigid body modes =3
min. number of zero energy modes 24 20 3 = 1 ZEM

The evaluation of a reduced (1x1) integrated 4-node element:

Number of degrees of freedom: 4x3 = 12


5 strain equations at each integration point 5x(1x1) =5
3 rigid body modes =3
min. number of zero energy modes 12 5 3 = 4 ZEM

Fig. 2.83: Two zero energy modes of the reduced (1x1) integrated 4-node plate element.

28
plate under uniform load
6x6 4-node elements
1x1 reduced Gauss-integration
Fig. 2.84: Example of a zero energy pattern.

2.8.5.5 Comparison of several Reissner/Mindlin plate elements


The behavior of rectangular R/M-elements will be demonstrated by the example of a fully
clamped square plate under uniform load. It is discretized by 4x4 subdivisions in a quarter.

system discretization

p = 10 kN/m2
8.00 m

8.00 m

In the sequel the following abbreviations will be used:

p = 10 kN / m 2 uniform area load


E t3 bending rigidity
B=
12(1 2 )
E = 1.2 10 4 kN / m 2 elasticity modulus
= 0.0 Poissons ratio
L = 8.00 m side length of plate
t, h plate thickness
fm max. displacement

29
4-node element.
Fully integrated the element shows extreme shear locking, the results are useless. Shear lock-
ing disappears if the element is reduced integrated, in addition is numerically very efficient.
However, there are 4 ZEM (as shown above). These modes can be controlled by so called
stabilizing matrices. The one point integrated 4-node elements are favorably used for large
dynamic analyses, e.g. crash simulations in automotive industries. The selectively reduced
integrated elements has only 2 ZEM, however, is considerably ineffective because of the nec-
essary four integration points (integration of bending part).

30
8-node serendipity element.
Although reduced integration improves the element behavior, locking cannot be completely
eliminated depending on the stiffness of supports (e.g. fully clamped, as in this example).
Slightly refined meshes improve results significantly. Reduced and selectively reduced inte-
gration give almost identical results. The only ZEM of the reduced integrated element is sup-
pressed in an element patch of at least two elements. All together the 2x2 reduced integrated
8-node element is very efficient and a good choice.

31
9-node Lagrange element.

This element locks moderately if fully integrated. Selective reduced integration avoids lock-
ing perfectly, however, a 3x3 point integration is necessary. It can compete with the 8-node
element for very thin plates (because the 8-node element needs a comparatively fine mesh).
For thick plates it less efficient. Reduced integration gives 4 ZEM which make additional
modifications necessary.

32
12-node element.
It behaves similar as the 8-node serendipity element. Shear locking cannot be avoided by any
integration scheme. ZEM do not appear, even reduced integrated. It is less efficient than the
8-node element since 9 integration points are necessary for reduced integration. Therefore, it
is not much used. If so it should be reduced integrated.

33
16-node Lagrange element.
This element almost does not lock if fully 4x4 integrated. It is very expensive. however, gives
reliable results. If reduced integrated it should be used with care because ZEM may appear.

general remark:
The quality of iso-parametric elements very much depends on the element geometry. They are
worse the more the geometry is distorted. Opposite side should be as close to parallel as
possible, mid-side nodes and center nodes should be placed as close as possible to the real
center of side or element, respectively.

34
2.8.6 Discrete Kirchhoff plate elements
A further alternative for the analysis of Kirchhoff plates is to consider a Reissner/Mindlin
plate at the Kirchhoff limit where the Kirchhoff constraint is identically satisfied. That means,
instead of considering the virtual work as previously by

W = m dA + Tq dA w p dA
A A A

= (L ) Db L dA + ( w + )T Ds ( w + ) dA w p dA
T

A A A

we now write

W = m dA w p dA
A A

= (L ) Db L dA w p dA
T

A A

subject to + w = 0

The shear part of virtual work vanished identically since the constraint + w = 0 is satisfied
at any point in the plate.

We have seen before that in a finite element formulation the Kirchhoff constraint cannot be
identically satisfied everywhere in the plate which led to shear locking. But it is possible to
satisfy it at several discrete locations. Elements generated by this approach are, therefore,
called Discrete Kirchhoff elements. Prominent species of that type are known as DKT (Dis-
crete Kirchhoff Triangle) and DKQ (Discrete Kirchhoff Quadrilateral). The elements combine
the advantages of both theories: shear locking free analysis of thin plates (Kirchhoff) and C0 -
continuity requirements for independently interpolated displacements and rotations (Reiss-
ner/Mindlin). The 3-node, 9 degree-of-freedom DKT element by Batoz is one of the best tri-
angular plate elements known.

How to construct a Discrete Kirchhoff element?


Again, that is conveniently explained by the example of a Timoshenko beam. First, consider a
three node beam. Displacements and rotations are independently approximated by quadratic
shape functions:

1 3 2 2
x
w(x ) = w 3 + (w 2 w1 ) + 2( w1 + w 2 2w 3 )
x
x L L
2
x
(x ) = 1 + (2 1 ) + 2(1 + 2 23 )
x
w1 w3 w2
L L
L

Next, the inner node variables (w3 ,3 ) are eliminated by proposing two discrete constraints at
the locations (x,x ):

w w
+ (x ) = 0 and + (x ) = 0
x x= x x x= x

35
1 L
If the Gauss points x / = m are chosen the inner variables are evaluated as:
32
w 3 = (w 1 + w 2 ) L(1 2 )
1 1
2 8
and 3 =
3
(w 1 w 2 ) 1 (1 + 2 )
2L 4

These are back substituted into the above approximations and a beam element remains with
all together 4 nodal parameters (w1 , 1 , w2 , 2 ) which satisfies the discrete Timoshenko
constraints at the two Gauss points. For the presented choice of (x,x ) it appears that the
stiffness matrix of the element is identical to that of a cubic Bernoulli element, i.e. the pre-
sented element is exact.

The construction of discrete Kirchhoff plate elements is analogous:

3 3 Development of the 3-node DKT element by elimina-


w
tion of the 9 variables at the mid-side nodes 4,5,6 of an
6 5 v i = x
y initially 6-node quadratic R/M element. 6 Kirchhoff
i
constraints are used w.r.t. the tangential side directions.
The normal slopes at the three mid-side nodes are line-
1 4 2 1 2
arly interpolated from the normal slopes at the corner
6-node R/M DKT
nodes.

2.8.7 Mixed and hybrid plate elements


Mixed elements can be derived for Kirchhoff as well as Reissner/Mindlin theories. The for-
mulation starts from the complementary energy of the problem. Considering the Kirchhoff
case, the complementary energy of the system writes as:

~ = m T dA 1 T D dA w p dA
A 2 A
b A
Inserting the geometrical as well as the inverse constitutive equations

= Lw and = D b Tm, respective ly

we get
~ = m T L w dA 1 m T D T m dA w p dA
A 2 A
b A
from which we derive by the first variation the stationary condition:

= m T (DbT m + L w )dA w ( TLTm + p) dA = 0


~
A A

Note, that in a weak form we can recognize the geometrical and the equilibrium equations
D b T m + L w = + L w = 0 and T LT m + p = Tq + p = 0 respectively.

36
The differential operators redistributed:

( ) (
= m T Db Tm + (L m ) w dA ( w ) LTm + w p dA = 0
~
T T
)
A A

the expression is used as the basis of a mixed finite element formulation. The unknown fields
are now the displacement w(x) and the moments m(x) which have to approximated by proper
shape functions:

mx
n n

wh = N w i w i and m h = N m i m y
i =1 i =1 m xy
i

Since displacement and stress resultants are approximated the element is called mixed. The
highest degree of derivatives of w and m is 1 which indicates that C0 -continuous shape func-
tions are sufficient which is of great advantage. Many very successful elements are available
which are based on that approach. The disadvantages are: (i) nodal variables are displacement
as well as moment components which makes it difficult to combine these elements with other
type of elements, and (ii) the system of equations shows zeros on the main diagonal why spe-
cial kind of equation solvers are necessary.

The disadvantages of mixed elements are overcome by special variants the so called hybrid
elements. Here, the nodal moment components are eliminated by special procedures which
adapt the element interior mixed fields to an overlaying displacement field interconnecting the
element patch. Mixed elements have only nodal displacement components, they can easily be
connected to other kind of elements. The ideas go mainly back to Pian. Hybrid elements are
very successfully used in the practical analysis of thin plates.

37

Das könnte Ihnen auch gefallen