Compurers & Srrucrures Vol. 43. No. 1. pp. lfA179.

Printed in Great Britain.

1992
0

W45-7949/92 55.00 + 0.00

1992 Pergamon Press plc

HYBRID-TREFFTZ p-METHOD ELEMENTS FOR

ANALYSIS OF FLAT SLABS WITH DROPS
J.

JIROUSEK~

and M. NDIAYE$

tLSC, Swiss Federal Institute of Technology, Lausanne, Switzerland

~Structural Division of Thies Institute of Technology, Senegal
(Received 15 January 1991)
Abstract-A simple method is presented by which various types of internal displacement fields involved
in the Hybrid-Trefftz (HT) element formulation for flat slabs with drops can be generated in a unified
way. Apart from a standard set of displacement functions for an isotropic plate element, such fields
involve, in particular, a special purpose set of functions representing the local solution in the vicinity of
a column. The elastic properties of the column and the effect of the sharp radial variation of the plate
thickness in the drop area are properly accounted for. The presented functions extend the library of
optional internal displacement fields with which an advanced HT element subroutine is typically provided.
Its use along with the p-extension of the HT elements enables the local effects in flat slabs with or without
drops to be efficiently handled through crude unrefined finite element meshes by using just a single element
for each column head as well as a large portion of the surrounding slab. The accuracy and the practical
efficiency of the approach are assessed on a series of numerical examples involving a comparison with a
three-dimensional analysis.

1. INTRODUCTION

A reluctance to abandon the obvious advantages of

the well-established conventional assumed displacement elements has often led to a sacrifice of both the
accuracy and the economy. In the case of the flat
slabs with drops, an accurate solution to the stress
concentration generated in the vicinity of the column
head is difficult to obtain and even if a plate theory
rather than a three-dimensional
(3-D) solution is
adopted, a tedious local mesh refinement is required
to obtain reliable moment concentrations. A further
costly refinement may be necessary if a reliable
prediction of shear forces is an important issue of
the analysis. Such refined discretisation is hardly
applicable for a full structure analysis of the slab.
For some time now, it has become obvious that
various problems involving stress singularities or
stress concentrations can efficiently be handled if use
is made of an alternative finite element (FE) formulation now known as the Hybrid-Trefftz
(HT)
model [ 11. Here the intraelement displacement field
accurately fulfils the governing differential equations
and the interelement continuity is enforced by a
stationary principle involving a conforming frame
function field. Different aspects of this comparatively
recent FE model involving p-method capabilities [2],
implementation of a curvilinear geometry [3], error
estimation [4], adaptivity [S] and other relevant topics
have been studied during the last five years. Though
initially [6-81 the HT elements have not been explicitly designed to handle areas of high stress gradients,
it has soon become obvious [9] that such problems
can be solved with a surprising efficiency if the

element subroutine is associated with a library of

optional special purpose functions representing
various local solutions applicable in the vicinity of,
for example, a singular corner, a crack, a circular or
elliptic hole [lO-131, etc. and involving both the
isotropic and the anisotropic materials [14, 151. This
attractive possibility has motivated the present
tendency to further extend the libraries of such
functions and to progressively building up a collection of local solutions covering the current needs of
the engineering practice. The present contribution
falls into this category and Fig. 1 shows the types of
solutions that will be provided.
Throughout this paper it will be assumed that the
reader is familiar with the HT element formulation of
which a short description has been presented in one
of the previous issues of this journal [15] and further
relevant details have been given elsewhere [l-3]. This
will make it possible to concentrate on the generation
of the element (index e) optional internal displacement field. For a HT element based on the classical
thin plate theory governed by the Lagrange plate
equation

where D = Et3/12(1 - v) is the plate rigidity, this

internal field will conventionally be assumed in the
form
m

w, = iJe+

1 QjC, = Ge+ Qec,

j=l

163

(2)

164

J. JIROUSEKand M. NDIAYE

displacement Z and that with M = 3, for example,

the normal slope will be cubic and the transverse
displacement quartic along the element side.
For practical implementation of the HT elements
it is useful to remember [l] that the necessary (but
not sufficient) condition (also known as stability
condition) for the resulting stiffness matrix to have
correct rank is
Fig. 1. Discretization of a plate into HT p-method
elements: A, elements with standard set of internal
functions. B, C, elements with optional special purpose
functions.

where tie stands for a particular

V4Ge=%
@, is a suitably truncated
homogeneous solutions
V4@,= 0

on

solution to (1)
R,,

part of a complete set of

on

f13

(2b)

and c, are underdetermined coefficients.

As in [15], a p-extension of the HT FE model will
be used for all numerical applications. As shown in
Fig. 2, such elements can have an optional number of
curved sides and the final unknowns-the
element
DOF-include
the conventional DOF (w, w,, wY) at
corner nodes 0, along with an optional number M of
hierarchic DOF (a,, u2,
, a,+,) belonging to the
element sides but associated for convenience with the
mid-side nodes, A. These DOF are used to define
along the element boundary an independent conforming frame function field including transverse
displacement
6, and Cartesian components
of
rotation 3,,, @Y;e
(appropriate frame function definitions are given for example in [2, 151). In the process
of the element formulation (variational principle [ 11)
the undetermined coefficients cj of the internal field w,
are expressed in terms of the frame function DOF
and since the resulting stiffness matrix is symmetric
and positive definite, the element can be implemented
without difficulty into the conventional FEM codes.
For the specification of the boundary conditions note
that, at A, the odd and the even DOF alternatively
concern the normal slope e,, and the transverse

m > NDOF - NRIG.

Here m is the number of independent homogeneous

solutions in (2), NDOF is the number of elements
degrees of freedom (frame function in Fig. 2) and
NRIG is the number of possible rigid body motion
modes of the element. Although (3) is the only
necessary condition, correct rank may always be
achieved by suitably augmenting m. Generally an
optimal value of m (not only warranting correct rank
but also optimally compensating the excess of rigidity, associated with a fixed number of frame function
DOF, by suitably relaxing the conformity) will be
used based on numerical experiments.
The following two sections present a unified
approach to the generation of the standard and
special purpose internal displacement functions (2)
for elements of either a constant (Sec. 2) or a radially
variable (Sec. 3) thickness. The special purpose functions include the plate-column and perforated plate
types (elements B and C in Fig. 1). These functions
have been implemented in the HT plate bending
element subroutine of the program SAFE [16] and all
the numerical examples and the assessment studies of
Sec. 4, including the comparison with a 3-D analysis
of a flat slab with drops, have been obtained with this
software. Section 5 brings concluding remarks.
2. DISPLACEMENT FUNCTIONS FOR HT PLATE
ELEMENT OF CONSTANT THICKNESS

For the class of problems considered in the present

paper a unified approach to generation of internal
displacement functions can be based on the general
solution of the Lagrange plate equation (1) written in
a local polar coordinate system (Fig. 3). Here the
non-dimensional radius

/)=;

3 DOF (w,wx,w,)

M DOF

(a,,a, ,..., aM1

(M optional)

0
6

r-ap
Y

Fig. 2. A HT p-version element.

(3)

Fig. 3. Local polar coordinate system.

165

Hybrid-Trefftz element formulation

has been normed by the average distance a between
the origin 0 and the corner nodes of the element. For
a standard element, 0 is the centre of gravity of the
element while for a plate-column element (Fig. 4) or
a perforated element, 0 is situated at the column axis
or at the hole centre. The main reason for using a
normed radius (presenting a value close to 1 at the
element boundary) is to avoid an over/underflow.
Indeed such problem may otherwise arise due to the
presence of large powers of the radius involved in the
boundary integrals used to evaluate the element
stiffness matrix.
Based on the Lagrange plate equation in polar
coordinates [17], the general solution in the p, 9 may
conveniently be written as

and a matrix of four fundamental

f,(p)=[l

p2 lnp P21np]

f,(p)=[p

p l/p ~lnp]

f,(p)=[p

solutions

in k=O
ifk=l

pZfk p-k p2-k]

if k > 1.

Based on (4) the expression of the form

w,= fi,,+ f ojcj= 3, + @$,
J=I

(5)

of the internal displacement field (2) of the element

will be generated through the use of suitable subsidiary conditions depending on the type of the
element.
2.1. Standard HT element

k = 0,1.2....

+ k=T2.,,,fk(P)Bk sin kg,

(4)

where wP stands for the particular

solution
(Vw, =p/D), which may also be written as

wp =

In the case of a standard element where in (4) the

constants A,,, Ati and B,,, B4k have to be set to zero
to prevent the displacement w, the rotation i3w/ar and
the shear force Q, to go to infinity as r = ap -+ 0, the
remaining constants may simply be assimilated to the
undetermined coefficients cj

(6)

w;&)

COS
k9
Consequently,
identified as

k = 0.1.2s..

w$(p) sin k9,

the functions

(4a)

G)eand Gj are readily

G,= wp

k = I,Z,...

(7)

[@,+I @j+2I=&QfCOSks,
and where the homogeneous solution involves for
each k two vectors of four integration constants

[@j+3@j+4]
= f,Qfsin

k9,

(8)

where
,:=,=[A

Bk= {B,kBx & t &k )

3

a)

Fig. 4. HT plate-column

(4b)
b)

element

(a) without

and (b) with drops.

:]

fork>

@a)

166

J. JIROUSEKand M. NDIAYE

and (since the rigid body motion modes have to be

removed [I] from we) the cases k = 0 and k = 1 yield
the three following independent functions only
@r=$

@,=gcos9,

@,=gsin9.

(8b)

where I,, I, and A are the principal moments of

inertia and area of the cross-section. Although the
cross-section may be arbitrary, it is assumed that the
short plate column intersection has a circular crosssection of 2r, in diameter, where the equivalent radius
r0 is equal to

Note that the particular term 3, = ids, e.g. for a

uniform load p = const is

r. = J(+n),

(11)

where s stands for the distance between the point of

application of P and the point considered. For a load
uniformly distributed over a given circular area (not
necessarily centred at the origin 0) the corresponding
3, may be found in [l]. More elaborate expressions
for an arbitrary patch load and line load are also
available and have been presented in [18].
For practical implementation of the above functions, it is useful to note that the minimum (but not
necessarily sufficient) number of c$-functions in w, is
defined by the stability condition (3) with NRIG = 3.

where A, = A if the column section is full and A, is the

area of a full section presenting the same external
contour if the actual section is hollow.
No attempt will be made to analyse the complex
3-D behaviour at the platecolumn
intersection
and the cylindrical extension of the column through
the plate thickness will be considered as rigid. If at the
same time the plate behaviour in the neighbourhood
of the intersection is assumed to obey the classical
Kirchhoff plate theory, it becomes easy to formulate
the kinematic and static conditions involved in
representing the plate-column interaction.
To derive from the general solution (4) a set
of special purpose functions for a plate-column
element, suitable kinematic and static conditions
(which represent the platecolumn
interaction)
have to be accounted for at r = r, = up,. With the
notation of Fig. 5, the following conditions will be
specified:

2.2. Plate-column element

k = 0 (simple compression

and for a concentrated

is

load P the singular solution

P
w =-slns.
16nD
0

A typical HT platecolumn
element is shown in
Fig. 4a. The plate is assumed free of membrane forces
(the small horizontal actions exerted by the columns
are neglected) and the plate displacements are assumed to vanish at middle surface level. The elastic
properties of the column are characterized by its

of the column)

dw
( ar >,=,,,=O

2nro(Q,),=,, = K(wL=~~,

(114

k = 1 (column flexure in the X-Z plane)

dW

ar
(->

r=r0.9=0

=h4r=,,).:9=o
r.
(lib)

~ k = 1 (column flexure in the y-z plane)

dW

(ar>

r=ro.g=n:2

=(wL,.s=r.2
r.

s
277

r0

(-M,sin9-M,,~0~9+Q,r~sin3),=,d~=K~~(w),=,.,~~-~:~,

(1lc)

k>l
flexural and extensional rigidities which for example
in the case of a column fixed at its base (as in Fig. 4)
are equal to

(w),=,=O

aw

(>
ar

I=,0

0.

(114

Hybrid-Trefftz

element

formulation

167

2.3. Perforated element

It is interesting to point out that a similar type of
formulation can also be used to generate special
purpose functions for a HT element with a circular
hole. If the hole boundary r = r,, = up, is free, then
the two conditions to be satisfied are, for example

M,dr

Fig. 5. Polar

coordinate
components
of internal
plate under bending.

forces in

If the moments M,, M,, and the shear force Q, in

these conditions are expressed in terms of w [17]
The corresponding matrices Qf , Q, and qf , qf are
listed in Appendix II.
Special purpose functions for other possible combinations of the boundary conditions around the circular hole can be obtained in the same manner.

)I
A4,8=-.$~~-~,$$~-~w),
M,,_D E!+y idw+iaw
a2 [

Q,= -;

ap2

( pap

p2aP2

3. EXTENSION

-$Vw,

(12)

and the general solution (4) is further substituted for

w, it becomes possible to eliminate again the constants A,k, Alk and Bjk, BJk, or alternatively, express
for each k the vectors A,, B, in terms of the linearly
independent constants (6)

(1% b)

(A,,

9 A,,

{BM

, $2,

1
k=

fkqf sin k9

(14a)

1.2,.
/,-

[~j++1j+2I=fkQ:COSk9,
[C~,+~@~+J=f~Qfsink9.

> A,,

iB,3,

B,,

.-

>,

},

(16)

,I,/

n
I

_,

j , , / /I//,/

(14b)

1
1

Note that the rigid body motion terms in (4), i.e. the
terms 1, p cos 9 and p sin 9, should be conserved in
w,. Indeed such terms result in non-vanishing column
strains and as such are now essential in representing
the plate-column
interaction.
Also the number
NRIG in the stability condition (3) has now to be set
to zero since the resulting plate-column element has
no rigid body motion modes.

,I

((

\I

r=ap

\\

\\\

;:i-

\\\\

43/l-L

1 A,,

Ir=J2..,fktiCoSk9
.,.
+

CAS

RADIALLY

The variation of the thickness considerably complicates the generation of the internal displacement
field since the Lagrange equation (1) has to be
replaced by a more general plate equation with
variable coefficients [ 171. For the class of application
addressed in this paper, an approximate but efficient
solution of this problem can be based (Fig. 6) on
replacing the continuous variation of the plate thickness by a stepwise one. This makes it possible to
assume, for any particular interval (p,, p, + , ) presenting a constant thickness t,, the general solution of the
form (4) with integration constant

The expressions of the 4 x 2 matrices Qf, Q, and

of the load dependent vectors QkA,Qkaare listed in
Appendix I. Finally, the substitution of (13a, b) into
(4) makes it easy to present the element internal
displacement functions ti, and @, in the following
simple form
*, = WP+

TO HT ELEMENTS WITH
VARIABLE THICKNESS

,.

.
.

h
Fig. 6. HT plate element

with radially

variable

thickness.

168

J. JIROUSEK

and

and to link these constants

to the constants
-A
jlBk
of
the
previous
interval
(p,_ , , pi) by
k,
the following four transition conditions written for

M.

NDIAYE

With (19a, b) the internal displacement (2) can

finally be defined again in a form similar to relations
(14a, b) of Sec. 2

P =P,
G,= WP+

lw=-Iw

fkiq; cos k9

c
k = 0,1,2,..

a.
a
--lw=---~w
ap
ap

[@,+

cj+I=

Ak,,

c,+z=&

cj+3=

Oh,,

5+4=

(18)

O&z,

for any i > 0, to

/+I

+I$,

{ c/+2 1

9+3

+1qk8,

(1%

i cj+41 1

which is a form analogous to the relations (13a, b) of

Sec. 2. The chain rule used to evaluate from the
known expressions (Sec. 2 and Appendix I) of
OQf=Qf,
OQf=Qf,
d=qkA and Oqf=QkB the
expressions of the same quantities for the following
intervals i = 1,2, . . . , is of the form

i+is~-*QkA+is,4,

k9

(21a)

,aj+,]

fkiQ; cos k9,

(17)

A similar approach has in the past been successfully applied by Benda [19] and Jirousek [20] for the
analysis of circular plates of variable thickness. They
have found that:
l a relatively
low number of constant thickness
intervals yields already a good approximation of the
displacements and the internal forces of a plate with
a continuous variation of the thickness,
l the application
of the conditions (17) makes it
possible to derive a simple chain rule by which the
integration constants of all intervals are expressed in
turn in terms of the independent constants of the
interval (pO, pi).
A similar technique can also be applied here and it
leads with

i~~=tQf

f,$sin

(21b)

iR, = i - R, + R.

lAk=iQ;

1
k = I.Z....

iQks=iski-l~+tsf,

(20)

where iSk and si, isf are the transfer matrices listed
in Appendix III.

The only difference is that the expressions of G,and

@, are now different for each interval i.
Though in the structural applications (Fig. 4b) the
plate thickness is seldom varied symmetrically with
respect to the assumed middle surface plane x-y, this
fact, under thin plate theory, is considered of little
importance. As a consequence, the thickness variation is assumed symmetric with respect to x-y plane.
In the following section the admissibility of this
assumption is assessed by means of a 3-D analysis.
4.

NUMERICAL

STUDIES AND ASSESSMENT

4.1. Standard HT element

The HT p-method elements have originally been
presented [2] based on the biharmonic polynomials.
The aim of this subsection is to ascertain that the use
of the alternative set (8) of T-functions leads to
comparable results and to complete the study of the
convergence properties of these elements. Both the hand the p-refinement methods will be considered. The
latter will be characterized by the number M of
hierarchic DOF associated with the mid-side nodes.
Only the uniform p-refinement with the same M for
all mid-side nodes will be used. As shown in [2]
only the odd values M = 1,3,5,7,
. should be
considered since they preserve the desirable uniformity in the degree of the polynomial interpolation of
the normal and the tangential slope of the frame
functions 9, and G,.
An important question which arises during the
implementation of the HT element into a standard
FE code is the number m of Trefftz functions 3
(j = 1,2, . , m) to be used with a given p-refinement
level M. Numerical studies have shown that for
quadrilateral HT plate elements with curved sides the
minimum m as given by the necessary but not sufficient stability condition (3) always yields a stiffness
matrix with correct rank. This minimal numbers
(m = 13,21,29, 37 for M = 1,3, 5, 7) incidentally
also preserve the desirable geometric isotropy (invariance of element properties with respect to coordinate
axes orientation since then for any k9 both the terms
involving cos k9 as well as sin k9 are always preserved
in the set of @-functions of the element).

169

Hybrid-Trefftz element formulation

ln Na

al
- 22

6
&

p- convergence

8
la'

Ine

- 10
Na

hl

-b-

50

5
t/t,

Fig. 7. HT h- and p-method (a) convergence rates and (b)

study of discretization minimizing computer time.

Figure 7(a) presents a numerical h- and p-convergence study of a centrally loaded square plate supported at corners. Only a symmetric plate quadrant
has been discretized and the error in energy norm
defined as

(22)
(U = exact value of strain energy and U,, = HT
element approximation)
has been plotted on a
log-log scale against the total number N, of active
DOF of the FE mesh. The superiority of the pmethod over the h-method is immediately obvious if
the accuracy is considered in terms of N,. This
measure, however, disregards the actual cost of the
computation which is adversely affected by the increasing cost of evaluation of element matrices and a
more densely populated stiffness matrix of the assembled elements. Therefore in Fig. 7(b) the error e has
also been plotted against the computer time? to
indicate that as a rule, even in simple cases a 3 x 3 FE
mesh will computationally be more efficient that just
a single element representing the whole domain.
The same results as shown in Fig. 7(a) have also
been obtained with the alternative @-function set of
[2] (biharmonic polynomials). The small differences
observed in the last decimal digits of the results have
been due to the rounding-off errors and a close
inspection has shown that each of the two alternative
t Program
solver.

SAFE [16] with standard

skyline equation

sets of functions is actually a linear combination of

the other.
It should be pointed out that the results in Fig. 7(a)
have been obtained by specifying in the customary
FE way the concentrated load P as a nodal load.
Though highly accurate, still better results may be
achieved if the concentrated loads are represented by
the particular load term (9b) rather than as nodal
loads. The additional advantage is that the load
location is not confined to the mesh nodes. As an
example, Tables l-3 show some results of engineering
interest (displacements,
moments,
shear forces)
obtained in the extreme case of a single HT element
(Fig. 8) covering the whole square plate area (no
symmetry condition used) with different boundary
conditions.
It is worth noting that the accuracy of HT element
may further be improved if an optimal rather than
minimal number m of @-functions is associated with
each p-refinement level M. The use of unnecessarily
large m deteriorates the accuracy: m + 00 restores
perfectly the conformity, but the beneficial effect of
compensating the excessive rigidity of the frame by a
slight lack of conformity between the frame and the
internal field is lost. On the other hand, the minimum
m (warranting that the stiffness matrix has full rank)
may allow too much of non-conformity.
Extensive
numerical studies have, however, show that the
improvement of results due to an optimally tuned m
is not sufficiently significant to overweigh the
increased computational cost associated with a larger
H matrix. Furthermore, the optimal number m has
been found mesh dependent and for finer meshes
(3 x 3 and more) likely to be used in practice, the
difference between the optimum and minimum m
tends to disappear.
4.2. HT elements with radially variable thickness
The number of constant thickness subintervals
necessary to adequately represent a smooth thickness
variation has been studied by considering a symmetrically loaded circular plate represented by a single
HT element with just two curved sides (Fig. 9) used
to represent the circular boundary. The lowest prefinement level M = 0 (no hierarchic DOF) is
adopted since both the frame function definition as
presented in [ 151 and the internal displacement field

M DOF

Fig.

8. Square plate (different boundary conditions)

represented by a single HT p-method element.

J.

170

JIROUSEK

and M. NDIAYE

Table 1. Study of p-convergence on a simply supported square plate (Fig. 8). M = number of hierarchic mid-side DOF
(constrained DOF included). A single HT p-element used over the whole plate
Finite element/theoretical
Load

Uniform p

Concent. central P

M=l

value

Point

Quantity

Theoretical value

Dw: pa4
M,: pa2
Q,: ia
M, : pa2
M2: pa2
V: pa2

1.053
1.025
0.740
0.331
2.313
1.322

1.002
1.000
1.038
0.856
0.900
1.078

1.000
1.000
1.019
0.937
0.937
1.023

1.000
1.000
0.997
0.960
1.004
0.982

0.00406235
0.0478864
0.337657
0.032483 1
-0.0324831
0.06496618

Dw: Pa2

B
A

Q,: Pia
M,: P

1.025
0.763
0.812
1.686
1.249

0.998
1.126
1.234
0.551
0.892

1.000
0.980
1.002
0.953
0.977

1.000
1.022
0.972
1.087
1.030

0.0116008
0.417307
0.0609527
-0.0609527
0.121905

M*: P
v: P

Table 2. Study ofp-convergence on a clampled square plate (Fig. 8). M = number of hierarchic mid-side DOF (constrained
DOF included). A single HT p-element used over the whole plate
Finite element/theoretical
Load

Point

Quantity

Dw: pa4
M,: pa*
M,y: pa2

Uniform
p

Concent. central P

C
B

M=l

value

Theoretical value

Q,: pa

1.470
1.182
0.568
0.566

1.036
1.017
0.888
0.962

1.001
1.001
1.005
1.002

1.000
1.000
1.014
1.021

0.00126532
0.022905 1
-0.0513338
0.441301

Dw: Pa2
M,: P
0,: Pla

1.194
0.592
0.401

1.021
0.823
0.790

1.003
0.929
0.839

1.000
0.972
0.970

0.00561202
-0.1257705
0.793827

Table 3. Study of p-convergence on a square plate supported at corners (Fig. 8). M = number of hierarchic mid-side DOF
(constrained DOF included). A single HT p-element used over the whole plate
Finite element/theoretical
Load

Point

Quantity

M=l

value

Theoretical value

Dw : pa4

M,: pa2
Dw : pa
M,: pa2

0.988
0.988
0.988
0.976

1.000
1.039
1.004
1.024

1.000
1.000
1.000
1.002

1.000
1.000
0.999
0.993

0.0255065
0.111711
0.0177474
0.150439

Concent. central P

C
B

Dw: Pa2
Dw: Pa*
My: P

0.981
0.972
0.925

1.000
1.003
1.012

1.000
1.000
0.995

1.000
1.000
0.994

0.0391419
0.0229130
0.203004

Uniform

M=O

M=O

Fig. 9. Symmetrically loaded circular plate represented by

a single HT element. The lowest p-refinement level (M = 0)
is sufficient to accurately represent any case of symmetry of
revolution.

are capable of accurately representing any axisymmetric behaviour with the minimum number of DOF.
Two cases of radially variable thickness and of
boundary conditions have been considered as shown
in Fig. 10. The plate with exponentially varying
rigidity (Fig. 10a) has been taken from [19] and the
converged results for n = 500 subintervals have been
used as reference solution. For the plate with linearly
variable thickness (Fig. lob) the theoretical solution
has been taken from Timoshenko and WoinowskyKrieger [17]. Tables 4 and 5 show how the accuracy
in displacement w and moments M, and MS improves

Hybrid-Trefftz element formulation

1,
r=af, _.j

a
n

a
n

aa
n 2n

Fig. 10. Two cases of circular plate of variable thickness: (a)

exponential variation and (b) linear variation.

with increasing number n of constant thickness

subintervals over the element. Though the linear
variation of t appears more demanding, even in this
case, for n = 50 the error in the largest IU, is inferior
to 0.5%. Since the chain rule used to generate the
@-fur&ions along the set of subintervals is very fast,
a default value of n = 50 has been adopted for all
applications with radially variable thickness.
4.3. Applicability

ofplate theory

to the analysis of

slabs on columns

To facilitate the generation of reliable reference

results, two cases of centrally supported and

b)

Fig. 11. Uniformly loaded @ = 10kN/m2) centrally supported circular slab (a) with drop and (b) without drop.
uniformly loaded circular slabs have been considered: slab with a drop (Fig. lla) and without
drop (Fig. 1lb). In both cases, the HT element
solution consisted in using a single element with
two semicircular sides each with vanishing number
of hierarchic DOF (M = 0) at the mid-side
nodes A.

Table 4. Circular plate (v = 0) with experimental variation of thickness (Fig. 10a)

D,w: pa4
n

p=o.o

10
20
40
500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1254
0.1253
0.1252

0.1241
0.1240
0.1240

0.1204
0.1203
0.1203

0.1141
0.1140
0.1140

0.1053
0.1052
0.1052

0.0939
0.0938
0.0938

0.0798
0.0798
0.0797

0.0632
0.0631
0.0631

0.0440
0.0440
0.0440

0.0227
0.0227
0.0227

0.0000
0.0000
0.0000

0.1252

0.1240

0.1203

0.1140

0.1052

0.0938

0.0797

0.0631

0.0439

0.0227

0.0000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

M,: paa2

0.0

0 =

10
20
40

0.2484
0.2484
0.2484

0.2451
0.2455
0.2456

0.2368
0.2371
0.2372

0.2231
0.2234
0.2235

0.2042
0.2045
0.2045

0.1804
0.1807
0.1807

0.1521
0.1523
0.1523

0.1196
0.1197
0.1198

0.0832
0.0833
0.0833

0.0433
0.0433
0.0433

0.0000
0.0000
0.0000

500

0.2484

0.2456

0.2373

0.2235

0.2046

0.1807

0.1524

0.1198

0.0833

0.0433

0.0000

0.1

0.2

0.3

0.4

MS: paa2

p=o.o

0.5

0.6

0.7

0.8

0.9

1.0

10
20
40

0.2484
0.2484
0.2484

0.2454
0.2451
0.2450

0.2355
0.2352
0.2351

0.2197
0.2194
0.2193

0.1992
0.1989
0.1988

0.1753
0.1750
0.1750

0.1495
0.1492
0.1492

0.1231
0.1228
0.1228

0.0973
0.0971
0.0971

0.0732
0.0731
0.0730

0.0513
0.0512
0.0512

500

0.2484

0.2450

0.2351

0.2193

0.1988

0.1749

0.1491

0.1228

0.0971

0.0730

0.0512

J. JIROUSEK and M. NDIAYE

172
Table

5. Circular

plate (v = l/3) with linearly

variable

of thickness

(Fig.

lob)

D,w: pa4
n

p = 0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

10
20
40
50

0.3785
0.3750
0.3738
0.3737

0.3481
0.3459
0.3452
0.3451

0.298 1
0.2967
0.2962
0.296 1

0.2453
0.2443
0.2439
0.2439

0.1930
0.1923
0.1921
0.1920

0.1423
0.1418
0.1416
0.1416

0.0932
0.0929
0.0928
0.0927

0.0458
0.0456
0.0456
0.0456

0.0000
0.0000
0.0000
0.0000

0.3734

0.3449

0.2960

0.2438

0.1920

0.1415

0.0927

0.0456

0.0000

p = 0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.10

10

0.0735
0.078 I
0.0795
0.0797

0.0516
0.0546
0.0557
0.0558

0.0403
0.0434
0.0440
0.0440

0.0344
0.0357
0.0364
0.0365

0.0290
0.0299
0.0302
0.0303

0.0229
0.0236
0.024 1
0.024 1

0.0157
0.0172
0.0173
0.0173

0.0089
0.0090
0.0094
0.0094

0.0000
0.0000
0.0000
0.0000

0.0800

0.0560

0.0442

0.0366

0.0303

0.0242

0.0173

0.0094

0.0000

Theory

0.1ozY

IV,: pa2

20
40
50
Theorv

iVg: paa2
n

n = 0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.10

10
20
40
50

0.0245
0.0261
0.0265
0.0266

0.0466
0.0545
0.0539
0.0525

0.0906
0.0882
0.0882
0.0882

0.1430
0.1291
0.1285
0.1313

0.1746
0.1742
0.1741
0.1740

0.2070
0.2250
0.2243
0.2206

0.2817
0.2791
0.2788
0.2788

0.3612
0.3377
0.3369
0.3413

0.4002
0.3989
0.3984
0.3983

Theory

0.0267

0.0539

0.0882

0.1284

0.1740

0.2243

0.2787

0.3368

0.3982

(a) Slab with drop (Fig. 11~). Since the thickness of the slab in the region of the slab is important
and the mid-surface is not plane (the slab is eccentric) some of the basic assumptions of the classical
thin plate theory have been violated. To assess
the practical consequence of this situation, the HT
single element solution has been compared with
results of a 3-D analysis performed by using a
fine mesh of 328 axisymmetric
isoparametric
quadratic elements leading to a total of 2430DOF.
To analyse the effect of eccentricity, two 3-D
analyses have been performed wherein the geometry

of the drop is either taken as shown in Fig. 11(a)

or with the same thickness variation applied symmetrically with respect to the middle plane of
the slab. The results of the two 3-D analyses are
displayed in Fig. 12 along with the HT element
solution
(a single element with 50 constant
thickness subintervals in the drop region). The
inspection of the results shows that the engineering
accuracy of the simple HT element solution is quite
satisfactory.
(b) Slab without drop (Fig. llb). A single HT
element now yields the exact Kirchhoff theory

cm1

ml

t?

M,( kNm/ml

-Q-

3-O with actual exrentrlc drop geometry

34~1th drop symmetnc wth

--v--

HTP-extension

respect to the rnlddk plane of the slab

With rpeclal purpose

T- functom

ml

Fig. 12. Comparison

of results

for circular

slab with drop of Fig. Ila.

Hybrid-Trefftz

element formulation

Fig.

__+L__
_-Q__

Slab wth drop-HT soluhon

Slab ulthout drop-HTsolutmn

Slab without drop-Plate

udh tranvene

theory

shear defnrmatwn

Fig. 13. Influence of drop on distribution of (a) displacements and (b) moments. Comparison of cases of Fig. 1I(a)

and (b).
solution of annular plate with the following boundary
conditions
intemalboundaryr=r,,...,--0,

aw
&

w=2-

r = a, . . . , w = M, = 0,

external boundary

QA
Era
(23)

where r. and 1 are respectively the radius of crosssection and the length of the column. As a matter of

173

15. Example of HT element discretization

rectangular flat slab with drops.

interest some results of the cases in Fig. 1l(a) and (b)

have been compared in Fig. 13 to highlight the
influence of the drops on the slab behaviour.
To ascertain the applicability of the classical
Kirchhoff plate theory which disregards the transverse shear deformations, the case of Fig. 1l(b) with
boundary conditions (22) has also been solved following the Mindlin plate theory. The comparison of
the results in Fig. 14 shows that the Kirchhoff theory
if sufficiently accurate in such applications
for
engineering practice.
The Mindlin plate theory raises the interesting
question of modelling slabs on columns in FE analysis based on the popular isoparametric Co plate
elements. Indeed, in sharp contrast with the classical
Kirchhoff plate theory, a concentrated load now
produces infinite displacements and as a consequence,

twtmm)
-6

plate thecq wth

_.w_

Mesh of 156 Conventional lsoparametw

quadratIc M~ddlm plate elements.

-L

transverse shear deform&on

Column zore represented

elarhr

-2
0

of

_-_

as slab on

found&on

S,ngle HT 2 rIded element ~11th

spew1 purpose T- functiom

Fig. 14. Comparison of different solutions for centrally loaded circular slab of Fig. 11(b).

174

J. JIROUSEK and M. NDIAYE

----

--y,

-----.

\
\,,

175

Hybrid-Trefftz element formulation

M,/M, (kN

m/m)

- --

Poshve

---

Negative

Fig. 17. Principal

moments

IO

I
20

kN m/m

in the slab with drops

the use of point supports is, in principle, illicit.

Whereas in a crude mesh of Co Reissner-Mindlin
elements a point support leads to results close to
those of the Kirchhoff theory (where a point support
is licit), the solution becomes unpredictable if the
mesh is strongly refined in the vicinity of the support.
To avoid this situation, a possible alternative consists
in simulating the columns by using over the area
corresponding to the column cross-section one or
more Co plate elements on elastic (Winkler) foundation. (This amounts to augmenting the customary
plate element stiffness matrix by the following area
integral: C, jA N%dx dy, where N is the matrix of
customary Co shape functions and C,,.is the Winkler
coefficient.) As a matter of interest, Fig. 14 shows the
results of such a calculation (156 isoparametric
quadratic Co plate elements over a symmetric plate
quadrant) which however exibits now a marked
difference with respect to the earlier exact Mindlin
theory calculation following the boundary conditions
(22). This difference is due to the fact that the Winkler
foundation leads to somewhat unrealistic deformation of the area simulating the cross-section of the
column. The remedy is to suitably increase the Young
modulus of the plate elements covering this subregion.
However, in contrast with the HT approach outlined
in this paper, this technique cannot correctly represent the effect of the flexural rigidity of the column.
4.4. Practical application
A symmetric quadrant of a rectangular slab with
drops has been solved by using a crude 5 x 7 mesh of

of Fig. IS. Uniform

load p = 10 kN/m2.

HT plate elements (Fig. 15). The four elements

including a drop and a column use the optional
special purpose Trefftz functions following the Sets
2.2 and 2.3 with 50 constant thickness subintervals in
the zone of the drop. The remaining HT elements use
the standard set of Trefftz functions (Sec. 2.1). Some
results for a uniformly distributed load are displayed
in Figs 16 and 17.
5. CONCLUDING

REMARKS

The concept of optional special purpose Trefftz

functions has been used along with p-extension of the
Hybrid-Trefftz (HT) plate elements to provide a
highly efficient approach to the analysis of flat slabs
with drops. The slab is discretized by a crude mesh
of HT p-method element where each column with the
adjacent part of the slab including the drop are
represented by a single element. Suitable set of Trefftz
functions accounts for the plate-column interaction
and accurately represents the local effect in the vicinity of the column head. As a consequence accurate
moment concentrations are obtained without mesh
refinement, as a part of the full structure analysis of
the slab.
Examples assessing the practical efficiency of the
advocated approach have been provided. Apart
from considerable economy in computer and data
preparation costs, the excellent accuracy and high
p-convergence rate (generally associated with HT
model) are valuable assets. The possibility of a simple
adaptive reliability assurance based on a uniform

176

J. JIROUSEKand M.

p-refinement
and specially devised for use with
the standard
FE codes,
has been
presented
elsewhere [4, 51.
The implementation
of the presented HT elements
is simDle since a single HT element subroutine Drovided with a library of optional Trefftz functions
covers the whole problem. As shown in the paper, a
variety of cases including a standard plate element
plate-column
elements
(with or without
drops)
and an element with a circular hole. have been
obtained in a unified manner from a complete set of
biharmonic
functions in polar coordinates.
1

Acknowledgement-M.

NDiaye was supported by the Swiss

Confederation.

9. J. Jirousek, A cqntribution to finite element and associated techniques for the analysis of problems with stress
singularities. Proc. 2nd Int. Conf. Num. Methods Engng
(GAMNI), Vol. 2, pp. 719-729. Dunod, Paris (1980).
10. J. Jirousek and P. Teodorescu. Laree finite element
method for the solution of problems-in the theory of
elasticity. Compuf. Sfruct. lj, 575-587 (1982).
_
11. R. Piltner, Spezielle finite Elemente mit Liichern,
Ecken und Rissen under Verwendung von analytischen
Teilliisungen.
Fortschritt-Berichte
der
VDIZeitschriften, Reihe 1, Nr. 96. VDI, Diisseldorf (1982).
12. J. Jirousek, Implementation of local effects into conventional and non-conventional
finite element formulations. In Local Eficts in the Analysis of Structures
(Edited by P. LadevBze), pp. 279-298. Elsevier
(1985).
13. i. Piltner, Special finite elements with holes and
internal cracks. Inr. J. Numer. Meth. Enpnp 21.
1471-1485 (1985).
14. T. G. Gerhardt, A hybrid/finite element approach for
stress analysis of notched anisotropic materials. Trans.
ASME 51. 804-810 (19841.
15. J. Jirousek and M. NDiaye, Solution of orthotropic
plates based on p-extension of the Hybrid-Trefftz finite
klement model. -Cornput. Strucf. 34, 51-62 (1990).
16. J. Jirousek, Manuel dutilisateur du programme
SAFE-RCdaction
provisoire,
LSC-DGC,
Swiss
Federal Institute of Technology, Lausanne (1990).
17. S. Timoshenko and S. Woinowsky-Krieger, Theory of
Plates and Shells, 2nd Edn. McGraw-Hill, New York
(1969).
18. A. Venkatesh and J. Jirousek, Accurate FE analysis of
thin plates under concentrated loads. IREM Internal
Report 89/6, Swiss Federal Institute of Technology,
Lausanne (1989).
19. J. Benda, Circular plates of variable thickness (in
Czech). Bulletin of the Brno Institute of Constructional
Engineering and Architecture, Czechoslovakia, Nos 39
and 47 (1954).
20. J. Jirousek, Solution of arbitrarily loaded circular plates
of variable thickness (in Czech). Bulletin of the Brno
Y

REFERENCES
1. J. Jirousek and Lan Guex. The Hvbrid-Trefftz finite
element model and its application to plate bending. Inf.
J. Numer. Meth. Engng 23, 651-693

(1986).

2. J. Jirousek, Hybrid-Trefftz plate bending elements with

p-method capabilities. Int. J. Numer. Meth. Engng 24,
1367-1393 (1987).
3. J. Jirousek and A. Venkatesh, Implementation of curvilinear geometry into p-version HT plate elements. Inf.
J. Numer. Meth. Engng u), 431-443

(1989).

4. J. Jirousek and A. Venkatesh, A simple stress error

estimator for Hybrid-Trefftz p-version elements. Int. J.
Numer. Meth. Engng 28, 21 l-236 (1989).
5. J. Jirousek and A. Venkatesh, Adaptivity in HybridTrefftz finite element formulation. In!. J. Numer. Meth.
Engng 29, 391-405

NDIAYE

(1990).

6. J. Jirousek and N. Leon, A powerful finite element for

plate bending. Comp. Meth. Appt. Mech. Engng 12,
77-96 (1977).

7. J. Jirousek, Basis for development of large finite elements locally satisfying all field equations. Comp.
Meth. Appl. Mech. Engng 14, 65-92 (1978).

Institute of Constructional Engineering and Architecture,

8. A. P. Zielinski and 0. C. Zienkiewicz, Generalized finite

element analysis with T-complete boundary solution
functions. Inr. J. Numer. Meth. Engng 21, 509-528
(1985).

Czechoslovakia, No. 106 (1956).

21. A. K. Noor and I. Babuska, Quality control and
assessment of finite element solutions. Finite Element
Analysis and Design 3, l-26 (1987).

APPENDIX

AUXILIARY MATRICES FOR GENERATION OF

INTERNAL DISPLACEMENT FIELD OF HT
PLATE-COLUMN
ELEMENT

k=O

-$(I

16~~0
~--

+2lnp,)

a2DKz

P:
D2

$(l-2lnp,)

DP,
0

q,=!

I:

-g(l+2lnp,)-+&$(l-2lnp,)-g(l
PO
x+%(1-4lnp,)+*

+4lnp,)

z-

Hybrid-Trefi

element formulation

177

where
2p, In *p,,

8n

c=Dz-J? = uniform load, P,,

= xa2p~p
+ P, P = concentrated

a2DK, PO
load (Fig. 4a).

k=l

Q: or Qf=-

-Kpi(l

-2lnp,)-8nDp2

->K

-4K

PO

where

e=i[K(l

+2lnp,)-8nD]

K =K, or K, for Tf or Tf
q!=qf=[O

0 0 oy

k>l

_kp#-

-(k

1)

1)~~ ;

where

APPENDIX

AUXILIARY MATRICES FOR GENERATION OF

INTERNAL DISPLACEMENT
FIELD OF HT ELEMENT
WlTH CIRCULAR HOLE (RADIUS r,=crp,)

k=O
0

!
0

Q,=

qt = t apg

If&p+v

2+?$
0
0
0

+4(* +v)lnp,]
-1

J. JIROUSEK and M. NDIAYE

178
k=l

q: =qf= [O 0 0 O]T.
k>l

k -2v

0
k - 2v

Q;=Qf=&

;(2+k)(k-2-2v)p;~k+

(k - l)(k --2h$

I
-k(k*-2+2v)p;-

y&k

qf=l$=[O

TRANSFER

O]?

APPENDIX

+ 1)(2+k)d

III

MATRICES

S, AND s,

k=O
D

-pf(D

- 1)(2lnp,[D(l -v)

l)(l +v)/2

+ 1 + v]/2

(D - l)(21npi-

1)(1 -v)/4

-pf(D

+ v) ln2p,]/4

(D - I)(1 - v)(l + 2 In p,)/4

(D - 1)(1 - v)l(4d)

S, =

- I)[1 - v +4(1

p;(D - l)(l + v)

[D(l + v) + 1 - v)]/2

pf(D

- 1)(1 + v)(l + 2 In p,)/2

1

k=l

S, =

pf(D - 1)(3 + v)/2

[D(l -v)

-pf(D

(D -

+ 3 + v]/4

-(D

- I)(3 + v)/4

1)(1- vY(2d)

(D - I)(1 +v +4lnp,)/4

- 1)(1 - v)l(4d)

(D - 1)(1 -v)/(W)
-p;(D

[D(3+v)+l-v)]/4

- 1)(3 + v)/8

k>l
[D(3+v)+l-v]/4

p;%(D

pf(D-l)(l+v)
x (k + 1)/W)

[D(l - v) + 3 + v]/4

- I)(1 -v)
x (k + I)/4

_p,-Y+k)(D - 1)

pf -(D

-p;%(D

- 1)[2(1 + v)
+ k2(l - v)]/(4k)
- l)(l - v)(k - I)/4

x(1 -v)k/4

is,=
-p(D

p;u

- 1)(1 -v)
x (k - 1)/4
+(D

_ 1)

x (1 - v)k/4

--P;+~(D

p(D

- 1)[2(1 + v)
+(l - v)k2/(4k)

- 1)(1 - v)(k + 1)/4

[D(3 + v) + I - v]/4

p:(D - l)(l + v)(k - 1)/(2k)

[D(l -v)

+ 3 + v]/4

Hybrid-Trefftz

element formulation

with

'D
D=i-D

st=

[~;;~;~l-21npi)K+api(lnpil)Rl 1 ,
-x

where

(1 + 2 1x1p,)Ap -z

1-W + ap,(l + In p,)RI

179