You are on page 1of 11

ht. 1.

Solids Structures,

Vol.
II,
pp,425-!35.
Pergamon
Press
1975.
Printed
inGreat
Britain

FINITE STRIP-DIFFERENCE
CALCULUS TECHNIQUE
FOR PLATE VIBRATION PROBLEMS
and D. V. REDDY~

C. SUNDARARAJAN?

Faculty of Engineering and Applied Science, Memorial University of Newfoundland,


St. Johns, Newfoundland, Canada
(Received 23 July 1973; revised 13 March 1974)
Abstract-The
paper presents the applicarion of the calculus of finite differences to obtain an explicit
expression for the natural frequencies of the finite strip model of a simply supported orthotropic rectangular
plate. This analytical solution not only involves far less computational work than the conventional finite strip
method, but also enables parametric studies for accuracy and convergence of the finite strip approximation.

INTRODUCTION

The finite strip method[l], which is a Kantorovich type formulation of combining the finite
element method and the Fourier series technique, has been used by many investigators for the
free vibration analysis of plates[2-4]. This method requires less computer storage and time,
compared to the finite element method, because of the reduced size of the stiffness and mass
matrices involved in vibration analysis. However, as the number of strips increases the amount of
computer work increases considerably.
For a rectangular plate of uniform thickness, simply supported along all the four edges,
approximated as an assemblage of a number of finite strips of equal width, it is possible to obtain
_an analytical solution using the calculus of finite differences[5,6].
The computational
work
involved is independent
of the number of strips. Hence the finite .strip-difference
calculus
technique can be used to test the accuracy of the finite strip approximation with increasing
number of strips.
STRIP

STIFFNESS

MATRIX

Consider a rectangular plate of uniform thickness and sides 1 and a simply supported along all
the four edges. It is approximated as an assemblage of N, strips (Fig. la). The lateral deflection of
a typical strip of width d (Fig. lb), defined by sides j and (j + 1) is assumed as

Wwi,~, [(1-$+$J

Wi,s

( -y+$)
X

0j.q

(s-s)

wj+t,q

($-$) &+t,q]
Y,(Y)

where the Y,(y) are functions

(1)

which satisfy the boundary

conditions

of the strip at y = 0 and

tPost-Doctoral Fellow.
$Professor.
A similar difference calculus approach has been used by Leckie [7] to test a Hrennikoff model approximation
425

S&S Vol. II. No. 4-D

of plates.

426

C.

SUNDARARAJAN

/STRIP

and

D. V.

REDD~

INTERFACE

L-

w
(a)

(bl

LAYOUT

TYPICAL

STRIP

Fig. I. Finite strip assemblage of the plate.


Y = a, ~i.~ and 19,,, are the lateral deflection and slope (aw,.,/ax) at the jth strip interface
corresponding to Y,(y). Since the strips are simply supported, the Y,(y) may be taken as

Y,(y) = J:sin$?

Equation

(2)

(1) can be written in the matrix form


W(X,Y I=

[Nl@I =

i
D%{A},
4=1.*...

(3)

where

[iv],=

[(,_$+$), (,_?z+C
#),(g-g,
d

{Al, = [wj.q4.q

wj+l.q

[Nl = [[~lI[m~

. [ml

{ii} = [[Al,[Al~.

. . [A]rT].

(g-g)]Yqiy)j

ej+l,,l
(4)

and

The curvature

matrix

(5)

+2-

can be expressed

in terms of the displacements

a%
axay
as

(6)

421

Difference calculus technique for plate vibration problems

Here
[Cl = [[Cl,[Cl*. . . [CL1

(7)

and

-(-$+$)Y'
[Cl,

(L_$+$)
2( -$+%)

-(-;+$)Y

y; _

-(-$-%)Y,

(x _zg+$)

y;

Y; 2(1-;+3

Y;

-($-f)Yq

_ (kp$)

2(p$)

y;

Y4 2($-3

(p_J

y:'

y;
i

where

Y=dY,

dy

yd-m

dy

The bending and twisting moments for an orthotropic plate are given by

= [Dl{xl

(9)

where

in which D,,D,,D,,D,,are the flexural rigidities of the plate.


The strip stiffness matrix is given by
[RI =

l I,[Cl~~l[Cl

dx dy

(10)

which yields
[klu [kln. . . [kl,r

(11)
where
[kl,, = I/ [~l,UX[~l,
0 0

dx dy.

(12)

428

C.SUNDARARAJAN
andD.V.REDDV

For strips simply supported at both ends with Y, taken as in (2) it can be shown that
[k],, = 0 for p #q.

(13)

[kl,, is given explicitly in Appendix 1.


STRIP MASS MATRIX

The consistent strip mass matrix has been used by earlier investigators for the plate vibration
problems[2-4]. In this paper a concentrated line mass matrix, in which the distributed mass is
assumed to be concentrated as line masses along the strip interfaces, is used because it is simpler
to use in conjunction with the difference calculus technique.
Let 1 be the uniform mass per unit area of the strip. For a typical strip the intensity of the line
mass along its edges is p(d/2) per unit length. So the inertial loading at the strip edges j and (j t 1)
during free vibration is

and

(14)

Equation (14) can be written in terms of [N] and {A}as

Using the principle of virtual work, as in the finite element method [8], the inertial force vector
is obtained by
[N]=Q dx dy
(Fl=~~d

(16)

where Q is the inertial force distribution. Expressing Q in terms of Qj and Qj+, and substituting
in (16) gives
{F} = ~2w2Ia Id [iif]=[Iif1{~}(A(x,o)+ A(x,d))dx dy
0 0

(17)

where A(x,a) is the Dirac delta function. We can also express the inertial forces in terms of the
frequency, displacement and mass as
{F} = 02[fi]{A}

(18)

429

Difference calculus technique for plate vibration problems

where [fi] is the strip mass matrix. Comparing (17) and (18)
Irfil=p;

II0a0d[~l[~l(~ko)

+ Ku!)) dx dy
. . [NIITLNlr1

[ml,,

[Nl*=[Nll

...

: [N],T[N],

. . . [Nl,va

[ml*2 . . .
.
...

=I
[ml21

[ml,,

where

=P$

Ml,
.

. . . [mlrr .

(19)

O([lOOO]T[lOOO]+[OO1O]TIOOIO]Y,Y,dy

(20)

Y,Y, dy.

Because of the orthonormal property of the Y, taken in (2)


[ml,, =0 for pfq

(21)

and

loo0
[ml,, = Pz

EQUILIBRIUM

0000

0010
pooo

EQUATIONS

(22)
I

FOR THE PLATE

The strip stiffness and mass matrices and the strip displacement vector are assembled to form
the plate stiffness and mass matrices, [I?],, and [&f],,, and the plate displacement vector {81q,,.
The boundary conditions in the x-direction are incorporated in the above matrices by deleting the

430

SUNDARARAJAN and

C.

D. V.

REDDY

rows and columns corresponding to the zero displacements. The resulting stiffness and mass
matrices and displacement vector are [K],,, [Ml,, and {S}, respectively.
Because of (13) and (21) we have
[K],, = 0 for p # q
[Ml,, = 0 for p # q 1 '

and

(23)

So the equilibrium equation can be written mode by mode (i.e. for each integer value of q
separately). The equilibrium equation for the qth mode is

[K],,(6),

- o[M],,{6}, = 0 for q = 1,2,. . .

(24)

The matrices [K],, and [Ml,, are of the order


N0 = [2(NS+ 1) - boundary conditions incorporated in the x-direction].
THE FINITE

STRIP

METHOD

For the nontrivial solution of (24)


Det([K],, - ~~[ivI],,) = 0.

(26)

This is a typical eigenvalue problem. The square of the frequencies of vibration, w2, are obtained
as the eigenvalues of the matrix

[a&a7,.
As the number of strips N, increases the order of the above matrix, NO,increases twice as fast
and hence the computation involved also increases considerably.
FINITE

STRIP-DIFFERENCE

CALCULUS

TECHNIQUE

Because of the repetitive pattern of the strip assemblage, it is possible to use the finite
difference calculus [5,6] in conjunction with the finite strip formulation for the frequency analysis
of rectangular plates.
The equation of motion of the q th mode, before the incorporation of the boundary conditions
in the x-direction (these will be incorporated later) is

([al, - dm,,)~~~,
This can be expanded as
(1, (1)
SII St2 s (1)
13
(1) (1) S::
SZl SZ?

s (1)
14

0
(1) (1)
12'
+ s::' s:'a+s:': s,,
s31 S3? s::'
(1)
S4I sg sgfs::' syj+s~l s:':

10

s ::'

s :la

s %'

s::'ts::'

tFor the sake of simplicity, the suffix q will be dropped henceforth.

= 0.

(27)1-

Difference calculus technique for plate vibration problems

431

(28)
where
s!! = k!!- w*m?. i = 1, 2, 3,4; j = 1, 2, 3,4 and I = 1,2.. . N, in which the superfix (I)
denotes yhe nlmber of ide strip and k!:',rn!,?are the elements of the strip stiffness and mass
matrices [kl,, and [ml,, respectively.
Since all the strips have identical elastic and geometric properties, the mass and stiffness
matrices are the same for all the strips, i.e.
(I)

mij

jf =

kij

m ii =
k

S{;=Sij

(29)
for 1=1,2,...Ns

Substituting (29) in (28) we find, on expanding the matrix equation into a system of simultaneous
algebraic equations, that any pair of equations except the first and the last are similar. A typical pair
of equations is
(k31- w*msl)Wj-1+
(ka -

w*msz)ej-1+
(km -

CL,- w*ma)wj-1+
(ka - wzm4z)ej-1
+ (kz1+(kzz -

02mz2+ k, -

02m,3 + kll-

o*m,l)wj

ozm2, + k43-

w*m43)wj

o*m4)Oj
+ (kz3 - w*mz3)wj+l+
(kz4- w*m.d&+1=
0.

WW

Because of the physical meaning of the stiffness coefficients, we have


k,, = km
klz = kz, = -k34 = --Lx,
k,, = ks,,
k,, = k4, = -kz, = -hz,

and

(3lYt

kzz= k,
kz, = ka.

The coefficients given in Appendix 1 verify equations (31).


Substituting (22) and (31) in (30a) and (30b) we get,

ktdwj-l+ Wj+l)+ k,,(ej+, - Oj--l)+ 2kl,wj = ,.h2wj,

Wa)

tFor example, k,, means the force required at the jth side of the strip to cause unit deflection (w) at the jth
side. This is same as k,,, the force required at the (j + 1)th side of the strip to cause unit deflection at the
(j + 1)th side.

C. SWDAR~RAJ~N

432

Introducing

the Boolean difference

operator

and D. V. REDDY

E, defined by
(33)

and
equation

(32) can be written as


(kll(E

+ E) + 2k1, (k14(E

Substituting

/.L~w]w~

Em)lOj=0,

CW

E-l) + 2kdi = 0.

Wb)

[kll(E

- E)]wj + [k&E +

(34b) in (34a), we get


[k:,(E - E-)2]
[k,,(E + E-J + 2k,, - Pdm*l+ [kz4(E + E-1) + 2ks2, wi = O.

(35)

wi may be assumed as
(36)
where (P,,(j) satisfies the displacement boundary conditions at x = 0 and x = 1 (i.e. j = 0 and
j = ids). Since the plate is simply supported at x = 0 and x = I, a sinusoidal form of (b(j) may be
assumed; i.e.
&(j) = sin?.

Substituting

(37)

(36) and (37) in (35) and using the relations

and

in which i = g-

Imag. &A,,

(38)

1, we get

(k13 (exP (g)+eXP

($)]+2kl~-~W2)

{ki,[exp

(z)

-exp

i{k21[eXP($)+exp($)]+2kr:}

(F)r]

.
-=
exp(

(39)

Difference calculus technique for plate vibration problems

433

where Imag. denotes the imaginary part. Simplifying (39) we get


2k,,co++2k,,-pdo2
s

so,
2k,, co+

2k,, - pdw2

1
nnlNS3
=ofor
n=1)2*..w
.
1 [%sin
-

. nnj
[2k:, sin* nn/Ns]
[kSdcos nr/N, + kZZ] ln % = *

lkZdcos m/N,

+ kzzl

(40)

(41)

From (41) the frequency w*,,corresponding to the q th sine mode along the strip (Y-direction) and
the nth sine mode across the strip (X-direction) is given by
pdw Zqn
=

-[2k& sin2 m/N,]


+ 2k,, cos m/N,
[kZdcos m/N, + krzl

+ 2k,,

(42)

where the kij correspond the q th sine mode along the strip.
By changing the value of q and n the various frequencies can be computed. It is interesting to
note that the computational work involved in calculating the above expression is independent of
the number of strips, N,, whereas it increases with N, in the conventional finite strip method.
Hence the present method is ideal to study the accuracy and convergence of finite strip
approximation.
If the plate has boundary conditions other than simply supports at x = 0 and x = I, then 4,,(j)
can not be assumed as a sine function. When other appropriate functions are chosen, the resulting
equations will not decouple in n as in equation (41). But a set of coupled linear algebraic equations
will result and they have to be solved numerically. Hence a simple analytical expression for the
frequencies, similar to (42), can not be obtained.
NUMERICAL

RESULTS

AND DISCUSSION

The natural frequencies of a rectangular simply supported plate of I/a = 2, and D, = 1, D, = 1,


Dx, = O-35and D, = 0.3 are calculated by the conventional finite strip method [equation (26)] and
the present method [equation (42)]. For different values of N,, the approximate execution time
required by an IBM 370-175 digital computer to calculate the (Ns - 1) frequencies are given in
Table 1. (In all cases q is taken equal to 1). It shows that as the number of strips, N,, increases the
execution time increases considerably for the conventional method, whereas the increase is very
little for the present method.
Table I. Execution time. (in l/IO,000of a second)
N.<

Conventional method

Present method

798
1863
10549
35310

432
665
865
865

IO
20
30

Squares of frequencies obtained from both the methods, along with the exact valuest, are
plotted in Fig. 2. The present method gives a lower bound in all the cases and converges towards
tFor a simply supported plate, the exact solution is obtained by assuming
w(x,y) = 2
2 II,, sin y
n=,.Z 4=1.2.

sin y.

434

C.

and D. V.

SLJNDARARAJAN

10
NUMBER

IPW
0

p,e*ent

I
30

1
20

30

20

I
30

OF STRlPS

10

20
OF STRIPS

10
NUMEE9

REDDY

NUMBER

OF STRlPS

method

-------

CcmMnt~onal

method

Fig. 2. Comparison of results.

the exact value as N, increases. The conventional method gives lower values in the beginning
increases above the exact value and then decreases towards the exact value. Moreover, in all
cases the conventional method gives a higher value than the present method.
The conventional method is a computer-oriented
eigenvalue analysis of equations (27), while
the present method takes advantage of the repetitive pattern of the equations and uses the Calculus

Difference calculus technique for plate vibration problems

435

of Finite Differences to obtain analytical expressions for the eigenvalues. The differences in the
results, as seen in Fig. 2 may be mainly due to the fact that the eigenvalues obtained by the
conventional method are approximate due to errors in the numerical analysis of eigenvalues of a
matrix, whereas the eigenvalues calculated from equation (42) are exact for the approximate
strip-assembled plate model.
CONCLUSIONS

The finite strip-difference calculus technique proves to be a powerful tool to study the
accuracy and convergence of the finite strip approximation. Whereas the conventional matrix
technique gives approximate results, the present method gives an exact solution for the
approximate structures, namely the strip-assemblage. Computational work is also considerably
reduced in the present method.
The difference calculus technique for a plate vibration problem described in this paper, can be
easily extended to static and stability problems of plates and shells.
Acknowledgements-The authors are grateful to Dr. A. A. Bruneau, Dean of the Faculty of Engineering and Applied
Science, Memorial University of Newfoundland, for his interest and encouragement. This research was supported by the
National Research Council of Canada through Grant No. A8119.
REFERENCES
1. Y. K. Cheung, The Finite Strip Method in the Analysis of Elastic Plates with Two Opposite Simply Supported Ends, Pm.
Inst. Civil Engrs. Great Britain, 40, I (1968).
2. M. S. Cheung, Y. K. Cheung and D. V. Reddy, Frequency Analysis of Certain Single and Continuous Span Bridges,
Developments in Bridge Design and Construction (Edited by K. C. Rockey, J. L. Bannister and H. R. Evans) p. 188.
Crosby, Lockwood and Sons Ltd., (1971).
3. Y. K. Cheung and M. S. Cheung, Flexural Vibrations of Rectangular and Other Polygonal Plates, Proc. AXE 97, EM2.391
(1971).
4. P. V. Thangam Babu and D. V. Reddy, Frequency Analysis of Skew Orthotropic Plates by the Finite Strip Method, J.
Sound and Vibration 18, 465 (1971).
5. C. Jordan, Calculus of Finite Diferences. Chelsea Publishing Company,
6. T. Wah and L. R. Calcote, Structural Analysis by Finite Diflerence Calculus. Von Nostrand Company (1970).
7. F. A. Leckie, The Application of Transfer Matrices to Plate Vibration, Ingenieur-Archiu XxX11, 100 (1963).
8. 0. C. Zienkiewicz, The Finite Element Method in Engineering Sciences, p. 322. McGraw-Hill, London (1971).
APPENDIX

$ D, t ; dc4D,
+cc2D
5d

+zc2D

Sd

D. + $, d3c4DY

$Dx+L$~Dy

+;cD~+;cD,,

ltISdc2D,tT5dc2DXy

fkl,, = -$Dx +;dcD,


12

symmetrical

--$ D, +$

d2c4D,

$Dx

-sdc2D,-$c2D~Y

--~c~D,-~c~D,,

+gc2D

60,
4

f D, -&

-$Dx

-$DY

+;cD,+$*D.,

-A

d3c4D,

dc2D, -is dc2D,,

5d

+;dcD,
+$D

5d
-&dcD,

=
$Dx +&dYDy
4
8
+i-jdc2D,tISdc2DxY

You might also like