Computers and Structures 89 (2011) 395410

Contents lists available at ScienceDirect

Computers and Structures

journal homepage: www.elsevier.com/locate/compstruc

Dynamic stiffness elements and their applications for plates using rst order
shear deformation theory
M. Boscolo , J.R. Banerjee
School of Engineering and Mathematical Sciences, City University London, Northampton Square, London EC1V 0HB, UK

a r t i c l e

i n f o

Article history:
Received 7 June 2010
Accepted 9 November 2010
Available online 4 December 2010
Keywords:
Dynamic stiffness method
Hamiltons principle
Free vibration analysis
Plates
First order shear deformation theory
Symbolic computation

a b s t r a c t
Dynamic stiffness elements for plates are developed using rst order shear deformation theory to carry
out exact free vibration analysis of plate assemblies. The analysis has been facilitated by the application
of Hamiltonian mechanics and symbolic computation. The WittrickWilliams algorithm has been used as
the solution technique. Results have been extensively validated using published literature for both uniform and non-uniform plates. Some nite element results are also provided. The accuracy and computational efciency of the method are demonstrated. In the nal part of the investigation, signicant plate
parameters are varied and their subsequent effects on the free vibration characteristics are studied.
2010 Elsevier Ltd. All rights reserved.

1. Introduction
Aircraft structures are generally modelled as assemblies of
thin-walled structural elements. In particular, the top and bottom
skins, torsion box, ribs and webs of the wing are idealised as plates.
Thus the free vibration analysis of such structures plays an important role in aircraft design. The analysis facilitates aeroelastic and
response analyses. The purpose of this paper is to develop the
dynamic stiffness method for an accurate and efcient free vibration analysis of plates and plate assemblies.
The usually adopted nite element method [1] (FEM) is a universal tool in structural analysis which can handle complex structures. With the advent of high speed computing, the tendency to
use FEM has increased enormously and many commercially available pre and post processing programs have broadened its appeal,
making it simple and straightforward to use. FEM is an approximate method, but it generally converges to the exact solution with
increasing number of elements. However, the accuracy of results
cannot be always guaranteed. This is particularly true in dynamic
analysis at high frequencies when the FEM may become unreliable.
Thus, there is, and there will always be a need to use analytical
methods based on classical theories, wherever possible, to validate
the FEM, provide further insights and importantly, restore condence in design. One such method is that of the dynamic stiffness

Corresponding author.
E-mail address: marco.boscolo.1@city.ac.uk (M. Boscolo).
0045-7949/$ - see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruc.2010.11.005

method [28] (DSM) which gives exact results that are independent of the number of elements used in the analysis. For instance,
one single structural element can be used in the DSM to compute
any number of natural frequencies to any desired accuracy, which
of course, is impossible in the FEM. In DSM [28], once initial
assumptions about the displacement eld have been made, no
inaccuracy occurs in the analysis. However, for the fundamental
mode there is generally very little discrepancy in the frequencies
computed using FEM and DSM, but with increasing mode number,
signicant differences can arise in both response and stability
analyses.
The DSM at present has been developed mainly for onedimensional elements such as bars and beams [39]. This is
generally accomplished by using the exact closed form solution
of their governing differential equations of motion for harmonic
oscillation, and relating a state vector of loads to the corresponding
state vector of responses at the nodes. The relationship between
the two vectors establishes the frequency dependent dynamic
stiffness matrix of the element. There are well established
computer programs such as BUNVIS-RG [10] and PFVIBAT [11]
which demonstrate the accuracy and computational efciency of
the method. A strong point about DSM is that it has all the essential
features of FEM such as coordinate transformation, offset connections, assembly procedure, etc., and yet it retains the exactness of
results through the use of exact solution of the governing
differential equation. However, the solution techniques for FEM
and DSM are different. Unlike the conventional FEM which leads
to a linear eigenvalue problem, the DSM leads to a non-linear

396

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

eigenvalue problem which is generally solved by applying the

WittrickWilliams algorithm [12,13].
The development of a dynamic stiffness (DS) matrix for a plate
element presents considerable difculties. Wittrick and Williams
[1417] are probably the earliest investigators who developed
DSM for simply supported (SS) plates using classical plate theory
(CPT). Their theory was later implemented in a computer program
called VICONOPT [9,18] which is well-suited to investigate the free
vibration as well as buckling behaviour of aircraft wings idealised
as prismatic plate assemblies. An important feature of their research is that explicit expressions for the DS elements for SS anisotropic plates were presented. However, the authors did not include
the effects of shear deformation and rotatory inertia in their work,
which are important when analysing thick plates. The inadequacy
of CPT when investigating the free vibration characteristics of thick
plates is well known and any method based on CPT will no-doubt
incur errors in modal analysis, particularly at high frequencies.
Anderson and Kennedy [19,20] advanced VICONOPT by including
the effect of shear deformation in their DS development. The explicit terms of the DS matrix were not obtained by them and the
problem was solved numerically.
Many higher order shear deformation theories [21,22] have also
been developed for thick plates and composite laminates. Reddy
and Phans higher order plate theory [21] has been used to develop
the DS matrix of a plate by Leung and Zhou [23]. Also in this case
explicit terms of the DS matrix have not been computed.
Using the rst order shear deformation theory [24] (FSDT, generally known as Mindlin plate theory in the literature), this paper
advances the aforementioned works [9,1420,23] by developing
DS matrix of thick plates by including the important effects of
shear deformation and rotatory inertia. Despite the complexity of
the problem as a result of the inclusion of these effects, it has been
possible to generate explicit expressions for the DS elements for
the rst time by using symbolic computation (Mathematica [25]).
Explicit terms of the DS matrix are essential for developing a quick
and efcient computer program which can study plate assemblies
as well as for optimisation purposes. Explicit terms of DS matrix of
a plate based on the FSDT have never been published before.
Although the dynamic stiffness (DS) development of a Mindlin
plate has apparently not been fully investigated earlier, some related works using classical resolution of differential equations with
subsequent imposition of boundary conditions (BC) have been
published. Reddy [21,26,27] amongst others, analysed the free
vibration behaviour of thick plates with the effects of shear deformation and rotatory inertia. These publications are not focused on
frequency dependent DS development as in the present case, but
an individual plate on its own was studied. For an individual plate,
it is possible to determine natural frequencies and mode shapes by
applying boundary conditions and eliminating the constants from
the general solutions, without resorting to the development of
the dynamic stiffness matrix. This procedure is termed as classical

method (CM) in the subsequent text. Clearly such as a procedure

cannot be easily extended to deal with plate assemblies and somehow lacks generality. Nevertheless, the results obtained by using
the CM are useful comparators for validation purposes. The need
to apply DSM principally arises to study free vibration behaviour
of complex structures. There are two important advantages of
DSM out of many. The rst one is that two sides other than the
SS ones can be constrained without having the need to reevaluate
and eliminate the constants as would be required in the CM. More
importantly, the second advantage is that the DSM has the capability to assemble element stiffness matrices of complex structures
consisting of plate assemblies. For instance, plates with stringers
can be analysed and yet exact results can be obtained.
The current investigation is carried out in following steps. First,
the fundamental equations of the CPT and FSDT are briey summarised and some salient features are discussed (Section 2.1). Secondly, the dynamic stiffness matrix based on the CPT is
formulated with and without the effect of rotatory inertia (Sections
2.2.1 and 2.2.2) as a precursor to the development of more advanced FSDT DSM which is dealt with in Section 2.2.3 by using
symbolic computation [25]. Subsequent to this development, the
assembly procedure and imposition of boundary conditions by
suppressing appropriate degrees of freedom (penalty method) are
explained in Section 2.3.1. This is followed by Section 2.3.2 which
highlights the application of the WittrickWilliams (WW) algorithm for computation of natural frequencies of thick plates with
various boundary conditions. The mode nding technique using
the DSM is then reported in Section 2.3.3.
Once the DS matrices using CPT and FSDT have been derived,
the results computed from the eigensolution procedure are validated in detail for rectangular plates with two opposite sides simply supported and the others having any generic boundary
conditions (BC) which can be in any combination of clamped (C),
free (F), or simply supported (SS) (Section 3.1). Section 3.2 essentially gives results for a variable thickness plate for which some
comparative results are available in the literature. An insight into
the advantages and disadvantages using different methods is given
and the future potential of DSM resulting form the current research
is highlighted. Finally the paper closes with some concluding
remarks.

2. Theory
2.1. Some basic preliminaries
Fig. 1 shows the notation for displacements and forces for a
thick plate in a rectangular Cartesian coordinate system. The displacement eld is described for both CPT and FSDT and the corresponding equations of motions with their natural boundary
conditions are briey summarised below. For brevity, only

Fig. 1. Coordinate system and notations for displacements and forces for a plate.

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

expressions that are directly relevant to the development of DS

theory are presented here. An exhaustive treatment of plate theories is not needed here, but can be found elsewhere [2629].
2.1.1. Classical plate theory (CPT)
Referring to Fig. 1(a) the displacement components of the plate
in the usual notation can be written as:

ux; y; z u0 x; y  z

@w0
;
@x

v x; y; z v 0 x; y  z

@w0
;
@y

wx; y; z w0 x; y:
The main focus of this work is on the out of plane vibratory motion
of the plate so that the displacements in the membrane mode
u0(x, y) and v0(x, y) are excluded. Clearly the only unknown in the
above expressions is the vertical displacement w0(x, y).
When deriving the DS matrix, there are some advantages of
using Hamiltons principle as opposed to Newtons second law.
The former provides not only a systematic deduction of the differential equations, but also links natural boundary conditions to the
displacement variables which facilitates the derivation of the DS
matrix greatly. The latter on the other hand is a direct approach
to derive the differential equations, but does not provide any additional information on the correspondence between forces and displacements which is essentially needed in DS formulation.
Although these advantages are not so apparent when CPT is used,
but for FSDT formulation, they are very useful (Section 2.2.3).
Application of Hamiltons principle to derive the governing differential equation of a plate undergoing free natural vibration
lends itself to the following equation and natural boundary conditions (details are not shown for brevity):

@ 4 w0
@ 4 w0
@ 4 w0
2 2 2
4
@x
@x @y
@y4

!
qh

@ 2 w0
0;
@t2

the BC are :

!
@ 3 w0
@ 3 w0
V x : D
2  m
dw0 ;
@x3
@x@y2
!
@ 2 w0
@ 2 w0
d/y ;
m
Mxx : D
@x2
@y2

where D = E h3/{12(1  m2)} is the plate bending or exural stiffness,

E the Youngs modulus of the plate material, h the plate thickness, m
the Poisson ratio, and q is the density of plate material.
It should be noted that the signs for forces and moments in Eq.
(3) have been changed in order to comply with the sign convention
used in Fig. 1(a). This is because the convention used in Fig. 1(a)
uses the positive faces of the plate (i.e. the ones with the coordinate system axes pointing out of them), to dene the positive
forces and moments. On the other hand, Hamiltons principle gives
positive forces and moments on the negative faces (i.e. the faces
located near the origin of the axis system). All derivations in this
paper are based on sign conventions shown in Fig. 1(a). Furthermore it can be seen from Eq. (3) how the transverse force or edge
reaction Vx is connected to w0 and the bending moment Mxx to the
rotation /y = @w0/@x. This piece of information would not be
available if Newtons second law was used. Eqs. (2) and (3) are
essential requirements to derive the DS matrix of a strip of isotropic plate based on the CPT.
The derivation of the DS matrix is deferred to Section 2.2.1 following the corresponding above developments for FSDT.

397

2.1.2. First order shear deformation theory (FSDT)

The displacement eld of FSDT, assuming no motion in the
membrane mode (u0, v0 = 0) is given by Eq. (4) where the symbols
are dened in Fig. 1(a):

ux; y; z z/y x; y; v x; y; z z/x x; y;

wx; y; z w0 x; y:

Based on the above displacements, Hamiltons principle is invoked

and the following three coupled partial differential equations are
obtained:

2

8
2 0
@/y
@/x
@ w0
@ 2 w0
>
kGh
 qh @@tw2 0
2 @x  @y @y2
>
@x
>
>
>
2



>
2
>
> D @ /2y 1m @ /2y  m1 @ 2 /x  kGh @w0 /
>
y
>
2
2
@y@x
@x
@x
@y
>
<
2
h3 @ /y
 q12
2 0
>
@t
>


>
> @2 /x 1m @2 /x m1 @2 /y 
0
>
>
 /x
D @y2 2 @x2  2 @y@x kGh @w
>
@y
>
>
>
>
:
h3 @ 2 /x
 q12
0
@t 2

where G = E/{2(1 + m)} is the shear modulus or modulus of rigidity, k

is the shear correction factor (5/6 used by Reissner [30], p2/12 used
by Mindlin [24]).
The BCs resulting form Hamiltonian formulation with conformity of sign conventions shown in Fig. 1 are as follows:

 0

@w
Q x : kGh
/y dw0 ;
@x


@/y
@/
M xx : D
 m x d/y ;
@x
@y

3
Gh @/y @/x

d/x :
M xy : 
12 @y
@x

The expressions of forces and moments shown in Eq. (6) routinely

resulted from the Hamiltonian formulation. The same equations
could have been derived using Newtons second law, i.e. the
force-equilibrium method through the use of displacements, the
constitutive equations and geometric considerations. Hamiltons
principle however, provides additional information on the connectivity of the displacement variables to the corresponding forces
and moments, i.e. w0 to Qx, /y to Mxx, /x to Mxy (see Eq. (6)) whcih
is a great advantage.
Eqs. (5) and (6) are essential requirements to derive the DS matrix based on FSDT as were Eqs. (2) and (3) when using CPT.
2.2. Dynamic stiffness development
2.2.1. CPT
The rst step in developing the DS matrix is to solve the governing differential Eq. (2). The solution is sought in the traditional Levi
form [27]. A plate, simply supported (SS) on two opposite sides
(y = 0 and y = L) is studied. As a consequence a solution which satises the BCs a priori is sought in the following form:

w0 x; y; t

1
X

W m xeixt sinam y;

m1

where x is the unknown frequency and am mLp (m = 1, 2, . . . , 1).

By substituting Eq. (7) into Eq. (2), the following fourth order
ordinary differential equations is obtained:



4
2
d Wm
d Wm
qhx2
 2a2m
a4m 
W m 0 m 1; 2; . . . ; 1:
4
2
dx
dx
D
8

Only the boundaries at x = 0 and x = b (Fig. 1(a)) are considered because a Levy
type solution of the differential equation is sought where two sides (y = 0 and y = L)
are assumed to be SS.

Using standard procedures, the solution of the above equation is obtained by solving the auxiliary equation from a trial solution [31]

398

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

and determining the four roots. Depending on the nature of the

roots two solutions of the differential equation are possible:
q
 Case 1. a2m P x qDh ) all roots are real (r1m ; r1m ; r2m ; r2m )

r 1m

s
r
qh
a2m x
;
D

r 2m

s
r
qh
a2m  x
:
D

The solution is:

W m x Am coshr 1m x Bm sinhr 1m x
C m coshr 2m x Dm sinhr 2m x:

10

 Case 2. a2m < x qDh )

two real and two imaginary roots (r1m ; r1m ; ir 2m ; ir 2m )

r 1m

s
r
qh
;
a2m x
D

r 2m

s
r
qh
:
a2m x
D

11
Fig. 2. Boundary conditions for displacements and forces for a plate element.

The solution is:

W m x Am coshr 1m x Bm sinhr 1m x
C m cosr2m x Dm sinr2m x:

12

The procedure to obtain the DS matrix for the rst case is given
below. The DS matrix for the second case follows exactly the same
procedure but is not shown for brevity.
Now from the known displacement w0 (Eqs. (10) and (7)), the
rotation /y, the edge reactions or net shear force Vx, and bending
moment Mxx can be expressed in the following form using Eq. (3).

/ym x; y Uym x sinam y r 1m Am sinhr1m x

Bm r 1m coshr 1m x C m r 2m sinhr 2m x
Dm r2m coshr 2m x sinam y:

13

 

V xm x; y V xm x sinam y D Am r 31m  2  ma2m r 1m sinhr1m x


Bm r31m  2  ma2m r1m coshr 1m x


C m r 32m  2  ma2m r 2m sinhr2m x



14
Dm r32m  2  ma2m r2m coshr 2m x sinam y:
 

M xxm x; y Mxxm x sinam y D Am r21m  ma2m coshr1m x




Bm r21m  ma2m sinhr1m x C m r22m  ma2m coshr2m x



Dm r 22m  ma2m sinhr 2m x sinam y:
15
The boundary conditions for displacements (Fig. 2) are:

x 0;

W m W 1;

Uym Uy1 ;

x b;

Wm W2;

Uym Uy2 ;

16

i.e.

d AC;

19

where

Chi coshrim b;
Ci cosr im b;

S hi sinhr im b;
S i sinr im b

20

with i = 1, 2.
Subsequently, by applying the BCs for forces, i.e. substituting
Eq. (17) into Eqs. (14) and (15), the following matrix relationship
is obtained:

3 2
0
V1
6 M 7 6 L1
1
7 6
6
76
6
4 V 2 5 4 R1 S h1

L2

R1 Ch1

R2 S h2

3
Am
7
7
6
0
76 Bm 7
76
7
R2 Ch2 54 C m 5

L1 Ch1

L1 S h1

L2 Ch2

L2 S h2

M2

R1

R2

32

21

Dm

i.e.

F RC;

22

where

Ri Dr3im  a2 rim 2  m;

Li Dr2im  a2 m

23

with i = 1, 2.
Using Eqs. (19) and (22) the dynamic stiffness matrix K for the
plate element based on the CPT can be obtained by eliminating the
constant vector C to give:

F Kd;

24

where
similarly the BC for the forces (Fig. 2) are:

x 0;

V xm V 1 ;

x b;

V xm V 2 ;

K RA1 :

Mxxm M1 ;

17

Mxxm M2 :

By applying the BCs for displacements, i.e. substituting Eq. (16) into
Eqs. (10) and (13), the following matrix relationship is obtained:

3 2
W1
7 6
6
6 Uy1 7 6
7 6
6
6W 7 6
4 25 4
2

Uy2

32

r 1m

r2m

C h1

S h1

C h2

S h2

3
Am
76
7
76 Bm 7
76
7
76
7
54 C m 5

r 1m S h1

r 1m Ch1

r 2m S h2

r 2m Ch2

Dm

The 4  4 dynamic stiffness matrix of Eq. (25) is symmetric with six

independent terms svv, svm, smm, fvv, fvm, fmm which describe the effect on shear and moment due to unit displacements imposed on
the same (s) nodal line, and on the far (f) nodal line. Thus K
can be expressed as:

2
18

25

6
6
K 6
4

svv

sv m

fvv

smm

fv m

Sym

svv

fv m

fmm 7
7
7:
sv m 5
smm

26

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

Explicit expressions of the elements are derived by extensive algebraic manipulation using Mathematica [25]. These are given in the
Appendix for both sets of solutions (case 1 and case 2).
2.2.2. Effect of rotatory inertia
The effect of rotatory inertia can now be incorporated in the
analysis to derive the DS matrix for better accuracy. The authors
have managed to accomplish this without altering much of the
above expressions because neither the order, nor the nature of
the differential equation changes as a result of the rotatory inertia
effect. Thus it has been possible to simply redene the roots of the
auxiliary equation stemming from the governing differential equation. The effect of rotatory inertia essentially changes the coefcients of the differential operators of Eq. (8), but the rest of the
procedure in the derivation of DS matrix remains unchanged. The
rotatory inertia term associated with plate bending (exure) is given by Mindlin [24] and Reddy [24,26]:
4

Irot

 Case 1.

v
!
u
u qx2 h3 2 qx2 h
qx h
t
2
;
am 

P
D
24D
24D
2 3

28

all roots are real (r1m ; r 1m ; r2m ; r2m ):

r 2m

2
6
6
4

v
v
u
!2
u
u
3
2
u
qx h u
qx2 h3
qx2 h
t
2
t
;
am 

D
24D
24D
v
v
u
!
u
u
u qx2 h3 2 qx2 h
2 h3
u
qx
t
2
t
:
am 


D
24D
24D

hx2 q GhkD2  a2
GhkD

aGhk

GhkD
 2

3
2
2 m1
D D a 2  Ghk x 12qh
D 12 m D

29

1
X

W m xeixt sinam y;

m1
1
X

Uym xeixt sinam y;

32

m1

/x x; y; t

1
X

Uxm xeixt cosam y;

m1

where x is the unknown frequency, am mLp and m = 1, 2, . . . , 1.

Note that unlike the previous case using CPT there are now three
independent series instead of one.
Substituting Eq. (32) into Eq. (5) gives a set of three coupled ordinary differential equations as follows:

8
2
dU
Ghk d dxW2m hx2 q  a2 GhkW m aGhkUxm Ghk dxym 0;
>
>
>
>
>


< d2 U
3
2
dU
m
aD 12 m dxxm a2 D m1
 Ghk x 12qh Uym 0;
D dx2ym  Ghk dW
2
dx
>
>
>


>
>
: D 1m d2 Uxm aGhkW  aD 1m dUym x2 qh3  a2 D  Ghk U 0:
m
xm
2
2
dx
12
dx2
33

3 2
3 2 3
0
Wm
7
7 6 7
7 6
aD 12 m D

0 5;
U
4
4
5
y
m
5
1m 2
 x2 qh3
0
U
xm
D
D a
 Ghk

v
!2
u
qx2 h3 u
qx2 h3
qx2 h
2
;
am 

6t
D
24D
24D

aGhk

30

where D is the differential operator d/dx.

By expanding the 3  3 determinant of Eq. (34) the following
governing differential equation satised by all of three variables
is obtained.

D6 a1 D4 a2 D2 a3 DW 0;

v
v
u
!
u
u
u qx2 h3 2 qx2 h
2 h3
u
qx
;
ta2m 

t
D
24D
24D
v
1
v
u 0
!
u
u
3
3 2
u
2
2
2
u B
qx h t qx h
qx hC
2
u

A:
t@am  24D 
D
24D

34

12

two real and two imaginary roots (r 1m ; r 1m ; ir 2m ; ir 2m ):

r 2m

w0 x; y; t

The above equations can be expressed in the following matrix form:

 Case 2.

r 1m

2.2.3. FSDT
Having established the methodology for DS formulation using
CPT, attention is now conned to the DS development using FSDT
which is a signicant step forward. The difcult arises due to the
complexity in the order of magnitude of the problem, exacerbated
by the increase in the number of variables.
As before, the solution of Eq. (5) is sought in the Levi form, i.e.
the two opposite sides of the plate are SS at y = 0 and y = L. Solutions complying with these conditions can be written as:

27

In essence, Eq. (27) should be included in Eq. (2) to account for the
effect of rotatory inertia. This will redene the roots of the auxiliary
Eqs. (9) and (11) as follows:

r 1m

Using the roots given by Eqs. (29) and (31) the DS matrix can
now be computed using the expressions of Eq. (26).

/y x; y; t

1
@ w
@ w

qh3

:
12
@x2 @t2 @y2 @t 2

399

35

where:

W W m or Uym
31

or Uxm :

36

Substituting a trial solution ek in Eq. (35) yields the following auxiliary equation:

k6 a1 k4 a2 k2 a3 0;

37

400

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

where:
2

24G hk Gh kx qm  3  36a DGk  12Dx qm  1

;
12DGkm  1




2
3
a2 hx2 q 12Gk  h x2 q Gh k  6Dm  3

2
3
12a2 D 24G2 hk Gh kx2 qm  3
 


12Dx2 qm  1 216a4 D2 Gkm  1 = 72D2 Gkm  1 ;




3
3
a3  12a4 DGk  12Ghk a2 12D Gh k x2 q h x4 q2

 

3
12Ghk h x2 q 6a2 Dm  1 = 72D2 Gkm  1 : 38

a1

The 6th order polynomial of Eq. (37) can be reduced to a cubic by

substituting l = k2 so that:

l3 a1 l2 a2 l a3 0;

39

1
3

l2 

1
3

l3

1

3

W m x A1m coshr 1m x A2m sinhr1m x

A3m coshr 2m x A4m sinhr 2m x
A5m cosr3m x A6m sinr 3m x;

Uym x B1m coshr 1m x B2m sinhr 1m x

B3m coshr 2m x B4m sinhr2m x

44

B5m cosr 3m x B6m sinr 3m x;

Uxm x C 1m coshr1m x C 2m sinhr 1m x

C 3m coshr 2m x C 4m sinhr 2m x
C 5m cosr3m x C 6m sinr 3m x:
 Case 3. l1 > 0 and l2, l3 < 0.
The roots are rearranged so that the negative roots are l2 and
p
p
l3, thus r1, r1, ir2, ir2, ir3, ir3 where r1 l1 ; r2 l2
p
and r 3 l3 .
Thus the solutions can be written as:

for which the three roots (l1, l2, l3) are given by:

l1 

 Case 2: l1, l2 > 0 and l3 < 0.

The roots are rearranged so that the negative root is l3, thus r1,
p
p
p
r1, r2, r2, i r3, ir3 where r 1 l1 ; r2 l2 and r3 l3 .
Thus the solutions can be written as:

r
p r
p!
3 p
q 3 p q
a1

;
2
2
r
r
p
p!
3 p
q
3 p 
q
;
a1 b2
b1
2
2
r
r
p
p!
3 p
3 p 
q
q
;
a1 b1
b2
2
2

40

W m x A1m coshr 1m x A2m sinhr1m x

A3m cosr2m x A4m sinr 2m x
A5m cosr3m x A6m sinr 3m x;

Uym x B1m coshr 1m x B2m sinhr 1m x

where:

p
p
i 31
i 31
; b2 
b1
2
2
3
p 2a1  9a1 a2 27a3 ; l a21  3a2 ;

B3m cosr 2m x B4m sinr 2m x

41
3

q p2  4l

C 5m cosr3m x C 6m sinr 3m x:

D 18a1 a2 a3  4a31 a3 a21 a22  4a22  27a23



1
2
2
2

h x2 qh x2 q  12Gk2 576DG2 hk
4
6 4 4
746496D G k m  1

3
3
gh k  12D2 x2 q Gh kx2 q1 m
2 
2
6Dx2 qm2  1  24G2 hk
:
42
The discriminant is always positive, i.e. D > 0, in the present case
which means that the three roots are real, and consequently, there
are no more than four possible solutions, i.e. (i) all roots are positive,
(ii) one negative and two positive, (iii) two negative and one positive and (iv) all roots are negative. Descartess rule of signs [32]
was applied to predict the sign of the roots in advance, but no denitive conclusion on their signs could be drawn. Thus, it can be said
that there cannot be less than the four aforementioned cases. These
are as follows:
 Case 1: l1, l2, l3 > 0.
The six roots are real r1, r1, r2, r2, r3, r3 where
p
p
p
r1 l1 ; r2 l2 and r3 l3 .
Thus the solutions can be written as:

W m x A1m coshr 1m x A2m sinhr 1m x

A3m coshr2m x A4m sinhr 2m x
A5m coshr3m x A6m sinhr 3m x;

Uym x B1m coshr 1m x B2m sinhr 1m x

B5m coshr 3m x B6m sinhr 3m x

Uxm x C 1m coshr 1m x C 2m sinhr 1m x

C 3m coshr2m x C 4m sinhr 2m x
C 5m coshr3m x C 6m sinhr 3m x:

Uxm x C 1m coshr1m x C 2m sinhr 1m x

C 3m cosr2m x C 4m sinr 2m x

The discriminant of Eq. (39) can be written as

B3m coshr 2m x B4m sinhr 2m x

45

B5m cosr 3m x B6m sinr 3m x;

43

 Case 4. l1, l2, l3 < 0.

The six roots are imaginary ir1, ir1, ir2, ir2, ir3, ir3 where
p
p
p
r1 l1 ; r 2 l2 and r3 l3 .
Thus the solutions can be written as:

W m x A1m cosr 1m x A2m sinr 1m x

A3m cosr2m x A4m sinr 2m x
A5m cosr3m x A6m sinr 3m x;

Uym x B1m cosr 1m x B2m sinr 1m x

B3m cosr 2m x B4m sinr 2m x

46

B5m cosr 3m x B6m sinr 3m x;

Uxm x C 1m cosr 1m x C 2m sinr1m x

C 3m cosr2m x C 4m sinr 2m x
C 5m cosr3m x C 6m sinr 3m x:
For presentational purposes, the derivation of the DS matrix is
presented only for the rst case. The same procedure has been used
to obtain the DS matrices for the other cases.
The solutions in Eq. (43) have three sets of six constants, namely
A1m  A6m ; B1m  B6m and C 1m  C 6m which are not all independent. It
is interesting to note that if the variables Uxm and Uym are successively removed from Eq. (33) instead of using the determinantal
equation (given by Eq. (34)), the resulting governing differential
equation in Wm becomes 4th order instead of 6th, unlike the case
with the differential equations using the other two variables. This
was further observed when the terms A5m and A6m in Eq. (43) during algebraic manipulation and numerical simulation became zero.
This made it impossible to express B5m ; B6m and C 5m ; C 6m in terms of
A5m and A6m . Therefore care should be exercised when expressing
the two sets of six constants in terms of the third to avoid any

401

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

numerical overow. In this case the constants A1m  A6m and

C 1m  C 6m were related to B1m  B6m .
Substituting Eq. (43) into Eq. (33) gives the following
relationship:

A1m d1 B2m ;

C 1m c1 B2m ;

A2m d1 B1m ;

C 2m c1 B1m ;

A3m d2 B4m ;

C 3m c2 B4m ;

A4m d2 B3m ;

C 4m c2 B3m ;

A5m d3 B6m ;

C 5m c3 B6m ;

A6m d3 B5m ;

C 6m c3 B5m ;

B1m r 1m  ac1 m sinhr 1m x B4m r 2m  ac3 mcoshr2m x

B3m r 2m  ac2 m sinhr 2m x B5m r 3m  ac2 mcoshr3m x
B5m r3m  ac3 msinhr 3m xsinam y;

47

52

1
3
Gh B1m a c1 r 1m coshr 1m x
12
B2m a c1 r 1m sinhr 1m x B3m a c2 r 2m coshr2m x

Mxym x; y Mxym x cosam y 

B4m a c2 r 2m sinhr 2m x B5m a c3 r 3m coshr3m x

B6m a c3 r 3m sinhr 3m x cosam y:

where:

di

Mxxm x;y Mxxm xsinam y DB2m r 1m  ac1 m coshr 1m x

53

The BCs for the displacements and rotations (Fig. 3) are:




3
Gk 12Ghk  h qx2  6D1 m a2  r 2i


;
2
6r i 2G2 hk  DGka2 1 m D1 mqx2 DGk1 mr2i

x0:

W m W 1;

xb:

Wm W2;

Uym Uy1 ;
Uym Uy2 ;

48

 


3
2
12Dr 2i  12Ghk 6Da2 m  1 h qx2 Gk a2  r 2i  qx2  12G2 hk r2i


ci




2
6ari 2G2 hk D1 m qx2 Gk a2  r2i

49

with i = 1, 2, 3.
Thus Eq. (43) can be rewritten in terms of only one set of constants as:

W m x B2m d1 coshr 1m x B1m d1 sinhr 1m x B4m d2 coshr 2m x

B3m d2 sinhr 2m x B6m d3 coshr 3m x B5m d3 sinhr3m x;

Uym x B1m coshr1m x B2m sinhr1m x B3m coshr 2m x

B4m sinhr2m x B5m coshr 3m x B6m sinhr 3m x;

Uxm Ux1 ;
Uxm Ux2 :

54

Similarly, the BCs for the forces are:

x0:

Qxm Q1 ;

xb:

Qxm Q2 ;

Mxxm Mxx1 ;
Mxxm Mxx2 ;

Mxym Mxy1 ;

By applying BCs for displacements (substituting Eq. (54) into Eq.

(50)), the following matrix relation is obtained:
2W 3

2 0
1
7 6
6
6 Uy1 7 6 1
7 6
6
7 6
6
6 Ux1 7 6 0
76
6
7 6
6
6 W 2 7 6 d1 S h1
7 6
6
7 6
6
4 Uy2 5 4 Ch1

Ux2

Uxm x B2m c1 coshr 1m x B1m c1 sinhr 1m x B4m c2 coshr 2m x

c1 S h1

d1

d2

c1

c2

d1 Ch1

d2 S h2

d2 Ch2

d3 S h3

S h1

C h2

S h2

Ch3

c1 Ch1

c2 S h2

c2 Ch2

 c3 S h 3

d3 32 B1 3
76 7
0 7 6 B2 7
76 7
76 7
c3 76 B3 7
76 7;
76 7
d3 Ch3 76 B4 7
76 7
6 7
S h3 7
5 4 B5 5
c3 Ch3

By substituting Eq. (50) into Eq. (6), the expressions for forces and
moments can be obtained as:

Q xm x; y Qxm x sinam y GhkB1m 1  d1 r 1m coshr 1m x

B2m 1  d1 r1m sinhr 1m x B3m 1  d2 r 2m coshr 2m x
B4m 1  d2 r2m sinhr 2m x B5m 1  d3 r 3m coshr 3m x
B6m 1  d3 r3m sinhr 3m x sinam y;

51

B6

56

B3m c2 sinhr 2m x B6m c3 coshr 3m x B5m c3 sinhr 3m x:

50

55

Mxym Mxy2 :

i.e.

d AC;

57

where Chi and S hi have already been dened in Eq. (20).

By applying BCs for forces (substituting Eq. (55) into Eqs. (51)
(53)), the following relationship is obtained:
3 2
32 3
Q x1
L1
0
L2
0
L3
0
B1
7 6
6
76 7
6 Mxx 7 6
0
R
0
R
0
R
7 6 B2 7
1
2
3
1 7
6
76 7
7 6
6
76 7
6 Mxy 7 6
0
T2
0
T3
0
7 6 B3 7
6 T1
1 7
6
76 7;
6

7 6
6
6 7
6 Qx 7 6 L1 Ch L1 S h L2 Ch L2 S h L3 Ch L3 S h 7
B
2 7
1
1
2
2
3
3 76 4 7
6
76 7
6
7 6
6
7
7
6 Mxx2 7 4 R1 S h R1 Ch R2 S h R2 Ch R3 S h R3 Ch 56
B5 5
4
1
1
2
2
3
3
5
4
Mxy2
T 1 Ch1 T 1 S h1 T 2 Ch2 T 2 S h2 T 2 Ch2 T 2 S h2
B6
2

58
i.e.

F RC;

59

where

Li Ghkdi r i  1; Ri Daci m  ri ; T i

1
3
Gh a ci ri
12

60

with i = 1, 2, 3.
From Eqs. (57) and (59) the dynamic stiffness matrix K for an
isotropic plate element based on FSDT is obtained as:

F Kd;

61

where
Fig. 3. Edge conditions of the plate element and sign conventions.

K RA1 :

62

402

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

Thus

2
6
6
6
6
K 6
6
6
6
4

sqq

sqm

sqt

fqq

fqm

smm

smt
stt

fqm
fqt

fmm
fmt

Sym

sqq

sqm
smm

fqt

7
7
7
7
7;
sqt 7
7
7
smt 5
fmt
ftt

63

stt
where the 12 independent terms sqq, sqm, sqt, smm, smt, stt, fqq, fqm, fqt,
fmm, fmt, ftt describe the relationships of shear, bending moment, and
twisting moment due to unitary displacements on the same (s)
nodal line, and on the far (f) nodal line. These are given in explicit
algebraic form in Appendix.
2.3. Application of the DSM
2.3.1. Assembly procedure and application of boundary conditions
The stiffness matrices given by Eqs. (26) and (63) are essential
requirements to compute exact natural frequencies of plates which
are simply supported on at least two opposite sides. For individual
plates, it is possible to determine the natural frequencies and mode
shapes by applying the CM dened in the introduction but clearly
that procedure cannot be easily extended to deal with plate assemblies as for the DSM. The assembly procedure in DSM is similar to
the nite element method and is schematically shown in Fig. 4.
Note that each strip of plate is connected through nodal lines instead of single points. Clearly the overall master stiffness matrix
will be banded as in the case of FEM.
The application of boundary conditions in DSM is also similar to
that of FEM. The penalty method is generally used to suppress any
particular degree of freedom. The method entails adding a large
stiffness to the appropriate term on the leading diagonal of the dynamic stiffness matrix.
The procedure for applying the boundary conditions is summarised as follows:
 CPT.
Free (F): no penalty is applied.
Simply supported (SS): Wi is penalised.
Clamped (C): Wi and Uyi are penalised.
 FSDT
Free (F): no penalty is applied.
Simply supported (SS): Wi and Uxi are penalised.
Clamped (C): W i ; Uyi , and Uxi are penalised.
where i the node to be constrained.
Because of the similarities, DS elements can be implemented in
FEM codes.
2.3.2. WittrickWilliams algorithm
Once the global dynamic stiffness matrix of a structure is
formed, the zeros of the determinant can be sought to determine
the natural frequencies. This procedure can be hugely cumbersome
because of the transcendental nature of the dynamic stiffness

elements. Thus the plot of the frequency determinant can cause

enormous difculties. Such a plot can also miss coincident natural
frequencies. The problem can be avoided by using the well known
Wittrick and Williams algorithm [12] which guarantees that no
natural frequencies of the structure are missed. The procedure is
briey summarised as follows.
A trial frequency x* is chosen to compute the dynamic stiffness
matrix K* of the nal structure which is then reduced to its upper
M
triangular form by the usual Gauss elimination to obtain K  . The
M
number of negative terms on the leading diagonal of K  was dened by Wittrick and Williams [12] as the sign count s(K*) of the
matrix. At this point, the number (j) of natural frequencies (x)
which are lower than the trial frequency (x*) is given by:

j j0 sK  ;

64

where j0 is the number of natural frequencies of single strip elements clamped on their opposite sides which are lower than the
trial frequency.
Assuming j0 is known, a suitable procedure can be devised, for
example the bi-section method, to bracket any natural frequency
between an upper and lower bound to any desired accuracy. This
allows quick and precise computations of the natural frequencies
without numerically searching for the zeros of the determinant
which may cause numerical difculties due to the transcendental
nature of stiffness element.
A drawback of the algorithm lays in computing the j0 values.
In this paper, j0 computation was avoided by using a sufciently
ne mesh to ensure that j0 = 0 at the frequency range of interest.
This was achieved by computing the rst CC natural frequency
of the largest strip by splitting it into narrower strips, applying
the DSM and subsequently compute the natural frequencies of
the global structure that are below the CC frequency computed
earlier.
2.3.3. Mode shape computation
The mode shapes are computed by using Eq. (61) and setting
the force vector to zero and one of the nodal displacements to an
arbitrary value. Then, the rest of the nodal displacements are determined in terms of the chosen one. For a ne mesh, the nodes will
be so close to each other that an accurate plot of the mode shapes
is possible from the nodal values of the displacements without the
need to process the results any further. Otherwise, for a coarse
mesh, once the nodal line values of the displacements have been
computed, the constants of the analytical solution (Eq. (50)) can
be computed by solving the system of algebraic Eq. (56) simultaneously. In this way, the analytical expressions of the mode shapes
can be obtained for any complex geometry and it can be plotted to
any desired accuracy.
Although the numerical values of the natural frequencies are
not inuenced by the mesh size since each element is analytically
exact, sometimes it is advisable to use a ner mesh if the plots of
the mode shapes are required. The reason for a ner mesh is essentially twofold. First, the nodal values can be used to plot the modes
and the calculation of the constants can be avoided since it

Fig. 4. Assembly of dynamic stiffness matrices.

403

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

increases the computational time. Secondly and more importantly,

higher natural frequencies of the global structure can be computed
before encountering the CC frequencies of an individual strip
element and thus, without having the need to compute j0 or
re-meshing for higher modes.
3. Results
3.1. Analysis of individual plates
The procedure described above has been implemented in a
MATLAB program to allow computation of natural frequencies of
plates and plate assemblies. The rst step taken was to validate
the natural frequencies obtained by the DSM against the results
available in the literature obtained by using classical methods
(CM) based on Navier or Levi solution.
The rst 13 natural frequencies of a simply supported (SS)
square plate (b/L = 1, b/h = 10) have been computed by the DSM
theory of this paper based on the classical plate theory (CPT), with
and without rotatory inertia (Irot) effect, and rst order shear deformation theory (FSDT). These are shown in Table 1 along side the
results by Reddy and Phan [21]. It can be seen that the results
are in exact agreement with the classical method as expected.
The error incurred by the CPT, and CPT with rotatory inertia when
compared with the FSDT is shown in brackets. It can be observed
that the error is larger for higher natural frequencies. Furthermore,
the effect of rotatory inertia is more pronounced in higher natural
frequencies, as expected. The error from the CPT on the 5th natural

frequency is more than 20% when compared to FSDT. If only the

rotatory inertia without the effect of shear deformation is considered, the error goes down to 14% which is still very considerable.
It is thus clear that if higher frequencies are sought for plates with
b/h = 10, the FSDT should be used.
Next,
p the natural frequencies of a rectangular plate
b=L 2; b=h 10 have been computed and compared against
classical method based results reported by Reddy and Phan [21]
(see Table 2). Also in this case the agreement is total. The errors
using CPT are shown in brackets and similar observations are made
as those in Table 1.
Table 3 shows the rst 12 natural frequencies of the aforementioned square plate with different boundary conditions, namely,
simply supported (SS), free (F), and clamped (C). The natural frequencies have been computed by using CPT and FSDT and the relative errors are shown in brackets. Representative mode shapes
corresponding to m=2 and n = 2 are plotted in Fig. 5. The error
using CPT is minimum when the plate is free on two sides (SS
FSSF), whereas it is maximum when the two sides are clamped
(SSCSSC). For all other cases the error is somewhere in between. By analysing the errors shown in Table 3, some engineering
judgment can be made on the suitability of using FSDT or CPT
depending on the boundary conditions and the number of natural
frequencies to be sought. For example, if a plate is clamped on two
sides (SSCSSC), it is necessary to use the FSDT even for computing the fundamental natural frequency to avoid a substantial error
of almost 10%. By contrast, if the plate is free on two sides (SSF
SSF), the CPT gives the rst two natural frequencies within

Table 1
p
^ xh q=G) for an isotropic SSSSSSSS square plate: b/L = 1, b/h = 10, m = 0.3, k = 5/6. The errors incurred by the CPT and CPT with rotatory
Validation of natural frequencies (x
inertia when compared with the FSDT are reported in brackets. Classical method results (CM) by Reddy and Phan [21].
mn

FSDT
CM

FSDT
DSM

CPT + Irot
CM

CPT + Irot (error %)

DSM

CPT
CM

CPT (error %)
DSM

1
1
2
1
2
1
3
2
3
1
2
4
3

0.0930
0.2219
0.3406
0.4149
0.5206
0.6520
0.6834
0.7446
0.8896
0.9174
0.9984
1.0764
1.1268

0.09303
0.22193
0.34056
0.41494
0.52056
0.65201
0.68341
0.74465
0.88965
0.91740
0.99844
1.07642
1.12685

0.0955
0.2360
0.3732
0.4629
0.5951
0.7668
0.8090
0.8926
1.0965
1.1365
1.2549
1.3716
1.4457

0.09553
0.23599
0.37319
0.46293
0.59506
0.76684
0.80904
0.89256
1.09650
1.13649
1.25494
1.37117
1.44747

0.0963
0.2408
0.3853
0.4816
0.6261
0.8187
0.8669
0.9632
1.2040
1.2521
1.3966
1.5411
1.6374

0.09632
0.24079
0.38527
0.48159
0.62606
0.81870
0.86686
0.96318
1.20397
1.25213
1.39660
1.54108
1.63740

1
2
2
3
3
4
3
4
4
5
5
4
5

(2.69)
(6.34)
(9.58)
(11.57)
(14.31)
(17.61)
(18.38)
(19.86)
(23.25)
(23.88)
(25.69)
(27.38)
(28.45)

(3.54)
(8.50)
(13.13)
(16.06)
(20.27)
(25.57)
(26.84)
(29.35)
(35.33)
(36.49)
(39.88)
(43.17)
(45.31)

Table 2
p
p
^ xh q=G) for an isotropic SSSSSSSS rectangular plate: b=L 2; b=h 10; m 0:3; k 5=6. The errors incurred by the CPT and CPT with
Validation of natural frequencies (x
rotatory inertia when compared with the FSDT are reported in brackets. Classical method results (CM) by Reddy and Phan [21].
mn

FSDT
CM

FSDT
DSM

CPT + Irot
CM

CPT + Irot (error %)

DSM

CPT
CM

CPT (error %)
DSM

1
2
1
3
2
3
4
1
2
4
3
5
5

0.07036
0.13729
0.20123
0.24235
0.26250
0.35948
0.37818
0.39666
0.45089
0.48608
0.53754
0.53754
0.63609

0.07036
0.13729
0.20123
0.24236
0.26251
0.35948
0.37818
0.39667
0.45090
0.48609
0.53754
0.53754
0.63610

0.07180
0.14273
0.21281
0.25908
0.28207
0.39575
0.41822
0.44062
0.50729
0.55133
0.61680
0.61680
0.74563

0.07180
0.14273
0.21281
0.25908
0.28208
0.39575
0.41823
0.44062
0.50730
0.55134
0.61680
0.61680
0.74563

0.07224
0.14448
0.21671
0.26487
0.28895
0.40935
0.43343
0.45751
0.52974
0.57790
0.65014
0.65014
0.79462

0.07224
0.14448
0.21671
0.26487
0.28895
0.40935
0.43343
0.45751
0.52975
0.57791
0.65014
0.65014
0.79462

1
1
2
1
2
2
1
3
3
2
3
1
2

(2.04)
(3.96)
(5.76)
(6.90)
(7.46)
(10.09)
(10.59)
(11.08)
(12.51)
(13.42)
(14.74)
(14.74)
(17.22)

(2.67)
(5.23)
(7.70)
(9.29)
(10.07)
(13.87)
(14.61)
(15.34)
(17.49)
(18.89)
(20.95)
(20.95)
(24.92)

404

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

Table 3
p
^ xL2 qh=D) for isotropic square plates with various boundary conditions: b/L = 1, b/h = 10, m = 0.3, k = 5/6. The errors incurred by the CPT and CPT with
Natural frequencies (x
rotatory inertia when compared with the FSDT are reported in brackets. CPT frequencies are validated against results published by Leissa [33].
SSSSSSSS

SSFSSSS

SSCSSSS

mn

FSDT

CPT (error %)

mn

FSDT

CPT (error %)

mn

FSDT

CPT (error %)

1
1
2
2
1
3
2
3
4
1
3
4

19.0650
45.4827
45.4827
69.7944
85.0380
85.0380
106.683
106.683
133.621
133.621
140.057
152.608

19.7392 (3.54)
49.3480 (8.50)
49.3480 (8.50)
78.9568 (13.13)
98.6960 (16.06)
98.6960 (16.06)
128.304 (20.27)
128.304 (20.27)
167.783 (25.57)
167.783 (25.57)
177.652 (26.84)
197.392 (29.35)

1
1
2
2
1
3
2
3
1
3
2
4

11.3731
26.1545
38.2861
53.2462
55.6234
78.3664
81.0756
92.2072
97.3551
117.908
119.844
127.461

11.6845 (2.74)
27.7563 (6.12)
41.1967 (7.60)
59.0655 (10.93)
61.8606 (11.21)
90.2941 (15.22)
94.4837 (16.54)
108.918 (18.12)
115.685 (18.83)
145.638 (23.52)
148.509 (23.92)
159.080 (24.81)

1
2
1
2
3
1
3
2
4
1
3
4

22.3886
47.1039
52.1496
74.1051
85.8759
93.2268
109.259
112.741
134.084
141.936
144.117
154.145

23.6463 (5.62)
51.6743 (9.70)
58.6464 (12.46)
86.1345 (16.23)
100.269 (16.76)
113.228 (21.45)
133.791 (22.45)
140.845 (24.93)
168.958 (26.01)
187.436 (32.06)
188.113 (30.53)
201.724 (30.87)

26.6683
49.1129
59.2102
78.8130
86.8440
101.371
112.058
118.922
134.595
148.316
149.990
155.788

28.9508 (8.56)
54.7431 (11.46)
69.3270 (17.09)
94.5853 (20.01)
102.216 (17.70)
129.095 (27.35)
140.204 (25.12)
154.775 (30.15)
170.346 (26.56)
199.810 (34.72)
206.697 (37.81)
208.391 (33.77)

1
2
1
2
3
1
3
2
1
4
3
2

SSFSSF
1
1
1
2
2
2
1
3
3
2
3
1

1
2
3
1
2
3
4
1
2
4
3
5

1
2
1
2
3
1
3
2
4
3
4
1

SSFSSC
9.4406
15.3893
33.8599
36.3570
42.7927
62.1467
66.1965
76.6330
82.4088
92.6128
101.029
110.024

9.6314
16.1348
36.7256
38.9450
46.7381
70.7401
75.2834
87.9867
96.0405
111.0254
122.0400
133.7046

(2.02)
(4.84)
(8.46)
(7.12)
(9.22)
(13.83)
(13.73)
(14.82)
(16.54)
(19.88)
(20.80)
(21.52)

1
1
2
2
1
3
2
3
1
3
2
4

1
2
1
2
3
1
3
2
4
3
4
1

1
1
2
2
1
3
2
3
1
4
3
2

SSCSSC
12.2492
30.4086
38.6343
55.8017
62.7260
78.5265
85.8935
93.6825
105.613
120.987
126.029
127.540

12.6874 (3.58)
33.0651 (8.74)
41.7019 (7.94)
63.0148 (12.93)
72.3976 (15.42)
90.6114 (15.39)
103.161 (20.10)
111.896 (19.44)
131.428 (24.44)
152.773 (26.27)
159.302 (26.40)
162.371 (27.31)

1
2
1
2
3
1
3
2
4
3
1
4

1
1
2
2
1
3
2
3
1
3
4
2

Fig. 5. Mode shapes (mode m = 2, n = 2) for square plates with different boundary conditions.

reasonable accuracy (error is around 5%.). These assertions are

valid for a plate with a thickness ratio of b/h = 10 and for smaller
ratios, the errors will be signicantly higher.
The percentage error incurred by the CPT frequencies when
compared to the FSDT for different plate thickness ratios is plotted
in Fig. 6 for different modes and BCs. Fig. 6(a) shows the convergence of the rst natural frequency. It can be seen that the BCs
inuence the convergence of the natural frequency with the thickness ratio signicantly. The clamped BC shows the slowest convergence rate. It can be seen that for a thickness ratio of 5 the CPT
causes substantial error for all boundary conditions. When the
thickness ratio is equal to 20 there is hardly any difference between CPT and FSDT results. Fig. 6(b)(d) show the corresponding
results for the 2nd, 7th, and 11th natural frequencies, respectively.
Understandably, the higher the natural frequency, the slower the
convergence. For thickness ratio in excess of 50 there is practically
no difference between CPT and FSDT results for any BC.

For completeness, some additional results have been obtained

using the FE method. The rst 10 natural frequencies of the aforementioned simply supported (SS) square plate (b/L = 1, b/h = 10)
computed using NASTRAN, CQUAD4, PSHELL elements have been
compared with the exact frequencies obtained by DSM based on
both CPT and FSDT. The results are reported in Table 4. For the
FE models, two different mesh sizes were used, namely a coarse
mesh (10  10) and a ner mesh (100  100). It can be observed
that the CPT based nite element result converges to the exact
solution whereas when FSDT elements are used, the FE solution
converges to a higher value than the exact one obtained by the
DS method which was validated against the results of Reddy and
Phan [21]. These errors are not caused by different assumptions
on the FSDT implemented in NASTRAN but, more likely, they are
due to numerical factors. One of these factors could be shear locking. It is also important to note that the accuracy of FE decreases for
higher modes, as expected.

405

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

SS-F-SS-F
SS-SS-SS-SS
SS-C-SS-C

140
120

120

100

100
Error, %

Error, %

SS-F-SS-F
SS-SS-SS-SS
SS-C-SS-C

140

80
60

80
60

40

40

20

20

0
10

100

10

b/h

(a) 1st Mode

(b) 2nd Mode

SS-F-SS-F
SS-SS-SS-SS
SS-C-SS-C

140
120

SS-F-SS-F
SS-SS-SS-SS
SS-C-SS-C

140
120
100
Error, %

100
Error, %

100
b/h

80
60

80
60

40

40

20

20

10

100

10

b/h

100
b/h

(c) 7 th Mode

(d) 11th Mode

Fig. 6. Convergence of CPT natural frequencies towards FSDT against thickness ratio for different boundary conditions.

Table 4
p
^ xL2 qh=D) for an isotropic SSSSSSSS square plate: b/L = 1, b/h = 10, m = 0.3, k = 5/6. The exact value obtained by the DS elements is compared to the
Natural frequencies (x
values obtained by FEM with different mesh sizes.
CPT

FSDT

FEM
10  10

100  100

DSM

FEM

11

10  10

DSM
100  100

11

^
x

Error %

^
x

Error %

^
x

^
x

Error %

^
x

Error %

^
x

0.09511
0.23676
0.23676
0.37027
0.47503
0.47503
0.59761
0.59761
0.80420
0.81289

1.3
1.7
1.7
3.9
1.4
1.4
4.5
4.5
1.8
0.7

0.09630
0.24074
0.24074
0.38505
0.48146
0.48146
0.62556
0.62556
0.81848
0.81848

0.0
0.0
0.0
0.1
0.0
0.0
0.1
0.1
0.0
0.0

0.09632
0.24079
0.24079
0.38527
0.48159
0.48159
0.62606
0.62606
0.81870
0.81870

0.09237
0.22020
0.22020
0.32916
0.41248
0.41248
0.49571
0.49571
0.62596
0.64391

0.7
0.8
0.8
3.3
0.6
0.6
4.8
4.8
4.0
1.2

0.09370
0.22537
0.22537
0.34781
0.42520
0.42520
0.53512
0.53512
0.67281
0.67281

0.7
1.5
1.5
2.1
2.5
2.5
2.8
2.8
3.2
3.2

0.09303
0.22193
0.22193
0.34056
0.41494
0.41494
0.52056
0.52056
0.65201
0.65201

3.2. Analysis of stepped plates

Fig. 7. Geometry of stiffened plate studied by Chopra [34].

One of the major advantages of the DSM is that the DS matrices

can be assembled in order to analyse complex structures modelled
as assemblies of plates. Classical methods can be applied to complex structures with great difculties and have very limited applications. A non-uniform plate shown in Fig. 7 was studied by
Chopra [34] using the classical method.
Apparently there was an inadvertent error in Chopras work
[34] which considered the same exural stiffness for all parts of
the plate even though it had different thicknesses. This was later
corrected by Guo et al. [35]. Despite the error, Chopra [34] showed
a clear methodology to solve the problem by obtaining the values

406

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

Table 5
Natural frequencies (x) for isotropic stepped plates (SSSSSSSS) with various
stringer thicknesses ratios. E = 71,200 MPa, m = 0.3,q = 2700 kg/m3, b/L = 1, b1 = 0.3b,
b2 = 0.1b, b3 = 0.2b, b/t1 = 1000. CM: classical method by Chopra [34] and corrected by
Guo et al. [35], DFSM: dynamic nite strip method [35], DSM (CPT): dynamic stiffness
method with classical plate theory developed in this paper.
t2/t1 = 1.2

t2/t1 = 1.5

mn

CM

DFSM

DSM (CPT)

mn

CM

DFSM

DSM (CPT)

1
1
2
2
1
3
2
3
1
4
3
2

31.74
79.30
80.60
127.3
156.4
162.5
207.1
207.8
/
/
/
/

31.74
79.29
80.60
127.3
156.3
162.4
207.0
207.7
/
/
/
/

31.738
79.298
80.606
127.309
156.362
162.490
207.137
207.833
273.604
277.029
288.246
319.387

11
12
21
22
13
31
23
32
14
41
33
24

33.17
81.69
87.64
134.2
161.5
179.0
221.1
222.0
/
/
/
/

33.17
81.69
87.67
134.2
161.5
179.0
221.0
221.8
/
/
/
/

33.171
81.695
87.678
134.230
161.495
179.096
221.104
221.959
284.582
305.75
309.201
331.024

1
2
1
2
3
1
3
2
4
1
3
4

of the constants of integration by imposing the BC between the

parts of the plate with different thicknesses and the boundaries.
Guo et al. [35] made an improvement on nite strip method [36]

in this respect by using dynamic shape functions for the strip

and validated their results against those of Chopra [34]. The natural frequencies obtained by Guo et al. [35] were still approximate
but more accurate than the ones obtained by the nite element
method and importantly, more accurate than the traditional nite
strip method. The advantage of the dynamic nite strip method
(DFSM) developed by Guo et al. [35] is that any natural frequencies
can be obtained by assembling the strip elements, but the method
is neither as accurate nor efcient as DSM. As a matter of fact, the
DSM developed in this paper has all the advantages of the nite
strip method and additionally gives exact natural frequencies and
mode shapes of a structure without making any approximations
en route.
The natural frequencies computed by the DSM in this paper are
compared with the ones obtained by Chopra [34] and Guo et al.
[35]. Table 5 shows the comparison. The dimensions of the panel
used are: b/L = 1, b1 = 0.3b, b2 = 0.1b, b3 = 0.2b, b/t1 = 1000 with
L = 1 m and the plate material is taken to be aluminium with
E = 71200 MPa, m = 0.3, q = 2700 kg/m3. The natural frequencies
have not been non-dimensionalised in order to make a direct comparison with the results given by Guo et al. [35]. It can be seen that
the DSM is in complete agreement with the classical method (CM).
Any new geometry (which can be modelled by plates) can be

Table 6
Natural frequencies (x) for an isotropic thick stepped plate computed by DSM based on different plate theories: CPT, CPT with rotatory inertia, and FSDT. E = 71200 MPa, m = 0.3,
q = 2700 kg/m3, b/L = 1, b1 = 0.3b, b2 = 0.1b, b3 = 0.2b, b/t1 = 10, t2/t1 = 1.5, k = 5/6.
SSSSSSSS
CPT

SSCSSC
CPT + Irot

FSDT

CPT

CPT + Irot

FSDT

mn

mn

mn

mn

mn

mn

1
1
2
2
1
3
2
3
1
4
3
2
4
3
1
4

3317.10
8169.43
8767.77
13422.98
16149.55
17909.59
22110.42
22195.91
28458.24
30574.96
30920.08
33102.40
34340.33
41239.72
42361.48
42513.24

1
1
2
2
1
3
3
2
1
4
3
2
4
3
1
4

3283.27
7995.98
8545.86
12937.92
15919.35
17040.96
20909.12
21449.62
26007.54
28202.95
29279.49
29846.95
31447.46
36342.73
38910.39
39056.25

1
1
2
2
1
3
2
3
1
4
3
2
4
3
4
1

3154.61
7339.71
7854.54
11433.49
13481.87
14698.52
17404.82
17630.01
21566.83
22796.97
22976.23
24377.75
25168.74
28997.32
29710.50
29778.97

1
2
1
2
3
1
3
2
4
3
1
4
2
4
3
5

4715.32
9661.77
11366.75
15895.44
18604.82
20908.31
24267.81
26520.18
31254.01
34988.73
35222.01
36343.84
39574.41
46150.98
47114.01
47342.38

1
2
1
2
3
1
3
2
4
1
4
3
2
3
5
4

4663.71
9405.74
11033.20
15196.79
17674.04
20750.18
22677.83
25931.32
28770.11
32352.55
33018.56
33403.57
35825.95
41637.54
42015.25
42722.08

1
2
1
2
3
1
3
2
4
1
3
4
2
3
4
5

4247.36
8395.51
9416.91
12775.79
15004.08
15939.68
18475.41
19298.84
22995.49
24131.01
24323.97
25733.40
26500.53
30559.19
30622.63
31740.18

1
2
1
2
3
1
3
2
4
1
3
4
2
4
5
3

1
2
1
2
3
1
2
3
4
1
3
4
2
4
5
3

1
2
1
2
3
1
3
2
4
1
3
4
2
4
3
5

1
1
2
2
1
3
2
3
1
3
4
2
4
3
4
1

1
1
2
2
1
3
2
3
1
4
2
3
4
4
1
3

Fig. 8. Natural modes for the stepped plates. Modes computed by DSM and FSDT.

1
1
2
2
1
3
2
3
1
4
3
2
4
4
3
1

407

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

Table 7
Natural frequencies (x) for an isotropic thick stepped plates computed by FEM (NASTRAN) compared with the exact values computed by the DSM. E = 71200 MPa, m = 0.3,
q = 2700 kg/m3, b/L = 1, b1 = 0.3b, b2 = 0.1b, b3 = 0.2b, t2/t1 = 1.5.
CPT, b/t1 = 1000, (SSSSSSSS)

FSDT, b/t1 = 10, k = 5/6, (SSCSSC)

FEM
10  10

100  100

DSM

FEM

51

10  10

DSM
100  100

51

^
x

Error %

^
x

Error %

^
x

^
x

Error %

^
x

Error %

^
x

32.80
80.35
86.56
129.54
157.40
177.53
210.09
213.26
285.33
287.65
304.52
324.37
332.01

1.12
1.65
1.27
3.49
2.54
0.87
4.98
3.92
0.26
5.92
1.51
2.01
3.32

33.17
81.67
87.66
134.15
161.42
179.06
220.89
221.79
284.52
305.68
308.75
330.74
343.09

0.02
0.03
0.02
0.06
0.04
0.02
0.10
0.08
0.02
0.02
0.15
0.09
0.09

33.17
81.69
87.68
134.23
161.50
179.10
221.10
221.96
284.58
305.75
309.20
331.02
343.40

4217.26
8348.32
9358.28
12385.63
14973.60
15936.29
17666.76
18483.53
22292.85
22700.80
23536.93
24231.91
24819.87

0.71
0.56
0.62
3.05
0.20
0.02
4.38
4.22
3.06
5.93
3.24
5.83
6.34

4287.44
8561.59
9572.40
13084.94
15445.34
16636.62
19053.29
20115.94
23801.80
24765.00
25318.64
26644.37
27265.66

0.94
1.98
1.65
2.42
2.94
4.37
3.13
4.23
3.51
2.63
4.09
3.54
2.89

4247.36
8395.51
9416.91
12775.79
15004.08
15939.68
18475.41
19298.84
22995.49
24131.01
24323.97
25733.40
26500.53

investigated to determine any required number of natural frequencies and mode shapes to any desired accuracy.
The panel studied by Chopra had a thickness ratio of b/t1 = 1000
and, as a consequence, there is no reason to use FSDT as opposed to
CPT. The same panel with a thickness ratio b/t1 = 10 and t2/t1 = 1.5
has been studied with two different sets of boundary conditions
SSSSSSSS and SSCSSC, respectively. The rst 16 natural frequencies are shown in Table 6. It can be seen again that the effect
of shear deformation is not negligible for this thickness ratio and
the presence of the two intermediate thick strips shown in Fig. 7
has an even more pronounced effect. In fact the error on the fundamental natural frequency for the simply supported panel due to
using CPT is around 5% while it is only 3% for a constant thickness
plate (Table 3). Clearly, the error on the rst frequency worsens at
11% for the clamped BCs. Two mode shapes (m = 2, n = 1 and
m = 4, n = 1) for the two boundary conditions are plotted in Fig. 8
for demonstration purposes.
Also for this case, results obtained by using the FEM (NASTRAN)
have been compared with the exact results computed by using the
DSM. The results are summarised in Table 7.
A coarse mesh (10  10) and a ner mesh (100  100) were
used as before. CPT based nite element results converge to the exact solution. On the other hand, when FSDT elements are used, the
FE solution converges to a higher value than the exact one obtained
by the DSM. Accuracy of the FEM decreases for higher natural frequencies, as in the previous case. Dynamic stiffness elements, not
only provide the exact value for any number of natural frequencies
but also they offer signicant savings in computer time.

4. Concluding remarks
An exact dynamic stiffness method for plate elements with two
opposite sides simply supported has been developed using classical plate theory with and without the inclusion of rotatory inertia
effect. The theory has then been signicantly enhanced by applying
the rst order shear deformation theory. Explicit expressions for
the terms of the dynamic stiffness matrix for all cases have been
derived by extensive use of symbolic computation. The dynamic
stiffness elements are sufciently general to be assembled to
investigate the free vibration behaviour of complex structures following a procedure similar to that used in the nite element method. Once the dynamic stiffness matrix for the overall structure is
formulated the eigenvalue problem is solved by using a modied
version of the Wittrick and Williams algorithm.

The complete procedure starting from the development of the

dynamic stiffness matrix and nishing with the calculation of natural frequencies and mode shapes has been implemented in a computer program using MATLAB. This enables computation of any
number of exact natural frequencies and mode shapes of plates
or plates assemblies to any desired accuracy.
Numerical results for a wide range of problems have been computed. These include prismatic plates with two opposite sides simply supported and the other two having any combination of
boundary condition, such as simple support, clamped (built-in)
support, or free edge. A stiffened plate with different boundary
conditions has also been analysed. The computed natural frequencies have been rigorously validated against available literature.
Exactness of the theory has been veried by published results, that
are restricted to single plates only, and complete agreement was
found.
Using the developed dynamic stiffness elements, a detailed
parametric study was undertaken, by varying signicant plate
parameters such as width to thickness ratio and boundary conditions. The predictable effects of shear deformation and rotatory
inertia to reduce the natural frequencies have been discovered.
With increasing values of width to thickness ratio, the classical
plate theory converges to rst order shear deformation theory, as
expected. Plates with a width to thickness ratio in excess of 50
can be probably investigated by using classical plate theory with
sufcient accuracy, whereas when the ratio is less than 15, it is
advisable to use the rst order shear deformation theory when
establishing even the fundamental natural frequency. Quite obviously, the higher naturally frequencies will be considerably less
accurate when the thickness ratio is small. For ratios within the
range 15  50, the boundary conditions of the plate and the order
of the mode play important roles in deciding which theory to use.
Of course, each case has to be treated individually by making some
engineering judgement on the suitability of the theory to use. As
expected, the effects of shear deformation and rotatory inertia
are more pronounced when the edges of the plate are clamped
whereas for free edge conditions the effects are less severe. The results for stiffened plates have shown that the effects of shear deformation and rotatory inertia are more signicant than those found
for uniform plates.
To put the analysis in a broader context, some comparative results using the nite element method have also been reported. Use
of FEM plate elements based on rst order shear deformation theory shows that natural frequencies convergence with increasing
mesh size to higher values than the exact ones obtained by DSM.

408

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

The investigation paves the way for further research to include

in-plane deformations and use of anisotropic materials, thus
broadens the scope for analysing complex structures such as composite wing boxes.
Acknowledgements

(56) and (58)). Numerical inversion of matrix A should be

avoided because it may cause ill-conditioning. After simplications the following explicit terms are obtained to generated the
dynamic stiffness matrix without causing any numerical error.

sqq

The authors thank the EPSRC (grant ref: EP/F03606X/1) which

made this work possible.
Appendix A

Ghk 
 Ch3 c1 d3 r2 c2 d2 r 3 c3 d2  2r3 c2 d3
D
d1 r 1 c1 r 2 c2 c3 d2 c2 r 1 c1 r3 c3 d3
 


 
S h1 S h2 d3 r 2 c21 r 3 c2 c3 d2  r 3 c21 c22 d3


 
d1 r2 c1 r 1 c2 c3 d2 r1 c22 r 3 c1 c3 d3
Ch2 d1 r 1 c1 r 2 c2 c3 d2 c2 r 1 c1 r 3 c3 d3

Explicit expressions of the elements of the dynamic stiffness

matrix follow:

c1 d2 2r 2 c3 d2  r2 c2 r 3 c3 d3
  
 

S h1 S h3 d1  r 2 c1 c2 r 1 c23 d2 c2 r3 c1 r 1 c3 d3
 


 
d2 r 2 c21 c23 d2  r3 c21 r 2 c2 c3 d3

Ch1 2r1 c2 c3 d21  r 1 c1 c3 d1 d2  r 2 c2 c3 d1 d2  r 1 c1 c2 d1 d3

CLT
 Case 1



svv r 2m R1  r 1m R2 r1m Ch2 S h1  r 2m Ch1 S h2 =D;
65
 



2
2
sv m r1m R2 Ch1  Ch1 Ch2  S h1  R1 S h1 S h2 ;
66
 




=D;
r2m  R2 S h1 S h2 R1 Ch1  Ch2 Ch2 S 2h2


smm L2  L1 r 2m Ch2 S h1  r 1m Ch1 S h2 =D;


fvv r 2m R1  r 1m R2 r 2m S h2  r 1m S h1 =D;


fv m Ch2  Ch1 r 2m R1  r 1m R2 =D;


fmm L1  L2 r 2m S h1  r 1m S h2 =D;

67
68
69
70
71

where r im ; Chi ; S hi ; Li ; Ri (with i = 1, 2) are dened in Eqs. (9), (20),

and (23) respectively and D:

D r21m S h1 S h2 r 22m S h1 S h2


2
r1m r 2m Ch1  Ch2  S 2h1  S 2h2 :

 r 3 c2 c3 d1 d3 r 2 c1 c2 d2 d3 r3 c1 c3 d2 d3
 

S h2 S h3 r 1 c22 c23 d21 c1 r 3 c2 r2 c3 d2 d3


 d1 r 1 c1 c2 d2 r 2 c23 d2 r 3 c22 d3 r1 c1 c3 d3

 Ch2 Ch3 2r 1 c2 c3 d21  d1 r 1 c1 r 2 c2 c3 d2

c2 r 1 c1 r3 c3 d3 c1 2r 2 c3 d22  r 2 c2 r3 c3 d2 d3


2r 3 c2 d23
;
83
sqm

S h2 S h3 c2 d1 c1 d2 c1 d3

 Ch2 Ch3 c3 d1 c1 d3


1 r 2 d2 c3 d1 c1 d3 1 Ch1 Ch2

72

 Case 2



svv r 2m R1 r 1m R2 r2m Ch1 S 2 r 1m C2 S h1 =D;
 


sv m r 2m R1 C22  C2 Ch1 S 22 R2 S 2 S h1 ;




 r1m R1 S 2 S h1 R2 C2  Ch1 Ch1 S 2h1 =D;


smm L1 L2 r 1m Ch1 S 2  r2m C2 S h1 =D;


fvv r 2m R1 r 1m R2 r 2m S 2 r 1m S h1 =D;


fv m C2  Ch1 r 2m R1 r 1m R2 =D;


fmm L1 L2 r 1m S 2  r 2m S h1 =D;

75
76
77
78





R1 D r 31m  a2 r1m 2  m ; R2 D r 32m a2 r 2m 2  m ;
2

 a m; L2 D

D r 21m S 2 S h1

 r 22m S 2 S h1

r 22m


 Ch2 Ch3 c2 d1 c1 d2 S h2 S h3 c3 d1 c1 d3


1 r 1 d1 c3 d2 c2 d3 1 Ch1 Ch2

 S h3 c3 d1  c1 d3 S h2 c2 d1  c1 d2 Ch1 Ch3 c2 d1


84
c1 d2 S h1 S h3 c3 d2 c2 d3 ;

74

where r im ; Chi ; S hi ; Ci ; S i (with i = 1, 2) are dened in Eqs. (11) and

(20). The other parameters are as follows:

L1 Dr 21m

 S h3 c3 d2  c2 d3 S h1 c2 d1 c1 d2

73

79

80

a m :
81


2
2
2
r 1m r 2m C2  Ch1 S 2  S h1 :
82

FSDT
Explicit expressions for the coefcients of the DS matrix are
given only for case 1 for brevity. The coefcients for the other 3
cases can be obtained by following the procedure reported in
Section 2.2.3. Given the complexity of these coefcient, the
use of a symbolic computation program such as Mathematica
[25] is essential. In order to avoid numerical instabilities and
overows, the expressions need to be simplied and carefully
constructed. Full simplication is achieved by using the explicit
expressions of the terms appearing in the matrix A and R (Eqs.



Ghk
c2 d1 c1 d2 1 r 3 d3 1 Ch1 Ch3
D
 S h2 c3 d2 c2 d3 S h1 c3 d1

sqt

Ghk 
Ch3 c1 d2 d3 r2 d2  r3 d3 r 1 d21 c3 d2 c2 d3
D


 d1 r 2 c3 d22  2r 3 c3 d2 d3 r 3 c2 d23

 S h1 S h2 c3 d21 r 2 d2  r 3 d3 r 3 d2 d3 c3 d2 c2 d3


d1 r 1 c3 d22  r 2 c1 r 1 c2 d2 d3 r 3 c1 d23

 Ch2 c1 d2 d3 r 2 d2  r 3 d3 r1 d21 c3 d2 c2 d3


d1 r 2 c3 d22  2r 2 c2 d2 d3 r 3 c2 d23

S h1 S h3 r 2 d2 d3 c3 d2  c2 d3 d21 r 2 c2 d2 r 3 c2 d3


d1 r 2 c1 d22  r 3 c1 r 1 c3 d2 d3 r 1 c2 d23
 

Ch1  r 1 c3 d21 d2 r 2 c3 d1 d22  r 1 c2 d21 d3 2r 1 c1 d1 d2 d3
 r 2 c1 d22 d3 r 3 c2 d1 d23  r 3 c1 d2 d23

 S h2 S h3 c1 d2 d3 r 3 d2 r 2 d3 r 1 d21 c2 d2 c3 d3


 d1 r 1 c1 d22 r3 c2 d2 d3 r 2 c3 d2 d3 r 1 c1 d23

Ch2 Ch3 c1 d2 d3 r 2 d2 r 3 d3 r 1 d21 c3 d2 c2 d3



d1 r 2 c3 d22  2r 1 c1 r2 c2 r 3 c3 d2 d3 r3 c2 d23
;
85

409

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410



D
r 3 c2 d1 r2 c3 d1 r 3 c1 d2  r 1 c3 d2  r 2 c1 d3 r 1 c2 d3 Ch3 S h1 S h2 c2 d1 c1 d2 S h3 Ch2 S h1 c3 d1  c1 d3
D

Ch1 S h2 c3 d2 c2 d3 ;

smm





D
r 3 amc3 1 Ch1 Ch3 S h2 d2 c2 d1 c1 d2  1 Ch1 Ch2 S h3 2c3 d1 d2 c2 d1 d3 c1 d2 d3
D



 

S h1 Ch2 Ch3 d1 c2 d1  c1 d2 d1 c2 d1 c1 d2  S h2 S h3 c3 d21 c3 d22  c1 d1 d3  c2 d2 d3
r 1  amc1 S h2 d2  Ch1 Ch3 d2



1 Ch1 Ch2 S h3 d3 c3 d2 c2 d3 S h1 c3 d1 d2 c2 d1 d3  2c1 d2 d3 Ch2 Ch3 c3 d1 d2  c2 d1 d3 2c1 d2 d3




S h2 S h3 c2 d1 d2 c1 d22  c3 d1 d3 c1 d23  r 2  amc2 1 Ch1 Ch2 S h3 d3 c3 d1 c1 d3




1 Ch1 Ch3 S h2 c3 d1 d2 2c2 d1 d3  c1 d2 d3 S h1 Ch2 Ch3 d1 c3 d1  c1 d3 d1 c3 d1 c1 d3  S h2 S h3 c2 d21


 c1 d1 d2  c3 d2 d3 c2 d23
;

86

smt

87

Gh 
Ch2 c1 r 2 c2  r3 c3 d2 d3  d1 r 1 c1 r 2 c2 c3 d2 c2 r 1 c1  2r2 c2 r 3 c3 d3  S h1 S h3 r2 c2  r 3 c3 d1 c2 d1 c1 d2
12D

r 1 c1 r 2 c2 c3 d2 d3 c2 r1 c1  r 2 c2 d23  Ch3 c1 r 2 c2 r 3 c3 d2 d3 d1 c3 r 1 c1 r 2 c2  2r 3 c3 d2 r 1 c1 c2 r 3 c2 c3 d3


S h1 S h2 c3 r 1 c1  r 3 c3 d22 c2 r 1 c1 r3 c3 d2 d3 r 2 c2 r3 c3 d1 c3 d1 c1 d3 Ch1 r 1 c1 c3 d1 d2 r2 c2 c3 d1 d2

 r 1 c1 c2 d1 d3 r 3 c2 c3 d1 d3 2r 1 c21 d2 d3  r 2 c1 c2 d2 d3  r 3 c1 c3 d2 d3 S h2 S h3 c2 r1 c1 r 3 c3 d1 d2 c1 r 1 c1  r 3 c3 d22

stt

r1 c1  r 2 c2 d3 c3 d1 c1 d3  Ch2 Ch3 c2 r 1 c1 2r 2 c2  r 3 c3 d1 d3 d2 c3 r 1 c1  r 2 c2

2r3 c3 d1 c1 2r 1 c1  r2 c2  r 3 c3 d3 g;

88


Ghk 
 r 1 S h2 S h3 c22 d21  2r 1 c2 c3 d21  r1 S h2 S h3 c23 d21 r 2 S h1 S h3 c1 c2 d1 d2 r1 S h2 S h3 c1 c2 d1 d2 r 1 c1 c3 d1 d2  r 2 S h1 S h2 c1 c3 d1 d2
D

fqq

r2 c2 c3 d1 d2  r 1 S h1 S h2 c2 c3 d1 d2 r 1 S h1 S h3 c23 d1 d2 r 2 S h2 S h3 c23 d1 d2  r 2 S h1 S h3 c21 d22  2r 2 c1 c3 d22  r 2 S h1 S h3 c23 d22 r 1 c1 c2 d1 d3

 r3 S h1 S h3 c1 c2 d1 d3 r 1 S h1 S h2 c22 d1 d3 r3 S h2 S h3 c22 d1 d3 r 3 S h1 S h2 c1 c3 d1 d3 r1 S h2 S h3 c1 c3 d1 d3 r 3 c2 c3 d1 d3
 r1 S h1 S h3 c2 c3 d1 d3 r 2 S h1 S h2 c21 d2 d3 r3 S h1 S h3 c21 d2 d3 r 2 c1 c2 d2 d3  r 3 S h2 S h3 c1 c2 d2 d3 r 3 c1 c3 d2 d3  r 2 S h2 S h3 c1 c3 d2 d3

r3 S h1 S h2 c2 c3 d2 d3 r 2 S h1 S h3 c2 c3 d2 d3  r 3 S h1 S h2 c21 d23  2r3 c1 c2 d23  r 3 S h1 S h2 c22 d23 Ch2 Ch3 2r1 c2 c3 d21 c1 r 2 c2

r3 c3 d2 d3  d1 r 1 c1 c3 d2 r2 c2 c3 d2 r 1 c1 c2 d3 r 3 c2 c3 d3  Ch1 Ch3 c1 d2 2r 2 c3 d2 r2 c2 d3 r 3 c3 d3 d1 r 1 c1 r 2 c2 c3 d2


 c2 r 1 c1 r 3 c3 d3 Ch2 c1 d3 r2 c2 d2 r 3 c3 d2  2r3 c2 d3 d1 r 1 c1 r2 c2 c3 d2 c2 r1 c1 r 3 c3 d3 ;
fqm




Ghk 
 Ch1 c3 d2 c2 d3 S h2 c2 d1 c1 d2 r 1 d1  r 3 d3 S h3 r 1 d1  r 2 d2 c3 d1  c1 d3 Ch2 c3 d1 c1 d3 S h1 c2 d1
D


c1 d2 r 2 d2  r 3 d3 S h3 r 1 d1 r 2 d2 c3 d2 c2 d3 Ch3 c2 d1 c1 d2 S h1 r 2 d2 r 3 d3 c3 d1 c1 d3 S h2 r 1 d1
 r 3 d3 c3 d2 c2 d3 g;

fqt

89

90


Ghk 
 r 2 S h1 S h3 c2 d21 d2 r 1 S h2 S h3 c2 d21 d2 r 1 c3 d21 d2 r 2 S h1 S h2 c3 d21 d2 r2 S h1 S h3 c1 d1 d22  r 1 S h2 S h3 c1 d1 d22 r 2 c3 d1 d22
D
r1 S h1 S h2 c3 d1 d22 r 1 c2 d21 d3 r 3 S h1 S h3 c2 d21 d3  r3 S h1 S h2 c3 d21 d3 r 1 S h2 S h3 c3 d21 d3  2r1 c1 d1 d2 d3  r 2 S h1 S h2 c1 d1 d2 d3
 r3 S h1 S h3 c1 d1 d2 d3  2r 2 c2 d1 d2 d3  r1 S h1 S h2 c2 d1 d2 d3  r 3 S h2 S h3 c2 d1 d2 d3  2r3 c3 d1 d2 d3  r 1 S h1 S h3 c3 d1 d2 d3
 r2 S h2 S h3 c3 d1 d2 d3 r 2 c1 d22 d3 r 3 S h2 S h3 c1 d22 d3  r3 S h1 S h2 c3 d22 d3 r 2 S h1 S h3 c3 d22 d3 r 3 S h1 S h2 c1 d1 d23  r1 S h2 S h3 c1 d1 d23
r3 c2 d1 d23 r 1 S h1 S h3 c2 d1 d23 r 3 c1 d2 d23 r 2 S h2 S h3 c1 d2 d23 r 3 S h1 S h2 c2 d2 d23  r 2 S h1 S h3 c2 d2 d23  Ch2 Ch3 c1 d2 d3 r 2 d2 r 3 d3
  


r1 d21 c3 d2 c2 d3  d1 r2 c3 d22 2r 1 c1 d2 d3 r3 c2 d23 Ch1  Ch3 c1 d2 d3 r 2 d2  r 3 d3 r 1 d21 c3 d2 c2 d3






d1 r 2 c3 d22  2r 2 c2 d2 d3 r 3 c2 d23
Ch2 c1 d2 d3 r 2 d2  r 3 d3 r 1 d21 c3 d2 c2 d3  d1 r2 c3 d22  2r 3 c3 d2 d3 r 3 c2 d23
;




D
r 3 c2 d1 r 2 c3 d1 r 3 c1 d2  r1 c3 d2  r 2 c1 d3 r 1 c2 d3 S h2 S h3 c3 d2  c2 d3 S h1 S h2 c2 d1  c1 d2 S h3 c3 d1 c1 d3 ;
D
  




 


D
 r 3 c2 d1  r 2 c3 d1  r 3 c1 d2 r 1 c3 d2 r2 c1 d3  r 1 c2 d3 Ch3  S h1 d1 S h2 d2 Ch2 S h1 d1  S h3 d3 Ch1  S h2 d2 S h3 d3 ;
D

91

fmm

92

fmt

93

ftt

3

Gh 
 r 2 S h1 S h3 c22 d21 r2 S h1 S h2 c2 c3 d21 r 3 S h1 S h3 c2 c3 d21  r 3 S h1 S h2 c23 d21 r 2 S h1 S h3 c1 c2 d1 d2 r 1 S h2 S h3 c1 c2 d1 d2 r1 c1 c3 d1 d2
12D
 r3 S h1 S h3 c1 c3 d1 d2 r 2 c2 c3 d1 d2  r 3 S h2 S h3 c2 c3 d1 d2  2r 3 c23 d1 d2  r 1 S h2 S h3 c21 d22 r1 S h1 S h2 c1 c3 d22 r 3 S h2 S h3 c1 c3 d22

 r3 S h1 S h2 c23 d22 r 1 c1 c2 d1 d3  r2 S h1 S h2 c1 c2 d1 d3  2r2 c22 d1 d3 r 3 S h1 S h2 c1 c3 d1 d3 r1 S h2 S h3 c1 c3 d1 d3 r 3 c2 c3 d1 d3

 r2 S h2 S h3 c2 c3 d1 d3  2r1 c21 d2 d3 r 2 c1 c2 d2 d3  r1 S h1 S h2 c1 c2 d2 d3 r 3 c1 c3 d2 d3  r 1 S h1 S h3 c1 c3 d2 d3 r3 S h1 S h2 c2 c3 d2 d3
r2 S h1 S h3 c2 c3 d2 d3  r 1 S h2 S h3 c21 d23 r1 S h1 S h3 c1 c2 d23 r 2 S h2 S h3 c1 c2 d23  r 2 S h1 S h3 c22 d23 Ch2 Ch3 c2 r 1 c1 r3 c3 d1 d3

d2 r 1 c1 r 2 c2 c3 d1 c1 2r 1 c1  r2 c2  r 3 c3 d3 Ch1 Ch2 c1 r 2 c2  r3 c3 d2 d3 d1 c3 r 1 c1  r 2 c2 2r 3 c3 d2 c2 r 1 c1


 r3 c3 d3  Ch3 c1 r 2 c2  r3 c3 d2 d3 d1 r 1 c1 r 2 c2 c3 d2 c2 r 1 c1  2r 2 c2 r 3 c3 d3 ;

94

410

M. Boscolo, J.R. Banerjee / Computers and Structures 89 (2011) 395410

where






D 2c3 d2 c2 d3 1 Ch1 Ch3 S h2 c2 d1  c1 d2 1 Ch1 Ch2 S h3 c3 d1 c1 d3 S h1 2c2 d1 c1 d2 c3 d1







c1 d3  2Ch2 Ch3 c2 d1 c1 d2 c3 d1 c1 d3 S h2 S h3 c22 c23 d21 c21 c23 d22  2c1 c3 d1 d3 c21 c22 d23  2c2 d2 c1 d1 c3 d3
95
References
[1] Zienkiewicz OC, Taylor, RL. The nite element method. The basis. 5th ed. vol. 1.
Butterworth-Heinemann, 2000.
[2] Banerjee JR. Dynamic stiffness formulation for structural elements: a general
approach. Comput Struct 1997;63(1):1013.
[3] Banerjee JR. Coupled bending-torsional dynamic stiffness matrix for beam
elements. Int J Numer Methods Eng 1989;28:128398.
[4] Banerjee JR. Free vibration analysis of a twisted beam using the dynamic
stiffness method. Int J Solids Struct 2001;38(3839):670322.
[5] Banerjee JR. Free vibration of sandwich beams using the dynamic stiffness
method. Comput Struct 2003;81(1819):191522.
[6] Banerjee JR. Development of an exact dynamic stiffness matrix for free
vibration analysis of a twisted Timoshenko beam. J Sound Vib 2004;270(1
2):379401.
[7] Banerjee JR, Su H, Jackson DR. Free vibration of rotating tapered beams using
the dynamic stiffness method. J Sound Vib 2006;298(4-5):103454.
[8] Banerjee JR, Cheung CW, Morishima R, Perera M, Njuguna J. Free vibration of a
three-layered sandwich beam using the dynamic stiffness method and
experiment. Int J Solids Struct 2007;44(2223):754363.
[9] Williams FW, Wittrick WH. An automatic computational procedure for
calculating natural frequencies of skeletal structures. Int J Mech Sci
1970;12(9):78191.
[10] Anderson MS, Williams FW. BUNVIS-RG: exact frame buckling and vibration
program, with repetitive geometry and substructuring. J Spacecraft Rockets
1987;24:35361.
[11] Akkeson B. A computer program for plane frame vibration analysis by an exact
method. Int J Numer Methods Eng 1976;10:122131.
[12] Wittrick WH, Williams FW. A general algorithm for computing natural
frequencies of elastic structures. Quarterly J Mech Appl Math
1970;24(3):26384.
[13] Williams FW, Howson WP. Compact computation of natural frequencies and
buckling loads for plane frames. Int J Numer Methods Eng 1977;11:106781.
[14] Wittrick WH. A unied approach to initial buckling of stiffened panels in
compression. Int J Numer Methods Eng 1968;11:106781.
[15] Wittrick WH. General sinusoidal stiffness matrices for buckling and vibration
analyses of thin at-walled structures. Int J Mech Sci 1968;10:94966.
[16] Williams FW, Wittrick WH. Computational procedures for a matrix analysis of
the stability and vibration of thin at-walled structures in compression. Int J
Mech Sci 1969;11(12):97998.

[17] Wittrick WH, Williams FW. Buckling and vibration of anisotropic or isotropic
plate assemblies under combined loadings. Int J Mech Sci 1974;16(4):20939.
[18] Williams FW, Kennedy D, Butler R, Anderson MS. VICONOPT: program for
exact vibration and buckling analysis or design of prismatic plate assemblies.
AIAA J 1991;29:19278.
[19] Anderson MS, Kennedy D. Inclusion of transverse shear deformation in exact
buckling and vibration of composite plate assemblies. Tech. Rep. NASA/CR1993-4510, Langley, Hampton; 1993.
[20] Anderson MS, Kennedy D. Transverse shear deformation in exact buckling and
vibration of composite plate assemblies. AIAA J 1993;31(10):19635.
[21] Reddy JN, Phan ND. Stability and vibration of isotropic, orthotropic and
laminated plates according to a higher-order shear deformation theory. J
Sound Vib 1985;98(2):15770.
[22] Carrera E, Boscolo M. Classical and mixed nite elements for static and
dynamic analysis of piezoelectric plates. Int J Numer Methods Eng
2007;70:113581.
[23] Leung AYT, Zhou WE. Dynamic stiffness analysis of laminated composite
plates. Thin-Walled Strcut 1996;25(2):10933.
[24] Mindlin RD. Inuence of rotatory inertia and shear on exural vibrations of
isotropic, elastic plates. J Appl Mech 1951;18:10316.
[25] Wolfram S. The mathematica book. 4th ed. Cambridge University Press; 1999.
[26] Reddy JN. Theory and analysis of elastic plates. CRC Press; 2006.
[27] Reddy JN, Miravete A. Practical analysis of composite laminates. CRC Press;
2000.
[28] Timoshenko S, Woinowsky-Krieger S. Theory of plates and shells. McGraw-Hill
Book Company; 1959.
[29] Lowe PG. Basics principles of plate theory. Surrey University Press; 1982.
[30] Reissner E. On the theory of bending of elastic plates. J Math Phys
1944;23(4):18491.
[31] Ince EL. Ordinary differential equations. Dover Publications; 1920.
[32] Ferrar WL. Higher algebra. Oxford: Clarendon Press; 1948.
[33] Leissa AW. Vibration of plates. Tech. Rep. NASA SP-160, National Aeronautics
and Space Administration, Washington; 1969.
[34] Chopra I. Vibration of stepped thickness plates. Int J Mech Sci 1974;16:33744.
[35] Guo SJ, Keane Aj, Moshere-Torbati M. Vibration analysis of stepped thickness
plates. J Sound Vib 1997;204(4):64557.
[36] Cheung YK. Finite strip method in structural analysis. Pergamon Press; 1976.