Engineering Structures 26 (2004) 171–186

www.elsevier.com/locate/engstruct

Application of differential quadrature (DQ) and harmonic

differential quadrature (HDQ) for buckling analysis of thin isotropic
plates and elastic columns
Ömer Civalek

Dokuz Eylül University, Engineering Faculty, Civil Engineering Department, Ízmir, Turkey

Received 7 January 2003; received in revised form 9 September 2003; accepted 15 September 2003

Abstract

The main purpose of this study is to compare the methods of differential quadrature (DQ) and harmonic differential quadrature
(HDQ). For this purpose, DQ and HDQ methods are presented for buckling, bending, and free vibration analysis of thin iso-
tropic plates and columns. Plates of different shapes such as rectangular, circular, square, skew, trapezoidal, annular, and sec-
torial plate subjected to different boundary conditions are selected to demonstrate the accuracy of the method. Four different
support conditions are taken into consideration for columns. Numerical results are presented to illustrate the method and demon-
strate its efficiency. It is emphasized that the HDQ method gives more accurate results and needs less grid points than the DQ
method.
# 2003 Elsevier Ltd. All rights reserved.

Keywords: Harmonic differential quadrature; Plates and columns; Buckling; Curvilinear domain; Numerical method

1. Introduction In addition to this, in a large number of practical

It is well known that analysis of engineering systems
includes two main stages, such as, construction of a
mathematical model for a given physical phenomenon
and the solution to this mathematical equation. Real
physical systems or engineering problems are often
described by partial differential equations, either linear
or non-linear and in most cases, their closed form solu-
tions are extremely difficult to establish. As a result,
approximate numerical methods have been widely used
to solve partial differential equations that arise in
almost all engineering disciplines. The most commonly
used numerical methods for such applications are the
finite element, finite difference, Ritz, and boundary
element method, and most engineering problems can be
solved by these methods to adequate accuracy if a
proper and sufficient number of grid points are used.


Corresponding address. Eryaman Eston Blokları, Güven Apt. 18,
Daire 6, Eryaman-Ankara, Turkey. Tel.: +90-312-280-5036; fax:
+90-312-309-3665.
E-mail address: civalek@yahoo.com (Ö. Civalek).
0141-0296/$ - see front matter # 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2003.09.005
applications where only reasonably accurate solutions
at few specified physical coordinates are of interest,
conventional numerical methods such as finite element
or finite difference method require a large number of
grid points and a large computer capacity. Among a
variety of numerical methods, the finite element
method is by far the most effectively and widely used
method. Furthermore, the finite element method is still
an effective method especially for systems with complex
geometry and load conditions or applications with
non-linear behavior and it has many successful applica-
tions. In seeking a more efficient numerical method
that requires fewer grid points yet achieves acceptable
accuracy, the method of differential quadrature (DQ)
was introduced by Bellman et al. [1]. Since then, appli-
cations of the differential quadrature method to various
engineering problems have been investigated and their
success has shown the potential of the method as an at-
tractive numerical analysis technique [6,26,28,30]. The
stability analysis of plates may be either in closed form
or approximate. The closed form solutions consist of
172 Ö. Civalek / Engineering Structures 26 (2004) 171–186
techniques for seeking direct solutions to the governing
differential equation of plates. A closed form or rigorous
solution of plates can be obtained for only a limited
number of cases. For the majority of practical pro-
blems, a closed form namely analytical solution either
cannot be obtained or is of such a complicated nature
that it can be applied only with great difficulty in a
practical computation. For many situations, numerical
solutions are the only method that can be employed.
The idea of the differential quadrature method is to
quickly compute the derivative of a function at any
grid point within its bounded domain by estimating a
weighted linear sum of values of the function at a small
set of points related to the domain. In this study, criti-
cal buckling loads of circular, square, skew, and rec-
tangular plates and elastic columns are investigated by
using DQ and HDQ.
There are many methods available in the literature to
study the buckling, bending, and dynamic behavior of
thin plates with different boundary and loading con-
ditions. For details, one may refer to Timoshenko and
Gere [40], Szilard [55], Leissa [16], Reddy [54], Ugural
[52], etc. Exact solutions for plate problems are rather
difficult to obtain, except for a few simple cases. In
many cases, one may have to resort to various approxi-
mate, namely numerical methods. Each method has its
own advantages and disadvantages. Of the various
methods proposed in recent times, one can cite the
finite differences, finite element and boundary element
methods as the most efficient and universal methods
for solving variant type plate problems.
The focus of this paper is on the comparison study
of the DQ and HDQ methods. For this reason, the
applications are limited to those problems having
smooth solutions for simplicity. In addition to this,
since only thin plates are considered in this paper, there
are some assumptions regarding the behavior of thin
plates. These include the transverse deflections of the
plates being small when compared to the thickness of
the plate. Thus, middle-surface stretching caused by
bending can be neglected; that is, membrane action
resulting from flexure is negligible compared to the
flexure. The material of the plate is homogeneous, iso-
tropic, and obeys Hooke’s law.
2. Differential quadrature method (DQM)
The numerical solution of partial differential equa-
tions plays a considerable role in the areas of engineer-
ing. In many cases, all that is desired is a moderately
accurate solution at a few points which can be calcu-
lated rapidly. Therefore, an effective numerical tech-
nique for the solution of non-linear partial equations is
very desirable. The differential quadrature method is
one which can satisfy the above purpose. The basic
idea of the differential quadrature method is that the
derivative of a function, with respect to a space variable
at a given sampling point, is approximated as a weigh-
ted linear sum of the sampling points in the domain of
that variable [2]. As with other numerical analysis tech-
niques, such as finite element or finite difference meth-
ods, the DQM also transforms the given differential
equation into a set of analogous algebraic equations in
terms of the unknown function values at the reselected
sampling points in the field domain. The problem areas
in which applications of the differential quadrature
method may be found in the available literature include
fluid mechanics, static and dynamic structural mech-
anics, static aeroelasticity, and some problems in
chemical engineering [3,5,7,8]. Mingle applied the dif-
ferential quadrature method to the solutions of the
non-linear diffusion equation [29]. Civan and Sliepce-
vich extended and generalized the method of differen-
tial quadrature and applied it to transport processes
[20], the Poisson equation [22], multi-dimensional pro-
blems [21] and Thomas–Fermi equation [23]. In recent
years, the DQM has been largely promoted by Bert
and associates who were the first to introduce the
method as a tool for structural analysis [4]. Liew and
his co-workers also made most important studies on
DQ and HDQ [25,50]. In fact, Liew and his associates
had been made most effective in supplementing to the
theory and application of the DQM. Recent work of
Liew and associates was mainly on the three-dimen-
sional vibration, bending and stability analysis of
plates [36,44,50]. Various problems in structural mech-
anics have been solved successfully using this method
[9,24,25]. The authors applied the DQM to the stab-
ility, vibration and bending analysis of elastic bars
[12,13], circular plates and guided beams [14,15]. It has
been claimed that the DQM has the capability of pro-
ducing highly accurate solutions with minimal compu-
tational effort. Recently, Liew and his co-workers also
proposed a new kind of DQ method [57]. This new
proposed method has been called the moving least
squares differential quadrature (MLSDQ). This new
approach (MLSDQ) exploits the merits of both the DQ
and meshless method. The method seemingly has a
high potential as an alternative to the above mentioned
conventional numerical solution techniques such as
the finite element and finite difference methods
[8,17,28,32,37]. All these studies have demonstrated
that the application of the DQ methods leads to accu-
rate results with less computational effort and that
there is a potential that the method may become an
alternative to conventional methods such as finite dif-
ferences and finite element [34,38,39]. Therefore,
research on extensions and applications of the method
becomes an important endeavor. In the differential
quadrature method, a partial derivative of a function
with respect to a space variable at a discrete point is
approximated as a weighted linear sum of the function
 Ö. Civalek / Engineering Structures 26 (2004) 171–186 173

values at all discrete points in the region of that variable.

For simplicity, we consider a one-dimensional function
u(x) in the ½1; 1 domain, and N discrete points. Then,
the first derivatives at point i, at x ¼ xi , are given by:


@u

XN
ux ðxi Þ ¼
¼ Aij uðxj Þ; i ¼ 1;2; . . . ;N ð1Þ
@x x¼xi j¼1

where xj are the discrete points in the variable domain,

u(xj) are the function values at these points and Aij are
the weighting coefficients for the first-order derivative
attached to these function values. Bellman et al. [1,2]
suggested two methods to determine the weighting
coefficients. The first one is to let Eq. (1) be exact for
the test functions
Fig. 1. Rectangular domain and grid distribution.
k1
uk ðxÞ ¼ x ; k ¼ 1;2; . . . ;N ð2Þ
which leads to a set of linear algebraic equations lem, we can solve a two-dimensional problem with
basic variables in a rectangular domain (Fig. 1). Sup-
X
N posing there are Nx points in the x-direction and Ny
ðk  1Þxk2
i ¼ Aij xk1
j ; points in the y-direction.
j¼1
The rth-order partial derivative of uðx;yÞ with
for i ¼ 1;2; . . . ;N and k ¼ 1;2; . . . ;N ð3Þ respect to x, the sth-order partial derivative of uðx;yÞ
which represents N sets of N linear algebraic equations. with respect to y and the (r þ s)th-order partial deriva-
This equation system has a unique solution because its tive of uðx;yÞ with respect to both x and y can be dis-
matrix is of Vandermonde form. This equation may be cretely expressed at the point ðxi ;yj Þ as
solved for the weighting coefficients analytically using

@ r u

XNx
ðrÞ
Hamming’s method [27] or numerical method using ¼ A uðxk ;yj Þ; r ¼ 1;2; . . . ;Nx  1 ð7Þ
@xr
x¼xi k¼1 ik
certain special algorithms for Vandermonde equations,
such as the method of Bjorck and Pereyra [10]. In
Ny
@ s u

X ðsÞ
order to reduce the complexity of the derivative ¼ B uðxi ;yk Þ; s ¼ 1;2; . . . ;Ny  1 ð8Þ
approximation formulae and thereby reduce the com- @ys
y¼yj k¼1 jk
putational effort, it is advantageous to use quadrature
 
Ny
@ ðrþsÞ u

@ r @ s u

X
Nx
ðrÞ
X ðsÞ
approximation formulae for also the second-, third-
¼
¼ Aik Bjm uðxk ;ym Þ;
and higher-order derivatives. Thus, the weighting coef- @xr @ys xi yj @xr @ys xi yj k¼1 m¼1
ficients for each formula will be different from those for i ¼ 1;2; . . . ;Nx and j ¼ 1;2; . . . ;Ny ð9Þ
the first-order derivative. Similar to the first order, the
ðrÞ ðsÞ
second-order derivative can be written as: where Aij and Bij are the weighting coefficients asso-

ciated with the rth-order partial derivative of uðx;yÞ
@ 2 u

XN
uxx ðxi Þ ¼ 2
¼ Bij uðxj Þ; i ¼ 1;2; . . . ;N ð4Þ with respect to x at the discrete point xi and the sth-
@x x¼xi j¼1 order partial derivative of uðx;yÞ with respect to y at yj,
respectively. A comprehensive and chronological
where the Bij is the weighting coefficients for the second- review of the work related to the DQ method can be
order derivative. Eq. (4) also can be written as: found in the well-known paper of Bert and Malik [8].


@ 2 u

XN X
N
uxx ðxi Þ ¼ 2
¼ Aij Ajk uðxk Þ;
@x x¼xi j¼1 k¼1
3. Harmonic differential quadrature
i ¼ 1;2; . . . ;N ð5Þ
Another way to determine the weighting coefficients
Again, the function given by Eq. (2) is used so that is to employ harmonic functions, named the harmonic
the second-order derivative is: differential quadrature (HDQ). Harmonic differential
X
N quadrature has been proposed by Striz et al. [19]. Un-
ðk  1Þðk  2Þxk3
i ¼ Bij xk1
j ð6Þ like the differential quadrature that uses the polynomial
j¼1 functions, such as power functions, Lagrange inter-
This can be solved in the same manner as indicated polated, and Legendre polynomials as the test func-
for Eq. (3) above. Similar to the one-dimensional prob- tions, harmonic differential quadrature uses harmonic
174 Ö. Civalek / Engineering Structures 26 (2004) 171–186

or trigonometric functions as the test functions. Thus, choice of grid points. It should be mentioned that in
this method is called the HDQ method. Shu and Xue the differential quadrature solutions, the sampling
proposed an explicit means of obtaining the weighting points in various coordinate directions may be different
coefficients for the HDQ [18]. The harmonic test func- in number as well as in their type. Three different types
tion hk ðxÞ used in this approach is defined as: of sampling grids are taken into consideration in this
ðx  x0 Þp ðx  xk1 Þp ðx  xkþ1 Þp ðx  xN Þp study. A natural, and often convenient, choice for sam-
sin sin sin sin pling points is that of equally spaced points. These
hk ðxÞ ¼ 2 2 2 2 ;
ðxk  x0 Þp ðxk  xk1 Þp ðxk  xkþ1 Þp ðxk  xN Þp points are given by:
sin sin sin sin
2 2 2 2
for k ¼ 0;1;2;. . . ;N ð10Þ i1
Type-I : xi ¼ ; i ¼ 1;2; . . . ;Nx ð16aÞ
Nx  1
According to the HDQ, the weighting coefficients of i1
yi ¼ ; i ¼ 1;2; . . . ;Ny ð16bÞ
the first-order derivatives Aij for i 6¼ j can be obtained Ny  1
by using the following formula: in the related directions. Sometimes, the differential
ðp=2ÞPðxi Þ quadrature solutions deliver more accurate results with
Aij ¼ ; i;j ¼ 1;2;3; . . . ;N ð11Þ unequally spaced sampling points. A better choice for
Pðxj Þsin½ðxi  xj Þ=2p
the positions of the grid points between the first and
where: the last points at the opposite edges is that correspond-
Y
N x  x  ing to the zeros of orthogonal polynomials such as the
i j
Pðxi Þ ¼ sin p ; for j ¼ 1;2;3; . . . ;N zeros of Chebyshev polynomials [36]. Another choice
2
j¼1;j6¼i that is found to be even better than the Chebyshev and
The weighting coefficients of the second-order deri- Legendre polynomials is the set of points proposed by
vatives Bij for i 6¼ j can be obtained using following Shu and Richards [31,32]. These points are given as:
formula:
  
h x  x  i 1 2i  1
ð1Þ i j Type-II : xi ¼ 1  cos p ;
Bij ¼ Aij 2Aii  pctg p ; 2 Nx  1
2
i;j ¼ 1;2;3; . . . ;N ð12Þ i ¼ 1;2; . . . ;Nx ð17aÞ

  
The weighting coefficients of the first-order and 1 2i  1
ðpÞ yi ¼ 1  cos p ; j ¼ 1;2; . . . ;Ny ð17bÞ
second-order derivatives Aij for i ¼ j are given as: 2 Ny  1

ðpÞ
X
N
ðpÞ
in the x- and y-directions, respectively. These type grid
Aii ¼  Aij ; p ¼ 1 or 2; points are known the Chebyshev–Gauss–Lobatto
j¼1;j6¼i
points.
and for i ¼ 1;2; . . . ;N ð13Þ Type-III: Equally spaced sampling points with adjac-
The weighting coefficient of the third- and fourth- ent-d points. This practical method proposed by Jang
order derivatives can be computed easily from Aij and et al. [28] employs points at a very small distance
Bij by [18] (d ¼ 0:00001) from the boundary points. The sampling
grid points are defined as:
X
N
Cij ¼ Aik Bkj ð14Þ x1 ¼ 0; x2 ¼ d; xN1 ¼ 1  d;
k¼1
xN ¼ 0 for adjacent points ð18Þ
X
N
Dij ¼ Bik Bkj ð15Þ i1
xi ¼ ; for i ¼ 3;4; . . . ; ðN  2Þ ð19Þ
k¼1 N 3
The main advantage of HDQ over the differential These three types of sampling points are more general,
quadrature is its ease of the computation of the weight- well-known, and widely used by the DQ researchers.
ing coefficients without any restriction on the choice of More detailed information for choices of sampling grid
grid points. points can be found in the literature [8,13,14,33,53].

3.1. Choices of sampling grid points

A decisive factor for the accuracy of the differential
quadrature solution is the choice of the sampling or
grid points. The main advantage of HDQ over the dif-
ferential quadrature is its ease of the computation of
the weighting coefficients without any restriction on the
4. Numerical applications and results
To verify the analytical formulation presented in the
previous section, elastic columns, circular, rectangular,
skew, trapezoidal, eccentric sectorial, and square plates
are considered. Plates of different types of boundary
 Ö. Civalek / Engineering Structures 26 (2004) 171–186 175

conditions are selected as test samples to demonstrate
the applicability and accuracy of the HDQ method.
The governing differential equations for free vibration,
bending and buckling of plates and columns are pre-
sented. The present formulations are based on classical
small deflection and thin plate theory. Then, the DQ
and HDQ methods are applied to these differential
equations. The results are obtained for each case using
various numbers of grid points. Subsequently, several
test samples for different support conditions are selec-
ted to demonstrate the convergence properties, accu-
racy and the simplicity in numerical implementation of
the HDQ procedures. The numerical results for various
example plate and column problems are tabulated
(Tables 1–9) and figured (Figs. 4–10), and a compari-
son of the present results with exact or other numerical
values available in the literature, when possible, is
made. In order to simplify the presentation, S, C, and
F represent simply supported, clamped, and free sup-
ports, respectively.
4.1. Buckling of circular plates under a uniform
compressive edge load
Consider a thin, circular plate of uniform thickness
subject to a uniform compressive radial load Fc dis-
tributed around the edge of the plate. The grid spacing
employed for circular plates with radius a is shown in
Fig. 2. There were two different governing equations in
the literature [9]. Both the third-order and the fourth-
order differential equations can be used to obtain the
buckling load. In this study, only the third-order gov-
erning equation is used. This equation is given by:
     
@3u 1 @2u 1 @u Fc a2 @u
þ  2 ¼ ; both the DQ and the exact method. The harmonic dif-
@r3 r @r2 r @r D @r
i ¼ 2;3; . . . ; ðN  1Þ
where a is known as the outside radius of the plate and
D denotes the flexural rigidity of plates, and is given as
D ¼ Eh3 =12ð1  m2 Þ, m is the Poisson ratio, E is the
modulus of elasticity of the plate material, h is the uni-
form plate thickness, u is the displacement in the z-
direction. Applying the differential quadrature method
to Eq. (20), one obtains
X
N
1 X N
1 XN
Cij Uj þ Bij Uj  2 Aij Uj
j¼1
Ri j¼1 Ri j¼1
Fc a2
þ ðAij Uj Þ ¼ 0; for i ¼ 2;3; . . . ;N  1 ð21Þ
D
Due to lowest buckling mode being axisymmetric,
the relation
Aij Uj ¼ 0
must be applied to find the buckling load.
Table 1 tabulates the critical buckling loads obtained
by the HDQ method for thin, isotropic circular plates
with clamped and simply supported boundary con-
ditions using the third-order governing equations.
Table 1 includes both the different numerical solutions
and the exact solutions. The authors used the finite
element method (FEM) for this problem earlier [11].
The results obtained from finite element method are
indicated by FEM. DQ results are obtained by using
the third-order differential equations for N ¼ 7 grid
points. In this case, Type-I sampling grid points are
used. For N ¼ 9, Bert et al. [9] obtained the buckling
loads. These results are also given in Table 1 for com-
parison. For N ¼ 9, HDQ and DQ results are also
obtained once again for this problem. Type-II grid
stretching is chosen for this second case. The obtained
buckling loads are compared with those calculated with
ferential quadrature results are generally in agreement
ð20Þ
with the results produced from the analytical [35] and
the DQ results [9]. It is found that the HDQ method

Table 1
 cr ¼ Ncr a2 =D)
Critical buckling load of the circular plates (m ¼ 0:30; N

Support DQ (N ¼ 7), FEM DQ HDQ (N ¼ 7), HDQ (N ¼ 9), Exact

conditions this study (N ¼ 11) [11] (N ¼ 9) [9] this study this study [35]
S–S 4.18 4.12 4.20 4.19 4.20 4.20
C–C 14.57 14.74 14.68 14.66 14.68 14.68

Table 2
 cr ¼ Ncr a2 =D)
Buckling loads of square plates (m ¼ 0:30; N

Support DQ (7  7), HDQ (7  7), DQ HDQ (9  9), Exact

conditions this study this study (9  9) [9] this study [35]
S–S–S–S 4.19 4.18 4.20 4.20 4.20
C–C–C–C 14.57 14.62 14.68 14.68 14.68
C–S–C–S 63.78 64.85 66.36 66.30 66.32
S–C–S–C 6.82 7.05 7.856 7.67 7.69
176 Ö. Civalek / Engineering Structures 26 (2004) 171–186

Table 3
 cr ¼ Ncr a2 =D)
Comparison of buckling loads of square plates (m ¼ 0:30; N

Support HDQ HDQ HDQ HDQ FEM FEM FEM FEM Exact
conditions (5  5) (7  7) (9  9) (11  11) (5  5) [11] (7  7) [11] (9  9) [11] (11  11), [35]
this study
S–S–S–S 4.12 4.18 4.20 4.20 4.02 4.08 4.12 4.16 4.20
C–C–C–C 14.18 14.62 14.68 14.68 13.55 13.82 13.98 14.56 14.68
C–S–C–S 64.01 64.85 66.30 66.32 70.11 69.75 68.79 68.34 66.32
S–C–S–C 6.93 7.05 7.67 7.69 8.84 – – 7.63 7.69

Table 4
Non-dimensional buckling load of rectangular plates for various aspect ratios

a/b C–C–C–S C–S–S–S

HDQ DQ FEM HDQ HDQ DQ HDQ FEM
(5  5) (5  5) [42] (9  9) [11] (7  7) (5  5) (5  5) [42] (7  7) (9  9) [11]
1/5 42.36 40.02 38.75 41.18 22.78 20.95 20.52 21.13
2/5 43.78 42.82 41.10 42.80 21.71 23.52 24.01 24.79
4/5 59.65 60.21 57.49 61.09 37.40 36.13 35.96 38.45
5/5 81.54 78.27 80.04 79.56 49.82 48.03 49.10 48.68

Table 5
Comparison of column buckling loads

FD (N ¼ 5) [41] Exact [40] HDQ (N ¼ 5) HDQ (N ¼ 7) HDQ (N ¼ 9) DQ (N ¼ 7) DQ (N ¼ 9)

C–C 41.360 39.478 38.745 39.034 39.476 40.067 39.450
C–F 2.333 2.467 2.556 2.304 2.466 2.288 2.461
S–S 11.548 9.869 10.184 9.621 9.869 9.614 9.868
C–S 22.296 20.142 22.960 21.025 20.142 21.239 20.133

Table 6
Comparison of column buckling loads for different numerical methods

Support FD FEM FEM FEM HDQ HDQ DQ DQ Exact

conditions (N ¼ 5) [41] (N ¼ 5) (N ¼ 7) (N ¼ 9) (N ¼ 5) (N ¼ 7) (N ¼ 5) (N ¼ 7) [40]
C–C 41.360 40.254 39.984 39.614 39.547 39.478 39.244 39.478 39.478
C–F 2.333 2.514 2.547 2.482 2.441 2.467 2.419 2.466 2.467
S–S 11.548 10.376 9.816 9.897 9.851 9.869 9.843 9.871 9.869
C–S 22.296 21.946 20.664 20.285 20.205 20.141 20.218 20.140 20.142

Table 7
Buckling coefficients of clamped skew plates [b=a ¼ 1; Ncr ¼ ðFx a2 Þ=D] for different type grid points

Skew DQ (9  9), DQ (11  11), HDQ (7  7), HDQ (9  9), HDQ (9  9), FEM Wang et al. DQM
angle (h) present studya present studyb present studyc present studyd present studye (11  11) [11] (11  11) [42]
45 21.35 20.29 20.01 21.01 20.25 18.65 20.23
60 12.89 13.49 13.66 12.94 13.52 11.73 13.54
75 11.82 10.90 10.80 11.13 10.82 12.44 10.84
90 9.96 11.23 10.02 10.02 10.06 13.08 10.07
a
Type-I grid sampling points (DQ—9  9).
b
Type-II grid sampling points (DQ—11  11).
c
Type-II grid sampling points (HDQ—7  7).
d
Type-I grid sampling points (HDQ—9  9).
e
Type-II grid sampling points (HDQ—9  9).
 Ö. Civalek / Engineering Structures 26 (2004) 171–186 177

Table 8
Buckling coefficients of trapezoidal plate [a=h ¼ 1; Ncr ¼ ðFx a2 Þ=p2 D; m ¼ 0:3; a ¼ 90]

Support DQ Ref. [49] Ref. [48] FEM [47] HDQ FEM FEM
conditions (11  11) (11  11) (15  15) [11] (11  11) [11]
S–S–S–S 4.03 4.00 4.00 4.00 4.01 4.02 3.98
C–C–C–C 9.97 10.08 10.07 – 10.08 10.06 11.07
S–C–S–Ca 7.82 7.70 – – 7.68 7.72 –
a
Exact (Timoshenko and Gere, 1959): 7.690.

Table 9
Comparison for HDQ and DQ solutions for first three non-dimensionala vibration frequencies of sectorial plates (r=b ¼ 2:0; a ¼ 45; S–S–S–S)

DQ (Nn ¼ Ng ¼ 17) HDQ (Nn ¼ Ng ¼ 15) Polar mesh FD [52] Ref. [46]
X1 X2 X3 X1 X2 X3 X1 X2 X3 X1 X2 X3
e=b ¼ 0 68.410 151.820 189.622 68.380 150.964 189.594 153.851 148.730 – 68.379 150.98 189.60
e=b ¼ 1:0 108.214 143.208 210.875 107.450 144.964 209.784 111.254 140.476 – – – –
a pffiffiffiffiffiffiffiffiffi
X ¼ xa2 qhD.

possesses both the advantages of DQ and the flexibility ing non-dimensional form
of the FEM.
@4U @4U @4U a2 @ 2 U
4
þ 2k2 2 2
þ k4 4
¼ Hx ð23Þ
@X @X @Y @Y D @X 2
4.2. Buckling analysis of thin, isotropic rectangular
plates where U is the dimensionless mode function of the
deflection, X ¼ x=a, Y ¼ y=b are the dimensionless
The governing differential equation of buckling of a coordinates, a and b are the dimensions of the plate
thin rectangular plate is given by: parallel to x-axis and y-axis, k ¼ a=b is ratio of the
plate edge length or aspect ratio, and Hx is the uniaxial
@4u @4u @4u 1 @2u compression load. D denotes the flexural rigidity of
þ 2 þ ¼ H x ð22Þ
@x4 @x2 @y2 @y4 D @x2 plates and it is given as D ¼ Eh3 =12ð1  m2 Þ, m is the
Poisson ratio, E is the modulus of elasticity of the plate
where u is the transverse displacement of the midsur- material, h is the uniform plate thickness. Eq. (23) can
face of the plate. Eq. (22) may be written in the follow- be given by applying the DQM as

X
Nx X
Nx Ny
X
2
Dik Ukj þ 2k Bjm Bik Ukm
k¼1 k¼1 m¼1

Ny
X Hx X
Nx
þ k4 Djk Uik ¼ a2 Bik Ukj ;
k¼1
D k¼1

for i ¼ 1; 2; . . . ;Nx and j ¼ 1; 2; . . . ;Ny ð24Þ

where Nx and Ny are the number of grid points along

the x- and y-directions, respectively, as shown in Fig. 1,
and Dik, Dik, Bik, Bjm represent the weighting coeffi-
cients of the fourth- and second-order derivatives along
x-and y-directions for the differential quadrature
approximation. Eq. (23) is a fourth-order partial differ-
ential equation, so we have to write two boundary con-
ditions at each edge for the stability analysis of plates.

4.2.1. Boundary conditions for rectangular plates

4.2.1.1. Four edges clamped (C–C–C–C). The bound-
Fig. 2. Circular plates and typical grid sampling points. ary conditions for a plate clamped on all four edges are
178 Ö. Civalek / Engineering Structures 26 (2004) 171–186

that the displacement and rotation must be zero on the increasing the number of grid points within the range
edge. 5 Nx ¼ Ny 9, the HDQ results monotonically
UðX ;0Þ ¼ UðX ;1Þ ¼ 0 and Uð0;Y Þ ¼ Uð1;Y Þ ¼ 0 approach the corresponding exact results. Type-II grid
points are used as the grid points. It should be noted
@U @U that the HDQ solution converges at a smaller grid size
ðX ;0Þ ¼ ðX ;1Þ ¼ 0
@Y @Y as compared to the DQ solution. Buckling loads
@U @U obtained for square plates are presented in Table 3
and ð0;Y Þ ¼ ð1;Y Þ ¼ 0
@X @X together with the exact solutions [35] and finite element
Applying the differential quadrature to these bound- solutions obtained by using four different grid numbers
ary conditions in each direction are presented in the table. The results
U1j ¼ UNj ¼ 0 and Ui1 ¼ UiN ¼ 0 ð25aÞ obtained from the finite element method are indicated
by FEM. Reasonably accurate results can be achieved
U1j ¼ UNj ¼ 0 and Ui1 ¼ UiN ¼ 0 ð25bÞ by using only 7  7 grid points for HDQ. The variation
X
Nx X
Nx of the error with the number of grid points was shown
A1k Ukj ¼ ANk Ukj ¼ 0 ð25cÞ in Fig. 3 for the HDQ and FE methods. The percent-
k¼1 k¼1 age of error had been reduced in parallel to the
Ny
X Ny
X increase of the grid points. In this figure, S–S–S–S and
A1k Uik ¼ ANk Uik ¼ 0; C–C–C–C supports are taken as boundary conditions.
k¼1 k¼1 The best solution is obtained for 9  9 grid sizes by
for i ¼ 1; 2; . . . ;Nx and j ¼ 2; 3; . . . ;Ny  1: ð25dÞ using the HDQ method. However, reasonably accurate
results can be achieved by using 13  13 grid points for
the FEM in this case. From the table, the convergence
4.2.1.2. Four edges simply supported (S–S–S–S). Dis- of the HDQ method is seen to be very good. It is also
placement and moment must be zero on the edge shown in this table that HDQ method produces better
UðX ;0Þ ¼ UðX ;1Þ ¼ 0 and Uð0;Y Þ ¼ Uð1;Y Þ ¼ 0 convergent solutions than the FEM when a similar
2 2
number of grid points are used.
@ U @ U In the case of the rectangular plate, the obtained
ðX ;0Þ ¼ ðX ;1Þ ¼ 0
@Y 2 @Y 2 results are presented for aspect ratios of k ¼ a=b ¼ 1=5,
@2U @2U 2/5, 4/5, 5/5. Two different type plate configurations
and 2
ð0;Y Þ ¼ ð1;Y Þ ¼ 0
@X @X 2 (C–C–C–S and C–S–S–S) are taken into consideration.
Applying the differential quadrature to these bound- Table 4 presents the non-dimensional buckling load for
ary conditions rectangular plates. For the harmonic differential quad-
U1j ¼ UNj ¼ 0 and Ui1 ¼ UiN ¼ 0 ð26aÞ rature method, the number of discrete points con-
sidered along the non-dimensional X-axis and Y-axis
U1j ¼ UNj ¼ 0 and Ui1 ¼ UiN ¼ 0 ð26bÞ was taken to be five and seven of these cases (i.e.,
X
Nx X
Nx Nx ¼ Ny ¼ 5; Nx ¼ Ny ¼ 7). These results are also
B1k Ukj ¼ BNk Ukj ¼ 0 ð26cÞ found by Wang et al. [43]. Wang had used five grid
k¼1 k¼1 points in each direction for this problem. Reasonably
Ny
X Ny
X accurate results can be achieved by using 9  9 grid
B1k Uik ¼ BNk Uik ¼ 0; points for the FEM in this case. For the 9  9 grid
k¼1 k¼1
for i ¼ 1; 2; . . . ;Nx and j ¼ 2; 3; . . . ;Ny  1: ð26dÞ
and

X Ny
Nx X
Aik Bjm Ukm ¼ 0 ð27Þ
k¼1 m¼1

at the corner. The other possible support conditions

can be obtained similar to the equations above. Table 2
summarizes the numerical results of non-dimensiona-
lized buckling loads by DQ and HDQ for cases of
square plates with four different support conditions.
As can be seen, the HDQ results compare very well
with the analytical solutions from references
[35] for only 9  9 grid points. It is observed that by Fig. 3. Percentage error with grid numbers for square plates.
 Ö. Civalek / Engineering Structures 26 (2004) 171–186 179

In the following, boundary conditions are given for

these four cases, respectively.

4.3.1. Clamped–simply supported (C–S)

W ¼ 0 and ðdW =dX Þ ¼ 0 at X ¼ 0
W ¼ 0 and ðd2 W =dX 2 Þ ¼ 0 at X ¼ 1

4.3.2. Clamped–clamped supported (C–S)

W ¼ 0 and dW =dX ¼ 0 at X ¼ 0
W ¼ 0 and dW =dX ¼ 0 at X ¼ 1

Fig. 4. Percentage error with grid numbers for column. 4.3.3. Simply supported–simply supported (S–S)
W ¼ 0 and d2 W =dX 2 ¼ 0 at X ¼ 0
sizes (8  8 mesh) in the FEM, the dimension of the
resulting system matrix is 147  147. However, only W ¼ 0 and d2 W =dX 2 ¼ 0 at X ¼ 1
7  7 grid points can produce quite accurate results by
HDQ. The results are obtained by solving an eigenva- 4.3.4. Clamped supported–free end (C–F)
lue problem of 9  9 matrix for HDQ. The computing
W ¼ 0 and ðdW =dX Þ ¼ 0 at X ¼ 0
time on a Pentium-II is less than 1 s for this case.
2 2 3
ðd W =dX ¼ 0Þ and ðd W =dX 3 ¼ 0Þ at X ¼ 1
4.3. Buckling of linear elastic columns The differential quadrature approximation to the
above equations has to be applied at each discrete
The non-dimensional governing differential equation point on the grid. For the case of (C–S) support con-
for buckling behavior of an elastic column is given [40] ditions,
by:
X
N
d4 W d2 W W1 ¼ 0 and A1j Wj ¼ 0 ð30aÞ
EI 4
þ PL2 ¼0 ð28Þ j¼1
dX dX 2
Differential quadrature form of this equation is X
N
WN ¼ 0 and BNj Wj ¼ 0 ð30bÞ
X
N X
N
j¼1
Dij Wj þ k Bij Wj ¼ 0; i ¼ 1; 2; 3; . . . ;N ð29Þ
j¼1 j¼1 For (C–C) support conditions,
where X ¼ x=L, W ¼ w=L and k ¼ PL2 =EI. X
N
W1 ¼ 0 and A1j Wj ¼ 0 ð31aÞ
Numerical applications have been performed for a
j¼1
linearly elastic column under four sets of different
boundary conditions, namely simply supported–simply X
N

supported (S–S), clamped–simply supported (C–S), WN ¼ 0 and ANj Wj ¼ 0 ð31bÞ

j¼1
clamped–clamped (C–C), and cantilever beam (C–F).

Fig. 5. (a) An actual arbitrary quadrilateral domain; (b) standard computational domain.
180 Ö. Civalek / Engineering Structures 26 (2004) 171–186

Fig. 6. (a) Curvilinear quadrilateral domain; (b) standard square domain.

For (S–S) boundary conditions, rewritten as:

X
N
X
N X
N
W1 ¼ 0 and B1j Wj ¼ 0 ð32aÞ Dij Wj þ k Bij Wj ¼ 0;
j¼1 j¼1 j¼1

X
N for i ¼ 3;4; . . . ; ðN  2Þ ð34Þ
WN ¼ 0 and BNj Wj ¼ 0 ð32bÞ
j¼1
Thus, the buckling load of column under a given
and for (C–F) boundary conditions axial load P can be found by solving above eigenvalue
equations with appropriate boundary conditions. For
X
N
W1 ¼ 0 and A1j Wj ¼ 0 ð33aÞ this case, adjacent-d points (Type-III) are used as
j¼1 the sampling grid points. Tables 5 and 6 summarize the
numerical results of buckling loads by HDQ for the
X
N X
N
BNj Wj ¼ 0 and CNj Wj ¼ 0 ð33bÞ case of linear elastic columns with four different
j¼1 j¼1 boundary conditions. It is shown in this table that the
HDQ results using five grid points are more accurate
The set of Eqs. (30), (31), (32) and (33) or other
than those of the DQ for seven grid points. The buck-
equations for beams having different boundary con- ling load by Chajes [41] using the finite difference (FD)
ditions are redundant, because for each column, four method is also presented in Table 5 for comparison.
equations in their supports and N equations in (29) are The results obtained from finite difference method are
given. In order to remove this redundancy, we have to indicated by FD. For the buckling load of columns,
drop certain equations such as i ¼ 1;2; ðN  1Þ and (N)
in the essence Eq. (29). Consequently, Eq. (29) can be

Fig. 7. Trapezoidal plate. Fig. 8. Eccentric sectorial plate.

 Ö. Civalek / Engineering Structures 26 (2004) 171–186 181

ing differential equations for skew plates under uniaxial

compression Fx along the x-direction and its differen-
tial quadrature form are given, respectively, as:

uxxxx  ð4kcoshÞuxxxy þ 2k2 ð1 þ 2cos2 hÞuxxyy

a2
 ð4k3 coshÞuxyyy þ k4 uyyyy ¼  Fx sin4 ðh=4Þux ð35Þ
D

Fig. 9. Simply supported annular plates.

X
Nx X
Nx Ny
X
Din unj  4kcosh Cin Ajm umn
N ¼ 9 grid points provide acceptable results with a n¼1 n¼1 m¼1

maximum discrepancy of 0.005% for the clamped– X

Nx Ny
X
clamped (C–C) boundary conditions and a maximum þ 2k2 ð1 þ 2cos2 hÞ Bin Bjm uim
discrepancy of 0.04% for clamped–free (C–F) support n¼1 m¼1
case for HDQ. In addition to this, if we use the DQ X
Nx Ny
X Ny
X
method with N ¼ 9 grid points, it provides acceptable  2k2 ð1 þ 2cos2 hÞ Ain Cjm umn þ k2 Djm umi
results with a maximum discrepancy of 0.07% for the n¼1 m¼1 m¼1
C–C support condition and 0.25% for C–F support X
Nx
a2
condition. From Table 6, one can conclude that for all ¼ Fx sin4 ðh=4Þ Ain umj ; i ¼ 1;2; . . . ;Nx
numerical methods, the solutions converge as the grid D n¼1
number is increased. A reasonably converged solution
and j ¼ 1;2; . . . ;Ny ð36Þ
may be achieved for 5 grids by the HDQ. In addition
to this, a reasonably converged solution may be
where h is the skew angle, k ¼ a=b is the aspect ratio, u
obtained for 9 grid points using FEM. The variation of
is the displacement in the z-direction, and D is the flex-
the error with the number of grid points is shown in
Fig. 4 for the DQ, HDQ, and FEM methods. The per- ural rigidity. Clamped support condition with movable
centage of error had been reduced in parallel to the edges is considered. In accordance with this condition,
increase of the grid points. In this figure (Fig. 4), S–S the plate is prevented from moving in the z-direction or
support case is taken as boundary conditions. The best rotating at the boundaries. In this case, both the deflec-
solution is obtained for N ¼ 7 grid points by using the tion and slope must vanish. The edges of the plate are,
HDQ method. however, free to move in the x–y plane. These support
conditions are given as:
4.4. Buckling analysis of skew plates @u
u ¼ 0 and ¼ 0 at x ¼ 0 and x¼1
@x
It is known that there are no closed-form solutions
for the buckling behavior of skew plates. Therefore,
@u
numerical methods must be utilized to solve the prob- u ¼ 0 and ¼0 at y ¼ 0 and y¼1
@y
lem. Consider a thin isotropic skew plate. The govern-
and their DQ form can be written as
X
Nx
u1 ¼ 0 and A1n unj ¼ 0; uN ¼ 0
n¼1
X
Nx
and ANn un ¼ 0 ð37aÞ
n¼1

Ny
X
u1 ¼ 0 and A1m umi ¼ 0; uN ¼ 0
m¼1
Ny
X
and ANm umi ¼ 0 ð37bÞ
m¼1

These boundary conditions are incorporated into

Fig. 10. Non-dimensional fundamental frequency of annular plates. Eq. (36). Since the first and last displacements are
182 Ö. Civalek / Engineering Structures 26 (2004) 171–186

known, Eq. (36) is rewritten for boundary conditions Using the chain rule, the first-order, and second-
Ny order derivatives of a function are given as:
X
Nx X
Nx X  
Din unj  4kcosh Cin Ajm umn ux u
n¼1 n¼1 m¼1 ¼ J11 n ð41Þ
uy ug
X
Nx Ny
X 8 9 8 9
2 2
þ 2k ð1 þ 2cos hÞ Bin Bjm uim < uxx = < unn = 
u
n¼1 m¼1 uyy ¼ J31 ugg  J31 J2 J11 n ð42Þ
: ; : ; ug
X
Nx Ny
X Ny
X 2uyx 2ung
 2k2 ð1 þ 2cos2 hÞ Ain Cjm umn þ k2 Djm umi
n¼1 m¼1 m¼1 where ni and gi are the coordinates of node i in the n–g
XNx plane, and J1, J2, and J3 are the Jacobian matrices.
a2
¼ Fx sin4 ðh=4Þ Ain umj ; These are expressed as follows:
D n¼1 2 3
i ¼ 2; . . . ;Nx  1 and j ¼ 2; . . . ;Ny  1 ð38Þ
 xnn ynn
xn y n 6 7
J1 ¼ ; J2 ¼ 4 xgg ygg 5;
Consequently, we solve the remaining eigenvalue xg yg
problem to obtain the natural frequencies. Numerical xng yng
2 2 2
3
results had been obtained for k ¼ 1. Four different xn yn x n yn
skew angles h are taken into consideration. The buck- 6 2 7
6
J3 ¼ 6 xg y2g x g yg 7
ð43Þ
ling coefficients for clamped skew plates are listed in 7
4 1 5
Table 7 together with those given by Wang et al. [42]. xn xg yn yg ðxn yg þ xg yn Þ
2
HDQ and DQ results are obtained for C–C–C–C sup-
port conditions using various numbers of grid points
for comparison. Reasonably accurate results can be 5.2. Curvilinear quadrilateral plates
achieved by using 9 grid points in HDQ for Type-I and
Furthermore, let the field of interest be a curvilinear
Type-II grid samplings in the related directions. Both 7
quadrilateral domain in the Cartesian x–y plane, as
and 1 grid points are used in HDQ for Type-II in each
shown in Fig. 6(a). The geometry of this plate can be
direction. Chebyshev–Gauss–Lobatto grid points
mapped into a square region in the natural n–g plane,
(Type-II) have been shown to be consistently better
as shown in Fig. 6(b). In this approach, the irregular
than any other choice under consideration. Hence, it
physical domain is transformed into a regular domain
can be concluded that by using same grid sizes, Type-II in a curvilinear coordinate system and, accordingly, the
grid solutions furnish higher accuracy for this problem. governing equation and boundary conditions are trans-
formed into relevant forms in the curvilinear coordi-
nate system. Then, all computations are based on the
5. HDQ method for irregular domains computational domain. Since the computational
domain is regular, the application of the differential
5.1. Straight-sided quadrilateral plates
quadrature method to irregular plates in the computa-
Consider an arbitrary straight-sided quadrilateral tional space is exactly the same as the application of
plate in the Cartesian x–y plane, as shown in Fig. 5(a). DQ method to regular plates in the physical domain.
The geometry of this plate can be mapped into a rec- In this case, interpolation or shape functions are given
tangular plate in the natural n–g plane, as shown in for grid i as:
Fig. 5(b). By employing the following transformation 1
equations, the physical domain is mapped into the Ui ðn;gÞ ¼ ð1 þ nni Þð1 þ ggi Þð  1 þ nni þ ggi Þ;
4
computational domain [45]: for i ¼ 1;3;5;7 ð44aÞ
X
N X
N
1
x¼ xi Ui ðn;gÞ and y¼ yi Ui ðn;gÞ ð39Þ Ui ðn;gÞ ¼ ð1  g2 Þð1 þ nni Þ; for i ¼ 4 and 8 ð44bÞ
i¼1 i¼1
2
1
where xi and yi are the coordinates of node i in the Ui ðn;gÞ ¼ ð1  n2 Þð1 þ ggi Þ; for i ¼ 2 and 6 ð44cÞ
2
physical domain, N is the number of grid points, and
Thus, the following equations are used for the coor-
Ui ðn;gÞ : i ¼ 1;2;3; . . . ;N are the interpolation or shape
dinate transformation:
functions. These are given for node i:
X
8 X
8
1 x¼ xi Ui ðn;gÞ and y¼ yi Ui ðn;gÞ ð45Þ
Ui ðn;gÞ ¼ ð1 þ nni Þð1 þ ggi Þ ð40Þ
4 i¼1 i¼1
 Ö. Civalek / Engineering Structures 26 (2004) 171–186 183

Bert and Malik give detailed information about (n–g) is given as:
irregular element for DQ [3]. In addition to this, more Nn Nn Ng Ng
X X X X
detailed formulation on the method of DQ for straight- Dnik Ukj þ 2k2 Bnjm Bgik Ukm þ k4 Dgjk Uik
sided quadrilateral domain can also be found in the k¼1 k¼1 m¼1 k¼1
paper by Liew and Han [51]. Details of the transform- Nn
Hx X
ation process of the DQ method as applied to straight- ¼ a2 Bn Ukj ; for i ¼ 1; 2; . . . ;Nn
sided quadrilateral and curvilinear domains are not D k¼1 ik
elaborated here but can be found elsewhere in the and j ¼ 1; 2; . . . ;Ng ð52Þ
literature [3,45,49,51]. In the numerical applications,
The considered boundary conditions are incorpor-
simply supported (SS) and clamped (C) boundary con-
ated into Eq. (52) similar to the other examples. The
ditions are considered. At any point of either type of buckling loads of a symmetric trapezoidal plate are
support, the deflection is zero. presented in Table 8 with three different boundary con-
U ¼0 ð46Þ ditions. One of the most important study of this prob-
For the simply supported edge, the normal moment lem is that of Wang et al. [49]. So, these results are
must be zero, chosen to compare the accuracy of HDQ results. It is
shown that in this table, HDQ results using 11  11
@2U @2U grid points are more accurate than those of the DQ.
ðcos2 a þ msin2 aÞ 2
þ ðsin2 a þ mcos2 aÞ
@X @Y 2 The buckling load by Civalek [11] using the FEM
2
@ U method is also presented in Table 8 for comparison.
þ ½2ð1  mÞcosa sina ¼0 ð47Þ
@X @Y Fifteen grid points are used in each direction for FEM.
and at the clamped edge, there is no rotation normal to A reasonably converged solution may be achieved for
the edge, thus: 11 grids by HDQ. In addition to this, a reasonably
converged solution may be obtained for 15 grid points
@U @U using FEM.
cosa þ sina ¼0 ð48Þ
@X @Y
where a is the angle between the normal to the plate 5.2.2. Free vibration analysis of eccentric sectorial plate
boundary and the x-axis. In the DQ form, Eqs. (46)– Consider free vibration analysis of eccentric sectorial
(48) are given as: plate as shown in Fig. 8. The governing differential
equations of free vibration of a thin plate are given in a
Ui ¼ 0 ð49Þ non-dimensional form as:
Nn
X Ng
X @4U @4U 4
4@ U
ðcos2 a þ msin2 aÞ Bnmj Uj þ ðsin2 a þ mcos2 aÞ Bgnj Uj þ 2k 2
þ k ¼ X2 U ð53Þ
@X 4 @X 2 @Y 2 @Y 4
j¼1 j¼1
Nn Ng
where U is the dimensionless mode function of the
X X
þ ½2ð1  mÞcosa sina Anmj Agnj Uj ¼0 ð50Þ vibration, X is the dimensionless frequency and is given
j¼1 j¼1 as X ¼ x2 a4 qh=D, X ¼ x=a, Y ¼ y=b are the dimen-
sionless coordinates, a and b are the dimensions of the
Nn
X Ng
X plate as parallel to x-axis and y-axis, k ¼ a=b is the
cosa Anmj þ sina Agnj Uj ¼ 0 ð51Þ
ratio of the plate edge length or aspect ratio, q is
j¼1 j¼1
the mass density of using material for plate, h is the
For the straight-sided quadrilateral and curvilinear uniform plate thickness, x is the natural frequency,
domains, the sampling points are taken as and D denotes the flexural rigidity of plates and is

  
  
1 i1 1 j1 given as D ¼ Eh3 =12ð1  m2 Þ, m is the Poisson ratio, E is
ni ¼ 1  cos p ; gj ¼ 1  cos p the modulus of elasticity of the plate material. Eq. (53)
2 Nn  1 2 Ng  1
can be given by applying the DQM as:
i ¼ 1;2;.. .;Nn and j ¼ 1;2;.. .;Ng
Nn
X Nn
X Ng
X
Dnik Ukj þ 2k2 Bnik Bgjm Ukm
k¼1 k¼1 m¼1
5.2.1. Buckling of trapezoidal plate Ng
X
Let us consider a trapezoidal plate as shown in þ k4 Dgjk Uik ¼ X2 Uij ; for i ¼ 1;2; . . . ;Nn
Fig. 7. Since the geometry of this plate can be mapped k¼1
into a rectangular plate in the natural n–g plane, we and j ¼ 1;2; . . . ;Ng ð54Þ
can use the governing equation for buckling of thin
plates under uniform compressive force as given in where Dnik Dgjk Bnik Bgjm represent the weighting coeffi-
Eq. (22). Its DQ form in the computational domain cients of the fourth- and second-order derivatives along
184 Ö. Civalek / Engineering Structures 26 (2004) 171–186
n- and g-directions for the differential quadrature
approximation. For the simply supported all over the
four edges plate 15  15 grid points can produce quite
accurate results. It is concluded that the HDQ solution
converge at lesser grid points when compared with the
DQ solution. The polar finite differences (FD) results
of the vibration for first two frequencies are also given
in Table 9 for the purpose of comparison. For the FD
solution, 19  19 grid size is used. The required time on
a Pentium-II with 64 MB RAM is less than 1 s for
HDQ and DQ. But, the required computing time is 3 s
for the FD solution. It should be noted that the HDQ
solution converges at a smaller grid size as compared
to the DQ and FD solution.
5.3. Free vibration analysis of annular plates
As the last example, we solve the free vibration of
simply supported circular plates with a concentric cir-
cular hole, the so-called annular plates (Fig. 9). The
inner and outer radii of an annular plate are denoted
by b and a, respectively. The governing differential
equation of circular annular plates undergoing axisym-
metric free vibration and its DQ form is given by:
2 1 1 buckling of elastic columns and plates under uniform
URRRR þ URRR  2 URR þ 3 UR  W2 U ¼ 0 ð55Þ
R R R axial compression, and vibration of plates has been
XN
2 X N
1 XN
1 XN
Dij Uj þ Cij Uj  2 Bij Uj þ 3 Aij Uj
j¼1
Ri j¼1 Ri j¼1 Ri j¼1
 W2 Ui ¼ 0; i ¼ 1;2; . . . ; ðN  2Þ
where R ¼ r=a, W is the non-dimensional frequency
pffiffiffiffiffiffiffiffiffiffiffiffi
and is given by W ¼ xa2 qh=D, D is the flexural rigid-
ity of the annular plate, h is the thickness, q is the den-
sity of the plate material, and x is the circular
frequency. Considering the simply supported annular
plate, that is,
U ¼0 at R ¼ 1
and
@2U m @U
þ at R ¼ 1 ð58Þ
@R2 R @R
Notice that we only keep the discretized equations
for i ¼ 1 to (N2) in Eq. (56) because there are two
boundary conditions at R ¼ 1 point. The non-dimen-
sional fundamental frequencies of annular plates with
simply supported boundary conditions obtained by
HDQ and DQ are shown in Fig. 10 for N ¼ 11. This
figure demonstrates the influence of non-dimensional
radial coordinate b/a on the non-dimensional funda-
mental frequencies. The non-dimensional fundamental
frequency given by Leissa [16] is also presented in this
figure for comparison. It is shown that the results com-
pare very well with the solution of Leissa [16]. As
observed from Fig. 10, the obtained fundamental fre-
quencies found by using 11 uniformly spaced grid
points are very accurate.
The HDQ method was found to require less than
two seconds of CPU time for almost all cases on a
standard personal computer (Pentium-II processor hav-
ing 64 MB RAM). An attractive advantage of the
HDQ method is that it can produce acceptable accu-
racy of the numerical results with very few grid points
in the solution domain and therefore can be very useful
for rapid evaluation in engineering design.
6. Concluding remarks
In the past 10 years, many researchers have been
interested in various types of DQ methods. These
include the DQ method, generalized differential quad-
rature (GDQ), differential quadrature element (DQE)
method, and the HDQ. Harmonic differential quadra-
ture was recently proposed and there are only a few
papers on this novel kind of differential quadrature. In
the present study, the HDQ method was introduced to
study the buckling analysis of thin, isotropic plates
under various support conditions. The problem of the
successfully solved using the HDQ and DQ approa-
ches. The conventional small deflection theory is used
in the study with the governing differential equations
ð56Þ transformed into a typical eigenvalue equation by the
HDQ formulation. The method of HDQ that was used
in the paper proposes a very simple algebraic formula
to determine the connection weighting coefficients
required by differential quadrature approximation
without restricting the choice of mesh grids. The
known boundary conditions are easily incorporated in
the HDQ as well as in the other type of DQ. A good
comparative accuracy of DQ and HDQ methods for
ð57Þ
static and vibration analysis of plates is presented by
Liew et al. [50,56]. More detailed information can be
found in these references. The discretizing and pro-
gramming procedures are straightforward and easy.
Several test examples for different plate shapes have
been selected to demonstrate the convergence proper-
ties, accuracy and simplicity in numerical implemen-
tation of the HDQ procedures. This has verified the
accuracy and applicability of the HDQ method to the
class of problem considered in this study. Furthermore,
in the numerical examples in the scope of the study, the
results obtained with HDQ method are more accurate
than the values calculated by using finite elements and
finite differences and a coarser grid is used. But it is
impossible to state that the HDQ method is superior to
finite elements in all cases or problems by only depend-
ing on this study. Each method has its own advantages
 Ö. Civalek / Engineering Structures 26 (2004) 171–186
and application areas. Finite element method is still an
effective way especially in systems with complex
geometry and load conditions or applications with
non-linear behavior and has many successful applica-
tions.
Acknowledgements
The author is grateful to the referees whose thor-
ough reviews and precise comments were very helpful
in the revision of the original manuscript of the present
paper. The author also would like to express his
appreciation to Mr. Ali K. Baltacioğlu for his assist-
ance during the computer programming.
References
[1] Bellman R, Casti J. Differential quadrature and long-term inte-
gration. J Math Anal Appl 1971;34:235–8.
[2] Bellman R, Kashef BG, Casti J. Differential quadrature: a tech-
nique for the rapid solution of nonlinear partial differential
equation. J Comput Phys 1972;10:40–52.
[3] Bert CW, Malik M. The differential quadrature method for
irregular domains and application to plate vibration. Int J Mech
Sci 1996;38(6):589–606.
[4] Bert CW, Jang SK, Striz AG. Two new approximate methods
for analyzing free vibration of structural components. AIAA J
1987;26(5):612–8.
[5] Bert CW, Wang Z, Striz AG. Differential quadrature for static
and free vibration analysis of anisotropic plates. Int J Solids
Struct 1993;30(13):1737–44.
[6] Bert CW, Malik M. Free vibration analysis of tapered rectangu-
lar plates by differential quadrature method: a semi-analytical
approach. J Sound Vib 1996;190(1):41–63.
[7] Bert CW, Wang Z, Striz AG. Convergence of the DQ method in
the analysis of an isotropic plates. J Sound Vib 1994;170(1):140–
4.
[8] Bert CW, Malik M. Differential quadrature method in computa-
tional mechanics: a review. Appl Mech Rev 1996;49(1):1–28.
[9] Bert CW, Wang Z, Striz AG. Static and free vibrational analysis
of beams and plates by differential quadrature method. Acta
Mech 1994;102:11–24.
[10] Björck A, Pereyra V. Solution of Vandermonde system of equa-
tions. Math Comput 1970;24:893–903.
[11] Civalek Ö. Finite element analysis of plates and shells. Elazığ:
Fırat University; 1998 [in Turkish].
[12] Civalek Ö. Static, dynamic and buckling analysis of elastic bars
using differential quadrature, XVI. National Technical Engineer-
ing Symposium 2001. Ankara: METU; 2001 [Paper no. 5].
[13] Civalek Ö, Çatal HH. Dynamic analysis of one and two dimen-
sional structures by the method of generalized differential quad-
rature. Turkish Bull Eng 2002;417:39–46.
[14] Civalek Ö. Differential quadrature (DQ) for static and dynamic
analysis of structures. Elazığ: Fırat University; 2002 [in Turkish].
[15] Civalek Ö, Çatal HH. Static analysis of thin isotropic circular
plates using harmonic differential quadrature method. T J Eng
Environ Sci, TUBITAK 2002 [Evaluation by referee].
[16] Leissa AW. Vibration of plates. NASA, SP-160; 1969.
185
[17] Du H, Lim MK, Lin RM. Application of generalized differential
quadrature method to structural problems. Int J Numer Meth-
ods Eng 1994;37:1881–96.
[18] Shu C, Xue H. Explicit computations of weighting coefficients in the
harmonic differential quadrature. J Sound Vib 1997;204(3):549–55.
[19] Striz AG, Wang X, Bert CW. Harmonic differential quadrature
method and applications to analysis of structural components.
Acta Mech 1995;111:85–94.
[20] Civan F, Sliepcevich CM. Application of differential quadrature
to transport process. J Math Anal Appl 1983;93:206–21.
[21] Civan F, Sliepcevich CM. Differential quadrature for multi
dimensional problems. J Math Anal Appl 1984;101:423–43.
[22] Civan F, Sliepcevich CM. Solution of the Poisson equation by
differential quadrature. Int J Numer Methods Eng 1983;19:
711–24.
[23] Civan F, Sliepcevich CM. On the solution of the Thomas–Fermi
equation by differential quadrature. J Comput Phys
1984;56:343–8.
[24] Du H, Lim MK, Lin RM. Application of generalized differential
quadrature method to vibration analysis. J Sound Vib
1995;181(2):279–93.
[25] Du H, Liew KM, Lim MK. Generalized differential quadrature
method for buckling analysis. J Eng Mech ASCE 1996;22(2):
95–100.
[26] Farsa J, Kukreti AR, Bert CW. Fundamental frequency analysis
of laminated rectangular plates by differential quadrature
method. Int J Numer Methods Eng 1993;36:2341–56.
[27] Hamming RW. Numerical methods for scientists and engineers.
New York: McGraw-Hill; 1973.
[28] Jang SK, Bert CW, Striz AG. Application of differential quadra-
ture to static analysis of structural components. Int J Numer
Methods Eng 1989;28:561–77.
[29] Mingle JO. The method of differential quadrature for transient
nonlinear diffusion. J Math Anal Appl 1977;60:559–69.
[30] Sherbourne AN, Pandey MD. Differential quadrature method in
the buckling analysis of beams and composite plates. Comput
Struct 1991;40(4):903–13.
[31] Shu C, Richards BE. Application of generalized differential
quadrature to solve two-dimensional incompressible Navier–
Stokes equations. Int J Numer Methods Fluids 1992;15:791–8.
[32] Shu C, Chew YT. On the equivalence of generalized differential
quadrature and highest order finite difference scheme. Comput
Methods Appl Mech Eng 1998;155:249–60.
[33] Quan JR, Chang CT. New insights in solving distributed system
equations by the quadrature method—I analysis. Comput Chem
Eng 1989;13(7):779–88.
[34] Striz AG, Chen W, Bert CW. Static analysis of structures by the
quadrature element method. Int J Solids Struct 1994;31(20):
2807–18.
[35] Iyengar NGR. Structural stability of columns and plates.
Chichester: Ellis Horwood; 1988.
[36] Liew KM, Teo TM. Three dimensional vibration analysis of rec-
tangular plates based on differential quadrature method. J Sound
Vib 1999;220(4):577–99.
[37] Chen WL, Striz AG, Bert CW. High-accuracy plane stress and
plate elements in the quadrature element method. Int J Solids
Struct 2000;37:627–47.
[38] Wu TY, Liu GR. The generalized differential quadrature rule for
fourth-order differential equations. Int J Numer Methods Eng
2001;50:1907–29.
[39] Striz AG, Chen WL, Bert CW. Free vibration of plates by the
high accuracy quadrature element method. J Sound Vib
1997;202(5):689–702.
[40] Timoshenko SP, Gere JM. Theory of elastic stability, 2nd ed.
Tokyo: McGraw-Hill; 1959.
[41] Chajes A. Principles of structural stability theory. New Jersey:
Prentice-Hall; 1974.
186 Ö. Civalek / Engineering Structures 26 (2004) 171–186
[42] Wang X, Striz AG, Bert CW. Buckling and vibration analysis of
skew plates by the differential quadrature method. AIAA J
1994;32(4):886–9.
[43] Wang X, Bert CW, Striz AG. Differential quadrature analysis of
deflection, buckling and free vibration of beams and rectangular
plates. Comput Struct 1993;48(3):473–9.
[44] Liew KM, Teo TM. Modeling via differential quadrature
method: three-dimensional solutions for rectangular plates.
Comput Methods Appl Mech Eng 1998;159:369–81.
[45] Zienkiewicz OC. The finite element method in engineering
science, 3rd ed. London: McGraw-Hill; 1977.
[46] Kim CS, Dickinson SM. On the free, transverse vibration of
annular and circular, thin sectorial plates subject to certain com-
plicating effects. J Sound Vib 1989;134:407–15.
[47] Fried I, Schimitt KH. Numerical results from the application of
gradients iterative techniques to the finite element vibration and
stability of skew plates. J Aeronaut 1972;76:166–9.
[48] York CB, Williams FW. Buckling analysis of skew plate assem-
blies, classical plate theory results incorporating Lagrange multi-
pliers. Comput Struct 1995;56:625–35.
[49] Wang CM, Liew KM, Alwis WAM. Buckling of skew plates and
corner conditions for simply supported edges. J Eng Mech
1991;118:651–62.
[50] Liew KM, Teo TM, Han JB. Comparative accuracy of DQ
and HDQ methods for three-dimensional vibration analysis
of rectangular plates. Int J Numer Methods Eng 1999;45:
1831–48.
[51] Liew KM, Han JB. A four-node differential quadrature method
for straight-sided quadrilateral reissner/mindlin plates. Commun
Numer Methods Eng 1997;13:73–81.
[52] Ugural AC. Stress in plates and shells, 2nd ed. McGraw Hill
Companies; 1999.
[53] Shu C, Chen W, Xue H, Du H. Numerical study of grid distri-
bution effect on accuracy of DQ analysis of beams and plates by
error estimation of derivative approximation. Int J Numer
Methods Eng 2001;51:159–79.
[54] Reddy JN. Improved finite difference analysis of bending the rec-
tangular plates. Comput Struct 1979;10:431–8.
[55] Szilard R. Theory and analysis of plates: classical and numerical
methods. Englewood Cliffs (NJ): Prentice-Hall; 1974.
[56] Liew KM, Teo TM, Han JB. Three-dimensional static solutions
of rectangular plates by variant differential quadrature method.
Int J Mech Sci 2001;43:1611–28.
[57] Liew KM, Huang YQ, Reddy JN. A hybrid moving least
squares and differential quadrature (MLSDQ) meshfree method.
Int J Comput Eng Sci 2002;3(1):1–12.