Jtam.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 46, 1, pp. 123-139, Warsaw 2008 METHODS BASED ON THE DIFFERENTIAL QUADRATURE IN VIBRATION ANALYSIS OF PLATES Artur Krowiak Institute of Computing Science, Cracow University of Technology, Cracow, Poland e-mail: krowiak@mech.pk.edu.pl The paper deals with the methods based on the differential quadrature and their application to the free vibration analysis of plates. The spline-based dif- ferential quadratureMethod (SDQM) is presented as an alternative to known methods based on the interpolation polynomial (PDQM). The SDQMuses a polynomial piecewise function to approximate the wanted solution of a go- verning equation. The way of determining the spline functions as well as the way of computing weighting coefficients for the method are presented in the paper. Then the SDQM is applied to determine natural frequencies of plates. The influence of the spline degree, number of nodes and grid point distribu- tion on the accuracy, convergence and stability is investigated in an example. All results are comparedwith values obtained by the conventional differential quadraturemethod (PDQM). Keywords:Differential quadraturemethod, spline interpolation, freevibration 1. Introduction Most engineering problems are described by partial differential equations. It is very difficult, if possible at all, to find closed-form solutions to them. A great increase in the computational power in recent years enables one to use numerical methods for solving very complex physical tasks. It is the reason that stimulates the development of known methods and the search for new, more efficient ones. Most of these methods rely on the conversion of a phy- sical model of a system from continuous to discrete one. The approximate solution is searched at some special points in the domain. Among currently used discretisation methods, the most popular are: finite difference method, finite element method and finite volume method. These discretisation tech- niques use a low degree interpolation to determine function values at a node 124 A. Krowiak and, therefore, they are numbered among low order methods. They allow one to achieve high accuracy by using a large number of nodes. In many prac- tical engineering problems, the numerical solution is required only at a few discrete points in the domain. The modal analysis could be an example. By discretizing apartial differential equation, an algebraic eigenvalue problemcan be obtained. Next, the latter is solved giving approximation values of natural frequencies (or/andnaturalmodes) of a continuous system.Thenumber of ob- tained frequencies corresponds to the number of nodes of the imposed mesh. Among these frequencies, only a few first are interesting from the practical point of view. However, in order to achieve high accuracy for these values by low order methods, one has to use a large number of nodes. It requires much virtual storage and computational effort. This drawback can be overcome by using so-called global methods, which take muchmore information than only local neighborhood of a node to approximate the function. Itmakes the rate of convergence of thesemethodsmuch higher than low ordermethods and allows one to achieve very accurate results with only a few discrete points. The differential quadraturemethod (DQM) falls under this category. The- oretical foundations of the DQMwere given by Bellman and Casti (1971). In recent years, onehas beenable tonotice significant development of themethod and its application in many fields of mechanics. Somemain works are quoted by Bert and Malik (1996). Higher efficiency of the DQM for linear problems than the finite element and finite difference method was proved by Bert et al. (1988, 1993),Wang andBert (1993), Malik andCiven (1994). In addition, the DQM ismuchmore effective for nonlinear problems comparingwith low order methods, which was presented by Bellman and Casti (1971), Bellman et al. (1972), Bert et al. (1989), Feng and Bert (1992), Malik and Civen (1994). The idea of the DQM is based on the approximation of spatial derivatives of a function at each node by a linear combination of function values at all discrete points in the domain along the coordinate lines. Considering a two- dimensional case, where the domain of the function f(x,y) is a rectangular area, partial derivatives with respect to spatial variables at each point (xi,yi) can be expressed in the method as ∂nf ∂xn ∣ ∣ ∣ x=xi y= yj = Nx ∑ k=1 a (n) k (xi)f(xk,yj)= Nx ∑ k=1 a (n) ik f(xk,yj) (1.1) ∂mf ∂ym ∣ ∣ ∣ x=xi y= yj = Ny ∑ k=1 b (m) k (yj)f(xi,yk)= Ny ∑ k=1 b (m) jk f(xi,yk) Methods based on the differential quadrature... 125 for i= 1, . . . ,Nx, j = 1, . . . ,Ny, where Nx, Ny are the numbers of nodes in the x and y directions, and a (n) ik , b (m) jk are the weighting coefficients for the nth and mth order derivatives with respect to appropriate variables. In order to approximate the mixed derivatives, the weighting coefficient matrices for appropriate derivatives with respect to x and y have to bemultiplied. Using formula (1.1), one has to collocate a governing equation at each grid point and imposeboundary conditions in order to reduce the partial differential equation to a system of algebraic or ordinary differential equations, respectively of the case under consideration. Themain stage of the method is the determination of the weighting coefficients. 2. Polynomial and Spline-based differential quadrature method Values of the weighting coefficients depend on the way the solution is ap- proximated (selection of trial functions), and they influence the accuracy, co- nvergence and stability of the method. The interpolation polynomial is the most often used to approximate the solution in the differential quadrature method (PDQM). Using Lagrange base functions, the rth order derivatives of the interpolation polynomial at the ith discrete point can be expressed as follows P(r)(xi)= N ∑ j=1 l (r) j (xi)fj (2.1) where lj(x) denotes Lagrange’s base polynomial of the (N −1)th degree. It is easy to notice that the derivatives of the appropriate Lagrange base functions are the weighting coefficients for the polynomial based differential quadrature method. Theanalytical formulas for the coefficients of thefirst order derivativewere given by Quan and Chang (1989) and have following forms a (1) ij = 1 xj −xi N ∏ k=1,k 6=i,j xi−xk xj −xk for j 6= i (2.2) a (1) ii = N ∑ k=1,k 6=i 1 xi−xk 126 A. Krowiak Due todifficultieswithderivation of explicit formulas for higher order derivati- ves of Lagrange’s base functions, Shu andRichards (1992) gave recurrence re- lationship (2.3) to overcome the problem, i,j =1,2, . . . ,N, r=2,3, . . . ,N−1 a (r) ij = r ( a (1) ij a (r−1) ii − a (r−1) ij xi−xj ) for i 6= j (2.3) a (r) ii =− N ∑ j=1,j 6=i a (r) ij There are also otherways based on theFourier series expansion orB-spline functions to approximate the wanted solution in the DQM. But especially PDQM allows one to achieve very high accuracy by using only a few nodes. However, this method is very sensitive to the type of imposed mesh and the number of nodes.When a uniform grid distribution is imposed and too many sampling points are used, the results are inaccurate or the method is not convergent. The computational instability of the method is the result from the manner of the interpolation polynomial. This polynomial oscillates much when its degree N−1 (N –numberofnodes) is toohigh,particularlywhen the uniform grid is imposed. In most problems of computational mechanics, the only way to estimate the accuracy of results is to carry out the computation again using larger number of nodes. Therefore, the computational stability should be the important feature of a numerical method. To improve stability of theDQM,Zhong(2004) usedquinticB-spline as trial functions todetermine theweighting coefficients. It considerably improves stability of themethodbut the convergence rate is less comparing to the PDQM. In the present work, another way of the approximation of the solution is shown. In the presentedmethod, the solution of a partial differential equation is approximated by the nth degree polynomial piecewise function (SDQM). The approximation was first applied in the DQM by Krowiak (2006). The work, that has been done so far, indicates that using a suitably high spli- ne degree the convergence rate of the method is similar to PDQM and the computational stability is much better even when using uniformly spaced nodes. When the degree n of the spline is odd then the function is approximated in the following way f(x)≈{si(x), x∈ [xi,xi+1], i=1, . . . ,N −1} (2.4) where N is the number of nodes. Methods based on the differential quadrature... 127 When n is even, the auxiliary spline knots are defined at themidpoints of the nodes in order to be able to meet the conditions for the determination of the spline function z1 =x1 zi+1 = 1 2 [xi+xi+1] i=1, . . . ,N−1 zN+1 =xN (2.5) and the interpolation function has the form f(x)≈{si(x), x∈ [zi,zi+1], i=1, . . . ,N} (2.6) In Equations (2.4) and (2.6), the ith spline section is defined as si(x)= n ∑ j=0 cijx j (2.7) The coefficients cij in Equation (2.7) are determined using the interpolation conditions, the continuity conditions of the derivatives at the nodes and the so-called natural end conditions at the domain boundaries. In the case of an odd degree, these conditions have the following form si(xi)= fi si(xi+1)= fi+1 i=1, . . . ,N−1 s (k) i (xi+1)= s (k) i+1(xi+1) i=1, . . . ,N−2, k=1, . . . ,n−1 s (k) 1 (x1)= 0 s (k) N−1(xN)= 0 k= n+1 2 , . . . ,n−1 (2.8) Their number is (n+1)(N −1), which corresponds to the number of coeffi- cients cij. In the case of an even spline degree, where the number of unknown coefficients is (n+1)N, the auxiliary knots are also used in interpolation conditions (2.9) and the continuity conditions for derivatives (2.10) si(xi)= fi i=1, . . . ,N si(zi+1)= si+1(zi+1) i=1, . . . ,N−1 (2.9) and s (k) i (zi+1)= s (k) i+1(zi+1) i=1, . . . ,N−1, k=1, . . . ,n−1 (2.10) To complete the set of equations, the natural end conditions are introduced at the end points s (k) 1 (x1)= 0 s (k) N (xN)= 0 k= n 2 , . . . ,n−1 (2.11) 128 A. Krowiak It insures that in every case of an even spline degree the number of conditions matches the number of coefficients cij. These coefficients depend on the grid distribution and unknown function values according to the formula cij = N ∑ k=1 Cijk(x1, . . . ,xN)fk i=1, . . . ,N, j=0, . . . ,n (2.12) where N =N−1 when n is odd and N =N when n is even. Taking advantage of the symbolic computation system, one can easily de- termine the weighting coefficients for the differential quadrature method ba- sed on the approximation given by Eqs. (2.4) and (2.6). Bymanipulating the expressions in which the unknown function values f(xi) are defined as sym- bols fi, it is possible to computederivatives of anarbitrarydegree r (r <n−1) for function (2.4) or (2.6) at the ith discrete point f(r)(xi)≈ s (r) i (xi) i=1, . . . ,N s (r) i−1(xi) i=N when n is odd (2.13) In Eq. (2.13), s (r) i (xi) and s (r) N−1(xN) have the forms s (r) i (xi)= n ∑ j=r [( N ∑ k=1 Cijkfk ) x j−r i j ∏ l=j−r+1 l ] i=1, . . . ,N (2.14) s (r) N−1(xN)= n ∑ j=r [( N ∑ k=1 CN−1jkfk ) x j−r N j ∏ l=j−r+1 l ] where the advantage has been taken of Equation (2.12). Equations (2.14) can be expressed in other forms s (r) i (xi)= N ∑ k=1 [ n ∑ j=r ( Cijkx j−r i j ∏ l=j−r+1 l )] fk i=1, . . . ,N (2.15) s (r) N−1(xN)= N ∑ k=1 [ n ∑ j=r ( CN−1jkx j−r N j ∏ l=j−r+1 l )] fk ComparingEq. (1.1) to (2.15), it is clear that theexpressions in squarebrackets in the above formula are theweighting coefficients for the rth order derivative Methods based on the differential quadrature... 129 in the differential quadraturemethod based on the spline functions which can be written as follows a (r) ik = n ∑ j=r ( Cijkx j−r i j ∏ l=j−r+1 l ) i=1, . . . ,N (2.16) a (r) Nk = n ∑ j=r ( CN−1jkx j−r N j ∏ l=j−r+1 l ) when n is odd 3. Free vibration analysis of rectangular plates Plates belong to basic structural elements in civil andmechanical engineering and, therefore, they are often subjects of static and dynamic research. Many numericalmethodshave beenused in dynamical analysis of plateswith various boundary conditions. The obtained results have been used in real structures or as the comparison of accuracy and convergence for applied methods. The conventional differential quadraturemethodhas alsobeenapplied to thevibra- tion analysis of plates (Shu andDu, 1997; Wang and Bert, 1993). The results show that the convergence rate of the PDQM is very high. Very accurate re- sults can be obtained applying a grid with points densely concentrated near boundaries. The use of an arbitrary grid, for example a uniform one, makes the results inaccurate or the method not convergent. This drawback can be overcome by using themethod presented in this paper. Since the results and conclusions coming from application of the PDQM to vibration analysis of plates are common and well know, the SDQM has been applied to check its versatility in using various grid point distributions. In the paper, the SDQMhas been applied to determine natural frequencies of a thin, isotropic, rectangular plate. The dimensionless governing equation for free vibration of the plate is as follows ∂4W ∂X4 +2λ2 ∂4W ∂X2∂Y 2 +λ4 ∂4W ∂Y 4 =Ω2W (3.1) In the above equation W denotes dimensionless mode shape function, X = x/a and Y = y/b are dimensionless coordinates, a and b are leng- ths of the plate edges, λ= a/b is the aspect ratio and Ω is the dimensionless frequency. Its relation with the dimensional circular frequency is following Ω=ωa2 √ ρ D (3.2) 130 A. Krowiak where ρ is the density of the plate material and D=Eh3/[12(1−ν2)] is the flexural rigidity (E, ν, h are Young’s modulus, Poisson’s ratio and the plate thickness, respectively). Calculations have beendone for squareplates (λ=1) with following configurations of boundary conditions: (a) SS-F-SS-F — for X =0 and X =1 W =0 ∂2W ∂X2 =0 (3.3) — for Y =0 and Y =1 λ2 ∂2W ∂Y 2 +ν ∂2W ∂X2 =0 λ2 ∂3W ∂Y 3 +(2−ν) ∂3W ∂X2∂Y =0 (3.4) (b) C-F-SS-F — for X =0 W =0 ∂W ∂X =0 (3.5) — for X =1 W =0 ∂2W ∂X2 =0 (3.6) — for Y =0 and Y =1 λ2 ∂2W ∂Y 2 +ν ∂2W ∂X2 =0 λ2 ∂3W ∂Y 3 +(2−ν) ∂3W ∂X2∂Y =0 (3.7) (c) C-C-C-C — for X =0 and X =1 W =0 ∂W ∂X =0 (3.8) — for Y =0 and Y =1 W =0 ∂W ∂Y =0 (3.9) whereC denotes the clamped edge, SS – simply supported edge andF – free edge. Chebyshev-Gauss-Lobatto grid (3.10) and a uniform grid have been used to discretize the area 0¬X ¬ 1, 0¬Y ¬ 1. In both cases, the same number of points has been used in the X and Y directions Xi =Yi = 1 2 [ 1− cos ( i−1 N−1 π )] i=1, . . . ,N (3.10) Methods based on the differential quadrature... 131 Equation (3.1),written inadiscrete form following fromtheuseof themethod, is as follows N ∑ k=1 a (4) ik Wkj +2λ 2 N ∑ k1=1 N ∑ k2=1 a (2) ik1 a (2) jk2 Wk1k2 +λ 4 N ∑ k=1 a (4) jk Wik =Ω 2Wij (3.11) where a (r) ik denote theweighting coefficients for the rth order derivative in the SDQM and Wij are unknown nodal values. Identical coefficients a (r) ik appro- ximate the derivatives in both directions due to the use of the same grid and number of nodes in these directions. The implementation of boundary conditions is the important stage of the method. Various approaches to this problem were presented by Shu and Du (1997). In the present study, the boundary conditions have been used to cal- culate function values at the boundary points and points adjacent to the bo- undaries as a linear combination of the values at the interior points. Details are presented for theSS-F-SS-F plate configuration. According to theDQM, the discrete form of the boundary conditions described by Eq. (3.3) and (3.4) is following W1j =0 j=1, . . . ,N N ∑ k=1 a (2) 1kWkj =0 j=2, . . . ,N −1 (3.12) WNj =0 j=1, . . . ,N N ∑ k=1 a (2) Nk Wkj =0 j=2, . . . ,N −1 (3.13) λ2 N ∑ k=1 a (2) 1kWik+ν N ∑ k=1 a (2) ik Wk1 =0 i=2, . . . ,N−1 λ2 N ∑ k=1 a (3) 1kWik+(2−ν) N ∑ k1=1 N ∑ k2=1 a (2) ik1 a (1) 1k2 Wk1k2 =0 i=3, . . . ,N−2 (3.14) λ2 N ∑ k=1 a (2) Nk Wik+ν N ∑ k=1 a (2) ik WkN =0 i=2, . . . ,N −1 λ2 N ∑ k=1 a (3) Nk Wik+(2−ν) N ∑ k1=1 N ∑ k2=1 a (2) ik1 a (1) Nk2 Wk1k2 =0 i=3, . . . ,N −2 (3.15) 132 A. Krowiak The function values at the plate edges parallel to the Y axis are defined by the first equation from equation sets (3.12) and (3.13). The rest from (3.12) and (3.13) are used to determine function values at points adjacent to the boundaries in the Y direction as a linear combination of values at the interior points.Thediscretisation of plate edges parallel to the X axis is similar.Using function values at the boundary points and points adjacent to the boundaries in Eq. (3.1) one have to solve a standard eigenvalue problem to obtain natural frequencies of the plate. Tabela 1. SS-F-SS-F plate – mesh of the Chebyshev-Gauss-Lobatto type N Ω1 Ω2 Ω3 Ω4 Ω5 n=11 16 9.634 16.153 36.778 38.987 46.877 (0.031%) (0.112%) (0.142%) (0.108%) (0.297%) 20 9.632 16.139 36.740 38.957 46.778 (0.010%) (0.025%) (0.038%) (0.031%) (0.086%) n=14 16 9.632 16.139 36.741 38.961 46.789 (0.010%) (0.025%) (0.041%) (0.041%) (0.109%) 20 9.631 16.136 36.729 38.948 46.748 (0.000%) (0.006%) (0.008%) (0.008%) (0.021%) PDQM 16 9.631 16.135 36.726 38.945 46.738 (0.000%) (0.000%) (0.000%) (0.000%) (0.000%) 20 9.631 16.134 36.725 38.945 46.738 (0.000%) (−0.006%) (−0.003%) (0.000%) (0.000%) The results obtained by the DQM based on spline functions of various degrees n are presented in Table 1, where the grid described by Equation (3.10) has been applied, and inTable 2, where the uniformgrid has been used. The tables contain also the results obtained by the conventional differential quadrature method (PDQM). When the uniform grid is used, the PDQM shows computational instability. The application of too many nodes makes the results very inaccurate. The percentage relative error δ= ΩSDQM −Ωreference Ωreference ·100% (3.16) Methods based on the differential quadrature... 133 Tabela 2. SS-F-SS-F plate – uniform grid N Ω1 Ω2 Ω3 Ω4 Ω5 n=11 16 9.666 16.349 37.271 39.201 47.964 (0.363%) (1.326%) (1.484%) (0.657%) (2.623%) 20 9.651 16.254 37.039 39.147 47.508 (0.208%) (0 .738%) (0.852%) (0.519%) (1.647%) 24 9.643 16.208 36.923 39.088 47.242 (0.125%) (0.452%) (0.536%) (0.367%) (1.078%) n=14 16 9.642 16.203 36.867 39.228 47.455 (0.114%) (0.421%) (0 .384%) (0 .727%) (1.534%) 20 9.636 16.165 36.809 39.072 47.084 (0.052%) (0.186%) (0.226%) (0.326%) (0.740%) 24 9.633 16.150 36.775 39.010 46.924 (0.021%) (0.093%) (0.133%) (0.167%) (0.398%) PDQM 10 9.638 16.157 37.740 38.925 47.029 (0.067%) (0.136%) (2.761%) (−0.051%) (0.623%) 12 0 0 0 0 9.627 16 0 0 0 0 0 has been calculated for the frequencies on the basis of Leissa’s (1973) re- sults, which are exact for the SS-F-SS-F plate. The remaining two sets of Leissa’s results (C-F-SS-F, C-C-C-C) were obtained by the Rayleigh-Ritz method with beam functions for the displacement, taking nine terms into account. The presented results show that the convergence rate of the SDQM is satisfactory when the weighting coefficients determined from the high degree spline functions are used. The accuracy can be improved by increasing the number of grid points. Analysing the results for the C-F-SS-F plate configuration (Table 3 and Table 4), one can notice that the error of calculated frequencies does not de- creasemonotonicallywhen the splinedegree is higher and the number of nodes increases. It is especially noticeable when non-uniform grid (3.6) is applied to insure high rate of convergence. The calculations for another plate with clam- ped edges (Table 5 and Table 6) confirm that results obtained by the SDQM and PDQM converge to lower values than the reference results obtained by 134 A. Krowiak Leissa (1973). It seems that the results obtained by differential quadrature methods are closer to exact values since the approximate solutions from the Rayleigh-Ritz method are upper bounds on the exact values. It should be no- ted that in the case of the SS-F-SS-F plate, where the reference results are the exact solutions, the values from the differential quadrature methods are in very good agreement. Tabela 3. C-F-SS-F plate – mesh of the Chebyshev-Gauss-Lobatto type N Ω1 Ω2 Ω3 Ω4 Ω5 n=8 16 15.434 21.682 43.083 50.286 58.753 (0.975%) (4.881%) (8.317%) (1.118%) (3.773%) 20 15.344 21.259 41.763 49.985 57.833 (0.386%) (2.835%) (4.998%) (0.513%) (2.148%) 24 15.297 21.043 41.099 49.821 57.346 (0.079%) (1.790%) (3.329%) (0.183%) (1.288%) n=11 16 15.220 20.677 39.885 49.588 56.632 (−0.425%) (0.019%) (0.277%) (−0.286%) (0.026%) 20 15.201 20.617 39.793 49.497 56.404 (−0.550%) (−0.271%) (0.045%) (−0.469%) (−0.376%) 24 15.194 20.596 39.760 49.467 56.328 (−0.595%) (−0.372%) (−0.038%) (−0.529%) (−0.510%) n=14 16 15.209 20.644 39.831 49.545 56.514 (−0.497%) (−0.140%) (0.141%) (−0.372%) (−0.182%) 20 15.197 20.604 39.771 49.477 56.353 (−0.576%) (−0.334%) (−0.010%) (−0.509%) (−0.466%) 24 15.193 20.591 39.750 49.458 56.304 (−0.602%) (−0.397%) (−0.063%) (−0.547%) (−0.553%) PDQM 16 15.193 20.602 39.776 49.545 56.515 (−0.602%) (−0.343%) (0.003%) (−0.372%) (−0.180%) 20 15.190 20.585 39.748 49.484 56.372 (−0.622%) (−0.426%) (−0.068%) (−0.495%) (−0.433%) 24 15.191 20.582 39.738 49.464 56.320 (−0.615%) (−0.440%) (−0.093%) (−0.535%) (−0.525%) Methods based on the differential quadrature... 135 Tabela 4. C-F-SS-F plate – uniform grid N Ω1 Ω2 Ω3 Ω4 Ω5 n=8 16 16.114 25.226 52.439 55.282 65.627 (5.424%) (22.024%) (31.839%) (11.164%) (15.914%) 20 15.901 24.082 50.942 51.994 63.494 (4.030%) (16.490%) (28.075%) (4.553%) (12.147%) 24 15.773 23.395 48.721 51.416 62.185 (3.193%) (13.167%) (22.492%) (3.390%) (9.834%) n=11 16 15.408 21.223 40.706 50.204 58.501 (0.805%) (2.660%) (2.341%) (0.953%) (3.328%) 20 15.333 21.002 40.359 50.011 57.777 (0.314%) (1.591%) (1.468%) (0.565%) (2.049%) 24 15.290 20.877 40.171 49.862 57.337 (0.033%) (0.987%) (0.996%) (0.265%) (1.271%) n=14 16 15.359 21.023 40.252 50.221 57.951 (0.484%) (1.693%) (1.199%) (0.987%) (2.356%) 20 15.295 20.865 40.087 49.905 57.290 (0.065%) (0.929%) (0.784%) (0.352%) (1.189%) 24 15.260 20.778 39.987 49.749 56.953 (−0.164%) (0.508%) (0.533%) (0.038%) (0.593%) PDQM 10 15.492 21.204 41.353 50.585 58.559 (1.354%) (2.569%) (3.967%) (1.719%) (3.430%) 12 0 0 15.406 21.051 27.976 16 0 0 0 0 0 4. Concluding remarks The problem of free vibration analysis of plates has been undertaken by the author inorder to examine the computational stability of theproposedmethod on various grid distributions. The presented results show that the convergence rate of the SDQM is high when a high degree spline is used to approximate the solution. The accuracy can be improved by increasing the number of grid points without concern for losing stability of themethod.Unlike thePDQM, themethod is not limited to 136 A. Krowiak Tabela 5. C-C-C-C plate – mesh of the Chebyshev-Gauss-Lobatto type N Ω1 Ω2 Ω3 Ω4 Ω5 n=8 12 36.021 73.498 73.498 108.549 131.895 (0.081%) (0.116%) (0.116%) (0.258%) (0.194%) 15 35.995 73.422 73.422 108.308 131.653 (0.008%) (0.012%) (0.012%) (0.035%) (0.001%) 20 35.987 73.399 73.399 108.233 131.593 (−0.014%) (−0.019%) (−0.019%) (−0.034%) (−0.036%) 25 35.986 73.395 73.395 108.221 131.584 (−0.017%) (−0.025%) (−0.025%) (−0.045%) (−0.043%) 30 35.985 73.394 73.394 108.218 131.582 (−0.019%) (−0.026%) (−0.026%) (−0.048%) (−0.044%) n=11 12 35.990 73.417 73.417 108.288 131.659 (−0.006%) (0.005%) (0.005%) (0.017%) (0.014%) 15 35.986 73.397 73.397 108.228 131.595 (−0.017%) (−0.022%) (−0.022%) (−0.039%) (−0.034%) 20 35.985 73.394 73.394 108.217 131.582 (−0.019%) (−0.026%) (−0.026%) (−0.049%) (−0.044%) 25 35.985 73.394 73.394 108.217 131.581 (−0.019%) (−0.026%) (−0.026%) (−0.049%) (−0.045%) 30 35.985 73.394 73.394 108.217 131.581 (−0.019%) (−0.026%) (−0.026%) (−0.049%) (−0.045%) PDQM 12 35.986 73.399 73.399 108.230 131.418 (−0.017%) (−0.019%) (−0.019%) (−0.037%) (−0.169%) 15 35.985 73.394 73.394 108.217 131.580 (−0.019%) (−0.026%) (−0.026%) (−0.049%) (−0.046%) 20 35.985 73.394 73.394 108.217 131.581 (−0.019%) (−0.026%) (−0.026%) (−0.049%) (−0.045%) 25 35.985 73.394 73.394 108.216 131.581 (−0.019%) (−0.026%) (−0.026%) (−0.050%) (−0.045%) 30 35.985 73.394 73.394 108.217 131.581 (−0.019%) (−0.026%) (−0.026%) (−0.049%) (−0.045%) Methods based on the differential quadrature... 137 Tabela 6. C-C-C-C plate – uniform grid N Ω1 Ω2 Ω3 Ω4 Ω5 n=8 12 36.577 75.061 75.061 112.497 136.144 (1.625%) (2.245%) (2.245%) (3.904%) (3.421%) 15 36.308 74.271 74.271 110.668 133.710 (0.878%) (1.169%) (1.169%) (2.215%) (1.572%) 20 36.132 73.788 73.788 109.407 132.445 (0.389%) (0.511%) (0.511%) (1.050%) (0.612%) 25 36.065 73.608 73.608 108.888 132.029 (0.203%) (0.266%) (0.266%) (0.571%) (0.296%) 30 36.033 73.523 73.523 108.633 131.847 (0.114%) (0.150%) (0.150%) (0.335%) (0.157%) n=11 12 36.109 73.680 73.680 109.419 128.917 (0.325%) (0.364%) (0.364%) (1.061%) (−2.069%) 15 36.037 73.602 73.602 108.887 131.528 (0.125%) (0.257%) (0.257%) (0.570%) (−0.085%) 20 36.002 73.475 73.475 108.466 131.795 (0.028%) (0.084%) (0.084%) (0.181%) (0.118%) 25 35.992 73.428 73.428 108.323 131.699 (0%) (0.020%) (0.020%) (0.049%) (0.045%) 30 35.988 73.410 73.410 108.267 131.642 (−0.011%) (−0.004%) (−0.004%) (−0.003%) (0.002%) PDQM 10 36.021 72.996 72.996 108.348 130.891 (0.081%) (−0.568%) (−0.568%) (0.072%) (−0.569%) 12 35.992 73.474 73.474 78.331 78.331 (0%) (0.083%) (0.083%) (−27.651%) (−40.496%) 15 0 0 0 0 0 special typesof grids.Various typesof gridpointdistributionscanbeapplied in theSDQMgivingdifferences only in the convergence rate of themethod.These make the SDQMmore versatile than the PDQMand in author’s opinion it is a good starting point to applying themethod tomore challenging engineering problems, even nonlinear ones. In the latter, the high rate of convergence and stability of the SDQM should be a great advantage. 138 A. Krowiak References 1. Bellman R., Casti J., 1971, Differential quadrature and long term integra- tion, Journal of Mathematical Analysis and Application, 34, 235-238 2. BellmanR., Kashef B.G., Casti J., 1972,Differential quadrature: A tech- nique for the rapid solution of nonlinear partial differential equations, Journal of Computational Physics, 10, 40-52 3. Bert C.W., Jang S.K., Striz A.G., 1988, Two newmethods for analyzing free vibration of structure components,AIAA Journal, 26, 612-6182 4. Bert C.W., Malik M., 1996, Differential quadraturemethod in computatio- nal mechanics,Applied Mechanics Review, 49, 1-28 5. Bert C.W., Striz A.G., Jang S.K., 1989, Nonlinear bending analysis of orthotropic rectangular plates by the method of differential quadrature,Com- putational Mechanics, 5, 217-226 6. Bert C.W., Wang X., Striz A.G., 1993, Differential quadrature for static and free vibration analysis of anisotropic plates, International Journal of Solid Structures, 30, 1737-1744 7. FengY., BertC.W., 1992,Application of the quadraturemethod to flexural vibration analysis of a geometrically nonlinear beam, Nonlinear Dynamics, 3, 13-18 8. KrowiakA., 2006,Symbolic computing in spline-baseddifferential quadrature method,Communications in NumericalMethods in Engineering, 22, 1097-1107 9. Leissa A.W., 1973,The free vibration of rectangular plates, Journal of Sound and Vibration, 31, 257–293 10. Malik M., Civan F., 1994, A comparative study of differential quadratu- re and cubature method vis-à-vis some conventional techniques in context of convection-diffusion-reaction problems, Chemical Engineering Science, 50, 531-547 11. Quan J.R., Chang C.T., 1989, New insight in solving distributed system equations by the quadrature methods I,Computer and Chemical Engineering, 13, 779-788 12. Shu C., 2000, Differential Quadrature and its Application in Engineering, Springer-Verlag, London 13. Shu C., Du H., 1997, Implementation of clamped and simply supported bo- undary conditions in GDQ free vibration analysis of beams and plates, Inter- national Journal of Solids and Structures, 34, 819-835 14. Shu C., Richards B.E., 1992, Application of generalized differential quadra- ture to solve two-dimensional incompressibleNanier-Stokes equations, Interna- tional Journal for Numerical Methods in Fluids, 15, 791-798 Methods based on the differential quadrature... 139 15. WangX., BertC.W., 1993,A new approach in applying differential quadra- ture to static and free vibration analysis of beams and plates, Journal of Sound and Vibration, 162, 566-572 16. Zhong H., 2004, Spline-based differential quadrature for fourth order diffe- rential equations and its application to Kirchhoff plates,Applied Mathematical Modelling, 28, 353-366 Metody oparte na kwadraturach różniczkowych w zastosowaniu do drgań płyt Streszczenie Praca dotyczy metod opartych na kwadraturach różniczkowych i ich aplikacji do zagadnienia drgań własnych płyt. W pracy, jako alternatywę do znanych me- tod kwadratur różniczkowych, opartych na wielomianie interpolacyjnym (PDQM), przedstawionometodę bazującą na funkcjach sklejanych (SDQM). W SDQM poszu- kiwane rozwiązanie przybliżane jest funkcją wielomianową, przedziałami zmienną.W pracy przedstawiono sposób wyznaczenia takiej funkcji interpolacyjnej, jak również sposób obliczenia współczynnikówwagowych, używanychwmetodzie kwadratur róż- niczkowych. Następnie SDQM użyto do wyznaczenia częstości drgań własnych płyt, gdzie analizowano wpływ stopnia wielomianu, liczby węzłów i ich rozmieszczenia na zbieżność, dokładność i stabilność metody. Otrzymane rezultaty porównano z wy- nikami uzyskanymi przy pomocy konwencjonalnej metody kwadratur różniczkowych (PDQM). Manuscript received February 21, 2007; accepted for print April 4, 2007