Journal of Sound and Vibration (1989) 131(2), 249-258
ANTICLASTIC CURVATURE CORR_ECTION OF A SQUARE
PLATE FINITE ELEMENT APPLIED TO THE TRANSVERSE VIBRATION OF FLAT PLATES B. DOWNS Department of Mechanical Engineering, Loughborough University of Technology, Loughborough, LE 11 3 TlJ, England (Received 23 May 1988, and in revised form 24 October 1988) Commencing with a conforming square plate finite element, based on the product of beam functions, anticlastic curvature considerations are applied to develop a modified element which satisfies more closely the requirement of the static biharmonic equation throughout the plate The modification results in some loss of conformity and derives from theory which is only approximate. The element is used to calculate the first nine transverse vibration frequencies of a thin square plate under eight different sets of edge constraints with 5x 5 modelling. For six cases comparisons are available with exact solutions and, overall, the modified element enables vibration frequencies to be predicted with considerably improved accuracy. 1. INTRODUCTION Dynamic discretization [ 11, in which dynamic displacement functions are used to generate an equivalent mass matrix, overcomes certain shortcomings of the consistent mass matrix which is derived from static displacement functions. When applied to beam systems, use of the equivalent mass matrix enables extremely accurate vibration analysis to be made even when the beams are of asymmetric section and have both taper and twist. The dynamic displacement functions are readily obtainable by repeated application of Stodola iteration, commencing with the corresponding static displacement functions. A first attempt [2] at using this method to obtain the equivalent mass matrix for the transverse deflection of a thin rectangular plate element has revealed serious shortcomings. The vibration analysis results are far less accurate than those for beam vibration. Two distinct sets of static deflection functions are used to obtain equivalent mass matrices for the plate. In the first set a non-conforming function is used [3], which fully satisfies the static biharmonic equation but permits kinking at inter-element boundaries. This kinking results in some stiffness reduction for the system and hence a tendency to underestimate vibration frequencies. This tendency is counteracted by the undershoot of the consistent mass below the true dynamic mass, which results in an overestimate of vibration frequency. The partial cancellation of these two errors enables better vibration predictions to be made by using this non-conforming element than by use of the conforming element referred to subsequently. As dynamic discretization would tend to the elimination of only the mass error, leaving unaltered the kinking at element interfaces, no further consideration is given to the non-conforming element. The second set of static deflection functions [4] satisfies both deflection and slope continuity across element interfaces and designates a conforming element or BffiNER element. This element does not satisfy the static biharmonic equation throughout the 249 0022-460X/89/110249+ 10 %03.00/O @ 1989 Academic Press Limited 250 B. DOWNS plate, but offers the possibility of embodying a correction so that the biharmonic equation is better satisfied. Such a corrected element might enable better vibration prediction when using dynamic discretization, and an examination is now made of one method of correc- tion. The BOGNER element comprises the product of raw beam functions and takes no account of the anticlastic curvature effects which influence plate deflections. The correction procedure adopted makes some allowance for anticlastic curvature and, for the sake of simplicity, the analysis and application is restricted to square plate elements and the use of the consistent mass matrix. 2. THEORY The corners of a square plate of side a are located at (-a/2, -a/2); (-a/2, a/2), (u/2, a/2) and (u/2, -u/2) with the origin of the axes located at the centre of the plate. Twelve degrees of freedom, on which the mass and stiffness matrices are based, comprise s, aslay and -as/ax at each of the four comers in sequence, where s denotes deflection normal to the plate. For static conditions, the basic plate equation [5] requires that the biharmonic expression equates to zero at all points throughout the plate, i.e., a4s/ax4+2 a4s/ax2 ay2+a4s/ay4= 0, (1) or, expressed in dimensionless parameters 6 = x/a, n = y/u, a4s/at4+2 a4s/a~2a~2+a4s/a7)4=0, (2) with the plate comers located at (-l/2, -l/2); (-l/2, l/2), etc. A typical raw beam function product is given by where s =f(5)f(n), (3) f([)=u+b~+c~2+d[3, f(77)=e+f17+gn2+hn3, (495) and the expanded values of equation (3) for s1 to s 12, which correspond to unit displace- ment (or rotation) on degrees of freedom 1 to 12 respectively, are presented in Appendix 1 of reference [2]. Partial differentiation of equation (3) gives a2s/at2 =f(S).W, a20n2 =f(W(n). (67) To consider profiles in the 77 direction only, anticlastic effects will introduce some correction to the curvature as/a# because of the transverse curvature a2s/at2. Correction is likewise required to the curvature a2s/at2 because of the transverse curvature a2s/an2. Any correction which is made to the curvature a2s/aq2 will induce a further correction to the curvature a2s/at2, so that a more accurate correction to a2s/aq2 will involve both the initial curvature a2s/av2 and the initial curvature a2s/ae2. Likewise, a correction to a2s/a[* should also involve both the initial curvatures. While the Poisson ratio may be used to relate strains in orthogonal directions where lateral movement is free and not influenced by surrounding layers of material, it cannot be used for thin sections where practically no transverse curvature occurs except near the edges. Anticlastic effects for a thin plate therefore imply that a correction to a2s/aq2 is an unknown function of a2s/a.f2 and a2s/an2. Likewise, a correction to a2s/af2 is also an unknown, but different, function of the same variables. ANTICLASTIC INFLUENCE ON VIBRATION To proceed, it is necessary to make simplifying assumptions, the correction to &/a~* = A a*s/a~*+ B a*s/aT* and correction to a*s/at* = C a*s/aq*+ D a*s/at* where A, B, C and D are constants. 251 one used here being (8) (9) A and C are due to primary correction while B and D arise from secondary correction. Upon neglecting the effects of secondary correction and writing (a2s/an2)correction = k,f( t)f( n), this requires a profile correction in the n direction of S = V(5) II f(n) dn drl, (IO) or S,, = k,(2c+6dS)(AI + B1n+en2/2+fq3/6+gn4/12+hn5/20), (II) the integration constants A, and B1 being obtained by setting S, =0 at n = -l/2 and at n = l/2. The corresponding profile correction in the 5 direction is S~=k2(C1+D1~+a~2/2+b~3/6+c~/12+d~5/20)(2g+6h~); (12) the values C, and D, are likewise obtained by setting S, = 0 at 5 = -l/2 and at 5 = l/2. Upon combining the two profile corrections, the deflection correction becomes S, = S&l 9 or S,=k(2c+6d~)(C,+D,~+a~2/2+b~3/6-tc~4/12+d~5/20) x(2g+6h~)(A,+B,~+eq2/2+_f~3/6+g~4/12+h~5/20), (13) with the constant k = k, k2 yet to be determined. S, is zero along all four edges of the square and therefore aS,/ag=aS,/a~ =0 at all four comers. The addition of the correction S, to a particular raw beam product function leaves unaltered the deflections on all 12 degrees of freedom. This property simplifies the subsequent evaluation of the corrected mass and stiffness matrices, even though the correction S, is not fully conforming. Upon adding in the correction, the displacement function becomes s=f(5)f(n)+S,, (14) and the unknown constant k of equation (13) is obtained by minimizing the integral over the area of the plate of (a4s/@+2 a4s/a~*an2+a4s/an4)*. The k values, together with the percentage reduction of the above integral resulting from including the correction profile S,, are given in the Appendix, together with the 6 and 71 polynomials of equation (13). 3. APPLICATION AND RESULTS Stiffness and consistent mass matrices corresponding to the s values of equation (14) are readily obtained by the application of standard integration procedures [3]. The square plate is subdivided into five equal strips in both the x and y directions for vibration analysis. The 5 x 5 mesh provides 108 degrees of freedom for the whole plate without external boundary constraints, and standard matrix vibration analysis predicts the first 252 B. DOWNS nine vibration frequencies for eight combinations of boundary conditions. For six of these, where two opposite edges are simply supported, exact mathematical solutions are available [6] and direct comparison with these exact values is made in Tables l-6. Some values obtained by other methods are also included in the tables, which are explained as follows. Table 1 is related to the plate which is simply supported on all four edges (S-S-S-S) and includes frequencies obtained by 5 x 5 modelling with the non-conforming and the conforming functions. Although the dimensionless frequencies are independent of the Poisson ratio, the values resulting from the anticlastic correction correspond to v = 0.3. Identical methods of frequency evaluation provide the values in Table 2 for S-F-S-F boundary conditions, but in all cases values correspond to v = 0.3 since with one or more free edges the dimensionless frequencies are dependent on V. In Table 3 for S-C-S-C boundary conditions, two sets of evaluations for the anticlastic corrected shape functions are given with values of u = 0.2 and v = 0.3, respectively. Frequency comparisons in this table and in Tables 4, 5 and 6 are limited to exact values and those obtained using the TABLE 1 Dimensionless frequency parameters A = oa2m for S-S-S-S square plate, based on 5 x 5 modelling and consistent mass, for (a) anticlastic corrected element (v = O-3); (b) uncorrected conforming element and (c) non-conforming element Frequency Exact [6] (4 (b) (cl 19.739 19.351 20.326 19.347 49.348 48.921 50.612 47,963 49.348 48.921 50,612 47.963 78.957 79.676 82.637 73.850 98.696 98.699 101.18 96.467 98.696 98.699 101.18 96467 128.30 130.59 135.07 118.68 128.30 130*59 135.07 118.68 167.78 170.27 173.67 159.11 TABLE 2 Dimensionless frequency parameters A for S-F-S-F square plate, based on 5 x 5 modelling, consistent mass analysis with v = 0.3, for (a) anticlastic corrected element; (b) uncorrected conforming element and (c) non-conforming element Frequency Exact [6] (4 (b) (c) 9.6314 9.286 9.645 9.685 16.135 16.454 16.842 16.135 36.726 37.109 38.099 36.078 38.945 38.008 39.068 39.395 46.738 46.911 48.122 46.780 70.740 72.357 74443 68.527 75.283 75-377 77.242 73.501 87.987 87.240 88.867 89.532 96*041 96.892 98.866 %-371 ANTICLASTIC INFLUENCE ON VIBRATION 253 TABLE 3 Dimensionless frequency parameters h for S-C-S-C square plate, based on 5 x 5 modelling, consistent mass for (a) anticlastic corrected element (v = O-2); (b) anticlastic corrected element (v = 0.3) and (c) uncorrected conforming element Frequency Exact [6] (a) v=O.2 (b) v=Oe3 (c) A, 28.951 28.574 28.465 29.597 A2 54.743 54.399 54.229 56.284 A3 69.327 68.825 68,642 70.734 A4 94.585 95.099 94.795 98.424 A5 102.22 102.14 101.87 104.86 h6 129.10 129.18 128.89 132.15 Al 140.20 141.62 141.26 146.71 AS 154.78 155.94 155.57 161.08 A9 170.35 172.40 172.01 176.04 TABLE 4 Dimensionless frequency parameters A for S-S-S-C squareplate, based on 5 x 5 modelling and consistent mass, for (a) anticlastic corrected element (v =0*3) and (b) uncorrected conforming element Frequency Exact [6] (a) (b) 23.646 23.141 24.269 51.674 51.105 53.079 58.646 58dlOO 59.977 86.134 86501 89.946 100.27 99.972 102.85 113.23 113.03 115.99 133.79 135.34 140.52 140.85 142.33 147.52 168.96 170.87 174.78 TABLE 5 Dimensionless frequency parameters h for S-S-S-F square plate, based on 5 x 5 modelling and consistent mass analysis with v = 0.3, for (a) anticlastic corrected element and (b) uncorrected conforming eZement Frequency Exact [6] (a) (b) A, 11.684 11.433 11.987 A2 27.756 27.731 28.770 A3 41.197 40.363 41.666 A4 59.066 59.413 61.527 A5 61.861 61.689 63.501 A6 90.294 89.653 91.660 A7 94.484 96.130 99.470 A8 108.92 110.16 113.51 A9 115.69 115w 118.59 254 B. DOWNS TABLE 6 Dimensionless frequency parameters h for S-C-S-F square plate, based on 5 x 5 modelling, consistent mass analysis with v = O-3, for (a) anticlastic corrected element; (b) fully conforming anticlastic corrected element and (b) uncorrected conforming element Frequency Exact [6] (a) (h) (c) 12.687 33.065 41.702 63.015 72.398 90.611 103.16 111.90 131.43 12.553 13.085 12.997 33.086 33.806 34.122 40.934 42.319 42.238 63.503 65.259 65.671 72.170 73.453 74.098 90@09 92.142 92.081 10470 107.23, 108.19 113.16 116.12 116.62 131.49 133.50 134.54 conforming beam product (BOGNER) functions. For Table 4, corresponding to S-S-S-C boundaries, the dimensionless frequencies, as in Table 3, are independent of v. However, the values for the anticlastic corrected functions correspond to v = 0.3. Evaluations for the S-S-S-F case, the results of which appear in Table 5, all correspond to v = 0.3. Similarly, v = 0.3 for the S-C-S-F results in Table 6 which includes two evaluations for the anticlastic correction method. The second of these is based on the addition of two additional terms with unknown coefficients, one in t6, the other in r7 in equation (12) and similarly for r) in equation (11). These two unknown coefficients are obtained by setting slopes to zero at 5 = -l/2 and at [= l/2. This method forces the anticlastic correction of equation (13) to be fully conforming in both deflection and slope at element interfaces. Two other sets of boundary conditions were selected for presentation. The fully clamped boundary (C-C-C-C) forms the basis for a large amount of experimental investigation (e.g., [7]) and hence analytical study (e.g., [8]) for purposes of comparison. Table 7 permits comparison of anticlastic correction values (v = 0.3) with Rayleigh-Ritz val _ 2s [6,8] and BOGNER values. Investigation of the completely free plate (F-F-F-F) reveals that, with 5 x 5 modelling, the three zero frequencies are not correctly predicted. The influence of interfacing on the artificial internal constraint introduced by anticlastic correction may be examined from the inclusion in Table 8 of evaluations based on progressively increasing subdivision of the plate from undivided whole plate (1 x 1) modelling to 5 x 5 modelling. This table includes comparisons with non-conforming and BOGNER analysis, and also with some Rayleigh-Ritz predictions and values obtained by superposition [9]. 4. DISCUSSION The correction procedure outlined in the theory lacks mathematical exactitude since it examines variation in one parameter independently of the other and vice versa, although the controlling partial differential equation establishes an interdependence between the 6 and 7 variables. Some errors must therefore be introduced by this procedure and it is not surprising that those dimensionless frequencies which should be independent of v prove to have some dependence on the Poisson ratio. Fortunately, this error is small, as may be seen from Table 3, which reveals that changing v from 0.2 to 0.3 results in ANTICLASTIC INFLUENCE ON VIBRATION 255 TABLE 7 Dimensionless frequency parameters A for a fully clamped (C- C-C-C) square plate based on 5 x 5 modelling and consistent mass for (c) anticlastic corrected element (Y = 0.3) and (d) uncorrected conforming element; reference values are (a) 36- term Rayleigh-Ritz evaluations [6] and (b) independent 36- term Rayleigh - Ritz eoaluations [ 81 Frequency (a) [61 (b) PI (c) (4 35.992 35.985 35446 36.694 73.413 73.394 72639 74.93 1 73.413 73.394 72643 74.93 1 108.27 108.22 107.883 111.98 131.64 131.58 131.10 134.60 132.24 132.20 131.85 135.31 165.00 164*&t 171.04 165.00 164+36 171.04 210.52 210.35 215.59 TABLE 8 Dimensionless frequency parameters A for a completely free (F-F-F-F) square plate based on consistent mass and Y = 0.3. (c) anticlastic corrected element for jive types of modelling ranging from whole plate to 5 x 5; (d) uncorrected conforming element and 5 x 5 modelling; (e) non-conforming element with 5 x 5 modelling; reference values are (a) Rayleigh-Ritz and (b) superposition of exact solutions (c) Frequency (a) [6] (b) [9] .l x 1 2x2 3x3 4x4 5x5 (d) (e) A4 A5 A6 A7 A8 A9 A LO A 11 A 12 o-0 o-0 o-0 13.489 19.789 24.432 35.024 35.024 61.526 61.526 o-0 0.119 0.0 0.298 0.0 0.299 13468 13.665 19.596 22.892 24.292 30.146 34.800 39.676 34.800 39.706 61.080 91.882 61.080 92.079 92.079 11740 0442 0.997 1.776 2.781 o-0 0.0 0.718 1.323 2.140 3.177 0.0 0.0 0.723 1.335 2.159 3.204 0.0 0.0 14.364 14.202 14.211 14.326 14.262 13+4K 19.637 19.634 19.601 19.636 196o!J 19643 23.836 23.828 23.719 23.680 24.298 24.374 35.686 36.158 35.851 35.694 36.214 34.749 35.689 36.160 35.852 35.694 36.214 34.749 63.362 60.612 60.603 60.235 61.361 61.619 69.636 60.616 60.607 60.240 61.361 61.619 69.658 67.999 66.753 66.056 67.355 62.913 81.154 70.268 71.028 70.619 71.370 69.605 frequency predictions which differ at the most by 0.4% and on average by less than 0.3%. Table 8 reveals that another error, the failure to correctly predict zero frequencies, becomes greater as the number of interfaces in the modelling is increased. This error may be expected to influence lower frequency values also. Any justification for the anticlastic correction method depends on whether or not non-zero frequencies are predicted with greater accuracy than when they are derived from conforming shape functions to which anticlastic correction has not been applied. 256 B. DOWNS Tables l-6 are therefore invaluable since they enable comparisons to be made with exact values for both the corrected and uncorrected cases. In all the tables the nearest value to exact, or other reference values, is displayed in bold type. Table 1 shows that, for all nine frequencies, correction produces significant improvement in predicted values over the uncorrected BOGNER values, and also values which are significantly superior to those obtaining from the non-conforming shape functions with the exception of the first frequency where only marginal superiority is shown. In Table 2 the first and fourth frequencies are predicted with less accuracy when correction is applied although the other seven frequency predictions are improved. Comparison with predictions based on non-conforming functions shows that use of anticlastic correction produces inferior results. Only in the case of frequencies 3, 6, 7 and 8 does correction provide more accurate prediction. In Table 3, with v = 0.3, correction produces remarkably superior frequency prediction and the same is true for v = 0.2. Tables 4, 5 and 6, with the exception of the third frequency in Table 5, all demonstrate superior vibration frequency predictions when correction is applied. Overall, it may be seen that quite remarkable improvements in predicting the first nine frequencies are obtained by the method of anticlastic correction described. For the 54 vibration modes encompassed in Tables l-6 anticlastic correction results in improved frequency prediction in no less than 47 cases. For the remaining two boundary conditions comparison has been made with Rayleigh- Ritz solutions since exact frequency values are not available. Table 7 demonstrates that all the first six frequencies of the fully clamped plate are predicted more accurately when anticlastic correction is applied. This is not so for the free plate (F-F-F-F) and it may be seen from Table 8 that with correction and 5 x 5 modelling almost all frequency predictions are inferior to BOGNER 5 x 5 modelling, to non-conforming modelling and to Gorman [9] superposition. The use of high power polynomial terms to force conformity on the correction function of equation (13) can be expected to result in an overstiff element. This expectation is fully confirmed by the values in Table 6 which show that frequencies obtained in this way are all higher than the corresponding exact values. By contrast, several of the frequencies obtained by correction, but without forcing full conformity, are underesti- mates. This is attributable to the limited kinking which such correction introduces at element interfaces. Throughout all the frequencies predicted with consistent mass, the error due to the underestimate of the equivalent mass, associated with dynamic deflection shapes, is present. The error in frequency prediction so introduced results in an overestimate of frequency which increases with frequency. No attempt has been made to apply dynamic discretization to the mass matrices resulting from anticlastic correction and it is only possible to give some small indication of what might result. Table 2 of reference [2] shows that dynamic discretization reduces the consistent mass frequency predictions for 5 x 5 modelling of a S-S-S-S square plate from values of 20.326, 50.612, 82.637, 101.18, 135.07 and 173.67 by 0.012, O-182,0*740, 1.312, 2.870 and 5.630, respectively. The same reductions applied to the results are Table 1 for anticlastic correction give frequencies of 19.339, 48.739, 78.364, 97.387, 127.72 and 164.64. All these values are lower than the corresponding exact values of 19.739, 49.348, 78.957, 98.696, 128.30 and 167.70. This undershoot of frequency prediction may be ascribed with some confidence to the limited inter-element kinking referred to earlier. 5. CONCLUSION The use of consistent mass matrices based on anticlastic correction of conforming beam product functions enables, in the vast majority of cases, very considerable improvement ANTICLASTIC INFLUENCE ON VIBRATION 257 in frequency prediction to be obtained for square plates were constraint exists on two or more edges. The correction results in some loss of conformity but the method adopted to force conformity is unsuitable. Little difficulty is foreseen in incorporating these corrected consistent mass and stiffness matrices into standard finite element programs. ACKNOWLEDGMENTS The author wishes to thank Loughborough University of Technology for use of its facilities, and Mrs J. Smith for her meticulous help in the preparation of the manuscript. 9. REFERENCES B. DOWNS 1980 Journal of Mechanical Design, American Society of Mechanical Engineers 102, 391-398. Vibration analysis of continuous systems by dynamic discretization. B. DOWNS 1988 Journal of Sound and Vibration 122,43-68. The generation of dynamical corrected flat plate finite elements and their application to the transverse vibration of flat plates. J. S. PRZEMIENIECKI 1968 Theory of Matrix Structural Analysis. New York: McGraw-Hill. F. K. BOGNER, R. L. FOX and L. A. SCHMIT 1965 Proceedings of Conference on Matrix Methods in Structural Mechanics, Wright- Patterson Air Force Base, Ohio, AFFDL TR 66-80. The generation of interelement-compatible stiffness and mass matrices by the use of interpolation formulas. A. W. LEISSA 1969 NASA SP-160. Vibration of plates. A. W. LEISSA 1973 Journal of Sound and Vibration 31 257-293. The free vibration of rectangular plates. C. R. HAZELL and A. K. MITCHELL 1986 Experimental Mechanics 26,337-344. Experimental eigenvalues and mode shapes for flat clamped plates. K. VIJAYAKUMAR and G. K. RAMAIAH 1978 Journal of Sound and Vibration 56, 127-135. Analysis of vibration of clamped square plates by the Rayleigh-Ritz method with asymptotic solutions from a modified Bolotin method. D. J. GORMAN 1978 Journal of Sound and Vibration 57,437-447. Free vibration analysis of the completely free rectangular plate by the method of superposition. APPENDIX The k values of equation (11) and percentage reduction of the integral over the plate of (13*s)/ag+ 2 84s/~~23n2+ c~~s/~T~)~, resulting from the corrected displacement function of equation (12) are shown in Table Al. TABLE Al The k values of equation (11) and percentage reductions of the integral over the plate when the equation (12) correction is used Degree of freedom k Percentage reduction 2 3 4 5 6 7 8 9 10 11 12 -0.289 65 -0.5 36 1.0 59 -0.289 65 1.0 59 1.0 59 -0.289 65 1.0 59 -0.5 36 -0.289 65 -0.5 36 -0.5 36 258 B. DOWNS With equation (11) written as SC = kf(t)f( q), the values of the correction polynomials f(e) and f(7) are as follows: d.o.f. 1: f(t) = f ( T) = d.o.f. 2: f(t) f(a) = d.o.f. 3: f(t) = f(s) d.o.f. 4: f(t) f(v) = d.o.f. 5: f(t) f(v) = d.o.f. 6: f(t) f(s) d.o.f. 7: f(t) = f(l)) d.o.f. 8: f(t) f(v) d.o.f. 9: f(t) = f(v) d.o.f. 10: f(t) f(T) d.o.f. 11: f(7) d.o.f. 12: f(t) f(9) -0.752 -0.7517 +0.6755= +3.053 + 0.675q2 +3.oq same as for d.o.f. 1, -0.0187q2 +0.4167q3 - 0*018722 + 0.4167@ same as for d.o.f. 1; same as for d.o.f. 1, +0.675q2 -3.017 same as for d.o.f. 1, -O.O187q* -0.416711 -3*op -3.014 0.013 -0.0854q 0.013 - 0.08545 -0.2083q4 - 0.208324 + 1.256, + 1.2q6; -0.3175 +0.3q6; -0.355 +0.356; 0.7511 0.013 + 0.085417 -3.oq4 -O*2083q4 + 1.2776; +0.3q5 +o.3qe; 0.752 0.013+0.08545 same as for d.o.f. 3, same as for d.o.f. 4; + 0.6755 - 3.053 same as for d.o.f. 4; same as for d.o.f. 7, same as for d.o.f. 5; -0.018752 -0.41675 same as for d.o.f. 4, same as for d.o.f. 7, same as for d.o.f. 1; same as for d.o.f. 7; same as for d.o.f. 2; same as for d.o.f. 9; same as for d.o.f. 1. -3.054 -0.208324 + 1.2&+,