ASA 129th Meeting - Washington, DC - 1995 May 30 .. Jun 06

4pSAb5. Sensitivity analysis of structural eigenfrequency to plate design parameters by using neural network.

Z. Q. Yu

L. H. Yam

T. P. Leung

Dept. of Mech. & Marine Eng., Hong Kong Polytech. Univ., Hung Hom, Kowloon, Hong Kong

The dynamic characteristics of plate structure can be improved by arranging the natural frequencies of the system in specified ranges. The sensitivities of structural dynamic response to design parameters can provide the essential gradient information for deriving efficient optimization procedures. They are used to form an approximate problem for solving the original structural optimization problem, while the sensitivity of geometric parameters to frequency can provide the essential gradient information for deriving efficient design procedures. In this paper, first the training set of neural network is generated by the finite-element method; second, the original design of a cantilever rectangular plate with multifrequency constraints is obtained by using the neural network; third, the sensitivities of the plate eigenvalue to design parameters (i.e., (cursive beta) f/(cursive beta)R[sub i]) around the point of original geometric design are obtained by using the trained neural network. Finally, the sensitivities of design parameters to eigenvalue (i.e., (cursive beta)R/(cursive beta)f[sub i]) are also obtained by using the trained neural network. The influence of frequency on geometric parameters, as well as the influence of geometric parameters on frequency, is also discussed.