RESPONSE-SURFACE APPROXIMATIONS FOR STRUCTURAL OPTIMIZATION

Response surface methodology can be used to construct global and midra nge approximations to functions in structural optimization. Since stru ctural optimization requires expensive function evaluations, it is imp ortant to construct accurate function approximations so that rapid con vergence may be achieve;l. In this paper techniques to find the region of interest containing the optimal design, and techniques for finding more accurate approximations are reviewed and investigated. Aspects c onsidered are experimental design techniques, the selection of the 'be st' regression equation, intermediate response functions and the locat ion and size of the region of interest. Standard examples in structura l optimization are used to show that the accuracy is largely dependent on the choice of the approximating function with its associated subre gion size, while the selection,of a larger number of points is not nec essarily cost-effective. In a further attempt to improve efficiency, d ifferent regression models were investigated. The results indicate tha t the use of the two methods investigated does not significantly impro ve the results. Finding: an accurate global approximation is challengi ng, and sufficient accuracy could only be achieved in,the,example prob lems by considering-a smaller region of the design space. (C) 1998 Joh n Wiley & Sons, Ltd.