Use of Direct Search in Automated Optimal Design

Abstract

A direct search penalty function procedure is described for treating a rather general mathematical programming form of the optimal design problem. A rotating coordinate modification of the pattern search of Hooke and Jeeves is proposed and compared with the original method, and an earlier variable step modification, as the basic optimal search method. The algorithm is successfully applied to several design examples which indicate that both modifications are comparable and are significantly superior to the original pattern search procedure.