Optimization in aircraft engineering including approximate structural evaluations by support vector machines

Abstract

AbstractThe minimum weight design of structures made of fiber reinforced composite materials leads to a class of mixed‐integer optimization problems for which evolutionary algorithms (EA) are well suited. Based on these algorithms the optimization tool package GEOPS has been developed at TU Dresden.For each design generated by an EA the structural response has to be evaluated. This is often based on a finite element analysis which results in a high computational complexity for each single design. Typical runs of EA require the evaluation of thousands of designs. Thus, an efficient approximation of the structural response could improve the performance considerably. To achieve this aim the constraints on the structural response are approximated by means of support vector machines (SVM). It is trained by means of exact structural evaluations for selected design alternatives only. Several ways to enhance the efficiency of such an optimization procedure are presented.As an example for a typical aircraft structure, a stiffened composite panel under compressive and shear loading is considered. The SVM is trained on geometrical and material data. Representing the design space of composite panels by ABD matrices turned out to be a valuable means for obtaining well trained SVMs. (© 2004 WILEY‐VCH Verlag GmbH & Co. KGaA, Weinheim)