Adaptation of soft computing methods in multidisciplinary and structural optimization.

摘要

Multidisciplinary and structural design problems typically involve a large number of design constraints and variables, where the latter may be discrete, integer, or continuous. The analysis required to compute the objective and constraint functions are usually complex, and coupled and sometimes, they are defined by imprecise information. The resulting design domain may well be nonconvex, disjointed or even non-crisply defined. Furthermore, multiple conflicting criteria from many disciplines may need to be optimized at the same time. In such problems, traditional optimization techniques based on principles of mathematical programming have often shown to be inadequate. Therefore, in this study, newly emergent computational techniques that are broadly defined as soft computing are presented as an approach to address some of these problems.; Soft computing techniques include genetic algorithms, neural networks, and fuzzy logic. In this study, various applications of these techniques in regards to the optimization and modeling aspects of multidisciplinary design are investigated. First, enhancements to genetic algorithms are explored to both improve the convergence characteristics and adapt the approach to handle multicriterion design problems. These modifications to genetic algorithms are derived from an integrated simulation of the biological immune network system. This approach has shown to be effective in handling multicriterion design problems of non-convex front of Pareto optimal solutions. Secondly, neural networks and fuzzy logic are presented as viable global approximation tools that can significantly reduce the high computational cost associated with genetic algorithms. Neural networks provide a robust response surface-like function approximation tool, whereas, fuzzy logic based approximation is an imperative technique when modeling a system (e.g. manufacturing process) that is described by imprecise information.; A collective contribution of each of these soft computing techniques is summarized in optimal design of a composite wing panel. This problem consists of a moderate size of design variables of mixed type, and several constraints involving disciplines of structure, dynamics, and manufacturing that are typically analyzed in design of composite parts. The results have shown the proposed solution approach based on soft computing techniques to be more efficient in solving such type of design problems than the traditional approach.