Multidisciplinary Analysis and Optimization of Discrete Problems Using Response Surface Methods

Abstract

The objective of this paper is to present a new algorithm to efficiently optimize multidisciplinary, coupled nonhierarchic systems with discrete variables. The algorithm decomposes the system into contributing disciplines, and uses designed experiments within the disciplines to build local response surface approximations to the discipline analysis. First and second order Global Sensitivity Equations are formulated and approximated by experimental data to build approximations to the global design space. The global approximation is optimized using branch and bound or simulated annealing. Convergence is rapid for systems with near quadratic behavior. The algorithm is demonstrated on a unique multidisciplinary learning tool, the Design and Manufacturing Learning Environment. This environment provides multimedia simulation for product life cycle disciplines, including design, manufacturing, marketing, and sales.