A Concurrent Subspace Optimization Procedure Based on Multidisciplinary Active Regional Crossover Optimization

Abstract

Concurrent Subspace Optimization (CSSO) procedure has been widely studied in the field of Multidisciplinary Design Optimization (MDO). It can realize decoupled concurrent design and optimization of complex systems, but it has difficulties in coordinating subspace optimization results and guaranteeing robust convergence of the whole optimization procedure. To address this problem and improve the CSSO optimization efficiency, especially to enhance the coordination efficiency of subspace optimizations, an improved CSSO procedure is proposed based on Multidisciplinary Active Regional Crossover Optimization (CSSO-MARCO). In the subspace optimization, this method utilizes an active region for each design variable from other subspaces instead of a constant value as done in the traditional CSSO so that the specialists can have more flexibility in finding the optimum of their subspace with variability of other subspace design variables. After optimization of each subspace, the specialists not only feedback the optimization results to the system level, but also the active regions for the design variables in this subspace within which the optimization design performance will change within given tolerance compared to the current optimum. Then the system optimization is run with all the design variables bounded within the active regions so as to reduce computation burden and improve system optimization efficiency. The capability of the proposed method is validated and demonstrated in two examples.