Coupled-analysis assisted gradient-enhanced kriging method for global multidisciplinary design optimization

Abstract

ABSTRACT A coupled-analysis assisted gradient-enhanced kriging (CAGEK) method is introduced to improve the quality and efficiency in solving global multidisciplinary design optimization (MDO) problems when multiple disciplines are coupled and expensive computations are required to evaluate these disciplines. In this method, the multidisciplinary feasible architecture is employed to effectively obtain the values of coupled variables. The CAGEK method is an adaptive metamodelling-based optimization method with the gradient-enhanced kriging (GEK) model as the metamodel for improving optimization efficiency by using fewer data samples. A coupled analysis approach is used to calculate the gradient efficiently for the GEK model. Besides, a multiple-point infill method is used to obtain new samples at each optimization iteration considering convergence rate and global optimization capability. The CAGEK method is compared with three traditional methods using four MDO problems to demonstrate its effectiveness.