A Novel Reliability-Based Design Optimization Method Using Ensemble of Metamodels

Abstract

Metamodels have been introduced in reliability-based design optimization (RBDO) to approximate the complex black-box performance function widely existed in today’s engineering design problems. To further improve the suitability and effectiveness of existing methods, a novel ensemble of metamodels based RBDO method (EM-RBDO) is proposed in this paper. In EM-RBDO, polynomial response surface (PRS), support vector regression (SVR) and Kriging are combined to construct a weighted average surrogate model (ensemble of metamodels). The weight coefficients are calculated according to prediction ability of different metamodels. Inherited Latin Hypercube Sampling (ILHS) method with the initial design as inherited point is applied to generate the samples to build metamodels. Then monte carlo simulation (MCS) combined with sequential approximate programming (SAP) are used to calculate the optimal design. The proposed EM-RBDO model is compared with the single-metamodel based RBDO method on three analytical examples. The comparison results demonstrate that RBDO using the proposed method is superior to other methods in terms of robustness.