A unified framework for two-level reliability-based design optimization using metamodels

Abstract

Reliability-based design optimization (RBDO) is an active field of research that combines reliability analysis and optimization techniques. The early RBDO techniques were limited in applications because of the approximation techniques (FORM) used in the reliability part. More sophisticated techniques that rely on simulations were not of practical interest until the recent introduction of surrogate modeling to the field of structural reliability. Many approaches that couple surrogate modeling, simulation-based reliability analysis and optimization for the solution of RBDO problems have been recently developed. This paper proposes a global and unified framework for surrogate-assisted RBDO that allows for the solution of various types of RBDO formulations. This framework is built using three distinct and independent blocks which are coupled non-intrusively. To enhance the overall efficiency, the surrogate models are built adaptively in a single augmented space prior to starting optimization. The proposed framework is illustrated using different techniques in each of the three blocks.