A novel adaptive Kriging method: Time-dependent reliability-based robust design optimization and case study

Abstract

The computational efficiency and accuracy of the time-dependent reliability-based robust design optimization (TRBRDO) directly rely on the capability to handle the time-dependent reliability analysis (TRA). Some TRA methods use ordinary efficient global optimization (EGO) to identify the extreme samples, and the Kriging model is utilized to approximate the implicit extreme value functions. However, the significant limitation of these methods lies in the unavailability for the parallelized reliability analysis, resulting from the point-to-point nature, which indicates the computational efficiency can be further improved. To construct a more efficient model for the TRA, this paper proposes an adaptive Kriging method, i.e., integrated parallel efficient global optimization (PEGO) and adaptive Kriging-Monte Carlo simulation (AK-MCS), which transforms the TRBRDO problem into an equivalent time-independent one. The proposed adaptive Kriging method was proven to be superior to existing TRBRDO methods in computing efficiency and accuracy, verified by the performance comparison via three cases, including a limit state function with only a time parameter, a two-dimensional function generator, and an engineering application.