Hybrid Response Surface Function-Based Metamodeling of Response Approximation for Reliability Analysis

Abstract

Response surface method (RSM) is mostly adopted to overcome the computational challenge of Monte Carlo simulation (MCS)-based reliability assessment of system comprising implicit limit state function (LSF). In the present study, a hybrid response surface function (HRSF) based on exponential approximation and second-order polynomial estimation is investigated for improved response approximation for reliability analysis. The method calibrates each random variable according to the response function using an exponential function having a varying parameter that attempts to accommodate the nonlinearity of each variable in the LSF. The exponential function is further regressed using conventional quadratic polynomial model. The effectiveness of the proposed HRSF-based approach is elucidated numerically by considering several saturated designs and uniform design schemes. The performance of the proposed procedure is demonstrated by comparing the proposed response approximation and the usual polynomial RSM-based approximation with that of obtained by the most accurate direct MCS technique.