An efficient method employing sequentially-updated surrogate model for repeated finite element reliability analysis

Abstract

As risk-informed design and maintenance is used in various ways for structures, it is often necessary to perform finite element reliability analysis (FERA) repeatedly with various values of load intensity. Although the first order reliability method (FORM) has been widely adopted to perform FERA efficiently, the computational cost can still be large, particularly when the target structure is nonlinear or complicated. In this research, a new method is proposed to perform FORM-based FERA more efficiently. In the proposed method, a kriging-based surrogate model is constructed using the results of prior FORM analyses, and the surrogate model allows the next FORM analysis to start from the optimal values of random variables. It can reduce the computational cost significantly, compared with the conventional method where the FORM analysis generally starts from the mean values of random variables. In addition, as FORM-based FERA is repeated with multiple load intensity values, the surrogate model is sequentially updated and further optimized starting values can be provided. The proposed method was applied to calculate flood fragility estimates for a bridge with various values of water velocity, and it was observed that the analysis cost was less than 60% of the original one, with the similar level of accuracy of the analysis results.