An Extended Kriging Model-Based Method for Efficient Reliability Analysis Under Parametric Variation of Probability Distribution

For traditional surrogate model-based methods, the distribution parameters of random variables will change with the enrichment of the data, ultimately leading to the reconstruction of surrogate model and reassessment of reliability, which often results in a wastage of computational resources. In response, an extended Kriging model-based method is proposed in this paper for efficient reliability analysis under parameters change. Different from the standard Kriging model, in addition to random variables, variable distribution parameters are also included in the proposed method to constitute the extended input variables. In this way, the updated Kriging model with prior parameter information can be directly treated as the initial surrogate model for the posterior problem. Consequently, there is no need to reconstruct the surrogate model if the distribution parameters change, and only one Kriging model is required for this method. The efficiency, robustness and accuracy of the proposed method are demonstrated through three examples. This study contributes to dynamic assessment techniques for structural reliability analysis.