Error-informed parallel adaptive Kriging method for time-dependent reliability analysis

Active learning single-loop Kriging methods have gained significant attention for time-dependent reliability analysis. However, it still remains a challenge to estimate the time-dependent failure probability efficiently and accurately in practical engineering problems. This study proposes a new method, called ‘Error-informed Parallel Adaptive Kriging’ (EPAK) for efficient time-dependent reliability analysis. First, a sequential variance-amplified importance sampling technique is developed to estimate the time-dependent failure probability based on the trained global response Kriging model of the true performance function. Then, the maximum relative error of the time-dependent failure probability is derived to facilitate the construction of stopping criterion. Finally, a parallel sampling strategy is proposed through combining the relative error contribution and an influence function, which enables parallel computing and reduces the unnecessary limit state function evaluations caused by excessive clustering. The superior performance of the proposed method is validated through several examples. Numerical results demonstrate that the method can accurately estimate the time-dependent failure probability with higher efficiency than several compared methods.