An innovative adaptive Kriging-based parallel system reliability method under error stopping criterion for efficiently analyzing the global reliability sensitivity index

Global reliability sensitivity index measures the effect of uncertainty of model input on model failure probability, which is critical for simplifying analysis model and the reliability-based design optimization model. For efficiently estimating the global reliability sensitivity index of each model input, this paper transforms it into estimating an unconditional failure probability and a two failure modes-based parallel system failure probability from the perspective of single-loop estimation method. Furthermore, the relationship of computational accuracy among the global reliability sensitivity index, the unconditional failure probability, and the two failure modes-based system failure probability is constructed, on which the error stopping criterion-based sequentially adaptive Kriging model approach is developed to significantly decrease the number of calls to the actual limit state functions and the corresponding computational time under the sufficient accuracy. Results of three case studies covering explicit and implicit limit state functions demonstrate the accuracy and efficiency of the proposed method.