A sequential metamodel-based importance sampling coupled with adaptive Kriging model method for efficiently estimating the global reliability sensitivity indices

Global reliability sensitivity analysis plays a critical role in identifying both important and unimportant variables affecting reliability, thus providing guidance for the simplification of reliability-based design optimization. Developing an efficient algorithm for estimating global reliability sensitivity indices is essential for the practical application of this theory in engineering contexts. This paper proposes an effective algorithm leveraging a metamodel-based importance sampling method combined with an adaptive Kriging model and a new single-loop estimation formula. Firstly, global reliability sensitivity analysis is equivalently transformed into an unconditional failure probability analysis and a two failure modes-based parallel failure probability analysis, utilizing the new single-loop estimation formula. Secondly, by sequentially constructing the importance sampling probability density functions for the variables within the global reliability sensitivity indices, both the unconditional failure probability and the two failure modes-based parallel failure probability can be efficiently estimated through the integrated metamodel-based importance sampling approach with the adaptive Kriging method. Finally, the efficiency and accuracy of the proposed method are methodically validated through analyzing a numerical analysis of a roof truss structure and a finite element model-based turbine shaft engineering structure.