open access

https://hdl.handle.net/2115/100480

Adaptive radial sampling combined with stratified mixture importance sampling for efficiently estimating extremely small failure probability under random-interval uncertainty

英語

Adaptive Kriging

2026-07-01

Proceedings of the 10th International Symposium on Reliability Engineering and Risk Management (ISRERM 2026)

To efficiently estimate the extremely small failure probability under random-interval uncertainty, this paper proposes an adaptive radial sampling combined with stratified mixture importance sampling method. The proposed method first constructs a hypersphere with relatively large radius in the standard normal space, outside which the samples are generated by radial sampling to explore the region far away from the coordinate origin. Then the rare failure domain can be quickly approached by adaptively stratifying. By clustering analysis for the failure samples of each layer, an explicit and regular mixture importance sampling probability density function (MIS-PDF) can be gradually constructed. Since the constructed explicit and regular MIS-PDF can efficiently explore the rare failure domain and greatly reduce the variance of estimating the extremely small failure probability with relatively small additional computational cost, it shows higher efficiency compared to the most update-to-date competing methods. To further reduce the performance function evaluations, the adaptive Kriging model of performance function is employed, thereby efficiently constructing the MIS-PDF and identifying the state of candidate sample pool. Several examples are investigated to validate the superiority of the proposed method.

会議発表論文

Version of Record