open access

https://hdl.handle.net/2115/100478

A weighted importance sampling method for estimating local reliability sensitivity

英語

Importance sampling method

2026-07-01

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

Local reliability sensitivity (LRS) quantifies the influence of distribution parameters of random inputs on the failure probability. Existing numerical simulation methods can simultaneously estimate both the failure probability and LRS using a shared set of samples. However, achieving convergence for all LRS estimates typically requires substantially more samples than needed for the failure probability alone, leading to high computational costs. To address this challenge, this paper proposes a weighted importance sampling method. First, a Gaussian mixture model (GMM) is constructed to approximate the optimal importance sampling density for failure probability estimation. This GMM is then used to compute the coefficient of variation for each estimated LRS. Subsequently, these coefficients are employed as weights to formulate an enhanced importance sampling density, which targets to reduce the variance of LRS with large coefficients of variation. The enhanced importance sampling strategy in the proposed method improves the efficiency of local sensitivity estimation. This improvement is clearly demonstrated by the presented examples.

会議発表論文

Version of Record