An efficient method for predictive-failure-probability-based global sensitivity analysis
Zhao Zhao,Zhao-Hui Lu,Yan-Gang Zhao
DOI: https://doi.org/10.1007/s00158-022-03434-3
IF: 4.279
2022-11-12
Structural and Multidisciplinary Optimization
Abstract:Predictive failure probability (PFP)-based global sensitivity analysis (GSA) can reasonably measure the effect of overall uncertainty, inherent uncertainty, and distribution parameter uncertainty of input variables on PFP. However, estimating the PFP-based global sensitivity measures with respect to these three uncertainties efficiently and accurately remains challengeable. Therefore, this paper presents a novel method for PFP-based GSA. Firstly, the original performance function is converted into an augmented performance function so that the overall uncertainty of input variable is decomposed into the corresponding inherent uncertainty and distribution parameter uncertainty. Then, an efficient estimation method is proposed for the unconditional and conditional moments of the augmented performance function based on the univariate dimension reduction method (UDRM). After obtaining the unconditional and conditional moments, the PFP-based global sensitivity measures with respect to three uncertainties can be computed directly. In the proposed method, the information of the integration grid of UDRM is reused to ensure the high computational efficiency. Four numerical examples are employed to illustrate the accuracy and efficiency of the proposed method.
mechanics,computer science, interdisciplinary applications,engineering, multidisciplinary