Structural reliability sensitivity analysis based on classification of model output

Abstract

In structural reliability analysis, sensitivity analysis can be used to measure how the input variable influences the failure of structure. In this work, a new reliability sensitivity analysis method is proposed. In the proposed method, the model output is separated into two classes (failure domain and safe domain). The basic idea is that if the failure-conditional probability density function of input variable is significantly different from its unconditional probability density function, then the input variable is sensitive to the failure of structure. The proposed reliability sensitivity indices contain both individual sensitivity index and interaction sensitivity index. The individual sensitivity index can measure the individual effect of input variable on the failure of structure. The asymmetrical interaction sensitivity index can measure how one input variable influences the effect of another input variable on the failure of structure. Additionally, the meanings of the proposed reliability sensitivity indices are also interpreted explicitly, and a data-driven estimation method is also proposed to estimate the proposed reliability sensitivity indices. Finally, a numerical example and two engineering examples are presented to illustrate the rationality of the proposed sensitivity indices and the feasibility of the proposed estimation method.