扩展的输入变量对失效概率重要性测度及其积分算法

Abstract

To fully analyze the effects of input variables on failure probability in reliability, an extended moment-independent importance measure based on the traditional moment-independent one is proposed. The computational cost of the moment-independent important measure is too high as direct Monte Carlo simulation is used. To overcome the difficulty, an integral solution is established by combining the highly efficient and exact Kernel density estimation. Results of several examples are used to demonstrate that the proposed importance measure can describe the effects of input variables on failure probability more fully and the established method can overcome the problem of "curse of dimensionality", which reduces the computational cost significantly.