New Importance Measures Based on Failure Probability in Global Sensitivity Analysis of Reliability

This article presents new sensitivity measures in reliability-oriented global analysis. The obtained results show that the contrast and newly proposed (entropy two others) effectively describe influence of input random variables on probability failure Pf. measure builds Sobol, using variance binary outcome as either a success (0) or (1). In Bernoulli distribution, Pf(1 ? Pf) discrete entropy—Pfln(Pf) (1 Pf)ln(1 are similar to dome functions. By replacing with entropy, alternative is obtained, then additional derived. It shown desired property all shape; rise not important. Although decomposition indices proven, case studies suggest rationale structure analysis small ranking based total approximately same, but proportions first-order higher-order very different. Discrete entropy gives significantly higher than other measures, presenting an interesting engineering reliability.