A moment independent based importance measure with hybrid uncertainty

Abstract

Input uncertainty always exists in most engineering problems and leads to output response uncertainty for model predictions. Several global sensitivity analysis methods are utilized to measure the importance of input aleatory uncertainty which influence output the most. However, the aleatory uncertainty often involves epistemic uncertainty in the distribution parameters due to the lack of knowledge. In this paper, an improved moment independent approach coupled with auxiliary variable method is presented to separate aleatory and epistemic terms of hybrid uncertainty. The importance measure is derived to compute the individual contributions of aleatory and epistemic parameters to model output’s uncertainty. Considering the high computation costs of moment independent method, a double loop sampling method is applied in the numerical codes to alleviate simulation. A modified Ishigami function is take for instance for demonstrating the effectiveness and rationality of proposed method and high efficiency of sampling algorithm.