Relative Entropy Based Method for Global and Regional Sensitivity Analysis in Probabilistic Design

To overcome the limitations of existing variance-based methods for Probabilistic Sensitivity Analysis (PSA) in design under uncertainty, a new PSA approach based on the concept of relative entropy is proposed. The relative entropy based method evaluates the impact of a random variable by measuring the divergence between two probability density functions of a response. The method can be applied both globally over the whole distribution of a performance response (called global response probabilistic sensitivity analysis-GRPSA) and in any regional range of a response distribution (called regional response probabilistic sensitivity analysis-RRPSA). The former is the most useful for studying variable impact on robust design objective, while the latter provides insight into reliability constraints. The proposed method is applicable to both the prior-design and post-design stages, for variable screening and uncertainty reduction, respectively. The proposed method is verified by numerical examples and industrial design cases. 1