Structural sensitivity analysis based on a hybrid parametric and non-parametric approach

Structural sensitivity analysis is used to rank the influence of different model design variables on a set of response features of interest. It can be used qualitatively in a preliminary screening stage, to identify the dominant physical effects, or quantitatively in uncertainty quantification, to characterize the relative contributions of the most influential design variables to the global observed variability. This article will focus on procedures for effects screening for which local first-order and global sampling-based approaches can be found in the literature. They both generally require the definition of a set of active design variables and this can be a time-consuming process for complex multi-level models. The objective of this study is to combine a nonparametric structural modification technique with a hypersensitivity analysis in order to identify the most influential subdomains of the model without detailing the specific design variables. The proposed strategy is illustrated on academic test cases and the results compared to more classical local and sampling procedures.