Reliability-oriented sensitivity analysis in presence of data-driven epistemic uncertainty

Reliability assessment in presence of epistemic uncertainty leads to consider the failure probability as a quantity depending on the state of knowledge about uncertain input parameters. The input joint distribution is often learnt from a small-sized dataset provided by operating experience. The computed failure probability depends on the estimated marginal distributions and the estimated copula distribution. This paper develops a reliability-oriented sensitivity analysis procedure in order to measure the influence exerted by the data-driven modeling of both the margins and the copula. The proposed methodology is validated for both deterministic and stochastic reliability methods through an extensive simulation study including several analytical performance functions as well as a real-life simulation code dealing with the buckling of a laminated composite plate.

Automated summary

This paper investigates reliability assessment in the presence of epistemic uncertainty and proposes a sensitivity analysis method to quantify the impact of data-driven modeling on both margins and Copula. The proposed method is validated through simulation studies.