Perturbed-Law based sensitivity Indices for sensitivity analysis in structural reliability

Roman Sueur (1, 2) , Nicolas Bousquet (1, 2) , Bertrand Iooss (1, 2) , Julien Bect (3, 2)

1 EDF - EDF
2 GdR MASCOT-NUM - Méthodes d'Analyse Stochastique des Codes et Traitements Numériques
3 L2S - Laboratoire des signaux et systèmes

In the field of sensitivity analysis of model outputs, the most popular methods are those based on the variance decomposition of the output, such as the Sobol’ indices, as they allow defining easy-to-interpret indices measuring the contribution of each input variable to the overall output dispersion. However, these indices are not adapted to the analysis of the impact of inputs on output quantities characterizing extreme events, such as a failure probability, a quantile or a failure domain.

Domains: Methodology [stat.ME]; Computation [stat.CO]

Roman Sueur, Nicolas Bousquet, Bertrand Iooss, Julien Bect. Perturbed-Law based sensitivity Indices for sensitivity analysis in structural reliability. 8th International Conference on Sensitivity Analysis of Model Output (SAMO 2016), Nov 2016, Le Tampon, Réunion, France. pp.89-90. ⟨hal-01569578⟩