摘要：AbstractThe aim of this paper is to present a sensitivity statistic developed in the context of the design of a new accelerated Monte-Carlo method. In the field of structural reliability, we elaborated the “Adaptive Directional Stratification” method (ADS), in order to estimate small failure probabilities in a robust manner with a limited number of simulations. In order to break the curse of dimensionality, we propose an efficient statistic, evaluated at the end of the learning stage of the ADS method, to detect the input variables which are the most influential on the failure event. Thereby, we can focus the computational effort by stratifying only the most influential variables, which allows to better deal with high-dimensional spaces in the estimation step of the ADS method.
