Title: Reliability sensitivity analysis with dependent inputs
Author: Iason Papaioannou (TU Munich)

In reliability assessments, it is often useful to compute importance measures that provide information on the influence of the uncertain input variables on the probability of failure. Most existing reliability sensitivity measures are mainly applicable to problems with statistically independent uncertain inputs. This presentation discusses reliability sensitivity measures for dependent inputs. Thereby, one needs to distinguish between variable interactions due to the probabilistic model of the uncertain inputs and the deterministic model describing the system response. This results in two sets of sensitivity measures, the first reflecting the total contribution of each input due to interactions in both probabilistic and deterministic model and the second representing the independent contribution of each input due to interactions in the deterministic model only. Estimation of these indices is challenging on account of the typically low magnitude of the probability of failure. We discuss computational strategies to address this problem that leverage existing approximation and Monte Carlo simulation-based reliability methods to obtain estimates of the sensitivity indices without additional runs of the deterministic model.