Seismic Reliability Analysis of NPP's Nonstructural Components using Surrogate Models

Abstract

This study evaluates the accuracy and computational efficiency of seismic probabilistic risk assessment (SPRA) using a new surrogate model-based reliability analysis and compares the results with the conventional Monte Carle Simulation (MCS) method. The newly developed Kriging-based reliability analysis approach, called REAK, is in the class of Adaptive Kriging-based MCS (AK-MCS) methods, but with the ability to estimate error in failure probability estimates, which is used as the stopping criterion in the reliability analysis. It is shown that REAK can reliably capture the failure probability of nonstructural components located in buildings of nuclear power plants, however, with significantly smaller number of simulations compared to MCS. Results also indicate that the probability of acceleration of nonstructural components in the first floor exceeding that on the second floor is small but not zero, indicating that it cannot be neglected as commonly done in current risk analysis procedures.