A Bayesian Network Approach for Modeling Dependent Seismic Failures in a Nuclear Power Plant Probabilistic Risk Assessment

Abstract

The importance of modeling dependency between seismic failures of multiple components in a nuclear power plant (NPP) probabilistic risk assessment (PRA) has been discussed since the 1980s. In NUREG/CR-7237, Budnitz et al. found the Reed-McCann method to be the most promising method for modeling dependent seismic failures in NPP PRA. However, there are issues with the Reed-McCann method's quantification of the seismic fragility of a system of multiple components. To address this issue and to facilitate an overall realism increase in modeling dependencies in seismic PRA, this paper proposes a Bayesian network (BN) approach to model dependent seismic failures. To illustrate the proposed approach, we calculate the fragility of a parallel system and a series system using the Reed-McCann method, the BN approach, the First-Order Reliability Method (FORM) and Monte Carlo simulation (MCS). Then, we compare the system fragility results from these four approaches/methods to the lower and upper bounds of the system fragility. We found that the BN approach performed better than the Reed-McCann method with respect to providing results that stay within the lower and upper bounds of the system fragility. Further, the BN approach gives similar results to FORM and MCS. This paper proposes a BN approach because, in combination with our previous work about extending a probabilistic seismic hazard analysis to account for the spatial variability of ground motion at an NPP hard-rock site, it can be used to simultaneously and realistically account for dependent seismic failures and spatial variability of ground motion in both single-unit and multi-unit seismic PRAs.