Multi-unit seismic probabilistic risk assessment: A Bayesian network perspective

Abstract

The Fukushima Dai-ichi highlighted multi-unit accidents in nuclear power plant (NPP) probabilistic risk assessment (PRA). This paper focuses on multi-unit seismic PRA because seismic events are contributors to NPP site risk. This paper integrates works on modeling ground motion spatial variability at a NPP site and dependent seismic failures into a coherent and rigorous framework using Bayesian networks, and explores the relative importance of different types of correlation on NPP site-based risk metrics. A study of a hypothetical dual unit NPP site considered 15 cases with varying degrees of correlation in spatial ground motion (GM) and seismic failures. Based on the results, the current assumption of perfect GM correlation across a NPP site underestimates risk metrics that consider either unit experiencing core damage (CD) (site CD), which is not conservative. The risk metric that considers both units experiencing CD (concurrent CD) shows differences between the partial and perfect GM correlation assumptions when multiple component failures contribute to the CD in each unit. Within a GM correlation assumption, the site CD increases as the fragility correlation increases, a counterintuitive trend. The risk assessment case outlined in this paper provides more realistic results than the cases reflecting the current state of practice.