Bayesian network-based surrogate model for physical simulation in probabilistic risk assessment of nuclear power plants

In recent years, physical simulations, such as computational fluid dynamics (CFD) and finite element analysis (FEA), have been increasingly utilized to support probabilistic risk assessment (PRA) of nuclear power plants (NPPs). However, these simulations are often computationally intensive, particularly when used to explore broad scenario spaces and quantify uncertainty in PRA. This paper proposes a Bayesian network (BN)-based surrogate modeling approach to reduce the computational burden of physical simulations in PRA. Methodological steps for developing, validating, and applying the BN-based surrogate model (specifically, to support screening analysis in PRA) are presented. The implementation of the proposed approach is demonstrated through a case study of NPP fire modeling, where a BN-based surrogate model is constructed and validated for the Consolidated Model of Fire Growth and Smoke Transport (CFAST) code. Compared to other machine learning-based surrogate models previously studied in the PRA domain, BNs offer two key advantages: (i) representing causal relationships among variables explicitly, and (ii) providing a transparent model structure and inputs that can be effectively communicated using diagrams and tables. Additionally, the widespread adoption of BNs in the PRA domain can facilitate broader applications of the proposed approach.