Safety assessment of passive safety systems in nuclear reactors using artificial neural networks

Abstract

This study investigates the application of Artificial Neural Networks (ANNs) for the safety assessment of Passive Safety Systems (PSSs) in nuclear reactors, focusing on mitigating Loss of Coolant Accidents (LOCAs). Using the BWRX-300 Small Modular Reactor (SMR) as an example, the research demonstrates how ANNs can enhance traditional Probabilistic Safety Assessment (PSA) methods. By training ANN models with failure probability data derived from Fault Tree Analysis (FTA), the study predicts failure probabilities of key systems, including the Reactor Isolation (RI) system, Reactor Scram (RS) system, and Isolation Condenser System (ICS). The ANN models successfully captured nonlinear interactions and complex failure scenarios, achieving high prediction accuracy. Additionally, intentional errors introduced into Basic Event (BE) probabilities highlight the ANN's advanced error-handling capabilities, with the models identifying and mitigating discrepancies that FTA failed to address. These findings underscore the potential of ANNs to improve the reliability and safety assessment of nuclear PSSs, offering valuable insights for the development of next-generation reactors.