Integration of artificial intelligence within an advanced filtering framework for real-time system state estimation and risk prediction with application to a nuclear microreactor

Abstract

Novel nuclear reactor designs, such as Small Modular Reactors and Microreactors, require advanced safety assessment methods to analyze potential threats and hazards, and evaluate the efficacy of the safety systems introduced to cope with them. Then, safety assessment frameworks, embedding simulation models, software tools and artificial intelligence algorithms are envisioned to be integrated in operation for autonomous and semiautonomous activities, ensuring plant reliability and availability while reducing operational and maintenance costs at the same time. For this, Condition-Based Probabilistic Safety Assessment (CB-PSA) must be enabled to provide real-time estimation of the system health state for up-to-date risk evaluation and operation situational awareness. In this paper, we propose a framework of integration consisting of: i) a Neural Network (NN) surrogate model-based Bayesian filter for real-time state estimation during operation using monitoring data; and ii) a NN model for real-time prediction of safety-related parameters. The effectiveness of the proposed framework is demonstrated through its application to a synthetic case study and to a Nuclear Battery design, with reference to a loss of heat sink (LOHS) accident.