Design and Development of an AI-Based Security and Safety System for Hydrogen Explosion Hazard Mitigation in Nuclear Fission Power Plants
Abstract
Mohammad Ali
This paper presents the design and conceptual development of an artificial-intelligence-based Security and Safety System (AI-S3) for early detection, prediction, and mitigation of hydrogen explosion hazards in nuclear fission power plants (NPPs), together with an integrated cybersecurity layer that protects the safety function itself from cyber- physical attack. The hazard addressed is chemical rather than nuclear-yield: a commercial fission core cannot sustain the fast supercritical chain reaction required for a weapon-type detonation, so the credible explosion hazard is combustible-gas—principally hydrogen— deflagration in auxiliary systems and, under severe-accident conditions, in the containment atmosphere, generated in part by high-temperature zirconium–steam oxidation of fuel cladding. The proposed architecture integrates five functional layers—field sensing, edge acquisition, an AI analytics core, decision and arbitration logic, and mitigation actuation—designed to satisfy the independence and redundancy criteria of IEEE Std 384 while applying long short-term memory networks, autoencoder-based anomaly detection, and Bayesian belief updating to obtain early warning ahead of conventional fixed-setpoint alarms. Mathematical models are developed for hydrogen generation and transport, recombiner reaction kinetics, the anomaly-detection and intrusion-classification sub-models, and a probabilistic riskreduction formulation linking AI-S3 to established fault-tree and Bayesian-network practice. A qualitative, illustrative-simulationbased evaluation using notional data indicates that AI-based early warning can identify anomalous hydrogen accumulation and anomalous network behavior appreciably earlier than fixedthreshold instrumentation alone, extending operator response margin. The paper closes with the verification and validation requirements, the cybersecurity assurance case, and the experimental and regulatory work needed before such a system could be qualified for plant deployment.
