Machine Learning Integration in Nuclear Safety Analysis: A Regulatory Framework for Nuclear Safety Applications

The use of machine learning (ML) in nuclear power plant safety analysis is becoming an important development in the assessment of risk, the detection of abnormal conditions, and the improvement of operational safety. Rather than replacing established safety analysis methods, ML is increasingly being examined as a supporting tool that can strengthen data interpretation, pattern recognition, and decision-making within the nuclear safety framework. This article develops a structured framework for incorporating ML-based methods into Safety Analysis Reports (SARs) for nuclear power plants, with particular attention to International Atomic Energy Agency (IAEA) safety standards and recent academic research. Drawing on recent technical discussions, relevant safety guidance, and peer-reviewed studies, the methodological foundations, regulatory expectations, and practical applications of ML in nuclear safety are examined. Particular attention is given to challenges that are especially significant in nuclear applications, including limited data availability, model transparency, and the verification and validation of ML-based tools. A structured approach is proposed for the inclusion of physics-informed machine learning, probabilistic risk assessment methods, and human factors considerations within SAR documentation. In addition, emerging applications in piping integrity assessment, fault diagnosis, and reactor performance monitoring are reviewed, and future research needs are identified in relation to the IAEA's developing safety framework for artificial intelligence (AI) in nuclear installations.