Sarcouncil Journal of Engineering and Computer Sciences

Abstract: Nuclear power plants are among the most complex and safety-sensitive facilities in the world. The ability to quickly detect and diagnose faults within their control systems is critical for ensuring safe, stable, and uninterrupted operations. While traditional fault detection methods, like rule-based systems or physics-driven models, have served the industry for decades, they often fall short in responding to the increasingly dynamic, data-rich, and nonlinear nature of today’s nuclear power environments. This growing complexity has created an opportunity for Artificial Intelligence (AI) to step in. Recent research shows that AI techniques, particularly those based on machine learning, offer powerful tools for identifying subtle anomalies, recognizing unusual system behaviors, and even predicting faults before they occur. From deep learning models and ensemble classifiers to digital twins and unsupervised anomaly detection, these innovations are helping engineers and operators improve reliability and reduce response times across a range of nuclear plant subsystems. This review explores the progress made so far in applying AI to fault detection and diagnostics (FDD) in nuclear control systems. It organizes the key techniques in use, highlights notable case studies and applications, and discusses both the benefits and challenges associated with deploying AI in this high-stakes setting. Issues such as data limitations, model transparency, regulatory concerns, and cybersecurity are also considered. The review concludes by outlining promising future directions, including the integration of explainable AI, edge computing, and collaborative learning methods. By consolidating existing knowledge and offering strategic insights into the state-of-the-art, this review aims to guide researchers, engineers, and policymakers in advancing safer, smarter, and more resilient fault detection frameworks for the next generation of nuclear power systems.