Optimizing Operations and Maintenance Through Machine Learning: Generative Adversarial Networks, Physics-Informed Machine Learning, and Reinforcement Learning

Safety and reliability are of the utmost importance in safety-critical systems, such as nuclear power plants. Anomalies in these systems can arise from various sources, including sensor faults, environmental factors, and human error, undermining plant integrity. If left undetected, these anomalies can lead to unexpected failures, costly downtime, and potentially significant safety incidents. Therefore, effective anomaly detection plays a crucial role in maintaining reliability and optimizing operation and maintenance efforts. Given the significance of the topic, there is a pressing need to develop robust tools to capture these anomalies to allow for effective error detection and mitigation. To this end, machine learning models, particularly anomaly detection algorithms, offer promising solutions by detecting outliers within complex data. Thus, the objective of this dissertation is to develop artificial intelligence-based tools for anomaly detection in nuclear power plants. Specifically, it explores the use of advanced machine learning techniques, including generative adversarial networks, physics-informed models, and reinforcement learning, in detecting sensor anomalies and human error in nuclear power plants, with the overarching goal of optimizing operation and maintenance strategies in safety-critical industries.