The applicability of AI/ML in this field offers significant benefits, including enhanced anomaly detection, failure prediction, and optimized maintenance schedules, all of which contribute to reducing accident risks and ensuring strict safety compliance. Assistant Professor and Associate PUR-1 Director Stylianos Chatzidakis and a team of researchers from the School of Nuclear Engineering conducted a project titled, Characterizing Nuclear Cybersecurity States Using Artificial Intelligence/Machine Learning. Their final report, now published as an official NRC Technical Letter Report, evaluates the feasibility of AI/ML technologies to characterize cyber events within nuclear systems. The project focuses on how these tools can distinguish between normal, abnormal, and cybersecurity-related states in nuclear environments.

 

The research identified nine potential use cases, with one fully implemented using real-world data from Purdue University’s PUR-1 digital reactor. Extensive operational technology (OT) and information technology (IT) data were collected to train AI/ML models, which were then assessed through a composite classifier architecture. This system demonstrated the ability to effectively identify different system states and detect cybersecurity events, offering critical insights into AI/ML's practical applications in nuclear safety and cybersecurity.

 

Among the various algorithms tested, Random Forest models emerged as the most effective, showing superior performance and explainability compared to alternatives such as Super Vector Machines and Logistic Regression. This approach proved to be effective in differentiating between cyber-threat-related, normal, abnormal, and abnormal conditions in the nuclear environment.  

 

The broader goal of this initiative is to leverage AI/ML technologies to predict system failures, detect anomalies, and enhance responses to cyber threats. By doing so, nuclear facilities can reduce the likelihood of accidents, improve maintenance practices, and strengthen operational safety. Integrating AI/ML into these systems will help contribute to the overarching objectives of strengthening critical infrastructure and representing the future of safer and more secure nuclear reactor operations.

 

This research project marks a milestone in nuclear cybersecurity and AI/ML research, being the first of its kind to use real-time reactor data. The Purdue team, led by Stylianos Chatzidakis, includes graduate research assistants Vasileios Theos, Konstantinos Gkouliaras, Zachery Dahm, William Richards, Kostas Vasili, and True Miller, Reactor Supervisor and Brian Jowers, Nuclear Electronics Technician, expects to publish several papers based on this groundbreaking work.