Predictions of the Maneuvers of Spent Fuel Elements in a Nuclear Power Plant to a Dry Storage Unit Using Artificial Intelligence

Transfer operations of irradiated fuel elements (IFEs) from storage pools to dry storage units (DSUs) involve heavy equipment and complex multibody dynamics, where failures may result in significant radiological and economic consequences. Incidents at nuclear power plants, such as those at San Onofre in 2018 and Vermont Yankee in 2008, involving prolonged canister suspensions and crane brake failures, highlight critical vulnerabilities in conventional reactive monitoring approaches and the need for advanced predictive monitoring systems.This study proposes a simulation-based methodology to evaluate the feasibility of the FastTree model for predictive anomaly detection during IFE transfer operations, from initial handling to final canister positioning in DSUs. The methodology integrates a physics-based simulation framework employing PyBullet with a FastTree-based prediction model, whose performance is benchmarked against random forest, eXtreme Gradient Boosting, Light Gradient Boosting Machine, and support vector machine models.The results show that FastTree achieved 97.2% accuracy, 95.8% precision, 97.1% recall, and a 95.6% F1 score on a data set comprised of four operational scenarios. The model provided 5 to 15 s of advance warning before critical events while exhibiting lower computational cost than the more complex models evaluated, thereby demonstrating that the proposed methodology is suitable for real-time implementation in predictive monitoring applications.Although the proposed framework was assessed using physics-based simulation data, these findings establish a foundation for future validation using operational data from nuclear facilities.