Impact of Cooldown on Pebble Fuel Burnup Prediction Using Machine Learning

The Pebble-Bed Reactor design is unique in its capacity for continuous multi-pass circulation of its fuel elements, informed by a burnup measurement system.The need for a lengthy cooldown period following the ejection of a fuel pebble from the reactor core to perform an accurate measurement of burnup is detrimental to pebble throughput and makes continuous fuel circulation logistically complicated.In a previous work, the authors analyzed the proficiency of common ML regression algorithms while assuming no cooldown period as a baseline comparison.It was found that reliable prediction was possible but that limitations in the model decreased prediction accuracy.In this work, the impact of a cooldown period on ML prediction accuracy was analyzed for a database of gamma spectra generated using GADRAS with irradiated material data calculated from the Monte Carlo neutronics code Serpent 2 to provide training and testing data for ML supervised regression algorithms.It has been demonstrated that this results in a data-driven measurement approach that boasts high prediction accuracy with minimal cooldown time needed.The inclusion of a cooldown period of 2 hours resulted in ML model prediction RMSE being decreased significantly (∼50%) resulting in a mean RMSE of <5 MWd/kgU when the BoL fuel enrichment is known and <9 MWd/kgU otherwise.