High-resolution predictions of the coolant properties for the 3D PWR core with artificial neural networks based on CTF

PSI and North Carolina State University are developing a high-resolution multi-physics core solver for Pressurized Water Reactor (PWR) analysis in Cartesian geometry, using the neutron transport code nTRACER and two Machine Learning (ML) models providing thermal-hydraulic (T/H) feedback. This work focuses on the ML models, trained with CTF data to predict PWR subchannel coolant properties during normal operation. The methodology presented can be applied to different PWR core, to produce ML models capable of high-resolution T/H predictions. The ML model's performance is evaluated on quarter-core CTF calculations achieving an average temperature difference of 1 degrees C from CTF and equivalent density accuracy. Their verification is extended outside configurations seen in their training, with varying mass flux, a water liner and different lattice geometry. They are also compared to a simplified one-dimensional T/H solver, showing significantly lower discrepancies, almost half, with similar computational cost, while being up to 15 times faster than CTF.