Verification of the PWR core solver coupling the neutron code nTRACER with artificial neural networks for thermal–hydraulic feedback

The Paul Scherrer Institute (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 presents the coupling of nTRACER/ML, comparing the solver with the verified core solver nTRACER/CTF, in terms of accuracy and computation costs, for steady-state and cycle analysis, with the presented results and their interpretation focusing primarily on performance improvements. nTRACER/ML shows strong agreement with nTRACER/CTF in power and coolant predictions for hot full power, with maximum deviations of 1.01% in assembly power and 2.63 °C in coolant temperature (root mean square error [RMSE]: 0.65 °C). Fuel temperature predictions are also good, with centerline temperature differences reaching up to 32.96 °C and an RMSE of 5.55 °C. As exposure increases, power deviations grow up to 3.33% axially and 4.13% in 3D assembly power at 392.3 effective full power days (EFPDs). Coolant temperature discrepancies decrease with burnup, while fuel temperature errors increase, with centerline differences peaking at 36.24 °C (RMSE: 8.30 °C). Despite its limitations, nTRACER/ML shows sufficient agreement with nTRACER/CTF in both neutronic and T/H metrics, making it a practical low-fidelity alternative for high-resolution simulations, particularly because it runs 3–4 times faster than CTF while using only approximately 1% of the CPU time.