Machine learning integrated higher-order model application for critical heat flux investigations in pressurized water reactors

Nuclear power station disasters like those at Chernobyl, Three Mile Island, and Fukushima Daiichi have highlighted how urgently improved nuclear safety is needed. This usually happened due to impeded cooling systems, resulting in heat accumulation, coolant boiling, and phase transformation leading to critical heat flux (CHF) events. Understanding bubble nucleation and dynamics during boiling heat transfer is crucial for ensuring the safety and reliability of pressurized water reactors (PWRs), particularly during postulated severe accident scenarios. Existing numerical models often struggle to accurately capture the complex multifluid interfaces and non-isothermal flow conditions inherent in these events, leading to potential inaccuracies in accident progression predictions. To address this gap, this study presents a novel numerical approach combining a high-order discontinuous Galerkin method (CVFEM), a conservative adaptive interface capturing method (CAICM), and a machine learning (ML) model (CVFEM+CAICM+ML/EoS). The ML component significantly enhances the accuracy of multifluid interface capturing in non-isothermal flows through precise fluid density evaluation, a key improvement over traditional methods. An adaptive mesh algorithm was implemented to optimize computational resource allocation, focusing on critical material interfaces. The model was validated against experimental data on single rising bubble dynamics, demonstrating its reliability. Analysis of dimensionless parameters, specifically the Galileo and Eötvös numbers, revealed the transition from laminar liquid flow to mixed vapor regimes, indicative of severe accident progression. This research provides a robust and validated tool for understanding complex boiling heat transfer mechanisms and bubble nucleation dynamics in PWRs, contributing to enhanced reactor safety.