Data-Driven Deep Learning Model for Critical Heat Flux Occurence Prediction Based on Bubble Dynamics in Pressurized Water Reactors

Prediction of critical heat flux (CHF) in thermal systems, such as heat pipes, heat exchangers, and pressurized water reactors (PWRs) is crucial for ensuring reactor safety and operational efficiency. Traditional CHF prediction methods often rely on empirical correlations with limited applicability, failing to capture the complex, multi-scale physics of boiling phenomena. This study addresses this challenge by introducing a novel deep learning (DL) framework that directly leverages detailed bubble dynamics extracted from high-fidelity computational multifluid dynamics (CMFD) simulations to predict the likelihood of CHF occurrence. This approach moves beyond simplified correlations by harnessing the extensive information contained within the simulated boiling process.
The novelty of this work is the development of an artificial intelligence (AI)-based predictive algorithm that directly integrates advanced bubble flow and heat dynamics. It achieves this by combining a machine-learning parameterized equation of state (EoS) with a robust coupled control volume and finite element method (CVFEM) model and a compressive advection interface capturing scheme (CAICM), forming self-adaptive, high-order accurate methods for detailed and precise CMFD simulations. Simulations provide a comprehensive dataset of bubble dynamics, which are then analyzed using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to extract key features. A DL model, trained on these extracted features, predicts CHF occurrence. This data-driven approach offers a significant advancement over existing methods by directly learning the complex relationships between bubble dynamics and CHF, providing a more accurate and physically grounded prediction. This high-accuracy framework improves PWR safety analysis and boiling heat transfer prediction for efficient reactor design.