A Physics-Informed Machine Learning-Aided Framework for Predicting Departure from Nucleate Boiling in Rod Bundles

摘要

This summary extends prior work on the physics-informed ML-aided (i.e., hybrid) framework and evaluates its performance against that of the standalone DK and ML models for the prediction of DNB in rod bundles. Both NN and RF methods are trained and cross-validated with the PSBT bundle DNB dataset that covers various boundary conditions and spacer grid configurations. The predictive capabilities are clearly in favor of the hybrid approach, as it yields reduced validation errors and corrects significant scatter or bias found with its standalone counterparts.