Machine learning in critical heat flux studies in nuclear systems: A detailed review

Abstract

Accurate prediction and assessment of Critical Heat Flux(CHF) events are essential for reducing Departure From Nucleate Boiling ratio(DNBr), thereby ensuring reactor safety and enhancing the economic efficiency of nuclear power plants. The application of Machine Learning(ML) models for predicting CHF has garnered significant attention in recent research. These models are not only more accessible but also provide a broader applicability and greater accuracy than traditional empirical CHF correlations, making them a hot topic in this field. To disseminate the latest advancements in the application of ML techniques for predicting CHF and to refine these models for superior predictive performance. This paper reviews recent advancements in the application of ML techniques to CHF predicting, categorizing the research into CHF value prediction and CHF event identification. The synthesis covers the performance of various ML models under different conditions, examining input feature selection, model training, performance evaluation, addressing current deficiencies in data generation and model performance assessment. Additionally, the paper in CHF event identification summarizes the strengths and weaknesses of approaches. Finally, the perspectives of challenges and future directions in CHF studies are proposed. The review emphasizing the potential significance of these advancements for further exploration and application of ML in CHF research.