Development and Uncertainty Quantification of Machine Learning Models for Critical Heat Flux predictions

Critical Heat Flux (CHF) represents a concern for nuclear safety, as it leads to a rapid drop down in the heat transfer between a heated surface and the liquid coolant in the core of nuclear reactors. This could cause several issues to the system, including structural damage and release of radioactive material. The main challenges related to CHF predictions are the highly non-linear relationships with the physical features it depends on, and the two different underlying, microscopic phenomena that lead to CHF: Departure from Nucleate Boiling (DNB) and dryout, occurring under different conditions but not distinguishable externally, both resulting in the same outcome. Both for these reasons and for the inaccuracy and limited applicability of current physical correlations, the OECD/NEA Expert Group on Reactor Systems Multi-Physics (EGMUP) has established a task force to develop machine learning (ML) strategies for CHF regression, along with uncertainty quantification (UQ) techniques to assess the reliability and the robustness of these models. In this research, the CHF database provided by the U.S. Nuclear Regulatory Commission (NRC) is utilized to develop machine learning (ML) methods for CHF prediction and robust uncertainty quantification (UQ) techniques. The performance of the ML models is assessed against established data-driven strategies, while a coverage-based approach is considered for UQ methods by using conformal prediction, (Bayesian) heteroscedastic regression, and a quality-driven loss function. It is found that it is possible to confidently estimate the prediction uncertainties with a well-calibrated 95% coverage of experimental CHF values.