PREDICTION OF CRITICAL HEAT FLUX FOR LIQUID HELIUM USING MACHINE LEARNING MODELS ASSISTED BY PHYSICS-BASED CORRELATION

Liquid helium plays an important role in fields such as superconductivity, quantum computing, medical cryopreservation, and space exploration. Prediction of pool boiling heat transfer, especially the critical heat flux, is crucial for all these applications. To date, no effective tool is available to give accurate predictions of critical heat flux specific for liquid helium. In this study, we use both traditional machine learning models and physics-assisted machine learning models to predict critical heat flux of liquid helium with 585 data sourced from 22 studies. While the traditional machine learning models possess no information on critical heat flux, the physics-assisted machine learning models obtain physical information of critical heat flux from Zuber [2]'s correlation by predicting the multiplier of the hydrodynamic instability term. Recursive feature elimination and stepwise selection are used to remove the redundant features while multivariate adaptive regression splines, support vector machine, random forest, and extreme gradient boosting are used for modeling. Support vector machine, random forest, and extreme gradient boosting result in better performance than traditional correlations. Physicsassisted extreme gradient boosting model achieves the best result with 14% mean absolute percentage error. This work is the first attempt in reported literatures to predict critical heat flux for liquid helium using machine learning models.