Physics and correlations informed deep learning to foresee various regimes of the pool boiling curve

Pool boiling curve is the key feature for performance evaluation of phase-change equipment such as Kettle reboiler, flooded evaporator, falling film evaporator, immersion cooling, and the like. The performance evaluation typically relies on empirical correlations or some semi-analytically based theoretical approaches. However, it has not been possible to establish a good relationship between heat flux and temperature until now. The pool boiling curve of smooth and roughened surfaces, including the nucleate boiling region, critical heat flux (CHF), and transition boiling, have been precisely predicted for the first time using a generalized physics-based deep learning (DL) model that considers physics and correlation-based features. To predict the higher order pool boiling curve for a specific liquid-surface combination, the model is fed the initial experimental data from the lower order boiling curve in addition to the most important variables of the surface morphology, thermophysical properties of the working fluid, and pool boiling conditions. Various regimes of pool boiling curve, such as fully developed nucleate boiling, CHF, and transition boiling for the investigated liquid-surface combinations and testing conditions, are predicted with an accuracy of R² (correlation coefficient) = 0.98. The reason for such a high accuracy to a wide range of surface morphologies, substrate materials, operating pressures, heater inclination angles, and working fluids is due to the inclusion of the initial pool boiling data along with the relevant, impactful surface, and liquid parameters into its framework.