Rapid 2-Dimensional prediction of supercritical CO2 heat transfer behaviors in inclined tubes based on deep learning

Abstract

At present, there is a dearth of empirical correlations for the prediction of the inclined supercritical heat transfer. Considering the shortage of the rapid prediction method in the initial design stages for the supercritical fluid heat exchanger with inclined tubes or under inclined conditions, a 2-Dimensional rapid prediction method based on the numerical method combining deep learning method was proposed to reconcile the accuracy and the efficiency of the inclined supercritical heat transfer behaviors prediction. In order to train the deep learning model, 200 numerical cases with 520,000 data were conducted. After that, 7 new cases were chosen to analyze the inclined heat transfer behaviors and validate the current model. The results were indicative of a significant circumferential inner wall temperature difference in the inclined supercritical heat transfer. The temperature difference was larger when the inclined angle was closer to 90°, even though the circumferentially averaged temperature is smooth and acceptable. This potential temperature difference would not only cause large thermal stress which may accelerate fatigue failure, but also increase the risk of over-temperature. Fortunately, the current deep learning model could predict the 2D temperature distribution of these new cases with the max absolute relative error of 16.60%. Therefore, the circumferential inner wall temperature distribution of the inclined supercritical CO2 heat transfer could be evaluated during the initial heat exchanger design. Moreover, the trained model and the relative files are available at Supplementary materials.