Thermal characteristics of in-tube upward supercritical CO₂ flows and a new heat transfer prediction model based on artificial neural networks (ANN)

The potential employment of supercritical carbon dioxide (sCO₂) flows in heated tubes in many applications requires accurate and reliable predictions of the thermal characteristics of these flows. However, the ability to predict such flows remains limited due to a lack of a complete fundamental understanding, with traditional prediction capabilities relying on either simple empirical correlations or highly complex and computationally demanding simulation methods both of which limit the design of next-generation systems. To overcome this challenge, a prediction model based on artificial neural network (ANN) is proposed and trained by 5780 sets of experimental wall temperature data from upward flows with a very satisfactory root mean square error (RMSE) and mean relative error that are less than 1.9 °C and 1.8%, respectively. The results confirm that the structured model can provide satisfactory prediction capabilities overall, as well specific performance with mean relative error under the normal, enhanced and deteriorated heat transfer (NHT, EHT and DHT) conditions of 1.8%, 1.6% and 1.7%, respectively. The proposed model's ability to predict the heat transfer coefficient in these flows is also considered, and it is shown that the mean relative error is<2.8%. Thus, it is confirmed that it has a better prediction accuracy than traditional empirical correlations. This work indicates that such ANN methods can provide a real alternative for adoption in select thermal science and engineering applications, shedding a new light and giving added insight into the thermal characteristics of heated supercritical fluids.