A wall-temperature-independent heat transfer model and dominant dimensionless numbers for supercritical CO2 in vertical upward flow

Abstract

The heat transfer coefficient is a critical parameter for the design and optimization of supercritical carbon dioxide (sCO₂) heat exchange systems. Based on 20723 data points collected from 22 published studies, existing heat transfer correlations are evaluated, and a new model with improved stability and predictive accuracy is developed. Current correlations for upward vertical sCO₂ flow typically include wall-temperature-dependent parameters, which often lead to issues such as non-uniqueness or the absence of solutions during wall temperature prediction, along with limited accuracy. To address these issues, this study integrates the Buckingham Pi theorem and constructs a wall-temperature-independent heat transfer model using a dimensionless neural network. This approach avoids the problems of multiple or no solutions and achieves high predictive performance, with mean absolute relative errors of 2.83 % for wall temperature and 5.81 % for Nusselt number predictions. Furthermore, the active subspace method is employed to identify four dominant dimensionless groups governing the heat transfer process. These groups can be decomposed into products of commonly known dimensionless numbers. During this process, a new dimensionless number—the buoyancy generation (BG) number—is proposed and defined. This study leverages data-driven dimensional analysis to explore key influencing parameters and dominant dimensionless numbers for supercritical heat transfer, offering new insights into the underlying physical mechanisms.