Prediction of heat transfer of bubble condensation in subcooled liquid using machine learning methods

Abstract

Owning to the complex behaviors of phase change, precise modeling of bubble condensation is challenging. In this study, nine machine learning (ML) methods were applied to develop a generalized model of bubble condensation. First, experimental investigation on the bubble condensation in a subcooled pool was performed to collect data points for bubble Nusselt number (Nuc). Afterward, combining the data from the present experiments and six previous sources, a dataset of the bubble Nusselt number with 5170 data points was compiled. Using this database, ML models were compared with the existing correlations, showing significant superiority in prediction performance. Among the ML models used, extreme gradient boosting has the lowest mean absolute error (MAE) of 16.5%, which is about half of the lowest MAE obtained by the existing correlations. Parametric analysis showed that ML models provide accurate predictions while offering insight into the physical variation trend of Nuc for various experimental conditions.