Investigation of Universal Consolidated Database for Heat Transfer Coefficient in Flow Condensation and Machine Learning Modelling

Abstract

Abstract Miniature condensers have been recognized as an effective tool for developing compact heat rejection devices. However, due to complex physical behaviors resulting from phase change, predicting necessary quantities like heat transfer coefficient has proven to be a difficult task. This study will use a database of 37 flow condensation studies for development of machine learning tools to predict two-phase heat transfer coefficient. In comparison to previous work performed by Zhou et. al. [1], this study will utilize thermodynamic data, dimensionless flow data, channel type, channel geometry, single and multichannel systems, and fluid species, as a more comprehensive means of predicting heat transfer coefficient. This study finds that in conjunction with the dimensionless quantities used in previous studies [1], the variables; mass velocity, quality, latent heat, specific volume, and hydraulic diameter are all of important significance to accurately predicting heat transfer coefficient. This study is a comprehensive Exploratory Data Analysis (EDA) as well as a machine learning based model for accurately predicting heat transfer coefficient. We find Mean Absolute Percent Error (MAPE), using simple multi-linear regression, of significantly less magnitude using our more comprehensive database as compared to previous works. Machine Learning methods also show a reduction in MAPE as compared to previous studies, as well as compared to previous semi-empirical correlations for heat transfer coefficient.