Machine-learning-based modeling of saturated flow boiling in pin-fin micro heat sinks with expanding flow passages

Abstract

For high potential flow-boiling-based thermal management systems, to better understand the underlying flow physics and to present an effective predictive approach have critical importance. Different from the existing literature, this study, for the first time, takes the machine learning (ML) algorithms into consideration for flow boiling in expanding type micro-pin-fin heat sinks (ETMPFHS). A new database including saturated flow boiling data in ETMPFHS is obtained for various operational conditions. Mass flux (G = 150, 210, 270 and 330 kg m−2 s−1), inlet temperature (Ti = 40, 49, 58, 67 and 76 °C) and effective heat flux (approximately, qeff″= 241 to 460 kW m−2) are the variable parameters. In this study, advanced ML algorithms including Support Vector Machine (SVM), Artificial Neural Network (ANN), Regression Trees (RT) and Linear Regression (LR) are used. It is concluded that, for flow boiling in ETMPFHS, the ANN emerges as the most effective model for prediction of htp, ΔT, and ΔP, followed by SVM, while RT and LR present poorer results in terms of predictive accuracy and reliability. Trends of predictions of both the ANN and SVM nearly overlap the experimental data; while both the RT and LR show different trends against the experimental results.