MACHINE LEARNING ALGORITHM FOR PREDICTING HEAT TRANSFER COEFFICIENT AND PRESSURE DROP IN DIMPLED DUCTS

Recent advances in machine learning (ML) techniques have led to a shift in strategy for predicting the hydrothermal performance of thermal management solutions. This study presents the ML-based prediction of hydrothermal performances of water-cooled dimpled ducts using an artificial neural network (ANN). The significance of the present study is to develop the ANN model using a limited number of performance data without any existing relations/correlations between input variables and outputs. Thermal and hydrodynamic performances of the ducts are represented by heat transfer coefficient and pressure drop, respectively. The input dataset for training the ANN model was prepared through a computational fluid dynamics (CFD) approach. The accuracy of the ANN model was demonstrated as such it predicted heat transfer coefficients and pressure drops of new dimpled ducts within +/- 17% and +/- 19% of true values, respectively. The present study provides a practical insight to predict the hydrothermal performance of a thermal management solution subject to limited available datapoints, and without detailed knowledge about the complex thermo-fluid physics behind the operation of the cooling system.