Machine Learning Assisted Microchannel Geometric Optimization—A Case Study of Channel Designs

Abstract

At present, microchannel heat exchangers are widely applied in the fields of air-conditioning and heat pumping applications given their high heat transfer performance, compact size, and low material cost. However, designing and optimizing the channel geometries remain challenging, as they require balancing multiple competing objectives to achieve the optimal performance. This study investigates various parameters, including the channel count, wetted perimeter, cross-sectional area, and mass flow rate for each channel, to achieve the optimal efficiency. The optimization objectives include maximizing the heat transfer rate, minimizing the refrigerant convective thermal resistance, maximizing the refrigerant heat transfer coefficient, and minimizing the pressure drop. A multi-objective genetic optimization algorithm, in conjunction with artificial neural network (ANN)-based machine learning models, was used to predict the heat transfer rate to speed up the calculation process during the optimization. We identified that a gradient reduction in the wetted perimeter from the air inlet along the airflow direction could enhance the heat transfer rate. Additionally, the results indicate that an increase in the number of channels leads to an enhanced heat transfer efficiency rate. However, with the increase in the number of channels, the cross-sectional area of each channel is correspondingly reduced to maintain a consistent overall cross-sectional area. This reduction increases the fluid resistance, leading to an increased pressure drop across the system. This observation is critical for a microchannel design optimization, highlighting the importance of attaining a balance between achieving a higher heat transfer efficiency and maintaining a favorable fluid dynamic performance.