Multi-objective optimization of porous microchannel heat sinks with a novel T-shaped configuration based on machine learning

Abstract

Porous microchannel heat sinks can address high-power chips’ increasing cooling demands, thereby significantly boosting the performance of chips. In this paper, a novel T-shaped configuration is proposed for the first time to enhance the overall performance of porous microchannel heat sinks. In limiting cases, the T-shaped configuration can be converted into special configurations, including the traditional fully filled configuration, the partially filled configuration, and the sandwich configuration, which proves the extensive representativeness of the T-shaped configuration. Parametric analysis is conducted to examine the impact of the design parameters. The results show that the novel T-shaped configuration can enhance heat transfer while reduce flow resistance, and the performance evaluation criterion can be increased by up to 43.66% compared with the traditional partially filled configuration and up to 18.22% compared with the sandwich configuration when the filling ratio is 0.5. Based on machine learning, Design of Experiment and Particle Swarm Optimization-Back Propagation neural network model are utilized to quickly and accurately predict the flow and heat transfer performance. A comprehensive multi-objective optimization is conducted by employing non-dominated sorting genetic algorithm II to obtain optimal configuration variables. The optimization results reveal trade-offs between thermal performance and hydraulic losses, and demonstrate that a better overall performance can be achieved using a lower filling ratio based on the T-shaped configuration. Notably, this study improves the understanding and optimization of heat transfer in porous microchannel heat sinks.