Multi-Objective Optimization of a Diamond-Based U-Type Counter-Flow Manifold Microchannel Heat Sink Using Neural Network-Assisted Nsga-Ii Algorithm

Abstract

This study presents a comprehensive multi-objective optimization of a diamond-based U-type counter-flow manifold microchannel heat sink (U-type CMMC) for ultra-high heat flux applications. A systematic optimization framework integrating Sobol sensitivity analysis, surrogate modeling, and NSGA-II algorithm was developed to simultaneously optimize thermal resistance, pumping power, and entropy generation. Analysis of five key geometric parameters revealed that fin width dominated thermal resistance (71.49%), while microchannel width exhibited the strongest influence on entropy generation (75.24%) and pumping power (52.18%). The established BP neural network surrogate model achieved prediction accuracy exceeding 99%. The tri-objective optimization incorporating entropy generation achieved 46.7% reduction in thermal resistance and 25.2% decrease in entropy generation compared to the initial design, while the bi-objective optimization scheme reduced thermal resistance by 44.1% with only 13.3% increase in pumping power. Temperature uniformity analysis showed that the optimized design significantly improved temperature distribution, with non-uniformity reduced by 34.82%. This study provides valuable insights into the design optimization of high-performance microchannel heat sinks and establishes a robust methodology for multi-objective thermal management optimization.

Keywords: Manifold microchannel heat sink, Neural network surrogate model, Multi-objective optimization, Sobol sensitivity analysis, Thermal-hydraulic performance