Structural optimization of multi-heat-source cold plates using neural network and genetic algorithm

This study proposes an integrated framework coupling neural networks with the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) for the independent multi-zone optimization of fin structures in both straight-fin and hybrid-fin cold plates, addressing the thermal management challenge of multi-heat-source configurations. The framework systematically investigates how zonal fin parameters regulate thermo-hydraulic performance, with optimization conducted under two distinct objectives. Results vs. the initial design show: minimum-pressure-drop designs cut flow resistance by 47.4% (straight-fin) and 52.9% (hybrid-fin), while balanced designs reduced it by 28% and 38.4% respectively, with improved temperature uniformity. Direct comparison with traditional global optimization (uniform fin parameters) shows the traditional method only cut flow resistance by 19.3%, versus 34.9–41.7% for ours, plus targeted thermal management unavailable to uniform designs. Hybrid-fin designs exhibit superior overall performance, particularly in zone R2 (high-power devices), enhancing heat dissipation without hydraulic compromise. Mechanistic analysis reveals outlet zone ( R1 ) parameters dominate pressure drop due to local resistance losses, inlet zone ( R3 ) is sensitive to downstream undeveloped flow, and intermediate zone ( R2 ) behaves thermally independently with fully developed flow. Fin count modulates convective intensity via area-velocity effects, while fin thickness regulates thermal conduction and uniformity. This zonal control methodology cuts pumping power while ensuring thermal safety for uneven heat sources. The framework provides scalable optimization for multi-heat-source cold plates, enabling flexible strategy formulation per application-specific thermo-hydraulic priorities.