Optimizing finned-microchannel heat sink design for enhanced overall performance by three different approaches: Numerical simulation, artificial neural network, and multi-objective optimization
Sahar Nekahi,Farhad Sadegh Moghanlou,Kourosh Vaferi,Hadi Ghaebi,Mohammad Vajdi,Hossein Nami
DOI: https://doi.org/10.1016/j.applthermaleng.2024.122835
IF: 6.4
2024-03-04
Applied Thermal Engineering
Abstract:Microelectronic devices with multifunctional capabilities have become an indispensable part of the modern life. These devices generate significant thermal energy during continuous use and elevate the chip temperature. Therefore, the need for high-efficient microchannel heat sinks as an innovative cooling solution has become more crucial than ever. This study utilized a multi-nozzle microchannel heat sink incorporating six distinct fin shapes to determine the optimal fin design. Initially, all six shapes were numerically simulated, and the one that provides the highest heat transfer and the lowest pressure drop was selected as the candidate for further optimization. Then, 27 tests were designed to examine the effect of the optimal fin's geometric parameters on the Nusselt number and pressure drop, and the obtained data were utilized for training the artificial neural network and response surface methodology models. Using these model, three geometric parameters of the chosen fin such as length ( L f ), horizontal pitch ( W bf ), and vertical pitch ( H bf ) were optimized in specified ranges. In the last step of optimization process, a single-objective optimization with three different goals: maximizing thermal performance index, maximizing Nusselt number, and minimizing pressure drop , and a multi-objective optimization aiming to find the right balance between Nusselt number and pressure drop were carried out by the neural network model and genetic algorithm. Besides, Pareto fronts of the Nusselt number and pressure drop were presented to show the simultaneous impacts of these objectives. Finally, three optimal designs for different conditions were anticipated. R-squared values near 1 illustrated that the trained neural network model had high accuracy in predicting the performance of the device with straight-slot fins. Among the suggested designs, D1 ( W bf = 145 μm, H bf = 15 μm, and L f = 110 μm) improved the overall system performance by 13.52 % compared to the reference heat sink.
energy & fuels,engineering, mechanical,thermodynamics,mechanics