Artificial neural network-based optimization of baffle geometries for maximized heat transfer efficiency in microchannel heat sinks

Abstract

Microchannel heat sinks (MCHSs) are widely utilized in various industries, including electronics, power systems, and aerospace, to dissipate heat generated by high-power components effectively. The efficient cooling of these devices is crucial for maintaining their optimal performance, reliability, and longevity. This research aimed to augment the overall efficiency of the MCHS by utilizing 25 baffles with varying geometric parameters. The aim was to identify the optimum geometry for these baffles using an Artificial Neural Network (ANN) model. The ANN model was applied to propose two optimal designs for the baffles, specifically targeting the maximum Nusselt number and overall performance of the MCHS. It was noticed that the vertical pitch of the baffles "d" had the most substantial influence on the device's behavior among the investigated geometric parameters. The higher value of "d" allowed the fluid to spread over the entire space of the heat sink rather than being confined to a specific area. This spreading effect enabled the fluid to contact all the baffles and channel walls, facilitating efficient heat transfer. The applied ANN model with R2 values of about 0.98 and lower values of MAE and RMSE successfully fitted the data. The findings of this study demonstrated that applying the specified input variables (horizontal pitch of 8.496 mm, vertical pitch of 5 mm, and attack angle of 210. 560°), as defined in the optimal overall design, in manufacturing of the baffles resulted in a remarkable enhancement of about 46% in the overall performance of the MCHS in comparison with the design proposed in the reference paper. The heat transfer improvement of the MCHS with baffles were due to the vortices in the baffles bring about chaotic advection and could greatly enhance the convective fluid mixing.