Machine learning approach to predict heat transfer and fluid flow characteristics of integrated pin fin-metal foam heat sink

Abstract

Integrated pin fin-metal foam (PF-MF) heat sink offers enhanced thermal performance owing to the higher surface area and flow mixing. By realizing the strong influence of geometrical parameters of PFs, morphological properties of MF, and coolant hydrodynamics on the thermal-hydraulic performance of the PF-MF heat sink, this article presents a machine learning (ML) surrogated multiparameter and multi-objective optimization framework to determine optimal PF-MF heat sink layouts within the laminar flow regime to enhance the thermal performance of the heat sink with a minimal pumping power penalty. Four supervised regression ML models, namely random forest, extreme gradient boosting, support vector regressor, and artificial neural networks (ANN), were considered while searching for the best surrogate. Four geometrical parameters of PFs, such as PFs diameter (D), height (h), longitudinal pitch, transverse pitch, two morphological properties of MF, i.e., pore density porosity and coolant flow condition, i.e., Reynolds number have been considered as design parameters. The heat transfer and fluid flow performance of the heat sink are evaluated using nondimensional Nusselt number and friction factor A high-fidelity 3-D computational fluid dynamics/heat transfer (CFD/HT) model has been developed to evaluate the thermal-hydraulic performance of all PF-MF heat sink layouts. Furthermore, utilizing the numerical data, ML models have been trained and fine-tuned to accurately predict the overall performance of the PF-MF heat sink. Upon testing the prediction and interpolation capability of all considered ML models, ANN outperforms other models, hence, ANN model has been coupled with genetic algorithm (GA) optimizer to identify three different optimized PF-MF heat sink layouts at three different Finally, the result shows that the ANN surrogated GA-optimized PF-MF heat sink layout can improve the overall thermal-hydraulic performance by approximately 19–35% compared to the best PF-MF heat sink layout among the used input feature space for ML model.