Finned Heat Sink Optimization Method with the Use of Computational Intelligence

With the technological advancements of recent years, electrical and electronic equipment has become increasingly faster, incorporating new functions while drastically reducing in size. This has led to higher power densities and heat fluxes in their components. To maintain the operating temperature of such equipment within acceptable limits - without compromising safety or reliability - the use of finned heat sinks is essential. In this context, the present work conducts a theoretical and numerical optimization analysis on the impact of geometric variables - such as fin height, width, and number -on the thermal performance of finned heat sinks used in the cooling of electrical and electronic devices. For this analysis, a straight-fin heat sink configuration was validated using experimental data available in the literature. The Generalized Subset Design (GSD) method, a Design of Experiments (DOE) approach, was employed to select simulation scenarios efficiently and reduce computational cost. All cases were simulated using commercial Computational Fluid Dynamics (CFD) software. The resulting data was used to train Machine Learning (ML) models developed in Python, an open-source programming language. The selected ML models included decision tree regressors, support vector regressors, multilayer perceptrons, and a neural network developed from scratch using Keras. A theoretical analysis of each variable was performed, and its impacts on the key thermal performance metrics were presented graphically. It was observed that both the number and width of the fins have an intermediate optimal value within the tested range, indicating non-linear behavior. The trained ML models were applied to identify behavior patterns for each geometric variable and to estimate their effect on the thermal performance of finned heat sinks.