Metamodel-based design optimization for heat transfer enhancement of finned heat sinks

Abstract

Finned heat sinks are a highly efficient means of dissipating heat from electronic devices. Under constant power and operating temperature, it is ideal to choose the heat sink with the minimum thermal resistance. However, in some instances the desired heat sink is not suitable due to space constraints. This paper explores a heat sink optimization strategy that optimizes the heat transfer coefficient in order to achieve the compromise between heat sink temperature and heat sink size. The optimization strategy employs computational fluid dynamics simulations to examine the impact of heat sink dimensions, including length, width, fin spacing, and height, on heat sink thermal performance. A Latin hypercube sampling method is used to generate 100 heat sink variations of height, width, length and spacing between the fins. The width and length of heat sink are varied between 42 mm and 46 mm. The fin height varies between 4 mm and 11 mm and the fin spacing varies between 4 mm and 6 mm. The metamodel used for this study is a decision tree model called Random Forest. This metamodel is constructed by running numerical simulations of the 100 heat sink variations and coupled to an optimizer algorithm. The goal of the optimization algorithm is to search for the optimal heat sink design with maximum heat transfer coefficient. The optimal solution is validated by conducting an experiment to measure the heat transfer coefficient of the optimized heat sink and compared against the baseline model. Experimental results show that the optimized model exhibits a 35 % increase in heat transfer coefficient compared to the baseline model. Furthermore, the fin height was reduced by 43 %. The volume of the heat sink is decreased by about 26 %, resulting in a space-saving effect. On the other hand, the temperature increase penalty occurred due to space reduction is about 3 %.