Machine learning and computational fluid dynamics based optimization of finned heat pipe radiator performance

This study integrates machine learning, genetic algorithm and computational fluid dynamics to explore the influence of geometry on the finned heat pipe radiator performance. It also proposes an optimal geometry that enhances the heat transfer of the radiator significantly. Firstly, orthogonal test were designed, coupled the thermal resistance network method and the CFD method to model the finned heat pipe heat sink. The analysis of variance yielded the following qualitative conclusions: the height, number and thickness of the fins had different effects on the heat source temperature in descending order; the heat source temperature decreased with increasing number and thickness of the fins and increased slightly with increasing height. Secondly, a prediction model of the heat source temperature under different heat source distributions was built using four MLAs. The artificial neural network was chosen as the optimal model by comparison. The model was analysed with correlation analysis, SHAP and PDP analysis and consistent conclusions with the orthogonal test were obtained. Finally, GA was used to obtain the optimal geometric structure. The results showed that the heat source temperature was reduced to 41.75 degrees C and the heat sink mass was only 0.2231 kg when the number of fins was 18, the length was 38 mm, and the thickness was 0.6 mm. This solution decreased the heat source temperature by 8.7% and the heat sink mass by 18.42% compared with the optimal solution from the orthogonal test. This study demonstrated the advantages of machine learning methods in computational fluid dynamics.