Machine Learning Approach to Predict the Effect of Metal Foam Heat Sinks Discretely Placed in a Cavity on Surface Temperature

Abstract

Metal foam heat sinks are preferred in electronic cooling systems with their advantages such as superior properties in heat transfer, light weight and ability to mix the cooling fluid. It is very difficult to conduct extensive experimental studies with metal foam heat sinks due to the difficulty of production and high cost. In addition, due to the complex structure of metal foam heat sinks, difficulties may arise in the creation of numerical simulations. In the present study, various machine learning methods were used, taking into account the mean surface temperature values obtained by using metal foam heat sinks discretely placed in a partially open volume. The pore density of metal foam heat sink, Reynolds number, modified Grashof number and distance to aperture were taken as input parameters. When the results were examined, it was determined which of the inlet parameters were more effective on the mean surface temperature. It was determined that modified Grashof number was the most effective parameter on mean surface temperatures, but L was the weakest parameter. The models were ranked according to 3 different evaluation metrics. It was observed that the top three most successful machine learning algorithms were eXtreme gradient boosting, support vector machine and random forest.