Abstract
In this study, artificial neural network (ANN) was used to predict forced convection heat transfer enhancement from microscale pin fin heat sinks under various operating conditions and at different geometries. Experimental data acquired from the literature were employed to train the ANN model. Different numbers of training cycles, layers, and neurons in the network were investigated to design an effective neural network. The performance of the trained network was tested using the mean squared error and the most accurate one was selected to predict the average Nusselt number. Mean absolute percentage errors between the ANN output and the experimental datapoints were calculated and compared with the existing correlations. This machine learning method offers significant improvement for thermal performance prediction and could be a reliable tool for determining convective heat transfer in microscale pin fin heat sinks.