Prediction and analysis of thermal-hydraulic performance of tubes with teardrop dimples based on artificial neural networks

In this study, the prediction and analysis of the thermal-hydraulic performance of tubes with teardrop dimples were carried out. First, a numerical model of dimpled tubes using ANSYS 17.0 was established and verified by comparison with the Nusselt number (Nu) and friction factor (f) in the literature. Second, artificial neural networks (ANNs) were developed with five neurons in the input layer, the hidden layers of various neurons, and one neuron in the output layer. The five neurons are the diameter of the windward sphere (d(1)), diameter of the leeward sphere (d(2)), dimple depth (h(d)), spacing between two axially adjacent dimples (l), and quantity of dimples in a transverse cross-section (N). After careful comparison, the model with the 5-2-6-1 structure was selected for predicting both f and Nu, while the model with the 5-6-1 structure was selected for performance evaluation criteria (PEC). Finally, the influence of d(2)/d(1) on the heat transfer and pressure drop is discussed when l/d(1) or N changes. The results showed when N is moderate or l/d(1) is large within the scope of the design, the heat transfer performance in the windward part of the teardrop dimples is better than that of the spherical dimples. In addition, as N increases or l/d(1) decreases, the influence of dimples on the main flow increases, making the mixture of hot and cold working fluid better but causing a greater pressure drop. This study may provide ideas and guidance for the design, selection, and optimization of dimpled tubes.