Artificial Neural Networks Application on Average and Stagnation Nusselt Number Prediction for Impingement Cooling of Flat Plate With Helically Coiled Air Jet

Abstract

In order to estimate the average and stagnation Nusselt numbers for turbulent flow for impingement cooling of a flat plate with a helically coiled air jet, a new artificial neural network (ANN) model is presented in the present study. A new dataset of stagnation and average Nusselt numbers as a function of Reynolds number (Re) varied from 5000 to 30,000, nozzle plate spacing ratio changed from 2 to 8, and jet helical angles of 0 deg, 20 deg, 30 deg, 40 deg, and 60 deg was created based on an experimental investigation. The ANN structure is composed of three layers with hidden neurons of 14–10–8. The training process comprises feed-forward propagation of the selected input parameters, back-propagation with biases and weight adjustments, and loss function evaluation for the training and validation datasets. The activation function of the output layer is a linear function, and the rectified linear unit activation function is utilized in the hidden layers. The adaptive moment estimation algorithm is employed to minimize the loss function to accelerate the ANN training. To prevent an increase in training time caused by the marked discrepancy in the gradients of loss function considering the values of the weights, the “MinMax” normalization strategy was used. For the ANN model, the mean absolute percent error values were 2.35% for the average Nusselt number and 2.52% for the stagnation Nusselt number. According to the comparison of projected data with the outcomes of earlier experiments, the derived model’s performance was validated and the findings showed outstanding accuracy.