Selection of the Training Function of Artificial Neural Network to Predict the Behavior of the Vortex Tube

One of the powerful tools in predicting system behavior and responses in the field of engineering is the neural network. In the present work, using a time-dependent, dynamic neural network, the available data from the laboratory work is processed to study the turbulence parameters. The neural network model is a feed forward model and the Lunberg-Marquardt algorithm is used for the training function. The data obtained from the two-dimensional CFD modeling of the vortex tube are used as input data and neural network training. The network is selected and trained with appropriate and relatively accurate conditions, and the results of the cold outlet temperature are predicted and compared according to the flow rate of the vortex tube inlet at different pressure ratios. The results show the adaptability and high accuracy of the defined neural network.