چكيده
This study reports the development and application of Artificial Neural Network (ANN) model for prediction of initial dilution of multi-port tee diffusers. In this study, four basic parameters of initial dilution were given to neural networks as inputs and the value of initial dilution obtained from laboratory measurements reported by Seo et al. was given to neural network as a target. Then, the second data set (non-dimensional) were used for training and testing the networks too. In second data set a one input network was created to compare with previous studies in which the equations had one variable. In this study, the standard three-layer feed forward back propagation network with a non-dimensional differentiable log-sigmoid transfer function in the hidden layer is used. The network programming was done by using the computer software, Matlab and the Neural Network Toolbox. Furthermore, a sensitivity study was conducted to examine the effect of the network size (number of neurons in hidden layer) on the model performance. Sensitivity analysis in the first data set (dimensional parameters) shows that the ambient current has a greater influence on initial dilution than the other independent parameters and in second data set (non-dimensional parameters) shows that Rv , suggested by Azimi and Etemad-Shahidi has a greater influence on initial dilution than the other non-dimensional parameters. The results indicate that the ANN model is capable of predicting the initial dilution of multiport diffusers successfully. The results of the ANN model were also compared with an empirical model (CORMIX), showing the superiority of the ANN approach.
