Using of Artificial Neural Network and COMSOL Multiphysics for Modeling of Pervaporation Process

Authors: Mansoor Kazemimoghadam, Zahra Amiri,
Abstract: The researcher in this study used a Feed Forward multilayer perceptron neural network with a back propagation algorithm and Levenberg-Marquardt function with two inputs and two outputs. The Tansig transfer function was used for the hidden layer and Purelin was used for the output layer; five neurons were defined for the hidden layer. After data processing, 70 percent of the data was allocated for learning, 15 percent was allocated for validation, and 25 percent was allocated for testing. The output values of Artificial Neural Network modelling were compared with the real values of pervaporation for separation of water from Ethanol, Acetone, and Butanol. The results revealed that the proposed model had a good performance. Moreover, the output of COMSOL software for pervaporation of five different alcohols was compared with the real values, and the error percentage of the actual amount of flux was calculated with the modeling value by means of related membranes. The results of COMSOL modeling showed that the error percentages of 3.049, 3.7, 3.51, 2.88, and 3.82 were respectively achieved for dehydration process of Acetone, Butanol, Ethanol, Isopropanol and Methanol.