Analysis of mean permeate flux for ultrafiltration process by two methods: resistance-in-series model and artificial neural network
Volume Title: 1
Chemical engineering dept., Faculty of engineering, University of Guilan, Rasht, Iran.
Two resistance-in-series and artificial neural network (ANN) methods have been studied to compute the amount of mean permeate flux of a hollow-fiber module for ultrafiltration of PVP (360) aqueous solution. The present resistance-in-series model includes two types of pressure distribution equations. In model 1, momentum balance with term of convection has been considered. In model 2, the pressure distribution has been obtained by Hagen Poiseuille equation. The total AAD% and MSE in model 1 are 7.47 and 0.0966. These numbers are a bit lower than model 2 with AAD% 7.53 and MSE equals to 0.0975. The calculated amounts of AAD% and MSE in the lowest velocity, 0.0723 (m/s), are 5.2 and 0.042 and in the highest velocity, 0.2195 (m/s), are 12.5 and 0.4. Thus in both models, the errors and deviations has been arised by increasing feed velocities. In ANN method, input data has been chosen feed velocity, feed concentration and transmembrane pressure and the output data has been selected mean permeate flux. Back-propagation ANN model has been developed with a tree-layer network and Levenberg-Marquardt (LM) as learning algorithm. The transfer function is tansig and six different structures are examined which the number of neurons in hidden layer are varying between 15 and 20. The network with 17 hidden neurons have showed good performance with least deviation and error (AAD%=6.14 and MSE=0.0004). As the amounts of R2 are between 0.87 and 0.99, all structures have got the acceptable proportion of variances.
Ultrafiltration Membrane; Resistance-in-Series Model; Artificial Neural Network; Momentum Balance; Hagen Poiseuille Equation