Development of intelligent system models for prediction of licorice concentration during nanofiltration/reverse osmosis process

Abstract

Reverse osmosis (RO) and nanofiltration (NF) membranes in spiral wound configurations have been widely used in food processing ranging from dairy to fruit juice for concentration, purification and recovering valuable components. In this work, intelligent systems, i.e., back-propagation artificial neural network (BPNN), radial basis function (RBF), fuzzy inference system (FIS) and adaptive Neuro-fuzzy inference system (ANFIS) were employed to predict the water flux and solute rejection of RO and NF membrane during concentration of licorice solution. To develop the intelligent systems, normalized membrane type, temperature, pressure, pH and cross-flow velocity are taken as inputs while normalized permeate flux and rejection are as outputs of the models. The proposed intelligent systems have been compared based on statistical parameters of the coefficient of determination (R 2 ) and the mean absolute error (MAE). The results indicate that the ANFIS model is more accurate and reliable compared to the BPNN, RBF and FIS approaches. It was found that the predictions using ANFIS model were usually in good agreement with the experimental data, showing the R 2 values within the range of 0.932–0.997 and the MAE values in the range of 0.01–1.7%. On the basis of comparison among the results obtained from this investigation, it is suggested that the ANFIS model could be potentially utilized to forecast the rejection and permeate flux of membrane during the concentration process of a licorice solution.