ARTIFICIAL NEURAL NETWORK BASED MODELING OF MICROFILTRATION PROCESS FOR WHEY PROTEIN CONCENTRATION

Abstract

Microfiltration is one of the most suitable processes for protein recovery from whey due to low energy consumption and no use of heat and chemicals. However, membrane fouling is one of the limiting factors in the microfiltration process preventing its commercial use. In this study, an Artificial Neural Network (ANN) based model was employed to study the effects of different operating parameters on the membrane fouling in whey concentration. Transmembrane pressure, Reynolds number, and temperature of feed were selected as the input parameters. Experimental data from the available studies were used to train ANN. ANN with 23 neurons gave minimum mean squared error (MSE) for trans-membrane pressure and Reynolds number. ANN with 7 neurons gave minimum MSE for feed temperature. Predicted values from both ANNs well fitted with the experimental results with R2 < 0.99. Simulations showed that membrane fouling increased as flux reduction increased from 36.3 % to 76.39 % when trans-membrane pressure increased from 0.5 to 2 bar. While a 19.96 % reduction in flux was observed by increasing the Reynolds number from 750 to 2500. An increment of 77.37 % of flux was observed with increasing feed temperature from 30 ºC 40 ºC. Simulations confirmed that trans-membrane pressure, Reynolds number, and temperature of feed all three operating parameters strongly influence the membrane fouling. ANN based approach was found most accurate results in comparison to theoretical models. Among all theoretical models, the intermediate blocking model gave the most accurate results with a mean relative error of 0.185.