Reduction and control of flux decline in cross-flow membrane processes modeled by artificial neural networks

The aim of this paper is to present an innovative methodology for the control of permeate flux decay in membrane separation processes. An artificial neural network (ANN) has been built on the basis of the experimental results collected during ultrafiltration of BSA solutions under pulsating conditions. ANN can predict very accurately real system behavior with relative errors at most reaching 5% in post-simulation analysis. The observed reliability of neural network prediction, suggested an optimal control application of ANN aimed at searching a pulsation frequency profile that could maximize the permeate flux. The control system has been developed by the integration of two different computational environments, that allow actuating a specific control action so that UF experiments are performed in the exact conditions suggested by neural network. This control action is accomplished by a procedure that is invoked whenever the “best” operating time value, function of the current value of feed flow rate, is to be calculated. The generation of suitable pulses obtained on the basis of neural network prediction is, therefore, attained. Several experimental tests aimed at the validation of the present methodology have been performed on a lab-scale UF module. It was found that the utilization of an operating time profile allows obtaining a significant improvement of UF performance as compared to the cases in which either no pulse or constant pulse frequency were used.