Neural network models for ultrafiltration and backwashing

Abstract The possibility of predicting the time dependence of flux evolution for ultrafiltration systems is one of the most important factor that influences large-scale applications, optimization and automation, because flux affects productivity and thus all cost components. This paper presents the development and exploitation of mathematical models based on artificial neural networks, trained with experimental data obtained in a laboratory scale ultrafiltration system. Hollow-fibre ultrafiltration membranes and dead-end mode of operation were used for secondary refinery effluent treatment, at constant transmembrane pressure and fixed loss of the initial flux. For the training procedure of neural networks, ultrafiltration tests followed by backwashing with demineralized water, performed in the same operating conditions have been selected. Two neural network models were constructed to predict the flux at any time instant during ultrafiltration and after backwashing for arbitrary cycles, within the ultrafiltration-backwashing process. The trained networks are able to accurately capture the non-linear dynamics for initial fluxes in the range 80–145 l h −1 m −2 and for time horizons up to 2500 s.