Dynamic modelling of crossflow microfiltration of bentonite suspension using recurrent neural networks

Abstract Using a set of experimental results (permeate flux and deposit thickness) as a function of different operating conditions, obtained during crossflow microfiltration of a bentonite suspension with a laboratory pilot, a dynamic modelisation of this process by means of recurrent neural networks is proposed. The elaborated neural network is able to describe the evolution of permeate flux and deposit thickness from the process variables (transmembrane pressure, crossflow velocity, concentration of the suspension) and the starting point values for permeate flux and deposit thickness. The simulation of the evolution by such a model for experiments limited to a certain timespan, allows us to obtain coherent limit values for both permeate flux and deposit thickness.