DYNAMIC MODELING OF CROSS-FLOW MICROFILTRATION USING NEURAL NETWORKS

The neural network theory was used to dynamically model membrane fouli ng for a raw cane sugar syrup feed stream. The use of neural networks enabled us to integrate the effects of hydrodynamic conditions on the time evolution of the total hydraulic resistance of the membrane under constant temperature and feed stream concentration. The results obtai ned satisfactorily model the effects of both constant and variable tra nsmembrane pressure and crossflow velocity as the filtration was follo wed through time. The effects of the hidden network structure as well, as the scatter of data on the quality of modeling are discussed in th is paper.