Predicting membrane fouling during municipal drinking water nanofiltration using artificial neural networks

A robust artificial neural network (ANN) model requiring minimal training, closely predicted membrane fouling during nanofiltration (NF) of ground and surface water. Neural networks accurately simulated the total resistance to water permeation across NF membranes during bench-scale experiments with flat membrane sheets, tests with single spiral-wound elements, as well as pilot- and full-scale tests with multiple spiral-wound elements arranged in two stages. ANN inputs included physically meaningful and independent variables including flow rates and feed water quality parameters (pH, UV254, and total dissolved solids (TDS)) that are commonly monitored during water treatment thereby facilitating their implementation. Therefore, under the experimental conditions investigated, colloidal fouling and biofouling appeared to be negligible because accurate ANN predictions were possible without using feed water turbidity and bacteria concentrations as inputs. One emphasis during this work was to minimize the data employed for ANN training while simultaneously performing simulations in purely predictive mode for entire cycles (experiments). Cumulatively, using only 10% of experimental data for ANN training allowed prediction of 93% of them with <5% absolute relative error. Hence, simple to implement ANNs are capable of capturing changes in feed water quality, flux, and recovery and can successfully overcome difficulties associated with mechanistic models to accurately predict long-term fouling during municipal drinking water nanotiltration. (C) 2003 Elsevier Science B.V. All rights reserved.