Fouling control and optimization of a drinking water membrane filtration process with real-time model parameter adaptation using fluorescence and permeate flux measurements

Abstract

The implementation of fouling control strategies is crucial for the effective maintenance and long-term application of ultrafiltration (UF) membranes in drinking water treatment. Membrane cleaning methods such as membrane back-washing protocols are typically employed to remove the build-up of natural water species (foulants) on the surface and/or in the pores of the membranes, which causes membrane fouling, contributes to increased trans-membrane pressure and shortens membrane lifetime. In a previous work by the authors, a fluorescence-based principal component (PC) modelling approach, which was able to forecast membrane fouling behaviour of a bench-scale UF membrane set-up over a future time horizon, was introduced. This approach also proved suitable for estimating the optimum future membrane back-washing times required for controlling fouling and minimizing the energy demand for drinking water production. In this study, the forecasting ability of this model was improved by updating the model parameters with current process measurements. The Extended Kalman filter (EKF) approach was used to achieve this objective. The EKF approach accomplished real-time adaptive estimation of key model parameters based on either real-time UF flux measurements or PC scores related to fluorescence measurements of membrane permeate. The model predictions and the corresponding experimental UF flux data for different membrane fouling situations revealed that on-line permeate flux-based parameter adaptation showed improved model predictions as compared to PC score-based adaptation, especially for longer filtration times. The improvements in the accuracy of the fluorescence-based model forecasts also aided the estimation of optimum back-washing times with better accuracy resulting in considerable energy savings compared to back-washing estimates obtained without real-time parameter adaptation.