The M¿ry-sur-Oise drinking water treatment plant consists of two parallel treatment paths: a conventional biological path, and a nanofiltration path that has been operating since September 99, producing 37 MGD and is equipped with NF200B-400 8-inch spiral elements (DOW). Part of the challenge was the operation of a large nanofiltration unit on treated surface water, containing high levels of particles and dissolved materials. Indeed, since the full-scale step was brought in service, a slight loss in water permeability has been observed and the effectiveness of the cleaning has not been constant. In order to understand this phenomenon, a statistical modelling of the fouling has been carried out. The input parameters are the running time after cleaning and the feedwater characteristics. The water quality after each treatment step and the water permeability of each stage of the eight units are recorded in a database. The running time since the last cleaning was also added to these data. On this basis, a model of water permeability (PLS regression) has been calculated for each of the eight units. For all models, the running time after cleaning was the main influent parameter on the membrane fouling. An increase in particle counts and Total Organic Carbon content accelerated the fouling phenomenon, while changes in chemical parameters such as hardness, sulphate content, seemed to have a slight impact on the membrane clogging. The modelling of the membrane fouling will be useful to estimate the lifetime of the membranes and to program their replacement. Includes reference, tables, figures.
