Can nanofiltration be fully predicted by a model?

There is an increasing need for model-based tools to design membrane processes for new industrial applications or to optimise existing membrane installations. The advantage of such tools is that costs can be saved by reducing the number of experiments. In this study, the requirements for a membrane filtration model, suitable for practical use, are summarised. It is investigated to what extent it is possible to set-up such a model with the current available literature and knowledge. A membrane filtration model has been set-up based on the Maxwell-Stefan transport equations. A Freundlich equation is used to describe the membrane charge by means of adsorption of ions. With the model the permeate flux and rejections of multi-component liquid feeds can be calculated as a function of membrane properties (mean pore size, porosity, thickness, surface charge characteristic) and feed pressure. With two NF-membranes (Desal 5DK and a prototype capillary type 2 membrane) rejection experiments have been carried out with glucose, single salt solutions (NaCl, CaCl2, Na2SO4) and ternary ion mixtures of these salts. With the model the experimental flux-rejection curves can be fitted reasonably well. However, each salt mixture needs its own set of fitted parameters for the membrane charge isotherms. Furthermore, the fitted membrane charges are in contradiction with values from the literature obtained by electrokinetic measurements. Obviously, the membrane charge parameters have lost their physical meaning and are used to compensate for physical phenomena not included in the model. Extending the model with an electrostatic free energy term will be a step forward in development. Further research is needed to fulfil all requirements for the wide scope of industrial applications. (C) 2002 Elsevier Science B.V. All rights reserved