Prediction of nanofiltration rejection performance in brackish water reverse osmosis brine treatment processes

Abstract

A predictive nanofiltration model was built in Python to be used in reverse osmosis brine treatment processes. The model was fitted to rejection data obtained from NF trains functioning as a brine concentrator upstream of a gypsum precipitation reactor at a full-scale mine water treatment plant in Ahafo, Ghana. Over the six month operational period considered (September 2017 – March 2018) the rejection capability of the installed elements deteriorated considerably. This was reflected by a 13% increase in membrane pore radius, 40% decrease in effective active layer thickness and an 18% decrease in absolute value of the feed-membrane Donnan potential. Performance of elements from other manufacturers was simulated by loading their respective properties into the model. Cases modelled included a tightly wound Dow NF 90, a loosely wound Desal DS-5 DL and a Koch TFC SR-2 element. Rejections obtained from the installed MDS elements most closely approximated the performance of a loose NF element. These modelling studies have shown that the NF model built is capable of modelling nanofiltration in brackish water treatment processes. The current disadvantage of the model is the number of membrane—specific input parameters which need to be verified with independent experimentation. As a start, it is recommended that electro-kinetic data be obtained for similar solutions to enable a membrane charge density sub-module to be incorporated into the model.