Investigation of the factors affecting reverse osmosis membrane performance using machine-learning techniques

Abstract

Several factors affecting the RO membrane performance can be investigated mainly in two groups; feed characteristics and operational parameters. In this work, the effect of feed characteristics of a municipal wastewater such as conductivity, oxidation reduction potential (ORP), total suspended solids (TSS), turbidity, chemical oxygen demand (COD) and feed operational parameters such as feed pressure, flow rate, temperature on RO membrane performance monitoring parameters (pressure difference across membranes, salt passage and permeate flow rate) were analyzed in a wastewater recovery plant using different machine learning techniques. Artificial neural networks (ANNs) were found to perform better to predict pressure difference whereas random forest and multiple linear regression models performed better in salt passage and permeate flow rate prediction, respectively. Association rule mining was also employed, and it was found that high feed flow rate, low feed conductivity and low temperature operation are desired for effective RO membrane operation.