Towards machine learning in water treatment: a diagnostic tool for assessing water quality

Abstract

Saltwater from the ocean constitutes 96.5% of the total water available on the Earth. The remaining 3.5% is freshwater, which is a crucial resource for living organisms. Increasing population and activ-ities related to climate change have led researchers to develop new methods to maximize freshwater resources. Solar desalination is an environment-benign method that can fulfil the requirement of freshwater. However, the efficiency of desalination cells is limited by the fouling phenomenon. The efficiency of the desalting process decreases because the pores of the membrane are clogged by fouling. Therefore, methods for detecting and diagnosing the fouling phenomenon by using mathematical models are required. We propose a machine learning modelling framework comprising of K-nearest neighbor, random forest, artificial neural network, and support vector machine algorithms to monitor the onset of fouling in desalination cells individually. Permeate datapoints from the filtration process were collected using a lab on a chip device. The datapoints were used to validate all four models. Furthermore, model responses for permeate data points were used as an indicator or soft sensor to grade the fouling level and potability of the treated water. The modelling framework can be used to detect the onset of fouling and erosion in desalination cells with high precision.