LIME: Long-Term Forecasting Model for Desalination Membrane Fouling to Estimate the Remaining Useful Life of Membrane

Abstract

Membrane fouling is one of the major problems in desalination processes as it can cause a severe drop in the quality and quantity of the permeate water. This paper presents a data-driven approach for long-term forecasting of fouling behavior in membrane-based desalination processes. The proposed Long-term forecastIng ModEl (LIME) consists of two intertwined machine learning models trained separately by historical operating conditions of ultrafiltration for pretreatment of reverse osmosis seawater where transmembrane pressure is used as a fouling indicator. The first model predicts the increase in fouling due to filtration. This output is fed to the second model to predict the fouling reduction due to membrane cleaning. In turn, this output is used as the initial fouling condition for predicting the next filtration cycle. The forecasted fouling is used to estimate the membrane's remaining useful life (RUL), which ends when cleaning no longer reduces the fouling below a safety threshold. Evaluation results show that the model can predict the membrane fouling for 1400 cycles with an R-squared score of 0.8. Moreover, the RUL is estimated for various thresholds with an average percentage error of 7%.