Machine learning-guided underlying decisive factors of high-performance membrane distillation system: Membrane properties, operation conditions and solution composition

Abstract

Membrane distillation (MD) is considered as one of the promising membrane technologies with the potential to effectively produce freshwater from high concentration brines. Increasing demand for freshwater necessitates a deep understanding of the high-performance MD systems. Traditional experimental approaches are limited in their ability to comprehensively explore factors from multiple perspectives. Herein, a comprehensive machine learning (ML) workflow comprising of four distinct modules was devised to elucidate the decisive factors of high-performance MD systems. A comprehensive database was constructed consisting of 25 input features with membrane properties, operating conditions, and solution composition, along with the inclusion of three output performance indices, namely flux, wetting, and fouling. Leveraging automated machine learning (AutoML) algorithms, three ML models have been developed for accurately predicting the performance of MD system. We interpreted the ML models and extracted meaningful insights pertaining to the contributions of important factors on performances. The results indicated that ML can capture the important roles of the temperature difference between feed and permeate (ΔT). Furthermore, the water contact angle (WCA) made considerable contributions to membrane wetting, and module size attached more importance to membrane fouling. Based on the predictive models, the particle swarm optimization (PSO) effectively inferred 6 optimal parameters to achieve high-performance for the MD system. Our work represents a paradigm shift in the field of membrane technologies, highlighting the potential of ML-guided methods to elucidate the fundamental mechanisms of high-performance MD systems.