Keynote: Integration of Artificial Intelligence into Membrane-Based Water Treatment: From Mechanisms to Processes

Membrane technology has been widely used in water and wastewater treatment industries. However, the complexity and existence of many confounding variables in membrane systems hinder an in-depth mechanistic understanding of membrane-based processes. In this study, we developed a method for modeling and mechanism analysis of membrane-based water treatment processes based on explainable artificial intelligence (AI), and explored its application in three key processes: membrane separation, membrane catalysis, and membrane fouling. We integrated AI algorithms with domain knowledge for modeling these processes and utilized the SHAP method to overcome the ‘black-box’ nature of conventional AI methods. As a result, data-knowledge co-driven AI model enhanced the predictive accuracy and understanding of trace organic contaminants (TrOC) rejection by polyamide membranes. The contributions of key mechanisms, including size exclusion, charge effect, hydrophobic interaction, etc., that dominate the rejections of TrOCs were quantified. Moreover, we predicted and elucidated oxidation kinetics in catalytic membranes at different pH levels, unraveling the quantitative relationship between molecular structure of contaminants and their reaction rates. Furthermore, we identified critical fouling factors and predicted fouling behavior in anaerobic membrane bioreactors. The feature importance and SHAP analysis indicated SMPp/SMPc (0.281) > EPSp/EPSc (0.110) > organic loading rate (0.106), which were the most critical factors affecting membrane fouling. Overall, this study established a novel AI-based approach to model and understand membrane-based water treatment processes, advancing the integration of Interpretable AI into membrane science research.