Applying machine learning and AI for nanofiltration membranes water applications: a review

Nanofiltration (NF) has long been the focus of researchers and operators in the field of water and wastewater treatment and desalination. Characterised by both high rejection of diverse substances depending on their molecular composition and charge, and high flux, NF membranes have demonstrated broad applicability while maintaining long-lasting performance. However, NF membrane scalability and reliability are intrinsically constrained by non-linear phenomena like complex solute-rejection mechanisms, long-term fouling dynamics, and the inherent flux-selectivity trade-off. Traditional mechanistic models relying on simplifying assumptions within the solution-diffusion framework, often fail to accurately predict performance in heterogeneous, real-time water matrices, leading to a critical divergence between predictive capability and operational reality. This review addresses this gap by comprehensively assessing the integration of machine learning (ML) and artificial intelligence (AI) across three critical domains: fabrication optimisation, performance prediction, and fouling diagnosis and mitigation, through the analysis of 100 NF-ML publications over the past 10 years. Additionally, it highlights key methodological challenges including data scarcity and heterogeneity, and lack of robust external validation. Finally, the review emphasises that future advancements lie in reinforcement learning for real-time adaptive control and in hybrid ML-mechanistic frameworks bridging the existing gap between data-driven prediction and transparent mechanistic understanding.