Machine learning for membrane-based distributed water treatment and desalination systems: Current advancements and future prospects

Reverse osmosis (RO) and nanofiltration (NF) water treatment and desalination (WTD) systems are increasingly deployed to provide fit-for-purpose potable and non-potable water for small communities and industrial facilities. WTD systems must operate reliably under temporally variable feed water quality and demand and autonomous operation is essential for distributed WTD (DWTD) systems. Accordingly, this review critically addresses the role of machine learning (ML) and artificial intelligence (AI) in enabling real-time performance forecasting, soft sensing, fault detection and isolation, performance degradation, and model-predictive control. The review discusses supervised, unsupervised, deep-learning, transfer-learning, and cooperative-learning approaches with respect to dataset size relative to process model attributes, model complexity, interpretability, robustness, and applicability to intermittent and non-stationary DWTD operation. Key challenges are identified in relation to multimode system operation, performance degradation, sensor faults, and the need for high-frequency data streams for real-time decision support. Future research needs are identified in relation to domain-informed model development, model generalizability and explainability, and cloud-enabled virtual water district architectures for resilient, scalable, and cost-effective autonomous operation of geographically distributed water treatment and desalination systems.