This research effort led by Carollo Engineering and NAWI is bringing artificial intelligence and advanced controls to potable water reuse, aiming to cut energy and costs while boosting reliability at reverse osmosis (RO)-based advanced treatment facilities. The project targeted “pipe parity” with conventional water supplies by testing and deploying machine learning (ML)–enabled fault detection and process control across integrated RO treatment trains.
Utilities in drought-prone regions increasingly rely on advanced purification of municipal wastewater to produce high-quality drinking water. RO-based advanced treatment (RBAT) is proven to robustly remove pathogens and chemicals, but in many locations remains nearly ten times more expensive and ten times more energy-intensive than conventional sources. While some processes feature embedded advanced controls, these are often not integrated across the full treatment train and can miss upstream–downstream interactions—leaving an opportunity for system-wide optimization using ML and AI.
The project’s approach began with a desktop evaluation of five fault detection and process control methods using high-frequency sensor data from partner utilities. Methods that could deliver at least 10% energy savings, 20% cost savings, or 50% improved reliability were advanced to demonstration at two RO-based potable reuse facilities operated by Las Virgenes Municipal Water District (LVMWD) and Orange County Water District (OCWD). A cloud-based, semi-autonomous architecture generated alerts for safer, more efficient operations.
Results show the promise of smart monitoring that reduces nuisance alarms while maintaining public health protection. For RO integrity, the team developed a non-parametric Shewhart Sign Chart to track pathogen removal performance using online surrogates, which can be skewed by outlier readings. With a rolling 12-hour window, the method detects true events within three hours and cuts false alarms by more than 50% compared with single-point alarms—meeting a key Reliability & Availability pipe parity metric. The approach was implemented via a real-time cloud dashboard at OCWD.
The researchers also tackled a long-standing operational challenge: real-time management of N-nitrosodimethylamine (NDMA), a critical disinfection byproduct with no commercial online sensor. Because UV/advanced oxidation (UV/AOP) has typically been operated conservatively based on historical NDMA highs, the team trained supervised ML “soft sensors” to predict influent NDMA using existing plant data. Using a hybrid statistical–ML approach and principal component analysis to handle highly correlated RO features, support vector machines successfully predicted NDMA levels. Simulations indicate the site could reduce UV dose by 21–29% while still achieving a 0.69 ng/L NDMA target, depending on the chosen factor of safety. Data from a second site revealed differences in seasonal and daily NDMA patterns between sites, underscoring the importance of site-specific control.
Beyond cutting energy and chemical consumption, the project aimed to increase uptime and operator support, improve detection of declining water quality, and avoid unnecessary shutdowns—outcomes that strengthen public confidence in purified water. Scale-up via cloud dashboards and semi-autonomous controls could make potable reuse systems more resilient and responsive to changing influent conditions across the full treatment train.
The effort brought together partners from Baylor University, Oregon State University, OCWD, LVMWD, National Water Research Institute, West Basin Municipal Water District, Yokogawa, and others, and was coordinated with NAWI’s Advanced Process Controls initiative.
Check out the recording of our informational webinar and the project poster for more information on this project.

