The operator's decision
Gradual membrane fouling begins before it causes serious performance deterioration. AquaSentinel recognizes the developing trajectory and forecasts when it may reach a documented intervention criterion. That forecast gives an operator time to inspect the affected train and decide what to do.
The forecast and its proof system
Early-warning research faces a reproducibility problem: plant histories are proprietary and rarely contain independently adjudicated fault onsets. AquaSentinel addresses that benchmark gap; it does not claim that RO digital twins or early-warning methods did not previously exist.
AquaSentinel is the forecasting system. Its transparent, physics-based generator and locked protocol form the proof system behind it. They create SWRO runs with labeled fault onsets and measure the modeled intervention window, alarms on controls, and diagnosis at the first advisory.
Why it matters
On 300 held-out synthetic runs across three regimes, AquaSentinel detected all 180 modeled process faults and warned earlier than the primary comparator in 175. Across gradual fouling cases, the average modeled intervention window was about 1.3 days. No process alarms occurred in 120 declared synthetic controls.
The GPIW prize
This entry targets Track A - Research and Proof of Concept in the GPIW Digitalization, Automation, and Smart Water Management category. AquaSentinel is self-assessed at TRL 3: the method functions as a computational proof of concept, but independent design-tool review, real historian evaluation, and operational testing remain ahead. A Gulf / Red Sea preset represents warm, high-salinity conditions relevant to regional desalination; it is not evidence from an SWA plant.
Team
Alain Trottier — independent researcher and software engineer with a background spanning physical modeling, data science, and applied AI. AquaSentinel's twin, detectors, learned model, diagnosis, benchmark, and interactive demo are original work, packaged for an open-source release.
Honesty
Explore
The backend application includes the simulator, tests, protocol, validation templates, license, citation metadata, and dataset generator. A public canonical repository and DOI are release-account steps still pending.