Over three years, a consortium of eight organisations across Spain, Belgium, the Netherlands and Germany developed a modular AI platform built around three complementary technology pillars. All designed to support operators, not replace them. On 12 August 2026, the DARROW project marked the end with a full-day final event at Waterschap De Dommel Headquarters in Boxtel, The Netherlands.
Researchers, water sector professionals, plant operators, and industry partners came together to reflect on what was achieved, share the technical outcomes of the project, and look ahead to what AI-assisted water treatment could mean for the sector as a whole.
A day of presentations, demonstrations, and real-world insight
The morning sessions centered around presenting the goals and outcomes of the DARROW project: from the current state of data quality and communication challenges in wastewater resource recovery facilities (WWRF), through the practical application of soft sensors and short-term predictive tools, to the more advanced territory of reinforcement learning and mechanistic model-supported AI for process optimisation.
The afternoon took participants to the Tilburg WWRF, one of the largest treatment plants in the Netherlands and the heart of the DARROW demonstration work. Participants received a tour through the facility, including a glimpse into the control room, where the plant operators are using and testing the DARROW tools. The day concluded with a networking reception, giving attendees the chance to continue the conversations started in the sessions.
The DARROW solution: A modular AI ecosystem
Smarter data
Wastewater treatment plants generate large volumes of operational data. Turning that data into reliable intelligence is challenging. Sensors fail, measurements are incomplete, and key process variables cannot always be measured continuously. DARROW addressed this through data augmentation and enrichment tools, including automatic sensor anomaly detection, real-time plant status classification, software sensors based on Extended Kalman Filters, and 24-hour influent forecasting models. At the Tilburg WWRF, these tools successfully detected a failing NO₃ sensor, identified abnormal NH₄ concentrations caused by an industrial pollution discharge, and provided continuous BOD concentration forecasts to support aeration planning.
Smarter control
Conventional control systems optimise individual loops independently, leaving much of the available process information unused. DARROW’s Reinforcement Learning controllers take a different approach: trained across millions of simulated scenarios, they consider the full plant state simultaneously and continuously adapt to changing conditions. Controllers were developed for both secondary treatment and anaerobic digestion at Tilburg, optimising aeration, internal recirculation, and sludge feeding in a coordinated, plant-wide manner. Before any live deployment, each controller was tested extensively in a calibrated digital simulator to ensure safe and reliable operation.
Smarter decisions
Testing different operational strategies directly on a live treatment plant is neither practical nor risk-free. DARROW’s digital twins, built on fast, machine-learning-based Reduced Order Models derived from high-fidelity mechanistic simulators, give operators the ability to simulate “what-if” scenarios in real time, explore trade-offs between water quality and energy cost, and receive recommendations they can understand and verify.
Results at Tilburg: What the demonstration showed
The Tilburg WWRF served as the primary demonstration site throughout the project, operated by Waterschap De Dommel. The plant, with a design capacity of 350,000 population equivalent and wastewater flows of up to 12,585 m³/h during wet weather, provided a genuine operational environment in which to test and validate the DARROW tools under realistic conditions.
The demonstration confirmed the technical feasibility of deploying the full DARROW platform in a live treatment plant, with performance assessed across five key indicators: nitrogen removal efficiency, sludge production, energy consumption, energy generation from biogas recovery, and greenhouse gas emissions.
Looking ahead: From Tilburg to Europe and beyond
If deployed at 60 large water resource recovery facilities across Europe within five years, the DARROW solution is projected to deliver:
🔹 92,000 MWh reduction in energy consumption per year
🔹 27,600 MWh increase in renewable energy generation per year
🔹 25,000 tonnes of CO₂ equivalent reduction in greenhouse gas emissions per year
🔹 4,000 tonnes of additional biological phosphorus recovered per year
🔹 238,900 m³ reduction in sludge waste production per year
The consortium is already building on the DARROW knowledge base, with digital twin deployments underway at further sites across Europe.