Water utilities face growing pressure to manage aging distribution networks more efficiently, but the tools to do so are limited. Leaks and losses often go undetected for long periods, and IoT sensor deployments across water systems have to cope with energy constraints, unreliable connectivity, and the difficulty of turning fragmented, multi-stakeholder sensor data into something a predictive model can actually use. Even where AI could help, most water utilities lack the specialised techniques and integrated infrastructure needed to move from raw sensor readings to real operational decisions.
In this context, the EU-funded WATERDEAL project will build an integrated communication, digital twin, and AI framework for water distribution networks. It has three main components: a low-power, secure IoT communication layer for gathering sensor data across multi-stakeholder networks, digital twins that turn that data into structured representations of the water system, and physics-informed AI models — including graph and topological neural networks — that run on top of the twins for leak detection, anomaly monitoring, demand prediction, and adaptive control. A federated learning framework lets utilities train these models collaboratively without centralising sensitive data, keeping results privacy-preserving and compliant with European regulations.
The project validates this pipeline at multiple levels, from digital-twin simulation through controlled testbeds to real-world pilots run with utility partners, and pairs the technical work with staff exchanges that build cross-sector expertise in network engineering, environmental science, and data analytics across the consortium.
