Pipelines—like people—are full of life. They change with pressure, flow, temperature, and sometimes with leaks or breaks. Distributed fiber optics gives us a privileged window to observe these variations along kilometers of pipeline in real time, at a resolution electronic point sensors can’t match. The challenge is no longer measuring, but interpreting: separating noise and interference from the relevant event—quickly and reliably. 

That’s where artificial intelligence comes in. Within the ADRIATICO project, we are developing and training neural networks on real-world data captured by sensors installed in water networks across several countries, to characterize incidents such as leaks, bursts, or tampering and to avoid false alarms—with meter-level precision. 

Early results in live operational conditions: 

• Detection precision >90% for relevant events 

• Recall >84%, leaving very few incidents unflagged 

• Reliability >96% across rupture detection, pressure-control anomalies, and leaks 

In practice, the system accurately understands what’s happening in the network, misses very few events, and distinguishes multiple incidents within a single measurement (e.g., a flow variation, a water hammer, and an incipient leak occurring at the same time). 

“Our goal isn’t just to detect—it’s to understand the network and prevent false alarms. With AI + fiber optics, we deliver context, precision, and operational action.” — ADRIATICO Team 

Building on these strong first results and looking to early 2026, phase two will focus on hard scenarios (high noise, complex urban environments, seasonal variability) and on optimizing the false-alarm rate, preparing the solution for continuous operation with a 99.9% reliability target. 

———————-

This work is supported by NextGenerationEU funds through the RETECH-IA program of the Generalitat Valenciana, in collaboration with UPV and Artikode, accelerating the translation from research to operational impact in the field.