PUB's Anomaly Leak Finder — Singapore's AI Digital-Twin System for Underground Water-Pipe Leak Detection

PUB, Singapore's national water agency, needed better tools to detect hidden underground pipe leaks across its distribution network, since earlier acoustic-only sensor deployments could not reliably distinguish genuine leaks from other anomalies like sensor failures or pressure shifts.

To build a cloud-hosted AI system combining data-driven prediction with physics-based hydraulic simulation, recalibrated daily against live sensor data, to detect and localise leaks before they become disruptive.

PUB built the Anomaly Leak Finder (ALF) on Bentley Systems' iTwin and OpenFlows software, combining a data-driven prediction (DDP) model with a physics-based simulation (PBS) model. Hydraulic models are recalibrated daily using data from over 90 smart sensors, creating a digital twin of more than 1,000km of pipelines across four major distribution networks, run 24/7. ALF sits within PUB's longer-running leak-management programme (330km of pipes renewed since 2016; 267 pipe leaks repaired in 2024).

Since going live in January 2024, ALF has detected and localised two significant underground leaks to within roughly one kilometre, alerting leak-detection crews before they became disruptive, and also flags related anomalies such as sensor failures and demand-driven pressure shifts.

The two confirmed leak detections are an early-stage result rather than a mature performance track record; PUB has not yet published network-wide non-revenue-water savings attributable specifically to ALF.