Smart Management of Leaks in Underground Pipelines Using Machine Learning

Undetected leakages in water distribution pipelines lead to a significant loss of freshwater supplies. Effective monitoring of these pipelines to curb water loss is a global concern. To this end, numerous works have been done previously. However, underground deployment of water pipelines, sensor sensitivity, and complexity of the pipeline network introduce challenges for accurate water leakage detection and localization. Leakages can be categorised as one of two types: reported or unreported leakages. While reported leakages like pipe bursts could be easily spotted with visual inspections, hidden and noiseless leaks creeping through joints are background leakages that require more sophisticated monitoring. Some of the advanced works relied on optical fibres and magnetic induction techniques. Being expensive, they were limited to high-risk areas and could not be generalised for larger deployment. Some other works use field sensors like temperature, pressure, and acoustic sensors. These sensor-based methods have gained sufficient accuracy for detecting leaks, but localizing the exact leaking part is still an open problem. This work proposes a novel leakage detection and localization methodology that applies machine learning to the data generated by water flow sensors. A simple pipeline network was set up consisting of three water flow sensors across different pipeline segments. The readings from the flow sensors were recorded and used to train a machine learning classifier. The results demonstrate that our method can efficiently detect and localize background leakages with 99.3% accuracy.