Detecting and Localizing Leakages in Water Distribution Systems Using a Two-Phase Model

Abstract

Rapid and accurate detection/localization of leakage is important to preventing severe consequences in water distribution systems, such as loss of water resources and sinkholes. This study presents a novel two-phased approach to leakage detection and localization as a part of the ‘Battle of Leakage Detection and Isolation Methods’ (BattLeDIM). During Phase 1, data classification, data feature scaling, and data verification were performed, where individual leakage events were detected using the K-mean clustering algorithm. In Phase 2, the location of the leakage was identified using a trial-and-error optimization procedure. During the process, an emitter coefficient is assigned to each node, and the node that has the minimum error between the supervisory control and data acquisition (SCADA) measurements and the simulation results is determined as the leakage location. The results show that there were 23 real leakage events, none of which were detected or localized. Finally, ideas for real leak detection are provided, and a direction to overcome the limitations of existing leak detection techniques is presented.