Leak Localization in Water Distribution Networks - A Case Study

Abstract

Leakage in municipal water distribution systems leads to significant economic and operational losses. Resources are spent on the field-based leak localization techniques, such as listening for leaks in various parts of the pipe networks. A traditional method of leak detection is by partitioning the network into District Metered Areas (DMA). These are hydraulically isolated zones monitored by sensors, which enable leak detection by analyzing night flows. Once a leak is detected, the next step is to localize it. This Master’s thesis investigates the use of virtual DMAs in a real-world water distribution network located in Göteborg, retaining the benefits of DMAs without requiring physical network isolation. A hydraulic model was used to reproduce the network conditions during 10 leak scenarios with leak magnitudes between 0.5 and 4 l/s. Measured flows were compared to simulated flows using three evaluation metrics: Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and mean flow difference. The metrics were used to rank candidate leak locations and identify regions of similar hydraulic response. In most cases, the metrics perform similarly. In 8 out of 10 scenarios, the predicted leak node is within 270 m of the true node.