Author(s): Mostafa Rahmanshahi; Huan Feng Duan; Alireza Keramat And Vincent Tjuatja
Linked Author(s): Huan-Feng Duan
Keywords: Water distribution system leak detection hydraulic transient machine learning
Abstract: The non-invasive assessment of water pipelines using fluid transient waves has been studied for three decades. While various methods can identify anomalies, they often require detailed modeling. However, machine learning (ML) offers a flexible solution by learning from examples rather than relying on explicit programming. This paper introduces an ML framework to predict leaks in a real pipe network with multiple excitation sources. ML was effectively trained and tested for leak detection and sizing, achieving notable results. Specifically, 90% of the tested cases successfully identified leak locations with an error of less than 5 m, while the average error in predicting leak size was about 5%. Furthermore, over 90% of the tested data had size errors lower than 10%. These findings highlight the promising potential of integrating fluid transient pressure waves with machine learning for leak detection in real pipe networks.
Year: 2025