The international research team "Leakbusters" has won 3rd place in the BattLeDIM 2020, an international competition on leakage detection and localization in water distribution systems. ECDF scientists Ivo Daniel and Prof. Dr. Andrea Cominola participated in the competition together with scientists from the Machine Learning Group at Technische Universität Berlin and and from the North Carolina State University.

Water losses are one of the main consequences of infrastructure failures in water distribution networks (WDNs), accounting for more than 50% in some WDNs worldwide. Methods that support prompt detection and accurate localization of leakages are crucial to help water utilities implement timely mitigation measures and avoid unnecessary losses of water and revenues.

The research team has developed a high-resolution, pressure-based method for leak detection and localization in WDNs and tested it against the benchmark data set provided as part of the “BattLeDIM”. “Our method is composed of two modules that operate sequentially. The first module performs leakage identification by processing pressure data observed at different sensor nodes in a WDN and by analyzing pressure differences between pairs of nodes,” says Ivo Daniel.

The first part of the method, the detection module, is trained using pressure data observed over a “normal” period of time, i.e. without any existing leakage events. “The reconstruction error of the model is then analyzed to detect the anomalies in a time series that may contain one or more simultaneous leakage events,” says Ivo Daniel.

The identified leakage events from the identification module are then used by the second module to perform the localization. This is done by first selecting the pipes in question in a search area around the most affected sensor based on the pressure drop reported by the sensors. Then the final localization is performed in a simulation optimization framework. “Preliminary experiments show that our pressure-driven method can promptly detect and localize most of the labeled leakages and it is suitable for real-time applications", reports Andrea Cominola.

//Further information is available in a Video on Youtube by Ivo Daniel.