2026/07/01 by Wei Zeng, Mark Stephens, Xiang Wang +5
Engineering · Environmental Science · #Water Systems and Optimization #Water Quality Monitoring Technologies #Underwater Vehicles and Communication Systems
paper · doi:10.1029/2026wr043758
Abstract Distributed wireless acoustic sensors, such as accelerometers, in smart water networks are increasingly used to detect leaks before they escalate into pipe breaks. In addition to leak‐induced signals, these sensors also capture acoustic noise generated by pumping and valve operations, pedestrians, traffic, and other environmental sources, leading to a large number of false‐positive alerts in operational water asset monitoring systems. In this study, a new set of tailored signal features is developed, for the first time, to characterize specific types of acoustic signals collected from real urban water networks. Unlike conventional acoustic features widely adopted in other application domains, the proposed features are specifically designed to represent both leak noise and common non‐leak environmental noise in water distribution systems. Explicit characterization of non‐leak noise, which has often been overlooked in existing studies, significantly improves the accuracy and reliability of leak detection under realistic urban conditions. Using these tailored features, leak noise is effectively distinguished from environmental noise within a support vector machine framework. The proposed model is validated using large‐scale IoT acoustic data sets from multiple operational urban water network. The results demonstrate a high leak detection rate while maintaining a substantially reduced false alarm rate of 3.2%, thereby minimizing unnecessary field inspections and operational disruptions in real‐world water network management. By enabling earlier and more reliable identification of leaks, the proposed approach supports timely intervention, reduces unnecessary water losses, and contributes to the sustainable management and conservation of water resources in aging urban water infrastructure.