Multifractal analysis of acoustic signals for leak detection in urban water distribution networks

Leak detection and pipe condition assessment in urban water distribution networks (WDNs) is challenging due to the underground location of most pipes, making visual inspection difficult. Non-intrusive, acoustic-based methods for continuous monitoring have gained growing attention due to advances in sensing and data processing technologies. Although many studies have explored acoustic leak detection, relatively few have been validated through experiments in real-world systems that assess both the capabilities and limitations of these methods. This paper introduces a leak detection approach based on multifractal analysis of acoustic signals recorded by hydrophones deployed in an operational WDN. The central hypothesis is that leaks generate distinct multifractal patterns in the acoustic signal, allowing for detection without the need for labeled data. The method was tested using real-world data from leak experiments conducted at two different sites within a WDN. Results show that the singularity exponent, a key feature in multifractal analysis, can support continuous acoustic monitoring and robust leak detection, even under variable conditions and temporal fluctuations. The study also highlights how factors such as distance, network layout, and leak size affect detection limits. This research offers valuable insights for advancing the application of continuous acoustic monitoring in leak detection and water loss control in urban water networks.