Development of acoustic data analysis software tool for near real-time anomaly detection

Anomaly detection is one of the crucial tasks in daily operational of water utility company as it can prevent the water service disruption and the water infrastructure damage, which may then lead to catastrophic outcomes. In this work, we present a general data analysis software solution for acoustic leak detection for the real-world data. The first step in the analysis procedure is acoustic data quality assessment. It is important to verify that the data used for analysis does not contain any errors because bad data can lead to wrong decisions even when the analytical algorithms applied are correct. The second step is acoustic data preprocessing that consists of two sub steps, including acoustic signal decomposition using empirical mode decomposition (EMD) algorithm and the acoustic feature identification. The final step is to build acoustic energy profile time series to detect any outliers, which are classified into leak or non-leak events. Unlike pressure or flow data, which are recorded at relatively large time step, namely one data record in every 5 minutes or 10 minutes interval. Instead, acoustic data is a high-resolution or high-frequency data and contains thousands of data points in one second. Consequently, each of the previously mentioned steps is computationally intensive. Therefore, an efficient software tool is required for automating and facilitating the acoustic data-based anomaly detection. In collaboration with PUB, Singapore’s National Water Agency, a software tool is developed and tested on a large-scale water distribution network using hydrophone sensors installed in the water supply network.