AI-based smart water leak detection using hydrophones ​

Abstract

Acoustic technologies are popular for the detection of leak detriments in water pipelines. However, problems of false alarms, missed leaks, limited site information, and the high cost of long-term monitoring remain prevalent. These issues demand a more sophisticated testing approach suitable for real-world applications. Hydrophone technology has a strong promise for precision leak detection. However, acoustic leak detection is mostly focused on detection using controlled testbed experiments. The practical application of hydrophones for leak detection has not been well reported in the literature. The current study presents a smart real-time leak detection system that uses real-time acoustic data collection. AI-based data-driven models were developed to identify leaks based on limited site information. Different classification models were trained using various feature combinations to identify the most significant model and feature set. ensemble-based classifiers of Adaboost, and Random Forest demonstrated the most promising performance for the leak detection application. Results reveal hydrophones to be more effective as compared to other acoustic devices like accelerometers and noise loggers in detecting leaks.