Vibro-Acoustic Distributed Sensing for Large-Scale Data-Driven Leak Detection on Urban Distribution Mains

Area of Science:

• Environmental Engineering
• Data Science
• Acoustics

Background:

• Non-surfacing leaks are a major cause of water loss globally.
• Effective monitoring of underground water infrastructure is crucial for utilities.

Purpose of the Study:

• To develop and evaluate advanced data-driven methods for leak detection using vibro-acoustic sensors.
• To assess the performance of these methods across diverse urban pipeline conditions.

Main Methods:

• Utilized commercial, field-deployable semi-permanent vibro-acoustic sensors for data collection.
• Applied Convolutional Neural Networks (CNNs) trained with Short-Time Fourier Transforms (STFTs) for signal analysis.
• Evaluated methods on live data from extensive multi-sensor deployments in a metropolitan city.

Main Results:

• Achieved excellent leak/no-leak classification accuracy exceeding 94%.
• Demonstrated consistent performance across sensors from different manufacturers.
• Confirmed the robustness of the method over time and in densely populated areas.

Conclusions:

• Data-driven vibro-acoustic analysis with CNNs offers a highly accurate and reliable method for non-surfacing leak detection.
• The proposed approach is sensor-agnostic and effective in complex urban environments.
• This technology can significantly improve structural pipe health monitoring and reduce water losses.