Related Experiment Video
Updated: Jul 16, 2026
12:45
Detection and Quantification of Tunneling Nanotubes Using 3D Volume View Images
Published on: August 31, 2022
Lightweight Graph Neural Network-Driven Acoustic Anomaly Detection Method for Gas Pipeline Leakage Levels in
Wei Sun1, Yang Li2,3, Jinghu Yang2,3
1China Coal Technology and Engineering Group, Chongqing Research Institute, Chongqing 400037, China.
Sensors (Basel, Switzerland)
|July 15, 2026
Summary
This study introduces a new method using a lightweight Spatial-Temporal Graph Neural Network (ST-GNN) to accurately identify gas pipeline leakage sizes in utility tunnels. The approach enhances public safety by distinguishing different risk levels based on acoustic signals.
Area of Science:
- Engineering
- Computer Science
- Public Safety
Background:
- Gas pipeline leakages in urban underground utility tunnels present significant public safety risks.
- Accurate identification of leakage hole size is crucial for effective risk management, but traditional methods struggle in complex environments.
Purpose of the Study:
- To develop a novel method for precise identification of gas pipeline leakage risk levels in underground utility tunnels.
- To address the limitations of traditional acoustic analysis in complex utility tunnel settings.
Main Methods:
- A lightweight Spatial-Temporal Graph Neural Network (ST-GNN) was proposed for leakage risk level identification.
- Acoustic signals from a utility tunnel simulation platform were collected and transformed into graph-structured data using Short-Time Fourier Transform (STFT).
- A Chebyshev graph convolutional network was employed to extract discriminative features for classifying leakage levels.
Main Results:
- The proposed ST-GNN method achieved excellent performance in a three-level leakage classification task.
- t-SNE visualization confirmed effective separation of features corresponding to different leakage levels.
- Ablation experiments validated the model's robustness with limited data and the efficiency of its lightweight design.
Conclusions:
- The developed method offers a feasible solution for automated and refined identification of gas pipeline leakage levels in underground utility tunnels.
- The ST-GNN approach effectively distinguishes acoustic features related to different leakage sizes, improving safety management.