Identification of acoustic spectral characteristics and propagation attenuation-based localization for gas pipeline leakage

Abstract

To address the challenge of traditional detection techniques failing due to the material characteristics of polyethylene (PE) gas pipelines, this paper proposes an acoustic-deep learning fusion leak diagnosis method. An experimental platform simulating medium and low-pressure gas pipelines with looped and branched configurations was established to systematically collect acoustic signals under diverse leakage conditions, including varying aperture sizes, pressures, and pipe network topologies. Analysis revealed that the main energy of the leakage signals is concentrated in the 0–20 Hz infrasound band, and the pipe network topology systematically modulates the signal's frequency domain characteristics, causing attenuation in the 0–10 Hz range and enhancement in the 70–90 Hz range. To enhance recognition accuracy in complex pipe networks, a leakage state recognition model based on a one-dimensional convolutional neural network (1D-CNN) was developed, and variational mode decomposition (VMD) was introduced for noise-signal separation. The results demonstrate that input signals optimized by VMD enable the model to achieve an average recognition accuracy of 88 % in looped and branched pipe networks, representing an 11.8 % improvement over raw signals. Furthermore, a single-sensor localization method based on waveform energy attenuation was proposed, achieving average localization errors of 0.72 % and 1.15 % in straight pipe sections and branch pipe scenarios, respectively, with an effective detection distance of up to 20 m. This work offers a solution for PE gas pipeline leak detection that integrates theoretical innovation with practical engineering value.