Feature extraction and pattern recognition of gas pipeline flow noise signals in a strong noisy background

Enbin Liu, Chang Lu, Zhaorong Wen, Tianshu Hao, Xudong Lu, Lidong WangPeerJ Comput. Sci 2024 · DOI

TL;DR

This paper introduces a novel method for accurately detecting gas pipeline flow noise signals amidst significant background noise, a critical challenge for urban infrastructure safety. By combining Principal Component Analysis (PCA) for feature extraction and a Backpropagation Neural Network (BPNN) for pattern recognition, the authors achieve a remarkable 97% accuracy, offering a robust solution for real-time monitoring.

What Problem Does This Paper Address?

Urban natural gas pipelines are expanding rapidly, creating a complex underground environment with active and abandoned lines, diverse materials, and coexisting utility networks. This complexity, coupled with aging infrastructure, increases the risk of gas leaks and pipeline ruptures, posing a significant threat to public safety. Existing methods for detecting these issues often struggle with the pervasive background noise from urban activities (traffic, construction, etc.), which can mask the faint acoustic signals of gas flow. The paper aims to address this by developing a method that can reliably identify gas pipeline flow noise even in these challenging, noisy conditions, thereby enhancing the safety and efficiency of urban gas pipeline networks.

Methods

The core of this research lies in a two-pronged approach:

  • Feature Extraction with PCA: The authors employ Principal Component Analysis (PCA) to reduce the dimensionality of the extracted acoustic features. PCA is a standard technique for dimensionality reduction, but here it's specifically applied to the sound pressure level features of gas pipeline flow noise. The goal is to retain the most informative components of the signal while discarding noise and redundant information, thus simplifying the subsequent classification task.
  • Pattern Recognition with BPNN: A Backpropagation Neural Network (BPNN) is then used for classification. BPNN is a well-established supervised learning algorithm for neural networks. The authors detail their optimization process for the BPNN, including selecting the number of input, hidden, and output nodes, determining initial weights, and tuning the learning rate and desired error. This tailored BPNN is trained to distinguish between gas pipeline flow noise and strong background noise signals.

The study also involves:

  • Experimental Platform: A closed-loop gas circulation system was built to simulate various pipeline conditions and generate flow noise.
  • Numerical Simulation: LMS Virtual.Lab was used to simulate gas pipeline flow noise propagation through different pipe geometries and soil conditions.
  • Data Collection: Both experimental and simulated data were collected, along with real-world background noise samples.

Datasets

The paper utilizes a combination of simulated and experimental data:

  • Simulated Gas Pipeline Flow Noise: 75 sets of operating conditions were simulated using LMS Virtual.Lab to generate flow noise data.
  • Experimental Gas Pipeline Flow Noise: Data was collected from an experimental platform designed to mimic gas pipeline flow noise.
  • Experimental Strong Background Noise: 25 sets of strong background noise signals (e.g., footstep noise, speech, construction noise) were recorded.

The paper also mentions using the NOISEX-92 noise dataset in the literature review, but it's not explicitly used for training or evaluation in this study.

Key Insights and Ideas

  • PCA's Effectiveness in Noise Reduction: The study highlights PCA's ability to effectively reduce the dimensionality of acoustic features while retaining crucial signal information, leading to a lower information loss rate compared to other methods like Wavelet Transform (WT), Discrete Fourier Transform (DFT), and Fast Fourier Transform (FFT).
  • BPNN's Adaptability to Noise: By carefully optimizing the BPNN architecture and training parameters, the authors demonstrate its capability to learn and classify gas pipeline flow noise signals even in the presence of significant interference.
  • Synergy of PCA and BPNN: The combination of PCA for feature extraction and BPNN for classification proves to be a powerful approach, outperforming other feature extraction methods when integrated with BPNN. This synergy is key to achieving high accuracy in noisy environments.
  • Real-World Relevance: The study's focus on simulating and experimentally validating the method under realistic urban conditions (including soil porosity effects) lends significant practical value to the findings.

Key Results

  • High Recognition Accuracy: The proposed PCA + BPNN method achieved a recognition accuracy of 97% for classifying gas pipeline flow noise signals against strong background noise.
  • Robust Performance Metrics: The method also demonstrated a recall rate of 96% and an F1 score of 96.5%, indicating strong overall performance.
  • PCA Superiority: PCA showed the lowest information loss rate among the tested feature extraction methods (PCA, WT, DFT, FFT), indicating superior retention of important signal characteristics.
  • Simulation Accuracy: The comparison between experimental and simulated sound pressure levels showed a maximum discrepancy of only 2.68%, validating the accuracy of the numerical simulations.

Limitations and Future Work

  • Limited Scope of Background Noise: While the study included various background noises, the authors suggest integrating more diverse sensor data and exploring advanced fusion algorithms to further enhance recognition accuracy and adaptability.
  • Need for Large-Scale Field Tests: The authors acknowledge the need for large-scale field tests to confirm and optimize the practicability of the model in real-world, large-scale urban environments.
  • Exploration of Deep Learning: The paper suggests exploring advanced data analysis technologies like deep learning for potential improvements.
  • Extension to Other Pipelines: The proposed method has potential applications beyond gas pipelines, including water pipes and cable lines, and could be adapted for industrial process control and mechanical fault diagnosis.

Summary updated

BibTeX
@article{liu2024feature,
  author = {Enbin Liu and Chang Lu and Zhaorong Wen and Tianshu Hao and Xudong Lu and Lidong Wang},
  title = {{Feature extraction and pattern recognition of gas pipeline flow noise signals in a strong noisy background}},
  journal = {PeerJ Comput. Sci.},
  year = {2024},
  doi = {10.7717/peerj-cs.2087},
}
Figure