Leak Detection of Gas Pipelines Based on Characteristics of Acoustic Leakage and Interfering Signals

Meng, Lexuan; Liu, Cuiwei; Fang, Liping; Li, Yuxing; Fu, Juntao

doi:10.32604/sv.2019.03835

Cited by 3 publications

(1 citation statement)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Traditional intelligent diagnosis methods mainly include two processes: feature extraction and pattern classification. Feature extraction usually uses a variety of signal processing methods such as mode decomposition (Vashishtha et al, 2022a(Vashishtha et al, , 2022b(Vashishtha et al, , and 2022cVashishtha and Kumar, 2021), spectral analysis (Vashishtha and Kumar, 2022), wavelet transform and statistical features (Meng et al, 2019;Zadkarami et al, 2017) to extract time-domain, frequency-domain and time-frequency domain features from original data. The extracted features are then input into support vector machine (SVM), particle swarm optimization (PSO), and artificial neural network (ANN), and Bayesian network algorithms are used to classify the faults.…”

Section: Introductionmentioning

confidence: 99%

Acoustic detection and localization of gas pipeline leak based on residual connection and one-dimensional separable convolutional neural network

Yan

Liu

et al. 2023

Transactions of the Institute of Measurement and Control

View full text Add to dashboard Cite

The leakage of natural gas pipelines may cause significant safety accidents and cause severe economic losses to society. Therefore, timely detection and location of pipeline leak is an effective measure to reduce losses. Traditional methods need to extract the leak features from signals and then input them into convolutional neural networks to realize leakage diagnosis, which is a tedious process. This paper proposes a pipeline leak detection method based on residual connections and one-dimensional separable convolutional neural network (1D-RC-SCNN) and optimizes the structure and parameters of the model. The model uses one-dimensional separable convolution to replace the traditional convolutional neural network, which effectively reduces the number of model parameters. The improved residual connection can prevent the training data from overfitting and speed up the model convergence. The leakage signal after segmentation and preprocessing is directly sent to the 1D-RC-SCNN model, the leakage features are adaptively extracted, and then, the results are output through the Softmax classifier. The method directly performs end-to-end processing on the leaked signal, which improves the data processing speed. Finally, the cross-correlation method is used to locate the pipeline leak. The effectiveness of the proposed method is verified by building an experimental platform and compared with related models. The experimental results show that the method has high accuracy for the detection and localization of pipeline leak.

show abstract