Experimental research on in-pipe leaks detection of acoustic signature in gas pipelines based on the artificial neural network

Experimental research on in-pipe leaks detection of acoustic signature in gas pipelines based on the artificial neural network

Acoustic signature

Artificial neural network

In-pipe inspection

Leaks in gas pipelines

Gas pipe leakage is a common and significant problem around the word. To detect the leakages, an in-pipe detector mounted on an acoustic inspection module is a direct and reliable solution. In this study, an in-pipe detector is designed, and an experimental platform is built to record the acoustic signal of leaks in laboratory environment. The experimental results show the signal average amplitude in time domain under a leak is higher than that under no leak and would decrease as the increased distance between the detector and the leakage point. However, the effects including the environmental noise and detector driving audio will increase difficulty to distinguish the signal characteristic. Therefore, a band-pass filter which is 1500–3500 Hz is select to eliminate distribution through the filtering, firstly, which is used to improve judgment accuracy. Moreover, the signal is then transformed to frequency domain by the fast Fourier transformed (FFT) and the power spectral density (PSD). The leaks can be judged by the change of the existing peaks around 50 Hz, 1800 Hz and 3100 Hz. To predict the leakage more precise and reliable, the artificial neural network (ANN) is selected to analyze the data. The data in frequency domain is import into a BP neural network model, and there are 14 selected cases to validate whether the pipeline has leakage. Validation results show that the leakage can be recognized with a precise of 96.87% by the trained ANN model. It indicates that this model has achieved a reasonably good performance for leak recognition. This study will provide a detection reference for the in-pipe actual inspection.

109875

183

Wang, Wenming

Mao, Xingxiang

Liang, Haiguan

Yang, Dashan

Zhang, Jifeng

Liu, Shuhai

Measurement

2021

journalArticle

0263-2241

10.1016/j.measurement.2021.109875