ABSTRACT
In a submarine liquid pipeline, the timely detection of tiny amounts of leakage (less than 1 L/min) can have a critical impact on pipeline safety. Hence, a method based on a passive acoustic internal detection for pipeline small leakages was established. Firstly, the variation laws between leakage rates and sound power levels with different leakage apertures and internal pressures were simulated using Ansys Fluent 19.0 software. Furthermore, an experimental platform for pipeline small leakage detection was constructed. The acoustic experiments of small leakage detection were carried out under different leakage apertures and internal pressures using a high-sensitivity acoustic sensor placed into the pipeline. It was found that consistency between the experimental results and the finite element simulation results was established. A novel improved empirical mode decomposition (IEMD) was used to denoise the leakage signals. The characteristic parameters of different signal processing fields were extracted. A novel decision tree support vector machine (DTSVM), based on parameter optimization, was used to construct a classification model to identify small leakages through different apertures. The results show that compared with three commonly used models, the accuracy of this novel model of pipeline small leakage detection was 92.2%. This model was found to improve the small leakages detection rate of different apertures and also provided a theoretical basis for developing an acoustic internal detector for submarine liquid pipelines.
Highlights
A method based on a passive acoustic internal detector for the detection of submarine liquid pipeline small leakages was established.
A novel improved empirical mode decomposition (IEMD) was used to denoise the leakage signals. The SNR of signals was improved and the RMSE of signals was reduced compared to other noise reduction methods.
A novel K-DTSVM model was used to construct a classification model to identify small leakages through different apertures. The accuracy of this new method was 92.2%, compared with three traditional models.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Additional information
Funding
Notes on contributors
Yundong Ma
Yundong Ma is a doctor at China University of Petroleum (Beijing). His research interests are non-destructive testing, pipeline integrity, safety engineering and evaluation.
Shaohua Dong
Shaohua Dong is a professor at China University of Petroleum (Beijing). His research interests are pipeline integrity, safety engineering and evaluation.
Hang Zhang
Hang Zhang received his Ph.D. degree from China University of Petroleum (Beijing). Currently he works at China University of Petroleum (Beijing). His research interests are pipeline robot, pipeline integrity.
Qignqing Xu
Qingqing Xu received his Ph.D. degree from University of Alberta, Canada. Currently she works at China University of Petroleum (Beijing). Her research interests are pipeline integrity, process control.
Haotian Wei
Haotian Wei is a doctor at China University of Petroleum (Beijing). His research interests are non-destructive testing, pipeline integrity.
Weichao Qian
Weichao Qian is a doctor at China University of Petroleum (Beijing). His research interests are deep learning, image processing.