Pipelines have become principal means of oil and gas transportation. However, pipeline leakage takes place due to some natural or artificial damages, which may cause loss of life and properties along with environmental pollution. So scientific workers and engineers investigate and develop several approaches for leakage monitoring, such as acoustic method, pressure gradient method, pressure point analysis, negative pressure method, and mass balance. Above methods have some common shortcomings. They can only detect leakage after enough accumulative total amounts, which is too late for prediction. Besides, the activities to cause leakage and its locations are difficult to find and tell.
Now, more and more researchers try to develop a pre-warning system that can detect damage activities as soon as it begins to drill hole or excavate soil just before damage occurs. An acoustic detecting system being able to discriminate background noise and hammering signals has been developed. But it can only differentiate vibration signals with a limited distance of 2.4km.
Recently, the distributed optical fiber sensing system is regarded as an important direction to detect vibration events around pipeline. Meanwhile, a general long-distance monitoring system with more than 50km is proposed, and its hardware has been accomplished and deployed. Since long-distance oil pipeline may go through a variety of different environments, such as railways, rivers, roads, mountains and farmland, the interested vibration signal will be covered in many kinds of background noise. Therefore, a recognition and classification method to abnormal events with vibration analysis and back propagation neural network (BPNN) is proposed for distributed optical fiber sensing system along the pipeline.
In the novel algorithm, original vibration signal is pre-processed and segmented according to energy threshold and sliding window. Through statistical and short-time Fourier transform (STFT) analysis in time and frequency domain, energy ratios and frequency centroid are extracted as feature vectors, which can describe and distinguish distribution characteristics of each vibration event effectively. At classification, event set is divided firstly into discrete and continuous events with kurtosis, which can decrease classified event dimension and improve recognition accuracy. And then BP artificial neural network is applied to identify damage and non-threatening events. Experiment results show that proposed algorithm can differentiate discrete events with accuracy rate of 99%, while continuous events with 97.5%.
In this research, five types of damage activities and three kinds of non-threatening activities were classified accurately. The future work will improve the algorithm and identify other kinds of events, in particular, mixture of several kinds of events.
The work with the title of “Abnormal events recognition and classification for pipeline monitoring systems based on vibration analysis and artificial neural networks” has been published online: http://asadl.org/poma/resource/1/pmarcw/v19/i1/p055021_s1 and on Proceedings of Meetings on Acoustics (POMA - ICA 2013 Montreal) (Vol.19, June 2013, pp. 055021, 7 page).