Leak Detection For Gas Pipeline Based On Blind Source Separation

As a kind of clear energy source,natural gas is very important in the industrial production and people's daily life.Natural gas is usually transported by the pipelines.In the process of transmission,once leakage occurs,it may cause great environmental pollution and damage to people's lives and economic losses.Leak detection is one of great important technology to guarantee the safe operation of the pipelines,and it has been widely studied in the field of natural gas transmission.However so far,the accuracy of leak detection and location is not very high.Therefore,it is of great significance to improve the accuracy of the leak detection and location and detect the level of the leakage.Based on the laboratory device of acoustic system,leakage detection technology is studied in this paper.Leakage acoustic signals are collected and then filtered by combining the wavelet transform and blind source reparation.A large number of experiments shows that this method can greatly improve the signal-to-noise ratio of the leakage signals comparing with the traditional wavelet transform de-noising method.Therefore,when this method is used to locate leakage together with the cross-correlation function of time delay estimation method,the obtained location error is relatively smaller.By analyzing the statistic features of the filtered acoustic signals in the time and frequency domain,features are extracted for leakage identification and leakage detection is realized by using PCA method.For the case of small samples,semi-supervised fisher discriminant analysis(SFDA)method is introduced and successfully applied to classify the compressor start-stop,valve switch,leakage et al.Five different kinds of working conditions are effectively identified,and the accuracy of the leakage is improved.Experiments show that the classification of the SFDA method is more suitable for small sample data compared with the FDA method,and can obtain better classification result.By using fuzzy c-means clustering algorithm,classification feasibility is studied among the stable signals and the leakage acoustic signals with different leakage aperture.The different leakage acoustic signals are successfully classified by using SFDA method.Moreover,SFDA method can get better classification results than FDA method.