Study On Fault Diagnosis Of Pipeline Leakage Signal Based On Time Frequency Analysis

With the development of petroleum and natural gas industry, pipeline transportation has become one of the most important carriers of energy transportation, and its safety problems have been paid more and more attention. Leakage fault diagnosis is an important means to prevent pipeline failure and reduce the risk of pipeline. To realize the leakage of pipeline in different fault types and different levels of damage diagnosis is bound to choose treat the measured signal is preprocessed with better signal noise reduction method, and to build a more sensitive signal detection system. This paper is focused on the two points above selected improved wavelet threshold function to complete the signal denoising, and introduced a feature extraction technique based on the combination of empirical mode decomposition(EEMD) and sample entropy, through BP neural network the fault state recognition, aims to build a more effective more accurate pipeline leakage fault diagnosis system.Firstly, the characteristics and defects of several kinds of wavelet threshold denoising were studied, and an improved wavelet threshold denoising method was proposed, which can achieve a very good denoising effect by adjusting the parameters. Improved wavelet threshold function overcomes the deficiency existing in the soft and hard threshold function, denoising performance is superior, in a greater degree to retain the characteristic information of the original signal, is helpful to improve the accuracy of monitoring system of pipeline leak.Secondly, the paper introduces the empirical mode decomposition(EMD) algorithm in theory and its application in signal decomposition. For the EMD exist false component and modal mixed stack defects, EEMD method is introduced for the auxiliary noise, realize the decomposition of the original signal in accordance with the different frequency scale layer by layer, to extract local feature information of the signal, and through experiments demonstrate the superiority of EEMD and EMD algorithm compared, in a certain extent to restrain the mode mixed stack. Finally, analyzed the parameters of EEMD.Then, according to the characteristics of the complexity of pipeline leakage signal presents introduction of entropy theory is used to measure the degree of the signal complex samples, is to use the sample entropy to quantify the for the IMF component EEMD, access to the pipeline sound signals of different frequency scale, the complexity of the information as the input of pattern recognition. In the process of feature extraction, in order to reduce the computational complexity, based on Shannon entropy screening guidelines, to determine the need for the sample entropy weight calculation of the IMF, to provide a basis for the pipeline leakage signal fault diagnosis.Finally, with natural gas pipeline leak detection laboratory sound signal data as analysis object, pipeline is studied in different types and different damage degree in both cases the fault diagnosis. The results show that based on EEMD and sample entropy in time-frequency analysis method effectively under the two situations of pipeline fault feature can be described,using BP neural network for pattern recognition and comparison with other methods in classification accuracy and has good recognition effect.