The Research On Pipeline Leak Acoustic Emmission Detection Based On Support Vector Machine

Pipeline equipment plays an important role in social development. With the increase of pipeline equipment, acoustic emission (AE) detection of pipeline leakage is also growing concern.AE signal of the pipeline leakage exists the subjectivity on the feature extraction, the extracted parameters can not fully describe the features of pipeline leakage, and the process of state identification is very complicated, so the article proposes a wavelet packet decomposition to the frequency range of energy, as the feature vectors of pipeline leak AE signals, can effectively reflect the time-frequency characteristics. The article also proposes a Kernel Principal Components Analysis (KPCA) to compress the dimension of feature vectors. This method reduces the computational complexity of pipeline state identification.Traditional pattern recognition has some problems of small training samples and slow convergence. SVM solves the actual problems of the small sample, nonlinear, and high dimension when the support vector machine (SVM) is introduced into the field of pipeline leak detection. According to the failure priority of pipeline state, the article proposes a multi-classification algorithm based on "improved binary tree". This method improves accuracy and efficiency of the recognition process through building a multi-fault classification model.The experimental study and analysis show that through analyzing the characteristics of pipeline leak AE signals, and selecting the feature extraction of wavelet packet and feature selection of kernel principal component analysis, new characteristic parameters can be constructed as input vectors. The multi-classification algorithm based on "improved binary tree" can identify the pipeline state; this multi-classification method can achieve better results of classification.