Research On Classification Method Of Pipeline Defects Based On Support Vector Machines And Design Of Defects Visualization Software

Oil pipeline operation in good condition is crucial for its fault diagnosis with the rapid development of pipeline transportation. Because of the inevitable corrosion and man-made sabotage, the accidents of oil pipeline leakage, especially submarine pipeline, occurred frequently in recent years, they have terrible influence on people, causing the economic loss, the waste of resources, and the pollution of the environment. Therefore, it has theoretical significance and practical value, to do the research on the pipeline leak detection technology.During pipeline running, there are various types of defects on the pipelines’inner and outer. To identify all the abnormal signals of Magnetic Flux Leakage requires too long time and the recognition accuracy is not high enough. However, in general, it only needs to determine the damage of the pipelines and identify the defects’ types qualitatively in regular pipeline assessments. For this reason, aiming at the classification of the pipe defects, the characteristics of the Support Vector Machines (SVM) is analyzed, and a new multi classification method for pipeline defects is proposed based on the SVM in this thesis. As it is not convenient to look over the defects data, defects visualization software is designed in the thesis. The contents of this thesis include the following aspects. First, the thesis analyzes the characteristics of the classification for the pipe defects, compares the applicability of the traditional methods used for the issue of pipeline defects, such as the support vector machines and the neural networks and so on, and does some theoretical analysis of the SVM classification. Second, aiming at the deficiency of Binary Tree (BT) SVM method, a new method of hierarchical clustering based on the separation degree between classes is proposed, a model for classification is designed according to this new method, the flow chart of training and testing algorithm is given, and the simulation is described to test and verify. Third, the model above is applied to classification for defects of pipeline. The classification performance of the One Versus One, One Versus Rest and the improved hierarchical clustering BT SVM method, has been verified by simulation in this thesis. The result proves that the improved hierarchical clustering BT SVM multi classification method has good effects not only on the classification accuracy and the test rapidity, but also on the promotion capacity. Finally, defects visualization software is designed in the thesis, to display the defects data of the magnetic flux leakage graphical through the graphical types, it is convenient for monitoring workers to check the pipeline defects and analyze the security of the pipeline.For the improved BT SVM multi classification method which is adopted in this thesis, the theory and simulation experiments have fully proved the feasibility and the effectiveness for the classification of pipeline defects. The expected effect and requirements have been achieved, and the complexity of the pipeline defects recognition has been effectively reduced.