Study On Defect Recognition For Magnetic Flux Leakage Detection In Pipeline

The method of magnetic flux leakage (MFL) detection is one of the most effective methods for the defect detection of oil-gas pipeline used by domestic and overseas experts. The system of MFL mainly concludes the forepart MFL signal collection system, the data compress part and the defect recognition part. The technology of defect recognition will materialize the important information of geometry parameters viz. the length, the width of the pipeline defects in the mode of idiographic data based on the analysis of the detected MFL data. So it will provide a scientific foundation for the proprietor that whether there needs a pipeline replacing. On account of the pipeline defect recognition problem is a inverse relation problem that it is not a uniqueness relationship between the output and the input, and there is a complex nonlinear relationship between the MFL data and the dimension of the defect, so the technology of defect recognition is a technique difficulty and researching emphasis in the field of pipeline non-destroy testing presently.A new method called the support vector machine (SVM) which is based on the statistical learning theory and according to the principle of structural risk minimization is proposed in this paper to achieve the dimension recognition of the pipeline defect in the field of MFL detection.The MFL principle and the influence on the MFL signals result from the any element are firstly analyzed in this paper. The method of finite element analysis is used to the setting up of the MFL model and simulation of the MFL signals. Then the simulation signal that has the same character as the MFL signal which is really detected by the MFL detector is acquired for the using of samples. The recognition work is done for the bowl-shape defect as a example using the SVM. 40 groups data of MFL samples acquired from the defects with different dimensions by the finite element analysis software named ANSYS are used as the SVM study samples. The other 10 groups new data are used for the examination of the SVM model. The experiment declared that the recognition errors are below 5% using the SVM for the pipeline defect recognition, so it is a effective method that has high feasibility and perfect effect on condition that with finite samples.There are some advantages using SVM for the pipeline defect recognition such as needing minor samples, having strong generalizable ability, can conquering the local extremum and is a simple method of distinguish and so on. So it will relief the workload and can heighten the reliability of the recognition result which is resulted from the SVM model.