| As the most import energy sources in social development, oil and gas are badly and stimulate the transportation of them develop rapidly. For pipeline is the main instrument widely used, many pipeline defects appear at the same time. AS conventional method is operated mainly by experienced staffs, position of defect cannot be precisely located.To solve these problems, methods of pipeline defect detection are analyzed and experimentally validated. The main contents of the thesis are summarized as follows:1. ANSYS is used to modeling the theoretic experimental model so as to extract the experimental data. Defect identification methods studied in this paper need a great deal of accurate data so the traditional experimental platform can’t meet the end. By profit from strong modeling ability of ANSYS, theoretical platform of experiment, materials as well as defects which generate a large number of experimental data without noise and interference can be acquired.2. The BP (Back Propagation) neural network is utilized to detect defects. Function, error, training times and network layer the experiment needs are chosen after introducing the widely used BP neural network method. Subsequently, depth detection of the defect length and size is completed by using BP neural network method.3. The RBF (Radial Basis Function) neural network is used to detect defects. Defect identification method based on BP neural network in current research has the limitation of slow convergence, instable and need long time training. To conquer these shortcomings, RBF is implemented into research of MFL nondestructive testing. Through layer selection for RBF training function, researches in training function error and training steps as well as defection of the same defect magnetic flux leakage signal under BP and RBF, it is validated that MFL nondestructive testing based on RBF algorithm is feasible. Defecting recognition precision and training steps of RBF are improved compared with BP neural network. |
PDF Full Text Request