To detect rail top surface defects, NdFeB toroidal permanent magnets were firstly used to excite the rail detection area, and the magnetic excitation direction was perpendicular to the rail direction. So that, the magnetic circuit can be analyzed, and the model was verified by finite element simulation. Secondly, the artificial rectangular defect samples were machined on the top surface of the rail, and the magnetic field data acquisition system was developed and equipped on the inspection vehicle. With that, a linear array which consisted of three-dimensional hall sensors was designed to detect the magnetic leakage field signal from the defect samples. Finally, the features of the leakage field signal in each dimension were extracted. And then, the length, width, and depth of the defects were predicted using BP neural network algorithm. The results indicate that the method can improve the excitation efficiency, and the average absolute errors of prediction for defect length, width, and depth were 1.43 mm, 1.01 mm, and 0.63 mm respectively.
Read full abstract