The MFL detection method by the gradient hall sensors array for the cracks on the top surface of the rails

Abstract

Cracks on the top surface of the rails due to contact fatigue are usually narrow and variable in direction. Thus, more spatial features need to be explored when using the magnetic flux leakage detection method. In this article, we proposed a way to predict the size and direction of the cracks by the gradient hall sensors array. First, the experimental equipment was designed for leakage field signal detection. Then, the peak time differences of the leakage field signal were analyzed to predict the crack angle, and the magnetic gradient modulus was calculated to predict the crack width. Furthermore, the principal component analysis was performed to reduce the number of eigenvalues. With that, BP neural network algorithm was used to predict the length and depth of the cracks. The experimental results indicate that the method can improve the accuracy of crack prediction, and the average relative errors of the prediction for crack angle, width, length, and depth were 5.7%, 15.3%, 9.2%, and 11.2%, respectively.