Convolution neural network for classification of magnetic flux leakage response segments

Abstract

Magnetic flux leakage (MFL) inspection is one of the most commonly used nondestructive testing (NDE) technologies. This paper proposes a novel method for classifying the MFL response segments based on convolution neural network (CNN). In order to skip the procedure of saving the normalized MFL segment and save some computing time, a normalization layer is added to the proposed model. Moreover, the rectified linear units (ReLUs) is employed as the activation functions in the convolution layers to allow the proposed model to easily obtain sparse representations. The performance of the proposed model is tested by the real MFL data collected from the experimental pipelines. The results demonstrate that the presented method can achieve a satisfactory accuracy of MFL response segment classification and can be applied to practical application.