Research on Wheel Defect Recognition Based on VGG Network and Electromagnetic Ultrasound

In this paper, based on the principle of ultrasonic phased array detection, a two-dimensional simulation model of 16-channel electromagnetic ultrasonic phased array detection for wheels was established by COMSOL Multiphysics finite element simulation software. By using this model, the detection process of cracks at different angles of train wheels is simulated, and the echo signals received by 16 array elements are collected. These echo signals are preprocessed and used for the training of convolutional neural network, and the classification model of the electromagnetic ultrasonic phased array for the detection of train wheel defects is obtained. Through simulation data verification, the model can identify the inside of the wheel with or without defects with an accuracy of 80%. The method can quickly form the echo data into the image set for deep learning. The training speed and recognition speed are much higher than that of training and recognition with 16 element waveforms, which improves the efficiency of defect recognition.