Magnetic Flux Leakage Image Detection in Axial Alignment Based on ResNet

The state detection of shaft alignment is an important industrial problem, and the magnetic flux leakage data set of shaft alignment state can be established by using the magnetic flux leakage detection technology. Aiming at the question of whether there is parallel deviation and angle deviation in the shaft alignment problem, this paper proposes an improved ResNet model that can be used to detect the shaft alignment state. Firstly, the ANSYS simulation software is used to establish a shaft alignment state model, and then the shaft alignment magnetic flux leakage data sets under different parallel deviations and angle deviations are simulated. Finally, an algorithm that can be used for shaft alignment state detection is designed based on the ResNet model. Among MFL data, axial MFL data, radial MFL data and circumferential MFL data are of different importance to axial parallel deviation and angular deviation. In this paper, self-attention mechanism is introduced to learn the importance of different channels to solve this problem. At the same time, the Stacking method is used to integrate the improved ResNet model with the Xception model, which improves the accuracy and versatility of the algorithm. Finally, the accuracy of the model is verified by experiments, as well as the influence of different steel materials and excitation strengths on the improved algorithm. The experimental results show that the improved algorithm can achieve an accuracy rate of 96.14% in the detection of the shaft alignment state, and has good versatility for different specifications of steel.