As an important equipment in the sintering process，the health state of the wheel axle of the sintering machine is of great significance to the safety production of the sintering machine. As a typical feature of axle failure，wheel oscillation can be used to monitor and warn of axle failure. In view of the lack of effective means for early warning of wheel axle failure of sintering machine in the production site，a detection scheme based on wheel swing angle is proposed，and a detection algorithm for wheel swing of sintering machine based on improved DeepLabv3 + is constructed. The algorithm selects the model with the best wheel segmentation performance by comparison. According to the segmented wheel mask image，the minimum external rectangular frame between the reference line and the wheel is determined，the swing angle of the wheel is determined by its included angle，and early warning rules are set. By comparing the wheel swing angle with the early warning rules，the wheel swing is classified according to the risk level. The experimental results show that the average intersection and union ratio of the optimally improved DeepLabv3 + model is increased by 1. 7% and the model size is reduced by 92. 9% compared with the initial model. It is used as a segmentation model to detect the swing angle of the wheel，and the angle error reaches less than 0. 6° in the set range. The average correct detection rate of 96. 1% is measured by using the set early warning rules to determine the wheel swing fault，and the automatic detection system for wheel swing is designed on this basis，which could provide technical support for the diagnosis of wheel axle fault.