Research On Scratch Detection Technology Of Axle Inspection Surface Of Electric Multiple Units Based On Binocular Vision

At present,in the Electric Multiple Units（EMU）manufacturing enterprises,the surface scratch detection of the EMU axles is performed manually,which not only requires the inspection engineer to have a wealth of practical experience,but also may cause detection errors that are difficult to observe with the naked eye due to manual operations.With the development and progress of science and technology,machine vision has gradually become an important industrial system that can replace the human eye.So,the use of machine vision to detect axles is a direction worth studying.Although machine vision has great advantages in image recognition,scratch detection and other fields,there is no shaped solution that can be applied to the detection of various scenes.Therefore,for specific scenes,specific analysis is required.And an independent and specialized program is determined for implementation.The research content of this thesis is as follows:（1）A working platform for detecting scratches on the axles of the train is built.As for camera models,Intel REALSENSE D435 was selected as the camera of the binocular vision system after comparing.The influence of the light source’s illumination method and the shape of the light source on the subsequent image processing is studied.And the shape and lighting method of the light source are determined.（2）The imaging model of the camera is established.According to Zhang Zhengyou’s calibration method,the camera calibration experiment was carried out in the MATLAB R2018a operating environment.After calculating the internal parameter matrix of the binocular camera and the external parameter matrix of the vision system,the binocular camera was subjected to an epipolar correction experiment according to the parameters.（3）According to the characteristics of the scratches on the axles of the high-speed train and the characteristics of the collected images,the image preprocessing process is designed.An edge detection method based on gray correlation is proposed.A symmetric circular filter template is designed.The image is filtered with the symmetric filter template.The correlation of gray values in the target pixel template is calculated.Correlation is used to determine whether the pixel is on the edge of the scratch.（4）A scratch endpoint detection method based on unary gray entropy is proposed.According to the difference in entropy in the neighborhood of the scratch in the image,the location of the end of the scratch is determined.Finally,according to the obtained endpoint coordinates,triangulation is used to obtain the spatial coordinates of the scratches,and the three-dimensional extraction of the scratches is completed.The experimental results show that the edge detection algorithm proposed in this paper can detect the complete edges of the image without the phenomenon of discontinuous edges.The endpoints detection method based on unary gray entropy proposed in this paper can detect the actual endpoints of scratches,and the variance of multiple measurements of the length of scratches after 3D extraction are all less than10^-3mm²,which is in line with expectations.