Research on partition method of bearing circumferential surface based on machine vision

As a key component of industrial machinery, the surface defects of bearings will directly af-fect the safety of industrial production, so it is of great significance to carry out quality inspec-tion of bearings in actual industrial production. Because the defects in different positions on the cir-cumferential surface have different damage degrees to the bearing, the defects in the core area of the circumferential surface are the most harmful. Therefore, based on machine vision inspec-tion, which is the mainstream apparent quality inspection technology of industrial prod-ucts. The essence of the bearing circumferential partitioning task is to accurately extract the up-per and lower edges of the bearing in the image. In order to solve the problems of uneven image il-lumination, oil stain and bump at the bearing edge in the bearing edge extraction task, this the-sis proposes a bearing demarcation line detection algorithm based on adaptive Butterworth homomorphic filtering and multi-scale Gaussian line fusion. Firstly, to address the problem of uneven bearing illumination, this thesis proposes an image adaptive threshold segmentation al-gorithm based on Butterworth homomorphic filtering. The Gaussian filter in the homomor-phic filter is replaced by the Butterworth filter, which effectively improves uneven illumination. Then, in combination with the adaptive threshold segmentation algorithm, the appropriate pa-rameters were automatically selected to preprocess the image and enhance the edge feature in-formation. Subsequently, to eliminate the influence of defects, this thesis proposes a demarca-tion line coarse localisation algorithm based on a class-aware hashing algorithm. The algorithm auto-matically locates near the bearing demarcation line and selects the appropriate region of inter-est to minimise the false edge problem caused by defects. Finally, to address the issue of bearing demarcation line fracture, a multi-scale bearing edge detection algorithm based on Gaussian line is proposed. The algorithm accurately extracts the coordinates of the dividing line by fusing the convolution map of multi-scale Gaussian derivative images, and uses the least squares method for curve fitting. Through the above method, the accuracy of the final bearing partition reached 99.23%.