Research On Machine Vision Inspection Method For Deep Groove Ball Bearings With Embedded Balls

The key data detection of finished bearings is an effective means to ensure the quality of bearings after assembly.The ball measurement and lack of ball discrimination of deep groove ball bearings have not been well solved in the visual detection.In this paper,the machine vision inspection method of precise measurement and missing judgment of inner ball of deep groove ball bearing in local visual state is studied.This paper first addresses the problem of extracting partially visible ball edge images of deep groove ball bearings under structural occlusion.The image acquisition of the effective edge of the ball is solved through the research on the selection of light distribution and image acquisition methods.Then a semantic segmentation model based on improved U-Net was designed to complete the effective edge information extraction of the roller ball according to the requirement of high accuracy for detection and the imaging characteristics of the bearing.This method adds expanded convolution to the shrinkage path of the original U-Net network as well as attentive mechanism and multiscale feature integration to make the network more adaptable to the task of small target segmentation and decrease the loss of detail information due to noise disturbance during segmentation,finally achieving the requirement of fine segmentation of bearing balls.The loss of detailed information will eventually meet the requirements for fine segmentation of the bearing balls.Then,for the problem of short ball feature arcs and difficult circle fitting in segmented images,this paper designs a combination of Shi-Tomasi corner detection and modified Canny edge detection algorithm to determine ball feature arcs and coordinates.Finally,the least squares method is used to fit the complete ball shape and parameters with accurate and clear edges,and then the actual ball size is calculated by combining the system calibration to establish the relationship between the digital image and the actual object size.The paper concludes with an experimental verification of the above deep groove ball bearing embedded ball detection method.The outcome indicates that the method has an accurate measurement of 0.01 mm for measuring the ball size of assembled deep groove ball bearings,a false detection rate of 2.125%,and a 100% accuracy for the detection of the leakage of balls in the assembly process,and can measure all the balls in the whole bearing at the same time,which provides methods and ideas for the theoretical research and practical application of bearing assembly inspection.