Accurate Detection Method Of Aviation Bearing Ball Assembly Based On Semantic Segmentation

The accuracy of ball assembly of high precision aviation bearings is an important test index of its quality and performance,which determines whether it meets the factory requirements.This article focuses on the lack of the present state of air bearing ball finished product assembly,for ball is held serious shade air bearing,according to its local features visible only through forecast complete ball shape detection difficulties,and its high requirements for detecting precision,based on the image semantic segmentation of local ball with complete shape fitting,And by measuring the size of the ball ball to detect the wrong and missing ball.Firstly,a detection platform for ball assembly is built.According to the reflective characteristics of the smooth metal texture of the aviation bearing and the problem that the light in the slot where the ball is located is difficult to penetrate,an optical environment is built,and the hardware selection of the image acquisition system is carried out,as well as the design of software scheme.Then,the convolutional neural network based semantic segmentation method was selected to segment the local ball in the image of aviation bearing because the ball was seriously blocked and the background was complex.Aiming at the characteristics of small object segmentation of ball ball,and the poor contrast between boundary and background requires fine edge,based on the Deeplab V3+model,the feature extraction network is improved,and the decoder structure is designed to combine more underlying detail features,that is,the local occlusion ball segmentation model with stronger decoding ability.Finally,the similarity measure of edge position and gradient was carried out between the segmented local ball edge and the original ball edge,and the ball edge was further refined in the original image.In view of the short and disconnected edges of two arcs,a random circle detection algorithm based on the evaluation of circle fitting degree was used to determine the complete shape of the fitted circle,namely the ball,by looking for the optimal value and measuring it.According to the results of segmentation and size measurement,we can judge whether the ball is misloaded or not.Finally,the proposed algorithm is verified by experiments.The results show that the improved ball segmentation model in this paper can achieve accurate segmentation of partial occlusion ball,and the average crossover ratio of the segmentation results can reach 0.748,which can correctly judge whether the ball is missing or not.After the subsequent processing of the segmentation results to fit the complete ball,the ball size was accurately measured and the measurement accuracy was up to 26.73 micron,and the error detection rate of ball misloading was 1%.The method proposed in this paper can realize the accurate detection of the aviation bearing ball assembly,and provide a useful reference for the detection of the ball target with partial shielding,which has certain value in engineering application and theoretical research.