Machine Vision Based On-line Recognition For Defects On Multiple Surfaces Of The Universal Joint Bearing Outer Ring And Its Application

Bearing is a key component in mechanical assembly,and plays an important role in supporting the rotating body and reducing the coefficient of friction.Surface defects that occur during the processing of a bearing can cause vibration and abnormal noise,even will affect the service life of the machine.Therefore,it is of great significance to realize on-line detection recognition of bearing surface defects to ensure the quality of bearing products.Aiming at the special bearing of the universal joint bearing outer ring,based on the relevant technology of machine vision,a segmentation algorithm of the object area of the outer ring of the bearing is designed,the multi-modal associated shape descriptors are introduced to solve the problem of feature extraction,and the training of the object recognition model under the condition of incomplete samples is realized.Finally,the on-line recognition scheme of the surface defects is applied in an enterprise implement the deployment verification in the production line of the industry.The main contents of this article are as follows:The first chapter summarizes three techniques of surface defect task based on machine vision,and elaborates its current research status in bearing surface defects.This paper analyzes the difficulties caused by multiple surfaces in the recognition of the surface defects of the outer ring of the universal joint bearing,and puts forward the research content and organization structure of this paper.In the second chapter,aiming at the difficulty of defect target region segmentation caused by the different shape features of the bearing outer ring,I design a multi-shape target region segmentation algorithm suitable for the bearing outer ring.Based on the idea of region growing segmentation,the 2D Mallat algorithm is used to realize multi-scale defect edge preselection,and K-means clustering algorithm is used to cluster similar defect edge points,which solves the limitation of manual selection of seed points in traditional algorithm and realizes image segmentation of defect areas on different surfaces.The third chapter proposes a method for feature extraction of target regions based on multi-modal associated shape descriptors.Fourier shape descriptors and curvature histograms are used to extract shape feature information in global and partial modes.In order to prevent “redundant calculations” and “dimensional disasters” caused by associated features in different modalities,the correlation of feature variables is determined through canonical correlation analysis,and different modal features are optimally composed to construct a more robust feature extraction descriptors.In the fourth chapter,in order to solve the problem of incomplete image data acquisition in polyhedral defect recognition,on the basis of traditional support vector machine model construction,the generalization ability of the model is improved by transductive learning while reducing the cost of marking resources.For the problem that there are multiple types of surface defects in the outer ring of bearing,the traditional scheme of combining multiple binary support vector machines is abandoned,and the hypersphere structure is introduced to solve the multi classification problem of support vector machines.Relying on the actual project of the enterprise,based on the actual recognition requirements of the industrial scene and hardware equipment conditions,we developed the surface defect recognition software for the universal joint bearing outer ring.Through debugging and deploying at the production site,and the reliability of the surface defect recognition system of the outer ring of the bearing is verified through the trial operation.The sixth chapter summarizes the content of the full text,analyzes the limitations in the research process,and looks forward to future research directions.