Research On Machine Vision Detection Method For The Assembly Defect Of Deep Groove Ball Bearing

The pros and cons of product quality are influence the efficiency and reputation of the enterprise directly.The product quality must be inspected strictly after the assembly is completed.The technolog of machine vision is being taken seriously by all walks of life which as a representative of the non-contact detection and automatic detection.In this paper,taking 6201 deep groove ball bearing as the research object and doing some study systematically for the defection technology of machine vision of bearing defects.The main detection process of machine vision is image acquisition,image preprocessing,image location,image segmentation,feature extraction and final results analysis of experiment.1.Build bearing image acquisition platform,which included three parts named camera,light source and bracket respectively.After comparison and optimization,the experiment selected the Manta G-125 B industrial digital camera finaly.Because of it carried the algorithm of gamma could obtain the better clear image under the same conditions.The combination of backlight source and co-axial light source are used as the illumination system to obtain the bearing image information.Results show that the details of the rolling element and chamfer waiting detection area are clear,which contributes to the subsequent processing of the image information.2.Analyze the Hough algorithm and polar coordinate expansion algorithm,and locate the detected bearing with the improved Hough algorithm.Using polar coordinate expansion algorithm to expand the detected bearing from a toroidal state to a rectangular state.Then find the rolling element area.The number of modules with a gray value of zero in the rolling element area is used as a criterion to determine whether the bearing rolling elements are missing.Experimental results show that the detection algorithm proposed in this paper is very effective for the detection of bearing rolling element defects.3.The three algorithms of connected region extraction,Harris corner detection and least-squares fitting are used to detect the cage area in the bearing image.It shows that there were different recognition rate by using the three methods to recognized the bearings which cage is skew or not and the Harris corner detection algorithm had the best results.The recognition rate of intact bearing is 96.3%,and the recognition rate of cage skew bearings is95%.4.Using image segmentation combined with Hough algorithm to extract the chamfered region of the bearing image.The experimental results show that the improved Houghalgorithm can achieve accurate positioning of the chamfer inside and outside the bearing,and obtain the ring diameter data of them.The recognition rate of chamfered bearing is 95%.However,Hough algorithm is too sensitive to the circular region extraction.The correct recognition rate of the algorithm when detecting a chamfer-free bearing is only 86%.