Research On Surface Defect Detection Technique Of Micro Bearing Outer Ring Based On Computer Vision

The bearing is a very important part of the mechanical industry. As it is applied to many fields, the processing precision and quality of the bearing concern service function and life-span of mechanical product. Therefore, the detection of bearing's producing qualities has been the question which the bearing producers care a lot all the time. Ningbo is a base of our country's bearing. It has more than 1,000 bearing enterprises. Bearing has already become the important foreign exchange earning product in Ningbo city. In recent years, many producing enterprises have accomplished the automatic manufacture of bearing through the technological revolution. However, it is still the manpower visual measurement to detect to the surface defect and to reject them. The method is not only much workload and inefficiently, but also low reliable. It affects product quality, users often returned. This has not only caused economic losses to the enterprise, but also damaged to reputation. To solve the problem for a long time, the bearing producers need this automatic detection instrument urgently.As the enterprises demand on product quality and production efficiency is on the rise, computer vision technology based on digital image processing technology has been more and more attention to the people. Mass production practice shows, the pressure pits and scratches to the size, depth and location of micro bearing outer ring surface are random. Using contactless detection is difficult and inefficient, so using image measurement and recognition technology to non-contact detection is the best solution to this problem.The research and application of international computer vision system has begun from 80 years last century. Computer vision inspection is booming. New conception, theory, technology come forth continuously, which have been applied for vision inspection and automatic recognition of all kinds of products. In our country, industry vision system is still in a time of conception leading. Although the theory and experience of computer vision inspection have gotten some results, it has just started and dropped behind some countries. The lead corporations among every industry have begun to pay attention to automatic vision measurement, after they solved the problems of automatic production.Combining the project of industrial science and technology of Ningbo city(2005B100014)《The inspection system research of surface defect of micro part based on CCD image recognition technology》and the cooperative project of Ningbo academy-corporation《The study of detecting the surface defect of micro bearing based on CCD image recognition》, the paper profoundly studies the problem of visual inspection technology of the digital image segmentation of micro bearing outer ring surface of image pre-processing, extraction and the defects identified, it has micro bearing outer ring surface defect detection on-line.Based on the demand of the quality inspections, the paper discusses the structures of the visual detection's hardware and its selected principles, especially the design and selection of lamp-source system, and selects adapted industry digital vidicon, camera lens and lamp-source and so on. It also illustrates the function evaluation index of the visual detection technology. It points out that hardware equipment is the prerequisite and basis of the detection.Image preprocessing is an important part of the former of target detection and recognition, is also a crucial step of bearing outer ring surface defect detection system. The paper analyses some general methods to suppress noise, chooses median filtering as the image smoothing method of bearing outer ring surface defect detection system through experiments, and discusses the image threshold segmentation and edge detection. In the aspect of image threshold segmentation, it mainly discusses the peak-valley method and maximum variance threshold method, compares the advantages and the disadvantages of the threshold segmentation method. Finally the paper selects iterative threshold method. In the aspect of the edge detection, it mainly discusses the common edge detection operator, compares the pros and cons of the edge detection method through the experiment, so that the paper selects canny operator to detect edges.There are not only the existences of the goal of defect detection, but also frivolous image information on the image after pretreatment. Therefore, it derives on how to extract target objects accurately. The paper uses edge linking based on the edge growth operator. It solves the issue of edge closed, links a small gap between the two sections. Then Hough transform detects and eliminates the most beelines. Finally the paper uses 8 connection tagging objects, the goals are extracted and described, then calculates the corresponding parameters.The paper introduces the basic concepts and main methods of pattern recognition, selects the pattern recognition method—LVQ neural networks, which is suitable for the system. The paper carries out simulated training and testing to the purpose of classification. Through the detecting to the 500 bearings'experiments, the recognition rate of flawed bearings can reach up to 98.5%. It achieves the desired result.