Mechanical Parts Detection System Based On Computer Vision

Computer vision inspection technology is the combination of computer technology, optical information technology, AI technology. Computer vision inspection has the advantages of non-contact, flexible, intelligent, and not subject to physical limitations. It is the optimal means to achieve high-speed and high-precision product inspection on assemble line or in extreme environments, which is widely used in various industries on-site monitoring or product quality inspection.The traditional mechanical parts inspection tool includes inspection ruler, inspection tooling set,3D coordinate machine, all these tools are often flexible not enough for the reason of cost or design flaws. The more serious is that there is some inconsistent between design space and test space. The whole process of traditional inspection tools is under controlled by man. In a typical labor-intensive and capital-intensive machinery manufacturing enterprises, inspection personnel, which often accounted for20%or more for whole one. And manual inspection deeply depends on the individual’s professional skills, experience, and is not replicable, coupled with the physical sustainability which often leads to instability during the inspection. Finally, another motivation is to solve the difficulties in order to integrate inspection information on CIMS.Computer vision inspection is the mapping process from visual information to the product quality attributes. From the view of content and scope, computer vision inspection can be divided into two parts:quantitative measurement and qualitative check. Dimensions size measurement is the top content by analyzing the image feature points, fitting a regular line, surface, extract the dimensions information in quantitative measurement. Quantitative measurement is suitable for number properties while qualitative check for quality label. Computer vision inspection can be greatly shown to its advantage in the classification of fuzzy attribute label or classification task is not clear. The main works are enumerated as followed:the accurate3D parameter size value and process quality label check.In this paper, a mechanical parts measurement system are designed based on computer vision detection model, which integrated illumination, camera calibration, binocular stereo imaging, edge contour extraction and surface fitting, and finally to the inner hole parameter. The holder is an important part of the auto transmission, of which the angle between an axis of hole cylindrical and another axis of hole conical to be measurement. Traditional measurements tools are often inefficient or insufficient accuracy. The system in this paper achieved the expected results.Another work is to establish process quality check system learned from image classification. The technology includes interest point search, the description of points of interest, the visual dictionary generation and pattern recognition. The quality check of the welding process is important task for mechanical parts, while the variability of the welding surface determines the difficulty of welding process quality classification, the use of computer vision for the qualitative detection can improve the efficiency and accuracy of detection to a certain extent.