The Research On Products Defect Recognition Based On Geometry And Its Application In Mechanical Assembly

Image recognition is an emerging interdisciplinary technology which is widelyapplied in industrial field machine vision projects. Huge amounts of mechanicalproducts require fast and accurate testing and recognition to ensure and identify themanufacturing and assembly quality. However, faced to different tested object, currentimage recognition methods normally use experimental image pretreatment effect todesign recognition algorithms, which ignore the geometric characteristics of mechanicalproduct images’ shape and structure. This dissertation analyzes the geometric featuresand topological structures of mechanical products, constructs image description models,and designs a machine vision inspection system and recognition algorithms. Theprinciple work of the dissertation includes:(1) Current online based-MV inspection methods for a mechanical product areanalyzed. New models and descriptors based on image structural dimensions areproposed. A machine vision inspection system theoretical frame with efficientutilization of high-level knowledge and feedback function is improved.(2) An edge detection method of calculating image measurement contour withPower-Law theory and least squares fitting method instead of traditional image edgeoperators are proposed. It use transform of measurement scale of rulers on the edge ofcomplex mechanical image contour to fit a measurement contour which is mostapproximate to the real contour, in order to eliminate the image features calculationerror caused by pits, burrs and other defects of complex image.(3) A region of interests (ROI) dimensions-reducing image segmentation model isestablished according to the image structural dimension description In this model,image are converted from high complexity feature space to the lower one so that signalprocessing methods can be used to process the transformed image to improve thecapacity and computing speed of image recognition system.(4) The image geometric features calculation method and image recognition modelare researched. Characteristics library and templates library for image recognition arebuilt and applied to machine products recognition systems in the industrial field.(5) The research results of this dissertation are applied in two practical imagerecognition systems. The detection efficiency and product quality of the productionprocess is dramatically improved.