Research On Vision System Inspecting Instrument For Steel Ball Surface Defect

Steel ball is an important workpiece of bearing. Its surface defect has a great impact on the accuracy, revolving ability and service life of bearing. Currently, in bearing line, it is a common practice to detect defects of steel ball by means of human visual checking. So it is hard to guarantee the accuracy and stabilization of steel ball detection. This thesis utilized digital image technology to analyze and identify surface defect of steel ball. We developed a detection system for surface defect of steel ball, made some basic research for its real-time identification,and studied inspecting instrument for steel ball surface defect meeting item index.Due to the complexity of steel ball's mirror reflection, it is hard to get an ideal and clear image of steel ball defect. So we studied optical reflection features of steel ball, and then designed an illuminating system, combining with plenty of experiments. To get a full detection for any part of the steel ball surface we designed a developed mechanism for image detection of steel ball surface, and established an image acquisition platform applied to acquire images of steel ball surface.Before defect identification, it is necessary to process the images. This thesis adopts the algorithm based on HSI color space,it prevent missing a mass of information. This thesis optimize the sobel operators and then process the images with the new operators of edge extraction.and use the theory of invariant zernike moments in defect regional to segment the steel ball image,then we get the characteristic parameter of the steel ball image.We build BP neural network and deal with this characteristic parameter of the steel ball image by BP neural network.We use the processed result of the steel ball image to classify steel ball. Finally, this thesis developed detection software of a system for detection technology for surface defect of steel ball , taking LabVIEW as system platform,We get the test result after used the system to classify steel ball,and average percentage of accurancy could above 90%.