Research On Sprocket Burr Detection Method And Geometric Parameter Measurement Technology

In recent years,inspection technology based on machine vision has gradually matured.It has the advantages of non-destructive,high-precision,and high-efficiency,and has been widely used in geometric measurement and defect detection of parts.Due to the particularity of the location of the sprocket burr and the similarity with the surrounding environment,the traditional image processing method can not achieve good results in the burr detection,so this paper uses the deep learning method to detect the sprocket burr.Due to the relative lack of research directly on sprocket size detection,this paper uses machine vision-based gear detection methods for reference to achieve the measurement of the geometric parameters of the three-row sprocket.The main tasks of the research are:Complete the construction of the visual inspection platform,including the selection of camera and lens,the type of light source,and the lighting method;in order to enhance the quality of the image and increase the speed of subsequent image processing,preprocessing operations are performed on the collected images;in order to reduce lens distortion on the impact of detection accuracy,distortion correction is performed on the picture.Burr is one of the inevitable defects in sprocket machining,this paper improves the sprocket burr detection algorithm based on the YOLO v3 framework.By adjusting the number of residual network layers and optimizing the network structure,the performance of the improved network is further optimized and the detection efficiency is improved.Experimental results show that compared with the original YOLO v3 network,the improved network has a higher detection accuracy,reaching 96.41%,which is 0.73%higher than the original v3 network;the improved network size is reduced by 1/4,which is about 161M;The improved network detection speed is nearly 2 times faster than the original v3 network,reaching 0.42s/frame.Through repeated testing experiments,the improved network has a relatively stable performance in the detection of sprocket burrs.The geometric parameters of the sprocket are measured by machine vision.First,the number of teeth of the sprocket is obtained by the mask method;secondly,the diameter and center coordinates of the hub hole circle of the two end faces are obtained by the optimized Hough circle detection algorithm;then,the tooth profile is extracted by the improved Zernike moment sub-pixel edge detection algorithm The edge,specifically,first uses improved eight-neighbor edge tracking for coarse positioning,and then uses Zernike moments for fine positioning,and then calculates the distance from the tooth profile to the center of the hub hole circle,and then obtains the size of the addendum circle and the tooth root circle;,Fit the arc of the tooth groove of the sprocket by the least square method and get its center coordinates,and then calculate the pitch angle;finally,randomly collect 20 pictures,calculate the relevant parameters of the sprocket,and verify the reliability of the measurement system.The experimental results show that the measurement results meet the testing requirements of sprocket.