Research On Defect Detectiing Technique Of Cylindrical Surface Appearance Of Precision Metal Machining Parts Based On Convolutional Neural Network

Precision metal processing parts have high manufacturing precision and are one of the components of modern industrial equipment manufacturing.At present,the quality inspection methods of precision metal processing parts are seriously lagging behind the industry development and market demand.This article is titled"Research on Defect Detecting Technique of Cylindrical Surface Appearance of Precision Metal Machining Parts Based on Convolutional Neural Network",systematically analyzing the image classification and detection mechanism of precision metal machining parts,focusing on the acquisition of workpiece images based on grating self-rotation lighting Technology,Res Net-based precision hardware machining parts cylindrical surface appearance defect detection technology,build hardware machining parts cylindrical surface appearance defect detection platform.The project has great academic value and practical significance to promote the development of intelligent inspection and instrumentation of manufacturing engineering.The research work was supported by the Guangzhou City Science and Technology Plan Project?No.201802030006?.The paper studies the detection system for the appearance defects of cylindrical surfaces of precision metal machining parts,and summarizes the research progress at home and abroad from three aspects:classic workpiece defect detection methods,workpiece defect detection methods based on deep learning,and visual lighting technology.The main point of this paper are as follows:?Analyze the general requirements of cylindrical surface defect detection of precision hardware processing parts,including high-quality image acquisition function,defect location function,defect type identification function,automatic detection function.Design the overall structure framework and workflow of defect detection system.Focuse on the analyzation of key technologies such as workpiece image acquisition technology based on grating self rotation lighting,the detection technology of cylindrical surface appearance defects of precision hardware parts based on convolution neural network.?Study the work piece image acquisition technology based on grating self-rotation lighting:First,analyze the shortcomings of the traditional flaw detection system's lighting method,mechanical structure,and imaging results.By integrating the grating structure and designing multi-industrial camera imaging methods,this article design the grating self-rotation lighting structure,effectively simplifing the lighting mechanical structure;Secondly,a grating-type self-rotating surface light source model and a grating-type self-rotating grating model are built to analyze the influence of the light intensity on the surface of the workpiece,the horizontal and vertical defect imaging conditions,which guided the design of the key parameters of the lighting model,including the height distance between the workpiece and the light source surface h_l-w,the grating structure parameters,the angular speed of the grating rotating motorw_motor,the load inertia J_motor,the load torque T_motor,etc.;Finally,compare the traditional visual inspection and machine vision algorithm test results of the light-emitting method and the grating-type self-rotating method.The results show that the grating-type self-rotating method can form an image that is convenient to highlight the defect and filter the background,and the defect imaging effect is better.?With comparison of mainstream target detection algorithms in recent years,propose the detection technology of cylindrical surface appearance defects of precision hardware parts based on convolution neural network.Use YOLO-v3 and Res Net to locate and extract defect target with high precision.Build data set under the framework of Vi Di tools of Cognex.Use data enhancement methods such as rotation,scale,flip,aspect ratio,clipping,brightness,contrast to expand the data set.Carry out off-line training and analyze the key links of software and hardware of on-line detection.?Build metal processing parts defect detection system,mainly composed of hardware selection,software integration,PLC communication,then verify the detection system from three aspects,including metal processing parts cylindrical surface positioning detection,metal processing parts cylindrical surface defect detection,metal processing parts actual production inspection.The application results show that,compared with the manual detection results,the device detection results have fewer false and missed detections,and the detection accuracy is higher.The developed metal processing parts cylindrical surface detection system has good practical effect,high detection efficiency,and has practical engineering value.