Surface Defect Detection Of Valve Plate Without Template Matching

Valve plate is one of the important components of automobile air conditioning compressor.Its surface defects in the production process will seriously affect the normal operation of the compressor.At present,due to the low inspection speed of valve parts and defects of valve plate,it is easy to misjudge the output of valve plate in China,which will not only affect the manual inspection speed,but also lead to the misjudgment of valve plate defects.With the development of artificial intelligence,intelligent manufacturing and other fields,machine vision detection technology has been widely used in industrial production.Using machine vision instead of human eyes can effectively improve the efficiency and accuracy of defect detection.Therefore,based on machine vision technology,this paper studies and designs a set of automatic vision detection system for valve plate defects.The main research work is as follows:1.The machining process and the defects of valve plate are analyzed.According to the characteristics and distribution of valve plate surface defects,the defects are classified into indentation,misrun,edging defect and so on.The detection algorithms are designed respectively by analysing the defects.2.In order to more clearly display the defect characteristics of the valve plate image when collecting the image,two kinds of lighting methods are designed.One is to use the plane light,which is mainly used to detect the defects with obvious defect characteristics such as indentation and misrun under the plane light;The other is to use the directional light.Because the edging characteristics of the valve plate are not obvious under the plane light,the use of directional light can highlight the characteristics of edging,which is easy for subsequent detection.3.The detection method of indentation of valve plate is studied.The image is collected under the plane light,and the screening conditions of the indentation characteristics are obtained through the defect analysis.The defect characteristics of the indentation can be extracted by using the traditional image processing method.Because the area characteristics of the indentation are different from those of the normal edge of the valve plate,the indentation can be screened by using the area characteristics of the indentation as the screening conditions.4.The detection method of misrun of valve plate is studied.The clear image of the edge of the part can be obtained by collecting under the plane light.The edge contour of the valve plate part is divided into three categories,namely,the outermost edge,the innermost edge and the edge of the circular through hole in the middle layer.The convex hull and convex defects of various edge contours are calculated by using the traditional image processing method.The misrun of the valve plate part can be screened according to the convex defect characteristics of the contour.5.The edging defect detection method of valve plate parts is studied.Image acquisition under directional light can highlight the edge edging defect area.Because there are non defective machining knife marks on the surface of the valve plate,and the knife marks will reflect light under some directional light,the normal image of the valve plate parts will also have abnormal areas similar to the edging defects.It is difficult to distinguish the edging defects from the normal reflection areas by using the traditional image processing methods.In this paper,a target detection method based on YOLOv5 s neural network is used to detect edging defects.The neural network model is trained under the self-built data set.The model can quickly and accurately identify edging defects.6.Valve plate defect detection experimental system is designed.The system is divided into two parts: software and hardware.The hardware includes computer,single chip microcomputer,industrial camera,conveyor belt,black box,light source and so on.The software is divided into upper computer software and lower computer program.The upper computer software is written in C++ language,VS2017 is used as the development environment,MFC framework is used for man-machine interface design,Open CV is used as the basic algorithm library of image processing,and runs on Windows computer;The lower computer program is written in C++ language.Arduino IDE is used as the development environment and runs on the Arduino UNO development board.The experimental results show that the system can accurately and quickly detect the main defects such as indentation,misrun,edging defect,and meet the detection requirements of the factory.The detection algorithm used in the system has high practical value...