Design Of Dimensional Measurement And Defect Detection System For Automotive Lamp Parts Based On Machine Vison

Automotive lamp parts are an important part of automotive lamps such as headlights,interior lights,tail lamps and connecting harnesses.However,the size of automotive lamp parts affects whether they can be assembled and turned on in the future.Burrs,cracks and The problem of lack of glue in injection molding affects the appearance and use of products.Nowadays,most enterprises conduct quality inspection through manual visual inspection,which is a low-precision and lowefficiency method,which cannot meet the needs of high-efficiency and high-quality mass production.This paper takes auto lamp parts as the research object,and designs a machine vision-based dimension measurement and defect detection system for auto lamp parts,aiming to meet industrial needs and bring great value to enterprises.By analyzing the inspection requirements of the car lamp parts to be inspected,a machine vision inspection system was developed,and the camera and lens of the image acquisition system in the system hardware were strictly selected.The optical method is analyzed and compared,and the corresponding visual inspection system is built.At the same time,the principles of several camera calibration methods are analyzed,and matlab is used to conduct calibration experiments,and the internal and external parameters of the camera are obtained through analysis.Based on Open CV library and Py Qt5 framework,the system software is developed using Python language.By analyzing the image preprocessing algorithm,the appropriate image filter is selected for edge detection,and the appropriate edge detection operator is selected for the line feature detection technology.By introducing the common line detection algorithm,an improved LSD line detection algorithm is proposed.It solves the problem of inaccurate edge extraction or ignoring small edge information,and more effectively detects the contours of car lamp parts,so as to use the extracted contour to detect the straight line to be measured,and complete the calculation of the size of the car lamp parts Measurement,the single detection error is within 0.066 mm,and the uncertainty of product measurement is analyzed and calculated to be 0.015 mm.The principle and network structure of convolutional neural network are studied to detect defects in automotive lamp parts.By establishing a standard data set of defect samples,and applying Faster-RCNN,YOLOv3 and YOLOv4 algorithms to the defect data set,the experimental data shows that,The accuracy rate of Faster-RCNN and YOLOv3 algorithms can reach about 70%,while YOLOv4 can reach about 75%,and an improved YOLOv3 model is proposed to improve its accuracy rate to 86.2%,but due to the insufficient number of samples,it cannot meet the needs of industrial The accuracy rate is more than 95%,but to a certain extent,it verifies that the convolutional neural network can effectively and accurately realize the defect detection of automotive lamp parts.Using Python language and QTDesigner to design a set of human-computer interaction interface suitable for size measurement and defect detection,real-time display of detection results and statistical production data,etc.,which is convenient for employees to operate,and use the detection system designed in this article to test three products.The detection speed is 3-4 times higher than that of manual visual inspection,the accuracy of size measurement and detection is increased to 99%,and the defect recognition rate is up to 86.2%,which greatly meets the high precision and high automation requirements of enterprises and brings huge benefits to enterprises.