Research On Grasping And Surface Defect Detection Technology Of Workpiece Based On Machine Vision

It is a practical and feasible way to promote the transformation and upgrading of the manufacturing industry through the artificial intelligence + robot + machine vision model to remove labor,improve automation,and reduce the problems of production and quality instability caused by labor.Basic technologies such as robotics and machine vision are involved in this mode.In this paper,the robot grasping and surface defect detection links in simulated production scenarios are researched,and experiments are designed.Small metal workpieces are used as experimental objects to realize the grasping action based on machine vision and the detection and classification of custom defects.It is useful for practical industrial applications.The main work is as follows:In the part of grabbing workpiece,two grabbing methods based on point cloud data and2 D plane images are designed according to the given workpiece.First,model the selected ABB IRB120 robot,and calibrate the internal and external parameters of the selected ZED 2 camera;Then perform hand-eye calibration with the eye out of the hand and the camera in the lateral position,and perform a nine-point calibration with the eye out of the hand and the camera in the upper position to determine the conversion matrix from camera coordinates to robot coordinates;Finally,using the ZED 2 camera SDK interface,the upper computer algorithm to obtain the grab point coordinates and the communication program between the upper computer and the robot control cabinet are written in python language,and the workpiece grasping experiment is carried out on the physical platform,which verifies the visual grasping of this article.In the defect detection part,two defect detection classification schemes are designed for the 3 types of defects of the experimental workpiece.In the traditional defect detection and classification scheme based on image processing,a suitable image processing algorithm is designed according to the acquired surface image of the workpiece,and the SVM classifier is used to perform defect detection and classification on the defect image;In the scheme based on deep learning,the VGG16 model is used for migration learning,which solves the problem that the use of convolutional neural networks cannot achieve high detection accuracy when the amount of defect detection sample data is small.The experimental results show that the traditional method based on image processing has a comprehensive detection accuracy rate of80.5%,the detection method based on deep learning has a comprehensive accuracy rate of92.8%,and the defect detection classification accuracy rate is higher.