Research On Positioning Point Measurement Method Of Local Feature Processing For Large-scale Components

In order to meet the development needs of China ’s aerospace industry in recent years,new requirements are proposed for the efficient and high-precision machining of large-scale components,the digital measurement technology is the key to ensure the high-precision,highefficiency and high-robustness of the machining process.At present,the aerospace industry in domestic and abroad has conducted a lot of research on the measurement technology in industrial production.Among them,visual measurement,as a non-contact measurement method,because of the advantages of fast data acquisition,high precision and efficiency,easy to move,and capable of three-dimensional measurement,has been fully studied and valued by the industry.This paper mainly studies the high-precision measurement method of local positioning points in the processing of large aerospace components based on binocular vision.At the same time,it analyzes the difficulties brought by the complex light source interference in the processing site and the high reflection characteristics of the cabin metal surface,and provides a solution based on convolution neural network.In this paper,the measurement data splicing method based on public landmarks and robot control information is also studied.The main research contents are as follows:(1)The theory of binocular vision measurement is introduced and the characteristics of visual landmarks are analyzed,and the high-precision extraction,matching and 3D reconstruction methods of visual landmarks are studied.In view of the problem of false extraction and mismatch of visual landmarks caused by complex light source environment and local high reflection of cabin in visual measurement,a target detection method of visual landmarks based on convolutional neural network and a matching method of visual landmarks based on Siamese neural network are proposed.The method integrates deep learning method on the basis of traditional morphology and optical geometry.The field and laboratory experiments show that the method can effectively reduce the false extraction rate and false matching rate of visual landmarks,and improve the robustness of the whole visual measurement process.(2)For the positioning point measurement of large-scale components,it is necessary to splice the visual measurement data under multi-pose.In this paper,a vision measurement method combining binocular camera and industrial robot is adopted.The binocular vision is fixed at the end effector of the industrial robot.By controlling the motion of the robot,the binocular camera ’s measurement vision can sweep through the surface of the cabin by sliding window.Finally,the measurement results of visual landmarks under different postures of the robot are spliced into the same coordinate system based on the method of common landmarks,and the results are displayed in the form of three-dimensional point cloud.At the same time,in view of the problem that data splicing cannot be realized through public landmarks in crossregional measurement,a measurement data splicing method based on robot control information is adopted.By establishing the robot motion model and calibrating the parameters,and calculating the hand-eye relationship between the end effector of the robot and the camera,the conversion of visual measurement data to the robot base coordinate system under multi-pose can be realized.(3)The binocular vision measurement system is built,and the accuracy is verified in the laboratory and the processing field by experiments.The results show that the RMS of visual measurement accuracy can reach 0.120 mm after splicing ten postures based on common landmarks in the range of 1.5 m × 1 m.For multi-regional visual measurement,the RMS of visual measurement accuracy can reach 0.749 mm after splicing the measurement data of three regions based on robot control information.