| As one of the core components of power equipment,piston assembly has an important impact on the performance and service life of equipment.At present,the automatic production process has been established for the manufacturing of piston components,but the assembly work is still mainly manual assembly method,workers work intensity is high and assembly efficiency is low.However,robots using off-line programming method or direct teaching method can only grasp and assemble parts with fixed position and attitude.Once the position or attitude of the parts changes when they arrive at the assembly station,it is very likely that the robot fails to grasp and cannot complete the original assembly task,which seriously affects the production efficiency.In addition,multiple conveyors and robots are required to work at the same time,which is a complicated and costly process.In order to improve production efficiency,reduce production cost and accurately grasp parts with different poses at the same time,aiming at grasping and sorting tasks of piston components,this thesis proposed an intelligent sorting system assisted by machine vision with xArm6 robot,depth camera and end-actuator components as hardware platform and ROS system as software platform.The main research contents of this thesis are as follows:(1)A robot intelligent sorting system is built with xArm6 robot as the main body,Intel Realsense D435i binocular depth camera as the image acquisition device,and flexible hand claw as the end actuator.At the same time,the control function pack of camera and robot is constructed based on ROS system to realize the real-time communication between the hardware device and the upper computer,and the upper computer is used to control the positive and negative rotation of the motor of the flexible hand claw,thus realizing the automatic control of the working state of the end-effector.(2)The calculation method of conversion matrix between each coordinate system is studied,and the conversion matrix between image pixel coordinate system and camera coordinate system is obtained by using Zhang Zhengyou calibration method and checkerboard camera internal parameter calibration method.Based on ArUco calibration plate,an eye-in-hand robot hand-eye calibration system was established.Tsai-Lenz method was used to solve the calibration plate and robot pose data obtained in the experiment,and the transformation matrix between the camera coordinate system and the robot coordinate system was obtained.Finally,the accuracy of the obtained coordinate transformation matrix is verified by experiments.(3)In order to complete the identification and positioning of target objects,this thesis adopts the method of combining YOLOv3 deep learning target detection algorithm and OpenCV image processing library to build the target detection module.By collecting images of piston components to make data sets and combining ROS to complete the training of target data sets,weight files are obtained.The detection of different piston components is realized.At the same time,compared with the detection results of Faster R-CNN algorithm as the target detection method,it is found that the YOLOv3 method used in this thesis is superior in detection accuracy and detection speed of piston components.By using the conversion matrix obtained by calibration,the positioning point of the object in the pixel coordinate system was converted to the world coordinate system,and the positioning accuracy was verified by designing experiments.The positioning error on the X and Y axes was about 1mm,and the positioning error on the Z axis was about 0.5mm,which could fully meet the system requirements.(4)The improved D-H method is adopted to establish the model of xArm6 robot,and the forward and inverse kinematics analysis is carried out,and the kinematics analysis results are simulated by MATLAB.The 3D model of xArm6 robot is used to establish the URDF motion simulation model and the corresponding motion control function package,and the simulation environment of robot intelligent sorting system is built in the upper computer based on Rviz visualization plug-in.Based on MoveIt motion planning plug-in,RT-Connect algorithm and coordinate information of target points,the robot trajectory planning was realized.Several sorting experiments of piston components were carried out in simulation environment and actual environment,and the experimental results were analyzed.At the same time,the Faster R-CNN method was used to carry out the comparative experiment of grasping piston components,and it was finally found that the ROS combined with YOLOv3 system proposed in this thesis had a great improvement in the grasping and sorting success rate of piston components.Finally,the system can identify and sort piston components in complex environments.Experiments show that the detection accuracy rate of target parts reaches 97.67%,the detection speed reaches 0.048 seconds,and the success rate of sorting reaches 95.33%,which is two times Faster than the common deep learning method faster R-CNN.The target detection accuracy rate is increased by 1.34%,and the sorting success rate is increased by 4.66%,and it can fully realize the sorting and grasping task for different piston groups of parts in the field environment. |
PDF Full Text Request