Design And Implementation Of Factory Parts Sorting Robot System Based On ROS And YOLOv5

Intelligent manufacturing is the development direction of "Made in China 2025" and "Made in China 2035".Intelligent industrial production requires robots to replace manual work,reduce manual work,ensure repeated accuracy and improve product quality.Machine sorting system is widely used in the production process of assembly products.There are some problems in the existing machine sorting system,such as the control of teaching device,less movement in the assembly process and fixed program.In this paper,ROS,camera calibration,machine vision and other technologies are used to build a sorting system with visual tracking and control functions of the mechanical arm,so as to realize the intelligent and professional sorting process.The main work is as follows:(1)Based on the working environment and beat characteristics of the grasping object,the sorting system of the mechanical arm was designed,including six-axis robot,DSE12 S monocular camera,end-actuator and ROS platform;Through performance comparative analysis and experimental verification,equipment selection and ROS as the control system were determined,and the application software design of the system was completed.(2)In order to capture the object detection part accurately,the transformation matrix between the object detection part and the base of the manipulator was obtained by establishing camera imaging model,camera calibration,hand-eye calibration and other actions.Using the Moveit module in ROS,forward and inverse kinematics analysis was carried out to obtain the object detection position and pose of the manipulator system and the object detection Angle of the joint during motion.In ROS environment,the manipulator system model was built by compiling URDF model files and configuration;V-rep was used to obtain the weight and other parameters of the manipulator,and the system model of the manipulator was improved.Display the model using Rviz.(3)Select YOLOv5 object detection detection algorithm to identify parts,collect more than 7326 parts pictures,prepare part data set,train YOLOv5 model,and realize the retrieval of object detection parts;The BN layer of YOLOv5 was replaced by Mobilenetv3,the object detection model was improved and trained,after the improvement,the number of model parameters is reduced by 49.78%,the size of the model is reduced by 47.51%,the FPS speed is increased by 25%,and the calculation amount is reduced by 61.39%.and the lightweight of YOLOv5 model was realized.(4)Build the sorting system of mechanical arm,and use ROS special TF tools and Topic communication to complete the communication of object detection detection and grasping control,so as to realize parts sorting and system verification.The results show that the system has been equipped with sorting capability,realizing the functions of identifying the object detection parts,avoiding collisions,picking up the object detection parts and placing them in the specified position,with accurate grasping action and high precision.The improved YOLOv5 model takes up less space and achieves embedded system application.V-rep is used to obtain the weight information of the manipulator,which improves the accuracy of the manipulator control.