Research On Key Technologies Of Intelligent Material Sorting System Based On Industrial Robot Vision

With the gradual carrying out reforms of intelligent manufacturing,industrial robots are increasingly used in industrial manufacturing due to their versatility,programmability,and flexibility.At the same time,industrial robots combined with vision technology are more and more widely used in manufacturing,because of the vigorous development of artificial intelligence and image processing.Manufacturing enterprises in China generally still maintain the traditional production methods,and face problems such as low automation level and insufficient innovation ability.The most common job of industrial robots in modern automation is responsible for the operation of picking target objects in the production process,such as handling,mounting,sorting,etc.These behaviors can be unified into the operation of “recognition-pick-place”,Therefore,this project designs and develops a robot vision picking system with versatility,scalability and ease of use that can complete various “recognition-pick-place” related tasks in industrial production,users can use the system software to realize rapid production task switching by various functions such as process design and production plan management.First,through the analysis of the requirements of the “pick-and-place” tasks of the manufacturing enterprises,the overall framework of the system and the core use cases of the system is designed.The whole system software is divided into five modules: robot module,external equipment module,calibration module,intelligent algorithm module and production management module.The robot module and the external equipment module abstract the behavior mode of the automation equipment involved in the system,design the interfaces,and provide the ability of rapid equipment expansion.The calibration module provides various calibration algorithms required for the system,including camera calibration,hand-eye calibration and tool calibration.The intelligent algorithm module provides the system with the intelligent algorithms required to achieve visual grasping,including material recognition and intelligent viewpoint selection algorithms.The production management module is responsible for order-related operations in the system and provides data interaction interfaces.Then,the research on the material recognition algorithm is carried out,and the Linemod3 D algorithm is developed and improved.Using equal proportions of the spiral line to select viewpoints of render camera for creating the template,and the process of training and matching in the algorithm is introduced.A recognition experiment is carried out in a simulation data set.The recognition time is within 300 ms,the position error is within 4mm,the orientation error is within5°,and the success rate is over 90%,which proves that the algorithm has good performance.Next,aiming at the problem that the recognition accuracy decreases due to occlusion and other reasons when recognizing stacked materials,the robot active viewpoint selection is studied,and a viewpoint selection algorithm based on deep reinforcement learning is proposed.In order to tackle the problem of sparse rewards in the training process,a viewpoint experience enhancement algorithm is proposed.Then,simulation experiments and real-world robot experiments were carried out.The experiment results show that the viewpoint selection policy learned by the algorithm in this paper has good generalization ability of grasping targets,and the proposed method only needs to execute once to find the areas of high grasping possibility,and grasping success in cluttered scenes is increased by 22.8% against the single-view method,and mean picks per hour reached 294,which proves that the proposed viewpoint selection policy can effectively improve robot grasping performance.Finally,based on the software design and algorithm in this paper,the development of the system software is completed on the Windows platform,and the system is tested in two actual production scene: greeting card beads placing and paint bucket handling.Greeting card beads placing use the 2D eye-to-hand system,and the paint bucket handling experiment uses the 3D eyein-hand system.In the experiment of different application scenarios show that users only need to configure the system and complete the process design on the system software to enable the system to production,which proves that the system is able to meet actual industrial production requirements.