Intelligent Grasping Algorithm Of Multi-Category Object And Its Digital Twin Application

The demand for product diversification and customization in the industrial manufacturing field requires production enterprises to adjust production lines in a timely manner and carry out flexible production.However,the current post-adjustment method is to stop the operation of the production line first,and then adjust it according to the characteristics of the product,which has a longer adjustment time and reduces the efficiency of the production line task switching.At the same time,as an indispensable link connecting the upstream and downstream of the production line,the robot usually adopts the teaching method to grasp the objects,which is difficult to solve the practical problems of different sizes,different shapes and random angles of multi-category objects.In view of the above problems,this paper focuses on the robot to carry out multi-category object grasping algorithm research and robot digital twin modeling work,and improve the grasping accuracy and task switching efficiency of the robot without stopping production.The specific research content is as follows:(1)Starting from the robot grasping point representation method,we give full play to the feature point mining and tracking capabilities of the deep learning method,improve the Generative Grasping Convolutional Neural Network(GGCNN),and propose a multi-category object grasping detection algorithm based on the new convolution(Grasp-Conv Ne Xt).Grasp-Conv Ne Xt can perform real-time inference(～20ms),and the model generalization performance is strong.The test results on Jacquard,a publicly available multi-category object grabbing detection dataset,show that: Grasp-Conv Ne Xt improves grasp accuracy by 13.63%compared to GGCNN.In summary,Grass-Conv Ne Xt has strong multicategory object grasping ability and high handling efficiency.(2)In the application of digital twins,this paper proposes to incorporate the knowledge data model into the existing five-dimensional digital twin model,forming a six-in-one digital twin model.Apply the model to robot intelligent grasping work,the robot twin is first established based on the Gazebo digital twin modeling technology,and then the proposed Grasp-Conv Ne Xt algorithm(knowledge data model)is used in the robot twin to simulate the grasping process of the real robot;and finally the digital twin application verification is carried out on the actual production line.The results of digital twin application show that the six-in-one digital twin modeling method improves the start-up rate of robot collaborative production by 16%,and shortens the production deployment time by 1/4.