Research On Robotic Flexible Assembly Method Based On Contact State Perception

With the continuous development of the manufacturing industry,automated operations in the assembly industry has become an inevitable trend for the manufacturing industry.Improving assembly quality and efficiency and reducing product costs have become major challenges of manufacturing companies.In recent years,industrial robots have been gradually applied to assembly operations,providing the possibility of replacing manual labor and realizing intelligent assembly.At present,most robots are used in structured environments.Manual teaching and offline programming are adopted to complete the pre-planned process,which makes the perception of assembly objects and environment very weak and lacks flexibility and adaptability.Therefore,howto extract features from sensor data and perceptual learning autonomously online to make robots adapt to complex and variable assembly environments is the main research in this paper.Low-voltage electrical components have the characteristics of small size,complex appearance,different models,complex coordination,etc.and the robot's assembly operation space is relatively insufficient,which put higher requirements on the flexible assembly operations.This paper takes the intelligent assembly process of circuit breakers as the research background,simulates the cognitive thoughts of human experts,and conducts in-depth research on the contact state perception,assembly state recognition,and assembly states evaluation in the assembly process of complex structural workpieces.Algorithm verification experiment was carried out in the small circuit breaker flexible assembly platform,and the experimental result show the achieved performance.The main work of this paper is listed as following.(1)An offline recognition method for contact state of component assembly based on extreme learning machine is proposed.Aiming at the problem that it is difficult to describe the contact state between the robot operating end effector and the operating object under visual occlusion,the ELM-kernel-based network model is established by using force sensor information(force and torque)and the pose of the end effector to describe the assembly contact state.The comparison results show that the ELM-kernel-based method can effectively identify the offline assembly contact state.(2)An online sensing method for contact states of component assembly based on SVDD-ELM kernel is proposed.Aiming at the problems of poor flexibility,insufficient robustness and lacking of self-adaptability in offline learning during assembly,an experience knowledge base of "contact state-manipulator movement" mapping is established,using the SVDD-ELM kernel algorithm for online perception to continuously accumulate the experience knowledge and to update the base.(3)An assembly states evaluation method based on Faster R-CNN is proposed.Aiming at the problem of misjudgment of assembly success solely based on assembly depth,combined with the visual system to collect the final assembly state image,the assembly quality was evaluated by deep learning method to judge whether the assembly was successful.(4)A flexible assembly platform for circuit breakers was built.Based on the contact state perception module and assembly states evaluation module,a flexible assembly method was proposed and verified on a real assembly system.By analyzing the success rate of assembly and the change of assembly process state,the results show that the assembly method proposed in this paper can effectively perceive the assembly state,so that the robot can complete the assembly task more accurately.Finally,we summarize the work of this paper and get results and experience and the direction of further research is analyzed.The flexible assembly method based on contact state perception proposed in this paper has important research significance for the flexibility and real-time performance of the robot's execution tasks.