Robot Vision And Force Control Technology For Peg-in-hole Assembly

Assembly is a common process in industrial production.Production efficiency and product quality is affect by assembly accuracy and time cost.With the development of control hardware and software in recent years,especially the development of automation control technology and vision sensor manufacturing,robots have acquired the ability to complete complex assembly tasks.The iron and steel industry is a pillar industry of the national economy,and its gross domestic product accounts for about 10% of China's GDP.In the steelmaking area with severe working conditions,there are a large number of molten iron temperature measurement and sampling operations.The metallurgical probes need to attech to the contact rod before immersing into the molten iron and steel for sampling.Due to the insufficient level of automation,the assembling and sampling is still operated by workers.Workers need to face splash,dust,molten iron and steelof of more than 1000 degrees frequently and for a long time.The job safety needs to be solved urgently.The current status of the steelmaking area restricts the level of intelligent manufacturing in the iron and steel industry,and robotic system for peg-in-hole assembly is urgently needed.This Paper focuses on the assembly of contact rod and metallurgical probes on the steelmaking site,a peg-in-hole assembly strategies is designed to solve the difficulties of multiple assembly working positions and insufficient camera field of view.To solve the problems of low repeated positioning accuracy of the rotary additional axis and the deformation of the contact rod,an alignment method based on adaptive visual servoing was proposed.Aiming at the problems of low precision machining of the inner wall of the metallurgical probe and long-distance assembly,a method based on reinforcement was designed.The contributions of this papper are as follows:1.During peg-in-hole alignment process,the high temperature of the steelmaking site will accelerate the drift speed of the camera's internal parameters,making it difficult to achieve precise positioning control of the tooltip of the contact rod.The contact rod deforms over time,and an accurate tool model cannot be obtained.This paper proposes an uncalibrated method for peg-in-hole alignment based on adaptive visual servoing.The internal and external parameters of the camera and the tool geometry are treated as unknown parameters,which are estimated online by minimizing the error function.This method use two point features and a vanishing point features on the contact rod.Corresponding depthindependent interaction matrix are derived and a control algorithm that does not require feature depth was designed.To solve the problem of coupled position and attitude control during the visual servoing process and the slow convergence speed of image errors,a multiple phases alignment strategy was designed to decouple position control from attitude control to achieve fast positioning control of the tip of contact rod.2.During insertion process,in order to solve the problems of low precision machining of the inner wall of the metallurgical probe and the complex contact state of the long-distance assembly,an insertion algorithm based on reinforcement learning was proposed.The proposed algorithm directly outputs the movement instructions of the robot end-effector based on the contact force obtained by the force/torque sensor on the robot end effector.By using a PD controller to guide the deep strategy gradient algorithm,the training time of the algorithm is effectively reduced.In this paper,simulation on Matlab and experiments are conducted on the proposed alignment method based on adaptive visual servoing,and dynamic simulation on Gazebo is conducted on the reinforcement learning based insertion algorithm.The results of the simulation and experiment verify the effectiveness of the proposed algorithms.