Research On Identification And Compensation Of Dynamic Parameters Of Multi-DoF Series Manipulator Based On Deep Learning

The multi-degree-of-freedom series manipulator is widely used in various industrial production processes,which greatly improves work efficiency and promotes economic benefits.Through technological progress and industrial development requirements,intelligence,integration,and efficient movement put forward new requirements for the development of industrial robots.An industrial robot is a highly complex nonlinear system.Factors such as joint crossing,friction characteristics,and load changes can all affect the precise control of the robot.Advanced motion control of models has become a hot and difficult direction in technical research.Manipulators have complex coupling relationships and serious nonlinearities,such as nonlinear friction,joint flexibility,and motor torque fluctuations.The inverse dynamic takes the coupling and nonlinear relationship into consideration,and solves the force or torque of each joint through the joint displacement,velocity and acceleration of the known trajectory.However,the robot dynamics system has structural and non-structural uncertainties.Structural uncertainty refers to model errors caused by parameter identification,and unstructured uncertainty refers to dynamic characteristics without model description.Because the mathematical model cannot express all the influencing factors in the robot motion control,no matter how complex the modeling method is used,there is always uncertainty in the robot dynamics,resulting in torque prediction errors.In recent years,artificial intelligence learning methods are in the ascendant.Deep learning has strong nonlinear fitting capabilities and good fault tolerance,which provides an opportunity to solve the problem of robot parameter identification accuracy.In this regard,this paper conducts research on the robot's high-precision joint torque prediction technology based on dynamics and deep learning.The research content of this article is as follows:1.Research on the identification of dynamic parameters of industrial robots with six Do F.The experimental equipment is the Universal Robot 5(UR5)six-axis rotating manipulator,the dynamic model is established by Newton-Euler method,the minimum parameter set of identifiable parameters is derived,and the zero-phase digital filter and Butterworth filter are used.The processing method is to process the experimental data,and then to identify the dynamic parameters.By verifying the trajectory to check the accuracy of the identification parameters,the theoretical dynamics model can obtain relatively accurate torque prediction.The prediction torque accuracy satisfies the torque prediction error of each joint less than 10% of the maximum torque,indicating good identification accuracy.At the same time,Simulink is used to establish the whole process of dynamic parameter identification,and the UR5 robot arm parameter identification panel is developed using GUI,which makes the parameter identification operation process easier and the identification result display more intuitive.2.Research on motion trajectory planning method based on robot operating systemThe robot developed by the Robot Operation System(ROS)has good versatility.It uses the open source code model to encapsulate a large number of abstract hardware packages and functions commonly used in robot control,which improves code reusability and development efficiency.The use of ROS for trajectory planning and motion control has important theoretical and practical significance.In this paper,The trajectory and control of the UR5 manipulator are based on Move It!.At the same time,Gazebo and RVIZ are used for joint simulation,and RVIZ and real manipulator are used for experimental verification.The experimental results show that the ROS-based motion planning and control method has good results.3.Propose a prediction torque error compensation model based on deep learningIn this paper,deep learning is applied for the first time to the compensation of uncertain factors to assist the identification of dynamic parameters of a 6-DOF manipulator.An error compensation model based on deep neural network—Uncertainty Compensation Model(UCM)is proposed to compensate the torque error caused by dynamic uncertainty.UCM is mainly composed of the proposed Input Control Module(ICM)and the proposed Error Learning Model(ELM)based on long and short-term memory units and attention mechanism.Taking into account the coupling characteristics of robot dynamics parameters,ICM is proposed to control the effective input for ELM,which can effectively avoid unnecessary interference of input data on model learning.The ELM unit is an error learning model.Its input is the data processed by ICM,and its output is an estimate of the torque error for a specific joint.The ELM composed of ELM units is used to extract significant data features from sequence data to predict the torque errors of the six joints of the robot.This article summarizes the impact of effective input,time step and attention mechanism on UCM performance.The torque compensation method proposed in this paper is verified by the UR5 manipulator.Experimental results show that,compared with the predicted torque of the theoretical dynamics model,the proposed UCM has a good ability to capture friction characteristics and compensate the local maximum torque error.The accuracy of the torque prediction error of each joint is less than 6% of the maximum torque,which effectively solves the shortcomings of the theoretical dynamics model and improves the prediction accuracy.