Research On High Order Differential State Estimation And Control Method Of Flexible Joint Based On Recurrent Neural Network

Due to the poor interaction and safety of the traditional rigid joint manipulator,it can not be effectively applied to the service,special maintenance and other fields,flexible joint has become a research hotspot at this stage.Different from the mature control theory of rigid joint system,the control effect of flexible joint still has some limitations due to its dynamic complexity,underactuated,nonlinear coupling and other characteristics.The current research shows that the use of high-order information can improve the control performance of flexible joints.Therefore,this paper focuses on the estimation of high-order differential state variables and the control based on high-order estimation terms.Aiming at the complex dynamic modeling,nonlinear coupling and underactuated problems of flexible joint,a high-order differential state estimation and control strategy based on recurrent neural network is proposed.Through the deep mining ability of time series recurrent neural network for sample information,the forward dynamic equation of nonlinear complex system is learned and approximated,the learning results are recorded,the high-order information at the current time is predicted by mathematical method,and the predicted high-order information is fed back to the control system.In view of the mature application of the traditional PID control algorithm in the control field,this paper obtains the control rate with high-order information,so as to construct a simple structure flexible joint control algorithm which needs a end encoder of no high-performance,which is convenient for engineering application.The determination of neural network structure,the verification of generalization performance,the comparison of high-order information prediction methods and the verification of the overall control algorithm are all determined by the control system simulation experiment.In order to meet the needs of the research on the control algorithm of flexible joint based on recurrent neural network,this paper improves the hardware on the existing single degree of freedom flexible joint experimental platform in the laboratory.On this platform,experiments are carried out to track sinusoidal trajectory excitation signals with different amplitudes and frequencies under different terminal loads,which verify the effectiveness of the proposed high-order differential state estimation and control algorithm based on recurrent neural network;By comparing with the traditional PID algorithm,the effectiveness of the control algorithm is proved.Finally,compared with the traditional zero force drag experiment without high-order differential state information,the correctness of the predicted high-order information and the importance of high-order information for the flexible joint system are verified.Finally,through collision detection experiments,the potential of high-order differential state information prediction in the field of safety control of flexible joint manipulator system is verified.