Research On Motion Analysis And Interactive Control Method Of Upper Limb Rehabilitation Robot

Early physical rehabilitation training has a positive effect on the treatment of patients with upper-limb functional injury.Active training which reflects the intention of patients can more effectively stimulate the motor learning ability of the nervous system.The rapid development of rehabilitation medical field and robot control field has brought new possibilities for the treatment of patients.The reasonable use of rehabilitation robot for limb adjuvant treatment can relieve the excessive workload of the therapist and ensure the safe and effective rehabilitation training.The existing upper limb rehabilitation robots mostly adopts repetitive passive training mode in use,and there is less research on active training with stronger interaction with patients.To solve the above problems,this paper studies the interactive control method of upper limb rehabilitation robot based on surface electromyography(sEMG).The research content of this paper comes from the National Natural Science Foundation of Beijing"Transmodal Coupling Mechanism and Bio-Driving Strategy of Upper Extremity Exoskeleton Robot" under grant 3202021.The main research work is as follows.Firstly,considering the control requirements,four degrees of freedom which conclude shoulder joint and elbow joint are selected to establish the upper limb rehabilitation robot model.The joint coordinate system is defined to obtain the D-H parameters,and then the homogeneous transformation matrix is solved to derive the velocity Jacobian.The mathematical model of the upper limb rehabilitation robot is established by MATLAB robotics Toolbox,and the simulation results are compared with the theoretical results to prove the reliability of the kinematic model.Based on the principle of energy,the Lagrange dynamic model of the upper limb rehabilitation robot is established.And the dynamic simulation analysis is carried out by using the virtual prototype model in the ADAMS/view environment,which provides the mechanical basis for the research of the interactive control method of the rehabilitation robot.Secondly,in order to accurately identify the human movement intention,the formation mechanism and characteristics of sEMG are explored,and the feasibility of sEMG as the recognition signal of movement intention is analyzed.On this basis,the motion mode of upper limb and the position of EMG signal acquisition with high correlation are selected.In order to reduce the noise caused by sensors and circuits,a suitable sEMG signal acquisition platform is built.In order to reduce the influence of external conditions on sEMG acquisition process,a reasonable acquisition scheme is designed to obtain extensive and reliable EMG sample data.According to different noise sources,notch filter and wavelet threshold de-noising method are used to remove the aliasing noise in sEMG.The time-frequency features of sEMG are extracted and the maximum and average absolute value feature matrices of wavelet coefficients are obtained.Then,to solve the problem that traditional action classifiers do not perform well in classifying large sample datasets,a gated convolution neural network(G-CNN)classifier is constructed.The gate control convolution layer is used instead of the ordinary convolution layer.It can control the flow of hidden features in the gate control convolution neural network,effectively extract useful information and block noise.The appropriate network structure and learning algorithm are designed to improve the accuracy of action recognition.Regularization method is proposed to optimize the network to avoid the over fitting problem caused by excessive learning of training data.Adam is used to update the weights and biases in G-CNN,which can avoid the problems of gradient dispersion and local optimization,and accelerate the convergence speed of the model.The indexes to judge the performance of classifier are accuracy and loss value.Experimental results show that compared with other classifiers,the proposed G-CNN model has obvious advantages in action recognition performance.Finally,according to the identified upper limb movements,the desired trajectory representing the human motion intention is obtained,and the upper limb rehabilitation robot system is controlled by the combination of biological signal and motion signal.Due to the strong robustness of nonsingular terminal sliding mode controller(NTSM),it can solve the influence of system parameter uncertainty and external disturbance.Then,the control law is derived to realize the accurate tracking of the trajectory.And the stability of the control system is proved by using the Laypunov function.In order to ensure the flexibility of upper limb active rehabilitation training,a nonsingular terminal sliding mode impedance control method is proposed.According to the human-computer interaction force,the motion trajectory is dynamically adjusted to ensure that the interaction force between human and computer is kept in a suitable range.The reliability and effectiveness of the control method are verified by MATLAB simulation platform.