Intention Recognition And Interaction Control For Lower Limb Rehabilitation Robot

In recent years,with China moving into an aging society,and the number of people with physically disabled suffered by stroke has increased.A heavy burden to the family and society has been brought by stroke,which seriously endangered the physical and mental health of patients.Therefore,society should pay attention to a major problem that how to provide an effective assistance for stroke patients.This paper focuses on the study of the lower limb rehabilitation robot,and is constructed by three aspects as follows: the lower limb movement intention recognition,the construction and stability of the noise-tolerant zeroing neural network,and the model predictive control for passive and active rehabilitation.The research purpose is dedicated to solve issues include that movement intention recognition and human-robot interactive control.Based on these problems,this paper conducts study more in depth for lower limb rehabilitation robots,and the major research contents are shown follows:(1)In order to achieve intention recognition,the first step that the input signals of intention recognition model are set as surface electromyography(sEMG)signals,and the information and characteristics of sEMG signals are analyzed.Then the Hill-based muscle model is built,which include the active dynamics,the muscle contraction dynamics and the musculoskeletal geometry,to calculate the muscle-tendon force and the torque of joint.Finally,according to the skeleton structure of the lower limb,the rigid skeleton model of the lower extremity is established,and the lower limb muscleskeleton model is established by Hill-based muscle model and to complete the intention recognition modeling.(2)For the time-varying nonlinear optimization problem,a novel noise-tolerant zeroing dynamic system is proposed.Then,a continuous zeroing neural network(ZNN)model and a noise-tolerant zeroing neural network(NTZNN)model are presented,analyzed and investigated for online solving time-varying nonlinear optimization problem.Combined with monotone increasing odd activation function,a kind of generalized noise-tolerant type zeroing neural network(GNTZNN)model is proposed for online solving time-varying nonlinear optimization problem.In addition,from the perspective of control,the problem of NTZNN is transformed into the problem of the nonlinear control system,and the time-varying nonlinear optimization problem is quickly solved through the design of the generalized PID controller.According to the Lyapunov stability theorem,the asymptotic stability and exponential convergence are studied for time-varying nonlinear optimization problem with different measurement noises.Moreover,combined the theory of ZNN with the lower limb muscle-skeleton model,a closed-loop intention recognition model is presented,and ensures the accuracy of closed-loop motion estimation of human lower limbs.(3)For the human-robot interaction control,the model predictive control(MPC)technology is utilized.A new projective active set conjugate gradient algorithm is primarily proposed to online solving the optimal controller by using the linearization/discretization first-order derivative of constrained conditions lower limb rehabilitation robot through a third-order Taylor type difference formula based on the MPC technology.The feasibility and global convergence of the algorithm are proved by strict theory.The simulation results show that the human-machine interaction optimal controller can realize the active/passive rehabilitation training activities of the affected limb.