Research On Control Strategy Of Upper Limb Rehabilitation Robot Based On Muscle Strength Estimation

With the development of hemiplegia medicine and robot control,rehabilitation robots have made progress.Now,the upper limb rehabilitation robot mainly adopts passive training mode,while the active training mode is relatively few.The patient’s personal participation in the rehabilitation task has a great effect on the recovery of the limb function.However,most rehabilitation robot systems indirectly perceive the patient’s main force/moment through force/moment sensors,instead of directly starting from the patient’s muscle strength,and are affected by many unpredictable effects such as interference and system error.At the same time,the active resistance training mode can not be carried out according to the specific recovery of muscle strength of patients,which is divided into stages and intensity training.The understanding of surface electromyography signal(sEMG)is getting deeper and deeper,and the relationship between sEMG and human motor intention has been identified.Therefore,this paper used sEMG to quantitatively estimate the patient’s motivation,and studied the control strategy of the resistance training mode of rehabilitation robot based on muscle strength estimation.The research involved in this paper includes the following:Firstly,a six-degree-of-freedom upper limb rehabilitation robot model was built after analyzing hemiplegia medicine and anthropology.The kinematics and dynamics of rehabilitation robot were modeled by D-H method and Lagrange method respectively.MATLAB/Robotics Toolbox was used to simulate the rehabilitation robot model,and the correctness of the model was verified.Secondly,the collection system of surface myoelectromyography signals and referenced muscle force signals(r MS)was built by using HKJ-15 C type sEMG collection equipment and DS2-100 N type push and pull collection instrument.SEMG and r MS of biceps were designed and completed.After denoising the original sEMG,root mean square value,absolute mean value and variance were extracted.Thirdly,in order to solve the problem that BPNN is easy to fall into local optimal solution and poor prediction ability,an upper limb muscle strength estimation algorithm based on Adaboost improved BPNN was designed with BPNN as the weak learner and sEMG time-domain eigenvalues and reference muscle strength values as samples.The number of hidden layer nodes and learning rate of BPNN and the number of BP weak learners in Adaboost strong learner model are determined by trial and error method.Muscle strength estimation results of BP model and Adaboost BP model were quantitatively analyzed by relative root mean square error(RMS)and R square muscle strength evaluation models.Finally,the force-based impedance control was applied to the controller of the rehabilitation robot,and the control strategy of the resistance training mode of the upper limb rehabilitation robot was designed based on the muscle force estimation.The simulation of elbow flexion task was carried out on MATLAB/Simulink platform,and the feasibility of the control strategy was verified.