Estimation Of Joint Moment For Elbow And Wrist Based On Muscle Synergy And Neuromusculoskeletal Model

Due to the increasing demand for intelligent rehabilitation aids and the continuous expansion of functions,human-machine interaction(HMI)has become a research hotspot in the field of rehabilitation in recent years.In the process of rehabilitation,rehabilitation aids such as intelligent prosthetic limbs will interact directly with the patient’s body.Surface electromyography(sEMG)is widely used in the design of HMI system because of its ability to reflect human motion intention.Joint moment can reflect the specific information of human movement from the physiological point of view.Estimation of joint moment has become one of the key research fields of HMI technology based on sEMG.In the process of rehabilitation training,using musculoskeletal model to obtain the joint moment/muscle force of patients can provide physiological control variable for intelligent rehabilitation equipment,thereby helping the HMI interface to quantitatively unscramble the motion intention and the physical activity state of human body.In this paper,elbow joint and wrist joint of human upper limb are studied.The joint moments in the synchronous motion of elbow and wrist joints are estimated.An on-line estimation system of elbow and wrist joint moment is designed,and the control strategy based on joint moment is studied to realize multi-joint continuous motion control of intelligent robot arm.The main research contents of this paper include:(1)sEMG signal and joint angle were collected during the flexion and extension of elbow and wrist.In order to obtain the activation degree of muscles for different joint movements during synchronous motion of two joints,an improved non-negative matrix decomposition(NMF)method was proposed based on muscle synergy theory.Aiming at the problem of poor stability caused by unsupervised feature of traditional NMF methods,semi-supervised analysis method was adopted.Firstly,the contribution matrix of each muscle was obtained when the single joint moved independently.Then,the activation coefficient sequence of synchronous motion of two joints was calculated based on Moore-Penrose generalized inverse.Aiming at the problem of low reconstruction accuracy due to the mandatory non-negativity of traditional NMF algorithm,a new decomposition method was proposed.The process of non-negative matrix decomposition was divided into two steps: matrix decomposition in real number range and converting to nonnegative matrix.It can improve the accuracy of matrix reconstruction while still guaranteeing the non-negativeness of the matrix.After the decomposition results are obtained by the improved algorithm,the activation degree of each muscle to different joint movements can be obtained by combining the muscle coordination model.(2)Based on the obtained muscle activation,a musculoskeletal model of elbow and wrist joint was built to estimate joint moment from the perspective of forward dynamics,and the reference joint moment was obtained through the inverse dynamic model of the two joints.Aiming at the problem of individual differences,the parameters of the model were identified by combining forward and reverse dynamics analysis to optimize the estimation results of joint moment.In order to solve the time-consuming problem of common bio-evolutionary algorithms when there are many parameters to be identified,the unscented Kalman filtering algorithm(UKF)optimized by differential evolution(DE)algorithm is proposed to obtain the optimal values of parameters,which can effectively improve the efficiency of parameter identification while optimizing the estimation results.(3)An on-line joint moment estimation system was designed when the elbow and wrist joints move synchronously.Combined with the proportional control strategy based on sEMG signal,the estimated joint moment was converted into the motion control command for robot arm.Then,the control command was applied to the motion control of the robot arm to realize the multi-joint motion control.The active motion intention of the human body can be reflected through the trajectory of the robot arm.