Research On 2-DOF Simultaneous Myoelectric Control Based On Sparse Constrained Nonnegative Matrix Factorization

With the increase of the active degree of freedom(DOF)of dexterous prosthesis,how to realize the simultaneous,continuous and proportional control is the primary issue for realizing the rehabilitation function of prosthesis and improving the acceptance among users,and it's also a central issue in the field of robotics and biomedical engineering.Supported by the project,the decoding of multiple DOFs for the hand/wrist movements is studied in this paper.An unsupervised myoelectric control strategy is the research subject which has good decoupling effect and strong robustness.Offline validation and online evaluation is completed in this paper.Firstly,the estimation method for 2-DOF joint motion parameters of human wrist is studied.From the neurophysiological perspective,the transformation relationship between sEMG and joint control force signal is expounded.The linearization assumptions of the muscle synergy matrix model are analyzed.An improved NMF algorithm is proposed by adding sparse constraints to the matrices,which is used for estimating the joint motion parameters of wrist.The decomposition results of the proposed algorithm are analyzed with the muscle distribution figure.Then,the advantages of the proposed method are discussed.Secondly,Multiple subjects' data collection experiment is carried out,and the four algorithms(NMF,SCNMF,NMF-HP,CNMF-HP)are evaluated offline.The experimental platform is established and used for simultaneously collecting 2-DOF wrist motions and sEMG signals of the subject.The dataset is divided by 6-fold offline cross-validation,and the coefficient of regular term is determined by the results of the test.The sparse significance and decoding accuracy of each algorithm is evaluated in terms of the metrics,average signal-to-noise ratio and global correlation coefficient.Then,online target-tracking experimental platform is established,and three myoelectric control methods(NMF,SCNMF,CNMF-HP)are evaluated online.The software program is compiled in LabVIEW,which can collect sEMG signals,calibrate models,predict and drive the virtual cursor to track the target.Multiple subjects are invited to attend the online experiment.The performance of the online experiment is evaluated by using Fitts' Law metrics system,and three methods are compared.Moreover,the robustness of the proposed model is analyzed with multiple clinical compounding factors.The grasping process of objects in different initial postures is simulated by virtual hand and wrist model,and the online performance of the control method is verified preliminarily.Finally,the online experimental platform for the control of dexterous prosthesis is established,and the simultaneous control method is verified in real-time.The mapping between human wrist motion and the finger motion of prosthesis is built by grasping mode division.Two grasp strategies are proposed based on the decoding model.The dexterous prosthesis is controlled to grasp four kinds of objects(ball,cylinder,lateral and three-finger),and the grasping processes under the two strategies are analyzed and compared.The metrics,grasping time and flexibility is used to evaluate the experimental performance,which verified the priority of the proposed model.