Design And Simulation On Myoelectric Control System Based On Pattern Recognition For Biomimetic Dexterous Hand

Using human EMG signal to control the intelligent biomimetic dexterous hand not only promotes the development of the cause of persons with disabilities, but also has great value in the field of the rehabilitation medicine and invasive surgery. The existing myoelectric control ways of intelligent biomimetic dexterous hand are: the way bases on threshold decisions, the way bases on amplitude coding, the way bases on hierarchical control decisions and the way bases on the pattern recognition. The myoelectric control of the biomimetic dexterous hand which bases on the way of pattern recognition includes the eigenvalue’s extraction and classification of EMG signal.This dissertation studies the design of the myoelectric control system for biomimetic dexterous hand. It is includes the 8 gesture’s signal processing and pattern recognition bases on the s EMG signal of human’s forearm, the myoelectric control system algorithm and myoelectric control system simulation of the biomimetic dexterous hand.Firstly, there is a detail overview of the research status and development process, it includes the dexterous hands, s EMG signal, dexterous hands’ myoelectric control system and virtual reality technology.Secondly, some questions have been analyzed, those are the EMG signal’s characteristics, production mechanism, acquisition methods and the methods of eigenvalue’s extraction and classification. The pattern recognition experiment for 8 gestures has been designed. The wireless s EMG signal acquisition system with high performance has been used to collect the s EMG signal witch bases on the subject’s forearm. The time autoregressive(TD-AR) model has been designed to obtain the eigenvalue with the ratio of average absolute value and 1-4 orders AR coefficients for each channel. The main element analysis method has been used to implement the eigenvalue matrix’s dimensions, and the integrated variable’s cumulative contribution rate is 99%. The function net=newpnn(P,T,SPREAD) in MATLAB has been used to build Probabilistic Neural Network and implement the movements’ recognition of 8 gestures.Thirdly, the virtual model of the biomimetic dexterous hand has been established bases on the Muscle Skeleton Modeling Software, and the myoelectric control system of virtual biomimetic dexterous hand has been implemented bases on the Simulink.The advantages on signal processing and pattern recognition method is that the TD-AR eigenvalue’s extraction model has made full use of the preponderance with simply count, quickly obtain in time domain eigenvalue analysis method, and made up for its shortcomings with lower stability and incomplete spectrum information of s EMG signal in terms of the eigenvalue’s extraction. The gesture recognition model based on Probabilistic Neural Network has any advantages with simple structure, strong stability, fast convergence, shorter training time and excellent fault tolerance. The average recognition rate for 5 subjects’ 8 movement gestures state is 92.2%.The innovation in virtual control system is use a new kind of musculo-skeletal simulation software to build the Shadow Hand virtual control model. The 6-channel subject’s forearm s EMG signal is used as the input of the recognizer, and finite state machine is used to drive the virtual Shadow Hand model. The myoelectric control system of virtual biomimetic dexterous hand has been implemented.The measurement, estimation, and visualization system described here can facilitate the development and testing of new dexterous hand system, which could lead to important application in medical robotics, intelligent prosthetics and new human-machine interactive system.