Research On Prosthetic Hand Control Based On Mechanomyography And Continuous SEMG

With the progress of society,people with disabilities’ living conditions have been paid more attention.Research on assistive products for the disabled in the field of scientific research is gradually developing.Aiming at solving the problems such as bloated structure,complex control and poor human-robot interaction experience of existing prosthetic hand,this thesis studies the control system of prosthetic hand and the way of human-robot interaction.In the control system,a set of low-delay prosthetic hand control system based on independent continuous surface electromyogram(sEMG)signal is presented,which ensures the correct rate and real-time performance of sEMG control of the prosthetic hand.In the aspect of humanrobot interaction,an airbag arm ring sensory system based on the acting signal is designed,and the precise recognition of common gestures is completed by using pattern recognition and tranfer learning methods.Firstly,a hardware platform for the prosthetic hand control system and human-robot interaction is established.According to the principle of gesture synergy,this thesis uses principal component analysis to reduce the complex structure of human hands to the four-dimensional principal component and complete the mechanical reproduction.At the same time,according to the design requirements of the function,energy consumption and volume of the controller,the embedded controller of the prosthetic hand is designed.In addition,this paper designs an airbag arm ring sensory system based on the the mechanomyography,and completes the design of hardware and software program.Secondly,this thesis achieves the classification of gestures based on mechanomyography.Linearity analysis proves that the mechanomyography is proportional to the muscle force,and verifies that the mechanomyography is more anti-noise ability than the sEMG.At the same time,a simple classifier is used to realize the offline recognition rate of up to 98% for six common gestures,and its practicability is verified by online test experiments on virtual platforms.Finally,to solve the problem that the training time is too long,this thesis presents a semi-supervised tranfer learning algorithm based on subspace alignment domain adaptation,which can guarantee the accuracy of gesture classification on 1/10 training data,and greatly shorten the training time.Finally,a low-delay control method for prosthetic hand based on independent continuous sEMG is presented.Based on independent component analysis,an algorithm for extracting sEMG independent signals is proposed,which can reduce sEMG interference caused by improper positioning of surface myoelectric electrodes,and can improve the control accuracy by up to 47%.To reduce the delay in the control process,a correlated vector autoregression algorithm was proposed to reduce the control delay of the prosthetic hand to less than 100 ms.Finally,the dexterity of the control system is verified by the dexterity test;the real-time of the control system is improved by measuring the motion delay time of the manipulator;and the functionality of the prosthetic hand is further verified by successfully completing 30 daily grabbing movements with the prosthetic hand operated by the disabled.