Algorithms And Systems Of Multi-DOF Prosthetic Based On Surface Electromyography

The myoelectric prosthesis hand is a type of anthropomorphic hand that based on the surface electromyography(sEMG) signals for amputee rehabilitation. It has important practical significance on improving the daily lives and mental for amputees. Despite some progress has been made on multi-DOF myoelectric prostheses in recent years, it still cannot meet amputees' requirements, such as few degrees of freedom, poorly stability, lack of real-time effects and bad intuitive control feelings. This paper focuses on the algorithm of patteren recognition and the real-time system, studying the virtual hand-platform and multi-degree of freedom prosthetic hand. The specific procedures are as follows.By comparing different pattern recognition control algorithm, this paper selected SVM as the real –time pattern recognition algorithm. In order to validate it, an experiment was carried out that the subject carry out five common hand movements(open, closed, lateral, cylindrical, hook), when the sEMG was acquired from his muscle. Then the data was used to build and test classifier. The result shows that the overall recognition rate is 89.67%. And later the algorithm was transplanted on the DSP real-time system and the virtual hand platform. Through experimental tests, the recognition rate was 80% and 85%, illustrating that the support vector machine can be better applied to real-time control of myoelectric prostheses.This paper studies a virtual hand-platform. The platform is Mixed-Language Programming based on Labwindows and Matlab, wherein Labwindows use Data Acquisition Card to acquire the multi-channel surface EMG and transport the data to Matlab workspace making use of the Active Control, then the Matlab use the data complete the pattern recognition and let the virtual hand carried out the movement. The virtual hand model is made by VR Builder, according to the size and motion parameters of human hand. Finally, experiments show that the virtual hand can be a good platform to perform different gestures. And the real-time tests found that the system can complete every action at 300 ms since the starting of the movement.The real-time mechanical prosthetic hand is based on the DSP. The mechanical prosthetic hand has five fingers which each with three joints can complete independence individually driven by a motor. The system uses DSP as the signal acquisition and analysis platform, and slave computer is in charge of getting the pattern and the drive motor control. Wherein the AD Acquisition by DMA manner to ensure the high speed signal read.The SVM recognition algorithm which has been tested on the virtual platform is transplant performed in C language using the Matlab Coder. And then been implemented on the DSP platform.Finally, through the simulation tests, we found that the delay time of the system is about 200 ms.To cope with the problem that some factors, such as the individual properties of anatomy, physiological state can cause different sEMG even in same motor tasks, the bilinear model representing the activities with motion factor and individual factors provides a novel approach for myoelectric prosthetic control. Here the bilinear model was introduced to extract neuromuscular characters of the Extensor Digitorum's activities. Firstly, the activities of the muscle were described by the multi-dimensional features vector including integrated sMEG amplitude and spatial distribution. Secondly, the vectors were decomposed into user-dependent matrices and motion-dependent matrices. To test the model, 6 subjects have been recruited for finger force-tracking tasks. The index, middle and ring finger produced force of 20% MVC, 40% MVC and 60% MVC respectively. At the same time, sEMG was collected from Extensor Digitorum with 32-channel flexible electrodes' array. Then, the bilinear model was built to recognize the task fingers and force levels. The results showed that it can classify the fingers and the force levels through only a few interactions, which suggested that the bilinear model can simplify the training procedures, and can be potentially applied to control the finger motions and force levels of myoelectric prosthetic hand.