Study On The Pattern Recognition Technique And Control Strategy Of Knee Joint Rehabilitation

With the frequent traffic accidents and advent of aging population society, the number of patients with knee injuries or residual in the lower limbs rises sharply. The lower limb rehabilitation robot designed by our group uses self-developed pneumatic artificial muscles, avoiding secondary damage in the process of rehabilitation. Traditional lower limb rehabilitation robot uses passive control strategy, lack of participation initiative of trained people. Surface EMG of lower limbs can not only react leg muscle activity, but also act as a control signal, which has a quite bright future in the field of medical rehabilitation.This paper aims to analyze the surface EMG, achieves EMG pattern recognition in order to control lower limb rehabilitation training system, research topics include the following aspects:(1) An anti-interference bypass has been added to the main circuit. The main circuit consists of front differential amplifier,20Hz high-pass filter,50Hz notch filter,500Hz low-pass filter and the post variable gain amplifier, the main circuit magnifies EMG between 1020～ 6120 in theory, multi-stage amplifier avoids signal distortion caused by over-saturation. The bypass is made up with "shield guard", "right leg driving" and "floating power supply" circuit, playing the role of suppressing common mode interference of human body. To build up the whole signal acquisition system and write acquisition program with Labview. It has solved the problem with SEMG that amplitude is small, noise ratio is low, it is easily influenced by ambient noise and gotten hard.(2) Using RMS, fourth-order AR model parameters, power spectral feature extraction ratio K as parameters, designing lower limb movement patterns under different state of motion, different load conditions, acquiring EMG and extracting feature values. Under a particular action mode, Observing discrimination case reflected by three kinds of features values, obtaining information conclusion each characteristic reflects in lower limb motion. It has solved characteristic parameter selection under specific action.(3) Designing 3-layer BP neural network classifier, using signal eigenvalues RMS and fourth-order AR model parameters to build up a new feature vector, using it as the input of BP neural network. Designing comparison test, when the new training feature vectors is treated as input, When training error reaches 9.5245-6, the number of iterations is 246, the recognition rate is 96.9%, confirming the superiority of the new feature vectors in convergence speed and recognition rates. It has solved the problem of movement pattern recognition of Lower limb.(4) Completing the communication interface of lower limb rehabilitation training system, proposing EMG mode recognition control strategy, using Labview to write system control program. Completing pattern recognition control experiment, the trainers can go on rehabilitation training according to the predict, it has completed the demonstration of the thesis project.