A Study For SEMG Control Of Lower-Limb Rehabilitation Robot

There is a study on Surface electromyography signal (SEMG) in this paper. SEMG is a kind of bio-electricity signal collected from the surface of human skin by surface EMG electrodes, which can reflect the subjective intention of human activity. The lower limb rehabilitation robotics, as the application platform, can be controlled by the EMG to improve the patients’rehabilitation by leading the movement of the injured limb. The main work of this paper is to analyze the SEMG collected from the lower limbs, to realize the identification and classification of the various stages of lower limb knee flexion movement, and to establish the relative reflection relationship between SEMG and the subjective intention of human activity in varying degrees.The research background and lower limb rehabilitation robot research status at home and abroad was investigated and discussed in this paper. Then the point is focused on expounding of the production mechanism, characteristics and detection principle of SEMG, which is a foundation laid for the design of the SEMG acquisition system.The SEMG processing generally includes two steps:feature extraction and pattern classification. Based on the study of classical weak signal processing system and combined with the characteristics of the target signal SEMG, the main parts of system structure are determined:electrodes, signal conditioning circuit, AD acquisition circuit and PC software. Signal conditioning circuit consists of: front-end amplifier circuit, band-pass filter circuit, power frequency trap circuit. Every part of this acquisition circuit made was introduced and analyzed correspondingly before the design of Signal conditioning circuit. After comparative analysis of all kinds of methods for SEMG processing, a suitable method is selected:time domain characteristic of RMS combined with frequency domain characteristics of MF as the input of BP neural network, and then a best structure of the BP neural network is established.In this paper, quadriceps are chosen as the main signal collection area based on the structure of human lower limb skeletal muscle, and the collecting solution is determined:three channels of SEMG of lower limb quadriceps and one channel of knee joint angle. At last, the design of experiment was introduced, and the experimental data is analyzed using time domain method, frequency domain method and the HHT analysis with the EMD decomposition. Finally the six dimensional vector is extracted that combined time domain feature with frequency domain feature as input of BP neural network, and successfully realize the identification and classification of the four stages of the knee flexed movement, that reflecting the subjective intention of human activity in varying degrees.