Artificial Elbow Joint Classification Using Upper Arm Based on Surface-EMG Signal

Abstract

This thesis proposes a method of elbow joint motions recognition using surface electromyography (sEMG) signal for disable people with below-elbow amputation. It solves the situation that forearm without muscle cannot control forearm pronation. The complete system could be categorised into 4 components: (1) signal measurement, (2) pre-processing, (3) classification and (4) control system. First, the signal measurement includes sEMG data collection and the relationship of motions and muscles. Second, the pre-processing component denoises the sEMG signals by soft threshold method. It reduces not only the electrical noise but also the white Gaussian noise. Third, the classification system recognizes elbow joint motions. Five characteristic features, Mean Absolute Value (MAV), Root Mean Square (RMS), Slope Change (SC), Signal Length (SL) and Zero Crossing (ZC), are extracted from denoised SEMG signals of each channel. The 5 features are used in back propagation neural network (BPNN) for the classifier of 3 channels, which outputs the 99.54% of the elbow joint motions accuracy from 8 healthy subjects. Furthermore, the results of classifier are tested on a subject-by-subject basis. It means that the classification system is a user- dependent system. It can detect the motions of healthy subject based on his or her own SEMG signals. At the end, the control system is designed in MATLAB and it demonstrates that the recognition results from the classifier is sent to the controller correctly. Moreover, the system is tested when the subjects lifts the weight of 1.5 kg and the accuracy is 96.64% that does not change significantly.