Upper Limb Elbow Joint Angle Estimation Based on Electromyography Using Artificial Neural Network

Abstract

This paper proposed a time delay artificial neural network (TDANN) to estimate elbow joint angle based on electromyography (EMG) signal. One channel EMG signal was recorded from biceps using disposable surface electrode while the upper limb elbow joint performed a flexion and extension motion randomly. The EMG signal was extracted using Wilson amplitude feature with windows length of 100 sample points. In order to identify the EMG features, the TDANN was constructed as follows: input, hidden, and an output layer consists of 5 time-delays of input nodes, 15 hidden nodes, and 1 output node, respectively. The proposed method reveals that by using single channel EMG from biceps, it is able to estimate the elbow joint angle. The performance of the elbow joint angle estimation in the random motion is $18.87^{\circ}\pm 3.46^{\circ}$ and $0.80\pm 0.09$ for RMSE and Pearson's correlation coefficient, respectively.