Two Methods of EMG Analysis for the Purpose of Exoskeletons and Robotic Rehabilitation Devices

Abstract

The paper introduces new methods for EMG signals analysis. Two types of artificial neural networks have been applied for this purpose, namely the learning vector quantization (LVQ) neural network and the competitive neural network. The sets of EMG signals were recorded for different walking conditions. The gait of a healthy person without muscle disorders history were recorded. The study proved that the LVQ neural network provides a good tool for detecting the differences in the EMG trajectories coherent with the different effort of the muscles work, and the competitive neural network performs well the EMG clustering. The last property allows to identify the motion fragments with similar muscles effort and to predict the farther continuation of movements after recording the initial fragments of EMG. Obtained results are useful for robotic applications and could use for the reference for exoskeleton design and rehabilitation purposes.