Design and development of electromyography signal driven prosthetic limb

Abstract

The work concentrates on the characterization of the bio-electric (sEMG) signal generated newlineduring the dynamics of human limb through experiments. The aim is to excavate the sEMG signal newlineproperties with respect to biological physical actions and to utilize the above obtained knowledge newlinefor the development of assistive limbs for afflicted human beings. newlineThe experiment is conducted by taking sEMG signals from forearm muscles (both biceps newlineand triceps muscle) during the dynamics of forearm and also from hamstring muscles (both leg) newlineduring normal gait cycle using surface electrodes. Different time domain features are extracted newlinefrom the sEMG signal after segmenting it into 250 msec window. These features are then analyzed newlinewith respect to dynamics (both angular displacement and angular velocity) of human limb. It is newlineobserved from the experimental results that two time domain features, IEMG (integrated EMG) newlineand ZC (zero crossing) are giving better correlation with the dynamics of forearm. Also observed newlinethat the time domain features, IEMG and SSC (sign slope change) are giving better correlation newlinewith the dynamics of gait during normal walking. newlineThe relationship obtained between the IEMG and ZC with the dynamics of forearm from the newlineabove experiment is made use to estimate the forearm dynamics using different models. A fuzzy newlinelogic model is proposed in this work for estimating the angular displacement of the forearm. An newlineempirical model (average value based model) is also proposed for the estimation of angular newlinedisplacement of forearm utilizing sEMG signal features and also using raw sEMG signals. The newlinemodels are validated using regression value and RMSE value between the estimated and actual newlineangular displacement of the forearm. Regression value obtained from the fuzzy model is 0.7975. newlineThe estimation of forearm dynamics is also done using several multiple input multiple output (MIMO) neural network models.