A preliminary investigation of the effect of force variation for myoelectric control of hand prosthesis

Abstract

The myoelectric control of prostheses has been an important area of research for the past 40 years. Significant advances have been achieved with Pattern Recognition (PR) systems regarding the number of movements to be classified with high accuracy. However, practical robustness still needs further research. This paper focuses on investigating the effect of the change in force levels by transradial amputee persons on the performance of PR systems. Two below-elbow amputee persons participated in the study. Three levels of forces (low, medium, and high) were recorded for different hand grips with the help of visual feedback from the Electromyography (EMG) signals. Results showed that changing the force level degraded the performance of the myoelectric control system by up to 60% with 12 EMG channels for 4 hand grips and a rest position. We investigated different EMG feature sets in combination with a Linear Discriminant Analysis (LDA) classifier. The performance was slightly better with Time Domain (TD) features compared to Auto Regression (AR) coefficients and Root Mean Square (RMS) features. Finally, the error of the classification was considerably reduced to approximately 17% when the PR system was trained with all force levels.