Development of SEMG and artificial neural networks based powered ankle prosthesis control algorithms for stair ascending and descending motions

Abstract

Amputation is the surgical removal of a limb due to various reasons, e.g trauma. Prosthesis is a device which is a replacement for the missing part of the limb. Ankle joint can have loads of 10-13 times of the body weight during power demanding activities. Since energetically-passive prostheses cannot generate net power output, powered ones become essential for demanding tasks. Surface electromyography (sEMG) is a non-invasive method which measures neuromuscular activity. The aim of this study was to develop artificial neural network models to predict ankle moment and position using only sEMG input for control algorithms of stair ascending and descending tasks. Time delay neural network and long short-term memory were compared for this aim. Features that represent sEMG signals better were investigated. Minimizing the number of sEMG signals from lower leg muscles can make prosthesis flexible while reducing the number of sEMG sensors required can make the prosthesis economic. Correlation of 0.90 between the predicted and actual values was set as the performance threshold. Long short-term memory based algorithms achieved significantly higher performances. 0.91 and 0.93 correlations were achieved for both motion tasks0 position and moment, respectively. The minimum number of sEMG sensors was 2 for moment and 3 for position estimation. The minimum number of lower leg muscles required was 1 for moment and 2 for position estimation. The results show that there are promising EMG sensor combinations for the specified targets.|Keywords : Ankle Prosthesis, Algorithm, Electromyography, Stair Climbing, Artificial Neural Networks.