Design of an Intelligent Controller for Myoelectric Prostheses based on Multilayer Perceptron Neural Network

Abstract

Myoelectric prostheses have been researched widely, and some cases have been implemented to be used by amputees in real life. However, natural control of an active prothesis remains a challenge. This work presents an exploration of an intelligent controller for upper prostheses based on myoelectric signals. A simple intelligent classifier for a small control system is designed and incorporated into a hand prosthesis to be used by the amputees in Iraq and similar developing countries. To achieve this, a Multi-Layer Perceptron Neural Networks (MLPNN) classification system is developed. The proposed system uses pattern recognition based on features extracted from eight raw EMG signals collected using a Myo armband. Five different classes of hand gestures are recognised. The system also applies remove silence process and overlapped segmentation to the collected EMG data. Continuous real values that represent class types are sent to the controller to move the prosthesis. This work shows that, by adding appropriate pre-processing, a considerable increase in the accuracy of the proposed MLP classifier can be obtained. The required hardware circuits were assembled and software scripts written to implement the intelligent myoelectric hand prosthesis.