Protocolo Experimental para el Entrenamiento y Evaluación de Algoritmos Mioeléctricos en el Control de Prótesis Transfemoral

An experimental protocol for training and testing the myoelectric algorithms based on neural networks for the transfemoral prosthesis control is presented. The experiments are supported by a bioinstrumentation module and real time acquisition software. The first experiment is based on the acquisition and pre-processing of myoelectric signals for evaluating algorithms based solely in electromyographyc signals. The second experiment adds gyroscope sensors and allows evaluating myoelectric algorithms with fusion the electromyographic signals and propioceptive sensors. With the information obtained from the training and testing records various myoelectric algorithms for the knee angle estimation based on the time domain, time-frequency domain, time-scale domain and data fusion were implemented in off-line mode. The experimental protocol allowed the error events conformation, including the number of error events, the maximum error event duration and the maximum error amplitude in the angle continuous estimation in off-line mode. These indicators can be applicable with modifications to discreet movement estimation. A sufficient amount of data carefully recorded and a correct experimental architecture represents the keys step of a very good performance in the pattern myoelectric classification.