Estimation of electrical stimulation pattern in FES cycling using neural network

Abstract

In this study, we suggested a stimulation pattern estimation system using a neural network as a method of determining electrical stimulation pattern during FES cycling. Generally, stimulation pattern decided from crank angle read from encoder. Therefore, the bicycle not attached encoder is not able to FES cycling. However, if stimulation pattern estimation is performed by a neural network, it is possible to perform FES cycling without remodeling the bicycle. Therefore, backpropagation was performed on the input signal using the data of the inertial sensor and the teaching signal using the encoder angle, and the weight and the threshold value were determined. The obtained weights and threshold values were reflected in the system and the stimulation timing was estimated. As a result, stimulation pattern estimation system was possible to estimate the stimulation pattern equivalent to that when using the encoder.