Development of a Closed Loop FES System Based on NARX Radial Based Network

Abstract

Abstract-In this paper design of a closed loop FES system for torque control is presented. Snap power worker s used for measuring muscle torque. Using this system torque is proportional to angle of a flexion so by controlling angle of a flexion torque is controlled too. During functional electrical stimulation 3 parameters can be changed: pulse width, pulse amplitude and time between two impulses. In this paper pulse amplitude and frequency are constant and system is controlled by changing pulse width. PI regulator is used to minimize difference between desired trajectory and system output. Parameters of a PI controller are obtained by minimizing mean square error on simulated system. In this paper NARX Radial based neural network model of system is used for simulation. Result on a real system show that this type of control system can be used for torque control.

Functional Electrical Stimulation System for Drop Foot Correction Using a Dynamic NARX Neural Network

“…NARX is a class of discrete-time non-linear systems that establish non-linear relationships between past observations and future outputs [18]. These models are useful for FES research purposes due to the small number of required parameters and their ability to represent the nonlinear dynamic behaviour of the electrically stimulated muscle [35].…”

“…A neural network is a type of black-box model that can be trained with the input and output data of the stimulated muscle to output the correct stimulation pulse depending on the desired trajectory [17]. Multiple studies have proposed the use of recurrent neural networks for modelling stimulated muscle regarding the upper [7,18] and lower limb movements [5,6,17,19], given their higher ability to identify dynamic systems with a smaller network structure and less number of parameters than other neural network types. Considering lower limb applications, Yassin et al [5] developed a study to compare the performance of using a NARX or a Cascade Forward Neural Network (CFNN) to model the quadriceps muscle behaviour (torque) based on stimulation frequency, pulse width, pulse, and duration of muscle excitation.…”

Functional Electrical Stimulation for Gait Rehabilitation

“…NNs have been extensively used in FES research projects to model electrically stimulated muscles, be it for upper limb movements [8][9][10] or for lower limb movements [11][12][13][14][15][16], given they can model the non-linear behavior of the electrically stimulated muscles. Although in the past, it was more common to use feedforward neural networks with FES [8,12,13], the most recent projects use recurrent neural networks, in particular Non-Linear Autoregressive Neural Networks with Exogenous inputs (NARX NN) [9][10][11][14][15][16]. These models are useful for FES research purposes because of the small number of required parameters and their ability to represent the nonlinear dynamic behavior of the electrically stimulated muscle.…”