Recurrent neuro-fuzzy model of pneumatic artificial muscle position

Abstract

In this paper, a dynamic neuro-fuzzy system is proposed toward modeling the pneumatic artificial muscle, which are widely used in robotics and rehabilitation. To benefit from the outstanding advantages of the pneumatic actuators such as high softness and low weight-to-force ratio, efficient control of the actuator force as well as its displacement is essential. Attaining a comprehensive model with a satisfactory accuracy in the entire course of the muscle is the most important challenge regarding utilization of the pneumatic artificial muscle in a wide range of the applications. Therefore, an adaptive neuro-fuzzy inference system has been developed for pneumatic artificial muscle modeling. The subtractive clustering method is applied to reduce the number of fuzzy rules without loss of accuracy. To verify the effectiveness of the proposed modeling approach, an experimental setup has been constructed using a vertically suited pneumatic artificial muscle which holds a mass. Input-output data are collected for training and testing the recurrent neuro-fuzzy model. The experimental results demonstrate the desirable performance of the proposed adaptive neuro-fuzzy inference system method in modeling the pneumatic artificial muscle as well as its superiority compared to the mathematical model.