Neural Network-Based Gain Scheduled Position Control of a Pneumatic Artificial Muscle

Abstract

The pneumatic artificial muscle (PAM) is considered one of the most preferred actuators in a variety of robotic and industrial applications. However, due to their inherent nonlinearities and hysteretic properties, they are difficult to model and the controller’s design becomes more sophisticated. The position control problem of a PAM having different regions of operations at various axial loads is considered in this paper. A neural network-based gain scheduled proportional-integral-derivative (PID-NN) control scheme has been synthesized and compared to the classical linear PID controllers. The PID gains for different operating regions at different loads are determined using Zeigler Nichols sustained oscillation method. These sets of PID gains are then used to determine the neural network (NN) model that schedules them based on the region of operations and axial loads. To validate the efficacy of the proposed control scheme with regards to different step inputs and a sinusoidal input reference tracking performance, experimental studies are conducted, and comparisons have been made with the PID controller. The experimental results for position control confirm the efficacy of the proposed control strategy.