Adaptive displacement online control of shape memory alloys actuator based on neural networks and hybrid differential evolution algorithm

Abstract

Shape memory alloys (SMAs) are smart metallic materials, which have the ability to recover their shape when heated, even under high-applied load and large inelastic deformation. This characteristic helps SMA provide an interesting alternative to replace conventional actuator. This paper proposes an adaptive online displacement control of an SMA actuator that is created by combining an adaptive feed-forward neural networks ( AFNNs) model and a PID feedback controller to increase the accuracy and to eliminate the steady state error in displacement position control process of the SMA actuator. The AFNN model, which is created by combining a multi-layers perceptron neural networks (MLPNNs) structure and an auto regressive with exogenous input (ARX) model, is used for modeling and identifying the hysteresis inverse model of the SMA actuator. Then, a new hybrid differential evolution (HDE) algorithm, which is a combination between a traditional differential evolution algorithm and a back-propagation algorithm, is used to optimally generate the best weights of the AFNN model. Due to the offline identification, the proposed adaptive online displacement control can learn the hysteresis behavior of the SMA actuator in advance and then provide online control signal efficiently. Consequently, the displacement of SMA actuator is controlled robustly and more precisely.