Neural-Network-Based Iterative Learning Control for Hysteresis in a Magnetic Shape Memory Alloy Actuator

Abstract

The typical characteristic of a magnetic shape memory alloy (MSMA)-based actuator is rate-dependent and stress-dependent hysteresis. In addition, the hysteresis in an MSMA-based actuator includes memory, multivalued mapping, and asymmetry properties, which seriously degrade the positioning accuracy. In this article, a feedforward neural network (FNN)-based nonlinear autoregressive moving average with exogenous inputs (NARMAX) model is employed to describe the hysteresis in an MSMA-based actuator. Through a Taylor expansion, it is revealed how the FNN constructs the nonlinear function of the NARMAX model. For the controller implementation, an FNN-based iterative learning control (ILC) strategy is proposed to achieve the desired tracking performance. The contributions of this study include exploring the update rule of the FNN-based ILC and analyzing the convergence properties of the proposed controller when the initial state of the system varies. To verify the effectiveness of the proposed method, experiments on an MSMA-based actuator were conducted. The results illustrate that the proposed modeling and control methods achieve an excellent performance and that the theoretical results are correct.