Neural Network Based Iterative Learning Control for Dynamic Hysteresis and Uncertainties in Magnetic Shape Memory Alloy Actuator

Magnetic shape memory alloy-based actuator (MSMA-BA) is constructed based on the strain mechanism of MSMA material and the magnetic effect of electric current. It can generate macroscopic deformation with micro-nano scale resolution. However, the rate-dependent and load-dependent hysteresis characteristics in MSMA-BA will reduce the positioning accuracy and hinder its applications. In this study, a long short-term memory (LSTM)-based U model with exogenous inputs is proposed to describe the complex dynamic hysteresis characteristics. Then, an LSTM-based iterative learning control (ILC) scheme is proposed to realize the reference trajectory tracking control of the MSMA-BA. Additionally, a dynamic expansion compression factor (DECF) is introduced in the controller to accelerate the convergence speed of system. The convergence of the proposed LSTM-based ILC scheme is analyzed with the consideration of state uncertainty, output disturbance, and the initial state error. It will promote the further applications of ILC in practical situations. Experiments are carried out on MSMA-BA to validate the effectiveness of the proposed method. The experimental results indicate that the proposed modeling and control methods exhibit excellent performance.