Reinforcement Learning Fuzzy Neural Network Control for Magnetic Shape Memory Alloy Actuator

Abstract

Magnetic shape memory alloy actuator is a new type of actuator that can offer big travel and high resolution of output displacement, which makes it suitable for driving task. However, its output displacement represents the hysteresis applied to input magnetic field. Hysteresis restricts its application in the high precision positioning. In order to eliminate the hysteresis of magnetic shape memory alloy actuator, a reinforcement learning fuzzy neural network controller is proposed. Network structure and special learning algorithm of reinforcement learning fuzzy neural network controller are introduced in detail. The proposed control system adopts the generalized approximate reasoning-based intelligent control architecture, which is mainly consisted of three parts: Action Selection Network, Action Evaluation Network and Stochastic Action Modifier. Finally, in order to verify the effectiveness of the proposed control method, the simulation experiment is researched. The experimental results show that the proposed control method can obtain the smaller tracking error, and the controller’s maximum tracking error is less than 0.95%, hysteresis loop is less than 2.66%.