Reinforcement Learning for Characterizing Hysteresis Behavior of Shape Memory Alloys

The ability to actively control the shape of aerospace structures has spawned the use of shape memory alloy actuators. These actuators can be used for morphing or shape control by modulating their temperature, which is effectively done by applying a voltage difference across their length. Characterization of this temperature–strain relationship is currently done using constitutive models, which is time and labor intensive. Shape memory alloys also contain both major and minor hysteresis loops. Understanding the hysteresis is crucial for practical applications, and characterization of the minor hysteresis loops, which map the behavior of a wire that is not fully actuated, is not possible using the constitutive method. Numerical simulation using reinforcement learning has been used for determining the temperature–strain relationship and characterizing the major and minor hysteresis loops, and determining a control policy relating applied voltage to desired strain. This paper extends and improves upon the numerical simulation results, using an experimental hardware apparatus and improved reinforcement learning algorithms. Results presented in the paper verify the numerical simulation results for determining the temperature–strain major hysteresis loop behavior, and also determine the relationships of the minor hysteresis loops.