Neural Network-Based Sensorless Control of a Shape Memory Alloy Actuator

Abstract Shape memory alloy (SMA) wires can be used as actuators when heated above their austenite finish temperature, as their length reduces because of a phase transformation from martensite to austenite. During phase transformation, changes in electrical properties such as the wire resistance can be correlated to changes in wire length, thus enabling the self-sensing operation mode. This effect can be used to implement sensorless closed-loop control, where the mechanical feedback is provided by a position estimation obtained through the wire resistance. Self-sensing-based control has a significant advantage over sensor-based control, as it removes the need of additional electro-mechanical transducers into the control system. Due to strong nonlinearities intrinsic to SMA behavior, the relationship between resistance and wire elongation is often complex and hysteretic, making the practical implementation of self-sensing architectures in SMA rather challenging. In this paper, we present a self-sensing-based control of an SMA wire actuator in which a neural network is used to estimate the displacement by using electrical properties. The estimated displacement is further used as feedback to a PI control system, allowing in turn to implement sensorless closed-loop control on an experimental prototype. The performance of the self-sensing-based PI control is compared to laser sensor based PI control, in which position feedback is provided by a laser sensor.