Estimation of displacement of a Shape Memory Alloy (SMA) actuator from electrical properties, is one of the attractive attributes of SMA. To utilize this feature, it is necessary to model the correlation between electrical properties and strain of the wire. Identification of the relationship between resistance and displacement is not always straightforward in SMA actuators, as the complexity of the resistance-displacement characteristics may vary depending upon various factors, e.g., biasing mechanism, additional external loads, actuation frequency, as well as the electrical activation strategy for heating SMA wires. In this poster, we present an experimental setup that incorporates a spring-loaded SMA wire mechanism. Resistance across the SMA wire has been measured for different amplitude and actuation frequencies. In general resistance behavior has been observed and analyzed, which motivates the use of data-based approaches for estimation of the displacement. Neural networks represent a possible solution for identifying the complex correlation between electrical input and displacement for different configurations of SMA-actuated systems. The displacement estimated by the trained model is then compared against the displacement measured by a laser displacement sensor, for validation purpose.

An embedded system is developed to collect characterization data. Power and Resistance are calculated from the measured Current and Voltage, to use them as inputs to the neural network for estimation of the displacement as an output of the system. Practically, it has been observed that SMA actuators has hysteretic correlation between resistance and displacement as well as power and displacement.

In this poster, we present a data-based approach for self-sensing in SMA actuator. Based on collected characterization data of a SMA actuator with a linear spring bias system for various amplitude and actuation frequencies, we have trained proposed neural network model for the estimation of displacement using measured electrical properties. The performance of the neural network on separate training and validation data confirms that it has good generalization capability, which motivates the use of data-based approach for self-sensing in SMA actuators.