Self-sensing SMA Actuator Using Extended Kalman Filter and Artificial Neural Network

Abstract

In this paper, self-sensing capability of Shape Memory Alloy wire actuator has been explored using Extended Kalman Filter assisted Artificial Neural Network. The change in length of a linear spring actuated using a Shape Memory Alloy wire is first estimated from the variation of its electrical resistance using Extended Kalman Filter. Though the estimation is qualitatively in agreement with the experiment, the quantitative mismatch makes it difficult to control the stretch of the spring solely based on the Extended Kalman Filter estimation. An Artificial Neural Network has been used to bridge the gap between the Extended Kalman Filter estimation and actual stretch of the spring. To evaluate the effectiveness of the Extended Kalman Filter based Artificial Neural Network model, the responses of the same are compared with that of the another Artificial Neural Network model, trained only using the experimental data. It has been observed that for the same number of neurons and same training data, Extended Kalman Filter based Artificial Neural Network model yields better result at higher frequencies.