Self-sensing shape memory alloy wire actuators using Kalman filters
No Thumbnail Available
Shape Memory Alloy wire actuators are very popular for offering large actuation force as well as displacement among other smart materials. The response of a SMA wire actuator is non-linear and hysteretic. In literature, sophisticated feedback controllers are being proposed to tame the behaviour of SMA actuators. Thus feedback sensors, measuring the response of the system being actuated by SMA wires, become essential part of the system. To get rid of the external sensors, the fact that the electrical resistance of a SMA wire varies during actuation, has been harnessed as a measured of the system output. This implies the use of SMA wire as a self-sensing actuator. In literature, this has been implemented mainly, by developing empirical relations, from the series of experimental data obtained for a given system, relating the change in electrical resistance of the SMA wire and the system response. Though, these approaches yield reasonably good response; however, are limited to a specific system and under specific loading. Moreover, no methods are proposed to obviate the noise that are available in the measured resistance data, particularly at low voltage level. For soft system, where hysteresis is significant, obtaining a unique relation is questionable. As a solution, in this thesis, different Kalman filters, as estimators are developed for SMA actuated linear and nonlinear systems. For a given time varying voltage signal, the stress and temperature of the SMA wire actuator are estimated using a dynamic model of the SMA actuated system and measured electrical resistance of the SMA wire. An Extended Kalman filter model of the system is developed and implemented in a real-time hardware, and is found to produce qualitative estimation of the system response. The Unscented Kalman filter is also explored to improve the estimation accuracy. Finally, a modified version of the Extended Kalman filter is proposed by augmenting some of the uncertain parameters of the model as state variable. This approach is found to estimate the system response with appreciable accuracy. In fine, the EKF model has also been implemented in a low cost hardware, Arduino Uno, to demonstrate the applicability of the developed approach, in practice, using the less resourceful hardware.
Supervisor: Atanu Banerjee