Adaptive neural Preisach model and model predictive control of Shape Memory Alloy actuators

Shape memory alloy (SMA) actuators are potential alternatives to conventional actuators in many surgical, aeronautic and biorobotic applications. However, their highly nonlinear hysteretic behavior between temperature and strain makes them difficult to use in real-time applications. The Preisach model is a well-known phenomenological method to accurately model the hysteresis of many physical system. In this paper, the numerical Preisach model is modified such that it can adapt to changes in the operating conditions, and can easily be used in real-time for controlling the strain in the SMA actuators. For this purpose, both the first-order descending and ascending curves are used and each of these curves is approximated by an artificial neural network (ANN). Weights of the ANNs are then updated online using the extended Kalman filter algorithm. To control the strain in the SMA actuator, a model predictive controller is implemented that uses the temperature dynamics and the adaptive modified Preisach model to calculate the optimal control signal using the Levenberg-Marquardt algorithm. Performance of the proposed model and the control scheme are validated using an experimental setup.