Adaptive control of a shape memory alloy actuator using neural-network feedforward and RISE feedback

A This paper presents a position tracking control system for a shape memory alloy (SMA) actuator using neural network (NN) feedforward and robust integral of signum of error (RISE) feedback. Nonlinear control of SMA actuators is difficult due to model uncertainties and unknown disturbances. Discontinuous control techniques such as sliding mode control have conventionally been used to achieve asymptotic tracking in the presence of model uncertainties. However, such discontinuous controllers usually result in increased power loss due to high frequency switching. With the recent development of the continuous RISE feedback control, semi-global asymptotic tracking can be achieved. Furthermore, the NN-RISE control leads to better tracking performance and lower power losses caused by input signal switching/chattering when compared to discontinuous controllers. In order to apply the NN-RISE, a state-space model of the SMA actuator is derived, which has been overlooked in many previous works, using Taylor series expansion and exploiting the nature of SMA dynamics. Experimental results show that the proposed control system works well even in the absence of an accurate model of the SMA actuator.