Neural network-based micropositioning control of smart shape memory alloy actuators

Shape memory alloys (SMA) are a special kind of smart materials whose dimensions change because of a temperature-dependent structural phase transition. This property can be used to generate motion or force in electromechanical devices and micromachines. However, their highly nonlinear hysteretical stimulus–response characteristic fundamentally limits the accuracy of SMA actuators. The purpose of this work is to design nonlinear control methods suitable for SMA-based positioning applications. To account for the hysteresis effects, inverse hysteresis models are inserted in proportional integral with antiwindup control loops. The inverse hysteresis models are obtained both using a linear phase shift approximation and by training neural networks using experimental data. It is found that neural networks are excellent tools perfectly capable of learning the hysteresis effects. Several control strategies, with and without compensation, are experimented on a laboratory SMA actuator and it is found that neural networks successfully improve the closed-loop response, leading to position accuracies close to the micron.