Model-Based Control of SMA Actuators with a Recurrent Neural Network in the Shape Control of an Airfoil

Abstract

This paper describes the development of a neural network control for shape memory alloy (SMA) actuators as well as the control tests. Precise control of SMA actuators that are integrated in the composite structure is difficult because of the hysteretic behaviour of SMA and time delays in the control system. The weakness of the static model of SMA is that it does not take into account the actuator dynamics. The NARMA-L2 neural network model enables us to solve the inverse dynamic model of the nonlinear discrete-time dynamical system. In this study the NARMA-L2 model of the airfoil (a cross section of wind turbine blade) was created using Matlab Simulink software and neural networks. Data for network training was measured from the specimen. The performance of the constituted NARMA-L2 controller was tested in control tests. According to the control tests the NARMA-L2 controller is well suited for controlling embedded SMA actuators. The constituted controller enabled fast and stable actuation in step responses and it was also able to compensate for changes in cooling conditions.

Analysis, optimization and testing of morphing structures employing shape memory alloy actuators

“…To this end two rows of SMA actuators were used to drive the displacement of the two control points. Ahola et al, (2009) developed a neural network to control the response of a morphing wind turbine airfoil based on SMA actuators. The developed model was trained based on data acquired by prototype testing, aiming to supply the required current to induce the phase transformation of the embedded SMA actuators.…”