Gaussian Process optimization for fixed-structure control applied to a morphing airfoil

More Info


With the current view on sustainability and global warming, renewable energy sources are becoming more and more important. One of these environmental friendly energy sources is wind energy. To make the price of wind energy more competitive compared to other energy sources, the current trend in the industry is to increase the rotor diameter of the turbines. However, with these larger turbines the loads and oscillations due to instabilities such as flutter also increase. By reducing this, the required maintenance is decreased and the lifetime increased, thus lowering the cost of wind energy. One of the solutions to obtain load reduction is by using a smart rotor, which is an airfoil with the ability to actively influence the airflow. One type of smart rotors that is used in this research is a morphing airfoil, which can morph its shape using piezoelectric actuators. However, this system is highly non-linear and therefore the classical method of control is not possible. A data-driven approach is used in the form of Gaussian Process regression. This is a versatile method that can be used in machine learning. It has stochastic properties and is able to deal with uncertainties and non-linearities. This algorithm is then used to obtain the optimal gains of a fixed-structure controller. This algorithm is used on a flutter model, which simulates a smart rotor using a flap to influence the airfoil. This model is not as non-linear as a morphing airfoil, but is the most relevant theoretical model available to test the algorithm on. With the GP controller tuning algorithm, an optimal controller was obtained that could remove the instability of flutter. To do experiments with the morphing airfoil prototype, an experimental setup was designed and attached to a small closed-loop wind tunnel. Unfortunately, this setup had lots of friction, which makes it not possible to obtain instabilities such as flutter in the wind tunnel. Therefore, the performed experiments are reference tracking and disturbance rejection, where the disturbance was generated by one side of the airfoil and countered by the other side. The Gaussian Process controller tuning algorithm was able to tune the gains for a PI and PID controller for these objectives with reasonable performance. Concluding, it can be said that morphing airfoils shows potential and can influence the wind flow in such a way that it is comparable to other smart rotor designs that, for example, make use of flaps.