What do healthcare advances and grand chess masters have in common? Both are being shaped by machine learning in some way. In the past few decades, machine learning has made large strides in solving more and more complex problems. The technology is used in many different fields and has led to multiple innovative products such as Netflix’s recommendation engine and self-driving cars.
The field of fluid mechanics has used reinforcement learning (RL) to increase the performance of control on objects moving in a fluid environment. One of the key requirements of the RL is the ability to observe the environment. In this research, a 3D printed flow sensor performs the environmental observations in an experimental setup. RL and a 3D printed flow sensor are combined with the objective of actuating the flaps of an airfoil in such a way that the lift of the airfoil remains constant for changing wind speeds. The performed experiments include looking into the sensing abilities of the flow sensor in the airfoil at different wind speeds and training of the RL algorithm.
The overall design of the system, which include the RL scheme, the 3D printed flow sensor, airfoil and the education wind tunnel are presented. Following that the results of the experiments and the encountered drawbacks of the setup are discussed. A recommendation is given on how to solve some of the problems this research encountered and for future research to increase the performance of RL in combination with a 3D printed flow sensor.
To joing the presentation via Microsoft Teams click here