Machine learning vortices (IMAGE)
Caption
The deep reinforcement learning agent is able to rediscover large-scale streamwise vortices strategy on its own for this high-Reynolds-number 3D flow, while also learning to adjust the wall-normal jets to sustain the vortices for long time periods and obtain effective flow reattachment. The visualization shows in the first row the flow with the machine learning control and in the second one the original flow without control. The two panels on the top right show that the machine learning creates a particular type of structure (streamwise vortices) to reduce the separation region. The creation of this type of structure is discovered by the machine learning itself.
Credit
Bernat Font, Francisco Alcántara-Ávila, Jean Rabault, Ricardo Vinuesa & Oriol Lehmkuhl
Usage Restrictions
Credit must be given to the creator. Adaptations must be shared under the same terms.
License
CC BY-SA