Machine learning control of airfoil lift increase using co-flow jet

Co-flow jet with zero-net-mass-flux is an effective method of increasing the lift of an airfoil. However, the optimal control law and the associated altered flow structures remain unclear. This paper describes a machine learning control law based on the ant colony algorithm for optimizing the control parameters, where the cost function involves lift increase, drag variation, and control power input. This technique was applied to an RAE2822 airfoil with an angle of attack of 10° at a Reynolds number of 6.5×10⁶. The co-flow jet was produced by blowing and suction at the airfoil leading and trailing edges, and was parameterized by their locations, angles, and momentum coefficients. MLC identified a forcing that increases the lift coefficient by 372%, corresponding to a remarkable lift-to-drag ratio of 32. This greatly exceeds the previously reported lift increase for this airfoil (26%). The discovered forcing also provides alternative solutions, such as a tremendously increased lift-to-drag ratio and reduced control power input given a small sacrifice in lift increase. Systematic flow analysis indicated significant alterations in the coherent structures over the airfoil upper surface and wake under the optimal control, such as low-pressure separation regions on the upper surface and periodic vortices at the trailing edge. These changes are linked to the significantly extended low-pressure region and substantially enhanced suction peak on the upper surface. This mechanism was compared with that which occurred under the efficient control law. The results offer valuable guidance for future research and engineering applications.