Gradient-enriched machine learning control of wingtip vortices via online S-PIV and synthetic jets

This work presents a novel approach to control wingtip vortices, with the aim of mitigating their detrimental effects through the application of synthetic jets positioned at the wingtips. A Machine Learning-driven control strategy is developed using the gradient-enriched Machine Learning Control (gMLC) algorithm to optimize the electric signal driving the synthetic jets. High-speed online time-averaged Stereoscopic Particle Image Velocimetry (S-PIV) measurements in a crosswise plane in the near-field wake of the wing are employed in the optimization loop to define a proper figure of merit of the control based on the vorticity reduction of the mean flow. The most effective control law generated by the gMLC algorithm exhibits a square-like waveshape of the voltage signal with a dominant frequency equal to twice the Crow instability frequency. In addition, it achieves a reduction in vorticity of the mean flow of approximately 65% compared to the uncontrolled flow. This control policy overcomes the corresponding reduction achieved by a sinusoidal control tuned the same actuation frequency, demonstrating the effectiveness of the optimization procedure over conventional approaches. Additionally, the resulting optimal control policies are characterized a posteriori by analysing the synthetic-jet exit velocity response via hot-wire anemometry and the behaviour of the controlled wingtip vortices with the aid of phase-averaged S-PIV measurements thus providing a deeper insight into the impact of the control on the flow field. • Use of genetic programming to optimize synthetic-jet control of wingtip vortices. • Use of online high-speed S-PIV measurements to build a proper cost function. • A posteriori S-PIV investigation of the vortex evolution in controlled cases. • Vorticity reduction correlated to vortex displacement and blowing phase. • Control promotes vortex geometry reshaping and confined phase-average trajectory.