Genetically-based active flow control of a circular cylinder wake via synthetic jets

The present work investigates the use of Machine Learning methods for optimizing the control of the wake behind a circular cylinder with the aim of reducing the associated aerodynamic drag using a single synthetic jet located at the rear stagnation point. Initially, a parametric study on sinusoidal shapes is performed to assess the control authority of the synthetic jet and to identify suitable initial configurations for the subsequent optimization study. This optimization leverages gradient-enriched Machine Learning (gMLC), which is based on Linear Genetic Programming, to determine the optimal waveshape for the input driving signal to the synthetic jet actuator, aiming at aerodynamic drag reduction. Machine Learning is thus exploited to overcome limitations inherent to canonical waveshapes. All the experiments are performed at a Reynolds number Re = 1.9 x 104. Four different optimization runs are conducted to study the effect of increasing the complexity of the genetic recombination process and including a power penalty in the cost function on the control effectiveness. The maximum drag reduction is achieved when no penalty for the power consumption is included in the cost function and amounts to 9.77% with respect to the baseline case. The addition of the power penalty results in control laws comparable in both waveshape and performance to the canonical sinusoidal control laws. In the second part of this work, the ML-derived control policies are investigated via hot-wire anemometry and Particle Image Velocimetry (PIV) to understand and characterize the mechanisms responsible for the drag reduction and the control effects on the wake evolution. For this purpose, a modal analysis based on Proper Orthogonal Decomposition is performed to comparatively assess the control laws and evaluate their capability of weakening and mitigating the most energetic flow structures associated with the vortex shedding phenomenon.