An active learning strategy to study the flow control of a stationary cylinder with two asymmetrically attached rotating cylinders.

We numerically investigate the flow control problem of the flow passing a stationary cylinder at a fixed Reynold number 500 using two attached control cylinders with different rotation rates. Compared to the traditional uniform (lattice) sampling method, we developed an active learning strategy based on Gaussian Process Regression (GPR), drastically reducing the number of simulations and accelerating the scientific findings. We also discussed the effects of parameters on different hydrodynamic coefficients, and verified the feasibility of this strategy. The mechanism of this asymmetric flow control model was also further studied by analyzing flow patterns.