Data-driven control of very-large-scale motions in a turbulent boundary layer using wall deformation

In previous experiments, we showed that a circular active deformable surface driven by a simple feedforward control algorithm can reduce streamwise velocity fluctuations by up to 7% in a turbulent boundary layer at a friction Reynolds number of 2600. The active deformable surface has a diameter roughly equal to the boundary layer thickness and is driven by real-time velocity measurements from a particle image velocimetry (PIV) system at 400 Hz. The previous results also revealed that surface deformations of less than 4% of the boundary layer thickness suppress very-large-scale motions (VLSMs) with little effect on the small scales, whereas larger deformations achieve greater VLSM attenuation but intensify the small scales. To further suppress VLSMs, we now investigate data-driven strategies. One approach derives a transfer function for actuation by identifying upstream fluctuation and surface deformation pairs that maximize the reduction in downstream streamwise velocity fluctuations. In addition, linear Wiener filters and temporal convolutional networks are trained using PIV data to compute actuator signals that suppress downstream fluctuations. These evaluations will determine whether additional suppression of VLSMs can be achieved using data-driven methods compared to the previously implemented feedforward algorithm.