Aerodynamic Physical Reservoir Computer Training Using Oscillating Vortex Generators

The ability to quickly sense and respond is paramount for numerous control systems and engineering applications. Physical reservoir computing (PRC) aims to advance this process by combining sensing physical inputs and information processing into a nonlinear structure to reduce time delays caused by data transmission. In this work, we develop training methods based on wind tunnel experimental data for PRC structures that allow for the ability to influence flow phenomena to cover a wide regime of flow conditions at the PRC surface for PRC training. To accomplish this, a piezoelectric vortex generator (VG) structure is employed to influence flow downstream. The VG fins are actuated in the direction normal to the aerodynamic surface by a piezoelectric actuator, allowing for the height of the VG fins to be controlled. To analyze how actuation of the VG height propagates to the flow phenomena, an information theoretic measure using Rényi entropy is used to compute multifractal spectra. This provides a measure of multiscale complexities in spatial or temporal data which is important to understanding information processing in PRCs. The shape and width of the multifractal spectrum may vary depending on these multiscale complexities within the dataset. The relationships between the multifractal spectrum of the VG piezoelectric actuator input and the flow phenomena at the aerodynamic surface of the PRC input are analyzed to provide a deeper understanding to design PRC structures on an aerodynamic surface such that the information transfer from input flow dynamics to the PRC input is maximized. The piezoelectric actuated VG also allows for a diverse training regime with a range of different flow conditions, and thus improving PRC training and potentially expanding the sensing capability to a wider set of flow conditions.