Session: SYMP S-1: Integrated Systems

Paper Number: 141256

141256 - Cavity Flow Dynamics as a Source of Information Processing 

As advances in flow sensing and active flow control technology are enabling increased sensor density and more effective flow inputs, novel computing concepts are needed to close the loop on self-contained local flow control systems while minimizing size, weight, and power requirements. Recent advances in sensor technology have opened the door to applying distributed sensing to bodies in complex flow environments. However, the increased computational burden of processing the resulting information is not trivial. Using the mathematical framework of reservoir computing, it is possible to offload some of this burden to the dynamics of the loading environment. The local flow environment may be treated as a non-linear operator that maps an input signal, a perturbation of the mean external flow, into a high dimensional latent space. A complex non-linear output function, such as a control signal or a classification designation, may then be computed by strategically sampling this space with a distributed sensor array and combining the readouts using a simple weighted sum. In this work, we demonstrate the concept using computational simulations and experimental validation of flow in an open-cavity. We find that for an incoming laminar incompressible boundary layer flow, placing sensors inside shear-driven open cavity flow performs much better on benchmark tasks requiring memory than placing many sensors directly in the boundary layer flow. This effect results from increased signal diversity in the cavity dynamics, but is attenuated by instability and loss of synchrony at higher Reynolds numbers. Experimental validation of the computational results was pursued using a water tunnel study with custom-designed, upstream disturbance generators.  We employ flow in a water tunnel disturbed by a frequency modulated rotating cylinder with a flap to deliver a sufficiently realistic and complex disturbance to the flow inside a cavity embedded in a plate. We find that the vortex gusts generated by a rotating cylinder with a flap can provide a sufficiently complex signal for physical reservoir computing within the parameter ranges investigated computationally. We also find that the experimental apparatus is able to collect flow data from the baseline unperturbed flow which is consistent with previous 3D Direct Numerical Simulation predictions. We further investigated the role of readout location, particularly between surface and bulk flow locations, and the role of experimental readout noise on the success of reservoir training. Leveraging flow-based information processing could reduce the need for electronic computing hardware, potentially reducing cost, size, weight, and power and shortening control cycle times.

Presenting Author: Alexander Pankonien Air Force Research Laboratory

Presenting Author Biography: Dr. Alexander Pankonien is a research aerospace engineer for the Air Vehicles Division of the Air Force Research laboratory in Dayton Ohio. He works in the conceptual vehicle design group and focuses on the physical realization of novel concepts for adaptive structures. In pursuit of this goal, he has contributed to several basic science and applied research programs including the development of the artificial hair sensor and novel mechanical computing concepts. He holds a Doctorate in Aerospace Engineering from the University of Michigan, and a Masters of Science and Bachelors of Science from Texas A&M University.

Authors:

Alexander Pankonien