Artificial neural network-based modeling and intelligent control of transitional flows

Abstract

Empirical eigenfunctions of transitional flow in a grooved channel are extracted by proper orthogonal decomposition (POD). POD is applied to numerical solutions of the governing Navier-Stokes partial differential equations at Reynolds numbers Re = 430, 750, 1050 and at Prandtl number Pr = 0.71 (air flow). For each value of Re, a low-dimensional set of nonlinear ordinary differential equations is derived by Galerkin projection. The Galerkin projection-based low-order dynamical models are used to generate the data required to efficiently train artificial neural networks in the range 400 less than or equal to Re less than or equal to 1200. Accurate artificial neural network-based models of the flow system are obtained. The study demonstrates the potential use of Galerkin projection-based and artificial neural network-based low-order models as valuable tools for flow modeling and for prediction of short-and long-time behavior of transitional flow systems. A possible real-time intelligent flow control scheme is briefly discussed.