Reduced-Order Modelling and Closed-Loop Control of the Cylinder Wake

TL;DR

This paper develops a reduced-order, model-based closed-loop control strategy to suppress vortex shedding in cylinder wakes by using a generalized Stuart-Landau model and model predictive control, achieving flow stabilization at Re=50.

Contribution

It introduces a generalized Stuart-Landau model with variable forcing amplitude for closed-loop flow control, extending previous fixed-amplitude approaches.

Findings

Successfully suppresses wake oscillations at Re=50.

Uses a model predictive controller with partial flow measurements.

Achieves flow stabilization with spatially dense forcing.

Abstract

We present a model-based approach for the closed-loop control of vortex shedding in the cylinder wake. The control objective is to suppress the unsteadiness of the flow, which arises at a critical Reynolds number $Re_c$ through a supercritical Hopf bifurcation. In the vicinity of $Re_c$ the flow is well described by a forced Stuart-Landau equation derived via a global weakly nonlinear analysis. This Stuart-Landau equation governs the evolution of the amplitude $A$ of the global mode on the slow time scale. In this paper, we generalize the approach from [Sipp 2012], which considers a fixed-amplitude harmonic forcing, by allowing the forcing amplitude E0 to vary on the slow time scale. This enables the design of closed-loop controllers for multiple surrogate Stuart-Landau models, which we obtain for different classes of forcing frequencies. When these frequencies are near the global mode oscillation frequency at Rec, we can bring both $A$ and $E'$ to zero, which fully suppresses the unsteady part of the flow. We also show that near this frequency, the optimal forcing structure is in the direction of the adjoint global mode. Assuming partial velocity measurements of the flow, we design an output-feedback control law that stabilizes the flow. The approach hinges on a model predictive controller for the surrogate model, which exploits the full-order model measurements to determine the necessary forcing amplitudes while respecting the modelling constraints. We achieve suppression of the wake oscillations with spatially dense volume forcing and two-point velocity measurement at $Re=50$.