Control of the Fluidic Pinball using the Quadratic-Quadratic Regulator

This study demonstrates that a model-based control framework combining interpolatory model order reduction with a quadratic-quadratic regulator (QQR) effectively stabilizes the fluidic pinball's unstable wake at Reynolds numbers of 30 and 50, outperforming traditional linear controllers by achieving faster convergence and successfully suppressing vortex shedding where linear methods fail.

The Gist

Imagine you are trying to balance a spinning top on a wobbly table. If the table shakes too much, the top falls over. Now, imagine that instead of a top, you have a complex dance of water swirling around three cylinders (like balls) arranged in a triangle. This is the "Fluidic Pinball."

The water naturally wants to swirl chaotically around these balls, creating a messy wake (the trail of water behind them). The goal of this paper is to teach the water how to stop dancing and sit still in a calm, steady state, even when it wants to be chaotic.

Here is how the researchers did it, explained simply:

1. The Problem: Too Much Math for a Computer

The water follows rules called the "Navier-Stokes equations." These are like a massive, complicated instruction manual for how fluids move. To simulate this on a computer, you have to break the water down into millions of tiny puzzle pieces. Trying to control the water using all those pieces at once is like trying to steer a ship by controlling every single drop of water in the ocean—it takes too long and is too hard for computers to handle in real-time.

2. The Solution: A "Cheat Sheet" (Model Reduction)

To make the math manageable, the authors created a "cheat sheet" called a Reduced-Order Model (ROM).

• The Analogy: Imagine you are trying to predict the weather. Instead of tracking every single molecule of air, you just track the big patterns (like high and low-pressure systems).
• The Method: They used a technique called IMOR (Interpolatory Model Order Reduction). Think of this as taking a few very smart snapshots of how the water usually behaves and how it reacts when you push it. They used these snapshots to build a tiny, simplified version of the water flow that acts exactly like the big, complicated version but is much faster to calculate.

3. The Controller: The "Smart Driver"

Once they had their simplified model, they needed a way to steer the water. They tested two types of "drivers":

• Driver A (Linear Controller): This driver is like a new student driver. They only understand straight lines and simple turns. If the water starts to swirl in a simple way, this driver can fix it. But if the water gets really wild and starts doing complex loops (nonlinear behavior), this driver gets confused and fails.
• Driver B (QQR - Quadratic-Quadratic Regulator): This driver is an expert race car driver. They understand that the water doesn't just move in straight lines; it curves, spins, and interacts with itself in complex ways. This driver uses a "quadratic" strategy, meaning they can predict and correct for those complex, curvy movements.

4. The Race: Testing at Two Speeds

The researchers tested both drivers at two different speeds of water flow (Reynolds numbers 30 and 50).

• At the slower speed (Re = 30): Both drivers could eventually calm the water down. However, the QQR driver was much faster. It got the water to a steady state 40% faster than the linear driver and used less energy to do it. It was like the expert driver taking the perfect racing line while the student driver took the long way around.
• At the faster speed (Re = 50): This is where the difference became huge. The water was swirling so wildly that the Linear Driver completely failed. It couldn't handle the complexity and the water kept spinning out of control. The QQR Driver, however, successfully tamed the chaos and brought the water to a calm, steady state.

5. The Result: A Calmer Wake

When the QQR driver was in charge, two good things happened:

1. No more shaking: The water stopped creating "vortex shedding" (those rhythmic swirls that make things shake). This is like stopping a bridge from swaying in the wind.
2. Less drag: The water flowed more smoothly past the cylinders, reducing the resistance (drag). This is like a car becoming more fuel-efficient because the air flows better over it.

Summary

The paper shows that for complex fluid problems, a "smart" controller that understands the complex, curvy nature of the flow (QQR) is much better than a "simple" controller that only looks at straight lines. By using a smart "cheat sheet" (the reduced model) to run the calculations quickly, they were able to stabilize a chaotic water flow that a simpler method couldn't handle at all.

Technical Summary

Technical Summary: Control of the Fluidic Pinball using the Quadratic-Quadratic Regulator

Problem Statement
The paper addresses the challenge of stabilizing the "fluidic pinball" flow, a benchmark problem involving the complex wake interactions of three cylinders arranged in an equilateral triangle. The objective is to stabilize the flow to its unstable steady-state solution using cylinder rotation as the actuation mechanism. The governing physics are described by the incompressible Navier-Stokes equations (NSE). While linear feedback control is a standard approach, the authors note that for flows with significant nonlinearity—particularly when stabilizing solutions far from the linearization point—linear controllers often fail. The specific challenge lies in the high dimensionality of the full-order model (FOM) derived from finite element discretization, which makes direct control design computationally prohibitive, and the inherent quadratic nonlinearity of the NSE convective term.

Methodology
The authors propose a model-based nonlinear feedback control framework that integrates two primary components:

1. Interpolatory Model Order Reduction (IMOR): To overcome the computational cost of the full-order system, a Reduced-Order Model (ROM) is constructed. Unlike snapshot-based methods like Proper Orthogonal Decomposition (POD), which require expensive controlled simulations for training data, the authors utilize IMOR. This method constructs a reduced basis by enforcing tangential Hermite interpolation conditions on the system's transfer function. The ROM preserves the input-output dynamics of the high-fidelity FEM model and explicitly accounts for the control inputs and observed outputs. The resulting ROM retains the quadratic structure of the NSE, taking the form of a differential-algebraic equation with a quadratic nonlinearity ($N(\mathbf{x}_r \otimes \mathbf{x}_r)$).

2. Quadratic-Quadratic Regulator (QQR): A nonlinear control strategy is designed specifically for the quadratic ROM. The control problem is formulated as minimizing a quadratic cost function (state error and control effort) subject to quadratic state dynamics. The optimal control law is approximated as a polynomial feedback function, specifically truncated to second order: $\mathbf{u} = \mathbf{k}_1 \mathbf{x}_r + \mathbf{k}_2 (\mathbf{x}_r \otimes \mathbf{x}_r)$. The coefficients are derived by solving a sequence of linear systems involving Kronecker sums, extending the standard Linear Quadratic Regulator (LQR) solution to include quadratic terms.

The control gains derived from the ROM are "lifted" back to the full-order state space to evaluate performance. The sensors are defined as spatial averages of velocity perturbations in 12 regions within the wake (specifically the vortex formation length), and the actuators are the tangential velocities of the three cylinders.

Key Contributions

• Framework Integration: The paper presents a unified framework combining IMOR with QQR, specifically tailored to exploit the quadratic nonlinearity of the Navier-Stokes equations for flow control.
• Nonlinear Control Design: It demonstrates the derivation and application of a second-order polynomial feedback controller (QQR) for a fluid dynamics problem, moving beyond linear approximations.
• Lifting Strategy: The methodology includes a rigorous procedure for lifting the reduced-order control law back to the full-order finite element state space for performance evaluation.

Results
The performance of the QQR controller was evaluated against a traditional linear feedback controller (LQR) at two Reynolds numbers ($Re_D = 30$ and $Re_D = 50$) using the full-order model for simulation.

• At $Re_D = 30$: Both controllers successfully stabilized the wake. However, the QQR controller achieved the desired performance criteria (stabilizing the wake to within a relative error of 2%) approximately 40.1% faster than the linear controller. Furthermore, the total running cost (combining state error and control effort) for the QQR was 13.5% lower than that of the linear controller. Both controllers reduced the drag coefficient by approximately 2.9% and suppressed lift oscillations.
• At $Re_D = 50$: A significant divergence in performance was observed. The linear controller failed to stabilize the flow to the desired steady state, instead driving the system into a different, undesired limit cycle. In contrast, the QQR controller successfully stabilized the wake to the target steady state. The QQR controller reduced drag by roughly 3.6% and suppressed lift oscillations from a magnitude of 0.073 to 0.008.

Significance and Claims
The paper claims that the IMOR-QQR framework provides an effective model-based control strategy for managing nonlinear hydrodynamic instabilities in complex wake flows. The primary significance lies in the demonstration that linear control strategies are insufficient for higher Reynolds numbers where nonlinear effects dominate, whereas the quadratic regulator can successfully stabilize the system. The results suggest that incorporating the quadratic nonlinearity directly into the control design allows for faster convergence and lower control costs compared to linear approaches, and is essential for stabilization in regimes where linearization fails. The authors conclude that the fluidic pinball serves as a robust testbed for such techniques and that future work should focus on sensitivity analysis regarding sensor placement and the exploration of output feedback control.