[Paper Review] Frequency lock closed-loop control of a separated flow using visual feedback

N. Gautier, Jean‐Luc Aider|arXiv (Cornell University)|Nov 4, 2013

Fluid Dynamics and Turbulent Flows6 references3 citations

TL;DR

This study presents a real-time, model-based closed-loop control system that uses visual feedback from PIV-like velocity fields to lock actuation to the natural Kelvin-Helmholtz frequency of a separated flow behind a backward-facing step. By continuously monitoring flow state via optical velocity measurements and adjusting actuation frequency in response to changes, the method reduces recirculation area by 70–85% across Reynolds numbers from 1400 to 2800, demonstrating robustness and adaptability without prior flow modeling.

ABSTRACT

In this study, a simple model based closed-loop algorithm is used to control the separated flow downstream a backward-facing step. It has been shown in previous studies that the recirculation bubble can be minimized when exciting the shear layer at its natural Kelvin-Helmholtz instability frequency. In this experiment, the natural shedding frequency is identified through real-time analysis of 2D velocity fields. Actuation (pulsed jet) is then locked on this frequency. If flow characteristics stray too far from a set point, shedding frequency is updated and actuation changed. The present work demonstrates the efficacy and robustness of this approach in reducing recirculation while Reynolds number is randomly varied between 1400 and 2800.

Motivation & Objective

• To develop a robust, adaptive closed-loop control strategy for separating flows that dynamically tracks the natural shedding frequency.
• To enable real-time adaptation to changing flow conditions, such as Reynolds number variations, without relying on pre-calibrated models.
• To validate the effectiveness of optical flow-based feedback for identifying and locking onto the Kelvin-Helmholtz instability frequency in a separated shear layer.
• To demonstrate that simple model-based control using visual feedback can achieve significant recirculation reduction in a practical experimental setup.

Proposed method

• A real-time optical flow algorithm (FOLKI-GPU) computes instantaneous 2D velocity fields at up to 100 Hz from PIV-like particle images.
• A monitoring variable, $ u_{\text{check}}(t) $, is defined as the spatially averaged longitudinal velocity in the upper corner of the flow field to detect flow stability.
• The shedding frequency $ f_{KH} $ is computed during periods of stable $ u_{\text{check}} $, using a defined computation window $ \Delta T_{\text{computation}} $.
• Actuation is applied via a pulsed jet at the identified $ f_{KH} $, locking the actuation frequency to the natural instability frequency.
• A feedback loop continuously evaluates $ \Delta u_{\text{check}}(t) $ using equation (3) to detect significant flow changes and trigger re-computation of $ f_{KH} $.
• The threshold for $ \Delta u_{\text{check}} $ is set to distinguish noise from meaningful flow state changes, ensuring adaptive responsiveness.

Experimental results

Research questions

• RQ1 Can real-time visual feedback from instantaneous velocity fields reliably identify the natural Kelvin-Helmholtz shedding frequency in a separated shear layer?
• RQ2 How effectively does frequency-locking actuation reduce recirculation area under varying Reynolds numbers?
• RQ3 Can a simple model-based closed-loop control system maintain performance when flow conditions such as Reynolds number are randomly varied?
• RQ4 What is the trade-off between control responsiveness and reliability when tuning the stability and computation time windows?

Key findings

• The control system successfully reduced the recirculation area by 70–85% compared to uncontrolled conditions across Reynolds numbers from 1400 to 2800.
• The method demonstrated robustness by maintaining effective control despite random variations in Reynolds number, with actuation frequency dynamically adjusted in real time.
• Re-computation of $ f_{KH} $ occurred only when significant changes in $ u_{\text{check}} $ were detected, ensuring minimal unnecessary recalibration.
• The recirculation area normalized by $ h^2 $ showed consistent reduction during each controlled phase, with mean values clearly below uncontrolled levels.
• The system achieved reliable frequency tracking and control using only a few tunable parameters, based on time-scale considerations of the flow process.
• The approach proved effective without requiring prior knowledge of the flow model, relying solely on real-time optical velocity measurements.