Stabilization of the fluidic pinball with gradient-enriched machine learning control

Guy Y Cornejo Maceda,Yiqing Li,Marek Morzyński,François Lusseyran,Bernd R Noack

doi:10.1017/jfm.2021.301

Abstract

Highlights

We stabilize the wake behind a fluidic pinball using a hierarchy of model-free self-learning control methods from a one-parametric study of open-loop control to a gradient-enriched machine learning feedback control
Traffic alone profits from flow control via drag reduction of transport vehicles (Choi, Jeon & Kim 2008), lift increase of wings (Semaan et al 2016), mixing control for more efficient combustion (Dowling & Morgans 2005) and noise reduction (Jordan & Colonius 2013)
This study focuses on the stabilization of the unstable symmetric steady solution of the fluidic pinball in the pitchfork regime, i.e. for asymmetric vortex shedding

Summary

Introduction

We stabilize the wake behind a fluidic pinball using a hierarchy of model-free self-learning control methods from a one-parametric study of open-loop control to a gradient-enriched machine learning feedback control. Control theory offers powerful methods for control design with large success for model-based stabilization of low-Reynolds-number flows or simple firstand second-order dynamics (Rowley & Williams 2006). Transport-related engineering applications are at high Reynolds numbers and, associated with turbulent flows. Examples relate to first- and second-order dynamics, e.g. the quasi-steady response to quasi-steady actuation (Pfeiffer & King 2012), opposition control near walls (Choi, Moin & Kim 1994; Fukagata & Nobuhide 2003), stabilizing phasor control of oscillations (Pastoor et al 2008) and two-frequency crosstalk (Glezer, Amitay & Honohan 2005; Luchtenburg et al 2009). Control design is challenged by the high dimensionality of the dynamics, the nonlinearity with many frequency crosstalk mechanisms and the large time delay between actuation and sensing

Objectives

Findings

Methods

Conclusion