Gradient-based machine learning control for the stabilization of the fluidic pinball and the open cavity experiment - Archive ouverte HAL Access content directly
Conference Papers Year :

Gradient-based machine learning control for the stabilization of the fluidic pinball and the open cavity experiment

Abstract

We optimized control laws for stabilization of two shear flows: a DNS of a flow past a cluster of three rotating cylinders-the fluidic pinball-and the cavity flow experiment. The fluidic pinball is stabilized in increasingly reacher control law spaces employing 9 sensors downstream and 3 actuators (the cylinders). As for the cavity, two regimes are controlled in single-input single-output manner: a single-mode driven regime and a mode-switching regime. Key enablers are automated machine learning algorithms augmented with gradient search: explorative gradient method for the open-loop parameter optimization and a gradientenriched machine learning control (gMLC, [1]) for the feedback optimization. For both plants, the need of feedback for control is demonstrated. gMLC learns the control law significantly faster than previously employed genetic programming control both in DNS and experiment.
Fichier principal
Vignette du fichier
IUTAM2022_CornejoMaceda.pdf (455.94 Ko) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03895510 , version 1 (12-12-2022)

Identifiers

  • HAL Id : hal-03895510 , version 1

Cite

Guy Y. Cornejo Maceda, Yiqing Li, François Lusseyran, Marek Morzyński, Bernd R. Noack. Gradient-based machine learning control for the stabilization of the fluidic pinball and the open cavity experiment. IUTAM Symposium on Data-driven modeling and optimization in fluid mechanics, Jun 2022, Aarhus, Denmark. ⟨hal-03895510⟩
4 View
0 Download

Share

Gmail Facebook Twitter LinkedIn More