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

Abstract

We stabilize the flow past a cluster of three rotating cylinders, the fluidic pinball, with automated gradient-enriched machine learning algorithms. The control laws command the rotation speed of each cylinder in an open- and closed-loop manner. These laws are optimized with respect to the average distance from the target steady solution in three successively richer search spaces. First, stabilization is pursued with steady symmetric forcing. Second, we allow for asymmetric steady forcing. And third, we determine an optimal feedback controller employing nine velocity probes downstream. As expected, the control performance increases with every generalization of the search space. Surprisingly, both open- and closed-loop optimal controllers include an asymmetric forcing, which surpasses symmetric forcing. Intriguingly, the best performance is achieved by a combination of phasor control and asymmetric steady forcing. We hypothesize that asymmetric forcing is typical for pitchfork bifurcated dynamics of nominally symmetric configurations. Key enablers are automated machine learning algorithms augmented with gradient search: explorative gradient method for the open-loop parameter optimization and a gradient-enriched machine learning control (gMLC) for the feedback optimization. gMLC learns the control law significantly faster than previously employed genetic programming control. The gMLC source code is freely available online.

45 pages, 23 figures

References in corpus (3)

• Closed-loop separation control using machine learning
• Robust active flow control over a range of Reynolds numbers using an artificial neural network trained through deep reinforcement learning
• Artificial intelligence control of a turbulent jet

Cited by in corpus (10)

• Comparative analysis of machine learning methods for active flow control
• Effective control of two-dimensional Rayleigh--Bénard convection: invariant multi-agent reinforcement learning is all you need
• Machine learning flow control with few sensor feedback and measurement noise
• From snapshots to manifolds - A tale of shear flows
• Stabilization of a multi-frequency open cavity flow with gradient-enriched machine learning control
• Machine-learned control-oriented flow estimation for multiactuator multi-sensor systems exemplified for the fluidic pinball
• Genetically-inspired convective heat transfer enhancement in a turbulent boundary layer
• Actuation manifold from snapshot data
• How to Control Hydrodynamic Force on Fluidic Pinball via Deep Reinforcement Learning
• Self-tuning model predictive control for wake flows