Closed-loop turbulence control: From human to machine learning (and retour)
Résumé
Closed-loop turbulence control is a rapidly evolving, interdisciplinary field of research. The
range of current and future engineering applications has truly epic proportions, including cars,
trains, airplanes, noise, air conditioning, medical applications, wind turbines, combustors, and
energy systems. A key feature, opportunity and technical challenge of closed-loop turbulence
control is the inherent nonlinearity of the actuation response. For instance, excitation at a given
frequency will affect also other frequencies. This frequency crosstalk is not accessible in any
linear control framework. This talk will address these nonlinear actuation mechanisms. First,
success stories of âhuman learningâ in turbulence control are presented, i.e. cases in which the
nonlinear actuation mechanism has been modelled and understood. A large class of literature
studies can be categorized in terms of surprisingly few mechanisms. Second, we discuss model-
free machine learning control (MLC) and selected applications. MLC detects and exploits the winning actuation mechanisms in the experiment in an unsupervised manner. Recent examples
include the stabilization of the fluidic pinball, drag reduction of a car model at high Reynolds
numbers and mixing optimization of a turbulent jet. In all studies, MLC has reproduced or
outperformed existing optimized control strategies, as will also be elaborated in other talks at
this Minisymposium. Finally, future directions of turbulence control are outlined. Methods of
machine learning are a disruptive technology will contribute to rapidly accelerating progress in
turbulence controlâboth for actuation performance and for physical understanding.