Artificial intelligence control applied to drag reduction of the fluidic pinball

Created by W.Langdon from gp-bibliography.bib Revision:1.8010

@InProceedings{Maceda:2018:jGDR,

• author = "Guy Yoslan {Cornejo Maceda} and Bernd R. Noack and Francois Lusseyran and Marek Morzynski and Luc Pastur and Nan Deng",
• title = "Artificial intelligence control applied to drag reduction of the fluidic pinball",
• booktitle = "Journees du GDR Controle Des Decollements",
• year = "2018",
• address = "Toulouse, France",
• publisher = "HAL CCSD",
• month = nov # "~01",
• keywords = "genetic algorithms, genetic programming, AI, artificial intelligence control, fluidic pinball, control, machine learning, physics, mechanics of the fluids",
• type = "info:eu-repo/semantics/conferenceObject",
• URL = "https://hal.archives-ouvertes.fr/hal-02387548",
• abstract = "Feedback turbulence control is at the core of engineering challenges and have to face high-dimensionality, time-delays, strong nonlinearities and frequency crosstalks, making modelling and linear control theory impractical.The aim of this project is a general, model-free, self-learning control strategy to tame/stabilize nonlinear dynamics and real world turbulence in the plant.The control problem is solved as a regression problem thanks to machine learning control (MLC) (Duriez et al. 2016 Springer).MLC is based on genetic programming (GP), it is a biological inspired method that mimickes the Darwinian process of natural selection to learn the control.Focus of current efforts is to understand the learning process, to reduce this learning time and to include real-world imperfections.Our genetic programming control has been demonstrated on a direct numerical simulation of the fluidic pinball taken as a drag reduction benchmark.It consists of three cylinders in a two-dimensional flow where the actuators are the spinning cylinders and feedback is provided by sensors downstream.Despite the simple configuration, the fluidic pinball shares characteristics with real flows such a bifurcations, nonlinear frequency crosstalk and multiple input multiple output (MIMO) control.We carried out a parameter optimisation study on GP and managed to reduce the learning rate by a factor 5 by avoiding the evaluation of redundant control laws.After 1000 evaluations, GP managed to find a non-trivial solution comprising two distinct actuation mechanisms : boat-tailing (open-loop) and phasor control (closed-loop) reducing even more the net drag power (46percent) than the best boat-tailing configuration (43percent).",
• annote = "Laboratoire d'Informatique pour la Mecanique et les Sciences de l'Ingenieur (LIMSI) ; Universite Paris Saclay (COmUE)-Centre National de la Recherche Scientifique (CNRS)-Sorbonne Universite - UFR d'Ingenierie (UFR 919) ; Sorbonne Universite (SU)-Sorbonne Universite (SU)-Universite Paris-Saclay-Universite Paris-Sud - Paris 11 (UP11)",
• bibsource = "OAI-PMH server at api.archives-ouvertes.fr",
• contributor = "Laboratoire d'Informatique pour la Mecanique et les Sciences de l'Ingenieur",
• coverage = "Toulouse, France",
• identifier = "hal-02387548",
• language = "en",
• oai = "oai:HAL:hal-02387548v1",

}

Genetic Programming entries for Guy Yoslan Cornejo Maceda Bernd R Noack Francois Lusseyran Marek Morzynski Luc Pastur Nan Deng

Citations