Deep reinforcement learning for industrial applications: taking flow control to the real world

In this project, we will bring DRL for AFC to the real world by implementing the first experimental demonstration of this technique.

Active flow control (AFC) bears many promises, including fuel and energy savings, optimization of industrial processes, and smarter control of complex non-linear systems. However, it is a famously difficult problem due to the combination of non-linearity, time dependence, and high dimensionality implied by the Navier Stokes equations.

Despite these challenges, recent progresses have been achieved by leveraging techniques from the Machine Learning / Artificial Intelligence (ML/AI) fields. In particular, Deep Reinforcement Learning (DRL) has proven efficient at performing AFC on several benchmarks in simulations.

In the present project, we will bring Deep Reinforcement Learning (DRL) for Active Flow Control (AFC) to the real world by implementing the first experimental demonstration of this technique. For this, we will start by reproducing in laboratory experiments the results recently obtained in simulations. This will include developing a novel, flexible Open Source framework for real-world DRL applications. Then, we will consider DRL applications to real-world industrial problems, such as wave and instability damping, and perform model ship stabilization in wave tank experiments.

This project is at the interface between AI/ML, Fluid Mechanics, and non-linear control. The host institution is world-leading in DRL applications to Fluid Mechanics, and has longstanding experience in many advanced experimental techniques.

Requirements

  • Candidates must hold a MSc in Fluid Mechanics, Computational Science, or Physics, preferably with some experience in Machine Learning in addition to key qualifications in fluid mechanics / numerical simulations.
  • Good abilities in Python and C++ programming is a requirement.

Supervisors

Professor Atle Jensen

Researcher Jean Rabault

Call 1: Project start autumn 2021

This project is in call 1, starting autumn 2021. Read about how to apply

Published Aug. 20, 2020 1:33 PM - Last modified Nov. 17, 2020 3:13 PM