Skip to main content Start main content

ME Seminar - Reinforcement-learning-based control of convectively-unstable flows

Event and Seminar

  • Date

    09 Dec 2022

  • Organiser

    Department of Mechanical Engineering, PolyU

  • Time

    16:00 - 17:00

  • Venue

    BC302, PolyU Campus Map  


e-Certificate of attendance will be provided. Latecomer or early leaver of the seminar might NOT be eligible for an attendance certificate.

Guest Speaker: Dr Zhang Mengqi

Assistant Professor, Department of Mechanical Engineering, National University of Singapore

Dr Zhang Mengqi is an Assistant Professor in the Department of Mechanical Engineering, National University of Singapore since 2018. He received his PhD in Université de Poitiers, France in 2016. During 2016-2018, he was a postdoctoral researcher in the University of Twente. His research interest includes numerical analysis and simulations of various flow problems, such as non-Newtonian flows, electrohydrodynamic flows and the flow in wind farms. He is also interested in flow control and reduced-order modelling of complex flows, including machine-learning-aided flow manipulation.

Reinforcement-learning-based control of convectively-unstable flows


Flow control is an important interdisciplinary topic in fluid mechanics. It deals with optimization problems in flows, such as maximally reducing the resistance in a flow or extremizing the mixing efficiency. This talk reports the application of a machine-learning method to study flow-control problems in convectively unstable flows. More specifically, we apply the deep reinforcement learning (DRL) method to suppress perturbations in the 1-D linearized Kuramoto-Sivashinsky (KS) equation and in the 2-D boundary layer flows over a flat plate. We found that the control mechanism identified by the DRL agent in the boundary layer flow is the classical control method, i.e., the opposition control. With the gradient-free Particle Swarm Optimization algorithm, we also determined the best strategy for the sensor placement in the DRL framework. This work demonstrates the ability of the reinforcement-learning method to control the convectively unstable flows.

Your browser is not the latest version. If you continue to browse our website, Some pages may not function properly.

You are recommended to upgrade to a newer version or switch to a different browser. A list of the web browsers that we support can be found here