Objective: Assess the feasibility of closed-loop active control for turbulent flows using actuation manifolds.

Project Summary

The importance of turbulent flows in physical and engineering systems has motivated countless studies attempting to control them. Closed-loop turbulence control has the potential to be a key enabler for efficiency improvement and reduction of the environmental footprint of transportation means, energy production/distribution, and of a wide variety of industrial processes. Nevertheless, the chaotic, multi-scale nature of turbulence represents a challenge for any real-time sensing strategy which requires a low-rank model plant to select efficiently the most suitable control action. Dimensionality reduction becomes here a key enabler, even when targeting the application of machine-learning methods, which are already obtaining promising results in numerical simulations. While for uncontrolled flows it is often possible to identify low-order surfaces, called manifolds, which embed the flow dynamics, this is still a relatively unexplored pathway for flow-control cases.

The main objective of EXCALIBUR is to assess the feasibility of closed-loop control for turbulent flows, leveraging the search for actuation manifolds to ease the identification of optimal control laws. EXCALIBUR is a continuation of previous projects that have studied the role of coherent structures in turbulent convective heat transfer and the identification of interpretable and scalable strategies for active flow control in wall-bounded and jet flows. The current project aims to build on this work by addressing the following research questions:

1) Is it possible to build an actuation manifold of a fully turbulent flow in the presence of control actuation?

2) Is it possible to identify the state of the flow in the actuation manifold using a limited number of sensors?

3) Is it possible to design suitable manifold-based control architectures for wall-bounded and free-shear turbulent flows?

4) Is it possible to design efficient high-control-authority actuators for boundary layers and jets?

The project will focus on using machine learning techniques, such as genetic programming or deep reinforcement learning, to find optimal or sub-optimal control strategies. It will also aim to make the control laws as simple and interpretable as possible to provide physical insights from laboratory experiments.

The project is expected to produce novel fundamental knowledge in system identification for chaotic turbulent systems and in sensor/actuator theory and design. The control strategies developed throughout the project will also pave the way toward a novel understanding of the flow physics of jet and wall-bounded turbulence, with potential applications in aviation, wind energy, and other industrial sectors. EXCALIBUR will have in the medium/long term socio-economic impact in the areas of climate-neutral and environment-friendly mobility, and clean and sustainable transition of the energy and transport sectors towards climate neutrality.

EXCALIBUR in a nutshell

Title: Extraction of machine learning strategies for turbulent flow control

Goal: Assess the feasibility of closed-loop active control for turbulent flows using actuation manifolds.

Duration: 36 months (01/09/2023 – 31//08/2026)

Budget: 289.250,00 €

Grant No PID2022-138314NB-I00), funded by MCIU/AEI/ 10.13039/501100011033 and by“ERDF A way of making Europe

Team

Results

Journal Papers

L. Marra, G.Y. Cornejo Maceda, A. Meilán-Vila, V. Guerrero, S. Rashwan, B.R. Noack, S. Discetti, A. Ianiro, “Actuation manifold from snapshot data”, Journal of Fluid Mechanics, 996:A26, 2024

Conference Contributions

 A.Rodríguez, S. Discetti, A. Ianiro, “Comparative analysis of manifold learning techniques for controlled flows”. 77th Annual Meeting of the APS Division of Fluid Dynamics, Salt Lake City, November 24- 26, 2024.

 L. Marra, G.Y. Cornejo Maceda, A. Meilán-Vila, V. Guerrero, S. Rashwan, B.R. Noack, S. Discetti, A. Ianiro, “Actuation manifold from snapshot data”. 77th Annual Meeting of the APS Division of Fluid Dynamics, Salt Lake City, November 24-26. 2024.

 A. Rodríguez, S. Discetti, A. Ianiro, “Application of manifold learning techniques to several actuated flow configurations”. 2nd International Conference on Mathematical Modelling in Mechanics and Engineering, Belgrade, September 12-14, 2024.

MSc and PhD Theses

 A. Cuéllar, “AI-based sensing of turbulent wall-bounded flows”. (December 2024)

 A. Solera, “title” (expected defense xxx)

 M.A. Gómez, “title” (expected defense xxx)

 A. Rodríguez “Identification of actuation manifolds in wall-bounded turbulent flows with active contro” (expected defense xxx)

 L. Marra, “title” (expected defense xxx)

Outreach Activities

 A. Cuéllar, “Aprende con una cámara térmica”, Madrid Science and Innovation Week, Madrid, November 7, 2024.

 A.Rodríguez, A. Cuellar, C. Cobos, “¿En qué se parece un globo a un motor de avión? Descubre la magia de un aerorreactor”, Madrid Science and Innovation Week, Madrid, November 14, 2024.

A. Rodriguez, P. Moreno, A. Ianiro, S. Discetti “Descubriendo patrones desconocidos en la turbulencia: secretos detrás de los datos”, European Researchers Night`, Madrid, September 27, 2024.

A. Cuéllar “Experimenta con cámaras térmicas utilizadas en investigación aeronáutica: ¡ven a una escuela de calor!”, European Researchers Night`, Madrid, September 27, 2024.

Open Science (datasets and codes)

L. Marra, G.Y. Cornejo Maceda, A. Meilán-Vila, V. Guerrero, S. Rashwan, B.R. Noack, S. Discetti, A. Ianiro, “Actuation manifold from snapshot data” (2024)

Data Set: https://zenodo.org/records/12802192

Github Codes: https://github.com/Lmarra1/Actuation-manifold-from-snapshot-data

Photo Gallery