The Project

AEROMATIC (Control activo de flujos aerodinámicos con aprendizaje automático – Active flow control of aerodynamic flows with machine learning) is a project funded by “Becas Leonardo a Investigadores y Creadores Culturales” of Fundación BBVA, with funding of 40.000 € and total duration of 18 months. The main objective of the project is the development of active flow control strategies for aerodynamics based on machine learning, and adapting such techniques to experimental scenarios.

Motivation

Transport is responsible for more than one quarter of the total CO2 emissions produced by man-made activities. The common denominator is that all means of transportation move in fluids, thus they have to pay the price of aerodynamic drag. For instance, friction drag account for approximately 30−50 % of the total drag of air transport. 

The means of transportation travel in turbulent flows, characterized by a chaotic unsteady behaviour, difficult to predict beforehand. Shape modification (passive control), such as mounting turbulence promoters on the surface, are simple and effective solution, but produce limited improvement, normally introduce an intrinsic loss and cannot adapt to the unsteady fluctuations of turbulent flows. Active flow control, although more complex, can be actuated on-demand; on the other hand, estabilishing control laws is difficult due to the complex dynamic of turbulent flows. Recent developments in the field of control of dynamical systems using machine-learning methods is now paving the way to efficient computation of control laws on the basis only of measured data, thus disclosing new horizons for closed-loop flow control.

Objectives and roadmap

The main objective of AEROMATIC is to investigate methods to determine control laws to reduce aerodynamic drag using machine learning. The main focus is to establish techniques that are sufficiently robust to be applied in experimental conditions, i.e. with limited data affected by measurement noise.

The workplan develops on 3 lines:

• Investigation of machine-learning strategies to establish control laws to reduce drag based on low-cost simulations
• Study of effects of realistic constraints in experimental conditions. , such as for instancec limited number of sensors and effect of measurement noise.
• Demonstration of active flow control in wind-tunnel experiments.