Résumé:
Closed-loop separated flow control on backward facing step
This thesis proposes a novel approach to closed-loop flow control using artificial intelligence
(AI), focusing on the prediction and manipulation of dynamic flow behaviors in fluid
dynamics applications. The research begins with a numerical investigation of flow separation
over a backward-facing step (BFS) using the Detached Eddy Simulation (DES) model,
validated against experimental data to ensure fidelity in capturing critical flow features.
We introduce a new class of neural networks designed to predict dynamic flow behavior
over the (BFS) using wall pressure measurements. These networks achieve high accuracy,
enabling real-time flow control applications. The study further explores the influence of
frequency information and shear layer dynamics on effective flow control strategies. Leveraging
these insights, AI-driven methods are applied to predict and manipulate shear layer
behavior, achieving significant improvements in flow control performance. Correlation
analyses demonstrate the predictability of future and past flow states within a defined
time horizon, offering valuable insights for optimizing both open-loop and closed-loop
control systems. These findings contribute to the development of adaptive, efficient flow
control strategies with broad implications for aerospace, automotive, and energy systems.
