Ofertes treballs fi d'estudis

Títol: Deep Reinforcement Learning for Discovering Optimal Active Flow Control Coflow Strategies on an Airfoil

Director/a: Rodriguez Perez, Ivette maria

Email del professor/a: ivette.rodriguez@upc.edu

Departament del professor/a: 724-MMT

Codirector/a: Montalà Sales, Ricard

Email del codirector/a: ricard.montala@upc.edu

Departament del professor/a: 724-MMT

• 2009 - GRAU EN ENGINYERIA MECÀNICA
• 2010 - GRAU EN ENGINYERIA EN TECNOLOGIES AEROESPACIALS
• 2010 - GRAU EN ENGINYERIA EN VEHICLES AEROESPACIALS
• 2013 - MÀSTER UNIVERSITARI EN ENGINYERIA INDUSTRIAL
• 2014 - MÀSTER UNIVERSITARI EN ENGINYERIA AERONÀUTICA
• 2016 - MÀSTER UNIVERSITARI EN ENGINYERIA ESPACIAL I AERONÀUTICA
• 2024 - MÀSTER UNIVERSITARI EN RECERCA EN ENGINYERIA MECÀNICA

Identificador de l'oferta: 205-06065

Modalitat: Universitat

Possibilitat de beca/finançament: No

Paraules clau:

Descripció:

This thesis aims to explore the use of Deep Reinforcement Learning (DRL) to discover and optimize active flow control (AFC) coflow strategies for improving the aerodynamic performance of an airfoil. Traditional AFC approaches often rely on manual tuning of actuation parameters, which can be time-consuming and limited in scope. By combining DRL with high-fidelity Large Eddy Simulations (LES), this study seeks to enable autonomous learning of effective control policies that can adapt to complex, unsteady flow environments. The work will focus on designing a DRL framework that interacts with the LES solver, defines suitable states, actions, and rewards, and evaluates the aerodynamic improvements achieved through learned actuation strategies. The outcomes are expected to provide new insights into AFC optimization and demonstrate the potential of machine learning in aerodynamic control applications.

Objectius:

1. Develop and validate a coupled LES–DRL framework for airfoil flow simulations with AFC coflow actuation. 2. Define the DRL agent’s control space (actions), observation space (states), and reward functions tailored to aerodynamic objectives (e.g., lift enhancement, drag reduction). 3. Train and evaluate the DRL agent on various flow conditions to identify robust and effective coflow control strategies. 4. Analyze the aerodynamic performance and underlying flow physics of the DRL-discovered solutions and compare them to conventional AFC approaches.

Tasques a realitzar / Característiques:

1. Introduction– Background, motivation, and objectives; combining AFC and DRL for flow control 2. Literature Review– Review of AFC coflow techniques, reinforcement learning in fluid mechanics, and relevant DRL algorithms 3. Methodology– Description of LES framework, DRL setup (agent, environment, reward function), and training procedure 4. Design of DRL-AFC System– Integration of DRL agent with LES, definition of control actions, states, and performance metrics 5. Results and Discussion– Analysis of DRL-discovered actuation strategies, aerodynamic performance, and flow physics 6. Conclusions – Summary of main findings, advantages over traditional AFC approaches 7. Recommendations and Future Work – Suggestions for improving DRL methods and experimental or computational extensions

Tema: ENGINYERIA AERONÀUTICA

Tipus: Estudi

Càrrega de treball:

La que correspongui a la titulacio

Temàtica ambiental: No

Confidencial (informatiu): Si

Ambit de cooperació: No

Publicació: 12/05/2025

Caducitat: 10/05/2026