Comparteix:

Ofertes treballs fi d'estudis

Quines ofertes cerques?

Loading...

Títol: Advanced Data-Driven Methods for Reduced Order Modeling in Computational Fluid Dynamics

Director/a: Miro Jane, Arnau

Email del professor/a: arnau.miro@upc.edu

Departament del professor/a: 748-FIS

Codirector/a: Rodriguez Perez, Ivette maria

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

Departament del professor/a: 724-MMT

Titulacions:
  • 2014 - MÀSTER UNIVERSITARI EN ENGINYERIA AERONÀUTICA

Identificador de l'oferta: 205-06401

Modalitat: Universitat

Possibilitat de beca/finançament: No

Idiomes d'elaboració del treball:
  • Català
  • Espanyol
  • Anglès

Paraules clau:

PROGRAMACIÓ CFD INTEL.LIGÈNCIA ARTIFICAL

Descripció:

Reduced order models are fundamental in computational fluid mechanics (CFD) in order to understand the underlying physics of the flow. Data-driven methods take advantage of the large existing databases to develop models of the flow. Traditional methods, such as proper orthogonal decomposition (POD) or dynamic mode decomposition (DMD) rely on the decomposition of the database through a singular value decomposition (SVD). This allows to find linear relationships between the flow structures, or modes, and project them through time. These methods, however, find difficulties when representing naturally chaotic and non-linear phenomenon such as turbulence. Recently, new methodologies such as the geometrically agnostic variational autoencoders (GAVI) have proven to be able to overcome these difficulties and provide better representations of turbulent flows.

Objectius:

To assess the possibility of extending the GAVI methodology towards the DMD by means of a Koopman penalization in the loss function.

Tasques a realitzar / Característiques:

1. Literature Review: conduct and familiarize with the literature on reduced order modeling, DMD algorithms and the GAVI methodology. 2. Training: familiarize with the DMD and GAVI methodology with the database of a cylinder at Re=100. 3. GAVI-Koopman: extend the GAVI autoencoder with a Koopman loss function. Validate at the database of a cylinder at Re=100. 4. GAVI-DMD: perform the full DMD algorithm exchanging the SVD by the GAVI-Koopman procedure. Validate at the database of a cylinder at Re=100. 5. Validation: perform the validation of the full algorithm in a turbulent case to be determined (e.g., Windsor body).

Tema: ENGINYERIA AERONÀUTICA

Tipus: Estudi

Càrrega de treball:

300h

Temàtica ambiental: Si

Confidencial (informatiu): Si

Ambit de cooperació: Si

Publicació: 09/12/2025

Caducitat: 02/12/2026