Title :

Reconstruction de l'écoulement autour d'un bâtiment de grande hauteur basée sur des mesures de sillages utilisant les techniques d'apprentissage automatique

Author(s) :

Diop, Moussa [Auteur]
DAAA, ONERA [Lille]
Dubois, Pierre [Auteur]
DAAA, ONERA [Lille]
Toubin, Hélène [Auteur]
DAAA, ONERA [Lille]
Planckaert, Laurent [Auteur]
DAAA, ONERA [Lille]
Laboratoire de Mécanique des Fluides de Lille - Kampé de Fériet [LMFL]
Le Roy, Jean-François [Auteur]
DAAA, ONERA [Lille]
Laboratoire de Mécanique des Fluides de Lille - Kampé de Fériet [LMFL]
Garnier, Eric [Auteur]
DAAA, ONERA [Lille]

Journal title :

Journal of Wind Engineering and Industrial Aerodynamics

Pages :

105149

Publisher :

Elsevier

Publication date :

2022-10-01

ISSN :

0167-6105

Keyword(s) :

RESEAU NEURONE
POD
TURBULENCE ATMOSPHERIQUE

English keyword(s) :

NEURAL NETWORK
POD
ATMOSPHERIC TURBULENCE

HAL domain(s) :

Physique [physics]
Sciences de l'ingénieur [physics]

English abstract : [en]

This paper investigates the unsteady flow around a high-rise building using OpenFoam. A Vortex Method is developed to generate upstream unsteady fluctuations that are validated considering the numerical simulation of a ...
Show more >This paper investigates the unsteady flow around a high-rise building using OpenFoam. A Vortex Method is developed to generate upstream unsteady fluctuations that are validated considering the numerical simulation of a neutral atmospheric boundary layer around a high-rise building. The Vortex Method parameters are tuned to produce an upstream velocity profile whose turbulence characteristics are comparable to experimental data. Comparison between mean and fluctuating velocity profiles of the numerical data measured in the building’s wake with experimental data shows satisfactory results. The resulting database is used to reconstruct the flow from limited velocity measurements inside the wake and static wall pressure measurements on the building surface. Machine learning techniques such as linear regressions (Ridge, Lasso, ElastikNet, MultiTaskLasso regression) and Artificial Neural Network are tested and compared. The flow reconstruction using velocity measurements inside the wake leads to a better result compared to the flow reconstruction from the wall pressure measurements. At the same time, it was noticed that the Artificial Neural Network regression does not lead to more satisfactory results compared to linear regression techniques.Show less >

Language :

Anglais

Popular science :

Non

Collections :

• Laboratoire de Mécanique des Fluides de Lille Kampé de Fériet (LMFL) - UMR 9014

Source :

Harvested from HAL