Skip to Main content Skip to Navigation
Theses

Quantification et réduction d'incertitudes liées à la modélisation de la turbulence dans les écoulements de turbomachines par méthodes multi-modèles bayésiennes

Abstract : Certification requirements and the constant pursuit of performance pushes aerospace manufacturers to thoroughly monitor the uncertainties inherent in their products.For this reason, Uncertainty Quantification (UQ) methods must now be integrated as early as possible in the design process, in order to guarantee reliability and performance.In this thesis, we focus on the numerical simulation of turbomachinery flows, and we present two methods for the quantification and reduction of epistemic uncertainties associated with turbulence closure models for the Reynolds Averaged Navier-Stokes (RANS) equations. These arise both from model-form inadequacy and from imperfect knowledge of model parameters.To make robust predictions under RANS model uncertainty, and to estimate and reduce uncertainties on the resulting solution, we investigate Bayesian multi-model ensembles techniques and, more specifically, Bayesian Model Averaging (BMA). This approach consists in using a set of competing model to make separate predictions of a turbulent flow of interest. Such predictions are then averaged together by using their posterior marginal probabilities, and the resulting model mixture is used to estimate expectancy and confidence intervals of the predicted flow properties.The first method, named Bayesian Model-Scenario Averaging (BMSA), extends BMA to account for the uncertainty in the choice of the flow configurations used to calibrate the model parameters.The second method, named space-dependent BMA (XBMA), produces space-dependent combinations of models by leveraging local information about the flow.Both methods demonstrate good generalization properties when predicting an unseen flow, while retaining the benefit of being non-intrusive, easy to implement, computationally affordable and general.Numerical examples focus on the quantification and reduction of turbulence modeling uncertainties for flows through a compressor cascade at various operating conditions
Complete list of metadata

https://pastel.archives-ouvertes.fr/tel-03652868
Contributor : ABES STAR :  Contact
Submitted on : Wednesday, April 27, 2022 - 10:37:21 AM
Last modification on : Wednesday, September 28, 2022 - 5:51:58 AM
Long-term archiving on: : Friday, July 29, 2022 - 9:17:19 AM

File

107627_DE_ZORDO-BANLIAT_2022_a...
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-03652868, version 1

Citation

Maximilien de Zordo-Banliat. Quantification et réduction d'incertitudes liées à la modélisation de la turbulence dans les écoulements de turbomachines par méthodes multi-modèles bayésiennes. Mécanique des matériaux [physics.class-ph]. HESAM Université, 2022. Français. ⟨NNT : 2022HESAE014⟩. ⟨tel-03652868⟩

Share

Metrics

Record views

88

Files downloads

17