Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / In this work a parameterization for the transport terms of the turbulent kinetic energy (TKE) budget equation, valid for a convective boundary layer (CBL) is presented. This is a hard task to acomplish from experimental data, especially because of the difficulty associated to the measurements of pressure turbulent fluctuations, which are necessary to determine the pressure correlation TKE transport term. Thus, employing a large eddy symulation (LES), all terms of the TKE budget equation were determined for a CBL. From these data, polynomials that describe the TKE transport terms vertical profiles were adjusted for a CBL. The found polynomial fits are a good description of the LES data, and from them it is shown that a simple formulation that directly relates the transport terms to the TKE magnitude has advantages on other parameterizations commonly used in CBL numerical models. Furthermore, the present study shows that the TKE turbulent transport term dominates over the TKE transport by pressure perturbations and that for most of the CBL these two terms have apposite signs. The simulation consists of a full diurnal PBL cycle utilizing, at the surface, a forcing obtained from experimental data, so that the numerical experiment represents a more realistic case than a idealized CBL. / Neste trabalho, uma parametrização para os termos de transporte da equação de balanço de energia cinética turbulenta (ECT), válida para uma camada limite convectiva (CLC), é apresentada. Esta é uma tarefa complicada de ser realizada a partir de dados experimentais, especialmente devido a dificuldade associada às medidas das flutuações de pressão, que são necessárias para a determinação do termo de correlação de pressão. Desta forma, empregando a simulação dos grandes turbilhões (LES, do inglês Large Eddy Simulation), todos os termos da equação de balanço de ECT foram determinados para a CLC. A partir desses dados, foram ajustados polinômios que descrevem os perfis verticais dos termos de transporte para a CLC. Os polinômios obtidos fornecem uma boa descrição dos dados da simulação LES, e em função deles é mostrado que uma formulação simples, que se relaciona com os termos de transporte a partir da ECT, apresenta vantagens em relação a outras paramametrizações comumente empregadas em modelos numéricos para a CLC. Além disso, o presente estudo mostra que o termo de transporte turbulento domina sobre o transporte devido a flutuações de pressão, e que para a maior parte da extensão vertical CLC estes dois termos tem sinais opostos. A simulação consiste em um ciclo diário da CLP, utilizando como forçante de superfície dados obtidos experimentalmente, assim o experimento numérico representa um caso mais realista que uma simulação de CLC estacionária.

L'objectif de la présente étude est d'analyser la modélisation de la turbulence de simulations en moyenne de Reynolds (RANS) dans le cadre d'écoulements de type turbomachines, en utilisant des simulations aux grandes échelles (SGE) comme référence. L'étude porte sur deux cas test: un décollement de coin dans une grille d'aubes rectiligne, et un écoulement de jeu pour un aubage isolé dans un jet. Deux lois de comportement, la loi de comportement de Boussinesq et la loi de comportement quadratique (quadratic constitutive relation ou QCR), sont analysées, avec deux versions du modèle de turbulence k-omega de Wilcox. Les lois de comportement étudiées reposent sur deux hypothèses: une hypothèse d'alignement entre le tenseur de Reynolds et un tenseur construit à partir de l'écoulement moyen, et une hypothèse sur la viscosité turbulente. L'hypothèse d'alignement est étudiée à partir de la SGE, pour laquelle les deux tenseurs sont indépendamment connus, en utilisant un indicateur construit sur le produit scalaire des tenseurs. Les résultats sont présentés sous forme d'une fonction de répartition de la valeur de l'indicateur pour le domaine complet, puis pour trois sous-domaines d'intérêt: l'entrée, une région où l'écoulement interagit fortement avec les parois, et une région où l'écoulement est fortement tourbillonnaire. L'hypothèse d'alignement n'est que rarement valide pour la loi de comportement de Boussinesq. Pour la QCR, les résultats sont meilleurs en entrée, comparé à la loi de Boussinesq. Il ne sont cependant pas meilleurs pour les régions où l'écoulement est plus tourbillonnaire. Une amélioration de la loi de comportement est nécessaire pour pouvoir faire progresser la modélisation turbulente en RANS. En revanche, l'utilisation de l'énergie cinétique turbulente et du taux de dissipation spécifique semble correcte pour estimer la valeur de la viscosité turbulente. L'analyse de la modélisation de l'équation d'énergie cinétique turbulente (ECT) est réalisée au travers d'une comparaison terme à terme avec l'équation d'ECT résolue par la SGE. Les résultats SGE présentent une turbulence qui n'est pas à l'équilibre : la production et la dissipation ne sont pas superposées, et le terme de transport est important. Pour le RANS, la turbulence est à l'équilibre : la production et la dissipation sont superposées, et le terme de transport est de faible intensité. Un modèle de turbulence qui prend en compte le déséquilibre est nécessaire pour améliorer ce point. En dernier lieu, une nouvelle formulation hybride RANS/SGE est proposée, fondée sur la distance à la paroi en unités de paroi. La formulation est validée dans un canal bi-périodique et un premier essai est réalisé sur le cas de décollement de coin, mais d'autres analyses sont nécessaires avant que cette formulation ne soit fonctionnelle. / The present study aims at analysing turbulence modelling in Reynolds-averaged Navier-Stokes (RANS) simulations, in the context of turbomachinery flows, using large-eddy simulations (LES) as references. Two test cases are considered: a corner separation (CS) flow in a linear compressor cascade, and a tip-leakage (TL) flow of a single blade in a jet. Two constitutive relations, the Boussinesq constitutive relation and the quadratic constitutive relation (QCR), are investigated, with two versions of Wilcox's $k-\omega$ turbulence model. The studied constitutive relations rely on two hypotheses: an alignment hypothesis between the Reynolds stress tensor and a mean flow tensor, and an hypothesis on the turbulent viscosity. The alignment hypothesis is investigated using LES, where both the tensors are known independently, with an indicator built on the inner product of the tensors. The results are presented as probability density functions of the indicator value for the entire domain first, and then for three specific areas of interest: the inlet area, similar to a boundary-layer flow, an area of strong interaction between the flow and the walls (CS: passage area, TL: tip clearance) and an area of highly vortical flow (CS: separation wake, TL: tip-leakage vortex). The alignment hypothesis is rarely verified in any area for the Boussinesq constitutive relation. For the QCR, the results are improved for the inlet areas compared to the Boussinesq constitutive relation, but no significant improvement is found in the highly vortical regions. An improvement of the constitutive relation is needed in order to improve the RANS turbulence modelling. In contrast, the use of the turbulent kinetic energy and the specific dissipation rate appears quite correct to estimate the turbulent viscosity. The modelling of the RANS turbulent kinetic energy (TKE) budget equation is investigated through a term to term comparison with the resolved LES TKE budget equation. The LES presents a turbulence that is not at equilibrium, with the production and the dissipation not superimposed, and an important amount of transport. This differs from the RANS models, at equilibrium: the production and the dissipation are superimposed, with a small amount of transport. The development of a non-equilibrium turbulence model for RANS simulations could improve this aspect of turbulence modelling. Finally, a new hybrid RANS-LES formulation, based on the wall distance in wall units, is also proposed. It is validated on a bi-periodical channel flow, and a first attempt is made on the corner separation case, but further investigations are still needed for the model to be fully operational.