<p>  </p> <p>In gas turbines, the hot gas exiting the combustor can be as high as 2000 ^oC, and some of this hot gas enter into the space between the stator and rotor disks (wheelspace). Since the hot gas entering with its high temperatures could damage the disks, hot-gas ingestion must be minimized. This is done by using rim seals and by introducing a flow of cooler air from the compressor (sealing flow) into the wheelspace. </p> <p>Ingress and egress into rim seals are driven by the stator vanes, the rotor and its rotation, and the rotor blades. This study focuses on the first-stage turbine, where ingress could cause the most damage and has two parts. The first part focuses on understanding ingress and egress driven by the rotor and its rotation, known as rotationally-induced ingress, by studying ingress about an axial seal in a stator-rotor configuration without vanes and without blades. The second part focuses on understanding ingress and egress driven by stator vanes, known as externally-induced ingress, by studying a stator-rotor configuration with vanes but no blades, where the ratio of the external Reynolds number to the rotational Reynolds number is 0.538. For both parts, solutions were generated by wall-resolved large-eddy simulation (LES) based on the WALE subgrid model and by Reynolds-averaged Navier-Stokes (RANS) based on the SST model. For both stator-rotor configurations, the grid-independent solutions obtained were compared with available experimental data.   </p> <p>Results obtained for the configuration without vanes and blades show Kelvin-Helmholtz instability (KHI) to form even without swirl from the hot-gas flow and to create a wavy shear layer on the rotor. Also, Vortex shedding (VS) occurs on the backward-facing side of the seal and impinges on the rotor side of the seal. The KHI and VS produce alternating regions of high and low pressures about the rotor-side of the axial seal, which cause ingress to start on the rotor side of the seal. Results obtained for the configuration with vanes but no blades show both LES and RANS to correctly predict the coefficient of pressure, C_p, upstream of the axial seal. However, only LES was able to correctly predict the sealing effectiveness. This shows C_p by itself maybe is inadequate in quantifying externally-induced ingress. One reason why RANS was unable to predict sealing effectiveness is significantly under predicting the pressure drop on the rotor surface, which affected the pressure variation along the hot-gas path and hence the pressure difference across the axial seal, which ultimately drives ingress. </p>

Les différentes structures composant les moteurs aéronautiques requièrent des analyses dynamiques afin de prédire leur durée de vie. Pour des raisons d'allègement, les roues aubagées fixes de turbomachines, appelées redresseurs, sont conçus comme des ensembles de secteurs comportant plusieurs aubes. Cette architecture rompt la symétrie cyclique empêchant l'application des méthodes numériques l'exploitant. De plus, les dispersions géométriques et matériaux génèrent un désaccordage involontaire impliquant des zones de forte densité modale, dans lesquelles est observée une amplification de la réponse vibratoire, accrue par le caractère monobloc, et donc peu amorti, des secteurs. Une méthodologie statistique de prédiction du niveau vibratoire d'un secteur de redresseur désaccordé aléatoirement est développée ici. La modélisation des incertitudes est basée sur une approche paramétrique de la théorie probabiliste : des paramètres matériaux aléatoires suivant une loi uniforme sont associés à différentes parties du secteur. Une expansion de Karhunen-Loève permet de réduire le champ stochastique à un petit nombre de variables aléatoires et donc de diminuer les temps de calcul. Les modes stochastiques sont ensuite projetés sur ces espaces aléatoires par le biais de deux méthodes d'interpolation non-intrusives. La première est basée sur une projection sur une base du chaos polynomial tandis que la deuxième est une méthode de régression non-paramétrique (méthode MARS). Afin d'appliquer les deux méthodes de calcul à un modèle industriel, une méthode de double synthèse modale est appliquée permettant de diviser le temps de calcul des modes par un facteur d'environ 300. La sous-structuration adoptée s'adapte à la méthode de modélisation des incertitudes et s'avère robuste vis-à-vis du désaccordage. De plus, les deux méthodes permettent d'obtenir des résultats prédictifs en termes de moments statistiques tout en réduisant les temps de calculs. Enfin, la méthodologie est validée expérimentalement puisque l'enveloppe vibratoire numérique encadre la réponse fréquentielle expérimentale au niveau de la zone des modes d'intérêt. Une stratégie de positionnement des jauges de déformation est proposée à partir d'une distribution statistique des déplacements maximaux à mi-hauteur de veine sur une plage fréquentielle donnée. / Aircraft engine components necessitate extensive dynamical analyses in order to obtain life cycle prediction. In order to lighten the structure, turbomachinery stator bladed disks, called stator vanes, are designed as a set of multiple blades clusters. This architecture implies a loss of cyclic symmetry condition and prevents the use of numerical methods using it. Moreover, geometric dispersions and materials defaults generate an involuntary mistuning involving high modal density areas, in which is observed an amplification of the vibratory response, enhanced by the monobloc character - and hence low damped - of stator vanes. A statistical methodology for predicting the vibratory level of a randomly mistuned industrial stator vanes is developed here. Uncertainties modelization is based on a parametric approach of the probability theory : material random parameters following a uniform distribution are associated with different cluster's parts. A Karhunen-Loeve expansion reduces the stochastic field to a small number of random variables and therefore reduces the computation time. Stochastic modes are then projected on these random spaces through two non-intrusive methods of interpolation. The first is based on a projection on a polynomial chaos basis while the second is non-parametric regression method (MARS method). In order to implement both numerical methods to an industrial model, a double modal synthesis method is applied to divide the calculation time of modes by a factor around 300. The sub-structuring way adopted fits the uncertainties modelization method and is robust towards mistuning. Moreover, both methods yield predictive results in terms of statistical moments while reducing computation time. Finally, the methodology is experimentally validated because the numerical vibratory envelope frames the experimental frequency response at the area of the modes of interest. A positioning strategy of strain gauges is proposed based on a statistical distribution of the maximum displacements in vein halfway over a given frequency range.