L’aspirateur d’une centrale hydroélectrique est l’organe hydraulique se situant en aval de la roue. Il a une forme divergente afin de récupérer l’énergie cinétique résiduelle en sortie de roue sous forme de pression statique et augmenter ainsi la chute nette de la centrale. Dans le cas des turbines de basse chute de type bulbe, les pertes de charge dans l’aspirateur influencent fortement le rendement global de la centrale. La prédiction correcte de ces pertes de charge au cours du dimensionnement de la turbine représente donc un enjeu majeur. La prédiction numérique des pertes de charge dans l’aspirateur est un réel challenge car l’écoulement dans l’aspirateur est dynamiquement complexe avec des nombres de Reynolds élevés, la présence de swirl et d’un gradient adverse de pression. Ces caractéristiques font que les approches de modélisation classiquement utilisées dans l’industrie sont mises en défaut. L’objectif de ce travail est double : (i) améliorer la prédiction de l’écoulement turbulent dans l’aspirateur en utilisant des approches instationnaires URANS et LES et en portant une attention particulière à la description des conditions d’entrée de l’aspirateur et (ii) réaliser une analyse fine des échanges énergétiques dans l’aspirateur pour mieux comprendre l’origine des pertes de charge. Une condition d’entrée instationnaire représentative de l’écoulement en sortie de roue est élaborée pour ces calculs. Les résultats de simulation sont comparés avec des mesures expérimentales afin d’évaluer la capacité prédictive de chaque approche de modélisation de la turbulence (URANS et LES). Cette étape de validation met en évidence l’importance d’une définition correcte des trois composantes de la vitesse en entrée d’aspirateur. L’influence des conditions aux limites du domaine de calcul, à savoir la rugosité de la paroi et la condition de sortie de l’aspirateur, sur les résultats de simulation est évaluée, notamment dans le cas d’une résolution LES. Grâce à une analyse détaillée du bilan d’énergie cinétique moyenne dans l’aspirateur, les phénomènes hydrodynamiques responsables des pertes de charge sont identifiés. Ceci permet d'analyser en détail les différences de prédiction de pertes de charge entre les calculs URANS et LES et d’identifier les pistes d’amélioration de la prédiction numérique de ces pertes. Enfin, cette analyse permet de comprendre l’évolution des pertes de charge observée entre plusieurs points de fonctionnement de la turbine. / The draft tube of a hydraulic turbine is the turbine element located downstream of the runner. It has a divergent shape in order to convert the residual kinetic energy leaving the runner into pressure and thus increase the effective head of the turbine. The performances of low head bulb turbines are highly influenced by the head losses in the draft tube. The prediction of these head losses in a design process is thereby a major issue. The numerical prediction of the head losses in the draft tube is a real challenge because the flow in the draft tube is dynamically complex with high Reynolds numbers, a swirl and an adverse pressure gradient. These characteristics render conventional industrial approaches not appropriate. The objective of this work is twofold: (i) to improve the numerical prediction of the turbulent flow in the draft tube by using URANS and LES unsteady approaches and paying special attention to the description of the inlet boundary conditions of the draft tube and (ii) to conduct a detailed analysis of the energy transfers in the draft tube in order to better understand the origin of the head losses. An unsteady inlet boundary condition for the simulations reproducing the flow field at the runner outlet is developed. Numerical results are compared to experimental measurements in order to evaluate the predictive capacity of each turbulence modelling approach (URANS and LES). This validation step highlights the importance of defining properly the three velocity components at the draft tube inlet. The influence on the numerical results of boundary conditions of the calculation domain, such as wall roughness and the outlet boundary condition, is evaluated, in particular in case of LES. Thanks to a detailed analysis of the mean kinetic energy balance in the draft tube, the hydrodynamic phenomena responsible for head losses are identified. The head losses prediction differences between URANS and LES are thus analyzed in detail and possible improvements for the head losses prediction are identified. Finally, this analysis enables to understand the head losses evolution observed between several operating points of the turbine.

Global renewable energy requirements rapidly increase with the transition to a fossil-free society. As a result, intermittent energy resources, such as wind- and solar power, have become increasingly popular. However, their energy production varies over time, both in the short- and long term. Hydropower plants are therefore utilized as a regulating resource more frequently to maintain a balance between production and consumption on the electrical grid. This means that they must be operated away from the design point, also known as the best-efficiency-point (BEP), and often are operated at part-load (PL) with a lower power output. Moreover, some plants are expected to provide a spinning reserve, also referred to as speed-no-load (SNL), to respond rapidly to power shortages. During this operating condition, the turbine rotates without producing any power. During the above mentioned off-design operating conditions, the flow rate is restricted by the closure of the guide vanes. This changes the absolute velocity of the flow and increases the swirl, which is unfavorable. The flow field can be described as chaotic, with separated regions and recirculating fluid. Shear layer formation between stagnant- and rotating flow regions can be an origin for rotating flow structures. Examples are the rotating-vortex-rope (RVR) found during PL operation and the vortical flow structures in the vaneless space during SNL operation, which can cause the flow between the runner blades to stall, also referred to as rotating stall. The flow structures are associated with pressure pulsations throughout the turbine, which puts high stress on the runner and other critical parts and shortens the turbine's lifetime. Numerical models of hydraulic turbines are highly coveted to investigate the detrimental flow inside the hydraulic turbines' different sections at off-design operating conditions. They enable the detailed study of the flow and the origin of the instabilities. This knowledge eases the design and assessment of mitigation techniques that expand the turbines' operating range, ultimately enabling a wider implementation of intermittent energy resources on the electrical grid and a smoother transition to a fossil-free society. This thesis presents the numerical study of the Porjus U9 model, a scaled-down version of the 10 MW prototype Kaplan turbine located along the Luleå river in northern Sweden. The distributor contains 20 guide vanes, 18 stay vanes and the runner is 6-bladed. The numerical model is a geometrical representation of the model turbine located at Vattenfall Research and Development in Älvkarleby, Sweden. The commercial software ANSYS CFX 2020 R2 is used to perform the numerical simulations. Firstly, the draft tube cone section of the U9 model is numerically studied to investigate the sensitivity of a swirling flow to the GEKO (generalized kω) turbulence model. The GEKO model aims to consolidate different eddy viscosity turbulence models. Six free coefficients are changeable to tune the model to flow conditions and obtain results closer to an experimental reference without affecting the calibration of the turbulence model to basic flow test cases. Especially, the coefficients affecting wall-bounded flows are of interest. This study aims to analyze if the GEKO model can be used to obtain results closer to experimental measurements and better predict the swirling flow at PL operation compared to other eddy viscosity turbulence models. Results show that the near-wall- and separation coefficients predict a higher swirl and give results closer to experimentally obtained ones. Secondly, a simplified version of the U9 model is investigated at BEP and PL operating conditions and includes one distributor passage with periodic boundary conditions, the runner and the draft tube. The flow is assumed axisymmetric upstream of the runner, hence the single distributor passage. Previous studies of hydraulic turbines operating at PL show difficulties predicting the flow's tangential velocity component as it is often under predicted. Therefore, a parametric analysis is performed to investigate which parameters affect the prediction of the tangential velocity in the runner domain. Results show that the model predicts the flow relatively well at BEP but has problems at PL; the axial velocity is overpredicted while the tangential is underpredicted. Moreover, the torque is overpredicted. The root cause for the deviation is an underestimation of the head losses. Another contributing reason is that the runner extracts too much swirl from the flow, hence the low tangential velocity and the high torque. Sensitive parameters are the blade clearance, blade angle and mass flow. Finally, the full version of the U9 model is analyzed at SNL operation, including the spiral casing, full distributor, runner and draft tube. During this operating condition, the flow is not axisymmetric; vortical flow structures extend from the vaneless space to the draft tube and the flow stalls between the runner blades. A mitigation technique with independent control of each guide vane is presented and implemented in the model. The idea is to open some of the guidevanes to BEP angle while keeping the remaining ones closed. The aim is to reduce the swirl and prevent the vortical flow structures from developing. Results show that the flow structures are broken down upstream the runner and the rotating stall between the runner blades is reduced, which decreases the pressure- and velocity fluctuations. The flow down stream the runner remains mainly unchanged.