Flow separation and reattachment happens in many fluid mechanical situations occurring in engineering applications as well as in nature. The flow over a backward-facing step represents a geometrically simple flow situation exhibiting both flow separation and reattachment. Broadly speaking there are only two important parameters in the problem, the Reynolds number(Re) based on the step height(h),and a geometrical parameter, referred to as the Expansion ratio(ER), defined as the downstream channel height to the upstream channel height. In spite of the relative simplicity of this geometry, the flow downstream is quite complex. The main focus of the present work is to elucidate the unsteady three-dimensional coherent structures present in this flow at large Re, Re>36,000,based on the step height(h). For this, we use velocity field measurements from Particle Image Velocimetry (PIV)in conjunction with hotwire anemometry measurements. The time-averaged structure of this flow is first studied in detail, including the effect of Reynolds number(Re) and Expansion Ratio(ER), on it. These studies show that at sufficiently large Re (Re>20,000), the reattachment length becomes independent of Re. The detailed internal structure of the separation bubble is also found to be independent of Re, but for Revalues that are relatively larger(Re>36,000). At large Re, the main effect of ER ,is found to be on the reattachment length, which increases with ER and saturates for ER values greater than about 1.8. The detailed internal structure of the separation bubble has been mapped at high Re and is found to be nearly the same for all ER, when the streamwise length is normalized by the reattachment length. In order to elucidate the unsteady coherent vortical structures, PIV measurements are done in two orthogonal planes downstream of the backward-facing step. These measurements are done for ER= 1.50 at large Re(Re=36,000) and in a large aspect ratio facility(AR= span length/step height= 24); the latter being important to avoid any effects due to span-wise confinement. In the spanwise plane parallel to the lower wall(x-z plane),instantaneous velocity fields show counter rotating vortex pairs, which is a signature of the three-dimensional vortical structures in this plane. Using conditional averaging, this counter-rotating vortex pair signature is captured right from upstream of the step, to well after reattachment. Spatial correlations are used to get the length scale of these coherent vortical structures, which varies substantially from the attached boundary layer before separation to the region after reattachment. The variation of these structures in the cross-stream (vertical) direction at reattachment and beyond gives an idea about their three dimensional shape. The circulation of these counter-rotating pairs is measured from the conditionally aver-aged fields, and is found to increase with streamwise distance reaching normalized circulation values (Γ/Uoh) of about 0.5 around reattachment. Velocity spectra downstream of the step show peaks corresponding to both the shear layer frequency(Stsl)and a relatively lower frequency that corresponds to large-scale shedding from the separation bubble (Stb); the latter in particular being quasi-periodic. Small amplitude sinusoidal forcing at the shedding frequency(Stb) is applied close to the step, by blowing and suction, to make the quasi-periodic shedding more regular. Measurements show that this has a very small effect on both the mean separation bubble and on the counter-rotating structures in the x-z plane. This mild forcing however enables phase locked PIV measurements to be made which shows the bubble shedding phenomenon in the cross-stream plane(side view or x-y plane). The phase-averaged velocity fields show significant variations from phase to phase. Although there is some hint of structures being shed, from these phase-averaged fields, it is not very clear. One of the primary reasons is the fact that the flow is effectively spanwise averaged, as the three-dimensional structures are not locked in the spanwise direction. To get a three dimensional view of the sheddin gphenomenon, it is necessary to lock the spanwise location with respect to the three-dimensional vortical structures before averaging across the different phases. We use the condition, u’<- urms, to locate the central plane between the counter-rotating structures, which in effect are the “legs” of the three-dimensional structure. With this condition, we effectively get a slice of the shedding cycle cutting through the “head” of the three-dimensional structure. Apart from this cut, we also get a cut between adjacent structures from the weak sweep events, with the condition u’<- urms. Using these conditions, on the phase-locked velocity fields, we effectively lock the structures in time, as well as in the spanwise direction. With this ,a clearer picture of the shedding process emerges. The flow is highly three-dimensional near reattachment and the shedding of the separation bubble is modulated in the spanwise direction owing to the three-dimensional hairpin like vortical structures in the flow. The separation bubble is seen bulged out and lifted high at locations where the head of the hairpin vortex passes, owing to the strong ejection of fluid caused by the vortical structure. On the other hand, outside the hairpin vortices, weak sweep events push the flow towards the wall and make it shallow and less prominent, with the shedding being very weak in this plane. From these observations, a three-dimensional picture of the flow is proposed.

Les stratégies de contrôle d’écoulement, telles que le soufflage / aspiration, ont prouvé leur efficacité à modifier les caractéristiques d’écoulement à des fins diverses en cas de configurations usuellement simples. Pour étendre cette approche sur des cas industriels, la simulation de dispositifs à échelle réelle et l’optimisation des paramètres de contrôle s’avèrent nécessaires. L’objectif de cette thèse est de mettre en place une procédure d’optimisation pour résoudre cette catégorie de problèmes. Dans cette perspective, l’organisation de la thèse est divisé en trois parties. Tout d’abord, le développement et la validation d’un solveur d’écoulement turbulent compressible instationnaire, résolvant les équations de Navier-Stokes moyennées (RANS) dans le cadre d’une discrétisation mixte de type éléments finis / volumes finis (MEV) sont présentés. Une attention particulière est portée sur la mise en œuvre de modèles numériques de jet synthétique à l’aide de simulations sur une plaque plane. Le deuxième axe de la thèse décrit et valide la mise en œuvre d’une méthode d’optimisation globale basée sur un modèle réduit du type processus gaussien (GP), incluant une approche de filtrage d’erreurs numériques liées aux observations. Cette méthode EGO (Efficient Global Optimization), est validée sur des cas analytiques bruités 1D et 2D. Pour finir, l’optimisation de paramètres de contrôle de jet synthétique sur deux cas test pertinents pour les industriels : un profil d’aile NACA0015, avec objectif de maximiser la portance moyenne et une marche descendante avec objectif de minimiser la longueur de recirculation moyenne. / Active flow control strategies, such as oscillatory blowing / suction, have proved their efficiency to modify flow characteristics for various purposes (e.g. skin friction reduction, separation delay, etc.) in case of rather simple configurations. To extend this approach to industrial cases, the simulation of a large number of devices at real scale and the optimization of parameters are required. The objective of this thesis is to set up an optimization procedure to solve this category of problems. In this perspective, the organization of the thesis is split into three main parts. First, the development and validation of an unsteady compressible turbulent flow solver using the Reynolds-Averaged Navier-Stokes (RANS) using a Mixed finite-Element/finite-Volume (MEV) framework is described. A particular attention is drawn on synthetic jet numerical model implementation by comparing different models in the context of a simulation over a flat plate. The second axis of the thesis describes and validates the implementation of a Gaussian Process surrogate model based global optimization method including an approach to account for some numerical errors during the optimization. This EGO (Efficient Global Optimization) method, is validated on noisy 1D and 2D analytical test cases. Finally, the optimization of two industrial relevant test cases using a synthetic jet actuator are considered: a turbulent flow over a NACA0015 for which the time-averaged lift is regarded as the control criterion to be maximized, and an incompressible turbulent flow over a Backward Facing Step for which the time-averaged recirculation length is minimized.