Search

A direct numerical simulation of a moderate adverse pressure gradient turbulent boundary layer flow on a flat plate was performed on a fairly long domain with a Reynolds number from $Re_\theta=2000$ to $Re_\theta=8000$. The mean streamwise velocity profile does not exhibit a clear log region even at the highest Reynolds number. The Reynolds stresses exhibit a well defined peak in the outer region which moves away from the wall before to stabilise at a wall distance of 0.45 boundary layer thickness after evolving over 20 local boundary layer thicknesses. These outer peaks are the consequences of an excess of production over dissipation and they all scale with the shear stress velocity for wall distance $0.2 \delta < y < 0.85 \delta$ . The different terms of the Reynolds stresses equations are analysed in the outer region and the analysis confirms that the leading terms are the same as for a zero pressure gradient turbulent boundary layer except that the production and dissipation are not negligible in the outer region. The most significant contribution of scales smaller than two Taylor scales is on the pressure strain term of the shear stress equation. The probability of the occurrence of small scale vortices conditioned by the presence of an ejection or a sweep shows that the shear stress contained in ejections is associated with the production and transport of near wall vortices away from the wall. The local mean shear generated by the sweeps increases the production of near wall vortices. Lastly, the large scale structures analysed by means of two point correlations of the streamwise fluctuating velocity are shown to be strongly modified by the adverse pressure gradient. The correlation reveals a decoupling between the near wall streaks and the large scale streamwise structures enhanced by the adverse pressure gradient.

International audience; Flow control consists in modifying a flow natural state in order to converge towards another state which is considered as favorable, as drag or noise radiation might be reduced. In this paper, open-loop flow control experiments are carried out on a subsonic open-cavity flow. In the case of unstable flow control, the control focus is brought onto the flow fluctuations modifications rather than modification of the mean flow properties. Therefore, the forcing flexibility using arbitrary signals and the forcing linearity are essential for such flow control cases. In that sense, a linear array of Micro Magneto-Electro-Mechanical Systems actuators has been implemented to perform open-loop flow control experiments on an open-cavity. The actuators are able to generate both quasi-steady and pulsed jets with linear behavior. We proved the microvalves efficiency to damp the cavity oscillations. The quasi-steady jets reached a reduction of 20 dB in the cavity fundamental amplitude sound pressure level. Pulsed jets enabled an additional cavity tone amplitude reduction, which depends on the pulsating frequency and on the forcing amplitude. These results are a first step towards the implementation of the closed-loop control of the open-cavity flow.

International audience; The tip gap region of an axial compressor rotor is the source of complex flows inducing losses and stability issues. Recent works have proven the ability of blowing high-speed jets in the tip region to improve the surge margin of an axial compressor stage with a narrow tip gap configuration. Nevertheless, the tip gap size can evolve during the compressor lifetime, possibly affecting its performance and operability. The objective is to evaluate the performance of an active flow control system on a compressor with different tip gap sizes. The present work is based on the single stage compressor CME2 located at the Laboratory of Fluid Mechanics of Lille and equipped with actuators blowing at the rotor tip leading edge. Configurations with two different values of the tip gap to chord ratio (0.6% and 2.4%) have been experimentally tested. RANS simulations have also been performed. The effect of tip gap sizes and of the tip-blowing is evaluated on the flow topology and compressor performance.; La région du jeu en extrémité d'aubes de rotor de compresseur est la source d'écoulements complexes à l'origine de pertes et d'instabilités. De récents travaux ont montré la capacité de jet fluidiques haute-vitesse à améliorer la marge au pompage d'un compresseur axial monoétagé, avec un jeu réduit. Néanmoins, la taille de jeu évolue au cours de la vie du compresseur. ce qui affecte ses performances et son domaine d'opérabilité. L'objectif est d'évaluer la performance d'un dispositif de contrôle actif pour différentes tailles de jeu, sur le compresseur CME2. Ce travail est basé sur le banc d'essais localisé au Laboratoire de Mécanique des Fluides de Lille, et équipé d'actionneurs soufflant à l'amont du bord d'attaque, en tête de rotor. Deux tailles de jeu ont été évaluées expérimentalement (0,6% et 2,4%) ainsi que numériquement à l'aide de simulations RANS. L'effet de la taille de jeu et de l'injection est évalué en termes de topologie d'écoulement et de performances du compresseur.

International audience; In this study, airfoil shape optimization of a wind turbine blade is performed using the ANSYS Fluent Adjoint Solver. The aim of this optimization process is to increase the wind turbine output power, and the objective function is to maximize the airfoil lift to drag ratio (CL/CD). This study is applied to the NREL phase VI wind turbine, therefore, the S809 airfoil is used as a reference profile. First, for the validation of the applied numerical model, steady-state simulations are carried out for the S809 airfoil at various angles of attack. Then, the optimization is performed with the airfoil set at a fixed angle of attack, AOA= 6.1°, considering three Reynolds numbers, RE = 300 000, 480 000, 1 000 000 . Next, computations are performed for the fluid flow around the optimized airfoils at angles of attack ranging from 0° to 20°. The results show that (i) the lift to drag ratios ofthe optimized airfoils are significantly improved compared to the baseline S809 airfoil, (ii) this improvement is sensitive to the Reynolds number, and (iii) the CL/CD ratios are also improved for another angle of attack values. Thereafter, the optimized airfoils are used for the design of the NREL Phase VI blade and the aerodynamic performances of this new wind turbine are assessed using the open-source code QBlade. These latter results indicate that when the blades are designed with the optimized airfoils, the wind turbine aerodynamic performances increase significantly. Indeed, at a wind speed of 10 m/s, the power output of the wind turbine is improved by about 38% compared to that of the original turbine.

Air jets for active flow control have proved effective in postponing the onset of stall phenomenon in axial compressors. In this paper, we use a combination of machine learning and genetic algorithm to explore the optimal parameters of air jets to control rotating stall in the axial compressor CME2. Three control parameters are investigated: the absolute injection angle, the number of injector pairs and the injection velocity. Given an experimental dataset, the influence of the air jet parameters on the surge margin improvement and power balance is modeled using two shallow neural networks. Parameters of the air jets are then optimized using a genetic algorithm for three rotational velocities, i.e., Ω=3200RPM,4500RPMand6000RPM. First, surge margin improvement and power balance are being maximized independently. Then, a bi-objective optimization problem is posed to explore the trade-off between the two competing objectives. Based on the Pareto front, results suggest that a globally optimal set of parameters is obtained for a velocity ratio (defined as the ratio of the injection velocity to the rotor tip speed) ranging from 1.1 to 1.6 and an injection angle attack varying from 1∘to11∘. These outcomes point out a potential generalization of the control strategy applicable to other compressors.

International audience; The present paper is devoted to characterizing the three-dimensional effects in a cavitating flow generated in a Venturi-type profile. Experimental measurements based on 2D3C(Two-dimensionalthree-component) stereoscopic PIV(Particle Image Velocimetry) are conducted to obtain the three components of the velocity field in multiple vertical planes aligned with the main flow direction, from the center of the channel to the side walls. Time-resolved acquisitions are conducted, so not onlytime-averaged quantities but also velocity fluctuations can be discussed. The attention was focused on configurations of cloud cavitation, where the attached cavity experiences large-scale periodical oscillations and shedding of clouds of vapor. Although the water channel is purely two-dimensional, some significant flow velocities in the third direction (depth of the test section) were measured. Some of those velocities were found to be related to small differences between the boundary conditions on the two sides, such as minor gaps between the sides and the bottom wall, while others reflect intrinsic three-dimensional mechanisms inside the cavitation area, such as side jets that contribute to the periodical instability process. These mechanisms are discussed, and a possible 3D(Threedimensional) structure of the cavitating flow is proposed.

International audience; Homogeneous turbulence and turbulence in scale-by-scale equilibrium, played a leading role in the turbulence research of the second half of the twentieth century, and Jack Herring was an important contributor to these developments. The research activity which has followed these developments over the past ten to fifteen years concerns turbulence, which is out of scale-by-scale equilibrium either because it is non-stationary or because it is non-homogeneous or both. This paper is a short review of recent progress in this relatively new direction of turbulence research.

Helicopters are widely used for civil and military needs. The assets of the rotorcrafts are numerous: vertical take-off and landing, hovering and mobility in all directions … However, amongst other limitations, the conventional configurations have a speed limited of about 300 km/h. In this context, the high-speed helicopters were created. Equipped with additional rotors and/or lifting devices, these configurations allow to unload the rotor and to increase the aerodynamic efficiency of the aircraft. However, the multiplication of rotating elements generates an increase of the interactions that can lead to a loss of performances, and even to critical flight situations. Thus, this thesis proposes to experimentally study the rotor/propeller interactions depending on the flight conditions for compound helicopters, of which the Eurocopter X3 and the Airbus RACER are examples. To date, numerical studies have been conducted but the complex phenomenon could not be transcribed entirely. In parallel, only few experimental data is available.Experimental studies are lead to underline the influence of the advance ratio, of the rotational velocity of the propeller and of its position on the performances of the rotors. The tests are conducted on a 1/7.7 scale Dauphin 365N equipped with a propeller whose position is modifiable in all three directions. The entire campaign is conducted in the ONERA large-size L2 wind tunnel.Efforts and velocity fields measurements highlighted the absence of direct rotor/propeller interaction at high speed. In hover, however, the propeller is entirely immersed in the rotor wake, which generates an increase of its performances. At low speed, the propeller is partially immersed in the rotor wake. In this context, a translation of the propeller towards the tail of the helicopter leads to increased interactions and propeller thrust. The absence of benefits consequent to the vertical or lateral translation has also been shown. Finally, the negligibility interactions of the propeller on the rotor has been highlighted for all the flight configurations.; Les hélicoptères sont largement utilisés pour des besoins civils et militaires. Les atouts des Appareils à Atterrissage et Décollage Vertical sont nombreux : capacité d’atterrissage et de décollage vertical, vol stationnaire et mobilité dans toutes les directions, … Cependant, parmi d’autres limites, les configurations classiques ont une vitesse limitée à environ 300 km/h. Dans ce cadre, les hélicoptères grande-vitesse ont été créés. Dotés de rotors et/ou d’aile de portance supplémentaires, ces configurations permettent de décharger le rotor et d’augmenter l’efficacité aérodynamique de l’appareil. Cependant la multiplication d’éléments tournants génère une augmentation des interactions qui peuvent mener à une perte de performance, voire à des situations de vol critiques. Ainsi, cette thèse propose d’étudier expérimentalement les interactions rotor/propulseur en fonction des conditions de vol pour les hélicoptères hybrides, dont l’Eurocopter X3 et l’Airbus Helicopters RACER sont des exemples. A ce jour, des études numériques ont été menées, mais les phénomènes d’interaction complexes ne sont pas entièrement retranscrits. En parallèle, peu de données expérimentales sont disponibles.Des études expérimentales sont menées pour souligner l’influence du paramètre d’avancement, de la vitesse de rotation du propulseur et de sa position sur les performances des rotors. Les essais sont menés sur une maquette à l’échelle 1/7,7 du Dauphin 365N dotée d’un propulseur dont la position est modifiable dans les trois directions. L’ensemble de la campagne est conduit dans la soufflerie de grande dimension L2 de l’ONERA.Des mesures d’efforts et de champs de vitesse ont mis en avant l’absence d’interactions directes rotor/propulseur à grande vitesse. En vol stationnaire, cependant, le propulseur est entièrement immergé dans le sillage rotor, ce qui génère une augmentation de sa traction. A faible vitesse, l’hélice est partiellement immergée dans le sillage rotor. Dans ce cadre, une translation du propulseur vers la queue de l’appareil génère une augmentation des interactions et des performances du propulseur. L’absence de bénéfices observés lors de la translation verticale ou latérale de l’hélice a aussi été soulignée. Enfin, la négligeabilité des interactions du propulseur sur le rotor a été montrée pour l’ensemble des configurations de vol.

The fundamental study of turbulence is of significant importance due to its dominant role in many fields of physics, engineering sciences, astrophysics, aeronautics, etc. Turbulent dissipation plays a central role in turbulent flow and is therefore an important quantity in turbulence models. With the aim of characterizing this dissipation, the present work is based on a particle tracking technique called "Shake-The Box" to measure three-dimensional velocity fields as function of time at a resolution which can approach dissipative scale. The experimental setup and processing of the Shake the Box technique have been optimized and pushed to their limits in order to improve the spatial resolution and the accuracy of the measured Lagrangian trajectories. The approach used in this thesis was based not only on experimental data but also on the synthetic PIV images based on direct numerical simulation of the Navier Stokes equations. The quantitative and qualitative investigation allows us to define the optimal parameters for our targeted application on a Giant Von Karmen flow and gives the limits of the method; L'étude fondamentale de la turbulence revêt d’une importance significative en raison de son rôle dominant dans de nombreux domaines de la physique, des sciences de l'ingénieur, de l'astrophysique, de l'aéronautique, etc. La dissipation turbulente joue un rôle central dans l'écoulement turbulent et est donc une quantité importante dans les modèles de turbulence. Dans le but de caractériser cette dissipation, le présent travail est basé sur une technique de suivi de particules appelée "Shake-The Box" pour mesurer les champs de vitesse tridimensionnels en fonction du temps à une résolution pouvant avoisiner l'échelle dissipative. Le dispositif expérimental et le traitement de la technique Shake the Box ont été optimisés et poussés à leurs limites afin d'améliorer la résolution spatiale et la précision des trajectoires lagrangiennes mesurées. L'approche utilisée dans cette thèse est basée non seulement sur des données expérimentales mais aussi sur les images synthétiques PIV basées sur une simulation numérique directe des équations de Navier Stokes. L'investigation quantitative et qualitative nous permet de définir les paramètres optimaux pour notre application ciblée sur un écoulement de Von Karman Géant et donne les limites de la méthode

The goal of this paper is to experimentally study the effect of the propeller position on the interaction between the rotor and the propeller on a compound helicopter configuration. Experiments are conducted in the ONERA L2 large-size low-speed wind tunnel with a 1/7.7 Dauphin 365 N model and a four-bladed small-scale propeller. PIV measurements have been used to visualize the velocity fields around the helicopter and to study changes to the flow field under the rotor up- and downstream of the propeller. Different flight conditions (wind speeds and propeller position) are tested. At low speed, the propeller is partially immersed and the flow on the propeller disk is highly asymmetrical. In this flight configuration, moving the propeller toward the tail of the device increases the interactions and the thrust of the propeller. No direct interactions of the two rotating parts are measured at high speed for all the configurations tested.; Cet article se concentre sur l'évaluation expérimentale des interactions entre le rotor et l'hélice pour les configurations hybrides composées. Les expériences sont menées dans la soufflerie de grande taille à basse vitesse ONERA L2 avec un modèle Dauphin 365N au 1/7.7 et une hélice quadripale à petite échelle. Des mesures PIV sont effectuées pour visualiser les champs de vitesse autour de l'hélicoptère. Différentes conditions de vol (vitesse du vent et position de l'hélice) sont testées.En vol stationnaire, l'hélice est complètement immergée dans le sillage du rotor pour toutes les configurations, et la poussée est donc maximale. A basse vitesse, l'hélice est partiellement immergée et l'écoulement sur le disque de l'hélice est fortement asymétrique. Dans cette configuration de vol, le déplacement de l'hélice vers la queue de l'appareil augmente les interactions et donc les performances de l'hélice. Aucune interaction directe des deux parties tournantes n'est mesurée à haute vitesse pour toutes les configurations testées.

A non-intrusive method to get a multi-element polynomial chaos model is developed. This method is called ME-ACD, for Multi-Element based on Agglomerative Clustering on Derivatives. It aims at approximating a Quantity of Interest which presents discontinuities or irregularities making it difficult to be accurately approximated by standard Polynomial Chaos models. The method permits to efficiently split the parameter space and to train local polynomial models of lower degrees on every element where the local pieces of the Quantity of Interest are smoother. The algorithm is based on agglomerative clustering of the observations in a well-chosen abstract space taking into account the value of the Quantity of Interest and of its derivatives with respect to the stochastic input parameters. The same observations are used for both partitioning the space and training the local models. Several partitions of the parameter space are tested, and the one leading to local models minimizing a crossvalidation error is selected. Once the training observations are labelled with a class number indicating the element they are located in, a neural network classifier is trained to determine which local model to use for further evaluations. The method has proven to efficiently split the parameter space for a set of applications and the piecewise chaos model is compared with a standard Polynomial Chaos non-intrusive method and a Gradient Tree Boosting in terms of accuracy., Une méthode non-intrusive ayant pour but d’obtenir un modèle par chaos polynomial multi-élément est développée. La méthode se nomme ME-ACD, acronyme Anglais signifiant « méthode multi-élément basée sur la classification hiérarchique ascendante des dérivées ». Le modèle a pour objectif d’approcher les Quantités d’intérêt présentant des discontinuités et/ou des irrégularités altérant la précision des modèles par chaos polynomial standards qui l’approchent. La méthode permet de partitionner efficacement l’espace des paramètres et d’entrainer des méta-modèles par chaos polynomial locaux de degrés inférieurs sur chaque sous-ensemble de l’espace des paramètres où les morceaux de la Quantité d’intérêt est plus régulière localement. L’algorithme repose sur le regroupement hiérarchique ascendant des observations représentées par des points dans un espace abstrait prenant en compte la valeur de la Quantité ainsi que les valeurs de ses dérivées par rapport aux paramètres d’entrée stochastiques. Les mêmes observations sont utilisées pour la partition de l’espace et pour l’entrainement des modèles locaux. Plusieurs partitions de l’espace sont testées, et celle menant à des modèles locaux minimisant une erreur de validation croisée est sélectionnée. Une fois que les observations d’entrainement sont associées à un numéro indiquant dans quel élément elles sont localisées, un réseau de neurones est entrainé pour déterminer quel modèle local sera utilisé pour les évaluations à venir. L’efficacité de la méthode est montrée sur un ensemble d’applications et la précision du modèle par morceaux est comparée avec celle d’un modèle par chaos polynomial standard obtenu par une méthode non-intrusive et celle d’un modèle « Gradient Tree Boosting ».

Among the new technologies identified as promising for a reduction of civil airplane fuel consumption is Boundary Layer Ingestion (BLI). The technology is based on the ingestion of the fuselage boundary layer into the jet engine and different studies show that a reduction between 5% and 10% of the fuel consumption is theoretically possible. However it is known that the performance and the stall margin of fans and compressors are badly affected in the presence of inlet flow distortion (as the one created by an ingested boundary layer). The effect of the flow distorsion on the phenomena leading to the loss of performance, and especially to rotating stall and surge, are still not known in sufficient detail by engine manufacturers to allow the production and operation of BLI engines. The present work is a part of the project SUBLIME led by ONERA and dealing with different aspects of BLI. In this frame, the aim of the present study is to increase knowledge about the phenomena that degrade the performance and stall margin of compressors in distorted flows, doing so by providing an experimental dataset of measurements in an axial compressor operated under distorted flow. The experiments have been performed on the axial flow low speed compressor CME2 operated in the Laboratoire de Mécanique des Fluides de Lille Kampé de Fériet (LMFL). A number of distortion grids of different porosities and angular extensions intended to represent very different types of inlet flow nonuniformities have been designed and mounted at the inlet of the compressor to influence its performance. Total pressure and static pressure measurements have been performed to identify the loss of performance, and unsteady pressure measurements and Particle Image Velocimetry (PIV) have been used to investigate in more detail the flow features, with a focus on the stability limit of the compressor.In particular, the study provides a classification into four flow regimes of the various tested distortions, which enriches the physics previously identified in distortion studies. It provides a new scaling of the stall margin based on a physical ground, that is an angle characteristic of the interaction between the distorted and undistorted region of the flow. Moreover, associated to these four flow regimes are identified four different stall regimes, suggesting a strong link between the stalling dynamics and the flow redistribution upstream of the compressor.; Parmi les technologies identifiées pour réduire la consommation de fuel, une des plus prometteuses est l’ingestion de couche limite (Boundary Layer Ingestion – BLI). Cette technologie se base sur l’ingestion, par le moteur, de la couche limite se développant sur le fuselage. Plusieurs études montrent qu’une réduction de 5 à 10 % de la consommation est alors théoriquement atteinte. Cependant, la présence de distorsion de l’écoulement à l’entrée des soufflantes et des compresseurs affecte négativement la performance et la marge au pompage de ces machines. Les effets de la distorsion sur les phénomènes qui amènent la perte de performance, et particulièrement le décrochage tournant et le pompage, sont encore mal connus par les constructeurs de moteurs aéronautiques et cela est un frein à la production et à l’utilisation de moteurs BL. Cette thèse s’inscrit dans le cadre du projet SUBLIME mené par l’ONERA et traite de différents aspects liés à la BLI. Elle a pour objectif de mieux comprendre les phénomènes amenant la dégradation de la performance et de la marge au pompage des compresseurs fonctionnant sous distorsion, et de produire une base de données expérimentale. Les expériences ont été réalisées sur le compresseur axial mono-étage CME2, opéré au Laboratoire de Mécanique des Fluides de Lille Kampé de Fériet (LMFL). Des grilles de porosité et d’extension angulaires différentes ont été conçues pour représenter différents types de non-uniformité en entrée de machine. Des mesures de pressions statiques et totales ont permis de caractériser les modifications de performance de l’étage et des mesures de pressions instationnaires et de vitesse par Vélocimétrie par Images de particules (PIV) ont été utilisées pour étudier en détail les écoulements internes à la machine, en se concentrant sur les conditions proche de limite de stabilité de la machine.Cette étude a permis de classifier les différentes configurations en quatre régimes d’écoulement, ce qui enrichit la physique connue, issue des précédentes études. Une nouvelle échelle a été introduite, sur la base d’arguments physiques, à savoir un angle d’interaction entre le compresseur, les régions distordue et non-distordue de l’écoulement. Par ailleurs, pour chacun des régimes, une forme différente de transition vers le décrochage tournant a été identifiée, suggérant un lien fort entre la dynamique de transition vers le décrochage et la redistribution de l’écoulement à l’entrée.

An experiment was performed using stereo particle image velocimetry (SPIV) in the Laboratoire de Mécanique des Fluides de Lille boundary layer facility to determine all the derivative moments needed to estimate the average dissipation rate of the turbulence kinetic energy $\epsilon = 2 \nu \langle {\mathsf{s}}_{ij}{\mathsf{s}}_{ij} \rangle$ , where ${\mathsf{s}}_{ij}$ is the fluctuating strain rate and $\langle ~\rangle$ denotes ensemble averages. Also measured were all the moments of the full average deformation rate tensor, as well as all of the first, second and third fluctuating velocity moments except those involving pressure. The Reynolds number was $Re_\theta = 7634$ or $Re_\tau = 2598$ . The present paper gives the measured average dissipation, $\epsilon$ and the derivative moments comprising it. The results are compared with the earlier measurements of Balint, Wallace & Vukolavcevic (J. Fluid Mech., vol. 228, 1991, pp. 53–86) and Honkan & Andreopoulos (J. Fluid Mech., vol. 350, 1997, pp. 29–96) at lower Reynolds numbers and to new results from a plane channel flow DNS at comparable Reynolds number. Of special interest is the prediction by George & Castillo (Appl. Mech. Rev., vol. 50, 1997, pp. 689–729) and Wosnik, Castillo & George (J. Fluid Mech., vol. 421, 2000, pp. 115–145) that $\epsilon ^+ \propto {x_2^+}^{-1}$ for streamwise homogeneous flows and a nearly indistinguishable power law, $\epsilon \propto {x_2^+}^{\gamma -1}$ , for boundary layers. In spite of the modest Reynolds number, the predictions seem to be correct. Then the statistical character of the velocity derivatives is examined in detail, and a particular problem is identified with the breakdown of local homogeneity inside $x_2^+ = 100$ . A more general alternative for partially homogeneous turbulence flows is offered which is consistent with the observations. With the help of DNS, the spatial characteristics of the dissipation very near the wall are also examined in detail.

In a turbulent fluid, the time-reversal symmetry is explicitly broken by viscosity, and spontaneously broken in the inviscid limit. Recently, Drivas [J. Nonlinear Sci. 29, 65 (2019).JNSCEK0938-897410.1007/s00332-018-9476-8] proved the equivalence of two different local indicators of time irreversibility: (i) an Eulerian one, based on regularity properties of the velocity field [Duchon and Robert, Nonlinearity 13, 249 (2000).NONLE50951-771510.1088/0951-7715/13/1/312]; (ii) a Lagrangian one, based on symmetry properties of the trajectories under time reversal [Jucha et al., Phys. Rev. Lett. 113, 054501 (2014).PRLTAO0031-900710.1103/PhysRevLett.113.054501]. We test this equivalence in a turbulent Von Kármán experiment at a resolution of the order of the Kolmogorov scale using a high resolution 4D-PTV technique. We use the equivalence to perform the first joined Eulerian-Lagrangian exploration of the dynamics leading to time irreversibility, and find that it is linked with vortex interaction, suggesting a link between irreversibility and singularity.

International audience; A synchronised multi-axis digital holographic interferometry set-up is presented for the study of 3-D flow fields with large density gradients. This optical configuration provides instantaneous interferograms with fine spatial resolution in six directions of projection. A regularised tomographic approach taking into account the presence of possible shock waves is furthermore considered to reconstruct 3-D density fields. Applied to a screeching under-expanded supersonic jet with helical dynamics, this set-up is used to provide dense optical phase measurements in the initial region of the jet. The jet mean density field is shown to be satisfactorily estimated with sharply resolved density gradients. In addition, an approach based on azimuthal Fourier transform and snapshot proper orthogonal decomposition (POD) applied to the instantaneous flow observations is proposed to study the main coherent dynamics of the jet. Relying on a cluster analysis of the azimuthal POD mode coefficients, a reduced dynamical model in the POD mode phase space is used as an approximation of the two observed limit cycles. A clear 3-D representation of the density field of a helical instability associated with screech mode C is then evidenced, with two equally probable directions of rotation. Switching between the two directions is reported, highlighting intermittency in the feedback loop. This helical structure is particularly seen to extend to the jet core, driving its internal dynamics and inducing out-of-phase density fluctuations between the outer and inner shear layers. These out-of-phase motions are related to the non-uniform radial distribution of fluctuation phase associated with the outer-layer Kelvin–Helmholtz instability wave.

In this work, high-speed drop impact onto deep volume of the same liquid is numerically studied in 3D using Direct Numerical Simulation (DNS). The numerical results are compared with the previous experimental works by Murphy et al. [1]. Most of the distinctive features during the impact, including the evolution of the crown and cavity, the ligaments emanating from the rim of the crown, the formation of the bubble canopy, the central spiral jet that pierces through the bottom of the cavity and the subsequent broad upward jet, are correctly reproduced. The trajectory of the upper rim of the crown as well as the depth and width of the subsurface cavity are tracked during the simulation. Very reliable quantitative agreements between numerical and experimental data are obtained with respect to the time evolution of the crown rim and cavity dimensions. The simulation shows very good qualitative and quantitative agreement against the available experimental data, which enables the further investigation of the internal mechanisms of drop impact in the future This work has been supported by the scholarship from China Scholarship Council (CSC) under the Grant CSC NO. 201908320462.

International audience; Flight Dynamics aims at studying the motion of an aircraft modelled as a solid. The equations of motion are gathered in a dynamical system of coupled Ordinary Differential Equations. Classically, behaviours of the aircraft are linked to equilibria of this system, and bifurcation theory is used. In the dynamical system, the forces and the moments from the fluid on the aircraft rely on models with a chosen structure, constructed with data coming from experiments. These models are therefore uncertain. It had be shown that the uncertainties of these models can have influence on some equilibria, notably on their stability. In this paper, a joint Uncertainty Quantification and Global Sensitivity Analysis study is conducted on the temporal evolution of a given predicted behaviour of a light aircraft. As different combinations of the uncertain parameters can lead to instability of the behaviour and other combinations to stability, the state variables of the dynamical system can present some irregularities or some discontinuities with respect to the uncertain parameters at a given instant. In this paper, a method to estimate Sobol indices of a solution of an Ordinary Differential Equation approximated with a multi-element polynomial chaos model is therefore used. It relies on two upcoming papers. The first paper presents a method relying on Machine learning methods to get a multi-element polynomial chaos model for a non-time dependent Quantity of Interest. The second one extends the analytical formula between the Sobol indices and the coefficients of a polynomial chaos surrogate model to the case of a multi-element polynomial chaos model. In the present paper, the time-dependent extension of the multi-element model is introduced. The analytical formula permits to find Sobol indices from the coefficients of the local polynomial models at every considered instants. Conversely to standard polynomial chaos models, the use of a multi-element model permits to accurately approximate solutions of Ordinary Differential Equation with bifurcations, being irregular or discontinuous with respect to the input parameters at some instants. This approach is then used to conduct the sensitivity study on the state variables of an aircraft behaviour predicted by the Flight Dynamics system.; La Quantification d’incertitude a pour but d’étudier l’impact de certains paramètres sur une Quantité d’intérêt (QOI). Les paramètres sélectionnés sont nommés les paramètres d’entrée. Ils sont supposés aléatoires, indépendants, et sont supposés suivre une loi de distribution stochastique de densité de probabilité connue. Les indices de Sobol, reposant sur la définition des variances partielles, classent les actions des paramètres d’entrée et les interactions entre ces paramètres d’après leur impact sur la QOI. Les indices donnent également un aspect quantitatif de ces impacts. Les indices de Sobol au premier ordre et les indices de Sobol totaux peuvent être estimés avec des méthodes de Monte-Carlo, mais nécessitent souvent un grand nombre de simulations. Approchant la relation entre le QOI et les paramètres d’entrée utilisant un méta-modèle par chaos polynomial avec le « generalised Polynomials Chaos » (gPC) framework permet d’établir un lien analytique entre les coefficients du métamodèle et les indices de Sobol. Une fois le métamodèle obtenu, les indices de Sobol sont accessibles. Néanmoins, les modèles par chaos polynomial ne sont pas précis en présence d’irrégularités ou de discontinuités. Un grand nombre de termes de la série est parfois nécessaire pour obtenir une précision acceptable, ce qui peut devenir problématique avec la célèbre malédiction de la dimensionnalité. De plus, le phénomène de Gibbs perdure et s’amplifie lorsque le degré de troncature de la série est augmenté alors que la QOI est discontinue.Les modèles multi-élément (ME) ont été développé pour résoudre ce problème. L’espace des paramètres, espace abstrait contenant toutes les valeurs possibles des paramètres d’entrée, est partitionné en sous-ensembles nommés éléments où la QOI est plus régulière et des modèles gPC locaux sont entrainés sur chacun de ces éléments. Le modèle global est donc un modèle par morceaux et contient une méthode afin de déterminer quel modèle local utiliser lors d’une évaluation. Néanmoins, avec ce modèle par morceaux, le lien analytique avec les indices de Sobol est perdu. Dans un article en cours d’écriture qui sera bientôt soumis, le lien analytique entre les indices de Sobol et les coefficients du chaos sera étendu dans les cas des modèles ME avec une QOI ne dépendant pas du temps. Dans cet article, une nouvelle approche est présentée. En premier lieu, les paramètres d’entrée subissent une bijection vers des nouvelles variables contenues dans l’hypercube unité. Ensuite, une matrice d’interaction contenant toutes les informations nécessaires pour obtenir les variances partielles et l’espérance est calculée. Les indices de Sobol sont ainsi accessibles à l’aide de cette matrice. Ce lien analytique entre les coefficients du métamodèle ME et les indices de Sobol est une extension de la formule effectuant le lien entre les indices de Sobol et les modèles gPC standards.Dans le présent article, nous étendons cette approche aux QOIs étant solutions de systèmes d’Équations Différentielles Ordinaires (ODE), et donc dépendantes du temps. La version dépendante du temps d’une nouvelle méthode ME appelée ME-ACD, développée dans un article qui sera également bientôt soumis, est utilisée. N’importe quelle méthode ME pourrait être utilisée à la place. La nouvelle approche est ainsi capable de donner une estimation des indices de Sobol avec un nombre de résolutions du système relativement bas, même si à un certain instant, la QOI est irrégulière ou discontinue (traduisant la présence de bifurcations) par rapport aux paramètres d’entrée. La précision de l’approche est étudiée sur plusieurs applications. Tout d’abord, une ODE classique est testée. Ensuite, la précision de la méthode est testée sur un système théorique modélisant l’apparition d’une bifurcation. En dernier lieu, l’approche est utilisée dans un cas pratique de Dynamique du vol.

International audience; Wind tunnel generation of flows representative of the Atmospheric Boundary Layer (ABL) still requires trial-and-error based design process for both passive and active devices. To address this issue, a new type of passive grids – Multiscale Inhomogeneous Grids (MIGs) — is presented. It takes advantage of obstruction variations and interacting wakes to tailor both mean flow profile and turbulent intensity profiles downstream of the grid.The present work derives theoretical models in order to predict the MIG-generated flows characteristics (mean velocity, turbulence intensities, integral length scales profiles), especially for applications to neutral atmospheric boundary layer. These prediction models are investigated experimentally downstream of MIGs using Hot-Wire Anemometry within the SCL-PIV subsonic wind tunnel at ONERA Lille (Section H×W = 290mm ×300mm, velocities up to 30 m/s).First, the mean velocity profile prediction, based on McCarthy (1954), is validated and applied successfully with MIGs to generate logarithmic mean profiles for two equivalent roughness length h0 (up to h0⁄H=0.001).A theoretical turbulence model is then derived based on the Turbulent Kinetic Energy (TKE) equation in the turbulence decaying region. Empirical models are suggested to close the main model, enabling a turbulence-based grid design.; La génération en soufflerie d'écoulements représentatifs de la Couche Limite Atmosphérique (CLA) requiert encore aujourd'hui l'utilisation de méthodes de design par essai-erreur, à reproduire pour toutes nouvelles installations et toutes nouvelles configurations souhaitées. Pour tenter de réduire cette difficulté, un nouveau type de grilles passives – les grilles multi-échelles inhomogènes (MIGs) – est proposé. Elles tirent parti d'une variation d'obstruction locale et d'interactions entre sillages pour controler à la fois le champs moyens et l'intensité turbulente en aval de grille.Le travail actuel décrit plusieurs modèles théoriques permettant une prédiction des caractéristiques des écoulements en aval de MIGs (profil de champs moyen, d'intensité turbulente et d'échelles intégrales), en vue d'une application à la génération de conditions atmosphèriques (CLA neutre). Ces modèles sont comparés à des mesures réalisés en soufflerie par Hot-Wire Anemometry (HWA) dans la soufflerie SCL-PIV de l'ONERA Lille (Section H×W = 290mm ×300mm, velocities up to 30 m/s).Tout d'abord, les prédictions de champs moyens, basées sur le modèle de McCarthy (1954), sont bien validés et appliqués avec succès pour permettre la génération de profil logarithmique pour deux longueurs de rugosités h0 différentes (jusqu'à h0/H=0.001).Un modèle d'évolution aval de la turbulence générée est ensuite développé, basé sur l'équation de l'énergie cinétique turbulente, pour décrire la région de déclin de turbulence. Des modèles empiriques sont enfin suggérés pour clore le modèle principale et ainsi permettre un design de grille pour un profil de turbulence souhaité.

We investigate scalings of turbulence dissipation and turbulence length/time scales in the fully developed turbulent channel flow region of wall distances $y$ where the ratio of turbulence production to turbulence dissipation oscillates close to 1. First, we study averages over both time and wall-parallel streamwise ( $x$ ) and spanwise ( $z$ ) planes at $y$ . Turbulent channel flow data with friction velocity $u_{\tau }$ , and global Reynolds number $Re_{\tau }$ ranging from $550$ to $5200$ , suggest that the integral length scales of streamwise fluctuating velocities along the streamwise direction, and of wall-normal fluctuating velocities along the transverse direction, tend towards scaling with $y$ , and that the respective turbulence dissipation coefficients tend towards being constant with increasing $Re_{\tau }$ . However, the data for integral lengths of transverse fluctuating velocities in the transverse direction suggest that these lengths obey an asymptotic scaling $\sqrt {\delta y}$ (where $\delta$ is the channel half-width) with increasing $Re_{\tau }$ . The corresponding turbulence dissipation's scaling seems to tend towards $\sqrt {Re_{\tau }}/Re_{\lambda }$ , which is reminiscent of the non-equilibrium turbulence dissipation scaling found in boundary-free turbulent flows, $Re_{\lambda }$ being a $y$ -local Taylor-length-based Reynolds number. The data do not exclude minor corrections from these asymptotic scalings, and in fact, suggest finite Reynolds number deviations to them. Second, we remove time averaging and study time-fluctuating averages over wall-parallel planes at $y$ . We find that the time fluctuations of the turbulence dissipation coefficients and the Taylor-length-based Reynolds number are very strongly anti-correlated at all wall distances $y$ considered, reflecting a dominance of turbulent kinetic energy fluctuations at the lower frequencies, but a dominance of both turbulent kinetic energy and turbulence dissipation at the higher frequencies. In the case of the turbulence dissipation coefficient corresponding to the integral length of the wall-normal velocity along the transverse direction, it is possible to determine the cross-over frequency $f_c^*$ between these two behaviours, and we find $f_c^{*}\sim u_{\tau }/y$ for $Re_{\tau } = 950$ , but $f_c^* \sim u_{\tau }/\delta$ for $Re_{\tau }=2000$ , where there is evidence of very-large-scale motions.

Rare events play a crucial role in many physics, chemistry, and biology phenomena, when they change the structure of the system, for instance in the case of multistability, or when they have a huge impact. Rare event algorithms have been devised to simulate them efficiently, avoiding the computation of long periods of typical fluctuations. We consider here the family of splitting or cloning algorithms, which are versatile and specifically suited for far-from-equilibrium dynamics. To be efficient, these algorithms need to use a smart score function during the selection stage. Committor functions are the optimal score functions. In this work we propose a new approach, based on the analogue Markov chain, for a data-based learning of approximate committor functions. We demonstrate that such learned committor functions are extremely efficient score functions when used with the adaptive multilevel splitting algorithm. We illustrate our approach for a gradient dynamics in a three-well potential, and for the Charney–DeVore model, which is a paradigmatic toy model of multistability for atmospheric dynamics. For these two dynamics, we show that having observed a few transitions is enough to have a very efficient data-based score function for the rare event algorithm. This new approach is promising for use for complex dynamics: the rare events can be simulated with a minimal prior knowledge and the results are much more precise than those obtained with a user-designed score function.

International audience; As a new type of the propulsive devices, cycloidal rotor has been attracting more attention recently. The presentwork concentrates on the analysis of the performance and internal flow field of a two-bladed cycloidal rotoroperating at low Reynolds numbers, with special emphasis on understanding how the flow structure affects theperformance at different working conditions. The symmetrical/asymmetrical pitching motions are evaluatedinitially and the primary results show that the asymmetrical pitching with a mean pitching angle of 5◦ improvesthe efficiency of the rotating system. Then, the effect of the chord-to-radius ratio at three different conditions isdiscussed, which shows that the chord-to-radius ratio of 0.45 is the best value to maximise the performance. Theflow pattern, involving the blade-wake interaction, wake-wake interaction, three vortex structures on the bladesurface, roll-up vortices inside the laminar boundary layer, separation bubble, leading-edge vortex, trailing-edgevortex and flow separation vortex, are discussed in detail under various operating conditions. The study on theperformance of a single blade, the forces (lift and drag) distributions, the blade loadings and the near-wall flowson two blades, provide a new understanding of the global performance change of the propeller.

International audience; The goal of the present work is to investigate the influence of several parameters on the performance and flow structures of reversed pitching airfoils. Firstly, the effect of the turbulence model is evaluated, and the results show that the SST transition model has a better prediction in the instantaneous lift coefficient of a reversed pitching airfoil and the transition locations of a stationary airfoil. Then, effects of the pitching angle, pitch-pivot-point, blade profile and morphed leading edge with various deflection angles and positions on the performance, unsteady vortical flow, near-wall transition and trajectory of main vortices, are analyzed. The main results show that both the mean pitching angle and pitching amplitude have the impact on the vortical flows, but depends on the reduced frequency. Then, the delayed flow structure by shifting the pitch-pivot-point from the leading edge (LE) to trailing edge (TE) can be explained by the distribution of the effective attack-of-angle. Moreover, the symmetrical, asymmetrical and inverse asymmetrical airfoils have great effect on the first (FMLC) and second maximal lift coefficients (SMLC). Finally, upward deflected LE decreases the negative lift coefficient while downward morphed LE improves it considerably due to the geometry curvature leading to the large flow separation. In addition, it is observed that the generation of vortices is earlier when the deflection position close to the middle surface. It is believed that this work can provide some guidelines to have a better design of energy devices with oscillating airfoils/hydrofoils.

An earlier version of this paper was presented at the IEEE Sensors 2020 conference and was published in its Proceedings with the title “Aerodynamic wall pressure measurement using a high temperature gradient Pirani micro-sensor” (DOI: 10.1109/SENSORS47125.2020.9278842); International audience; This paper presents and discusses the first demonstration of aerodynamic wall pressure measurement using a thermal Pirani effect based micro-sensor. The thermal micro-sensor is designed as a suspended micro-hot-wire separated from the substrate by a nanoscale gap, and mechanically supported by perpendicular micro-bridges. The micro-sensor was calibrated for pressure measurement from 10 kPa to 800 kPa. It was then used as aerodynamic wall pressure sensor in a turbulent boundary layer wind tunnel and successfully measured the 1.1 kPa depression occurring in the wind tunnel when the flow velocity goes up to 40 m/s.

International audience; In recent years, the idea of using a pump as a left ventricle assist device is being well developedby several groups. Meanwhile, one of the challenges in this field is the occurrence of biologicalphenomena such as hemolysis. From an engineering point of view, a solution to this problem is toprovide an accurate and efficient numerical method to predict the phenomenon. Hemolysis modelsare typically based on equivalent scalar stress and exposure time. This paper aims to study the impactof considering extensional stresses as the main reason for hemolysis in the FDA Nozzle benchmark.The idea comes from an experimental article. First of all, flow’s hemodynamic was validated bycomparing the results of normalized axial velocity in several sections with the experimental data. Inthis case, three different RANS models k− , k−ω and k−ω SST were employed. As expected, it isclear that the k−ω SST is the most accurate model. In the next step, hemolysis simulations performedfor different equivalent stresses. In this case, the impact of scaling up the extensional stresses onpredicted hemolysis is studied by adding a coefficient Cn to equivalent stress. It is concluded that byapplying these new modifications the hemolysis index would be in a reliable range.

In aircraft engines, compressor stages can encounter situations in which the flow is distorted at rotor inlet, for example during particular flight maneuvers, or due to the shape of the inlet of the airframe. The main objective of this paper is to investigate experimentally the effect of an inlet flow distortion on the internal flow dynamic and its consequences on the performance and operability of these machines. The distortion was generated by a porous plate grid installed upstream of the compressor. Eight total pressure rakes placed downstream of the grid were used to evaluate the distortion. Unsteady pressure measurements were performed on the casing at different axial and azimuthal positions to investigate the dynamic flow behavior at nominal conditions, at an operating point close to stall and during the transition to stall. 2D-2C PIV maps, synchronized with the runner position, were obtained on rotor blade-to-blade planes at three different spanwise positions (79%, 51%, 18% of the blade span). Three relative angular positions of the grid to the laser sheet were investigated at two different flow rates, namely the nominal flow rate and a flow rate close to stall. These three positions are corresponding to the laser sheet cutting through i/ the center of the grid’s wake and ii/ the two edges of the grid. The impact of the distortion on the performance of the compressor is analyzed and compared to existing models. The impact of the flow dynamic will also be considered, especially in operations close to the stall limit. The work is supported by DGAC (the French Agency For Civil Aviation) through the SUBLIME research program.

A non-intrusive method to get a multi-element polynomial chaos model is developed. This method is called ME-ACD, for Multi-Element based on Agglomerative Clustering on Derivatives. It aims at approximating a Quantity of Interest which presents discontinuities or irregularities making it difficult to be accurately approximated by standard Polynomial Chaos models. The method permits to efficiently split the parameter space and to train local polynomial models of lower degrees on every element where the local pieces of the Quantity of Interest are smoother. The algorithm is based on agglomerative clustering of the observations in a well-chosen abstract space taking into account the value of the Quantity of Interest and of its derivatives with respect to the stochastic input parameters. The same observations are used for both partitioning the space and training the local models. Several partitions of the parameter space are tested, and the one leading to local models minimizing a crossvalidation error is selected. Once the training observations are labelled with a class number indicating the element they are located in, a neural network classifier is trained to determine which local model to use for further evaluations. The method has proven to efficiently split the parameter space for a set of applications and the piecewise chaos model is compared with a standard Polynomial Chaos non-intrusive method and a Gradient Tree Boosting in terms of accuracy.; Une méthode non-intrusive ayant pour but d’obtenir un modèle par chaos polynomial multi-élément est développée. La méthode se nomme ME-ACD, acronyme Anglais signifiant « méthode multi-élément basée sur la classification hiérarchique ascendante des dérivées ». Le modèle a pour objectif d’approcher les Quantités d’intérêt présentant des discontinuités et/ou des irrégularités altérant la précision des modèles par chaos polynomial standards qui l’approchent. La méthode permet de partitionner efficacement l’espace des paramètres et d’entrainer des méta-modèles par chaos polynomial locaux de degrés inférieurs sur chaque sous-ensemble de l’espace des paramètres où les morceaux de la Quantité d’intérêt est plus régulière localement. L’algorithme repose sur le regroupement hiérarchique ascendant des observations représentées par des points dans un espace abstrait prenant en compte la valeur de la Quantité ainsi que les valeurs de ses dérivées par rapport aux paramètres d’entrée stochastiques. Les mêmes observations sont utilisées pour la partition de l’espace et pour l’entrainement des modèles locaux. Plusieurs partitions de l’espace sont testées, et celle menant à des modèles locaux minimisant une erreur de validation croisée est sélectionnée. Une fois que les observations d’entrainement sont associées à un numéro indiquant dans quel élément elles sont localisées, un réseau de neurones est entrainé pour déterminer quel modèle local sera utilisé pour les évaluations à venir. L’efficacité de la méthode est montrée sur un ensemble d’applications et la précision du modèle par morceaux est comparée avec celle d’un modèle par chaos polynomial standard obtenu par une méthode non-intrusive et celle d’un modèle « Gradient Tree Boosting ».

International audience; This paper focuses on the experimental evaluation of the rotor/propeller interactions for hybrid compound configurations. Experiments are conducted in the ONERA L2 large size-low speed wind tunnel with a 1/7.7 Dauphin 365N model and a four-bladed small-scale propeller. Measurements are realized using two six-component balances, accelerometers, and blade pitch, flap and lead-lag angle sensors. PIV measurements are performed to visualize the velocity fields around the helicopter. Different flight conditions (wind speeds and propeller rotational speeds) have been tested. The comparison of the results obtained with an isolated main rotor and an isolated propeller with the complete assembly highlighted the influence of their interactions on their performances. In hover, the propeller is completely immersed in the rotor wake, and the thrust is therefore maximal. At low speed, the propeller is partially immersed and the flow on the propeller disk is highly asymmetrical. No direct interactions are measured at high speed, where the interactions increase the performances of the propeller due to the increased angle of attack of the flow on the propeller disk.

International audience; As an academic platform, the French Naval Academy Research Institute (IRENav) developed a large-scale experimental cycloidal propeller (800 kg, 0.4 m radius) with the aim of running diverse pitch motions to evaluate performances of cross-flow propellers. Blades’ pitching is here performed by servo-motors in order to control each blade independently. Blade pitch regulation has been improved to obtain a quick response of auxiliary servo-motors. However, pitch tracking validation shows a systemic error due the PI regulation loop which depends on the rotational speed and the considered pitch law (the sharper the law, the higher the error). The platform is widely instrumented with load and torque sensors to measure inflow hydrodynamic forces during the rotation. First experiments have been performed at the IFREMER wave–current flume tank for a range of sinusoidal pitch laws to confirm the post process procedure and to determine performances for a range of advance parameters. Results show that depending on the advance parameter the more different efficient law is: the higher the advance parameter, the lower the sinusoidal amplitude must be for a better efficiency. These results confirm the requirement of an adaptable pitch control for cycloidal propeller to boost their performances.

Stall and surge are strong limitations in the operating range of compressors and thus one of the major limits of jet engine performance. A promising way to push the stability limit of compression machines is to inject a small amount of flow at the blade tip to alter the physical mechanism responsible for stall onset. This study focuses on the experimental performance of such a system. To do so, an axial compressor test bench was equipped with 40 actuators connected to an auxiliary pressurised air supply system. They were able to generate high-speed jet blowing just at the tip of the rotor blades. The opening of each actuator was controlled by an electromagnetic valve. This allowed generating continuous or pulsed jets with frequencies up to 500 Hz at different duty cycles. The performance of the control system was investigated for various control strategies, where the injected flow rate, the injection angle, the number of injectors, the jet frequency and the duty cycle were systematically varied. This paper is concluded by a study of the energy balance of the system for various configurations. To the best of the authors’ knowledge, this constitutes a rarely seen analysis in the literature. This project (ACONIT) has received funding from the Clean Sky 2 Joint Undertaking under the European Union’s Horizon 2020 research and innovation programme under grant agreement No. 886352.

The operating range of axial compressors can be limited by the presence of unstable flow phenomena such as rotating stall, which can severely reduce the surge margin of the engine. These phenomena prevent the compressor to reach high-pressure ratio, and often to obtain better efficiency. They are thus one of the main limitations of jet engines performances [1]. To push the stability limit of the compression machines and improve the Surge Margin, recent studies have shown the benefit of active control or passive control like casing treatments [2] [3]. More recently, one of the promising ways is to inject a small amount of flow at blade tip to alter the physical mechanism of the stall onset. Using active flow control systems, which can easily be turned off and on, is thus a very attractive alternative [4]. Before including control device in a real compressor configuration, investigations on CFD strategies are required to correctly capture the effects of the control system, due to the complexity of the flows involved and the size and geometric complexity of the device. As the final configuration (active flow control in a compressor) is the single stage compressor CME2, located at the Lille Laboratory of Fluid Mechanics (LMFL) [4], the injector device must be studied in similar conditions. Thus, the test bench dedicated to fluid actuator characterization located at ONERA Lille is used for CFD validation. The present study aims at determining a methodology taking account computational domain, mesh, numerical methods, and turbulence modeling. The goal is to optimally capture the behavior of internal flows within the control device consisting in an isolated fluidic actuator on a flat plate, blowing air. As the actuator geometry corresponds to the ones located on the CME2 test bench, numerical set up is characterized to be similar to the experimental conditions ensuring the same actuator mass flow rate and boundary conditions. Another operating condition consists in a jet in free flow configuration, i.e. without boundary layer over the flat plate. This paper is presenting a first research focusing on different types of mesh such as structured, unstructured, and hybrid mesh. Mesh convergence study is carried out for each mesh strategy with specific refinement. Reynolds Averaged Navier-Stokes (RANS) simulations are performed with CFD elsA code, with the same boundary conditions. Then, an analysis on turbulence model effect is also provided, by assessing different models such as Spalart-Allmaras, k-l, k-?, and k-? turbulence models to determine the best results with acceptable convergence. Results of the simulation are compared with those of injector characterization experimental tests obtained on the ONERA test bench, and representative of the case of the CME2. Satisfying agreements between numerical and experimental data are provided with similar flow behavior. In addition, this paper presents a parametric study on the effect of the mesh on the numerical flow fields in the case of a jet blowing in a boundary layer (case corresponding to the final application) and a jet blowing in free flow. These simulations permit to investigate different mesh strategies and the effect on the flow behavior.

International audience; This paper investigates the use of data-driven methods for the reconstruction of unsteady fluid flow fields. The proposed framework is based on the combination of machine learning tools: dimensionality reduction to extract dominant spatial directions from data, reconstruction algorithm to recover encoded data by limited measurements and cross-validation for hyperparameter optimization. For the encoding part, linear and nonlinear extraction of patterns are considered: proper orthogonal decomposition (POD), linear autoencoder (LAE) and variational autoencoder (VAE). For the reconstruction part, regressive reconstruction (neural network, linear, support vector, gradient boosting) and library-based reconstruction are compared, each method being cross-validated to ensure good generalization on testing data. The position of sensors is optimized using an enhanced clustering algorithm. The robustness of regressive reconstructions to noise measurements is also addressed, showing the benefits of variational approaches in the reduction phase. The strategy is tested for three increasing complexity flows: 2D vortex shedding (Re = 200), 2D spatial mixing layer and 3D vortex shedding (Re = 20000). The results suggest that proper machine learning approaches to fluid flow data can lead to effective reconstruction models that can be used for the rapid estimation of complex flows.; Ce papier s'intéresse à l'utilisation de méthodes orientée données pour la reconstruction de champs de vitesse instationnaires. La stratégie proposée est basée sur la combinaison d'outils du machine learning : réduction de dimension pour extraire les directions dominantes de l'écoulement, algorithme de reconstruction pour estimer les données encodées à partir de mesures et validation croisée pour l’optimisation des hyperparamètres. Pour la partie «encodage», des transformations linéaires et non linéaires sont considérées : décomposition en modes propres orthogonaux, auto-encodeur linéaire et auto encodeur variationnel. Pour la partie «reconstruction», une méthode régressive (réseau de neurones, linéaire, machine à vecteurs support, boosting du gradient) et une méthode basée sur une bibliothèque de référence sont comparées. La position des capteurs est optimisée par clustering amélioré. La robustesse au bruit de mesures est étudiée, montrant l'intérêt des approches variationnelles lors de l'étape de réduction. La stratégie est testée sur trois cas de mécanique des fluides : écoulement derrière un cyclindre (Re 200), couche de mélange spatiale (Re 500) et écoulement derrière un cylindre 3D (Re 20000). Les résultats montrent qu'une bonne approche machine learning permet d'apprendre des modèles robustes et efficients pour l'estimation rapide d'écoulements complexes.

International audience; Flow control devices are used within complex intakes to reduce the flow distortion which can adversely impact the stability and performance of embedded engines. There is a need to assess the capability of modern computational methods such as detached eddy based models to compute the unsteady flowfield and to evaluate the potential benefits of flow control devices on the unsteady distortion. This paper investigates the unsteady flowfield for an S-duct using Zonal Detached Eddy Simulations (ZDES) with passive flow control devices modelled with an overlapping Chimera grid method. The ability of ZDES to evaluate the impact of passive flow control devices on the unsteady flow distortion was assessed. The computed unsteady flowfield at the Aerodynamic Interface Plane was compared with experimental data based on total pressure and velocity field measurements. For the baseline configuration, the ZDES model has proven to be able to simulate the unsteady flowfield at the AIP, to provide the time averaged and fluctuating levels of swirl distortion within 1% and 13% respectively of the measurements. The strong impact of the flow control devices on the AIP flowfield was also captured by the ZDES. The overall increase of pressure ratio (PR) at the AIP due to the flow control devices was predicted with less than 1% error. The 65% reduction in swirl distortion fluctuation when the flow control devices are used was predicted within less than 8% error by the ZDES compared with S-PIV measurements. Overall it was determined that the ZDES method is able to simulate the unsteady flow and distortion characteristics for both the baseline reference configuration as well as the case with flow control.

Active flow control is an aerodynamic improvement method that could be applied to turbomachinery to improve their performance. In particular, it could be used to mitigate the long-lasting issue of rotating stall and surge instabilities occurring at some points on the performance curve of axial compressors [1]. Practically speaking, active flow control could be used here to stabilize the flow at high pressure ratios, allowing to achieve better performance, or using less compressor stages [2]. In any cases, interesting gains could be achieved in both fuel consumption and pollutant emission. This paper presents the early phase of the deployment of such an active flow control system in a semi-industrial axial compressor. The machine is the CME2 compressor operated in the Lille Laboratory of Fluid Mechanics (LMFL), and consists in a single stage, 550 mm diameter, axial compressor driven by a 200 kW DC motor. The paper describes first the design, the implementation, and the characterization of the control system itself. As in any flow control study, the parametric space is large, and if there is a near consensus for some parameters in the considered application, there is still some debates about the influence of many of them [3]. Consequently, present control system is intended to be modular, consisting in 20 blocks of two injectors flush mounted on the compressor casing. Each injector consists in 3D printed parts designed to exploit Coand? effect to direct airflow right into the tip gap region at the top of the blades. Various injector shapes can be tested, and yaw angle injection can be investigated as the injectors are able to rotate by 15 steps around their axis. These injectors are feed by magnetic valves, reaching an actuation frequency up to 500Hz, allowing studying both continuous and pulsed jets actuations. When performing flow control experiments, a good knowledge of the control system is needed to estimate its effect on the flow. Second part of this paper present then a detailed characterization of the actuator behavior. First, some results achieved in quiescent air with an isolated actuator are presented. These results have been obtained thank to a dedicated test bench at ONERA Lille, using hot wire anemometry. Some interesting physics link to the coupling between pulsed actuation and Coand? effect have been observed in transient regime, but also an impressive vectorization of the jet along the casing wall. These detailed tests have been completed by an in situ calibration of the whole system with the 40 injection ports, allowing a global verification of their homogeneity. Finally, this paper is concluded by the presentation of some of the results achieved with this system on the compressor in terms of stall margin improvement. The effect of several parameters is presented, with, in particular, an interesting demonstration of the duty cycle as a command parameter allowing a precise control of the surge margin. In addition, the role of the time without blowing occurring with pulsed actuation is also investigated. To conclude this paper, some original physics are presented concerning the stalling pattern of the compressor with and without control, as some clues seem to indicate a shift from a spike type stall inception to a modal one. References 1. Day, I.J. Stall, Surge, and 75 Years of Research. J. Turbomach. 2015, 138, 011001. 2. Tahara, N.; Kurosaki, M.; Ohta, Y.; Outa, E.; Nakajima, T.; Nakakita, T. Early Stall Warning Technique for Axial-Flow Compressors. J. Turbomach. 2007, 129, 448456. 3. Li, J.; Du, J.; Nie, C.; Zhang, H. Review of tip air injection to improve stall margin in axial compressors. Prog. Aerosp. Sci. 2019, 106, 1531.

The open cavity is a common geometry in weapon bays, wheel wells or landing gear doors. The flow developing over such a geometry has been extensively studied since Roshko [1] and complex dynamic phenomena develop in it. The shear layer destabilises over the cavity and impacts the downstream corner of the cavity. The acoustic waves generated from this location propagate upstream the cavity exciting the shear layer instabilities. This results in self-sustained oscillations in the flow, which can induce structural vibrations and noise radiation [2] [3].Therefore damping these oscillations is of interest. We propose a design of micro-valve based on MMMS technology and integrate an array of 15 of these microfluidic device to the upstream wall of an open-cavity, placed in a wind tunnel at the ONERA (Office National dEtudes et de Recherches Aerospatiales), in order to implement an open-loop flow control strategy. The MMMS design consist in a pressurised air inlet, a channel surmounted by a flexible resonator made in elastomer material (Silastic) and a submillimetric outlet. A NdFeB permanent magnet is fixed on the flexible resonator enabling the channel height modulation and hence modulating the outlet velocity, when an electrical current runs in the coil surrounding the magnets. The microvalve can deliver a high-speed jet with an outlet velocity of 150 m/s (with a submillimetric outlet) pulsed at frequencies up to 500Hz [4]. The actuators outlet design was enhanced and adapted to the control of cavity flow. Therefore, the submillimetric outlet was modified in order to obtain an outlet velocity spread over a slot. Performances of these adapted actuators were assessed with hot wire measurements performed at the ONERA Lille. For a chosen flow regime, effects of quasi-steady and pulsed jets on the cavity oscillations were assessed through hot wire and unsteady pressure measurements, performed at the cavity downstream corner, where the cavity dynamics is the most amplified. Depending on the microvalves performances and the flow regime, quasi-steady jets enabled a decrease of 20dB in the cavity oscillations sound pressure level, deduced from the unsteady pressure measurements. Regarding pulsed jets, depending on the forcing induced by the actuators a locking phenomenon was observed. Therefore, the actuators authority proved to be sufficient to shift the cavity natural oscillation frequency towards the microvalves actuation frequency. The open-loop experiments campaign aims at preparing the closed-loop flow control campaign. Therefore, the system transfer function identification was performed with the pulsed jets in the linear limit assumption. Harmonic forcings were used to identify the transfer function in a given flow regime with different forcing amplitudes. References: [1] A. Roshko, Some measurements in a rectangular cutout, National Advisory committee for aeronautics, Technical note 3488, 1955. [2] J. E. Rossiter, Wind tunnel experiments on the flow over rectangular cavities at subsonic and transonic speeds, RAE Technical Report No. 64037, 1964. [3] C.K.W. Tam and P.J.W. Block, On the tones and pressure oscillations induced by flow over rectangular cavities, J. Fluid Mech., vol. 89, peed 2, pp. 373-399, 1978 [4] VIARD R., TALBI A., PERNOD P., PREOBRAZHENSKY V., MERLEN A., Magnetostatic microvalve for high momentum rate pulsed jet generation, Eurosensors XXIII, Lausanne, Switzerland, september 6-9, 2009, Procedia Chem. 1, 1 (2009) 421-424.

International audience; A direct numerical simulation of a moderate adverse pressure gradient turbulent boundary layer flow on a flat plate was performed on a fairly long domain with a Reynolds number from Re θ ∼ 2000 to Re θ ∼ 8000. The mean flow and Reynolds stress statistics are analysed. The mean streamwise velocity profile does not exhibit a clear log region even at the highest Reynolds number. The Reynolds stresses exhibit a well defined peak in the outer region which moves away from the wall before to stabilise near 0.45 boundary layer thickness. These outer peaks are the consequences of an excess of production over dissipation. The large scales structures analysed by means of two point correlations of streamwise fluctuating velocity are shown to be strongly modified by the adverse pressure gradient.

International audience; The leakage flow in the gap between the rotor and the casing of an axial compressor is very complex. It is inducing losses and is at the origin of the arising of rotating stall. In addition, the tip clearance can increase with the machine aging and the induced flow will then be modified. A good prediction of the tip leakage flow and of its evolution with gap size is thus essential to predict correctly the performance and operability of a compressor stage. The support of the study is a single stage compressor for which three different ratio R between the gap and the axial chord are considered (R = 0.6 %, 1.2 %, 2.4 %). Reynolds Averaged Navier-Stokes (RANS) simulations are performed on the three configurations. The results provided by two computational domains (a single blade-channel and one tenth of the full annulus geometry) are compared. The aim is to define the best configuration, which provides reliable results at the most reasonable computational cost. Then, a numerical study on the effect of tip gap size on the performance and flow topology of the machine is conducted. A special attention will be paid on the flow at near stall operating condition.; L'écoulement apparaissant dans le jeu entre le rotor et le carter d'un compresseur axial est très complexe. Il induit des pertes et est à l'origine de l'apparition du décollement tournant. De plus, le jeu peut augmenter avec le vieillissement de la machine et l'écoulement sera alors modifié. Une bonne prédiction de cet écoulement de jeu et de son évolution avec la taille de jeu est donc essentielle pour prévoir correctement les performances et l'opérabilité d'un étage de compresseur. Le support de l'étude est un compresseur à un étage pour lequel trois rapports, R, définissant la taille de jeu sur la corde axiale sont considérés (R = 0.6 %, 1.2 %, 2.4 %). Des simulations Reynolds Averaged Navier-Stokes (RANS) sont effectuées sur les trois configurations. Les résultats fournis par deux domaines de calcul (un seul canal d'aubes et un dixième de la géométrie du compresseur complet) sont comparés. L'objectif est de définir la meilleure configuration, qui fournit des résultats fiables en terme de coût de calcul raisonnable. Ensuite, une étude numérique sur l'effet de la taille de jeu sur les performances et la topologie d'écoulement du compresseur est menée. Une attention particulière sera portée sur l'écoulement en condition de fonctionnement proche du décrochage.

A new method of compressed air energy storage at sea developed by the company SEGULA technologies is presented. REMORA is a storage system based on the quasi isothermal compression of air carried out by liquid pistons in the compression chambers. The latter are supplied by reversible centrifugal pumps which also act as turbines. The global system consists of elements forming a reversible energy cascade that can store energy in the form of compressed air in the underwater reservoirs or to return energy in the form of electricity in the electricity grid. The operating principle of REMORA is to always work at the nominal operating point of the pumps. This is why pumps operate at variable speed to continuously approach its nominal operating point. To optimize the operating range, a low-pressure pump and a high-pressure pump will have to take turns to supply the compression chambers. In addition, hydraulic valves are present at the inlet of each compression chamber and are connected to each pump by pipes. These valves open and close successively over a very short time depending on the state of compression or expansion of the air in each chamber. These valve switching events occur at the same time as rapid accelerations and decelerations of pump rotation speed that can induce transient flow into the system. The ODySEA demonstrator of the REMORA project with a few kW average power was built. Its objective is to analyse and simulate the various functions of the circuit during transient period. The experimental results are analysed in order to observe transient phenomena related to the rapid dynamics of the system, to verify the reliability of the REMORA concept, to detect possible problems and to remove the technological barriers. CFD modelling of the low-pressure pump is performed using STAR ccm+ calculation code in order to analyse in details the fluid flow in both modes (pump and turbine). The CFD model makes it possible to set up an elaborate pump model taking into account the different losses in the pump intended for the bond graph model. Finally, a dynamic model of the operation of the ODySEA hydraulic system, using the Bond Graph code, is proposed. A partial model, built on a simplified hydraulic circuit with two pumps and two liquids pistons, was developed and validated using the experimental results., Une nouvelle méthode de stockage d’énergie par air comprimé basé en mer développé par l’entreprise Segula technologies est présentée. REMORA est un système de stockage basé sur la compression de l’air de manière quasi isotherme effectuée par des pistons liquides dans des chambres de compression. Ces dernières sont alimentées par des pompes centrifuges réversibles qui vont également travailler en mode récepteur en assumant le rôle de turbines. Le système global est composé d’éléments formant une cascade d’énergie réversible pouvant stocker de l’énergie sous forme d’air comprimé dans les réservoirs sous-marins ou de restituer l’énergie sous forme d’électricité dans le réseau électrique. Le principe de fonctionnement de REMORA est de toujours travailler au point de fonctionnement nominal des pompes. C’est pourquoi, les pompes fonctionnent à vitesse variable pour se rapprocher de manière continue de son point de fonctionnement nominal. Pour optimiser la plage de fonctionnement, une pompe basse pression et une pompe haute pression vont devoir se relayer pour alimenter les chambres de compression. De plus, des vannes hydrauliques sont présentes à l’entrée de chaque chambre de compression et sont reliées à chaque pompe par des conduites. Ces vannes s’ouvrent et se ferment successivement sur un temps très court en fonction de l’état de compression ou de détente de l’air dans chaque chambre. Ces commutations de vannes se produisent en même temps que des accélérations et des décélérations rapides de la vitesse de rotation des pompes pouvant induire l’apparition d’un écoulement transitoire dans le système. Le démonstrateur ODySEA du projet REMORA, d’une puissance de quelques kilowatts, a été construit. Son objectif est d’analyser et de simuler les différents fonctionnements du circuit en mode instationnaire. Les résultats expérimentaux sont analysés afin de d’observer les phénomènes instationnaires liés à la dynamique rapide du système, de vérifier la fiabilité du concept REMORA, de détecter les éventuels problèmes et de lever les verrous technologiques. Une modélisation CFD de la pompe basse pression est faite à l’aide du code de calcul STAR ccm+ dans le but d’analyser finement l’écoulement du fluide dans la pompe dans les deux modes (pompe et turbine). Le modèle CFD permet de mettre en place un modèle de pompe élaboré prenant en compte les différentes pertes dans la pompe destinée au modèle bondgraph. Enfin un modèle dynamique du fonctionnement du circuit hydraulique de ODySEA, à l’aide du code Bond Graph, est proposé. Un modèle partiel, construit sur un circuit simplifié comprenant deux pompes et deux pistons a été élaboré et validé à l’aide des résultats expérimentaux.

Nowadays, the variable-pitch turbines and propellers are extensively employed to many engineering equipment, such as the vertical-axis wind turbine, underwater propulsion system and unmanned air vehicle. As a novel type of variable-pitch devices, cycloidal propellers, comprising several blades oscillating with sinusoidal motions, have been attracting more attention. The principle of the cycloidal rotor includes two basic motions: one is the rotation based on the centre of the shaft and the other is the blade pitching motion over a pivot-point along the chord line. The lift and propulsive force of the cycloidal propeller is provided by the components of the lift and drag on the blades, due to the variation of the relative angle-of-attack. The flow structures inside the cycloidal propeller are extremely complicated, involving several kinds of blade-wake and wake-wake interactions and different vortices on the blade surfaces, which depends on the geometrical parameters and working conditions significantly. For a better design of such device, it is of great importance to understand the flow physics of the cycloidal propeller and how it affects the performance under different working conditions. In this thesis, the unsteady vortical flows over a two-bladed cycloidal rotor are simulated using the SST γ-(Re) ̃_θttransition model. The main contributions of the present work can be summarized as follows: (1) calibration of the SST γ-(Re) ̃_θt transition model over a single airfoil, in terms of the mesh distribution, inlet turbulence condition, correlations and parameters in the transition model, and then comparing the results with the experiments at different incidences and Reynolds numbers; (2) applying the optimized transition model to a cycloidal rotor at two advance coefficients, to investigate the internal flow structures in detail, mainly including the blade-wake interaction, wake-wake interaction, near-wall vo rtex flows and laminar-turbulence transition; (3) optimizing the cycloidal propeller, including the pitching kinematic, chord-to-radius ratio, pitch-pivot-point and blade profile, and explaining how the global performance changes at different Reynolds numbers and advance coefficients, based on the analysis of the performance of the single blade, forces (lift and drag), pressure coefficients and near-wall flows on two blades.; De nos jours, les turbines et hélices à pas variable sont largement utilisées dans de nombreux équipements d'ingénierie, tels que l'éolienne à axe vertical, le système de propulsion sous-marin et les véhicules aériens sans pilote. En tant que nouveau type de dispositifs à pas variable, les hélices cycloïdales, comprenant plusieurs pales oscillant avec des mouvements sinusoïdaux, ont attiré plus d'attention. Le principe du rotor cycloïdal comprend deux mouvements de base : l'un est la rotation basée sur le centre de l'arbre et l'autre est le mouvement de tangage de la pale sur un point de pivot le long de la ligne de corde. La portance et la force de propulsion de l'hélice cycloïdale sont fournies par les composants de la portance et de la traînée sur les pales, en raison de la variation de l'angle d'attaque relatif. Les structures d'écoulement à l'intérieur de l'hélice cycloïdale sont extrêmement compliquées, impliquant plusieurs types d'interactions pale-sillage et sillage-sillage et différents tourbillons sur les surfaces des pales, ce qui dépend de manière significative des paramètres géométriques et des conditions de travail. Pour une meilleure conception d'un tel dispositif, il est d'une grande importance de comprendre la physique de l'écoulement autour du propulseur cycloïdal et de déterminer comment elle affecte les performances du propulseur, dans différentes conditions de travail. Dans cette thèse, les écoulements tourbillonnaires instationnaires sur un rotor cycloïdal bipale sont modélisés en utilisant le modèle de turbulence de transition SST k-(Re) ̃_θt. Les principales contributions du présent travail peuvent être résumées comme suit : (1) étalonnage du modèle de transition SST k-(Re) ̃_θt sur un profil aérodynamique fixe isolé par étude d'influence des paramètres de distribution du maillage, des conditions limites de turbulence à l'entrée, des corrélations et des paramètres du modèle de transition et pa r comparaison aux résultats expérimentaux à différentes incidences et différents nombres de Reynolds ; (2) application du modèle de transition optimisé à un rotor cycloïdal pour deux coefficients d'avance, pour étudier en détail les structures d'écoulement internes, notamment l'interaction pale-sillage, l'interaction sillage-sillage, les écoulements vortex près de la paroi et la transition laminaire-turbulence ; (3) optimisation des paramètres du rotor cycloïdal : de la cinématique de tangage, du rapport corde/rayon, de la position du point de pivot, du pas et du profil de pale. Analyse des performances globales pour différents nombres de Reynolds et coefficients d'avance, sur la base de l'analyse des performances de la pale unique, forces (portance et traînée), coefficients de pression et écoulements près de la paroi sur deux pales.

The main objective of this work is to analyze the effects of a moderate adverse pressure gradient on the dynamics of turbulent boundary layer flows. For that purpose, a direct numerical simulation (DNS) of the turbulent boundary layer (TBL) subjected to a moderate adverse pressure gradient (APG) out of equilibrium has been performed using the open-source code Incompact3d up to a Reynolds number of 8000 based on momentum thickness. A large database resolved in time and space was collected and used to analyze the turbulence statistics. Special attention has been paid to the existence and evolution of the outer peak of Reynolds stresses observed in APG wall-bounded flows. Different velocity scalings have been investigated and tested against the numerical results. The velocity scale based on the shear stress is shown to scale all the Reynolds stresses profiles for different Reynolds numbers, indicating that all Reynolds stresses are associated with a single dynamics of turbulent structures.The large-scale coherent structures of the streamwise velocity fluctuations have been investigated using two-point spatial correlation. A comparison with a zero pressure gradient case at an equivalent Reynolds number allows us to further investigate the effect of the pressure gradient on the size and inclination of attached coherent structures. A deeper investigation of the coherent structures was also performed, where each structure was detected separately based on a thresholding method to distinguish between the effects of large and small scales and to better understand the mechanisms controlling the dynamics of these structures. The contribution of large-scale motions (LSM) on the Reynolds stresses comparing with ZPG case was also analyzed.; L'objectif principal de ce travail est d'analyser les effets d'un gradient de pression défavorable modéré sur la dynamique d'écoulement d'une couche limite turbulente. Dans ce contexte, une simulation numérique directe (DNS) de la couche limite turbulente (TBL) soumise à un gradient de pression défavorable modéré (APG) hors équilibre a été réalisée jusqu'à un Reynolds de 8000 en utilisant le code open-source Incompact3d. Une large base de données résolues en temps et en espace a été collectée et utilisée pour analyser les statistiques de la turbulence. Une attention particulière a été consacrée à l'existence et à l'évolution du pic de contraintes de Reynolds observé dans la zone externe de la couche limite. Différentes échelles de vitesse ont été étudiées, testées et confrontées aux résultats numériques. L'échelle de vitesse basée sur la contrainte de cisaillement permet de mettre à l'échelle tous les profils de contraintes de Reynolds pour plusieurs nombres de Reynolds, ce qui indique que toutes les contraintes de Reynolds sont associées à une dynamique unique des structures turbulentes.Les structures cohérentes à grande échelle des fluctuations de vitesse longitudinales ont été étudiées en utilisant la corrélation spatiale en deux points. Une comparaison avec un cas sans gradient de pression à un nombre de Reynolds équivalent nous permet d'étudier l'effet du gradient de pression sur la taille et l'inclinaison des structures cohérentes attachées. Une étude approfondie sur les structures cohérentes a également été réalisée, où chaque structure a été détectée séparément en utilisant une méthode de seuillage afin de distinguer les effets des grandes et petites échelles et de mieux comprendre les mécanismes qui contrôlent la dynamique de ces structures. La contribution des mouvements de grande échelle (LSM) sur les contraintes de Reynolds en comparaison avec le cas ZPG a également été analysée.

International audience; The present paper is devoted to the analysis of the various instabilities of cavitation attached to a two-dimensional (2D) profile. Time resolved stereo Particle Image Velocimetry (PIV) was conducted in a small-scale 2D venturi type section, in different vertical planes in the streamwise direction, located at varying positions in the depth of the channel. These experiments enabled to obtain the time evolution of the three components of the velocity field in the cavitation area, and to derive the time-averaged gradients in the spanwise direction. Test cases at various Reynolds number were conducted, maintaining either the pressure or the cavitation number constant, to discuss the impact of these parameters on the flow. Then, the attention was focused on three distinct flowdynamics, namely sheet cavitation, where no large-scale instability can be detected, single cloud cavitation, where a large cloud of vapor is shed periodically at the rear of the cavity, and multi-cloud cavitation, where the process is more complex, as more than one clouds are shed downstream. The data reveal that the structure and the structure of the re-entrant jet, which is one of the primary mechanisms of cloud cavitation, is more complex than reported in the previous studies. Although the jet can be detected as an intermittent low speed reverse flow in the streamwise direction, it is actually made of successive vortices about the channel depth, which are convected downstream while expanding in the vertical direction, causing the cavity lift and thus contributing to its final split and the cloud shedding

International audience; The pitching airfoils, applied to the vertical axis turbines and propellers, are critical to extract more energy from the environment. At retreating side, when the airfoil blunt leading edge becomes the trailing edge, the transition and vortex dynamics are quite different from that at advancing side. The goal of the present work is to investigate the transition and vortex evolution over the reversed pitching airfoil, with main focus on the parametrical effect, including the mean pitching angle and pitching amplitude, reduced frequency and Reynolds number. The main results show that the flow structure on the reversed airfoil is more complex compared with that over the forward airfoil due to the earlier flow separation near the sharp leading edge. Then, the transition on the reversed airfoil firstly occurs within the separated shear layer near the sharp leading edge, and then the flow reattaches, leading to the generation of the leading-edge vortex. Near the blunt trailing edge, the second transition appears on two sides, resulting in the asymmetrical boundary layer as the incidence increases continuously. This event is totally different from that on the forward airfoil, shown by the transition always moving from the trailing edge to the leading edge. The flow unsteadiness of the reversed airfoil is mainly induced by the separated shear layer and leading-edge vortex, which is greatly affected by different parameters. Besides, the trajectory of some specific vortices also depends on the working conditions significantly. It is believed that this work can deepen the understandings of underlying flow physics of the reversed airfoils.

Fluid mechanics is an important part in industrial questions. The modelisation, reduction and control of fluid flows require to solve a nonlinear and multiscale optimisation problem. In the age of artificial intelligence, there is a craze to solve these optimisation problems using the wealth of experimental and numerical data. In this context, the objective of the manuscript is to present how machine learning tools can be used for the data-driven estimation of fluid flow velocity fields. In particular, we aim at reducing, reconstructing and predicting four increasing complexity flows: the laminar wake of a cylinder, a spatial mixing layer, the turbulent wake of a square cylinder and the flow around an isolated tower. To do so, we start by reformulating the reduction, the reconstruction and the prediction problems. For the reduction, we use autoencoders to find a latent space and rewrite the flow with with a low rank approximation. Different methods are used: proper orthogonal decomposition, linear autoencoder, linear variational autoencoder and variational autoencoder. For the reconstruction, we use supervised learning tools to estimate the latent state from limited measurements of the fluctuating velocity field. The methods include the linear regression (in its multitask and single task formulation), the support vector regression (linear regression in higher dimensional space), the shallow neural network and gradient boosted trees. For the prediction, we search for a finite approximation of the Koopman operator to linearly advance in time observations of the latent state. Different refinements of the core dynamical mode decomposition algorithm are used: direct models, higher order DMD (use of delayed coordinates), extended DMD and DMD with dictionnary learning. The results show that machine learning is a promising direction to establish fluid flow estimation models. However, the deployment of models for highly turbulent flows remains questionable mainly because of robustness issues.; La mécanique des fluides est présente dans de nombreuses thématiques industrielles telles que la santé, le transport et l'énergie. Pour modéliser, contrôler et réduire les écoulements d'intérêt, il faut résoudre des problèmes d'optimisation non linéaires et multi-échelles. Face aux difficultés de résolution analytique et dans l'âge du Big Data, on cherche à utiliser la disponibilité des données numériques et expérimentales pour faire l'optimisation à partir des données. Dans ce contexte, la thèse s'intéresse aux outils de l'apprentissage automatique pour l'estimation orientée données d'un écoulement. En particulier, on cherche à réduire, reconstruire et prédire le champ de vitesse d'écoulements de complexité croissante: le sillage laminaire d'un cylindre, une couche de mélange spatiale, le sillage turbulent d'un cylindre rectangulaire et l'écoulement autour d'une tour isolée. Pour ce faire, on commence par reformuler les problèmes de réduction, de reconstruction et de prédiction. Pour la réduction, on s'intéresse à l'utilisation d'auto-encodeurs pour déterminer un espace latent de dimension réduite dans lequel on peut réécrire le film de l'écoulement. On utilise en particulier la décomposition en modes propres orthogonaux et différents auto-encodeurs neuronaux. Concernant la reconstruction, on utilise l'apprentissage supervisé pour estimer l'état latent à un instant donné à partir de mesures du champ de vitesse fluctuante au même instant. Les méthodes incluent la régression linéaire (monô et multi tâches), la régression par vecteurs supports, la régression par réseau de neurones superficiel et les arbres de décisions boostés. Pour la prédiction, on cherche une approximation finie de l'opérateur de Koopman pour avancer linéairement dans le temps des observations choisies de l'état latent. Les méthodes sont des raffinements de l'algorithme DMD classique: utilisation de modèles directs, DMD étendue, DMD d'ordre élevé et apprentissage des observables. Les résultats mettent en évidence que l'apprentissage automatique est une piste pertinente pour établir des modèles d'estimation d'écoulements. Toutefois, les développements restent académiques et le déploiement de modèles sur des configurations très turbulentes reste peu envisageable, notamment pour des questions de robustesses (à de nouvelles configurations, au bruit, etc.)

International audience; The denoising of Lagrangian trajectories based on regularized B-spline is investigated. The aim is to find systematic criteria for optimization of algorithms used in 4D-PTV in order to optimize the quality of 4D-PTV measurements of turbulent flows as well as high-order of turbulence statistics. We introduce and adapt to this context two innovative tuning strategies which are commonly used in the Tikhonov regularization of inverse problems based on -curve shape and Normalized Cumulative Periodogram (NCP). The corresponding strategies are tested on synthetic Lagrangian trajectories computed from Direct Numerical Simulation with additional white Gaussian noise. Error-based quantities like Signal-to-Noise Ratio as well as statistical Lagrangian quantities are investigated to compare the different strategies. We then apply the algorithm to experimental data from a 4D-PTV Lagrangian measurements in a turbulent Von Kármán flow. We show the ability of those strategies to optimize the quality of the signal compared to conventional methods. Moreover, the strategies are more adaptable to real experimental noise.

Numerical modeling of Turbulence is one of the classical approaches for studying active scales dynamic in a turbulent flow. The interactions between the different scales of motion featured by any turbulent flow is currently mathematically and numerically intractable. Instead numerical reduced models have been proposed : large-scale motions are resolved using the large eddy equations whereas small-scale influence is modeled and injected into the large-scale dynamic.Numerical modeling is brought to grow in complexity along the progress made in connected fields (such as Algorithmics), the computer power increase and theoretical understanding of the physical phenomena lying beneath the Navier-Stokes equations. This background leads to the coupling of computational fluid mechanics (CFD) and to methods based on artificial intelligence (AI) aiming to model non solved scales impacts on large scale motions. This work brings to the CFD community a corpus of definitions and methods promoting a more optimal use of the AI potential on one hand. On the other hand, we provide a framework allowing to deal with the variety of scales. We discuss of the neural network main components such as nonlinearities induced by activation function, cost functions as well as network's weights optimization algorithms. Properties, expectations and limitations of these hyper-parameters are studied. Based on that work, we combine the Mish activation function with a custom multi-scale cost function with the Nadam network optimization algorithm. Deep convolutionnal neural network models have to be cautiously designed. We therefore provide elements to understand the complexity of the learning task. We go from pattern abstractions to information flow, dealing with vanishing gradients and so on. This discussion is based on the U-net architecture we used. Our works highlight the fact that learning the stress tensor that describes the non resolved - resolved/non resolved interactions from the almost raw 3D filtered velocity field is possible and is one the most accurate solutions that's been proposed to date. This is true both on the prediction itself and on the derived quantities - its divergence and its contraction with the filtered strain rate tensor. The AI based quantities correlations with their expected counterparts oscillates between 90 and 99% outperforming the Clark model. This performance isn't bound to the simulation used for the learning nor to its Reynolds; we show that the machine learning based model accuracy doesn't seem to be affected by the increase on the Reynolds number of 50 or almost 200%. To perform a deeper analysis and try to ``physically'' interpret results, we perform Leonard's decomposition to identify three parts in the stress tensor. The interactions between resolved and non resolved scales carried by the two first terms are better apprehended by the neural network than the Clark model; the predicted tensor captures much more information about interactions between non resolved scales than the algebraic model. This confirms that the machine learning model found a physics compatible mapping between filtered velocity field and stress tensor. Throughout the manuscript, we tried to give a critical point of view on the methods and their results, theirs involvements or their validity. This coupling widens CFD scope and proposes to overcome the limitation of our current understanding of the physics, of current computer capacity to construct faster and more reliable numerical models., La modélisation numérique de la turbulence est une des approches classiques pour étudier la complexité de la dynamique des échelles actives dans un écoulement turbulent. La richesse des mécanismes œuvrant dans ces écoulements rend quasiment impossible la prise en compte de l'ensemble des phénomènes inter-échelles. Des méthodes numériques simplifiées ne résolvant que la dynamique aux grandes échelles sont proposées ; l'influence des petites échelles sur la dynamique des grandes échelles étant introduite par une modélisation physique « sous mailles ». Ces modèles numériques se sont complexifiés et améliorés parallèlement aux progrès de l'algorithmie, des capacités des ordinateurs et de la compréhension des phénomènes physiques décrits par les équations de Navier-Stokes. C'est dans ce contexte qu'est né le développement de méthodes numériques couplées avec l'intelligence artificielle (IA) afin de tenir compte de manière plus fiable de la physique non résolue. Les travaux développés dans ce manuscrit apportent à la communauté de la mécanique des fluides un corpus de définitions et de méthodes permettant une exploitation plus optimale du potentiel des IA en adéquation avec la pluralité et la richesse des échelles que les écoulements turbulents exhibent. Nous discutons des composants principaux des réseaux de neurones tels que les non-linéarités induites par les fonctions d'activation, les fonctions de coût ainsi que les algorithmes d'optimisation des poids du réseau. Les propriétés, les attentes et les limites de ces hyperparamètres sont étudiées pour les optimiser. Nous avons ainsi combiné l'activation Mish avec une fonction coût multi-échelle tout en optimisant les poids du réseau avec l'algorithme Nadam. Les réseaux à convolutions profonds nécessitent une construction minutieuse. À ce titre, nous apportons les éléments nécessaires pour comprendre la complexité de l'apprentissage. Nous traitons de l'abstraction des patterns identifiés, à la propagation de l'information dans le réseau, en passant par d'autres concepts. Cette conversation s'articule autour des particularités de l'architecture U-net utilisée dans ce manuscrit. Ces travaux montrent qu'il est possible d'apprendre la relation entre le champ de vitesse 3D filtré brute, et les composantes du tenseur de stress décrivant les interactions entre échelles résolues et non résolues. C'est aussi une des solutions les plus précises proposée à ce jour. En effet, que nous regardions la prédiction ou les grandeurs dérivées - sa divergence et sa contraction avec le champ de déformations - il apparaît que les corrélations de ces grandeurs avec leurs homologues théoriques oscillent entre 90 et 99% surclassant le modèle de Clark. Il s'avère également que les performances ne sont que très peu perturbées par une augmentation de 50 ou de quasiment 200% du Reynolds. Afin d’affiner l'analyse vers l'interprétation physique des résultats, nous décomposons le tenseur prédit en trois tenseurs conformément à l'idée Leornard. Les interactions entre les échelles résolues et non résolues portées par les deux premiers termes sont mieux appréhendées par le réseau de neurones que par la modélisation de Clark. Le tenseur prédit montre une meilleure appréhension des interactions entre les échelles non résolues que le modèle de Clark ne le permet. Cela confirme que l'IA a construit une correspondance physiquement viable entre champ de vitesse filtré et tenseur de stress. Au fil de ce manuscrit, nous tentons d'apporter un point de vue critique aussi bien sur les méthodes que sur les résultats, leurs implications et leurs étendues. Le modèle hybride présenté s'inscrit dans un domaine en pleine expansion autour de la mécanique des fluides. Il élargit l'application de la modélisation numérique de la Turbulence au-delà de notre compréhension, intuition de la Physique, ou encore des capacités des ordinateurs actuels en construisant des modèles numériques plus rapides et plus fiables.

International audience; Flow control techniques can be defined as methods changing a flow natural dynamics. The control of a flow can be achieved with actuators affecting the flow momentum. Therefore, a MMMS microvalve was designed and characterized. The micro-actuators have then been integrated to a wind tunnel on the upstream corner of a cavity. Open-loop flow control experiments have been performed.

International audience; We experimentally study how the turbulent energy dissipation rate scales in the cross-stream direction of turbulent wake flows generated by two side-by-side square prisms. We consider three different such turbulent flows with gap ratios $G/H=1.25$ , $2.4$ and $3.5$ , where $G$ is the distance between the prisms and $H$ is the prism width. These three flows have a very different dynamics, inhomogeneities and large-scale features. The measurements were taken with a multi-camera particle image velocimetry system at several streamwise locations between $2.5H$ and $20H$ downstream of the prisms. After removing the large-scale most energetic coherent structures, the normalised turbulence dissipation coefficient $C'_{\epsilon }$ of the remaining incoherent turbulence is found to scale as $C'_{\epsilon } \sim (\sqrt {Re_{L}}/Re'_{\lambda })^{3/2}$ along the highly inhomogeneous cross-stream direction for all streamwise locations tested in all three flows and for all three inlet Reynolds numbers considered; $Re'_{\lambda }$ and $Re_{L}$ are, respectively, a Taylor length-based and an integral length-based Reynolds number of the remaining incoherent turbulence.

International audience; The unsteady vortical flows and separation-induced transition over a 2-bladed cycloidal rotor are investigated numerically at two advance coefficients, with special emphasis on the influence of two different turbulence models, namely the original SST k-ω model and SST γ-Reθt transition model. Firstly, the applicability of the transition model in predicting the transitional flows over a single airfoil at different incidences is evaluated, and the main results show a good agreement with the experiments, in terms of the mean pressure coefficients and transition locations. Then, the global performance and detailed internal flow structures of the cycloidal rotor are also used to compare with the existing numerical and experimental data. The primary results show that increasing the advanced coefficient can’t change the transition location of the performance of the single blade, but the magnitudes of these variables (vertical force, propulsive force and power coefficients). Then, combining the forces (lift and drag) acting on two blades, the blade loadings and the detailed near-wall flows, the difference of the vertical force and propulsive force of the rotating system and single blade are clarified clearly for two turbulence models. Finally, at advancing side, the transition and its evolution on a single blade is elaborated. It shows that the SST γ-Reθt transition model is superior in predicting the overall performance, but is highly subjected to the disturbances, characterized by the large-scale vortex structures and massive flow separation, compared with SST k-ω model. Simultaneously, it has the capability to capture the transition process, from growing waves of the laminar boundary layer induced by the roll-up vortices to the fully generation of the separation bubble. It believes that this work can deepen the understandings of underlying flow physics inside the cycloidal rotor at low Reynolds number.

This work aims to investigate the scale-by-scale energy cascade in wall-bounded turbulentflows. Karman-Howarth-Monin-Hill (KHMH) equation is an evolution equation of δu2,which is directly linked to energy content in the scale space and incorporates different processes associated with energy transfers in both physical and scale-space (cascade).The peak of the Spatio-temporal average of cascade term scales with the modified Taylor microscale in the region away from the wall. The instantaneous modified time derivative term and pressure term correlates strongly with the two energy transfer terms. A positive value of the Spatio-temporal average of cascade term observed at the buffer layer moves tohigher spanwise scales as the wall distance increases, suggesting a combination of inverse cascade and wall-attached eddy in the physics of the flow. The use of two independent Stereoscopic PIV experiments allowed to denoise the statistics, which helped to compute some parts of KHMH equation terms in the ZPG-TBL flows experiment, thereby enablingmeasurements close to the wall at higher Reynolds numbers.; Ce travail consiste à étudier la cascade d’ énergie échelle par échelle dans les écoulements turbulents limités par des parois. L’ équation de Karman-Howarth-Monin-Hill (KHMH)est une équation d’ évolution de δu2, qui est directement liée au contenu énergétique dans l’espace des échelles et intègre différents processus associés aux transferts d’énergie dans l’espace physique et l’espace d’échelle (cascade). Le pic de la moyenne spatio-temporelle du terme cascade se met `a l’ échelle avec la micro- échelle de Taylor modifié dans une région éloignée de la paroi. Le terme de dérivée temporelle modifiée instantanée et le terme de pression sont fortement corrélés avec les deux termes de transfert d’ énergie. Une valeur positive de la moyenne spatio-temporelle du terme de cascade observée au niveau de la zone tampon se déplace vers des échelles plus élevées dans le sens de l’envergure à mesure que la distance à la paroi augmente, ce qui suggère une combinaison de cascade inverse et de tourbillon attaché à la paroi dans la physique de l’ écoulement. L’utilisation d’une expérience de double PIV stéréoscopiques indépendantes a permis de d ébruiter les statistiques, ce qui a aidé à calculer certaines parties des termes de l’équation KHMHdans l’expérience des flux ZPG-TBL, permettant ainsi des mesures près de la paroi à des nombres de Reynolds plus élevés.

International audience; In this paper we investigate whether the features of the non-equilibrium cascade, which have been identified in recent studies using high-fidelity tools, can be captured in the case of the classical dissipation scaling by turbulence closures based on the statistical description of freely decaying isotropic turbulence. Numerical results obtained using the EDQNM model over a very large range