The ultimate goal of this work is to facilitate the design of gas turbine engine particle separators by reducing the computational expense to accurately simulate the fluid flow and particle motion inside the separator. It has been well-documented that particle ingestion yields many detrimental impacts for gas turbine engines. The consequences of ice particle ingestion can range from surface-wear abrasion to engine power loss. It is known that sufficiently small particles, characterized by small particle response times (τp), closely follow the fluid trajectory whereas large particles deviate from the streamlines. Rather than manually deriving how the particle acceleration varies from the fluid acceleration, this work chooses to implicitly derive this relationship using machine learning (ML). Inertial particle separators are devices designed to remove particles from the engine intake flow, which contributes to both elongating the lifespan and promoting safer operation of aviation gas turbine engines. Complex flows, such as flow through a particle separator, naturally have rotation and strain present throughout the flow field. This study attempts to understand if the motion of particles within rotational and strained canonical flows can be accurately predicted using supervised ML. This report suggests that preprocessing the ML training data to the fluid streamline coordinates can improve model training. ML models were developed for predicting particle acceleration in laminar, fully rotational/irrotational flows and combined laminar flows with rotation and strain. Lastly, the ML model is applied to particle data extracted from a Computational Fluid Dynamics (CFD) study of particle-laden flow around a louver-geometry. However, the model trained with particle data from combined canonical flows fails to accurately predict particle accelerations in the CFD flow field. / Master of Science / Aviation gas turbine engine particle ingestion is known to reduce engine lifespans and even pose a threat to safe operation in the worst case. Particles being ingested into an engine can be modeled using multiphase flow techniques. Devices called inertial particle separators are designed to remove particles from the flow into the engine. One challenge with designing such a separator is figuring out how to efficiently expel the small particles from the flow while not unnecessarily increasing pressure loss with excessive twists and turns in the geometry. Designers usually have to develop such geometries using multiphase flow computational fluid dynamics (CFD) that solve the fluid and particle dynamics. The abundance of data associated with CFD, and especially multiphase flows make it an ideal application to study with machine learning (ML). Because such multiphase simulations are very computationally expensive, it is desirable to develop "cheaper" methods. This is the long term goal of this work; we want to create ML surrogates that decrease the computational cost of simulating the particle and fluid flow in particle separator geometries such that designs can be iterated more quickly. In this work we introduce how artificial neural networks (ANNs), which are a tool used in ML, can be used to predict particle acceleration in fluid flow. The ANNs are shown to learn the acceleration predictions with acceptable accuracy for the training data generated with canonical flow cases. However, the ML model struggles to become generalizable to actual CFD simulations.

No description available.

Acoustic noise pollution is an environmental aggressor in everyday life. Aero- dynamically generated noise annoys and was linked with health issues. It may be caused by high-speed turbulent free flows (e.g. aircraft jet exhausts), by airflow interacting with solid surfaces (e.g. fan noise, wind turbine noise), or it may arise within a confined flow environment (e.g. air ventilation systems, refrigeration systems). Hence, reducing the acoustic noise levels would result in a better life quality, where a systematic approach to decrease the acoustic noise radiation is required to guarantee optimal results. Computational predic- tion methods able to provide all the required flow quantities with the desired temporal and spatial resolutions are perfectly suited in such application areas, when supplementing restricted experimental investigations. This thesis focuses on the use of numerical methodologies in compressible flow applications to understand aerodynamically noise generation mechanisms and to assess technologies used to suppress it. Robust and fast steady-state Reynolds Averaged Navier-Stokes (RANS) based formulations are employed for the optimal design process, while the high fidelity Large Eddy Simulation (LES) approach is utilized to reveal the detailed flow physics and to investigate the acoustic noise production mechanisms. The employment of fast methods on a wide range of cases represents a brute-force strategy used to scrutinize the optimization parameter space and to provide general behavioral trends. This in combination with accurate simulations performed for particular condi- tions of interest becomes a very powerful approach. Advance post-processing techniques (i.e. Proper Orthogonal Decomposition and Dynamic Mode Decomposition) have been employed to analyze the intricate, highly turbulent flows. The impact of using fluidic injection inside a convergent-divergent nozzle for acoustic noise suppression is analyzed, first using steady-state RANS simulations. More than 250 cases are investigated for the optimal injection location and angle, amount of injected flow and operating conditions. Based on a-priori established criteria, a few optimal candidate solutions are detected from which one geometrical configuration is selected for being thoroughly investigated by using detailed LES calculations. This allows analyzing the unsteady shock pattern movement and the flow structures resulting with fluidic injec- tion. When investigating external fluidic injection configurations, some lead to a high amplitude shock associated noise, so-called screech tones. Such unsteady phenomena can be captured and explained only by using unsteady simulations. Another complex flow scenario demonstrated using LES is that of a high ve- locity jet ejected into a confined convergent-divergent ejector (i.e. a jet pump). The standing wave pattern developed in the confined channel and captured by LES, significantly alters the acoustic noise production. Steady-state methods failed to predict such events. The unsteady highly resolved simulations proved to be essential for analyzing flow and acoustics phenomena in complex problems. This becomes a very powerful approach when is used together with steady-state, low time-consuming formulations and when complemented with experimental measurements. / <p>QC 20141202</p>

Turbochargers play an important role in increasing the energetic efficiency andreducing emissions of modern power-train systems based on downsized recipro-cating internal combustion engines (ICE). The centrifugal compressor in tur-bochargers is limited at off-design operating conditions by the inception of flowinstabilities causing rotating stall and surge. They occur at reduced enginespeeds (low mass flow rates), i.e. typical operating conditions for a betterengine fuel economy, harming ICEs efficiency. Moreover, unwanted unsteadypressure loads within the compressor are induced; thereby lowering the com-pressors operating life-time. Amplified noise and vibration are also generated,resulting in a notable discomfort. The thesis aims for a physics-based understanding of flow instabilities andthe surge inception phenomena using numerical methods. Such knowledge maypermit developing viable surge control technologies that will allow turbocharg-ers to operate safer and more silent over a broader operating range. Therefore,broadband turbulent enabled compressible Large Eddy Simulation (LES) cal-culations have been performed and several flow-driven instabilities have beencaptured under unstable conditions. LES produces large amounts of 3D datawhich has been post-processed using Fourier spectra and Dynamic Mode De-composition (DMD). These techniques are able to quantify modes in the flowfield by extracting large coherent flow structures and characterize their relativecontribution to the total fluctuation energy at associated. Among the mainfindings, a dominant mode was found which describes the filling and emptyingprocess during surge. A narrowband feature at half of the rotating order wasidentified to correspond to co-rotating vortices upstream of the impeller faceas well as elevated velocity magnitude regions propagating tangentially in thediffuser and the volute. Dominant mode shapes were also found at the rotatingorder frequency and its harmonics, which manifest as a spinning mode shapelocalized at the diffuser inlet. From the compressible LES flow solution one can extract the acoustic infor-mation and the noise affiliated with the compressor. This enable through datacorrelation quantifying the flow-acoustics coupling phenomena in the compres-sor. Detailed comparison of flow (pressure, velocity) and aeroacoustics (soundpressure levels) predictions in terms of time-averaged, fluctuating quantities,and spectra is carried out against experimental measurements. / <p>QC 20160406</p>

No description available.

Les nouveaux acteurs, les nouveaux services et les nouveaux contenus multimédias qui transitent sur le réseau internet génèrent un trafic et des débits de plus en plus élevés. Ceci peut occasionner une congestion, source de latence et de dépréciation de la qualité de service ressentie par les utilisateurs. Un fournisseur d'accès à internet dont l'objectif est de garantir un réseau d'excellence doit donc prendre des mesures pour améliorer sans cesse la fluidité de son réseau. Cela passe notamment par la mise en place d'un réseau de distribution de contenus (déploiement de dispositifs sur le réseau existant). Dans un premier temps cette thèse s'articule à présenter des approches de programmation dynamique de localisation de serveurs optimales dans des arborescences. Nous présentons également un approche pour résoudre le problème de déploiement de CDN et de k serveurs/caches à l'aide de l'algorithme exact et polynomial d'intersection de matroïdes. Nous explicitons ensuite ce qu'est un cache et quelles sont ses caractéristiques. Nous définissons ensuite les hypothèses effectuées et la modélisation associée pour le déploiement de caches transparents dans une arborescence, et le liens avec les algorithmes existants présentés précédemment. Nous présentons alors un modèle complet pour un programme linéaire en nombres entiers (PLNE) et un nouveau paradigme de programmation dynamique pour résoudre ce même problème. Nous montrons alors en quoi cette approche se généralise à des problèmes connexes de localisation dans les arborescences, ainsi que les performances pratiques d'une telle approche. D'un regard plus théorique, nous mesurons la capacité d'un réseau donné par le routage optimal de ses demandes, et, de ce fait, ses liens critiques. Nous manipulons alors le problème de flot concurrent maximal (FCM), un problème classique de la littérature de recherche opérationnelle. Nous exhibons alors de nouvelles formulations exactes pour résoudre ce problème, ainsi que les problèmes de multi-flots de manière plus générale. Une heuristique de construction de formulation pour le FCM est également proposée, pour tirer parti de la distribution spécifique des capacités d'une instance. Nous montrons alors la supériorité des performances de ces nouvelles formulations par le biais de comparaisons. Enfin, nous décrivons le premier algorithme exact et fortement polynomial pour résoudre le problème de flot concurrent maximal dans le cas d'une seule source; et nous montrons l'efficacité pratique d'une telle approche, comparée aux meilleures formulations explicitées précédemment / Streaming requirements on internet network are even more driven by new actors, new services and new digital contents. This leads to high probability of congestion, latency and therefore, a critical decrease of quality of service and/or experience for customers. An internet service provider (ISP) whose goal is to guarantee a first-class performance, needs to take measures to constantly enhance the fluidity of the traffic streaming on its network. One way to face the problem, is to build a Content Delivery Network (CDN). A CDN mainly consists in the deployment of different devices on an existing network. First of all, this thesis presents dynamic programming approaches to tackle server location problems in tree networks. Then, we address a variation of the matroïd intersection algorithm to solve the k-server/cache location problem. We start by giving the definition and characteristics of transparent-caching, as well as the hypothesis that we will use it to build models for transparent cache location in tree network. We tract it to a Mixed Integer Program, and formulate a new paradigm of dynamic programming. We show the relevance of such approach for our problem, and to what extent it can be tractable in other related problems. From a more theoretical point of view, we manage to measure the capacity of a network which is given by the optimal routing strategy, and hence, to identify its critical links. We deal with the Maximum Concurrent Flow (MCF), a classical combinatorial optimization problem. We propose new models and formulations to solve this problem exactly, and more general multi-flows problems as well. A heuristic is also given, to adapt the model to the specific instance values. We experiment these formulations to show the improvements they can provide. Finally, we describe the first strongly polynomial algorithm to solve the maximum concurrent flow to optimality, in the single source case. We show the efficiency of such an approach, even compared to the best models previously presented

L’évolution de l’équipement des réseaux d’eau potable a considérablement amélioré le pilotage de ces derniers. Le telérelevé des compteurs d’eau est sans doute la technologie qui a créé la plus grande avancée ces dernières années dans la gestion de l’eau, tant pour l’opérateur que pour l’usager. Cette technologie a permis de passer d’une information le plus souvent annuelle sur les consommations (suite à la relève manuelle des compteurs d’eau) à une information infra-journalière. Mais le télérelevé, aussi performant soit-il, a un inconvénient : son coût. L’instrumentation complète d’un réseau engendre des investissements que certains opérateurs ne peuvent se permettre. Ainsi la création d’un échantillon de compteurs à équiper permet d’estimer la consommation totale d’un réseau tout en minimisant les coûts d’investissement. Cet échantillon doit être construit de façon intelligente de sorte que l’imprécision liée à l’estimation ne nuise pas à l’évaluation des consommations. Une connaissance précise sur les consommations d’eau permet de quantifier les volumes perdus en réseau. Mais, même dans le cas d’une évaluation exacte des pertes, cela ne peut pas suffire à éliminer toutes les fuites sur le réseau. En effet, si le réseau de distribution d’eau potable est majoritairement enterré, donc invisible, il en va de même pour les fuites. Une fraction des fuites est invisible et même indétectable par les techniques actuelles de recherche de fuites, et donc irréparable. La construction d’un modèle de fuite multi-états permet de décomposer le débit de fuite suivant les différents stades d’apparition d’une fuite : invisible et indétectable, invisible mais détectable par la recherche de fuite et enfin visible en surface. Ce modèle, de type semi-markovien, prend en compte les contraintes opérationnelles, notamment le fait que nous disposons de données de panel. La décomposition du débit de fuite permet de fait une meilleure gestion du réseau en ciblant et adaptant les actions de lutte contre les fuites à mettre en place en fonction de l’état de dégradation du réseau. / The evolution of equipment on drinking water networks has considerably bettered the monitoring of these lasts. Automatic meter reading (AMR) is clearly the technology which has brought the major progress these last years in water management, as for the operator and the end-users. This technology has allowed passing from an annual information on water consumption (thanks to the manual meter reading) toan infra-daily information. But as efficient as AMR can be, it has one main inconvenient : its cost. A complete network instrumentation generates capital expenditures that some operators can’t allowed themselves. The constitution of a sample of meters to equip enables then to estimate the network total consumption while minimizing the investments. This sample has to be built smartly so the inaccuracy of the estimator shouldn’t be harmful to the consumption estimation. A precise knowledge on water consumption allowsquantifying the water lost volumes on the network. But even an exact assessment of losses is still not enough to eliminate all the leaks on the network. Indeed, if the water distribution network is buried, and so invisible, so do the leaks. A fraction of leaks are invisible and even undetectable by the current technologies of leakage control, and so these leaks are un-reparable. The construction of a multi-state model enables us to decompose the leakage flow according to the different stages of appearance of a leak : invisible and undetectable, invisible but detectable with leakage control and finally detectable. This semi-Markovian model takes into account operational constrains, in particular the fact that we dispose of panel data. The leakage flow decomposition allows a better network monitoring but targeting and adapting the action of leakage reduction to set up according to the degradation state of the network.