Rock type is a key concept in modern reservoir characterization that straddles multiple scales and bridges multiple disciplines. Reservoir rock classification (or simply rock typing) has been recognized as one of the most effective description tools to facilitate large-scale reservoir modeling and simulation. This dissertation aims to integrate core data and well logs to enhance reservoir description by classifying reservoir rocks in a geologically and petrophysically consistent manner. The main objective is to develop scientific approaches for utilizing multi-physics rock data at different time and length scales to describe reservoir rock-fluid systems. Emphasis is placed on transferring physical understanding of rock types from limited ground-truthing core data to abundant well logs using fast log simulations in a multi-layered earth model. Bimodal log-normal pore-size distribution functions derived from mercury injection capillary pressure (MICP) data are first introduced to characterize complex pore systems in carbonate and tight-gas sandstone reservoirs. Six pore-system attributes are interpreted and integrated to define petrophysical orthogonality or dissimilarity between two pore systems of bimodal log-normal distributions. A simple three-dimensional (3D) cubic pore network model constrained by nuclear magnetic resonance (NMR) and MICP data is developed to quantify fluid distributions and phase connectivity for predicting saturation-dependent relative permeability during two-phase drainage. There is rich petrophysical information in spatial fluid distributions resulting from vertical fluid flow on a geologic time scale and radial mud-filtrate invasion on a drilling time scale. Log attributes elicited by such fluid distributions are captured to quantify dynamic reservoir petrophysical properties and define reservoir flow capacity. A new rock classification workflow that reconciles reservoir saturation-height behavior and mud-filtrate for more accurate dynamic reservoir modeling is developed and verified in both clastic and carbonate fields. Rock types vary and mix at the sub-foot scale in heterogeneous reservoirs due to depositional control or diagenetic overprints. Conventional well logs are limited in their ability to probe the details of each individual bed or rock type as seen from outcrops or cores. A bottom-up Bayesian rock typing method is developed to efficiently test multiple working hypotheses against well logs to quantify uncertainty of rock types and their associated petrophysical properties in thinly bedded reservoirs. Concomitantly, a top-down reservoir description workflow is implemented to characterize intermixed or hybrid rock classes from flow-unit scale (or seismic scale) down to the pore scale based on a multi-scale orthogonal rock class decomposition approach. Correlations between petrophysical rock types and geological facies in reservoirs originating from deltaic and turbidite depositional systems are investigated in detail. Emphasis is placed on the cause-and-effect relationship between pore geometry and rock geological attributes such as grain size and bed thickness. Well log responses to those geological attributes and associated pore geometries are subjected to numerical log simulations. Sensitivity of various physical logs to petrophysical orthogonality between rock classes is investigated to identify the most diagnostic log attributes for log-based rock typing. Field cases of different reservoir types from various geological settings are used to verify the application of petrophysical rock classification to assist reservoir characterization, including facies interpretation, permeability prediction, saturation-height analysis, dynamic petrophysical modeling, uncertainty quantification, petrophysical upscaling, and production forecasting. / text

Quantifying uncertainties in large-scale forward and inverse PDE simulations has emerged as a central challenge facing the field of computational science and engineering. The promise of modeling and simulation for prediction, design, and control cannot be fully realized unless uncertainties in models are rigorously quantified, since this uncertainty can potentially overwhelm the computed result. While statistical inverse problems can be solved today for smaller models with a handful of uncertain parameters, this task is computationally intractable using contemporary algorithms for complex systems characterized by large-scale simulations and high-dimensional parameter spaces. In this dissertation, I address issues regarding the theoretical formulation, numerical approximation, and algorithms for solution of infinite-dimensional Bayesian statistical inverse problems, and apply the entire framework to a problem in global seismic wave propagation. Classical (deterministic) approaches to solving inverse problems attempt to recover the “best-fit” parameters that match given observation data, as measured in a particular metric. In the statistical inverse problem, we go one step further to return not only a point estimate of the best medium properties, but also a complete statistical description of the uncertain parameters. The result is a posterior probability distribution that describes our state of knowledge after learning from the available data, and provides a complete description of parameter uncertainty. In this dissertation, a computational framework for such problems is described that wraps around the existing forward solvers, as long as they are appropriately equipped, for a given physical problem. Then a collection of tools, insights and numerical methods may be applied to solve the problem, and interrogate the resulting posterior distribution, which describes our final state of knowledge. We demonstrate the framework with numerical examples, including inference of a heterogeneous compressional wavespeed field for a problem in global seismic wave propagation with 10⁶ parameters.

This study presents different aspects on the use of deterministic methods including Artificial Neural Networks (ANNs), and linear and nonlinear regression, as well as probabilistic methods including Bayesian inference and Monte Carlo methods to develop reliable solutions for challenging problems in geotechnics. This study addresses the theoretical and computational advantages and limitations of these methods in application to: 1) prediction of the stiffness and strength of stabilized organic soils, 2) determination of unknown foundations for bridges vulnerable to scour, and 3) uncertainty quantification for one-dimensional diffusion processes. ANNs were successfully implemented in this study to develop nonlinear models for the mechanical properties of stabilized organic soils. ANN models were able to learn from the training examples and then generalize the trend to make predictions for the stiffness and strength of stabilized organic soils. A stepwise parameter selection and a sensitivity analysis method were implemented to identify the most relevant factors for the prediction of the stiffness and strength. Also, the variations of the stiffness and strength with respect to each factor were investigated. A deterministic and a probabilistic approach were proposed to evaluate the characteristics of unknown foundations of bridges subjected to scour. The proposed methods were successfully implemented and validated by collecting data for bridges in the Bryan District. ANN models were developed and trained using the database of bridges to predict the foundation type and embedment depth. The probabilistic Bayesian approach generated probability distributions for the foundation and soil characteristics and was able to capture the uncertainty in the predictions. The parametric and numerical uncertainties in the one-dimensional diffusion process were evaluated under varying observation conditions. The inverse problem was solved using Bayesian inference formulated by both the analytical and numerical solutions of the ordinary differential equation of diffusion. The numerical uncertainty was evaluated by comparing the mean and standard deviation of the posterior realizations of the process corresponding to the analytical and numerical solutions of the forward problem. It was shown that higher correlation in the structure of the observations increased both parametric and numerical uncertainties, whereas increasing the number of data dramatically decreased the uncertainties in the diffusion process.

The dissertation focuses on estimating the uncertainty associated with stress/strain prediction procedures from dynamic test data used in turbine blade analysis. An accurate prediction of the maximum response levels for physical components during in-field operating conditions is essential for evaluating their performance and life characteristics, as well as for investigating how their behavior critically impacts system design and reliability assessment. Currently, stress/strain inference for a dynamic system is based on the combination of experimental data and results from the analytical/numerical model of the component under consideration. Both modeling challenges and testing limitations, however, contribute to the introduction of various sources of uncertainty within the given estimation procedure, and lead ultimately to diminished accuracy and reduced confidence in the predicted response. The objective of this work is to characterize the uncertainties present in the current response estimation process and provide a means to assess them quantitatively. More specifically, proposed in this research is a statistical methodology based on a Bayesian-network representation of the modeling process which allows for a statistically rigorous synthesis of modeling assumptions and information from experimental data. Such a framework addresses the problem of multi-directional uncertainty propagation, where standard techniques for unidirectional propagation from inputs' uncertainty to outputs' variability are not suited. Furthermore, it allows for the inclusion within the analysis of newly available test data that can provide indirect evidence on the parameters of the structure's analytical model, as well as lead to a reduction of the residual uncertainty in the estimated quantities. As part of this work, key uncertainty sources (i.e., material and geometric properties, sensor measurement and placement, as well as noise due data processing limitations) are investigated, and their impact upon the system response estimates is assessed through sensitivity studies. The results are utilized for the identification of the most significant contributors to uncertainty to be modeled within the developed Bayesian inference scheme. Simulated experimentation, statistically equivalent to specified real tests, is also constructed to generate the data necessary to build the appropriate Bayesian network, which is then infused with actual experimental information for the purpose of explaining the uncertainty embedded in the response predictions and quantifying their inherent accuracy.

Prospection de pétrole utilise un équipement appelé tube de forage pour forer le sol jusqu'au le niveau du réservoir. Cet équipement est une longue colonne rotative, composée par une série de tiges de forage interconnectées et les équipements auxiliaires. La dynamique de cette colonne est très complexe parce que dans des conditions opérationnelles normales, elle est soumise à des vibrations longitudinales, latérales et de torsion, qui présentent un couplage non linéaire. En outre, cette structure est soumise à effets de frottement et à des chocs dûs aux contacts mécaniques entre les paires tête de forage/sol et tube de forage/sol. Ce travail présente un modèle mécanique-mathématique pour analyser un tube de forage en configuration horizontale. Ce modèle utilise la théorie des poutres qui utilise l'inertie de rotation, la déformation de cisaillement et le couplage non linéaire entre les trois mécanismes de vibration. Les équations du modèle sont discrétisées par la méthode des éléments finis. Les incertitudes des paramètres du modèle d'interaction tête de forage/sol sont prises en compte par l'approche probabiliste paramétrique, et les distributions de probabilité des paramètres aléatoires sont construits par le principe du maximum d'entropie. Des simulations numériques sont réalisées afin de caractériser le comportement dynamique non linéaire de la structure, et en particulier, de l'outil de forage. Des phénomènes dynamiques non linéaires par nature, comme le slick-slip et le bit-bounce, sont observés dans les simulations, ainsi que les chocs. Une analyse spectrale montre étonnamment que les phénomènes slick-slip et bit-bounce résultent du mécanisme de vibration latérale, et ce phénomène de choc vient de la vibration de torsion. Cherchant à améliorer l'efficacité de l'opération de forage, un problème d'optimisation qui cherche à maximiser la vitesse de pénétration de la colonne dans le sol, sur ses limites structurelles, est proposé et résolu / Oil prospecting uses an equipment called drillstring to drill the soil until the reservoir level. This equipment is a long column under rotation, composed by a sequence of connected drill-pipes and auxiliary equipment. The dynamics of this column is very complex because, under normal operational conditions, it is subjected to longitudinal, lateral, and torsional vibrations, which presents a nonlinear coupling. Also, this structure is subjected to friction and shocks effects due to the mechanical contacts between the pairs drill-bit/soil and drill-pipes/borehole. This work presents a mechanical-mathematical model to analyze a drillstring in horizontal configuration. This model uses a beam theory which accounts rotatory inertia, shear deformation, and the nonlinear coupling between three mechanisms of vibration. The model equations are discretized using the finite element method. The uncertainties in bit-rock interaction model parameters are taken into account through a parametric probabilistic approach, and the random parameters probability distributions are constructed by means of maximum entropy principle. Numerical simulations are conducted in order to characterize the nonlinear dynamic behavior of the structure, specially, the drill-bit. Dynamical phenomena inherently nonlinear, such as slick-slip and bit-bounce, are observed in the simulations, as well as shocks. A spectral analysis shows, surprisingly, that slick-slip and bit-bounce phenomena result from the lateral vibration mechanism, and that shock phenomena comes from the torsional vibration. Seeking to increase the efficiency of the drilling process, an optimization problem that aims to maximize the rate of penetration of the column into the soil, respecting its structural limits, is proposed and solved

La simulation numérique fiable des performances de turbines hydrauliques suppose : i) de pouvoir inclure dans les calculs RANS (Reynolds-Averaged Navier-Stokes) traditionnellement mis en œuvre l'effet des incertitudes qui existent en pratique sur les conditions d'entrée de l'écoulement; ii) de pouvoir faire appel à une stratégie de type SGE (Simulation des Grandes Echelles) pour améliorer la description des effets de la turbulence lorsque des écarts subsistent entre calculs RANS et résultats d'essai de référence même après prise en compte des incertitudes. Les présents travaux mettent en oeuvre une démarche non intrusive de quantification d'incertitude (NISP pour Non-Intrusive Spectral Projection) pour deux configurations d'intérêt pratique : un distributeur de turbine Francis avec débit et angle d'entrée incertains et un aspirateur de turbine bulbe avec conditions d'entrée (profils de vitesse,en particulier en proche paroi, et grandeurs turbulentes) incertaines. L'approche NISP est utilisée non seulement pour estimer la valeur moyenne et la variance de quantités d'intérêt mais également pour disposer d'une analyse de la variance qui permet d'identifier les incertitudes les plus influentes. Les simulations RANS, vérifiées par une démarche de convergence en maillage, ne permettent pas pour la plupart des configurations analysées d'expliquer les écarts calcul / expérience grâce à la prise en compte des incertitudes d'entrée.Nous mettons donc également en ouvre des simulations SGE en faisant appel à une stratégie originale d'évaluation de la qualité des maillages utilisés dans le cadre d'une démarche de vérification des calculs SGE. Pour une majorité des configurations analysées, la combinaison d'une stratégie SGE et d'une démarche de quantification des incertitudes permet de produire des résultats numériques fiables. La prise en compte des incertitudes d'entrée permet également de proposer une démarche d'optimisation robuste du distributeur de turbine Francis étudié. / The reliable numerical simulation of hydraulic turbines performance requires : i) to includeinto the conventional RANS computations the effect of the uncertainties existing in practiceon the inflow conditions; ii) to rely on a LES (Large Eddy Simulation) strategy to improve thedescription of turbulence effects when discrepancies between RANS computations and experimentskeep arising even though uncertainties are taken into account. The present workapplies a non-intrusive Uncertainty Quantification strategy (NISP for Non-Intrusive SpectralProjection) to two configurations of practical interest : a Francis turbine distributor, with uncertaininlet flow rate and angle, and a draft-tube of a bulb-type turbine with uncertain inflowconditions (velocity distributions, in particular close to the wall boundaries, and turbulentquantities). The NISP method is not only used to compute the mean value and variance ofquantities of interest, it is also applied to perform an analysis of the variance and identify inthis way the most influential uncertainties. The RANS simulations, verified through a gridconvergence approach, are such the discrepancies between computation and experimentcannot be explained by taking into account the inflow uncertainties for most of the configurationsunder study. Therefore, LES simulations are also performed and these simulations areverified using an original methodology for assessing the quality of the computational grids(since the grid-convergence concept is not relevant for LES). For most of the flows understudy, combining a SGE strategy with a UQ approach yields reliable numerical results. Takinginto account inflow uncertainties also allows to propose a robust optimization strategy forthe Francis turbine distributor under study.

Cette thèse s’inscrit dans le cadre du projet ANR « Uncertain flow optimization » (UFO) consacré au développement et à l’application de méthodes efficaces de quantification d’incertitudes pour l’analyse et l’optimisation d’écoulements. Dans ce contexte, ces méthodes sont appliquées à des gazosiphons de grande échelle utilisés comme pompe. Plus particulièrement, on s’intéresse à de tels gazosiphons choisis pour constituer l’organe central d’un système géothermique innovant,  compatible avec un environnement urbain. On souhaite en quantifier le potentiel énergétique par voie numérique avec la recherche d’un compromis entre justesse des résultats et efficacité optimale. La simulation de l’écoulement diphasique produit dans le gazosiphon est fondée sur un modèle quasi-1D à flux de dérive et s’appuie sur une démarche de résolution implicite. Les résultats sont validés sur les études expérimentales les plus pertinents de la littérature, dont aucune toutefois n’atteint les longueurs requises de l'ordre du kilomètre. Le code de simulation du gazosiphon fait ensuite l’objet d’une démarche de prise en compte d’incertitudes physiques et de modélisation, précédée par une analyse de deux méthodes de quantification d’incertitude : une méthode non-intrusive de type chaos polynomial, et une méthode plus récente dite semi-intrusive qui fut développée en amont du projet UFO. Cet outil est intégré dans une modélisation simplifiée du système géothermique urbain dans son ensemble impliquant les composants en surface, notamment le compresseur d'air. Il en résulte une optimisation énergétique robuste préliminaire de deux variantes du système géothermique urbain proposé, respectivement de récupération de chaleur et de production d’électricité. / This PhD thesis is part of the ANR project « Uncertain Flow Optimization » (UFO). The project is devoted to the development and application of efficient uncertainty quantification methods for flow analysis and optimization. In this framework, these methods are applied to the study of a large-scale airlift pump. The airlift pump is selected to be part of an innovative geothermal system, which can be exploited within an urban environment. We wish to quantify and optimize the energy potential of this new system with numerical tools. They provide both good accuracy and efficiency properties. The airlift two-phase flow simulation is based on a quasi one-dimensional drift flux model, which is implicitly solved. The solver is validated by comparison with relevant experimental airlift studies from the literature. However, these studies remain below the kilometric-targeted pipe length. Thanks to the analysis of two uncertainty quantification methods, a non-intrusive approach relying on polynomial chaos expansion and a new semi-intrusive method developed ahead of the UFO project, we perform airlift pump simulations taking into account physical and modelling uncertainties. This numerical tool is inserted into a simplified model of the complete urban geothermal system that involves surface devices, such as an air compressor. Finally, a robust preliminary optimization process is performed for two versions of the proposed geothermal urban system. They are designed respectively for heat recovery and electricity production.

Formulations of optimization under uncertainty (OUU) commonly used in aerospace design—those based on treating statistical moments of the quantity of interest (QOI) as separate objectives—can result in stochastically dominated designs. A stochastically dominated design is undesirable, because it is less likely than another design to achieve a QOI at least as good as a given value, for any given value. As a remedy to this limitation for the multi-objective formulation of moments, a novel OUU formulation is proposed—dominance optimization. This formulation seeks a set of solutions and makes use of global optimizers, so is useful for early stages of the design process when exploration of design space is important. Similarly, to address this limitation for the single-objective formulation of moments (combining moments via a weighted sum), a second novel formulation is proposed—horsetail matching. This formulation can make use of gradient- based local optimizers, so is useful for later stages of the design process when exploitation of a region of design space is important. Additionally, horsetail matching extends straightforwardly to different representations of uncertainty, and is flexible enough to emulate several existing OUU formulations. Existing multi-fidelity methods for OUU are not compatible with these novel formulations, so one such method—information reuse—is generalized to be compatible with these and other formulations. The proposed formulations, along with generalized information reuse, are compared to their most comparable equivalent in the current state-of-the-art on practical design problems: transonic aerofoil design, coupled aero-structural wing design, high-fidelity 3D wing design, and acoustic horn shape design. Finally, the two novel formulations are combined in a two-step design process, which is used to obtain a robust design in a challenging version of the acoustic horn design problem. Dominance optimization is given half the computational budget for exploration; then horsetail matching is given the other half for exploitation. Using exactly the same computational budget as a moment-based approach, the design obtained using the novel formulations is 95% more likely to achieve a better QOI than the best value achievable by the moment-based design.

Ce travail de recherche présente une extension de la construction classique des modèles réduits (ROMs) obtenus par analyse modale, en dynamique numérique des structures linéaires. Cette extension est basée sur une stratégie de projection multi-niveau, pour l'analyse dynamique des structures complexes en présence d'incertitudes. De nos jours, il est admis qu'en dynamique des structures, la prévision sur une large bande de fréquence obtenue à l'aide d'un modèle éléments finis doit être améliorée en tenant compte des incertitudes de modèle induites par les erreurs de modélisation, dont le rôle croît avec la fréquence. Dans un tel contexte, l'approche probabiliste non-paramétrique des incertitudes est utilisée, laquelle requiert l'introduction d'un ROM. Par conséquent, ces deux aspects, évolution fréquentielle des niveaux d'incertitudes et réduction de modèle, nous conduisent à considérer le développement d'un ROM multi-niveau, pour lequel les niveaux d'incertitudes dans chaque partie de la bande de fréquence peuvent être adaptés. Dans cette thèse, on s'intéresse à l'analyse dynamique de structures complexes caractérisées par la présence de plusieurs niveaux structuraux, par exemple avec un squelette rigide qui supporte diverses sous-parties flexibles. Pour de telles structures, il est possible d'avoir, en plus des modes élastiques habituels dont les déplacements associés au squelette sont globaux, l'apparition de nombreux modes élastiques locaux, qui correspondent à des vibrations prédominantes des sous-parties flexibles. Pour ces structures complexes, la densité modale est susceptible d'augmenter fortement dès les basses fréquences (BF), conduisant, via la méthode d'analyse modale, à des ROMs de grande dimension (avec potentiellement des milliers de modes élastiques en BF). De plus, de tels ROMs peuvent manquer de robustesse vis-à-vis des incertitudes, en raison des nombreux déplacements locaux qui sont très sensibles aux incertitudes. Il convient de noter qu'au contraire des déplacements globaux de grande longueur d'onde caractérisant la bande BF, les déplacements locaux associés aux sous-parties flexibles de la structure, qui peuvent alors apparaître dès la bande BF, sont caractérisés par de courtes longueurs d'onde, similairement au comportement dans la bande hautes fréquences (HF). Par conséquent, pour les structures complexes considérées, les trois régimes vibratoires BF, MF et HF se recouvrent, et de nombreux modes élastiques locaux sont entremêlés avec les modes élastiques globaux habituels. Cela implique deux difficultés majeures, concernant la quantification des incertitudes d'une part et le coût numérique d'autre part. L'objectif de cette thèse est alors double. Premièrement, fournir un ROM stochastique multi-niveau qui est capable de rendre compte de la variabilité hétérogène introduite par le recouvrement des trois régimes vibratoires. Deuxièmement, fournir un ROM prédictif de dimension réduite par rapport à celui de l'analyse modale. Une méthode générale est présentée pour la construction d'un ROM multi-niveau, basée sur trois bases réduites (ROBs) dont les déplacements correspondent à l'un ou l'autre des régimes vibratoires BF, MF ou HF (associés à des déplacements de type BF, de type MF ou bien de type HF). Ces ROBs sont obtenues via une méthode de filtrage utilisant des fonctions de forme globales pour l'énergie cinétique (par opposition aux fonctions de forme locales des éléments finis). L'implémentation de l'approche probabiliste non-paramétrique dans le ROM multi-niveau permet d'obtenir un ROM stochastique multi-niveau avec lequel il est possible d'attribuer un niveau d'incertitude spécifique à chaque ROB. L'application présentée est relative à une automobile, pour laquelle le ROM stochastique multi-niveau est identifié par rapport à des mesures expérimentales. Le ROM proposé permet d'obtenir une dimension réduite ainsi qu'une prévision améliorée, en comparaison avec un ROM stochastique classique / This work deals with an extension of the classical construction of reduced-order models (ROMs) that are obtained through modal analysis in computational linear structural dynamics. It is based on a multilevel projection strategy and devoted to complex structures with uncertainties. Nowadays, it is well recognized that the predictions in structural dynamics over a broad frequency band by using a finite element model must be improved in taking into account the model uncertainties induced by the modeling errors, for which the role increases with the frequency. In such a framework, the nonparametric probabilistic approach of uncertainties is used, which requires the introduction of a ROM. Consequently, these two aspects, frequency-evolution of the uncertainties and reduced-order modeling, lead us to consider the development of a multilevel ROM in computational structural dynamics, which has the capability to adapt the level of uncertainties to each part of the frequency band. In this thesis, we are interested in the dynamical analysis of complex structures in a broad frequency band. By complex structure is intended a structure with complex geometry, constituted of heterogeneous materials and more specifically, characterized by the presence of several structural levels, for instance, a structure that is made up of a stiff main part embedding various flexible sub-parts. For such structures, it is possible having, in addition to the usual global-displacements elastic modes associated with the stiff skeleton, the apparition of numerous local elastic modes, which correspond to predominant vibrations of the flexible sub-parts. For such complex structures, the modal density may substantially increase as soon as low frequencies, leading to high-dimension ROMs with the modal analysis method (with potentially thousands of elastic modes in low frequencies). In addition, such ROMs may suffer from a lack of robustness with respect to uncertainty, because of the presence of the numerous local displacements, which are known to be very sensitive to uncertainties. It should be noted that in contrast to the usual long-wavelength global displacements of the low-frequency (LF) band, the local displacements associated with the structural sub-levels, which can then also appear in the LF band, are characterized by short wavelengths, similarly to high-frequency (HF) displacements. As a result, for the complex structures considered, there is an overlap of the three vibration regimes, LF, MF, and HF, and numerous local elastic modes are intertwined with the usual global elastic modes. This implies two major difficulties, pertaining to uncertainty quantification and to computational efficiency. The objective of this thesis is thus double. First, to provide a multilevel stochastic ROM that is able to take into account the heterogeneous variability introduced by the overlap of the three vibration regimes. Second, to provide a predictive ROM whose dimension is decreased with respect to the classical ROM of the modal analysis method. A general method is presented for the construction of a multilevel ROM, based on three orthogonal reduced-order bases (ROBs) whose displacements are either LF-, MF-, or HF-type displacements (associated with the overlapping LF, MF, and HF vibration regimes). The construction of these ROBs relies on a filtering strategy that is based on the introduction of global shape functions for the kinetic energy (in contrast to the local shape functions of the finite elements). Implementing the nonparametric probabilistic approach in the multilevel ROM allows each type of displacements to be affected by a particular level of uncertainties. The method is applied to a car, for which the multilevel stochastic ROM is identified with respect to experiments, solving a statistical inverse problem. The proposed ROM allows for obtaining a decreased dimension as well as an improved prediction with respect to a classical stochastic ROM

La dispersion d'agents biologiques hautement pathogène dans une zone urbanisée après un acte terroriste est l'une des situations que les agences de sécurité nationales ont besoin d'évaluer en termes des risques et de la prise de décision. La simulation numérique des écoulements turbulents dans les zones urbaines, y compris la surveillance de la dispersion des polluants, a atteint un niveau de maturité suffisant pour faire des prédictions sur les zones urbaines réalistes jusqu'à 4 km2. Les simulations existantes sont déterministes dans le sens que tous les paramètres qui définissent le cas étudié (l'intensité et la direction du vent, la stratification atmosphérique, l'emplacement de la source des émissions, etc.) devraient être bien connu. Cette précision ne peut être atteint dans la pratique en raison d'un manque de connaissances sur la source d'émissions et de l'incertitude aléatoire intrinsèque des conditions météorologiques.Pour augmenter la contribution de la simulation numérique pour l'évaluation des risques et la prise de décision, il est essentiel de mesurer quantitativement l'impact d'un manque de connaissances en termes de résolution spatiale et temporelle des zones de danger.L'objet de cette thèse est d'appliquer des méthodes de quantification d'incertitude pour quantifier l'impact de ces incertitudes dans l'évaluation des zones de danger à moyenne portée dans des scénarios de dispersion de gaz toxiques. Une méthode hybride c-ANOVA et POD/Krigeage permet d'envisager jusqu'à 5 paramètres incertains dans une simulation 3D-CFD haute fidélité non-stationnaire de la dispersion d'un gaz toxique provenant d'une source type flaque dans une zone urbaine de 1km2. / The dispersion of highly pathogenic biological agents in an urbanized area following a terrorist act is one of the situations that national security agencies need to evaluate in terms of risk assessment and decision-making. The numerical simulation of turbulent flows in urban areas, including monitoring the dispersion of pollutants, has reached a sufficient level of maturity to make predictions on realistic urban zones up to 4 square kilometers. However, the existing simulations are deterministic in the sense that all the parameters that define the case studied (intensity and wind direction, atmospheric stratification, source of emissions location, quantity of injected toxic agent, etc.) should be well known. Such precision cannot be achieved in practice due to a lack of knowledge about the source of emission and the intrinsic aleatoric uncertainty of the meteorological conditions. To significantly increase the contribution of numerical simulation for risk assessment and decision-making, it is essential to quantitatively measure the impact of a lack of knowledge especially in terms of spatial and temporal resolution of the danger zones. The object of this thesis is to apply uncertainty quantification methods to quantify the impact of these uncertainties in the evaluation of the danger zones in medium range toxic gas dispersion scenarios. A hybrid method merging c-ANOVA and POD/Kriging allows to consider up to 5 uncertain parameters in a high-fidelity unsteady 3D-CFD simulation of the dispersion of a toxic gas from a pond-like source in an urban area of 1km2.