La conception de lanceurs est un problème d’optimisation multidisciplinaire dont l’objectif est de trouverl’architecture du lanceur qui garantit une performance optimale tout en assurant un niveau de fiabilité requis.En vue de l’obtention de la solution optimale, les phases d’avant-projet sont cruciales pour le processus deconception et se caractérisent par la présence d’incertitudes dues aux phénomènes physiques impliqués etaux méconnaissances existantes sur les modèles employés. Cette thèse s’intéresse aux méthodes d’analyse et d’optimisation multidisciplinaire en présence d’incertitudes afin d’améliorer le processus de conception de lanceurs. Trois sujets complémentaires sont abordés. Tout d’abord, deux nouvelles formulations du problème de conception ont été proposées afin d’améliorer la prise en compte des interactions disciplinaires. Ensuite, deux nouvelles méthodes d’analyse de fiabilité, permettant de tenir compte d’incertitudes de natures variées, ont été proposées, impliquant des techniques d’échantillonnage préférentiel et des modèles de substitution. Enfin, une nouvelle technique de gestion des contraintes pour l’algorithme d’optimisation ”Covariance Matrix Adaptation - Evolutionary Strategy” a été développée, visant à assurer la faisabilité de la solution optimale. Les approches développées ont été comparées aux techniques proposées dans la littérature sur des cas tests d’analyse et de conception de lanceurs. Les résultats montrent que les approches proposées permettent d’améliorer l’efficacité du processus d’optimisation et la fiabilité de la solution obtenue. / Launch vehicle design is a Multidisciplinary Design Optimization problem whose objective is to find the launch vehicle architecture providing the optimal performance while ensuring the required reliability. In order to obtain an optimal solution, the early design phases are essential for the design process and are characterized by the presence of uncertainty due to the involved physical phenomena and the lack of knowledge on the used models. This thesis is focused on methodologies for multidisciplinary analysis and optimization under uncertainty for launch vehicle design. Three complementary topics are tackled. First, two new formulations have been developed in order to ensure adequate interdisciplinary coupling handling. Then, two new reliability techniques have been proposed in order to take into account the various natures of uncertainty, involving surrogate models and efficient sampling methods. Eventually, a new approach of constraint handling for optimization algorithm ”Covariance Matrix Adaptation - Evolutionary Strategy” has been developed to ensure the feasibility of the optimal solution. All the proposed methods have been compared to existing techniques in literature on analysis and design test cases of launch vehicles. The results illustrate that the proposed approaches allow the improvement of the efficiency of the design process and of the reliability of the found solution.

Ce travail de thèse traite du développement d'une méthode de télédiagnostique de l'état de santé des suspensions des trains à grande vitesse à partir de mesures de la réponse dynamique du train en circulation par des accéléromètres embarqués. Un train en circulation est un système dynamique dont l'excitation provient des irrégularités de la géométrie de la voie ferrée. Ses éléments de suspension jouent un rôle fondamental de sécurité et de confort. La réponse dynamique du train étant dépendante des caractéristiques mécaniques des éléments de suspension, il est possible d'obtenir en inverse des informations sur l'état de ces éléments à partir de mesures accélérométriques embarquées. Connaître l'état de santé réel des suspensions permettrait d'améliorer la maintenance des trains. D’un point de vue mathématique, la méthode de télédiagnostique proposée consiste à résoudre un problème statistique inverse. Elle s'appuie sur un modèle numérique de dynamique ferroviaire et prend en compte l'incertitude de modèle ainsi que les erreurs de mesures. Les paramètres mécaniques associés aux éléments de suspension sont identifiés par calibration Bayésienne à partir de mesures simultanées des entrées (les irrégularités de la géométrie de la voie) et sorties (la réponse dynamique du train) du système. La calibration Bayésienne classique implique le calcul de la fonction de vraisemblance à partir du modèle stochastique de réponse et des données expérimentales. Le modèle numérique étant numériquement coûteux d'une part, ses entrées et sorties étant fonctionnelles d'autre part, une méthode de calibration Bayésienne originale est proposée. Elle utilise un métamodèle par processus Gaussien de la fonction de vraisemblance. Cette thèse présente comment un métamodèle aléatoire peut être utilisé pour estimer la loi de probabilité des paramètres du modèle. La méthode proposée permet la prise en compte du nouveau type d'incertitude induit par l'utilisation d'un métamodèle. Cette prise en compte est nécessaire pour une estimation correcte de la précision de la calibration. La nouvelle méthode de calibration Bayésienne a été testée sur le cas applicatif ferroviaire, et a produit des résultats concluants. La validation a été faite par expériences numériques. Par ailleurs, l'évolution à long terme des paramètres mécaniques de suspensions a été étudiée à partir de mesures réelles de la réponse dynamique du train / The work presented here deals with the development of a health-state monitoring method for high-speed train suspensions using in-service measurements of the train dynamical response by embedded acceleration sensors. A rolling train is a dynamical system excited by the track-geometry irregularities. The suspension elements play a key role for the ride safety and comfort. The train dynamical response being dependent on the suspensions mechanical characteristics, information about the suspensions state can be inferred from acceleration measurements in the train by embedded sensors. This information about the actual suspensions state would allow for providing a more efficient train maintenance. Mathematically, the proposed monitoring solution consists in solving a statistical inverse problem. It is based on a train-dynamics computational model, and takes into account the model uncertainty and the measurement errors. A Bayesian calibration approach is adopted to identify the probability distribution of the mechanical parameters of the suspension elements from joint measurements of the system input (the track-geometry irregularities) and output (the train dynamical response).Classical Bayesian calibration implies the computation of the likelihood function using the stochastic model of the system output and experimental data. To cope with the fact that each run of the computational model is numerically expensive, and because of the functional nature of the system input and output, a novel Bayesian calibration method using a Gaussian-process surrogate model of the likelihood function is proposed. This thesis presents how such a random surrogate model can be used to estimate the probability distribution of the model parameters. The proposed method allows for taking into account the new type of uncertainty induced by the use of a surrogate model, which is necessary to correctly assess the calibration accuracy. The novel Bayesian calibration method has been tested on the railway application and has achieved conclusive results. Numerical experiments were used for validation. The long-term evolution of the suspension mechanical parameters has been studied using actual measurements of the train dynamical response

Lors de la conception d’un frein, le confort et le bien-être du consommateur font partie des critères principaux. En effet, les instabilités de crissement, qui engendrent une des pollutions acoustiques les plus importantes, représentent un challenge actuel pour la communauté scientifique et les industriels du domaine. Dans le cadre de cette thèse, nous nous sommes intéressés à la mise en évidence du caractère variable du crissement, observé pour deux systèmes de freinage d’un même véhicule, grâce à des plans d’expériences, expérimental et numérique. Pour être représentatif d’une famille de structures, il est désormais indéniable qu’il faille prendre en compte les variabilités observées sur de multiples paramètres liés au système étudié dès la phase de conception. L’enrichissement des simulations déterministes actuelles nécessite la mise en place d’outils non déterministes rapides et respectant le conservatisme des solutions étudiées. Pour ce faire, nous avons contribué au développement de méthodes numériques dédiées à la propagation des données floues dans le cas des graphes de coalescence, à la détermination des positions d’équilibre de corps en contact frottant à partir d’une méthode de régulation basée sur la logique floue. Cette solution permet d’appliquer une technique de projection pour réduire le coût numérique en utilisant des bases modales des composants réanalysées par un développement homotopique. / During a brake design, consumer comfort and well-being are the main criteria. Indeed, squeal instabilities, that produce main acoustic pollution, represent a current challenge in the scientific community and for industrials. In this thesis, we interest first in the highlight of the variability of squeal, observed for two brake systems of a same vehicle, thanks to experimental and numerical designs of experiments. To be representative of a structure family, it is now undeniable that we must take into account variability observed in various parameters of the studied system, from the design phase. To enrich existing deterministic simulations, quick non deterministic tools must be established, respecting the studied solutions conservatism. For this, we have contributed to the development of numerical methods to propagate fuzzy data in the case of diagram of coalescence, to determine the equilibrium position of frictional contact bodies with a fuzzy logic controller. This solution allows to apply a projection technique for reducing the computational cost. The modal bases of components are reanalyzed by homotopy perturbation.

Les simulations numériques représentent un élément central du processus de conception d’un avion complétant les tests physiques et essais en vol. Elles peuvent notamment bénéficier de méthodes innovantes, telle que l’intelligence artificielle qui se diffuse largement dans l’aviation. Simuler une mission de vol complète pour plusieurs disciplines pose d’importants problèmes à cause des coûts de calcul et des conditions d’opérations changeantes. De plus, des phénomènes complexes peuvent se produire. Par exemple, des chocs peuvent apparaître sur l’aile pour l’aérodynamique alors que le mélange entre les écoulements du moteur et de l’air extérieur impacte fortement l’aérothermie autour de la nacelle et du mât. Des modèles de substitution peuvent être utilisés pour remplacer les simulations haute-fidélité par des approximations mathématiques afin de réduire le coût de calcul et de fournir une méthode construite autour des données de simulations. Deux développements sont proposés dans cette thèse : des modèles de substitution utilisant l’apprentissage automatique pour approximer des calculs aérodynamiques et l’intégration de modèles de substitution classiques dans un processus aérothermique industriel. La première approche sépare les solutions en sous-ensembles selon leurs formes grâce à de l’apprentissage automatique. En outre, une méthode de reéchantillonnage complète la base d’entrainement en ajoutant de l’information dans des sous-ensembles spécifiques. Le deuxième développement se concentre sur le dimensionnement du mât moteur en remplaçant les simulations aérothermiques par des modèles de substitution. Ces deux développements sont appliqués sur des configurations avions afin de combler l’écart entre méthode académique et industrielle. On peut noter que des améliorations significatives en termes de coût et de précision ont été atteintes. / Numerical simulations provide a key element in aircraft design process, complementing physical tests and flight tests. They could take advantage of innovative methods, such as artificial intelligence technologies spreading in aviation. Simulating the full flight mission for various disciplines pose important problems due to significant computational cost coupled to varying operating conditions. Moreover, complex physical phenomena can occur. For instance, the aerodynamic field on the wing takes different shapes and can encounter shocks, while aerothermal simulations around nacelle and pylon are sensitive to the interaction between engine flows and external flows. Surrogate models can be used to substitute expensive high-fidelitysimulations by mathematical and statistical approximations in order to reduce overall computation cost and to provide a data-driven approach. In this thesis, we propose two developments: (i) machine learning-based surrogate models capable of approximating aerodynamic experiments and (ii) integrating more classical surrogate models into industrial aerothermal process. The first approach mitigates aerodynamic issues by separating solutions with very different shapes into several subsets using machine learning algorithms. Moreover, a resampling technique takes advantage of the subdomain decomposition by adding extra information in relevant regions. The second development focuses on pylon sizing by building surrogate models substitutingaerothermal simulations. The two approaches are applied to aircraft configurations in order to bridge the gap between academic methods and real-world applications. Significant improvements are highlighted in terms of accuracy and cost gains

Knowledge regarding a nuclear power plants potential and limitations is of utmost importance when working in the nuclear field. One way to extend the knowledge is using fuel performance codes that to its best ability mimics the real-world phenomena. Fuel performance codes involve a system of interlinked and complex models to predict the thermo-mechanical behaviour of the fuel rods. These models use several different model parameters that can be imprecise and therefore the parameters need to be fitted/calibrated against measurement data. This thesis presents two methods to calibrate model parameters in the presence of unknown sources of uncertainty. The case where these methods have been tested are the oxidation and hydrogen pickup of the zirconium cladding around the fuel rods. Initially, training and testing data were sampled by using the Dakota software in combination with the nuclear simulation program TRANSURANUS so that a Gaussian process surrogate model could be built. The model parameters were then calibrated in a Bayesian way by a MCMC algorithm. Additionally, two models are presented to handle unknown sources of uncertainty that may arise from model inadequacies, nuisance parameters or hidden measurement errors, these are the Marginal likelihood optimization method and the Margin method. To calibrate the model parameters, data from two sources were used. One source that only had data regarding the oxide thickness but the data was extensive, and another that had both oxide data and hydrogen concentration data, but less data was available.  The model parameters were calibrated by the use of the presented methods. But an unforeseen non-linearity for the joint oxidation and hydrogen pick-up case when predicting the correlation of the model parameters made this result unreliable.

No description available.

With modern high-throughput technologies, scientists can now collect high-dimensional data of various forms, including brain images, medical spectrum curves, engineering signals, etc. These data provide a rich source of information on disease development, cell evolvement, engineering systems, and many other scientific phenomena. To achieve a clearer understanding of the underlying mechanism, one needs a fast and reliable analytical approach to extract useful information from the wealth of data. The goal of this dissertation is to develop novel methods that enable scalable estimation, testing, and analysis of complex, high-dimensional data. It contains three parts: parameter estimation based on complex data, powerful testing of functional data, and the analysis of functional data supported on manifolds. The first part focuses on a family of parameter estimation problems in which the relationship between data and the underlying parameters cannot be explicitly specified using a likelihood function. We introduce a wavelet-based approximate Bayesian computation approach that is likelihood-free and computationally scalable. This approach will be applied to two applications: estimating mutation rates of a generalized birth-death process based on fluctuation experimental data and estimating the parameters of targets based on foliage echoes. The second part focuses on functional testing. We consider using multiple testing in basis-space via p-value guided compression. Our theoretical results demonstrate that, under regularity conditions, the Westfall-Young randomization test in basis space achieves strong control of family-wise error rate and asymptotic optimality. Furthermore, appropriate compression in basis space leads to improved power as compared to point-wise testing in data domain or basis-space testing without compression. The effectiveness of the proposed procedure is demonstrated through two applications: the detection of regions of spectral curves associated with pre-cancer using 1-dimensional fluorescence spectroscopy data and the detection of disease-related regions using 3-dimensional Alzheimer's Disease neuroimaging data. The third part focuses on analyzing data measured on the cortical surfaces of monkeys' brains during their early development, and subjects are measured on misaligned time markers. In this analysis, we examine the asymmetric patterns and increase/decrease trend in the monkeys' brains across time. / Doctor of Philosophy / With modern high-throughput technologies, scientists can now collect high-dimensional data of various forms, including brain images, medical spectrum curves, engineering signals, and biological measurements. These data provide a rich source of information on disease development, engineering systems, and many other scientific phenomena. The goal of this dissertation is to develop novel methods that enable scalable estimation, testing, and analysis of complex, high-dimensional data. It contains three parts: parameter estimation based on complex biological and engineering data, powerful testing of high-dimensional functional data, and the analysis of functional data supported on manifolds. The first part focuses on a family of parameter estimation problems in which the relationship between data and the underlying parameters cannot be explicitly specified using a likelihood function. We introduce a computation-based statistical approach that achieves efficient parameter estimation scalable to high-dimensional functional data. The second part focuses on developing a powerful testing method for functional data that can be used to detect important regions. We will show nice properties of our approach. The effectiveness of this testing approach will be demonstrated using two applications: the detection of regions of the spectrum that are related to pre-cancer using fluorescence spectroscopy data and the detection of disease-related regions using brain image data. The third part focuses on analyzing brain cortical thickness data, measured on the cortical surfaces of monkeys’ brains during early development. Subjects are measured on misaligned time-markers. By using functional data estimation and testing approach, we are able to: (1) identify asymmetric regions between their right and left brains across time, and (2) identify spatial regions on the cortical surface that reflect increase or decrease in cortical measurements over time.

Code validation and uncertainty quantification are important tasks in nuclear reactor safety analysis. Code users have to deal with large number of uncertain parameters, complex multi-physics, multi-dimensional and multi-scale phenomena. In order to make results of analysis more robust, it is important to develop and employ procedures for guiding user choices in quantification of the uncertainties.   The work aims to further develop approaches and procedures for system analysis code validation and application to practical problems of safety analysis. The work is divided into two parts.   The first part presents validation of two reactor system thermal-hydraulic (STH) codes RELAP5 and TRACE for prediction of two-phase circulation flow instability.   The goals of the first part are to: (a) develop and apply efficient methods for input calibration and STH code validation against unsteady flow experiments with two-phase circulation flow instability, and (b) examine the codes capability to predict instantaneous thermal hydraulic parameters and flow regimes during the transients.   Two approaches have been developed: a non-automated procedure based on separate treatment of uncertain input parameters (UIPs) and an automated method using genetic algorithm. Multiple measured parameters and system response quantities (SRQs) are employed in both calibration of uncertain parameters in the code input deck and validation of RELAP5 and TRACE codes. The effect of improvement in RELAP5 flow regime identification on code prediction of thermal-hydraulic parameters has been studied.   Result of the code validations demonstrates that RELAP5 and TRACE can reproduce qualitative behaviour of two-phase flow instability. However, both codes misidentified instantaneous flow regimes, and it was not possible to predict simultaneously experimental values of oscillation period and maximum inlet flow rate. The outcome suggests importance of simultaneous consideration of multiple SRQs and different test regimes for quantitative code validation.   The second part of this work addresses core degradation and relocation to the lower head of a boiling water reactor (BWR). Properties of the debris in the lower head provide initial conditions for vessel failure, melt release and ex-vessel accident progression.   The goals of the second part are to: (a) obtain a representative database of MELCOR solutions for characteristics of debris in the reactor lower plenum for different accident scenarios, and (b) develop a computationally efficient surrogate model (SM) that can be used in extensive uncertainty analysis for prediction of the debris bed characteristics.   MELCOR code coupled with genetic algorithm, random and grid sampling methods was used to generate a database of the full model solutions and to investigate in-vessel corium debris relocation in a Nordic BWR. Artificial neural networks (ANNs) with classification (grouping) of scenarios have been used for development of the SM in order to address the issue of chaotic response of the full model especially in the transition region.   The core relocation analysis shows that there are two main groups of scenarios: with relatively small (&lt;20 tons) and large (&gt;100 tons) amounts of total relocated debris in the reactor lower plenum. The domains are separated by transition regions, in which small variation of the input can result in large changes in the final mass of debris.  SMs using multiple ANNs with/without weighting between different groups effectively filter out the noise and provide a better prediction of the output cumulative distribution function, but increase the mean squared error compared to a single ANN. / Validering av datorkoder och kvantifiering av osäkerhetsfaktorer är viktiga delar vid säkerhetsanalys av kärnkraftsreaktorer. Datorkodanvändaren måste hantera ett stort antal osäkra parametrar vid beskrivningen av fysikaliska fenomen i flera dimensioner från mikro- till makroskala. För att göra analysresultaten mer robusta, är det viktigt att utveckla och tillämpa rutiner för att vägleda användaren vid kvantifiering av osäkerheter.Detta arbete syftar till att vidareutveckla metoder och förfaranden för validering av systemkoder och deras tillämpning på praktiska problem i säkerhetsanalysen. Arbetet delas in i två delar.Första delen presenterar validering av de termohydrauliska systemkoderna (STH) RELAP5 och TRACE vid analys av tvåfasinstabilitet i cirkulationsflödet.Målen för den första delen är att: (a) utveckla och tillämpa effektiva metoder för kalibrering av indatafiler och validering av STH mot flödesexperiment med tvåfas cirkulationsflödeinstabilitet och (b) granska datorkodernas förmåga att förutsäga momentana termohydrauliska parametrar och flödesregimer under transienta förlopp.Två metoder har utvecklats: en icke-automatisk procedur baserad på separat hantering av osäkra indataparametrar (UIPs) och en automatiserad metod som använder genetisk algoritm. Ett flertal uppmätta parametrar och systemresponser (SRQs) används i både kalibrering av osäkra parametrar i indatafilen och validering av RELAP5 och TRACE. Resultatet av modifikationer i hur RELAP5 identifierar olika flödesregimer, och särskilt hur detta påverkar datorkodens prediktioner av termohydrauliska parametrar, har studerats.Resultatet av valideringen visar att RELAP5 och TRACE kan återge det kvalitativa beteende av två-fas flödets instabilitet. Däremot kan ingen av koderna korrekt identifiera den momentana flödesregimen, det var därför ej möjligt att förutsäga experimentella värden på svängningsperiod och maximal inloppsflödeshastighet samtidigt. Resultatet belyser betydelsen av samtidig behandling av flera SRQs liksom olika experimentella flödesregimer för kvantitativ kodvalidering.Den andra delen av detta arbete behandlar härdnedbrytning och omfördelning till reaktortankens nedre plenumdel i en kokarvatten reaktor (BWR). Egenskaper hos härdrester i nedre plenum ger inledande förutsättningar för reaktortanksgenomsmältning, hur smältan rinner ut ur reaktortanken och händelseförloppet i reaktorinneslutningen.Målen i den andra delen är att: (a) erhålla en representativ databas över koden MELCOR:s analysresultat för egenskaperna hos härdrester i nedre plenum under olika händelseförlopp, och (b) utveckla en beräkningseffektiv surrogatsmodell som kan användas i omfattande osäkerhetsanalyser för att förutsäga partikelbäddsegenskaper.MELCOR, kopplad till en genetisk algoritm med slumpmässigt urval användes för att generera en databas av analysresultat med tillämpning på smältans omfördelning i reaktortanken i en Nordisk BWR.Analysen av hur härden omfördelas visar att det finns två huvudgrupper av scenarier: med relativt liten (&lt;20 ton) och stor (&gt; 100 ton) total mängd omfördelade härdrester i nedre plenum. Dessa domäner är åtskilda av övergångsregioner, där små variationer i indata kan resultera i stora ändringar i den slutliga partikelmassan. Flergrupps artificiella neurala nätverk med klassificering av händelseförloppet har använts för utvecklingen av en surrogatmodell för att hantera problemet med kaotiska resultat av den fullständiga modellen, särskilt i övergångsregionen. / <p>QC 20170309</p>

This dissertation focuses on the introduction and development of an integrated model-based design and optimization platform to solve the optimal design and optimal control, or hardware and software co-design, problem for hybrid electric propulsion systems. Specifically, the hybrid and plug-in hybrid electric powertrain systems with diesel and natural gas (NG) fueled compression ignition (CI) engines and large Li-ion battery energy storage system (ESS) for propelling a hybrid electric marine vessel are investigated. The combined design and control optimization of the hybrid propulsion system is formulated as a bi-level, nested optimization problem. The lower-level optimization applies dynamic programming (DP) to ensure optimal energy management for each feasible powertrain system design, and the upper-level global optimization aims at identifying the optimal sizes of key powertrain components for the powertrain system with optimized control. Recently, Li-ion batteries became a promising ESS technology for electrified transportation applications. However, these costly Li-ion battery ESSs contribute to a large portion of the powertrain electrification and hybridization costs and suffer a much shorter lifetime compared to other key powertrain components. Different battery performance modelling methods are reviewed to identify the appropriate degradation prediction approach. Using this approach and a large set of experimental data, the performance degradation and life prediction model of LiFePO4 type battery has been developed and validated. This model serves as the foundation for determining the optimal size of battery ESS and for optimal energy management in powertrain system control to achieve balanced reduction of fuel consumption and the extension of battery lifetime. In modelling and design of different hybrid electric marine propulsion systems, the life cycle cost (LCC) model of the cleaner, hybrid propulsion systems is introduced, considering the investment, replacement and operational costs of their major contributors. The costs of liquefied NG (LNG), diesel and electricity in the LCC model are collected from various sources, with a focus on present industrial price in British Columbia, Canada. The greenhouse gas (GHG) and criteria air pollutant (CAP) emissions from traditional diesel and cleaner NG-fueled engines with conventional and optimized hybrid electric powertrains are also evaluated. To solve the computational expensive nested optimization problem, a surrogate model-based (or metamodel-based) global optimization method is used. This advanced global optimization search algorithm uses the optimized Latin hypercube sampling (OLHS) to form the Kriging model and uses expected improvement (EI) online sampling criterion to refine the model to guide the search of global optimum through a much-reduced number of sample data points from the computationally intensive objective function. Solutions from the combined hybrid propulsion system design and control optimization are presented and discussed. This research has further improved the methodology of model-based design and optimization of hybrid electric marine propulsion systems to solve complicated co-design problems through more efficient approaches, and demonstrated the feasibility and benefits of the new methods through their applications to tugboat propulsion system design and control developments. The resulting hybrid propulsion system with NG engine and Li-ion battery ESS presents a more economical and environmentally friendly propulsion system design of the tugboat. This research has further improved the methodology of model-based design and optimization of hybrid electric marine propulsion systems to solve complicated co-design problems through more efficient approaches, and demonstrated the feasibility and benefits of the new methods through their applications to tugboat propulsion system design and control developments. Other main contributions include incorporating the battery performance degradation model to the powertrain size optimization and optimal energy management; performing a systematic design and optimization considering LCC of diesel and NG engines in the hybrid electric powertrains; and developing an effective method for the computational intensive powertrain co-design problem. / Graduate

Cette thèse porte sur la construction de modèles de substitution en régimes transitoire et permanent pour la simulation thermique, en présence de peu d'observations et de plusieurs sorties.Nous proposons dans un premier temps une construction robuste de perceptron multicouche bouclé afin d'approcher une dynamique spatio-temporelle. Ce modèle de substitution s'obtient par une moyennisation de réseaux de neurones issus d'une procédure de validation croisée, dont le partitionnement des observations associé permet d'ajuster les paramètres de chacun de ces modèles sur une base de test sans perte d'information. De plus, la construction d'un tel perceptron bouclé peut être distribuée selon ses sorties. Cette construction est appliquée à la modélisation de l'évolution temporelle de la température en différents points d'une armoire aéronautique.Nous proposons dans un deuxième temps une agrégation de modèles par processus gaussien dans un cadre multifidélité où nous disposons d'un modèle d'observation haute-fidélité complété par plusieurs modèles d'observation de fidélités moindres et non comparables. Une attention particulière est portée sur la spécification des tendances et coefficients d'ajustement présents dans ces modèles. Les différents krigeages et co-krigeages sont assemblés selon une partition ou un mélange pondéré en se basant sur une mesure de robustesse aux points du plan d'expériences les plus fiables. Cette approche est employée pour modéliser la température en différents points de l'armoire en régime permanent.Nous proposons dans un dernier temps un critère pénalisé pour le problème de la régression hétéroscédastique. Cet outil est développé dans le cadre des estimateurs par projection et appliqué au cas particulier des ondelettes de Haar. Nous accompagnons ces résultats théoriques de résultats numériques pour un problème tenant compte de différentes spécifications du bruit et de possibles dépendances dans les observations. / This PhD thesis deals with the construction of surrogate models in transient and steady states in the context of thermal simulation, with a few observations and many outputs.First, we design a robust construction of recurrent multilayer perceptron so as to approach a spatio-temporal dynamic. We use an average of neural networks resulting from a cross-validation procedure, whose associated data splitting allows to adjust the parameters of these models thanks to a test set without any information loss. Moreover, the construction of this perceptron can be distributed according to its outputs. This construction is applied to the modelling of the temporal evolution of the temperature at different points of an aeronautical equipment.Then, we proposed a mixture of Gaussian process models in a multifidelity framework where we have a high-fidelity observation model completed by many observation models with lower and no comparable fidelities. A particular attention is paid to the specifications of trends and adjustement coefficients present in these models. Different kriging and co-krigings models are put together according to a partition or a weighted aggregation based on a robustness measure associated to the most reliable design points. This approach is used in order to model the temperature at different points of the equipment in steady state.Finally, we propose a penalized criterion for the problem of heteroscedastic regression. This tool is build in the case of projection estimators and applied with the Haar wavelet. We also give some numerical results for different noise specifications and possible dependencies in the observations.