La complexité croissante des systèmes employés dans l'industrie automobile, en termes de fonctions réalisées et de méthodes de mise en œuvre, mais aussi en terme de norme d'homologation, amène à envisager des outils toujours plus innovants lors de la conception d'un véhicule. On observe d'ailleurs depuis quelques années une forte augmentation du nombre de brevets déposés, en particulier dans le domaine des systèmes électroniques, dont l'importance ne cesse de croître au sein d'un véhicule automobile moderne. Cette complexité croissante des fonctions réalisées requiert une précision de description accrue pour les dispositifs impliqués, notamment pour les systèmes complexes où une approche analytique est difficilement envisageable. Aux impératifs de précision de la description, qui imposent souvent de prendre en considération les non-linéarités des processus, s'ajoute donc la complexité d'analyse des phénomènes physiques à l'origine des observations que l'on souhaite modéliser. Les développements qu'ont connus ces dernières années les techniques de modélisation non linéaires par apprentissage (notamment les réseaux de neurones formels et les machines à vecteurs supports), alliés à la croissance de la capacité des ordinateurs et des calculateurs embarqués dans les véhicules automobiles, justifient donc l'intérêt porté par Renault à ces outils. C'est dans cette optique qu'a été envisagée une étude portant sur les méthodes de modélisation non linéaire par apprentissage, dont l'objectif était d'en tester les secteurs d'applications possibles dans le contexte automobile, et d'en évaluer les difficultés de mise en œuvre ainsi que les gains attendus. Cette étude a fait l'objet d'une collaboration, sous forme d'un contrat de thèse CIFRE, avec le Laboratoire d'Electronique de l'Ecole Supérieure de Physique et de Chimie Industrielles de la Ville de Paris (ESPCI), dirigé par le Professeur Gérard Dreyfus. De manière générale, les techniques de modélisation par apprentissage permettent d'aborder la modélisation de phénomènes physiques dont la description est ardue, élargissant ainsi le champ des possibles en matière de modélisation, mais également de s'affranchir d'une description physique détaillée pour des processus connus, réduisant ainsi le temps de développement d'un modèle particulier. En contrepartie, l'élaboration de tels modèles par apprentissage requiert la réalisation de mesures sur ledit processus, ce qui implique des coûts qui sont parfois loin d'être négligeables. Notre objectif a donc été d'identifier certains problèmes correspondant à la première approche, c'est-à-dire pour lesquels la réalisation de modèles de connaissance est soit inenvisageable soit particulièrement ardue. Le premier chapitre de ce mémoire s'attache à rappeler les concepts de base relatifs à la modélisation de processus par apprentissage. Nous y introduirons les notions essentielles que nous serons amenés à employer par la suite. Dans le deuxième chapitre, nous décrivons les principaux outils d'optimisation nécessaires à l'élaboration de modèles par apprentissage. Le troisième chapitre regroupe l'ensemble des travaux menés, au cours de cette thèse, sur le thème des réseaux de neurones. Après avoir rappelé la méthodologie d'élaboration de modèles neuronaux, en particulier dans le cas récurrent, nous présentons les résultats obtenus sur deux applications industrielles: l'estimation de la température en un point particulier de la ligne d'échappement, et l'estimation des émissions de différents polluants en sortie d'échappement. Ces deux applications participent à la maîtrise des émissions polluantes, soit durant l'utilisation habituelle d'un véhicule, car la connaissance de cette température est indispensable à la mise en œuvre des stratégies de dépollution actives, soit au stade de la mise au point du moteur, qui sera facilitée par l'utilisation d'un modèle de prédiction des débits de polluants en fonction des réglages du moteur. Nous décrivons également un système de commande optimale en boucle ouverte, associé à un modèle neuronal, et destiné à réduire les variations rapides de la sortie d'un processus: ce système est susceptible d'être utilisé pour contrôler les à-coups de couple d'un véhicule, consécutifs à une variation rapide de l'enfoncement de la pédale d'accélérateur. Une méthode de calcul exact de la matrice Hessienne, dans le cas de modèles décrits par des équations récurrentes, est alors introduite pour permettre l'utilisation de ce système de commande dans le cas de processus dynamiques. Dans le quatrième chapitre, nous nous intéressons aux méthodes de modélisation par noyaux, dont font partie les machines à vecteurs supports, et tentons de les adapter à la modélisation de processus dynamiques, d'abord par un traitement analytique (pour l'une particulière de ces méthodes), avant de proposer un approche itérative du problème d'apprentissage, inspirée de l'algorithme d'apprentissage semi dirigé utilisé pour les réseaux de neurones récurrents.

The aim of this research is to develop an optimisation tool to support the preliminary design of gas turbine combustors by providing a partial automation of the design process. This tool is to enable better design to be obtained faster, providing a reduction in the development costs and time to market of new engines. The first phase of this work involved the analysis of the combustor design process with the aim of identifying the critical tasks that are suitable for being automated and most importantly identifying the key parameters describing the performance of a combustor. During the second phase of this work an adequate design methodology for this problem was defined. This led to the development of a design optimisation Toolbox based on genetic algorithms, containing the tools required for it's proper integration into the combustor preliminary design environment. For the development of this Toolbox, extensive work was performed on genetic algorithms and derived techniques in order to provide the most efficient and robust optimisation method possible. The optimisation capability of the Toolbox was first validated and metered on analytical problems of known solution, where it demonstrated excellent optimisation performance especially for higher-dimensional problems. In a second step of the testing and validation process the combustor design capability of the Toolbox was demonstrated by applying it to diverse combustor design test cases. There the Toolbox demonstrated its capacity to achieve the required performance targets and to successfully optimise some key combustor parameters such as liner wall cooling flow and NOx emissions. In addition, the Toolbox demonstrated its ability to be applied to different types of engineering problems such as wing profile optimisation.

Lean burn engines are more fuel efficient than standard stoichiometric-burn engines but at the same time, the conventional three-way catalyst is not effective in reducing the NOx in oxygen-rich exhaust. One of the recent advancements in exhaust after treatment technologies for lean burn engines is the NOx storage and reduction (NSR) methodology. In this mechanism, NOx is stored on the storage component of a NSR catalyst during normal engine operation. However, before the catalyst reaches its saturation capacity, an excess of fuel is injected to the engine for a very short period resulting in reductant rich exhaust and during this period, NOx is released and subsequently reduced to N2, therefore, restoring the storage capacity of the catalyst. The operation is cyclic in nature, with the engine operating between an oxygen rich feed for long periods and a fuel rich feed for relatively shorter periods. To implement this technology in the most efficient way, a detailed understanding of the NSR chemistry under different operating conditions is required. For the past few years, several authors have studied the NSR systems using both experimental and modeling techniques. However, most of the models proposed in the literature were calibrated against the steady cyclic operation where the NOx profiles are similar for each cycle. In real life situations, the engine operation changes with different driving conditions, occurring due to sudden acceleration, roads in hilly areas, non-uniform braking, etc., which results in operation with a number of different transient cycle-to-cycle regimes depending upon the frequency with which the engine operation is altered. Due to such varying conditions, it is very important to investigate the significance of transients observed between the two different steady cycle-to-cycle operations for the optimization and control purposes. Also, the models in the literature are specific to the catalyst used in the study and therefore, their adaptation to other NSR catalysts is not straightforward. Therefore, one of the main motivations behind this research work is to develop a general approach to explain the storage dynamics. Moreover, the existing models have not studied the regeneration mechanisms, which is very important to explain the cyclic data in complete operation including both transients and steady state cycles. In this study, a pseudo one-dimensional model of a commercial NOx storage/release (NSR) catalyst is presented. The NOx storage is considered to be mass transfer limited, where as the storage proceeds, the barium carbonate particle is converted into the nitrate and for further storage, the NOx has to diffuse through this growing nitrate layer and a after certain depth, this penetration becomes nearly impossible. To explain the transient nature of the cyclic NOx profile, it is hypothesized that when incomplete regeneration occurs, only part of the nitrate is converted back to carbonate. Therefore, the nitrate layer increases in thickness with each cycle, thus making further storage increasingly more difficult. The shrinking core concept with incomplete storage in the lean phase followed by incomplete regeneration of the nitrate layer during the regeneration phase accounts for a net drop in storage capacity of the catalyst in each cycle, which continues decreasing until the amount of sites regenerated equal the amount used in NOx storage. The number of unknown parameters used for fitting were reduced by parameter sensitivity analysis and then fitted against a NOx profile at the reactor exit. The overall amount of NOx that can be stored in the lean phase of the cycle depends on the extent of regeneration that can be achieved during the previous rich phase, which in turn depends directly on the concentration of reductants in the feed. Therefore, there is a trade-off between the amount of fuel used and the NOx emissions. The proposed model can be potentially used to improve this trade-off by using model-based optimization techniques.

Lean burn engines are more fuel efficient than standard stoichiometric-burn engines but at the same time, the conventional three-way catalyst is not effective in reducing the NOx in oxygen-rich exhaust. One of the recent advancements in exhaust after treatment technologies for lean burn engines is the NOx storage and reduction (NSR) methodology. In this mechanism, NOx is stored on the storage component of a NSR catalyst during normal engine operation. However, before the catalyst reaches its saturation capacity, an excess of fuel is injected to the engine for a very short period resulting in reductant rich exhaust and during this period, NOx is released and subsequently reduced to N2, therefore, restoring the storage capacity of the catalyst. The operation is cyclic in nature, with the engine operating between an oxygen rich feed for long periods and a fuel rich feed for relatively shorter periods. To implement this technology in the most efficient way, a detailed understanding of the NSR chemistry under different operating conditions is required. For the past few years, several authors have studied the NSR systems using both experimental and modeling techniques. However, most of the models proposed in the literature were calibrated against the steady cyclic operation where the NOx profiles are similar for each cycle. In real life situations, the engine operation changes with different driving conditions, occurring due to sudden acceleration, roads in hilly areas, non-uniform braking, etc., which results in operation with a number of different transient cycle-to-cycle regimes depending upon the frequency with which the engine operation is altered. Due to such varying conditions, it is very important to investigate the significance of transients observed between the two different steady cycle-to-cycle operations for the optimization and control purposes. Also, the models in the literature are specific to the catalyst used in the study and therefore, their adaptation to other NSR catalysts is not straightforward. Therefore, one of the main motivations behind this research work is to develop a general approach to explain the storage dynamics. Moreover, the existing models have not studied the regeneration mechanisms, which is very important to explain the cyclic data in complete operation including both transients and steady state cycles. In this study, a pseudo one-dimensional model of a commercial NOx storage/release (NSR) catalyst is presented. The NOx storage is considered to be mass transfer limited, where as the storage proceeds, the barium carbonate particle is converted into the nitrate and for further storage, the NOx has to diffuse through this growing nitrate layer and a after certain depth, this penetration becomes nearly impossible. To explain the transient nature of the cyclic NOx profile, it is hypothesized that when incomplete regeneration occurs, only part of the nitrate is converted back to carbonate. Therefore, the nitrate layer increases in thickness with each cycle, thus making further storage increasingly more difficult. The shrinking core concept with incomplete storage in the lean phase followed by incomplete regeneration of the nitrate layer during the regeneration phase accounts for a net drop in storage capacity of the catalyst in each cycle, which continues decreasing until the amount of sites regenerated equal the amount used in NOx storage. The number of unknown parameters used for fitting were reduced by parameter sensitivity analysis and then fitted against a NOx profile at the reactor exit. The overall amount of NOx that can be stored in the lean phase of the cycle depends on the extent of regeneration that can be achieved during the previous rich phase, which in turn depends directly on the concentration of reductants in the feed. Therefore, there is a trade-off between the amount of fuel used and the NOx emissions. The proposed model can be potentially used to improve this trade-off by using model-based optimization techniques.

Internal combustion engines have dealt with increasingly restricted emissions requirements. After-treatment devices are successful in bringing emissions into compliance, but in-cylinder combustion control can reduce their burden by reducing engine out emissions. For example, oxides of nitrogen (NOx) are diesel combustion exhaust species that are notoriously difficult to remove by after-treatment. In-cylinder conditions can be controlled for low levels of NOx, but this produces high levels of soot potentially leading to increased particulate matter (PM). The simultaneous reduction of NOx and PM can be realized through a combustion process known as low temperature combustion (LTC). In this study, the typical definition of LTC as the defeat of the inverse relationship between soot and NOx is not applicable as a return to the soot-NOx tradeoff is observed with increasing exhaust gas recirculation (EGR). It is postulated that this effect is the result of an increase in the hot ignition equivalence ratio, moving the combustion event into a slightly higher soot formation region. This is important because a simple emissions based definition of LTC is no longer helpful. In this study, the manifestation of LTC in the calculated heat release profile is investigated. The conditions classified as LTC undergo a two-stage ignition process. Two-stage ignition is characterized by an initial cool-flame reaction followed by typical hot ignition. In traditional combustion conditions, the ignition is fast enough that a cool-flame is not observed. By controlling initial conditions (pressure, temperature, and composition), the creation and duration of the cool-flame event is predictable. Further, the effect that injection timing and the exhaust gas recirculation level have on the controlling factors of the cool-flame reaction is well correlated to the duration of the cool-flame event. These two results allow the postulation that the presence of a sufficiently long cool-flame reaction indicates a combustion event that can be classified as low temperature combustion. A potential method for identifying low temperature combustion events using only the rate of heat release profile is theorized. This study employed high levels of EGR and late injection timing to realize the LTC mode of ordinary petroleum diesel fuel. Under these conditions, and based on a 90 percent reduction in nitric oxide and no increase in smoke output relative to the chosen baseline condition, a two part criteria is developed that identifies the LTC classified conditions. The criteria are as follow: the combustion event of conventional petroleum diesel fuel must show a two-stage ignition process; the first stage (cool-flame reaction) must consume at least 2 percent of the normalized fuel energy before the hot ignition commences.

Energy need in transportation and industry is mainly met by fossil fuels. This causes consumption of resources and some environmental problems. Diesel and gasoline engines are developed to consume fuel efficiently in vehicles. Since these engines work in a low fuel to air ratio, it becomes difficult to reduce nitrogen oxide emission. For this reason NO x storage/reduction (NSR) catalysts have been developed. While engines are operating under lean conditions alkaline or alkaline-earth component of NSR catalysts capture nitrogen oxides and during fuel rich period stored nitrates are reduced to nitrogen and oxygen gases. To develop this technology, different system parameters, for example system components and reaction environments have been widely investigated experimentally. To supplement the experimental findings, binding energies and structural properties of NO x on different catalyst components have been investigated theoretically. It has been experimentally observed that adding TiO2 to other conventional support materials increases resistance against sulfur poisoning, which is one of the main problems concerning NSR catalysts. For this reason, in this thesis (001) and (101) anatase surfaces have been investigated. Moreover, the effects of barium oxide units and layers on the electronic properties of the (001) anatase surface have been studied. To observe the effects of TiO2 as a support component, interactions of NO2 and SO2 on the unsupported and anatase supported (100) BaO surfaces have been compared. A clear increase in sulfur resistance has been observed in the presence of TiO2 in the catalyst under certain conditions.

A comprehensive model, previously developed for a rectangular parallelepiped shaped 0.3 MWt circulating fluidized bed combustor (CFBC) fired with high calorific value coal burning in sand and validated against experimental data is adapted to cylindrical configuration and is extended to incorporate NOx formation and reduction reactions and pressure drops around cyclone, downcomer and loop seal. Its predictive accuracy is tested by applying it to the simulation of Middle East Technical University (METU) 150 kWt CFBC burning low calorific value indigenous lignite with high Volatile Matter/Fixed Carbon (VM/FC) ratio in its own ash and comparing its predictions with measurements. Favorable comparisons are obtained between the predicted and measured temperatures, pressure profiles and emissions of gaseous species. Results reveal that predictive accuracy in pressure profile strongly depends on the correlation utilized for entrainment in dilute zone and that accuracy in NO emission requires data on partitioning of coal nitrogen into char-N and volatile-N and is affected significantly by dilute zone oxygen content.

No description available.

This study utilized Cu-catalysts to catalyze a NO reduction reaction using CH4 as a reductant. Due to CH4 being a weak reductant and is easily affected by O2 concentration, we undertook a series experiments with O2 concentration and promoters, so that we could better understand their influence. The experiment conditions were as follows : reaction temperature between 150¢J- 800¢J¡Fa catalysts weight of 0.5 g¡F total gas flow rate of 1000 ml/min¡Frelative humidity at 0.9 %¡Fan O2 concentration between 0 - 6%¡Fand CH4 concentration between 1000 - 10000 ppm. First, we sorted out the best metal carriers and calcining temperature, from this we decide to use £^-Al2O3 as a carrier with a calcining temperature under 500¢Jto produce our catalysts. During the O2 concentration experiment, when the inflow O2 concentration was below 1% (including 0% O2), Cu-catalysts reduce NO above 550¢J.The conversion reached a rate of 95 % at a temperature of 750¢J¡Fwhen the oxygen concentration was between 3% and 6% O2, catalysts reacted within 300 - 500¢J with NO converting to NO2¡Fat a concentration between 1.5% and 2% O2, NOx underwent reduction at 750¢J,and NOx conversion raised from 0 % to above 90%. Therefore in analyzing the experiment results, it shows that NOx will reduce violently when the O2 concentration is below 0.7% and while using CH4 as a reductant. This result was also apparent in O2 concentrations between 1.5 % and 2%. In the experiments of M/O ratio (the ratio of CH4 and O2 inflow), we found M/O ratio was not a deciding factor within the reaction mechanics, furthermore the limiting factor of O2 concentration decreases under 0.7%¡Fin addition it was also found that adding large amounts of CH4 could quicken the reduction process. Lastly, a mass balance was performed, which had a result over 70 %. In the experiments where Y¡BLa¡BSr¡BCo were added as promoters to the Cu-catalysts, we found that Cu-La/£^-Al2O3¡BCu-Sr/£^-Al2O3 and Cu-Co/£^-Al2O3 can accelerate O2 depletion. Henceforth, it is possible to deduce promoters will be a useful method in solving O2 limiting. In the comparison of metals loading methods, we found no difference in activity using separate-impregnation and co-impregnation methods, whereas in the BET and SEM co-impregnation experiments, there was a larger surface-area and dispersion.

Two flow reactor studies, using an electrically heated laminar flow reactor over Vanadia based (V2O5-WO3/TiO2) honeycomb catalyst, were performed at 1 atm pressure and various temperatures. The experiments were conducted using simulated exhaust gas compositions for different exhaust gases. A quartz tube was used in order to establish inert conditions inside the reactor. The experiments utilized a Fourier transform infrared (FTIR) spectrometer in order to perform both qualitative and quantitative analysis of the reaction products. Urea-water solution decomposition was investigated over V2O5-WO3/TiO2 catalyst over the entire SCR temperature range using the temperature controlled flow reactor. The solution was preheated and then injected into pure nitrogen (N2) stream. The decomposition experiments were conducted with a number of oxygen (O2) compositions (0, 1, 10, and 15%) over the temperature range of 227oC to 477oC. The study showed ammonia (NH3), carbon-dioxide (CO2) and nitric oxide (NO) as the major products of decomposition along with other products such as nitrous oxide (N2O) and nitrogen dioxide (NO2). The selective catalytic reduction (SCR) of nitric oxide (NO) with urea-water solution over V2O5-WO3/TiO2 catalyst using a laboratory laminar-flow reactor was investigated. Urea-water solution was injected at a temperature higher than the vaporization temperature of water and the flow reactor temperature was varied from 127oC to 477oC. A FTIR spectrometer was used to determine the concentrations of the product species. The major products of SCR reduction were NH3, NO and CO2 along with the presence of other minor products NO2 and N2O. NO removal of up to 87% was observed. The aim of the urea-water decomposition experiments was to study the decomposition process as close to the SCR configuration as possible. The aim of the SCR experiments was to delineate the effect of various parameters including reaction temperature and O2 concentration on the reduction process. The SCR investigation showed that changing parameter values significantly affected the NO removal, the residual NH3 concentration, the temperature of the maximum NO reduction, and the temperature of complete NH3 conversion. In the presence of O2, the reaction temperature for maximum NO reduction was 377?C for ratio of 1.0.