A otimização em tempo real (RTO: Real Time Optimization) permite fazer correções, com o menor atraso possível, nas condições de operação de um processo, buscando manter um desempenho ótimo. A RTO, na abordagem clássica, requer um ciclo constante de análise e correção do estado do processo que inclui múltiplas etapas: i) identificação do estado estacionário, ii) identificação e correção dos erros grosseiros, iii) reconciliação dos dados, iv) ajuste dos parâmetros do modelo, v) estimação das condições operacionais ótimas e, vi) implementação das condições no sistema de controle. Um sistema de análise de dados é necessário na aquisição das variáveis medidas da planta que classifique o estado da operação como válido para atualizar um modelo em estado estacionário. O sistema deve fornecer um modelo atualizado que seja representativo do comportamento real da operação para que seja otimizado em um passo posterior. Este trabalho é focado na análise de metodologias de detecção de estado estacionário, reconciliação de dados e estimação de parâmetros com as características necessárias que um sistema de RTO requer. Como caso de estudo considera-se uma coluna depropenizadora da PETROBRAS, em que foi feita uma análise da variabilidade associada à instrumentação para melhorar o conhecimento da operação. Além disso, a análise e validação do modelo do processo, permitiu estabelecer faixas de convergência nas especificações do processo e parâmetros. Finalmente, sugere-se que a composição da corrente de alimentação e a queda de pressão na coluna sejam classificadas como parâmetros importantes no ajuste de modelo. / Real-time optimization (RTO) allows making corrections in process operation conditions, with the smallest possible delay, in order to provide an optimal performance. RTO, in the classical approach, requires a constant cycle of analysis and correction of process conditions, that includes multiple steps: i) steady state identification, ii) gross errors detection and correction, iii) data reconciliation, iv) parameter estimation, v) economical optimization and vi) implementation of the optimal conditions in the control system. A data analysis system is required in the plant to classify the operating conditions as valid in order to update a steady state model. The system shall provide an updated model that can represent the real behavior of the operation that wi ll be optimized in a next step. This work is focused on the analysis of methodologies for steady-state detection, data reconciliation and parameter estimation with the required characteristics that an RTO system requires. As a case-study, a depropenizer column, owned by Petrobras is considered. An analysis of the variability of the instrumentation was performed to allow a better understanding of the process. Moreover, the analysis and validation of the process model enabled drawing convergence boundaries on process specifications and parameters. Finally, it is suggested that the feed composition and the column pressure drop should be considered as important parameters in model fitting.

A otimização em tempo real (RTO: Real Time Optimization) permite fazer correções, com o menor atraso possível, nas condições de operação de um processo, buscando manter um desempenho ótimo. A RTO, na abordagem clássica, requer um ciclo constante de análise e correção do estado do processo que inclui múltiplas etapas: i) identificação do estado estacionário, ii) identificação e correção dos erros grosseiros, iii) reconciliação dos dados, iv) ajuste dos parâmetros do modelo, v) estimação das condições operacionais ótimas e, vi) implementação das condições no sistema de controle. Um sistema de análise de dados é necessário na aquisição das variáveis medidas da planta que classifique o estado da operação como válido para atualizar um modelo em estado estacionário. O sistema deve fornecer um modelo atualizado que seja representativo do comportamento real da operação para que seja otimizado em um passo posterior. Este trabalho é focado na análise de metodologias de detecção de estado estacionário, reconciliação de dados e estimação de parâmetros com as características necessárias que um sistema de RTO requer. Como caso de estudo considera-se uma coluna depropenizadora da PETROBRAS, em que foi feita uma análise da variabilidade associada à instrumentação para melhorar o conhecimento da operação. Além disso, a análise e validação do modelo do processo, permitiu estabelecer faixas de convergência nas especificações do processo e parâmetros. Finalmente, sugere-se que a composição da corrente de alimentação e a queda de pressão na coluna sejam classificadas como parâmetros importantes no ajuste de modelo. / Real-time optimization (RTO) allows making corrections in process operation conditions, with the smallest possible delay, in order to provide an optimal performance. RTO, in the classical approach, requires a constant cycle of analysis and correction of process conditions, that includes multiple steps: i) steady state identification, ii) gross errors detection and correction, iii) data reconciliation, iv) parameter estimation, v) economical optimization and vi) implementation of the optimal conditions in the control system. A data analysis system is required in the plant to classify the operating conditions as valid in order to update a steady state model. The system shall provide an updated model that can represent the real behavior of the operation that wi ll be optimized in a next step. This work is focused on the analysis of methodologies for steady-state detection, data reconciliation and parameter estimation with the required characteristics that an RTO system requires. As a case-study, a depropenizer column, owned by Petrobras is considered. An analysis of the variability of the instrumentation was performed to allow a better understanding of the process. Moreover, the analysis and validation of the process model enabled drawing convergence boundaries on process specifications and parameters. Finally, it is suggested that the feed composition and the column pressure drop should be considered as important parameters in model fitting.

No description available.

According to the World Health Organization, almost all human population (99%) lives in 117 countries with over 6000 cities, where air pollutant concentration exceeds recommended thresholds. The most common, so-called criteria, air pollutants that affect human lives, are particulate matter (PM) and gas-phase (SO2, CO, NO2, O3 and others). Therefore, many countries or regions worldwide have imposed regulations or interventions to reduce these effects. Whenever an intervention occurs, air quality changes due to changes in ambient factors, such as weather characteristics and human activities. One approach for assessing the effects of interventions or events on air quality is through the use of the Weather Normalized Model (WNM). However, current deterministic models struggle to accurately capture the complex, non-linear relationship between pollutant concentrations and their emission sources. Hence, the primary objective of this thesis is to examine the power of machine learning (ML) and deep learning (DL) techniques to develop and improve WNMs. Subsequently, these enhanced WNMs are employed to assess the impact of events on air quality. Furthermore, these ML/DL-based WNMs can serve as valuable tools for conducting exploratory data analysis (EDA) to uncover the correlations between independent variables (meteorological and temporal features) and air pollutant concentrations within the models.  It has been discovered that DL techniques demonstrated their efficiency and high performance in different fields, such as natural language processing, image processing, biology, and environment. Therefore, several appropriate DL architectures (Long Short-Term Memory - LSTM, Recurrent Neural Network - RNN, Bidirectional Recurrent Neural Network - BIRNN, Convolutional Neural Network - CNN, and Gated Recurrent Unit - GRU) were tested to develop the WNMs presented in Paper I. When comparing these DL architectures and Gradient Boosting Machine (GBM), LSTM-based methods (LSTM, BiRNN) have obtained superior results in developing WNMs. The study also showed that our WNMs (DL-based) could capture the correlations between input variables (meteorological and temporal variables) and five criteria contaminants (SO2, CO, NO2, O3 and PM2.5). This is because the SHapley Additive exPlanations (SHAP) library allowed us to discover the significant factors in DL-based WNMs. Additionally, these WNMs were used to assess the air quality changes during COVID-19 lockdown periods in Ecuador. The existing normalized models operate based on the original units of pollutants and are designed for assessing pollutant concentrations under “average” or consistent weather conditions. Predicting pollution peaks presents an even greater challenge because they often lack discernible patterns. To address this, we enhanced the Weather Normalized Models (WNMs) to boost their performance specifically during daily concentration peak conditions. In the second paper, we accomplished this by developing supervised learning techniques, including Ensemble Deep Learning methods, to distinguish between daily peak and non-peak pollutant concentrations. This approach offers flexibility in categorizing pollutant concentrations as either daily concentration peaks or non-daily concentration peaks. However, it is worth noting that this method may introduce potential bias when selecting non-peak values. In the third paper, WNMs are directly applied to daily concentration peaks to predict and analyse the correlations between meteorological, temporal features and daily concentration peaks of air pollutants.

No description available.

L’énergie photovoltaïque jouera un rôle déterminant dans la transition énergétique future. Bien que les cellules solaires à base de silicium dominent encore le marché, leur coût de fabrication et le poids des modules limitent leur développement. Depuis quelques années, les industriels s’intéressent de plus en plus aux dispositifs à base de couches minces en raison de leurs procédés de fabrication rapides et peu onéreux sur de larges substrats. Cette technologie utilise une large variété de matériaux; les chalcopyrites tels que Cu(In,Ga)Se2 sont les plus prometteurs. Le procédé de fabrication de couches chalcopyrites le plus répandu est la coévaporation mais l’utilisation de vides très poussés rende cette technique peu adaptée à la production à grande échelle de modules bon marché. La solution alternative décrite dans ce travail est un procédé en deux étapes basé sur le recuit sous atmosphère réactive de précurseurs métalliques électrodéposés. Le développement de cette technologie passe par une meilleure compréhension des mécanismes d’incorporation et d’homogénéisation du gallium dans les couches formées et par une optimisation des étapes de recuit. Le premier objectif de ce travail de thèse est une étude des mécanismes réactionnels mis en jeu lors du procédé de recuit à travers l’étude de différents types de précurseur. Par la suite ces connaissances sont utilisées pour modéliser et optimiser un recuit industriel innovant. Ce travail est réalisé à l’aide de plans d’expérience (DOE) où l’influence de certains paramètres, les plus critiques est mise en évidence. Des voies d’optimisation sont proposées et des hypothèses faites afin d’expliquer les phénomènes observés. / Solar energy is promised to be a major actor in the future of energy production. Even if silicon based solar cells remain the main product their fabrication is energy consuming and requires heavy cover glass for protection, which reduce their development. For several years, commercial interest has shifted towards thin-film cells for which manufacturing time, large scale production, fabrication costs and weight savings are the main advantages. For thin film technology, a wide variety of materials can be used but chalcopyrite such as Cu(In,Ga)Se2 is one of the most promising. The most current method used for chalcopyrite formation is co- evaporation but this process is very expensive and not well suitable for large scale production due to high vacuum requirements. One alternative solution described in this work consists of a two-step technology based on the sequential electro-deposition of a metallic precursor followed by a rapid reactive annealing. However to reach its full potential this technology needs a better understanding of the Ga incorporation mechanism and of the selenization/sulfurization step. This work focuses first on formation mechanisms through the study of several kinds of precursor. This knowledge is then used to explain and to optimize innovative annealing processes. This study is achieved by observing the impact of some process parameters using designs of experiment (DOE). A link between process parameters and properties of these thin films is obtained using electrical, structural and diffusion characterization of the devices. Finally we propose hypothesis to explain observed phenomena and also some improvements to meet the challenges of this process.

The aim of this thesis is the investigation of modeling and optimization particularities of tubular polymerization reactors. The original work is divided in two sections, the first treating a modeling and optimization study of tubular reactors for methyl methacrylate polymerization in solution, and the second, an experimental and theoretical study of L-lactide reactive extrusion. In the first section, reactor simulations in similar operating conditions were performed in order to select a representative kinetic model among the published kinetic models for MMA solution polymerization. Two widely used numerical algorithms, one based on Pontryagin's Minimum Principle and the other a Genetic Algorithm, were compared for an average-complexity optimization problem. The results showed a superior robustness of the Genetic Algorithm for this category of problems. The second part of the thesis deals with the modeling and optimization of L-lactide reactive extrusion. A kinetic model is proposed and its parameters estimated using nonlinear estimation numerical procedures based on experimentally measured data. Reactive extrusion experiments were performed in representative operating conditions. The Llactide/ polylactide flow in the extruder was characterized by simulation using the commercial software LUDOVIC®. The simulated residence time distributions characteristics are used to model the reactive extrusion process of two approaches, an axial dispersion model and a compartment model, based on compartments whose characteristics are deduced from the simulations using LUDOVIC®. The modeling results are in good agreement with the measured data in the same operating conditions.

Dans un contexte ou la question de la préservation de l'environnement est devenue un sujet sociétal majeur, la recherche de nouvelles technologies pour remplacer la voiture à essence constitue un véritable enjeu industriel. Les véhicules hybrides et électriques sont une alternative prometteuse aux véhicules conventionnels. Ce travail de thèse porte sur la conception et l'optimisation des machines électriques pour la motorisation de ces voitures.Au cours de ces travaux, nous avons développé une nouvelle méthodologie de dimensionnement et d'optimisation des machines synchrones à double excitation. L'intérêt de cette méthode est son bon compromis entre la précision et le temps de calcul et sa capacité d'adaptation à plusieurs types de machines. Le second volet de la thèse est consacré à la proposition d'une nouvelle structure de machine synchrone à rotor bobiné. Une technique originale de compensation de la réaction magnétique d'induit est mise en place, elle consiste en l'insertion d'aimants secondaires permettant de créer un flux dans l'axe q de la machine. Ce dernier a pour rôle d'affaiblir le flux de la réaction magnétique d'induit. / In a context where the question of the environmental protection has become a major social problem, research new technologies to replace the gasoline car is a real industrial challenge. The hybrid and electric vehicles are a promising alternative to conventional vehicles. This thesis focuses on the design and optimization of electrical machines for electric and hybrid cars.In this work, we developed a new methodology for design and optimization of hybrid excitation synchronous machines. The advantage of this method is its good compromise between accuracy and computation time and its ability to be adapted to a wide range of machines. The second part of this thesis is devoted to the development of a new structure of a wound rotor synchronous machine. A novel technique for compensating the armature reaction of this machine is introduced, it involves the insertion of secondary magnets to produce a quadratic axis flux (q-axis), this latter has the function of weakening the armature reaction flux.

Le but de cette thèse est l’investigation des particularités des problèmes d’optimisation et modélisation des réacteurs tubulaires de polymérisation. La partie originale du travail est divisé en deux sections : la première traitant de l'étude théorique de la modélisation et de l’optimisation des réacteurs tubulaires de polymérisation du méthacrylate de méthyle en solution, et la deuxième, une étude expérimentale et théorique de l'extrusion réactive de L-lactide. Dans la première partie, afin de sélectionner un modèle cinétique représentatif, parmi les modèles publiés pour le processus de polymérisation de MMA, des simulations ont été effectuées en conditions identiques de fonctionnement. Deux algorithmes numériques, l’un basé sur le Principe du Minimum de Pontriaguine et l’autre de type Génétique, ont été comparés pour un problème d'optimisation de complexité moyenne. Les résultats ont montré une robustesse supérieure de l’Algorithme Génétique pour cette catégorie de problèmes. La deuxième partie de la thèse est consacrée à la modélisation et à l’optimisation de l'extrusion réactive du Llactide. Nous avons proposé un modèle cinétique et ses paramètres ont été estimés en utilisant des procédures numériques basées sur les données cinétiques expérimentales. Les expériences d'extrusion réactives ont été exécutées dans les conditions de fonctionnement représentatives. L'écoulement de L-lactide/polylactide dans l'extrudeuse a été caractérisé par la simulation en utilisant un logiciel commercial, LUDOVIC®. Les caractéristiques de la distribution des temps de séjour simulées sont utilisées pour modéliser le processus d'extrusion réactive en utilisant deux approches, un modèle à dispersion axiale et un modèle à base de compartiments, dont les caractéristiques sont déduites des simulations effectuées avec LUDOVIC®. Les résultats de la modélisation du processus sont en bon accord avec des données mesurées en mêmes conditions opératoires. / The aim of this thesis is the investigation of modeling and optimization particularities of tubular polymerization reactors. The original work is divided in two sections, the first treating a modeling and optimization study of tubular reactors for methyl methacrylate polymerization in solution, and the second, an experimental and theoretical study of L-lactide reactive extrusion. In the first section, reactor simulations in similar operating conditions were performed in order to select a representative kinetic model among the published kinetic models for MMA solution polymerization. Two widely used numerical algorithms, one based on Pontryagin’s Minimum Principle and the other a Genetic Algorithm, were compared for an average-complexity optimization problem. The results showed a superior robustness of the Genetic Algorithm for this category of problems. The second part of the thesis deals with the modeling and optimization of L-lactide reactive extrusion. A kinetic model is proposed and its parameters estimated using nonlinear estimation numerical procedures based on experimentally measured data. Reactive extrusion experiments were performed in representative operating conditions. The Llactide/ polylactide flow in the extruder was characterized by simulation using the commercial software LUDOVIC®. The simulated residence time distributions characteristics are used to model the reactive extrusion process of two approaches, an axial dispersion model and a compartment model, based on compartments whose characteristics are deduced from the simulations using LUDOVIC®. The modeling results are in good agreement with the measured data in the same operating conditions.

Desde a prospecção do minério bauxita, passando pelo seu beneficiamento até a sua entrada no processo Bayer, tem-se como principais índices de qualidade e de processo os parâmetros químicos: alumina aproveitável (Al2O3Ap) e sílica reativa (SiO2Re), determinados segundo um procedimento que simula a digestão Bayer em escala de laboratório. Uma grande inovação para a indústria da bauxita seria fazer o controle por parâmetros mineralógicos, % gibbsita e % caulinita, via difratometria de raios X, intenção buscada nesse trabalho pela proposta de um método combinado Rietveld-Le Bail-Padrão Interno, cujos resultados são bem promissores para bauxitas gibbsíticas tipo-Paragominas, matriz para qual foi desenvolvido. Tal combinação não só melhorou a qualidade da quantificação de gibbsita e caulinita, como diminuiu o peso de cálculo tornando o procedimento mais prático e rápido. A alta correlação (r2=0,99) entre os resultados mineralógicos pelo método combinado e os resultados químicos pelo método tradicional, os deixam em igual escolha, pois foram iguais estatisticamente. No entanto, ressalta-se que o método tradicional subestima o valor de caulinita pela conversão da SiO2Re, enquanto o método combinado se aproxima mais do valor verdadeiro. Obter um resultado pelo método combinado mostrou ser mais prático e rápido que pelo método tradicional. Enquanto o tempo total estimado pelo combinado é < 3 h, pelo tradicional é de no mínimo 6 h. Como proposta de validação do método combinado, um segundo foi desenvolvido para quantificação de Al-goethita por DSC, o qual mostrou boa precisão. E muito embora o uso da técnica no controle industrial seja pouco provável por questões de praticidade e tempo de análise, usá-la na validação de antigos e novos métodos de quantificação mineralógica de bauxitas pode ser muito útil. A ordem crescente de substituição de Fe por Al pretendida pelas sínteses planejadas (7 variedades) foi confirmada pelos resultados de DRX, FRX, DSC e MEV, e assim um pequeno banco de dados de entalpias padrão de desidroxilação de Al-goethitas foi estabelecido. A produção de padrões complexos, misturas de variedades goethíticas, é tão importante quanto produzir uma só goethita, pois tais misturas são termodinamicamente comuns na natureza e, portanto, comuns em bauxitas. Após uma identificação clara da limitação do método tradicional para estimar caulinita pela conversão de SiO2Re em bauxitas tipo-Paragominas, um estudo de otimização do método Alcan foi realizado com base em um planejamento fatorial completo 23. As variáveis escolhidas foram temperatura, concentração cáustica e tempo para duas situações: bauxita com baixa SiO2Re e bauxita com alta SiO2Re. A temperatura foi a variável mais importante, apresentando um efeito positivo sobre a quantidade de SiO2Re, uma vez que o aumento na temperatura aumentou a taxa de conversão completa de caulinita em sodalita. Modelos empíricos de 1ª ordem foram apropriadamente obtidos para predição da quantidade de SiO2Re como função da temperatura, concentração cáustica e tempo, os quais responderam com as seguintes condições ótimas: (1) sem presença significante de quartzo - temperatura de 180 °C, concentração cáustica de 10 % com tempo de 60 min para baixa SiO2Re e 25 min para alta SiO2Re, e (2) com presença significante de quartzo - temperatura de 150 °C, concentração cáustica de 20 % e tempo de 60 min, para ambas as situações estudadas. / In the bauxite industry - exploration, beneficiation and refinery - two main chemical parameters are used for the quality control: available alumina (AvAl2O3) and reactive silica (RxSiO2). They are determined by a procedure that simulates the Bayer process in laboratory scale. A great innovation for this industry would be to make this control by mineralogical parameters, i.e., the % of gibbsite and % of kaolinite via Powder X-ray Diffraction Analysis. This is one of the main purposes of this work by means of a combined Rietveld-Le Bail-Internal Standard Method, whose results were very promising for the Paragominas-type bauxites. This combination not only improved the quality of gibbsite and kaolinite quantification, as decreased computer processing time, making it a more convenient and fast procedure. The high correlation (r2=0.99) between the mineralogical results from the combined method and chemical results by the traditional method, leave them the same choice, as they were statistically equal. However, it is noteworthy that the traditional method underestimates the kaolinite value obtained from the conversion of RxSiO2, while the combined method is closer to the true value. Obtaining a result by the combined method proved to be more convenient and faster (< 3 hours) than the traditional method (at least 6 hours). As a validation for the proposed combined method, a second method was developed to quantify Al-goethite by DSC, which showed good accuracy. Although the use of DSC technique in industrial control is unlikely for practical reasons and analysis time, its use can be very helpful in the validation of old and new methods for the mineralogical quantification of bauxites. XRD, XRF, DSC and SEM results confirmed the increasing order of Al for Fe replacement intended for the planned synthesis (7 types). Thus, a small database of standard enthalpies of Al-goethites dehydroxylation was built. The production of standards of goethites mixtures is as important as producing a single goethite standard, because these are thermodynamically common in nature and thus bauxites with complex mixtures of goethites are also common. After clearly identifying the limitations of the traditional method to estimate kaolinite from the conversion of RxSiO2 in the Paragominas-type bauxites, an optimization study of the Alcan method was carried out based on a 23 full factorial design. The chosen variables were temperature, caustic concentration and time, for two main situations: bauxite with low RxSiO2 and bauxite with high RxSiO2. The temperature was the most important variable, with a positive effect on the amount of RxSiO2, since the increase in temperature increased the rate of full kaolinite to sodalite conversion. First-order empirical models were properly obtained to predict the amount of RxSiO2 as a function of temperature, caustic concentration and time, which responded to the following optimal conditions: (1) without significant amount of quartz - 180 °C, NaOH 10 % and 60 min for low RxSiO2 and 25 min for high SiO2Re, and (2) with significant amount of quartz - 150 °C, NaOH 20 % and 60 min for both situations.