A combretastatina A4 é um promissor agente anticâncer. Na célula, inibe a polimerização dos microtúbulos, os quais são fundamentais nos processos de motilidade, manutenção estrutural e mitose. Essa inibição se dá a partir do sítio de interação da αβ-tubulina bloqueando o fluxo do sangue que alimenta os tumores, o que resulta na morte dos mesmos. Com estrutura semelhante às combretastatinas, as chalconas constituem uma classe de compostos que atuam no mesmo sítio de interação na tubulina. Baseando-se nos trabalhos experimentais de Ducki e colaboradores, estudou-se a estrutura molecular de 87 chalconas análogas à combretastatina A4 por meio do método quântico DFT com o propósito de desenvolver modelos de Relações Quantitativas Estrutura-Atividade (QSAR) aplicados a tais antagonistas. A partir dos métodos dos Mínimos Quadrados Parciais (PLS) e de Redes Neurais Artificiais (ANN), foram gerados modelos que conduzem à elucidação da relação dos compostos estudados com suas respectivas atividades biológicas. Os descritores eletrônicos e moleculares selecionados apresentam alta concordância com as características das moléculas, havendo predominância de comportamento linear com a atividade biológica, podendo, eventualmente, apresentar comportamento não-linear, o que torna o modelo gerado altamente consistente. / Combretastatin A4 is a promising anticancer agent. It inhibits the polymerization of microtubules in the cell, which are essential in the process of motility, structural maintenance and mitosis. This inhibition is given from the interaction site of αβ-tubulin blocking the blood flow that feeds the tumor, what results in its death. The chalcones, sharing a similar structure of the combretastatin, are also a class of compounds that act in the same site of interaction in the tubulin. Based on the experimental work of Ducki and co-workers, we proposed a molecular structure study of 87 chalcones similar to combretastatin A4 using the DFT method in order to develop Quantitative Structure-Activity Relationships (QSAR) applied to the given antagonists. Through Partial Least Squares (PLS) and Artificial Neural Network (ANN) methods, some models has been generated to lead the understanding on the relationship between the compounds studied and their respective biological activities. The electronic and molecular descriptors selected have high correlation with the molecule features, being linear most of the time, although with eventual non-linear behavior, which makes the generated model highly consistent.

Les turbomachines aéronautiques sont composées de plusieurs roues aubagées dont la fonction estde transférer l’énergie de l’air au rotor. Les roues aubagées des modules compresseur et turbine sontdes pièces particulièrement sensibles car elles doivent répondre à des impératifs de performanceaérodynamique, de tenue mécanique, de tenue thermique et de performance acoustique. L’optimisation aéro-méca-acoustique ou aéro-thermo-mécanique des aubages consiste à chercher, pourun ensemble de formes aérodynamiques paramétrées (par plusieurs dizaines de variables), celleassurant le meilleur compromis entre la performance aérodynamique du moteur et la satisfactionde plusieurs dizaines de contraintes souvent contradictoires. Cette thèse introduit une méthode d’optimisation basée sur les métamodèles et adaptée à la grande dimension pour répondre à la problématique industrielle des aubages. Les contributions de cettethèse portent sur deux aspects : le développement de modèles de krigeage, et l’adaptation d’unestratégie d’optimisation pour la gestion du grand nombre de variables et de contraintes.La première partie de ce travail traite des modèles de krigeage. Nous avons proposé une nouvelleformulation du noyau de covariance permettant de réduire le nombre de paramètres du modèleafin d’accélérer sa construction. Une des limitations connues du modèle de krigeage concernel’estimation de ses paramètres. Cette estimation devient de plus en plus difficile lorsque nousaugmentons la dimension du phénomène à approcher. En particulier, la base de données nécessitedavantage de points et par conséquent la matrice de covariance du modèle du krigeage est de plusen plus coûteuse à inverser. Notre approche consiste à réduire le nombre de paramètres à estimer en utilisant la méthode de régression des moindres carrés partiels (PLS pour Partial Least Squares). Cette méthode de réduction dimensionnelle fournit des informations sur la relation linéaire entre les variables d’entrée et la variable de sortie. Ces informations ont été intégrées dans les noyaux du modèle de krigeage tout en conservant les propriétés de symétrie et de positivité des noyaux. Grâce à cette approche, la construction de ces nouveaux modèles appelés KPLS est très rapide étant donné le faible nombre de paramètres nécessaires à estimer. La validation de ces modèles KPLS sur des cas test académiques ou industriels a démontré leur qualité de prédiction équivalente voire même meilleure que celle des modèles de krigeage classiques. Dans le cas de noyaux de covariance de type exponentiel, laméthode KPLS peut être utilisée pour initialiser les paramètres du krigeage classique, afin d’accélérerla convergence de l’estimation des paramètres du modèle. La méthode résultante, notée KPLS+K, a permis d’améliorer la qualité des modèles dans le cas de fonctions fortement multimodales. La deuxième contribution de la thèse a consisté à développer une stratégie d’optimisation globale sous contraintes pour la grande dimension, en s’appuyant sur les modèles KPLS ou les modèlesKPLS+K. En effet, nous avons étendu la méthode d’optimisation auto-adaptative connue dans lalittérature sous le nom "Efficient Global Optimisation, EGO" pour gérer les problèmes d’optimisationsous contraintes en grande dimension. Différents critères d’enrichissement adaptatifs ont pu êtreexplorés. Cette stratégie a permis de retrouver l’optimum global sur des problèmes académiquesjusqu’à la dimension 50. La méthode proposée a été confrontée à deux types de problèmes industriels, le cas test MOPTA issu de l’industrie automobile (124 variables d’entrée et 68 fonctions contraintes) et le cas test Snecma des aubes de turbomachines (50 variables d’entrée et 31 fonctions contraintes). Les résultats ont permis de montrer la validité de la démarche ainsi que les limites de la méthode pour une application dans un cadre industriel. / Aerospace turbomachinery consists of a plurality of blades. Their main function is to transfer energybetween the air and the rotor. The bladed disks of the compressor are particularly important becausethey must satisfy both the requirements of aerodynamic performance and mechanical resistance.Mechanical and aerodynamic optimization of blades consists in searching for a set of parameterizedaerodynamic shape that ensures the best compromise solution between a set of constraints.This PhD introduces a surrogate-based optimization method well adapted to high-dimensionalproblems. This kind of high-dimensional problem is very similar to the Snecma’s problems. Ourmain contributions can be divided into two parts : Kriging models development and enhancementof an existing optimization method to handle high-dimensional problems under a large number ofconstraints. Concerning Kriging models, we propose a new formulation of covariance kernel which is able toreduce the number of hyper-parameters in order to accelerate the construction of the metamodel.One of the known limitations of Kriging models is about the estimation of its hyper-parameters.This estimation becomes more and more difficult when the number of dimension increases. Inparticular, the initial design of experiments (for surrogate modelling construction) requires animportant number of points and therefore the inversion of the covariance matrix becomes timeconsuming. Our approach consists in reducing the number of parameters to estimate using the Partial LeastSquares regression method (PLS). This method provides information about the linear relationshipbetween input and output variables. This information is integrated into the Kriging model kernelwhile maintaining the symmetry and the positivity properties of the kernels. Thanks to this approach,the construction of these new models called KPLS is very fast because of the low number of newparameters to estimate. When the covariance kernel used is of an exponential type, the KPLS methodcan be used to initialize parameters of classical Kriging models, to accelerate the convergence of theestimation of parameters. The final method, called KPLS+K, allows to improve the accuracy of themodel for multimodal functions. The second main contribution of this PhD is to develop a global optimization method to tacklehigh-dimensional problems under a large number of constraint functions thanks to KPLS or KPLS+Kmethod. Indeed, we extended the self adaptive optimization method called "Efficient Global Optimization,EGO" for high-dimensional problems under constraints. Several enriching criteria have been tested. This method allows to estimate known global optima on academic problems up to 50 inputvariables. The proposed method is tested on two industrial cases, the first one, "MOPTA", from the automotiveindustry (with 124 input variables and 68 constraint functions) and the second one is a turbineblade from Snecma company (with 50 input variables and 31 constraint functions). The results showthe effectiveness of the method to handle industrial problems.We also highlight some importantlimitations.

Made available in DSpace on 2017-05-12T14:36:24Z (GMT). No. of bitstreams: 1 DISSERTAcaO_ Final.pdf: 7044304 bytes, checksum: 2c043ef6d78e2a532d7a87d8a9b74434 (MD5) Previous issue date: 2015-12-04 / The computer-aided drug design field have attracted attention with regard to the discovery of new antiretroviral for use in AIDS pharmacotherapy. Thus, the aim of this study was to develop mathematical models that can predict the anti HIV1-IN activity of analogues not synthesized from pyrimidones, pyrimidines and pyridopyrazine carboxamides, reducing the time to obtain new prototypes of drugs. Therefore, it was selected from the literature, a set of 199 compounds described as strand transfer reaction inhibitors and developed 2D and 3D QSAR studies. For the 2D QSAR was used the Ordered Predictors Selection (OPS) methodology in order to develop the variable selection and Fractional Factorial Design (FFD) and OPS for the 3D QSAR. The models in both studies were constructed using partial least squares (PLS). Regarding the 3D QSAR studies, were employed two different methodologies for the calculation of the molecular interaction fields (MIFs): GRID and GRIND (GRid-INdependent descriptors). In addition, they were also used descriptors of pharmacokinetic properties from the Volsurf+ software (extracted from MIFs generated with GRID) for building models in combination with 2D descriptors. The statistical models with good quality were used to predict the activities of a second data set (prediction set, N=145). The resultants predictions were satisfactory. The Williams and Euclidean applicability domains for the set of samples reveal that the predictions did not occur by extrapolation. Moreover, the descriptor ET (total energy of the molecule), selected by the QSAR 2D model is consistent with earlier findings. Further, for the 3D study, the most relevant characteristics (N1-N1 interactions and DRY-O) are coherent with the pharmacophore of the class under study. / O campo do planejamento de fármacos auxiliado por computador tem atraído atenção no que se refere à descoberta de novos antirretrovirais para uso na farmacoterapia da AIDS. Com isso, o objetivo deste trabalho foi desenvolver modelos matemáticos que possam prever a atividade anti HIV1-IN de análogos ainda não sintetizados de pirimidonas, pirimidinas e piridopirazinas carboxamidas, agilizando assim a obtenção de novos compostos protótipos de fármacos. Desta maneira, foi selecionado da literatura um conjunto de 199 compostos descritos como inibidores da reação de transferência de fita e desenvolvido estudos de QSAR 2D e 3D. Para o primeiro utilizou-se a metodologia de seleção de variáveis Ordered Predictors Selection (OPS) e para o segundo Planejamento Fatorial Fracionário (Fractional Factorial Design, FFD) e OPS. Os modelos em ambos os estudos foram construídos através da regrressão por quadrados mínimos parciais (PLS). Em relação aos estudos de QSAR 3D, foram empregadas duas diferentes metodologias para o cálculo dos campos de interação molecular (Molecular Interaction Fields, MIFs): GRID e GRIND (GRId-INdependent descriptors). Ademais, foram também utilizados descritores de propriedades farmacocinéticas provenientes do software Volsurf+ (extraídos de MIFs gerados com GRID) para construção de modelos em associação com descritores 2D. Os modelos com boa qualidade estatística, foram empregados para predição das atividades de um segundo conjunto de dados (conjunto de predição, N=145). Os resultados para as previsões foram satisfatórios. Os domínios de aplicabilidade de Willians e Euclidiano em relação as amostras do conjunto de predição revelam que as predições não ocorreram por extrapolação. Além disto, o descritor ET (energia total da molécula), selecionado para o modelo QSAR 2D é consistente com achados bibliográficos prévios. Em relação ao estudo 3D, as características mais relevantes (interações N1-N1 e DRY-O) se mostram coerentes com o farmacóforo da respectiva classe sob estudo.

Made available in DSpace on 2017-07-10T13:59:27Z (GMT). No. of bitstreams: 1 DISSERTAcaO_ Final.pdf: 7044304 bytes, checksum: 2c043ef6d78e2a532d7a87d8a9b74434 (MD5) Previous issue date: 2015-12-04 / Sim (não especificado) / The computer-aided drug design field have attracted attention with regard to the discovery of new antiretroviral for use in AIDS pharmacotherapy. Thus, the aim of this study was to develop mathematical models that can predict the anti HIV1-IN activity of analogues not synthesized from pyrimidones, pyrimidines and pyridopyrazine carboxamides, reducing the time to obtain new prototypes of drugs. Therefore, it was selected from the literature, a set of 199 compounds described as strand transfer reaction inhibitors and developed 2D and 3D QSAR studies. For the 2D QSAR was used the Ordered Predictors Selection (OPS) methodology in order to develop the variable selection and Fractional Factorial Design (FFD) and OPS for the 3D QSAR. The models in both studies were constructed using partial least squares (PLS). Regarding the 3D QSAR studies, were employed two different methodologies for the calculation of the molecular interaction fields (MIFs): GRID and GRIND (GRid-INdependent descriptors). In addition, they were also used descriptors of pharmacokinetic properties from the Volsurf+ software (extracted from MIFs generated with GRID) for building models in combination with 2D descriptors. The statistical models with good quality were used to predict the activities of a second data set (prediction set, N=145). The resultants predictions were satisfactory. The Williams and Euclidean applicability domains for the set of samples reveal that the predictions did not occur by extrapolation. Moreover, the descriptor ET (total energy of the molecule), selected by the QSAR 2D model is consistent with earlier findings. Further, for the 3D study, the most relevant characteristics (N1-N1 interactions and DRY-O) are coherent with the pharmacophore of the class under study. / O campo do planejamento de fármacos auxiliado por computador tem atraído atenção no que se refere à descoberta de novos antirretrovirais para uso na farmacoterapia da AIDS. Com isso, o objetivo deste trabalho foi desenvolver modelos matemáticos que possam prever a atividade anti HIV1-IN de análogos ainda não sintetizados de pirimidonas, pirimidinas e piridopirazinas carboxamidas, agilizando assim a obtenção de novos compostos protótipos de fármacos. Desta maneira, foi selecionado da literatura um conjunto de 199 compostos descritos como inibidores da reação de transferência de fita e desenvolvido estudos de QSAR 2D e 3D. Para o primeiro utilizou-se a metodologia de seleção de variáveis Ordered Predictors Selection (OPS) e para o segundo Planejamento Fatorial Fracionário (Fractional Factorial Design, FFD) e OPS. Os modelos em ambos os estudos foram construídos através da regrressão por quadrados mínimos parciais (PLS). Em relação aos estudos de QSAR 3D, foram empregadas duas diferentes metodologias para o cálculo dos campos de interação molecular (Molecular Interaction Fields, MIFs): GRID e GRIND (GRId-INdependent descriptors). Ademais, foram também utilizados descritores de propriedades farmacocinéticas provenientes do software Volsurf+ (extraídos de MIFs gerados com GRID) para construção de modelos em associação com descritores 2D. Os modelos com boa qualidade estatística, foram empregados para predição das atividades de um segundo conjunto de dados (conjunto de predição, N=145). Os resultados para as previsões foram satisfatórios. Os domínios de aplicabilidade de Willians e Euclidiano em relação as amostras do conjunto de predição revelam que as predições não ocorreram por extrapolação. Além disto, o descritor ET (energia total da molécula), selecionado para o modelo QSAR 2D é consistente com achados bibliográficos prévios. Em relação ao estudo 3D, as características mais relevantes (interações N1-N1 e DRY-O) se mostram coerentes com o farmacóforo da respectiva classe sob estudo.

A cefalexina é um dos antibióticos mais prescritos no Brasil e a sua disponibilidade nas instituições de saúde é de extrema importância para o tratamento adequado e seguro dos pacientes. Porém a distribuição dos medicamentos é prejudicada pela dificuldade de liberação pelos laboratórios de controle de qualidade que utilizam métodos convencionais como: o biológico (Difusão em Ágar) e o cromatográfico (Cromatografia Líquida de Alta Eficiência), que demandam tempo de análise e não permitem a obtenção de resultados em tempo real. O maior desafio dos centros produtivos é desenvolver métodos capazes de realizarem as análises dos produtos de forma rápida e eficaz durante as fases mais críticas da produção. As técnicas espectroscópicas por infravermelho são alternativas aos métodos convencionais, tendo como vantagens a rapidez, a capacidade de análise de amostras nas suas formas físico-químicas originais e a possibilidade de projetar a sua utilização na Tecnologia Analítica de Processos (PAT). Com esse propósito foi desenvolvido um método por espectroscopia no infravermelho médio (MID) por reflectância difusa (DRIFTS) combinado à técnica de regressão múltipla por mínimos quadrados parciais (PLS) para análise de cefalexina na forma pó para suspensão oral. A avaliação da eficácia do método desenvolvido foi feita mediante cálculo da precisão e exatidão do método e posteriormente os resultados experimentais foram submetidos à análise estatística. A verificação da equivalência dos métodos foi feita utilizando análise de variância (ANOVA) e, para finalizar o trabalho, foi determinado o intervalo de tolerância (δ) do método. O método espectroscópico apresentou significativa variabilidade de resultados, porém uma técnica padronizada de preparo das amostras de calibração capaz de produzir misturas homogêneas do fármaco com placebo, aliado à seleção criteriosa das regiões espectrais e tratamento espectral adequado, foi determinante para o desenvolvimento de um método equivalente aos convencionais, atendendo aos requisitos da PAT. Portanto, este trabalho pode servir de ferramenta para o avanço da PAT pelos produtores e agência regulatória no Brasil (ANVISA). / Cephalexin is the most prescribed antibiotics in Brazil and its availability in health institutions is of utmost importance for the proper and safe treatment of patients. But the distribution of drugs is hindered to difficulty of release by quality control laboratories using conventional methods such as biological (Agar Diffusion) and HPLC (High Performance Liquid Chromatography), which require analysis time and not allow to obtain results in real time. The biggest challenge of manufacturing centers is to develop analytical methods to perform quickly and effectively analyzes of the products during the most critical phases of production. The infrared spectroscopic technique is alternative to conventional methods, with the advantages of speed, the ability to analyze samples in their original physical-chemical forms and the possibility to use in designing the Process Analytical Technology (PAT). For this purpose, a method was developed in the mid-infrared spectroscopy (MID) by diffuse reflectance (DRIFTS) combined to technique of partial least squares (PLS) regression for cephalexin analysis in oral suspension product. Assessing the effectiveness of the method was done by calculation of precision and of the method and then the experimental results were subjected to statistical analysis. The equivalence of methods was verified using analysis of variance (ANOVA) and, to complete the work, the tolerance interval (δ) was determined. The spectroscopic method showed significant variability in results, but a standardized technique for preparation of calibration samples able to produce homogeneous mixtures of the drug with placebo, combined to careful selection of spectral regions and appropriate spectral treatment, was crucial for the development of an equivalent method to conventional, meeting the requirements of PAT. Therefore, this work can serve as a tool for advancing the PAT by manufacturers and regulatory agency in Brazil (ANVISA).

Der statistische Diskussionbeitrag untersucht, ob und wie sich Erwartungen und Stimmungen in der Wirtschaft bilden bzw. von welchen volkswirtschaftlichen Größen sie abhängen. Als Methodik werden Partial Least Squares (PLS) Modelle genutzt, eine Modellklasse der Pfadanalyse mit latenten Variablen. Die verwendeten Daten wurden vom Ifo-Institut und aus der amtlichen Statistik entnommen. / This paper analyses the development of sentiments and expectations in the German economy. The issue is how these expectatons are influenced by major macroeconomic variables like investments or unemployment. Several Partial Least Squares models (PLS) are used to estimate the relations. The data is derived from the German Ifo Institut and the official statistic.

Diese Forschungsarbeit widmet sich der Analyse von Stimmungen und Erwartungen im System der Märkte mit Dynamic Partial Least Squares (DPLS) Modellen. Die Analyse komplexer Systeme mit umfangreichen Datensätzen und die Erkennung relevanter Muster erfordern die Verwendung moderner statistischer Verfahren. DPLS-Modelle, eine Variante der Strukturgleichungs-modelle mit Latenten Variablen, werden methodisch erweitert, um mehrere zeitliche Verzögerungsstufen gleichzeitig modellieren zu können. Die ökonometrischen Modelle versuchen, zahlreiche latente Einflussfaktoren und ihre verdeckten Beziehungen zu identifizieren. Als Daten werden rund 80 Indikatoren verwendet von Januar 1991 bis Juni 2010, um Stimmungen, Erwartungen und wirtschaftlich relevanten Größen zu operationalisieren und die Zusammenhänge detailliert zu untersuchen. Die Modellergebnisse zeigen, dass Stimmungen, also die Einschätzung der aktuellen wirtschaftlichen Lage, deutlich mit wirtschaftlichen Größen zusammenhängen, unter anderem mit Investitionen, Auftragseingängen oder Aktienmarktentwicklungen. Die Erwartungen, also die Einschätzung der zukünftigen Entwicklung, bieten eine mittlere bis schwache Prognosekraft für sechs bis maximal achtzehn Monate in die Zukunft. Für kürzere Zeiträume von sechs Monaten sind Stimmungen und Erwartungen die besten verfügbaren Prognosevariablen. Die Analyse der Modellabweichungen erlaubt Rückschlüsse auf wirtschaftstheoretische Konzepte, wie Rationalität der Erwartungen. Auffällig sind scheinbare Phasen der systematischen Über- und Unterbewertungen der aktuellen Situation oder zukünftigen Entwicklung, insbesondere vor und nach Krisensituationen. Die Ergebnisse dieser Arbeit geben somit einen erweiterten Einblick in die empirischen Zusammenhänge subjektiver Einschätzungen mit realen wirtschaftlichen Größen. / This research thesis is devoted to the analysis of sentiments and expectations in the system of economic markets with Dynamic Partial Least Squares (DPLS) models with time-series. The analysis of complex systems with large data sets and discovering relevant patterns require the use of modern statistical methods. The DPLS models, a variant of structural equation models with latent variables, are methodically extended to include multiple lags simultaneously. The econometric models try to identify a large number of latent factors and their unobservable relations. The data consists of about 80 indicators from January 1991 to June 2010 to quantify sentiments, expectations and economically relevant variables and examine the relations in detail. The results show, that the sentiments, the assessment of the current economic situation, are clearly linked with economic variables like investments, incoming orders and stock market developments. The expectations, the assessment of future development, provide a weak to moderate predictive power for up to 18 months in the future. For shorter periods of six months sentiments and expectations are the best available forecast variables. The analysis of model deviations allows to draw conclusions about theoretical concepts such as rationality of expectations. There seem to be phases of systematic over- and undervaluation of the current situation and future development, especially before and after economic crisis. The results should provide a deeper insight into the empirical relationships between subjective assessments and real economic variables.

Le développement de la maturité supply chain est une préoccupation croissante des entreprises évoluant dans un environnement qui se transforme profondément et simultanément dans plusieurs domaines. S’adossant à plusieurs initiatives proposées sous forme de référentiels, les entreprises progresser quelque peu avec difficulté. S’inscrivant dans un paradigme post-positiviste, cette recherche se concentre sur la conception d’un modèle qui permet d’évaluer la maturité, sur plusieurs dimensions, de la supply chain. Ces objectifs ont été atteints à partir d’une revue approfondie de la littérature, mais également à partir de la réalisation d’une étude empirique. D’une part, les concepts de supply chain, de supply chain management, de maturité supply chain, ainsi que des capacités relatives ont été revus en profondeur, et d’autre part, les référentiels d’évaluation dans le domaine de la logistique, de la gestion des opérations et du supply chain management ont été étudiés et confrontés aux capacités de maturité supply chain établies. Cela a donné lieu à la conception d’un modèle qui s’appuie sur les modèles antérieurs pour proposer des critères qui les complètent. Ce modèle a été validé auprès d’un échantillon de 115 répondants (Directeurs logistique, supply chain managers, etc.). L’analyse des données collectées s’est appuyée sur les modèles d’équations structurelles, en l’occurrence PLS-PM (régressions aux moindres carrées partiels). Cette analyse, précédée d’une analyse univariée et d’une analyse en composantes principales a permis de constater la validité convergente ainsi que la validité discriminante des échelles de mesures, mais également de l’existence de liens d’impacts variables entre les différentes dimensions et la maturité supply chain. Les résultats de cette recherche montrent que la maturité est principalement constituée par les dimensions opérationnelles et stratégiques, suivies par les dimensions informationnelles, structurelles, organisationnelles et humaines. La thèse établit le constat selon lequel la dimension de risques et de résilience, la dimension relationnelle et la dimension de développement durable et de responsabilité sociale ont le moins d’impact dans la structuration de la maturité supply chain. Une fois le modèle validé, la maturité effective des entreprises de l’échantillon a été évaluée en utilisant l’analyse Importance/Maturité. Les résultats émanant de cette analyse montrent une forte maturité supply chain des entreprises par rapport aux dimensions stratégiques, structurelles, organisationnelles, de risques et de résilience, ce qui n’est pas le cas par rapport aux autres dimensions, où l’analyse a décelé une maturité moyenne à faible.Cela suggère aux entreprises de concentrer leurs efforts sur les aspects opérationnels, relationnels, informationnels et relatifs au développement durable et à la responsabilité sociale. / Supply chain maturity development is a growing concern for companies operating in an environment that is changing profoundly and simultaneously in several areas. Teaming up with several initiatives proposed as standards, companies make some progress with difficulty. As part of a post-positivist paradigm, this research focuses on the design of a model, which assesses the maturity of several dimensions of the supply chain. These objectives have been achieved from a thorough review of the literature, but also from the realization of an empirical study. On the one hand, the concepts of supply chain, supply chain management, supply chain maturity, as well as related capacity were reviewed in depth.On the other hand, assessment reference models in logistics, operations management and supply chain management have been studied and compared to capabilities of supply chain maturity established. This led to the design of a model based on earlier ones to propose criteria that complement them. This model was validated with a sample of 115 respondents (Directors logistics, supply chain managers, etc.). The analysis of the data collected was based on structural equation models, namely PLS-PM (Partial Least Squares -Path Modeling). This analysis was preceded by univariate analysis and principal component analysis, showed convergent validity and discriminant validity of measurement scales, but also the existence of links with variable impact between dimensions and supply chain maturity. Results of this research show that maturity is mainly constituted by the operational and strategic dimensions, followed by informational, structural, organizational and human resourcesdimensions. The thesis establishes that the maturity level of risk and resilience, relational and sustainable development and social responsibility dimensions have the least impact on the structuring of the supply chain maturity. Once validated, the effective maturity of the firms in the sample was evaluated using the Importance/Maturity analysis. Results from this analysis show a strong supply chain maturity of the company in relation to strategic dimensions, structural, organizational, risk and resilience, which is not the case with respect to other dimensions, where the analysis has identified low average maturity. This suggests that companies focus their efforts on the operational, relational, informational aspects and those related to sustainable development and social responsibility.

Les systèmes ERP ont été largement étudiés au cours des dernières décennies, mais ils échouent souvent à offrir les avantages prévus initialement attendus. L'une des raisons est le manque de compréhension comment la personnalisation influence l’acceptation de l’utilisateur ERP par manque d'ajustement du système-à-business, ce qui peut conduire à des résultats négatifs de business. Pour certaines raisons, beaucoup ont fait valoir qu'une mise en oeuvre de ‘vanille’, à savoir sans personnalisation, est la «meilleure» façon de mettre en œuvre des systèmes ERP. Cependant, grâce à la recherche quantitative sur la base de sondage sur le Web, cette thèse a révélé que, en Chine, la personnalisation est une condition nécessaire dans l'acceptation de l'ERP et la réussite du projet. Étant donné d’une variété de risques dans le projet ERP, financier, technique, fonctionnel et politique, les fournisseurs et les consultants sont prêts à aider le chef de projet pour respecter le budget et le temps cible plutôt que de récolter plus d'avantages pour la performance des entreprises. Ainsi, la personnalisation habituellement avait été évitée, et la personnalisation insuffisante est plus fréquente que la personnalisation excessive. Nous avons proposé 15 hypothèses et 11 ont été soutenus, et la conclusion générale que, le niveau de personnalisation a une influence positive considérable sur l'intention comportementale, plus le niveau de personnalisation est élevé, plus l'intention comportementale (BI) à utiliser le système est élevée. C’est une instruction pour les entreprises dans lesquelles le bon niveau de personnalisation (CL) devrait être réalisé au lieu d'éviter la personnalisation. Et seulement la personnalisation importante stratégique plus élevée devrait être inclue dans le projet quand il y a la limitation dans le budget et le temps. Nous avons également confirmé que la facilité de personnalisation est un facteur important dans le choix de la solution ERP correcte. En tant que modérateur, différents rôles ont différentes perceptions sur la personnalisation, les utilisateurs normaux ont une espérance plus forte dans la personnalisation, il est donc essentiel d'expliquer aux utilisateurs, quelle personnalisation est stratégique, et quelle personnalisation est à des fins de cohérence et pourrait compromettre le bénéfice réel de l'ERP système. S'il ne convient pas de faire la personnalisation dans la phase de mise en oeuvre du projet, un plan pour la personnalisation future du système doit être préparé et il améliorera finalement le succès du système à long terme. / ERP systems have been widely studied during the past decades, yet they often fail to deliver the intended benefits originally expected. One notable reason is the lack of understanding how the customization influence the ERP user acceptance when there is lack of system-to-business fit, which can lead to negative business outcomes. For some reasons, many have argued that a "vanilla" implementation, i.e. without customization, is the "best" way to implement ERP systems. However, through quantitative research based on web survey, this dissertation revealed that, in China, customization is a must in ERP acceptance and project success. Because of the various risks in ERP project, financial, technical, functional and political, vendors and consultants are keen on helping the project manager to meet the budget and time target rather than to reap more benefit for business performance. Thus, customization usually had been avoided, and insufficient customization are more common than over customization. We proposed 15 hypothesis and 11 were supported, and the general conclusion that, customization level has significant positive influence on behavioral intention, the higher customization done, the higher behavioral intention (BI) to use the system. It is an instruction for companies that, right level of customization (CL) should be achieved instead of avoiding customization. And only higher strategic important customization should be included in the project when there is limitation in budget and time. We also confirmed that ease of customization is an important factor in selecting the right ERP solution. As a moderator, different role have different perception on customization, normal users have stronger expectancy in customization, so it is critical to explain to the users, which customization is strategic, and which one is for consistency purpose and could jeopardize the real benefit of ERP system. If it is not appropriate to do the customization in the project implementation phase, a plan for future system customization should be prepared and it will ultimately improve the system long term success.

La complexification des intégrations sur les puces électroniques et la course à la miniaturisation sont les deux moteurs actuels de la microélectronique. Les limites optiques de la lithographie sont déjà atteintes depuis longtemps. Ainsi, la fabrication doit aussi être contrôlée de plus en plus étroitement afin d’éviter des variabilités qui nuiraient au bon fonctionnement du produit. Cette thèse présente une approche holistique du contrôle d’un des paramètres les plus importants de la photolithographie : le focus. Celui-ci est directement lié à la qualité de l’image transférée dans la résine photosensible pendant l’exposition. Son contrôle est donc primordial. Les sources de variabilités du focus sur le wafer sont multiples et diverses mais le cas particulier de la topographie du substrat a été privilégié dans cette étude. L’approche holistique de cet effet en particulier a conduit à l’utilisation d’outils de « data mining » telle la régression par la méthode des moindres carrés partiels qui a permis de pointer les principales causes de cette topographie, de créer un modèle prédictif de la topologie mais aussi d’évaluer des solutions d’améliorations comme l’amélioration des corrections qu’effectue le scanner permettant un meilleur contrôle généralisé de toutes les technologies sans toutefois changer l’intégration et le design ou encore la mise en place d’une méthode qui permet d’évaluer les erreurs de focus sur le wafer sans pour autant avoir recours à des mesures intensives sur silicium. D’autres solutions permettent de corriger les facteurs de risques à la source en modifiant le design afin de limiter la formation de la topologie de surface / The increasing complexity in chip integration (co-integration, increasing diversity of matérials…) and the race to dimension shrinkage are the two main drivers of research in microelectronics today. The optical limitations of lithography have been reached some years ago so that double patterning is now a typical process flow in production and helps reducing pattern size and increasing design density. Because of these, the manufacturing itself needs to be more tightly controlled in order to avoid marginalities. Which will affect the chip operation. The cross-effects between these elements are more numerous and their ratio in the total budget is larger whereas the needs for tighter process control are rising. This thesis presents a holistic approach of the control of one of the main parameters for photolithography: focus. It is directly linked to the quality of the image transferred into the photoresist during exposure. Its control is then essential. Variability sources for focus are manifold and diverse: laser, mask, optical column, servo-controllers, wafer flatness, integration, design, substrate reflectivity, material quality etc. All these are added to each other, leading to the creation of defects which can be catastrophic such as shorts. The first objective of this work was to show current challenges raised by STMicroelectronics new technologies, specifically photolithography-wise and focus-wise. A budget breakdown of two critical processes (Metal line patterning in 28nm FD-SOI and Contact patterning for 14nm FD-SOI) has been established which gives the impact of every effect. The product layout effects were evaluated to represent up to 20% of the complete budget and 50% of its intra-chip component. Topography contributes to a large part of these effects and offline measurements showed up to 32nm 3s of height variation in a single field. This may lead to local defocuses of the same order of magnitude. The usable depth of field being about 60 to 70nm for the studied layers, it is clear that focus control is really tight here. The holistic approach of topology leaded to the use of data mining tooling as PLS regression (Partial least Square). It allowed the highlighting of main causes of topography, the creation of a predictive model of topology and the evaluation of several improvement solutions. One may distinguish “palliative” and “curative” solutions. In the first category, on may put scanner levelling improvements which might be effective for every technology without any modification to make on integration and design. The emulated wafer map methodology providing on-product focus non-uniformities without any measurements is also a solution for investigation. “Curative” solutions may concern the mitigation of risk factors by modifying the design topography built-up main factors