Mémoire numérisé par la Division de la gestion de documents et des archives de l'Université de Montréal.

<p>This thesis concerns optimization of structures considering various uncertainties. The overall objective is to find methods to create solutions that are optimal both in the sense of handling the typical load case and minimising the variability of the response, i.e. robust optimal designs.</p><p>Traditionally optimized structures may show a tendency of being sensitive to small perturbations in the design or loading conditions, which of course are inevitable. To create robust designs, it is necessary to account for all conceivable variations (or at least the influencing ones) in the design process.</p><p>The thesis is divided in two parts. The first part serves as a theoretical background to the second part, the two appended articles. This first part includes the concept of robust design, basic statistics, optimization theory and meta modelling.</p><p>The first appended paper is an application of existing methods on a large industrial example problem. A sensitivity analysis is performed on a Scania truck cab subjected to impact loading in order to identify the most influencing variables on the crash responses.</p><p>The second paper presents a new method that may be used in robust optimizations, that is, optimizations that account for variations and uncertainties. The method is demonstrated on both an analytical example and a Finite Element example of an aluminium extrusion subjected to axial crushing.</p> / ROBDES

<p>Analyzing software contains two different tasks. First of all we are analyzing the software and try to calculate some metrics for software quality. Then those results have to be presented to the software engineers. VizzAnalyzer is a tool for analysis and visualization of software. It visualization allow not for a standardized diagram representation. Therefore it is difficult for others to understand, and we need to explain the meaning of our non-standard diagram elements. The solution is to use a standardized representation which can be understood by both sides. UML is such a collection of intuitively diagrams with standardized elements. Their meaning is clear to most software engineers.</p><p>We extended our analysis tool, the VizzAnalyzer, allowing it to view software systems as UML Class diagrams. We reused the existing plug-in architecture to connect our analysis tool with yEd, a graph visualization program. This plug-in is responsible for exchanging the data between the two applications.</p><p>We solve this conversion defining an UML Class Diagram Model and the mapping function between this model and the Common Meta-Model used by VizzAnalyzer and our Class Diagram Model. After that, we export this Class Diagram Model to a format suitable for yEd to display.</p><p>Now we can generate Class Diagrams with the VizzAnalyzer tool. This will allow a better communication of the results derived by different analysis with the software engineers.</p><p>This thesis describes the evolution of different alternatives and the design and implementation of our solution.</p>

<p>The VizzAnalyzer is a powerful software analysis tool. It is able to extract information from various software representations like source code  but also other specifications like UML. The extracted information is input to static analysis of these software projects. One programming language the VizzAnalyzer can extract information from is Java source code.</p><p>Analyzing the source code is sufficient for most of the analysis. But, sometimes it is necessary to analyze compiled classes either because the program is only available in byte-code, or the scope of analysis includes libraries that exist usually in binary form. Thus, being able to extract information from Java byte-code is paramount for the extension of some analyses, e.g., studying the dependecy structure of a project and the libraries it uses.</p><p>Currently, the VizzAnalyzer does not feature information extraction from Java byte-code. To allow, e.g., the analysis of the project dependency structure, we extend the VizzAnalyzer tool with a bytecode front-end that will allow the extraction of information from Java bytecode.</p><p>This thesis describes the design and implementation of the bytecode front-end. After we implemented and integrated the new front-end with the VizzAnalyzer, we are now able to perform new analyses that work on data extracted from both, source- and bytecode.</p>

Mémoire numérisé par la Division de la gestion de documents et des archives de l'Université de Montréal

The use of models for designing and developing Information Systems (IS) has changed in recent years. Models are no longer considered a peripheral documentation medium that is poorly maintained and often neglected. Rather, models are increasingly seen as essential parts of the final product—as central artifacts that drive and guide the development efforts. The knowledge that modelers rely on when designing models is represented as formal models and clearly defined rules for transforming the models. The flexibility, reliability, and effectiveness offered by the formal models and the transformations are making Model Driven Development (MDD) a popular choice for building IS. Models also serve in describing enterprise design, where enterprise-level models capture organizational knowledge and aid in understanding, improving, and growing the enterprise. Enterprise Modeling (EM) offers a structured and unified view of the enterprise, thereby enabling more informed and accurate decisions to be made. Many MDD approaches have been proposed to tackle a wide range of IS-related issues, but little attention is being paid to the source of the knowledge captured by the IS models. EM approaches capture organizational knowledge and provide the necessary input and underlying context for designing IS. However, the results produced by EM approaches need to be manually analyzed by modelers to create the initial MDD model. This interruption of the MDD process represents a gap between enterprise models and MDD models. Limited research has been done to connect EM to MDD in a systematic and structured manner based on the principles of model-driven development. This thesis proposes a unifying meta-model for integrating EM and MDD. The meta-model captures the inherent links that exist between organizational knowledge and IS design. This helps to improve the alignment between organizational goals and the IS that are created to support them. The research presented herein follows the guidelines of the design science research methodology. It starts with a state-of-the-art survey of the current relationship between MDD and prior stages of development. The findings of the survey are used to elicit a set of necessary properties for integrating EM and MDD. The unifying meta-model is then proposed as the basis for an integrated IS development approach that applies the principles of MDD and starts on the enterprise level by considering enterprise models in the development process. The design of the meta-model supports the elicited integration properties. The unifying meta-model is based on the Enterprise Knowledge Development (EKD) approach to EM. A prototype tool is developed to support the unifying meta-model, following a study to choose a suitable implementation environment. The use of the unifying meta-model is demonstrated through the implemented tool platform using an example case study, revealing its advantages and highlighting the potential for improvement and future development.

La recherche de la performance énergétique dans les bâtiments est devenue un objectif sociétal et réglementaire. De nos jours, un bâtiment doit assurer des conditions de confort, tout en limitant son impact énergétique et environnemental. Les demandes de performances sont telles qu’il n’est plus possible d’appliquer des solutions standards à l’ensemble du bâtiment. Un critère énergétique doit donc être pris en compte au plus tôt dans la conception, ce qui exige dès l’esquisse un travail de plus en plus lourd et complexe. Pour faciliter la prise en compte du critère énergétique, nous avons choisi une approche d’optimisation valorisant la mise en œuvre de la simulation énergétique du bâtiment (quelque soit la méthode de simulation choisie) afin de fournir un outil d’aide à la décision au concepteur à partir du métré et de l’esquisse. La thèse consiste au développement d’une méthodologie commençant par le développement d’un algorithme d’optimisation valorisant la construction séquentielle d’un Kriging sur des paramètres globaux du bâtiment qui permet de rendre le temps de calcul nécessaire compatible avec le contexte de conception d’un bâtiment. Le méta-modèle de Kriging construit peut ensuite être valorisé pour restituer les solutions optimales, permettre l’exploration de l’espace de solutions respectant les contraintes, et mettre en valeur la sensibilité énergétique du bâtiment face aux différents paramètres d’entrée. Le concepteur peut alors faire des choix constructifs optimisés, tout en conservant un certain espace de liberté, ou encore réaliser des modifications à postériori. La dernière étape consiste à inverser la relation liant les paramètres globaux choisis à l’ensemble des données élémentaires de l’ouvrage. L’ensemble de la démarche est appliqué au cas d’une cellule monozone de référence. / Contribution to the use of surrogate models for constrained optimization of energy criteria: application to building design

Analyzing software contains two different tasks. First of all we are analyzing the software and try to calculate some metrics for software quality. Then those results have to be presented to the software engineers. VizzAnalyzer is a tool for analysis and visualization of software. It visualization allow not for a standardized diagram representation. Therefore it is difficult for others to understand, and we need to explain the meaning of our non-standard diagram elements. The solution is to use a standardized representation which can be understood by both sides. UML is such a collection of intuitively diagrams with standardized elements. Their meaning is clear to most software engineers. We extended our analysis tool, the VizzAnalyzer, allowing it to view software systems as UML Class diagrams. We reused the existing plug-in architecture to connect our analysis tool with yEd, a graph visualization program. This plug-in is responsible for exchanging the data between the two applications. We solve this conversion defining an UML Class Diagram Model and the mapping function between this model and the Common Meta-Model used by VizzAnalyzer and our Class Diagram Model. After that, we export this Class Diagram Model to a format suitable for yEd to display. Now we can generate Class Diagrams with the VizzAnalyzer tool. This will allow a better communication of the results derived by different analysis with the software engineers. This thesis describes the evolution of different alternatives and the design and implementation of our solution.

The VizzAnalyzer is a powerful software analysis tool. It is able to extract information from various software representations like source code  but also other specifications like UML. The extracted information is input to static analysis of these software projects. One programming language the VizzAnalyzer can extract information from is Java source code. Analyzing the source code is sufficient for most of the analysis. But, sometimes it is necessary to analyze compiled classes either because the program is only available in byte-code, or the scope of analysis includes libraries that exist usually in binary form. Thus, being able to extract information from Java byte-code is paramount for the extension of some analyses, e.g., studying the dependecy structure of a project and the libraries it uses. Currently, the VizzAnalyzer does not feature information extraction from Java byte-code. To allow, e.g., the analysis of the project dependency structure, we extend the VizzAnalyzer tool with a bytecode front-end that will allow the extraction of information from Java bytecode. This thesis describes the design and implementation of the bytecode front-end. After we implemented and integrated the new front-end with the VizzAnalyzer, we are now able to perform new analyses that work on data extracted from both, source- and bytecode.

Les langages de spécifications basés et orientés objets (comme UML/OCL, JML, Spec#, ou Eiffel) permettent la création et destruction, la conversion et tests de types dynamiques d'objets statiquement typés. Par dessus, les invariants de classes et les opérations de contrat peuvent y être exprimés; ces derniers représentent les éléments clés des spécifications orientées objets. Une sémantique formelle des structures de données orientées objets est complexe : des descriptions imprécises mènent souvent à différentes interprétations dans les outils qui en résultent. Dans cette thèse, nous démontrons comment dériver un environnement de preuves moderne comme un méta-outil pour la définition et l'analyse de sémantique formelle de langages de spécifications orientés objets. Étant donné une représentation d'un langage particulier plongé en Isabelle/HOL, nous construisons pour ce langage un environnement étendu d'Isabelle, à travers une méthode de génération de code particulière, qui implique notamment plusieurs variantes de génération de code. Le résultat supporte l'édition asynchrone, la vérification de types, et les activités de déduction formelle, tous "hérités" d'Isabelle. En application de cette méthode, nous obtenons un outil de modélisation orienté objet pour du UML/OCL textuel. Nous intégrons également des idiomes non nécessairement présent dans UML/OCL --- en d'autres termes, nous développons un support pour des dialectes d'UML/OCL à domaine spécifique. En tant que construction méta, nous définissons un méta-modèle d'une partie d'UML/OCL en HOL, un méta-modèle d'une partie de l'API d'Isabelle en HOL, et une fonction de traduction entre eux en HOL. Le méta-outil va alors exploiter deux procédés de générations de code pour produire soit du code raisonnablement efficace, soit du code raisonnablement lisible. Cela fournit donc deux modes d'animations pour inspecter plus en détail la sémantique d'un langage venant d'être plongé : en chargeant à vitesse réelle sa sémantique, ou simplement en retardant à un autre niveau "méta" l'expérimentation précédente pour un futur instant de typage en Isabelle, que ce soit pour des raisons de performances, de tests ou de prototypages. Remarquons que la génération de "code raisonnablement efficace", et de "code raisonnablement lisible" incluent la génération de code tactiques qui prouvent une collection de théorèmes formant une théorie de types de données orientés objets d'un modèle dénotationnel : étant donné un modèle de classe UML/OCL, les preuves des propriétés pertinentes aux conversions, tests de types, constructeurs et sélecteurs sont traitées automatiquement. Cette fonctionnalité est similaire aux paquets de théories de types de données présents au sein d'autres prouveurs de la famille HOL, à l'exception que certaines motivations ont conduit ce travail présent à programmer des tactiques haut-niveaux en HOL lui-même. Ce travail prend en compte les plus récentes avancées du standard d'UML/OCL 2.5. Par conséquent, tous les types UML/OCL ainsi que les types logiques distinguent deux éléments d'exception différents : invalid (exception) et null (élément non-existant). Cela entraîne des conséquences sur les propriétés aussi bien logiques qu'algébriques des structures orientées objets résultant des modèles de classes. Étant donné que notre construction est réduite à une séquence d'extension conservative de théorie, notre approche peut garantir la correction logique du langage entier considéré, et fournit une méthodologie pour étendre formellement des langages à domaine spécifique. / Object-based and object-oriented specification languages (likeUML/OCL, JML, Spec#, or Eiffel) allow for the creation and destruction, casting and test for dynamic types of statically typed objects. On this basis, class invariants and operation contracts can be expressed; the latter represent the key elements of object-oriented specifications. A formal semantics of object-oriented data structures is complex: imprecise descriptions can often imply different interpretations in resulting tools. In this thesis we demonstrate how to turn a modern proof environment into a meta-tool for definition and analysis of formal semantics of object-oriented specification languages. Given a representation of a particular language embedded in Isabelle/HOL, we build for this language an extended Isabelle environment by using a particular method of code generation, which actually involves several variants of code generation. The result supports the asynchronous editing, type-checking, and formal deduction activities, all "inherited" from Isabelle. Following this method, we obtain an object-oriented modelling tool for textual UML/OCL. We also integrate certain idioms not necessarily present in UML/OCL --- in other words, we develop support for domain-specific dialects of UML/OCL. As a meta construction, we define a meta-model of a part of UML/OCL in HOL, a meta-model of a part of the Isabelle API in HOL, and a translation function between both in HOL. The meta-tool will then exploit two kinds of code generation to produce either fairly efficient code, or fairly readable code. Thus, this provides two animation modes to inspect in more detail the semantics of a language being embedded: by loading at a native speed its semantics, or just delay at another "meta"-level the previous experimentation for another type-checking time in Isabelle, be it for performance, testing or prototyping reasons. Note that generating "fairly efficient code", and "fairly readable code" include the generation of tactic code that proves a collection of theorems forming an object-oriented datatype theory from a denotational model: given a UML/OCL class model, the proof of the relevant properties for casts, type-tests, constructors and selectors are automatically processed. This functionality is similar to the datatype theory packages in other provers of the HOL family, except that some motivations have conducted the present work to program high-level tactics in HOL itself. This work takes into account the most recent developments of the UML/OCL 2.5 standard. Therefore, all UML/OCL types including the logic types distinguish two different exception elements: invalid (exception) and null (non-existing element). This has far-reaching consequences on both the logical and algebraic properties of object-oriented data structures resulting from class models. Since our construction is reduced to a sequence of conservative theory extensions, the approach can guarantee logical soundness for the entire considered language, and provides a methodology to soundly extend domain-specific languages.