Advanced development environments are essential for efficient realization of complex industrial products. Powerful equation-based object-oriented (EOO) languages such as Modelica are successfully used for modeling and virtual prototyping complex physical systems and components. The Modelica language enables engineers to build large, sophisticated and complex models. Modelica environments should scale up and be able to handle these large models. This thesis addresses the scalability of Modelica tools by employing incremental compilation and dynamic loading. The design, implementation and evaluation of this approach is presented. OpenModelica is an open-source Modelica environment developed at PELAB in which we have implemented our strategy for incremental compilation and dynamic loading of functions. We have tested the performance of these strategies in a number of different scenarios in order to see how much of an impact they have on the compilation and execution time. Our solution contains an overhead of one or two hash calls during runtime as it uses dynamic hashes instead of static arrays.

Context. Heat load forecasting is an important part of district heating optimization. In particular, energy companies aim at minimizing peak boiler usage, optimizing combined heat and power generation and planning base production. To achieve resource efficiency, the energy companies need to estimate how much energy is required to satisfy the market demand. Objectives. We suggest an online machine learning algorithm for heat load forecasting. Online algorithms are increasingly used due to their computational efficiency and their ability to handle changes of the predictive target variable over time. We extend the implementation of online bagging to make it compatible to regression problems and we use the Fast Incremental Model Trees with Drift Detection (FIMT-DD) as the base model. Finally, we implement and incorporate to the algorithm a mechanism that handles missing values, measurement errors and outliers. Methods. To conduct our experiments, we use two machine learning software applications, namely Waikato Environment for Knowledge Analysis (WEKA) and Massive Online Analysis (MOA). The predictive ability of the suggested algorithm is evaluated on operational data from a part of the Karlshamn District Heating network. We investigate two approaches for aggregating the data from the nodes of the network. The algorithm is evaluated on 100 runs using the repeated measures experimental design. A paired T-test is run to test the hypothesis that the the choice of approach does not have a significant effect on the predictive error of the algorithm. Results. The presented algorithm forecasts the heat load with a mean absolute percentage error of 4.77\%. This means that there is a sufficiently accurate estimation of the actual values of the heat load, which can enable heat suppliers to plan and manage more effectively the heat production. Conclusions. Experimental results show that the presented algorithm can be a viable alternative to state-of-the-art algorithms that are used for heat load forecasting. In addition to its predictive ability, it is memory-efficient and can process data in real time. Robust heat load forecasting is an important part of increased system efficiency within district heating, and the presented algorithm provides a concrete foundation for operational usage of online machine learning algorithms within the domain.

Nous proposons dans cette thèse une architecture pour le raisonnement incrémental sur des flux de triples. Afin de passer à l’échelle, elle est conçue sous la forme de modules indépendants, permettant l’exécution parallèle du raisonnement. Plusieurs instances d’une même règle peuvent être exécutées simultanément afin d’améliorer les performances. Nous avons également concentré nos efforts pour limiter la dispersion des doublons dans le système, problème récurrent du raisonnement. Pour cela, un triplestore partagé permet à chaque module de filtrer au plus tôt les doublons. La structure de notre architecture, organisée en modules indépendants par lesquels transitent les triples, lui permet de recevoir en entrée des flux de triples. Enfin, notre architecture est indépendante du fragment utilisé. Nous présentons trois modes d’inférence pour notre architecture. Le premier consiste à inférer l’ensemble des connaissances implicites le plus rapidement possible. Le second priorise l'inférence de certaines connaissances prédéterminées. Le troisième vise à maximiser la quantité de triples inférés par seconde. Nous avons implémenté l’architecture présentée à travers Slider, un raisonneur incrémental prenant nativement en charge les fragments ρdf et RDFS. Il peut être facilement étendu à des fragments plus complexes. Nos expérimentations ont montré une amélioration des performances de plus de 65% par rapport au raisonneur OWLIM-SE. Nous avons également mené des tests montrant que l’utilisation du raisonnement incrémental avec Slider apporte un avantage systématique aux performances par rapport au raisonnement par lots, quels que soient l’ontologie utilisée et le fragment appliqué / In this thesis, we propose an architecture for incremental reasoning on triple streams. To ensure scalability, it is composed of independent modules; thus allowing parallel reasoning. That is, several instances of a same rule can be simultaneously executed to enhance performance. We also focused our efforts to limit the duplicates spreading in the system, a recurrent issue for reasoning. To achieve this, we design a shared triplestore which allows each module to filter duplicates as soon as possible. The triples passes through the different independent modules of the architecture allows the reasoner to receive triple streams as input. Finally, our architecture is of agnostic nature regarding the fragment used for the inference. We also present three inference modes for our architecture: the first one infers all the implicit knowledge as fast as possible; the second mode should be used when the priority has to be defined for the inference of a specific type of knowledge; the third one proposes to maximize the amount of triples inferred per second. We implemented this architecture through Slider, an incremental reasoning natively supporting the fragments ρdf and RDFS: It can easily be extended to more complex fragments. Our experimentations show a 65% improvement over the reasoner OWLIM-SE. However, the recently published reasoner RDFox exhibits better performance, although this one does not provide prioritized inference. We also conducted experimentations showing that the use of incremental reasoning over batch-based reasoning offers systematically better performance for all the ontologies and fragments used

This thesis consists of three chapters. The first chapter explores the effects of prenatal nutritional deficiency on depression in adulthood. It is well established that maternal behaviour during pregnancy has a lasting effect on the child for years to come. Studies show that in utero nutritional shocks can have prolonged effects on health and labour market outcomes later in life of the offspring. In this paper I investigate whether such nutritional deficiencies during gestation can have an extended impact on mental health in adulthood. Using the fourth wave of Indonesian Family Life Survey (IFLS), I find that Muslim individuals who were potentially exposed to Ramadan in the first and third trimester have significantly higher scores on the depression scale than those who were not exposed. This effect is particularly significant among Muslim males who were exposed in the first trimester and Muslim females who were exposed in the third trimester. Similar effects of exposure are also found on the probability of being depressed in the Muslim population. The absence of such impact of exposure in the non-Muslim population suggests that nutritional deficiencies during the gestation period can have lasting effects on mental health and may increase the possibility of developing depression later in life. Next, the literature on socioeconomic health inequality uses individuals' socioeconomic rank (p) to develop the concentration index. In the second chapter of the thesis, I construct an alternative framework by directly using individuals' income level (y) to rank them and develop stochastic dominance conditions to investigate whether this method leads to the same conclusion as using the socioeconomic ranks (p). Using World Health Survey data for five South Asian countries, I conclude that using the socioeconomic ranks (p) and income levels (y) to rank individuals lead to different results in dominance tests adjusted for different equivalence scales. Lastly, to address the arbitrariness problem of the health concentration index's value caused by assuming the existence of a ratio-scaled variable, Makdissi and Yazbeck (2014) adopted a counting approach to measure health inequality. In the third chapter of the thesis, I apply this counting approach in a two-fold way. Firstly, I estimate the values of population health status and health inequality in United States using the National Health Interview Survey (2010) data. Then, assuming increased government expenditure on health awareness, I simulate the effects such policy interventions and see what improvements in the public health can be achieved. Also, I propose the count-approach incremental cost effectiveness ratio (C-ICER) which is a simple measure to assess the cost effectiveness of public health awareness campaigns.

L'apprentissage statistique propose un vaste ensemble de techniques capables de construire des modèles prédictifs à partir d'observations passées. Ces techniques ont montré leurs capacités à traiter des volumétries importantes de données sur des problèmes réels. Cependant, de nouvelles applications génèrent de plus en plus de données qui sont seulement visibles sous la forme d'un flux et doivent être traitées séquentiellement. Parmi ces applications on citera : la gestion de réseaux de télécommunications, la modélisation des utilisateurs au sein d'un réseau social, le web mining. L'un des défis techniques est de concevoir des algorithmes permettant l'apprentissage avec les nouvelles contraintes imposées par les flux de données. Nous proposons d'abord ce problème en proposant de nouvelles techniques de résumé de flux de données dans le cadre de l'apprentissage supervisé. Notre méthode est constituée de deux niveaux. Le premier niveau utilise des techniques incrémentales de résumé en-ligne pour les flux qui prennent en compte les ressources mémoire et processeur et possèdent des garanties en termes d'erreur. Le second niveau utilise les résumés de faible taille, issus du premier niveau, pour construire le résumé final à l'aide d'une méthode supervisée performante hors-ligne. Ces résumés constituent un prétraitement qui nous permet de proposer de nouvelles versions du classifieur bayésien naïf et des arbres de décision fonctionnant en-ligne sur flux de données. Les flux de données peuvent ne pas être stationnaires mais comporter des changements de concept. Nous proposons aussi une nouvelle technique pour détecter ces changements et mettre à jour nos classifieurs. / Statistical learning provides numerous algorithms to build predictive models on past observations. These techniques proved their ability to deal with large scale realistic problems. However, new domains generate more and more data which are only visible once and need to be processes sequentially. These volatile data, known as data streams, come from telecommunication network management, social network, web mining. The challenge is to build new algorithms able to learn under these constraints. We proposed to build new summaries for supervised classification. Our summaries are based on two levels. The first level is an online incremental summary which uses low processing and address the precision/memory tradeoff. The second level uses the first layer summary to build the final sumamry with an effcient offline method. Building these sumamries is a pre-processing stage to develop new classifiers for data streams. We propose new versions for the naive-Bayes and decision trees classifiers using our summaries. As data streams might contain concept drifts, we also propose a new technique to detect these drifts and update classifiers accordingly.

Le présent travail porte sur la modélisation 3D du comportement hydromécanique du bois en général et du sapin blanc en particulier, avec la prise en compte des couplages entre les effets orthotrope, hydrique, élastique, viscoélastique et mécanosorptifs, y compris l’effet hygroverrou qui est un phénomène de blocage temporaire de la déformation en phase de séchage sous contrainte. Ce mémoire est scindé en trois grandes parties découpées en sept chapitres. La première partie examine le contexte et la problématique du comportement hydromécanique du bois. Les aspects traités vont de la structure, du phénomène hygroscopique et de l’effet de retrait/gonflement, aux divers aspects du comportement hydromécanique du bois sous humidité constante ou variable comme l’orthotropie, la viscoélasticité et les effets mécanosorptifs qui traduisent l’interaction complexe entre le chargement mécanique et les variations d’humidité. Les bases pour la modélisation sont présentées dans le deuxième chapitre, telles que la formulation incrémentale à pas de temps fini pour modéliser le comportement viscoélastique orthotrope 3D et les modèles mécanosorptifs intéressants de la littérature. A partir de cette étude bibliographique, on propose une voie pour modéliser dans ce travail l’effet mécanosorptif comme la somme de trois effets élémentaires : effet mécanosorptif irréversible, fluage mécanosorptif et effet hygroverrou. Les deux premiers effets sont modélisés par des modèles existants tandis que la modélisation de l’effet hygroverrou est l’objet d’une démarche originale dans ce travail. La deuxième partie, décomposée en deux chapitres, est consacrée à la construction du modèle de comportement 3D. Le premier chapitre présente les développements mathématiques pour l’élaboration d’un modèle analytique. Ce modèle est basé sur l’hypothèse de partition de la déformation totale en une somme de six déformations élémentaires : hydrique, élastique instantanée, viscoélastique pure, hygroverrou, mécanosorptive irréversible et de fluage mécanosorptif. Les variations de ces déformations élémentaires sont établies de manière séparée. En particulier, la loi d’évolution de la déformation hygroverrou construite sur la base d’observations expérimentales, est différente en phase de séchage et d’humidification. Une contrainte auxiliaire, introduite en respectant les principes thermodynamiques, permet de résoudre le problème de la récupération de la déformation hygroverrou en phase d’humidification en cas de contrainte nulle ou insuffisante. En parallèle, un nouveau modèlerhéologique est également proposé pour modéliser le comportement viscoélastique à humidité variable. Ce modèle, équivalent à un modèle de Maxwell généralisé et/ou à un modèle de Kelvin-Voigt généralisé, est capable de décrire le fluage aussi bien que la relaxation. Le deuxième chapitre de cette partie est consacré à la transformation du modèle analytique en une forme incrémentale à pas de temps fini. La contribution de chaque partie élémentaire est établie par résolution exacte à partir d’équations différentielles ou d’intégrales de Boltzmann. La somme des formes élémentaires ainsi obtenues conduit à la loi de comportement du modèle complet qui est similaire à celle d’un comportement thermo-élastique équivalent. Du fait de la procédure d’intégration, le pas de temps de calcul est fini mais pas nécessairement petit. Cette propriété est très importante car elle permet de réduire considérablement le temps de calcul tout en préservant une très bonne précision. La dernière partie est divisée en trois chapitres. Elle présente la mise en œuvre numérique du modèle hydromécanique à l’aide du code d’éléments finis Cast3m, suivie de la validation et d’applications à diverses classes de problèmes. L’algorithme numérique est organisé en modules indépendants. Des procédures élémentaires sont construites pour réaliser des fonctions spécifiques ; elles sont appelées selon un ordre précis par un programme principal. (...) / This work concerns 3D modeling of hydro-mechanical behavior of wood in general and the silver fir (Abies alba Mill.) in particular with taking account of the couplings between the effects: orthotropie, hydric, elastic, viscoelastic and mechano-sorptive including hydro-lock effect that is a temporary locking of the mechanical strain during a period of drying under stress. This memory is divided into three parts divided into seven chapters. The first part examines the background and the problem of hydro-mechanical behavior of wood. The aspects go from the structure, hygroscopic phenomenon and the effect of swelling/shrinkage, to various aspects of the hydro-mechanical behavior of wood under constant or variable moisture as orthotropie, viscoelasticity and mechano-sorptive effects that is the interaction complex between mechanical loading and moisture variations. The bases for modeling are presented in the second chapter, such as incremental formulation on finite time step to model the 3D orthotropic viscoelastic behavior and interesting mechano-sorptive models of literature. From this literature review, we propose a way to model in this work mechano-sorptive effect as the sum of three elementary effects: irrecoverable mechanosorptive, mechano-sorptive creep and hydro-lock effect. The first and the second effects are modeled by existing models while modeling hygroverrou effect is an original subject in this work. The second part, divided into two chapters, is dedicated to building the 3D model of behavior. The first chapter presents the mathematical developments for the development of an analytical model. This model is based on the assumption partition of the total strain by a sum of six elementary strains: hydric, instant elastic, viscoelastic pure, hydro-lock, irrecoverable mechano-sorptive and mechano-sorptive creep. Variations of these elementary strains are established separately. In particular, the evolution law of the hydro-lock strain constructed on the basis of experimental observations is different in phase of drying and moistening. An auxiliary stress introduced in accordance with thermodynamic principles, solves the problem of recovering the hydro-lock strain in the moistening phase in case with zero or little stress. In parallel, a new rheological model is proposed to model the viscoelastic behavior at variable humidity. This model, equivalent to a generalized Maxwell model and / or a generalized Kelvin-Voigt model, is able to describe the creep as well as relaxation. The second chapter of this part is devoted to the transformation of the analytical model in an incremental form on finite time step. The contribution of each elementary part is established by exact resolutionfrom differential equations or Boltzmann’s integrals. The sum of elementary forms thus obtained leads to the complete model behavior law which is similar to that of an equivalent thermo-elastic behavior. Because of the integration process, the time step calculation is finished but not necessarily small. This property is very important because it significantly reduces the computation time while maintaining very good accuracy. The last part is divided into three chapters. It presents numerical implementation of hydro- mechanical model using the finite element code Cast3m, followed by validation and applications to various classes of problems. The numerical algorithm is organized into independent modules. Elementary procedures are built to perform specific functions; they are called in a specific order by the main program. Model validation is made by comparison between simulated results and experimental data available in tension and bending. The last chapter of the thesis presents applications of solid wood reconstituted silver fir. They show the ability of the model to predict the states of stress and strain in timber structures under mechanical loading and variable humidity.

Incremental sheet forming (ISF) is a flexible process where an indenter moves over the surface of a sheet of metal to form a 3D shell incrementally by a progression of localised deformation. Despite extensive research into the process, the deformation mechanics is not fully understood. This thesis presents new insights into the mechanics of ISF applied to two groups of materials: sheet metals and sandwich panels. A new system for measuring tool forces in ISF is commissioned. The system uses six loadcells to measure reaction forces on the workpiece frame. Each force signal has an uncertainty of ±15 N. This is likely to be small in comparison to tool forces measured in ISF. The mechanics of ISF of sheet metals is researched. Through-thickness deformation and strains of copper plates are measured for single-point incremental forming (SPIF) and two-point incremental forming (TPIF). It is shown that the deformation mechanisms of SPIF and TPIF are shear parallel to the tool direction, with both shear and stretching perpendicular to the tool direction. Tool forces are measured and compared throughout the two processes. Tool forces follow similar trends to strains, suggesting that shear parallel to the tool direction is a result of friction between the tool and workpiece. The mechanics of ISF of sandwich panels is investigated. The mechanical viability of applying ISF to various sandwich panel designs is evaluated by observing failure modes and damage under two simple tool paths. ISF is applicable to metal/polymer/metal sandwich panels. This is because the cores and faceplates are ductile and largely incompressible, and therefore survive local indentation during ISF without collapse. Through-thickness deformation, tool forces and applicability of the sine law for prediction of wall thickness are measured and compared for a metal/polymer/metal sandwich panel and a monolithic sheet metal. The mechanical results for ISF of sheet metals transfer closely to sandwich panels. Hence, established knowledge and process implementation procedures derived for ISF of monolithic sheet metals may be used in the future for ISF of sandwich panels.

Unreinforced masonry (URM) structures have shown tobe susceptible to significant damage during strong earthquakes. Vulnerability assessment of URM buildings is needed so that appropriate mitigation strategies can be implemented. The existing Canadian practice consists of rapid seismic screening of buildings to assign priorities for further and more refined assessments, followed by refined analysis of individual critical buildings. The current seismic screening procedure, from 1992, is based on qualitative observations of seismic vulnerability, enabling the assignment of seismic priority indices, quantified on the basis of expert opinion and experience. More refined tools are needed for seismic vulnerability assessment of URM buildings in Canada, based on the current Canadian seismic hazard values. The objective of the research project is to fulfill these needs by developing fragility curves that provide a probabilistic assessment of different levels of building performance under different intensities ofeastern and western seismicity. Using an inventory of over 50,000 structures, a seismic assessment of typical low-rise and mid-rise URM structures located in eastern and western Canada was carried out. The required analyses were done using applied element method software which effectively modeled the in-plane and out-of-plane behaviour of masonry walls. Using incremental dynamic analysis, fragility curves were developed to reflect the capacity of URM structures with a wide variety of selected structural and ground motion parameters. The results were verified against available fragility information in the literature. They show the significance of selected parameters, while providing effective tools for seismic vulnerability assessment of URM buildings in eastern and western Canada.

abstract: In the last decade deep learning based models have revolutionized machine learning and computer vision applications. However, these models are data-hungry and training them is a time-consuming process. In addition, when deep neural networks are updated to augment their prediction space with new data, they run into the problem of catastrophic forgetting, where the model forgets previously learned knowledge as it overfits to the newly available data. Incremental learning algorithms enable deep neural networks to prevent catastrophic forgetting by retaining knowledge of previously observed data while also learning from newly available data. This thesis presents three models for incremental learning; (i) Design of an algorithm for generative incremental learning using a pre-trained deep neural network classifier; (ii) Development of a hashing based clustering algorithm for efficient incremental learning; (iii) Design of a student-teacher coupled neural network to distill knowledge for incremental learning. The proposed algorithms were evaluated using popular vision datasets for classification tasks. The thesis concludes with a discussion about the feasibility of using these techniques to transfer information between networks and also for incremental learning applications. / Dissertation/Thesis / Masters Thesis Computer Science 2020

Existing office buildings’ embodied energy, history and culture offer something a newly constructed building cannot. On the other hand, new office buildings’ adoption of new technologies and building philosophies offer a range of sustainable efficiencies previously unavailable. Combining these efficiencies with elements that embrace human diversity and well- being offer the opportunity to not only mend our existing buildings’ deteriorating physical bodies but aid in creating workplaces that promote good physical and mental health. This project provides recommendation on how an existing high-rise commercial building can incorporate a number of incremental improvements that continually evolve to meet rapidly changing market demands. This design approach allows for ease of installation and modification to meet the needs of the tenants and the building owner.