Pose estimation and tracking is essential for applications involving human controls. Specifically, as the primary operating tool for human activities, hand pose estimation plays a significant role in applications such as hand tracking, gesture recognition, human-computer interaction and VR/AR. As the field develops, there has been a trend to utilize deep learning to estimate the 2D/3D hand poses using color-based information without depth data. Within the depth-based as well as color-based approaches, the research community has primarily focused on single-hand scenarios in a localized/normalized coordinate system. Due to the fact that both hands are utilized in most applications, we propose to push the frontier by addressing two-hand pose estimation in the global coordinate system using only color information. Our first chapter introduces the first system capable of estimating global 3D joint locations for both hands via only monocular RGB input images. To enable training and evaluation of the learning-based models, we propose to introduce a large-scale synthetic 3D hand pose dataset Ego3DHands. As knowledge in synthetic data cannot be directly applied to the real-world domain, a natural two-hand pose dataset is necessary for real-world applications. To this end, we present a large-scale RGB-based egocentric hand dataset Ego2Hands in two chapters. In chapter 2, we address the task of two-hand segmentation/detection using images in the wild. In chapter 3, we focus on the task of two-hand 2D/3D pose estimation using real-world data. In addition to research in hand pose estimation, chapter 4 includes our work on interactive refinement that generalizes the backpropagating refinement technique for dense prediction models.

Analyses of nuclear reactor safety have increasingly required coupling of full three dimensional neutron kinetics (NK) core models with system transient thermal-hydraulics (TH) codes. To produce results within a reasonable computing time, the coupled codes use different spatial description of the reactor core. The TH code uses few, typically 5 to 20 TH channels, which represent the core. The NK code uses explicit node for each fuel assembly. Therefore, a spatial mapping of coarse grid TH and fine grid NK domain is necessary. However, improper mappings may result in loss of valuable information, thus causing inaccurate prediction of safety parameters. The purpose of this thesis is to study the sensitivity of spatial coupling (channel refinement and spatial mapping) and develop recommendations for NK-TH mapping in simulation of safety transients – Control Rod Drop, Turbine Trip, Feedwater Transient combined with stability performance (minimum pump speed of recirculation pumps). The research methodology consists of spatial coupling convergence study, as increasing number of TH channels and different mapping approach the reference case. The reference case consists of one TH channel per one fuel assembly. The comparison of results has been done under steady-state and transient conditions. Obtained results and conclusions are presented in this licentiate thesis.

Cement production is responsible for around 7% of the global anthropogenic carbon dioxide emissions. More than half of these emissions are due to the unavoidable release of carbon dioxide upon thermal decomposition of the main raw material limestone. Many different options for carbon capture are currently being investigated to lower emissions, and one potential route to facilitate carbon capture could be the implementation of an electrified separate calcination step. However, potential effects on the phase composition of cement clinker need to be investigated, which is the aim of the present study. Phases of special interest are alite, belite, aluminate, ferrite, calcite, and lime.  The phase composition during clinker formation was examined through HT-XRD lab-scale experiments, allowing the phase transformations to be observed in situ. Two different methods of separate calcination were investigated, one method in which the raw meal was calcined separately, and one method where the limestone was calcined separately. The former yielded an alite amount similar to that of the reference experiments, whereas the latter method yielded a lower amount. It could, unfortunately, not be excluded that the difference was due to poor experimental conditions, and additional experiments are needed to investigate the matter further. The study does, however, indicate that a calcined raw meal might be used to produce a clinker of similar phase composition concerning major phases belite, aluminate, ferrite, alite, and free lime. A raw meal containing calcined limestone might, however, need longer residence time at clinkering temperature too obtain similar phase composition. In addition, a raw meal containing calcined limestone was observed to be carbonated to a greater extent upon reheating than a calcined raw meal. Further experiments are needed to fully understand the effects on clinker composition of an electrified separate calcination step, and several improvements to the experimental method are given in the study.

The use cases of Information Extraction (IE) are more or less endless, often consisting of a combination of Named Entity Recognition (NER) and Relation Extraction (RE). One use case of IE is the extraction of transaction information from Norwegian insider transaction Press Releases (PRs), where a transaction consists of at most four entities: the name of the owner performing the transaction, the number of shares transferred, the transaction date, and the price of the shares bought or sold. The relationships between the entities define which entity belongs to which transaction, and whether shares were bought or sold. This report has investigated how a pair of supervised NER and RE models extract this information. Since these Norwegian PRs were not labeled, two different approaches to annotating the transaction entities and their associated relations were investigated, and it was found that it is better to annotate only entities that occur in a relation than annotating all occurrences. Furthermore, the number of PRs needed to achieve a satisfactory result in the IE pipeline was investigated. The study shows that training with about 400 PRs is sufficient for the results to converge, at around 0.85 in F1-score. Finally, the report shows that there is not much difference between a complex RE model and a simple rule-based approach, when applied on the studied corpus.

Dans cette thèse, nous étudions le problème de l’analyse de signatures de pannes dans le domaine de la maintenance avionique, afin d’identifier les défaillances au sein d’équipements en panne et suggérer des actions correctives permettant de les réparer. La thèse a été réalisée dans le cadre d’une convention CIFRE entre Thales Research & Technology et l’Université Paris-Sud. Les motivations sont donc à la fois théoriques et industrielles. Une signature de panne devrait fournir toutes les informations nécessaires pour identifier, comprendre et réparer la panne. Pour comprendre le mécanisme la panne son identification doit donc être explicable. Nous proposons une approche à base d’ontologies pour modéliser le domaine d’étude, permettant une interprétation automatisée des tests techniques réalisés afin d’identifier les pannes et obtenir les actions correctives associées. Il s’agit d’une approche d’apprentissage de concepts permettant de découvrir des concepts représentant les signatures de pannes tout en fournissant des explications sur les choix de propositions de réparations. Comme les signatures ne sont pas connues a priori, un algorithme d’apprentissage automatique non supervisé approxime les définitions des concepts. Les signatures apprises sont fournies sous forme de définitions de la logique de description (DL) et ces définitions servent d’explications. Contrairement aux techniques courantes d’apprentissage de concepts conçues pour faire de l’apprentissage supervisé ou basées sur l’analyse de patterns fréquents au sein de gros volumes de données, l’approche proposée adopte une perspective différente. Elle repose sur une construction bottom-up de l’ontologie. Le processus d’apprentissage est réalisé via un opérateur de raffinement appliqué sur l’espace des expressions de concepts et le processus est guidé par les données, c’est-à-dire les individus de l’ontologie. Ainsi, les notions de justifications, de concepts plus spécifiques et de raffinement de concepts ont été révisées et adaptées pour correspondre à nos besoins. L’approche a ensuite été appliquée au problème de la maintenance avionique. Un prototype a été implémenté et mis en œuvre au sein de Thales Avionics à titre de preuve de concept. / In this thesis we explore the problem of signature analysis in avionics maintenance, to identify failures in faulty equipment and suggest corrective actions to resolve the failure. The thesis takes place in the context of a CIFRE convention between Thales R&T and the Université Paris-Sud, thus it has both a theoretical and an industrial motivation. The signature of a failure provides all the information necessary to understand, identify and ultimately repair a failure. Thus when identifying the signature of a failure it is important to make it explainable. We propose an ontology based approach to model the domain, that provides a level of automatic interpretation of the highly technical tests performed in the equipment. Once the tests can be interpreted, corrective actions are associated to them. The approach is rooted on concept learning, used to approximate description logic concepts that represent the failure signatures. Since these signatures are not known in advance, we require an unsupervised learning algorithm to compute the approximations. In our approach the learned signatures are provided as description logics (DL) definitions which in turn are associated to a minimal set of axioms in the A-Box. These serve as explanations for the discovered signatures. Thus providing a glass-box approach to trace the reasons on how and why a signature was obtained. Current concept learning techniques are either designed for supervised learning problems, or rely on frequent patterns and large amounts of data. We use a different perspective, and rely on a bottom-up construction of the ontology. Similarly to other approaches, the learning process is achieved through a refinement operator that traverses the space of concept expressions, but an important difference is that in our algorithms this search is guided by the information of the individuals in the ontology. To this end the notions of justifications in ontologies, most specific concepts and concept refinements, are revised and adapted to our needs. The approach is then adapted to the specific avionics maintenance case in Thales Avionics, where a prototype has been implemented to test and evaluate the approach as a proof of concept.

Ce travail de thèse présente l'implémentation de la simulation d'écoulements diphasiques dans des conditions de réacteurs nucléaires à caloporteur eau, à l'échelle de bulles individuelles. Pour ce faire, nous étudions plusieurs modèles d'écoulements thermohydrauliques et nous focalisons sur une technique de capture d'interface mince entre phases liquide et vapeur. Nous passons ainsi en revue quelques techniques possibles de maillage adaptatif (AMR) et nous fournissons des outils algorithmiques et informatiques adaptés à l'AMR par patchs dont l'objectif localement la précision dans des régions d'intérêt. Plus précisément, nous introduisons un algorithme de génération de patchs conçu dans l'optique du calcul parallèle équilibré. Cette approche nous permet de capturer finement des changements situés à l'interface, comme nous le montrons pour des cas tests d'advection ainsi que pour des modèles avec couplage hyperbolique-elliptique. Les calculs que nous présentons incluent également la simulation du système de Navier-Stokes incompressible qui modélise la déformation de l'interface entre deux fluides non-miscibles. / This PhD work presents the implementation of the simulation of two-phase flows in conditions of water-cooled nuclear reactors, at the scale of individual bubbles. To achieve that, we study several models for Thermal-Hydraulic flows and we focus on a technique for the capture of the thin interface between liquid and vapour phases. We thus review some possible techniques for Adaptive Mesh Refinement (AMR) and provide algorithmic and computational tools adapted to patch-based AMR, which aim is to locally improve the precision in regions of interest. More precisely, we introduce a patch-covering algorithm designed with balanced parallel computing in mind. This approach lets us finely capture changes located at the interface, as we show for advection test cases as well as for models with hyperbolic-elliptic coupling. The computations we present also include the simulation of the incompressible Navier-Stokes system, which models the shape changes of the interface between two non-miscible fluids.

Dans cette thèse, nous nous intéressons à la résolution des équations SPN du transport de neutrons au sein des cœurs de réacteurs nucléaires à eau pressurisée. Ces équations forment un problème aux valeurs propres généralisé. Dans notre étude nous commençons par le problème source associé et ensuite nous étudions le problème aux valeurs propres. Un cœur de réacteur est composé de différents milieux: le combustible, le fluide caloporteur, le modérateur... à cause de ces hétérogénéités de la géométrie, le flux solution du problème source peut être peu régulier. Nous proposons l’analyse numérique de l’approximation de la solution par la méthode des éléments finis du problème source dans le cas où la solution est peu régulière. Pour le problème aux valeurs propres, dans le cas mixte, les théories déjà développées ne s’appliquent pas. Nous proposons ici une nouvelle méthode pour étudier la convergence de la méthode des éléments finis mixtes pour les problèmes aux valeurs propres. Pour les solutions peu régulières, la montée en ordre de la méthode des éléments finis n’améliore pas l’approximation du problème, il faut raffiner le maillage aux alentours des singularités de la solution. La géométrie des cœurs de réacteur se prête bien aux maillages cartésiens, mais leur raffinement augmente vite leur nombre de degrés de liberté. Pour palier à cette augmentation, nous proposons ici une méthode de décomposition de domaine qui permet d’utiliser des maillages globalement non-conformes. / In this thesis, we investigate the resolution of the SPN neutron transport equations in pressurized water nuclear reactor. These equations are a generalized eigenvalue problem. In our study, we first considerate the associated source problem and after we concentrate on the eigenvalue problem. A nuclear reactor core is composed of different media: the fuel, the coolant, the neutron moderator... Due to these heterogeneities of the geometry, the solution flux can have a low-regularity. We propose the numerical analysis of its approximation with finite element method for the low regular case. For the eigenvalue problem under its mixed form, we can not rely on the theories already developed. We propose here a new method for studying the convergence of the SPN neutron transport eigenvalue problem approximation with mixed finite element. When the solution has low-regularity, increasing the order of the method does not improve the approximation, the triangulation need to be refined near the singularities of the solution. Nuclear reactor cores are well-suited for Cartesian grids, but the refinement of these sort of triangulations increases rapidly their number of degrees of freedom. To avoid this drawback, we propose domain decomposition method which can handle globally non-conforming triangulations.

L’objectif de ce travail est d’étudier les capacités de la méthode de Boltzmann sur réseau (LBM) dans un cadre numériquement contraignant : celui de la simulation aéroacoustique en maillage non-uniforme, à très haut nombre de Reynolds et à nombre de Mach non négligeable (Ma > 0.1), appliquée à l’automobile. La problématique industrielle est celle du calcul du bruit intérieur d’origine aérodynamique, dont le calcul du champ de pression pariétal instationnaire sur le vitrage conducteur est la première étape décisive. Il a été constaté qu’un manque de précision sur la faible part acoustique du champ de pression total sur le vitrage, provenant très probablement d’erreurs au niveau des transitions de résolution du maillage, était la cause d’une surestimation du bruit intérieur. Nous présentons d’abord une construction cohérente et unifiée de la méthode de Boltzmann sur réseau à partir de l’équation de Boltzmann, dans un cadre athermal faiblement compressible. Nous étudions ensuite en détail les propriétés aéroacoustiques de la LBM, en parcourant toutes les grandes familles d’opérateurs de collision de la littérature. Une variante de modèle à temps de relaxation multiples, utilisable pour l’aéroacoustique, est présentée et testée. Un modèle alternatif simplifié de filtrage sélectif, rapide et compact, est développé et validé. La problématique des maillages non-uniformes est abordée. Un recensement exhaustif des études LBM menées dans ce cadre dans la littérature montre qu’aucune ne correspond à nos contraintes. Des algorithmes alternatifs aux transitions sont développés. Enfin, des applications industrielles sont réalisées à l’aide des modèles développés dans le mémoire. / The main goal of this work is to study the capacities of the Lattice Boltzmann Method in a constrained numerical framework : that of numerical simulation in automotive aeroacoustics with non-uniform meshes, at high Reynolds number and non egligible Mach number (Ma > 0.1). The industrial problem is the computation of the interior aerodynamic noise, which includes as its first decisive step the computation of the unsteady wall pressure field on the car windows. It was observed that a lack of precision on the weak acoustic part of the total pressure field on the driver-side window, which is most probably due to errors at mesh refinement interfaces, caused an overestimation of the interior noise. We first present a coherent and unified construction of the Lattice BoltzmannMethod from the Boltzmann equation, in an athermal weakly compressible framework. Then, we study in details the aeroacoustic properties of the LBM by reviewingall the main families of collisional operators that exist in the literature. A variant of multiple relaxation time operator that can be used for aeroacoustics is presented and tested. A simplified alternative selective filter, fast and compact, is developped and numerically validated. The problem of non-uniform meshes is discussed. An exhaustive review of the LBM studies that have been carried out within that framework shows that none of them corresponds to our constraints. Alternative transition nodes algorithms are developed. Finally, all the developed models of this work are applied to industrial cases.

In the modern world, Scrum is widely adopted by many companies for software development, no matter the size of the company. As Scrum provides rapid development and meets the customer needs in a faster and efficient way, many companies want to move towards Scrum but moving towards Scrum is difficult and challenging due to different barriers for example cost, lack of experience, and so on. This thesis is to address the barriers which are involved during the Scrum adoption for software start-up companies such as communication, team mindset, backlog refining, daily stand up, and retrospective meetings. In this research report, these issues are divided based on the factors such as organization, people, process, and tools. We have figured out these barriers using action research based on the observation done during the Scrum team meetings and through interviews and surveys.

No description available.