Cette thèse est consacrée à l'étude des propriétés des transports diphasiques au sein des couches de diffusions (Gas Diffusion Layer = GDL) des piles à combustible PEMFC (Proton Exchange Membrane Fuel Cells). La GDL est faite d'une structure fibreuse (dont l'épaisseur est de quelques centaines de micromètres) traitée généralement avec une matière hydrophobe. Des images numériques de la GDL réelle obtenues par tomographie X sont d'abord analysées afin d'étudier des propriétés telles que la porosité, la perméabilité, ou le tenseur de diffusion. L'écrasement de la GDL est ensuite simulé en utilisant un algorithme comprimant les fibres dans un plan transversal. Les transports diphasiques (invasion quasi statique d'eau liquide) sont modélisés dans des réseaux de pores, milieux représentatifs de l'espace poreux de la GDL, en relation avec le problème de la gestion de l'eau dans les piles PEMFC. Deux algorithmes d'invasion, dénommés algorithmes séquentiel et cinétique, sont développés et comparés pour analyser les distributions de phases au sein des GDL. Un point clé est que l'eau rentre dans la couche poreuse par divers points d'injection indépendants, conduisant à la possibilité de multiples points de percée. Des expériences sur un système microfluidique sont conduites pour valider les algorithmes utilisés. Une étude statistique est menée pour caractériser le nombre de points de percée, les profils de saturation, l'accès au gaz, le transport diffusif, de même que l'influence du piégeage et de la mouillabilité mixte. / This thesis is devoted to the study of transport properties and two-phase flow in the Gas Diffusion Layer (GDL) of Proton Exchange Membrane Fuel Cells (PEMFC). A GDL is a thin fibrous structure (a few hundreds μm thick) treated generally with a hydrophobic agent. Numerical images obtained from X-ray computed tomography X are first exploited to study properties such as the porosity, permeability and diffusion tensors of a real GDL microstructure. The effect of GDL compression is also investigated using an algorithm mimicking the compression in GDL through plane direction. Then two phase flow (quasi-static water invasion) is studied in relation with the water management problem in PEMFC, using a structured pore network representation of the pore space. Two invasion algorithms, referred to as the sequential and the kinetic algorithm respectively, are developed and compared to study the fluid distributions within the GDL. A key point is that water enters the porous layer through multiple independent inlet injection points, leading to the possibility of many breakthrough points. Experiments are conducted on a microfluidic device to validate the algorithms. A numerical statistical study is performed to characterize the breakthrough point statistics, saturation profiles, gas access, diffusion transport as well as the influence of trapping and mixed wettability.

In this thesis, we have proposed a new computing platform called probabilistic spin logic (PSL) based on probabilistic bits (p-bit) using low barrier nanomagnets (LBM) whose thermal barrier is of the order of a kT unlike conventional memory and spin logic devices that rely on high thermal barrier magnets (40-60 kT) to retain stability. p-bits are tunable random number generators (TRNG) analogous to the concept of binary stochastic neurons (BSN) in artificial neural network (ANN) whose output fluctuates between a +1 and -1 states with 50-50 probability at zero input bias and the stochastic output can be tuned by an applied input producing a sigmoidal characteristic response. p-bits can be interconnected by a synapse or weight matrix [J] to build p-circuits for solving a wide variety of complex unconventional problems such as inference, invertible Boolean logic, sampling and optimization. It is important to update the p-bits sequentially for proper operation where each p-bit update is informed of the states of other p-bits that it is connected to and this requires the use of sequencers in digital clocked hardware. But the unique feature of our probabilistic hardware is that they are autonomous that runs without any clocks or sequencers.<br>To ensure the necessary sequential informed update in our autonomous hardware it is important that the synapse delay is much smaller than the neuron fluctuation time.<br>We have demonstrated the notion of this autonomous hardware by SPICE simulation of different designs of low barrier nanomagnet based p-circuits for both symmetrically connected Boltzmann networks and directed acyclic Bayesian networks. It is interesting to note that for Bayesian networks a specific parent to child update order is important and requires specific design rule in the autonomous probabilistic hardware to naturally ensure the specific update order without any clocks. To address the issue of scalability of these autonomous hardware we have also proposed and benchmarked compact models for two different hardware designs against SPICE simulation and have shown that the compact models faithfully mimic the dynamics of the real hardware.<br>

Thesis (Ph.D. (Statistics)) -- University of Limpopo, 2017 / Operations Research (OR) is a scientific method for developing quantitatively well-grounded recommendations for decision making. While it is true that it uses a variety of mathematical techniques, OR has a much broader scope. It is in fact a systematic approach to solving problems, which uses one or more analytical tools in the process of analysis. Over the years, OR has evolved through different stages. This study is motivated by new real-world challenges needed for efficiency and innovation in line with the aims and objectives of OR – the science of better, as classified by the OR Society of the United Kingdom. New real-world challenges are encountered on a daily basis from problems arising in the fields of water, energy, agriculture, mining, tourism, IT development, natural phenomena, transport, climate change, economic and other societal requirements. To counter all these challenges, new techniques ought to be developed. The growth of global markets and the resulting increase in competition have highlighted the need for OR techniques to be improved. These developments, among other reasons, are an indication that new techniques are needed to improve the day-to-day running of organisations, regardless of size, type and location. The principal aim of this study is to modify and develop new OR techniques that can be used to solve emerging problems encountered in the areas of linear programming, integer programming, mixed integer programming, network routing and travelling salesman problems. Distribution models, resource allocation models, travelling salesman problem, general linear mixed integer ii programming and other network problems that occur in real life, have been modelled mathematically in this thesis. Most of these models belong to the NP-hard (non-deterministic polynomial) class of difficult problems. In other words, these types of problems cannot be solved in polynomial time (P). No general purpose algorithm for these problems is known. The thesis is divided into two major areas namely: (1) network models and (2) resource allocation and distribution models. Under network models, five new techniques have been developed: the minimum weight algorithm for a non-directed network, maximum reliability route in both non-directed and directed acyclic network, minimum spanning tree with index less than two, routing through 0k0 specified nodes, and a new heuristic to the travelling salesman problem. Under the resource allocation and distribution models section, four new models have been developed, and these are: a unified approach to solve transportation and assignment problems, a transportation branch and bound algorithm for the generalised assignment problem, a new hybrid search method over the extreme points for solving a large-scale LP model with non-negative coefficients, and a heuristic for a mixed integer program using the characteristic equation approach. In most of the nine approaches developed in the thesis, efforts were done to compare the effectiveness of the new approaches to existing techniques. Improvements in the new techniques in solving problems were noted. However, it was difficult to compare some of the new techniques to the existing ones because computational packages of the new techniques need to be developed first. This aspect will be subject matter of future research on developing these techniques further. It was concluded with strong evidence, that development of new OR techniques is a must if we are to encounter the emerging problems faced by the world today. Key words: NP-hard problem, Network models, Reliability, Heuristic, Largescale LP, Characteristic equation, Algorithm.

The investigation of the hidden metric space behind complex network topologies is a fervid topic in current network science and the hyperbolic space is one of the most studied, because it seems associated to the structural organization of many real complex systems. The popularity-similarity-optimization (PSO) model simulates how random geometric graphs grow in the hyperbolic space, generating realistic networks with clustering, small-worldness, scale-freeness and rich-clubness. However, it misses to reproduce an important feature of real complex networks, which is the community organization. The geometrical-preferential-attachment (GPA) model was recently developed in order to confer to the PSO also a soft community structure, which is obtained by forcing different angular regions of the hyperbolic disk to have a variable level of attractiveness. However, the number and size of the communities cannot be explicitly controlled in the GPA, which is a clear limitation for real applications. Here, we introduce the nonuniform PSO (nPSO) model. Differently from GPA, the nPSO generates synthetic networks in the hyperbolic space where heterogeneous angular node attractiveness is forced by sampling the angular coordinates from a tailored nonuniform probability distribution (for instance a mixture of Gaussians). The nPSO differs from GPA in other three aspects: it allows one to explicitly fix the number and size of communities; it allows one to tune their mixing property by means of the network temperature; it is efficient to generate networks with high clustering. Several tests on the detectability of the community structure in nPSO synthetic networks and wide investigations on their structural properties confirm that the nPSO is a valid and efficient model to generate realistic complex networks with communities.

Social network analysis is a fruitful approach to the study of relations and interaction between actors involved in organized crime. This dissertation utilizes network perspective to study several cases of organized criminal groups. It is divided into eight chapters. The first introductory chapter is followed by a chapter reviewing the most important network concepts, measures, and models, and their application in the study of organized crime. The four subsequent chapters are empirical studies. The third chapter is a case study of a political corruption network, known as the Rath affair. The study shows that the network consists of different, sometimes overlapping, relations (multiplexity), namely collaboration, resource transfer, and pre-existing ties. The network shows a clear core-periphery structure with politicians forming a dense core and businesspeople occupying periphery. The following chapter studies a case of counterfeit alcohol distribution network, known as the methanol affair. The network structure is composed of two subgroups bridged by one tie, permitting relatively efficient distribution of the beverages. Furthermore, statistical models point out the importance of triadic closure and pre-existing ties for the formation of ties in the network. The fifth chapter tests an influential theory in...

No description available.

[pt] A formação e o acúmulo de condensado em reservatórios de gás retrógrado, especialmente na vizinhança de poços de produção, obstruem parcialmente o fluxo de gás e afetam negativamente a composição dos fluidos produzidos. Entretanto, a previsão de bloqueio por condensado é comumente imprecisa, visto que experimentos raramente reproduzem as condições extremas e composições complexas dos fluidos dos reservatórios, enquanto a maioria dos modelos em escala de poros simplificam demasiadamente os fenômenos físicos associados à transição de fases entre gás e condensado. Para corrigir essa lacuna, um modelo de rede de poros isotérmico composicional e totalmente implícito é apresentado. As redes de poros propostas consistem em estruturas tridimensionais de capilares constritos circulares. Modos de condensação e padrões de escoamento são atrubuídos aos capilares de acordo com a molhabilidade do meio, as saturações locais e a influência de forças viscosas e capilares. Nos nós da rede, pressão e conteúdo molar são determinados através da solução acoplada de equações de balanço molar e consistênc ia de volumes. Concomitantemente, um cálculo de flash à pressão e à temperatura constantes, baseado na equação de estado de Peng e Robinson, é realizado em cada nó, atualizando as saturações e composições das fases. Para a validação do modelo proposto, análises de escoamento foram executadas baseadas em experimentos de escoamento em testemunho reportados na literatura, usando composição dos fluidos e condições de escoamento correspondentes, e geometria do meio poroso aproximada. Curvas de permeabilidade relativa medidas nos experimentos e previstas pelo modelo mostraram boa concordância quantitativa, para dois valores de tensão interfacial e três valores de velocidade de escoamento de gás. Após a validação, o modelo foi usado para avaliar alteração de molhabilidade e injeção de gás como possíveis métodos de recuperação avançada para reservatórios de gás retrógrado. Os resultados exibiram tendências similares àquelas observadas em experimentos de escoamento em testemunhos, e condições ótimas para melhoramento do escoamento foram identificadas. / [en] Liquid dropout and accumulation in gas-condensate reservoirs, especially in the near wellbore region, hinder gas flow and affect negatively the produced fluid composition. Yet, condensate banking forecasting is commonly inaccurate, as experiments seldom reproduce reservoir extreme conditions and complex fluid composition, while most pore-scale models oversimplify the physical phenomena associated with phase transitions between gas and condensate. To address this gap, a fully implicit isothermal compositional pore-network model for gas and condensate flow is presented. The proposed pore-networks consist of 3D structures of constricted circular capillaries. Condensation modes and flow patterns are attributed to the capillaries according to the medium s wettability, local saturations and influence of viscous and capillary forces. At the network nodes, pressure and molar contents are determined via the coupled solution of molar balance and volume consistency equations. Concomitantly, a PT-flash based on the Peng-Robinson equation of state is performed for each node, updating the local phases saturations and compositions. For the proposed model validation, flow analyses were carried out based on coreflooding experiments reported in the literature, with matching fluid composition and flow conditions, and approximated pore-space geometry. Predicted and measured relative permeability curves showed good quantitative agreement, for two values of interfacial tension and three values of gas flow velocity. Following the validation, the model was used to evaluate wettability alteration and gas injection as prospect enhanced recovery methods for gas-condensate reservoirs. Results exhibited similar trends observed in coreflooding experiments and conditions for optimal flow enhancement were identified.

No / An analytic framework is devised, based on the principle of maximum entropy (ME), for the performance modelling and evaluation of a wireless GSM/GPRS cell supporting bursty multiple class traffic of voice calls and data packets under complete partitioning (CPS), partial sharing (PSS) and aggregate sharing (ASS) traffic handling schemes. Three distinct open queueing network models (QNMS) under CPS, PSS and ASS, respectively, are described, subject to external compound Poisson traffic processes and generalised exponential (GE) transmission times under a repetitive service blocking mechanism and a complete buffer sharing management rule. Each QNM generally consists of three building block stations, namely a loss system with GSM/GPRS traffic and a system of access and transfer finite capacity queues in tandem dealing with GPRS traffic under head-of-line and discriminatory processor sharing scheduling disciplines, respectively. The analytic methodology is illustrated by focusing on the performance study of the GE-type tandem queueing system for GPRS under a CPS. An ME product-form approximation is characterised leading into a decomposition of the tandem system into individual queues and closed-form ME expressions for state and blocking probabilities are presented. Typical numerical examples are included to validate the ME solutions against simulation and study the effect of external GPRS bursty traffic upon the performance of the cell. Moreover, an overview of recent extensions of the work towards the analysis of a GE-type multiple server finite capacity queue with preemptive resume priorities and its implications towards the performance modelling and evaluation of GSM/GPRS cells with PSS and ASS are included. / ,

This thesis focuses on the link between financial system and sovereign debt crises through sovereign support to banks on one hand and banks' exposures to weak sovereigns on the other. After illustrating the main relationships on the recent financial crisis, we construct an agent-based network model of an artificial financial system allowing us to analyse the effects of state support on systemic stability and the feedback loops of risk transfer back into the financial system. First, the model is tested with various parameter settings in Monte Carlo simulations and second, it is calibrated to the real world data using a unique dataset put together from various sources. Our analyses yield the following key results: Firstly, in the short term, all the support measures improve the systemic stability. Secondly, in the longer run, the effects of state support depend on several parameters but still there are settings in which it significantly mitigates the systemic crisis. Finally, there are differences among the effects of the different types of support measures.

In recent years, deep neural network models have shown to outperform many state of the art algorithms. The reason for this is, unsupervised pretraining with multi-layered deep neural networks have shown to learn better features, which further improves many supervised tasks. These models not only automate the feature extraction process but also provide with robust features for various machine learning tasks. But the unsupervised pretraining and feature extraction using multi-layered networks are restricted only to the input features and not to the output. The performance of many supervised learning algorithms (or models) depends on how well the output dependencies are handled by these algorithms [Dembczy´nski et al., 2012]. Adapting the standard neural networks to handle these output dependencies for any specific type of problem has been an active area of research [Zhang and Zhou, 2006, Ribeiro et al., 2012]. On the other hand, inference into multimodal data is considered as a difficult problem in machine learning and recently ‘deep multimodal neural networks’ have shown significant results [Ngiam et al., 2011, Srivastava and Salakhutdinov, 2012]. Several problems like classification with complete or missing modality data, generating the missing modality etc., are shown to perform very well with these models. In this work, we consider three nontrivial supervised learning tasks (i) multi-class classification (MCC), (ii) multi-label classification (MLC) and (iii) label ranking (LR), mentioned in the order of increasing complexity of the output. While multi-class classification deals with predicting one class for every instance, multi-label classification deals with predicting more than one classes for every instance and label ranking deals with assigning a rank to each label for every instance. All the work in this field is associated around formulating new error functions that can force network to identify the output dependencies. Aim of our work is to adapt neural network to implicitly handle the feature extraction (dependencies) for output in the network structure, removing the need of hand crafted error functions. We show that the multimodal deep architectures can be adapted for these type of problems (or data) by considering labels as one of the modalities. This also brings unsupervised pretraining to the output along with the input. We show that these models can not only outperform standard deep neural networks, but also outperform standard adaptations of neural networks for individual domains under various metrics over several data sets considered by us. We can observe that the performance of our models over other models improves even more as the complexity of the output/ problem increases.