Rozprawy doktorskie na temat „NON-LINEAR BLOCKS”

L'instabilité des blocs situés à la surface des excavations souterraines est un problème courant dans les milieux rocheux fracturés. Comme les méthodes exactes prenant en compte tous les blocs et leurs interactions sont très lourdes, l'approche des Blocs Isolés est souvent adoptée. Elle consiste à étudier chaque bloc en considérant qu'il est rigide et que le reste de la masse rocheuse est rigide et fixe. Néanmoins, aucune des méthodes existantes adoptant cette approche ne prend en compte de façon rigoureuse les contraintes initiales et le comportement des joints. Dans cette thèse, on développe une nouvelle méthode qui apporte des améliorations importantes aux méthodes conventionnelles de Blocs Isolés. Connaissant les contraintes initiales, on rend compte du processus d'excavation par le déchargement de la face libre du bloc. Les efforts sur les faces du bloc en contact avec la roche sont alors modifiés en respectant l'équilibre des forces et des moments, la loi de comportement des joints et le mouvement de corps solide du bloc. On aboutit ainsi à un système linéaire où les seules inconnues sont les vecteurs translation et rotation du bloc. Deux modèles sont proposés : le premier considère un comportement linéaire élastique des joints et, par conséquent, la stabilité est évaluée a posteriori. Le deuxième modèle, plus pertinent, considère un comportement hyperbolique des joints dans la direction normale et élastoplastique dans la direction tangentielle avec prise en compte de la dilatance. La méthode numérique adoptée pour la résolution du problème non linéaire est une intégration explicite dans le temps cinématique avec des pas de déchargement constants. La technique d'intégration surfacique utilisée permet d'étudier toute forme géométrique de bloc. La méthode proposée a été validée puis comparée aux méthodes conventionnelles. Des études paramétriques ont montré l'influence des contraintes initiales et des propriétés mécaniques des joints sur la stabilité. Le soutènement a été aussi intégré dans le code développé. Finalement, la nouvelle méthode a été appliquée pour l'étude d'un assemblage de blocs autour d'une excavation souterraine et comparée à un modèle prenant en compte tous les blocs via la méthode des Éléments Distincts. Elle a été aussi utilisée pour restituer un cas réel de chute de blocs.

Made available in DSpace on 2014-07-29T15:08:18Z (GMT). No. of bitstreams: 1 Previsao de Vazoes Naturais Diarias.pdf: 3835466 bytes, checksum: f927e5c8c3a89c73430512243b55c36c (MD5) Previous issue date: 2012-09-05
The forecast of natural flows to hydroelectric plant reservoirs is an essential input to the planning and programming of the SIN´s operation. Various computer models are used to determine these forecasts, including physical models, statistical models and the ones developed with the RNA´s techniques. Currently, the ONS performs daily forecasts of natural flows to the UHE Tucuruí based on the univariate stochastic model named PREVIVAZH, developed by Electric Energy Research Center - Eletrobras CEPEL. Throughout the last decade, several papers have shown evolution in the application of neural networks methodology in many areas, specially in the prediction of flows on a daily, weekly and monthly basis. The goal of this dissertation is to present and calibrate a model of natural flow forecast using the RNA´s methodology, more specifically the NSRBN (Non-Linear Sigmoidal Regression Blocks Networks) (VALENCA; LUDERMIR, 2001), on a time lapse from 1 to 12 days forward to the Tucuruí Hydroelectric Plant, considering the hydrometric stations data located upstream from it s reservoir. In addition, a comparative analysis of results found throughout the calibrated neural network and the ones released by ONS is performed. The results show the advantage of the methodology of artificial neural networks on autoregressive models. The Mean Absolute Percentage Error - MAPE values obtained were, on average, 48 % lower than those released by the ONS.
A previsão de vazões naturais aos reservatórios das usinas hidrelétricas é insumo fundamental para o planejamento e operação do SIN. Diversos modelos são utilizados na determinação dessas previsões, entre os quais podem ser citados os modelos físicos, os estatísticos e aqueles baseados na técnica de Redes Neurais Artificiais. Atualmente, o ONS realiza as previsões diárias de vazões naturais para a Usina Hidrelétrica Tucuruí com base no modelo estocástico univariado denominado PREVIVAZH, desenvolvido pelo CEPEL. Ao longo da última década, muitos trabalhos têm mostrado a evolução da aplicação da metodologia de Redes Neurais Artificiais em diversas áreas e em particular na previsão de vazões naturais, para intervalos de tempo diários, semanais e mensais. O objetivo deste trabalho foi calibrar e avaliar um modelo de previsão de vazões naturais, utilizando a metodologia de RNA, mais especificamente as redes construtivas do tipo NSRBN(Non-Linear Sigmoidal Regression Blocks Networks) (VALENCA; LUDERMIR, 2001), no horizonte de 1 até 12 dias à frente, para a Usina Hidrelétrica Tucuruí, considerando as informações advindas de postos hidrométricos localizados à montante do seu reservatório. Adicionalmente, foi realizada uma análise comparativa dos resultados encontrados pela rede neural calibrada e aqueles obtidos e divulgados pelo ONS. Os resultados obtidos mostram a vantagem da metodologia de redes neurais artificiais sobre os modelos auto-regressivos. Os valores do Erro Percentual Médio Absoluto - MAPE foram, em média, 48% inferiores aos divulgados pelo ONS.

Cette thèse porte sur l’analyse de graphes dynamiques, définis en temps discret ou continu. Nous introduisons une nouvelle extension dynamique du modèle a blocs stochastiques (SBM), appelée dSBM, qui utilise des processus de Poisson non homogènes pour modéliser les interactions parmi les paires de nœuds d’un graphe dynamique. Les fonctions d’intensité des processus ne dépendent que des classes des nœuds comme dans SBM. De plus, ces fonctions d’intensité ont des propriétés de régularité sur des intervalles temporels qui sont à estimer, et à l’intérieur desquels les processus de Poisson redeviennent homogènes. Un récent algorithme d’estimation pour SBM, qui repose sur la maximisation d’un critère exact (ICL exacte) est ici adopté pour estimer les paramètres de dSBM et sélectionner simultanément le modèle optimal. Ensuite, un algorithme exact pour la détection de rupture dans les séries temporelles, la méthode «pruned exact linear time» (PELT), est étendu pour faire de la détection de rupture dans des données de graphe dynamique selon le modèle dSBM. Enfin, le modèle dSBM est étendu ultérieurement pour faire de l’analyse de réseau textuel dynamique. Les réseaux sociaux sont un exemple de réseaux textuels: les acteurs s’échangent des documents (posts, tweets, etc.) dont le contenu textuel peut être utilisé pour faire de la classification et détecter la structure temporelle du graphe dynamique. Le modèle que nous introduisons est appelé «dynamic stochastic topic block model» (dSTBM)
This thesis focuses on the statistical analysis of dynamic graphs, both defined in discrete or continuous time. We introduce a new extension of the stochastic block model (SBM) for dynamic graphs. The proposed approach, called dSBM, adopts non homogeneous Poisson processes to model the interaction times between pairs of nodes in dynamic graphs, either in discrete or continuous time. The intensity functions of the processes only depend on the node clusters, in a block modelling perspective. Moreover, all the intensity functions share some regularity properties on hidden time intervals that need to be estimated. A recent estimation algorithm for SBM, based on the greedy maximization of an exact criterion (exact ICL) is adopted for inference and model selection in dSBM. Moreover, an exact algorithm for change point detection in time series, the "pruned exact linear time" (PELT) method is extended to deal with dynamic graph data modelled via dSBM. The approach we propose can be used for change point analysis in graph data. Finally, a further extension of dSBM is developed to analyse dynamic net- works with textual edges (like social networks, for instance). In this context, the graph edges are associated with documents exchanged between the corresponding vertices. The textual content of the documents can provide additional information about the dynamic graph topological structure. The new model we propose is called "dynamic stochastic topic block model" (dSTBM).Graphs are mathematical structures very suitable to model interactions between objects or actors of interest. Several real networks such as communication networks, financial transaction networks, mobile telephone networks and social networks (Facebook, Linkedin, etc.) can be modelled via graphs. When observing a network, the time variable comes into play in two different ways: we can study the time dates at which the interactions occur and/or the interaction time spans. This thesis only focuses on the first time dimension and each interaction is assumed to be instantaneous, for simplicity. Hence, the network evolution is given by the interaction time dates only. In this framework, graphs can be used in two different ways to model networks. Discrete time […] Continuous time […]. In this thesis both these perspectives are adopted, alternatively. We consider new unsupervised methods to cluster the vertices of a graph into groups of homogeneous connection profiles. In this manuscript, the node groups are assumed to be time invariant to avoid possible identifiability issues. Moreover, the approaches that we propose aim to detect structural changes in the way the node clusters interact with each other. The building block of this thesis is the stochastic block model (SBM), a probabilistic approach initially used in social sciences. The standard SBM assumes that the nodes of a graph belong to hidden (disjoint) clusters and that the probability of observing an edge between two nodes only depends on their clusters. Since no further assumption is made on the connection probabilities, SBM is a very flexible model able to detect different network topologies (hubs, stars, communities, etc.)

One of the facets of the mobile or wireless environment is that errors quite often occur in bursts. Thus, strong codes are required to provide protection against such errors. This in turn motivates the employment of decoding algorithms which are simple to implement, yet are still able to attempt to take the dependence or memory of the channel model into account in order to give optimal decoding estimates. Furthermore, such algorithms should be able to be applied for a variety of channel models and signalling alphabets. The research presented within this thesis describes a number of algorithms which can be used with linear block codes. Given the received word, these algorithms determine the symbol which was most likely transmitted, on a symbol-by-symbol basis. Due to their relative simplicity, a collection of algorithms for memoryless channels is reported first. This is done to establish the general style and principles of the overall collection. The concept of matrix diagonalisation may or may not be applied, resulting in two different types of procedure. Ultimately, it is shown that the choice between them should be motivated by whether storage space or computational complexity has the higher priority. As with all other procedures explained herein, the derivation is first performed for a binary signalling alphabet and then extended to fields of prime order. These procedures form the paradigm for algorithms used in conjunction with finite state channel models, where errors generally occur in bursts. In such cases, the necessary information is stored in matrices rather than as scalars. Finally, by analogy with the weight polynomials of a code and its dual as characterised by the MacWilliams identities, new procedures are developed for particular types of Gilbert-Elliott channel models. Here, the calculations are derived from three parameters which profile the occurrence of errors in those models. The decoding is then carried out using polynomial evaluation rather than matrix multiplication. Complementing this theory are several examples detailing the steps required to perform the decoding, as well as a collection of simulation results demonstrating the practical value of these algorithms.

This master's thesis describes the design of effective working artificial neural network in FPGA Virtex-5 series with the maximum use of the possibility of parallelization. The theoretical part contains basic information on artificial neural networks, FPGA and VHDL. The practical part describes the used format of the variables, creating non-linear function, the principle of calculation the single layers, or the possibility of parameter settings generated artificial neural networks.

Spectral properties of linear operators and operator functions can be used to analyze models in nature. When dispersion and damping are taken into account, the dependence of the spectral parameter is in general non-linear and the operators are not selfadjoint. In this thesis non-selfadjoint operator functions are studied and several methods for obtaining properties of unbounded non-selfadjoint operator functions are presented. Equivalence is used to characterize operator functions since two equivalent operators share many significant characteristics such as the spectrum and closeness. Methods of linearization and other types of equivalences are presented for a class of unbounded operator matrix functions. To study properties of the spectrum for non-selfadjoint operator functions, the numerical range is a powerful tool. The thesis introduces an optimal enclosure of the numerical range of a class of unbounded operator functions. The new enclosure can be computed explicitly, and it is investigated in detail. Many properties of the numerical range such as the number of components can be deduced from the enclosure. Furthermore, it is utilized to prove the existence of an infinite number of eigenvalues accumulating to specific points in the complex plane. Among the results are proofs of accumulation of eigenvalues to the singularities of a class of unbounded rational operator functions. The enclosure of the numerical range is also used to find optimal and computable estimates of the norm of resolvent and a corresponding enclosure of the ε-pseudospectrum.

Lors des dernières décennies, la Séparation Aveugle de Sources (BSS) est devenue un outil de premier plan pour le traitement de données multi-valuées. L’objectif de ce doctorat est cependant d’étudier les cas grande échelle, pour lesquels la plupart des algorithmes classiques obtiennent des performances dégradées. Ce document s’articule en quatre parties, traitant chacune un aspect du problème: i) l’introduction d’algorithmes robustes de BSS parcimonieuse ne nécessitant qu’un seul lancement (malgré un choix d’hyper-paramètres délicat) et fortement étayés mathématiquement; ii) la proposition d’une méthode permettant de maintenir une haute qualité de séparation malgré un nombre de sources important: iii) la modification d’un algorithme classique de BSS parcimonieuse pour l’application sur des données de grandes tailles; et iv) une extension au problème de BSS parcimonieuse non-linéaire. Les méthodes proposées ont été amplement testées, tant sur données simulées que réalistes, pour démontrer leur qualité. Des interprétations détaillées des résultats sont proposées
During the last decades, Blind Source Separation (BSS) has become a key analysis tool to study multi-valued data. The objective of this thesis is however to focus on large-scale settings, for which most classical algorithms fail. More specifically, it is subdivided into four sub-problems taking their roots around the large-scale sparse BSS issue: i) introduce a mathematically sound robust sparse BSS algorithm which does not require any relaunch (despite a difficult hyper-parameter choice); ii) introduce a method being able to maintain high quality separations even when a large-number of sources needs to be estimated; iii) make a classical sparse BSS algorithm scalable to large-scale datasets; and iv) an extension to the non-linear sparse BSS problem. The methods we propose are extensively tested on both simulated and realistic experiments to demonstrate their quality. In-depth interpretations of the results are proposed

This thesis is divided into three main parts. In the first, we discuss that, although permutation tests can provide exact control of false positives under the reasonable assumption of exchangeability, there are common examples in which global exchangeability does not hold, such as in experiments with repeated measurements or tests in which subjects are related to each other. To allow permutation inference in such cases, we propose an extension of the well known concept of exchangeability blocks, allowing these to be nested in a hierarchical, multi-level definition. This definition allows permutations that retain the original joint distribution unaltered, thus preserving exchangeability. The null hypothesis is tested using only a subset of all otherwise possible permutations. We do not need to explicitly model the degree of dependence between observations; rather the use of such permutation scheme leaves any dependence intact. The strategy is compatible with heteroscedasticity and can be used with permutations, sign flippings, or both combined. In the second part, we exploit properties of test statistics to obtain accelerations irrespective of generic software or hardware improvements. We compare six different approaches using synthetic and real data, assessing the methods in terms of their error rates, power, agreement with a reference result, and the risk of taking a different decision regarding the rejection of the null hypotheses (known as the resampling risk). In the third part, we investigate and compare the different methods for assessment of cortical volume and area from magnetic resonance images using surface-based methods. Using data from young adults born with very low birth weight and coetaneous controls, we show that instead of volume, the permutation-based non-parametric combination (NPC) of thickness and area is a more sensitive option for studying joint effects on these two quantities, giving equal weight to variation in both, and allowing a better characterisation of biological processes that can affect brain morphology.

Non-Linear signal processing circuits are used in various streams of engineering such as telecommunication, control system, digital design, biomedical, information security etc. Various types of signal processing methods such as analog or digital are used according to applications and requirements. CMOS non-linear signal processing circuits consume less power whereas bipolar non-linear circuits provide faster response and also provide wide operating range. In this dissertation linear term generator and higher order non-linear blocks such as squarer generator, cubic generator, power four generator and power five generator blocks have been implemented. The functions like sine, cosine, exponential, logarithmic, tangent hyperbolic inverse functions and the other two functions viz. Gaussian and sigmoid, which are extensively used in artificial neural network are also implemented in this dissertation. These functions are implemented in current mode since this mode is found to be more advantageous as compared to voltage mode implementation. The analog implementation of elliptic curve using these non-linear blocks is also presented in this dissertation which can be further used in Cryptography.

In structural mechanics there are several occasions where a linearized formulation of the original nonlinear problem reduces considerably the computational effort for the response analysis. In a broader sense, a linearized formulation can be viewed as a first-order expansion of the dynamic equilibrium of the system about a `static¿ configuration; yet caution should be exercised when identifying the `correct¿ static configuration. This paper uses as a case study the rocking response of a rigid block stepping on viscoelastic supports, whose non-linear dynamics is the subject of the companion paper, and elaborates on the challenge of identifying the most appropriate static configuration around which a first-order expansion will produce the most dependable results in each regime of motion. For the regime when the heel of the block separates, a revised set of linearized equations is presented, which is an improvement to the unconservative equations published previously in the literature. The associated eigenvalues demonstrate that the characteristics of the foundation do not affect the rocking motion of the block once the heel separates.

A "rheinforce'' composite consists of a laminar sheet of compacted micro-agglomerated cork engraved by laser with a network of microchannels and filled with a concentrated aqueous solution of shear-thickening fluid. Using it for personal protection equipment, as an energy-absorbing material layer, is one of its important applications. An accurate numerical modeling of such material is not currently available and would be a valuable tool during design and manufacturing processes. A virtual dynamic drop tower test can be used to shed lights on the dynamics of the energy dissipation of the "rheinforce" composite. From the continuum mechanics point of view, it can be translated into an initial-boundary value problem with enclosed-Fluid-Structure Interaction, contact boundary and rigid body dynamics. The open-source OpenFOAM toolbox called Solids4foam (S4F) offers a very attractive starting point to build a solver for that initial-boundary value problem. Not only because no commercial software available (as far as the author is aware) is as customizable as needed, but also because S4F offers a strongly-coupled Fluid-Structure Interaction with fully customizable and replaceable solid and fluid solvers. Furthermore, these solvers are constructed based on only one numerical framework, the Finite Volume Method. Unfortunately, numerical simulations revealed that the current finite volume methodologies for solid mechanics implemented in S4F, i.e. Segregated (SEG) and the recently developed Block-Coupled (BC), cannot simulate the solid part of the "rheinforce'' composite under the finite strain regime. Moreover, the SEG method supports general materials (in theory any material law can be used), but it can be very slow. The BC method is fast, but only Hooke's law (an infinitesimal strain model) is supported. The principal aim of the thesis is to present the development of a new BC methodology that preserves, in a great extent, the fast convergence of the original BC and material generality (only requiring the elasticity tensor to have right-minor symmetry) of SEG. This work also demonstrates that the Field Operation and Manipulation (FOAM) concept should be implemented in a high-level programming environment to fill an important "prototyping gap'' left untouched by OpenFOAM and S4F, that is, between concept development and concept implementation in a high-performance programming language.

碩士
國立中興大學
通訊工程研究所
105
In this thesis, we propose combination of a modified PTS algorithm by partitioning an OFDM block into non-disjoint OFDM sub-block with linear block codes as phase factors. Through the concept of Gray code, the linear block codes are grown in parallel and serial, compare conventional PTS with our algorithms, our algorithms which are low computational complexity for searching phase factors, and simulation results show that the modified PTS with a non-disjoint partition achieves an obvious improvement of PAPR reduction. In addition, the side information provide extra error-correction property.

With the volume of data generated in today's internet-of-things, learning algorithms to extract and understand the underlying relations between the various attributes of data have gained momentum. This thesis is focused on learning algorithms to extract meaningful relations from the data using both unsupervised and supervised learning algorithms. Vector quantization techniques are popularly used for applications in contextual data clustering, data visualization and high-dimensional data exploration. Existing vector quantization techniques, such as, the K-means and its variants and those derived from the self-organizing maps consider the input data vector as a whole without prioritizing over individual coordinates. Motivated by applications requiring priorities over data coordinates, we develop a theory for clustering data with different priorities over the coordinates called the data-dependent priority-based soft vector quantization. Based on the input data distribution, the priorities over the data coordinates are learnt by estimating the marginal distributions over each coordinate. The number of neurons approximating each coordinate based on the priority are determined through a reinforcement learning algorithm. Analysis on the convergence of the proposed algorithm and the probability of misclassification are presented along with simulation results on various data sets. Self-organizing maps (SOM) are popularly used for applications in learning features, vector quantization, and recalling spatial input patterns. The adaptation rule in SOMs is based on the Euclidean distance between the input vector and the neuronal weight vector along with a neighborhood function that brings in topological arrangement of the neurons in the output space. It is capable of learning the spatial correlations among the data but fails to capture temporal correlations present in a sequence of inputs. We formulate a potential function based on a spatio-temporal metric and create hierarchical vector quantization feature maps by embedding memory structures similar to long short-term memories across the feature maps to learn the spatio-temporal correlations in the data across clusters. Error correction codes such as low density parity check codes are popularly used to enhance the performance of digital communication systems. The current decoding framework relies on exchanging beliefs over a Tanner graph, which the encoder and decoder are aware of. However, this information may not be available readily, for example, in covert communication. The main idea is to build a neural network to learn the encoder mappings in the absence of knowledge of the Tanner graph. We propose a scheme to learn the mappings using the back-propagation algorithm. We investigate into the choice of different cost functions and the number of hidden neurons for learning the encoding function. The proposed scheme is capable of learning the parity check equations over a binary field towards identifying the validity of a codeword. Simulation results over synthetic data show that our algorithm is indeed capable of learning the encoder mappings. We also propose an approach to identify noisy codes using uncertainty estimation and to decode them using autoencoders. In the next work, we consider the convolutional neural networks which are widely used in natural language processing, video analysis, and image recognition. However, the popularly used max-pooling layer discards most of the data, which is a drawback in applications, such as, prediction of video frames. We propose an adaptive prediction and classification network based on a data-dependent pooling architecture. We formulate a combined cost function for minimizing the prediction and classification errors. We also detect the presence of an unseen class during testing for digit prediction in videos.

A high-order central essentially non-oscillatory (CENO) finite-volume scheme in combination with a block-based adaptive mesh refinement (AMR) algorithm is proposed for solution of hyperbolic and elliptic systems of conservation laws on body- fitted multi-block mesh. The spatial discretization of the hyperbolic (inviscid) terms is based on a hybrid solution reconstruction procedure that combines an unlimited high-order k-exact least-squares reconstruction technique following from a fixed central stencil with a monotonicity preserving limited piecewise linear reconstruction algorithm. The limited reconstruction is applied to computational cells with under-resolved solution content and the unlimited k-exact reconstruction procedure is used for cells in which the solution is fully resolved. Switching in the hybrid procedure is determined by a solution smoothness indicator. The hybrid approach avoids the complexity associated with other ENO schemes that require reconstruction on multiple stencils and therefore, would seem very well suited for extension to unstructured meshes. The high-order elliptic (viscous) fluxes are computed based on a k-order accurate average gradient derived from a (k+1)-order accurate reconstruction. A novel h-refinement criterion based on the solution smoothness indicator is used to direct the steady and unsteady refinement of the AMR mesh. The predictive capabilities of the proposed high-order AMR scheme are demonstrated for the Euler and Navier-Stokes equations governing two-dimensional compressible gaseous flows as well as for advection-diffusion problems characterized by the full range of Peclet numbers, Pe. The ability of the scheme to accurately represent solutions with smooth extrema and yet robustly handle under-resolved and/or non-smooth solution content (i.e., shocks and other discontinuities) is shown for a range of problems. Moreover, the ability to perform mesh refinement in regions of smooth but under-resolved and/or non-smooth solution content to achieve the desired resolution is also demonstrated.