[pt] Materiais compósitos são a base de muitos produtos, devido à sua capacidade de aperfeiçoar certas propriedades. Recentemente, a utilização de nanocargas na fabricação de compósitos vem sendo amplamente estudada, pois a partir de concentrações baixas de nanocargas, as propriedades começam a melhorar, possibilitando a criação de materiais leves e com uma grande gama de propriedades. Uma das propriedades mecânicas mais estudadas é o módulo de Young, que mensura a rigidez de um material. Alguns dos modelos existentes para essa propriedade em nanocompósitos pecam na precisão ou são limitados em função da fração máxima de nanopartículas admissível no modelo. Outros se adequam apenas a uma determinada combina ção de matriz/carga preestabelecida. O objetivo deste trabalho é utilizar Redes Neurais Artificiais como um aproximador capaz de modelar tal propriedade para diversas matrizes/cargas, levando em consideração suas características, sem perder a precisão. A validação do aproximador é realizada comparando o resultado com outros modelos propostos na literatura. Uma vez validada, utiliza-se Algoritmos Genéticos em conjunto com tal rede para definir qual seria a configuração ideal para três casos de estudo: um que maximize o valor do módulo de Young, outro que maximize o módulo relativo e um terceiro que maximize o módulo relativo e minimize a quantidade de carga utilizada, diminuindo os custos de projeto. As técnicas de Inteligência Computacional empregadas na modelagem e síntese de materiais nanoestruturados se mostraram boas ferramentas, uma vez que geraram uma boa aproximação dos dados utilizados com erros inferiores a 5%, além de possibilitarem a determinação dos parâmetros de síntese de um material com o módulo de Young desejado. / [en] Composite materials became very popular due to its improvements on certain properties achieved from the mixture of two different components. Recently, the use of nanofillers in the manufacture of composites has been widely studied due to the improvement of properties at low concentrations of nanofillers, enabling the creation of lightweight materials. Some of the existing models for the Young modulus of the nanocomposites have low accuracy or are limited in terms of the maximum filler fraction possible. Others are appropriate only for a given combination of matrix and filler. The objective of this work is to use Artificial Neural Networks as a function approximation method capable of modeling such property for various matrix/nanofillers, taking into account their characteristics, without losing accuracy. The validation of this approximator is performed comparing its results with other models proposed in the literature. Once validated, a Genetic Algorithm is used with the Neural Network to define which would be the ideal setting for three case studies: one that maximizes the value of composite’s Young’s modulus, other that maximizes the relative modulus and a third one that maximizes the relative modulus and minimizes the amount of load used, reducing the cost of project. Computational Intelligence techniques employed on the modeling and synthesis of nanostructured materials proved to be adequate tools, since it generated a good approximation of the data with errors lower than 5%, and determined the material’s parameters for synthesis with the desired Young’s modulus.

Survival analysis today is widely implemented in the fields of medical and biological sciences, social sciences, econometrics, and engineering. The basic principle behind the survival analysis implies to a statistical approach designed to take into account the amount of time utilized for a study period, or the study of time between entry into observation and a subsequent event. The event of interest pertains to death and the analysis consists of following the subject until death. Events or outcomes are defined by a transition from one discrete state to another at an instantaneous moment in time. In the recent years, research in the area of survival analysis has increased greatly because of its large usage in areas related to bio sciences and the pharmaceutical studies. After identifying the probability density function that best characterizes the tumors and survival times of breast cancer women, one purpose of this research is to compare the efficiency between competing estimators of the survival function. Our study includes evaluation of parametric, semi-parametric and nonparametric analysis of probability survival models. Artificial Neural Networks (ANNs), recently applied to a number of clinical, business, forecasting, time series prediction, and other applications, are computational systems consisting of artificial neurons called nodes arranged in different layers with interconnecting links. The main interest in neural networks comes from their ability to approximate complex nonlinear functions. Among the available wide range of neural networks, most research is concentrated around feed forward neural networks called Multi-layer perceptrons (MLPs). One of the important components of an artificial neural network (ANN) is the activation function. This work discusses properties of activation functions in multilayer neural networks applied to breast cancer stage classification. There are a number of common activation functions in use with ANNs. The main objective in this work is to compare and analyze the performance of MLPs which has back-propagation algorithm using various activation functions for the neurons of hidden and output layers to evaluate their performance on the stage classification of breast cancer data. Survival analysis can be considered a classification problem in which the application of machine-learning methods is appropriate. By establishing meaningful intervals of time according to a particular situation, survival analysis can easily be seen as a classification problem. Survival analysis methods deals with waiting time, i.e. time till occurrence of an event. Commonly used method to classify this sort of data is logistic regression. Sometimes, the underlying assumptions of the model are not true. In model building, choosing an appropriate model depends on complexity and the characteristics of the data that affect the appropriateness of the model. Two such strategies, which are used nowadays frequently, are artificial neural network (ANN) and decision trees (DT), which needs a minimal assumption. DT and ANNs are widely used methodological tools based on nonlinear models. They provide a better prediction and classification results than the traditional methodologies such as logistic regression. This study aimed to compare predictions of the ANN, DT and logistic models by breast cancer survival. In this work our goal is to design models using both artificial neural networks and logistic regression that can precisely predict the output (survival) of breast cancer patients. Finally we compare the performances of these models using receiver operating characteristic (ROC) analysis.

The topic of this thesis is the prediction of production of ash as a side product in the coal power plant Ledvice. The goal of the first two chapters was to present the development of consumption of electricity and the resources of electric energy in Europe, show the trends in energy industry at the beginning of the 21st century, introduce the energetic company ČEZ, summarize the issue of side products in power plants and present the new block of the coal power plant Ledvice. In the theory section, the methods of linear programming, simulation models, linear regression model and artificial neural networks were presented. These methods were used in the application section to create an optimization model for the processing of the side products of burning coal, to create a simulation of the production of side products and to use the results to optimize the processing of the side products of burning coal and to create a model to predict the production of bottom ash to coal burned ratio based on the characteristics of the coal burned moisture, ash and lower heating value using artificial neural networks.

Road damage identification is still largely based on visual inspection methods and profilometer data. Visual inspection methods heavily rely on expert knowledge which is often very subjective. They also result in traffic flow interference due to the need for redirection of traffic to alternative routes during inspection. In addition to this, accurate high-speed profilometers, such as scanning vehicles, are extremely expensive often requiring strong economic justifications for their acquisition. The low-cost profilometers are very slow, typically operating at or less than walking speeds, causing their use to be labour-intensive if applied to large networks.This study aims at developing a road damage identification methodology for both paved and unpaved roads based on modelling the road-vehicle interaction system with an artificial neural network. The artificial neural network is created and trained with vehicle acceleration data as inputs and road profiles as targets. Then the trained neural network is consequently used for reconstruction of road profiles upon simulating it with vertical vehicle accelerations. The simulation process is very fast and can often be completed in a very short time thus making it possible to implement the methodology in real-time. Three case studies were used to demonstrate the feasibility of the methodology and the results on field tests carried out on mine vehicles with crudely measured road profiles showed a majority of the tested roads were reconstructed to within a fitting accuracy of less than 40% at a correlation level of greater than 55% which in this study was found to be practically acceptable considering the limitations imposed by the sizes of the haul trucks and their tyres as well as the quality of the road profiles and lack of control in the vehicle operation. / Thesis (PhD)--University of Pretoria, 2015. / Mechanical and Aeronautical Engineering / Unrestricted

Orientador: Marcos Eduardo Ribeiro do Valle Mesquita / Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Matemática Estatística e Computação Científica / Made available in DSpace on 2018-08-27T12:11:26Z (GMT). No. of bitstreams: 1 Souza_AlineCristinade_M.pdf: 2392301 bytes, checksum: 8b771e0b97302577daaf84624b0c4751 (MD5) Previous issue date: 2015 / Resumo: Memórias associativas são modelos matemáticos inspirados pela capacidade do cérebro humano de armazenar e recordar informações por meio de associações. Tais modelos são projetados para armazenar um conjunto finito de associações chamado de conjunto das memórias fundamentais. Além disso, espera-se que a memória associativa seja capaz de recuperar uma informação armazenada mesmo a partir de um item incompleto ou ruidoso. As Memórias Associativas Recorrentes Exponenciais Fuzzy (REFAMs, acrônimo do termo em inglês Recurrent Exponential Fuzzy Associative Memories) podem ser efetivamente utilizadas para o armazenamento e recordação de uma família finita de conjuntos fuzzy. Em geral, uma REFAM define recursivamente uma sequência de conjuntos fuzzy obtidos usando médias ponderadas e exponenciais dos valores de medida de similaridade. Experimentos computacionais relacionados à recuperação de imagens em tons de cinza ruidosas mostraram que os novos modelos podem apresentar ótima capacidade absoluta de armazenamento bem como excelente tolerância a ruído / Abstract: Associative memories are mathematical models inspired by the human brain ability to store and recall information by means of associations. Such models are designed for the storage of a finite set of associations called the fundamental memories set. Furthermore, the associative memory is expected to be able to retrieve a stored information even from an incomplete or noisy item. The Recurrent Exponential Fuzzy Associative Memories (REFAMs) can be effectively used for storage and recall of a finite family of fuzzy sets. In general, a REFAM defines recursively a sequence of fuzzy sets obtained using weighted averages and exponentials of similarity measure values. Computational experiments concerning the retrieval of noisy gray-scale images revealed that the novel models may exhibit optimal absolute storage capacity as well as excellent noise tolerance / Mestrado / Matematica Aplicada / Mestra em Matemática Aplicada

Esta dissertação de mestrado tem por objetivo analisar os conjuntos-K, uma hierarquia de redes neurais biologicamente mais plausíveis, e aplicá-los ao problema de classificação de imagética motora através do eletroencefalograma (EEG). A imagética motora consiste no ato de processar um movimento motor da memória humana de longo tempo para a memória de curto prazo. A imagética motora deixa um rastro no sinal do EEG que torna possível a identificação e classificação dos diferentes movimentos motores. A tarefa de classificação de imagética motora através do EEG é reconhecida como complexa devido à não linearidade e quantidade de ruído da série temporal do EEG e da pequena quantidade de dados disponíveis para aprendizagem. Os conjuntos-K são um modelo conexionista que simula o comportamento dinâmico e caótico de populações de neurônios do cérebro e foram modelados com base em observações do sistema olfatório feitas por Walter Freeman. Os conjuntos-K já foram aplicados em diversos domínios de classificação diferentes, incluindo EEG, tendo demonstrado bons resultados. Devido às características da classificação de imagética motora, levantou-se a hipótese de que a aplicação dos conjuntos-K na tarefa pudesse prover bons resultados. Um simulador para os conjuntos-K foi construído para a realização dos experimentos. Não foi possível validar a hipótese levantada no trabalho, dado que os resultados dos experimentos realizados com conjuntos-K e imagética motora não apresentaram melhorias significativas para a tarefa nas comparações realizadas. / This dissertation aims to examine the K-sets, a hierarchy of biologically plausible neural networks, and apply them to the problem of motor imagery classification through electroencephalogram (EEG). Motor imagery is the act of processing a motor movement from long-term to short-term memory. Motor imagery leaves a trail in the EEG signal, which makes possible the identification and classification of different motor movements. Motor imagery classification is a complex problem due to non-linearity of the EEG time series, low signal-to-noise ratio, and the small amount of data typically available for learning. K-sets are a connectionist model that simulates the dynamic and chaotic behavior of populations of neurons in the brain, modeled based on observations of the olfactory system by Walter Freeman. K-sets have already been used in several different classification domains, including EEG, showing good results. Due to the characteristics of motor imagery classification, a hypothesis that the application of K-sets in the task could provide good results was raised. A simulator for K-sets was created for the experiments. Unfortunately, the hypothesis could not be validated, as the results of the conducted experiments with K-sets and motor imagery showed no significant improvements in comparison in the task performed.

A navegação autônoma é um dos problemas fundamentais da robótica móvel. Para um robô executar esta tarefa, é necessário determinar a região segura para a navegação. Este trabalho propõe um sistema de identificação de superfícies navegáveis baseado em visão computacional utilizando redes neurais artificiais. Mais especificamente, é realizado um estudo sobre a utilização de diferentes atributos de imagem, como descritores estatísticos e elementos de espaços de cores, para serem utilizados como entrada das redes neurais artificiais que tem como tarefa a identificação de superfícies navegáveis. O sistema desenvolvido utiliza resultados de classificação de múltiplas configurações de redes neurais artificiais, onde a principal diferença entre elas é o conjunto de atributos de imagem utilizados como entrada. Essa combinação de diversas classificações foi realizada visando maior robustez e melhor desempenho na identificação de vias em diferentes cenários / Autonomous navigation is a fundamental problem in mobile robotics. In order to perform this task, a robot must identify the areas where it can navigate safely. This dissertation proposes a navigable terrain identification system based on computer vision and neural networks. More specifically, it is presented a study of image attributes, such as statistical decriptors and elements of different color spaces, that are used as neural neworks inputs for the navigable surfaces identification. The system developed combines the classification results of multiple neural networks topologies with different image attributes. This combination of classification results allows for improved efficient and robustenes in different scenarios

Um desafio recorrente em processos produtivos é o desenvolvimento de sistemas de monitoração e diagnóstico. Esses sistemas ajudam na detecção de mudanças inesperadas e interrupções, prevenindo perdas e mitigando riscos. Redes Neurais Artificiais (RNA) têm sido largamente utilizadas na criação de sistemas de monitoração. Normalmente as RNA utilizadas para resolver este tipo de problema são criadas levando-se em conta apenas parâmetros como o número de entradas, saídas e quantidade de neurônios nas camadas escondidas. Assim, as redes resultantes geralmente possuem uma configuração onde há uma total conexão entre os neurônios de uma camada e os da camada seguinte, sem que haja melhorias em sua topologia. Este trabalho utiliza o algoritmo de Otimização por Colônia de Formigas (OCF) para criar redes neurais otimizadas. O algoritmo de busca OCF utiliza a técnica de retropropagação de erros para otimizar a topologia da rede neural sugerindo as melhores conexões entre os neurônios. A RNA resultante foi aplicada para monitorar variáveis do reator de pesquisas IEA-R1 do IPEN. Os resultados obtidos mostram que o algoritmo desenvolvido é capaz de melhorar o desempenho do modelo que estima o valor de variáveis do reator. Em testes com diferentes números de neurônios na camada escondida, utilizando como comparativos o erro quadrático médio, o erro absoluto médio e o coeficiente de correlação, o desempenho da RNA otimizada foi igual ou superior ao da tradicional. / A recurring challenge in production processes is the development of monitoring and diagnosis systems. Those systems help on detecting unexpected changes and interruptions, preventing losses and mitigating risks. Artificial Neural Networks (ANN) have been extensively used in creating monitoring systems. Usually the ANN used to solve this kind of problem are created by taking into account only parameters as the number of inputs, outputs, and number of neurons in the hidden layers. This way, the result networks are generally fully connected and have no improvements in its topology. This work uses an Ant Colony Optimization (ACO) algorithm to create a tuned neural networks. The ACO search algorithm uses Back Error Propagation (BP) to optimize the network topology by suggesting the best neuron connections. The outcome ANN was applied to monitoring the IEA-R1 research reactor at IPEN. The results show that the algorithm is able to improve the performance of the model which estimates the values of the reactor variables. In tests with different numbers of neurons in the hidden layer, using as comparison the mean squared error, the mean absolute error, and the correlation coefficient, the performance of the optimized ANN proved equal or better than the equivalent traditional neural networks.

Esta Tesis propone el uso de técnicas multidisciplinares como una alternativa viable a los procedimientos actuales de evaluación del calzado los cuales, normalmente, consumen muchos recursos humanos y técnicos. Estas técnicas son Realidad Virtual, Biomecánica Computacional y Redes Neuronales Artificiales. El marco de esta tesis es el análisis virtual del confort mecánico en el calzado, es decir, el análisis de las presiones de confort en el calzado y su principal objetivo es predecir las presiones ejercidas por el zapato sobre la superficie del pie al caminar mediante la simulación del contacto en esta interfaz. En particular, en esta tesis se ha desarrollado una aplicación software que usa el Método de los Elementos Finitos para simular la deformación del calzado. Se ha desarrollado un modelo preliminar que describe el comportamiento del corte del calzado, se ha implementado un proceso automático para el ajuste pie-zapato y se ha presentado una metodología para obtener una animación genérica del paso de cada individuo. Además, y con el fin de mejorar la aplicación desarrollada, se han propuesto nuevos modelos para simular el comportamiento del corte del calzado al caminar. Por otro lado, las Redes Neuronales Artificiales han sido aplicadas en esta tesis a la predicción de la fuerza ejercida por una esfera, que simulando un hueso, empuja a una muestra de material. Además, también han sido utilizadas para predecir las presiones ejercidas por el corte del calzado sobre la superficie del pie (presiones dorsales) en un paso completo. Las principales contribuciones de esta tesis son: el desarrollo de un innovador simulador que permitirá a los fabricantes de calzado realizar evaluaciones virtuales de las características de sus diseños sin tener que construir el prototipo real, y el desarrollo de una también innovadora herramienta que les permitirá predecir las presiones dorsales ejercidas por el calzado sobre la superficie del pie al caminar. / Rupérez Moreno, MJ. (2011). MULTIDISCIPLINARY TECHNIQUES FOR THE SIMULATION OF THE CONTACT BETWEEN THE FOOT AND THE SHOE UPPER IN GAIT: VIRTUAL REALITY, COMPUTATIONAL BIOMECHANICS, AND ARTIFICIAL NEURAL NETWORKS [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/11235 / Palancia

Diploma thesis deals with the creation of an automated trading system for foreign exchange market with the usage of artificial intelligence and elements of technical analysis. In the custom solution design a brokerage company for trading and backtesting is selected with the help of fuzzy logic. For selecting currency pairs for backtesting the strategy on is used method of clustering called self-organizing map. The particular ATS is created in MetaTrader4 platform with the usage of a programming language MQL4 and a FANN library for creating artificial neural networks.