Projeto otimizado de redes neurais artificiais para predição da rugosidade em processos de usinagem com a utilização da metodologia de projeto de experimentos

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)

Imageamento da porosidade através de perfis geofísicos de poço

O imageamento da porosidade é uma representação gráfica da distribuição lateral da porosidade da rocha, estimada a partir de dados de perfis geofísicos de poço. Apresenta-se aqui uma metodologia para produzir esta imagem geológica, totalmente independente da intervenção do intérprete, através de um algoritmo, dito, interpretativo baseado em dois tipos de redes neurais artificiais. A primeira parte do algoritmo baseia-se em uma rede neural com camada competitiva e é construído para realizar uma interpretação automática do clássico gráfico o P_b - Φ_N, produzindo um zoneamento do perfil e a estimativa da porosidade. A segunda parte baseia-se em uma rede neural com função de base radial, projetado para realizar uma integração espacial dos dados, a qual pode ser dividida em duas etapas. A primeira etapa refere-se à correlação de perfis de poço e a segunda à produção de uma estimativa da distribuição lateral da porosidade. Esta metodologia ajudará o intérprete na definição do modelo geológico do reservatório e, talvez o mais importante, o ajudará a desenvolver de um modo mais eficiente as estratégias para o desenvolvimento dos campos de óleo e gás. Os resultados ou as imagens da porosidade são bastante similares às seções geológicas convencionais, especialmente em um ambiente deposicional simples dominado por clásticos, onde um mapa de cores, escalonado em unidades de porosidade aparente para as argilas e efetiva para os arenitos, mostra a variação da porosidade e a disposição geométrica das camadas geológicas ao longo da seção. Esta metodologia é aplicada em dados reais da Formação Lagunillas, na Bacia do Lago Maracaibo, Venezuela.

Aplicação de redes NeuroFuzzy ao processamento de peças automotivas por meio de injeção de polímeros

The injection molding of automotive parts is a complex process due to the many non-linear and multivariable phenomena that occur simultaneously. Commercial software applications exist for modeling the parameters of polymer injection but can be prohibitively expensive. It is possible to identify these parameters analytically, but applying classical theories of transport phenomena requires accurate information about the injection machine, product geometry, and process parameters. However, neurofuzzy networks, which achieve a synergy by combining the learning capabilities of an artificial neural network with a fuzzy set's inference mechanism, have shown success in this field. The purpose of this paper was to use a multilayer perceptron artificial neural network and a radial basis function artificial neural network combined with fuzzy sets to produce an inference mechanism that could predict injection mold cycle times. The results confirmed neurofuzzy networks as an effective alternative to solving such problems.

Aplicações de redes neurais e neuro fuzzy em engenharia biomédica e agronomia

Pós-graduação em Engenharia Elétrica - FEIS

Optimum-Path Forest Applied for Breast Masses Classification

In Computer-Aided Diagnosis-based schemes in mammography analysis each module is interconnected, which directly affects the system operation as a whole. The identification of mammograms with and without masses is highly needed to reduce the false positive rates regarding the automatic selection of regions of interest for further image segmentation. This study aims to evaluate the performance of three techniques in classifying regions of interest as containing masses or without masses (without clinical findings), as well as the main contribution of this work is to introduce the Optimum-Path Forest (OPF) classifier in this context, which has never been done so far. Thus, we have compared OPF against with two sorts of neural networks in a private dataset composed by 120 images: Radial Basis Function and Multilayer Perceptron (MLP). Texture features have been used for such purpose, and the experiments have demonstrated that MLP networks have been slightly better than OPF, but the former is much faster, which can be a suitable tool for real-time recognition systems.

Automatic classification of prostate stromal tissue in histological images using Haralick descriptors and Local Binary Patterns

In this paper we presente a classification system that uses a combination of texture features from stromal regions: Haralick features and Local Binary Patterns (LBP) in wavelet domain. The system has five steps for classification of the tissues. First, the stromal regions were detected and extracted using segmentation techniques based on thresholding and RGB colour space. Second, the Wavelet decomposition was applied in the extracted regions to obtain the Wavelet coefficients. Third, the Haralick and LBP features were extracted from the coefficients. Fourth, relevant features were selected using the ANOVA statistical method. The classication (fifth step) was performed with Radial Basis Function (RBF) networks. The system was tested in 105 prostate images, which were divided into three groups of 35 images: normal, hyperplastic and cancerous. The system performance was evaluated using the area under the ROC curve and resulted in 0.98 for normal versus cancer, 0.95 for hyperplasia versus cancer and 0.96 for normal versus hyperplasia. Our results suggest that texture features can be used as discriminators for stromal tissues prostate images. Furthermore, the system was effective to classify prostate images, specially the hyperplastic class which is the most difficult type in diagnosis and prognosis.

Segurança em redes sem fio: estudo sobre o desenvolvimento de conjuntos de dados para comparação de IDS

Pós-graduação em Engenharia Elétrica - FEIS

Fault detection, diagnosis and active fault tolerant control for a satellite attitude control system

Modern control systems are becoming more and more complex and control algorithms more and more sophisticated. Consequently, Fault Detection and Diagnosis (FDD) and Fault Tolerant Control (FTC) have gained central importance over the past decades, due to the increasing requirements of availability, cost efficiency, reliability and operating safety. This thesis deals with the FDD and FTC problems in a spacecraft Attitude Determination and Control System (ADCS). Firstly, the detailed nonlinear models of the spacecraft attitude dynamics and kinematics are described, along with the dynamic models of the actuators and main external disturbance sources. The considered ADCS is composed of an array of four redundant reaction wheels. A set of sensors provides satellite angular velocity, attitude and flywheel spin rate information. Then, general overviews of the Fault Detection and Isolation (FDI), Fault Estimation (FE) and Fault Tolerant Control (FTC) problems are presented, and the design and implementation of a novel diagnosis system is described. The system consists of a FDI module composed of properly organized model-based residual filters, exploiting the available input and output information for the detection and localization of an occurred fault. A proper fault mapping procedure and the nonlinear geometric approach are exploited to design residual filters explicitly decoupled from the external aerodynamic disturbance and sensitive to specific sets of faults. The subsequent use of suitable adaptive FE algorithms, based on the exploitation of radial basis function neural networks, allows to obtain accurate fault estimations. Finally, this estimation is actively exploited in a FTC scheme to achieve a suitable fault accommodation and guarantee the desired control performances. A standard sliding mode controller is implemented for attitude stabilization and control. Several simulation results are given to highlight the performances of the overall designed system in case of different types of faults affecting the ADCS actuators and sensors.

Artificial analysis of molecular marker loci linked to tree resistance response by an artificial neural network

One of the biggest challenges that software developers face is to make an accurate estimate of the project effort. Radial basis function neural networks have been used to software effort estimation in this work using NASA dataset. This paper evaluates and compares radial basis function versus a regression model. The results show that radial basis function neural network have obtained less Mean Square Error than the regression method.

Modelización de la respuesta de elementos estructurales lineales de vidrio plano : vigas y pilares pretensados

El vidrio se trata de un material muy apreciado en la arquitectura debido a la transparencia, característica que pocos materiales tienen. Pero, también es un material frágil, con una rotura inmediata cuando alcanza su límite elástico, sin disponer de un período plástico, que advierta de su futura rotura y permita un margen de seguridad. Por ambas razones, el vidrio se ha utilizado en arquitectura como elemento de plementería o relleno, desde tiempos antiguos, pero no como elemento estructural o portante, pese a que es un material interesante para los arquitectos para ese uso, por su característica de transparencia, ya que conseguiría la desmaterialización visual de la estructura, logrando espacios más ligeros y livianos. En cambio, si se tienen en cuenta las propiedades mecánicas del material se puede comprobar que dispone de unas características apropiadas para su uso estructural, ya que su Módulo elástico es similar al del aluminio, elemento muy utilizado en la arquitectura principalmente en las fachadas desde los últimos años, y su resistencia a compresión es muy superior incluso al hormigón armado; aunque su principal problema es su resistencia a tracción que es muy inferior a su resistencia a compresión, lo que penaliza su resistencia a flexión. En la actualidad se empieza a utilizar el vidrio como elemento portante o estructural, pero debido a su peor resistencia a flexión, se utilizan con grandes dimensiones que, a pesar de su transparencia, tienen una gran presencia. Por ello, la presente investigación pretende conseguir una reducción de las secciones de estos elementos estructurales de vidrio. Entonces, para el desarrollo de la investigación es necesario responder a una serie de preguntas fundamentales, cuyas respuestas serán el cuerpo de la investigación: 1. ¿Cuál es la finalidad de la investigación? El objetivo de esta investigación es la optimización de elementos estructurales de vidrio para su utilización en arquitectura. 2. ¿Cómo se va a realizar esa optimización? ¿Qué sistemas se van a utilizar? El sistema para realizar la optimización será la pretensión de los elementos estructurales de vidrio 3. ¿Por qué se va a utilizar la precompresión? Porque el vidrio tiene un buen comportamiento a compresión y un mal comportamiento a tracción lo que penaliza su utilización a flexión. Por medio de la precompresión se puede incrementar esta resistencia a tracción, ya que los primeros esfuerzos reducirán la compresión inicial hasta comenzar a funcionar a tracción, y por tanto aumentará su capacidad de carga. 4. ¿Con qué medios se va a comprobar y justificar ese comportamiento? Mediante simulaciones informáticas con programas de elementos finitos. 5. ¿Por qué se utilizará este método? Porque es una herramienta que arroja ventajas sobre otros métodos como los experimentales, debido a su fiabilidad, economía, rapidez y facilidad para establecer distintos casos. 6. ¿Cómo se garantiza su fiabilidad? Mediante el contraste de resultados obtenidos con ensayos físicos realizados, garantizando de ésta manera el buen comportamiento de los programas utilizados. El presente estudio tratará de responder a todas estas preguntas, para concluir y conseguir elementos estructurales de vidrio con secciones más reducidas gracias a la introducción de la precompresión, todo ello a través de las simulaciones informáticas por medio de elementos finitos. Dentro de estas simulaciones, también se realizarán comprobaciones y comparaciones entre distintas tipologías de programas para comprobar y contrastar los resultados obtenidos, intentando analizar cuál de ellos es el más idóneo para la simulación de elementos estructurales de vidrio. ABSTRACT Glass is a material very appreciated in architecture due to its transparency, feature that just a few materials share. But it is also a brittle material with an immediate breakage when it reaches its elastic limit, without having a plastic period that provides warning of future breakage allowing a safety period. For both reasons, glass has been used in architecture as infill panels, from old times. However, it has never been used as a structural or load‐bearing element, although it is an interesting material for architects for that use: because of its transparency, structural glass makes possible the visual dematerialization of the structure, achieving lighter spaces. However, taking into account the mechanical properties of the material, it is possible to check that it has appropriate conditions for structural use: its elastic modulus is similar to that of aluminium, element widely used in architecture, especially in facades from recent years; and its compressive strength is much higher than even the one of concrete. However, its main problem consists in its tensile strength that is much lower than its compressive strength, penalizing its resistance to bending. Nowadays glass is starting to be used as a bearing or structural element, but due to its worse bending strength, elements with large dimensions must be used, with a large presence despite its transparency. Therefore this research aims to get smaller sections of these structural glass elements. For the development of this thesis, it is necessary to answer a number of fundamental questions. The answers will be the core of this work: 1. What is the purpose of the investigation? The objective of this research is the optimization of structural glass elements for its use in architecture. 2. How are you going to perform this optimization? What systems will be implemented? The system for optimization is the pre‐stress of the structural elements of glass 3. Why are you going to use the pre‐compression? Because glass has a good resistance to compression and a poor tensile behaviour, which penalizes its use in bending elements. Through the pre‐compression it is possible to increase this tensile strength, due to the initial tensile efforts reducing the pre‐stress and increasing its load capacity. 4. What are the means that you will use in order to verify and justify this behaviour? The means are based on computer simulations with finite element programs (FEM) 5. Why do you use this method? Because it is a tool which gives advantages over other methods such as experimental: its reliability, economy, quick and easy to set different cases. 6. How the reliability is guaranteed? It’s guaranteed comparing the results of the simulation with the performed physical tests, ensuring the good performance of the software. This thesis will attempt to answer all these questions, to obtain glass structural elements with smaller sections thanks to the introduction of the pre‐compression, all through computer simulations using finite elements methods. In these simulations, tests and comparisons between different types of programs will also be implemented, in order to test and compare the obtained results, trying to analyse which one is the most suitable for the simulation of structural glass elements.

Efficient training of RBF networks for classification.

Radial Basis Function networks with linear outputs are often used in regression problems because they can be substantially faster to train than Multi-layer Perceptrons. For classification problems, the use of linear outputs is less appropriate as the outputs are not guaranteed to represent probabilities. We show how RBFs with logistic and softmax outputs can be trained efficiently using the Fisher scoring algorithm. This approach can be used with any model which consists of a generalised linear output function applied to a model which is linear in its parameters. We compare this approach with standard non-linear optimisation algorithms on a number of datasets.

Efficient training of RBF networks for classification

Radial Basis Function networks with linear outputs are often used in regression problems because they can be substantially faster to train than Multi-layer Perceptrons. For classification problems, the use of linear outputs is less appropriate as the outputs are not guaranteed to represent probabilities. In this paper we show how RBFs with logistic and softmax outputs can be trained efficiently using algorithms derived from Generalised Linear Models. This approach is compared with standard non-linear optimisation algorithms on a number of datasets.

The impact of architecture on the performance of artificial neural networks

A number of researchers have investigated the impact of network architecture on the performance of artificial neural networks. Particular attention has been paid to the impact on the performance of the multi-layer perceptron of architectural issues, and the use of various strategies to attain an optimal network structure. However, there are still perceived limitations with the multi-layer perceptron and networks that employ a different architecture to the multi-layer perceptron have gained in popularity in recent years, particularly, networks that implement a more localised solution, where the solution in one area of the problem space does not impact, or has a minimal impact, on other areas of the space. In this study, we discuss the major architectural issues affecting the performance of a multi-layer perceptron, before moving on to examine in detail the performance of a new localised network, namely the bumptree. The work presented here examines the impact on the performance of artificial neural networks of employing alternative networks to the long established multi-layer perceptron. In particular, networks that impose a solution where the impact of each parameter in the final network architecture has a localised impact on the problem space being modelled are examined. The alternatives examined are the radial basis function and bumptree neural networks, and the impact of architectural issues on the performance of these networks is examined. Particular attention is paid to the bumptree, with new techniques for both developing the bumptree structure and employing this structure to classify patterns being examined.

The Usage of Neural Networks for the Medical Diagnosis

The problem of cancer diagnosis from multi-channel images using the neural networks is investigated. The goal of this work is to classify the different tissue types which are used to determine the cancer risk. The radial basis function networks and backpropagation neural networks are used for classification. The results of experiments are presented.

Data-Driven Dynamic Modeling of Coupled Thermal and Electric Outputs of Microturbines

Microturbines are among the most successfully commercialized distributed energy resources, especially when they are used for combined heat and power generation. However, the interrelated thermal and electrical system dynamic behaviors have not been fully investigated. This is technically challenging due to the complex thermo-fluid-mechanical energy conversion processes which introduce multiple time-scale dynamics and strong nonlinearity into the analysis. To tackle this problem, this paper proposes a simplified model which can predict the coupled thermal and electric output dynamics of microturbines. Considering the time-scale difference of various dynamic processes occuring within microturbines, the electromechanical subsystem is treated as a fast quasi-linear process while the thermo-mechanical subsystem is treated as a slow process with high nonlinearity. A three-stage subspace identification method is utilized to capture the dominant dynamics and predict the electric power output. For the thermo-mechanical process, a radial basis function model trained by the particle swarm optimization method is employed to handle the strong nonlinear characteristics. Experimental tests on a Capstone C30 microturbine show that the proposed modeling method can well capture the system dynamics and produce a good prediction of the coupled thermal and electric outputs in various operating modes.