O presente trabalho tem como objetivo principal o desenvolvimento de um sistema de identificação do surgimento de vazamentos (rupturas) em dutos, através da análise do sinal de sensores de pressão de resposta rápida (frequência de corte superior a 1 kHz). O reconhecimento do sinal de vazamento se realiza através de uma rede neural artificial feed-foward do tipo Perceptron Multi Camadas, previamente treinada. Neste trabalho, a implementação para tal operação foi feita off-line, mas devido ao baixo custo computacional pode ser facilmente implementada em eletrônica embarcada, em tempo real (on-line). Os resultados experimentais foram obtidos no oleoduto piloto do NETeF - Núcleo de Engenharia Térmica e Fluidos da USP - Universidade de São Paulo, com uma seção de testes com 1500 metros e diâmetro de 51,2 mm. Especificamente, os resultados foram obtidos com escoamento monofásico de água. Os resultados mostram-se promissores, visto que o sistema de redes neurais artificiais foi capaz de discriminar 2 universos linearmente separáveis, para sinais de vazamento e de não vazamento, para diversas vazões e localizações de vazamentos simulados. / The present dissertation deals with the development of a system to identify abrupt leaks (ruptures) in pipelines, by analyzing the signal of fast response pressure sensors (cutoff frequency over then 1kHz). The recognition of the leak signal is established by an artificial neural network feed-forward Perceptron Multi Layer, previously trained. In the present work the implementation was performed off-line, but due to low computational costs, the neural network can be easily implemented in real time embedded electronics (online). The experimental results were obtained in a 1500 meter-long and 51.2 millimeter-diameter pilot pipeline at the Center of Thermal Engineering and Fluids. Specifically, the results were obtained with single-phase flow of water. The results have proven to be promising, as the trained neural network was capable of classifying the 2 types of signals into 2 linearly separable regions, for leakage signals and no leakage signals, for various flow rates and locations of simulated leaks.

A área de dinâmica de aeronaves atingiu um alto nível de desenvolvimento e devido à crescente disponibilidade de computadores cada vez mais rápidos e com maior capacidade de processamento; a aplicação de técnicas numéricas de identificação nesta área também teve grande avanço. Este trabalho apresenta uma metodologia para predição de respostas de aeronaves dentro de envelopes de vôo pré-estabelecidos usando redes neurais recorrentes e uma metodologia para estimação das suas derivadas aerodinâmicas usando redes neurais feedforward. Para obter os conjuntos de dados para treinar as redes neurais, foi implementado um modelo não linear de dinâmica de vôo e simulado o comportamento de uma aeronave de combate em nove pontos de um envelope de vôo. Foram usadas as respostas simuladas correspondentes a quatro pontos para treinar a rede neural e depois disto, esta capturou satisfatoriamente a dinâmica da aeronave, identificando com grande sucesso as respostas do movimento longitudinal da aeronave por todo o envelope de vôo considerado. Após a simulação e identificação das respostas da aeronave dentro do envelope de vôo, é apresentada a resolução do problema inverso, ou seja, usando velocidades escalares e angulares da aeronave juntamente com seus dados geométricos como entradas para a rede neural feedforward, é obtido um modelo neural estimador de derivadas aerodinâmicas. Para mostrar a capacidade deste modelo neural estimador, este é aplicado na estimação das derivadas da aeronave simulada e também aplicado na estimação das derivadas aerodinâmicas da aeronave militar a jato Xavante AT-26 da Força Aérea Brasileira. Estas metodologias propostas reduzem custo de obtenção das derivadas aerodinâmicas e mostram a eficácia das redes neurais em estimar as respostas de aeronaves dentre de um envelope de vôo pré-definido. / The area of aircraft dynamics has reached a high level of development and due to the increasing availability of computers continuously faster and with bigger processing capacity, the application of numerical identification techniques in this area also had great advance. This work presents two methodologies, one for prediction of aircraft responses within a pre-established flight envelope using recurrent neural networks and another one for estimation of its aerodynamic derivatives using feedforward neural networks. To get data sets to train the neural networks, a combat aircraft flight dynamics non-linear model was implemented and simulated in nine points of the flight envelope to obtain its behavior. The simulated responses corresponding to a four points of the flight envelope were used to train the neural network and after that, it was possible to verify that this net satisfactorily captured the dynamics of the aircraft, identifying with great success the longitudinal motion responses of the aircraft at all the considered flight envelope positions. After the simulation and identification of the aircraft responses inside the flight envelope, the solution of the inverse problem is presented, i.e., using scalar and angular aircraft velocities together with its geometric data as input to the feedforward neural network, a neural estimator model of aerodynamic derivatives is obtained. In order to show the capacity of this neural estimator model, this model is applied to the estimation of the derivatives of the simulated aircraft as well as to the estimation of the aerodynamic derivatives of a brazilian air force military jet aircraft, the Xavante AT-26. These proposed methodologies reduce the cost of obtaining the aerodynamic derivatives and show the estimation effectiveness of the neural networks to estimate the responses of an aircraft inside a pre-defined flight envelope.

Dados reais apontam um expressivo aumento do consumo de combustível no Brasil e no mundo, além de um crescimento acelerado da população urbana. Ambos os processos vem ocorrendo sem um controle adequado no país e, como conseqüência, têm surgido grandes deseconomias urbanas, tais como: congestionamentos, poluição ambiental, consumo exagerado de combustíveis e uso inadequado do espaço viário. Neste contexto, quaisquer iniciativas no intuito de frear estas deseconomias são relevantes e oportunas, tanto que pesquisas nacionais e internacionais vêm sendo realizadas buscando entender melhor os fatores que mais interferem na energia gasta com transportes. O objetivo deste trabalho é investigar a relação entre o consumo de energia com transportes e algumas variáveis espaciais e sócio-econômicas dos municípios do estado de São Paulo com população superior a 50 mil habitantes. A caracterização dos padrões de forma das áreas urbanizadas foi viabilizada graças aos recursos de um Sistema de Informações Geográficas, que possibilitaram determinar com relativa precisão as variáveis espaciais das manchas urbanas a partir de imagens de satélite georeferenciadas. Uma vez levantados todos os dados possíveis, procedeu-se a uma análise através do emprego de Redes Neurais Artificiais, ferramenta que possibilita identificar e classificar as variáveis de acordo com suas importâncias relativas no consumo de energia, que é a variável dependente do modelo. Os resultados encontrados para as cidades paulistas pesquisadas confirmam a tendência internacional, sobretudo no que concerne à grande relevância da densidade populacional urbana, juntamente com outras características sócio-econômicas, sobre o consumo de energia com transportes. Variáveis como a população urbana, a densidade populacional e o nível de empregos no comércio revelaram-se como as de maior importância relativa no contexto analisado. / The world has been experiencing in recent years an unprecedented increase in the amount of fuel consumed for transportation purposes, in addition to a fast growth of the urban population. Those conditions were also found in Brazil, where they have produced several problems for urban areas, such as: traffic congestion, environmental pollution, high fuel consumption, and an improper use of the urban space. In such a context, any attempt to reduce those problems and their consequences is relevant and opportune. That is the reason why a considerable research effort is being directed to the issue at both national and international levels, in order to better understand the factors that most significantly contribute for the high levels of energy use for transportation.The aim of this work is to investigate the relationship between energy consumption for transportation and a few selected variables related to urban form and socioeconomic characteristics of urbanized areas with more then 50,000 inhabitants located in the state of São Paulo. The boundaries of the urbanized areas were obtained from satellite images georeferenced in a Geographic Information System environment, which also offered the tools for the analysis of some spatial attributes. After the spatial and socioeconomic data were combined in a single database, they were then analyzed using Artificial Neural Network models, in order to identify variables that are relevant to energy consumption for transportation, along with their relative weights.The results found with the Brazilian cities selected for the current study confirmed the trend observed in several countries worldwide, in which urban density played an important role influencing energy use for transportation. In the case studied here, other relevant input variables that considerably influenced the energy consumed for transportation were population and employment level.

Este trabalho apresenta um sistema de trading que toma decisões de compra e de venda do índice Standard & Poors 500, na modalidade seguidor de tendência, mediante o emprego de redes neurais artificiais multicamadas com propagação para frente, no período de 5 anos, encerrado na última semana do primeiro semestre de 2012. Geralmente o critério usual de escolha de redes neurais nas estimativas de preços de ativos financeiros é o do menor erro quadrático médio entre as estimativas e os valores observados. Na seleção das redes neurais foi empregado o critério do menor erro quadrático médio na amostra de teste, entre as redes neurais que apresentaram taxas de acertos nas previsões das oscilações semanais do índice Standard & Poors 500 acima de 60% nessas amostras de teste. Esse critério possibilitou ao sistema de trading superar a taxa anual de retorno das redes neurais selecionadas pelo critério usual e, por larga margem, a estratégia de compre e segure no período. A escolha das variáveis de entrada das redes neurais recaiu entre as que capturaram o efeito da anomalia do momento dos preços do mercado de ações no curto prazo, fenômeno amplamente reconhecido na literatura financeira. / This work presents a trend follower system that makes decisions to buy and sell short the Standard & Poors 500 Index, by using multilayer feedforward neural networks. It was considered a period of 5 years, ending in the last week of the first half of 2012. Usually a neural networks choice criterion to forecast financial asset prices is based on the least mean square error between the estimated and observed prices in the test samples. In this work we also adopted another criterion based on the least mean square error for those neural networks that had a hit rate above 60% of the Standard & Poors 500 Index weekly change in the test sample. This criterion was shown to be the most appropriate one. The neural networks input variables were chosen among those technical indicators that better captured the anomaly of the short term momentum of prices. The annual rate of return of the trading system based on those criteria surpassed those selected by the usual criteria, and by a wide margin the buy-and-hold strategy. The neural networks inputs were chosen to capture the momentum anomaly of the prices on the short term that is fully recognized in the financial literature.

The high complexity existent in businesses has required managers to rely on accurate and up to date information. Over the years, many tools have been created to give support to decision makers, such as discrete event simulation and artificial neural networks. Both tools have been applied to improve business performance; however, most of the time they are used separately. This research aims to interpret artificial neural network models that are applied to the data generated by a simulation model and determine which inputs have the most impact on the output of a business. This would allow prioritization of the variables for maximized system performance. A connection weight approach will be used to interpret the artificial neural network models. The research methodology consisted of three main steps: 1) creation of an accurate simulation model, 2) application of artificial neural network models to the output data of the simulation model, and 3) interpretation of the artificial neural network models using the connection weight approach. In order to test this methodology, a study was performed in the raw material receiving process of a manufacturing facility aiming to determine which variables impact the most the total time a truck stays in the system waiting to unload its materials. Through the research it was possible to observe that artificial neural network models can be useful in making good prediction about the system they model. Moreover, through the connection weight approach, artificial neural network models were interpreted and helped determine the variables that have the greatest impact on the modeled system. As future research, it would be interesting to use this methodology with other data mining algorithms and understand which techniques have the greatest capabilities of determining the most meaningful variables of a model. It would also be relevant to use this methodology as a resource to not only prioritize, but optimize a simulation model.

Three dimensional (3D) ground reaction force (GRF) are an essential component for gait analysis. Current methods for measuring 3D GRF involve using a stationary force plate embedded in the ground, which captures the forces as subjects walk across the platform. This approach has several limitations, a few being: it can only capture a few steps at a time, it is expensive to purchase and maintain, it can't reflect forces caused by natural uneven surfaces, etc. Previous research has attempted to develop wearable force sensors to overcome these problems; however, these endeavors have resulted in devices that are expensive, bulky, and fail to accurately measure forces when compared to static force plates. This thesis presents the implementation and validation of novel nanocomposite piezoresponsive foam (NCPF) sensors for measuring 3D GRF. Four NCPF sensors were embedded in a shoe sole at four locations: heel, arch, ball, and toe. The signals from each sensor were used in a functional data analysis (FDA) to develop a statistical model for estimating 3D GRF. The process of calibrating the sensors to model GRF was validated through a study where 9 subjects (4 females, 5 males) walked on a force-sensing treadmill for two minutes. Two approaches were used to model the GRF response. The first approach was based on functional decomposition of the data. Using a tenfold cross validation process a statistical model was developed for each subject with the ability to predict walking 3D GRF with less than 7% error. The second approach used machine learning to model 3D GRF. Using the same walking data for the statistical models, an artificial neural network (ANN) was used to create subject-specific models that could predict walking 3D GRF with less than 11% error. The predictive capabilities of ANN were tested using a pilot study where a single subject performed a calibration procedure by running at seven different speeds for thirty seconds each on the force-sensing treadmill. This calibration data was used to train a model, which was then used to estimate vertical GRF (VGRF) for three additional running trials at randomly selected speeds from within the calibration range. The ANN model was able to predict VGRF for three running speeds after calibration with less than 4% error. The use of NCPF sensors to estimate 3D GRF was shown to be a viable alternative to static force plates. It is recommended, in future work, that 3D GRF and subsequent sensor data be collected from a large sample of subjects to create a baseline of 3D GRF characteristics for a population that will enable a robust cross-subject model capable of performing real-time ground reaction force analysis across the general population, which will greatly benefit our understanding of human gait.

Developing advanced semantic models is important in building computational systems that can not only understand language but also convey ideas and concepts to others. Semantic models can allow a creative image-producing-agent to autonomously produce artifacts that communicate an intended meaning. This notion of communicating meaning through art is often considered a necessary part of eliciting an aesthetic experience in the viewer and can thus enhance the (perceived) creativity of the agent. Computational creativity, a subfield of artificial intelligence, deals with designing computational systems and algorithms that either automatically create original and functional products, or that augment the ability of humans to do so. We present work on DARCI (Digital ARtist Communicating Intention), a system designed to autonomously produce original images that convey meaning. In order for DARCI to automatically express meaning through the art it creates, it must have its own semantic model that is perceptually grounded with visual capabilities.The work presented here focuses on designing, building, and incorporating advanced semantic and perceptual models into the DARCI system. These semantic models give DARCI a better understanding of the world and enable it to be more autonomous, to better evaluate its own artifacts, and to create artifacts with intention. Through designing, implementing, and studying DARCI, we have developed evaluation methods, models, frameworks, and theories related to the creative process that can be generalized to other domains outside of visual art. Our work on DARCI has even influenced the visual art community through several collaborative efforts, art galleries, and exhibits. We show that the DARCI system is successful at autonomously producing original art that is meaningful to human viewers. We also discuss insights that our efforts have contributed to the field of computational creativity.

Extreme Learning Machine (ELM) is a recently discovered way of training Single Layer Feed-forward Neural Networks with an explicitly given solution, which exists because the input weights and biases are generated randomly and never change. The method in general achieves performance comparable to Error Back-Propagation, but the training time is up to 5 orders of magnitude smaller. Despite a random initialization, the regularization procedures explained in the thesis ensure consistently good results. While the general methodology of ELMs is well developed, the sheer speed of the method enables its un-typical usage for state-of-the-art techniques based on repetitive model re-training and re-evaluation. Three of such techniques are explained in the third chapter: a way of visualizing high-dimensional data onto a provided fixed set of visualization points, an approach for detecting samples in a dataset with incorrect labels (mistakenly assigned, mistyped or a low confidence), and a way of computing confidence intervals for ELM predictions. All three methods prove useful, and allow even more applications in the future. ELM method is a promising basis for dealing with Big Data, because it naturally deals with the problem of large data size. An adaptation of ELM to Big Data problems, and a corresponding toolbox (published and freely available) are described in chapter 4. An adaptation includes an iterative solution of ELM which satisfies a limited computer memory constraints and allows for a convenient parallelization. Other tools are GPU-accelerated computations and support for a convenient huge data storage format. The chapter also provides two real-world examples of dealing with Big Data using ELMs, which present other problems of Big Data such as veracity and velocity, and solutions to them in the particular problem context.

Attention Deficit Hyperactivity Disorder (ADHD) is a neurodevelopmental disorder characterized by difficulty paying attention, impulsivity, and hyperactivity. Diagnosis of ADHD rose 42% from 2003–2004 to 2011–2012. In 2011, 3.5 million children were treated with drugs. Optimizing therapy can take a year, and may not be completely effective. A clinical trial is currently being conducted of a device/drug combination using the computer game Minecraft, to determine how certain activities affect executive function, working memory, and restraint in patients diagnosed with ADHD. The human subjects’ responses are being modeled using artificial neural networks (ANNs), an artificial intelligence method that can be utilized to interpret highly complex data. We propose using ANNs to optimize drug and Minecraft therapy for individual patients based on the initial NICHQ Vanderbilt assessment scores. We are applying ANNs in the development of computational models for executive function deficiencies in ADHD. These models will then be used to develop a therapeutic video game as a drug/device combination with stimulants for the treatment of ADHD symptoms in Fragile X Syndrome. As a first step towards the design of virtual subjects with executive function deficits, computational models of the core executive functions working memory and fluid intelligence were constructed. These models were combined to create healthy control and executive function-deficient virtual subjects, who performed a Time Management task simulation that required the use of their executive functions to complete. The preliminary working memory model utilized a convolutional neural network to identify handwritten digits from the MNIST dataset, and the fluid intelligence model utilized a basic recurrent neural network to produce sequences of integers in the range 1-9 that can be multiplied together to produce the number 12. A simplified Impulsivity function was also included in the virtual subject as a first step towards the future inclusion of the core executive function inhibition.

No description available.