Los sobrecostos en la construcción de edificaciones de concreto armado representan pérdidas de un 28% de la inversión (Flyvbjerg, 2002), esto se debe a que las secciones de los elementos estructurales son sobredimensionadas y generan mayor costo en el concreto y acero. Por ello, se realizó la presente investigación en la que se empleó una metodología capaz de optimizar las dimensiones de los elementos estructurales (columnas, vigas y placas) en edificios multifamiliares regulares de 6 pisos. La metodología empleada se basó en el uso de redes neuronales del tipo feedforward, en la que se estableció como variables de entrada, los datos preliminares que se tienen de una edificación y como variables de salida las dimensiones de cada elemento estructural. Para ello, se elaboraron 30 edificios de 6 pisos como base de datos y en cada uno de estos se realizaron las verificaciones de derivas según la Norma Técnica Peruana E 0.30 y la resistencia de cada elemento estructural. De la base de datos se usaron 22 como entrenamiento y 8 para la validación interna de la red neuronal. La estructura de la red neuronal se estableció luego de ejecutar 10 diferentes redes neuronales y se seleccionó la red con un coeficiente de correlación más homogéneo y cercano a 1, en esta investigación fue de 0.98. Finalmente, se realizó la comparación del volumen de concreto que se emplea en una edificación dimensionada con métodos convencionales con el uso del software ETABS y los obtenidos con el uso de la metodología empleando redes neuronales artificiales, según esto, se calculó la diferencia de concreto entre ambos casos. Con los resultados obtenidos se comprobó que la metodología aplicada en esta investigación brinda un ahorro eficaz cercano al 10%. / Cost overruns in the construction of reinforced concrete buildings represent losses of 28% of the investment (Flyvbjerg, 2002), this is due to the fact that the sections of the structural elements are oversized and generate higher costs in concrete and steel. Therefore, the present research was carried out using a methodology capable of optimizing the dimensions of structural elements (columns, beams and slabs) in regular 6-story multifamily buildings. The methodology used was based on the use of feedforward neural networks, in which the preliminary data of a building were established as input variables and the dimensions of each structural element as output variables. For this purpose, 30 6-story buildings were prepared as a database and in each one of them the drift verifications were performed according to the Peruvian Technical Standard E 0.30 and the resistance of each structural element. From the database, 22 were used for training and 8 for the internal validation of the neural network. The structure of the neural network was established after running 10 different neural networks and the network with the most homogeneous correlation coefficient close to 1 was selected; in this research it was 0.98. Finally, a comparison was made between the volume of concrete used in a building dimensioned with conventional methods with the use of ETABS software and those obtained with the use of the methodology employing artificial neural networks, according to this, the difference of concrete between both cases was calculated. With the results obtained, it was proved that the methodology applied in this research provides an effective saving close to 10%. / Tesis

No description available.

No description available.

Computer vision object detection is the task of detecting and identifying objects present in an image or a video sequence. Models based on artificial convolutional neural networks are commonly used as detector models. Object detection precision and inference efficiency are crucial for surveillance-based applications. A decrease in the detector model complexity as well as in the complexity of the post-processing computations promotes increased inference efficiency. Modern object detectors for surveillance applications usually make use of a regression algorithm and bounding box priors referred to as anchor boxes to compute bounding box proposals, and the proposal selection algorithm contributes to the computational cost at inference. In this study, an anchor-free and low complexity deep learning detector model was implemented within a surveillance applications setting, and was evaluated and compared to a reference baseline state-of-the-art anchor-based object detector. A key-point-based detector model (CenterNet), predicting Gaussian distribution based object centers, was selected for the evaluation against the baseline. The surveillance applications adapted anchor-free detector exhibited a factor 2.4 lower complexity than the baseline detector. Further, a significant redistribution to shorter post-processing times was demonstrated at inference for the anchor-free surveillance adapted CenterNet detector, giving a modal values factor 0.6 of the baseline detector post-processing time. Furthermore, the surveillance adapted CenterNet model was shown to outperform the baseline in terms of detection precision for several surveillance applications relevant classes and for objects of smaller spatial scale.

[pt] O objetivo desta dissertação é desenvolver modelos matemáticos que permitam estimar o tempo médio dos banhos com a utilização de chuveiros elétricos e a curva de carga desses aparelhos, considerando as informações das Pesquisas de Posses e Hábitos de Consumo (PPH) e medições realizadas com o auxílio de medidores eletrônicos com memória de massa, em residências com chuveiros elétricos. A motivação do estudo advém de uma exigência da ANEEL que determina que as distribuidoras de energia elétrica realizem a cada 2 (dois) ciclos de revisão tarifária a PPH em suas unidades consumidoras. Os métodos empregados foram: estatística descritiva (para a obtenção do tempo médio de banho); aplicação da regressão linear e de redes neurais (para corrigir a curva de carga horária obtida com a PPH, com base nos dados das medições). Os resultados foram promissores, pois o tempo médio de banho se encontra próximo às estimativas do PROCEL (que são de 8 (oito) a 10 (dez) minutos) e a curva de carga estimada se encontra próxima à da medição, sendo esta última o consumo real. Conclui-se que a abordagem desta dissertação resultou em melhorias na estimativa dos coeficientes de ajustes e que o método de redes neurais foi relativamente melhor que o método de regressão linear simples. / [en] The aim of this dissertation is to develop mathematical models that would allow the estimation of the average time of baths using electric showers and the load shape curves for these devices, obtained from two sources: the information of Electrical Appliances Ownership Survey and measurements of electric shower usage in households carried out with electronic meters with storage capacity. The motivation stems from a requirement of ANEEL that determines that the electric energy distributors periodically should hold a PPH in their consumer units. Concerning the average time of shower baths, the last PPH survey conducted by PROCEL in 2005 estimated this time between 8 (eight) and 10 (ten) minutes. The methods employed in this work were: descriptive statistics (for obtaining the average bath time); application of linear regression and neural networks (to estimate the correction factors to approximate the load shape curves obtained by PPH to those obtained by measurements). The obtained results are rather promising due to the following reasons: the average time of bath is next to the estimates of PROCEL and the corrected load shape curve estimated is quite close to the measured curve, the latter being the actual consumption. This approach has resulted in improvements in the estimation of the coefficients of adjustments and the method of neural networks was relatively better than the simple linear regression method.

[pt] Aplicações em Internet das Coisas (IoT) baseadas em agentes têm surgido como aplicações que podem envolver sensores, dispositivos sem fio, máquinas e softwares que podem compartilhar dados e que podem ser acessados remotamente. Essas aplicações vêm sendo propostas em vários domínios de aplicação, incluindo cuidados em saúde, cidades inteligentes e agricultura. Uma terminologia comumente utilizada para representar agentes embarcados inteligentes é embodied agents, a qual é proposta esse trabalho para projetar agentes para o domínio de IoT. Embodied agents significa agentes que possuem corpo, o qual pode ser definido pelos tipos de sensores e atuadores, e controlador, normalmente representada por uma rede neural artificial. Apesar da capacidade de reconfiguração ser essencial para embodied agents inteligentes, existem poucas tecnologias para suportar sistemas reconfigurfuaveis. Além disso, é necessário novas abordagens para lidar com as variabilidades dos agentes e do ambiente, e novos procedimentos para investigar a relação o entre o corpo e o controlador de um embodied agent, assim como as interações entre as mudanças do agente e do ambiente. Além da variabilidade do corpo e do controlador desses agentes, a exemplo do número e tipos de sensores, assim como o número de camadas e tipos de função de ativação para a rede neural, também é preciso lidar com a variabilidade do ambiente em que esses agentes estão situados. A fifim de entender melhor e esclarecer os conceitos de embodied agents, este trabalho apresenta um modelo de referência para embodied agents autoconfifiguráveis de IoT. A partir desse modelo de referência, três abordagens foram criadas para projetar e testar agentes embarcados reconfifiguráeis: i) um software framework para o desenvolvimento de embodied agents no domínio de internet das coisas; ii) uma arquitetura para configurar o corpo e controlador dos agentes de acordo com as variantes do ambiente; e iii) uma ferramenta para testar embodied agents. As abordagens foram avaliadas através de estudos de caso e experimentos em diferentes domínios de aplicação. / [en] Agent-based Internet of Things (IoT) applications have recently emerged as applications that can involve sensors, wireless devices, machines and software that can exchange data and be accessed remotely. Such applications have been proposed in several domains including health care, smart cities and agriculture. Embodied Agents is a term used to denote intelligent embedded agents, which we use to design agents to the IoT domain. Each agent is provided with a body that has sensors to collect data from the environment and actuators to interact with the environment, and a controller that is usually represented by an artificial neural network. Because reconfigurable behavior is key for autonomous embodied agents, there is a spectrum of approaches to support system reconfigurations. However, there is a need for approaches to handle agents and environment variability, and for a broad spectrum of procedures to investigate the relationship between the body and the controller of an embodied agent, as the interaction between the agent and the environment changes. In addition to the body and controller variability of these agents, such as those variations related to the number and types of sensors as well as the number of layers and types of activation function for the neural network, it is also necessary to deal with the variability of the environment in which these agents are situated. A discussion of the embodied agents should have some formal basis in order to clarify these concepts. Notwithstanding, this thesis presents a reference model for selfcon figurable IoT embodied agents. Based on this reference model, we have created three approaches to design and test self-configurable IoT embodied agents: i) a software framework for the development of embodied agents to the Internet of Things (IoT) applications; ii) an architecture to configure the body and controller of the agents based on environment variants; and iii) a tool for testing embodied agents. To evaluate these approaches, we have conducted diffierent case studies and experiments in difierent application domains.

[pt] O mundo vive hoje uma era de transformação digital resultante da chamada indústria 4.0 ou quarta revolução industrial. Nesta fase, a tecnologia tem exercido um papel cada vez mais estratégico no desempenho das organizações. Estes avanços tecnológicos têm revolucionado o processo de tomada de decisão na gestão e operação de cadeias de suprimentos. Neste contexto, esta dissertação apresenta uma metodologia de apoio à decisão na gestão de estoques, que combina multi-criteria decision making (MCDM) e machine learning (ML). A princípio, é realizada uma revisão sistemática da literatura para analisar como estas duas abordagens são aplicadas na gestão de estoques. Os resultados são complementados com um scoping review abrangendo a previsão de demanda. Inicia-se então um estudo de caso, aplicado em uma ferrovia de transporte de cargas. É aplicado, inicialmente, o método MCDM combinado Fuzzy AHP Vikor para ranquear os stock keeping units (SKUs) em ordem de criticidade. O passo seguinte é a aplicação do método de ML combinado GA-ANN, artificial neural network com genetic algorithm, com o objetivo de realizar a previsão de demanda em um piloto com alguns dos itens mais críticos. A etapa final consiste em estruturar um dashboard gerencial, integrando os resultados das etapas anteriores. Dentre os resultados alcançados, a partir do modelo proposto, observa-se considerável melhora na performance da previsão de demanda dos SKUs selecionados. Além disso, a integração entre as abordagens e implementação em um dashboard gerencial permitiu o desenvolvimento de um modelo semiautomático de tomada de decisão na gestão de estoques. / [en] The world is experiencing an era of digital transformation resulting from the industry 4.0 or fourth industrial revolution. In this period, technology has played an increasingly strategic role in the performance of organizations. These technological advances have revolutionized the decision-making process in the management and operation of supply chains. In this context, this dissertation presents a methodology to support decision-making in inventory management, which combines multi-criteria decision-making (MCDM) and machine learning (ML). At first, there is a systematic literature review in order to analyze how these two approaches are applied in inventory management. The results are complemented with a scoping review that includes the demand forecasting. A case study is then applied to a freight transport railway. Initially, the MCDM combined Fuzzy AHP Vikor method is applied to rank stock keeping units (SKUs) in degrees of criticality. The next step is the application of the ML combined GA-ANN method, artificial neural network with genetic algorithm, for the purpose of demand forecasting in a pilot with some of the most critical items. The final step is to structure a management dashboard, integrating the results of the previous steps. Among the results achieved from the proposed model, there is a considerable improvement in the performance of the demand forecasting for the selected SKUs. In addition, the integration between approaches and implementation in a management dashboard allowed the development of a semiautomatic model for decision-making in inventory management.

[pt] As estruturas civis durante toda a sua vida útil estão sujeitas a diversas ações de deterioração, desgastes ou corrosão de seus membros, que podem gerar variações em suas características físicas. Estas ações podem causar danos ao seu funcionamento, podendo chegar até ao colapso, em casos mais extremos. Além disso, o avanço tecnológico que permite a concepção de estruturas cada vez mais esbeltas, e que geram assim possíveis vibrações excessivas, elevam o monitoramento estrutural a um patamar de extrema importância e atenção na ótica dos gestores desses sistemas. Particularmente, no caso de obras de infraestrutura como pontes e passarelas, as grandes dimensões são características significativas que tornam as práticas de monitoramento e inspeção mais difíceis. Dessa forma, com o objetivo auxiliar no monitoramento estrutural e direcionar inspeções visuais, diversos métodos de identificação de danos são estudados com base nas características dinâmicas das estruturas, como as frequências naturais e os modos de vibração. A revisão de literatura, porém, demonstra que há uma dificuldade na aplicação desta identificação em estruturas mais complexas de grande porte. Assim, este trabalho visa estudar esta dificuldade e propor uma solução baseada na construção de um índice, composto pelos modos de vibração. Além disso, através da aplicação de algoritmos de aprendizado de máquina e de reconhecimento de padrões, como as Redes Neurais Artificiais (RNAs), propõese aumentar a eficiência do processo de localização espacial e quantificação dos danos. Em seguida, a metodologia proposta é, então, aplicada em um modelo de passarela metálica inspirado em uma estrutura real presente na região do Terminal Centro Olímpico da cidade do Rio de Janeiro – RJ. A identificação de danos é estudada através da aplicação do índice proposto, incorporando as redes neurais e avaliando do impacto da variação dos parâmetros da RNA na eficiência global da detecção. / [en] Civil structures are subjected to different deterioration and corrosion actions throughout their entire service life, which can generate variations in their physical characteristics. These actions can cause damage to its functioning, and possibly leading to collapse in more severe cases. In addition, technology development which allows the design of increasingly slender structures, can produce excessive vibrations, which elevates the importance ofstructural monitoring to a higher level from the perspective of infrastructure managers. Particularly, in the case of bridges and walkaways, due to their large dimensions make monitoring and inspection even more difficult. Thus, with the aim of providing methods to assist in structural monitoring and facilitate visual inspections, several damage identification methods are investigated, which are based on structures dynamic characteristics, such as natural frequencies and mode shapes. The conducted literature review revealed that there is a difficulty in applying these identification methods in large-scale and complex structures. Thus, this research aims to study these barriers and propose a solution based on the development of a new damage index based on the structure s mode shapes. Furthermore, through the application of machine learning algorithms and pattern recognition, such as Artificial Neural Networks (ANN), it is proposed to increase the efficiency of the damage identification and quantification process. Then, the proposed methodology is tested numerically on a steel footbridge model inspired by a real structure located in the region of the Olympic Center Terminal, in the city of Rio de Janeiro – RJ. The damage identification method is studied through the application of the proposed damage index, incorporating the neural network and assessing the impact of ANNs parameters variation in the global efficiency of the damage detection method.

Intra-prediction is a method for coding standalone frames in video coding. Until now, this has mainly been done using linear formulae. Using an Artificial Neural Network (ANN) may improve the prediction accuracy, leading to improved coding efficiency. In this degree project, Fully Connected Networks (FCN) and Convolutional Neural Networks (CNN) were used for intra-prediction. Experiments were done on samples from different image sizes, block sizes, and block contents, and their effect on the results were compared and discussed. The results show that ANN methods have the potential to perform better or on par with the video coder High Efficiency Video Coding (HEVC) in the intra-prediction task. The proposed ANN designs perform better on smaller block sizes, but different designs could lead to better performance on larger block sizes. It was found that training one network for each HEVC mode and using the most suitable network to predict each block improved performance of the ANN approach. / Intra-prediktion är en metod för kodning av stillbilder i videokodning. Hittills har detta främst gjorts med hjälp av linjära formler. Användning av artificialla neuronnät (ANN) skulle kunna öka prediktionsnoggrannheten och ge högre effektivitet vid kodning. I detta examensarbete användes fully connected networks (FCN) och convolutional neural networks (CNN) för att utföra intra-prediktion. Experiment gjordes på prover från olika bildstorlekar, blockstorlekar och blockinnehåll, och de olika parametrarnas effekt på resultaten jämfördes och diskuterades. Resultaten visar att ANN-metoder har potential att prestera bättre eller lika bra som videokodaren High Efficiency Video Coding (HEVC) för intra-prediktion. De föreslagna ANN-designerna presterar bättre på mindre blockstorlekar, men andra ANN-designs skulle kunna ge bättre prestanda för större blockstorlekar. Det konstaterades att prestandan för ANN-metoderna kunde ökas genom att träna ett nätverk för varje HEVC-mode och använda det mest passande nätverket för varje block.

<p>Artificial neural networks (ANNs) have achieved extraordinary performance in various domains in recent years. However, some studies reveal that ANNs may be vulnerable in three aspects: label scarcity, perturbations, and open-set emerging classes. Noisy labeling and self-supervised learning approaches address the label scarcity issues, but most of the work couldn't handle the perturbations. Adversarial training methods, topological denoising methods, and mechanism designing methods aim to mitigate the negative effects caused by perturbations. However, adversarial training methods can barely train a robust model under the circumstance of extensive label scarcity; topological denoising methods are not efficient on dynamic data structures; and mechanism designing methods often depend on heuristic explorations. Detection-based methods devote to identifying novel or anomaly instances for further downstream tasks. Nonetheless, such instances may belong to open-set new emerging classes. To embrace the aforementioned challenges, we address the robustness issues of ANNs from two aspects. First, we propose a series of Bayesian label transition models to improve the robustness of Graph Neural Networks (GNNs) in the presence of label scarcity and perturbations in the graph domain. Second, we propose a new non-exhaustive learning model, named NE-GM-GAN, to handle both open-set problems and class-imbalance issues in network intrusion datasets. Extensive experiments with several datasets demonstrate that our proposed models can effectively improve the robustness of ANNs.</p>