Orientador: Carlos Roberto Minussi / Banca: Francisco Villarreal Alvarado / Banca: Maria do Carmo Gomes da Silveira / Banca: Luciana Cambraia Leite / Banca: Ricardo Menezes Salgado / Resumo: Esta pesquisa visa o desenvolvimento de um método para análise da estabilidade transitória de sistemas de energia eletrica multimaquinas, por meio de uma rede neural ART-ARTMAP Fuzzy Euclidiana Modificada com Treinamento Continuado. Esta arquitetura apresenta tres diferenciais em e relação a outras já utilizadas para abordar tal problema: (1) a rede iniciada com apenas um neuronio ativado e vai se expandindo durante todo o o treinamento/análise, (2) possui um módulo de treinamento continuado e (3) a o possui um módulo de deteção de intruso. No primeiro diferencial, a redeé iniciada com um neuronio e vai se expandindo de acordo com a aquisição de conhecimento, isto faz com que esta se torne muito mais rápida e que o gasto computacional se torne mínimo. Com o módulo de treinamento continuado, a rede neural consegue armazenar novos dados sem a necessidade de realizar o retreinamento. Já o módulo de detecção de intruso faz com que, ao ser apresentada a rede uma configuração "estranha", a rede execute um treinamento específico para que esta configuração, com um número mínimo de entradas, seja incorporada definitivamente à rede neural. A aplicação para a rede proposta nesta pesquisa, foi a análise de estabilidade transitória, considerando-se o modelo clássico (estabilidade de primeira oscilação), para um sistema composto por 10 máquinas síncronas, 45 barras e 73 linhas de transmissão / Abstract: This doctoral research aims to develop a method to analyze the transient stability of multimachine eletric power systems, through a neural network Modified Euclidean Fuzzy ART-ARTMAP with Continuous Training. The architecture presented has three differences in relation to others used to deal with this problem: (1) the network starts with only one neuron activated and expands throughout the training/analysis, (2) has a continuous training module and (3) has an intrusion detection module. The first difference, is the fact that it starts with a neuron and expands according to knowledge acquisition of the network, and causes it to become much faster and the computational expenses becomes minimum. With continuous training mod- ule, the neural network can store the new data without the need for the retraining. The intrusion detection module causes, when presented to the network a strange configuration, the network to carry out a specific training for this configuration with a minimum total of inputs so that the configu- ration is definitely incorporated to the neural network. The application for this network, in this research, was to analyze the transient stability consid- ering the classical model (stability of first oscillation) to a system composed of 10 synchronous machines, 45 buses and 73 transmission lines / Doutor

Networks of distributed, remote sensors are providing ecological scientists with a view of our environment that is unprecedented in detail. However, these networks are subject to harsh conditions, which lead to malfunctions in individual sensors and failures in network communications. This behavior manifests as corrupt or missing measurements in the data. Consequently, before the data can be used in ecological models, future experiments, or even policy decisions, it must be quality controlled (QC'd) to flag affected measurements and impute corrected values. This dissertation describes a probabilistic modeling approach for real-time automated QC that exploits the spatial and temporal correlations in the data to distinguish sensor failures from valid observations. The model adapts to a site by learning a Bayesian network structure that captures spatial relationships among sensors, and then extends this structure to a dynamic Bayesian network to incorporate temporal correlations. The final QC model contains both discrete and continuous variables, which makes inference intractable for large sensor networks. Consequently, we examine the performance of three approximate methods for inference in this probabilistic framework. Two of these algorithms represent contemporary approaches to inference in hybrid models, while the third is a greedy search-based method of our own design. We demonstrate the results of these algorithms on synthetic datasets and real environmental sensor data gathered from an ecological sensor network located in western Oregon. Our results suggest that we can improve performance over networks with less sensors that use exhaustive asynchronic inference by including additional sensors and applying approximate algorithms. / Graduation date: 2013

In order to make computers proactive and assistive, we must enable them to perceive, learn, and predict what is happening in their surroundings. This presents us with the challenge of formalizing computational models of everyday human activities. For a majority of environments, the structure of the in situ activities is generally not known a priori. This thesis therefore investigates knowledge representations and manipulation techniques that can facilitate learning of such everyday human activities in a minimally supervised manner. A key step towards this end is finding appropriate representations for human activities. We posit that if we chose to describe activities as finite sequences of an appropriate set of events, then the global structure of these activities can be uniquely encoded using their local event sub-sequences. With this perspective at hand, we particularly investigate representations that characterize activities in terms of their fixed and variable length event subsequences. We comparatively analyze these representations in terms of their representational scope, feature cardinality and noise sensitivity. Exploiting such representations, we propose a computational framework to discover the various activity-classes taking place in an environment. We model these activity-classes as maximally similar activity-cliques in a completely connected graph of activities, and describe how to discover them efficiently. Moreover, we propose methods for finding concise characterizations of these discovered activity-classes, both from a holistic as well as a by-parts perspective. Using such characterizations, we present an incremental method to classify a new activity instance to one of the discovered activity-classes, and to automatically detect if it is anomalous with respect to the general characteristics of its membership class. Our results show the efficacy of our framework in a variety of everyday environments.

Made available in DSpace on 2014-06-11T19:30:50Z (GMT). No. of bitstreams: 0 Previous issue date: 2010-10-22Bitstream added on 2014-06-13T20:00:53Z : No. of bitstreams: 1 moreno_al_dr_ilha.pdf: 923809 bytes, checksum: e8a55f496e6bf5bfbe0531f9211526e5 (MD5) / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / Esta pesquisa visa o desenvolvimento de um método para análise da estabilidade transitória de sistemas de energia eletrica multimaquinas, por meio de uma rede neural ART-ARTMAP Fuzzy Euclidiana Modificada com Treinamento Continuado. Esta arquitetura apresenta tres diferenciais em e relação a outras já utilizadas para abordar tal problema: (1) a rede iniciada com apenas um neuronio ativado e vai se expandindo durante todo o o treinamento/análise, (2) possui um módulo de treinamento continuado e (3) a o possui um módulo de deteção de intruso. No primeiro diferencial, a redeé iniciada com um neuronio e vai se expandindo de acordo com a aquisição de conhecimento, isto faz com que esta se torne muito mais rápida e que o gasto computacional se torne mínimo. Com o módulo de treinamento continuado, a rede neural consegue armazenar novos dados sem a necessidade de realizar o retreinamento. Já o módulo de detecção de intruso faz com que, ao ser apresentada a rede uma configuração estranha, a rede execute um treinamento específico para que esta configuração, com um número mínimo de entradas, seja incorporada definitivamente à rede neural. A aplicação para a rede proposta nesta pesquisa, foi a análise de estabilidade transitória, considerando-se o modelo clássico (estabilidade de primeira oscilação), para um sistema composto por 10 máquinas síncronas, 45 barras e 73 linhas de transmissão / This doctoral research aims to develop a method to analyze the transient stability of multimachine eletric power systems, through a neural network Modified Euclidean Fuzzy ART-ARTMAP with Continuous Training. The architecture presented has three differences in relation to others used to deal with this problem: (1) the network starts with only one neuron activated and expands throughout the training/analysis, (2) has a continuous training module and (3) has an intrusion detection module. The first difference, is the fact that it starts with a neuron and expands according to knowledge acquisition of the network, and causes it to become much faster and the computational expenses becomes minimum. With continuous training mod- ule, the neural network can store the new data without the need for the retraining. The intrusion detection module causes, when presented to the network a strange configuration, the network to carry out a specific training for this configuration with a minimum total of inputs so that the configu- ration is definitely incorporated to the neural network. The application for this network, in this research, was to analyze the transient stability consid- ering the classical model (stability of first oscillation) to a system composed of 10 synchronous machines, 45 buses and 73 transmission lines

Les systèmes navals représentent une infrastructure stratégique pour le commerce international et les activités militaires. Ces systèmes sont de plus en plus informatisés afin de réaliser une navigation optimale et sécurisée. Pour atteindre cet objectif, une grande variété de systèmes embarqués génèrent différentes informations sur la navigation et l'état des composants, ce qui permet le contrôle et le monitoring à distance. Du fait de leur importance et de leur informatisation, les systèmes navals sont devenus une cible privilégiée des pirates informatiques. Par ailleurs, la mer est un environnement rude et incertain qui peut produire des dysfonctionnements. En conséquence, la prise de décisions basée sur des fausses informations à cause des anomalies, peut être à l'origine de répercussions potentiellement catastrophiques.Du fait des caractéristiques particulières de ces systèmes, les méthodologies classiques de détection d'anomalies ne peuvent pas être appliquées tel que conçues originalement. Dans cette thèse nous proposons les mesures de qualité comme une potentielle alternative. Une méthodologie adaptée aux systèmes cyber-physiques a été définie pour évaluer la qualité des flux de données générés par les composants de ces systèmes. À partir de ces mesures, une nouvelle approche pour l'analyse de scénarios fonctionnels a été développée. Des niveaux d'acceptation bornent les états de normalité et détectent des mesures aberrantes. Les anomalies examinées par composant permettent de catégoriser les détections et de les associer aux catégories définies par le modèle proposé. L'application des travaux à 13 scénarios créés pour une plate-forme composée par deux cuves et à 11 scénarios pour deux drones aériens a servi à démontrer la pertinence et l'intérêt de ces travaux. / Naval systems represent a strategic infrastructure for international commerce and military activity. Their protection is thus an issue of major importance. Naval systems are increasingly computerized in order to perform an optimal and secure navigation. To attain this objective, on board vessel sensor systems provide navigation information to be monitored and controlled from distant computers. Because of their importance and computerization, naval systems have become a target for hackers. Maritime vessels also work in a harsh and uncertain operational environments that produce failures. Navigation decision-making based on wrongly understood anomalies can be potentially catastrophic.Due to the particular characteristics of naval systems, the existing detection methodologies can't be applied. We propose quality evaluation and analysis as an alternative. The novelty of quality applications on cyber-physical systems shows the need for a general methodology, which is conceived and examined in this dissertation, to evaluate the quality of generated data streams. Identified quality elements allow introducing an original approach to detect malicious acts and failures. It consists of two processing stages: first an evaluation of quality; followed by the determination of agreement limits, compliant with normal states to identify and categorize anomalies. The study cases of 13 scenarios for a simulator training platform of fuel tanks and 11 scenarios for two aerial drones illustrate the interest and relevance of the obtained results.

Nearly all present-day commercial intrusion detection systems are based on a hierarchical architecture. In such an architecture, the root node is responsible for detecting intrusions and for issuing responses. However, an intrusion detection system (IDS) based on a hierarchical architecture has many single points of failure. For example, by disabling the root node, the intrusion-detection function of the IDS will also be disabled. To solve this problem, an IDS inspired by the human immune system is proposed. The proposed IDS has no single component that is responsible for detecting intrusions. Instead, the intrusion-detection function is divided and placed within mobile agents. Mobile agents act similarly to white blood cells of the human immune system and travel from host to host in the network to detect intrusions. The IDS is fault-tolerant because it can continue to detect intrusions even when most of its components have been disabled. / Computer Science (School of Computing) / M. Sc. (Computer Science)

The use of magnetometers combined with unmanned aerial vehicles (UAVs) is an emerging market for commercial and military applications. This study presents the methodology used to magnetically characterize a novel fixed-wing vertical take-off and landing (VTOL) UAV. The most challenging aspect of integrating magnetometers on manned or unmanned aircraft is minimizing the amount of magnetic noise generated by the aircraft’s onboard components. As magnetometer technology has improved in recent years magnetometer payloads have decreased in size. As a result, there has been an increase in opportunities to employ small to medium UAV with magnetometer applications. However, in comparison to manned aviation, small UAVs have smaller distance scales between sources of interference and sensors. Therefore, more robust magnetic characterization techniques are required specifically for UAVs. This characterization determined the most suitable position for the magnetometer payload by evaluating the aircraft’s static-field magnetic signature. For each aircraft component, the permanent and induced magnetic dipole moment characteristics were determined experimentally. These dipole characteristics were used to build three dimensional magnetic models of the aircraft. By assembling the dipoles in 3D space, analytical and numerical static-field solutions were obtained using MATLAB computational and COMSOL finite element analysis frameworks. Finally, Tolles and Lawson aeromagnetic compensation coefficients were computed and compared to evaluate the maneuver noise for various payload locations. The magnetic models were used to study the sensitivity of the aircraft configuration and to simultaneously predict the effects at potential sensor locations. The study concluded by predicting that a wingtip location was the area of lowest magnetic interference. / Graduate

Le travail présenté dans cette thèse porte sur l'inspection de surfaces métalliques industrielles. Nous proposons de généraliser des méthodes de l'imagerie hyperspectrale à des données multimodales comme des images optiques multi-canales, et des images thermographiques multi-temporelles. Dans la première application, les cubes de données sont construits à partir d'images multi-composantes pour détecter des défauts de surface. Les meilleures performances sont obtenues avec les éclairages multi-longueurs d'ondes dans le visible et le proche IR, et la détection du défaut en utilisant l'angle spectral, avec le spectre moyen comme référence. La deuxième application concerne l'utilisation de l'imagerie thermique pour l'inspection de pièces métalliques nucléaires afin de détecter des défauts de surface et sub-surface. Une approche 1D est proposée, basée sur l'utilisation du kurtosis pour sélectionner la composante principale parmi les premières obtenues après réduction des données avec l’ACP. La méthode proposée donne de bonnes performances avec des données non-bruitées et homogènes, cependant la SVD avec les algorithmes de détection d'anomalies est très robuste aux perturbations. Finalement, une approche, basée sur les techniques d'analyse de franges et la lumière structurée est présentée, dans le but d'inspecter des surfaces métalliques à forme libre. Après avoir déterminé les paramètres décrivant les modèles de franges sinusoïdaux, l'approche proposée consiste à projeter une liste de motifs déphasés et à calculer l'image de phase des motifs enregistrés. La localisation des défauts est basée sur la détection et l'analyse des franges dans les images de phase. / The work presented in this thesis deals with the quality control and inspection of industrial metallic surfaces. The purpose is the generalization and application of hyperspectral imagery methods for multimodal data such as multi-channel optical images and multi-temporal thermographic images. In the first application, data cubes are built from multi-component images to detect surface defects within flat metallic parts. The best performances are obtained with multi-wavelength illuminations in the visible and near infrared ranges, and detection using spectral angle mapper with mean spectrum as a reference. The second application turns on the use of thermography imaging for the inspection of nuclear metal components to detect surface and subsurface defects. A 1D approach is proposed based on using the kurtosis to select 1 principal component (PC) from the first PCs obtained after reducing the original data cube with the principal component analysis (PCA) algorithm. The proposed PCA-1PC method gives good performances with non-noisy and homogeneous data, and SVD with anomaly detection algorithms gives the most consistent results and is quite robust to perturbations such as inhomogeneous background. Finally, an approach based on fringe analysis and structured light techniques in case of deflectometric recordings is presented for the inspection of free-form metal surfaces. After determining the parameters describing the sinusoidal stripe patterns, the proposed approach consists in projecting a list of phase-shifted patterns and calculating the corresponding phase-images. Defect location is based on detecting and analyzing the stripes within the phase-images.

Aujourd'hui, le contrôle des procédés de fabrication est une tâche essentielle pour assurer une production de haute qualité. A la fin du processus de fabrication du semi-conducteur, un test électrique, appelé test paramétrique (PT), est effectuée. PT vise à détecter les plaques dont le comportement électrique est anormal, en se basant sur un ensemble de paramètres électriques statiques mesurées sur plusieurs sites de chaque plaque. Le but de ce travail est de mettre en place un système de détection dynamique au niveau de PT, pour détecter les plaques anormales à partir d'un historique récent de mesures électriques. Pour cela, nous développons un système de détection en temps réel basé sur une technique de réapprentissage optimisée, où les données d'apprentissage et le modèle de détection sont mis à jour à travers une fenêtre temporelle glissante. Le modèle de détection est basé sur les machines à vecteurs supports à une classe (1-SVM), une variante de l'algorithme d'apprentissage statistique SVM largement utilisé pour la classification binaire. 1-SVM a été introduit dans le cadre des problèmes de classification à une classe pour la détection des anomalies. Pour améliorer la performance prédictive de l'algorithme de classification 1-SVM, deux méthodes de sélection de variables ont été développées. La première méthode de type filtrage est basé sur un score calculé avec le filtre MADe,une approche robuste pour la détection univariée des valeurs aberrantes. La deuxième méthode de type wrapper est une adaptation à l'algorithme 1-SVM de la méthode d'élimination récursive des variables avec SVM (SVM-RFE). Pour les plaques anormales détectées, nous proposons une méthode permettant de déterminer leurs signatures multidimensionnelles afin d'identifier les paramètres électriques responsables de l'anomalie. Finalement, nous évaluons notre système proposé sur des jeux de données réels de STMicroelecronics, et nous le comparons au système de détection basé sur le test de T2 de Hotelling, un des systèmes de détection les plus connus dans la littérature. Les résultats obtenus montrent que notre système est performant et peut fournir un moyen efficient pour la détection en temps réel. / Nowadays, control of manufacturing process is an essential task to ensure production of high quality. At the end of the semiconductor manufacturing process, an electric test, called Parametric Test (PT), is performed. The PT aims at detecting wafers whose electrical behavior is abnormal, based on a set of static electrical parameters measured on multiple sites of each wafer. The purpose of this thesis is to develop a dynamic detection system at PT level to detect abnormal wafers from a recent history of electrical measurements. For this, we develop a real time detection system based on an optimized learning technique, where training data and detection model are updated through a moving temporal window. The detection scheme is based on one class Support Vector Machines (1-SVM), a variant of the statistical learning algorithm SVM widely used for binary classification. 1-SVM was introduced in the context of one class classification problems for anomaly detection. In order to improve the predictive performance of the 1-SVM classification algorithm, two variable selection methods are developed. The first one is a filter method based on a calculated score with MADe filter, a robust approach for univariate outlier detection. The second one is of wrapper type that adapts the SVM Recursive Feature Elimination method (SVM-RFE) to the 1-SVM algorithm. For detected abnormal wafers, we propose a method to determine their multidimensional signatures to identify the electrical parameters responsible for the anomaly. Finally, we evaluate our proposed system on real datasets of STMicroelecronics and compare it to the detection system based on Hotelling's T2 test, one of the most known detection systems in the literature. The results show that our system yields very good performance and can provide an efficient way for real-time detection.

De nos jours, l’échange électronique est le seul média qui offre l’accès à l’information pour tous, partout et tout le temps, mais en même temps il s’est ouvert à de nouvelles formes de vulnérabilités. La régulation des systèmes numériques, en héritage de la régulation cybernétique, maintient les équilibres à l’aide d’une boucle de rétroaction négative. Ainsi, leurs sys-tèmes de défense, désignés sous le terme de zone démilitarisée (DMZ) suivent-ils une régulation cybernétique en émettant ce que l’on appelle des évènements de sécurité. De tels évènements sont issus de sondes de surveillance qui matérialisent la ligne de dé-fense du système régulé. Toutefois, de telles sondes sont des système-experts et ces évènements appris au préalable ne rendent pas toujours compte de la dynamique de l’environnement et plus encore de la psychologie des individus. Plus encore, la multi-plication des systèmes de surveillance a entrainé une production considérable de ces évènements rendant cet ensemble de plus en plus inefficace. Par ailleurs, les systèmes vivants obéissent à une régulation complexe, l’homéostasie, qui les guide dans l’incertain à l’aide de mécanismes de surveillance continue. La force de tels mécanismes repose sur la variété des points de vue qu’ils empruntent ce qui leur permet de conjuguer leurs connaissances préalables à leurs informations de contexte pour comprendre leur environnement et s’adapter. Dans notre thèse, nous proposons d’associer à chaque système communicant, un sys-tème de surveillance continue : Dangerousness Incident Management (DIM) qui rend compte des changements de l’environnement en collectant et analysant toutes les traces laissées par les activités des usagers ou systèmes, légitimes ou non ; de cette manière, un tel système accède à une information étendue et reste sensible à son contexte. Néan-moins, plusieurs difficultés surviennent liées à la compréhension des informations re-cueillies dont le sens est noyé dans une grande masse d’informations, elles sont deve-nues implicites. Notre contribution principale repose sur un mécanisme de fouille de données adapté aux informations implicites. Nous proposons une structure à fort pou-voir d’abstraction fondée sur le principe d’un treillis de concepts. À partir de ce modèle de référence adaptatif, il nous est possible de représenter tous les acteurs d’un échange afin de faire coopérer plusieurs points de vue et plusieurs systèmes, qu’ils soient hu-mains ou machine. Lorsque l’incertitude de ces situations persiste, nous proposons un mécanisme pour guider l’usager dans ses décisions fondé sur le risque et la confiance. Enfin, nous évaluons nos résultats en les comparant aux systèmes de références Com-mon Vulnerabilities and Exposures (CVE) proposés par le National Institute of Stan-dards and Technology (NIST). / Nowadays, the electronic form of exchanges offers a new media able to make easy all information access, ubiquitous access, everywhere and everytime. But, at the same time, such a media - new, opened and complex - introduces unknown threats and breaches. So, how can we start up trust exchanges? From the system theory point of view, the cybernetic regulation maintains the sys-tems equilibrium with negative feedback loops. In this way, the defense line is based on a set of defense components still named Demilitarized Zone (DMZ) in order to block flow, to control anomalies and give out alerts messages if deviances are detected. Nev-ertheless, most of these messages concern only anomalies of machines and very little of human. So, messages do not take into account neither psychological behavior nor the dynamic of the context. Furthermore, messages suffer of the "big data" problem and become confused due to too much velocity, volume and variety. Finally, we can limit this problem to the understanding difficulty during the access to the specific knowledge in connection with the message. For example, the identity theft with the XSS attack is an illustration of this unfriendly environment. On the contrary, the living sciences show that organisms follow a positive regulation by where each one itself adapts according to his complexity. For that, they deploy adapted and continuous environment monitoring process still named "homeostasis". During this cycle, inputs capture information, then outputs adjust in response corre-sponding actions : this is the feedback. The strength of such a mechanism lies on the information meaning and in particular on the clues they include. In fact, some of these information include clues by which organisms can explain situations. For example, the information « attention" alludes to dangerous situation. This faculty comes from ad-vanced knowledge having first explicit relationship with this information: this relation forms what we call the "cognitive loop". To illustrate this phenomenon, the cognitive sciences often evoke "a friend immediately recognized by her friend" despite he is swal-lowed up in the crowd. But, the cognitive loop should not be broken. Like the living beings functioning, our work propose a cognitive model named Diag-nostic And Incident Model (DIM). The main idea lies on the context-aware model in order to adapt itself like "homeostasis". DIM has been founded on the principle of the "cognitive loop" where the inputs are the "logs" of numerics systems. So, in order to make easier the comparison between contextual and known situation, we will design "logs" and advanced knowledge by a common model. DIM proposes a conceptual struc-ture to extract clues from massive and various "logs” issued from environment based on advanced knowledge acquisition. Then, we propose the cognitive structure will be applied to the anomaly detection, incident management and diagnosis process.