Cette thèse présente une méthode pour désagréger les appareils électriques dans le profil des bâtiments résidentiels de charge. Au cours des dernières années, la surveillance de l’énergie a obtenu beaucoup de popularité dans un environnement privé et industriel. Avec des algorithmes de la désagrégation, les données mesurées à partir de soi-disant compteurs intelligents peuvent être utilisés pour fournir de plus amples informations de la consommation d’énergie. Une méthode pour recevoir ces données est appelé non-intrusifs charge identification. La majeure partie de la thèse peut être divisée en trois parties. Dans un premier temps, un système de mesure propre a été développé et vérifié. Avec ce système, les ensembles de données réelles peuvent être générés pour le développement et la vérification des algorithmes de désagrégation. La deuxième partie décrit le développement d’un détecteur de flanc. Différentes méthodes sont présentées et évaluées, avec lequel les temps de commutation des appareils peuvent être détectés dans le profil de la charge. La dernière partie décrit un procédé de classification. Différents critères sont utilisés pour la classification. Le classificateur reconnaît et étiquette les appareils individuels de la courbe de charge. Pour les classifications différentes structures de réseaux de neurones artificiels sont comparés. / This thesis presents a method to disaggregate electrical appliances in the load profile of residential buildings. In recent years, energy monitoring has obtained significantly popularity in private and industrial environment. With algorithms of the disaggregation, the measured data from so-called smart meters can be used to provide more information of the energy usage. One method to receive these data is called non-intrusive appliance load monitoring.The main part of the thesis can be divided into three parts. At first, an own measurement system was developed and verified. With that system, real data sets can be generated for the development and verification of the disaggregation algorithms. The second part describes the development of an event detector. Different methods are presented and evaluated, with which the switching times of the appliances can be detected in the load profile. The last part describes a classification method. Different features are used for the classification. The classifier recognizes and labels the individual appliances in the load profile. For the classification different structures of artificial neural network (ANN) are compared.

Wireless sensor networks are application-specific networks that necessitate the development of specific network and information processing architectures that can meet the requirements of the applications involved. The most important challenge related to wireless sensor networks is the limited energy and computational resources of the battery powered sensor nodes. Although the central processing of information produces the most accurate results, it is not an energy-efficient method because it requires a continuous flow of raw sensor readings over the network. As communication operations are the most expensive in terms of energy usage, the distributed processing of information is indispensable for viable deployments of applications in wireless sensor networks. This method not only helps in reducing the total amount of packets transmitted and the total energy consumed by sensor nodes, but also produces scalable and fault-tolerant networks. Another important challenge associated with wireless sensor networks is that the possibility of sensory data being imperfect and imprecise is high. The requirement of precision necessitates employing expensive mechanisms such as redundancy or use of sophisticated equipments. Therefore, approximate computing may need to be used instead of precise computing to conserve energy. This thesis presents two schemes that distribute information processing for event-driven reactive applications, which are interested in higher-level information not in the raw sensory data of individual nodes, to appropriate nodes in sensor networks. Furthermore, based on these schemes, a fuzzy rule-based system is proposed that handles imprecision, inherently present in sensory data.

Detecting human actions using a camera has many possible applications in the security industry. When a human performs an action, his/her body goes through a signature sequence of poses. To detect these pose changes and hence the activities performed, a pattern recogniser needs to be built into the video system. Due to the temporal nature of the patterns, Hidden Markov Models (HMM), used extensively in speech recognition, were investigated. Initially a gesture recognition system was built using novel features. These features were obtained by approximating the contour of the foreground object with a polygon and extracting the polygon's vertices. A Gaussian Mixture Model (GMM) was fit to the vertices obtained from a few frames and the parameters of the GMM itself were used as features for the HMM. A more practical activity detection system using a more sophisticated foreground segmentation algorithm immune to varying lighting conditions and permanent changes to the foreground was then built. The foreground segmentation algorithm models each of the pixel values using clusters and continually uses incoming pixels to update the cluster parameters. Cast shadows were identified and removed by assuming that shadow regions were less likely to produce strong edges in the image than real objects and that this likelihood further decreases after colour segmentation. Colour segmentation itself was performed by clustering together pixel values in the feature space using a gradient ascent algorithm called mean shift. More robust features in the form of mesh features were also obtained by dividing the bounding box of the binarised object into grid elements and calculating the ratio of foreground to background pixels in each of the grid elements. These features were vector quantized to reduce their dimensionality and the resulting symbols presented as features to the HMM to achieve a recognition rate of 62% for an event involving a person writing on a white board. The recognition rate increased to 80% for the &quotseen" person sequences, i.e. the sequences of the person used to train the models. With a fixed lighting position, the lack of a shadow removal subsystem improved the detection rate. This is because of the consistent profile of the shadows in both the training and testing sequences due to the fixed lighting positions. Even with a lower recognition rate, the shadow removal subsystem was considered an indispensable part of a practical, generic surveillance system.

Este trabalho apresenta o desenvolvimento de um sensor de corrente transiente destinado a detectar a ocorrência de um evento transiente causado pela incidência de radiação ionizante em um circuito integrado. Iniciando com uma descrição dos efeitos da radiação sobre os circuitos integrados e dos tipos de radiação de interesse, os fundamentos da técnica Bulk- BICS são apresentados e as propostas existentes na literatura são expostas e avaliadas, com ênfase no sensor que utiliza a célula de memória dinâmica DynBICS, resultado de um trabalho prévio e do qual se dispõe de amostras fabricadas. Sobre essas amostras são efetuados testes elétricos, um ensaio de dose total irradiada TID e um ensaio de estimulação laser, cujos resultados são apresentados e confirmam a funcionalidade da topologia da célula de memória dinâmica aplicada a circuitos Bulk-BICS. Em seguida, é apresentada a topologia da célula de memória integrativa como uma evolução da célula de memória dinâmica e propõe-se o circuito de um novo sensor Bulk-BICS baseado na nova célula. O funcionamento elétrico do circuito desse novo sensor TRIBICS é avaliado através de simulação de circuitos determinando-se a sensibilidade e o tempo de resposta do sensor utilizando-se pulsos de corrente em dupla exponencial. É feita uma análise do funcionamento da célula de memória estática e, através de uma comparação de desempenho entre as células de memória estáticas utilizadas em três circuitos propostos e a célula de memória integrativa, utilizando um modelo simplificado, mostra-se que a célula de memória integrativa é mais rápida e sensível do que as contrapartes estáticas O sensor TRIBICS é então simulado em conexão com um modelo de dispositivo, sendo antes apresentados os modelos TCAD do inversor utilizado como alvo da incidência da radiação nas simulações. São apresentados resultados obtidos individualmente para o transistor NMOS e para o transistor PMOS, nos quais se mostra a formação de um canal condutivo entre dreno e fonte durante o SET. Mostra-se, também, que os resultados obtidos com a simulação de dispositivos não concorda com aqueles proporcionados pela simulação de circuitos no tocante à divisão das correntes transitórias entre dreno, fonte e substrato. O resultado das simulações de dispositivo efetuadas com os modelos TCAD em modo misto com o circuito TRIBICS descrito em SPICE mostram a relação entre a transferência de energia da irradiação LET e a efetiva deteção do SET provocado, em função da distância entre os contatos de bulk ou substrato, permitindo determinar a máxima distância entre contatos para 100% de certeza na deteção do SET. Com isso, obtém-se uma estimativa do número de transistores que pode ser monitorado pelos Bulk-BICS. É proposta a estratégia de implementação dos Bulk-BICS na forma de uma standard cell a ser posicionada entre os grupos de transistores sob monitoração, e uma estimativa da relação entre as áreas dos transistores monitorados e do Bulk-BICS é apresentada. Por fim, é estudada a questão da fabricação dos Bulk-BICS no mesmo substrato dos transistores monitorados e uma maneira de fazê-la é proposta. Os resultados encontrados permitem definir a viabilidade e a eficácia da técnica Bulk-BICS como forma de deteção de eventos transientes em sistemas digitais. / A current sensor to detect the occurrence of a single-event transient that is caused by the incidence of ionizing radiation in an integrated circuit is presented. Radiation of interest and their effects on the integrated circuits are discussed. Fundamentals of the Bulk-BICS technique and the circuits proposed in the literature to implement this technique are discussed and evaluated, with emphasis on the dynamic memory cell-based circuit DynBICS, which was developed as a previous work and with fabricated samples available. Experimental results obtained from a series of electrical tests, a TID test, and a laser-stimulated test that were conducted on a number of fabricated and packaged samples are presented. The results confirm that the dynamic memory cell is suitable and robust enough to be used in Bulk-BICS circuits. Next, evolution of the dynamic memory cell into an integrative memory cell is discussed and the circuit of a Bulk-BICS using this new memory cell topology is presented. The electrical operation of this new sensor TRIBICS is evaluated using circuit simulations. By using double-exponential current pulses, both the sensitivity and the response time are determined. The static memory cell operation is analyzed and a comparison of performance between static and integrative cells is performed using a simplified model. The results show that the integrative memory cell is faster and more sensitive than the static cells used in three state-ofthe- art sensors published in literature Then the TRIBICS sensor is simulated connected to a TCAD-modeled device, comprising an inverter, which is used as a target for radiation impact. TCAD models are previously presented and the results obtained when the PMOS and NMOS transistors are separately excited by radiation show the formation of a conductive link between drain and source regions during the occurrence of SET. The simulations also show that the results obtained by using TCAD simulations do not agree with the ones obtained by using circuit simulation regarding the current share among drain, source and bulk during the SET. Mixed-mode simulations using the TCAD models in conjunction of TRIBICS circuits described in SPICE show the relationship between LET and the effective SET-detection with the inter-tap distance as a parameter, and allows to determine the inter-tap distance for 100% of SET detection efficiency. Based on these results, an estimate of how many transistors can be monitored by the Bulk-BICS is obtained. It is proposed to implement the Bulk-BICS as a standard cell, to be positioned in between the standard cell that compose a digital circuit and the area overhead necessary to implant the sensors in a real circuit is estimated. The problem on how to manufacture the Bulk-BICS circuit in the same substrate of the monitored transistors is studied and a solution is proposed. The results show the viability and effectiveness of the Bulk-BICS technique, as a means to detect single-event transients in digital systems.

La maintenance d’un système complexe est souvent segmentée par sous-système. Chacun de sous-systèmes faisant intervenir des compétences pointues et variées, déterminer l’état de santé globale du système s’avère être une tâche compliquée. Cependant, les systèmes complexes sont aujourd’hui surveillés avec attention ce qui permet d’enregistrer un nombre important de données hétérogènes permettant à la maintenance d’être efficace sur chaque sous-systèmes. La variété de ces données permet d’avoir une vision d’ensemble sur l’état de santé du système mais du fait de leur quantité et de leur hétérogénéité leur analyse est un exercice complexe. Pour pallier aux problématiques de traitement de données de masse, ces dernières décennies ont vu se développer des outils informatiques et mathématiques permettant d’extraire des informations pertinentes d’un ensemble de données : le data mining. La mise en application d’outils issus des méthodes de data mining peut être une solution à l’identification de l’état santé globale du système.Pour rendre cela possible, la thèse présente dans un premier temps comment construire un modèle de comportement du système hélicoptère en se basant sur des données relatives au bon fonctionnement de l’appareil. Ce modèle de comportement considéré comme normal va ensuite servir de référence. Il permettra donc dans un deuxième temps de répondre à comment détecter et caractériser une déviance, un événement anormal, vis-à-vis du modèle du comportement normal du système. Cette méthode cherche à maximiser l’utilisation des données et minimiser l’introduction d’information relative à la connaissance du système. De cette manière, elle permet de fournir des résultats complètement objectifs qui pourront être comparés à des analyses physiques. La méthode est supportée par un ensemble d’outils mathématiques implémentés dans une infrastructure industrielle permettant l’utilisation de données réelles associées aux appareils d’Eurocopter.Pour accompagner la mise en place de cette méthode, cette thèse présente une application sur des données réelles issues d’hélicoptères de type EC225 de la gamme d’Eurocopter. / The maintenance of complex system is often segmented by subsystem. Each subsystem involving specialized and various skills, assess the global health of the system is a difficult problem. However, today, complex systems are carefully monitored that allows the records of a substantial amount of heterogeneous data that leads to accurate subsystem maintenance. Diversity of data provides an overview of the global health of the system but in the reason of quantity and heterogeneity their analyses is a difficult exercise. To tackle these data treatment difficulties, computational and mathematical tools have been developed. They allows extraction of relevant information in a substantial amount of data, it is the data mining. The implementation of data mining method could be a solution to the assessment of global health of the system.To make that possible, in a first time, this thesis present how define a helicopter behavior model by using data which are recorded in a good way of running. This behavior model considered as normal will be used as a reference. In a second time, this model will allow to answer to how detect and characterize a drift, an abnormal event, from the normal system behavior model. This method tries to maximize the data usage and minimize the expert knowledge. By this way, it provides results totally objective which could be compare to physical analyses. This approach is supported by a set of mathematical tools implemented in an industrial infrastructure which allows the use of Eurocopter aircraft operational data.To support the implementation of this method, this thesis presents an application of the method on real data from the EC225 helicopter of Eurocopter

Este trabalho apresenta o desenvolvimento de um sensor de corrente transiente destinado a detectar a ocorrência de um evento transiente causado pela incidência de radiação ionizante em um circuito integrado. Iniciando com uma descrição dos efeitos da radiação sobre os circuitos integrados e dos tipos de radiação de interesse, os fundamentos da técnica Bulk- BICS são apresentados e as propostas existentes na literatura são expostas e avaliadas, com ênfase no sensor que utiliza a célula de memória dinâmica DynBICS, resultado de um trabalho prévio e do qual se dispõe de amostras fabricadas. Sobre essas amostras são efetuados testes elétricos, um ensaio de dose total irradiada TID e um ensaio de estimulação laser, cujos resultados são apresentados e confirmam a funcionalidade da topologia da célula de memória dinâmica aplicada a circuitos Bulk-BICS. Em seguida, é apresentada a topologia da célula de memória integrativa como uma evolução da célula de memória dinâmica e propõe-se o circuito de um novo sensor Bulk-BICS baseado na nova célula. O funcionamento elétrico do circuito desse novo sensor TRIBICS é avaliado através de simulação de circuitos determinando-se a sensibilidade e o tempo de resposta do sensor utilizando-se pulsos de corrente em dupla exponencial. É feita uma análise do funcionamento da célula de memória estática e, através de uma comparação de desempenho entre as células de memória estáticas utilizadas em três circuitos propostos e a célula de memória integrativa, utilizando um modelo simplificado, mostra-se que a célula de memória integrativa é mais rápida e sensível do que as contrapartes estáticas O sensor TRIBICS é então simulado em conexão com um modelo de dispositivo, sendo antes apresentados os modelos TCAD do inversor utilizado como alvo da incidência da radiação nas simulações. São apresentados resultados obtidos individualmente para o transistor NMOS e para o transistor PMOS, nos quais se mostra a formação de um canal condutivo entre dreno e fonte durante o SET. Mostra-se, também, que os resultados obtidos com a simulação de dispositivos não concorda com aqueles proporcionados pela simulação de circuitos no tocante à divisão das correntes transitórias entre dreno, fonte e substrato. O resultado das simulações de dispositivo efetuadas com os modelos TCAD em modo misto com o circuito TRIBICS descrito em SPICE mostram a relação entre a transferência de energia da irradiação LET e a efetiva deteção do SET provocado, em função da distância entre os contatos de bulk ou substrato, permitindo determinar a máxima distância entre contatos para 100% de certeza na deteção do SET. Com isso, obtém-se uma estimativa do número de transistores que pode ser monitorado pelos Bulk-BICS. É proposta a estratégia de implementação dos Bulk-BICS na forma de uma standard cell a ser posicionada entre os grupos de transistores sob monitoração, e uma estimativa da relação entre as áreas dos transistores monitorados e do Bulk-BICS é apresentada. Por fim, é estudada a questão da fabricação dos Bulk-BICS no mesmo substrato dos transistores monitorados e uma maneira de fazê-la é proposta. Os resultados encontrados permitem definir a viabilidade e a eficácia da técnica Bulk-BICS como forma de deteção de eventos transientes em sistemas digitais. / A current sensor to detect the occurrence of a single-event transient that is caused by the incidence of ionizing radiation in an integrated circuit is presented. Radiation of interest and their effects on the integrated circuits are discussed. Fundamentals of the Bulk-BICS technique and the circuits proposed in the literature to implement this technique are discussed and evaluated, with emphasis on the dynamic memory cell-based circuit DynBICS, which was developed as a previous work and with fabricated samples available. Experimental results obtained from a series of electrical tests, a TID test, and a laser-stimulated test that were conducted on a number of fabricated and packaged samples are presented. The results confirm that the dynamic memory cell is suitable and robust enough to be used in Bulk-BICS circuits. Next, evolution of the dynamic memory cell into an integrative memory cell is discussed and the circuit of a Bulk-BICS using this new memory cell topology is presented. The electrical operation of this new sensor TRIBICS is evaluated using circuit simulations. By using double-exponential current pulses, both the sensitivity and the response time are determined. The static memory cell operation is analyzed and a comparison of performance between static and integrative cells is performed using a simplified model. The results show that the integrative memory cell is faster and more sensitive than the static cells used in three state-ofthe- art sensors published in literature Then the TRIBICS sensor is simulated connected to a TCAD-modeled device, comprising an inverter, which is used as a target for radiation impact. TCAD models are previously presented and the results obtained when the PMOS and NMOS transistors are separately excited by radiation show the formation of a conductive link between drain and source regions during the occurrence of SET. The simulations also show that the results obtained by using TCAD simulations do not agree with the ones obtained by using circuit simulation regarding the current share among drain, source and bulk during the SET. Mixed-mode simulations using the TCAD models in conjunction of TRIBICS circuits described in SPICE show the relationship between LET and the effective SET-detection with the inter-tap distance as a parameter, and allows to determine the inter-tap distance for 100% of SET detection efficiency. Based on these results, an estimate of how many transistors can be monitored by the Bulk-BICS is obtained. It is proposed to implement the Bulk-BICS as a standard cell, to be positioned in between the standard cell that compose a digital circuit and the area overhead necessary to implant the sensors in a real circuit is estimated. The problem on how to manufacture the Bulk-BICS circuit in the same substrate of the monitored transistors is studied and a solution is proposed. The results show the viability and effectiveness of the Bulk-BICS technique, as a means to detect single-event transients in digital systems.

Twitter est un réseau social très répandu en Amérique du Nord, offrant aux autorités policières une opportunité pour détecter les événements d’intérêt public. Les messages Twitter liés à un événement contiennent souvent les noms de rue où se déroule l’événement, ce qui permet la géolocalisation en temps réel. Plusieurs logiciels commerciaux sont offerts pour effectuer la vigie des réseaux sociaux. L’efficacité de ces outils pour les autorités policières pourrait être grandement améliorée avec un accès à un plus grand échantillon de messages Twitter, avec un tri préalable pour dégager les événements pertinents en moins de temps et avec une mesure de la fiabilité des événements détectés. Ce mémoire vise à proposer une démarche afin de détecter, à partir du flux de messages Twitter, les événements de sécurité publique d’un territoire, automatiquement et avec un niveau de fiabilité acceptable. Pour atteindre cet objectif, un modèle informatisé a été conçu, basé sur les quatre composantes suivantes: a) la cueillette de tweets à partir de mots clés avec un filtrage géographique, b) l’analyse linguistique et l’utilisation d’un répertoire de rues pour déceler les tweets localisables et pour trouver leurs coordonnées à partir des noms de rue et de leur intersection, c) une méthode spatio-temporelle pour former des grappes de tweets, et d) la détection des événements en identifiant les grappes contenant au moins deux (2) tweets communs touchant le même sujet. Ce travail de recherche diffère des articles scientifiques recensés car il combine l’analyse textuelle, la recherche et le géocodage de toponymes à partir d’un répertoire de noms de rue, la formation de grappes avec la géomatique et l’identification de grappes contenant des tweets communs pour détecter localement des événements de sécurité publique. L’application du modèle aux 90 347 tweets cueillis dans la région de Toronto-Niagara au Canada a résulté en l’identification et la géolocalisation de 1 614 tweets ainsi qu’en la formation de 172 grappes dont 79 grappes d’événements contenant au moins deux (2) tweets touchant le même sujet, soit un taux de fiabilité de 45,9 %. / Abstract : Twitter is a social media that is very popular in North America, giving law enforcement agencies an opportunity to detect events of public interest. Twitter messages (tweets) tied to an event often contain street names, indicating where this event takes place, which can be used to infer the event's geographical coordinates in real time. Many commercial software tools are available to monitor social media. The performance of these tools could be greatly improved with a larger sample of tweets, a sorting mechanism to identify pertinent events more quickly and to measure the reliability of the detected events. The goal of this master‟s thesis is to detect, from a public Twitter stream, events relative to public safety of a territory, automatically and with an acceptable level of reliability. To achieve this objective, a computer model based on four components has been developed: a) capture of public tweets based on keywords with the application of a geographic filter, b) natural language processing of the text of these tweets, use of a street gazetteer to identify tweets that can be localized and geocoding of tweets based on street names and intersections, c) a spatio-temporal method to form tweet clusters and, d) event detection by isolating clusters containing at least two tweets treating the same subject. This research project differs from existing scientific research as it combines natural language processing, search and geocoding of toponyms based on a street gazetteer, the creation of clusters using geomatics and identification of event clusters based on common tweets to detect public safety events in a Twitter public stream. The application of the model to the 90,347 tweets collected for the Toronto-Niagara region in Ontario, Canada has resulted in the identification and geocoding of 1,614 tweets and the creation of 172 clusters from which 79 event clusters contain at least two tweets having the same subject showing a reliability rate of 45.9 %.

This diploma thesis deals with automatic soccer event detection . Its goal is to introduce reader to this issue , discuss possible ways of solution of this task and then implement event detection . This work aims at event recognition using spatio - temporal data of gaming objects . Introduced way of dealing with event detection lies in its converting to sequence labeling task . Then such task is solved using LSTM recurrent neural networks . Lastly , result of sequence labeling is interpreted as detected events . Library for event detection has been created as the output of this work . This library allow user to experiment with different variants how to formulate event detection as sequence labeling task .

This thesis follows the trend of last decades in using neural networks in order to detect speech in noisy data. The text begins with basic knowledge about discussed topics, such as audio features, machine learning and neural networks. The network parameters are examined in order to provide the most suitable background for the experiments. The main focus of the experiments is to observe the influence of various sound events on the speech detection on a small, diverse database. Where the sound events correlated to the speech proved to be the most beneficial. In addition, the accuracy of the acoustic events, previously used only as a supplement to the speech, is also a part of experimentation. The experiment of examining the extending of the datasets by more fairly distributed data shows that it doesn't guarantee an improvement. And finally, the last experiment demonstrates that the network indeed succeeded in learning how to predict voice activity in both clean and noisy data.

No description available.