Made available in DSpace on 2014-06-11T19:27:14Z (GMT). No. of bitstreams: 0 Previous issue date: 2011-12-21Bitstream added on 2014-06-13T20:35:24Z : No. of bitstreams: 1 fernandesjunior_ja_me_ilha.pdf: 4603291 bytes, checksum: 36699db66d480afb9c0cd31e5f7b4427 (MD5) / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / Os métodos de diagnose de falhas buscam evitar que estas ocorram sem predição, de modo que medidas preventivas possam ser tomadas para garantir a integridade física dos equipamentos e pessoas envolvidas no processo em que o sistema monitorado está inserido. Os eixos rotativos são elementos passíveis de fenômenos prejudiciais causados pela própria condição de movimento, que são refletidas na ciclagem das tensões atuantes em seu material. Este trabalho propõe a detecção em tempo real de trincas em eixos rotativos, pelo monitoramento do sistema em operação, usando a metodologia dos observadores de estados. Inicialmente, o projeto de observadores de estado via Regulador Linear Quadrático e Filtro de Kalman é introduzido. Em seguida, os observadores de estados com entradas desconhecidas são apresentados. Os modelos numéricos são obtidos por discretização através do Método dos Elementos Finitos e pelo ajuste da resposta em freqüência. A validação numérica do método proposto é feita pela identificação de alterações de parâmetros de rigidez de um sistema massa-mola-amortecedor e de trincas simuladas em um eixo rotativo numérico através de um modelo de falha empírico. A validação experimental da técnica proposta é obtida pela identificação de um entalhe transversal causado no eixo de um sistema rotativo experimental / The faults diagnosis methods seeks to avoid them not to occur without prediction, so that preventive tasks can be taken to ensure the physical integrity of equipments and people involved in the process where the monitored system is inserted. The rotating shafts are elements capable of harmful phenomena caused by its own movement condition, which are reflected on the cycling of the acting stresses on its material. This work proposes the real-time structural crack detection in rotating shafts, by the monitoring of the system in operation, using the methodology of the state observers. Initially, the project of state observers via Linear Quadratic Regulator and Kalman Filter are introduced. Then, the unknown input state observers are presented. The numerical models are obtained by discretization through the Finite Elements Method and the frequency response fitting. The numerical validation of the proposed method if verified by the identification of stiffness parameters change of a mass-spring-damper system and flaws simulated in a numerical rotating shaft using an empirical model of failure. The experimental validation is obtained by the identification of a transversal dent caused in the shaft of an experimental rotating system

Orientador: Gilberto Pechoto de Melo / Coorientador: Gustavo Luiz Chagas Manhães de Abreu / Banca: Luiz de Paula do Nascimento / Banca: Kátia Luchese Cavalca Dedini / Resumo: Os métodos de diagnose de falhas buscam evitar que estas ocorram sem predição, de modo que medidas preventivas possam ser tomadas para garantir a integridade física dos equipamentos e pessoas envolvidas no processo em que o sistema monitorado está inserido. Os eixos rotativos são elementos passíveis de fenômenos prejudiciais causados pela própria condição de movimento, que são refletidas na ciclagem das tensões atuantes em seu material. Este trabalho propõe a detecção em tempo real de trincas em eixos rotativos, pelo monitoramento do sistema em operação, usando a metodologia dos observadores de estados. Inicialmente, o projeto de observadores de estado via Regulador Linear Quadrático e Filtro de Kalman é introduzido. Em seguida, os observadores de estados com entradas desconhecidas são apresentados. Os modelos numéricos são obtidos por discretização através do Método dos Elementos Finitos e pelo ajuste da resposta em freqüência. A validação numérica do método proposto é feita pela identificação de alterações de parâmetros de rigidez de um sistema massa-mola-amortecedor e de trincas simuladas em um eixo rotativo numérico através de um modelo de falha empírico. A validação experimental da técnica proposta é obtida pela identificação de um entalhe transversal causado no eixo de um sistema rotativo experimental / Abstract: The faults diagnosis methods seeks to avoid them not to occur without prediction, so that preventive tasks can be taken to ensure the physical integrity of equipments and people involved in the process where the monitored system is inserted. The rotating shafts are elements capable of harmful phenomena caused by its own movement condition, which are reflected on the cycling of the acting stresses on its material. This work proposes the real-time structural crack detection in rotating shafts, by the monitoring of the system in operation, using the methodology of the state observers. Initially, the project of state observers via Linear Quadratic Regulator and Kalman Filter are introduced. Then, the unknown input state observers are presented. The numerical models are obtained by discretization through the Finite Elements Method and the frequency response fitting. The numerical validation of the proposed method if verified by the identification of stiffness parameters change of a mass-spring-damper system and flaws simulated in a numerical rotating shaft using an empirical model of failure. The experimental validation is obtained by the identification of a transversal dent caused in the shaft of an experimental rotating system / Mestre

L’objectif de cette thèse était de valider l’ensemble des informations délivrées par les capteurs utiles à la commande d’une station de traitement des eaux usées. Pour cela, nous avons utilisé l’analyse en composantes principales (ACP) pour effectuer la détection et localisation de défauts de capteurs de la station de traitement des eaux usées. Afin de construire un modèle ACP, nous avons eu recours à une matrice de données constituée de l’ensemble des mesures disponibles (obtenues lors du fonctionnement normal de la station de traitement des eaux usées) dans l’installation. Cependant, afin d’appliquer l’ACP, nous avons rencontré plusieurs difficultés : 1. Présence dans les données de valeurs aberrantes (valeurs obtenues durant des périodes de démarrage, d’arrêt, de fonctionnement dégradé, erreurs de mesure, ...) perturbant la construction d’un modèle ACP. 2. Présence de défauts multiples, ce qui entraîne une explosion combinatoire des scénarii de défauts à considérer. Afin de résoudre le premier point, nous nous sommes intéressé aux variantes robustes de l’ACP. L’estimateur robuste MCD (Minimum Covariance Determinant), méthode de référence pour ses performances, nécessite un temps de calcul important, et une connaissance a priori de la quantité de valeurs aberrantes présente dans les données (inconnue). C’est la raison pour laquelle nous avons proposé une nouvelle méthode robuste, basée sur l’utilisation de MM-estimateur, nommée MMRPCA (MM-estimator Robust Principal Component Analysis). Concernant le point 2, une méthode d’analyse du modèle en terme de capacité de détection et de localisation a été appliquée afin de réduire le nombre de défauts à considérer. Les différentes méthodes développées ont été menées avec succès afin de valider les mesures issues des différents capteurs de la station d’épuration des eaux usées / This thesis deals with the validation of the information provided by the sensors to the control of a wastewater treatment plant. For this purpose, Principal Component Analysis (PCA) approach is used in order to accomplish sensor fault detection and isolation of the wastewater treatment plant. This approach is well adapted to cope with diagnosis of complex systems because no a priori theoretical model of the plant must be considered. A data matrix, obtained by taking into consideration the available measurements in normal behaviour of the wastewater treatment plant, is used in order to build a PCA model. However, two major problems must be taking into consideration when PCA is implemented: 1. Outliers appear naturally in the collection data (caused for example by faulty data, data obtained during shutdown or startup periods or data issued from different operating mode) and consequently the PCA model can seriously be affected. 2. Multiple sensor faults introduce unavoidably a combinatory explosion of the different fault scenarios to be considered. The first problem is solved by introducing a robustness degree in the PCA methodology. Among the existing robust methods proposed in the literature, the robust estimator MCD (Minimum Covariance Determinant) is the most popular. However, this method needs a large computing time on the one hand and a priori knowledge of the quantity of outliers present (generally unknown) in the data on the other hand. To avoid these difficulties, a new robust method is proposed in this thesis. Our method, namely MMRPCA (MM-estimator Robust Principal Component Analysis), is based on MM-estimators. The second mentioned problem is tackled by reducing the considered number of faults thanks to a new analysis method of the capacities of detection and isolation of the PCA model. The efficiency of the proposed methodologies is verified by considering the real wastewater treatment plant data

O projeto VoIPFix surgiu da necessidade de uma ferramenta que complementasse as demais existentes no ramo de análise de redes de computadores para telefonia IP. Ele foi construído para ser uma ferramenta de gerenciamento eficiente e exclusiva para VoIP, com funcionalidades necessárias para dar suporte ao profissional de rede de computadores e telefonia IP a observar e diagnosticar problemas de VoIP. / The VoIPFix project arose from the need for a tool to complement similar tools in the analysis of computer networks for IP telephony. It was built to be an efficient and unique management tool for VoIP, with advanced features required to support the computer network and IP telephony professionals to observe and diagnose problems related to VoIP.

Orientador: Gilberto Pechoto de Melo / Banca: Vicente Lopes Júnior / Banca: Raquel Santini Leandro Rade / Resumo: Hoje em dia um dos fatores de interesse das indústrias no desenvolvimento de novas técnicas de detecção e localização de falhas é a preocupação com a segurança de seus sistemas, tendo-se a necessidade de supervisão e monitoramento de modo que as falhas sejam detectadas e corrigidas o mais rápido possível. Verifica-se na prática que determinados parâmetros dos sistemas podem variar durante o processo, devido a características específicas ou o desgaste natural de seus componentes. Sabe-se também que, mesmo nos sistemas bem projetados, a ocorrência de trincas em alguns componentes pode provocar perdas econômicas ou conduzir a situações perigosas. Os observadores de estado podem reconstruir os estados não medidos do sistema, desde que os mesmos sejam observáveis, tornando possível, desta forma, estimar as medidas nos pontos de difícil acesso. A técnica dos observadores de estado consiste em desenvolver um modelo para o sistema em análise e comparar a estimativa da saída com a saída medida, a diferença entre os dois sinais presentes resulta em um resíduo que é utilizado para análise. Neste trabalho foi montado um banco de observadores associado a um modelo de trinca de modo a acompanhar o progresso da mesma. Os resultados obtidos através de simulações computacionais em uma placa engastada discretizada pela técnica dos elementos finitos e as análises experimentais realizadas foram bastante satisfatórios, validando a metodologia desenvolvida. / Abstract: Nowadays a main factor of interest in industries in the development of new techniques for detection and localization of faults is the concern with the security of its systems. There is the need of supervising and monitoring the machines to detect and correct the fault as soon as possible. In practice it is verified that some determined parameters of the systems can vary during the process, due to the specific characteristics or the natural wearing of its components. It is known that even in well-designed systems the occurrence of cracks in some components can induce economic losses or lead to dangerous situations. The state observers methodology can reconstruct the unmeasured states of the system, since they are observable, becoming possible in this way to estimate the measures at points of difficult access. The technique of state observers consists of developing a model for the system under analysis and to compare the estimated with the measured exit, and the difference between these two signals results in a residue that is used for analysis. In this work was set up a bank of observers associated to a model of crack in order to follow its progress. The results obtained through computational simulations in a cantilever plate discretized by using the finite elements technique and the accomplished experimental analysis were sufficiently satisfactory, validating the developed methodology. / Mestre

O projeto VoIPFix surgiu da necessidade de uma ferramenta que complementasse as demais existentes no ramo de análise de redes de computadores para telefonia IP. Ele foi construído para ser uma ferramenta de gerenciamento eficiente e exclusiva para VoIP, com funcionalidades necessárias para dar suporte ao profissional de rede de computadores e telefonia IP a observar e diagnosticar problemas de VoIP. / The VoIPFix project arose from the need for a tool to complement similar tools in the analysis of computer networks for IP telephony. It was built to be an efficient and unique management tool for VoIP, with advanced features required to support the computer network and IP telephony professionals to observe and diagnose problems related to VoIP.

To allow monitoring and control of chemical processes, a sensor network has to be installed. It must allow the estimation of all important variables of the process. However, all measurements are erroneous, it is not possible to measure every variable and some types of sensors are expensive. Data reconciliation allows to correct the measurements, to estimate the values of unmeasured variables and to compute a posteriori uncertainties of all variables. However, a posteriori standard deviations are function of the number, the location and the precision of the measurement tools that are installed. A general method to design the cheapest sensor network able to estimate all process key variables within a prescribed accuracy in the case of steady-state processes has been developed. That method uses a posteriori variances estimation method based on the analysis of the sensitivity matrix. The goal function of the optimization problem depends on the annualized cost of the sensor network and on the accuracies that can be reached for the key variables. The problem is solved by means of a genetic algorithm. To reduce the computing time, two parallelization techniques using the message passing interface have been examined: the global parallelization and the distributed genetic algorithms. Both methods have been tested on several examples. To extend the method to dynamic processes, a dynamic data reconciliation method allowing to estimate a posteriori variances was necessary. Kalman filtering approach and orthogonal collocation-based moving horizon method have been compared. A posteriori variances computing has been developed using a similar method than the one used for the steady-state case. The method has been reconciled on several small examples. On the basis of the variances estimation an observability criterion has been defined for dynamic systems so that the sensor network design algorithm could be modified for the dynamic case. Another problem that sensor networks have to allow to solve is process faults detection and localisation. The method has been adapted to generate sensor networks that allow to detect and locate process faults among a list of faults in the case of steady-state processes./ Afin de permettre le suivi et le contrôle des procédés chimiques, un réseau de capteurs doit être installé. Il doit permettre l'estimation de toutes les variables importantes du procédé. Cependant, toutes les mesures sont entachées d'erreurs, toutes les variables ne peuvent pas être mesurées et certains types de capteurs sont onéreux. La réconciliation de données permet de corriger les mesures, d'estimer les valeurs des variables non mesurées et de calculer les incertitudes a posteriori de toutes les variables. Cependant, les écarts-types a posteriori sont fonction du nombre, de la position et de la précision des instruments de mesure qui sont installés. Une méthode générale pour réaliser le design du réseau de capteur le moins onéreux capable d'estimer toutes les variables clés avec une précision déterminée dans le cas des procédés stationnaires a été développée. Cette méthode utilise une technique d'estimation des variances a posteriori basée sur l'analyse de la matrice de sensibilité. La fonction objectif du problème d'optimisation dépend du coût annualisé du réseau de capteurs et des précisions qui peuvent être obtenues pour les variables clés. Le problème est résolu au moyen d'un algorithme génétique. Afin de réduire le temps de calcul, deux techniques de parallélisation utilisant une interface de passage de messages (MPI) ont été examinées: la parallélisation globale et les algorithmes génétiques distribués. Les deux méthodes ont été testées sur plusieurs exemples. Afin d'étendre la méthode aux procédés fonctionnant de manière dynamique, une méthode de réconciliation dynamique des données permettant le calcul des variances a posteriori est nécessaire. La méthode des filtres de Kalman et une technique de fenêtre mobile basée sur les collocations orthogonales ont été comparées. Le calcul des variances a posteriori a été développé grâce à une méthode similaire à celle utilisée dans le cas stationnaire. La méthode a été validée sur plusieurs petits exemples. Grâce à la méthode d'estimation des variances a posteriori, un critère d'observabilité a été défini pour les systèmes dynamiques de sorte que l'algorithme de design de réseaux de capteurs a pu être adapté aux systèmes dynamiques. Un autre problème que les réseaux de capteurs doivent permettre de résoudre est la détection et la localisation des erreurs de procédé. La méthode a été adaptée afin de générer des réseaux de capteurs permettant de détecter et de localiser les erreurs de procédé parmi une liste d'erreurs dans le cas des procédés fonctionnant de manière stationnaire.

Cette thèse a pour but l’étude de la détection basée sur modèle de défauts pour lesmoteurs Diesel produits en grande série. La nécessité d’une surveillance continue del’état de santé des véhicules est maintenant renforcée par la législation Euro VI sur lesémissions polluantes, qui sera probablement rendue encore plus contraignante dans sesprochaines révisions. Dans ce contexte, le développement de stratégies robustes, faciles àcalibrer et valides pour des systèmes dispersés (car produits en grande série) procureraitun avantage considérable aux constructeurs automobile. L’étude développée ici tentede répondre à ces besoins en proposant une méthodologie générique. On utilise desobservateurs adaptatifs locaux pour des systèmes scalaires non linéaires et affines parrapport à l’état, pour résoudre les problèmes de la détection de défauts, de son isolationet de son estimation d’une façon compacte. De plus, les incertitudes liées aux biais demesure et de modèle et aux dérives temporelles nécessitent d’améliorer les méthodes dedétection par l’utilisation de seuils robustes pour éviter les fausses détections. Dans cettethèse, on propose un seuil variable basé sur la condition d’observabilité du paramètreimpacté par le défaut et sur une étude de sensibilité par rapport aux incertitudes surles entrées ou sur le modèle. Cette méthode permet, entre autres, de fournir un outild’analyse pour la sélection des conditions de fonctionnement du système pour lesquelsle diagnostic est plus fiable et plus robuste par rapport aux incertitudes sur les entrées.L’approche présentée a été appliquée avec succès et validée de façon expérimentale surun moteur Diesel pour le problème de détection de fuite dans le système d’admissiond’air, puis dans un environnement de simulation pour le problème de détection dedérive d’efficacité turbine. On montre ainsi ses avantages en termes de fiabilité dedétection, d’effort de calibration, et pour l’analyse des conditions de fonctionnementmoteur adaptées au diagnostic. / The study of model-based fault detection for mass production Diesel engines isthe aim of this thesis. The necessity of continuous vehicles health monitoring is nowenforced by the Euro VI pollutant legislation, which will probably be tightened in itsfuture revisions. In this context developing a robust strategy that could be easilycalibrated and work with different systems (due to production variability) would bea tremendous advantage for car manufacturers. The study developed here tries toanswer to those necessities by proposing a generic methodology based on local adaptiveobservers for scalar nonlinear state-affine systems. The fault detection, isolation andestimation problems are thus solved in a compact way. Moreover, the uncertaintiesdue to measurement or model biases and time drifts lead to the necessity of improvingthe detection methodology by the use of robust thresholds that could avoid undesiredfalse alarms. In this thesis a variable threshold is proposed based on the observabilitycondition and the sensitivity analysis of the parameter impacted by the fault withrespect to input or model uncertainties. This approach allows, among other things, tobe used as an analysis tool for the individuation of the system operating points for whichthe diagnosis is more reliable and more robust to inputs uncertainties. The discussedapproach has been successfully implemented and experimentally tested on a real Dieselengine for the intake leak detection and for the turbine efficiency loss drift detectionin a co-simulation environment showing its advantages in term of detection reliability,calibration effort and engines diagnosis operating condition analysis.

Made available in DSpace on 2014-06-11T19:27:14Z (GMT). No. of bitstreams: 0 Previous issue date: 2008-04-19Bitstream added on 2014-06-13T18:55:45Z : No. of bitstreams: 1 zacarias_at_me_ilha.pdf: 587117 bytes, checksum: 4a907603292d52ff41cf7df3b7718be5 (MD5) / Hoje em dia um dos fatores de interesse das indústrias no desenvolvimento de novas técnicas de detecção e localização de falhas é a preocupação com a segurança de seus sistemas, tendo-se a necessidade de supervisão e monitoramento de modo que as falhas sejam detectadas e corrigidas o mais rápido possível. Verifica-se na prática que determinados parâmetros dos sistemas podem variar durante o processo, devido a características específicas ou o desgaste natural de seus componentes. Sabe-se também que, mesmo nos sistemas bem projetados, a ocorrência de trincas em alguns componentes pode provocar perdas econômicas ou conduzir a situações perigosas. Os observadores de estado podem reconstruir os estados não medidos do sistema, desde que os mesmos sejam observáveis, tornando possível, desta forma, estimar as medidas nos pontos de difícil acesso. A técnica dos observadores de estado consiste em desenvolver um modelo para o sistema em análise e comparar a estimativa da saída com a saída medida, a diferença entre os dois sinais presentes resulta em um resíduo que é utilizado para análise. Neste trabalho foi montado um banco de observadores associado a um modelo de trinca de modo a acompanhar o progresso da mesma. Os resultados obtidos através de simulações computacionais em uma placa engastada discretizada pela técnica dos elementos finitos e as análises experimentais realizadas foram bastante satisfatórios, validando a metodologia desenvolvida. / Nowadays a main factor of interest in industries in the development of new techniques for detection and localization of faults is the concern with the security of its systems. There is the need of supervising and monitoring the machines to detect and correct the fault as soon as possible. In practice it is verified that some determined parameters of the systems can vary during the process, due to the specific characteristics or the natural wearing of its components. It is known that even in well-designed systems the occurrence of cracks in some components can induce economic losses or lead to dangerous situations. The state observers methodology can reconstruct the unmeasured states of the system, since they are observable, becoming possible in this way to estimate the measures at points of difficult access. The technique of state observers consists of developing a model for the system under analysis and to compare the estimated with the measured exit, and the difference between these two signals results in a residue that is used for analysis. In this work was set up a bank of observers associated to a model of crack in order to follow its progress. The results obtained through computational simulations in a cantilever plate discretized by using the finite elements technique and the accomplished experimental analysis were sufficiently satisfactory, validating the developed methodology.

Les fautes systèmes peuvent conduire à des conséquences sérieuses pour l’humain, l’environnement et le matériel. Or, y remédier peut s’avérer coûteux voire même dangereux. Ainsi, afin d’éviter ces situations, il est devenu essentiel pour les systèmes complexes modernes de détecter et d’identifier tout changement dans leur fonctionnement nominal avant que cela ne devienne critique. De ce fait, plusieurs méthodes de détection et de diagnostic ont été proposées ou améliorées durant les dernières décennies. Parmi ces méthodes, celles présentant un fort intérêt se basent sur un outil statistique et probabiliste nommé réseau Bayésien. Toutefois, la majorité d’entre elles ne tiennent pas compte du risque de fausse alarme dans leur prise de décision. L’intérêt de cette thèse est alors d’introduire sous réseau Bayésien des limites probabilistes permettant le respect d’un niveau de signification considéré. Plus exactement, nous proposons une modélisation des statistiques quadratiques et les limites leurs correspondant sur réseau Bayésien. Ceci nous permet de généraliser sous réseau Bayésien des schémas de détection de fautes comme par exemple ceux basés sur l’analyse en composantes principale. Cette modélisation nous permet également de proposer une famille de réseaux Bayésiens permettant de faire de la détection et du diagnostic de façon simultanée, tout en tenant compte d’un rejet de distance. Enfin, nous proposons un cadre probabiliste permettant d’unifier les différents réseaux Bayésiens pouvant être utilisés pour la détection ou le diagnostic de fautes. / Systems failures can potentially lead to serious consequences forhuman, environment and material, and sometimes fixing them could be expensive and even dangerous. Thus, in order to avoid these undesirable situations, it becomes very important and essential for modern complex systems to detect and identify any changes in their nominal operations before they become critical. To do so, several detection and diagnosis methods have been proposed or enhanced during the last decades. Among these methods, those with a great interest are based on a statistical and probabilistic tool named Bayesian network. However, the majority of these methods do not handle the risk of false alarm in their decision-making. The interest of this thesis is to introduce, under Bayesian network, probabilistic limits able to respect a given significance level. More precisely, we propose to model the quadratic statistics and their limits in Bayesian network. This allows us to generalize under Bayesian network fault detection schemes as those associated to the principal component analysis. This modeling allows us also to propose a family of Bayesian networks that can make detection and diagnosis simultaneously, while taking into account the distance rejection.Finally, we propose a probabilistic framework able to unify different BNs dedicated to the detection or diagnosis of systems faults.