This thesis studies and develops an agent-based on-line monitoring and diagnosis system. With the advent of industry facilities toward high speed and automation, the on-line monitoring and diagnosis mechanism becomes important. The traditional time-based maintenance strategy cannot consider all practical conditions of machines. The accidental breakdowns may occur unexpectedly. In order to make production line automation, yield high production rate, and enhance the ability of market competition, an effective way is to ensure facilities running smoothly and to reduce the manpower monitoring. Meanwhile in order to reduce the cost of development and enhance the feature of reuse, we use a communication interface to integrate and interact information among developed softwares. Through this system, users may easily monitor and diagnose the machines on production lines through user interfaces, such as display of machine status, malfunction alarming, trend chart, waveform, spectrum, cepstrum and envelop spectrum, etc. A model consisting of domain knowledge and behaviors of cooperative multi-agents to integrate various facilities and functions in different production lines is proposed. In addition, via a Web-based environment, it releases the restrictions by of temporal and regional isolation. The system provides a real-time and dynamical operation circumstance and can meet the needs of different levels of users. Consequently, it is of greatly help to lower the cost of personnel and enhance the ability of market competition.

The gas turbine engine has a wide range of applications, these include industrial and aerospace applications on locomotive, ferry, compressor and power generation, and the most popular application will be for the air transportation. The application for air transportation including military and commercial aircraft is highly sensitive to safety concerns. The engine health monitoring system plays a major role for addressing this concern, a good engine monitoring system will not only to provide immediate and correct information to the engine user but also provide useful information for managing the maintenance activities. Without a reliable performance diagnosis module involved, there will be not possible to build a good health monitoring system. There are many methodologies had been proposed and studied during past three decades, and yet still struggling to search for some good techniques to handle instrumentation errors. In order to develop a reliable engine performance diagnosis technique, a fully understanding and proper handling of the instrumentation is essential. A engine performance fault pattern matching method has been proposed and developed in this study, two fault libraries contains a complete defined set of 51963 faults was created by using a newly serviced fighter engine component data. This pattern matching system had been verified by different approaches, such as compares with linear and nonlinear diagnosis results and compares with performance sensitivity analysis results by using LTF program engine data. The outcomes from the verications indicate an encouraging result for further exploring this method. In conclusion, this research has not only propose a feasible performance diagnosis techniques, but also developed and verified through different kind of approaches for this techniques. In addition to that, by proper manipulating the created fault library, a possible new tool for analyzing the application of instruments' implementation was discovered. The author believes there will be more to study by using this created fault pattern library. For instance, this fault pattern library can be treated as a very good initial training sets for neural networking to develop a neural diagnosis technique. This study has put a new milestone for further exploring gas turbine diagnosis technique by using fault pattern related methods.

Os estudos na área de Monitoração e Diagnóstico de Falhas têm sido estimulados devido ao aumento crescente em qualidade, confiabilidade e segurança nos processos de produção, onde a interrupção da produção por alguma anomalia imprevista pode colocar em risco a segurança do operador, além de provocar perdas econômicas, aumentando os custos com a reparação de algum equipamento danificado. Tendo em vista estes dois fatores, o fator econômico e a própria questão de segurança do operador, torna-se necessário a implementação de Sistemas de Monitoração e Detecção de Falhas. Neste trabalho foi desenvolvido um Sistema de Monitoração e Detecção de Falhas usando a metodologia de Redes Neurais Artificiais que foi aplicado ao reator de pesquisas IEA-R1. O desenvolvimento deste sistema foi dividido em três etapas: sendo a primeira etapa dedicada à monitoração, a segunda a detecção, e a terceira ao diagnóstico de falhas. Na primeira etapa, foram treinadas diversas Redes Neurais Artificiais para a monitoração das variáveis de temperatura, potência e taxa de dose. Para tanto foram utilizadas duas bases dados: uma contendo dados gerados por um modelo teórico do reator, e outra contendo dados referentes a uma semana típica de operação. Na segunda etapa, as redes treinadas para realizar a monitoração das variáveis, foram testadas com uma base de dados contendo falhas inseridas artificialmente nos sensores de temperatura. Como o limite máximo de erro de calibração para termopares especiais é de , foram inseridas falhas de ± nos sensores responsáveis pela leitura das variáveis T3 e T4. Na terceira etapa foi desenvolvido um Sistema Fuzzy para realizar o diagnóstico de falhas, onde foram consideradas 3 condições possíveis de falhas: condição normal, falha de −, e falha de , sendo que o sistema desenvolvido indicará qual o sensor de temperatura está com falha. Cº5,0±Cº1Cº1Cº1+ / The increasing demand on quality in production processes has encouraged the development of several studies on Monitoring and Diagnosis Systems in industrial plant, where the interruption of the production due to some unexpected change can bring risk to the operator\'s security besides provoking economic losses, increasing the costs to repair some damaged equipment. Because of these two points, the economic losses and the operator\'s security, it becomes necessary to implement Monitoring and Diagnosis Systems. In this work, a Monitoring and Diagnosis Systems was developed based on the Artificial Neural Networks methodology. This methodology was applied to the IEA-R1 research reactor at IPEN. The development of this system was divided in three stages: the first was dedicated to monitoring, the second to the detection and the third to diagnosis of failures. In the first stage, several Artificial Neural Networks were trained to monitor the temperature variables, nuclear power and dose rate. Two databases were used: one with data generated by a theoretical model and another one with data to a typical week of operation of the IEA-R1 reactor. In the second stage, the neural networks used to monitor the variables was tested with a fault database. The faults were inserted artificially in the sensors signals. As the value of the maximum calibration error for special thermocouples is , it had been inserted faults of in the sensors for the reading of the variables T3 and T4. In the third stage a Fuzzy System was developed to carry out the faults diagnosis, where were considered three conditions: a normal condition, a fault of , and a fault of . This system will indicate which thermocouple is faulty. Cº5,0±Cº1Cº1±−Cº1+

Os estudos na área de Monitoração e Diagnóstico de Falhas têm sido estimulados devido ao aumento crescente em qualidade, confiabilidade e segurança nos processos de produção, onde a interrupção da produção por alguma anomalia imprevista pode colocar em risco a segurança do operador, além de provocar perdas econômicas, aumentando os custos com a reparação de algum equipamento danificado. Tendo em vista estes dois fatores, o fator econômico e a própria questão de segurança do operador, torna-se necessário a implementação de Sistemas de Monitoração e Detecção de Falhas. Neste trabalho foi desenvolvido um Sistema de Monitoração e Detecção de Falhas usando a metodologia de Redes Neurais Artificiais que foi aplicado ao reator de pesquisas IEA-R1. O desenvolvimento deste sistema foi dividido em três etapas: sendo a primeira etapa dedicada à monitoração, a segunda a detecção, e a terceira ao diagnóstico de falhas. Na primeira etapa, foram treinadas diversas Redes Neurais Artificiais para a monitoração das variáveis de temperatura, potência e taxa de dose. Para tanto foram utilizadas duas bases dados: uma contendo dados gerados por um modelo teórico do reator, e outra contendo dados referentes a uma semana típica de operação. Na segunda etapa, as redes treinadas para realizar a monitoração das variáveis, foram testadas com uma base de dados contendo falhas inseridas artificialmente nos sensores de temperatura. Como o limite máximo de erro de calibração para termopares especiais é de , foram inseridas falhas de ± nos sensores responsáveis pela leitura das variáveis T3 e T4. Na terceira etapa foi desenvolvido um Sistema Fuzzy para realizar o diagnóstico de falhas, onde foram consideradas 3 condições possíveis de falhas: condição normal, falha de −, e falha de , sendo que o sistema desenvolvido indicará qual o sensor de temperatura está com falha. Cº5,0±Cº1Cº1Cº1+ / The increasing demand on quality in production processes has encouraged the development of several studies on Monitoring and Diagnosis Systems in industrial plant, where the interruption of the production due to some unexpected change can bring risk to the operator\'s security besides provoking economic losses, increasing the costs to repair some damaged equipment. Because of these two points, the economic losses and the operator\'s security, it becomes necessary to implement Monitoring and Diagnosis Systems. In this work, a Monitoring and Diagnosis Systems was developed based on the Artificial Neural Networks methodology. This methodology was applied to the IEA-R1 research reactor at IPEN. The development of this system was divided in three stages: the first was dedicated to monitoring, the second to the detection and the third to diagnosis of failures. In the first stage, several Artificial Neural Networks were trained to monitor the temperature variables, nuclear power and dose rate. Two databases were used: one with data generated by a theoretical model and another one with data to a typical week of operation of the IEA-R1 reactor. In the second stage, the neural networks used to monitor the variables was tested with a fault database. The faults were inserted artificially in the sensors signals. As the value of the maximum calibration error for special thermocouples is , it had been inserted faults of in the sensors for the reading of the variables T3 and T4. In the third stage a Fuzzy System was developed to carry out the faults diagnosis, where were considered three conditions: a normal condition, a fault of , and a fault of . This system will indicate which thermocouple is faulty. Cº5,0±Cº1Cº1±−Cº1+

Le secteur de l’énergie éolienne est en pleine expansion depuis les 10 dernières années. Le nombre et la taille des éoliennes ont été multipliés, ce qui accroît la difficulté et la criticité de la maintenance, et impose aux industriels de passer d’une maintenance corrective et systématique à une maintenance conditionnelle et prédictive. L'objectif de ces travaux est de développer des indicateurs de dégradation, en utilisant les données numériques fournies par le SCADA, disponibles à faible coût mais enregistrées à une faible fréquence d'échantillonnage (10 min) dans un objectif de suivi de production. Une analyse bibliographique approfondie des travaux réalisés sur la surveillance de parcs éoliens à partir de données SCADA montre que deux types d’approches sont généralement proposés. Les méthodes dites mono-turbines où un modèle de comportement d’une turbine est appris sur des périodes de bon fonctionnement, à partir duquel il est possible de créer des résidus mesurant la différence entre la valeur prédite par le modèle et la mesure en ligne, qui servent d’indicateurs de défaut. Les modèles mono-turbines ont la particularité d’utiliser des variables provenant de la même turbine du parc. Les deuxièmes méthodes, dites multi-turbines, sont des méthodes où l'effet parc et la similarité entre machines sont utilisés. Là où les recherches les plus récentes proposent principalement de créer des courbes de performances pour chaque machine du parc pendant une période de temps et de comparer ces courbes entre elles, nous faisons la proposition originale de combiner les deux approches et de comparer les résidus mono-turbines à une référence parc traduisant le comportement majoritaire des turbines du parc. Nous validons de manière extensive ces indicateurs en analysant leurs performances sur une base de données composée d’enregistrements de variables SCADA d’une durée de 4 ans sur un parc de 6 machines. Nous proposons aussi des critères de performances pertinents permettant d’évaluer de manière réaliste les gains et éventuels surcoûts que généreraient ces indicateurs s’ils étaient intégrés dans un outil de maintenance. Ainsi, nous montrons que le taux d’interventions inutiles associées à des fausses alarmes produites par les indicateurs de défaut, et qui provoquent un surcoût très important pour l’entreprise, peut être fortement diminué par la fusion d’indicateurs mono-turbines que nous proposons, tout en conservant une avance à la détection suffisante pour planifier la mise en place d’interventions par les équipes de maintenance. / The wind energy sector has rapidly gown in the last 10 years. The number and the size of wind turbines have multiplied, which increases the difficulty and the criticality of the maintenance, and forces the wind turbine industry to change from a corrective and systematic maintenance to a conditional and predictive maintenance. The objective of this research is to develop failure indicators using numerical SCADA data, available at a low price but with a very low sampling frequency (10 min), in order to make online monitoring.A thorough bibliographical analysis on the surveillance of wind farms using SCADA data shows that two types of approaches are usually suggested. The first approach, called mono-turbine, where a good behaviour model of a turbine is learnt over unfaulty periods. With this approach, it is possible to create residuals measuring the difference between the predicted value by the model and the on-line measure, which serves as failure indicators. The mono-turbine models have the peculiarity in that they use variables coming from the same turbine as the farms. The second approach, called multi-turbine, are methods where the similarity between machines is used. Where the most recent researches mostly suggest creating performance curves for every machine on the farm during a period of time and comparing these curves between each other, we make the original proposal to combine both approaches and compare mono-turbine residuals with a farm reference representing the behaviour of the turbines of the farm.We validate in an extensive way those failure indicators by analysing their performances on a database made up of SCADA variable recordings of a duration of 4 years on a windfarm of 6 machines. We also propose relevant performance criteria allowing an estimation in a realistic way of the gains and possible additional costs which would generate these indicators if they were integrated into a tool of maintenance. Therefore, we show that the rate of useless interventions associated with false alarms produced by the failure indicators, which cause a heavy additional cost for the company, can be strongly decreased by the mono-turbines indicators merging that we propose, while preserving a sufficient detection time for the maintenance teams to plan interventions.